* [patch,avr, 0/3] Support 64-bit (long) double.
@ 2019-12-16 16:43 Georg-Johann Lay
2019-12-16 16:46 ` [patch,avr, 1/3] Support 64-bit (long) double: The gcc part Georg-Johann Lay
` (5 more replies)
0 siblings, 6 replies; 22+ messages in thread
From: Georg-Johann Lay @ 2019-12-16 16:43 UTC (permalink / raw)
To: GCC Patches
Now that the avr backend can support 64-bit floats by means of
configure-options --with-double= and --with-long-double=, this patch
series adds some routines to support it.
It's an ad-hoc, avr-specific implementation in assembly and GNU-C which
is added as a new subfolder in libgcc/config/avr/libf7.
Patch 1/3 is the GCC changes: Documentation and new avr-specific
configure options:
--with-libf7 selects to which level double support from libf7 is added
to libgcc.
--with-double-comparison select what FLOAT_LIB_COMPARE_RETURNS_BOOL returns.
I wrote the libf7 code from scratch and put it under GPL v3 + library
exception, so it should be no problem to have it as part of libgcc.
Patch 2/3 is the libgcc additions:
--with-libf7 selects which makefile-snips from libf7 to use.
Patch 3/3 is the actual libf7 implementation. A great deal of which is
assembly, together with C + inline assembly for higher routines.
Ok for trunk?
Johann
^ permalink raw reply [flat|nested] 22+ messages in thread* [patch,avr, 1/3] Support 64-bit (long) double: The gcc part. 2019-12-16 16:43 [patch,avr, 0/3] Support 64-bit (long) double Georg-Johann Lay @ 2019-12-16 16:46 ` Georg-Johann Lay 2019-12-28 13:20 ` [PING^1][patch,avr, " Georg-Johann Lay ` (2 more replies) 2019-12-16 16:49 ` [patch,avr, 2/3] Support 64-bit (long) double: The libgcc changes Georg-Johann Lay ` (4 subsequent siblings) 5 siblings, 3 replies; 22+ messages in thread From: Georg-Johann Lay @ 2019-12-16 16:46 UTC (permalink / raw) To: GCC Patches [-- Attachment #1: Type: text/plain, Size: 853 bytes --] Am 16.12.19 um 17:40 schrieb Georg-Johann Lay: Patch 1/3 is the GCC changes: Documentation and new avr-specific configure options: --with-libf7 selects to which level double support from libf7 is added to libgcc. --with-double-comparison select what FLOAT_LIB_COMPARE_RETURNS_BOOL returns. Johann gcc/ * config.gcc (tm_defines) [target=avr]: Support --with-libf7, --with-double-comparison. * doc/install.texi: Document them. * config/avr/avr-c.c (avr_cpu_cpp_builtins) <WITH_LIBF7_LIBGCC, WITH_LIBF7_MATH, WITH_LIBF7_MATH_SYMBOLS> <WITH_DOUBLE_COMPARISON>: New built-in defines. * doc/invoke.texi (AVR Built-in Macros): Document them. * config/avr/avr-protos.h (avr_float_lib_compare_returns_bool): New. * config/avr/avr.c (avr_float_lib_compare_returns_bool): New function. * config/avr/avr.h (FLOAT_LIB_COMPARE_RETURNS_BOOL): New macro. [-- Attachment #2: m64-gcc.diff --] [-- Type: text/x-patch, Size: 8726 bytes --] Index: gcc/config/avr/avr-c.c =================================================================== --- gcc/config/avr/avr-c.c (revision 278667) +++ gcc/config/avr/avr-c.c (working copy) @@ -390,6 +390,20 @@ start address. This macro shall be used cpp_define (pfile, "__WITH_AVRLIBC__"); #endif /* WITH_AVRLIBC */ + // From configure --with-libf7={|libgcc|math|math-symbols|yes|no} + +#ifdef WITH_LIBF7_LIBGCC + cpp_define (pfile, "__WITH_LIBF7_LIBGCC__"); +#endif /* WITH_LIBF7_LIBGCC */ + +#ifdef WITH_LIBF7_MATH + cpp_define (pfile, "__WITH_LIBF7_MATH__"); +#endif /* WITH_LIBF7_MATH */ + +#ifdef WITH_LIBF7_MATH_SYMBOLS + cpp_define (pfile, "__WITH_LIBF7_MATH_SYMBOLS__"); +#endif /* WITH_LIBF7_MATH_SYMBOLS */ + // From configure --with-double={|32|32,64|64,32|64} #ifdef HAVE_DOUBLE_MULTILIB @@ -438,7 +452,23 @@ start address. This macro shall be used #error "align this with config.gcc" #endif - + // From configure --with-double-comparison={2|3} --with-libf7. + +#if defined (WITH_DOUBLE_COMPARISON) +#if WITH_DOUBLE_COMPARISON == 2 || WITH_DOUBLE_COMPARISON == 3 + /* The number of states a DFmode comparison libcall might take and + reflects what avr.c:FLOAT_LIB_COMPARE_RETURNS_BOOL returns for + DFmode. GCC's default is 3-state, but some libraries like libf7 + implement true / false (2-state). */ + cpp_define_formatted (pfile, "__WITH_DOUBLE_COMPARISON__=%d", + WITH_DOUBLE_COMPARISON); +#else +#error "align this with config.gcc" +#endif +#else +#error "align this with config.gcc" +#endif + /* Define builtin macros so that the user can easily query whether non-generic address spaces (and which) are supported or not. This is only supported for C. For C++, a language extension is needed Index: gcc/config/avr/avr-protos.h =================================================================== --- gcc/config/avr/avr-protos.h (revision 278667) +++ gcc/config/avr/avr-protos.h (working copy) @@ -128,6 +128,8 @@ extern bool avr_xload_libgcc_p (machine_ extern rtx avr_eval_addr_attrib (rtx x); extern bool avr_casei_sequence_check_operands (rtx *xop); +extern bool avr_float_lib_compare_returns_bool (machine_mode, enum rtx_code); + static inline unsigned regmask (machine_mode mode, unsigned regno) { Index: gcc/config/avr/avr.c =================================================================== --- gcc/config/avr/avr.c (revision 278667) +++ gcc/config/avr/avr.c (working copy) @@ -14575,6 +14575,23 @@ avr_fold_builtin (tree fndecl, int n_arg return NULL_TREE; } + +/* Worker function for `FLOAT_LIB_COMPARE_RETURNS_BOOL'. */ + +bool +avr_float_lib_compare_returns_bool (machine_mode mode, enum rtx_code) +{ + if (mode == DFmode) + { +#if WITH_DOUBLE_COMPARISON == 2 + return true; +#endif + } + + // This is the GCC default and also what AVR-LibC implements. + return false; +} + \f /* Initialize the GCC target structure. */ Index: gcc/config/avr/avr.h =================================================================== --- gcc/config/avr/avr.h (revision 278667) +++ gcc/config/avr/avr.h (working copy) @@ -107,6 +107,9 @@ These two properties are reflected by bu #define BYTES_BIG_ENDIAN 0 #define WORDS_BIG_ENDIAN 0 +#define FLOAT_LIB_COMPARE_RETURNS_BOOL(mode, comparison) \ + avr_float_lib_compare_returns_bool (mode, comparison) + #ifdef IN_LIBGCC2 /* This is to get correct SI and DI modes in libgcc2.c (32 and 64 bits). */ #define UNITS_PER_WORD 4 Index: gcc/config.gcc =================================================================== --- gcc/config.gcc (revision 278552) +++ gcc/config.gcc (working copy) @@ -1303,6 +1303,46 @@ avr-*-*) tm_file="${tm_file} ${cpu_type}/avrlibc.h" tm_defines="${tm_defines} WITH_AVRLIBC" fi + # Work out avr_double_comparison which is 2 or 3 and is used in + # target hook FLOAT_LIB_COMPARE_RETURNS_BOOL to determine whether + # DFmode comparisons return 3-state or 2-state results. + case y${with_double_comparison} in + y | ytristate) + avr_double_comparison=3 + ;; + ybool | ylibf7) + avr_double_comparison=2 + ;; + *) + echo "Error: --with-double-comparison= can only be used with: 'tristate', 'bool', 'libf7'" 1>&2 + exit 1 + ;; + esac + case "y${with_libf7}" in + yno) + # avr_double_comparison as set above. + ;; + ylibgcc) + avr_double_comparison=2 + tm_defines="${tm_defines} WITH_LIBF7_LIBGCC" + ;; + y | yyes | ymath-symbols) + avr_double_comparison=2 + tm_defines="${tm_defines} WITH_LIBF7_LIBGCC" + tm_defines="${tm_defines} WITH_LIBF7_MATH" + tm_defines="${tm_defines} WITH_LIBF7_MATH_SYMBOLS" + ;; + ymath) + avr_double_comparison=2 + tm_defines="${tm_defines} WITH_LIBF7_LIBGCC" + tm_defines="${tm_defines} WITH_LIBF7_MATH" + ;; + *) + echo "Error: --with-libf7=${with_libf7} but can only be used with: 'libgcc', 'math', 'math-symbols', 'yes', 'no'" 1>&2 + exit 1 + ;; + esac + tm_defines="${tm_defines} WITH_DOUBLE_COMPARISON=${avr_double_comparison}" case y${with_double} in y | y32) avr_double=32 @@ -1332,7 +1372,7 @@ avr-*-*) ;; esac case y${with_long_double} in - y | y32) + y32) avr_long_double=32 tm_defines="${tm_defines} HAVE_LONG_DOUBLE32" ;; @@ -1340,7 +1380,7 @@ avr-*-*) avr_long_double=64 tm_defines="${tm_defines} HAVE_LONG_DOUBLE64" ;; - y64,32) + y | y64,32) avr_long_double=64 avr_long_double_multilib=1 tm_defines="${tm_defines} HAVE_LONG_DOUBLE32" Index: gcc/doc/install.texi =================================================================== --- gcc/doc/install.texi (revision 278552) +++ gcc/doc/install.texi (working copy) @@ -2306,9 +2306,10 @@ as a multilib option. If @option{--with-long-double=double} is specified, @samp{double} and @samp{long double} will have the same layout. @item -If the configure option is not set, it defaults to @samp{32} which -is compatible with older versions of the compiler that use non-standard -32-bit types for @samp{double} and @samp{long double}. +If the configure option is not set, @option{-mdouble=} defaults to @samp{32} +which is compatible with older versions of the compiler that use non-standard +32-bit types for @samp{double}. The default for @option{-mlong-double=} +is 64. @end itemize Not all combinations of @option{--with-double=} and @option{--with-long-double=} are valid. For example, the combination @@ -2318,6 +2319,28 @@ multilibs for @samp{double}, whereas the that @samp{long double} --- and hence also @samp{double} --- is always 32@tie{}bits wide. +@item --with-double-comparison=@{tristate|3|bool|2|libf7@} +Only supported for the AVR target since version@tie{}10. +Specify what result format is returned by library functions that +compare 64-bit floating point values (@code{DFmode}). +The GCC default is @samp{tristate}. If the floating point +implementation returns a boolean instead, set it to @samp{bool}. + +@item --with-libf7=@{libgcc|math|math-symbols|no@} +Only supported for the AVR target since version@tie{}10. +Specify to which degree code from Libf7 is included in libgcc. +Libf7 is an ad-hoc, AVR-specific, 64-bit floating point emulation +written in C and (inline) assembly. @samp{libgcc} adds support +for functions that one would usually expect in libgcc like double addition, +double comparisons and double conversions. @samp{math} also adds routines +that one would expect in @file{libm.a}, but with @code{__} (two underscores) +prepended to the symbol names as specified by @file{math.h}. +@samp{math-symbols} also defines weak aliases for the functions +declared in @file{math.h}. However, @code{--with-libf7} won't +install no @file{math.h} header file whatsoever, this file must come +from elsewhere. This option sets @option{--with-double-comparison} +to @samp{bool}. + @item --with-nds32-lib=@var{library} Specifies that @var{library} setting is used for building @file{libgcc.a}. Currently, the valid @var{library} is @samp{newlib} or @samp{mculib}. Index: gcc/doc/invoke.texi =================================================================== --- gcc/doc/invoke.texi (revision 278552) +++ gcc/doc/invoke.texi (working copy) @@ -18772,6 +18772,15 @@ features like attribute @code{progmem} a The compiler is configured to be used together with AVR-Libc. See the @option{--with-avrlibc} configure option. +@item __WITH_LIBF7_LIBGCC__ +@itemx __WITH_LIBF7_MATH__ +@itemx __WITH_LIBF7_MATH_SYMBOLS__ +Reflects the @code{--with-libf7=@{libgcc|math|math-symbols@}} +configure option, see +@code{--with-libf7=math}, @code{--with-libf7=math-symbols} was +specified, respectively, see +@uref{http://gcc.gnu.org/@/install/@/configure.html#avr}. + @end table @node Blackfin Options ^ permalink raw reply [flat|nested] 22+ messages in thread
* [PING^1][patch,avr, 1/3] Support 64-bit (long) double: The gcc part. 2019-12-16 16:46 ` [patch,avr, 1/3] Support 64-bit (long) double: The gcc part Georg-Johann Lay @ 2019-12-28 13:20 ` Georg-Johann Lay 2020-01-06 13:08 ` [Ping^2][patch,avr, " Georg-Johann Lay 2020-01-06 16:25 ` [patch,avr, " Jeff Law 2 siblings, 0 replies; 22+ messages in thread From: Georg-Johann Lay @ 2019-12-28 13:20 UTC (permalink / raw) To: GCC Patches; +Cc: Denis Chertykov Ping #1 > Am 16.12.19 um 17:40 schrieb Georg-Johann Lay: > Patch 1/3 is the GCC changes: Documentation and new avr-specific > configure options: > > --with-libf7 selects to which level double support from libf7 is added > to libgcc. > > --with-double-comparison select what FLOAT_LIB_COMPARE_RETURNS_BOOL > returns. > > Johann > > gcc/ > * config.gcc (tm_defines) [target=avr]: Support --with-libf7, > --with-double-comparison. > * doc/install.texi: Document them. > * config/avr/avr-c.c (avr_cpu_cpp_builtins) > <WITH_LIBF7_LIBGCC, WITH_LIBF7_MATH, WITH_LIBF7_MATH_SYMBOLS> > <WITH_DOUBLE_COMPARISON>: New built-in defines. > * doc/invoke.texi (AVR Built-in Macros): Document them. > * config/avr/avr-protos.h (avr_float_lib_compare_returns_bool): New. > * config/avr/avr.c (avr_float_lib_compare_returns_bool): New function. > * config/avr/avr.h (FLOAT_LIB_COMPARE_RETURNS_BOOL): New macro. > ^ permalink raw reply [flat|nested] 22+ messages in thread
* [Ping^2][patch,avr, 1/3] Support 64-bit (long) double: The gcc part. 2019-12-16 16:46 ` [patch,avr, 1/3] Support 64-bit (long) double: The gcc part Georg-Johann Lay 2019-12-28 13:20 ` [PING^1][patch,avr, " Georg-Johann Lay @ 2020-01-06 13:08 ` Georg-Johann Lay 2020-01-06 16:25 ` [patch,avr, " Jeff Law 2 siblings, 0 replies; 22+ messages in thread From: Georg-Johann Lay @ 2020-01-06 13:08 UTC (permalink / raw) To: GCC Patches [-- Attachment #1: Type: text/plain, Size: 968 bytes --] Ping #2 Georg-Johann Lay schrieb: > Am 16.12.19 um 17:40 schrieb Georg-Johann Lay: > Patch 1/3 is the GCC changes: Documentation and new avr-specific > configure options: > > --with-libf7 selects to which level double support from libf7 is added > to libgcc. > > --with-double-comparison select what FLOAT_LIB_COMPARE_RETURNS_BOOL > returns. > > Johann > > gcc/ > * config.gcc (tm_defines) [target=avr]: Support --with-libf7, > --with-double-comparison. > * doc/install.texi: Document them. > * config/avr/avr-c.c (avr_cpu_cpp_builtins) > <WITH_LIBF7_LIBGCC, WITH_LIBF7_MATH, WITH_LIBF7_MATH_SYMBOLS> > <WITH_DOUBLE_COMPARISON>: New built-in defines. > * doc/invoke.texi (AVR Built-in Macros): Document them. > * config/avr/avr-protos.h (avr_float_lib_compare_returns_bool): New. > * config/avr/avr.c (avr_float_lib_compare_returns_bool): New function. > * config/avr/avr.h (FLOAT_LIB_COMPARE_RETURNS_BOOL): New macro. > [-- Attachment #2: m64-gcc.diff --] [-- Type: text/plain, Size: 8967 bytes --] Index: gcc/config/avr/avr-c.c =================================================================== --- gcc/config/avr/avr-c.c (revision 278667) +++ gcc/config/avr/avr-c.c (working copy) @@ -390,6 +390,20 @@ start address. This macro shall be used cpp_define (pfile, "__WITH_AVRLIBC__"); #endif /* WITH_AVRLIBC */ + // From configure --with-libf7={|libgcc|math|math-symbols|yes|no} + +#ifdef WITH_LIBF7_LIBGCC + cpp_define (pfile, "__WITH_LIBF7_LIBGCC__"); +#endif /* WITH_LIBF7_LIBGCC */ + +#ifdef WITH_LIBF7_MATH + cpp_define (pfile, "__WITH_LIBF7_MATH__"); +#endif /* WITH_LIBF7_MATH */ + +#ifdef WITH_LIBF7_MATH_SYMBOLS + cpp_define (pfile, "__WITH_LIBF7_MATH_SYMBOLS__"); +#endif /* WITH_LIBF7_MATH_SYMBOLS */ + // From configure --with-double={|32|32,64|64,32|64} #ifdef HAVE_DOUBLE_MULTILIB @@ -438,7 +452,23 @@ start address. This macro shall be used #error "align this with config.gcc" #endif - + // From configure --with-double-comparison={2|3} --with-libf7. + +#if defined (WITH_DOUBLE_COMPARISON) +#if WITH_DOUBLE_COMPARISON == 2 || WITH_DOUBLE_COMPARISON == 3 + /* The number of states a DFmode comparison libcall might take and + reflects what avr.c:FLOAT_LIB_COMPARE_RETURNS_BOOL returns for + DFmode. GCC's default is 3-state, but some libraries like libf7 + implement true / false (2-state). */ + cpp_define_formatted (pfile, "__WITH_DOUBLE_COMPARISON__=%d", + WITH_DOUBLE_COMPARISON); +#else +#error "align this with config.gcc" +#endif +#else +#error "align this with config.gcc" +#endif + /* Define builtin macros so that the user can easily query whether non-generic address spaces (and which) are supported or not. This is only supported for C. For C++, a language extension is needed Index: gcc/config/avr/avr-protos.h =================================================================== --- gcc/config/avr/avr-protos.h (revision 278667) +++ gcc/config/avr/avr-protos.h (working copy) @@ -128,6 +128,8 @@ extern bool avr_xload_libgcc_p (machine_ extern rtx avr_eval_addr_attrib (rtx x); extern bool avr_casei_sequence_check_operands (rtx *xop); +extern bool avr_float_lib_compare_returns_bool (machine_mode, enum rtx_code); + static inline unsigned regmask (machine_mode mode, unsigned regno) { Index: gcc/config/avr/avr.c =================================================================== --- gcc/config/avr/avr.c (revision 278667) +++ gcc/config/avr/avr.c (working copy) @@ -14575,6 +14575,23 @@ avr_fold_builtin (tree fndecl, int n_arg return NULL_TREE; } + +/* Worker function for `FLOAT_LIB_COMPARE_RETURNS_BOOL'. */ + +bool +avr_float_lib_compare_returns_bool (machine_mode mode, enum rtx_code) +{ + if (mode == DFmode) + { +#if WITH_DOUBLE_COMPARISON == 2 + return true; +#endif + } + + // This is the GCC default and also what AVR-LibC implements. + return false; +} + \f /* Initialize the GCC target structure. */ Index: gcc/config/avr/avr.h =================================================================== --- gcc/config/avr/avr.h (revision 278667) +++ gcc/config/avr/avr.h (working copy) @@ -107,6 +107,9 @@ These two properties are reflected by bu #define BYTES_BIG_ENDIAN 0 #define WORDS_BIG_ENDIAN 0 +#define FLOAT_LIB_COMPARE_RETURNS_BOOL(mode, comparison) \ + avr_float_lib_compare_returns_bool (mode, comparison) + #ifdef IN_LIBGCC2 /* This is to get correct SI and DI modes in libgcc2.c (32 and 64 bits). */ #define UNITS_PER_WORD 4 Index: gcc/config.gcc =================================================================== --- gcc/config.gcc (revision 278552) +++ gcc/config.gcc (working copy) @@ -1303,6 +1303,46 @@ avr-*-*) tm_file="${tm_file} ${cpu_type}/avrlibc.h" tm_defines="${tm_defines} WITH_AVRLIBC" fi + # Work out avr_double_comparison which is 2 or 3 and is used in + # target hook FLOAT_LIB_COMPARE_RETURNS_BOOL to determine whether + # DFmode comparisons return 3-state or 2-state results. + case y${with_double_comparison} in + y | ytristate) + avr_double_comparison=3 + ;; + ybool | ylibf7) + avr_double_comparison=2 + ;; + *) + echo "Error: --with-double-comparison= can only be used with: 'tristate', 'bool', 'libf7'" 1>&2 + exit 1 + ;; + esac + case "y${with_libf7}" in + yno) + # avr_double_comparison as set above. + ;; + ylibgcc) + avr_double_comparison=2 + tm_defines="${tm_defines} WITH_LIBF7_LIBGCC" + ;; + y | yyes | ymath-symbols) + avr_double_comparison=2 + tm_defines="${tm_defines} WITH_LIBF7_LIBGCC" + tm_defines="${tm_defines} WITH_LIBF7_MATH" + tm_defines="${tm_defines} WITH_LIBF7_MATH_SYMBOLS" + ;; + ymath) + avr_double_comparison=2 + tm_defines="${tm_defines} WITH_LIBF7_LIBGCC" + tm_defines="${tm_defines} WITH_LIBF7_MATH" + ;; + *) + echo "Error: --with-libf7=${with_libf7} but can only be used with: 'libgcc', 'math', 'math-symbols', 'yes', 'no'" 1>&2 + exit 1 + ;; + esac + tm_defines="${tm_defines} WITH_DOUBLE_COMPARISON=${avr_double_comparison}" case y${with_double} in y | y32) avr_double=32 @@ -1332,7 +1372,7 @@ avr-*-*) ;; esac case y${with_long_double} in - y | y32) + y32) avr_long_double=32 tm_defines="${tm_defines} HAVE_LONG_DOUBLE32" ;; @@ -1340,7 +1380,7 @@ avr-*-*) avr_long_double=64 tm_defines="${tm_defines} HAVE_LONG_DOUBLE64" ;; - y64,32) + y | y64,32) avr_long_double=64 avr_long_double_multilib=1 tm_defines="${tm_defines} HAVE_LONG_DOUBLE32" Index: gcc/doc/install.texi =================================================================== --- gcc/doc/install.texi (revision 278552) +++ gcc/doc/install.texi (working copy) @@ -2306,9 +2306,10 @@ as a multilib option. If @option{--with-long-double=double} is specified, @samp{double} and @samp{long double} will have the same layout. @item -If the configure option is not set, it defaults to @samp{32} which -is compatible with older versions of the compiler that use non-standard -32-bit types for @samp{double} and @samp{long double}. +If the configure option is not set, @option{-mdouble=} defaults to @samp{32} +which is compatible with older versions of the compiler that use non-standard +32-bit types for @samp{double}. The default for @option{-mlong-double=} +is 64. @end itemize Not all combinations of @option{--with-double=} and @option{--with-long-double=} are valid. For example, the combination @@ -2318,6 +2319,28 @@ multilibs for @samp{double}, whereas the that @samp{long double} --- and hence also @samp{double} --- is always 32@tie{}bits wide. +@item --with-double-comparison=@{tristate|3|bool|2|libf7@} +Only supported for the AVR target since version@tie{}10. +Specify what result format is returned by library functions that +compare 64-bit floating point values (@code{DFmode}). +The GCC default is @samp{tristate}. If the floating point +implementation returns a boolean instead, set it to @samp{bool}. + +@item --with-libf7=@{libgcc|math|math-symbols|no@} +Only supported for the AVR target since version@tie{}10. +Specify to which degree code from Libf7 is included in libgcc. +Libf7 is an ad-hoc, AVR-specific, 64-bit floating point emulation +written in C and (inline) assembly. @samp{libgcc} adds support +for functions that one would usually expect in libgcc like double addition, +double comparisons and double conversions. @samp{math} also adds routines +that one would expect in @file{libm.a}, but with @code{__} (two underscores) +prepended to the symbol names as specified by @file{math.h}. +@samp{math-symbols} also defines weak aliases for the functions +declared in @file{math.h}. However, @code{--with-libf7} won't +install no @file{math.h} header file whatsoever, this file must come +from elsewhere. This option sets @option{--with-double-comparison} +to @samp{bool}. + @item --with-nds32-lib=@var{library} Specifies that @var{library} setting is used for building @file{libgcc.a}. Currently, the valid @var{library} is @samp{newlib} or @samp{mculib}. Index: gcc/doc/invoke.texi =================================================================== --- gcc/doc/invoke.texi (revision 278552) +++ gcc/doc/invoke.texi (working copy) @@ -18772,6 +18772,15 @@ features like attribute @code{progmem} a The compiler is configured to be used together with AVR-Libc. See the @option{--with-avrlibc} configure option. +@item __WITH_LIBF7_LIBGCC__ +@itemx __WITH_LIBF7_MATH__ +@itemx __WITH_LIBF7_MATH_SYMBOLS__ +Reflects the @code{--with-libf7=@{libgcc|math|math-symbols@}} +configure option, see +@code{--with-libf7=math}, @code{--with-libf7=math-symbols} was +specified, respectively, see +@uref{http://gcc.gnu.org/@/install/@/configure.html#avr}. + @end table @node Blackfin Options ^ permalink raw reply [flat|nested] 22+ messages in thread
* Re: [patch,avr, 1/3] Support 64-bit (long) double: The gcc part. 2019-12-16 16:46 ` [patch,avr, 1/3] Support 64-bit (long) double: The gcc part Georg-Johann Lay 2019-12-28 13:20 ` [PING^1][patch,avr, " Georg-Johann Lay 2020-01-06 13:08 ` [Ping^2][patch,avr, " Georg-Johann Lay @ 2020-01-06 16:25 ` Jeff Law 2020-01-06 17:10 ` Georg-Johann Lay 2 siblings, 1 reply; 22+ messages in thread From: Jeff Law @ 2020-01-06 16:25 UTC (permalink / raw) To: Georg-Johann Lay, GCC Patches On Mon, 2019-12-16 at 17:43 +0100, Georg-Johann Lay wrote: > Am 16.12.19 um 17:40 schrieb Georg-Johann Lay: > Patch 1/3 is the GCC changes: Documentation and new avr-specific > configure options: > > --with-libf7 selects to which level double support from libf7 is added > to libgcc. > > --with-double-comparison select what FLOAT_LIB_COMPARE_RETURNS_BOOL > returns. > > Johann > > gcc/ > * config.gcc (tm_defines) [target=avr]: Support --with-libf7, > --with-double-comparison. > * doc/install.texi: Document them. > * config/avr/avr-c.c (avr_cpu_cpp_builtins) > <WITH_LIBF7_LIBGCC, WITH_LIBF7_MATH, WITH_LIBF7_MATH_SYMBOLS> > <WITH_DOUBLE_COMPARISON>: New built-in defines. > * doc/invoke.texi (AVR Built-in Macros): Document them. > * config/avr/avr-protos.h (avr_float_lib_compare_returns_bool): New. > * config/avr/avr.c (avr_float_lib_compare_returns_bool): New function. > * config/avr/avr.h (FLOAT_LIB_COMPARE_RETURNS_BOOL): New macro. I'd suggest "--with-libf77" rather than "--with-libf7". jeff ^ permalink raw reply [flat|nested] 22+ messages in thread
* Re: [patch,avr, 1/3] Support 64-bit (long) double: The gcc part. 2020-01-06 16:25 ` [patch,avr, " Jeff Law @ 2020-01-06 17:10 ` Georg-Johann Lay 2020-01-06 17:16 ` Jeff Law 0 siblings, 1 reply; 22+ messages in thread From: Georg-Johann Lay @ 2020-01-06 17:10 UTC (permalink / raw) To: law; +Cc: GCC Patches Jeff Law schrieb: > On Mon, 2019-12-16 at 17:43 +0100, Georg-Johann Lay wrote: >> Am 16.12.19 um 17:40 schrieb Georg-Johann Lay: >> Patch 1/3 is the GCC changes: Documentation and new avr-specific >> configure options: >> >> --with-libf7 selects to which level double support from libf7 is added >> to libgcc. >> >> --with-double-comparison select what FLOAT_LIB_COMPARE_RETURNS_BOOL >> returns. >> >> Johann >> >> gcc/ >> * config.gcc (tm_defines) [target=avr]: Support --with-libf7, >> --with-double-comparison. >> * doc/install.texi: Document them. >> * config/avr/avr-c.c (avr_cpu_cpp_builtins) >> <WITH_LIBF7_LIBGCC, WITH_LIBF7_MATH, WITH_LIBF7_MATH_SYMBOLS> >> <WITH_DOUBLE_COMPARISON>: New built-in defines. >> * doc/invoke.texi (AVR Built-in Macros): Document them. >> * config/avr/avr-protos.h (avr_float_lib_compare_returns_bool): New. >> * config/avr/avr.c (avr_float_lib_compare_returns_bool): New function. >> * config/avr/avr.h (FLOAT_LIB_COMPARE_RETURNS_BOOL): New macro. > I'd suggest "--with-libf77" rather than "--with-libf7". > > jeff Why that second 7? I called it "libf7" because the internal _f_loat representation has a mantissa of 7 bytes. So libf56 would also be indicative. Actually the name does not matter very much... It just needs to have /some/ name. "f77" however puts it close to Fortran, but that code has absolutely nothing to do with Fortran whatsoever. Johann p.s. You know anything about Denis Chertykov? He used to reply to avr patches within hours, but my latest patches (after a long period where I didn't propose any patches) where all approved by you; not a single mail from Denis. So I am concerned if he's well. Maybe he left you or some other global maintainer a note? He's still listed as maintainer though. ^ permalink raw reply [flat|nested] 22+ messages in thread
* Re: [patch,avr, 1/3] Support 64-bit (long) double: The gcc part. 2020-01-06 17:10 ` Georg-Johann Lay @ 2020-01-06 17:16 ` Jeff Law 2020-12-15 7:40 ` AVR maintainership (was: [patch,avr, 1/3] Support 64-bit (long) double: The gcc part.) Gerald Pfeifer 0 siblings, 1 reply; 22+ messages in thread From: Jeff Law @ 2020-01-06 17:16 UTC (permalink / raw) To: Georg-Johann Lay; +Cc: GCC Patches On Mon, 2020-01-06 at 18:12 +0100, Georg-Johann Lay wrote: > Jeff Law schrieb: > > On Mon, 2019-12-16 at 17:43 +0100, Georg-Johann Lay wrote: > > > Am 16.12.19 um 17:40 schrieb Georg-Johann Lay: > > > Patch 1/3 is the GCC changes: Documentation and new avr-specific > > > configure options: > > > > > > --with-libf7 selects to which level double support from libf7 is added > > > to libgcc. > > > > > > --with-double-comparison select what FLOAT_LIB_COMPARE_RETURNS_BOOL > > > returns. > > > > > > Johann > > > > > > gcc/ > > > * config.gcc (tm_defines) [target=avr]: Support --with-libf7, > > > --with-double-comparison. > > > * doc/install.texi: Document them. > > > * config/avr/avr-c.c (avr_cpu_cpp_builtins) > > > <WITH_LIBF7_LIBGCC, WITH_LIBF7_MATH, WITH_LIBF7_MATH_SYMBOLS> > > > <WITH_DOUBLE_COMPARISON>: New built-in defines. > > > * doc/invoke.texi (AVR Built-in Macros): Document them. > > > * config/avr/avr-protos.h (avr_float_lib_compare_returns_bool): New. > > > * config/avr/avr.c (avr_float_lib_compare_returns_bool): New function. > > > * config/avr/avr.h (FLOAT_LIB_COMPARE_RETURNS_BOOL): New macro. > > I'd suggest "--with-libf77" rather than "--with-libf7". > > > > jeff > > Why that second 7? Because I incorrectly thought you were referring to the Fortran library. Jakub corrected me on an internal IRC. Ignore my comments about changing the name. > p.s. You know anything about Denis Chertykov? He used to reply to avr > patches within hours, but my latest patches (after a long period where I > didn't propose any patches) where all approved by you; not a single mail > from Denis. So I am concerned if he's well. Maybe he left you or some > other global maintainer a note? He's still listed as maintainer though. I haven't heard much from him for a while. I'm trying to leave enough space that he can step in on this stuff, but I'm also hesitant to let patches linger for too long. jeff > ^ permalink raw reply [flat|nested] 22+ messages in thread
* AVR maintainership (was: [patch,avr, 1/3] Support 64-bit (long) double: The gcc part.) 2020-01-06 17:16 ` Jeff Law @ 2020-12-15 7:40 ` Gerald Pfeifer 2020-12-15 14:30 ` AVR maintainership (was: [patch, avr, " Segher Boessenkool 0 siblings, 1 reply; 22+ messages in thread From: Gerald Pfeifer @ 2020-12-15 7:40 UTC (permalink / raw) To: Jeff Law, Georg-Johann Lay; +Cc: gcc-patches On Mon, 6 Jan 2020, Jeff Law wrote: >> p.s. You know anything about Denis Chertykov? He used to reply to avr >> patches within hours, but my latest patches (after a long period where I >> didn't propose any patches) where all approved by you; not a single mail >> from Denis. So I am concerned if he's well. Maybe he left you or some >> other global maintainer a note? He's still listed as maintainer though. > I haven't heard much from him for a while. I'm trying to leave enough > space that he can step in on this stuff, but I'm also hesitant to let > patches linger for too long. I don't think I have seen mails from Denis since then (more than eleven months). Do we need to get a new maintainer for AVR in place? Any nominations? Gerald ^ permalink raw reply [flat|nested] 22+ messages in thread
* Re: AVR maintainership (was: [patch, avr, 1/3] Support 64-bit (long) double: The gcc part.) 2020-12-15 7:40 ` AVR maintainership (was: [patch,avr, 1/3] Support 64-bit (long) double: The gcc part.) Gerald Pfeifer @ 2020-12-15 14:30 ` Segher Boessenkool 2020-12-15 20:42 ` AVR maintainership (was: [patch,avr, " Jeff Law 2020-12-18 16:53 ` AVR maintainership Georg-Johann Lay 0 siblings, 2 replies; 22+ messages in thread From: Segher Boessenkool @ 2020-12-15 14:30 UTC (permalink / raw) To: Gerald Pfeifer; +Cc: Jeff Law, Georg-Johann Lay, gcc-patches On Tue, Dec 15, 2020 at 08:40:08AM +0100, Gerald Pfeifer wrote: > On Mon, 6 Jan 2020, Jeff Law wrote: > >> p.s. You know anything about Denis Chertykov? He used to reply to avr > >> patches within hours, but my latest patches (after a long period where I > >> didn't propose any patches) where all approved by you; not a single mail > >> from Denis. So I am concerned if he's well. Maybe he left you or some > >> other global maintainer a note? He's still listed as maintainer though. > > I haven't heard much from him for a while. I'm trying to leave enough > > space that he can step in on this stuff, but I'm also hesitant to let > > patches linger for too long. > > I don't think I have seen mails from Denis since then (more than > eleven months). > > Do we need to get a new maintainer for AVR in place? Any nominations? I nominate Georg-Johann Lay as the obvious choice, since he has authored 333 of the 511 patches that went into config/avr/ the last ten years. Segher ^ permalink raw reply [flat|nested] 22+ messages in thread
* Re: AVR maintainership (was: [patch,avr, 1/3] Support 64-bit (long) double: The gcc part.) 2020-12-15 14:30 ` AVR maintainership (was: [patch, avr, " Segher Boessenkool @ 2020-12-15 20:42 ` Jeff Law 2020-12-18 16:53 ` AVR maintainership Georg-Johann Lay 1 sibling, 0 replies; 22+ messages in thread From: Jeff Law @ 2020-12-15 20:42 UTC (permalink / raw) To: Segher Boessenkool, Gerald Pfeifer; +Cc: Georg-Johann Lay, gcc-patches On 12/15/20 7:30 AM, Segher Boessenkool wrote: > On Tue, Dec 15, 2020 at 08:40:08AM +0100, Gerald Pfeifer wrote: >> On Mon, 6 Jan 2020, Jeff Law wrote: >>>> p.s. You know anything about Denis Chertykov? He used to reply to avr >>>> patches within hours, but my latest patches (after a long period where I >>>> didn't propose any patches) where all approved by you; not a single mail >>>> from Denis. So I am concerned if he's well. Maybe he left you or some >>>> other global maintainer a note? He's still listed as maintainer though. >>> I haven't heard much from him for a while. I'm trying to leave enough >>> space that he can step in on this stuff, but I'm also hesitant to let >>> patches linger for too long. >> I don't think I have seen mails from Denis since then (more than >> eleven months). >> >> Do we need to get a new maintainer for AVR in place? Any nominations? > I nominate Georg-Johann Lay as the obvious choice, since he has authored > 333 of the 511 patches that went into config/avr/ the last ten years. I'd support that if Georg-Johann wants it. jeff ^ permalink raw reply [flat|nested] 22+ messages in thread
* Re: AVR maintainership 2020-12-15 14:30 ` AVR maintainership (was: [patch, avr, " Segher Boessenkool 2020-12-15 20:42 ` AVR maintainership (was: [patch,avr, " Jeff Law @ 2020-12-18 16:53 ` Georg-Johann Lay 1 sibling, 0 replies; 22+ messages in thread From: Georg-Johann Lay @ 2020-12-18 16:53 UTC (permalink / raw) To: Segher Boessenkool; +Cc: Gerald Pfeifer, Jeff Law, gcc-patches Segher Boessenkool schrieb: > On Tue, Dec 15, 2020 at 08:40:08AM +0100, Gerald Pfeifer wrote: >> On Mon, 6 Jan 2020, Jeff Law wrote: >>>> p.s. You know anything about Denis Chertykov? He used to reply to avr >>>> patches within hours, but my latest patches (after a long period where I >>>> didn't propose any patches) where all approved by you; not a single mail >>>> from Denis. So I am concerned if he's well. Maybe he left you or some >>>> other global maintainer a note? He's still listed as maintainer though. >>> I haven't heard much from him for a while. I'm trying to leave enough >>> space that he can step in on this stuff, but I'm also hesitant to let >>> patches linger for too long. >> I don't think I have seen mails from Denis since then (more than >> eleven months). >> >> Do we need to get a new maintainer for AVR in place? Any nominations? > > I nominate Georg-Johann Lay as the obvious choice, since he has authored > 333 of the 511 patches that went into config/avr/ the last ten years. > > > Segher Thanks, I feel honoured by this proposal. However, I have to decline it and cannot set in as avr maintainer because I do not have respective resources right now. Johann ^ permalink raw reply [flat|nested] 22+ messages in thread
* Re: [patch,avr, 2/3] Support 64-bit (long) double: The libgcc changes. 2019-12-16 16:43 [patch,avr, 0/3] Support 64-bit (long) double Georg-Johann Lay 2019-12-16 16:46 ` [patch,avr, 1/3] Support 64-bit (long) double: The gcc part Georg-Johann Lay @ 2019-12-16 16:49 ` Georg-Johann Lay 2019-12-28 12:00 ` [PING^1][patch,avr, " Georg-Johann Lay ` (2 more replies) 2019-12-16 17:09 ` [patch,avr, 3/3] Support 64-bit (long) double: libf7 Georg-Johann Lay ` (3 subsequent siblings) 5 siblings, 3 replies; 22+ messages in thread From: Georg-Johann Lay @ 2019-12-16 16:49 UTC (permalink / raw) To: GCC Patches [-- Attachment #1: Type: text/plain, Size: 430 bytes --] Am 16.12.19 um 17:40 schrieb Georg-Johann Lay: Patch 2/3 is the libgcc additions: --with-libf7 selects which makefile-snips from libf7 to use. libgcc/ * config.host (tmake_file) [target=avr]: Add t-libf7, t-libf7-math, t-libf7-math-symbols as specified by --with-libf7=. * config/avr/t-avrlibc: Don't copy libgcc.a if there are modules depending on sizeof (double) or sizeof (long double). * config/avr/libf7: New folder. [-- Attachment #2: m64-libgcc.diff --] [-- Type: text/x-patch, Size: 1890 bytes --] Index: libgcc/config.host =================================================================== --- libgcc/config.host (revision 278552) +++ libgcc/config.host (working copy) @@ -514,6 +514,29 @@ arm*-*-eabi* | arm*-*-symbianelf* | arm* avr-*-*) # Make HImode functions for AVR tmake_file="${cpu_type}/t-avr t-fpbit" + # Make some DFmode functions from libf7, part of avr-libgcc. + # This must be prior to adding t-avrlibc. + case "y${with_libf7}" in + yno) + # No libf7 support. + ;; + ylibgcc) + tmake_file="$tmake_file ${cpu_type}/libf7/t-libf7" + ;; + ymath) + tmake_file="$tmake_file ${cpu_type}/libf7/t-libf7-math" + tmake_file="$tmake_file ${cpu_type}/libf7/t-libf7" + ;; + ymath-symbols | yyes | y) + tmake_file="$tmake_file ${cpu_type}/libf7/t-libf7-math-symbols" + tmake_file="$tmake_file ${cpu_type}/libf7/t-libf7-math" + tmake_file="$tmake_file ${cpu_type}/libf7/t-libf7" + ;; + *) + echo "Error: --with-libf7=${with_libf7} but can only be used with: 'libgcc', 'math', 'math-symbols', 'yes', 'no'" 1>&2 + exit 1 + ;; + esac if test x${with_avrlibc} != xno; then tmake_file="$tmake_file ${cpu_type}/t-avrlibc" fi Index: libgcc/config/avr/t-avrlibc =================================================================== --- libgcc/config/avr/t-avrlibc (revision 278992) +++ libgcc/config/avr/t-avrlibc (working copy) @@ -65,6 +65,12 @@ LIB2FUNCS_EXCLUDE += \ _fixunssfdi \ _floatdisf _floatundisf +ifeq (,$(WITH_LIBF7_MATH_SYMBOLS)) + +# No modules depend on __SIZEOF_LONG_DOUBLE__ or __SIZEOF_DOUBLE__ +# which means we might have an opportunity to copy libgcc.a. +# WITH_LIBF7_MATH_SYMBOLS is set by libf7/t-libf7-math-symbols. + ifneq (,$(findstring avr,$(MULTISUBDIR))) # We are not in the avr2 (default) subdir, hence copying will work. @@ -95,3 +101,4 @@ Makefile: t-copy-libgcc.dep endif endif +endif ^ permalink raw reply [flat|nested] 22+ messages in thread
* [PING^1][patch,avr, 2/3] Support 64-bit (long) double: The libgcc changes. 2019-12-16 16:49 ` [patch,avr, 2/3] Support 64-bit (long) double: The libgcc changes Georg-Johann Lay @ 2019-12-28 12:00 ` Georg-Johann Lay 2020-01-06 13:08 ` [Ping^2][patch,avr, " Georg-Johann Lay 2020-01-06 16:26 ` [patch,avr, " Jeff Law 2 siblings, 0 replies; 22+ messages in thread From: Georg-Johann Lay @ 2019-12-28 12:00 UTC (permalink / raw) To: GCC Patches; +Cc: Denis Chertykov Ping #1 > Am 16.12.19 um 17:40 schrieb Georg-Johann Lay: > > Patch 2/3 is the libgcc additions: > > --with-libf7 selects which makefile-snips from libf7 to use. > > libgcc/ > * config.host (tmake_file) [target=avr]: Add t-libf7, > t-libf7-math, t-libf7-math-symbols as specified by --with-libf7=. > * config/avr/t-avrlibc: Don't copy libgcc.a if there are modules > depending on sizeof (double) or sizeof (long double). > * config/avr/libf7: New folder. > ^ permalink raw reply [flat|nested] 22+ messages in thread
* [Ping^2][patch,avr, 2/3] Support 64-bit (long) double: The libgcc changes. 2019-12-16 16:49 ` [patch,avr, 2/3] Support 64-bit (long) double: The libgcc changes Georg-Johann Lay 2019-12-28 12:00 ` [PING^1][patch,avr, " Georg-Johann Lay @ 2020-01-06 13:08 ` Georg-Johann Lay 2020-01-06 16:26 ` [patch,avr, " Jeff Law 2 siblings, 0 replies; 22+ messages in thread From: Georg-Johann Lay @ 2020-01-06 13:08 UTC (permalink / raw) To: GCC Patches; +Cc: Denis Chertykov [-- Attachment #1: Type: text/plain, Size: 508 bytes --] Ping #2 Georg-Johann Lay schrieb: > Am 16.12.19 um 17:40 schrieb Georg-Johann Lay: > > Patch 2/3 is the libgcc additions: > > --with-libf7 selects which makefile-snips from libf7 to use. > > libgcc/ > * config.host (tmake_file) [target=avr]: Add t-libf7, > t-libf7-math, t-libf7-math-symbols as specified by --with-libf7=. > * config/avr/t-avrlibc: Don't copy libgcc.a if there are modules > depending on sizeof (double) or sizeof (long double). > * config/avr/libf7: New folder. > [-- Attachment #2: m64-libgcc.diff --] [-- Type: text/plain, Size: 1946 bytes --] Index: libgcc/config.host =================================================================== --- libgcc/config.host (revision 278552) +++ libgcc/config.host (working copy) @@ -514,6 +514,29 @@ arm*-*-eabi* | arm*-*-symbianelf* | arm* avr-*-*) # Make HImode functions for AVR tmake_file="${cpu_type}/t-avr t-fpbit" + # Make some DFmode functions from libf7, part of avr-libgcc. + # This must be prior to adding t-avrlibc. + case "y${with_libf7}" in + yno) + # No libf7 support. + ;; + ylibgcc) + tmake_file="$tmake_file ${cpu_type}/libf7/t-libf7" + ;; + ymath) + tmake_file="$tmake_file ${cpu_type}/libf7/t-libf7-math" + tmake_file="$tmake_file ${cpu_type}/libf7/t-libf7" + ;; + ymath-symbols | yyes | y) + tmake_file="$tmake_file ${cpu_type}/libf7/t-libf7-math-symbols" + tmake_file="$tmake_file ${cpu_type}/libf7/t-libf7-math" + tmake_file="$tmake_file ${cpu_type}/libf7/t-libf7" + ;; + *) + echo "Error: --with-libf7=${with_libf7} but can only be used with: 'libgcc', 'math', 'math-symbols', 'yes', 'no'" 1>&2 + exit 1 + ;; + esac if test x${with_avrlibc} != xno; then tmake_file="$tmake_file ${cpu_type}/t-avrlibc" fi Index: libgcc/config/avr/t-avrlibc =================================================================== --- libgcc/config/avr/t-avrlibc (revision 278992) +++ libgcc/config/avr/t-avrlibc (working copy) @@ -65,6 +65,12 @@ LIB2FUNCS_EXCLUDE += \ _fixunssfdi \ _floatdisf _floatundisf +ifeq (,$(WITH_LIBF7_MATH_SYMBOLS)) + +# No modules depend on __SIZEOF_LONG_DOUBLE__ or __SIZEOF_DOUBLE__ +# which means we might have an opportunity to copy libgcc.a. +# WITH_LIBF7_MATH_SYMBOLS is set by libf7/t-libf7-math-symbols. + ifneq (,$(findstring avr,$(MULTISUBDIR))) # We are not in the avr2 (default) subdir, hence copying will work. @@ -95,3 +101,4 @@ Makefile: t-copy-libgcc.dep endif endif +endif ^ permalink raw reply [flat|nested] 22+ messages in thread
* Re: [patch,avr, 2/3] Support 64-bit (long) double: The libgcc changes. 2019-12-16 16:49 ` [patch,avr, 2/3] Support 64-bit (long) double: The libgcc changes Georg-Johann Lay 2019-12-28 12:00 ` [PING^1][patch,avr, " Georg-Johann Lay 2020-01-06 13:08 ` [Ping^2][patch,avr, " Georg-Johann Lay @ 2020-01-06 16:26 ` Jeff Law 2 siblings, 0 replies; 22+ messages in thread From: Jeff Law @ 2020-01-06 16:26 UTC (permalink / raw) To: Georg-Johann Lay, GCC Patches On Mon, 2019-12-16 at 17:46 +0100, Georg-Johann Lay wrote: > Am 16.12.19 um 17:40 schrieb Georg-Johann Lay: > > Patch 2/3 is the libgcc additions: > > --with-libf7 selects which makefile-snips from libf7 to use. > > libgcc/ > * config.host (tmake_file) [target=avr]: Add t-libf7, > t-libf7-math, t-libf7-math-symbols as specified by --with-libf7=. > * config/avr/t-avrlibc: Don't copy libgcc.a if there are modules > depending on sizeof (double) or sizeof (long double). > * config/avr/libf7: New folder. Same as #1/3, libf7->libf77. Otherwise this looks reasonable. jeff ^ permalink raw reply [flat|nested] 22+ messages in thread
* Re: [patch,avr, 3/3] Support 64-bit (long) double: libf7. 2019-12-16 16:43 [patch,avr, 0/3] Support 64-bit (long) double Georg-Johann Lay 2019-12-16 16:46 ` [patch,avr, 1/3] Support 64-bit (long) double: The gcc part Georg-Johann Lay 2019-12-16 16:49 ` [patch,avr, 2/3] Support 64-bit (long) double: The libgcc changes Georg-Johann Lay @ 2019-12-16 17:09 ` Georg-Johann Lay 2019-12-28 12:02 ` [PING^1][patch,avr, " Georg-Johann Lay ` (2 more replies) 2019-12-28 11:59 ` [PING^1][patch,avr, 0/3] Support 64-bit (long) double Georg-Johann Lay ` (2 subsequent siblings) 5 siblings, 3 replies; 22+ messages in thread From: Georg-Johann Lay @ 2019-12-16 17:09 UTC (permalink / raw) To: GCC Patches [-- Attachment #1: Type: text/plain, Size: 716 bytes --] Am 16.12.19 um 17:40 schrieb Georg-Johann Lay: Patch 3/3 is the actual libf7 implementation. A great deal of which is assembly, together with C + inline assembly for higher routines. Johann libgcc/config/avr/libf7/ * t-libf7: New file. * t-libf7-math: New file. * t-libf7-math-symbols: New file. * libf7-common.mk: New file. * libf7-asm-object.mk: New file. * libf7-c-object.mk: New file. * asm-defs.h: New file. * libf7.h: New file. * libf7.c: New file. * libf7-asm.sx: New file. * libf7-array.def: New file. * libf7-const.def: New file. * libf7-constdef.h: New file. * f7renames.sh: New script. * f7wraps.sh: New script. * f7-renames.h: New generated file. * f7-wraps.h: New generated file. [-- Attachment #2: m64-libf7.diff --] [-- Type: text/x-patch, Size: 190608 bytes --] Index: libgcc/config/avr/libf7/asm-defs.h =================================================================== --- libgcc/config/avr/libf7/asm-defs.h (nonexistent) +++ libgcc/config/avr/libf7/asm-defs.h (working copy) @@ -0,0 +1,237 @@ +/* Copyright (C) 2019 Free Software Foundation, Inc. + + This file is part of LIBF7, which is part of GCC. + + GCC is free software; you can redistribute it and/or modify it under + the terms of the GNU General Public License as published by the Free + Software Foundation; either version 3, or (at your option) any later + version. + + GCC is distributed in the hope that it will be useful, but WITHOUT ANY + WARRANTY; without even the implied warranty of MERCHANTABILITY or + FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + for more details. + + Under Section 7 of GPL version 3, you are granted additional + permissions described in the GCC Runtime Library Exception, version + 3.1, as published by the Free Software Foundation. + + You should have received a copy of the GNU General Public License and + a copy of the GCC Runtime Library Exception along with this program; + see the files COPYING3 and COPYING.RUNTIME respectively. If not, see + <http://www.gnu.org/licenses/>. */ + +#ifndef ASM_DEFS_H +#define ASM_DEFS_H + +#ifdef __AVR__ +#ifdef __ASSEMBLER__ +/*****************************************************************/ +/* Stuff for Assembler-only */ +/*****************************************************************/ + +#if defined (__AVR_TINY__) + #define __tmp_reg__ 16 + #define __zero_reg__ 17 +#else + #define __tmp_reg__ 0 + #define __zero_reg__ 1 +#endif /* AVR_TINY */ + +#define __SREG__ 0x3f +#define __SP_L__ 0x3d +#if defined (__AVR_HAVE_SPH__) +#define __SP_H__ 0x3e +#endif + +#if !defined ASM_DEFS_HAVE_DEFUN +.macro DEFUN name + .global \name + .func \name + \name: +.endm + +.macro ENDF name + .size \name, .-\name + .endfunc +.endm + +.macro LABEL name + .global \name + \name: +.endm +#endif /* HAVE_DEFUN */ + + +#if defined (__AVR_HAVE_JMP_CALL__) + #define XCALL call + #define XJMP jmp +#else + #define XCALL rcall + #define XJMP rjmp +#endif + +#if defined (__AVR_HAVE_EIJMP_EICALL__) + #define XICALL eicall + #define XIJMP eijmp + #define PC_SIZE 3 +#else + #define XICALL icall + #define XIJMP ijmp + #define PC_SIZE 2 +#endif + +.macro skipnext + cpse r16, r16 +.endm + +/* + Factor out support of MOVW. Usage is like + + wmov 30, 24 + + to move R25:R24 to R31:R30, i.e. plain register numbers + are required and no register prefix 'R'. +*/ + +#if defined (__AVR_HAVE_MOVW__) +#define wmov movw +#else + .macro wmov dst src + ..dst = \dst + ..src = \src + ..regno = 0 + .irp reg, \ + r0, r1, r2, r3, r4, r5, r6, r7, r8, r9, \ + r10, r11, r12, r13, r14, r15, r16, r17, r18, r19, \ + r20, r21, r22, r23, r24, r25, r26, r27, r28, r29, \ + r30, r31 + .ifc \reg,\dst + ..dst = ..regno + .endif + .ifc \reg,\src + ..src = ..regno + .endif + ..regno = ..regno + 1 + .endr + + ..regno = 0 + + .irp reg, \ + R0, R1, R2, R3, R4, R5, R6, R7, R8, R9, \ + R10, R11, R12, R13, R14, R15, R16, R17, R18, R19, \ + R20, R21, R22, R23, R24, R25, R26, R27, R28, R29, \ + R30, R31 + .ifc \reg,\dst + ..dst = ..regno + .endif + .ifc \reg,\src + ..src = ..regno + .endif + ..regno = ..regno + 1 + .endr + + ..regno = 0 + + .irp reg, \ + X, x, XL, xl, Xl, xL, x, x \ + Y, y, YL, yl, Yl, yL, y, y, \ + Z, z, ZL, zl, Zl, zL, z, z + .ifc \reg,\dst + ..dst = (..regno / 8) + 26 + .endif + .ifc \reg,\src + ..src = (..regno / 8) + 26 + .endif + ..regno = ..regno + 1 + .endr + + mov ..dst+0, ..src+0 + mov ..dst+1, ..src+1 + .endm +#endif /* MOVW */ + + +#if !defined (__AVR_TINY__) +/* + Convenience macro for easy use of __prologue_saves__ from libgcc. + Push the N_PUSHED callee-saved registers Y, R17, R16, R15, ... + with 0 <= N_PUSHED <= 18. The frame pointer (Y) is set up according + to a frame size of N_FRAME. Clobbers TMP_REG. + For the code of __prologue_saves__ from libgcc see + http://gcc.gnu.org/viewcvs/gcc/trunk/libgcc/config/avr/lib1funcs.S?revision=267494&view=markup#l2159 +*/ + +.macro do_prologue_saves n_pushed n_frame=0 + ldi r26, lo8(\n_frame) + ldi r27, hi8(\n_frame) + ldi r30, lo8(gs(.L_prologue_saves.\@)) + ldi r31, hi8(gs(.L_prologue_saves.\@)) + XJMP __prologue_saves__ + ((18 - (\n_pushed)) * 2) +.L_prologue_saves.\@: +.endm + +/* + Convenience macro for easy use of __epilogue_restores__ from libgcc. + Undo the effect of __prologue_saves__. Clobbers TMP_REG. + For the code of __epilogue_restores__ from libgcc see + http://gcc.gnu.org/viewcvs/gcc/trunk/libgcc/config/avr/lib1funcs.S?revision=267494&view=markup#l2216 +*/ + +.macro do_epilogue_restores n_pushed n_frame=0 + in r28, __SP_L__ +#ifdef __AVR_HAVE_SPH__ + in r29, __SP_H__ +.if \n_frame > 63 + subi r28, lo8(-\n_frame) + sbci r29, hi8(-\n_frame) +.elseif \n_frame > 0 + adiw r28, \n_frame +.endif +#else + clr r29 +.if \n_frame > 0 + subi r28, lo8(-\n_frame) +.endif +#endif /* HAVE SPH */ + ldi r30, \n_pushed + XJMP __epilogue_restores__ + ((18 - (\n_pushed)) * 2) +.endm + +#endif /* AVR_TINY */ + +#else /* Assembler */ +/*****************************************************************/ +/* Space for C/C++ only Stuff */ +/*****************************************************************/ +#endif /* Assembler */ + +/*****************************************************************/ +/* Space for Generic Stuff (Assembler, C, C++) */ +/*****************************************************************/ + +#ifdef __AVR_PM_BASE_ADDRESS__ + /* + Devices with a linear address space: Flash memory is seen in the + RAM address space at an offset of __AVR_PM_BASE_ADDRESS__ and can + be accessed by LD*. This is the case for devices like ATtiny40 + (avrtiny) or ATtiny1616 and ATmega4808 (avrxmega3). The default + linker script locates .rodata in the .text output section and + at the required offset. + */ + #define RODATA_SECTION .rodata.asm + #define USE_LD 1 + #define USE_LPM 0 +#else /* PM_BASE_ADDRESS */ + /* + No linear address space. As .rodata is located in RAM, we have to + use .progmem.data (located in flash) and LPM to read the data. + This will also work for devices from avrxmega3. + */ + #define RODATA_SECTION .progmem.data.asm + #define USE_LD 0 + #define USE_LPM 1 +#endif /* PM_BASE_ADDRESS */ + +#endif /* target AVR */ +#endif /* ASM_DEFS_H */ Property changes on: libgcc/config/avr/libf7/asm-defs.h ___________________________________________________________________ Added: svn:eol-style ## -0,0 +1 ## +native \ No newline at end of property Added: svn:mime-type ## -0,0 +1 ## +text/plain \ No newline at end of property Index: libgcc/config/avr/libf7/f7-renames.h =================================================================== --- libgcc/config/avr/libf7/f7-renames.h (nonexistent) +++ libgcc/config/avr/libf7/f7-renames.h (working copy) @@ -0,0 +1,234 @@ +/* + Auto-generated file, do not change by hand. + + Generated by: f7renames.sh. + Generated using: F7_PREFIX = __f7_ from t-libf7. + F7F, F7F_cst, F7F_asm from libf7-common.mk. + Included by: libf7.h. + Used by: libf7.c, libf7.h, libf7-asm.sx, f7-wraps.h. +*/ + +#ifndef F7_RENAMES_H +#define F7_RENAMES_H + +#define F7_(name) __f7_##name +#define F7P __f7_ + +/* Renames for libf7.c, libf7.h. */ + +#define f7_fabs __f7_fabs +#define f7_neg __f7_neg +#define f7_add __f7_add +#define f7_sub __f7_sub +#define f7_addsub __f7_addsub +#define f7_div __f7_div +#define f7_div1 __f7_div1 +#define f7_divx __f7_divx +#define f7_fmod __f7_fmod +#define f7_sqrt __f7_sqrt +#define f7_cbrt __f7_cbrt +#define f7_square __f7_square +#define f7_mul __f7_mul +#define f7_mulx __f7_mulx +#define f7_madd_msub __f7_madd_msub +#define f7_madd __f7_madd +#define f7_msub __f7_msub +#define f7_hypot __f7_hypot +#define f7_Ineg __f7_Ineg +#define f7_Iadd __f7_Iadd +#define f7_Isub __f7_Isub +#define f7_Imul __f7_Imul +#define f7_Idiv __f7_Idiv +#define f7_IRsub __f7_IRsub +#define f7_Isquare __f7_Isquare +#define f7_Ildexp __f7_Ildexp +#define f7_Isqrt __f7_Isqrt +#define f7_le __f7_le +#define f7_lt __f7_lt +#define f7_gt __f7_gt +#define f7_ge __f7_ge +#define f7_ne __f7_ne +#define f7_eq __f7_eq +#define f7_cmp __f7_cmp +#define f7_cmp_abs __f7_cmp_abs +#define f7_ordered __f7_ordered +#define f7_unordered __f7_unordered +#define f7_cmp_unordered __f7_cmp_unordered +#define f7_lt_impl __f7_lt_impl +#define f7_gt_impl __f7_gt_impl +#define f7_le_impl __f7_le_impl +#define f7_ge_impl __f7_ge_impl +#define f7_eq_impl __f7_eq_impl +#define f7_ne_impl __f7_ne_impl +#define f7_unord_impl __f7_unord_impl +#define f7_lrint __f7_lrint +#define f7_ldexp __f7_ldexp +#define f7_frexp __f7_frexp +#define f7_exp __f7_exp +#define f7_logx __f7_logx +#define f7_log __f7_log +#define f7_log10 __f7_log10 +#define f7_log2 __f7_log2 +#define f7_minmax __f7_minmax +#define f7_fmax __f7_fmax +#define f7_fmin __f7_fmin +#define f7_floor __f7_floor +#define f7_ceil __f7_ceil +#define f7_round __f7_round +#define f7_lround __f7_lround +#define f7_trunc __f7_trunc +#define f7_truncx __f7_truncx +#define f7_horner __f7_horner +#define f7_pow10 __f7_pow10 +#define f7_exp10 __f7_exp10 +#define f7_pow __f7_pow +#define f7_powi __f7_powi +#define f7_sin __f7_sin +#define f7_cos __f7_cos +#define f7_tan __f7_tan +#define f7_cotan __f7_cotan +#define f7_sincos __f7_sincos +#define f7_sinh __f7_sinh +#define f7_cosh __f7_cosh +#define f7_tanh __f7_tanh +#define f7_sinhcosh __f7_sinhcosh +#define f7_asinacos __f7_asinacos +#define f7_asin __f7_asin +#define f7_acos __f7_acos +#define f7_atan __f7_atan +#define f7_atan2 __f7_atan2 +#define f7_mul_noround __f7_mul_noround +#define f7_sqrt16_round __f7_sqrt16_round +#define f7_sqrt16_floor __f7_sqrt16_floor +#define f7_clr_mant_lsbs __f7_clr_mant_lsbs +#define f7_abscmp_msb_ge __f7_abscmp_msb_ge +#define f7_lshrdi3 __f7_lshrdi3 +#define f7_ashldi3 __f7_ashldi3 +#define f7_assert __f7_assert +#define f7_classify __f7_classify +#define f7_class_inf __f7_class_inf +#define f7_class_nan __f7_class_nan +#define f7_class_number __f7_class_number +#define f7_class_zero __f7_class_zero +#define f7_class_nonzero __f7_class_nonzero +#define f7_class_sign __f7_class_sign +#define f7_signbit __f7_signbit +#define f7_set_sign __f7_set_sign +#define f7_set_nan __f7_set_nan +#define f7_set_inf __f7_set_inf +#define f7_is_inf __f7_is_inf +#define f7_is_nan __f7_is_nan +#define f7_is_number __f7_is_number +#define f7_is_zero __f7_is_zero +#define f7_is_nonzero __f7_is_nonzero +#define f7_clr __f7_clr +#define f7_copy __f7_copy +#define f7_copy_P __f7_copy_P +#define f7_copy_mant __f7_copy_mant +#define f7_msbit __f7_msbit +#define f7_is0 __f7_is0 +#define f7_cmp_mant __f7_cmp_mant +#define f7_store_expo __f7_store_expo +#define f7_abs __f7_abs +#define f7_set_s64 __f7_set_s64 +#define f7_set_s32 __f7_set_s32 +#define f7_set_s16 __f7_set_s16 +#define f7_set_s16_impl __f7_set_s16_impl +#define f7_set_u16_worker __f7_set_u16_worker +#define f7_set_u64 __f7_set_u64 +#define f7_set_u32 __f7_set_u32 +#define f7_set_u16 __f7_set_u16 +#define f7_set_u16_impl __f7_set_u16_impl +#define f7_set_float __f7_set_float +#define f7_set_pdouble __f7_set_pdouble +#define f7_set_double_impl __f7_set_double_impl +#define f7_set_double __f7_set_double +#define f7_init_impl __f7_init_impl +#define f7_init __f7_init +#define f7_get_s16 __f7_get_s16 +#define f7_get_s32 __f7_get_s32 +#define f7_get_s64 __f7_get_s64 +#define f7_get_float __f7_get_float +#define f7_get_u16 __f7_get_u16 +#define f7_get_u32 __f7_get_u32 +#define f7_get_u64 __f7_get_u64 +#define f7_get_double __f7_get_double +#define f7_set_eps __f7_set_eps +#define f7_set_1pow2 __f7_set_1pow2 +#define f7_min __f7_min +#define f7_max __f7_max +#define f7_exp10 __f7_exp10 +#define f7_floatunsidf __f7_floatunsidf +#define f7_floatsidf __f7_floatsidf +#define f7_extendsfdf2 __f7_extendsfdf2 +#define f7_fixdfsi __f7_fixdfsi +#define f7_fixdfdi __f7_fixdfdi +#define f7_fixunsdfdi __f7_fixunsdfdi +#define f7_fixunsdfsi __f7_fixunsdfsi +#define f7_truncdfsf2 __f7_truncdfsf2 +#define f7_le_impl __f7_le_impl +#define f7_lt_impl __f7_lt_impl +#define f7_gt_impl __f7_gt_impl +#define f7_ge_impl __f7_ge_impl +#define f7_ne_impl __f7_ne_impl +#define f7_eq_impl __f7_eq_impl +#define f7_unord_impl __f7_unord_impl + +/* Renames for libf7.c, libf7.h. */ + +#define f7_const_1 __f7_const_1 +#define f7_const_1_P __f7_const_1_P +#define f7_const_2 __f7_const_2 +#define f7_const_2_P __f7_const_2_P +#define f7_const_1_2 __f7_const_1_2 +#define f7_const_1_2_P __f7_const_1_2_P +#define f7_const_1_3 __f7_const_1_3 +#define f7_const_1_3_P __f7_const_1_3_P +#define f7_const_m1 __f7_const_m1 +#define f7_const_m1_P __f7_const_m1_P +#define f7_const_pi __f7_const_pi +#define f7_const_pi_P __f7_const_pi_P +#define f7_const_ln2 __f7_const_ln2 +#define f7_const_ln2_P __f7_const_ln2_P +#define f7_const_ln10 __f7_const_ln10 +#define f7_const_ln10_P __f7_const_ln10_P +#define f7_const_1_ln2 __f7_const_1_ln2 +#define f7_const_1_ln2_P __f7_const_1_ln2_P +#define f7_const_1_ln10 __f7_const_1_ln10 +#define f7_const_1_ln10_P __f7_const_1_ln10_P +#define f7_const_sqrt2 __f7_const_sqrt2 +#define f7_const_sqrt2_P __f7_const_sqrt2_P + +/* Renames for libf7-asm.sx, f7-wraps.h. */ + +#define f7_classify_asm __f7_classify_asm +#define f7_store_expo_asm __f7_store_expo_asm +#define f7_clr_asm __f7_clr_asm +#define f7_copy_asm __f7_copy_asm +#define f7_copy_P_asm __f7_copy_P_asm +#define f7_copy_mant_asm __f7_copy_mant_asm +#define f7_cmp_mant_asm __f7_cmp_mant_asm +#define f7_normalize_asm __f7_normalize_asm +#define f7_store_expo_asm __f7_store_expo_asm +#define f7_set_u64_asm __f7_set_u64_asm +#define f7_set_s64_asm __f7_set_s64_asm +#define f7_addsub_mant_scaled_asm __f7_addsub_mant_scaled_asm +#define f7_mul_mant_asm __f7_mul_mant_asm +#define f7_to_integer_asm __f7_to_integer_asm +#define f7_to_unsigned_asm __f7_to_unsigned_asm +#define f7_clr_mant_lsbs_asm __f7_clr_mant_lsbs_asm +#define f7_div_asm __f7_div_asm +#define f7_sqrt_approx_asm __f7_sqrt_approx_asm +#define f7_sqrt16_round_asm __f7_sqrt16_round_asm +#define f7_sqrt16_floor_asm __f7_sqrt16_floor_asm +#define f7_lshrdi3_asm __f7_lshrdi3_asm +#define f7_ashldi3_asm __f7_ashldi3_asm +#define f7_class_D_asm __f7_class_D_asm +#define f7_call_ddd_asm __f7_call_ddd_asm +#define f7_call_xdd_asm __f7_call_xdd_asm +#define f7_call_ddx_asm __f7_call_ddx_asm +#define f7_call_dd_asm __f7_call_dd_asm +#define f7_call_xd_asm __f7_call_xd_asm +#define f7_call_dx_asm __f7_call_dx_asm + +#endif /* F7_RENAMES_H */ Property changes on: libgcc/config/avr/libf7/f7-renames.h ___________________________________________________________________ Added: svn:eol-style ## -0,0 +1 ## +native \ No newline at end of property Added: svn:mime-type ## -0,0 +1 ## +text/plain \ No newline at end of property Index: libgcc/config/avr/libf7/f7-wraps.h =================================================================== --- libgcc/config/avr/libf7/f7-wraps.h (nonexistent) +++ libgcc/config/avr/libf7/f7-wraps.h (working copy) @@ -0,0 +1,638 @@ +;; Copyright (C) 2019 Free Software Foundation, Inc. +;; +;; This file is part of LIBF7, which is part of GCC. +;; +;; GCC is free software; you can redistribute it and/or modify it under +;; the terms of the GNU General Public License as published by the Free +;; Software Foundation; either version 3, or (at your option) any later +;; version. +;; +;; GCC is distributed in the hope that it will be useful, but WITHOUT ANY +;; WARRANTY; without even the implied warranty of MERCHANTABILITY or +;; FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License +;; for more details. +;; +;; Under Section 7 of GPL version 3, you are granted additional +;; permissions described in the GCC Runtime Library Exception, version +;; 3.1, as published by the Free Software Foundation. +;; +;; You should have received a copy of the GNU General Public License and +;; a copy of the GCC Runtime Library Exception along with this program; +;; see the files COPYING3 and COPYING.RUNTIME respectively. If not, see +;; <http://www.gnu.org/licenses/>. */ + +;; Auto-generated file, do not change by hand. +;; +;; Wrappers for double and long double functions to use functions that +;; operate on f7_t, and get f7_t* and const f7_t*. +;; +;; Generated by: f7wraps.sh +;; Included by : libf7-asm.sx +;; WITH_LIBF7_MATH_FUNCTIONS=1 +;; WITH_LIBF7_MATH_SYMBOLS=1 + +;; Functions that usually live in libgcc: __<name>df3 for <name> in: +;; add sub mul div + +;; double __adddf3 (double, double) ; add +#ifdef F7MOD_D_add_ +_DEFUN __adddf3 + ALIAS __add + .global F7_NAME(add) + ldi ZH, hi8(gs(F7_NAME(add))) + ldi ZL, lo8(gs(F7_NAME(add))) + F7jmp call_ddd +_ENDF __adddf3 +#endif /* F7MOD_D_add_ */ + +;; double __subdf3 (double, double) ; sub +#ifdef F7MOD_D_sub_ +_DEFUN __subdf3 + ALIAS __sub + .global F7_NAME(sub) + ldi ZH, hi8(gs(F7_NAME(sub))) + ldi ZL, lo8(gs(F7_NAME(sub))) + F7jmp call_ddd +_ENDF __subdf3 +#endif /* F7MOD_D_sub_ */ + +;; double __muldf3 (double, double) ; mul +#ifdef F7MOD_D_mul_ +_DEFUN __muldf3 + ALIAS __mul + .global F7_NAME(mul) + ldi ZH, hi8(gs(F7_NAME(mul))) + ldi ZL, lo8(gs(F7_NAME(mul))) + F7jmp call_ddd +_ENDF __muldf3 +#endif /* F7MOD_D_mul_ */ + +;; double __divdf3 (double, double) ; div +#ifdef F7MOD_D_div_ +_DEFUN __divdf3 + ALIAS __div + .global F7_NAME(div) + ldi ZH, hi8(gs(F7_NAME(div))) + ldi ZL, lo8(gs(F7_NAME(div))) + F7jmp call_ddd +_ENDF __divdf3 +#endif /* F7MOD_D_div_ */ + +;; Functions that usually live in libgcc: __<name>df2 for <name> in: +;; le lt ge gt ne eq unord + +;; bool __ledf2 (double, double) ; le +#ifdef F7MOD_D_le_ +_DEFUN __ledf2 + .global F7_NAME(le_impl) + ldi ZH, hi8(gs(F7_NAME(le_impl))) + ldi ZL, lo8(gs(F7_NAME(le_impl))) + F7jmp call_xdd +_ENDF __ledf2 +#endif /* F7MOD_D_le_ */ + +;; bool __ltdf2 (double, double) ; lt +#ifdef F7MOD_D_lt_ +_DEFUN __ltdf2 + .global F7_NAME(lt_impl) + ldi ZH, hi8(gs(F7_NAME(lt_impl))) + ldi ZL, lo8(gs(F7_NAME(lt_impl))) + F7jmp call_xdd +_ENDF __ltdf2 +#endif /* F7MOD_D_lt_ */ + +;; bool __gedf2 (double, double) ; ge +#ifdef F7MOD_D_ge_ +_DEFUN __gedf2 + .global F7_NAME(ge_impl) + ldi ZH, hi8(gs(F7_NAME(ge_impl))) + ldi ZL, lo8(gs(F7_NAME(ge_impl))) + F7jmp call_xdd +_ENDF __gedf2 +#endif /* F7MOD_D_ge_ */ + +;; bool __gtdf2 (double, double) ; gt +#ifdef F7MOD_D_gt_ +_DEFUN __gtdf2 + .global F7_NAME(gt_impl) + ldi ZH, hi8(gs(F7_NAME(gt_impl))) + ldi ZL, lo8(gs(F7_NAME(gt_impl))) + F7jmp call_xdd +_ENDF __gtdf2 +#endif /* F7MOD_D_gt_ */ + +;; bool __nedf2 (double, double) ; ne +#ifdef F7MOD_D_ne_ +_DEFUN __nedf2 + .global F7_NAME(ne_impl) + ldi ZH, hi8(gs(F7_NAME(ne_impl))) + ldi ZL, lo8(gs(F7_NAME(ne_impl))) + F7jmp call_xdd +_ENDF __nedf2 +#endif /* F7MOD_D_ne_ */ + +;; bool __eqdf2 (double, double) ; eq +#ifdef F7MOD_D_eq_ +_DEFUN __eqdf2 + .global F7_NAME(eq_impl) + ldi ZH, hi8(gs(F7_NAME(eq_impl))) + ldi ZL, lo8(gs(F7_NAME(eq_impl))) + F7jmp call_xdd +_ENDF __eqdf2 +#endif /* F7MOD_D_eq_ */ + +;; bool __unorddf2 (double, double) ; unord +#ifdef F7MOD_D_unord_ +_DEFUN __unorddf2 + .global F7_NAME(unord_impl) + ldi ZH, hi8(gs(F7_NAME(unord_impl))) + ldi ZL, lo8(gs(F7_NAME(unord_impl))) + F7jmp call_xdd +_ENDF __unorddf2 +#endif /* F7MOD_D_unord_ */ + +;; Functions that usually live in libgcc: __<name> for <name> in: +;; fixdfsi fixdfdi fixunsdfdi fixunsdfsi truncdfsf2 + +;; type_t __fixdfsi (double) ; fixdfsi +#ifdef F7MOD_D_fixdfsi_ +_DEFUN __fixdfsi + .global F7_NAME(fixdfsi) + ldi ZH, hi8(gs(F7_NAME(fixdfsi))) + ldi ZL, lo8(gs(F7_NAME(fixdfsi))) + F7jmp call_xd +_ENDF __fixdfsi +#endif /* F7MOD_D_fixdfsi_ */ + +;; type_t __fixdfdi (double) ; fixdfdi +#ifdef F7MOD_D_fixdfdi_ +_DEFUN __fixdfdi + .global F7_NAME(fixdfdi) + ldi ZH, hi8(gs(F7_NAME(fixdfdi))) + ldi ZL, lo8(gs(F7_NAME(fixdfdi))) + F7jmp call_xd +_ENDF __fixdfdi +#endif /* F7MOD_D_fixdfdi_ */ + +;; type_t __fixunsdfdi (double) ; fixunsdfdi +#ifdef F7MOD_D_fixunsdfdi_ +_DEFUN __fixunsdfdi + .global F7_NAME(fixunsdfdi) + ldi ZH, hi8(gs(F7_NAME(fixunsdfdi))) + ldi ZL, lo8(gs(F7_NAME(fixunsdfdi))) + F7jmp call_xd +_ENDF __fixunsdfdi +#endif /* F7MOD_D_fixunsdfdi_ */ + +;; type_t __fixunsdfsi (double) ; fixunsdfsi +#ifdef F7MOD_D_fixunsdfsi_ +_DEFUN __fixunsdfsi + .global F7_NAME(fixunsdfsi) + ldi ZH, hi8(gs(F7_NAME(fixunsdfsi))) + ldi ZL, lo8(gs(F7_NAME(fixunsdfsi))) + F7jmp call_xd +_ENDF __fixunsdfsi +#endif /* F7MOD_D_fixunsdfsi_ */ + +;; type_t __truncdfsf2 (double) ; truncdfsf2 +#ifdef F7MOD_D_truncdfsf2_ +_DEFUN __truncdfsf2 + .global F7_NAME(truncdfsf2) + ldi ZH, hi8(gs(F7_NAME(truncdfsf2))) + ldi ZL, lo8(gs(F7_NAME(truncdfsf2))) + F7jmp call_xd +_ENDF __truncdfsf2 +#endif /* F7MOD_D_truncdfsf2_ */ + +;; Functions that usually live in libgcc: __<name> for <name> in: +;; floatunsidf floatsidf extendsfdf2 + +;; double __floatunsidf (type_t) ; floatunsidf +#ifdef F7MOD_D_floatunsidf_ +_DEFUN __floatunsidf + .global F7_NAME(floatunsidf) + ldi ZH, hi8(gs(F7_NAME(floatunsidf))) + ldi ZL, lo8(gs(F7_NAME(floatunsidf))) + F7jmp call_dx +_ENDF __floatunsidf +#endif /* F7MOD_D_floatunsidf_ */ + +;; double __floatsidf (type_t) ; floatsidf +#ifdef F7MOD_D_floatsidf_ +_DEFUN __floatsidf + .global F7_NAME(floatsidf) + ldi ZH, hi8(gs(F7_NAME(floatsidf))) + ldi ZL, lo8(gs(F7_NAME(floatsidf))) + F7jmp call_dx +_ENDF __floatsidf +#endif /* F7MOD_D_floatsidf_ */ + +;; double __extendsfdf2 (type_t) ; extendsfdf2 +#ifdef F7MOD_D_extendsfdf2_ +_DEFUN __extendsfdf2 + .global F7_NAME(extendsfdf2) + ldi ZH, hi8(gs(F7_NAME(extendsfdf2))) + ldi ZL, lo8(gs(F7_NAME(extendsfdf2))) + F7jmp call_dx +_ENDF __extendsfdf2 +#endif /* F7MOD_D_extendsfdf2_ */ + +;; Functions that usually live in libm: Depending on [long] double layout, +;; define <name> and <name>l as weak alias(es) of __<name> for <name> in: +;; pow fmin fmax fmod hypot atan2 + +;; double __pow (double, double) +#ifdef F7MOD_D_pow_ +_DEFUN __pow + DALIAS pow + LALIAS powl + .global F7_NAME(pow) + ldi ZH, hi8(gs(F7_NAME(pow))) + ldi ZL, lo8(gs(F7_NAME(pow))) + F7jmp call_ddd +_ENDF __pow +#endif /* F7MOD_D_pow_ */ + +;; double __fmin (double, double) +#ifdef F7MOD_D_fmin_ +_DEFUN __fmin + DALIAS fmin + LALIAS fminl + .global F7_NAME(fmin) + ldi ZH, hi8(gs(F7_NAME(fmin))) + ldi ZL, lo8(gs(F7_NAME(fmin))) + F7jmp call_ddd +_ENDF __fmin +#endif /* F7MOD_D_fmin_ */ + +;; double __fmax (double, double) +#ifdef F7MOD_D_fmax_ +_DEFUN __fmax + DALIAS fmax + LALIAS fmaxl + .global F7_NAME(fmax) + ldi ZH, hi8(gs(F7_NAME(fmax))) + ldi ZL, lo8(gs(F7_NAME(fmax))) + F7jmp call_ddd +_ENDF __fmax +#endif /* F7MOD_D_fmax_ */ + +;; double __fmod (double, double) +#ifdef F7MOD_D_fmod_ +_DEFUN __fmod + DALIAS fmod + LALIAS fmodl + .global F7_NAME(fmod) + ldi ZH, hi8(gs(F7_NAME(fmod))) + ldi ZL, lo8(gs(F7_NAME(fmod))) + F7jmp call_ddd +_ENDF __fmod +#endif /* F7MOD_D_fmod_ */ + +;; double __hypot (double, double) +#ifdef F7MOD_D_hypot_ +_DEFUN __hypot + DALIAS hypot + LALIAS hypotl + .global F7_NAME(hypot) + ldi ZH, hi8(gs(F7_NAME(hypot))) + ldi ZL, lo8(gs(F7_NAME(hypot))) + F7jmp call_ddd +_ENDF __hypot +#endif /* F7MOD_D_hypot_ */ + +;; double __atan2 (double, double) +#ifdef F7MOD_D_atan2_ +_DEFUN __atan2 + DALIAS atan2 + LALIAS atan2l + .global F7_NAME(atan2) + ldi ZH, hi8(gs(F7_NAME(atan2))) + ldi ZL, lo8(gs(F7_NAME(atan2))) + F7jmp call_ddd +_ENDF __atan2 +#endif /* F7MOD_D_atan2_ */ + +;; Functions that usually live in libm: Depending on [long] double layout, +;; define <name> and <name>l as weak alias(es) of __<name> for <name> in: +;; ldexp frexp + +;; double __ldexp (double, word_t) +#ifdef F7MOD_D_ldexp_ +_DEFUN __ldexp + DALIAS ldexp + LALIAS ldexpl + .global F7_NAME(ldexp) + ldi ZH, hi8(gs(F7_NAME(ldexp))) + ldi ZL, lo8(gs(F7_NAME(ldexp))) + F7jmp call_ddx +_ENDF __ldexp +#endif /* F7MOD_D_ldexp_ */ + +;; double __frexp (double, word_t) +#ifdef F7MOD_D_frexp_ +_DEFUN __frexp + DALIAS frexp + LALIAS frexpl + .global F7_NAME(frexp) + ldi ZH, hi8(gs(F7_NAME(frexp))) + ldi ZL, lo8(gs(F7_NAME(frexp))) + F7jmp call_ddx +_ENDF __frexp +#endif /* F7MOD_D_frexp_ */ + +;; Functions that usually live in libm: Depending on [long] double layout, +;; define <name> and <name>l as weak alias(es) of __<name> for <name> in: +;; sqrt cbrt exp exp10 pow10 log log10 log2 sin cos tan cotan asin acos atan ceil floor trunc round sinh cosh tanh + +;; double __sqrt (double) +#ifdef F7MOD_D_sqrt_ +_DEFUN __sqrt + DALIAS sqrt + LALIAS sqrtl + .global F7_NAME(sqrt) + ldi ZH, hi8(gs(F7_NAME(sqrt))) + ldi ZL, lo8(gs(F7_NAME(sqrt))) + F7jmp call_dd +_ENDF __sqrt +#endif /* F7MOD_D_sqrt_ */ + +;; double __cbrt (double) +#ifdef F7MOD_D_cbrt_ +_DEFUN __cbrt + DALIAS cbrt + LALIAS cbrtl + .global F7_NAME(cbrt) + ldi ZH, hi8(gs(F7_NAME(cbrt))) + ldi ZL, lo8(gs(F7_NAME(cbrt))) + F7jmp call_dd +_ENDF __cbrt +#endif /* F7MOD_D_cbrt_ */ + +;; double __exp (double) +#ifdef F7MOD_D_exp_ +_DEFUN __exp + DALIAS exp + LALIAS expl + .global F7_NAME(exp) + ldi ZH, hi8(gs(F7_NAME(exp))) + ldi ZL, lo8(gs(F7_NAME(exp))) + F7jmp call_dd +_ENDF __exp +#endif /* F7MOD_D_exp_ */ + +;; double __exp10 (double) +#ifdef F7MOD_D_exp10_ +_DEFUN __exp10 + DALIAS exp10 + LALIAS exp10l + .global F7_NAME(exp10) + ldi ZH, hi8(gs(F7_NAME(exp10))) + ldi ZL, lo8(gs(F7_NAME(exp10))) + F7jmp call_dd +_ENDF __exp10 +#endif /* F7MOD_D_exp10_ */ + +;; double __pow10 (double) +#ifdef F7MOD_D_pow10_ +_DEFUN __pow10 + DALIAS pow10 + LALIAS pow10l + .global F7_NAME(pow10) + ldi ZH, hi8(gs(F7_NAME(pow10))) + ldi ZL, lo8(gs(F7_NAME(pow10))) + F7jmp call_dd +_ENDF __pow10 +#endif /* F7MOD_D_pow10_ */ + +;; double __log (double) +#ifdef F7MOD_D_log_ +_DEFUN __log + DALIAS log + LALIAS logl + .global F7_NAME(log) + ldi ZH, hi8(gs(F7_NAME(log))) + ldi ZL, lo8(gs(F7_NAME(log))) + F7jmp call_dd +_ENDF __log +#endif /* F7MOD_D_log_ */ + +;; double __log10 (double) +#ifdef F7MOD_D_log10_ +_DEFUN __log10 + DALIAS log10 + LALIAS log10l + .global F7_NAME(log10) + ldi ZH, hi8(gs(F7_NAME(log10))) + ldi ZL, lo8(gs(F7_NAME(log10))) + F7jmp call_dd +_ENDF __log10 +#endif /* F7MOD_D_log10_ */ + +;; double __log2 (double) +#ifdef F7MOD_D_log2_ +_DEFUN __log2 + DALIAS log2 + LALIAS log2l + .global F7_NAME(log2) + ldi ZH, hi8(gs(F7_NAME(log2))) + ldi ZL, lo8(gs(F7_NAME(log2))) + F7jmp call_dd +_ENDF __log2 +#endif /* F7MOD_D_log2_ */ + +;; double __sin (double) +#ifdef F7MOD_D_sin_ +_DEFUN __sin + DALIAS sin + LALIAS sinl + .global F7_NAME(sin) + ldi ZH, hi8(gs(F7_NAME(sin))) + ldi ZL, lo8(gs(F7_NAME(sin))) + F7jmp call_dd +_ENDF __sin +#endif /* F7MOD_D_sin_ */ + +;; double __cos (double) +#ifdef F7MOD_D_cos_ +_DEFUN __cos + DALIAS cos + LALIAS cosl + .global F7_NAME(cos) + ldi ZH, hi8(gs(F7_NAME(cos))) + ldi ZL, lo8(gs(F7_NAME(cos))) + F7jmp call_dd +_ENDF __cos +#endif /* F7MOD_D_cos_ */ + +;; double __tan (double) +#ifdef F7MOD_D_tan_ +_DEFUN __tan + DALIAS tan + LALIAS tanl + .global F7_NAME(tan) + ldi ZH, hi8(gs(F7_NAME(tan))) + ldi ZL, lo8(gs(F7_NAME(tan))) + F7jmp call_dd +_ENDF __tan +#endif /* F7MOD_D_tan_ */ + +;; double __cotan (double) +#ifdef F7MOD_D_cotan_ +_DEFUN __cotan + DALIAS cotan + LALIAS cotanl + .global F7_NAME(cotan) + ldi ZH, hi8(gs(F7_NAME(cotan))) + ldi ZL, lo8(gs(F7_NAME(cotan))) + F7jmp call_dd +_ENDF __cotan +#endif /* F7MOD_D_cotan_ */ + +;; double __asin (double) +#ifdef F7MOD_D_asin_ +_DEFUN __asin + DALIAS asin + LALIAS asinl + .global F7_NAME(asin) + ldi ZH, hi8(gs(F7_NAME(asin))) + ldi ZL, lo8(gs(F7_NAME(asin))) + F7jmp call_dd +_ENDF __asin +#endif /* F7MOD_D_asin_ */ + +;; double __acos (double) +#ifdef F7MOD_D_acos_ +_DEFUN __acos + DALIAS acos + LALIAS acosl + .global F7_NAME(acos) + ldi ZH, hi8(gs(F7_NAME(acos))) + ldi ZL, lo8(gs(F7_NAME(acos))) + F7jmp call_dd +_ENDF __acos +#endif /* F7MOD_D_acos_ */ + +;; double __atan (double) +#ifdef F7MOD_D_atan_ +_DEFUN __atan + DALIAS atan + LALIAS atanl + .global F7_NAME(atan) + ldi ZH, hi8(gs(F7_NAME(atan))) + ldi ZL, lo8(gs(F7_NAME(atan))) + F7jmp call_dd +_ENDF __atan +#endif /* F7MOD_D_atan_ */ + +;; double __ceil (double) +#ifdef F7MOD_D_ceil_ +_DEFUN __ceil + DALIAS ceil + LALIAS ceill + .global F7_NAME(ceil) + ldi ZH, hi8(gs(F7_NAME(ceil))) + ldi ZL, lo8(gs(F7_NAME(ceil))) + F7jmp call_dd +_ENDF __ceil +#endif /* F7MOD_D_ceil_ */ + +;; double __floor (double) +#ifdef F7MOD_D_floor_ +_DEFUN __floor + DALIAS floor + LALIAS floorl + .global F7_NAME(floor) + ldi ZH, hi8(gs(F7_NAME(floor))) + ldi ZL, lo8(gs(F7_NAME(floor))) + F7jmp call_dd +_ENDF __floor +#endif /* F7MOD_D_floor_ */ + +;; double __trunc (double) +#ifdef F7MOD_D_trunc_ +_DEFUN __trunc + DALIAS trunc + LALIAS truncl + .global F7_NAME(trunc) + ldi ZH, hi8(gs(F7_NAME(trunc))) + ldi ZL, lo8(gs(F7_NAME(trunc))) + F7jmp call_dd +_ENDF __trunc +#endif /* F7MOD_D_trunc_ */ + +;; double __round (double) +#ifdef F7MOD_D_round_ +_DEFUN __round + DALIAS round + LALIAS roundl + .global F7_NAME(round) + ldi ZH, hi8(gs(F7_NAME(round))) + ldi ZL, lo8(gs(F7_NAME(round))) + F7jmp call_dd +_ENDF __round +#endif /* F7MOD_D_round_ */ + +;; double __sinh (double) +#ifdef F7MOD_D_sinh_ +_DEFUN __sinh + DALIAS sinh + LALIAS sinhl + .global F7_NAME(sinh) + ldi ZH, hi8(gs(F7_NAME(sinh))) + ldi ZL, lo8(gs(F7_NAME(sinh))) + F7jmp call_dd +_ENDF __sinh +#endif /* F7MOD_D_sinh_ */ + +;; double __cosh (double) +#ifdef F7MOD_D_cosh_ +_DEFUN __cosh + DALIAS cosh + LALIAS coshl + .global F7_NAME(cosh) + ldi ZH, hi8(gs(F7_NAME(cosh))) + ldi ZL, lo8(gs(F7_NAME(cosh))) + F7jmp call_dd +_ENDF __cosh +#endif /* F7MOD_D_cosh_ */ + +;; double __tanh (double) +#ifdef F7MOD_D_tanh_ +_DEFUN __tanh + DALIAS tanh + LALIAS tanhl + .global F7_NAME(tanh) + ldi ZH, hi8(gs(F7_NAME(tanh))) + ldi ZL, lo8(gs(F7_NAME(tanh))) + F7jmp call_dd +_ENDF __tanh +#endif /* F7MOD_D_tanh_ */ + +;; Functions that usually live in libm: Depending on [long] double layout, +;; define <name> and <name>l as weak alias(es) of __<name> for <name> in: +;; lrint lround + +;; type_t __lrint (double) +#ifdef F7MOD_D_lrint_ +_DEFUN __lrint + DALIAS lrint + LALIAS lrintl + .global F7_NAME(lrint) + ldi ZH, hi8(gs(F7_NAME(lrint))) + ldi ZL, lo8(gs(F7_NAME(lrint))) + F7jmp call_xd +_ENDF __lrint +#endif /* F7MOD_D_lrint_ */ + +;; type_t __lround (double) +#ifdef F7MOD_D_lround_ +_DEFUN __lround + DALIAS lround + LALIAS lroundl + .global F7_NAME(lround) + ldi ZH, hi8(gs(F7_NAME(lround))) + ldi ZL, lo8(gs(F7_NAME(lround))) + F7jmp call_xd +_ENDF __lround +#endif /* F7MOD_D_lround_ */ Property changes on: libgcc/config/avr/libf7/f7-wraps.h ___________________________________________________________________ Added: svn:eol-style ## -0,0 +1 ## +native \ No newline at end of property Added: svn:mime-type ## -0,0 +1 ## +text/plain \ No newline at end of property Index: libgcc/config/avr/libf7/f7renames.sh =================================================================== --- libgcc/config/avr/libf7/f7renames.sh (nonexistent) +++ libgcc/config/avr/libf7/f7renames.sh (working copy) @@ -0,0 +1,80 @@ +#!/usr/bin/env sh + +# The first command argument tells us which flavour to generate for +# the rest of the command line arguments. + +what=$1 +shift + +# The second command argument is the prefix to prepend to all functions. +# It is defined by F7_PREFIX in $2. + +PRE=$1 +shift + +case ${what} in + head) + cat << EOF +/* + Auto-generated file, do not change by hand. + + Generated by: `basename $0`. + Generated using: F7_PREFIX = ${PRE} from $1. + F7F, F7F_cst, F7F_asm from libf7-common.mk. + Included by: libf7.h. + Used by: libf7.c, libf7.h, libf7-asm.sx, f7-wraps.h. +*/ + +#ifndef F7_RENAMES_H +#define F7_RENAMES_H + +#define F7_(name) ${PRE}##name +#define F7P ${PRE} +EOF + ;; + + c) + if [ x${PRE} != xf7_ ]; then + echo " " + echo "/* Renames for libf7.c, libf7.h. */" + echo " " + for x in $*; do + echo "#define f7_$x ${PRE}$x" + done + fi + ;; + + cst) + if [ x${PRE} != xf7_ ]; then + echo " " + echo "/* Renames for libf7.c, libf7.h. */" + echo " " + for x in $*; do + echo "#define f7_const_${x} ${PRE}const_${x}" + echo "#define f7_const_${x}_P ${PRE}const_${x}_P" + done + fi + ;; + + asm) + if [ x${PRE} != xf7_ ]; then + echo " " + echo "/* Renames for libf7-asm.sx, f7-wraps.h. */" + echo " " + for x in $*; do + echo "#define f7_${x}_asm ${PRE}${x}_asm" + done + fi + ;; + + tail) + cat << EOF + +#endif /* F7_RENAMES_H */ +EOF + ;; + + *) + exit 1 + ;; +esac Property changes on: libgcc/config/avr/libf7/f7renames.sh ___________________________________________________________________ Added: svn:eol-style ## -0,0 +1 ## +native \ No newline at end of property Added: svn:executable ## -0,0 +1 ## +* \ No newline at end of property Added: svn:mime-type ## -0,0 +1 ## +text/plain \ No newline at end of property Index: libgcc/config/avr/libf7/f7wraps.sh =================================================================== --- libgcc/config/avr/libf7/f7wraps.sh (nonexistent) +++ libgcc/config/avr/libf7/f7wraps.sh (working copy) @@ -0,0 +1,244 @@ +#!/usr/bin/env sh + +# The first command argument $1 tells us which flavour to generate for +# the rest of the command line arguments. + +what=$1 +shift + +if [ "x$*" = "x" ]; then + none="(none)" +fi + +case ${what} in + header) + cat << EOF +;; Copyright (C) 2019 Free Software Foundation, Inc. +;; +;; This file is part of LIBF7, which is part of GCC. +;; +;; GCC is free software; you can redistribute it and/or modify it under +;; the terms of the GNU General Public License as published by the Free +;; Software Foundation; either version 3, or (at your option) any later +;; version. +;; +;; GCC is distributed in the hope that it will be useful, but WITHOUT ANY +;; WARRANTY; without even the implied warranty of MERCHANTABILITY or +;; FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License +;; for more details. +;; +;; Under Section 7 of GPL version 3, you are granted additional +;; permissions described in the GCC Runtime Library Exception, version +;; 3.1, as published by the Free Software Foundation. +;; +;; You should have received a copy of the GNU General Public License and +;; a copy of the GCC Runtime Library Exception along with this program; +;; see the files COPYING3 and COPYING.RUNTIME respectively. If not, see +;; <http://www.gnu.org/licenses/>. */ + +;; Auto-generated file, do not change by hand. +;; +;; Wrappers for double and long double functions to use functions that +;; operate on f7_t, and get f7_t* and const f7_t*. +;; +;; Generated by: `basename $0` +;; Included by : libf7-asm.sx +EOF + for n in $*; do + echo ";; $n" + done + ;; + + xd_libgcc) + cat << EOF + +;; Functions that usually live in libgcc: __<name> for <name> in: +;; $*${none} +EOF + for n in $*; do + cat << EOF + +;; type_t __${n} (double) ; $n +#ifdef F7MOD_D_${n}_ +_DEFUN __${n} + .global F7_NAME($n) + ldi ZH, hi8(gs(F7_NAME($n))) + ldi ZL, lo8(gs(F7_NAME($n))) + F7jmp call_xd +_ENDF __${n} +#endif /* F7MOD_D_${n}_ */ +EOF + done + ;; + + dx_libgcc) + cat << EOF + +;; Functions that usually live in libgcc: __<name> for <name> in: +;; $*${none} +EOF + for n in $*; do + cat << EOF + +;; double __${n} (type_t) ; $n +#ifdef F7MOD_D_${n}_ +_DEFUN __${n} + .global F7_NAME($n) + ldi ZH, hi8(gs(F7_NAME($n))) + ldi ZL, lo8(gs(F7_NAME($n))) + F7jmp call_dx +_ENDF __${n} +#endif /* F7MOD_D_${n}_ */ +EOF + done + ;; + + ddd_libgcc) + cat << EOF + +;; Functions that usually live in libgcc: __<name>df3 for <name> in: +;; $*${none} +EOF + for n in $*; do + cat << EOF + +;; double __${n}df3 (double, double) ; $n +#ifdef F7MOD_D_${n}_ +_DEFUN __${n}df3 + ALIAS __$n + .global F7_NAME($n) + ldi ZH, hi8(gs(F7_NAME($n))) + ldi ZL, lo8(gs(F7_NAME($n))) + F7jmp call_ddd +_ENDF __${n}df3 +#endif /* F7MOD_D_${n}_ */ +EOF + done + ;; + + xdd_libgcc_cmp) + cat << EOF + +;; Functions that usually live in libgcc: __<name>df2 for <name> in: +;; $*${none} +EOF + for n in $*; do + cat << EOF + +;; bool __${n}df2 (double, double) ; $n +#ifdef F7MOD_D_${n}_ +_DEFUN __${n}df2 + .global F7_NAME(${n}_impl) + ldi ZH, hi8(gs(F7_NAME(${n}_impl))) + ldi ZL, lo8(gs(F7_NAME(${n}_impl))) + F7jmp call_xdd +_ENDF __${n}df2 +#endif /* F7MOD_D_${n}_ */ +EOF + done + ;; + + dd_math) + cat << EOF + +;; Functions that usually live in libm: Depending on [long] double layout, +;; define <name> and <name>l as weak alias(es) of __<name> for <name> in: +;; $*${none} +EOF + for n in $*; do + cat << EOF + +;; double __${n} (double) +#ifdef F7MOD_D_${n}_ +_DEFUN __${n} + DALIAS $n + LALIAS ${n}l + .global F7_NAME($n) + ldi ZH, hi8(gs(F7_NAME($n))) + ldi ZL, lo8(gs(F7_NAME($n))) + F7jmp call_dd +_ENDF __${n} +#endif /* F7MOD_D_${n}_ */ +EOF + done + ;; + + xd_math) + cat << EOF + +;; Functions that usually live in libm: Depending on [long] double layout, +;; define <name> and <name>l as weak alias(es) of __<name> for <name> in: +;; $*${none} +EOF + for n in $*; do + cat << EOF + +;; type_t __${n} (double) +#ifdef F7MOD_D_${n}_ +_DEFUN __${n} + DALIAS $n + LALIAS ${n}l + .global F7_NAME($n) + ldi ZH, hi8(gs(F7_NAME($n))) + ldi ZL, lo8(gs(F7_NAME($n))) + F7jmp call_xd +_ENDF __${n} +#endif /* F7MOD_D_${n}_ */ +EOF + done + ;; + + ddd_math) + cat << EOF + +;; Functions that usually live in libm: Depending on [long] double layout, +;; define <name> and <name>l as weak alias(es) of __<name> for <name> in: +;; $*${none} +EOF + for n in $*; do + cat << EOF + +;; double __${n} (double, double) +#ifdef F7MOD_D_${n}_ +_DEFUN __${n} + DALIAS $n + LALIAS ${n}l + .global F7_NAME($n) + ldi ZH, hi8(gs(F7_NAME($n))) + ldi ZL, lo8(gs(F7_NAME($n))) + F7jmp call_ddd +_ENDF __${n} +#endif /* F7MOD_D_${n}_ */ +EOF + done + ;; + + ddx_math) + cat << EOF + +;; Functions that usually live in libm: Depending on [long] double layout, +;; define <name> and <name>l as weak alias(es) of __<name> for <name> in: +;; $*${none} +EOF + for n in $*; do + cat << EOF + +;; double __${n} (double, word_t) +#ifdef F7MOD_D_${n}_ +_DEFUN __${n} + DALIAS $n + LALIAS ${n}l + .global F7_NAME($n) + ldi ZH, hi8(gs(F7_NAME($n))) + ldi ZL, lo8(gs(F7_NAME($n))) + F7jmp call_ddx +_ENDF __${n} +#endif /* F7MOD_D_${n}_ */ +EOF + done + ;; + + *) + exit 1 + ;; +esac Property changes on: libgcc/config/avr/libf7/f7wraps.sh ___________________________________________________________________ Added: svn:eol-style ## -0,0 +1 ## +native \ No newline at end of property Added: svn:executable ## -0,0 +1 ## +* \ No newline at end of property Added: svn:mime-type ## -0,0 +1 ## +text/plain \ No newline at end of property Index: libgcc/config/avr/libf7/libf7-array.def =================================================================== --- libgcc/config/avr/libf7/libf7-array.def (nonexistent) +++ libgcc/config/avr/libf7/libf7-array.def (working copy) @@ -0,0 +1,45 @@ +/* Copyright (C) 2019 Free Software Foundation, Inc. + + This file is part of LIBF7, which is part of GCC. + + GCC is free software; you can redistribute it and/or modify it under + the terms of the GNU General Public License as published by the Free + Software Foundation; either version 3, or (at your option) any later + version. + + GCC is distributed in the hope that it will be useful, but WITHOUT ANY + WARRANTY; without even the implied warranty of MERCHANTABILITY or + FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + for more details. + + Under Section 7 of GPL version 3, you are granted additional + permissions described in the GCC Runtime Library Exception, version + 3.1, as published by the Free Software Foundation. + + You should have received a copy of the GNU General Public License and + a copy of the GCC Runtime Library Exception along with this program; + see the files COPYING3 and COPYING.RUNTIME respectively. If not, see + <http://www.gnu.org/licenses/>. */ + +static const F7_PGMSPACE f7_t ARRAY_NAME[] = +{ + #define F7_CONST_DEF(NAME, FLAGS, M6, M5, M4, M3, M2, M1, M0, EXPO) \ + { .flags = FLAGS, .mant = { M0, M1, M2, M3, M4, M5, M6 }, .expo = EXPO }, + #include "libf7-const.def" + #undef F7_CONST_DEF +}; + +// static const uint8_t n_ARRAY_NAME = <Entries in ARRAY_NAME[]>. + +#define F7_n_NAME2(X) n_##X +#define F7_n_NAME1(X) F7_n_NAME2(X) + +F7_UNUSED static const uint8_t F7_n_NAME1 (ARRAY_NAME) = + #define F7_CONST_DEF(NAME, FLAGS, M6, M5, M4, M3, M2, M1, M0, EXPO) \ + + 1 + #include "libf7-const.def" + #undef F7_CONST_DEF +; + +#undef F7_n_NAME1 +#undef F7_n_NAME2 Property changes on: libgcc/config/avr/libf7/libf7-array.def ___________________________________________________________________ Added: svn:eol-style ## -0,0 +1 ## +native \ No newline at end of property Added: svn:mime-type ## -0,0 +1 ## +text/plain \ No newline at end of property Index: libgcc/config/avr/libf7/libf7-asm-object.mk =================================================================== --- libgcc/config/avr/libf7/libf7-asm-object.mk (nonexistent) +++ libgcc/config/avr/libf7/libf7-asm-object.mk (working copy) @@ -0,0 +1,20 @@ +# This file is included several times in a row, once for each element of +# $(iter-items). On each inclusion, we advance $o to the next element. +# $(iter-labels) is also advanced. +# This works similar to $(srcdir)/siditi-object.mk. + +o := $(firstword $(iter-items)) +iter-items := $(filter-out $o,$(iter-items)) + +$o-label := $(firstword $(iter-labels)) +iter-labels := $(wordlist 2,$(words $(iter-labels)),$(iter-labels)) + +f7_asm_$o$(objext): f7_asm_%$(objext): $(libf7)/libf7-asm.sx + $(gcc_compile) -DF7MOD_$($*-label)_ $(F7_ASM_FLAGS) \ + -c $< + +ifeq ($(enable_shared),yes) +f7_asm_$(o)_s$(objext): f7_asm_%_s$(objext): $(libf7)/libf7-asm.sx + $(gcc_s_compile) -DF7MOD_$($*-label)_ $(F7_ASM_FLAGS) \ + -c $< +endif Property changes on: libgcc/config/avr/libf7/libf7-asm-object.mk ___________________________________________________________________ Added: svn:eol-style ## -0,0 +1 ## +native \ No newline at end of property Added: svn:mime-type ## -0,0 +1 ## +text/plain \ No newline at end of property Index: libgcc/config/avr/libf7/libf7-asm.sx =================================================================== --- libgcc/config/avr/libf7/libf7-asm.sx (nonexistent) +++ libgcc/config/avr/libf7/libf7-asm.sx (working copy) @@ -0,0 +1,1664 @@ +;; Copyright (C) 2019 Free Software Foundation, Inc. +;; +;; This file is part of LIBF7, which is part of GCC. +;; +;; GCC is free software; you can redistribute it and/or modify it under +;; the terms of the GNU General Public License as published by the Free +;; Software Foundation; either version 3, or (at your option) any later +;; version. +;; +;; GCC is distributed in the hope that it will be useful, but WITHOUT ANY +;; WARRANTY; without even the implied warranty of MERCHANTABILITY or +;; FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License +;; for more details. +;; +;; Under Section 7 of GPL version 3, you are granted additional +;; permissions described in the GCC Runtime Library Exception, version +;; 3.1, as published by the Free Software Foundation. +;; +;; You should have received a copy of the GNU General Public License and +;; a copy of the GCC Runtime Library Exception along with this program; +;; see the files COPYING3 and COPYING.RUNTIME respectively. If not, see +;; <http://www.gnu.org/licenses/>. */ + +#ifndef __AVR_TINY__ + +#define ASM_DEFS_HAVE_DEFUN + +#include "asm-defs.h" +#include "libf7.h" + +#define ZERO __zero_reg__ +#define TMP __tmp_reg__ + +#define F7(name) F7_(name##_asm) + +.macro F7call name + .global F7(\name\()) + XCALL F7(\name\()) +.endm + +.macro F7jmp name + .global F7(\name\()) + XJMP F7(\name\()) +.endm + +;; Just for visibility in disassembly. +.macro LLL name + .global LLL.\name + LLL.\name: + nop +.endm + +.macro DEFUN name + .section .text.libf7.asm.\name, "ax", @progbits + .global F7(\name\()) + .func F7(\name\()) + F7(\name\()) : +.endm + +.macro ENDF name + .size F7(\name\()), . - F7(\name\()) + .endfunc +.endm + +.macro LABEL name + .global F7(\name\()) + F7(\name\()) : +.endm + +.macro _DEFUN name + .section .text.libf7.asm.\name, "ax", @progbits + .weak \name + .type \name, @function + \name : +.endm + +.macro _ENDF name + .size \name, . - \name +.endm + +.macro _LABEL name + .weak \name + .type \name, @function + \name : +.endm + +#define F7_NAME(X) F7_(X) + +;; Make a weak alias. +.macro ALIAS sym + .weak \sym + .type \sym, @function + \sym: +.endm + +;; Make a weak alias if double is 64 bits wide. +.macro DALIAS sym +#if defined (WITH_LIBF7_MATH_SYMBOLS) && __SIZEOF_DOUBLE__ == 8 +ALIAS \sym +#endif +.endm + +;; Make a weak alias if long double is 64 bits wide. +.macro LALIAS sym +#if defined (WITH_LIBF7_MATH_SYMBOLS) && __SIZEOF_LONG_DOUBLE__ == 8 +ALIAS \sym +#endif +.endm + +#define Off 1 +#define Expo (Off + F7_MANT_BYTES) + +#ifdef F7MOD_classify_ +;; r24 = classify (*Z) +;; NaN -> F7_FLAG_nan +;; INF -> F7_FLAG_inf [ | F7_FLAG_sign ] +;; ==0 -> F7_FLAG_zero +;; ... -> 0 [ | F7_FLAG_sign ] + +;; Clobbers: None (no TMP, no T). +DEFUN classify + + ld r24, Z + lsr r24 + brne .Lnan_or_inf + + ldd r24, Z+6+Off + tst r24 + brpl 0f + sbc r24, r24 + andi r24, F7_FLAG_sign + ret + +0: ldi r24, F7_FLAG_zero + ret + +.Lnan_or_inf: + rol r24 + ret + +ENDF classify +#endif /* F7MOD_classify_ */ + +#ifdef F7MOD_clr_ +DEFUN clr + std Z+0, ZERO + std Z+0+Off, ZERO + std Z+1+Off, ZERO + std Z+2+Off, ZERO + std Z+3+Off, ZERO + std Z+4+Off, ZERO + std Z+5+Off, ZERO + std Z+6+Off, ZERO + std Z+0+Expo, ZERO + std Z+1+Expo, ZERO + ret +ENDF clr + +#endif /* F7MOD_clr_ */ + +#ifdef F7MOD_clz_ +;; The libcc CLZ implementations like __clzsi2 aka. __builtin_clzl are +;; not very well suited for out purpose, so implement our own. + +#define ZBITS r26 +.macro .test.byte reg + or ZERO, \reg + brne .Loop_bit + subi ZBITS, -8 +.endm + +;; R26 = CLZ (uint64_t R18); CLZ (0) = 64. +;; Unchanged: T +DEFUN clzdi2 + clr ZBITS + ;; Catch the common case of normalized .mant for speed-up. + tst r25 + brmi 9f + .test.byte r25 + .test.byte r24 + .test.byte r23 + .test.byte r22 + .test.byte r21 + .test.byte r20 + .test.byte r19 + .test.byte r18 +.Ldone: + clr ZERO +9: ret + +.Loop_bit: + lsl ZERO + brcs .Ldone + inc ZBITS + rjmp .Loop_bit + +ENDF clzdi2 +#undef ZBITS +#endif /* F7MOD_clz_ */ + +#ifdef F7MOD_cmp_mant_ +DEFUN cmp_mant + + adiw X, 6 + Off + ld r24, X $ ldd TMP, Z+6+Off $ SUB r24, TMP + brne .Lunequal + + sbiw X, 6 + ld r24, X+ $ ldd TMP, Z+0+Off $ SUB r24, TMP + ld r24, X+ $ ldd TMP, Z+1+Off $ sbc r24, TMP + ld r24, X+ $ ldd TMP, Z+2+Off $ sbc r24, TMP + ld r24, X+ $ ldd TMP, Z+3+Off $ sbc r24, TMP + ld r24, X+ $ ldd TMP, Z+4+Off $ sbc r24, TMP + ld r24, X+ $ ldd TMP, Z+5+Off $ sbc r24, TMP + ;; MSBs are already known to be equal + breq 9f +.Lunequal: + sbc r24, r24 + sbci r24, -1 +9: sbiw X, 6 + Off + ret +ENDF cmp_mant +#endif /* F7MOD_cmp_mant_ */ + +#define CA 18 +#define C0 CA+1 +#define C1 C0+1 +#define C2 C0+2 +#define C3 C0+3 +#define C4 C0+4 +#define C5 C0+5 +#define C6 C0+6 +#define Carry r16 +#define Flags 18 + +#ifdef F7MOD_store_ +;; Z->flags = CA. +;; Z->mant = C[7]. +DEFUN store_mant.with_flags + st Z, CA + +;; Z->mant = C[7]. +LABEL store_mant + std Z+0+Off, C0 + std Z+1+Off, C1 + std Z+2+Off, C2 + std Z+3+Off, C3 + std Z+4+Off, C4 + std Z+5+Off, C5 + std Z+6+Off, C6 + ret +ENDF store_mant.with_flags +#endif /* F7MOD_store_ */ + +#ifdef F7MOD_load_ +;; CA = Z->flags +;; C[7] = Z->mant +DEFUN load_mant.with_flags + ld CA, Z + skipnext + +;; CA = 0 +;; C[7] = Z->mant +LABEL load_mant.clr_CA +LABEL load_mant.clr_flags + clr CA ; May be skipped + +;; C[7] = Z->mant +LABEL load_mant + ldd C0, Z+0+Off + ldd C1, Z+1+Off + ldd C2, Z+2+Off + ldd C3, Z+3+Off + ldd C4, Z+4+Off + ldd C5, Z+5+Off + ldd C6, Z+6+Off + ret +ENDF load_mant.with_flags +#endif /* F7MOD_load_ */ + +#ifdef F7MOD_copy_ +DEFUN copy + cp XL, ZL + cpc XH, ZH + breq 9f + adiw XL, 10 + adiw ZL, 10 + set + bld ZERO, 1 + bld ZERO, 3 ; ZERO = 0b1010 = 10. +.Loop: + ld TMP, -X + st -Z, TMP + dec ZERO + brne .Loop +9: ret +ENDF copy +#endif /* F7MOD_copy_ */ + +#ifdef F7MOD_copy_P_ +DEFUN copy_P + set + bld ZERO, 1 + bld ZERO, 3 ; ZERO = 0b1010 = 10. +.Loop: +#ifdef __AVR_HAVE_LPMX__ + lpm TMP, Z+ +#else + lpm + adiw Z, 1 +#endif /* Have LPMx */ + st X+, TMP + dec ZERO + brne .Loop + sbiw X, 10 + sbiw Z, 10 + ret +ENDF copy_P +#endif /* F7MOD_copy_P_ */ + +#ifdef F7MOD_copy_mant_ +DEFUN copy_mant + cp XL, ZL + cpc XH, ZH + breq 9f + adiw XL, 1 + adiw ZL, 1 + set + bld ZERO, 3 + dec ZERO ; ZERO = 7 +.Loop: + ld TMP, X+ + st Z+, TMP + dec ZERO + brne .Loop + sbiw XL, 8 + sbiw ZL, 8 +9: ret +ENDF copy_mant +#endif /* F7MOD_copy_mant_ */ + + +#ifdef F7MOD_clr_mant_lsbs_ +DEFUN clr_mant_lsbs + push r16 + mov r16, r20 + wmov XL, r24 + + wmov ZL, r22 + F7call load_mant + + F7call lshrdi3 + + clr CA + + F7call ashldi3 + + pop r16 + + wmov ZL, XL + F7jmp store_mant + +ENDF clr_mant_lsbs +#endif /* F7MOD_clr_mant_lsbs_ */ + + +#ifdef F7MOD_normalize_with_carry_ +;; Z = &f7_t +;; C[] = .mant may be not normalized +;; Carry === r16 = Addend to Z->expo in [-64, 128). +;; Normalize C[], set Flags, and adjust Z->expo. +;; Return CA (after normalization) in TMP. +;; Unchanged: T +#define Addend r17 +#define Zbits r26 +#define expL r26 +#define expH r27 +DEFUN normalize_with_carry + mov Addend, Carry + tst C6 + brmi .Lshift.0 + ;; r26 = CLZ (uint64_t R18) + F7call clzdi2 + cpi Zbits, 64 + breq .Lclr + sub Addend, Zbits + mov r16, Zbits + + F7call ashldi3 + ;; Assert (R25.7 == 1) +.Lshift.0: + mov TMP, CA + ld Flags, Z + + ;; .expo += Addend + ldd expL, Z+0+Expo + ldd expH, Z+1+Expo + ;; Sign-extend Addend + clr r16 + sbrc Addend, 7 + com r16 + + ;; exp += (int8_t) Addend, i.e. sign-extend Addend. + add expL, Addend + adc expH, r16 + brvc .Lnormal + tst r16 + brmi .Lclr + ;; Overflow +#if F7_HAVE_Inf == 1 + ori Flags, F7_FLAG_inf +#else + ldi Flags, F7_FLAG_nan +#endif /* Have Inf */ + ret + +.Lnormal: + std Z+0+Expo, expL + std Z+1+Expo, expH + ret + +.Lclr: + ;; Underflow or Zero. + clr TMP + .global __clr_8 + XJMP __clr_8 + +LABEL normalize.store_with_flags + ;; no rounding + set + skipnext +LABEL normalize.round.store_with_flags + ;; with rounding + clt ; skipped ? +LABEL normalize.maybe_round.store_with_flags + F7call normalize_with_carry + ;; We have: + ;; Z = &f7_t + ;; X = .expo + ;; C[] = .mant + ;; R18 = .flags + ;; TMP = byte below .mant after normalization + ;; T = 1 => no rounding. + brts .Lstore + lsl TMP + adc C0, ZERO + brcc .Lstore + adc C1, ZERO + adc C2, ZERO + adc C3, ZERO + adc C4, ZERO + adc C5, ZERO + adc C6, ZERO + brcc .Lstore + ;; We only come here if C6 overflowed, i.e. C[] is 0 now. + ;; .mant = 1.0 by restoring the MSbit. + ror C6 + ;; .expo += 1 and override the .expo stored during normalize. + adiw expL, 1 + std Z+0+Expo, expL + std Z+1+Expo, expH + +.Lstore: + F7call store_mant.with_flags + + ;; Return the byte below .mant after normalization. + ;; This is only useful without rounding; the caller will know. + mov R24, TMP + ret +ENDF normalize_with_carry +#endif /* F7MOD_normalize_with_carry_ */ + + +#ifdef F7MOD_normalize_ +;; Using above functionality from C. +;; f7_t* normalize (f7_t *cc) +;; Adjusts cc->expo +;; Clears cc->flags +DEFUN normalize + push r17 + push r16 + wmov ZL, r24 + F7call load_mant.clr_CA + clr Carry + st Z, ZERO + F7call normalize.store_with_flags + wmov r24, Z + pop r16 + pop r17 + ret +ENDF normalize +#endif /* F7MOD_normalize_ */ + + +#ifdef F7MOD_store_expo_ +#define Done r24 +#define expLO r24 +#define expHI r25 +;; expo == INT16_MAX => *Z = Inf, return Done = true. +;; expo == INT16_MIN => *Z = 0x0, return Done = true. +;; else => Z->expo = expo, return Done = false. +DEFUN store_expo + cpi expHI, 0x80 + cpc expLO, ZERO + breq .Ltiny + adiw expLO, 1 + brvs .Lhuge + sbiw expLO, 1 + std Z+0+Expo, expLO + std Z+1+Expo, expHI + ldi Done, 0 + ret + +.Lhuge: +#if F7_HAVE_Inf == 1 + ld Done, Z + andi Done, F7_FLAG_sign + ori Done, F7_FLAG_inf +#else + ldi Done, F7_FLAG_nan +#endif /* Have Inf */ + st Z, Done + ldi Done, 1 + ret + +.Ltiny: + ldi Done, 1 + F7jmp clr +ENDF store_expo +#endif /* F7MOD_store_expo_ */ + + +#ifdef F7MOD_set_u64_ +DEFUN set_s64 + set + skipnext + ;; ... +LABEL set_u64 + clt ; Skipped? + wmov Zl, r16 + ;; TMP holds .flags. + clr TMP + brtc .Lnot.negative + + bst C6, 7 + brtc .Lnot.negative + bld TMP, F7_FLAGNO_sign + .global __negdi2 + XCALL __negdi2 + +.Lnot.negative: + st Z, TMP + std Z+0+Expo, ZERO + std Z+1+Expo, ZERO + ldi Carry, 63 + F7call normalize.round.store_with_flags + wmov r24, Z + wmov r16, Z ; Unclobber r16. + ret +ENDF set_s64 +#endif /* F7MOD_set_u64_ */ + + +#ifdef F7MOD_to_integer_ +#define Mask r26 +DEFUN to_integer + wmov ZL, r24 + mov Mask, r22 + + F7call load_mant.with_flags + + sbrc Flags, F7_FLAGNO_nan + rjmp .Lset_0x8000 + + sbrc Flags, F7_FLAGNO_inf + rjmp .Lsaturate + + sbrs C6, 7 + rjmp .Lset_0x0000 + + bst Flags, F7_FLAGNO_sign + ldd r27, Z+0+Expo + ;; Does .expo have bits outside Mask? ... + mov TMP, Mask + com TMP + and TMP, r27 + ldd r27, Z+1+Expo + tst r27 + brmi .Lset_0x0000 ; ...yes: .expo is < 0 => return 0 + or TMP, r27 + brne .Lsaturate.T ; ...yes: .expo > Mask => saturate + + ;; ...no: Shift right to meet .expo = 0. + PUSH r16 + ldd r16, Z+0+Expo + eor r16, Mask + and r16, Mask + clr CA + F7call lshrdi3 + POP r16 + tst C6 + brmi .Lsaturate.T ; > INTxx_MAX => saturate + + rcall .Lround + brmi .Lsaturate.T ; > INTxx_MAX => saturate + + brtc 9f ; >= 0 => return + sbrc Mask, 5 + .global __negdi2 + XJMP __negdi2 + sbrc Mask, 4 + .global __negsi2 + XJMP __negsi2 + neg C6 + neg C5 + sbci C6, 0 +9: ret + +.Lsaturate: + bst Flags, F7_FLAGNO_sign +.Lsaturate.T: + +#if F7_HAVE_Inf + brtc .Lset_0x7fff + ;; -Inf => return 1 + INTxx_MIN + mov ZL, Flags + .global __clr_8 + XCALL __clr_8 + ldi C6, 0x80 + + ldi CA+0, 0x01 + + sbrs Mask, 5 + ldi CA+4, 0x01 + + sbrs Mask, 4 + ldi CA+6, 0x01 + ret + +.Lset_0x7fff: + ;; +Inf => return INTxx_MAX + sec + .global __sbc_8 + XCALL __sbc_8 + ldi C6, 0x7f + ret +#endif /* F7_HAVE_Inf */ + +.Lset_0x8000: + ;; NaN => return INTxx_MIN + .global __clr_8 + XCALL __clr_8 + ldi C6, 0x80 + ret + +.Lset_0x0000: + ;; Small value => return 0x0 + .global __clr_8 + XJMP __clr_8 + +.Lround: + ;; C6.7 is known to be 0 here. + ;; Return N = 1 iff we have to saturate. + cpi Mask, 0xf + breq .Lround16 + cpi Mask, 0x1f + breq .Lround32 + + ;; For now, no rounding in the 64-bit case. This rounding + ;; would have to be integrated into the right-shift. + cln + ret + +.Lround32: + rol C2 + adc C3, ZERO + adc C4, ZERO + rjmp 2f + +.Lround16: + rol C4 +2: adc C5, ZERO + adc C6, ZERO + ret +ENDF to_integer +#endif /* F7MOD_to_integer_ */ + + +#ifdef F7MOD_to_unsigned_ +#define Mask r26 +DEFUN to_unsigned + wmov ZL, r24 + mov Mask, r22 + + F7call load_mant.with_flags + + sbrc Flags, F7_FLAGNO_nan + rjmp .Lset_0xffff + + sbrc Flags, F7_FLAGNO_sign + rjmp .Lset_0x0000 + + sbrc Flags, F7_FLAGNO_inf + rjmp .Lset_0xffff + + sbrs C6, 7 + rjmp .Lset_0x0000 + + ldd r27, Z+0+Expo + ;; Does .expo have bits outside Mask? ... + mov TMP, Mask + com TMP + and TMP, r27 + ldd r27, Z+1+Expo + tst r27 + brmi .Lset_0x0000 ; ...yes: .expo is < 0 => return 0 + or TMP, r27 + brne .Lset_0xffff ; ...yes: .expo > Mask => saturate + + ;; ...no: Shift right to meet .expo = 0. + PUSH r16 + ldd r16, Z+0+Expo + eor r16, Mask + and r16, Mask + clr CA + F7call lshrdi3 + POP r16 + + ;; Rounding + ;; ??? C6.7 is known to be 0 here. + cpi Mask, 0xf + breq .Lround16 + cpi Mask, 0x1f + breq .Lround32 + + ;; For now, no rounding in the 64-bit case. This rounding + ;; would have to be integrated into the right-shift. + ret + +.Lround32: + rol C2 + adc C3, ZERO + adc C4, ZERO + rjmp 2f + +.Lround16: + rol C4 +2: adc C5, ZERO + adc C6, ZERO + brcs .Lset_0xffff ; Rounding overflow => saturate + ret + +.Lset_0xffff: + ;; return UINTxx_MAX + sec + .global __sbc_8 + XJMP __sbc_8 + +.Lset_0x0000: + ;; Small value => return 0x0 + .global __clr_8 + XJMP __clr_8 + +ENDF to_unsigned +#endif /* F7MOD_to_unsigned_ */ + + +#ifdef F7MOD_addsub_mant_scaled_ +;; int8_t f7_addsub_mant_scaled_asm (f7_t *r24, const f7_t *r22, const f7_t 20*, +;; uint8_t r18); +;; R18.0 = 1 : ADD +;; R18.0 = 0 : SUB +;; R18[7..1] : Scale +;; Compute *R24 = *R22 + *R20 >> R18[7..1]. + +#define BA 10 +#define B0 BA+1 +#define B1 B0+1 +#define B2 B0+2 +#define B3 B0+3 +#define B4 B0+4 +#define B5 B0+5 +#define B6 B0+6 + +DEFUN addsub_mant_scaled + do_prologue_saves 10 + + bst r18, 0 ;; ADD ? + lsr r18 + mov r16, r18 + + wmov ZL, r20 + wmov YL, r22 + ;; C[] = bb >> shift + wmov XL, r24 + + F7call load_mant.clr_CA + F7call lshrdi3 + + wmov BA, CA + wmov B1, C1 + wmov B3, C3 + wmov B5, C5 + wmov ZL, YL + F7call load_mant.clr_CA + + wmov ZL, XL + + brts .Ladd + + .global __subdi3 + XCALL __subdi3 + + breq .Lzero + brcc .Lround + ;; C = 1: Can underflow happen at all ? +.Lzero: + F7call clr + rjmp .Lepilogue + +.Ladd: + .global __adddi3 + XCALL __adddi3 + brcc .Lround + ldi Carry, 1 + .global __lshrdi3 + XCALL __lshrdi3 + ori C6, 1 << 7 + skipnext +.Lround: + clr Carry ; skipped? + F7call normalize.round.store_with_flags + +.Lepilogue: + do_epilogue_restores 10 + +ENDF addsub_mant_scaled + +#if !defined (__AVR_HAVE_MOVW__) || !defined (__AVR_HAVE_JMP_CALL__) +DEFUN lshrdi3 + .global __lshrdi3 + XJMP __lshrdi3 +ENDF lshrdi3 +DEFUN ashldi3 + .global __ashldi3 + XJMP __ashldi3 +ENDF ashldi3 +#else + +# Basically just a wrapper around libgcc's __lshrdi3. +DEFUN lshrdi3 + ;; Handle bit 5 of shift offset. + sbrs r16, 5 + rjmp 4f + wmov CA, C3 + wmov C1, C5 + clr C6 $ clr C5 $ wmov C3, C5 +4: + ;; Handle bit 4 of shift offset. + sbrs r16, 4 + rjmp 3f + wmov CA, C1 + wmov C1, C3 + wmov C3, C5 + clr C6 $ clr C5 +3: + ;; Handle bits 3...0 of shift offset. + push r16 + andi r16, 0xf + breq 0f + + .global __lshrdi3 + XCALL __lshrdi3 +0: + pop r16 + ret +ENDF lshrdi3 + +# Basically just a wrapper around libgcc's __ashldi3. +DEFUN ashldi3 + ;; Handle bit 5 of shift offset. + sbrs r16, 5 + rjmp 4f + wmov C5, C1 + wmov C3, CA + clr C2 $ clr C1 $ wmov CA, C1 +4: + ;; Handle bit 4 of shift offset. + sbrs r16, 4 + rjmp 3f + wmov C5, C3 + wmov C3, C1 + wmov C1, CA + clr CA $ clr C0 +3: + ;; Handle bits 3...0 of shift offset. + push r16 + andi r16, 0xf + breq 0f + + .global __ashldi3 + XCALL __ashldi3 +0: + pop r16 + ret +ENDF ashldi3 +#endif /* Small device */ + +#endif /* F7MOD_addsub_mant_scaled_ */ + +#if defined F7MOD_mul_mant_ && defined (__AVR_HAVE_MUL__) + #define A0 11 + #define A1 A0+1 + #define A2 A0+2 + #define A3 A0+3 + #define A4 A0+4 + #define A5 A0+5 + #define A6 A0+6 + + #define TT0 26 + #define TT1 TT0+1 + #define TT2 28 + #define TT3 TT2+1 + + #define BB 10 + +;; R18.0 = 1: No rounding. + +DEFUN mul_mant + do_prologue_saves 10 + bst r18, 0 + push r25 + push r24 + movw ZL, r22 + LDD A0, Z+0+Off + LDD A1, Z+1+Off + LDD A2, Z+2+Off + LDD A3, Z+3+Off + LDD A4, Z+4+Off + LDD A5, Z+5+Off + LDD A6, Z+6+Off + movw ZL, r20 + + ;; 6 * 6 -> 6:5 + ;; 4 * 6 -> 4:3 + ;; 2 * 6 -> 2:1 + ;; 0 * 6 -> 0:a + ldd BB, Z+6+Off + mul A6, BB $ movw C5, r0 + mul A4, BB $ movw C3, r0 + mul A2, BB $ movw C1, r0 + mul A0, BB $ movw CA, r0 + + ;; 5 * 6 -> 5:4 + ;; 3 * 6 -> 3:2 + ;; 1 * 6 -> 1:0 + mul A5, BB $ movw TT2, r0 + mul A3, BB $ movw TT0, r0 + mul A1, BB + ADD C0, r0 $ adc C1, r1 + adc C2, TT0 $ adc C3, TT1 + adc C4, TT2 $ adc C5, TT3 $ clr ZERO + adc C6, ZERO + ;; Done B6 + + ;; 3 * 3 -> 0:a + ;; 4 * 4 -> 2:1 + ;; 5 * 5 -> 4:3 + ldd BB, Z+3+Off $ mul A3, BB $ movw TT0, r0 + ldd BB, Z+4+Off $ mul A4, BB $ movw TT2, r0 + ldd BB, Z+5+Off $ mul A5, BB + + ADD CA, TT0 $ adc C0, TT1 + adc C1, TT2 $ adc C2, TT3 + adc C3, r0 $ adc C4, r1 + brcc .+2 + adiw C5, 1 + + ;; 6 * 5 -> 5:4 + ;; 4 * 5 -> 3:2 + ;; 2 * 5 -> 1:0 + ;; 0 * 5 -> a:- + mul A0, BB + ;; A0 done +#define Atmp A0 + + mov Atmp, r1 + mul A6, BB $ movw TT2, r0 + mul A4, BB $ movw TT0, r0 + mul A2, BB + + ADD CA, Atmp + adc C0, r0 $ adc C1, r1 + adc C2, TT0 $ adc C3, TT1 + adc C4, TT2 $ adc C5, TT3 $ clr ZERO + adc C6, ZERO + + ;; 1 * 5 -> 0:a + ;; 3 * 5 -> 2:1 + ;; 6 * 4 -> 4:3 + mul A1, BB $ movw TT0, r0 + mul A3, BB $ movw TT2, r0 + ldd BB, Z+4+Off + mul A6, BB + + ADD CA, TT0 $ adc C0, TT1 + adc C1, TT2 $ adc C2, TT3 + adc C3, r0 $ adc C4, r1 $ clr ZERO + adc C5, ZERO $ adc C6, ZERO + ;; B5 done + + ;; 6 * 3 -> 3:2 + ;; 6 * 1 -> 1:0 + ;; 4 * 1 -> a:- + mov TT0, A6 $ ldd TMP, Z+3+Off + mov BB, A4 $ ldd Atmp, Z+1+Off + rcall .Lmul.help.3 + + ;; 5 * 4 -> 3:2 + ;; 5 * 2 -> 1:0 + ;; 3 * 2 -> a:- + mov TT0, A5 $ ldd TMP, Z+4+Off + mov BB, A3 $ ldd Atmp, Z+2+Off + rcall .Lmul.help.3 + + ;; 4 * -> 3:2 (=0) + ;; 4 * 3 -> 1:0 + ;; 2 * 3 -> a:- + mov TT0, A4 $ clr TMP + mov BB, A2 $ ldd Atmp, Z+3+Off + rcall .Lmul.help.3 + + ;; 3 * . -> 3:2 (=0) + ;; 3 * 4 -> 1:0 + ;; 1 * 4 -> a:- + mov TT0, A3 $ clr TMP + mov BB, A1 $ ldd Atmp, Z+4+Off + rcall .Lmul.help.3 + + ;; . * ? -> 3:2 (=0) + ;; . * 0 -> 1:0 (=0) + ;; 5 * 0 -> a:- + clr TT0 + mov BB, A5 $ ldd Atmp, Z+0+Off + rcall .Lmul.help.3 + + clr TT3 ;; Asserted by .Lmul.help.2 + ;; 6 * 2 -> 2:1 + ;; 6 * 0 -> 0:a + $ ldd TMP, Z+2+Off + mov BB, A6 ;$ ldd Atmp, Z+0+Off + rcall .Lmul.help.2 + + ;; 5 * 3 -> 2:1 + ;; 5 * 1 -> 0:a + $ ldd TMP, Z+3+Off + mov BB, A5 $ ldd Atmp, Z+1+Off + rcall .Lmul.help.2 + + ;; 4 * . -> 2:1 (=0) + ;; 4 * 2 -> 0:a + $ clr TMP + mov BB, A4 $ ldd Atmp, Z+2+Off + rcall .Lmul.help.2 + + ;; 2 * . -> 2:1 (=0) + ;; 2 * 4 -> 0:a + $ clr TMP + mov BB, A2 $ ldd Atmp, Z+4+Off + rcall .Lmul.help.2 + + ;; Finally... + + pop ZL + pop ZH + ;; The high byte is at least 0x40 and at most 0xfe. + ;; The result has to be left-shifted by one in order to scale it + ;; correctly. + + ldi Carry, 1 + F7call normalize.maybe_round.store_with_flags + + do_epilogue_restores 10 + +;; TT0 * Tmp -> 3:2 +;; TT0 * Atmp -> 1:0 +;; BB * Atmp -> a:- +;; +;; Clobbers : TMP, TT0...TT3. +;; Sets : ZERO = 0. +.Lmul.help.3: + mul TT0, TMP $ movw TT2, r0 + mul TT0, Atmp $ movw TT0, r0 + mul BB, Atmp + + ADD CA, r1 + adc C0, TT0 $ adc C1, TT1 + adc C2, TT2 +.Lmul.help.3.C3: $ adc C3, TT3 $ clr ZERO + adc C4, ZERO $ adc C5, ZERO + adc C6, ZERO + ret + +;; BB * TMP -> 2:1 +;; BB * Atmp -> 0:a +;; +;; Asserts : TT3 = 0 +;; Clobbers : TMP, TT0, TT1. +;; Sets : ZERO = 0. +.Lmul.help.2: + mul BB, TMP $ movw TT0, r0 + mul BB, Atmp + ADD CA, r0 $ adc C0, r1 + adc C1, TT0 $ adc C2, TT1 + rjmp .Lmul.help.3.C3 + +ENDF mul_mant +#endif /* F7MOD_mul_mant_ && MUL */ + + +#if defined (F7MOD_div_) + +;; Dividend is C[] + +;; Divisor +#define A0 9 +#define A1 10 +#define A2 11 +#define A3 12 +#define A4 13 +#define A5 14 +#define A6 15 + +;; Quotient +#define Q0 0 /* === TMP */ +#define Q1 Q0+1 /* === ZERO */ +#define Q2 26 +#define Q3 Q2+1 +#define Q4 28 +#define Q5 Q4+1 +#define Q6 16 +#define Q7 Q6+1 + +#define Cnt CA +#define QBits r8 + +DEFUN div + do_prologue_saves 12 + + ;; Number of bits requested for the quotient. + ;; This is usually 2 + F7_MANT_BITS. + mov QBits, r20 + wmov ZL, r22 + LDD A0, Z+0+Off + LDD A1, Z+1+Off + LDD A2, Z+2+Off + LDD A3, Z+3+Off + LDD A4, Z+4+Off + LDD A5, Z+5+Off + LDD A6, Z+6+Off + wmov ZL, r24 + F7call load_mant + + ;; Clear quotient Q[]. + clr Q0 ; === TMP + ;clr Q1 ; === ZERO + wmov Q2, Q0 + wmov Q4, Q0 + wmov Q6, Q0 + + ;; C[] and A[] are valid mantissae, i.e. their MSBit is set. Therefore, + ;; quotient Q[] will be in [0x0.ff..., 0x0.40...] and to adjust Q[] we + ;; need at most 1 left-shift. Compute F7_MANT_BITS + 2 bits of the + ;; quotient: One bit is used for rounding, and one bit might be consumed + ;; by the mentioned left-shift. + mov Cnt, QBits + rjmp .Loop_start + +.Loop: + ;; Shift dividend. + LSL C0 + rol C1 + rol C2 + rol C3 + rol C4 + rol C5 + rol C6 + brcs .Lfits + ;; Compare dividend against divisor. +.Loop_start: + CP C0, A0 + cpc C1, A1 + cpc C2, A2 + cpc C3, A3 + cpc C4, A4 + cpc C5, A5 + cpc C6, A6 + ;; Shift 0 into quotient. + brlo 1f +.Lfits: + ;; Divisor fits into dividend. + SUB C0, A0 + sbc C1, A1 + sbc C2, A2 + sbc C3, A3 + sbc C4, A4 + sbc C5, A5 + sbc C6, A6 + ;; Shift 1 into quotient. + sec + rol Q0 + skipnext +1: lsl Q0 + rol Q1 + rol Q2 + rol Q3 + rol Q4 + rol Q5 + rol Q6 + rol Q7 + dec Cnt + brne .Loop + + wmov CA, Q0 + wmov C1, Q2 + wmov C3, Q4 + wmov C5, Q6 + clr ZERO + + ldi Carry, 64 + sub Carry, QBits + F7call normalize.round.store_with_flags + + do_epilogue_restores 12 +ENDF div + +#endif /* F7MOD_div_ */ + + +#if defined (F7MOD_sqrt16_) && defined (__AVR_HAVE_MUL__) + +#define Mask C6 +#define Q0 C3 /* = R22 */ +#define Q1 C4 /* = R23 */ + +;; uint16_t R24 = sqrt16_XXX (uint16_t R24); +;; Clobbers: R22, R23, TMP. +;; +;; XXX = floor: Return integral part of square-root of R25:R24 with R25 = 0. +;; Error is in [0, -1 LSB). +;; XXX = round: Return quare-root of R25:R24 rounded to nearest integer. +;; R25 = (Q[] >= 65281) = (Q > 0xff00), i.e. if Q[] is not +;; bigger than 0xff00, then the result fits in 8 bits. +;; Return C = 0 if the result is the same as for XXX = floor, +;; error in [0, -1/2 LSB) +;; Return C = 1 if the result is one higher than for XXX = floor, +;; error in [1/2 LSB, 0). +DEFUN sqrt16_round + set + skipnext + ;; ... +LABEL sqrt16_floor + clt ; Skipped? + movw Q0, r24 + clr C5 + ldi Mask, 1 << 7 + +.Loop_mask: + add C5, Mask + mul C5, C5 + cp Q0, R0 + cpc Q1, R1 + brsh 1f + sub C5, Mask +1: lsr Mask + brne .Loop_mask + + brtc .Ldone ; No rounding => C6 will be 0. + + ;; Rounding: (X + 1/2)^2 = X^2 + X + 1/4, thus probing + ;; for bit -1 is testing Q[] against C5^2 + C5. + mul C5, C5 + add R0, C5 + adc R1, C6 ; Exploit C6 === Mask = 0. + cp R0, Q0 + cpc R1, Q1 + brcc .Ldone + ;; If C5^2 + C5 + 1/4 fits into Q[], then round up and C = 1. + adiw C5, 1 ; Exploit C6 === Mask = 0. + sec + +.Ldone: + clr __zero_reg__ + ret +ENDF sqrt16_round +#undef Mask +#undef Q0 +#undef Q1 +#endif /* F7MOD_sqrt16_ && MUL */ + +#ifdef F7MOD_sqrt_approx_ +DEFUN sqrt_approx + push r17 + push r16 + wmov XL, r24 + wmov ZL, r22 + + ;; C[] = 0. + .global __clr_8 + XCALL __clr_8 + + ldd C5, Z+5+Off + ldd C6, Z+6+Off + + ldd Carry, Z+0+Expo + ldd TMP, Z+1+Expo + wmov ZL, XL + + st Z, ZERO + + asr TMP + ror Carry + std Z+1+Expo, TMP + std Z+0+Expo, Carry + + ;; Re-interpreting our Q-format 1.xx mantissa as Q2.yy, we have to shift + ;; the mantissa to the right by 1. As we need an even exponent, multiply + ;; the mantissa by 2 for odd exponents, i.e. only right-shift if .expo + ;; is even. + + brcs 1f + lsr C6 + ror C5 + +1: + F7call sqrt16_round + + ;; sqrt16_round() returns: C = 0: error in [0, -1/2 LSB). + ;; C = 1: error in [1/2 LSB, 0) + + brcc 2f + ;; Undo the round-up from sqrt16_round(); this will transform to + ;; error in [-1/2 LSB, -1 LSB). + sbiw C5, 1 + ;; Together with the correct bit C4.7, the error is in [0, -1/2 LSB). + ori C4, 1 << 7 + +2: ;; Setting C4.6 adds 1/4 LSB and the error is now in [1/4 LSB, -1/4 LSB) + ;; in either case. + ori C4, 1 << 6 + + ;; ???????????? + ;; sqrt16_round() runs on integers which means that it computes the + ;; square root of mant * 2^14 if we regard mant as Q-format 2.yy, + ;; i.e. 2 integral bits. The result is sqrt(mant) * 2^7, + ;; and in order to get the same scaling like the input, .expo has to + ;; be adjusted by 7. ??????????????? + + ldi Carry, 8 + F7call normalize.store_with_flags + + pop r16 + pop r17 + ret + +ENDF sqrt_approx +#endif /* F7MOD_sqrt_approx_ */ + + +#undef CA +#undef C0 +#undef C1 +#undef C2 +#undef C3 +#undef C4 +#undef C5 +#undef C6 +#undef Carry + + +#ifdef F7MOD_D_fabs_ +_DEFUN __fabs + DALIAS fabs + LALIAS fabsl + andi R25, 0b01111111 + ret +_ENDF __fabs +#endif /* F7MOD_D_fabs_ */ + + +#ifdef F7MOD_D_neg_ +_DEFUN __neg +_LABEL __negdf2 + subi R25, 0b10000000 + ret +_ENDF __neg +#endif /* F7MOD_D_neg_ */ + + +#ifdef F7MOD_D_signbit_ +_DEFUN __signbit + DALIAS signbit + LALIAS signbitl + bst R25, 7 + clr R25 + clr R24 + bld R24, 0 + ret +_ENDF __signbit +#endif /* F7MOD_D_signbit_ */ + + +#ifdef F7MOD_D_copysign_ +_DEFUN __copysign + DALIAS copysign + LALIAS copysignl + bst R17, 7 + bld R25, 7 + ret +_ENDF __copysign +#endif /* F7MOD_D_copysign_ */ + + +#ifdef F7MOD_D_isinf_ +_DEFUN __isinf + DALIAS isinf + LALIAS isinfl + F7call class_D + ;; Inf: T = Z = 1. + brtc 0f + ldi R24, 1 + breq 1f +0: + clr R24 +1: + clr R25 + ret +_ENDF __isinf +#endif /* F7MOD_D_isinf_ */ + + +#ifdef F7MOD_D_isnan_ +_DEFUN __isnan + DALIAS isnan + LALIAS isnanl + F7call class_D + ;; NaN: T = 1, Z = 0. + brtc 0f + ldi R24, 1 + brne 1f +0: + clr R24 +1: + clr R25 + ret +_ENDF __isnan +#endif /* F7MOD_D_isnan_ */ + + +#ifdef F7MOD_D_isfinite_ +_DEFUN __isfinite + DALIAS isfinite + LALIAS isfinitel + F7call class_D + ;; Number <=> T = 0. + bld R24, 0 + com R24 + andi R24, 1 + clr R25 + ret +_ENDF __isfinite +#endif /* F7MOD_D_isfinite_ */ + + +#ifdef F7MOD_D_class_ +;; The encoded exponent has 11 Bits. +#define MAX_BIASED_EXPO 0b0111111111110000 + +;; Classify a double in R18[] +;; Number: T-Flag = 0. +;; +-Inf : T-Flag = 1, Z-Flag = 1. +;; NaN : T-Flag = 1, Z-Flag = 0. +DEFUN class_D + wmov R26, R24 + andi R26, lo8 (MAX_BIASED_EXPO) + andi R27, hi8 (MAX_BIASED_EXPO) + subi R26, lo8 (MAX_BIASED_EXPO) + sbci R27, hi8 (MAX_BIASED_EXPO) + clt + brne .L.number + set + ;; Set sign and expo to 0. + clr R25 + andi R24, lo8 (~MAX_BIASED_EXPO) + ;; What remains is the mantissa. + ;; Mantissa == 0 => +/-Inf. + ;; Mantissa != 0 => NaN. + ;; Compare R18[] against sign_extend(R26) with R26 = 0. + .global __cmpdi2_s8 + XJMP __cmpdi2_s8 +.L.number: + ret + +ENDF class_D +#endif /* F7MOD_D_class_ */ + + +#ifdef F7MOD_call_dd_ + +;; Provide double wrappers for functions that operate on f7_t and get f7_t*. +;; +;; We set up a frame of sizeof(f7_t), convert the input double in R18[] to +;; f7_t in that frame location, then call *Z and finally convert the result f7_t +;; to double R18[] if that's requested. +;; +;; call_dd: double func (double A) +;; void (*Z) (f7_t *aa, const f7_t *aa) +;; +;; call_dx: double func (type_t A) , sizeof(type_t) <= 4 +;; void (*Z) (f7_t *aa, type_t) +;; +;; call_xd: type_t func (double A) +;; type_t (*Z) (const f7_t *aa) +;; +;; call_ddx: double func (double A, word_t) , sizeof (word_t) <= 2 +;; void (*Z) (f7_t *aa, const f7_t *aa, word_t) + +#define WHAT R13 + +DEFUN call_dd ; WHAT = R13 = 3 + inc ZERO +LABEL call_xd ; WHAT = R13 = 2 + inc ZERO +LABEL call_ddx ; WHAT = R13 = 1 + inc ZERO +LABEL call_dx ; WHAT = R13 = 0 + push WHAT + mov WHAT, ZERO + clr ZERO + ;; R14/R15 hold Z, the address of the f7_worker function, until we need it. + push r14 + push r15 + wmov r14, Z + +#define n_pushed 4 +#define n_frame 10 + + do_prologue_saves n_pushed, n_frame + ;; Y = FramePointer + 1 + adiw Y, 1 + dec WHAT + brmi .Ldx ; WHAT was initially 0. + ;; FP + 1 = (f7_t) arg1 + wmov r16, Y + ;; The double argument is in R18[]. + XCALL F7_NAME (set_double_impl) + tst WHAT + brne .Lno.ddx ; WHAT was initially != 1. + ;; call_ddx: Set R20/21 to the 2-byte scalar / pointer argument. + ;; Fetch it from where prologue_saves put it. + ldd r20, Y + n_frame + 3 ; Saved R16 + ldd r21, Y + n_frame + 2 ; Saved R17 +.Lno.ddx: + wmov r22, Y ; &arg1 (input) +.Ldo.dx: + wmov r24, Y ; &arg1 (output) + wmov Z, r14 + XICALL + dec WHAT + breq .Lepilogue ; WHAT was initially 2: Return non-double. + wmov r24, Y ; &arg1 + XCALL F7_NAME (get_double) +.Lepilogue: + ;; + 3 to account for R13...R15 pushed prior to do_prologue_saves. + do_epilogue_restores n_pushed + 3, n_frame + +.Ldx: + ;; call_dx: Copy the 4-byte input scalar from R22[4] to R20[4]. + wmov r20, r22 + wmov r22, r24 + rjmp .Ldo.dx + +ENDF call_dd +#endif /* F7MOD_call_dd_ */ + + +#ifdef F7MOD_call_ddd_ + +;; Provide double wrappers for functions that operate on f7_t and get f7_t*. +;; +;; We set up a frame of 2 * sizeof(f7_t), convert the input doubles in R18[] +;; and R10[] to f7_t in these frame locations, then call *Z and finally +;; convert the result f7_t to double R18[] if that's requested. +;; +;; call_ddd: double func (double A, double B) +;; void (*Z) (f7_t *aa, const f7_t *aa, const f7_t *bb) +;; +;; call_xdd: type_t func (double A, double B) +;; type_t (*Z) (const f7_t *aa, const f7_t *bb) + +DEFUN call_ddd + inc ZERO +LABEL call_xdd + ;; R8/R9 hold Z, the address of the f7_worker function, until we need it. + push r9 + push r8 + wmov r8, Z + ;; This is an argument to call.2 and will be accessed by the arg pointer. + push ZERO + clr ZERO + rcall call.2 + pop TMP + pop r8 + pop r9 + ret + +#define n_pushed 4 +#define n_frame 20 + +call.2: + do_prologue_saves n_pushed, n_frame + ;; Y = FramePointer + 1 + adiw Y, 1 + ;; FP + 1 = (f7_t) arg1 + wmov r16, Y + ;; First double argument is already in R18[]. + XCALL F7_NAME (set_double_impl) + ;; FP + 11 = (f7_t) arg2 + wmov r16, Y + subi r16, lo8 (-10) + sbci r17, hi8 (-10) + ;; Move second double argument to R18[]. + wmov r18, r10 + wmov r20, r12 + wmov r22, r14 + ;; Get high word of arg2 from where prologue_saves put it. + ldd r24, Y + n_frame + 3 ; Saved R16 + ldd r25, Y + n_frame + 2 ; Saved R17 + XCALL F7_NAME (set_double_impl) + ;; Z (f7_t *arg1, const f7_t *arg1, const f7_t *arg2) + wmov Z, r8 + wmov r24, Y ; &arg1 + ;; WHAT == 0 => call_xdd + ;; WHAT != 0 => call_ddd + ldd TMP, Y + n_frame + n_pushed + PC_SIZE + tst TMP + breq .Lxdd + wmov r22, Y ; &arg1 + wmov r20, r16 ; &arg2 + XICALL + wmov r24, Y ; &arg1 + XCALL F7_NAME (get_double) +.Lepilogue: + do_epilogue_restores n_pushed, n_frame +.Lxdd: + wmov r22, r16 ; &arg2 + XICALL + rjmp .Lepilogue +ENDF call_ddd +#endif /* F7MOD_call_ddd_ */ + +#include "f7-wraps.h" + +#endif /* !AVR_TINY */ Property changes on: libgcc/config/avr/libf7/libf7-asm.sx ___________________________________________________________________ Added: svn:eol-style ## -0,0 +1 ## +native \ No newline at end of property Added: svn:mime-type ## -0,0 +1 ## +text/plain \ No newline at end of property Index: libgcc/config/avr/libf7/libf7-c-object.mk =================================================================== --- libgcc/config/avr/libf7/libf7-c-object.mk (nonexistent) +++ libgcc/config/avr/libf7/libf7-c-object.mk (working copy) @@ -0,0 +1,20 @@ +# This file is included several times in a row, once for each element of +# $(iter-items). On each inclusion, we advance $o to the next element. +# $(iter-labels) is also advanced. +# This works similar to $(srcdir)/siditi-object.mk. + +o := $(firstword $(iter-items)) +iter-items := $(filter-out $o,$(iter-items)) + +$o-label := $(firstword $(iter-labels)) +iter-labels := $(wordlist 2,$(words $(iter-labels)),$(iter-labels)) + +f7_c_$o$(objext): f7_c_%$(objext): $(libf7)/libf7.c + $(gcc_compile) -DF7MOD_$($*-label)_ $(F7_C_FLAGS) \ + -c $< + +ifeq ($(enable_shared),yes) +f7_c_$(o)_s$(objext): %_s$(objext): $(libf7)/libf7.c + $(gcc_s_compile) -DF7MOD_$($*-label)_ $(F7_C_FLAGS) \ + -c $< +endif Property changes on: libgcc/config/avr/libf7/libf7-c-object.mk ___________________________________________________________________ Added: svn:eol-style ## -0,0 +1 ## +native \ No newline at end of property Added: svn:mime-type ## -0,0 +1 ## +text/plain \ No newline at end of property Index: libgcc/config/avr/libf7/libf7-common.mk =================================================================== --- libgcc/config/avr/libf7/libf7-common.mk (nonexistent) +++ libgcc/config/avr/libf7/libf7-common.mk (working copy) @@ -0,0 +1,102 @@ +# f7_c_*.o modules from libf7.c. +F7_C_PARTS += set_s16 set_u16 set_s32 set_u32 init +F7_C_PARTS += get_s16 get_u16 get_s32 get_u32 get_s64 get_u64 +F7_C_PARTS += lrint ldexp frexp madd_msub madd msub hypot +F7_C_PARTS += addsub add sub mulx mul square divx div div1 fmod sqrt cbrt +F7_C_PARTS += Ineg Iadd Isub Imul Idiv IRsub Isquare Ildexp Isqrt +F7_C_PARTS += set_float get_float get_double set_double set_pdouble +F7_C_PARTS += fabs neg fmin fmax minmax truncx trunc floor ceil round lround +F7_C_PARTS += horner logx log log10 log2 exp pow10 pow powi +F7_C_PARTS += sin cos tan cotan sincos sinh cosh tanh sinhcosh +F7_C_PARTS += asinacos asin acos atan atan2 +F7_C_PARTS += abscmp_msb_ge cmp cmp_abs cmp_unordered + +F7_C_PARTS += const_1 const_1_2 const_1_3 +F7_C_PARTS += const_pi const_ln2 const_1_ln2 const_ln10 const_1_ln10 const_sqrt2 +F7_C_PARTS += # const_m1 const_2 const_sqrt2 + +# f7_asm_*.o modules from libf7-asm.sx. +F7_ASM_PARTS += classify clr mul_mant cmp_mant set_u64 +F7_ASM_PARTS += copy copy_P copy_mant clr_mant_lsbs +F7_ASM_PARTS += addsub_mant_scaled store load +F7_ASM_PARTS += to_integer to_unsigned clz normalize_with_carry normalize +F7_ASM_PARTS += store_expo sqrt16 sqrt_approx div + +F7_ASM_PARTS += D_class +F7_ASM_PARTS += D_isnan D_isinf D_isfinite D_signbit D_copysign D_neg D_fabs + +F7_ASM_PARTS += call_dd call_ddd + +# Stuff that will be wrapped in f7-wraps.h (included by libf7-asm.sx) +# and give f7_asm_D_*.o modules. +g_ddd += add sub mul div +g_xdd_cmp += le lt ge gt ne eq unord +g_dx += floatunsidf floatsidf extendsfdf2 +g_xd += fixdfsi fixdfdi fixunsdfdi fixunsdfsi truncdfsf2 + +m_ddd += pow fmin fmax fmod hypot atan2 +m_ddx += ldexp frexp +m_dd += sqrt cbrt exp exp10 pow10 log log10 log2 sin cos tan cotan asin acos atan +m_dd += ceil floor trunc round sinh cosh tanh +m_xd += lrint lround + +# -mcall-prologues +CALL_PROLOGUES += divx sqrt cbrt get_double set_double logx exp exp10 pow10 +CALL_PROLOGUES += put_C truncx round minmax sincos tan cotan pow powi fmod +CALL_PROLOGUES += atan asinacos madd_msub hypot init horner sinhcosh tanh + +# -mstrict-X +STRICT_X += log addsub truncx ldexp exp + +# Renames used when building f7-renames.h. +F7F += fabs neg add sub addsub div div1 divx fmod sqrt cbrt +F7F += square mul mulx madd_msub madd msub hypot +F7F += Ineg Iadd Isub Imul Idiv IRsub Isquare Ildexp Isqrt +F7F += le lt gt ge ne eq cmp cmp_abs ordered unordered cmp_unordered +F7F += lt_impl gt_impl le_impl ge_impl eq_impl ne_impl unord_impl + +F7F += lrint ldexp frexp exp logx log log10 log2 +F7F += minmax fmax fmin floor ceil round lround trunc truncx +F7F += horner pow10 exp10 pow powi +F7F += sin cos tan cotan sincos sinh cosh tanh sinhcosh +F7F += asinacos asin acos atan atan2 +F7F += mul_noround sqrt16_round sqrt16_floor +F7F += clr_mant_lsbs abscmp_msb_ge lshrdi3 ashldi3 +F7F += assert + +F7F += classify + +F7F += class_inf class_nan class_number class_zero class_nonzero class_sign +F7F += signbit set_sign set_nan set_inf +F7F += is_inf is_nan is_number is_zero is_nonzero +F7F += clr copy copy_P copy_mant msbit is0 cmp_mant store_expo +F7F += abs + +F7F += set_s64 set_s32 set_s16 set_s16_impl set_u16_worker +F7F += set_u64 set_u32 set_u16 set_u16_impl +F7F += set_float set_pdouble set_double_impl set_double init_impl init +F7F += get_s16 get_s32 get_s64 get_float +F7F += get_u16 get_u32 get_u64 get_double + +F7F += set_eps set_1pow2 + +# Renames for ALIASes without own module. +F7F += min max exp10 +F7F += floatunsidf floatsidf extendsfdf2 +F7F += fixdfsi fixdfdi fixunsdfdi fixunsdfsi truncdfsf2 + +# Renames for f7-const.def. +F7F_cst += 1 2 1_2 1_3 m1 pi ln2 ln10 1_ln2 1_ln10 sqrt2 + +F7F_asm += classify +F7F_asm += store_expo clr copy copy_P copy_mant +F7F_asm += cmp_mant normalize store_expo +F7F_asm += set_u64 set_s64 addsub_mant_scaled mul_mant +F7F_asm += to_integer to_unsigned clr_mant_lsbs +F7F_asm += div sqrt_approx sqrt16_round sqrt16_floor +F7F_asm += lshrdi3 ashldi3 + +F7F_asm += class_D + +F7F_asm += call_ddd call_xdd call_ddx +F7F_asm += call_dd call_xd call_dx Property changes on: libgcc/config/avr/libf7/libf7-common.mk ___________________________________________________________________ Added: svn:eol-style ## -0,0 +1 ## +native \ No newline at end of property Added: svn:mime-type ## -0,0 +1 ## +text/plain \ No newline at end of property Index: libgcc/config/avr/libf7/libf7-const.def =================================================================== --- libgcc/config/avr/libf7/libf7-const.def (nonexistent) +++ libgcc/config/avr/libf7/libf7-const.def (working copy) @@ -0,0 +1,201 @@ +/* Copyright (C) 2019 Free Software Foundation, Inc. + + This file is part of LIBF7, which is part of GCC. + + GCC is free software; you can redistribute it and/or modify it under + the terms of the GNU General Public License as published by the Free + Software Foundation; either version 3, or (at your option) any later + version. + + GCC is distributed in the hope that it will be useful, but WITHOUT ANY + WARRANTY; without even the implied warranty of MERCHANTABILITY or + FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + for more details. + + Under Section 7 of GPL version 3, you are granted additional + permissions described in the GCC Runtime Library Exception, version + 3.1, as published by the Free Software Foundation. + + You should have received a copy of the GNU General Public License and + a copy of the GCC Runtime Library Exception along with this program; + see the files COPYING3 and COPYING.RUNTIME respectively. If not, see + <http://www.gnu.org/licenses/>. */ + +#ifndef ONLY_CONST_WITH_ID +#if defined IN_LIBF7_H || defined (F7MOD_const_) || defined (F7MOD_const_1_2_) +F7_CONST_DEF (1_2, 0, 0x80,0,0,0,0,0,0, -1) +#endif /* 1/2 */ + +#if defined IN_LIBF7_H || defined (F7MOD_const_) || defined (F7MOD_const_1_) +F7_CONST_DEF (1, 0, 0x80,0,0,0,0,0,0, 0) +#endif /* 1 */ + +#if defined IN_LIBF7_H || defined (F7MOD_const_) || defined (F7MOD_const_2_) +F7_CONST_DEF (2, 0, 0x80,0,0,0,0,0,0, 1) +#endif /* 2 */ + +#if defined IN_LIBF7_H || defined (F7MOD_const_) || defined (F7MOD_const_1_3_) +F7_CONST_DEF (1_3, 0, 0xaa,0xaa,0xaa,0xaa,0xaa,0xaa,0xab, -2) +#endif /* 1/3 */ + +#if defined IN_LIBF7_H || defined (F7MOD_const_) || defined (F7MOD_const_1_ln2_) +F7_CONST_DEF (1_ln2, 0, 0xb8,0xaa,0x3b,0x29,0x5c,0x17,0xf1, 0) +#endif /* 1 / ln2 */ + +#if defined IN_LIBF7_H || defined (F7MOD_const_) || defined (F7MOD_const_1_ln10_) +F7_CONST_DEF (1_ln10, 0, 0xde,0x5b,0xd8,0xa9,0x37,0x28,0x72, -2) +#endif /* 1 / ln10 */ +#endif /* const with ID */ + +#if defined IN_LIBF7_H || defined (F7MOD_const_) || defined (F7MOD_const_m1_) +F7_CONST_DEF (m1, 1, 0x80,0,0,0,0,0,0, 0) +#endif /* -1 */ + +#if defined IN_LIBF7_H || defined (F7MOD_const_) || defined (F7MOD_const_pi_) +F7_CONST_DEF (pi, 0, 0xc9,0x0f,0xda,0xa2,0x21,0x68,0xc2, 1) +#endif /* pi */ + +#if defined IN_LIBF7_H || defined (F7MOD_const_) || defined (F7MOD_const_ln2_) +F7_CONST_DEF (ln2, 0, 0xb1,0x72,0x17,0xf7,0xd1,0xcf,0x7a, -1) +#endif /* ln2 */ + +#if defined IN_LIBF7_H || defined (F7MOD_const_) || defined (F7MOD_const_ln10_) +F7_CONST_DEF (ln10, 0, 0x93,0x5d,0x8d,0xdd,0xaa,0xa8,0xac, 1) +#endif /* ln10 */ + +#if defined IN_LIBF7_H || defined (F7MOD_const_) || defined (F7MOD_const_sqrt2_) +F7_CONST_DEF (sqrt2, 0, 0xb5,0x04,0xf3,0x33,0xf9,0xde,0x65, 0) +#endif /* sqrt2 */ + +#if !defined (IN_LIBF7_H) && !defined (F7MOD_const_) + +#if defined (F7MOD_logx_) +// 2.00000000000000000287938058543222037939 + 0.666666666666667653654896829047862109605x + 0.399999999996639180070480580779767357622x^2 + 0.285714286985476106491309914245597720749x^3 + 0.222222024077476110197597399876978069272x^4 + 0.181833876328594532366358057253631240618x^5 + 0.153181571233880298729095145342556944283x^6 + 0.147580071554569676223389696418304199218x^7 +F7_CONST_DEF (X, 0, 0x80,0x00,0x00,0x00,0x00,0x00,0x00, 1) +F7_CONST_DEF (X, 0, 0xaa,0xaa,0xaa,0xaa,0xaa,0xaa,0xf2, -1) +F7_CONST_DEF (X, 0, 0xcc,0xcc,0xcc,0xcc,0xc5,0x68,0xd4, -2) +F7_CONST_DEF (X, 0, 0x92,0x49,0x24,0x9d,0x34,0x85,0x25, -2) +F7_CONST_DEF (X, 0, 0xe3,0x8e,0x2b,0x97,0x74,0x69,0xb3, -3) +F7_CONST_DEF (X, 0, 0xba,0x32,0xa8,0xe0,0x8b,0x7d,0xcc, -3) +F7_CONST_DEF (X, 0, 0x9c,0xdb,0xa1,0x3b,0x32,0x98,0x9a, -3) +F7_CONST_DEF (X, 0, 0x97,0x1f,0x3a,0xf3,0x79,0x0d,0xc8, -3) + +#elif defined (F7MOD_exp_) +#ifdef ARRAY_NAME +// Intended for |X| < ln2 / 8 +F7_CONST_DEF (X, 0, 0x80,0x00,0x00,0x00,0x00,0x00,0x00, 0) +// f(x) = (e^x - 1) / x and f -> 1 + x*f by means of const_1 above. +// 0.99999999999999993104781692731238623644199 + 0.50000000000000018681311888335564572558345x + 0.16666666666696003026683736262180526042594x^2 + 0.041666666666489887893617892155331720764329x^3 + 0.0083333331383434741356242595738508060753717x^4 + 0.0013888889242082668539796357369837122917246x^5 + 0.00019845416752769938401628433087258224774248x^6 + 0.000024801569801689274979195020335136136418918x^7 +F7_CONST_DEF (X, 0, 0xff,0xff,0xff,0xff,0xff,0xff,0xfb, -1) +F7_CONST_DEF (X, 0, 0x80,0x00,0x00,0x00,0x00,0x00,0x0d, -1) +F7_CONST_DEF (X, 0, 0xaa,0xaa,0xaa,0xaa,0xab,0xf4,0xf7, -3) +F7_CONST_DEF (X, 0, 0xaa,0xaa,0xaa,0xaa,0xa7,0x8e,0x87, -5) +F7_CONST_DEF (X, 0, 0x88,0x88,0x88,0x52,0xef,0x57,0x81, -7) +F7_CONST_DEF (X, 0, 0xb6,0x0b,0x61,0x03,0xb6,0x6b,0x82, -10) +F7_CONST_DEF (X, 0, 0xd0,0x18,0x22,0x8c,0x6f,0xde,0x72, -13) +F7_CONST_DEF (X, 0, 0xd0,0x0c,0xf7,0x31,0x28,0xba,0xb5, -16) +#else +// Negative because f7_const_ln2 is ln2 rounded up. +F7_CONST_DEF (ln2_low, 1, 0xa8,0x6c,0x38,0x98,0xcf,0xf8,0x1a, -58) +#endif + +#elif defined (F7MOD_atan_) + +#if defined (MINIMAX_6_6_IN_0_1) +// https://www.mikrocontroller.net/topic/480840#6003520 +#if defined (FOR_NUMERATOR) +// 1209.7470017580907217240715 + 3031.0745956115083044212807x + 2761.7198246138834959053784x^2 + 1114.1290728455183546172942x^3 + 192.5792014481559613474286x^4 + 11.3221594116764655236245x^5 + 0.09762721591717633036983x^6 +// p = Poly ([Decimal('1209.7470017580907217240715'), Decimal('3031.0745956115083044212807'), Decimal('2761.7198246138834959053784'), Decimal('1114.1290728455183546172942'), Decimal('192.5792014481559613474286'), Decimal('11.3221594116764655236245'), Decimal('0.09762721591717633036983')]) +F7_CONST_DEF (X, 0, 0x97,0x37,0xe7,0x70,0x3b,0x21,0xbc, 10) +F7_CONST_DEF (X, 0, 0xbd,0x71,0x31,0x8b,0x2a,0xfd,0xa7, 11) +F7_CONST_DEF (X, 0, 0xac,0x9b,0x84,0x66,0xd0,0x77,0xc4, 11) +F7_CONST_DEF (X, 0, 0x8b,0x44,0x21,0x5d,0x60,0x49,0xb2, 10) +F7_CONST_DEF (X, 0, 0xc0,0x94,0x46,0x8b,0xcd,0xa0,0x2d, 7) +F7_CONST_DEF (X, 0, 0xb5,0x27,0x90,0xa0,0x93,0xfb,0xfc, 3) +F7_CONST_DEF (X, 0, 0xc7,0xf0,0xc7,0x1c,0x82,0xab,0x23, -4) +#elif defined (FOR_DENOMINATOR) +// 1209.7470017580907287514197 + 3434.3235961975351716547069x + 3664.5449563283749893504796x^2 + 1821.6003392918464941509225x^3 + 423.0716464809047804524206x^4 + 39.9178842486537981501999x^5 + x^6 +// q = Poly ([Decimal('1209.7470017580907287514197'), Decimal('3434.3235961975351716547069'), Decimal('3664.5449563283749893504796'), Decimal('1821.6003392918464941509225'), Decimal('423.0716464809047804524206'), Decimal('39.9178842486537981501999'), Decimal('1.0')]) +F7_CONST_DEF (X, 0, 0x97,0x37,0xe7,0x70,0x3b,0x21,0xbc, 10) +F7_CONST_DEF (X, 0, 0xd6,0xa5,0x2d,0x73,0x34,0xd8,0x60, 11) +F7_CONST_DEF (X, 0, 0xe5,0x08,0xb8,0x24,0x20,0x81,0xe7, 11) +F7_CONST_DEF (X, 0, 0xe3,0xb3,0x35,0xfa,0xbf,0x1f,0x81, 10) +F7_CONST_DEF (X, 0, 0xd3,0x89,0x2b,0xb6,0x3e,0x2e,0x05, 8) +F7_CONST_DEF (X, 0, 0x9f,0xab,0xe9,0xd9,0x35,0xed,0x27, 5) +F7_CONST_DEF (X, 0, 0x80,0x00,0x00,0x00,0x00,0x00,0x00, 0) +#endif + +#elif defined (SWIFT_3_4) +// My [3,4] MiniMax +#if defined (FOR_NUMERATOR) +// 0.999999999999999998080178351225003952632 + 1.51597040589722809277543885223133087789x + 0.636928974763539784860899545005247736093x^2 + 0.0638944455799886571709448345524306512048x^3 +F7_CONST_DEF (X, 0, 0x80,0x00,0x00,0x00,0x00,0x00,0x00, 0) +F7_CONST_DEF (X, 0, 0xc2,0x0b,0x51,0x79,0x84,0x27,0x00, 0) +F7_CONST_DEF (X, 0, 0xa3,0x0d,0xc6,0xfc,0x7b,0xf5,0x5d, -1) +F7_CONST_DEF (X, 0, 0x82,0xdb,0x17,0x51,0x4b,0xfc,0xad, -4) +#elif defined (FOR_DENOMINATOR) +// 1 + 1.84930373923056200945453682584178320362x + 1.05336355450697082895016644607427716580x^2 + 0.188012025422931152047197803304030906006x^3 + 0.00507310235929401206762490897042543192106x^4 +F7_CONST_DEF (X, 0, 0x80,0x00,0x00,0x00,0x00,0x00,0x00, 0) +F7_CONST_DEF (X, 0, 0xec,0xb5,0xfc,0x24,0x2e,0xd1,0xc0, 0) +F7_CONST_DEF (X, 0, 0x86,0xd4,0x9d,0xf0,0xb3,0xef,0xb8, 0) +F7_CONST_DEF (X, 0, 0xc0,0x86,0x39,0x71,0xc8,0xeb,0x3d, -3) +F7_CONST_DEF (X, 0, 0xa6,0x3c,0x44,0x5c,0x78,0x87,0x2d, -8) +#else +F7_CONST_DEF (1_sqrt3, 0, 0x93,0xCD,0x3A,0x2C,0x81,0x98,0xE2, -1) +F7_CONST_DEF (pi_6, 0, 0x86,0x0a,0x91,0xc1,0x6b,0x9b,0x2c, -1) +#endif +#endif // which MiniMax + +#elif defined (F7MOD_asinacos_) +// Relative error < 5.6E-18, quality = 1.00000037 (ideal = 1). +#if defined (FOR_NUMERATOR) +// 0.99999999999999999442491073135027586203 - 1.035234033892197627842731209x + 0.35290206232981519813422591897720574012x^2 - 0.04333483170641685705612351801x^3 + 0.0012557428614630796315205218507940285622x^4 + 0.0000084705471128435769021718764878041684288x^5 +// p = Poly ([Decimal('0.99999999999999999442491073135027586203'), Decimal('-1.0352340338921976278427312087167692142'), Decimal('0.35290206232981519813422591897720574012'), Decimal('-0.043334831706416857056123518013656946650'), Decimal('0.0012557428614630796315205218507940285622'), Decimal('0.0000084705471128435769021718764878041684288')]) +F7_CONST_DEF (X, 0, 0x80,0x00,0x00,0x00,0x00,0x00,0x00, 0) +F7_CONST_DEF (X, 1, 0x84,0x82,0x8c,0x7f,0xa2,0xf6,0x65, 0) +F7_CONST_DEF (X, 0, 0xb4,0xaf,0x94,0x40,0xcb,0x86,0x69, -2) +F7_CONST_DEF (X, 1, 0xb1,0x7f,0xdd,0x4f,0x4e,0xbe,0x1d, -5) +F7_CONST_DEF (X, 0, 0xa4,0x97,0xbd,0x0b,0x59,0xc9,0x25, -10) +F7_CONST_DEF (X, 0, 0x8e,0x1c,0xb9,0x0b,0x50,0x6c,0xce, -17) +#elif defined (FOR_DENOMINATOR) +// 1 - 1.118567367225532923662371649x + 0.42736600959872448854098334016758333519x^2 - 0.06355588484963171659942148390x^3 + 0.0028820878185134035637440105959294542908x^4 +// q = Poly ([Decimal('1'), Decimal('-1.1185673672255329236623716486696411533'), Decimal('0.42736600959872448854098334016758333519'), Decimal('-0.063555884849631716599421483898013782858'), Decimal('0.0028820878185134035637440105959294542908')]) +F7_CONST_DEF (X, 0, 0x80,0x00,0x00,0x00,0x00,0x00,0x00, 0) +F7_CONST_DEF (X, 1, 0x8f,0x2d,0x37,0x2a,0x4d,0xa1,0x57, 0) +F7_CONST_DEF (X, 0, 0xda,0xcf,0xb7,0xb5,0x4c,0x0d,0xee, -2) +F7_CONST_DEF (X, 1, 0x82,0x29,0x96,0x77,0x2e,0x19,0xc7, -4) +F7_CONST_DEF (X, 0, 0xbc,0xe1,0x68,0xec,0xba,0x20,0x29, -9) +#endif + +#elif defined (F7MOD_sincos_) +#if defined (FOR_SIN) +// sin(sqrt(x)) / sqrt(x) in [0, pi^2/16] ~ [0, 0.6169]. +// Error < 3.313E-18, quality@quot ~ 1. +// p = Poly ([Decimal('0.9999999999999999966879566228039012'), Decimal('-0.1666666666666661475365354492822562'), Decimal('0.008333333333320002444252560522892007'), Decimal('-0.0001984126982840212992698250499618594'), Decimal('0.000002755731329901509333692668814721451'), Decimal('-2.505070584638448064973439248538648E-8'), Decimal('1.589413637225924008237178592214358E-10')]) +// 0.9999999999999999966879566228039012 - 0.1666666666666661475365354492822562x + 0.008333333333320002444252560522892007x^2 - 0.0001984126982840212992698250499618594x^3 + 0.000002755731329901509333692668814721451x^4 - 2.505070584638448064973439248538648E-8x^5 + 1.589413637225924008237178592214358E-10x^6 +F7_CONST_DEF (X, 0, 0x80,0x00,0x00,0x00,0x00,0x00,0x00, 0) +F7_CONST_DEF (X, 1, 0xaa,0xaa,0xaa,0xaa,0xaa,0xaa,0x15, -3) +F7_CONST_DEF (X, 0, 0x88,0x88,0x88,0x88,0x87,0x98,0x63, -7) +F7_CONST_DEF (X, 1, 0xd0,0x0d,0x00,0xcd,0xc9,0x7e,0x34, -13) +F7_CONST_DEF (X, 0, 0xb8,0xef,0x1a,0x8f,0x9e,0x91,0xa2, -19) +F7_CONST_DEF (X, 1, 0xd7,0x2f,0x15,0xb0,0x7c,0x11,0x7a, -26) +F7_CONST_DEF (X, 0, 0xae,0xc2,0x04,0x43,0x51,0x29,0xe8, -33) +#elif defined (FOR_COS) +// cos(sqrt(x)) in [0, pi^2/16] ~ [0, 0.6169]. +// Error < 3.6E-20, quality@quot ~ 1. +// p = Poly ([Decimal('0.999999999999999999964159204335255989833'), Decimal('-0.499999999999999992843598630174036090975'), Decimal('0.0416666666666664302574165741342831192783'), Decimal('-0.00138888888888589604372397206213200172560'), Decimal('0.0000248015872828994642415389301648196400167'), Decimal('-2.75573128656963697259081300852869272237E-7'), Decimal('2.08755551457127352241183195408105664857E-9'), Decimal('-1.13521232075810109334454849167561526185E-11')]) +// 0.999999999999999999964159204335255989833 - 0.499999999999999992843598630174036090975x + 0.0416666666666664302574165741342831192783x^2 - 0.00138888888888589604372397206213200172560x^3 + 0.0000248015872828994642415389301648196400167x^4 - 2.75573128656963697259081300852869272237E-7x^5 + 2.08755551457127352241183195408105664857E-9x^6 - 1.13521232075810109334454849167561526185E-11x^7 +F7_CONST_DEF (X, 0, 0x80,0x00,0x00,0x00,0x00,0x00,0x00, 0) +F7_CONST_DEF (X, 1, 0xff,0xff,0xff,0xff,0xff,0xff,0xff, -2) +F7_CONST_DEF (X, 0, 0xaa,0xaa,0xaa,0xaa,0xaa,0xa9,0x9a, -5) +F7_CONST_DEF (X, 1, 0xb6,0x0b,0x60,0xb6,0x09,0xb1,0x66, -10) +F7_CONST_DEF (X, 0, 0xd0,0x0d,0x00,0xcd,0x6b,0xb3,0xf0, -16) +F7_CONST_DEF (X, 1, 0x93,0xf2,0x7b,0x7f,0x10,0xce,0x3d, -22) +F7_CONST_DEF (X, 0, 0x8f,0x74,0xaa,0x3c,0xcf,0x51,0x3d, -29) +F7_CONST_DEF (X, 1, 0xc7,0xb5,0x6a,0xf8,0x0e,0x32,0x07, -37) +#else +F7_CONST_DEF (pi_low,0, 0xd3,0x13,0x19,0x8a,0x2e,0x03,0x70, 1 - F7_MANT_BITS-2) +#endif + +#endif +#endif /* ! IN_LIBF7_H && ! F7MOD_const_ */ Property changes on: libgcc/config/avr/libf7/libf7-const.def ___________________________________________________________________ Added: svn:eol-style ## -0,0 +1 ## +native \ No newline at end of property Added: svn:mime-type ## -0,0 +1 ## +text/plain \ No newline at end of property Index: libgcc/config/avr/libf7/libf7-constdef.h =================================================================== --- libgcc/config/avr/libf7/libf7-constdef.h (nonexistent) +++ libgcc/config/avr/libf7/libf7-constdef.h (working copy) @@ -0,0 +1,43 @@ +/* Copyright (C) 2019 Free Software Foundation, Inc. + + This file is part of LIBF7, which is part of GCC. + + GCC is free software; you can redistribute it and/or modify it under + the terms of the GNU General Public License as published by the Free + Software Foundation; either version 3, or (at your option) any later + version. + + GCC is distributed in the hope that it will be useful, but WITHOUT ANY + WARRANTY; without even the implied warranty of MERCHANTABILITY or + FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + for more details. + + Under Section 7 of GPL version 3, you are granted additional + permissions described in the GCC Runtime Library Exception, version + 3.1, as published by the Free Software Foundation. + + You should have received a copy of the GNU General Public License and + a copy of the GCC Runtime Library Exception along with this program; + see the files COPYING3 and COPYING.RUNTIME respectively. If not, see + <http://www.gnu.org/licenses/>. */ + +#if !defined STATIC +#define STATIC /* empty */ +#endif + +#if USE_LPM +#define F7_CONST_DEF(NAME, FLAGS, M6, M5, M4, M3, M2, M1, M0, EXPO) \ + STATIC const __attribute__((__progmem__)) \ + f7_t F7_(const_ ## NAME ## _P) = \ + { .flags = FLAGS, .mant = { M0, M1, M2, M3, M4, M5, M6 }, .expo = EXPO }; + #include "libf7-const.def" +#undef F7_CONST_DEF +#else +#define F7_CONST_DEF(NAME, FLAGS, M6, M5, M4, M3, M2, M1, M0, EXPO) \ + STATIC const f7_t F7_(const_ ## NAME) = \ + { .flags = FLAGS, .mant = { M0, M1, M2, M3, M4, M5, M6 }, .expo = EXPO }; + #include "libf7-const.def" +#undef F7_CONST_DEF +#endif // USE_LPM + +#undef STATIC Property changes on: libgcc/config/avr/libf7/libf7-constdef.h ___________________________________________________________________ Added: svn:eol-style ## -0,0 +1 ## +native \ No newline at end of property Added: svn:mime-type ## -0,0 +1 ## +text/plain \ No newline at end of property Index: libgcc/config/avr/libf7/libf7.c =================================================================== --- libgcc/config/avr/libf7/libf7.c (nonexistent) +++ libgcc/config/avr/libf7/libf7.c (working copy) @@ -0,0 +1,2501 @@ +/* Copyright (C) 2019 Free Software Foundation, Inc. + + This file is part of LIBF7, which is part of GCC. + + GCC is free software; you can redistribute it and/or modify it under + the terms of the GNU General Public License as published by the Free + Software Foundation; either version 3, or (at your option) any later + version. + + GCC is distributed in the hope that it will be useful, but WITHOUT ANY + WARRANTY; without even the implied warranty of MERCHANTABILITY or + FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + for more details. + + Under Section 7 of GPL version 3, you are granted additional + permissions described in the GCC Runtime Library Exception, version + 3.1, as published by the Free Software Foundation. + + You should have received a copy of the GNU General Public License and + a copy of the GCC Runtime Library Exception along with this program; + see the files COPYING3 and COPYING.RUNTIME respectively. If not, see + <http://www.gnu.org/licenses/>. */ + +#include "libf7.h" + +#ifndef __AVR_TINY__ + +#define ALIAS(X, Y) \ + F7_WEAK __attribute__((__alias__(F7_STRINGY(X)))) __typeof__(X) Y; + +#define DALIAS(...) // empty +#define LALIAS(...) // empty + +#ifndef IN_LIBGCC2 + +#include <stdio.h> +#include <assert.h> + +#define in_libgcc false + +_Static_assert (sizeof (f7_t) == 10 && F7_MANT_BYTES == 7, + "libf7 will only work with 7-byte mantissa."); +#else + +#define in_libgcc true + +#if __SIZEOF_DOUBLE__ == 8 +#undef DALIAS +#define DALIAS(X,Y) \ + F7_WEAK __attribute__((__alias__(F7_STRINGY(X)))) __typeof__(X) Y; +#endif + +#if __SIZEOF_LONG_DOUBLE__ == 8 +#undef LALIAS +#define LALIAS(X,Y) \ + F7_WEAK __attribute__((__alias__(F7_STRINGY(X)))) __typeof__(X) Y; +#endif + +#endif // in libgcc + +static F7_INLINE +void f7_assert (bool x) +{ + if (!in_libgcc && !x) + __builtin_abort(); +} + +static F7_INLINE +int16_t abs_ssat16 (int16_t a) +{ + _Sat _Fract sa = __builtin_avr_rbits (a); + return __builtin_avr_bitsr (__builtin_avr_absr (sa)); +} + +static F7_INLINE +int16_t add_ssat16 (int16_t a, int16_t b) +{ + _Sat _Fract sa = __builtin_avr_rbits (a); + _Sat _Fract sb = __builtin_avr_rbits (b); + return __builtin_avr_bitsr (sa + sb); +} + +static F7_INLINE +int16_t sub_ssat16 (int16_t a, int16_t b) +{ + _Sat _Fract sa = __builtin_avr_rbits (a); + _Sat _Fract sb = __builtin_avr_rbits (b); + return __builtin_avr_bitsr (sa - sb); +} + +static F7_INLINE +int8_t ssat8_range (int16_t a, int8_t range) +{ + if (a >= range) + return range; + if (a <= -range) + return -range; + return a; +} + + +#define IN_LIBF7_H + #define F7_CONST_DEF(NAME, FLAGS, M6, M5, M4, M3, M2, M1, M0, EXPO) \ + F7_UNUSED static const uint8_t F7_(const_##NAME##_msb) = M6; \ + F7_UNUSED static const int16_t F7_(const_##NAME##_expo) = EXPO; + #include "libf7-const.def" + #undef F7_CONST_DEF +#undef IN_LIBF7_H + + +/* + libgcc naming converntions for conversions: + + __float<fmode><fmode> : Convert float modes. + __floatun<imode><fmode>: Convert unsigned integral to float. + __fix<fmode><imode> : Convert float to signed integral. + __fixuns<fmode><imode> : Convert float to unsigned integral. +*/ + + +#ifdef F7MOD_floatundidf_ +F7_WEAK +f7_double_t __floatundidf (uint64_t x) +{ + f7_t xx; + f7_set_u64 (&xx, x); + return f7_get_double (&xx); +} +#endif // F7MOD_floatundidf_ + + +#ifdef F7MOD_floatdidf_ +F7_WEAK +f7_double_t __floatdidf (int64_t x) +{ + f7_t xx; + f7_set_s64 (&xx, x); + return f7_get_double (&xx); +} +#endif // F7MOD_floatdidf_ + + +#ifdef F7MOD_init_ +f7_t* f7_init_impl (uint64_t mant, uint8_t flags, f7_t *cc, int16_t expo) +{ + flags &= F7_FLAGS; + if (f7_class_number (flags)) + { + uint8_t msb; + while ((__builtin_memcpy (&msb, (uint8_t*) &mant + 7, 1), msb)) + { + mant >>= 1; + expo = add_ssat16 (expo, 1); + } + *(uint64_t*) cc->mant = mant; + cc->expo = add_ssat16 (expo, F7_MANT_BITS-1); + + cc = f7_normalize_asm (cc); + } + + cc->flags = flags; + + return cc; +} +#endif // F7MOD_init_ + + +#ifdef F7MOD_set_s16_ +f7_t* f7_set_s16_impl (f7_t *cc, int16_t i16) +{ + uint16_t u16 = (uint16_t) i16; + uint8_t flags = 0; + if (i16 < 0) + { + u16 = -u16; + flags = F7_FLAG_sign; + } + f7_set_u16_impl (cc, u16); + cc->flags = flags; + return cc; +} +#endif // F7MOD_set_s16_ + + +#ifdef F7MOD_set_u16_ +f7_t* f7_set_u16_impl (f7_t *cc, uint16_t u16) +{ + f7_clr (cc); + F7_MANT_HI2 (cc) = u16; + cc->expo = 15; + return f7_normalize_asm (cc); +} +#endif // F7MOD_set_u16_ + + +#ifdef F7MOD_set_s32_ +f7_t* f7_set_s32 (f7_t *cc, int32_t i32) +{ + uint32_t u32 = (uint32_t) i32; + uint8_t flags = 0; + if (i32 < 0) + { + u32 = -u32; + flags = F7_FLAG_sign; + } + cc = f7_set_u32 (cc, u32); + cc->flags = flags; + return cc; +} +ALIAS (f7_set_s32, f7_floatsidf) +#endif // F7MOD_set_s32_ + + +#ifdef F7MOD_set_u32_ +f7_t* f7_set_u32 (f7_t *cc, uint32_t u32) +{ + f7_clr (cc); + F7_MANT_HI4 (cc) = u32; + cc->expo = 31; + return f7_normalize_asm (cc); +} +ALIAS (f7_set_u32, f7_floatunsidf) +#endif // F7MOD_set_u32_ + + +// IEEE 754 single +// float = s bbbbbbbb mmmmmmmmmmmmmmmmmmmmmmm +// 31 +// s = sign +// b = biased exponent, bias = 127 +// m = mantissa + +// +0 = 0 0 0 +// -0 = 1 0 0 +// Inf = S B 0 = S * Inf, B = 0xff +// NaN = S B M, B = 0xff, M != 0 +// Sub-normal = S 0 M = S * 0.M * 2^{1 - bias}, M != 0 +// Normal = S B M = S * 1.M * 2^{B - bias}, B = 1 ... 0xfe + +#define FLT_DIG_EXP 8 +#define FLT_DIG_MANT (31 - FLT_DIG_EXP) +#define FLT_MAX_EXP ((1 << FLT_DIG_EXP) - 1) +#define FLT_EXP_BIAS (FLT_MAX_EXP >> 1) + +#ifdef F7MOD_set_float_ +F7_WEAK +void f7_set_float (f7_t *cc, float f) +{ + uint32_t val32; + + _Static_assert (__SIZEOF_FLOAT__ == 4, ""); + _Static_assert (__FLT_MANT_DIG__ == 24, ""); + __builtin_memcpy (&val32, &f, __SIZEOF_FLOAT__); + + uint16_t val16 = val32 >> 16; + val16 >>= FLT_DIG_MANT - 16; + + uint8_t expo_biased = val16 & FLT_MAX_EXP; + bool sign = val16 & (1u << FLT_DIG_EXP); + + f7_clr (cc); + + uint32_t mant = val32 & ((1ul << FLT_DIG_MANT) -1); + + if (mant == 0) + { + if (expo_biased == 0) + return; + if (expo_biased >= FLT_MAX_EXP) + return f7_set_inf (cc, sign); + } + + if (expo_biased == 0) + expo_biased = 1; // Sub-normal: biased expo of 1 was encoded as 0. + else if (expo_biased < FLT_MAX_EXP) + mant |= (1ul << FLT_DIG_MANT); + else + return f7_set_nan (cc); + + __builtin_memcpy (& F7_MANT_HI4 (cc), &mant, 4); + + cc->expo = expo_biased - FLT_EXP_BIAS + 31 - FLT_DIG_MANT; + f7_normalize_asm (cc); + f7_set_sign (cc, sign); +} +ALIAS (f7_set_float, f7_extendsfdf2) +#endif // F7MOD_set_float_ + + +#ifdef F7MOD_get_float_ +static F7_INLINE +float make_float (uint32_t x) +{ + float ff; + __builtin_memcpy (&ff, &x, 4); + return ff; + +} + +F7_WEAK +float f7_get_float (const f7_t *aa) +{ + uint8_t a_class = f7_classify (aa); + + if (f7_class_nan (a_class)) + return make_float (0xffc00000 /* NaN: Biased expo = 0xff, mant != 0 */); + + uint32_t mant; + __builtin_memcpy (&mant, &F7_MANT_CONST_HI4 (aa), 4); + + uint8_t expo8 = 0; + uint8_t mant_offset = FLT_DIG_EXP; + int16_t c_expo = add_ssat16 (aa->expo, FLT_EXP_BIAS); + + if (f7_class_zero (a_class) || c_expo <= -FLT_DIG_MANT) + { + // Zero or tiny. + return 0.0f; + } + else if (c_expo >= FLT_MAX_EXP || f7_class_inf (a_class)) + { + // Inf or overflow. + expo8 = FLT_MAX_EXP; + mant = 0; + } + else if (c_expo > 0) + { + // Normal. + expo8 = c_expo; + } + else + { + // Sub-normal: -DIG_MANT < c_expo <= 0. + // Encoding of 0 represents a biased exponent of 1. + // mant_offset in 9...31. + expo8 = 0; + mant_offset += 1 - c_expo; + } + + uint16_t expo16 = expo8 << (FLT_DIG_MANT - 16); + + if (f7_class_sign (a_class)) + expo16 |= 1u << (FLT_DIG_EXP + FLT_DIG_MANT - 16); + + mant >>= mant_offset; + + __asm ("cbr %T0%t2, 1 << (7 & %2)" "\n\t" + "or %C0, %A1" "\n\t" + "or %D0, %B1" + : "+d" (mant) + : "r" (expo16), "n" (FLT_DIG_MANT)); + + return make_float (mant); +} +F7_PURE ALIAS (f7_get_float, f7_truncdfsf2) +#endif // F7MOD_get_float_ + +#define DBL_DIG_EXP 11 +#define DBL_DIG_MANT (63 - DBL_DIG_EXP) +#define DBL_MAX_EXP ((1 << DBL_DIG_EXP) - 1) +#define DBL_EXP_BIAS (DBL_MAX_EXP >> 1) + + +#ifdef F7MOD_set_double_ +void f7_set_double_impl (f7_double_t val64, f7_t *cc) +{ + f7_clr (cc); + register uint64_t mant __asm ("r18") = val64 & ((1ull << DBL_DIG_MANT) -1); + + uint16_t val16 = 3[(uint16_t*) & val64]; + val16 >>= DBL_DIG_MANT - 48; + + uint16_t expo_biased = val16 & DBL_MAX_EXP; + bool sign = val16 & (1u << DBL_DIG_EXP); + + if (mant == 0) + { + if (expo_biased == 0) + return; + if (expo_biased >= DBL_MAX_EXP) + return f7_set_inf (cc, sign); + } + __asm ("" : "+r" (mant)); + + if (expo_biased == 0) + expo_biased = 1; // Sub-normal: biased expo of 1 was encoded as 0. + else if (expo_biased < DBL_MAX_EXP) + mant |= (1ull << DBL_DIG_MANT); + else + return f7_set_nan (cc); + + *(uint64_t*) & cc->mant = mant; + + cc->expo = expo_biased - DBL_EXP_BIAS + 63 - DBL_DIG_MANT - 8; + f7_normalize_asm (cc); + f7_set_sign (cc, sign); +} +#endif // F7MOD_set_double_ + + +#ifdef F7MOD_set_pdouble_ +void f7_set_pdouble (f7_t *cc, const f7_double_t *val64) +{ + f7_set_double (cc, *val64); +} +#endif // F7MOD_set_pdouble_ + + +#ifdef F7MOD_get_double_ +static F7_INLINE +uint64_t clr_r18 (void) +{ + extern void __clr_8 (void); + register uint64_t r18 __asm ("r18"); + __asm ("%~call %x[f]" : "=r" (r18) : [f] "i" (__clr_8)); + return r18; +} + +static F7_INLINE +f7_double_t make_double (uint64_t x) +{ + register f7_double_t r18 __asm ("r18") = x; + __asm ("" : "+r" (r18)); + return r18; +} + +F7_WEAK +f7_double_t f7_get_double (const f7_t *aa) +{ + uint8_t a_class = f7_classify (aa); + + if (f7_class_nan (a_class)) + { + uint64_t nan = clr_r18() | (0x7fffull << 48); + return make_double (nan); + } + + uint64_t mant; + __builtin_memcpy (&mant, & aa->mant, 8); + + mant &= 0x00ffffffffffffff; + + // FIXME: For subnormals, rounding is premature and should be + // done *after* the mantissa has been shifted into place + // (or the round value be shifted left accordingly). + // Round. + mant += 1u << (F7_MANT_BITS - (1 + DBL_DIG_MANT) - 1); + + uint8_t dex; + register uint64_t r18 __asm ("r18") = mant; + // dex = Overflow ? 1 : 0. + __asm ("bst %T[mant]%T[bitno]" "\n\t" + "clr %0" "\n\t" + "bld %0,0" + : "=r" (dex), [mant] "+r" (r18) + : [bitno] "n" (64 - 8)); + + mant = r18 >> dex; + + uint16_t expo16 = 0; + uint16_t mant_offset = DBL_DIG_EXP - 8; + int16_t c_expo = add_ssat16 (aa->expo, DBL_EXP_BIAS + dex); + + if (f7_class_zero (a_class) || c_expo <= -DBL_DIG_MANT) + { + // Zero or tiny. + return make_double (clr_r18()); + } + else if (c_expo >= DBL_MAX_EXP || f7_class_inf (a_class)) + { + // Inf or overflow. + expo16 = DBL_MAX_EXP; + mant = clr_r18(); + } + else if (c_expo > 0) + { + // Normal. + expo16 = c_expo; + } + else + { + // Sub-normal: -DIG_MANT < c_expo <= 0. + // Encoding expo of 0 represents a biased exponent of 1. + // mant_offset in 5...55 = 63-8. + mant_offset += 1 - c_expo; + } + + expo16 <<= (DBL_DIG_MANT - 48); + + if (f7_class_sign (a_class)) + expo16 |= 1u << (DBL_DIG_EXP + DBL_DIG_MANT - 48); + + // mant >>= mant_offset; + mant = f7_lshrdi3 (mant, mant_offset); + + r18 = mant; + __asm ("cbr %T0%t2, 1 << (7 & %2)" "\n\t" + "or %r0+6, %A1" "\n\t" + "or %r0+7, %B1" + : "+r" (r18) + : "r" (expo16), "n" (DBL_DIG_MANT)); + + return make_double (r18); +} +#endif // F7MOD_get_double_ + + +#ifdef F7MOD_fabs_ +F7_WEAK +void f7_fabs (f7_t *cc, const f7_t *aa) +{ + f7_abs (cc, aa); +} +#endif // F7MOD_fabs_ + + +#ifdef F7MOD_neg_ +F7_WEAK +f7_t* f7_neg (f7_t *cc, const f7_t *aa) +{ + f7_copy (cc, aa); + + uint8_t c_class = f7_classify (cc); + + if (! f7_class_zero (c_class)) + cc->sign = ! f7_class_sign (c_class); + + return cc; +} +#endif // F7MOD_neg_ + + +#ifdef F7MOD_frexp_ +F7_WEAK +void f7_frexp (f7_t *cc, const f7_t *aa, int *expo) +{ + uint8_t a_class = f7_classify (aa); + + if (f7_class_nan (a_class)) + return f7_set_nan (cc); + + if (f7_class_inf (a_class) || aa->expo == INT16_MAX) + return f7_set_inf (cc, f7_class_sign (a_class)); + + if (! f7_msbit (aa)) + { + *expo = 0; + return f7_clr (cc); + } + + *expo = 1 + aa->expo; + cc->flags = a_class & F7_FLAG_sign; + cc->expo = -1; + f7_copy_mant (cc, aa); +} +#endif // F7MOD_frexp_ + +#ifdef F7MOD_get_s16_ +F7_WEAK +int16_t f7_get_s16 (const f7_t *aa) +{ + extern int16_t to_s16 (const f7_t*, uint8_t) F7ASM(f7_to_integer_asm); + return to_s16 (aa, 0xf); +} +#endif // F7MOD_get_s16_ + + +#ifdef F7MOD_get_s32_ +F7_WEAK +int32_t f7_get_s32 (const f7_t *aa) +{ + extern int32_t to_s32 (const f7_t*, uint8_t) F7ASM(f7_to_integer_asm); + return to_s32 (aa, 0x1f); +} +F7_PURE ALIAS (f7_get_s32, f7_fixdfsi) +#endif // F7MOD_get_s32_ + + +#ifdef F7MOD_get_s64_ + F7_WEAK + int64_t f7_get_s64 (const f7_t *aa) +{ + extern int64_t to_s64 (const f7_t*, uint8_t) F7ASM(f7_to_integer_asm); + return to_s64 (aa, 0x3f); +} +F7_PURE ALIAS (f7_get_s64, f7_fixdfdi) +#endif // F7MOD_get_s64_ + +#ifdef F7MOD_get_u16_ + F7_WEAK + uint16_t f7_get_u16 (const f7_t *aa) +{ + extern uint16_t to_u16 (const f7_t*, uint8_t) F7ASM(f7_to_unsigned_asm); + return to_u16 (aa, 0xf); +} +#endif // F7MOD_get_u16_ + + +#ifdef F7MOD_get_u32_ +F7_WEAK +uint32_t f7_get_u32 (const f7_t *aa) +{ + extern uint32_t to_u32 (const f7_t*, uint8_t) F7ASM(f7_to_unsigned_asm); + return to_u32 (aa, 0x1f); +} +F7_PURE ALIAS (f7_get_u32, f7_fixunsdfsi) +#endif // F7MOD_get_u32_ + + +#ifdef F7MOD_get_u64_ +F7_WEAK +uint64_t f7_get_u64 (const f7_t *aa) +{ + extern int64_t to_u64 (const f7_t*, uint8_t) F7ASM(f7_to_unsigned_asm); + return to_u64 (aa, 0x3f); +} +F7_PURE ALIAS (f7_get_u64, f7_fixunsdfdi) +#endif // F7MOD_get_u64_ + + +#ifdef F7MOD_cmp_unordered_ +F7_NOINLINE +static int8_t cmp_u8 (uint8_t a_class, uint8_t b_class, bool sign_a); + +F7_WEAK +int8_t f7_cmp_unordered (const f7_t *aa, const f7_t *bb, bool with_sign) +{ + uint8_t a_class = f7_classify (aa); + uint8_t b_class = f7_classify (bb); + + uint8_t a_sign = f7_class_sign (a_class) & with_sign; + uint8_t b_sign = f7_class_sign (b_class) & with_sign; + uint8_t ab_class = a_class | b_class; + ab_class &= with_sign - 2; + + if (f7_class_nan (ab_class)) + return INT8_MIN; + + if (a_sign != b_sign) + return b_sign - a_sign; + + if (f7_class_inf (ab_class)) + return cmp_u8 (a_class, b_class, a_sign); + + if (f7_class_zero (ab_class)) + return cmp_u8 (b_class, a_class, a_sign); + + if (aa->expo < bb->expo) + return a_sign ? 1 : -1; + + if (aa->expo > bb->expo) + return a_sign ? -1 : 1; + + return cmp_u8 (1 + f7_cmp_mant (aa, bb), 1, a_sign); +} + + +int8_t cmp_u8 (uint8_t a_class, uint8_t b_class, bool sign_a) +{ + int8_t c; + __asm ("sub %[a], %[b]" "\n\t" + "breq 1f" "\n\t" + "sbc %[c], %[c]" "\n\t" + "sbci %[c], -1" "\n\t" + "sbrc %[s], 0" "\n\t" + "neg %[c]" "\n\t" + "1:" + : [c] "=d" (c) + : [a] "0" (a_class), [b] "r" (b_class), [s] "r" (sign_a)); + return c; +} +#endif // F7MOD_cmp_unordered_ + + +#ifdef F7MOD_cmp_abs_ +F7_WEAK +int8_t f7_cmp_abs (const f7_t *aa, const f7_t *bb) +{ + return f7_cmp_unordered (aa, bb, false /* no signs */); +} +#endif // F7MOD_cmp_abs_ + + +#ifdef F7MOD_cmp_ +F7_WEAK +int8_t f7_cmp (const f7_t *aa, const f7_t *bb) +{ + return f7_cmp_unordered (aa, bb, true /* with signs */); +} +#endif // F7MOD_cmp_ + + +#ifdef F7MOD_abscmp_msb_ge_ +// Compare absolute value of Number aa against a f7_t represented +// by msb and expo. +F7_WEAK +bool f7_abscmp_msb_ge (const f7_t *aa, uint8_t msb, int16_t expo) +{ + uint8_t a_msb = aa->mant[F7_MANT_BYTES - 1]; + + if (0 == (0x80 & a_msb)) + // 0 or subnormal. + return false; + + return aa->expo == expo + ? a_msb >= msb + : aa->expo > expo; +} +#endif // F7MOD_abscmp_msb_ge_ + +#ifdef F7MOD_lt_ +F7_WEAK +bool f7_lt_impl (const f7_t *aa, const f7_t *bb) +{ + return f7_lt (aa, bb); +} +#endif // F7MOD_lt_ + +#ifdef F7MOD_le_ +F7_WEAK +bool f7_le_impl (const f7_t *aa, const f7_t *bb) +{ + return f7_le (aa, bb); +} +#endif // F7MOD_le_ + +#ifdef F7MOD_gt_ +F7_WEAK +bool f7_gt_impl (const f7_t *aa, const f7_t *bb) +{ + return f7_gt (aa, bb); +} +#endif // F7MOD_gt_ + +#ifdef F7MOD_ge_ +F7_WEAK +bool f7_ge_impl (const f7_t *aa, const f7_t *bb) +{ + return f7_ge (aa, bb); +} +#endif // F7MOD_ge_ + +#ifdef F7MOD_ne_ +F7_WEAK +bool f7_ne_impl (const f7_t *aa, const f7_t *bb) +{ + return f7_ne (aa, bb); +} +#endif // F7MOD_ne_ + +#ifdef F7MOD_eq_ +F7_WEAK +bool f7_eq_impl (const f7_t *aa, const f7_t *bb) +{ + return f7_eq (aa, bb); +} +#endif // F7MOD_eq_ + + +#ifdef F7MOD_unord_ +F7_WEAK +bool f7_unord_impl (const f7_t *aa, const f7_t *bb) +{ + return f7_unordered (aa, bb); +} +#endif // F7MOD_unord_ + + +#ifdef F7MOD_minmax_ +F7_WEAK +f7_t* f7_minmax (f7_t *cc, const f7_t *aa, const f7_t *bb, bool do_min) +{ + int8_t cmp = f7_cmp_unordered (aa, bb, true /* with signs */); + + if (cmp == INT8_MIN) + return (f7_set_nan (cc), cc); + + if (do_min) + cmp = -cmp; + + return f7_copy (cc, cmp >= 0 ? aa : bb); +} +#endif // F7MOD_minmax_ + + +#ifdef F7MOD_fmax_ +F7_WEAK +f7_t* f7_fmax (f7_t *cc, const f7_t *aa, const f7_t *bb) +{ + return f7_minmax (cc, aa, bb, false); +} +ALIAS (f7_fmax, f7_max) +#endif // F7MOD_fmax_ + + +#ifdef F7MOD_fmin_ +F7_WEAK +f7_t* f7_fmin (f7_t *cc, const f7_t *aa, const f7_t *bb) +{ + return f7_minmax (cc, aa, bb, true); +} +ALIAS (f7_fmin, f7_min) +#endif // F7MOD_fmin_ + + +#ifdef F7MOD_mulx_ +F7_WEAK +uint8_t f7_mulx (f7_t *cc, const f7_t *aa, const f7_t *bb, bool no_rounding) +{ + uint8_t a_class = f7_classify (aa); + uint8_t b_class = f7_classify (bb); + // From this point on, no more access aa->flags or bb->flags + // to avoid early-clobber when writing cc->flags. + + uint8_t ab_class = a_class | b_class; + // If either value is NaN, return NaN. + if (f7_class_nan (ab_class) + // Any combination of Inf and 0. + || (f7_class_zero (ab_class) && f7_class_inf (ab_class))) + { + cc->flags = F7_FLAG_nan; + return 0; + } + // If either value is 0.0, return 0.0. + if (f7_class_zero (ab_class)) + { + f7_clr (cc); + return 0; + } + // We have 2 non-zero numbers-or-INF. + + uint8_t c_sign = (a_class ^ b_class) & F7_FLAG_sign; + uint8_t c_inf = ab_class & F7_FLAG_inf; + cc->flags = c_sign | c_inf; + if (c_inf) + return 0; + + int16_t expo = add_ssat16 (aa->expo, bb->expo); + // Store expo and handle expo = INT16_MIN and INT16_MAX. + if (f7_store_expo (cc, expo)) + return 0; + + return f7_mul_mant_asm (cc, aa, bb, no_rounding); +} +#endif // F7MOD_mulx_ + + +#ifdef F7MOD_square_ +F7_WEAK +void f7_square (f7_t *cc, const f7_t *aa) +{ + f7_mul (cc, aa, aa); +} +#endif // F7MOD_square_ + + +#ifdef F7MOD_mul_ +F7_WEAK +void f7_mul (f7_t *cc, const f7_t *aa, const f7_t *bb) +{ + f7_mulx (cc, aa, bb, false); +} +#endif // F7MOD_mul_ + + +#ifdef F7MOD_Iadd_ +F7_WEAK void f7_Iadd (f7_t *cc, const f7_t *aa) { f7_add (cc, cc, aa); } +#endif + +#ifdef F7MOD_Isub_ +F7_WEAK void f7_Isub (f7_t *cc, const f7_t *aa) { f7_sub (cc, cc, aa); } +#endif + +#ifdef F7MOD_Imul_ +F7_WEAK void f7_Imul (f7_t *cc, const f7_t *aa) { f7_mul (cc, cc, aa); } +#endif + +#ifdef F7MOD_Idiv_ +F7_WEAK void f7_Idiv (f7_t *cc, const f7_t *aa) { f7_div (cc, cc, aa); } +#endif + +#ifdef F7MOD_IRsub_ +F7_WEAK void f7_IRsub (f7_t *cc, const f7_t *aa) { f7_sub (cc, aa, cc); } +#endif + +#ifdef F7MOD_Ineg_ +F7_WEAK void f7_Ineg (f7_t *cc) { f7_neg (cc, cc); } +#endif + +#ifdef F7MOD_Isqrt_ +F7_WEAK void f7_Isqrt (f7_t *cc) { f7_sqrt (cc, cc); } +#endif + +#ifdef F7MOD_Isquare_ +F7_WEAK void f7_Isquare (f7_t *cc) { f7_square (cc, cc); } +#endif + +#ifdef F7MOD_Ildexp_ +F7_WEAK f7_t* f7_Ildexp (f7_t *cc, int ex) { return f7_ldexp (cc, cc, ex); } +#endif + + +#ifdef F7MOD_add_ +F7_WEAK +void f7_add (f7_t *cc, const f7_t *aa, const f7_t *bb) +{ + f7_addsub (cc, aa, bb, false); +} +#endif // F7MOD_add_ + + +#ifdef F7MOD_sub_ +F7_WEAK +void f7_sub (f7_t *cc, const f7_t *aa, const f7_t *bb) +{ + f7_addsub (cc, aa, bb, true); +} +#endif // F7MOD_sub_ + + +#ifdef F7MOD_addsub_ +static void return_with_sign (f7_t *cc, const f7_t *aa, int8_t c_sign) +{ + __asm (";;; return with sign"); + f7_copy (cc, aa); + if (c_sign != -1) + f7_set_sign (cc, c_sign); +} + +F7_WEAK +void f7_addsub (f7_t *cc, const f7_t *aa, const f7_t *bb, bool neg_b) +{ + uint8_t a_class = f7_classify (aa); + uint8_t b_class = f7_classify (bb); + // From this point on, no more access aa->flags or bb->flags + // to avoid early-clobber when writing cc->flags. + + // Hande NaNs. + if (f7_class_nan (a_class | b_class)) + return f7_set_nan (cc); + + bool a_sign = f7_class_sign (a_class); + bool b_sign = f7_class_sign (b_class) ^ neg_b; + + // Add the mantissae? + bool do_add = a_sign == b_sign; + + // Handle +Infs and -Infs. + bool a_inf = f7_class_inf (a_class); + bool b_inf = f7_class_inf (b_class); + + if (a_inf && b_inf) + { + if (do_add) + return f7_set_inf (cc, a_sign); + else + return f7_set_nan (cc); + } + else if (a_inf) + return f7_set_inf (cc, a_sign); + else if (b_inf) + return f7_set_inf (cc, b_sign); + + int16_t shift16 = sub_ssat16 (aa->expo, bb->expo); + + // aa + 0 = aa. + // Also check MSBit to get rid of Subnormals and 0. + if (shift16 > F7_MANT_BITS || f7_is0 (bb)) + return return_with_sign (cc, aa, -1); + + // 0 + bb = bb. + // 0 - bb = -bb. + // Also check MSBit to get rid of Subnormals and 0. + if (shift16 < -F7_MANT_BITS || f7_is0 (aa)) + return return_with_sign (cc, bb, b_sign); + + // Now aa and bb are non-zero, non-NaN, non-Inf. + // shift > 0 ==> |a| > |b| + // shift < 0 ==> |a| < |b| + int8_t shift = (int8_t) shift16; + bool c_sign = a_sign; + if (shift < 0 + || (shift == 0 && f7_cmp_mant (aa, bb) < 0)) + { + const f7_t *p = aa; aa = bb; bb = p; + c_sign = b_sign; + shift = -shift; + } + uint8_t shift2 = (uint8_t) (shift << 1); + + cc->expo = aa->expo; + // From this point on, no more access aa->expo or bb->expo + // to avoid early-clobber when writing cc->expo. + + cc->flags = c_sign; _Static_assert (F7_FLAGNO_sign == 0, ""); + + // This function uses neither .expo nor .flags from either aa or bb, + // hence there is early-clobber for cc->expo and cc->flags. + f7_addsub_mant_scaled_asm (cc, aa, bb, shift2 | do_add); +} +#endif // F7MOD_addsub_ + + +#ifdef F7MOD_madd_msub_ +F7_WEAK +void f7_madd_msub (f7_t *cc, const f7_t *aa, const f7_t *bb, const f7_t *dd, + bool neg_d) +{ + f7_t xx7, *xx = &xx7; + uint8_t x_lsb = f7_mulx (xx, aa, bb, true /* no rounding */); + uint8_t x_sign = f7_signbit (xx); + int16_t x_expo = xx->expo; + f7_addsub (xx, xx, dd, neg_d); + // Now add LSB. If cancellation occured in the add / sub, then we have the + // chance of extra 8 bits of precision. Turn LSByte into f7_t. + f7_clr (cc); + cc->expo = sub_ssat16 (x_expo, F7_MANT_BITS); + cc->mant[F7_MANT_BYTES - 1] = x_lsb; + cc = f7_normalize_asm (cc); + cc->flags = x_sign; + f7_Iadd (cc, xx); +} +#endif // F7MOD_madd_msub_ + +#ifdef F7MOD_madd_ +F7_WEAK +void f7_madd (f7_t *cc, const f7_t *aa, const f7_t *bb, const f7_t *dd) +{ + f7_madd_msub (cc, aa, bb, dd, false); +} +#endif // F7MOD_madd_ + +#ifdef F7MOD_msub_ +F7_WEAK +void f7_msub (f7_t *cc, const f7_t *aa, const f7_t *bb, const f7_t *dd) +{ + f7_madd_msub (cc, aa, bb, dd, true); +} +#endif // F7MOD_msub_ + + +#ifdef F7MOD_ldexp_ +F7_WEAK +f7_t* f7_ldexp (f7_t *cc, const f7_t *aa, int delta) +{ + uint8_t a_class = f7_classify (aa); + + cc->flags = a_class; + + // Inf and NaN. + if (! f7_class_number (a_class)) + return cc; + + if (f7_msbit (aa) == 0) + return (f7_clr (cc), cc); + + int16_t expo = add_ssat16 (delta, aa->expo); + // Store expo and handle expo = INT16_MIN and INT16_MAX. + if (! f7_store_expo (cc, expo)) + f7_copy_mant (cc, aa); + + return cc; +} +#endif // F7MOD_ldexp_ + + +#if USE_LPM +#elif USE_LD +#else +#error need include "asm-defs.h" +#endif // USE_LPM + +/* + Handling constants: + + F7_PCONST (PVAR, X) + + Set f7_t [const] *PVAR to an LD address for one + of the f7_const_X[_P] constants. + PVAR might be set to point to a local auto that serves + as temporary storage for f7_const_X_P. PVAR is only + valid in the current block. + + const f7_t* F7_PCONST_U16 (PVAR, <ident> X) // USE_LD + f7_t* F7_PCONST_U16 (PVAR, uint16_t X) // USE_LPM + + Set f7_t [const] *PVAR to an LD address for one of the + f7_const_X[_P] constants. PVAR might be set to point to a + local auto that serves as temporary storage for X. PVAR is + only valid in the current block. + + F7_PCONST_VAR (PVAR, VAR) + + VAR is a pointer variable holding the address of some f7_const_X[_P] + constant. Set [const] f7_t *PVAR to a respective LD address. + PVAR might be set to point to a local auto that serves + as temporary storage for f7_const_X_P. PVAR is only + valid in the current block. + + F7_CONST_ADDR (<ident> CST, f7_t* PTMP) + + Return an LD address to for some f7_const_X[_P] constant. + *PTMP might be needed to hold a copy of f7_const_X_P in RAM. + + f7_t* F7_U16_ADDR (uint16_t X, f7_t* PTMP) // USE_LPM + const f7_t* F7_U16_ADDR (<cst-ident> X, <unused>) // USE_LD + + Return an LD address to compile-time constant uint16_t X which is + also known as f7_const_X[_P]. *PTMP might be set to (f7_t) X. + + f7_t* f7_const (f7_t *PVAR, <cst-ident> X) + + Copy f7_const_X[_P] to *PVAR. + + f7_t* f7_copy_flash (f7_t *DST, const f7_t *SRC) + + Copy to *DST with LD (from .rodata in flash) if the address + space is linear, or with LPM (from .progmem.data) if the + address space is not linear. + + f7_t* f7_copy (f7_t *DST, const f7_t* SRC) + + Copy to RAM using LD. + + f7_t* f7_copy_P (f7_t *DST, const f7_t *SRC) + + Copy to RAM using LPM. +*/ + +#if USE_LPM + #define F7_RAW_CONST_ADDR(CST) \ + & F7_(const_##CST##_P) + + #define F7_PCONST(PVAR, CST) \ + f7_t _var_for_##CST; \ + f7_copy_P (& _var_for_##CST, & F7_(const_##CST##_P)); \ + PVAR = & _var_for_##CST + + #define F7_PCONST_U16(PVAR, CST) \ + f7_t _var_for_##CST; \ + PVAR = f7_set_u16 (& _var_for_##CST, CST) + + #define F7_PCONST_VAR(PVAR, VAR) \ + f7_t _var_for_##VAR; \ + f7_copy_P (& _var_for_##VAR, VAR); \ + PVAR = & _var_for_##VAR + + #define MAYBE_const // empty + + #define F7_CONST_ADDR(CST, PTMP) \ + f7_copy_P ((PTMP), & F7_(const_##CST##_P)) + + #define F7_U16_ADDR(CST, PTMP) \ + f7_set_u16 ((PTMP), CST) + +#elif USE_LD + #define F7_RAW_CONST_ADDR(CST) \ + & F7_(const_##CST) + + #define F7_PCONST(PVAR, CST) \ + PVAR = & F7_(const_##CST) + + #define F7_PCONST_U16(PVAR, CST) \ + PVAR = & F7_(const_##CST) + + #define F7_PCONST_VAR(PVAR, VAR) \ + PVAR = (VAR) + + #define F7_CONST_ADDR(CST, PTMP) \ + (& F7_(const_##CST)) + + #define F7_U16_ADDR(CST, PTMP) \ + (& F7_(const_##CST)) + + #define MAYBE_const const +#endif + + + +#define DD(str, X) \ + do { \ + LOG_PSTR (PSTR (str)); \ + f7_dump (X); \ + } while (0) + +#undef DD +#define DD(...) (void) 0 + + +#ifdef F7MOD_sqrt_ +static void sqrt_worker (f7_t *cc, const f7_t *rr) +{ + f7_t tmp7, *tmp = &tmp7; + f7_t aa7, *aa = &aa7; + + // aa in [1/2, 2) => aa->expo in { -1, 0 }. + int16_t a_expo = -(rr->expo & 1); + int16_t c_expo = (rr->expo - a_expo) >> 1; // FIXME: r_expo = INT_MAX??? + + __asm ("" : "+r" (aa)); + + f7_copy (aa, rr); + aa->expo = a_expo; + + // No use of rr or *cc past this point: We may use cc as temporary. + // Approximate square-root of A by X <-- (X + A / X) / 2. + + f7_sqrt_approx_asm (cc, aa); + + // Iterate X <-- (X + A / X) / 2. + // 3 Iterations with 16, 32, 58 bits of precision for the quotient. + + for (uint8_t prec = 16; (prec & 0x80) == 0; prec <<= 1) + { + f7_divx (tmp, aa, cc, (prec & 64) ? 2 + F7_MANT_BITS : prec); + f7_Iadd (cc, tmp); + // This will never underflow because |c_expo| is small. + cc->expo--; + } + + // Similar: |c_expo| is small, hence no ldexp needed. + cc->expo += c_expo; +} + +F7_WEAK +void f7_sqrt (f7_t *cc, const f7_t *aa) +{ + uint8_t a_class = f7_classify (aa); + + if (f7_class_nan (a_class) || f7_class_sign (a_class)) + return f7_set_nan (cc); + + if (f7_class_inf (a_class)) + return f7_set_inf (cc, 0); + + if (f7_class_zero (a_class)) + return f7_clr (cc); + + sqrt_worker (cc, aa); +} +#endif // F7MOD_sqrt_ + + +#ifdef F7MOD_hypot_ +F7_WEAK +void f7_hypot (f7_t *cc, const f7_t *aa, const f7_t *bb) +{ + f7_t xx7, *xx = &xx7; + + f7_square (xx, aa); + f7_square (cc, bb); + f7_Iadd (cc, xx); + f7_Isqrt (cc); +} +#endif // F7MOD_hypot_ + + +#ifdef F7MOD_const_m1_ +#include "libf7-constdef.h" +#endif // -1 + +#ifdef F7MOD_const_1_2_ +#include "libf7-constdef.h" +#endif // 1/2 + +#ifdef F7MOD_const_1_3_ +#include "libf7-constdef.h" +#endif // 1/3 + +#ifdef F7MOD_const_ln2_ +#include "libf7-constdef.h" +#endif // ln2 + +#ifdef F7MOD_const_1_ln2_ +#include "libf7-constdef.h" +#endif // 1_ln2 + +#ifdef F7MOD_const_ln10_ +#include "libf7-constdef.h" +#endif // ln10 + +#ifdef F7MOD_const_1_ln10_ +#include "libf7-constdef.h" +#endif // 1_ln10 + +#ifdef F7MOD_const_1_ +#include "libf7-constdef.h" +#endif // 1 + +#ifdef F7MOD_const_sqrt2_ +#include "libf7-constdef.h" +#endif // sqrt2 + +#ifdef F7MOD_const_2_ +#include "libf7-constdef.h" +#endif // 2 + +#ifdef F7MOD_const_pi_ +#include "libf7-constdef.h" +#endif // pi + + +#ifdef F7MOD_divx_ + +// C /= A +extern void f7_div_asm (f7_t*, const f7_t*, uint8_t); + +F7_WEAK +void f7_divx (f7_t *cc, const f7_t *aa, const f7_t *bb, uint8_t quot_bits) +{ + uint8_t a_class = f7_classify (aa); + uint8_t b_class = f7_classify (bb); + // From this point on, no more access aa->flags or bb->flags + // to avoid early-clobber when writing cc->flags. + + // If either value is NaN, return NaN. + if (f7_class_nan (a_class | b_class) + // If both values are Inf or both are 0, return NaN. + || f7_class_zero (a_class & b_class) + || f7_class_inf (a_class & b_class) + // Inf / 0 = NaN. + || (f7_class_inf (a_class) && f7_class_zero (b_class))) + { + return f7_set_nan (cc); + } + + // 0 / B = 0 for non-zero, non-NaN B. + // A / Inf = 0 for non-zero numbers A. + if (f7_class_zero (a_class) || f7_class_inf (b_class)) + return f7_clr (cc); + + uint8_t c_sign = (a_class ^ b_class) & F7_FLAG_sign; + + if (f7_class_inf (a_class) || f7_class_zero (b_class)) + return f7_set_inf (cc, c_sign); + + cc->flags = c_sign; _Static_assert (F7_FLAGNO_sign == 0, ""); + int16_t expo = sub_ssat16 (aa->expo, bb->expo); + + // Store expo and handle expo = INT16_MIN and INT16_MAX. + if (f7_store_expo (cc, expo)) + return; + + f7_t ss7, *ss = &ss7; + ss->flags = cc->flags; + ss->expo = cc->expo; + + f7_copy_mant (ss, aa); + f7_div_asm (ss, bb, quot_bits); + f7_copy (cc, ss); +} +#endif // F7MOD_divx_ + + +#ifdef F7MOD_div_ +F7_WEAK +void f7_div (f7_t *cc, const f7_t *aa, const f7_t *bb) +{ + /* When f7_divx calls f7_div_asm, dividend and divisor are valid + mantissae, i.e. their MSBit is set. Therefore, the quotient will + be in [0x0.ff..., 0x0.40...] and to adjust it, at most 1 left-shift + is needed. Compute F7_MANT_BITS + 2 bits of the quotient: + One bit is used for rounding, and one bit might be consumed by the + mentioned left-shift. */ + + f7_divx (cc, aa, bb, 2 + F7_MANT_BITS); +} +#endif // F7MOD_div_ + + +#ifdef F7MOD_div1_ +F7_WEAK +void f7_div1 (f7_t *cc, const f7_t *aa) +{ + F7_PCONST_U16 (const f7_t *one, 1); + f7_div (cc, one, aa); +} +#endif // F7MOD_div_ + + +#ifdef F7MOD_fmod_ +F7_WEAK +void f7_fmod (f7_t *cc, const f7_t *aa, const f7_t *bb) +{ + uint8_t a_class = f7_classify (aa); + uint8_t b_class = f7_classify (bb); + + if (! f7_class_number (a_class) + || f7_class_nan (b_class) + || f7_class_zero (b_class)) + { + return f7_set_nan (cc); + } + + // A == 0 and B != 0 => 0. + if (f7_class_zero (a_class)) + return f7_clr (cc); + + f7_t zz7, *zz = & zz7; + + f7_div (zz, aa, bb); + + // Z in Z, |Z| <= |A/B|. + f7_trunc (zz, zz); + + // C = A - Z * B. + f7_msub (cc, zz, bb, aa); + cc->flags ^= F7_FLAG_sign; +} +#endif // F7MOD_fmod_ + + +#ifdef F7MOD_truncx_ +F7_WEAK +f7_t* f7_truncx (f7_t *cc, const f7_t *aa, bool do_floor) +{ + uint8_t a_class = f7_classify (aa); + + if (! f7_class_nonzero (a_class)) + return f7_copy (cc, aa); + + bool sign = f7_class_sign (a_class); + + int16_t a_expo = aa->expo; + + if (a_expo < 0) + { + // |A| < 1. + if (sign & do_floor) + return f7_set_s16 (cc, -1); + + f7_clr (cc); + return cc; + } + else if (a_expo >= F7_MANT_BITS - 1) + // A == floor (A). + return f7_copy (cc, aa); + + f7_t tmp7, *tmp = &tmp7; + + // Needed if aa === cc. + f7_copy (tmp, aa); + + cc->flags = sign; + cc->expo = a_expo; + f7_clr_mant_lsbs (cc, aa, F7_MANT_BITS - 1 - a_expo); + + if (do_floor && cc->sign && f7_cmp_mant (cc, tmp) != 0) + { + F7_PCONST_U16 (const f7_t *one, 1); + f7_Isub (cc, one); + } + + return cc; +} +#endif // F7MOD_truncx_ + + +#ifdef F7MOD_floor_ +F7_WEAK +f7_t* f7_floor (f7_t *cc, const f7_t *aa) +{ + return f7_truncx (cc, aa, true); +} +#endif // F7MOD_floor_ + + +#ifdef F7MOD_trunc_ +F7_WEAK +f7_t* f7_trunc (f7_t *cc, const f7_t *aa) +{ + return f7_truncx (cc, aa, false); +} +#endif // F7MOD_trunc_ + + +#ifdef F7MOD_ceil_ +F7_WEAK +void f7_ceil (f7_t *cc, const f7_t *aa) +{ + cc = f7_copy (cc, aa); + cc->flags ^= F7_FLAG_sign; + cc = f7_floor (cc, cc); + f7_Ineg (cc); +} +#endif // F7MOD_ceil_ + + +#ifdef F7MOD_round_ +F7_WEAK +void f7_round (f7_t *cc, const f7_t *aa) +{ + f7_t tmp; + (void) tmp; + const f7_t *half = F7_CONST_ADDR (1_2, &tmp); + + f7_addsub (cc, aa, half, f7_signbit (aa)); + f7_trunc (cc, cc); +} +#endif // F7MOD_round_ + + +#ifdef F7MOD_horner_ + +// Assertion when using this function is that either cc != xx, +// or if cc == xx, then tmp1 must be non-NULL and tmp1 != xx. +// In General, the calling functions have a spare f7_t object available +// and can pass it down to save some stack. +// Moreover, the power series must have degree 1 at least. + +F7_WEAK +void f7_horner (f7_t *cc, const f7_t *xx, uint8_t n_coeff, const f7_t *coeff, + f7_t *tmp1) +{ + f7_assert (n_coeff > 1); + + if (cc != xx) + tmp1 = cc; + else + f7_assert (tmp1 != NULL && tmp1 != xx); + + f7_t *yy = tmp1; + f7_t tmp27, *tmp2 = &tmp27; + + n_coeff--; + const f7_t *pcoeff = coeff + n_coeff; + + f7_copy_flash (yy, pcoeff); + + while (1) + { + --pcoeff; +#if 1 + f7_Imul (yy, xx); + const f7_t *cst = USE_LD ? pcoeff : f7_copy_P (tmp2, pcoeff); + if (coeff == pcoeff) + return f7_add (cc, yy, cst); + f7_Iadd (yy, cst); +#else + const f7_t *cst = USE_LD ? pcoeff : f7_copy_P (tmp2, pcoeff); + f7_madd (yy, yy, xx, cst); + if (coeff == pcoeff) + { + f7_copy (cc, yy); + return; + } +#endif + } + + __builtin_unreachable(); +} +#endif // F7MOD_horner_ + + +#ifdef F7MOD_log_ +F7_WEAK +void f7_log (f7_t *cc, const f7_t *aa) +{ + f7_logx (cc, aa, NULL); +} +#endif // F7MOD_log_ + + +#ifdef F7MOD_log2_ +F7_WEAK +void f7_log2 (f7_t *cc, const f7_t *aa) +{ + f7_logx (cc, aa, F7_RAW_CONST_ADDR (1_ln2)); +} +#endif // F7MOD_log2_ + + +#ifdef F7MOD_log10_ +F7_WEAK +void f7_log10 (f7_t *cc, const f7_t *aa) +{ + f7_logx (cc, aa, F7_RAW_CONST_ADDR (1_ln10)); +} +#endif // F7MOD_log10_ + + +#ifdef F7MOD_logx_ + +#define ARRAY_NAME coeff_artanh +#include "libf7-array.def" +#undef ARRAY_NAME + +// Compute P * ln(A) if P != NULL and ln(A), otherwise. +// P is a LD-address if USE_LD and a LPM-address if USE_LPM. +// Assumption is that P > 0. + +F7_WEAK +void f7_logx (f7_t *cc, const f7_t *aa, const f7_t *p) +{ + uint8_t a_class = f7_classify (aa); + + if (f7_class_nan (a_class) || f7_class_sign (a_class)) + return f7_set_nan (cc); + + if (f7_class_inf (a_class)) + return f7_set_inf (cc, 0); + + if (f7_class_zero (a_class)) + return f7_set_inf (cc, 1); + + f7_t *yy = cc; + f7_t xx7, *xx = &xx7; + f7_t tmp7, *tmp = &tmp7; + + // Y in [1, 2] = A * 2 ^ (-a_expo). + int16_t a_expo = aa->expo; + f7_copy (yy, aa); + yy->expo = 0; + + // Y in [1 / sqrt2, sqrt2]. + + if (f7_abscmp_msb_ge (yy, F7_(const_sqrt2_msb), F7_(const_sqrt2_expo))) + { + yy->expo = -1; + a_expo = add_ssat16 (a_expo, 1); + } + + const f7_t *one = F7_U16_ADDR (1, & tmp7); + + // X := (Y - 1) / (Y + 1), |X| <= (sqrt2 - 1) / (sqrt2 + 1) ~ 0.172. + f7_sub (xx, yy, one); + f7_Iadd (yy, one); + f7_Idiv (xx, yy); + + // Y := X^2, |Y| < 0.03. + f7_square (yy, xx); + + // Y := artanh (X^2) / X + f7_horner (yy, yy, n_coeff_artanh, coeff_artanh, tmp); + + // C = X * Y = ln A - a_expo * ln2. + f7_mul (cc, xx, yy); + + // X := a_expo * ln2. + f7_set_s16 (xx, a_expo); + f7_Imul (xx, F7_CONST_ADDR (ln2, & tmp7)); + + // C = ln A. + f7_Iadd (cc, xx); + + if (p && USE_LPM) + f7_Imul (cc, f7_copy_P (tmp, p)); + if (p && USE_LD) + f7_Imul (cc, p); +} +#endif // F7MOD_logx_ + + +#ifdef F7MOD_exp_ + +#define ARRAY_NAME coeff_exp +#include "libf7-array.def" +#undef ARRAY_NAME + +#define STATIC static +#include "libf7-constdef.h" // ln2_low +#undef STATIC + +F7_WEAK +void f7_exp (f7_t *cc, const f7_t *aa) +{ + uint8_t a_class = f7_classify (aa); + + if (f7_class_nan (a_class)) + return f7_set_nan (cc); + + /* The maximal exponent of 2 for a double is 1023, hence we may limit + to |A| < 1023 * ln2 ~ 709. We limit to 1024 ~ 1.99 * 2^9 */ + + if (f7_class_inf (a_class) + || (f7_class_nonzero (a_class) && aa->expo >= 9)) + { + if (f7_class_sign (a_class)) + return f7_clr (cc); + else + return f7_set_inf (cc, 0); + } + + f7_t const *cst; + f7_t qq7, *qq = &qq7; + + F7_PCONST (cst, ln2); + + // We limited |A| to 1024 and are now dividing by ln2, hence Q will + // be at most 1024 / ln2 ~ 1477 and fit into 11 bits. We will + // round Q anyway, hence only request 11 bits from the division and + // one additional bit that might be needed to normalize the quotient. + f7_divx (qq, aa, cst, 1 + 11); + + // Use the smallest (by absolute value) remainder system. + f7_round (qq, qq); + int16_t q = f7_get_s16 (qq); + + // Reducing A mod ln2 gives |C| <= ln2 / 2, C = -A mod ln2. + f7_msub (cc, qq, cst, aa); + + // Corrigendum: We added Q * ln2; now add Q times the low part of ln2 + // for better precision. Due to |C| < |A| this is not a no-op in general. + const f7_t *yy = F7_CONST_ADDR (ln2_low, &_var_for_ln2); + f7_madd (cc, qq, yy, cc); + + // Because we computed C = -A mod ... + cc->flags ^= F7_FLAG_sign; + + // Reduce further to |C| < ln2 / 8 which is the range of our MiniMax poly. + const uint8_t MAX_LN2_RED = 3; + int8_t scal2 = 0; + + while (f7_abscmp_msb_ge (cc, F7_(const_ln2_msb), + F7_(const_ln2_expo) - MAX_LN2_RED)) + { + scal2++; + cc->expo--; + } + + f7_horner (cc, cc, n_coeff_exp, coeff_exp, qq); + + while (--scal2 >= 0) + f7_Isquare (cc); + + f7_Ildexp (cc, q); +} +#endif // F7MOD_exp_ + + +#ifdef F7MOD_pow10_ +F7_WEAK +void f7_pow10 (f7_t *cc, const f7_t *aa) +{ + const f7_t *p_ln10; + F7_PCONST (p_ln10, ln10); + f7_mul (cc, aa, p_ln10); + f7_exp (cc, cc); +} +ALIAS (f7_pow10, f7_exp10) +#endif // F7MOD_pow10_ + + +#ifdef F7MOD_cbrt_ +F7_WEAK +void f7_cbrt (f7_t *cc, const f7_t *aa) +{ + f7_copy (cc, aa); + const f7_t *p_1_3; + uint8_t c_flags = cc->flags; + cc->flags &= ~F7_FLAG_sign; + f7_log (cc, cc); + F7_PCONST (p_1_3, 1_3); + f7_Imul (cc, p_1_3); + f7_exp (cc, cc); + + if (c_flags & F7_FLAG_sign) + cc->flags |= F7_FLAG_sign; +} +#endif // F7MOD_cbrt_ + + +#ifdef F7MOD_pow_ +F7_WEAK +void f7_pow (f7_t *cc, const f7_t *aa, const f7_t *bb) +{ +#if 0 + f7_t slots[cc == bb]; + f7_t *yy = cc == bb ? slots : cc; +#else + f7_t yy7, *yy = &yy7; +#endif + f7_log (yy, aa); + f7_Imul (yy, bb); + f7_exp (cc, yy); +} +#endif // F7MOD_pow_ + + +#ifdef F7MOD_powi_ +F7_WEAK +void f7_powi (f7_t *cc, const f7_t *aa, int ii) +{ + uint16_t u16 = ii; + f7_t xx27, *xx2 = &xx27; + + if (ii < 0) + u16 = -u16; + + f7_copy (xx2, aa); + + f7_set_u16 (cc, 1); + + while (1) + { + if (u16 & 1) + f7_Imul (cc, xx2); + + if (! f7_is_nonzero (cc)) + break; + + u16 >>= 1; + if (u16 == 0) + break; + f7_Isquare (xx2); + } + + if (ii < 0) + f7_div1 (xx2, aa); +} +#endif // F7MOD_powi_ + + +#ifdef F7MOD_sincos_ + +#define ARRAY_NAME coeff_sin + #define FOR_SIN + #include "libf7-array.def" + #undef FOR_SIN +#undef ARRAY_NAME + +#define ARRAY_NAME coeff_cos + #define FOR_COS + #include "libf7-array.def" + #undef FOR_COS +#undef ARRAY_NAME + +#define STATIC static +#include "libf7-constdef.h" // pi_low +#undef STATIC + +typedef union +{ + struct + { + bool neg_sin : 1; // Must be bit F7_FLAGNO_sign. + bool neg_cos : 1; + bool do_sin: 1; + bool do_cos: 1; + bool swap_sincos : 1; + uint8_t res : 3; + }; + uint8_t bits; +} sincos_t; + + +F7_WEAK +void f7_sincos (f7_t *ss, f7_t *cc, const f7_t *aa) +{ + uint8_t a_class = f7_classify (aa); + + sincos_t sc = { .bits = a_class & F7_FLAG_sign }; + if (ss != NULL) sc.do_sin = 1; + if (cc != NULL) sc.do_cos = 1; + + if (f7_class_nan (a_class) || f7_class_inf (a_class)) + { + if (sc.do_sin) f7_set_nan (ss); + if (sc.do_cos) f7_set_nan (cc); + return; + } + + f7_t pi7, *pi = &pi7; + f7_t xx7, *xx = &xx7; + f7_t yy7, *yy = &yy7; + f7_t *hh = sc.do_sin ? ss : cc; + + // X = |A| + f7_copy (xx, aa); + xx->flags = 0; + + // Y is how often we subtract PI from X. + f7_clr (yy); + f7_const (pi, pi); + + if (f7_abscmp_msb_ge (xx, F7_(const_pi_msb), F7_(const_pi_expo) + 1)) + { + pi->expo = 1 + F7_(const_pi_expo); // 2*pi + + // Y = X / 2pi. + f7_div (yy, xx, pi); + + // The integral part of |A| / pi mod 2 is bit 55 - x_expo. + if (yy->expo >= F7_MANT_BITS && !f7_is_zero (yy)) + { + // Too big for sensible calculation: Should this be NaN instead? + if (sc.do_sin) f7_clr (ss); + if (sc.do_cos) f7_clr (cc); + return; + } + + // X -= 2pi * [ X / 2pi ] + f7_floor (yy, yy); + + f7_msub (xx, yy, pi, xx); + xx->flags ^= F7_FLAG_sign; + + // We divided by 2pi, but Y should count times we subtracted pi. + yy->expo++; + } + + pi->expo = F7_(const_pi_expo); // pi + f7_sub (hh, xx, pi); + if (!f7_signbit (hh)) + { + // H = X - pi >= 0 => X >= pi + // sin(x) = -sin(x - pi) + // cos(x) = -cos(x - pi) + f7_copy (xx, hh); + // Y: We subtracted pi one more time. + f7_Iadd (yy, f7_set_u16 (hh, 1)); + sc.neg_sin ^= 1; + sc.neg_cos ^= 1; + } + + pi->expo = F7_(const_pi_expo) - 1; // pi/2 + if (f7_gt (xx, pi)) + { + // x > pi/2 + // sin(x) = sin(pi - x) + // cos(x) = -cos(pi - x) + pi->expo = F7_(const_pi_expo); // pi + f7_IRsub (xx, pi); + // Y: We subtracted pi one more time (and then negated). + f7_Iadd (yy, f7_set_u16 (hh, 1)); + yy->flags ^= F7_FLAG_sign; + sc.neg_cos ^= 1; + } + + pi->expo = F7_(const_pi_expo) - 2; // pi/4 + if (f7_gt (xx, pi)) + { + // x > pi/4 + // sin(x) = cos(pi/2 - x) + // cos(x) = sin(pi/2 - x) + pi->expo = F7_(const_pi_expo) - 1; // pi/2 + f7_IRsub (xx, pi); + // Y: We subtracted pi/2 one more time (and then negated). + f7_Iadd (yy, f7_set_1pow2 (hh, -1, 0)); + yy->flags ^= F7_FLAG_sign; + sc.swap_sincos = 1; + } + + if (!f7_is0 (yy)) + { + // Y counts how often we subtracted pi from X in order to + // get 0 <= X < pi/4 as small as possible (Y is 0 mod 0.5). + // Now also subtract the low part of pi: + // f7_const_pi_low = pi - f7_const_pi in order to get more precise + // results in the cases where the final result is close to 0. + const f7_t *pi_low = F7_CONST_ADDR (pi_low, pi); + //f7_const (pi, pi_low); + f7_Imul (yy, pi_low); + f7_Isub (xx, yy); + } + + // X in [0, pi/4]. + // X^2 in [0, pi^2/16] ~ [0, 0.6169] + + f7_square (yy, xx); + + f7_t *x_sin = xx; + f7_t *x_cos = yy; + + if ((sc.do_sin && !sc.swap_sincos) + || (sc.do_cos && sc.swap_sincos)) + { + f7_horner (hh, yy, n_coeff_sin, coeff_sin, NULL); + f7_mul (x_sin, hh, xx); + } + + if ((sc.do_cos && !sc.swap_sincos) + || (sc.do_sin && sc.swap_sincos)) + { + f7_horner (x_cos, yy, n_coeff_cos, coeff_cos, hh); + } + + if (sc.swap_sincos) + { + x_sin = yy; + x_cos = xx; + } + + if (sc.do_sin) + { + x_sin->flags ^= sc.bits; + x_sin->flags &= F7_FLAG_sign; + f7_copy (ss, x_sin); + } + + if (sc.do_cos) + { + x_cos->flags = sc.neg_cos; + f7_copy (cc, x_cos); + } +} +#endif // F7MOD_sincos_ + +#ifdef F7MOD_sin_ +F7_WEAK +void f7_sin (f7_t *ss, const f7_t *aa) +{ + f7_sincos (ss, NULL, aa); +} +#endif // F7MOD_sin_ + +#ifdef F7MOD_cos_ +F7_WEAK +void f7_cos (f7_t *cc, const f7_t *aa) +{ + f7_sincos (NULL, cc, aa); +} +#endif // F7MOD_cos_ + + +#ifdef F7MOD_tan_ +F7_WEAK +void f7_tan (f7_t *tt, const f7_t *aa) +{ + f7_t xcos; + f7_sincos (tt, & xcos, aa); + f7_Idiv (tt, & xcos); +} +#endif // F7MOD_tan_ + + +#ifdef F7MOD_cotan_ +F7_WEAK +void f7_cotan (f7_t *ct, const f7_t *aa) +{ + f7_t xcos; + f7_sincos (ct, & xcos, aa); + f7_div (ct, & xcos, ct); +} +#endif // F7MOD_cotan_ + + +#ifdef F7MOD_sinhcosh_ +F7_WEAK +void f7_sinhcosh (f7_t *cc, const f7_t *aa, bool do_sinh) +{ + f7_t xx7, *xx = &xx7; + // C = exp(A) + f7_exp (cc, aa); + // X = exp(-A) + f7_div (xx, f7_set_u16 (xx, 1), cc); + // sinh(A) = (exp(A) - exp(-A)) / 2 + // cosh(A) = (exp(A) + exp(-A)) / 2 + f7_addsub (cc, cc, xx, do_sinh); + cc->expo--; +} +#endif // F7MOD_sinhcosh_ + + +#ifdef F7MOD_sinh_ +F7_WEAK +void f7_sinh (f7_t *cc, const f7_t *aa) +{ + f7_sinhcosh (cc, aa, true); +} +#endif // F7MOD_sinh_ + + +#ifdef F7MOD_cosh_ +F7_WEAK +void f7_cosh (f7_t *cc, const f7_t *aa) +{ + f7_sinhcosh (cc, aa, false); +} +#endif // F7MOD_cosh_ + + +#ifdef F7MOD_tanh_ +F7_WEAK +void f7_tanh (f7_t *cc, const f7_t *aa) +{ + // tanh(A) = (exp(2A) - 1) / (exp(2A) + 1) + f7_t xx7, *xx = &xx7; + F7_PCONST_U16 (const f7_t *one, 1); + // C = 2A + f7_copy (cc, aa); + cc->expo++; + // C = exp(2A) + f7_exp (cc, cc); + // X = exp(2A) + 1 + f7_add (xx, cc, one); + // C = exp(2A) - 1 + f7_Isub (cc, one); + // C = tanh(A) + f7_Idiv (cc, xx); +} +#endif // F7MOD_tanh_ + + +#ifdef F7MOD_atan_ + +#define MINIMAX_6_6_IN_0_1 + +#define ARRAY_NAME coeff_atan_zahler +#define FOR_NUMERATOR +#include "libf7-array.def" +#undef FOR_NUMERATOR +#undef ARRAY_NAME + +#define ARRAY_NAME coeff_atan_nenner +#define FOR_DENOMINATOR +#include "libf7-array.def" +#undef FOR_DENOMINATOR +#undef ARRAY_NAME + +#include "libf7-constdef.h" + +F7_WEAK +void f7_atan (f7_t *cc, const f7_t *aa) +{ + uint8_t a_class = f7_classify (aa); + uint8_t flags = a_class & F7_FLAG_sign; + + if (f7_class_nan (a_class)) + return f7_set_nan (cc); + + f7_t yy7, *yy = &yy7; + f7_t zz7, *zz = &zz7; + + if (f7_class_inf (a_class)) + { + f7_set_u16 (cc, 0); + goto do_Inf; + } + + // C = |A| + f7_copy (cc, aa); + cc->flags = 0; + + if (!f7_class_zero (a_class) && cc->expo >= 0) + { + // C >= 1: use atan (x) + atan (1/x) = pi / 2 to reduce to [0, 1]. + flags |= 1 << 1; + f7_div (cc, f7_set_u16 (yy, 1), cc); + } +#if !defined (MINIMAX_6_6_IN_0_1) + const uint8_t const_a_msb = 0x89; + const int16_t const_a_expo = -2; + if (f7_abscmp_msb_ge (cc, const_a_msb, const_a_expo)) + { + // We have C in [0, 1] and we want to use argument reduction by means + // of addition theorem atan(x) - atan(y) = atan((x - y) / (1 + xy)). + // We want to split [0, 1] into [0, a] u [a, 1] in such a way that + // the upper interval will be mapped to [-a, a]. The system is easy + // to solve and yiels + // y = 1 / sqrt (3) ~ 0.57735 atan(y) = pi / 6 + // a = (1 - y) / (1 + y) ~ 0.26795 ~ 0x0.8930A2F * 2^-1. + flags |= 1 << 2; + // C <- (C - Y) / (1 + C * Y) in [-a, a] + const f7_t *cst = F7_CONST_ADDR (1_sqrt3, zz); + f7_mul (yy, cc, cst); + f7_Isub (cc, cst); + f7_Iadd (yy, F7_U16_ADDR (1, zz)); + f7_Idiv (cc, yy); + } +#endif + // C <- C * p(C^2) / q(C^2) + f7_square (yy, cc); + f7_horner (zz, yy, n_coeff_atan_zahler, coeff_atan_zahler, NULL); + f7_Imul (zz, cc); + f7_horner (cc, yy, n_coeff_atan_nenner, coeff_atan_nenner, NULL); + f7_div (cc, zz, cc); + +#if !defined (MINIMAX_6_6_IN_0_1) + if (flags & (1 << 2)) + f7_Iadd (cc, F7_CONST_ADDR (pi_6, yy)); +#endif + + if (flags & (1 << 1)) + { + do_Inf:; + // Y = pi / 2 + f7_const (yy, pi); + yy->expo = F7_(const_pi_expo) - 1; + f7_IRsub (cc, yy); + } + + cc->flags = a_class & F7_FLAG_sign; +} +#undef MINIMAX_6_6_IN_0_1 +#endif // F7MOD_atan_ + + +#ifdef F7MOD_asinacos_ + +#define ARRAY_NAME coeff_func_a_zahler +#define FOR_NUMERATOR +#include "libf7-array.def" +#undef FOR_NUMERATOR +#undef ARRAY_NAME + +#define ARRAY_NAME coeff_func_a_nenner +#define FOR_DENOMINATOR +#include "libf7-array.def" +#undef FOR_DENOMINATOR +#undef ARRAY_NAME + +typedef union +{ + struct + { + bool sign : 1; // Must be bit F7_FLAGNO_sign. + bool do_acos : 1; // From caller. + bool have_acos : 1; // What we compute from rational approx p/q. + uint8_t res : 5; + }; + uint8_t bits; +} asinacos_t; + +F7_WEAK +void f7_asinacos (f7_t *cc, const f7_t *aa, uint8_t what) +{ + f7_t xx7, *xx = &xx7; + f7_t xx27, *xx2 = &xx27; + + asinacos_t flags = { .bits = what | f7_signbit (aa) }; + + f7_abs (xx, aa); + + int8_t cmp = f7_cmp (xx, f7_set_u16 (cc, 1)); + + if (cmp == INT8_MIN + || cmp > 0) + { + return f7_set_nan (cc); + } + + if (xx->expo <= -2 || f7_is_zero (xx)) + { + // |A| < 1/2: asin(x) = x * a(2*x^2) + f7_square (xx2, xx); + xx2->expo ++; + } + else + { + // |A| > 1/2: acos (1-x) = sqrt(2*x) * a(x) + // C is 1 from above. + f7_IRsub (xx, cc); + f7_copy (xx2, xx); + flags.have_acos = 1; + } + + // MiniMax [5/4] numerator. + f7_horner (cc, xx2, n_coeff_func_a_zahler, coeff_func_a_zahler, NULL); + + if (flags.have_acos) + { + xx->expo ++; + f7_Isqrt (xx); + } + f7_Imul (cc, xx); + + // MiniMax [5/4] denominator. + f7_horner (xx, xx2, n_coeff_func_a_nenner, coeff_func_a_nenner, NULL); + + f7_Idiv (cc, xx); + + /* + With the current value of C, we have: + + | | do_asin | do_acos + | C | A <= 0 | A >= 0 | A <= 0 | A >= 0 + ----------+------------+-----------+----------+----------+---------- + have_asin | asin (|A|) | -C | C | pi/2 + C | pi/2 - C + have_acos | acos (|A|) | -pi/2 + C | pi/2 - C | pi - C | C + + Result = n_pi2 * pi/2 + C * (c_sign ? -1 : 1) + Result (A, do_asin) = asin (A) + Result (A, do_acos) = acos (A) + + with + c_sign = do_acos ^ have_acos ^ a_sign + n_pi2 = do_acos + have_acos * (a_sign ^ do_acos) ? (-1 : 1) + n_pi2 in { -1, 0, 1, 2 } + */ + + // All what matters for c_sign is bit 0. + uint8_t c_sign = flags.bits; + int8_t n_pi2 = flags.do_acos; + c_sign ^= flags.do_acos; + if (flags.have_acos) + { + n_pi2++; + __asm ("" : "+r" (n_pi2)); + if (c_sign & 1) // c_sign & 1 = a_sign ^ do_acos + n_pi2 -= 2; + c_sign++; + } + + cc->flags = c_sign & F7_FLAG_sign; + + if (n_pi2) + { + f7_const (xx, pi); + if (n_pi2 < 0) + xx->sign = 1; + if (n_pi2 != 2) + xx->expo = F7_(const_pi_expo) - 1; + + f7_Iadd (cc, xx); + } +} +#endif // F7MOD_asinacos_ + + +#ifdef F7MOD_asin_ +F7_WEAK +void f7_asin (f7_t *cc, const f7_t *aa) +{ + f7_asinacos (cc, aa, 0); +} +#endif // F7MOD_asin_ + + +#ifdef F7MOD_acos_ +F7_WEAK +void f7_acos (f7_t *cc, const f7_t *aa) +{ + f7_asinacos (cc, aa, 1 << 1); +} +#endif // F7MOD_acos_ + + +#ifndef IN_LIBGCC2 + +#ifdef F7MOD_put_C_ + +#include <stdio.h> +#include <avr/pgmspace.h> + +static F7_INLINE +uint8_t f7_hex_digit (uint8_t nibble) +{ + nibble = (uint8_t) (nibble + '0'); + if (nibble > '9') + nibble = (uint8_t) (nibble + ('a' - '0' - 10)); + return nibble; +} + +static void f7_put_hex2 (uint8_t x, FILE *stream) +{ + putc ('0', stream); + if (x) + { + putc ('x', stream); + putc (f7_hex_digit (x >> 4), stream); + putc (f7_hex_digit (x & 0xf), stream); + } +} + +#define XPUT(str) \ + fputs_P (PSTR (str), stream) + +// Write to STREAM a line that is appropriate for usage in libf7-const.def. + +F7_WEAK +void f7_put_CDEF (const char *name, const f7_t *aa, FILE *stream) +{ + char buf[7]; + XPUT ("F7_CONST_DEF ("); + fputs (name, stream); + XPUT (",\t"); + uint8_t a_class = f7_classify (aa); + if (! f7_class_nonzero (a_class)) + { + f7_put_hex2 (a_class & F7_FLAGS, stream); + XPUT (",\t0,0,0,0,0,0,0,\t0)"); + return; + } + putc ('0' + (a_class & F7_FLAGS), stream); + XPUT (",\t"); + + for (uint8_t i = 0; i < F7_MANT_BYTES; i++) + { + f7_put_hex2 (aa->mant[F7_MANT_BYTES-1 - i], stream); + putc (',', stream); + } + putc ('\t', stream); + + itoa (aa->expo, buf, 10); + fputs (buf, stream); + XPUT (")"); +} + +void f7_put_C (const f7_t *aa, FILE *stream) +{ + char buf[7]; + + uint8_t a_class = f7_classify (aa); + if (f7_class_nan (a_class)) + { + XPUT ("{ .is_nan = 1 }"); + return; + } + bool sign = a_class & F7_FLAG_sign; + + if (f7_class_inf (a_class)) + { + XPUT ("{ .is_nan = 1, .sign = "); + putc ('0' + sign, stream); + XPUT (" }"); + return; + } + + XPUT ("{ .sign = "); + putc ('0' + sign, stream); + + XPUT (", .mant = { "); + for (uint8_t i = 0; i < F7_MANT_BYTES; i++) + { + f7_put_hex2 (aa->mant[F7_MANT_BYTES-1 - i], stream); + if (i != F7_MANT_BYTES - 1) + putc (',', stream); + } + + XPUT (" }, .expo = "); + itoa (aa->expo, buf, 10); + fputs (buf, stream); + XPUT (" }"); +} +#endif //F7MOD_put_C_ + + +#ifdef F7MOD_dump_ + +#include <avr/pgmspace.h> + +#ifndef AVRTEST_H + +#include <stdio.h> + +static void LOG_PSTR (const char *str) +{ + fputs_P (str, stdout); +} + +static void LOG_PFMT_U16 (const char *fmt, uint16_t x) +{ + printf_P (fmt, x); +} + +static void LOG_PFMT_FLOAT (const char *fmt, float x) +{ + printf_P (fmt, x); +} + +#define LOG_X8(X) LOG_PFMT_U16 (PSTR (" 0x%02x "), (uint8_t)(X)) +#define LOG_PFMT_S16(FMT, X) LOG_PFMT_U16 (FMT, (unsigned)(X)) +#define LOG_PFMT_ADDR(FMT, X) LOG_PFMT_U16 (FMT, (unsigned)(X)) + +#endif // AVRTEST_H + +static void dump_byte (uint8_t b) +{ + LOG_PSTR (PSTR (" ")); + for (uint8_t i = 0; i < 8; i++) + { + LOG_PSTR ((b & 0x80) ? PSTR ("1") : PSTR ("0")); + b = (uint8_t) (b << 1); + } +} + +void f7_dump_mant (const f7_t *aa) +{ + LOG_PSTR (PSTR ("\tmant =")); + for (int i = F7_MANT_BYTES - 1; i >= 0; i--) + LOG_X8 (aa->mant[i]); + LOG_PSTR (PSTR ("\n\t =")); + + for (int i = F7_MANT_BYTES - 1; i >= 0; i--) + dump_byte (aa->mant[i]); + LOG_PSTR (PSTR ("\n")); +} + +void f7_dump (const f7_t *aa) +{ + LOG_PFMT_ADDR (PSTR ("\n0x%04x\tflags = "), aa); + dump_byte (aa->flags); + uint8_t a_class = f7_classify (aa); + LOG_PSTR (PSTR (" = ")); + LOG_PSTR (f7_class_sign (a_class) ? PSTR ("-") : PSTR ("+")); + if (f7_class_inf (a_class)) LOG_PSTR (PSTR ("Inf ")); + if (f7_class_nan (a_class)) LOG_PSTR (PSTR ("NaN ")); + if (f7_class_zero (a_class)) LOG_PSTR (PSTR ("0 ")); + if (f7_class_number (a_class)) LOG_PSTR (PSTR ("Number ")); + + LOG_PFMT_FLOAT (PSTR (" = %.10g\n"), f7_get_float (aa)); + LOG_PFMT_S16 (PSTR ("\texpo = %d\n"), aa->expo); + + f7_dump_mant (aa); +} +#endif // F7MOD_dump_ + +#endif // ! libgcc + +#endif // !AVR_TINY Property changes on: libgcc/config/avr/libf7/libf7.c ___________________________________________________________________ Added: svn:eol-style ## -0,0 +1 ## +native \ No newline at end of property Added: svn:mime-type ## -0,0 +1 ## +text/plain \ No newline at end of property Index: libgcc/config/avr/libf7/libf7.h =================================================================== --- libgcc/config/avr/libf7/libf7.h (nonexistent) +++ libgcc/config/avr/libf7/libf7.h (working copy) @@ -0,0 +1,687 @@ +/* Copyright (C) 2019 Free Software Foundation, Inc. + + This file is part of LIBF7, which is part of GCC. + + GCC is free software; you can redistribute it and/or modify it under + the terms of the GNU General Public License as published by the Free + Software Foundation; either version 3, or (at your option) any later + version. + + GCC is distributed in the hope that it will be useful, but WITHOUT ANY + WARRANTY; without even the implied warranty of MERCHANTABILITY or + FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + for more details. + + Under Section 7 of GPL version 3, you are granted additional + permissions described in the GCC Runtime Library Exception, version + 3.1, as published by the Free Software Foundation. + + You should have received a copy of the GNU General Public License and + a copy of the GCC Runtime Library Exception along with this program; + see the files COPYING3 and COPYING.RUNTIME respectively. If not, see + <http://www.gnu.org/licenses/>. */ + +#ifndef LIBF7_H +#define LIBF7_H +#define IN_LIBF7_H + +#include "f7-renames.h" + +#define F7_MANT_BYTES 7 +#define F7_MANT_BITS (8 * F7_MANT_BYTES) + +/* Using the following GCC features: + -- Unnamed structs / unions (GNU-C) + -- Fixed-point types (GNU-C) + -- Inline asm + -- Setting assembler names by means of __asm (GNU-C). + -- Attributes: alias, always_inline, const, noinline, unused, + progmem, pure, weak, warning + -- GCC built-ins: __builtin_abort, __builtin_constant_p + -- AVR built-ins: __builtin_avr_bitsr, __builtin_avr_rbits +*/ + +/* We have 2 kinds of flags: + + A) The flags that are stored in f7_t.flags: + -- f7_t.is_nan (NaN) + -- f7_t.is_inf (+Inf or -Inf) + -- f7_t.sign (negative or -Inf). + + B) The flags that are returned by f7_classify(). This are the + flags from A) together with + -- _zero: indicate that a number is zero. +*/ + +#define F7_FLAGNO_sign 0 +#define F7_FLAGNO_zero 1 +#define F7_FLAGNO_nan 2 +#define F7_FLAGNO_inf 7 + +#define F7_HAVE_Inf 1 + +// Flags that might be set by f7_classify(). +#define F7_FLAG_sign (1 << F7_FLAGNO_sign) +#define F7_FLAG_zero (1 << F7_FLAGNO_zero) +#define F7_FLAG_nan (1 << F7_FLAGNO_nan) +#define F7_FLAG_inf (F7_HAVE_Inf << F7_FLAGNO_inf) + +// Flags that might be set in f7_t.flags. +#define F7_FLAGS (F7_FLAG_inf | F7_FLAG_nan | F7_FLAG_sign) + +#if !defined __ASSEMBLER__ + +#ifndef IN_LIBGCC2 +#include <stdint.h> +#include <stdbool.h> +#include <stdlib.h> +#include <stdio.h> +#else +/* Do not assume that we have std headers when we build libgcc. */ + +typedef __UINT64_TYPE__ uint64_t; +typedef __UINT32_TYPE__ uint32_t; +typedef __UINT16_TYPE__ uint16_t; +typedef __UINT8_TYPE__ uint8_t; +typedef __INT64_TYPE__ int64_t; +typedef __INT32_TYPE__ int32_t; +typedef __INT16_TYPE__ int16_t; +typedef __INT8_TYPE__ int8_t; +typedef _Bool bool; +#define false 0 +#define true 1 +#define INT8_MIN (-1 - __INT8_MAX__) +#define INT16_MAX __INT16_MAX__ +#define NULL ((void*) 0) +#endif /* IN_LIBGCC2 */ + +#include "asm-defs.h" + +#ifdef __cplusplus +extern "C" { +#define _Static_assert(X, Y) static_assert (X) +#endif // C++ + +#define F7_INLINE inline __attribute__((__always_inline__)) +#define F7_NOINLINE __attribute__((__noinline__)) +#define F7_WEAK __attribute__((__weak__)) +#define F7_PURE __attribute__((__pure__)) +#define F7_UNUSED __attribute__((__unused__)) +#define F7_CONST __attribute__((__const__)) + +#define F7_STRINGY2(X) #X +#define F7_STRINGY(X) F7_STRINGY2(X) +#define F7ASM(X) __asm (F7_STRINGY2(X)) + +typedef struct f7_t +{ + union + { + struct + { + uint8_t sign :1; + uint8_t reserved1 :1; + uint8_t is_nan :1; + uint8_t reserved2 :4; + uint8_t is_inf :1; + }; + uint8_t flags; + }; + + uint8_t mant[7]; + int16_t expo; +} f7_t; + +typedef uint64_t f7_double_t; + +#define F7_MANT_HI4(X) \ + (*(uint32_t*) & (X)->mant[F7_MANT_BYTES - 4]) + +#define F7_MANT_CONST_HI4(X) \ + (*(const uint32_t*) & (X)->mant[F7_MANT_BYTES - 4]) + +#define F7_MANT_HI2(X) \ + (*(uint16_t*) & (X)->mant[F7_MANT_BYTES - 2]) + +static F7_INLINE F7_PURE +uint8_t f7_classify (const f7_t *aa) +{ + extern void f7_classify_asm (void); + register uint8_t rclass __asm ("r24"); + __asm ("%~call %x[f]" + : "=r" (rclass) + : [f] "i" (f7_classify_asm), "z" (aa)); + return rclass; +} + + +// +Inf or -Inf +static F7_INLINE +bool f7_class_inf (uint8_t c) +{ +#if defined (F7_HAVE_Inf) && F7_HAVE_Inf == 1 + return c >= F7_FLAG_inf; +#elif defined (F7_HAVE_Inf) && F7_HAVE_Inf == 0 + (void) c; + return false; +#else +#error macro F7_HAVE_Inf must be defined to 0 or to 1. +#endif // Have Inf +} + +static F7_INLINE +bool f7_is_inf (const f7_t *aa) +{ + return f7_class_inf (aa->flags); +} + +// Not-a-Number (NaN). +static F7_INLINE +bool f7_class_nan (uint8_t c) +{ + return c & F7_FLAG_nan; +} + +static F7_INLINE +bool f7_is_nan (const f7_t *aa) +{ + return f7_class_nan (aa->flags); +} + +// Some number +static F7_INLINE +bool f7_class_number (uint8_t c) +{ + return c <= (F7_FLAG_sign | F7_FLAG_zero); +} + +static F7_INLINE +bool f7_is_number (const f7_t *aa) +{ + return f7_class_number (f7_classify (aa)); +} + +// Zero +static F7_INLINE +bool f7_class_zero (uint8_t c) +{ + return c & F7_FLAG_zero; +} + +static F7_INLINE +bool f7_is_zero (const f7_t *aa) +{ + return f7_class_zero (f7_classify (aa)); +} + +// A non-zero number. +static F7_INLINE +bool f7_class_nonzero (uint8_t c) +{ + return c <= F7_FLAG_sign; +} + +static F7_INLINE +bool f7_is_nonzero (const f7_t *aa) +{ + return f7_class_nonzero (f7_classify (aa)); +} + +static F7_INLINE +bool f7_class_sign (uint8_t c) +{ + return c & F7_FLAG_sign; +} + +static F7_INLINE +bool f7_signbit (const f7_t *aa) +{ + return aa->flags & F7_FLAG_sign; +} + +static F7_INLINE +void f7_set_sign (f7_t *cc, bool sign) +{ + _Static_assert (F7_FLAGNO_sign == 0, ""); + cc->flags &= ~F7_FLAG_sign; + cc->flags |= sign; +} + +static F7_INLINE +void f7_set_nan (f7_t *cc) +{ + cc->flags = F7_FLAG_nan; +} + +static F7_INLINE +void f7_clr (f7_t *cc) +{ + extern void f7_clr_asm (void); + __asm ("%~call %x[f]" + : + : [f] "i" (f7_clr_asm), "z" (cc) + : "memory"); +} + +static F7_INLINE +f7_t* f7_copy (f7_t *cc, const f7_t *aa) +{ + extern void f7_copy_asm (void); + __asm ("%~call %x[f]" + : + : [f] "i" (f7_copy_asm), "z" (cc), "x" (aa) + : "memory"); + return cc; +} + +static F7_INLINE +f7_t* f7_copy_P (f7_t *cc, const f7_t *aa) +{ + extern void f7_copy_P_asm (void); + __asm ("%~call %x[f]" + : + : [f] "i" (f7_copy_P_asm), "x" (cc), "z" (aa) + : "memory"); + return cc; +} + +static F7_INLINE +void f7_copy_mant (f7_t *cc, const f7_t *aa) +{ + extern void f7_copy_mant_asm (void); + __asm ("%~call %x[f]" + : + : [f] "i" (f7_copy_mant_asm), "z" (cc), "x" (aa) + : "memory"); +} + +static F7_INLINE +void f7_set_inf (f7_t *cc, bool sign) +{ +#if F7_HAVE_Inf == 1 + cc->flags = F7_FLAG_inf | sign; +#else + (void) sign; + cc->flags = F7_FLAG_nan; +#endif // Have Inf +} + + +static F7_INLINE +bool f7_msbit (const f7_t *aa) +{ + return aa->mant[F7_MANT_BYTES - 1] & 0x80; +} + +// Quick test against 0 if A is known to be a number (neither NaN nor Inf). +static F7_INLINE +bool f7_is0 (const f7_t *aa) +{ + return 0 == f7_msbit (aa); +} + + +static F7_INLINE +int8_t f7_cmp_mant (const f7_t *aa, const f7_t *bb) +{ + extern void f7_cmp_mant_asm (void); + register int8_t r24 __asm ("r24"); + __asm ("%~call %x[f] ;; %1 %3" + : "=r" (r24) + : [f] "i" (f7_cmp_mant_asm), "x" (aa), "z" (bb)); + return r24; +} + +static F7_INLINE +bool f7_store_expo (f7_t *cc, int16_t expo) +{ + extern void f7_store_expo_asm (void); + register bool r24 __asm ("r24"); + register int16_t rexpo __asm ("r24") = expo; + __asm ("%~call %x[f] ;; %0 %2 %3" + : "=r" (r24) + : [f] "i" (f7_store_expo_asm), "z" (cc), "r" (rexpo)); + return r24; +} + +static F7_INLINE +f7_t* f7_abs (f7_t *cc, const f7_t *aa) +{ + f7_copy (cc, aa); + f7_set_sign (cc, 0); + + return cc; +} + + +F7_PURE extern int8_t f7_cmp (const f7_t*, const f7_t*); +F7_PURE extern bool f7_lt_impl (const f7_t*, const f7_t*); +F7_PURE extern bool f7_le_impl (const f7_t*, const f7_t*); +F7_PURE extern bool f7_gt_impl (const f7_t*, const f7_t*); +F7_PURE extern bool f7_ge_impl (const f7_t*, const f7_t*); +F7_PURE extern bool f7_ne_impl (const f7_t*, const f7_t*); +F7_PURE extern bool f7_eq_impl (const f7_t*, const f7_t*); +F7_PURE extern bool f7_unord_impl (const f7_t*, const f7_t*); + +static F7_INLINE +bool f7_lt (const f7_t *aa, const f7_t *bb) +{ + return 2 & f7_cmp (aa, bb); +} + +static F7_INLINE +bool f7_gt (const f7_t *aa, const f7_t *bb) +{ + return 1 == f7_cmp (aa, bb); +} + +static F7_INLINE +bool f7_le (const f7_t *aa, const f7_t *bb) +{ + int8_t c = f7_cmp (aa, bb); + return (uint8_t) (c + 1) <= 1; +} + +static F7_INLINE +bool f7_ge (const f7_t *aa, const f7_t *bb) +{ + return f7_cmp (aa, bb) >= 0; +} + +static F7_INLINE +bool f7_unordered (const f7_t *aa, const f7_t *bb) +{ + return INT8_MIN == f7_cmp (aa, bb); +} + +static F7_INLINE +bool f7_ordered (const f7_t *aa, const f7_t *bb) +{ + return INT8_MIN != f7_cmp (aa, bb); +} + +static F7_INLINE +bool f7_eq (const f7_t *aa, const f7_t *bb) +{ + return 0 == f7_cmp (aa, bb); +} + +static F7_INLINE +bool f7_ne (const f7_t *aa, const f7_t *bb) +{ + return 1 & f7_cmp (aa, bb); +} + +extern void f7_clr (f7_t*); + +__attribute__((warning ("foo_u16"))) void foo_u16 (void); +__attribute__((warning ("foo_s16"))) void foo_s16 (void); + +extern f7_t* f7_set_s16_impl (f7_t*, int16_t); +extern f7_t* f7_set_u16_impl (f7_t*, uint16_t); + +static F7_INLINE +f7_t* f7_set_u16_worker (f7_t *cc, uint16_t u16) +{ + if (__builtin_constant_p (u16)) + { + if (u16 == 0) + return cc; + + uint8_t off = __builtin_clz (u16); + if (15 - off) + * (uint8_t*) & cc->expo = (uint8_t) (15 - off); + u16 <<= off; + if (u16 & 0xff) + cc->mant[5] = (uint8_t) u16; + if (u16 & 0xff00) + cc->mant[6] = (uint8_t) (u16 >> 8); + + return cc; + } + else + { + foo_u16(); + __builtin_abort(); + return NULL; + } +} + +static F7_INLINE +f7_t* f7_set_u16 (f7_t *cc, uint16_t u16) +{ + if (__builtin_constant_p (u16)) + { + f7_clr (cc); + return f7_set_u16_worker (cc, u16); + } + + return f7_set_u16_impl (cc, u16); +} + +static F7_INLINE +f7_t* f7_set_s16 (f7_t *cc, int16_t s16) +{ + if (__builtin_constant_p (s16)) + { + f7_clr (cc); + + uint16_t u16 = (uint16_t) s16; + + if (s16 < 0) + { + u16 = -u16; + cc->flags = F7_FLAG_sign; + } + + return f7_set_u16_worker (cc, u16); + } + + return f7_set_s16_impl (cc, s16); +} + +static F7_INLINE +void f7_set_eps (f7_t *cc, uint8_t eps, bool sign) +{ + cc = f7_set_u16 (cc, 1); + if (!__builtin_constant_p (sign) || sign) + cc->flags = sign; + cc->mant[0] = eps; +} + +static F7_INLINE +f7_t* f7_set_1pow2 (f7_t *cc, int16_t expo, bool sign) +{ + cc = f7_set_u16 (cc, 1); + cc->expo = expo; + if (!__builtin_constant_p (sign) || sign) + cc->flags = sign; + return cc; +} + +static F7_INLINE +f7_t* f7_set_u64 (f7_t *cc, uint64_t u64) +{ + extern f7_t* f7_set_u64_asm (uint64_t, f7_t*); + return f7_set_u64_asm (u64, cc); +} + +static F7_INLINE +f7_t* f7_set_s64 (f7_t *cc, int64_t s64) +{ + extern f7_t* f7_set_s64_asm (int64_t, f7_t*); + return f7_set_s64_asm (s64, cc); +} + +extern void f7_set_double_impl (f7_double_t, f7_t*); +static F7_INLINE +void f7_set_double (f7_t *cc, f7_double_t val64) +{ + f7_set_double_impl (val64, cc); +} + +extern f7_t* f7_init_impl (uint64_t, uint8_t, f7_t*, int16_t); + +static F7_INLINE +f7_t* f7_init (f7_t *cc, uint8_t flags, uint64_t mant, int16_t expo) +{ + return f7_init_impl (mant, flags, cc, expo); +} + +extern f7_t* f7_set_s32 (f7_t*, int32_t); +extern f7_t* f7_set_u16 (f7_t*, uint16_t); +extern f7_t* f7_set_u32 (f7_t*, uint32_t); +extern void f7_set_float (f7_t*, float); +extern void f7_set_pdouble (f7_t*, const f7_double_t*); + +F7_PURE extern int16_t f7_get_s16 (const f7_t*); +F7_PURE extern int32_t f7_get_s32 (const f7_t*); +F7_PURE extern int64_t f7_get_s64 (const f7_t*); +F7_PURE extern uint16_t f7_get_u16 (const f7_t*); +F7_PURE extern uint32_t f7_get_u32 (const f7_t*); +F7_PURE extern uint64_t f7_get_u64 (const f7_t*); +F7_PURE extern float f7_get_float (const f7_t*); +F7_PURE extern f7_double_t f7_get_double (const f7_t*); + +#if USE_LPM == 1 + #define F7_PGMSPACE __attribute__((__progmem__)) + #define f7_copy_flash f7_copy_P + + #define f7_const(X, NAME) \ + f7_copy_P ((X), & F7_(const_ ## NAME ## _P)) + + #define F7_CONST_DEF(NAME, FLAGS, M0, M1, M2, M3, M4, M5, M6, EXPO) \ + extern const f7_t F7_(const_ ## NAME ## _P); + #include "libf7-const.def" + #undef F7_CONST_DEF +#else + #define F7_PGMSPACE // Empty + #define f7_copy_flash f7_copy + + #define f7_const(X, NAME) \ + f7_copy ((X), & F7_(const_ ## NAME)) + + #define F7_CONST_DEF(NAME, FLAGS, M0, M1, M2, M3, M4, M5, M6, EXPO) \ + extern const f7_t F7_(const_ ## NAME); + #include "libf7-const.def" + #undef F7_CONST_DEF +#endif // USE_LPM + + +// Basic floating point arithmetic: +// double output <=> f7_t* +// double input <=> const f7_t* +extern f7_t* f7_neg (f7_t*, const f7_t*); +extern void f7_add (f7_t*, const f7_t*, const f7_t*); +extern void f7_sub (f7_t*, const f7_t*, const f7_t*); +extern void f7_mul (f7_t*, const f7_t*, const f7_t*); +extern void f7_div (f7_t*, const f7_t*, const f7_t*); + +// Analogies of functions from math.h: +// double output <=> f7_t* +// double input <=> const f7_t* +extern void f7_fabs (f7_t*, const f7_t*); +extern void f7_fmod (f7_t*, const f7_t*, const f7_t*); +extern void f7_frexp (f7_t*, const f7_t*, int*); +extern void f7_exp (f7_t*, const f7_t*); +extern void f7_log (f7_t*, const f7_t*); +extern void f7_pow (f7_t*, const f7_t*, const f7_t*); +extern void f7_sqrt (f7_t*, const f7_t*); +extern void f7_cbrt (f7_t*, const f7_t*); +extern void f7_hypot (f7_t*, const f7_t*, const f7_t*); +extern f7_t* f7_ldexp (f7_t*, const f7_t*, int); +extern f7_t* f7_fmax (f7_t*, const f7_t*, const f7_t*); +extern f7_t* f7_fmin (f7_t*, const f7_t*, const f7_t*); +extern f7_t* f7_trunc (f7_t*, const f7_t*); +extern f7_t* f7_floor (f7_t*, const f7_t*); +extern void f7_ceil (f7_t*, const f7_t*); +extern void f7_round (f7_t*, const f7_t*); +extern void f7_sin (f7_t*, const f7_t*); +extern void f7_cos (f7_t*, const f7_t*); +extern void f7_tan (f7_t*, const f7_t*); +extern void f7_atan (f7_t*, const f7_t*); +extern void f7_asin (f7_t*, const f7_t*); +extern void f7_acos (f7_t*, const f7_t*); +extern void f7_tanh (f7_t*, const f7_t*); +extern void f7_sinh (f7_t*, const f7_t*); +extern void f7_cosh (f7_t*, const f7_t*); +extern void f7_log2 (f7_t*, const f7_t*); +extern void f7_log10 (f7_t*, const f7_t*); +extern void f7_exp10 (f7_t*, const f7_t*); +extern void f7_pow10 (f7_t*, const f7_t*); + +// Just prototypes, not implemented yet. +extern void f7_atan2 (f7_t*, const f7_t*, const f7_t*); +extern long f7_lrint (const f7_t*); +extern long f7_lround (const f7_t*); + +// Helper functions, aliases, convenience. +extern void f7_div1 (f7_t*, const f7_t*); +extern void f7_square (f7_t*, const f7_t*); + +extern void f7_powi (f7_t*, const f7_t*, int); +extern f7_t* f7_max (f7_t*, const f7_t*, const f7_t*); +extern f7_t* f7_min (f7_t*, const f7_t*, const f7_t*); +extern f7_t* f7_truncx (f7_t*, const f7_t*, bool); +extern void f7_cotan (f7_t*, const f7_t*); +extern void f7_sincos (f7_t*, f7_t*, const f7_t*); +extern void f7_asinacos (f7_t*, const f7_t*, uint8_t); +extern void f7_sinhcosh (f7_t*, const f7_t*, bool); + +extern void f7_horner (f7_t*, const f7_t*, uint8_t, const f7_t *coeff, f7_t*); +extern void f7_mul_noround (f7_t*, const f7_t*, const f7_t*); +extern void f7_clr_mant_lsbs (f7_t*, const f7_t*, uint8_t) F7ASM(f7_clr_mant_lsbs_asm); + +F7_PURE extern int8_t f7_cmp_unordered (const f7_t*, const f7_t*, bool); +F7_PURE extern int8_t f7_cmp_abs (const f7_t*, const f7_t*); + +F7_PURE extern bool f7_abscmp_msb_ge (const f7_t*, uint8_t msb, int16_t expo); +extern void f7_addsub (f7_t*, const f7_t*, const f7_t*, bool neg_b); +extern void f7_madd_msub (f7_t*, const f7_t*, const f7_t*, const f7_t*, bool); +extern void f7_madd (f7_t*, const f7_t*, const f7_t*, const f7_t*); +extern void f7_msub (f7_t*, const f7_t*, const f7_t*, const f7_t*); +extern uint8_t f7_mulx (f7_t*, const f7_t*, const f7_t*, bool); +extern void f7_divx (f7_t*, const f7_t*, const f7_t*, uint8_t); +extern void f7_logx (f7_t*, const f7_t*, const f7_t*); +extern f7_t* f7_minmax (f7_t*, const f7_t*, const f7_t*, bool); + +// Idem: +// f7_Ifunc (y) = f7_func (y, y) +// f7_Ifunc (y, x) = f7_func (y, y, x) +extern void f7_Iadd (f7_t*, const f7_t*); +extern void f7_Isub (f7_t*, const f7_t*); +extern void f7_Imul (f7_t*, const f7_t*); +extern void f7_Idiv (f7_t*, const f7_t*); +extern void f7_IRsub (f7_t*, const f7_t*); +extern void f7_Ineg (f7_t*); +extern void f7_Isqrt (f7_t*); +extern void f7_Isquare (f7_t*); +extern f7_t* f7_Ildexp (f7_t*, int); + +// Protoypes for some functions from libf7-asm.sx. +F7_CONST extern uint16_t f7_sqrt16_round (uint16_t) F7ASM(f7_sqrt16_round_asm); +F7_CONST extern uint8_t f7_sqrt16_floor (uint16_t) F7ASM(f7_sqrt16_floor_asm); +extern void f7_addsub_mant_scaled_asm (f7_t*, const f7_t*, const f7_t*, uint8_t); +extern uint8_t f7_mul_mant_asm (f7_t*, const f7_t*, const f7_t*, uint8_t); +extern void f7_sqrt_approx_asm (f7_t*, const f7_t*); +extern uint64_t f7_lshrdi3 (uint64_t, uint8_t) F7ASM(f7_lshrdi3_asm); +extern uint64_t f7_ashldi3 (uint64_t, uint8_t) F7ASM(f7_ashldi3_asm); +// Normalize a non-Inf, non-NaN value. Sets .sign to 0. +extern f7_t* f7_normalize_asm (f7_t*); + +// Dumping. +#ifndef IN_LIBGCC2 +extern void f7_dump (const f7_t*); +extern void f7_dump_mant (const f7_t*); +extern void f7_put_C (const f7_t*, FILE*); +extern void f7_put_CDEF (const char *name, const f7_t*, FILE*); +#endif /* IN_LIBGCC2 */ + +#ifdef __cplusplus +} // extern "C" +#include "libf7-class.h" +#endif // C++ + +#endif /* __ASSEMBLER__ */ +#undef IN_LIBF7_H +#endif /* LIBF7_H */ Property changes on: libgcc/config/avr/libf7/libf7.h ___________________________________________________________________ Added: svn:eol-style ## -0,0 +1 ## +native \ No newline at end of property Added: svn:mime-type ## -0,0 +1 ## +text/plain \ No newline at end of property Index: libgcc/config/avr/libf7/t-libf7 =================================================================== --- libgcc/config/avr/libf7/t-libf7 (nonexistent) +++ libgcc/config/avr/libf7/t-libf7 (working copy) @@ -0,0 +1,159 @@ +# Used except --with-libf7=no + +avrsrc := $(srcdir)/config/avr + +libf7 := $(avrsrc)/libf7 + +F7_PREFIX = __f7_ + +include $(libf7)/libf7-common.mk + +LIBF7_DF_CONV += floatundidf floatdidf # floatunsidf floatsidf + +# Wrappers like f7_lt_impl for f7_lt etc. because the latter is inline. +LIBF7_DF_CMP += lt le gt ge ne eq unord + +F7_C_PARTS += $(LIBF7_DF_CONV) $(LIBF7_DF_CMP) + +# -mcall-prologues +CALL_PROLOGUES += $(LIBF7_DF_CONV) + +# -Wno-missing-prototypes +NO_PROTO += $(LIBF7_DF_CONV) + +F7F += le_impl lt_impl gt_impl ge_impl ne_impl eq_impl unord_impl + + +$(libf7)/f7-renames.h: $(libf7)/f7renames.sh $(libf7)/libf7-common.mk + $< head $(F7_PREFIX) t-libf7 > $@ + $< c $(F7_PREFIX) $(F7F) >> $@ + $< cst $(F7_PREFIX) $(F7F_cst) >> $@ + $< asm $(F7_PREFIX) $(F7F_asm) >> $@ + $< tail $(F7_PREFIX) >> $@ + +# The right-hand sides like g_ddd come from libf7-common.mk. +# The _m_ wraps are added by t-libf7-math + +# __adddf3, ... +F7_ASM_WRAPS_g_ddd += $(g_ddd) + +# __ltdf2, ... +F7_ASM_WRAPS_g_xdd_cmp += $(g_xdd_cmp) + +# __floatsidf, ... +F7_ASM_WRAPS_g_dx += $(g_dx) + +# __fixdfsi, ... +F7_ASM_WRAPS_g_xd += $(g_xd) + +$(libf7)/f7-wraps.h: $(libf7)/f7wraps.sh \ + $(libf7)/libf7-common.mk $(libf7)/t-libf7-math + $< header "WITH_LIBF7_MATH_FUNCTIONS=$(WITH_LIBF7_MATH_FUNCTIONS)" "WITH_LIBF7_MATH_SYMBOLS=$(WITH_LIBF7_MATH_SYMBOLS)" > $@ + $< ddd_libgcc $(F7_ASM_WRAPS_g_ddd) >> $@ + $< xdd_libgcc_cmp $(F7_ASM_WRAPS_g_xdd_cmp) >> $@ + $< xd_libgcc $(F7_ASM_WRAPS_g_xd) >> $@ + $< dx_libgcc $(F7_ASM_WRAPS_g_dx) >> $@ + $< ddd_math $(F7_ASM_WRAPS_m_ddd) >> $@ + $< ddx_math $(F7_ASM_WRAPS_m_ddx) >> $@ + $< dd_math $(F7_ASM_WRAPS_m_dd) >> $@ + $< xd_math $(F7_ASM_WRAPS_m_xd) >> $@ + +F7_ASM_WRAPS += $(F7_ASM_WRAPS_g_xd) +F7_ASM_WRAPS += $(F7_ASM_WRAPS_g_dx) +F7_ASM_WRAPS += $(F7_ASM_WRAPS_g_ddd) +F7_ASM_WRAPS += $(F7_ASM_WRAPS_g_xdd_cmp) +F7_ASM_WRAPS += $(F7_ASM_WRAPS_m_ddd) +F7_ASM_WRAPS += $(F7_ASM_WRAPS_m_ddx) +F7_ASM_WRAPS += $(F7_ASM_WRAPS_m_dd) +F7_ASM_WRAPS += $(F7_ASM_WRAPS_m_xd) + +F7_ASM_PARTS += $(patsubst %, D_%, $(F7_ASM_WRAPS)) + +# Options +F7_FLAGS += -I $(libf7) -save-temps=obj + +# t-avr::HOST_LIBGCC2_CFLAGS sets -mcall-prologues which will inhibits +# tail-call optimizations. The user could get it with -mrelax, but we +# just switch it off here and then explicitly on again for the +# CALL_PROLOGUES modules. + +F7_C_FLAGS += $(F7_FLAGS) \ + -dp -g0 \ + -mno-call-prologues \ + -fno-lto -Os \ + -fdata-sections -ffunction-sections \ + -fno-reorder-blocks \ + -fno-tree-loop-optimize \ + -fno-tree-loop-im -fno-move-loop-invariants + +F7_ASM_FLAGS += $(F7_FLAGS) + +$(patsubst %, f7_c_%.o, $(CALL_PROLOGUES)) \ + : F7_C_FLAGS += -mcall-prologues + +$(patsubst %, f7_c_%.o, $(STRICT_X)) \ + : F7_C_FLAGS += -mstrict-X + +$(patsubst %, f7_c_%.o, $(NO_PROTO)) \ + : F7_C_FLAGS += -Wno-missing-prototypes + +# Depends will be worked out by the libgcc build system. + +F7_C_OBJECTS = $(patsubst %, f7_c_%$(objext), $(F7_C_PARTS)) +F7_ASM_OBJECTS = $(patsubst %, f7_asm_%$(objext), $(F7_ASM_PARTS)) + +$(F7_ASM_OBJECTS) $(F7_C_OBJECTS) : $(libf7)/t-libf7 +$(F7_ASM_OBJECTS) $(F7_C_OBJECTS) : $(libf7)/t-libf7-math +$(F7_ASM_OBJECTS) $(F7_C_OBJECTS) : $(libf7)/t-libf7-math-symbols + +.PHONY: log_vars + +all: log_vars + +log_vars: + $(info # libf7: WITH_LIBF7_MATH_FUNCTIONS = $(WITH_LIBF7_MATH_FUNCTIONS)) + $(info # libf7: WITH_LIBF7_MATH_SYMBOLS = $(WITH_LIBF7_MATH_SYMBOLS)) + $(info # libf7: F7_C_PARTS = $(F7_C_PARTS)) + $(info # libf7: F7_C_OBJECTS = $(F7_C_OBJECTS)) + $(info # libf7: F7_ASM_PARTS = $(F7_ASM_PARTS)) + $(info # libf7: F7_ASM_OBJECTS = $(F7_ASM_OBJECTS)) + +# Build the libf7 C objects and add them to libgcc.a. + +f7_parts := $(F7_C_PARTS) + +iter-items := $(f7_parts) +iter-labels := $(f7_parts) + +include $(srcdir)/empty.mk $(patsubst %,$(libf7)/libf7-c-object.mk,$(iter-items)) + +libgcc-objects += $(patsubst %,f7_c_%$(objext),$(F7_C_PARTS)) + +# Build the libf7 ASM objects and add them to libgcc.a. + +f7_parts := $(F7_ASM_PARTS) + +iter-items := $(f7_parts) +iter-labels := $(f7_parts) + +include $(srcdir)/empty.mk $(patsubst %,$(libf7)/libf7-asm-object.mk,$(iter-items)) + +libgcc-objects += $(patsubst %,f7_asm_%$(objext),$(F7_ASM_PARTS)) + +.PHONY: clean-f7 + +clean: clean-f7 + +clean-f7: + rm -f $(wildcard f7_*.i f7_*.s f7_*.o) + +# Get rid if any DFmode remains. + +LIB2FUNCS_EXCLUDE += \ + _sf_to_df \ + _fixdfdi \ + _fixunsdfsi \ + _floatundidf \ + _fixunsdfdi \ + _floatdidf \ + _powidf2 Property changes on: libgcc/config/avr/libf7/t-libf7 ___________________________________________________________________ Added: svn:eol-style ## -0,0 +1 ## +native \ No newline at end of property Added: svn:mime-type ## -0,0 +1 ## +text/plain \ No newline at end of property Index: libgcc/config/avr/libf7/t-libf7-math =================================================================== --- libgcc/config/avr/libf7/t-libf7-math (nonexistent) +++ libgcc/config/avr/libf7/t-libf7-math (working copy) @@ -0,0 +1,21 @@ +# Triggered by --with-libf7=math or --with-libf7=math-symbols +# +# We provide weak double wrappers for functions specified in math.h, +# but with __ prepended to the symbol name used for the double function. +# For example we provide double __sin (double) but neither sin nor sinl. +# To get weak symbols according to math.h, t-libf7-math-symbols has to +# be used which is triggered by --with-libf7=math-symbols. + +WITH_LIBF7_MATH_FUNCTIONS = 1 + +# __sin, ... +F7_ASM_WRAPS_m_dd += $(m_dd) + +# __pow, __fmin, ... +F7_ASM_WRAPS_m_ddd += $(m_ddd) + +# __ldexp, ... +F7_ASM_WRAPS_m_ddx += $(m_ddx) + +# __lrint, ... +F7_ASM_WRAPS_m_xd += $(m_xd) Property changes on: libgcc/config/avr/libf7/t-libf7-math ___________________________________________________________________ Added: svn:eol-style ## -0,0 +1 ## +native \ No newline at end of property Added: svn:mime-type ## -0,0 +1 ## +text/plain \ No newline at end of property Index: libgcc/config/avr/libf7/t-libf7-math-symbols =================================================================== --- libgcc/config/avr/libf7/t-libf7-math-symbols (nonexistent) +++ libgcc/config/avr/libf7/t-libf7-math-symbols (working copy) @@ -0,0 +1,11 @@ +# Triggered by --with-libf7=math-symbols +# +# We have at least one module in libgcc that depends on __SIZEOF_DOUBLE__ +# or __SIZEOF_LONG_DOUBLE__ which means that t-avrlibc must not copy +# double32/64 or long-double32/64 variants from the vanilla one. +# This occurs when some module(s) define(s) weak aliases for functions +# that usually live in libm. + +WITH_LIBF7_MATH_SYMBOLS = 1 + +F7_FLAGS += -DWITH_LIBF7_MATH_SYMBOLS Property changes on: libgcc/config/avr/libf7/t-libf7-math-symbols ___________________________________________________________________ Added: svn:eol-style ## -0,0 +1 ## +native \ No newline at end of property Added: svn:mime-type ## -0,0 +1 ## +text/plain \ No newline at end of property ^ permalink raw reply [flat|nested] 22+ messages in thread
* [PING^1][patch,avr, 3/3] Support 64-bit (long) double: libf7. 2019-12-16 17:09 ` [patch,avr, 3/3] Support 64-bit (long) double: libf7 Georg-Johann Lay @ 2019-12-28 12:02 ` Georg-Johann Lay 2020-01-06 13:09 ` [Ping^2][patch,avr, " Georg-Johann Lay 2020-01-06 16:26 ` [patch,avr, " Jeff Law 2 siblings, 0 replies; 22+ messages in thread From: Georg-Johann Lay @ 2019-12-28 12:02 UTC (permalink / raw) To: GCC Patches; +Cc: Denis Chertykov Ping #1 > Am 16.12.19 um 17:40 schrieb Georg-Johann Lay: > Patch 3/3 is the actual libf7 implementation. A great deal of which is > assembly, together with C + inline assembly for higher routines. > > Johann > > libgcc/config/avr/libf7/ > * t-libf7: New file. > * t-libf7-math: New file. > * t-libf7-math-symbols: New file. > * libf7-common.mk: New file. > * libf7-asm-object.mk: New file. > * libf7-c-object.mk: New file. > * asm-defs.h: New file. > * libf7.h: New file. > * libf7.c: New file. > * libf7-asm.sx: New file. > * libf7-array.def: New file. > * libf7-const.def: New file. > * libf7-constdef.h: New file. > * f7renames.sh: New script. > * f7wraps.sh: New script. > * f7-renames.h: New generated file. > * f7-wraps.h: New generated file. > ^ permalink raw reply [flat|nested] 22+ messages in thread
* [Ping^2][patch,avr, 3/3] Support 64-bit (long) double: libf7. 2019-12-16 17:09 ` [patch,avr, 3/3] Support 64-bit (long) double: libf7 Georg-Johann Lay 2019-12-28 12:02 ` [PING^1][patch,avr, " Georg-Johann Lay @ 2020-01-06 13:09 ` Georg-Johann Lay 2020-01-06 16:26 ` [patch,avr, " Jeff Law 2 siblings, 0 replies; 22+ messages in thread From: Georg-Johann Lay @ 2020-01-06 13:09 UTC (permalink / raw) To: GCC Patches; +Cc: Denis Chertykov [-- Attachment #1: Type: text/plain, Size: 851 bytes --] Ping #2 Georg-Johann Lay schrieb: > Am 16.12.19 um 17:40 schrieb Georg-Johann Lay: > Patch 3/3 is the actual libf7 implementation. A great deal of which is > assembly, together with C + inline assembly for higher routines. > > Johann > > libgcc/config/avr/libf7/ > * t-libf7: New file. > * t-libf7-math: New file. > * t-libf7-math-symbols: New file. > * libf7-common.mk: New file. > * libf7-asm-object.mk: New file. > * libf7-c-object.mk: New file. > * asm-defs.h: New file. > * libf7.h: New file. > * libf7.c: New file. > * libf7-asm.sx: New file. > * libf7-array.def: New file. > * libf7-const.def: New file. > * libf7-constdef.h: New file. > * f7renames.sh: New script. > * f7wraps.sh: New script. > * f7-renames.h: New generated file. > * f7-wraps.h: New generated file. > [-- Attachment #2: m64-libf7.diff --] [-- Type: text/plain, Size: 197795 bytes --] Index: libgcc/config/avr/libf7/asm-defs.h =================================================================== --- libgcc/config/avr/libf7/asm-defs.h (nonexistent) +++ libgcc/config/avr/libf7/asm-defs.h (working copy) @@ -0,0 +1,237 @@ +/* Copyright (C) 2019 Free Software Foundation, Inc. + + This file is part of LIBF7, which is part of GCC. + + GCC is free software; you can redistribute it and/or modify it under + the terms of the GNU General Public License as published by the Free + Software Foundation; either version 3, or (at your option) any later + version. + + GCC is distributed in the hope that it will be useful, but WITHOUT ANY + WARRANTY; without even the implied warranty of MERCHANTABILITY or + FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + for more details. + + Under Section 7 of GPL version 3, you are granted additional + permissions described in the GCC Runtime Library Exception, version + 3.1, as published by the Free Software Foundation. + + You should have received a copy of the GNU General Public License and + a copy of the GCC Runtime Library Exception along with this program; + see the files COPYING3 and COPYING.RUNTIME respectively. If not, see + <http://www.gnu.org/licenses/>. */ + +#ifndef ASM_DEFS_H +#define ASM_DEFS_H + +#ifdef __AVR__ +#ifdef __ASSEMBLER__ +/*****************************************************************/ +/* Stuff for Assembler-only */ +/*****************************************************************/ + +#if defined (__AVR_TINY__) + #define __tmp_reg__ 16 + #define __zero_reg__ 17 +#else + #define __tmp_reg__ 0 + #define __zero_reg__ 1 +#endif /* AVR_TINY */ + +#define __SREG__ 0x3f +#define __SP_L__ 0x3d +#if defined (__AVR_HAVE_SPH__) +#define __SP_H__ 0x3e +#endif + +#if !defined ASM_DEFS_HAVE_DEFUN +.macro DEFUN name + .global \name + .func \name + \name: +.endm + +.macro ENDF name + .size \name, .-\name + .endfunc +.endm + +.macro LABEL name + .global \name + \name: +.endm +#endif /* HAVE_DEFUN */ + + +#if defined (__AVR_HAVE_JMP_CALL__) + #define XCALL call + #define XJMP jmp +#else + #define XCALL rcall + #define XJMP rjmp +#endif + +#if defined (__AVR_HAVE_EIJMP_EICALL__) + #define XICALL eicall + #define XIJMP eijmp + #define PC_SIZE 3 +#else + #define XICALL icall + #define XIJMP ijmp + #define PC_SIZE 2 +#endif + +.macro skipnext + cpse r16, r16 +.endm + +/* + Factor out support of MOVW. Usage is like + + wmov 30, 24 + + to move R25:R24 to R31:R30, i.e. plain register numbers + are required and no register prefix 'R'. +*/ + +#if defined (__AVR_HAVE_MOVW__) +#define wmov movw +#else + .macro wmov dst src + ..dst = \dst + ..src = \src + ..regno = 0 + .irp reg, \ + r0, r1, r2, r3, r4, r5, r6, r7, r8, r9, \ + r10, r11, r12, r13, r14, r15, r16, r17, r18, r19, \ + r20, r21, r22, r23, r24, r25, r26, r27, r28, r29, \ + r30, r31 + .ifc \reg,\dst + ..dst = ..regno + .endif + .ifc \reg,\src + ..src = ..regno + .endif + ..regno = ..regno + 1 + .endr + + ..regno = 0 + + .irp reg, \ + R0, R1, R2, R3, R4, R5, R6, R7, R8, R9, \ + R10, R11, R12, R13, R14, R15, R16, R17, R18, R19, \ + R20, R21, R22, R23, R24, R25, R26, R27, R28, R29, \ + R30, R31 + .ifc \reg,\dst + ..dst = ..regno + .endif + .ifc \reg,\src + ..src = ..regno + .endif + ..regno = ..regno + 1 + .endr + + ..regno = 0 + + .irp reg, \ + X, x, XL, xl, Xl, xL, x, x \ + Y, y, YL, yl, Yl, yL, y, y, \ + Z, z, ZL, zl, Zl, zL, z, z + .ifc \reg,\dst + ..dst = (..regno / 8) + 26 + .endif + .ifc \reg,\src + ..src = (..regno / 8) + 26 + .endif + ..regno = ..regno + 1 + .endr + + mov ..dst+0, ..src+0 + mov ..dst+1, ..src+1 + .endm +#endif /* MOVW */ + + +#if !defined (__AVR_TINY__) +/* + Convenience macro for easy use of __prologue_saves__ from libgcc. + Push the N_PUSHED callee-saved registers Y, R17, R16, R15, ... + with 0 <= N_PUSHED <= 18. The frame pointer (Y) is set up according + to a frame size of N_FRAME. Clobbers TMP_REG. + For the code of __prologue_saves__ from libgcc see + http://gcc.gnu.org/viewcvs/gcc/trunk/libgcc/config/avr/lib1funcs.S?revision=267494&view=markup#l2159 +*/ + +.macro do_prologue_saves n_pushed n_frame=0 + ldi r26, lo8(\n_frame) + ldi r27, hi8(\n_frame) + ldi r30, lo8(gs(.L_prologue_saves.\@)) + ldi r31, hi8(gs(.L_prologue_saves.\@)) + XJMP __prologue_saves__ + ((18 - (\n_pushed)) * 2) +.L_prologue_saves.\@: +.endm + +/* + Convenience macro for easy use of __epilogue_restores__ from libgcc. + Undo the effect of __prologue_saves__. Clobbers TMP_REG. + For the code of __epilogue_restores__ from libgcc see + http://gcc.gnu.org/viewcvs/gcc/trunk/libgcc/config/avr/lib1funcs.S?revision=267494&view=markup#l2216 +*/ + +.macro do_epilogue_restores n_pushed n_frame=0 + in r28, __SP_L__ +#ifdef __AVR_HAVE_SPH__ + in r29, __SP_H__ +.if \n_frame > 63 + subi r28, lo8(-\n_frame) + sbci r29, hi8(-\n_frame) +.elseif \n_frame > 0 + adiw r28, \n_frame +.endif +#else + clr r29 +.if \n_frame > 0 + subi r28, lo8(-\n_frame) +.endif +#endif /* HAVE SPH */ + ldi r30, \n_pushed + XJMP __epilogue_restores__ + ((18 - (\n_pushed)) * 2) +.endm + +#endif /* AVR_TINY */ + +#else /* Assembler */ +/*****************************************************************/ +/* Space for C/C++ only Stuff */ +/*****************************************************************/ +#endif /* Assembler */ + +/*****************************************************************/ +/* Space for Generic Stuff (Assembler, C, C++) */ +/*****************************************************************/ + +#ifdef __AVR_PM_BASE_ADDRESS__ + /* + Devices with a linear address space: Flash memory is seen in the + RAM address space at an offset of __AVR_PM_BASE_ADDRESS__ and can + be accessed by LD*. This is the case for devices like ATtiny40 + (avrtiny) or ATtiny1616 and ATmega4808 (avrxmega3). The default + linker script locates .rodata in the .text output section and + at the required offset. + */ + #define RODATA_SECTION .rodata.asm + #define USE_LD 1 + #define USE_LPM 0 +#else /* PM_BASE_ADDRESS */ + /* + No linear address space. As .rodata is located in RAM, we have to + use .progmem.data (located in flash) and LPM to read the data. + This will also work for devices from avrxmega3. + */ + #define RODATA_SECTION .progmem.data.asm + #define USE_LD 0 + #define USE_LPM 1 +#endif /* PM_BASE_ADDRESS */ + +#endif /* target AVR */ +#endif /* ASM_DEFS_H */ Property changes on: libgcc/config/avr/libf7/asm-defs.h ___________________________________________________________________ Added: svn:eol-style ## -0,0 +1 ## +native \ No newline at end of property Added: svn:mime-type ## -0,0 +1 ## +text/plain \ No newline at end of property Index: libgcc/config/avr/libf7/f7-renames.h =================================================================== --- libgcc/config/avr/libf7/f7-renames.h (nonexistent) +++ libgcc/config/avr/libf7/f7-renames.h (working copy) @@ -0,0 +1,234 @@ +/* + Auto-generated file, do not change by hand. + + Generated by: f7renames.sh. + Generated using: F7_PREFIX = __f7_ from t-libf7. + F7F, F7F_cst, F7F_asm from libf7-common.mk. + Included by: libf7.h. + Used by: libf7.c, libf7.h, libf7-asm.sx, f7-wraps.h. +*/ + +#ifndef F7_RENAMES_H +#define F7_RENAMES_H + +#define F7_(name) __f7_##name +#define F7P __f7_ + +/* Renames for libf7.c, libf7.h. */ + +#define f7_fabs __f7_fabs +#define f7_neg __f7_neg +#define f7_add __f7_add +#define f7_sub __f7_sub +#define f7_addsub __f7_addsub +#define f7_div __f7_div +#define f7_div1 __f7_div1 +#define f7_divx __f7_divx +#define f7_fmod __f7_fmod +#define f7_sqrt __f7_sqrt +#define f7_cbrt __f7_cbrt +#define f7_square __f7_square +#define f7_mul __f7_mul +#define f7_mulx __f7_mulx +#define f7_madd_msub __f7_madd_msub +#define f7_madd __f7_madd +#define f7_msub __f7_msub +#define f7_hypot __f7_hypot +#define f7_Ineg __f7_Ineg +#define f7_Iadd __f7_Iadd +#define f7_Isub __f7_Isub +#define f7_Imul __f7_Imul +#define f7_Idiv __f7_Idiv +#define f7_IRsub __f7_IRsub +#define f7_Isquare __f7_Isquare +#define f7_Ildexp __f7_Ildexp +#define f7_Isqrt __f7_Isqrt +#define f7_le __f7_le +#define f7_lt __f7_lt +#define f7_gt __f7_gt +#define f7_ge __f7_ge +#define f7_ne __f7_ne +#define f7_eq __f7_eq +#define f7_cmp __f7_cmp +#define f7_cmp_abs __f7_cmp_abs +#define f7_ordered __f7_ordered +#define f7_unordered __f7_unordered +#define f7_cmp_unordered __f7_cmp_unordered +#define f7_lt_impl __f7_lt_impl +#define f7_gt_impl __f7_gt_impl +#define f7_le_impl __f7_le_impl +#define f7_ge_impl __f7_ge_impl +#define f7_eq_impl __f7_eq_impl +#define f7_ne_impl __f7_ne_impl +#define f7_unord_impl __f7_unord_impl +#define f7_lrint __f7_lrint +#define f7_ldexp __f7_ldexp +#define f7_frexp __f7_frexp +#define f7_exp __f7_exp +#define f7_logx __f7_logx +#define f7_log __f7_log +#define f7_log10 __f7_log10 +#define f7_log2 __f7_log2 +#define f7_minmax __f7_minmax +#define f7_fmax __f7_fmax +#define f7_fmin __f7_fmin +#define f7_floor __f7_floor +#define f7_ceil __f7_ceil +#define f7_round __f7_round +#define f7_lround __f7_lround +#define f7_trunc __f7_trunc +#define f7_truncx __f7_truncx +#define f7_horner __f7_horner +#define f7_pow10 __f7_pow10 +#define f7_exp10 __f7_exp10 +#define f7_pow __f7_pow +#define f7_powi __f7_powi +#define f7_sin __f7_sin +#define f7_cos __f7_cos +#define f7_tan __f7_tan +#define f7_cotan __f7_cotan +#define f7_sincos __f7_sincos +#define f7_sinh __f7_sinh +#define f7_cosh __f7_cosh +#define f7_tanh __f7_tanh +#define f7_sinhcosh __f7_sinhcosh +#define f7_asinacos __f7_asinacos +#define f7_asin __f7_asin +#define f7_acos __f7_acos +#define f7_atan __f7_atan +#define f7_atan2 __f7_atan2 +#define f7_mul_noround __f7_mul_noround +#define f7_sqrt16_round __f7_sqrt16_round +#define f7_sqrt16_floor __f7_sqrt16_floor +#define f7_clr_mant_lsbs __f7_clr_mant_lsbs +#define f7_abscmp_msb_ge __f7_abscmp_msb_ge +#define f7_lshrdi3 __f7_lshrdi3 +#define f7_ashldi3 __f7_ashldi3 +#define f7_assert __f7_assert +#define f7_classify __f7_classify +#define f7_class_inf __f7_class_inf +#define f7_class_nan __f7_class_nan +#define f7_class_number __f7_class_number +#define f7_class_zero __f7_class_zero +#define f7_class_nonzero __f7_class_nonzero +#define f7_class_sign __f7_class_sign +#define f7_signbit __f7_signbit +#define f7_set_sign __f7_set_sign +#define f7_set_nan __f7_set_nan +#define f7_set_inf __f7_set_inf +#define f7_is_inf __f7_is_inf +#define f7_is_nan __f7_is_nan +#define f7_is_number __f7_is_number +#define f7_is_zero __f7_is_zero +#define f7_is_nonzero __f7_is_nonzero +#define f7_clr __f7_clr +#define f7_copy __f7_copy +#define f7_copy_P __f7_copy_P +#define f7_copy_mant __f7_copy_mant +#define f7_msbit __f7_msbit +#define f7_is0 __f7_is0 +#define f7_cmp_mant __f7_cmp_mant +#define f7_store_expo __f7_store_expo +#define f7_abs __f7_abs +#define f7_set_s64 __f7_set_s64 +#define f7_set_s32 __f7_set_s32 +#define f7_set_s16 __f7_set_s16 +#define f7_set_s16_impl __f7_set_s16_impl +#define f7_set_u16_worker __f7_set_u16_worker +#define f7_set_u64 __f7_set_u64 +#define f7_set_u32 __f7_set_u32 +#define f7_set_u16 __f7_set_u16 +#define f7_set_u16_impl __f7_set_u16_impl +#define f7_set_float __f7_set_float +#define f7_set_pdouble __f7_set_pdouble +#define f7_set_double_impl __f7_set_double_impl +#define f7_set_double __f7_set_double +#define f7_init_impl __f7_init_impl +#define f7_init __f7_init +#define f7_get_s16 __f7_get_s16 +#define f7_get_s32 __f7_get_s32 +#define f7_get_s64 __f7_get_s64 +#define f7_get_float __f7_get_float +#define f7_get_u16 __f7_get_u16 +#define f7_get_u32 __f7_get_u32 +#define f7_get_u64 __f7_get_u64 +#define f7_get_double __f7_get_double +#define f7_set_eps __f7_set_eps +#define f7_set_1pow2 __f7_set_1pow2 +#define f7_min __f7_min +#define f7_max __f7_max +#define f7_exp10 __f7_exp10 +#define f7_floatunsidf __f7_floatunsidf +#define f7_floatsidf __f7_floatsidf +#define f7_extendsfdf2 __f7_extendsfdf2 +#define f7_fixdfsi __f7_fixdfsi +#define f7_fixdfdi __f7_fixdfdi +#define f7_fixunsdfdi __f7_fixunsdfdi +#define f7_fixunsdfsi __f7_fixunsdfsi +#define f7_truncdfsf2 __f7_truncdfsf2 +#define f7_le_impl __f7_le_impl +#define f7_lt_impl __f7_lt_impl +#define f7_gt_impl __f7_gt_impl +#define f7_ge_impl __f7_ge_impl +#define f7_ne_impl __f7_ne_impl +#define f7_eq_impl __f7_eq_impl +#define f7_unord_impl __f7_unord_impl + +/* Renames for libf7.c, libf7.h. */ + +#define f7_const_1 __f7_const_1 +#define f7_const_1_P __f7_const_1_P +#define f7_const_2 __f7_const_2 +#define f7_const_2_P __f7_const_2_P +#define f7_const_1_2 __f7_const_1_2 +#define f7_const_1_2_P __f7_const_1_2_P +#define f7_const_1_3 __f7_const_1_3 +#define f7_const_1_3_P __f7_const_1_3_P +#define f7_const_m1 __f7_const_m1 +#define f7_const_m1_P __f7_const_m1_P +#define f7_const_pi __f7_const_pi +#define f7_const_pi_P __f7_const_pi_P +#define f7_const_ln2 __f7_const_ln2 +#define f7_const_ln2_P __f7_const_ln2_P +#define f7_const_ln10 __f7_const_ln10 +#define f7_const_ln10_P __f7_const_ln10_P +#define f7_const_1_ln2 __f7_const_1_ln2 +#define f7_const_1_ln2_P __f7_const_1_ln2_P +#define f7_const_1_ln10 __f7_const_1_ln10 +#define f7_const_1_ln10_P __f7_const_1_ln10_P +#define f7_const_sqrt2 __f7_const_sqrt2 +#define f7_const_sqrt2_P __f7_const_sqrt2_P + +/* Renames for libf7-asm.sx, f7-wraps.h. */ + +#define f7_classify_asm __f7_classify_asm +#define f7_store_expo_asm __f7_store_expo_asm +#define f7_clr_asm __f7_clr_asm +#define f7_copy_asm __f7_copy_asm +#define f7_copy_P_asm __f7_copy_P_asm +#define f7_copy_mant_asm __f7_copy_mant_asm +#define f7_cmp_mant_asm __f7_cmp_mant_asm +#define f7_normalize_asm __f7_normalize_asm +#define f7_store_expo_asm __f7_store_expo_asm +#define f7_set_u64_asm __f7_set_u64_asm +#define f7_set_s64_asm __f7_set_s64_asm +#define f7_addsub_mant_scaled_asm __f7_addsub_mant_scaled_asm +#define f7_mul_mant_asm __f7_mul_mant_asm +#define f7_to_integer_asm __f7_to_integer_asm +#define f7_to_unsigned_asm __f7_to_unsigned_asm +#define f7_clr_mant_lsbs_asm __f7_clr_mant_lsbs_asm +#define f7_div_asm __f7_div_asm +#define f7_sqrt_approx_asm __f7_sqrt_approx_asm +#define f7_sqrt16_round_asm __f7_sqrt16_round_asm +#define f7_sqrt16_floor_asm __f7_sqrt16_floor_asm +#define f7_lshrdi3_asm __f7_lshrdi3_asm +#define f7_ashldi3_asm __f7_ashldi3_asm +#define f7_class_D_asm __f7_class_D_asm +#define f7_call_ddd_asm __f7_call_ddd_asm +#define f7_call_xdd_asm __f7_call_xdd_asm +#define f7_call_ddx_asm __f7_call_ddx_asm +#define f7_call_dd_asm __f7_call_dd_asm +#define f7_call_xd_asm __f7_call_xd_asm +#define f7_call_dx_asm __f7_call_dx_asm + +#endif /* F7_RENAMES_H */ Property changes on: libgcc/config/avr/libf7/f7-renames.h ___________________________________________________________________ Added: svn:eol-style ## -0,0 +1 ## +native \ No newline at end of property Added: svn:mime-type ## -0,0 +1 ## +text/plain \ No newline at end of property Index: libgcc/config/avr/libf7/f7-wraps.h =================================================================== --- libgcc/config/avr/libf7/f7-wraps.h (nonexistent) +++ libgcc/config/avr/libf7/f7-wraps.h (working copy) @@ -0,0 +1,638 @@ +;; Copyright (C) 2019 Free Software Foundation, Inc. +;; +;; This file is part of LIBF7, which is part of GCC. +;; +;; GCC is free software; you can redistribute it and/or modify it under +;; the terms of the GNU General Public License as published by the Free +;; Software Foundation; either version 3, or (at your option) any later +;; version. +;; +;; GCC is distributed in the hope that it will be useful, but WITHOUT ANY +;; WARRANTY; without even the implied warranty of MERCHANTABILITY or +;; FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License +;; for more details. +;; +;; Under Section 7 of GPL version 3, you are granted additional +;; permissions described in the GCC Runtime Library Exception, version +;; 3.1, as published by the Free Software Foundation. +;; +;; You should have received a copy of the GNU General Public License and +;; a copy of the GCC Runtime Library Exception along with this program; +;; see the files COPYING3 and COPYING.RUNTIME respectively. If not, see +;; <http://www.gnu.org/licenses/>. */ + +;; Auto-generated file, do not change by hand. +;; +;; Wrappers for double and long double functions to use functions that +;; operate on f7_t, and get f7_t* and const f7_t*. +;; +;; Generated by: f7wraps.sh +;; Included by : libf7-asm.sx +;; WITH_LIBF7_MATH_FUNCTIONS=1 +;; WITH_LIBF7_MATH_SYMBOLS=1 + +;; Functions that usually live in libgcc: __<name>df3 for <name> in: +;; add sub mul div + +;; double __adddf3 (double, double) ; add +#ifdef F7MOD_D_add_ +_DEFUN __adddf3 + ALIAS __add + .global F7_NAME(add) + ldi ZH, hi8(gs(F7_NAME(add))) + ldi ZL, lo8(gs(F7_NAME(add))) + F7jmp call_ddd +_ENDF __adddf3 +#endif /* F7MOD_D_add_ */ + +;; double __subdf3 (double, double) ; sub +#ifdef F7MOD_D_sub_ +_DEFUN __subdf3 + ALIAS __sub + .global F7_NAME(sub) + ldi ZH, hi8(gs(F7_NAME(sub))) + ldi ZL, lo8(gs(F7_NAME(sub))) + F7jmp call_ddd +_ENDF __subdf3 +#endif /* F7MOD_D_sub_ */ + +;; double __muldf3 (double, double) ; mul +#ifdef F7MOD_D_mul_ +_DEFUN __muldf3 + ALIAS __mul + .global F7_NAME(mul) + ldi ZH, hi8(gs(F7_NAME(mul))) + ldi ZL, lo8(gs(F7_NAME(mul))) + F7jmp call_ddd +_ENDF __muldf3 +#endif /* F7MOD_D_mul_ */ + +;; double __divdf3 (double, double) ; div +#ifdef F7MOD_D_div_ +_DEFUN __divdf3 + ALIAS __div + .global F7_NAME(div) + ldi ZH, hi8(gs(F7_NAME(div))) + ldi ZL, lo8(gs(F7_NAME(div))) + F7jmp call_ddd +_ENDF __divdf3 +#endif /* F7MOD_D_div_ */ + +;; Functions that usually live in libgcc: __<name>df2 for <name> in: +;; le lt ge gt ne eq unord + +;; bool __ledf2 (double, double) ; le +#ifdef F7MOD_D_le_ +_DEFUN __ledf2 + .global F7_NAME(le_impl) + ldi ZH, hi8(gs(F7_NAME(le_impl))) + ldi ZL, lo8(gs(F7_NAME(le_impl))) + F7jmp call_xdd +_ENDF __ledf2 +#endif /* F7MOD_D_le_ */ + +;; bool __ltdf2 (double, double) ; lt +#ifdef F7MOD_D_lt_ +_DEFUN __ltdf2 + .global F7_NAME(lt_impl) + ldi ZH, hi8(gs(F7_NAME(lt_impl))) + ldi ZL, lo8(gs(F7_NAME(lt_impl))) + F7jmp call_xdd +_ENDF __ltdf2 +#endif /* F7MOD_D_lt_ */ + +;; bool __gedf2 (double, double) ; ge +#ifdef F7MOD_D_ge_ +_DEFUN __gedf2 + .global F7_NAME(ge_impl) + ldi ZH, hi8(gs(F7_NAME(ge_impl))) + ldi ZL, lo8(gs(F7_NAME(ge_impl))) + F7jmp call_xdd +_ENDF __gedf2 +#endif /* F7MOD_D_ge_ */ + +;; bool __gtdf2 (double, double) ; gt +#ifdef F7MOD_D_gt_ +_DEFUN __gtdf2 + .global F7_NAME(gt_impl) + ldi ZH, hi8(gs(F7_NAME(gt_impl))) + ldi ZL, lo8(gs(F7_NAME(gt_impl))) + F7jmp call_xdd +_ENDF __gtdf2 +#endif /* F7MOD_D_gt_ */ + +;; bool __nedf2 (double, double) ; ne +#ifdef F7MOD_D_ne_ +_DEFUN __nedf2 + .global F7_NAME(ne_impl) + ldi ZH, hi8(gs(F7_NAME(ne_impl))) + ldi ZL, lo8(gs(F7_NAME(ne_impl))) + F7jmp call_xdd +_ENDF __nedf2 +#endif /* F7MOD_D_ne_ */ + +;; bool __eqdf2 (double, double) ; eq +#ifdef F7MOD_D_eq_ +_DEFUN __eqdf2 + .global F7_NAME(eq_impl) + ldi ZH, hi8(gs(F7_NAME(eq_impl))) + ldi ZL, lo8(gs(F7_NAME(eq_impl))) + F7jmp call_xdd +_ENDF __eqdf2 +#endif /* F7MOD_D_eq_ */ + +;; bool __unorddf2 (double, double) ; unord +#ifdef F7MOD_D_unord_ +_DEFUN __unorddf2 + .global F7_NAME(unord_impl) + ldi ZH, hi8(gs(F7_NAME(unord_impl))) + ldi ZL, lo8(gs(F7_NAME(unord_impl))) + F7jmp call_xdd +_ENDF __unorddf2 +#endif /* F7MOD_D_unord_ */ + +;; Functions that usually live in libgcc: __<name> for <name> in: +;; fixdfsi fixdfdi fixunsdfdi fixunsdfsi truncdfsf2 + +;; type_t __fixdfsi (double) ; fixdfsi +#ifdef F7MOD_D_fixdfsi_ +_DEFUN __fixdfsi + .global F7_NAME(fixdfsi) + ldi ZH, hi8(gs(F7_NAME(fixdfsi))) + ldi ZL, lo8(gs(F7_NAME(fixdfsi))) + F7jmp call_xd +_ENDF __fixdfsi +#endif /* F7MOD_D_fixdfsi_ */ + +;; type_t __fixdfdi (double) ; fixdfdi +#ifdef F7MOD_D_fixdfdi_ +_DEFUN __fixdfdi + .global F7_NAME(fixdfdi) + ldi ZH, hi8(gs(F7_NAME(fixdfdi))) + ldi ZL, lo8(gs(F7_NAME(fixdfdi))) + F7jmp call_xd +_ENDF __fixdfdi +#endif /* F7MOD_D_fixdfdi_ */ + +;; type_t __fixunsdfdi (double) ; fixunsdfdi +#ifdef F7MOD_D_fixunsdfdi_ +_DEFUN __fixunsdfdi + .global F7_NAME(fixunsdfdi) + ldi ZH, hi8(gs(F7_NAME(fixunsdfdi))) + ldi ZL, lo8(gs(F7_NAME(fixunsdfdi))) + F7jmp call_xd +_ENDF __fixunsdfdi +#endif /* F7MOD_D_fixunsdfdi_ */ + +;; type_t __fixunsdfsi (double) ; fixunsdfsi +#ifdef F7MOD_D_fixunsdfsi_ +_DEFUN __fixunsdfsi + .global F7_NAME(fixunsdfsi) + ldi ZH, hi8(gs(F7_NAME(fixunsdfsi))) + ldi ZL, lo8(gs(F7_NAME(fixunsdfsi))) + F7jmp call_xd +_ENDF __fixunsdfsi +#endif /* F7MOD_D_fixunsdfsi_ */ + +;; type_t __truncdfsf2 (double) ; truncdfsf2 +#ifdef F7MOD_D_truncdfsf2_ +_DEFUN __truncdfsf2 + .global F7_NAME(truncdfsf2) + ldi ZH, hi8(gs(F7_NAME(truncdfsf2))) + ldi ZL, lo8(gs(F7_NAME(truncdfsf2))) + F7jmp call_xd +_ENDF __truncdfsf2 +#endif /* F7MOD_D_truncdfsf2_ */ + +;; Functions that usually live in libgcc: __<name> for <name> in: +;; floatunsidf floatsidf extendsfdf2 + +;; double __floatunsidf (type_t) ; floatunsidf +#ifdef F7MOD_D_floatunsidf_ +_DEFUN __floatunsidf + .global F7_NAME(floatunsidf) + ldi ZH, hi8(gs(F7_NAME(floatunsidf))) + ldi ZL, lo8(gs(F7_NAME(floatunsidf))) + F7jmp call_dx +_ENDF __floatunsidf +#endif /* F7MOD_D_floatunsidf_ */ + +;; double __floatsidf (type_t) ; floatsidf +#ifdef F7MOD_D_floatsidf_ +_DEFUN __floatsidf + .global F7_NAME(floatsidf) + ldi ZH, hi8(gs(F7_NAME(floatsidf))) + ldi ZL, lo8(gs(F7_NAME(floatsidf))) + F7jmp call_dx +_ENDF __floatsidf +#endif /* F7MOD_D_floatsidf_ */ + +;; double __extendsfdf2 (type_t) ; extendsfdf2 +#ifdef F7MOD_D_extendsfdf2_ +_DEFUN __extendsfdf2 + .global F7_NAME(extendsfdf2) + ldi ZH, hi8(gs(F7_NAME(extendsfdf2))) + ldi ZL, lo8(gs(F7_NAME(extendsfdf2))) + F7jmp call_dx +_ENDF __extendsfdf2 +#endif /* F7MOD_D_extendsfdf2_ */ + +;; Functions that usually live in libm: Depending on [long] double layout, +;; define <name> and <name>l as weak alias(es) of __<name> for <name> in: +;; pow fmin fmax fmod hypot atan2 + +;; double __pow (double, double) +#ifdef F7MOD_D_pow_ +_DEFUN __pow + DALIAS pow + LALIAS powl + .global F7_NAME(pow) + ldi ZH, hi8(gs(F7_NAME(pow))) + ldi ZL, lo8(gs(F7_NAME(pow))) + F7jmp call_ddd +_ENDF __pow +#endif /* F7MOD_D_pow_ */ + +;; double __fmin (double, double) +#ifdef F7MOD_D_fmin_ +_DEFUN __fmin + DALIAS fmin + LALIAS fminl + .global F7_NAME(fmin) + ldi ZH, hi8(gs(F7_NAME(fmin))) + ldi ZL, lo8(gs(F7_NAME(fmin))) + F7jmp call_ddd +_ENDF __fmin +#endif /* F7MOD_D_fmin_ */ + +;; double __fmax (double, double) +#ifdef F7MOD_D_fmax_ +_DEFUN __fmax + DALIAS fmax + LALIAS fmaxl + .global F7_NAME(fmax) + ldi ZH, hi8(gs(F7_NAME(fmax))) + ldi ZL, lo8(gs(F7_NAME(fmax))) + F7jmp call_ddd +_ENDF __fmax +#endif /* F7MOD_D_fmax_ */ + +;; double __fmod (double, double) +#ifdef F7MOD_D_fmod_ +_DEFUN __fmod + DALIAS fmod + LALIAS fmodl + .global F7_NAME(fmod) + ldi ZH, hi8(gs(F7_NAME(fmod))) + ldi ZL, lo8(gs(F7_NAME(fmod))) + F7jmp call_ddd +_ENDF __fmod +#endif /* F7MOD_D_fmod_ */ + +;; double __hypot (double, double) +#ifdef F7MOD_D_hypot_ +_DEFUN __hypot + DALIAS hypot + LALIAS hypotl + .global F7_NAME(hypot) + ldi ZH, hi8(gs(F7_NAME(hypot))) + ldi ZL, lo8(gs(F7_NAME(hypot))) + F7jmp call_ddd +_ENDF __hypot +#endif /* F7MOD_D_hypot_ */ + +;; double __atan2 (double, double) +#ifdef F7MOD_D_atan2_ +_DEFUN __atan2 + DALIAS atan2 + LALIAS atan2l + .global F7_NAME(atan2) + ldi ZH, hi8(gs(F7_NAME(atan2))) + ldi ZL, lo8(gs(F7_NAME(atan2))) + F7jmp call_ddd +_ENDF __atan2 +#endif /* F7MOD_D_atan2_ */ + +;; Functions that usually live in libm: Depending on [long] double layout, +;; define <name> and <name>l as weak alias(es) of __<name> for <name> in: +;; ldexp frexp + +;; double __ldexp (double, word_t) +#ifdef F7MOD_D_ldexp_ +_DEFUN __ldexp + DALIAS ldexp + LALIAS ldexpl + .global F7_NAME(ldexp) + ldi ZH, hi8(gs(F7_NAME(ldexp))) + ldi ZL, lo8(gs(F7_NAME(ldexp))) + F7jmp call_ddx +_ENDF __ldexp +#endif /* F7MOD_D_ldexp_ */ + +;; double __frexp (double, word_t) +#ifdef F7MOD_D_frexp_ +_DEFUN __frexp + DALIAS frexp + LALIAS frexpl + .global F7_NAME(frexp) + ldi ZH, hi8(gs(F7_NAME(frexp))) + ldi ZL, lo8(gs(F7_NAME(frexp))) + F7jmp call_ddx +_ENDF __frexp +#endif /* F7MOD_D_frexp_ */ + +;; Functions that usually live in libm: Depending on [long] double layout, +;; define <name> and <name>l as weak alias(es) of __<name> for <name> in: +;; sqrt cbrt exp exp10 pow10 log log10 log2 sin cos tan cotan asin acos atan ceil floor trunc round sinh cosh tanh + +;; double __sqrt (double) +#ifdef F7MOD_D_sqrt_ +_DEFUN __sqrt + DALIAS sqrt + LALIAS sqrtl + .global F7_NAME(sqrt) + ldi ZH, hi8(gs(F7_NAME(sqrt))) + ldi ZL, lo8(gs(F7_NAME(sqrt))) + F7jmp call_dd +_ENDF __sqrt +#endif /* F7MOD_D_sqrt_ */ + +;; double __cbrt (double) +#ifdef F7MOD_D_cbrt_ +_DEFUN __cbrt + DALIAS cbrt + LALIAS cbrtl + .global F7_NAME(cbrt) + ldi ZH, hi8(gs(F7_NAME(cbrt))) + ldi ZL, lo8(gs(F7_NAME(cbrt))) + F7jmp call_dd +_ENDF __cbrt +#endif /* F7MOD_D_cbrt_ */ + +;; double __exp (double) +#ifdef F7MOD_D_exp_ +_DEFUN __exp + DALIAS exp + LALIAS expl + .global F7_NAME(exp) + ldi ZH, hi8(gs(F7_NAME(exp))) + ldi ZL, lo8(gs(F7_NAME(exp))) + F7jmp call_dd +_ENDF __exp +#endif /* F7MOD_D_exp_ */ + +;; double __exp10 (double) +#ifdef F7MOD_D_exp10_ +_DEFUN __exp10 + DALIAS exp10 + LALIAS exp10l + .global F7_NAME(exp10) + ldi ZH, hi8(gs(F7_NAME(exp10))) + ldi ZL, lo8(gs(F7_NAME(exp10))) + F7jmp call_dd +_ENDF __exp10 +#endif /* F7MOD_D_exp10_ */ + +;; double __pow10 (double) +#ifdef F7MOD_D_pow10_ +_DEFUN __pow10 + DALIAS pow10 + LALIAS pow10l + .global F7_NAME(pow10) + ldi ZH, hi8(gs(F7_NAME(pow10))) + ldi ZL, lo8(gs(F7_NAME(pow10))) + F7jmp call_dd +_ENDF __pow10 +#endif /* F7MOD_D_pow10_ */ + +;; double __log (double) +#ifdef F7MOD_D_log_ +_DEFUN __log + DALIAS log + LALIAS logl + .global F7_NAME(log) + ldi ZH, hi8(gs(F7_NAME(log))) + ldi ZL, lo8(gs(F7_NAME(log))) + F7jmp call_dd +_ENDF __log +#endif /* F7MOD_D_log_ */ + +;; double __log10 (double) +#ifdef F7MOD_D_log10_ +_DEFUN __log10 + DALIAS log10 + LALIAS log10l + .global F7_NAME(log10) + ldi ZH, hi8(gs(F7_NAME(log10))) + ldi ZL, lo8(gs(F7_NAME(log10))) + F7jmp call_dd +_ENDF __log10 +#endif /* F7MOD_D_log10_ */ + +;; double __log2 (double) +#ifdef F7MOD_D_log2_ +_DEFUN __log2 + DALIAS log2 + LALIAS log2l + .global F7_NAME(log2) + ldi ZH, hi8(gs(F7_NAME(log2))) + ldi ZL, lo8(gs(F7_NAME(log2))) + F7jmp call_dd +_ENDF __log2 +#endif /* F7MOD_D_log2_ */ + +;; double __sin (double) +#ifdef F7MOD_D_sin_ +_DEFUN __sin + DALIAS sin + LALIAS sinl + .global F7_NAME(sin) + ldi ZH, hi8(gs(F7_NAME(sin))) + ldi ZL, lo8(gs(F7_NAME(sin))) + F7jmp call_dd +_ENDF __sin +#endif /* F7MOD_D_sin_ */ + +;; double __cos (double) +#ifdef F7MOD_D_cos_ +_DEFUN __cos + DALIAS cos + LALIAS cosl + .global F7_NAME(cos) + ldi ZH, hi8(gs(F7_NAME(cos))) + ldi ZL, lo8(gs(F7_NAME(cos))) + F7jmp call_dd +_ENDF __cos +#endif /* F7MOD_D_cos_ */ + +;; double __tan (double) +#ifdef F7MOD_D_tan_ +_DEFUN __tan + DALIAS tan + LALIAS tanl + .global F7_NAME(tan) + ldi ZH, hi8(gs(F7_NAME(tan))) + ldi ZL, lo8(gs(F7_NAME(tan))) + F7jmp call_dd +_ENDF __tan +#endif /* F7MOD_D_tan_ */ + +;; double __cotan (double) +#ifdef F7MOD_D_cotan_ +_DEFUN __cotan + DALIAS cotan + LALIAS cotanl + .global F7_NAME(cotan) + ldi ZH, hi8(gs(F7_NAME(cotan))) + ldi ZL, lo8(gs(F7_NAME(cotan))) + F7jmp call_dd +_ENDF __cotan +#endif /* F7MOD_D_cotan_ */ + +;; double __asin (double) +#ifdef F7MOD_D_asin_ +_DEFUN __asin + DALIAS asin + LALIAS asinl + .global F7_NAME(asin) + ldi ZH, hi8(gs(F7_NAME(asin))) + ldi ZL, lo8(gs(F7_NAME(asin))) + F7jmp call_dd +_ENDF __asin +#endif /* F7MOD_D_asin_ */ + +;; double __acos (double) +#ifdef F7MOD_D_acos_ +_DEFUN __acos + DALIAS acos + LALIAS acosl + .global F7_NAME(acos) + ldi ZH, hi8(gs(F7_NAME(acos))) + ldi ZL, lo8(gs(F7_NAME(acos))) + F7jmp call_dd +_ENDF __acos +#endif /* F7MOD_D_acos_ */ + +;; double __atan (double) +#ifdef F7MOD_D_atan_ +_DEFUN __atan + DALIAS atan + LALIAS atanl + .global F7_NAME(atan) + ldi ZH, hi8(gs(F7_NAME(atan))) + ldi ZL, lo8(gs(F7_NAME(atan))) + F7jmp call_dd +_ENDF __atan +#endif /* F7MOD_D_atan_ */ + +;; double __ceil (double) +#ifdef F7MOD_D_ceil_ +_DEFUN __ceil + DALIAS ceil + LALIAS ceill + .global F7_NAME(ceil) + ldi ZH, hi8(gs(F7_NAME(ceil))) + ldi ZL, lo8(gs(F7_NAME(ceil))) + F7jmp call_dd +_ENDF __ceil +#endif /* F7MOD_D_ceil_ */ + +;; double __floor (double) +#ifdef F7MOD_D_floor_ +_DEFUN __floor + DALIAS floor + LALIAS floorl + .global F7_NAME(floor) + ldi ZH, hi8(gs(F7_NAME(floor))) + ldi ZL, lo8(gs(F7_NAME(floor))) + F7jmp call_dd +_ENDF __floor +#endif /* F7MOD_D_floor_ */ + +;; double __trunc (double) +#ifdef F7MOD_D_trunc_ +_DEFUN __trunc + DALIAS trunc + LALIAS truncl + .global F7_NAME(trunc) + ldi ZH, hi8(gs(F7_NAME(trunc))) + ldi ZL, lo8(gs(F7_NAME(trunc))) + F7jmp call_dd +_ENDF __trunc +#endif /* F7MOD_D_trunc_ */ + +;; double __round (double) +#ifdef F7MOD_D_round_ +_DEFUN __round + DALIAS round + LALIAS roundl + .global F7_NAME(round) + ldi ZH, hi8(gs(F7_NAME(round))) + ldi ZL, lo8(gs(F7_NAME(round))) + F7jmp call_dd +_ENDF __round +#endif /* F7MOD_D_round_ */ + +;; double __sinh (double) +#ifdef F7MOD_D_sinh_ +_DEFUN __sinh + DALIAS sinh + LALIAS sinhl + .global F7_NAME(sinh) + ldi ZH, hi8(gs(F7_NAME(sinh))) + ldi ZL, lo8(gs(F7_NAME(sinh))) + F7jmp call_dd +_ENDF __sinh +#endif /* F7MOD_D_sinh_ */ + +;; double __cosh (double) +#ifdef F7MOD_D_cosh_ +_DEFUN __cosh + DALIAS cosh + LALIAS coshl + .global F7_NAME(cosh) + ldi ZH, hi8(gs(F7_NAME(cosh))) + ldi ZL, lo8(gs(F7_NAME(cosh))) + F7jmp call_dd +_ENDF __cosh +#endif /* F7MOD_D_cosh_ */ + +;; double __tanh (double) +#ifdef F7MOD_D_tanh_ +_DEFUN __tanh + DALIAS tanh + LALIAS tanhl + .global F7_NAME(tanh) + ldi ZH, hi8(gs(F7_NAME(tanh))) + ldi ZL, lo8(gs(F7_NAME(tanh))) + F7jmp call_dd +_ENDF __tanh +#endif /* F7MOD_D_tanh_ */ + +;; Functions that usually live in libm: Depending on [long] double layout, +;; define <name> and <name>l as weak alias(es) of __<name> for <name> in: +;; lrint lround + +;; type_t __lrint (double) +#ifdef F7MOD_D_lrint_ +_DEFUN __lrint + DALIAS lrint + LALIAS lrintl + .global F7_NAME(lrint) + ldi ZH, hi8(gs(F7_NAME(lrint))) + ldi ZL, lo8(gs(F7_NAME(lrint))) + F7jmp call_xd +_ENDF __lrint +#endif /* F7MOD_D_lrint_ */ + +;; type_t __lround (double) +#ifdef F7MOD_D_lround_ +_DEFUN __lround + DALIAS lround + LALIAS lroundl + .global F7_NAME(lround) + ldi ZH, hi8(gs(F7_NAME(lround))) + ldi ZL, lo8(gs(F7_NAME(lround))) + F7jmp call_xd +_ENDF __lround +#endif /* F7MOD_D_lround_ */ Property changes on: libgcc/config/avr/libf7/f7-wraps.h ___________________________________________________________________ Added: svn:eol-style ## -0,0 +1 ## +native \ No newline at end of property Added: svn:mime-type ## -0,0 +1 ## +text/plain \ No newline at end of property Index: libgcc/config/avr/libf7/f7renames.sh =================================================================== --- libgcc/config/avr/libf7/f7renames.sh (nonexistent) +++ libgcc/config/avr/libf7/f7renames.sh (working copy) @@ -0,0 +1,80 @@ +#!/usr/bin/env sh + +# The first command argument tells us which flavour to generate for +# the rest of the command line arguments. + +what=$1 +shift + +# The second command argument is the prefix to prepend to all functions. +# It is defined by F7_PREFIX in $2. + +PRE=$1 +shift + +case ${what} in + head) + cat << EOF +/* + Auto-generated file, do not change by hand. + + Generated by: `basename $0`. + Generated using: F7_PREFIX = ${PRE} from $1. + F7F, F7F_cst, F7F_asm from libf7-common.mk. + Included by: libf7.h. + Used by: libf7.c, libf7.h, libf7-asm.sx, f7-wraps.h. +*/ + +#ifndef F7_RENAMES_H +#define F7_RENAMES_H + +#define F7_(name) ${PRE}##name +#define F7P ${PRE} +EOF + ;; + + c) + if [ x${PRE} != xf7_ ]; then + echo " " + echo "/* Renames for libf7.c, libf7.h. */" + echo " " + for x in $*; do + echo "#define f7_$x ${PRE}$x" + done + fi + ;; + + cst) + if [ x${PRE} != xf7_ ]; then + echo " " + echo "/* Renames for libf7.c, libf7.h. */" + echo " " + for x in $*; do + echo "#define f7_const_${x} ${PRE}const_${x}" + echo "#define f7_const_${x}_P ${PRE}const_${x}_P" + done + fi + ;; + + asm) + if [ x${PRE} != xf7_ ]; then + echo " " + echo "/* Renames for libf7-asm.sx, f7-wraps.h. */" + echo " " + for x in $*; do + echo "#define f7_${x}_asm ${PRE}${x}_asm" + done + fi + ;; + + tail) + cat << EOF + +#endif /* F7_RENAMES_H */ +EOF + ;; + + *) + exit 1 + ;; +esac Property changes on: libgcc/config/avr/libf7/f7renames.sh ___________________________________________________________________ Added: svn:eol-style ## -0,0 +1 ## +native \ No newline at end of property Added: svn:executable ## -0,0 +1 ## +* \ No newline at end of property Added: svn:mime-type ## -0,0 +1 ## +text/plain \ No newline at end of property Index: libgcc/config/avr/libf7/f7wraps.sh =================================================================== --- libgcc/config/avr/libf7/f7wraps.sh (nonexistent) +++ libgcc/config/avr/libf7/f7wraps.sh (working copy) @@ -0,0 +1,244 @@ +#!/usr/bin/env sh + +# The first command argument $1 tells us which flavour to generate for +# the rest of the command line arguments. + +what=$1 +shift + +if [ "x$*" = "x" ]; then + none="(none)" +fi + +case ${what} in + header) + cat << EOF +;; Copyright (C) 2019 Free Software Foundation, Inc. +;; +;; This file is part of LIBF7, which is part of GCC. +;; +;; GCC is free software; you can redistribute it and/or modify it under +;; the terms of the GNU General Public License as published by the Free +;; Software Foundation; either version 3, or (at your option) any later +;; version. +;; +;; GCC is distributed in the hope that it will be useful, but WITHOUT ANY +;; WARRANTY; without even the implied warranty of MERCHANTABILITY or +;; FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License +;; for more details. +;; +;; Under Section 7 of GPL version 3, you are granted additional +;; permissions described in the GCC Runtime Library Exception, version +;; 3.1, as published by the Free Software Foundation. +;; +;; You should have received a copy of the GNU General Public License and +;; a copy of the GCC Runtime Library Exception along with this program; +;; see the files COPYING3 and COPYING.RUNTIME respectively. If not, see +;; <http://www.gnu.org/licenses/>. */ + +;; Auto-generated file, do not change by hand. +;; +;; Wrappers for double and long double functions to use functions that +;; operate on f7_t, and get f7_t* and const f7_t*. +;; +;; Generated by: `basename $0` +;; Included by : libf7-asm.sx +EOF + for n in $*; do + echo ";; $n" + done + ;; + + xd_libgcc) + cat << EOF + +;; Functions that usually live in libgcc: __<name> for <name> in: +;; $*${none} +EOF + for n in $*; do + cat << EOF + +;; type_t __${n} (double) ; $n +#ifdef F7MOD_D_${n}_ +_DEFUN __${n} + .global F7_NAME($n) + ldi ZH, hi8(gs(F7_NAME($n))) + ldi ZL, lo8(gs(F7_NAME($n))) + F7jmp call_xd +_ENDF __${n} +#endif /* F7MOD_D_${n}_ */ +EOF + done + ;; + + dx_libgcc) + cat << EOF + +;; Functions that usually live in libgcc: __<name> for <name> in: +;; $*${none} +EOF + for n in $*; do + cat << EOF + +;; double __${n} (type_t) ; $n +#ifdef F7MOD_D_${n}_ +_DEFUN __${n} + .global F7_NAME($n) + ldi ZH, hi8(gs(F7_NAME($n))) + ldi ZL, lo8(gs(F7_NAME($n))) + F7jmp call_dx +_ENDF __${n} +#endif /* F7MOD_D_${n}_ */ +EOF + done + ;; + + ddd_libgcc) + cat << EOF + +;; Functions that usually live in libgcc: __<name>df3 for <name> in: +;; $*${none} +EOF + for n in $*; do + cat << EOF + +;; double __${n}df3 (double, double) ; $n +#ifdef F7MOD_D_${n}_ +_DEFUN __${n}df3 + ALIAS __$n + .global F7_NAME($n) + ldi ZH, hi8(gs(F7_NAME($n))) + ldi ZL, lo8(gs(F7_NAME($n))) + F7jmp call_ddd +_ENDF __${n}df3 +#endif /* F7MOD_D_${n}_ */ +EOF + done + ;; + + xdd_libgcc_cmp) + cat << EOF + +;; Functions that usually live in libgcc: __<name>df2 for <name> in: +;; $*${none} +EOF + for n in $*; do + cat << EOF + +;; bool __${n}df2 (double, double) ; $n +#ifdef F7MOD_D_${n}_ +_DEFUN __${n}df2 + .global F7_NAME(${n}_impl) + ldi ZH, hi8(gs(F7_NAME(${n}_impl))) + ldi ZL, lo8(gs(F7_NAME(${n}_impl))) + F7jmp call_xdd +_ENDF __${n}df2 +#endif /* F7MOD_D_${n}_ */ +EOF + done + ;; + + dd_math) + cat << EOF + +;; Functions that usually live in libm: Depending on [long] double layout, +;; define <name> and <name>l as weak alias(es) of __<name> for <name> in: +;; $*${none} +EOF + for n in $*; do + cat << EOF + +;; double __${n} (double) +#ifdef F7MOD_D_${n}_ +_DEFUN __${n} + DALIAS $n + LALIAS ${n}l + .global F7_NAME($n) + ldi ZH, hi8(gs(F7_NAME($n))) + ldi ZL, lo8(gs(F7_NAME($n))) + F7jmp call_dd +_ENDF __${n} +#endif /* F7MOD_D_${n}_ */ +EOF + done + ;; + + xd_math) + cat << EOF + +;; Functions that usually live in libm: Depending on [long] double layout, +;; define <name> and <name>l as weak alias(es) of __<name> for <name> in: +;; $*${none} +EOF + for n in $*; do + cat << EOF + +;; type_t __${n} (double) +#ifdef F7MOD_D_${n}_ +_DEFUN __${n} + DALIAS $n + LALIAS ${n}l + .global F7_NAME($n) + ldi ZH, hi8(gs(F7_NAME($n))) + ldi ZL, lo8(gs(F7_NAME($n))) + F7jmp call_xd +_ENDF __${n} +#endif /* F7MOD_D_${n}_ */ +EOF + done + ;; + + ddd_math) + cat << EOF + +;; Functions that usually live in libm: Depending on [long] double layout, +;; define <name> and <name>l as weak alias(es) of __<name> for <name> in: +;; $*${none} +EOF + for n in $*; do + cat << EOF + +;; double __${n} (double, double) +#ifdef F7MOD_D_${n}_ +_DEFUN __${n} + DALIAS $n + LALIAS ${n}l + .global F7_NAME($n) + ldi ZH, hi8(gs(F7_NAME($n))) + ldi ZL, lo8(gs(F7_NAME($n))) + F7jmp call_ddd +_ENDF __${n} +#endif /* F7MOD_D_${n}_ */ +EOF + done + ;; + + ddx_math) + cat << EOF + +;; Functions that usually live in libm: Depending on [long] double layout, +;; define <name> and <name>l as weak alias(es) of __<name> for <name> in: +;; $*${none} +EOF + for n in $*; do + cat << EOF + +;; double __${n} (double, word_t) +#ifdef F7MOD_D_${n}_ +_DEFUN __${n} + DALIAS $n + LALIAS ${n}l + .global F7_NAME($n) + ldi ZH, hi8(gs(F7_NAME($n))) + ldi ZL, lo8(gs(F7_NAME($n))) + F7jmp call_ddx +_ENDF __${n} +#endif /* F7MOD_D_${n}_ */ +EOF + done + ;; + + *) + exit 1 + ;; +esac Property changes on: libgcc/config/avr/libf7/f7wraps.sh ___________________________________________________________________ Added: svn:eol-style ## -0,0 +1 ## +native \ No newline at end of property Added: svn:executable ## -0,0 +1 ## +* \ No newline at end of property Added: svn:mime-type ## -0,0 +1 ## +text/plain \ No newline at end of property Index: libgcc/config/avr/libf7/libf7-array.def =================================================================== --- libgcc/config/avr/libf7/libf7-array.def (nonexistent) +++ libgcc/config/avr/libf7/libf7-array.def (working copy) @@ -0,0 +1,45 @@ +/* Copyright (C) 2019 Free Software Foundation, Inc. + + This file is part of LIBF7, which is part of GCC. + + GCC is free software; you can redistribute it and/or modify it under + the terms of the GNU General Public License as published by the Free + Software Foundation; either version 3, or (at your option) any later + version. + + GCC is distributed in the hope that it will be useful, but WITHOUT ANY + WARRANTY; without even the implied warranty of MERCHANTABILITY or + FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + for more details. + + Under Section 7 of GPL version 3, you are granted additional + permissions described in the GCC Runtime Library Exception, version + 3.1, as published by the Free Software Foundation. + + You should have received a copy of the GNU General Public License and + a copy of the GCC Runtime Library Exception along with this program; + see the files COPYING3 and COPYING.RUNTIME respectively. If not, see + <http://www.gnu.org/licenses/>. */ + +static const F7_PGMSPACE f7_t ARRAY_NAME[] = +{ + #define F7_CONST_DEF(NAME, FLAGS, M6, M5, M4, M3, M2, M1, M0, EXPO) \ + { .flags = FLAGS, .mant = { M0, M1, M2, M3, M4, M5, M6 }, .expo = EXPO }, + #include "libf7-const.def" + #undef F7_CONST_DEF +}; + +// static const uint8_t n_ARRAY_NAME = <Entries in ARRAY_NAME[]>. + +#define F7_n_NAME2(X) n_##X +#define F7_n_NAME1(X) F7_n_NAME2(X) + +F7_UNUSED static const uint8_t F7_n_NAME1 (ARRAY_NAME) = + #define F7_CONST_DEF(NAME, FLAGS, M6, M5, M4, M3, M2, M1, M0, EXPO) \ + + 1 + #include "libf7-const.def" + #undef F7_CONST_DEF +; + +#undef F7_n_NAME1 +#undef F7_n_NAME2 Property changes on: libgcc/config/avr/libf7/libf7-array.def ___________________________________________________________________ Added: svn:eol-style ## -0,0 +1 ## +native \ No newline at end of property Added: svn:mime-type ## -0,0 +1 ## +text/plain \ No newline at end of property Index: libgcc/config/avr/libf7/libf7-asm-object.mk =================================================================== --- libgcc/config/avr/libf7/libf7-asm-object.mk (nonexistent) +++ libgcc/config/avr/libf7/libf7-asm-object.mk (working copy) @@ -0,0 +1,20 @@ +# This file is included several times in a row, once for each element of +# $(iter-items). On each inclusion, we advance $o to the next element. +# $(iter-labels) is also advanced. +# This works similar to $(srcdir)/siditi-object.mk. + +o := $(firstword $(iter-items)) +iter-items := $(filter-out $o,$(iter-items)) + +$o-label := $(firstword $(iter-labels)) +iter-labels := $(wordlist 2,$(words $(iter-labels)),$(iter-labels)) + +f7_asm_$o$(objext): f7_asm_%$(objext): $(libf7)/libf7-asm.sx + $(gcc_compile) -DF7MOD_$($*-label)_ $(F7_ASM_FLAGS) \ + -c $< + +ifeq ($(enable_shared),yes) +f7_asm_$(o)_s$(objext): f7_asm_%_s$(objext): $(libf7)/libf7-asm.sx + $(gcc_s_compile) -DF7MOD_$($*-label)_ $(F7_ASM_FLAGS) \ + -c $< +endif Property changes on: libgcc/config/avr/libf7/libf7-asm-object.mk ___________________________________________________________________ Added: svn:eol-style ## -0,0 +1 ## +native \ No newline at end of property Added: svn:mime-type ## -0,0 +1 ## +text/plain \ No newline at end of property Index: libgcc/config/avr/libf7/libf7-asm.sx =================================================================== --- libgcc/config/avr/libf7/libf7-asm.sx (nonexistent) +++ libgcc/config/avr/libf7/libf7-asm.sx (working copy) @@ -0,0 +1,1664 @@ +;; Copyright (C) 2019 Free Software Foundation, Inc. +;; +;; This file is part of LIBF7, which is part of GCC. +;; +;; GCC is free software; you can redistribute it and/or modify it under +;; the terms of the GNU General Public License as published by the Free +;; Software Foundation; either version 3, or (at your option) any later +;; version. +;; +;; GCC is distributed in the hope that it will be useful, but WITHOUT ANY +;; WARRANTY; without even the implied warranty of MERCHANTABILITY or +;; FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License +;; for more details. +;; +;; Under Section 7 of GPL version 3, you are granted additional +;; permissions described in the GCC Runtime Library Exception, version +;; 3.1, as published by the Free Software Foundation. +;; +;; You should have received a copy of the GNU General Public License and +;; a copy of the GCC Runtime Library Exception along with this program; +;; see the files COPYING3 and COPYING.RUNTIME respectively. If not, see +;; <http://www.gnu.org/licenses/>. */ + +#ifndef __AVR_TINY__ + +#define ASM_DEFS_HAVE_DEFUN + +#include "asm-defs.h" +#include "libf7.h" + +#define ZERO __zero_reg__ +#define TMP __tmp_reg__ + +#define F7(name) F7_(name##_asm) + +.macro F7call name + .global F7(\name\()) + XCALL F7(\name\()) +.endm + +.macro F7jmp name + .global F7(\name\()) + XJMP F7(\name\()) +.endm + +;; Just for visibility in disassembly. +.macro LLL name + .global LLL.\name + LLL.\name: + nop +.endm + +.macro DEFUN name + .section .text.libf7.asm.\name, "ax", @progbits + .global F7(\name\()) + .func F7(\name\()) + F7(\name\()) : +.endm + +.macro ENDF name + .size F7(\name\()), . - F7(\name\()) + .endfunc +.endm + +.macro LABEL name + .global F7(\name\()) + F7(\name\()) : +.endm + +.macro _DEFUN name + .section .text.libf7.asm.\name, "ax", @progbits + .weak \name + .type \name, @function + \name : +.endm + +.macro _ENDF name + .size \name, . - \name +.endm + +.macro _LABEL name + .weak \name + .type \name, @function + \name : +.endm + +#define F7_NAME(X) F7_(X) + +;; Make a weak alias. +.macro ALIAS sym + .weak \sym + .type \sym, @function + \sym: +.endm + +;; Make a weak alias if double is 64 bits wide. +.macro DALIAS sym +#if defined (WITH_LIBF7_MATH_SYMBOLS) && __SIZEOF_DOUBLE__ == 8 +ALIAS \sym +#endif +.endm + +;; Make a weak alias if long double is 64 bits wide. +.macro LALIAS sym +#if defined (WITH_LIBF7_MATH_SYMBOLS) && __SIZEOF_LONG_DOUBLE__ == 8 +ALIAS \sym +#endif +.endm + +#define Off 1 +#define Expo (Off + F7_MANT_BYTES) + +#ifdef F7MOD_classify_ +;; r24 = classify (*Z) +;; NaN -> F7_FLAG_nan +;; INF -> F7_FLAG_inf [ | F7_FLAG_sign ] +;; ==0 -> F7_FLAG_zero +;; ... -> 0 [ | F7_FLAG_sign ] + +;; Clobbers: None (no TMP, no T). +DEFUN classify + + ld r24, Z + lsr r24 + brne .Lnan_or_inf + + ldd r24, Z+6+Off + tst r24 + brpl 0f + sbc r24, r24 + andi r24, F7_FLAG_sign + ret + +0: ldi r24, F7_FLAG_zero + ret + +.Lnan_or_inf: + rol r24 + ret + +ENDF classify +#endif /* F7MOD_classify_ */ + +#ifdef F7MOD_clr_ +DEFUN clr + std Z+0, ZERO + std Z+0+Off, ZERO + std Z+1+Off, ZERO + std Z+2+Off, ZERO + std Z+3+Off, ZERO + std Z+4+Off, ZERO + std Z+5+Off, ZERO + std Z+6+Off, ZERO + std Z+0+Expo, ZERO + std Z+1+Expo, ZERO + ret +ENDF clr + +#endif /* F7MOD_clr_ */ + +#ifdef F7MOD_clz_ +;; The libcc CLZ implementations like __clzsi2 aka. __builtin_clzl are +;; not very well suited for out purpose, so implement our own. + +#define ZBITS r26 +.macro .test.byte reg + or ZERO, \reg + brne .Loop_bit + subi ZBITS, -8 +.endm + +;; R26 = CLZ (uint64_t R18); CLZ (0) = 64. +;; Unchanged: T +DEFUN clzdi2 + clr ZBITS + ;; Catch the common case of normalized .mant for speed-up. + tst r25 + brmi 9f + .test.byte r25 + .test.byte r24 + .test.byte r23 + .test.byte r22 + .test.byte r21 + .test.byte r20 + .test.byte r19 + .test.byte r18 +.Ldone: + clr ZERO +9: ret + +.Loop_bit: + lsl ZERO + brcs .Ldone + inc ZBITS + rjmp .Loop_bit + +ENDF clzdi2 +#undef ZBITS +#endif /* F7MOD_clz_ */ + +#ifdef F7MOD_cmp_mant_ +DEFUN cmp_mant + + adiw X, 6 + Off + ld r24, X $ ldd TMP, Z+6+Off $ SUB r24, TMP + brne .Lunequal + + sbiw X, 6 + ld r24, X+ $ ldd TMP, Z+0+Off $ SUB r24, TMP + ld r24, X+ $ ldd TMP, Z+1+Off $ sbc r24, TMP + ld r24, X+ $ ldd TMP, Z+2+Off $ sbc r24, TMP + ld r24, X+ $ ldd TMP, Z+3+Off $ sbc r24, TMP + ld r24, X+ $ ldd TMP, Z+4+Off $ sbc r24, TMP + ld r24, X+ $ ldd TMP, Z+5+Off $ sbc r24, TMP + ;; MSBs are already known to be equal + breq 9f +.Lunequal: + sbc r24, r24 + sbci r24, -1 +9: sbiw X, 6 + Off + ret +ENDF cmp_mant +#endif /* F7MOD_cmp_mant_ */ + +#define CA 18 +#define C0 CA+1 +#define C1 C0+1 +#define C2 C0+2 +#define C3 C0+3 +#define C4 C0+4 +#define C5 C0+5 +#define C6 C0+6 +#define Carry r16 +#define Flags 18 + +#ifdef F7MOD_store_ +;; Z->flags = CA. +;; Z->mant = C[7]. +DEFUN store_mant.with_flags + st Z, CA + +;; Z->mant = C[7]. +LABEL store_mant + std Z+0+Off, C0 + std Z+1+Off, C1 + std Z+2+Off, C2 + std Z+3+Off, C3 + std Z+4+Off, C4 + std Z+5+Off, C5 + std Z+6+Off, C6 + ret +ENDF store_mant.with_flags +#endif /* F7MOD_store_ */ + +#ifdef F7MOD_load_ +;; CA = Z->flags +;; C[7] = Z->mant +DEFUN load_mant.with_flags + ld CA, Z + skipnext + +;; CA = 0 +;; C[7] = Z->mant +LABEL load_mant.clr_CA +LABEL load_mant.clr_flags + clr CA ; May be skipped + +;; C[7] = Z->mant +LABEL load_mant + ldd C0, Z+0+Off + ldd C1, Z+1+Off + ldd C2, Z+2+Off + ldd C3, Z+3+Off + ldd C4, Z+4+Off + ldd C5, Z+5+Off + ldd C6, Z+6+Off + ret +ENDF load_mant.with_flags +#endif /* F7MOD_load_ */ + +#ifdef F7MOD_copy_ +DEFUN copy + cp XL, ZL + cpc XH, ZH + breq 9f + adiw XL, 10 + adiw ZL, 10 + set + bld ZERO, 1 + bld ZERO, 3 ; ZERO = 0b1010 = 10. +.Loop: + ld TMP, -X + st -Z, TMP + dec ZERO + brne .Loop +9: ret +ENDF copy +#endif /* F7MOD_copy_ */ + +#ifdef F7MOD_copy_P_ +DEFUN copy_P + set + bld ZERO, 1 + bld ZERO, 3 ; ZERO = 0b1010 = 10. +.Loop: +#ifdef __AVR_HAVE_LPMX__ + lpm TMP, Z+ +#else + lpm + adiw Z, 1 +#endif /* Have LPMx */ + st X+, TMP + dec ZERO + brne .Loop + sbiw X, 10 + sbiw Z, 10 + ret +ENDF copy_P +#endif /* F7MOD_copy_P_ */ + +#ifdef F7MOD_copy_mant_ +DEFUN copy_mant + cp XL, ZL + cpc XH, ZH + breq 9f + adiw XL, 1 + adiw ZL, 1 + set + bld ZERO, 3 + dec ZERO ; ZERO = 7 +.Loop: + ld TMP, X+ + st Z+, TMP + dec ZERO + brne .Loop + sbiw XL, 8 + sbiw ZL, 8 +9: ret +ENDF copy_mant +#endif /* F7MOD_copy_mant_ */ + + +#ifdef F7MOD_clr_mant_lsbs_ +DEFUN clr_mant_lsbs + push r16 + mov r16, r20 + wmov XL, r24 + + wmov ZL, r22 + F7call load_mant + + F7call lshrdi3 + + clr CA + + F7call ashldi3 + + pop r16 + + wmov ZL, XL + F7jmp store_mant + +ENDF clr_mant_lsbs +#endif /* F7MOD_clr_mant_lsbs_ */ + + +#ifdef F7MOD_normalize_with_carry_ +;; Z = &f7_t +;; C[] = .mant may be not normalized +;; Carry === r16 = Addend to Z->expo in [-64, 128). +;; Normalize C[], set Flags, and adjust Z->expo. +;; Return CA (after normalization) in TMP. +;; Unchanged: T +#define Addend r17 +#define Zbits r26 +#define expL r26 +#define expH r27 +DEFUN normalize_with_carry + mov Addend, Carry + tst C6 + brmi .Lshift.0 + ;; r26 = CLZ (uint64_t R18) + F7call clzdi2 + cpi Zbits, 64 + breq .Lclr + sub Addend, Zbits + mov r16, Zbits + + F7call ashldi3 + ;; Assert (R25.7 == 1) +.Lshift.0: + mov TMP, CA + ld Flags, Z + + ;; .expo += Addend + ldd expL, Z+0+Expo + ldd expH, Z+1+Expo + ;; Sign-extend Addend + clr r16 + sbrc Addend, 7 + com r16 + + ;; exp += (int8_t) Addend, i.e. sign-extend Addend. + add expL, Addend + adc expH, r16 + brvc .Lnormal + tst r16 + brmi .Lclr + ;; Overflow +#if F7_HAVE_Inf == 1 + ori Flags, F7_FLAG_inf +#else + ldi Flags, F7_FLAG_nan +#endif /* Have Inf */ + ret + +.Lnormal: + std Z+0+Expo, expL + std Z+1+Expo, expH + ret + +.Lclr: + ;; Underflow or Zero. + clr TMP + .global __clr_8 + XJMP __clr_8 + +LABEL normalize.store_with_flags + ;; no rounding + set + skipnext +LABEL normalize.round.store_with_flags + ;; with rounding + clt ; skipped ? +LABEL normalize.maybe_round.store_with_flags + F7call normalize_with_carry + ;; We have: + ;; Z = &f7_t + ;; X = .expo + ;; C[] = .mant + ;; R18 = .flags + ;; TMP = byte below .mant after normalization + ;; T = 1 => no rounding. + brts .Lstore + lsl TMP + adc C0, ZERO + brcc .Lstore + adc C1, ZERO + adc C2, ZERO + adc C3, ZERO + adc C4, ZERO + adc C5, ZERO + adc C6, ZERO + brcc .Lstore + ;; We only come here if C6 overflowed, i.e. C[] is 0 now. + ;; .mant = 1.0 by restoring the MSbit. + ror C6 + ;; .expo += 1 and override the .expo stored during normalize. + adiw expL, 1 + std Z+0+Expo, expL + std Z+1+Expo, expH + +.Lstore: + F7call store_mant.with_flags + + ;; Return the byte below .mant after normalization. + ;; This is only useful without rounding; the caller will know. + mov R24, TMP + ret +ENDF normalize_with_carry +#endif /* F7MOD_normalize_with_carry_ */ + + +#ifdef F7MOD_normalize_ +;; Using above functionality from C. +;; f7_t* normalize (f7_t *cc) +;; Adjusts cc->expo +;; Clears cc->flags +DEFUN normalize + push r17 + push r16 + wmov ZL, r24 + F7call load_mant.clr_CA + clr Carry + st Z, ZERO + F7call normalize.store_with_flags + wmov r24, Z + pop r16 + pop r17 + ret +ENDF normalize +#endif /* F7MOD_normalize_ */ + + +#ifdef F7MOD_store_expo_ +#define Done r24 +#define expLO r24 +#define expHI r25 +;; expo == INT16_MAX => *Z = Inf, return Done = true. +;; expo == INT16_MIN => *Z = 0x0, return Done = true. +;; else => Z->expo = expo, return Done = false. +DEFUN store_expo + cpi expHI, 0x80 + cpc expLO, ZERO + breq .Ltiny + adiw expLO, 1 + brvs .Lhuge + sbiw expLO, 1 + std Z+0+Expo, expLO + std Z+1+Expo, expHI + ldi Done, 0 + ret + +.Lhuge: +#if F7_HAVE_Inf == 1 + ld Done, Z + andi Done, F7_FLAG_sign + ori Done, F7_FLAG_inf +#else + ldi Done, F7_FLAG_nan +#endif /* Have Inf */ + st Z, Done + ldi Done, 1 + ret + +.Ltiny: + ldi Done, 1 + F7jmp clr +ENDF store_expo +#endif /* F7MOD_store_expo_ */ + + +#ifdef F7MOD_set_u64_ +DEFUN set_s64 + set + skipnext + ;; ... +LABEL set_u64 + clt ; Skipped? + wmov Zl, r16 + ;; TMP holds .flags. + clr TMP + brtc .Lnot.negative + + bst C6, 7 + brtc .Lnot.negative + bld TMP, F7_FLAGNO_sign + .global __negdi2 + XCALL __negdi2 + +.Lnot.negative: + st Z, TMP + std Z+0+Expo, ZERO + std Z+1+Expo, ZERO + ldi Carry, 63 + F7call normalize.round.store_with_flags + wmov r24, Z + wmov r16, Z ; Unclobber r16. + ret +ENDF set_s64 +#endif /* F7MOD_set_u64_ */ + + +#ifdef F7MOD_to_integer_ +#define Mask r26 +DEFUN to_integer + wmov ZL, r24 + mov Mask, r22 + + F7call load_mant.with_flags + + sbrc Flags, F7_FLAGNO_nan + rjmp .Lset_0x8000 + + sbrc Flags, F7_FLAGNO_inf + rjmp .Lsaturate + + sbrs C6, 7 + rjmp .Lset_0x0000 + + bst Flags, F7_FLAGNO_sign + ldd r27, Z+0+Expo + ;; Does .expo have bits outside Mask? ... + mov TMP, Mask + com TMP + and TMP, r27 + ldd r27, Z+1+Expo + tst r27 + brmi .Lset_0x0000 ; ...yes: .expo is < 0 => return 0 + or TMP, r27 + brne .Lsaturate.T ; ...yes: .expo > Mask => saturate + + ;; ...no: Shift right to meet .expo = 0. + PUSH r16 + ldd r16, Z+0+Expo + eor r16, Mask + and r16, Mask + clr CA + F7call lshrdi3 + POP r16 + tst C6 + brmi .Lsaturate.T ; > INTxx_MAX => saturate + + rcall .Lround + brmi .Lsaturate.T ; > INTxx_MAX => saturate + + brtc 9f ; >= 0 => return + sbrc Mask, 5 + .global __negdi2 + XJMP __negdi2 + sbrc Mask, 4 + .global __negsi2 + XJMP __negsi2 + neg C6 + neg C5 + sbci C6, 0 +9: ret + +.Lsaturate: + bst Flags, F7_FLAGNO_sign +.Lsaturate.T: + +#if F7_HAVE_Inf + brtc .Lset_0x7fff + ;; -Inf => return 1 + INTxx_MIN + mov ZL, Flags + .global __clr_8 + XCALL __clr_8 + ldi C6, 0x80 + + ldi CA+0, 0x01 + + sbrs Mask, 5 + ldi CA+4, 0x01 + + sbrs Mask, 4 + ldi CA+6, 0x01 + ret + +.Lset_0x7fff: + ;; +Inf => return INTxx_MAX + sec + .global __sbc_8 + XCALL __sbc_8 + ldi C6, 0x7f + ret +#endif /* F7_HAVE_Inf */ + +.Lset_0x8000: + ;; NaN => return INTxx_MIN + .global __clr_8 + XCALL __clr_8 + ldi C6, 0x80 + ret + +.Lset_0x0000: + ;; Small value => return 0x0 + .global __clr_8 + XJMP __clr_8 + +.Lround: + ;; C6.7 is known to be 0 here. + ;; Return N = 1 iff we have to saturate. + cpi Mask, 0xf + breq .Lround16 + cpi Mask, 0x1f + breq .Lround32 + + ;; For now, no rounding in the 64-bit case. This rounding + ;; would have to be integrated into the right-shift. + cln + ret + +.Lround32: + rol C2 + adc C3, ZERO + adc C4, ZERO + rjmp 2f + +.Lround16: + rol C4 +2: adc C5, ZERO + adc C6, ZERO + ret +ENDF to_integer +#endif /* F7MOD_to_integer_ */ + + +#ifdef F7MOD_to_unsigned_ +#define Mask r26 +DEFUN to_unsigned + wmov ZL, r24 + mov Mask, r22 + + F7call load_mant.with_flags + + sbrc Flags, F7_FLAGNO_nan + rjmp .Lset_0xffff + + sbrc Flags, F7_FLAGNO_sign + rjmp .Lset_0x0000 + + sbrc Flags, F7_FLAGNO_inf + rjmp .Lset_0xffff + + sbrs C6, 7 + rjmp .Lset_0x0000 + + ldd r27, Z+0+Expo + ;; Does .expo have bits outside Mask? ... + mov TMP, Mask + com TMP + and TMP, r27 + ldd r27, Z+1+Expo + tst r27 + brmi .Lset_0x0000 ; ...yes: .expo is < 0 => return 0 + or TMP, r27 + brne .Lset_0xffff ; ...yes: .expo > Mask => saturate + + ;; ...no: Shift right to meet .expo = 0. + PUSH r16 + ldd r16, Z+0+Expo + eor r16, Mask + and r16, Mask + clr CA + F7call lshrdi3 + POP r16 + + ;; Rounding + ;; ??? C6.7 is known to be 0 here. + cpi Mask, 0xf + breq .Lround16 + cpi Mask, 0x1f + breq .Lround32 + + ;; For now, no rounding in the 64-bit case. This rounding + ;; would have to be integrated into the right-shift. + ret + +.Lround32: + rol C2 + adc C3, ZERO + adc C4, ZERO + rjmp 2f + +.Lround16: + rol C4 +2: adc C5, ZERO + adc C6, ZERO + brcs .Lset_0xffff ; Rounding overflow => saturate + ret + +.Lset_0xffff: + ;; return UINTxx_MAX + sec + .global __sbc_8 + XJMP __sbc_8 + +.Lset_0x0000: + ;; Small value => return 0x0 + .global __clr_8 + XJMP __clr_8 + +ENDF to_unsigned +#endif /* F7MOD_to_unsigned_ */ + + +#ifdef F7MOD_addsub_mant_scaled_ +;; int8_t f7_addsub_mant_scaled_asm (f7_t *r24, const f7_t *r22, const f7_t 20*, +;; uint8_t r18); +;; R18.0 = 1 : ADD +;; R18.0 = 0 : SUB +;; R18[7..1] : Scale +;; Compute *R24 = *R22 + *R20 >> R18[7..1]. + +#define BA 10 +#define B0 BA+1 +#define B1 B0+1 +#define B2 B0+2 +#define B3 B0+3 +#define B4 B0+4 +#define B5 B0+5 +#define B6 B0+6 + +DEFUN addsub_mant_scaled + do_prologue_saves 10 + + bst r18, 0 ;; ADD ? + lsr r18 + mov r16, r18 + + wmov ZL, r20 + wmov YL, r22 + ;; C[] = bb >> shift + wmov XL, r24 + + F7call load_mant.clr_CA + F7call lshrdi3 + + wmov BA, CA + wmov B1, C1 + wmov B3, C3 + wmov B5, C5 + wmov ZL, YL + F7call load_mant.clr_CA + + wmov ZL, XL + + brts .Ladd + + .global __subdi3 + XCALL __subdi3 + + breq .Lzero + brcc .Lround + ;; C = 1: Can underflow happen at all ? +.Lzero: + F7call clr + rjmp .Lepilogue + +.Ladd: + .global __adddi3 + XCALL __adddi3 + brcc .Lround + ldi Carry, 1 + .global __lshrdi3 + XCALL __lshrdi3 + ori C6, 1 << 7 + skipnext +.Lround: + clr Carry ; skipped? + F7call normalize.round.store_with_flags + +.Lepilogue: + do_epilogue_restores 10 + +ENDF addsub_mant_scaled + +#if !defined (__AVR_HAVE_MOVW__) || !defined (__AVR_HAVE_JMP_CALL__) +DEFUN lshrdi3 + .global __lshrdi3 + XJMP __lshrdi3 +ENDF lshrdi3 +DEFUN ashldi3 + .global __ashldi3 + XJMP __ashldi3 +ENDF ashldi3 +#else + +# Basically just a wrapper around libgcc's __lshrdi3. +DEFUN lshrdi3 + ;; Handle bit 5 of shift offset. + sbrs r16, 5 + rjmp 4f + wmov CA, C3 + wmov C1, C5 + clr C6 $ clr C5 $ wmov C3, C5 +4: + ;; Handle bit 4 of shift offset. + sbrs r16, 4 + rjmp 3f + wmov CA, C1 + wmov C1, C3 + wmov C3, C5 + clr C6 $ clr C5 +3: + ;; Handle bits 3...0 of shift offset. + push r16 + andi r16, 0xf + breq 0f + + .global __lshrdi3 + XCALL __lshrdi3 +0: + pop r16 + ret +ENDF lshrdi3 + +# Basically just a wrapper around libgcc's __ashldi3. +DEFUN ashldi3 + ;; Handle bit 5 of shift offset. + sbrs r16, 5 + rjmp 4f + wmov C5, C1 + wmov C3, CA + clr C2 $ clr C1 $ wmov CA, C1 +4: + ;; Handle bit 4 of shift offset. + sbrs r16, 4 + rjmp 3f + wmov C5, C3 + wmov C3, C1 + wmov C1, CA + clr CA $ clr C0 +3: + ;; Handle bits 3...0 of shift offset. + push r16 + andi r16, 0xf + breq 0f + + .global __ashldi3 + XCALL __ashldi3 +0: + pop r16 + ret +ENDF ashldi3 +#endif /* Small device */ + +#endif /* F7MOD_addsub_mant_scaled_ */ + +#if defined F7MOD_mul_mant_ && defined (__AVR_HAVE_MUL__) + #define A0 11 + #define A1 A0+1 + #define A2 A0+2 + #define A3 A0+3 + #define A4 A0+4 + #define A5 A0+5 + #define A6 A0+6 + + #define TT0 26 + #define TT1 TT0+1 + #define TT2 28 + #define TT3 TT2+1 + + #define BB 10 + +;; R18.0 = 1: No rounding. + +DEFUN mul_mant + do_prologue_saves 10 + bst r18, 0 + push r25 + push r24 + movw ZL, r22 + LDD A0, Z+0+Off + LDD A1, Z+1+Off + LDD A2, Z+2+Off + LDD A3, Z+3+Off + LDD A4, Z+4+Off + LDD A5, Z+5+Off + LDD A6, Z+6+Off + movw ZL, r20 + + ;; 6 * 6 -> 6:5 + ;; 4 * 6 -> 4:3 + ;; 2 * 6 -> 2:1 + ;; 0 * 6 -> 0:a + ldd BB, Z+6+Off + mul A6, BB $ movw C5, r0 + mul A4, BB $ movw C3, r0 + mul A2, BB $ movw C1, r0 + mul A0, BB $ movw CA, r0 + + ;; 5 * 6 -> 5:4 + ;; 3 * 6 -> 3:2 + ;; 1 * 6 -> 1:0 + mul A5, BB $ movw TT2, r0 + mul A3, BB $ movw TT0, r0 + mul A1, BB + ADD C0, r0 $ adc C1, r1 + adc C2, TT0 $ adc C3, TT1 + adc C4, TT2 $ adc C5, TT3 $ clr ZERO + adc C6, ZERO + ;; Done B6 + + ;; 3 * 3 -> 0:a + ;; 4 * 4 -> 2:1 + ;; 5 * 5 -> 4:3 + ldd BB, Z+3+Off $ mul A3, BB $ movw TT0, r0 + ldd BB, Z+4+Off $ mul A4, BB $ movw TT2, r0 + ldd BB, Z+5+Off $ mul A5, BB + + ADD CA, TT0 $ adc C0, TT1 + adc C1, TT2 $ adc C2, TT3 + adc C3, r0 $ adc C4, r1 + brcc .+2 + adiw C5, 1 + + ;; 6 * 5 -> 5:4 + ;; 4 * 5 -> 3:2 + ;; 2 * 5 -> 1:0 + ;; 0 * 5 -> a:- + mul A0, BB + ;; A0 done +#define Atmp A0 + + mov Atmp, r1 + mul A6, BB $ movw TT2, r0 + mul A4, BB $ movw TT0, r0 + mul A2, BB + + ADD CA, Atmp + adc C0, r0 $ adc C1, r1 + adc C2, TT0 $ adc C3, TT1 + adc C4, TT2 $ adc C5, TT3 $ clr ZERO + adc C6, ZERO + + ;; 1 * 5 -> 0:a + ;; 3 * 5 -> 2:1 + ;; 6 * 4 -> 4:3 + mul A1, BB $ movw TT0, r0 + mul A3, BB $ movw TT2, r0 + ldd BB, Z+4+Off + mul A6, BB + + ADD CA, TT0 $ adc C0, TT1 + adc C1, TT2 $ adc C2, TT3 + adc C3, r0 $ adc C4, r1 $ clr ZERO + adc C5, ZERO $ adc C6, ZERO + ;; B5 done + + ;; 6 * 3 -> 3:2 + ;; 6 * 1 -> 1:0 + ;; 4 * 1 -> a:- + mov TT0, A6 $ ldd TMP, Z+3+Off + mov BB, A4 $ ldd Atmp, Z+1+Off + rcall .Lmul.help.3 + + ;; 5 * 4 -> 3:2 + ;; 5 * 2 -> 1:0 + ;; 3 * 2 -> a:- + mov TT0, A5 $ ldd TMP, Z+4+Off + mov BB, A3 $ ldd Atmp, Z+2+Off + rcall .Lmul.help.3 + + ;; 4 * -> 3:2 (=0) + ;; 4 * 3 -> 1:0 + ;; 2 * 3 -> a:- + mov TT0, A4 $ clr TMP + mov BB, A2 $ ldd Atmp, Z+3+Off + rcall .Lmul.help.3 + + ;; 3 * . -> 3:2 (=0) + ;; 3 * 4 -> 1:0 + ;; 1 * 4 -> a:- + mov TT0, A3 $ clr TMP + mov BB, A1 $ ldd Atmp, Z+4+Off + rcall .Lmul.help.3 + + ;; . * ? -> 3:2 (=0) + ;; . * 0 -> 1:0 (=0) + ;; 5 * 0 -> a:- + clr TT0 + mov BB, A5 $ ldd Atmp, Z+0+Off + rcall .Lmul.help.3 + + clr TT3 ;; Asserted by .Lmul.help.2 + ;; 6 * 2 -> 2:1 + ;; 6 * 0 -> 0:a + $ ldd TMP, Z+2+Off + mov BB, A6 ;$ ldd Atmp, Z+0+Off + rcall .Lmul.help.2 + + ;; 5 * 3 -> 2:1 + ;; 5 * 1 -> 0:a + $ ldd TMP, Z+3+Off + mov BB, A5 $ ldd Atmp, Z+1+Off + rcall .Lmul.help.2 + + ;; 4 * . -> 2:1 (=0) + ;; 4 * 2 -> 0:a + $ clr TMP + mov BB, A4 $ ldd Atmp, Z+2+Off + rcall .Lmul.help.2 + + ;; 2 * . -> 2:1 (=0) + ;; 2 * 4 -> 0:a + $ clr TMP + mov BB, A2 $ ldd Atmp, Z+4+Off + rcall .Lmul.help.2 + + ;; Finally... + + pop ZL + pop ZH + ;; The high byte is at least 0x40 and at most 0xfe. + ;; The result has to be left-shifted by one in order to scale it + ;; correctly. + + ldi Carry, 1 + F7call normalize.maybe_round.store_with_flags + + do_epilogue_restores 10 + +;; TT0 * Tmp -> 3:2 +;; TT0 * Atmp -> 1:0 +;; BB * Atmp -> a:- +;; +;; Clobbers : TMP, TT0...TT3. +;; Sets : ZERO = 0. +.Lmul.help.3: + mul TT0, TMP $ movw TT2, r0 + mul TT0, Atmp $ movw TT0, r0 + mul BB, Atmp + + ADD CA, r1 + adc C0, TT0 $ adc C1, TT1 + adc C2, TT2 +.Lmul.help.3.C3: $ adc C3, TT3 $ clr ZERO + adc C4, ZERO $ adc C5, ZERO + adc C6, ZERO + ret + +;; BB * TMP -> 2:1 +;; BB * Atmp -> 0:a +;; +;; Asserts : TT3 = 0 +;; Clobbers : TMP, TT0, TT1. +;; Sets : ZERO = 0. +.Lmul.help.2: + mul BB, TMP $ movw TT0, r0 + mul BB, Atmp + ADD CA, r0 $ adc C0, r1 + adc C1, TT0 $ adc C2, TT1 + rjmp .Lmul.help.3.C3 + +ENDF mul_mant +#endif /* F7MOD_mul_mant_ && MUL */ + + +#if defined (F7MOD_div_) + +;; Dividend is C[] + +;; Divisor +#define A0 9 +#define A1 10 +#define A2 11 +#define A3 12 +#define A4 13 +#define A5 14 +#define A6 15 + +;; Quotient +#define Q0 0 /* === TMP */ +#define Q1 Q0+1 /* === ZERO */ +#define Q2 26 +#define Q3 Q2+1 +#define Q4 28 +#define Q5 Q4+1 +#define Q6 16 +#define Q7 Q6+1 + +#define Cnt CA +#define QBits r8 + +DEFUN div + do_prologue_saves 12 + + ;; Number of bits requested for the quotient. + ;; This is usually 2 + F7_MANT_BITS. + mov QBits, r20 + wmov ZL, r22 + LDD A0, Z+0+Off + LDD A1, Z+1+Off + LDD A2, Z+2+Off + LDD A3, Z+3+Off + LDD A4, Z+4+Off + LDD A5, Z+5+Off + LDD A6, Z+6+Off + wmov ZL, r24 + F7call load_mant + + ;; Clear quotient Q[]. + clr Q0 ; === TMP + ;clr Q1 ; === ZERO + wmov Q2, Q0 + wmov Q4, Q0 + wmov Q6, Q0 + + ;; C[] and A[] are valid mantissae, i.e. their MSBit is set. Therefore, + ;; quotient Q[] will be in [0x0.ff..., 0x0.40...] and to adjust Q[] we + ;; need at most 1 left-shift. Compute F7_MANT_BITS + 2 bits of the + ;; quotient: One bit is used for rounding, and one bit might be consumed + ;; by the mentioned left-shift. + mov Cnt, QBits + rjmp .Loop_start + +.Loop: + ;; Shift dividend. + LSL C0 + rol C1 + rol C2 + rol C3 + rol C4 + rol C5 + rol C6 + brcs .Lfits + ;; Compare dividend against divisor. +.Loop_start: + CP C0, A0 + cpc C1, A1 + cpc C2, A2 + cpc C3, A3 + cpc C4, A4 + cpc C5, A5 + cpc C6, A6 + ;; Shift 0 into quotient. + brlo 1f +.Lfits: + ;; Divisor fits into dividend. + SUB C0, A0 + sbc C1, A1 + sbc C2, A2 + sbc C3, A3 + sbc C4, A4 + sbc C5, A5 + sbc C6, A6 + ;; Shift 1 into quotient. + sec + rol Q0 + skipnext +1: lsl Q0 + rol Q1 + rol Q2 + rol Q3 + rol Q4 + rol Q5 + rol Q6 + rol Q7 + dec Cnt + brne .Loop + + wmov CA, Q0 + wmov C1, Q2 + wmov C3, Q4 + wmov C5, Q6 + clr ZERO + + ldi Carry, 64 + sub Carry, QBits + F7call normalize.round.store_with_flags + + do_epilogue_restores 12 +ENDF div + +#endif /* F7MOD_div_ */ + + +#if defined (F7MOD_sqrt16_) && defined (__AVR_HAVE_MUL__) + +#define Mask C6 +#define Q0 C3 /* = R22 */ +#define Q1 C4 /* = R23 */ + +;; uint16_t R24 = sqrt16_XXX (uint16_t R24); +;; Clobbers: R22, R23, TMP. +;; +;; XXX = floor: Return integral part of square-root of R25:R24 with R25 = 0. +;; Error is in [0, -1 LSB). +;; XXX = round: Return quare-root of R25:R24 rounded to nearest integer. +;; R25 = (Q[] >= 65281) = (Q > 0xff00), i.e. if Q[] is not +;; bigger than 0xff00, then the result fits in 8 bits. +;; Return C = 0 if the result is the same as for XXX = floor, +;; error in [0, -1/2 LSB) +;; Return C = 1 if the result is one higher than for XXX = floor, +;; error in [1/2 LSB, 0). +DEFUN sqrt16_round + set + skipnext + ;; ... +LABEL sqrt16_floor + clt ; Skipped? + movw Q0, r24 + clr C5 + ldi Mask, 1 << 7 + +.Loop_mask: + add C5, Mask + mul C5, C5 + cp Q0, R0 + cpc Q1, R1 + brsh 1f + sub C5, Mask +1: lsr Mask + brne .Loop_mask + + brtc .Ldone ; No rounding => C6 will be 0. + + ;; Rounding: (X + 1/2)^2 = X^2 + X + 1/4, thus probing + ;; for bit -1 is testing Q[] against C5^2 + C5. + mul C5, C5 + add R0, C5 + adc R1, C6 ; Exploit C6 === Mask = 0. + cp R0, Q0 + cpc R1, Q1 + brcc .Ldone + ;; If C5^2 + C5 + 1/4 fits into Q[], then round up and C = 1. + adiw C5, 1 ; Exploit C6 === Mask = 0. + sec + +.Ldone: + clr __zero_reg__ + ret +ENDF sqrt16_round +#undef Mask +#undef Q0 +#undef Q1 +#endif /* F7MOD_sqrt16_ && MUL */ + +#ifdef F7MOD_sqrt_approx_ +DEFUN sqrt_approx + push r17 + push r16 + wmov XL, r24 + wmov ZL, r22 + + ;; C[] = 0. + .global __clr_8 + XCALL __clr_8 + + ldd C5, Z+5+Off + ldd C6, Z+6+Off + + ldd Carry, Z+0+Expo + ldd TMP, Z+1+Expo + wmov ZL, XL + + st Z, ZERO + + asr TMP + ror Carry + std Z+1+Expo, TMP + std Z+0+Expo, Carry + + ;; Re-interpreting our Q-format 1.xx mantissa as Q2.yy, we have to shift + ;; the mantissa to the right by 1. As we need an even exponent, multiply + ;; the mantissa by 2 for odd exponents, i.e. only right-shift if .expo + ;; is even. + + brcs 1f + lsr C6 + ror C5 + +1: + F7call sqrt16_round + + ;; sqrt16_round() returns: C = 0: error in [0, -1/2 LSB). + ;; C = 1: error in [1/2 LSB, 0) + + brcc 2f + ;; Undo the round-up from sqrt16_round(); this will transform to + ;; error in [-1/2 LSB, -1 LSB). + sbiw C5, 1 + ;; Together with the correct bit C4.7, the error is in [0, -1/2 LSB). + ori C4, 1 << 7 + +2: ;; Setting C4.6 adds 1/4 LSB and the error is now in [1/4 LSB, -1/4 LSB) + ;; in either case. + ori C4, 1 << 6 + + ;; ???????????? + ;; sqrt16_round() runs on integers which means that it computes the + ;; square root of mant * 2^14 if we regard mant as Q-format 2.yy, + ;; i.e. 2 integral bits. The result is sqrt(mant) * 2^7, + ;; and in order to get the same scaling like the input, .expo has to + ;; be adjusted by 7. ??????????????? + + ldi Carry, 8 + F7call normalize.store_with_flags + + pop r16 + pop r17 + ret + +ENDF sqrt_approx +#endif /* F7MOD_sqrt_approx_ */ + + +#undef CA +#undef C0 +#undef C1 +#undef C2 +#undef C3 +#undef C4 +#undef C5 +#undef C6 +#undef Carry + + +#ifdef F7MOD_D_fabs_ +_DEFUN __fabs + DALIAS fabs + LALIAS fabsl + andi R25, 0b01111111 + ret +_ENDF __fabs +#endif /* F7MOD_D_fabs_ */ + + +#ifdef F7MOD_D_neg_ +_DEFUN __neg +_LABEL __negdf2 + subi R25, 0b10000000 + ret +_ENDF __neg +#endif /* F7MOD_D_neg_ */ + + +#ifdef F7MOD_D_signbit_ +_DEFUN __signbit + DALIAS signbit + LALIAS signbitl + bst R25, 7 + clr R25 + clr R24 + bld R24, 0 + ret +_ENDF __signbit +#endif /* F7MOD_D_signbit_ */ + + +#ifdef F7MOD_D_copysign_ +_DEFUN __copysign + DALIAS copysign + LALIAS copysignl + bst R17, 7 + bld R25, 7 + ret +_ENDF __copysign +#endif /* F7MOD_D_copysign_ */ + + +#ifdef F7MOD_D_isinf_ +_DEFUN __isinf + DALIAS isinf + LALIAS isinfl + F7call class_D + ;; Inf: T = Z = 1. + brtc 0f + ldi R24, 1 + breq 1f +0: + clr R24 +1: + clr R25 + ret +_ENDF __isinf +#endif /* F7MOD_D_isinf_ */ + + +#ifdef F7MOD_D_isnan_ +_DEFUN __isnan + DALIAS isnan + LALIAS isnanl + F7call class_D + ;; NaN: T = 1, Z = 0. + brtc 0f + ldi R24, 1 + brne 1f +0: + clr R24 +1: + clr R25 + ret +_ENDF __isnan +#endif /* F7MOD_D_isnan_ */ + + +#ifdef F7MOD_D_isfinite_ +_DEFUN __isfinite + DALIAS isfinite + LALIAS isfinitel + F7call class_D + ;; Number <=> T = 0. + bld R24, 0 + com R24 + andi R24, 1 + clr R25 + ret +_ENDF __isfinite +#endif /* F7MOD_D_isfinite_ */ + + +#ifdef F7MOD_D_class_ +;; The encoded exponent has 11 Bits. +#define MAX_BIASED_EXPO 0b0111111111110000 + +;; Classify a double in R18[] +;; Number: T-Flag = 0. +;; +-Inf : T-Flag = 1, Z-Flag = 1. +;; NaN : T-Flag = 1, Z-Flag = 0. +DEFUN class_D + wmov R26, R24 + andi R26, lo8 (MAX_BIASED_EXPO) + andi R27, hi8 (MAX_BIASED_EXPO) + subi R26, lo8 (MAX_BIASED_EXPO) + sbci R27, hi8 (MAX_BIASED_EXPO) + clt + brne .L.number + set + ;; Set sign and expo to 0. + clr R25 + andi R24, lo8 (~MAX_BIASED_EXPO) + ;; What remains is the mantissa. + ;; Mantissa == 0 => +/-Inf. + ;; Mantissa != 0 => NaN. + ;; Compare R18[] against sign_extend(R26) with R26 = 0. + .global __cmpdi2_s8 + XJMP __cmpdi2_s8 +.L.number: + ret + +ENDF class_D +#endif /* F7MOD_D_class_ */ + + +#ifdef F7MOD_call_dd_ + +;; Provide double wrappers for functions that operate on f7_t and get f7_t*. +;; +;; We set up a frame of sizeof(f7_t), convert the input double in R18[] to +;; f7_t in that frame location, then call *Z and finally convert the result f7_t +;; to double R18[] if that's requested. +;; +;; call_dd: double func (double A) +;; void (*Z) (f7_t *aa, const f7_t *aa) +;; +;; call_dx: double func (type_t A) , sizeof(type_t) <= 4 +;; void (*Z) (f7_t *aa, type_t) +;; +;; call_xd: type_t func (double A) +;; type_t (*Z) (const f7_t *aa) +;; +;; call_ddx: double func (double A, word_t) , sizeof (word_t) <= 2 +;; void (*Z) (f7_t *aa, const f7_t *aa, word_t) + +#define WHAT R13 + +DEFUN call_dd ; WHAT = R13 = 3 + inc ZERO +LABEL call_xd ; WHAT = R13 = 2 + inc ZERO +LABEL call_ddx ; WHAT = R13 = 1 + inc ZERO +LABEL call_dx ; WHAT = R13 = 0 + push WHAT + mov WHAT, ZERO + clr ZERO + ;; R14/R15 hold Z, the address of the f7_worker function, until we need it. + push r14 + push r15 + wmov r14, Z + +#define n_pushed 4 +#define n_frame 10 + + do_prologue_saves n_pushed, n_frame + ;; Y = FramePointer + 1 + adiw Y, 1 + dec WHAT + brmi .Ldx ; WHAT was initially 0. + ;; FP + 1 = (f7_t) arg1 + wmov r16, Y + ;; The double argument is in R18[]. + XCALL F7_NAME (set_double_impl) + tst WHAT + brne .Lno.ddx ; WHAT was initially != 1. + ;; call_ddx: Set R20/21 to the 2-byte scalar / pointer argument. + ;; Fetch it from where prologue_saves put it. + ldd r20, Y + n_frame + 3 ; Saved R16 + ldd r21, Y + n_frame + 2 ; Saved R17 +.Lno.ddx: + wmov r22, Y ; &arg1 (input) +.Ldo.dx: + wmov r24, Y ; &arg1 (output) + wmov Z, r14 + XICALL + dec WHAT + breq .Lepilogue ; WHAT was initially 2: Return non-double. + wmov r24, Y ; &arg1 + XCALL F7_NAME (get_double) +.Lepilogue: + ;; + 3 to account for R13...R15 pushed prior to do_prologue_saves. + do_epilogue_restores n_pushed + 3, n_frame + +.Ldx: + ;; call_dx: Copy the 4-byte input scalar from R22[4] to R20[4]. + wmov r20, r22 + wmov r22, r24 + rjmp .Ldo.dx + +ENDF call_dd +#endif /* F7MOD_call_dd_ */ + + +#ifdef F7MOD_call_ddd_ + +;; Provide double wrappers for functions that operate on f7_t and get f7_t*. +;; +;; We set up a frame of 2 * sizeof(f7_t), convert the input doubles in R18[] +;; and R10[] to f7_t in these frame locations, then call *Z and finally +;; convert the result f7_t to double R18[] if that's requested. +;; +;; call_ddd: double func (double A, double B) +;; void (*Z) (f7_t *aa, const f7_t *aa, const f7_t *bb) +;; +;; call_xdd: type_t func (double A, double B) +;; type_t (*Z) (const f7_t *aa, const f7_t *bb) + +DEFUN call_ddd + inc ZERO +LABEL call_xdd + ;; R8/R9 hold Z, the address of the f7_worker function, until we need it. + push r9 + push r8 + wmov r8, Z + ;; This is an argument to call.2 and will be accessed by the arg pointer. + push ZERO + clr ZERO + rcall call.2 + pop TMP + pop r8 + pop r9 + ret + +#define n_pushed 4 +#define n_frame 20 + +call.2: + do_prologue_saves n_pushed, n_frame + ;; Y = FramePointer + 1 + adiw Y, 1 + ;; FP + 1 = (f7_t) arg1 + wmov r16, Y + ;; First double argument is already in R18[]. + XCALL F7_NAME (set_double_impl) + ;; FP + 11 = (f7_t) arg2 + wmov r16, Y + subi r16, lo8 (-10) + sbci r17, hi8 (-10) + ;; Move second double argument to R18[]. + wmov r18, r10 + wmov r20, r12 + wmov r22, r14 + ;; Get high word of arg2 from where prologue_saves put it. + ldd r24, Y + n_frame + 3 ; Saved R16 + ldd r25, Y + n_frame + 2 ; Saved R17 + XCALL F7_NAME (set_double_impl) + ;; Z (f7_t *arg1, const f7_t *arg1, const f7_t *arg2) + wmov Z, r8 + wmov r24, Y ; &arg1 + ;; WHAT == 0 => call_xdd + ;; WHAT != 0 => call_ddd + ldd TMP, Y + n_frame + n_pushed + PC_SIZE + tst TMP + breq .Lxdd + wmov r22, Y ; &arg1 + wmov r20, r16 ; &arg2 + XICALL + wmov r24, Y ; &arg1 + XCALL F7_NAME (get_double) +.Lepilogue: + do_epilogue_restores n_pushed, n_frame +.Lxdd: + wmov r22, r16 ; &arg2 + XICALL + rjmp .Lepilogue +ENDF call_ddd +#endif /* F7MOD_call_ddd_ */ + +#include "f7-wraps.h" + +#endif /* !AVR_TINY */ Property changes on: libgcc/config/avr/libf7/libf7-asm.sx ___________________________________________________________________ Added: svn:eol-style ## -0,0 +1 ## +native \ No newline at end of property Added: svn:mime-type ## -0,0 +1 ## +text/plain \ No newline at end of property Index: libgcc/config/avr/libf7/libf7-c-object.mk =================================================================== --- libgcc/config/avr/libf7/libf7-c-object.mk (nonexistent) +++ libgcc/config/avr/libf7/libf7-c-object.mk (working copy) @@ -0,0 +1,20 @@ +# This file is included several times in a row, once for each element of +# $(iter-items). On each inclusion, we advance $o to the next element. +# $(iter-labels) is also advanced. +# This works similar to $(srcdir)/siditi-object.mk. + +o := $(firstword $(iter-items)) +iter-items := $(filter-out $o,$(iter-items)) + +$o-label := $(firstword $(iter-labels)) +iter-labels := $(wordlist 2,$(words $(iter-labels)),$(iter-labels)) + +f7_c_$o$(objext): f7_c_%$(objext): $(libf7)/libf7.c + $(gcc_compile) -DF7MOD_$($*-label)_ $(F7_C_FLAGS) \ + -c $< + +ifeq ($(enable_shared),yes) +f7_c_$(o)_s$(objext): %_s$(objext): $(libf7)/libf7.c + $(gcc_s_compile) -DF7MOD_$($*-label)_ $(F7_C_FLAGS) \ + -c $< +endif Property changes on: libgcc/config/avr/libf7/libf7-c-object.mk ___________________________________________________________________ Added: svn:eol-style ## -0,0 +1 ## +native \ No newline at end of property Added: svn:mime-type ## -0,0 +1 ## +text/plain \ No newline at end of property Index: libgcc/config/avr/libf7/libf7-common.mk =================================================================== --- libgcc/config/avr/libf7/libf7-common.mk (nonexistent) +++ libgcc/config/avr/libf7/libf7-common.mk (working copy) @@ -0,0 +1,102 @@ +# f7_c_*.o modules from libf7.c. +F7_C_PARTS += set_s16 set_u16 set_s32 set_u32 init +F7_C_PARTS += get_s16 get_u16 get_s32 get_u32 get_s64 get_u64 +F7_C_PARTS += lrint ldexp frexp madd_msub madd msub hypot +F7_C_PARTS += addsub add sub mulx mul square divx div div1 fmod sqrt cbrt +F7_C_PARTS += Ineg Iadd Isub Imul Idiv IRsub Isquare Ildexp Isqrt +F7_C_PARTS += set_float get_float get_double set_double set_pdouble +F7_C_PARTS += fabs neg fmin fmax minmax truncx trunc floor ceil round lround +F7_C_PARTS += horner logx log log10 log2 exp pow10 pow powi +F7_C_PARTS += sin cos tan cotan sincos sinh cosh tanh sinhcosh +F7_C_PARTS += asinacos asin acos atan atan2 +F7_C_PARTS += abscmp_msb_ge cmp cmp_abs cmp_unordered + +F7_C_PARTS += const_1 const_1_2 const_1_3 +F7_C_PARTS += const_pi const_ln2 const_1_ln2 const_ln10 const_1_ln10 const_sqrt2 +F7_C_PARTS += # const_m1 const_2 const_sqrt2 + +# f7_asm_*.o modules from libf7-asm.sx. +F7_ASM_PARTS += classify clr mul_mant cmp_mant set_u64 +F7_ASM_PARTS += copy copy_P copy_mant clr_mant_lsbs +F7_ASM_PARTS += addsub_mant_scaled store load +F7_ASM_PARTS += to_integer to_unsigned clz normalize_with_carry normalize +F7_ASM_PARTS += store_expo sqrt16 sqrt_approx div + +F7_ASM_PARTS += D_class +F7_ASM_PARTS += D_isnan D_isinf D_isfinite D_signbit D_copysign D_neg D_fabs + +F7_ASM_PARTS += call_dd call_ddd + +# Stuff that will be wrapped in f7-wraps.h (included by libf7-asm.sx) +# and give f7_asm_D_*.o modules. +g_ddd += add sub mul div +g_xdd_cmp += le lt ge gt ne eq unord +g_dx += floatunsidf floatsidf extendsfdf2 +g_xd += fixdfsi fixdfdi fixunsdfdi fixunsdfsi truncdfsf2 + +m_ddd += pow fmin fmax fmod hypot atan2 +m_ddx += ldexp frexp +m_dd += sqrt cbrt exp exp10 pow10 log log10 log2 sin cos tan cotan asin acos atan +m_dd += ceil floor trunc round sinh cosh tanh +m_xd += lrint lround + +# -mcall-prologues +CALL_PROLOGUES += divx sqrt cbrt get_double set_double logx exp exp10 pow10 +CALL_PROLOGUES += put_C truncx round minmax sincos tan cotan pow powi fmod +CALL_PROLOGUES += atan asinacos madd_msub hypot init horner sinhcosh tanh + +# -mstrict-X +STRICT_X += log addsub truncx ldexp exp + +# Renames used when building f7-renames.h. +F7F += fabs neg add sub addsub div div1 divx fmod sqrt cbrt +F7F += square mul mulx madd_msub madd msub hypot +F7F += Ineg Iadd Isub Imul Idiv IRsub Isquare Ildexp Isqrt +F7F += le lt gt ge ne eq cmp cmp_abs ordered unordered cmp_unordered +F7F += lt_impl gt_impl le_impl ge_impl eq_impl ne_impl unord_impl + +F7F += lrint ldexp frexp exp logx log log10 log2 +F7F += minmax fmax fmin floor ceil round lround trunc truncx +F7F += horner pow10 exp10 pow powi +F7F += sin cos tan cotan sincos sinh cosh tanh sinhcosh +F7F += asinacos asin acos atan atan2 +F7F += mul_noround sqrt16_round sqrt16_floor +F7F += clr_mant_lsbs abscmp_msb_ge lshrdi3 ashldi3 +F7F += assert + +F7F += classify + +F7F += class_inf class_nan class_number class_zero class_nonzero class_sign +F7F += signbit set_sign set_nan set_inf +F7F += is_inf is_nan is_number is_zero is_nonzero +F7F += clr copy copy_P copy_mant msbit is0 cmp_mant store_expo +F7F += abs + +F7F += set_s64 set_s32 set_s16 set_s16_impl set_u16_worker +F7F += set_u64 set_u32 set_u16 set_u16_impl +F7F += set_float set_pdouble set_double_impl set_double init_impl init +F7F += get_s16 get_s32 get_s64 get_float +F7F += get_u16 get_u32 get_u64 get_double + +F7F += set_eps set_1pow2 + +# Renames for ALIASes without own module. +F7F += min max exp10 +F7F += floatunsidf floatsidf extendsfdf2 +F7F += fixdfsi fixdfdi fixunsdfdi fixunsdfsi truncdfsf2 + +# Renames for f7-const.def. +F7F_cst += 1 2 1_2 1_3 m1 pi ln2 ln10 1_ln2 1_ln10 sqrt2 + +F7F_asm += classify +F7F_asm += store_expo clr copy copy_P copy_mant +F7F_asm += cmp_mant normalize store_expo +F7F_asm += set_u64 set_s64 addsub_mant_scaled mul_mant +F7F_asm += to_integer to_unsigned clr_mant_lsbs +F7F_asm += div sqrt_approx sqrt16_round sqrt16_floor +F7F_asm += lshrdi3 ashldi3 + +F7F_asm += class_D + +F7F_asm += call_ddd call_xdd call_ddx +F7F_asm += call_dd call_xd call_dx Property changes on: libgcc/config/avr/libf7/libf7-common.mk ___________________________________________________________________ Added: svn:eol-style ## -0,0 +1 ## +native \ No newline at end of property Added: svn:mime-type ## -0,0 +1 ## +text/plain \ No newline at end of property Index: libgcc/config/avr/libf7/libf7-const.def =================================================================== --- libgcc/config/avr/libf7/libf7-const.def (nonexistent) +++ libgcc/config/avr/libf7/libf7-const.def (working copy) @@ -0,0 +1,201 @@ +/* Copyright (C) 2019 Free Software Foundation, Inc. + + This file is part of LIBF7, which is part of GCC. + + GCC is free software; you can redistribute it and/or modify it under + the terms of the GNU General Public License as published by the Free + Software Foundation; either version 3, or (at your option) any later + version. + + GCC is distributed in the hope that it will be useful, but WITHOUT ANY + WARRANTY; without even the implied warranty of MERCHANTABILITY or + FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + for more details. + + Under Section 7 of GPL version 3, you are granted additional + permissions described in the GCC Runtime Library Exception, version + 3.1, as published by the Free Software Foundation. + + You should have received a copy of the GNU General Public License and + a copy of the GCC Runtime Library Exception along with this program; + see the files COPYING3 and COPYING.RUNTIME respectively. If not, see + <http://www.gnu.org/licenses/>. */ + +#ifndef ONLY_CONST_WITH_ID +#if defined IN_LIBF7_H || defined (F7MOD_const_) || defined (F7MOD_const_1_2_) +F7_CONST_DEF (1_2, 0, 0x80,0,0,0,0,0,0, -1) +#endif /* 1/2 */ + +#if defined IN_LIBF7_H || defined (F7MOD_const_) || defined (F7MOD_const_1_) +F7_CONST_DEF (1, 0, 0x80,0,0,0,0,0,0, 0) +#endif /* 1 */ + +#if defined IN_LIBF7_H || defined (F7MOD_const_) || defined (F7MOD_const_2_) +F7_CONST_DEF (2, 0, 0x80,0,0,0,0,0,0, 1) +#endif /* 2 */ + +#if defined IN_LIBF7_H || defined (F7MOD_const_) || defined (F7MOD_const_1_3_) +F7_CONST_DEF (1_3, 0, 0xaa,0xaa,0xaa,0xaa,0xaa,0xaa,0xab, -2) +#endif /* 1/3 */ + +#if defined IN_LIBF7_H || defined (F7MOD_const_) || defined (F7MOD_const_1_ln2_) +F7_CONST_DEF (1_ln2, 0, 0xb8,0xaa,0x3b,0x29,0x5c,0x17,0xf1, 0) +#endif /* 1 / ln2 */ + +#if defined IN_LIBF7_H || defined (F7MOD_const_) || defined (F7MOD_const_1_ln10_) +F7_CONST_DEF (1_ln10, 0, 0xde,0x5b,0xd8,0xa9,0x37,0x28,0x72, -2) +#endif /* 1 / ln10 */ +#endif /* const with ID */ + +#if defined IN_LIBF7_H || defined (F7MOD_const_) || defined (F7MOD_const_m1_) +F7_CONST_DEF (m1, 1, 0x80,0,0,0,0,0,0, 0) +#endif /* -1 */ + +#if defined IN_LIBF7_H || defined (F7MOD_const_) || defined (F7MOD_const_pi_) +F7_CONST_DEF (pi, 0, 0xc9,0x0f,0xda,0xa2,0x21,0x68,0xc2, 1) +#endif /* pi */ + +#if defined IN_LIBF7_H || defined (F7MOD_const_) || defined (F7MOD_const_ln2_) +F7_CONST_DEF (ln2, 0, 0xb1,0x72,0x17,0xf7,0xd1,0xcf,0x7a, -1) +#endif /* ln2 */ + +#if defined IN_LIBF7_H || defined (F7MOD_const_) || defined (F7MOD_const_ln10_) +F7_CONST_DEF (ln10, 0, 0x93,0x5d,0x8d,0xdd,0xaa,0xa8,0xac, 1) +#endif /* ln10 */ + +#if defined IN_LIBF7_H || defined (F7MOD_const_) || defined (F7MOD_const_sqrt2_) +F7_CONST_DEF (sqrt2, 0, 0xb5,0x04,0xf3,0x33,0xf9,0xde,0x65, 0) +#endif /* sqrt2 */ + +#if !defined (IN_LIBF7_H) && !defined (F7MOD_const_) + +#if defined (F7MOD_logx_) +// 2.00000000000000000287938058543222037939 + 0.666666666666667653654896829047862109605x + 0.399999999996639180070480580779767357622x^2 + 0.285714286985476106491309914245597720749x^3 + 0.222222024077476110197597399876978069272x^4 + 0.181833876328594532366358057253631240618x^5 + 0.153181571233880298729095145342556944283x^6 + 0.147580071554569676223389696418304199218x^7 +F7_CONST_DEF (X, 0, 0x80,0x00,0x00,0x00,0x00,0x00,0x00, 1) +F7_CONST_DEF (X, 0, 0xaa,0xaa,0xaa,0xaa,0xaa,0xaa,0xf2, -1) +F7_CONST_DEF (X, 0, 0xcc,0xcc,0xcc,0xcc,0xc5,0x68,0xd4, -2) +F7_CONST_DEF (X, 0, 0x92,0x49,0x24,0x9d,0x34,0x85,0x25, -2) +F7_CONST_DEF (X, 0, 0xe3,0x8e,0x2b,0x97,0x74,0x69,0xb3, -3) +F7_CONST_DEF (X, 0, 0xba,0x32,0xa8,0xe0,0x8b,0x7d,0xcc, -3) +F7_CONST_DEF (X, 0, 0x9c,0xdb,0xa1,0x3b,0x32,0x98,0x9a, -3) +F7_CONST_DEF (X, 0, 0x97,0x1f,0x3a,0xf3,0x79,0x0d,0xc8, -3) + +#elif defined (F7MOD_exp_) +#ifdef ARRAY_NAME +// Intended for |X| < ln2 / 8 +F7_CONST_DEF (X, 0, 0x80,0x00,0x00,0x00,0x00,0x00,0x00, 0) +// f(x) = (e^x - 1) / x and f -> 1 + x*f by means of const_1 above. +// 0.99999999999999993104781692731238623644199 + 0.50000000000000018681311888335564572558345x + 0.16666666666696003026683736262180526042594x^2 + 0.041666666666489887893617892155331720764329x^3 + 0.0083333331383434741356242595738508060753717x^4 + 0.0013888889242082668539796357369837122917246x^5 + 0.00019845416752769938401628433087258224774248x^6 + 0.000024801569801689274979195020335136136418918x^7 +F7_CONST_DEF (X, 0, 0xff,0xff,0xff,0xff,0xff,0xff,0xfb, -1) +F7_CONST_DEF (X, 0, 0x80,0x00,0x00,0x00,0x00,0x00,0x0d, -1) +F7_CONST_DEF (X, 0, 0xaa,0xaa,0xaa,0xaa,0xab,0xf4,0xf7, -3) +F7_CONST_DEF (X, 0, 0xaa,0xaa,0xaa,0xaa,0xa7,0x8e,0x87, -5) +F7_CONST_DEF (X, 0, 0x88,0x88,0x88,0x52,0xef,0x57,0x81, -7) +F7_CONST_DEF (X, 0, 0xb6,0x0b,0x61,0x03,0xb6,0x6b,0x82, -10) +F7_CONST_DEF (X, 0, 0xd0,0x18,0x22,0x8c,0x6f,0xde,0x72, -13) +F7_CONST_DEF (X, 0, 0xd0,0x0c,0xf7,0x31,0x28,0xba,0xb5, -16) +#else +// Negative because f7_const_ln2 is ln2 rounded up. +F7_CONST_DEF (ln2_low, 1, 0xa8,0x6c,0x38,0x98,0xcf,0xf8,0x1a, -58) +#endif + +#elif defined (F7MOD_atan_) + +#if defined (MINIMAX_6_6_IN_0_1) +// https://www.mikrocontroller.net/topic/480840#6003520 +#if defined (FOR_NUMERATOR) +// 1209.7470017580907217240715 + 3031.0745956115083044212807x + 2761.7198246138834959053784x^2 + 1114.1290728455183546172942x^3 + 192.5792014481559613474286x^4 + 11.3221594116764655236245x^5 + 0.09762721591717633036983x^6 +// p = Poly ([Decimal('1209.7470017580907217240715'), Decimal('3031.0745956115083044212807'), Decimal('2761.7198246138834959053784'), Decimal('1114.1290728455183546172942'), Decimal('192.5792014481559613474286'), Decimal('11.3221594116764655236245'), Decimal('0.09762721591717633036983')]) +F7_CONST_DEF (X, 0, 0x97,0x37,0xe7,0x70,0x3b,0x21,0xbc, 10) +F7_CONST_DEF (X, 0, 0xbd,0x71,0x31,0x8b,0x2a,0xfd,0xa7, 11) +F7_CONST_DEF (X, 0, 0xac,0x9b,0x84,0x66,0xd0,0x77,0xc4, 11) +F7_CONST_DEF (X, 0, 0x8b,0x44,0x21,0x5d,0x60,0x49,0xb2, 10) +F7_CONST_DEF (X, 0, 0xc0,0x94,0x46,0x8b,0xcd,0xa0,0x2d, 7) +F7_CONST_DEF (X, 0, 0xb5,0x27,0x90,0xa0,0x93,0xfb,0xfc, 3) +F7_CONST_DEF (X, 0, 0xc7,0xf0,0xc7,0x1c,0x82,0xab,0x23, -4) +#elif defined (FOR_DENOMINATOR) +// 1209.7470017580907287514197 + 3434.3235961975351716547069x + 3664.5449563283749893504796x^2 + 1821.6003392918464941509225x^3 + 423.0716464809047804524206x^4 + 39.9178842486537981501999x^5 + x^6 +// q = Poly ([Decimal('1209.7470017580907287514197'), Decimal('3434.3235961975351716547069'), Decimal('3664.5449563283749893504796'), Decimal('1821.6003392918464941509225'), Decimal('423.0716464809047804524206'), Decimal('39.9178842486537981501999'), Decimal('1.0')]) +F7_CONST_DEF (X, 0, 0x97,0x37,0xe7,0x70,0x3b,0x21,0xbc, 10) +F7_CONST_DEF (X, 0, 0xd6,0xa5,0x2d,0x73,0x34,0xd8,0x60, 11) +F7_CONST_DEF (X, 0, 0xe5,0x08,0xb8,0x24,0x20,0x81,0xe7, 11) +F7_CONST_DEF (X, 0, 0xe3,0xb3,0x35,0xfa,0xbf,0x1f,0x81, 10) +F7_CONST_DEF (X, 0, 0xd3,0x89,0x2b,0xb6,0x3e,0x2e,0x05, 8) +F7_CONST_DEF (X, 0, 0x9f,0xab,0xe9,0xd9,0x35,0xed,0x27, 5) +F7_CONST_DEF (X, 0, 0x80,0x00,0x00,0x00,0x00,0x00,0x00, 0) +#endif + +#elif defined (SWIFT_3_4) +// My [3,4] MiniMax +#if defined (FOR_NUMERATOR) +// 0.999999999999999998080178351225003952632 + 1.51597040589722809277543885223133087789x + 0.636928974763539784860899545005247736093x^2 + 0.0638944455799886571709448345524306512048x^3 +F7_CONST_DEF (X, 0, 0x80,0x00,0x00,0x00,0x00,0x00,0x00, 0) +F7_CONST_DEF (X, 0, 0xc2,0x0b,0x51,0x79,0x84,0x27,0x00, 0) +F7_CONST_DEF (X, 0, 0xa3,0x0d,0xc6,0xfc,0x7b,0xf5,0x5d, -1) +F7_CONST_DEF (X, 0, 0x82,0xdb,0x17,0x51,0x4b,0xfc,0xad, -4) +#elif defined (FOR_DENOMINATOR) +// 1 + 1.84930373923056200945453682584178320362x + 1.05336355450697082895016644607427716580x^2 + 0.188012025422931152047197803304030906006x^3 + 0.00507310235929401206762490897042543192106x^4 +F7_CONST_DEF (X, 0, 0x80,0x00,0x00,0x00,0x00,0x00,0x00, 0) +F7_CONST_DEF (X, 0, 0xec,0xb5,0xfc,0x24,0x2e,0xd1,0xc0, 0) +F7_CONST_DEF (X, 0, 0x86,0xd4,0x9d,0xf0,0xb3,0xef,0xb8, 0) +F7_CONST_DEF (X, 0, 0xc0,0x86,0x39,0x71,0xc8,0xeb,0x3d, -3) +F7_CONST_DEF (X, 0, 0xa6,0x3c,0x44,0x5c,0x78,0x87,0x2d, -8) +#else +F7_CONST_DEF (1_sqrt3, 0, 0x93,0xCD,0x3A,0x2C,0x81,0x98,0xE2, -1) +F7_CONST_DEF (pi_6, 0, 0x86,0x0a,0x91,0xc1,0x6b,0x9b,0x2c, -1) +#endif +#endif // which MiniMax + +#elif defined (F7MOD_asinacos_) +// Relative error < 5.6E-18, quality = 1.00000037 (ideal = 1). +#if defined (FOR_NUMERATOR) +// 0.99999999999999999442491073135027586203 - 1.035234033892197627842731209x + 0.35290206232981519813422591897720574012x^2 - 0.04333483170641685705612351801x^3 + 0.0012557428614630796315205218507940285622x^4 + 0.0000084705471128435769021718764878041684288x^5 +// p = Poly ([Decimal('0.99999999999999999442491073135027586203'), Decimal('-1.0352340338921976278427312087167692142'), Decimal('0.35290206232981519813422591897720574012'), Decimal('-0.043334831706416857056123518013656946650'), Decimal('0.0012557428614630796315205218507940285622'), Decimal('0.0000084705471128435769021718764878041684288')]) +F7_CONST_DEF (X, 0, 0x80,0x00,0x00,0x00,0x00,0x00,0x00, 0) +F7_CONST_DEF (X, 1, 0x84,0x82,0x8c,0x7f,0xa2,0xf6,0x65, 0) +F7_CONST_DEF (X, 0, 0xb4,0xaf,0x94,0x40,0xcb,0x86,0x69, -2) +F7_CONST_DEF (X, 1, 0xb1,0x7f,0xdd,0x4f,0x4e,0xbe,0x1d, -5) +F7_CONST_DEF (X, 0, 0xa4,0x97,0xbd,0x0b,0x59,0xc9,0x25, -10) +F7_CONST_DEF (X, 0, 0x8e,0x1c,0xb9,0x0b,0x50,0x6c,0xce, -17) +#elif defined (FOR_DENOMINATOR) +// 1 - 1.118567367225532923662371649x + 0.42736600959872448854098334016758333519x^2 - 0.06355588484963171659942148390x^3 + 0.0028820878185134035637440105959294542908x^4 +// q = Poly ([Decimal('1'), Decimal('-1.1185673672255329236623716486696411533'), Decimal('0.42736600959872448854098334016758333519'), Decimal('-0.063555884849631716599421483898013782858'), Decimal('0.0028820878185134035637440105959294542908')]) +F7_CONST_DEF (X, 0, 0x80,0x00,0x00,0x00,0x00,0x00,0x00, 0) +F7_CONST_DEF (X, 1, 0x8f,0x2d,0x37,0x2a,0x4d,0xa1,0x57, 0) +F7_CONST_DEF (X, 0, 0xda,0xcf,0xb7,0xb5,0x4c,0x0d,0xee, -2) +F7_CONST_DEF (X, 1, 0x82,0x29,0x96,0x77,0x2e,0x19,0xc7, -4) +F7_CONST_DEF (X, 0, 0xbc,0xe1,0x68,0xec,0xba,0x20,0x29, -9) +#endif + +#elif defined (F7MOD_sincos_) +#if defined (FOR_SIN) +// sin(sqrt(x)) / sqrt(x) in [0, pi^2/16] ~ [0, 0.6169]. +// Error < 3.313E-18, quality@quot ~ 1. +// p = Poly ([Decimal('0.9999999999999999966879566228039012'), Decimal('-0.1666666666666661475365354492822562'), Decimal('0.008333333333320002444252560522892007'), Decimal('-0.0001984126982840212992698250499618594'), Decimal('0.000002755731329901509333692668814721451'), Decimal('-2.505070584638448064973439248538648E-8'), Decimal('1.589413637225924008237178592214358E-10')]) +// 0.9999999999999999966879566228039012 - 0.1666666666666661475365354492822562x + 0.008333333333320002444252560522892007x^2 - 0.0001984126982840212992698250499618594x^3 + 0.000002755731329901509333692668814721451x^4 - 2.505070584638448064973439248538648E-8x^5 + 1.589413637225924008237178592214358E-10x^6 +F7_CONST_DEF (X, 0, 0x80,0x00,0x00,0x00,0x00,0x00,0x00, 0) +F7_CONST_DEF (X, 1, 0xaa,0xaa,0xaa,0xaa,0xaa,0xaa,0x15, -3) +F7_CONST_DEF (X, 0, 0x88,0x88,0x88,0x88,0x87,0x98,0x63, -7) +F7_CONST_DEF (X, 1, 0xd0,0x0d,0x00,0xcd,0xc9,0x7e,0x34, -13) +F7_CONST_DEF (X, 0, 0xb8,0xef,0x1a,0x8f,0x9e,0x91,0xa2, -19) +F7_CONST_DEF (X, 1, 0xd7,0x2f,0x15,0xb0,0x7c,0x11,0x7a, -26) +F7_CONST_DEF (X, 0, 0xae,0xc2,0x04,0x43,0x51,0x29,0xe8, -33) +#elif defined (FOR_COS) +// cos(sqrt(x)) in [0, pi^2/16] ~ [0, 0.6169]. +// Error < 3.6E-20, quality@quot ~ 1. +// p = Poly ([Decimal('0.999999999999999999964159204335255989833'), Decimal('-0.499999999999999992843598630174036090975'), Decimal('0.0416666666666664302574165741342831192783'), Decimal('-0.00138888888888589604372397206213200172560'), Decimal('0.0000248015872828994642415389301648196400167'), Decimal('-2.75573128656963697259081300852869272237E-7'), Decimal('2.08755551457127352241183195408105664857E-9'), Decimal('-1.13521232075810109334454849167561526185E-11')]) +// 0.999999999999999999964159204335255989833 - 0.499999999999999992843598630174036090975x + 0.0416666666666664302574165741342831192783x^2 - 0.00138888888888589604372397206213200172560x^3 + 0.0000248015872828994642415389301648196400167x^4 - 2.75573128656963697259081300852869272237E-7x^5 + 2.08755551457127352241183195408105664857E-9x^6 - 1.13521232075810109334454849167561526185E-11x^7 +F7_CONST_DEF (X, 0, 0x80,0x00,0x00,0x00,0x00,0x00,0x00, 0) +F7_CONST_DEF (X, 1, 0xff,0xff,0xff,0xff,0xff,0xff,0xff, -2) +F7_CONST_DEF (X, 0, 0xaa,0xaa,0xaa,0xaa,0xaa,0xa9,0x9a, -5) +F7_CONST_DEF (X, 1, 0xb6,0x0b,0x60,0xb6,0x09,0xb1,0x66, -10) +F7_CONST_DEF (X, 0, 0xd0,0x0d,0x00,0xcd,0x6b,0xb3,0xf0, -16) +F7_CONST_DEF (X, 1, 0x93,0xf2,0x7b,0x7f,0x10,0xce,0x3d, -22) +F7_CONST_DEF (X, 0, 0x8f,0x74,0xaa,0x3c,0xcf,0x51,0x3d, -29) +F7_CONST_DEF (X, 1, 0xc7,0xb5,0x6a,0xf8,0x0e,0x32,0x07, -37) +#else +F7_CONST_DEF (pi_low,0, 0xd3,0x13,0x19,0x8a,0x2e,0x03,0x70, 1 - F7_MANT_BITS-2) +#endif + +#endif +#endif /* ! IN_LIBF7_H && ! F7MOD_const_ */ Property changes on: libgcc/config/avr/libf7/libf7-const.def ___________________________________________________________________ Added: svn:eol-style ## -0,0 +1 ## +native \ No newline at end of property Added: svn:mime-type ## -0,0 +1 ## +text/plain \ No newline at end of property Index: libgcc/config/avr/libf7/libf7-constdef.h =================================================================== --- libgcc/config/avr/libf7/libf7-constdef.h (nonexistent) +++ libgcc/config/avr/libf7/libf7-constdef.h (working copy) @@ -0,0 +1,43 @@ +/* Copyright (C) 2019 Free Software Foundation, Inc. + + This file is part of LIBF7, which is part of GCC. + + GCC is free software; you can redistribute it and/or modify it under + the terms of the GNU General Public License as published by the Free + Software Foundation; either version 3, or (at your option) any later + version. + + GCC is distributed in the hope that it will be useful, but WITHOUT ANY + WARRANTY; without even the implied warranty of MERCHANTABILITY or + FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + for more details. + + Under Section 7 of GPL version 3, you are granted additional + permissions described in the GCC Runtime Library Exception, version + 3.1, as published by the Free Software Foundation. + + You should have received a copy of the GNU General Public License and + a copy of the GCC Runtime Library Exception along with this program; + see the files COPYING3 and COPYING.RUNTIME respectively. If not, see + <http://www.gnu.org/licenses/>. */ + +#if !defined STATIC +#define STATIC /* empty */ +#endif + +#if USE_LPM +#define F7_CONST_DEF(NAME, FLAGS, M6, M5, M4, M3, M2, M1, M0, EXPO) \ + STATIC const __attribute__((__progmem__)) \ + f7_t F7_(const_ ## NAME ## _P) = \ + { .flags = FLAGS, .mant = { M0, M1, M2, M3, M4, M5, M6 }, .expo = EXPO }; + #include "libf7-const.def" +#undef F7_CONST_DEF +#else +#define F7_CONST_DEF(NAME, FLAGS, M6, M5, M4, M3, M2, M1, M0, EXPO) \ + STATIC const f7_t F7_(const_ ## NAME) = \ + { .flags = FLAGS, .mant = { M0, M1, M2, M3, M4, M5, M6 }, .expo = EXPO }; + #include "libf7-const.def" +#undef F7_CONST_DEF +#endif // USE_LPM + +#undef STATIC Property changes on: libgcc/config/avr/libf7/libf7-constdef.h ___________________________________________________________________ Added: svn:eol-style ## -0,0 +1 ## +native \ No newline at end of property Added: svn:mime-type ## -0,0 +1 ## +text/plain \ No newline at end of property Index: libgcc/config/avr/libf7/libf7.c =================================================================== --- libgcc/config/avr/libf7/libf7.c (nonexistent) +++ libgcc/config/avr/libf7/libf7.c (working copy) @@ -0,0 +1,2501 @@ +/* Copyright (C) 2019 Free Software Foundation, Inc. + + This file is part of LIBF7, which is part of GCC. + + GCC is free software; you can redistribute it and/or modify it under + the terms of the GNU General Public License as published by the Free + Software Foundation; either version 3, or (at your option) any later + version. + + GCC is distributed in the hope that it will be useful, but WITHOUT ANY + WARRANTY; without even the implied warranty of MERCHANTABILITY or + FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + for more details. + + Under Section 7 of GPL version 3, you are granted additional + permissions described in the GCC Runtime Library Exception, version + 3.1, as published by the Free Software Foundation. + + You should have received a copy of the GNU General Public License and + a copy of the GCC Runtime Library Exception along with this program; + see the files COPYING3 and COPYING.RUNTIME respectively. If not, see + <http://www.gnu.org/licenses/>. */ + +#include "libf7.h" + +#ifndef __AVR_TINY__ + +#define ALIAS(X, Y) \ + F7_WEAK __attribute__((__alias__(F7_STRINGY(X)))) __typeof__(X) Y; + +#define DALIAS(...) // empty +#define LALIAS(...) // empty + +#ifndef IN_LIBGCC2 + +#include <stdio.h> +#include <assert.h> + +#define in_libgcc false + +_Static_assert (sizeof (f7_t) == 10 && F7_MANT_BYTES == 7, + "libf7 will only work with 7-byte mantissa."); +#else + +#define in_libgcc true + +#if __SIZEOF_DOUBLE__ == 8 +#undef DALIAS +#define DALIAS(X,Y) \ + F7_WEAK __attribute__((__alias__(F7_STRINGY(X)))) __typeof__(X) Y; +#endif + +#if __SIZEOF_LONG_DOUBLE__ == 8 +#undef LALIAS +#define LALIAS(X,Y) \ + F7_WEAK __attribute__((__alias__(F7_STRINGY(X)))) __typeof__(X) Y; +#endif + +#endif // in libgcc + +static F7_INLINE +void f7_assert (bool x) +{ + if (!in_libgcc && !x) + __builtin_abort(); +} + +static F7_INLINE +int16_t abs_ssat16 (int16_t a) +{ + _Sat _Fract sa = __builtin_avr_rbits (a); + return __builtin_avr_bitsr (__builtin_avr_absr (sa)); +} + +static F7_INLINE +int16_t add_ssat16 (int16_t a, int16_t b) +{ + _Sat _Fract sa = __builtin_avr_rbits (a); + _Sat _Fract sb = __builtin_avr_rbits (b); + return __builtin_avr_bitsr (sa + sb); +} + +static F7_INLINE +int16_t sub_ssat16 (int16_t a, int16_t b) +{ + _Sat _Fract sa = __builtin_avr_rbits (a); + _Sat _Fract sb = __builtin_avr_rbits (b); + return __builtin_avr_bitsr (sa - sb); +} + +static F7_INLINE +int8_t ssat8_range (int16_t a, int8_t range) +{ + if (a >= range) + return range; + if (a <= -range) + return -range; + return a; +} + + +#define IN_LIBF7_H + #define F7_CONST_DEF(NAME, FLAGS, M6, M5, M4, M3, M2, M1, M0, EXPO) \ + F7_UNUSED static const uint8_t F7_(const_##NAME##_msb) = M6; \ + F7_UNUSED static const int16_t F7_(const_##NAME##_expo) = EXPO; + #include "libf7-const.def" + #undef F7_CONST_DEF +#undef IN_LIBF7_H + + +/* + libgcc naming converntions for conversions: + + __float<fmode><fmode> : Convert float modes. + __floatun<imode><fmode>: Convert unsigned integral to float. + __fix<fmode><imode> : Convert float to signed integral. + __fixuns<fmode><imode> : Convert float to unsigned integral. +*/ + + +#ifdef F7MOD_floatundidf_ +F7_WEAK +f7_double_t __floatundidf (uint64_t x) +{ + f7_t xx; + f7_set_u64 (&xx, x); + return f7_get_double (&xx); +} +#endif // F7MOD_floatundidf_ + + +#ifdef F7MOD_floatdidf_ +F7_WEAK +f7_double_t __floatdidf (int64_t x) +{ + f7_t xx; + f7_set_s64 (&xx, x); + return f7_get_double (&xx); +} +#endif // F7MOD_floatdidf_ + + +#ifdef F7MOD_init_ +f7_t* f7_init_impl (uint64_t mant, uint8_t flags, f7_t *cc, int16_t expo) +{ + flags &= F7_FLAGS; + if (f7_class_number (flags)) + { + uint8_t msb; + while ((__builtin_memcpy (&msb, (uint8_t*) &mant + 7, 1), msb)) + { + mant >>= 1; + expo = add_ssat16 (expo, 1); + } + *(uint64_t*) cc->mant = mant; + cc->expo = add_ssat16 (expo, F7_MANT_BITS-1); + + cc = f7_normalize_asm (cc); + } + + cc->flags = flags; + + return cc; +} +#endif // F7MOD_init_ + + +#ifdef F7MOD_set_s16_ +f7_t* f7_set_s16_impl (f7_t *cc, int16_t i16) +{ + uint16_t u16 = (uint16_t) i16; + uint8_t flags = 0; + if (i16 < 0) + { + u16 = -u16; + flags = F7_FLAG_sign; + } + f7_set_u16_impl (cc, u16); + cc->flags = flags; + return cc; +} +#endif // F7MOD_set_s16_ + + +#ifdef F7MOD_set_u16_ +f7_t* f7_set_u16_impl (f7_t *cc, uint16_t u16) +{ + f7_clr (cc); + F7_MANT_HI2 (cc) = u16; + cc->expo = 15; + return f7_normalize_asm (cc); +} +#endif // F7MOD_set_u16_ + + +#ifdef F7MOD_set_s32_ +f7_t* f7_set_s32 (f7_t *cc, int32_t i32) +{ + uint32_t u32 = (uint32_t) i32; + uint8_t flags = 0; + if (i32 < 0) + { + u32 = -u32; + flags = F7_FLAG_sign; + } + cc = f7_set_u32 (cc, u32); + cc->flags = flags; + return cc; +} +ALIAS (f7_set_s32, f7_floatsidf) +#endif // F7MOD_set_s32_ + + +#ifdef F7MOD_set_u32_ +f7_t* f7_set_u32 (f7_t *cc, uint32_t u32) +{ + f7_clr (cc); + F7_MANT_HI4 (cc) = u32; + cc->expo = 31; + return f7_normalize_asm (cc); +} +ALIAS (f7_set_u32, f7_floatunsidf) +#endif // F7MOD_set_u32_ + + +// IEEE 754 single +// float = s bbbbbbbb mmmmmmmmmmmmmmmmmmmmmmm +// 31 +// s = sign +// b = biased exponent, bias = 127 +// m = mantissa + +// +0 = 0 0 0 +// -0 = 1 0 0 +// Inf = S B 0 = S * Inf, B = 0xff +// NaN = S B M, B = 0xff, M != 0 +// Sub-normal = S 0 M = S * 0.M * 2^{1 - bias}, M != 0 +// Normal = S B M = S * 1.M * 2^{B - bias}, B = 1 ... 0xfe + +#define FLT_DIG_EXP 8 +#define FLT_DIG_MANT (31 - FLT_DIG_EXP) +#define FLT_MAX_EXP ((1 << FLT_DIG_EXP) - 1) +#define FLT_EXP_BIAS (FLT_MAX_EXP >> 1) + +#ifdef F7MOD_set_float_ +F7_WEAK +void f7_set_float (f7_t *cc, float f) +{ + uint32_t val32; + + _Static_assert (__SIZEOF_FLOAT__ == 4, ""); + _Static_assert (__FLT_MANT_DIG__ == 24, ""); + __builtin_memcpy (&val32, &f, __SIZEOF_FLOAT__); + + uint16_t val16 = val32 >> 16; + val16 >>= FLT_DIG_MANT - 16; + + uint8_t expo_biased = val16 & FLT_MAX_EXP; + bool sign = val16 & (1u << FLT_DIG_EXP); + + f7_clr (cc); + + uint32_t mant = val32 & ((1ul << FLT_DIG_MANT) -1); + + if (mant == 0) + { + if (expo_biased == 0) + return; + if (expo_biased >= FLT_MAX_EXP) + return f7_set_inf (cc, sign); + } + + if (expo_biased == 0) + expo_biased = 1; // Sub-normal: biased expo of 1 was encoded as 0. + else if (expo_biased < FLT_MAX_EXP) + mant |= (1ul << FLT_DIG_MANT); + else + return f7_set_nan (cc); + + __builtin_memcpy (& F7_MANT_HI4 (cc), &mant, 4); + + cc->expo = expo_biased - FLT_EXP_BIAS + 31 - FLT_DIG_MANT; + f7_normalize_asm (cc); + f7_set_sign (cc, sign); +} +ALIAS (f7_set_float, f7_extendsfdf2) +#endif // F7MOD_set_float_ + + +#ifdef F7MOD_get_float_ +static F7_INLINE +float make_float (uint32_t x) +{ + float ff; + __builtin_memcpy (&ff, &x, 4); + return ff; + +} + +F7_WEAK +float f7_get_float (const f7_t *aa) +{ + uint8_t a_class = f7_classify (aa); + + if (f7_class_nan (a_class)) + return make_float (0xffc00000 /* NaN: Biased expo = 0xff, mant != 0 */); + + uint32_t mant; + __builtin_memcpy (&mant, &F7_MANT_CONST_HI4 (aa), 4); + + uint8_t expo8 = 0; + uint8_t mant_offset = FLT_DIG_EXP; + int16_t c_expo = add_ssat16 (aa->expo, FLT_EXP_BIAS); + + if (f7_class_zero (a_class) || c_expo <= -FLT_DIG_MANT) + { + // Zero or tiny. + return 0.0f; + } + else if (c_expo >= FLT_MAX_EXP || f7_class_inf (a_class)) + { + // Inf or overflow. + expo8 = FLT_MAX_EXP; + mant = 0; + } + else if (c_expo > 0) + { + // Normal. + expo8 = c_expo; + } + else + { + // Sub-normal: -DIG_MANT < c_expo <= 0. + // Encoding of 0 represents a biased exponent of 1. + // mant_offset in 9...31. + expo8 = 0; + mant_offset += 1 - c_expo; + } + + uint16_t expo16 = expo8 << (FLT_DIG_MANT - 16); + + if (f7_class_sign (a_class)) + expo16 |= 1u << (FLT_DIG_EXP + FLT_DIG_MANT - 16); + + mant >>= mant_offset; + + __asm ("cbr %T0%t2, 1 << (7 & %2)" "\n\t" + "or %C0, %A1" "\n\t" + "or %D0, %B1" + : "+d" (mant) + : "r" (expo16), "n" (FLT_DIG_MANT)); + + return make_float (mant); +} +F7_PURE ALIAS (f7_get_float, f7_truncdfsf2) +#endif // F7MOD_get_float_ + +#define DBL_DIG_EXP 11 +#define DBL_DIG_MANT (63 - DBL_DIG_EXP) +#define DBL_MAX_EXP ((1 << DBL_DIG_EXP) - 1) +#define DBL_EXP_BIAS (DBL_MAX_EXP >> 1) + + +#ifdef F7MOD_set_double_ +void f7_set_double_impl (f7_double_t val64, f7_t *cc) +{ + f7_clr (cc); + register uint64_t mant __asm ("r18") = val64 & ((1ull << DBL_DIG_MANT) -1); + + uint16_t val16 = 3[(uint16_t*) & val64]; + val16 >>= DBL_DIG_MANT - 48; + + uint16_t expo_biased = val16 & DBL_MAX_EXP; + bool sign = val16 & (1u << DBL_DIG_EXP); + + if (mant == 0) + { + if (expo_biased == 0) + return; + if (expo_biased >= DBL_MAX_EXP) + return f7_set_inf (cc, sign); + } + __asm ("" : "+r" (mant)); + + if (expo_biased == 0) + expo_biased = 1; // Sub-normal: biased expo of 1 was encoded as 0. + else if (expo_biased < DBL_MAX_EXP) + mant |= (1ull << DBL_DIG_MANT); + else + return f7_set_nan (cc); + + *(uint64_t*) & cc->mant = mant; + + cc->expo = expo_biased - DBL_EXP_BIAS + 63 - DBL_DIG_MANT - 8; + f7_normalize_asm (cc); + f7_set_sign (cc, sign); +} +#endif // F7MOD_set_double_ + + +#ifdef F7MOD_set_pdouble_ +void f7_set_pdouble (f7_t *cc, const f7_double_t *val64) +{ + f7_set_double (cc, *val64); +} +#endif // F7MOD_set_pdouble_ + + +#ifdef F7MOD_get_double_ +static F7_INLINE +uint64_t clr_r18 (void) +{ + extern void __clr_8 (void); + register uint64_t r18 __asm ("r18"); + __asm ("%~call %x[f]" : "=r" (r18) : [f] "i" (__clr_8)); + return r18; +} + +static F7_INLINE +f7_double_t make_double (uint64_t x) +{ + register f7_double_t r18 __asm ("r18") = x; + __asm ("" : "+r" (r18)); + return r18; +} + +F7_WEAK +f7_double_t f7_get_double (const f7_t *aa) +{ + uint8_t a_class = f7_classify (aa); + + if (f7_class_nan (a_class)) + { + uint64_t nan = clr_r18() | (0x7fffull << 48); + return make_double (nan); + } + + uint64_t mant; + __builtin_memcpy (&mant, & aa->mant, 8); + + mant &= 0x00ffffffffffffff; + + // FIXME: For subnormals, rounding is premature and should be + // done *after* the mantissa has been shifted into place + // (or the round value be shifted left accordingly). + // Round. + mant += 1u << (F7_MANT_BITS - (1 + DBL_DIG_MANT) - 1); + + uint8_t dex; + register uint64_t r18 __asm ("r18") = mant; + // dex = Overflow ? 1 : 0. + __asm ("bst %T[mant]%T[bitno]" "\n\t" + "clr %0" "\n\t" + "bld %0,0" + : "=r" (dex), [mant] "+r" (r18) + : [bitno] "n" (64 - 8)); + + mant = r18 >> dex; + + uint16_t expo16 = 0; + uint16_t mant_offset = DBL_DIG_EXP - 8; + int16_t c_expo = add_ssat16 (aa->expo, DBL_EXP_BIAS + dex); + + if (f7_class_zero (a_class) || c_expo <= -DBL_DIG_MANT) + { + // Zero or tiny. + return make_double (clr_r18()); + } + else if (c_expo >= DBL_MAX_EXP || f7_class_inf (a_class)) + { + // Inf or overflow. + expo16 = DBL_MAX_EXP; + mant = clr_r18(); + } + else if (c_expo > 0) + { + // Normal. + expo16 = c_expo; + } + else + { + // Sub-normal: -DIG_MANT < c_expo <= 0. + // Encoding expo of 0 represents a biased exponent of 1. + // mant_offset in 5...55 = 63-8. + mant_offset += 1 - c_expo; + } + + expo16 <<= (DBL_DIG_MANT - 48); + + if (f7_class_sign (a_class)) + expo16 |= 1u << (DBL_DIG_EXP + DBL_DIG_MANT - 48); + + // mant >>= mant_offset; + mant = f7_lshrdi3 (mant, mant_offset); + + r18 = mant; + __asm ("cbr %T0%t2, 1 << (7 & %2)" "\n\t" + "or %r0+6, %A1" "\n\t" + "or %r0+7, %B1" + : "+r" (r18) + : "r" (expo16), "n" (DBL_DIG_MANT)); + + return make_double (r18); +} +#endif // F7MOD_get_double_ + + +#ifdef F7MOD_fabs_ +F7_WEAK +void f7_fabs (f7_t *cc, const f7_t *aa) +{ + f7_abs (cc, aa); +} +#endif // F7MOD_fabs_ + + +#ifdef F7MOD_neg_ +F7_WEAK +f7_t* f7_neg (f7_t *cc, const f7_t *aa) +{ + f7_copy (cc, aa); + + uint8_t c_class = f7_classify (cc); + + if (! f7_class_zero (c_class)) + cc->sign = ! f7_class_sign (c_class); + + return cc; +} +#endif // F7MOD_neg_ + + +#ifdef F7MOD_frexp_ +F7_WEAK +void f7_frexp (f7_t *cc, const f7_t *aa, int *expo) +{ + uint8_t a_class = f7_classify (aa); + + if (f7_class_nan (a_class)) + return f7_set_nan (cc); + + if (f7_class_inf (a_class) || aa->expo == INT16_MAX) + return f7_set_inf (cc, f7_class_sign (a_class)); + + if (! f7_msbit (aa)) + { + *expo = 0; + return f7_clr (cc); + } + + *expo = 1 + aa->expo; + cc->flags = a_class & F7_FLAG_sign; + cc->expo = -1; + f7_copy_mant (cc, aa); +} +#endif // F7MOD_frexp_ + +#ifdef F7MOD_get_s16_ +F7_WEAK +int16_t f7_get_s16 (const f7_t *aa) +{ + extern int16_t to_s16 (const f7_t*, uint8_t) F7ASM(f7_to_integer_asm); + return to_s16 (aa, 0xf); +} +#endif // F7MOD_get_s16_ + + +#ifdef F7MOD_get_s32_ +F7_WEAK +int32_t f7_get_s32 (const f7_t *aa) +{ + extern int32_t to_s32 (const f7_t*, uint8_t) F7ASM(f7_to_integer_asm); + return to_s32 (aa, 0x1f); +} +F7_PURE ALIAS (f7_get_s32, f7_fixdfsi) +#endif // F7MOD_get_s32_ + + +#ifdef F7MOD_get_s64_ + F7_WEAK + int64_t f7_get_s64 (const f7_t *aa) +{ + extern int64_t to_s64 (const f7_t*, uint8_t) F7ASM(f7_to_integer_asm); + return to_s64 (aa, 0x3f); +} +F7_PURE ALIAS (f7_get_s64, f7_fixdfdi) +#endif // F7MOD_get_s64_ + +#ifdef F7MOD_get_u16_ + F7_WEAK + uint16_t f7_get_u16 (const f7_t *aa) +{ + extern uint16_t to_u16 (const f7_t*, uint8_t) F7ASM(f7_to_unsigned_asm); + return to_u16 (aa, 0xf); +} +#endif // F7MOD_get_u16_ + + +#ifdef F7MOD_get_u32_ +F7_WEAK +uint32_t f7_get_u32 (const f7_t *aa) +{ + extern uint32_t to_u32 (const f7_t*, uint8_t) F7ASM(f7_to_unsigned_asm); + return to_u32 (aa, 0x1f); +} +F7_PURE ALIAS (f7_get_u32, f7_fixunsdfsi) +#endif // F7MOD_get_u32_ + + +#ifdef F7MOD_get_u64_ +F7_WEAK +uint64_t f7_get_u64 (const f7_t *aa) +{ + extern int64_t to_u64 (const f7_t*, uint8_t) F7ASM(f7_to_unsigned_asm); + return to_u64 (aa, 0x3f); +} +F7_PURE ALIAS (f7_get_u64, f7_fixunsdfdi) +#endif // F7MOD_get_u64_ + + +#ifdef F7MOD_cmp_unordered_ +F7_NOINLINE +static int8_t cmp_u8 (uint8_t a_class, uint8_t b_class, bool sign_a); + +F7_WEAK +int8_t f7_cmp_unordered (const f7_t *aa, const f7_t *bb, bool with_sign) +{ + uint8_t a_class = f7_classify (aa); + uint8_t b_class = f7_classify (bb); + + uint8_t a_sign = f7_class_sign (a_class) & with_sign; + uint8_t b_sign = f7_class_sign (b_class) & with_sign; + uint8_t ab_class = a_class | b_class; + ab_class &= with_sign - 2; + + if (f7_class_nan (ab_class)) + return INT8_MIN; + + if (a_sign != b_sign) + return b_sign - a_sign; + + if (f7_class_inf (ab_class)) + return cmp_u8 (a_class, b_class, a_sign); + + if (f7_class_zero (ab_class)) + return cmp_u8 (b_class, a_class, a_sign); + + if (aa->expo < bb->expo) + return a_sign ? 1 : -1; + + if (aa->expo > bb->expo) + return a_sign ? -1 : 1; + + return cmp_u8 (1 + f7_cmp_mant (aa, bb), 1, a_sign); +} + + +int8_t cmp_u8 (uint8_t a_class, uint8_t b_class, bool sign_a) +{ + int8_t c; + __asm ("sub %[a], %[b]" "\n\t" + "breq 1f" "\n\t" + "sbc %[c], %[c]" "\n\t" + "sbci %[c], -1" "\n\t" + "sbrc %[s], 0" "\n\t" + "neg %[c]" "\n\t" + "1:" + : [c] "=d" (c) + : [a] "0" (a_class), [b] "r" (b_class), [s] "r" (sign_a)); + return c; +} +#endif // F7MOD_cmp_unordered_ + + +#ifdef F7MOD_cmp_abs_ +F7_WEAK +int8_t f7_cmp_abs (const f7_t *aa, const f7_t *bb) +{ + return f7_cmp_unordered (aa, bb, false /* no signs */); +} +#endif // F7MOD_cmp_abs_ + + +#ifdef F7MOD_cmp_ +F7_WEAK +int8_t f7_cmp (const f7_t *aa, const f7_t *bb) +{ + return f7_cmp_unordered (aa, bb, true /* with signs */); +} +#endif // F7MOD_cmp_ + + +#ifdef F7MOD_abscmp_msb_ge_ +// Compare absolute value of Number aa against a f7_t represented +// by msb and expo. +F7_WEAK +bool f7_abscmp_msb_ge (const f7_t *aa, uint8_t msb, int16_t expo) +{ + uint8_t a_msb = aa->mant[F7_MANT_BYTES - 1]; + + if (0 == (0x80 & a_msb)) + // 0 or subnormal. + return false; + + return aa->expo == expo + ? a_msb >= msb + : aa->expo > expo; +} +#endif // F7MOD_abscmp_msb_ge_ + +#ifdef F7MOD_lt_ +F7_WEAK +bool f7_lt_impl (const f7_t *aa, const f7_t *bb) +{ + return f7_lt (aa, bb); +} +#endif // F7MOD_lt_ + +#ifdef F7MOD_le_ +F7_WEAK +bool f7_le_impl (const f7_t *aa, const f7_t *bb) +{ + return f7_le (aa, bb); +} +#endif // F7MOD_le_ + +#ifdef F7MOD_gt_ +F7_WEAK +bool f7_gt_impl (const f7_t *aa, const f7_t *bb) +{ + return f7_gt (aa, bb); +} +#endif // F7MOD_gt_ + +#ifdef F7MOD_ge_ +F7_WEAK +bool f7_ge_impl (const f7_t *aa, const f7_t *bb) +{ + return f7_ge (aa, bb); +} +#endif // F7MOD_ge_ + +#ifdef F7MOD_ne_ +F7_WEAK +bool f7_ne_impl (const f7_t *aa, const f7_t *bb) +{ + return f7_ne (aa, bb); +} +#endif // F7MOD_ne_ + +#ifdef F7MOD_eq_ +F7_WEAK +bool f7_eq_impl (const f7_t *aa, const f7_t *bb) +{ + return f7_eq (aa, bb); +} +#endif // F7MOD_eq_ + + +#ifdef F7MOD_unord_ +F7_WEAK +bool f7_unord_impl (const f7_t *aa, const f7_t *bb) +{ + return f7_unordered (aa, bb); +} +#endif // F7MOD_unord_ + + +#ifdef F7MOD_minmax_ +F7_WEAK +f7_t* f7_minmax (f7_t *cc, const f7_t *aa, const f7_t *bb, bool do_min) +{ + int8_t cmp = f7_cmp_unordered (aa, bb, true /* with signs */); + + if (cmp == INT8_MIN) + return (f7_set_nan (cc), cc); + + if (do_min) + cmp = -cmp; + + return f7_copy (cc, cmp >= 0 ? aa : bb); +} +#endif // F7MOD_minmax_ + + +#ifdef F7MOD_fmax_ +F7_WEAK +f7_t* f7_fmax (f7_t *cc, const f7_t *aa, const f7_t *bb) +{ + return f7_minmax (cc, aa, bb, false); +} +ALIAS (f7_fmax, f7_max) +#endif // F7MOD_fmax_ + + +#ifdef F7MOD_fmin_ +F7_WEAK +f7_t* f7_fmin (f7_t *cc, const f7_t *aa, const f7_t *bb) +{ + return f7_minmax (cc, aa, bb, true); +} +ALIAS (f7_fmin, f7_min) +#endif // F7MOD_fmin_ + + +#ifdef F7MOD_mulx_ +F7_WEAK +uint8_t f7_mulx (f7_t *cc, const f7_t *aa, const f7_t *bb, bool no_rounding) +{ + uint8_t a_class = f7_classify (aa); + uint8_t b_class = f7_classify (bb); + // From this point on, no more access aa->flags or bb->flags + // to avoid early-clobber when writing cc->flags. + + uint8_t ab_class = a_class | b_class; + // If either value is NaN, return NaN. + if (f7_class_nan (ab_class) + // Any combination of Inf and 0. + || (f7_class_zero (ab_class) && f7_class_inf (ab_class))) + { + cc->flags = F7_FLAG_nan; + return 0; + } + // If either value is 0.0, return 0.0. + if (f7_class_zero (ab_class)) + { + f7_clr (cc); + return 0; + } + // We have 2 non-zero numbers-or-INF. + + uint8_t c_sign = (a_class ^ b_class) & F7_FLAG_sign; + uint8_t c_inf = ab_class & F7_FLAG_inf; + cc->flags = c_sign | c_inf; + if (c_inf) + return 0; + + int16_t expo = add_ssat16 (aa->expo, bb->expo); + // Store expo and handle expo = INT16_MIN and INT16_MAX. + if (f7_store_expo (cc, expo)) + return 0; + + return f7_mul_mant_asm (cc, aa, bb, no_rounding); +} +#endif // F7MOD_mulx_ + + +#ifdef F7MOD_square_ +F7_WEAK +void f7_square (f7_t *cc, const f7_t *aa) +{ + f7_mul (cc, aa, aa); +} +#endif // F7MOD_square_ + + +#ifdef F7MOD_mul_ +F7_WEAK +void f7_mul (f7_t *cc, const f7_t *aa, const f7_t *bb) +{ + f7_mulx (cc, aa, bb, false); +} +#endif // F7MOD_mul_ + + +#ifdef F7MOD_Iadd_ +F7_WEAK void f7_Iadd (f7_t *cc, const f7_t *aa) { f7_add (cc, cc, aa); } +#endif + +#ifdef F7MOD_Isub_ +F7_WEAK void f7_Isub (f7_t *cc, const f7_t *aa) { f7_sub (cc, cc, aa); } +#endif + +#ifdef F7MOD_Imul_ +F7_WEAK void f7_Imul (f7_t *cc, const f7_t *aa) { f7_mul (cc, cc, aa); } +#endif + +#ifdef F7MOD_Idiv_ +F7_WEAK void f7_Idiv (f7_t *cc, const f7_t *aa) { f7_div (cc, cc, aa); } +#endif + +#ifdef F7MOD_IRsub_ +F7_WEAK void f7_IRsub (f7_t *cc, const f7_t *aa) { f7_sub (cc, aa, cc); } +#endif + +#ifdef F7MOD_Ineg_ +F7_WEAK void f7_Ineg (f7_t *cc) { f7_neg (cc, cc); } +#endif + +#ifdef F7MOD_Isqrt_ +F7_WEAK void f7_Isqrt (f7_t *cc) { f7_sqrt (cc, cc); } +#endif + +#ifdef F7MOD_Isquare_ +F7_WEAK void f7_Isquare (f7_t *cc) { f7_square (cc, cc); } +#endif + +#ifdef F7MOD_Ildexp_ +F7_WEAK f7_t* f7_Ildexp (f7_t *cc, int ex) { return f7_ldexp (cc, cc, ex); } +#endif + + +#ifdef F7MOD_add_ +F7_WEAK +void f7_add (f7_t *cc, const f7_t *aa, const f7_t *bb) +{ + f7_addsub (cc, aa, bb, false); +} +#endif // F7MOD_add_ + + +#ifdef F7MOD_sub_ +F7_WEAK +void f7_sub (f7_t *cc, const f7_t *aa, const f7_t *bb) +{ + f7_addsub (cc, aa, bb, true); +} +#endif // F7MOD_sub_ + + +#ifdef F7MOD_addsub_ +static void return_with_sign (f7_t *cc, const f7_t *aa, int8_t c_sign) +{ + __asm (";;; return with sign"); + f7_copy (cc, aa); + if (c_sign != -1) + f7_set_sign (cc, c_sign); +} + +F7_WEAK +void f7_addsub (f7_t *cc, const f7_t *aa, const f7_t *bb, bool neg_b) +{ + uint8_t a_class = f7_classify (aa); + uint8_t b_class = f7_classify (bb); + // From this point on, no more access aa->flags or bb->flags + // to avoid early-clobber when writing cc->flags. + + // Hande NaNs. + if (f7_class_nan (a_class | b_class)) + return f7_set_nan (cc); + + bool a_sign = f7_class_sign (a_class); + bool b_sign = f7_class_sign (b_class) ^ neg_b; + + // Add the mantissae? + bool do_add = a_sign == b_sign; + + // Handle +Infs and -Infs. + bool a_inf = f7_class_inf (a_class); + bool b_inf = f7_class_inf (b_class); + + if (a_inf && b_inf) + { + if (do_add) + return f7_set_inf (cc, a_sign); + else + return f7_set_nan (cc); + } + else if (a_inf) + return f7_set_inf (cc, a_sign); + else if (b_inf) + return f7_set_inf (cc, b_sign); + + int16_t shift16 = sub_ssat16 (aa->expo, bb->expo); + + // aa + 0 = aa. + // Also check MSBit to get rid of Subnormals and 0. + if (shift16 > F7_MANT_BITS || f7_is0 (bb)) + return return_with_sign (cc, aa, -1); + + // 0 + bb = bb. + // 0 - bb = -bb. + // Also check MSBit to get rid of Subnormals and 0. + if (shift16 < -F7_MANT_BITS || f7_is0 (aa)) + return return_with_sign (cc, bb, b_sign); + + // Now aa and bb are non-zero, non-NaN, non-Inf. + // shift > 0 ==> |a| > |b| + // shift < 0 ==> |a| < |b| + int8_t shift = (int8_t) shift16; + bool c_sign = a_sign; + if (shift < 0 + || (shift == 0 && f7_cmp_mant (aa, bb) < 0)) + { + const f7_t *p = aa; aa = bb; bb = p; + c_sign = b_sign; + shift = -shift; + } + uint8_t shift2 = (uint8_t) (shift << 1); + + cc->expo = aa->expo; + // From this point on, no more access aa->expo or bb->expo + // to avoid early-clobber when writing cc->expo. + + cc->flags = c_sign; _Static_assert (F7_FLAGNO_sign == 0, ""); + + // This function uses neither .expo nor .flags from either aa or bb, + // hence there is early-clobber for cc->expo and cc->flags. + f7_addsub_mant_scaled_asm (cc, aa, bb, shift2 | do_add); +} +#endif // F7MOD_addsub_ + + +#ifdef F7MOD_madd_msub_ +F7_WEAK +void f7_madd_msub (f7_t *cc, const f7_t *aa, const f7_t *bb, const f7_t *dd, + bool neg_d) +{ + f7_t xx7, *xx = &xx7; + uint8_t x_lsb = f7_mulx (xx, aa, bb, true /* no rounding */); + uint8_t x_sign = f7_signbit (xx); + int16_t x_expo = xx->expo; + f7_addsub (xx, xx, dd, neg_d); + // Now add LSB. If cancellation occured in the add / sub, then we have the + // chance of extra 8 bits of precision. Turn LSByte into f7_t. + f7_clr (cc); + cc->expo = sub_ssat16 (x_expo, F7_MANT_BITS); + cc->mant[F7_MANT_BYTES - 1] = x_lsb; + cc = f7_normalize_asm (cc); + cc->flags = x_sign; + f7_Iadd (cc, xx); +} +#endif // F7MOD_madd_msub_ + +#ifdef F7MOD_madd_ +F7_WEAK +void f7_madd (f7_t *cc, const f7_t *aa, const f7_t *bb, const f7_t *dd) +{ + f7_madd_msub (cc, aa, bb, dd, false); +} +#endif // F7MOD_madd_ + +#ifdef F7MOD_msub_ +F7_WEAK +void f7_msub (f7_t *cc, const f7_t *aa, const f7_t *bb, const f7_t *dd) +{ + f7_madd_msub (cc, aa, bb, dd, true); +} +#endif // F7MOD_msub_ + + +#ifdef F7MOD_ldexp_ +F7_WEAK +f7_t* f7_ldexp (f7_t *cc, const f7_t *aa, int delta) +{ + uint8_t a_class = f7_classify (aa); + + cc->flags = a_class; + + // Inf and NaN. + if (! f7_class_number (a_class)) + return cc; + + if (f7_msbit (aa) == 0) + return (f7_clr (cc), cc); + + int16_t expo = add_ssat16 (delta, aa->expo); + // Store expo and handle expo = INT16_MIN and INT16_MAX. + if (! f7_store_expo (cc, expo)) + f7_copy_mant (cc, aa); + + return cc; +} +#endif // F7MOD_ldexp_ + + +#if USE_LPM +#elif USE_LD +#else +#error need include "asm-defs.h" +#endif // USE_LPM + +/* + Handling constants: + + F7_PCONST (PVAR, X) + + Set f7_t [const] *PVAR to an LD address for one + of the f7_const_X[_P] constants. + PVAR might be set to point to a local auto that serves + as temporary storage for f7_const_X_P. PVAR is only + valid in the current block. + + const f7_t* F7_PCONST_U16 (PVAR, <ident> X) // USE_LD + f7_t* F7_PCONST_U16 (PVAR, uint16_t X) // USE_LPM + + Set f7_t [const] *PVAR to an LD address for one of the + f7_const_X[_P] constants. PVAR might be set to point to a + local auto that serves as temporary storage for X. PVAR is + only valid in the current block. + + F7_PCONST_VAR (PVAR, VAR) + + VAR is a pointer variable holding the address of some f7_const_X[_P] + constant. Set [const] f7_t *PVAR to a respective LD address. + PVAR might be set to point to a local auto that serves + as temporary storage for f7_const_X_P. PVAR is only + valid in the current block. + + F7_CONST_ADDR (<ident> CST, f7_t* PTMP) + + Return an LD address to for some f7_const_X[_P] constant. + *PTMP might be needed to hold a copy of f7_const_X_P in RAM. + + f7_t* F7_U16_ADDR (uint16_t X, f7_t* PTMP) // USE_LPM + const f7_t* F7_U16_ADDR (<cst-ident> X, <unused>) // USE_LD + + Return an LD address to compile-time constant uint16_t X which is + also known as f7_const_X[_P]. *PTMP might be set to (f7_t) X. + + f7_t* f7_const (f7_t *PVAR, <cst-ident> X) + + Copy f7_const_X[_P] to *PVAR. + + f7_t* f7_copy_flash (f7_t *DST, const f7_t *SRC) + + Copy to *DST with LD (from .rodata in flash) if the address + space is linear, or with LPM (from .progmem.data) if the + address space is not linear. + + f7_t* f7_copy (f7_t *DST, const f7_t* SRC) + + Copy to RAM using LD. + + f7_t* f7_copy_P (f7_t *DST, const f7_t *SRC) + + Copy to RAM using LPM. +*/ + +#if USE_LPM + #define F7_RAW_CONST_ADDR(CST) \ + & F7_(const_##CST##_P) + + #define F7_PCONST(PVAR, CST) \ + f7_t _var_for_##CST; \ + f7_copy_P (& _var_for_##CST, & F7_(const_##CST##_P)); \ + PVAR = & _var_for_##CST + + #define F7_PCONST_U16(PVAR, CST) \ + f7_t _var_for_##CST; \ + PVAR = f7_set_u16 (& _var_for_##CST, CST) + + #define F7_PCONST_VAR(PVAR, VAR) \ + f7_t _var_for_##VAR; \ + f7_copy_P (& _var_for_##VAR, VAR); \ + PVAR = & _var_for_##VAR + + #define MAYBE_const // empty + + #define F7_CONST_ADDR(CST, PTMP) \ + f7_copy_P ((PTMP), & F7_(const_##CST##_P)) + + #define F7_U16_ADDR(CST, PTMP) \ + f7_set_u16 ((PTMP), CST) + +#elif USE_LD + #define F7_RAW_CONST_ADDR(CST) \ + & F7_(const_##CST) + + #define F7_PCONST(PVAR, CST) \ + PVAR = & F7_(const_##CST) + + #define F7_PCONST_U16(PVAR, CST) \ + PVAR = & F7_(const_##CST) + + #define F7_PCONST_VAR(PVAR, VAR) \ + PVAR = (VAR) + + #define F7_CONST_ADDR(CST, PTMP) \ + (& F7_(const_##CST)) + + #define F7_U16_ADDR(CST, PTMP) \ + (& F7_(const_##CST)) + + #define MAYBE_const const +#endif + + + +#define DD(str, X) \ + do { \ + LOG_PSTR (PSTR (str)); \ + f7_dump (X); \ + } while (0) + +#undef DD +#define DD(...) (void) 0 + + +#ifdef F7MOD_sqrt_ +static void sqrt_worker (f7_t *cc, const f7_t *rr) +{ + f7_t tmp7, *tmp = &tmp7; + f7_t aa7, *aa = &aa7; + + // aa in [1/2, 2) => aa->expo in { -1, 0 }. + int16_t a_expo = -(rr->expo & 1); + int16_t c_expo = (rr->expo - a_expo) >> 1; // FIXME: r_expo = INT_MAX??? + + __asm ("" : "+r" (aa)); + + f7_copy (aa, rr); + aa->expo = a_expo; + + // No use of rr or *cc past this point: We may use cc as temporary. + // Approximate square-root of A by X <-- (X + A / X) / 2. + + f7_sqrt_approx_asm (cc, aa); + + // Iterate X <-- (X + A / X) / 2. + // 3 Iterations with 16, 32, 58 bits of precision for the quotient. + + for (uint8_t prec = 16; (prec & 0x80) == 0; prec <<= 1) + { + f7_divx (tmp, aa, cc, (prec & 64) ? 2 + F7_MANT_BITS : prec); + f7_Iadd (cc, tmp); + // This will never underflow because |c_expo| is small. + cc->expo--; + } + + // Similar: |c_expo| is small, hence no ldexp needed. + cc->expo += c_expo; +} + +F7_WEAK +void f7_sqrt (f7_t *cc, const f7_t *aa) +{ + uint8_t a_class = f7_classify (aa); + + if (f7_class_nan (a_class) || f7_class_sign (a_class)) + return f7_set_nan (cc); + + if (f7_class_inf (a_class)) + return f7_set_inf (cc, 0); + + if (f7_class_zero (a_class)) + return f7_clr (cc); + + sqrt_worker (cc, aa); +} +#endif // F7MOD_sqrt_ + + +#ifdef F7MOD_hypot_ +F7_WEAK +void f7_hypot (f7_t *cc, const f7_t *aa, const f7_t *bb) +{ + f7_t xx7, *xx = &xx7; + + f7_square (xx, aa); + f7_square (cc, bb); + f7_Iadd (cc, xx); + f7_Isqrt (cc); +} +#endif // F7MOD_hypot_ + + +#ifdef F7MOD_const_m1_ +#include "libf7-constdef.h" +#endif // -1 + +#ifdef F7MOD_const_1_2_ +#include "libf7-constdef.h" +#endif // 1/2 + +#ifdef F7MOD_const_1_3_ +#include "libf7-constdef.h" +#endif // 1/3 + +#ifdef F7MOD_const_ln2_ +#include "libf7-constdef.h" +#endif // ln2 + +#ifdef F7MOD_const_1_ln2_ +#include "libf7-constdef.h" +#endif // 1_ln2 + +#ifdef F7MOD_const_ln10_ +#include "libf7-constdef.h" +#endif // ln10 + +#ifdef F7MOD_const_1_ln10_ +#include "libf7-constdef.h" +#endif // 1_ln10 + +#ifdef F7MOD_const_1_ +#include "libf7-constdef.h" +#endif // 1 + +#ifdef F7MOD_const_sqrt2_ +#include "libf7-constdef.h" +#endif // sqrt2 + +#ifdef F7MOD_const_2_ +#include "libf7-constdef.h" +#endif // 2 + +#ifdef F7MOD_const_pi_ +#include "libf7-constdef.h" +#endif // pi + + +#ifdef F7MOD_divx_ + +// C /= A +extern void f7_div_asm (f7_t*, const f7_t*, uint8_t); + +F7_WEAK +void f7_divx (f7_t *cc, const f7_t *aa, const f7_t *bb, uint8_t quot_bits) +{ + uint8_t a_class = f7_classify (aa); + uint8_t b_class = f7_classify (bb); + // From this point on, no more access aa->flags or bb->flags + // to avoid early-clobber when writing cc->flags. + + // If either value is NaN, return NaN. + if (f7_class_nan (a_class | b_class) + // If both values are Inf or both are 0, return NaN. + || f7_class_zero (a_class & b_class) + || f7_class_inf (a_class & b_class) + // Inf / 0 = NaN. + || (f7_class_inf (a_class) && f7_class_zero (b_class))) + { + return f7_set_nan (cc); + } + + // 0 / B = 0 for non-zero, non-NaN B. + // A / Inf = 0 for non-zero numbers A. + if (f7_class_zero (a_class) || f7_class_inf (b_class)) + return f7_clr (cc); + + uint8_t c_sign = (a_class ^ b_class) & F7_FLAG_sign; + + if (f7_class_inf (a_class) || f7_class_zero (b_class)) + return f7_set_inf (cc, c_sign); + + cc->flags = c_sign; _Static_assert (F7_FLAGNO_sign == 0, ""); + int16_t expo = sub_ssat16 (aa->expo, bb->expo); + + // Store expo and handle expo = INT16_MIN and INT16_MAX. + if (f7_store_expo (cc, expo)) + return; + + f7_t ss7, *ss = &ss7; + ss->flags = cc->flags; + ss->expo = cc->expo; + + f7_copy_mant (ss, aa); + f7_div_asm (ss, bb, quot_bits); + f7_copy (cc, ss); +} +#endif // F7MOD_divx_ + + +#ifdef F7MOD_div_ +F7_WEAK +void f7_div (f7_t *cc, const f7_t *aa, const f7_t *bb) +{ + /* When f7_divx calls f7_div_asm, dividend and divisor are valid + mantissae, i.e. their MSBit is set. Therefore, the quotient will + be in [0x0.ff..., 0x0.40...] and to adjust it, at most 1 left-shift + is needed. Compute F7_MANT_BITS + 2 bits of the quotient: + One bit is used for rounding, and one bit might be consumed by the + mentioned left-shift. */ + + f7_divx (cc, aa, bb, 2 + F7_MANT_BITS); +} +#endif // F7MOD_div_ + + +#ifdef F7MOD_div1_ +F7_WEAK +void f7_div1 (f7_t *cc, const f7_t *aa) +{ + F7_PCONST_U16 (const f7_t *one, 1); + f7_div (cc, one, aa); +} +#endif // F7MOD_div_ + + +#ifdef F7MOD_fmod_ +F7_WEAK +void f7_fmod (f7_t *cc, const f7_t *aa, const f7_t *bb) +{ + uint8_t a_class = f7_classify (aa); + uint8_t b_class = f7_classify (bb); + + if (! f7_class_number (a_class) + || f7_class_nan (b_class) + || f7_class_zero (b_class)) + { + return f7_set_nan (cc); + } + + // A == 0 and B != 0 => 0. + if (f7_class_zero (a_class)) + return f7_clr (cc); + + f7_t zz7, *zz = & zz7; + + f7_div (zz, aa, bb); + + // Z in Z, |Z| <= |A/B|. + f7_trunc (zz, zz); + + // C = A - Z * B. + f7_msub (cc, zz, bb, aa); + cc->flags ^= F7_FLAG_sign; +} +#endif // F7MOD_fmod_ + + +#ifdef F7MOD_truncx_ +F7_WEAK +f7_t* f7_truncx (f7_t *cc, const f7_t *aa, bool do_floor) +{ + uint8_t a_class = f7_classify (aa); + + if (! f7_class_nonzero (a_class)) + return f7_copy (cc, aa); + + bool sign = f7_class_sign (a_class); + + int16_t a_expo = aa->expo; + + if (a_expo < 0) + { + // |A| < 1. + if (sign & do_floor) + return f7_set_s16 (cc, -1); + + f7_clr (cc); + return cc; + } + else if (a_expo >= F7_MANT_BITS - 1) + // A == floor (A). + return f7_copy (cc, aa); + + f7_t tmp7, *tmp = &tmp7; + + // Needed if aa === cc. + f7_copy (tmp, aa); + + cc->flags = sign; + cc->expo = a_expo; + f7_clr_mant_lsbs (cc, aa, F7_MANT_BITS - 1 - a_expo); + + if (do_floor && cc->sign && f7_cmp_mant (cc, tmp) != 0) + { + F7_PCONST_U16 (const f7_t *one, 1); + f7_Isub (cc, one); + } + + return cc; +} +#endif // F7MOD_truncx_ + + +#ifdef F7MOD_floor_ +F7_WEAK +f7_t* f7_floor (f7_t *cc, const f7_t *aa) +{ + return f7_truncx (cc, aa, true); +} +#endif // F7MOD_floor_ + + +#ifdef F7MOD_trunc_ +F7_WEAK +f7_t* f7_trunc (f7_t *cc, const f7_t *aa) +{ + return f7_truncx (cc, aa, false); +} +#endif // F7MOD_trunc_ + + +#ifdef F7MOD_ceil_ +F7_WEAK +void f7_ceil (f7_t *cc, const f7_t *aa) +{ + cc = f7_copy (cc, aa); + cc->flags ^= F7_FLAG_sign; + cc = f7_floor (cc, cc); + f7_Ineg (cc); +} +#endif // F7MOD_ceil_ + + +#ifdef F7MOD_round_ +F7_WEAK +void f7_round (f7_t *cc, const f7_t *aa) +{ + f7_t tmp; + (void) tmp; + const f7_t *half = F7_CONST_ADDR (1_2, &tmp); + + f7_addsub (cc, aa, half, f7_signbit (aa)); + f7_trunc (cc, cc); +} +#endif // F7MOD_round_ + + +#ifdef F7MOD_horner_ + +// Assertion when using this function is that either cc != xx, +// or if cc == xx, then tmp1 must be non-NULL and tmp1 != xx. +// In General, the calling functions have a spare f7_t object available +// and can pass it down to save some stack. +// Moreover, the power series must have degree 1 at least. + +F7_WEAK +void f7_horner (f7_t *cc, const f7_t *xx, uint8_t n_coeff, const f7_t *coeff, + f7_t *tmp1) +{ + f7_assert (n_coeff > 1); + + if (cc != xx) + tmp1 = cc; + else + f7_assert (tmp1 != NULL && tmp1 != xx); + + f7_t *yy = tmp1; + f7_t tmp27, *tmp2 = &tmp27; + + n_coeff--; + const f7_t *pcoeff = coeff + n_coeff; + + f7_copy_flash (yy, pcoeff); + + while (1) + { + --pcoeff; +#if 1 + f7_Imul (yy, xx); + const f7_t *cst = USE_LD ? pcoeff : f7_copy_P (tmp2, pcoeff); + if (coeff == pcoeff) + return f7_add (cc, yy, cst); + f7_Iadd (yy, cst); +#else + const f7_t *cst = USE_LD ? pcoeff : f7_copy_P (tmp2, pcoeff); + f7_madd (yy, yy, xx, cst); + if (coeff == pcoeff) + { + f7_copy (cc, yy); + return; + } +#endif + } + + __builtin_unreachable(); +} +#endif // F7MOD_horner_ + + +#ifdef F7MOD_log_ +F7_WEAK +void f7_log (f7_t *cc, const f7_t *aa) +{ + f7_logx (cc, aa, NULL); +} +#endif // F7MOD_log_ + + +#ifdef F7MOD_log2_ +F7_WEAK +void f7_log2 (f7_t *cc, const f7_t *aa) +{ + f7_logx (cc, aa, F7_RAW_CONST_ADDR (1_ln2)); +} +#endif // F7MOD_log2_ + + +#ifdef F7MOD_log10_ +F7_WEAK +void f7_log10 (f7_t *cc, const f7_t *aa) +{ + f7_logx (cc, aa, F7_RAW_CONST_ADDR (1_ln10)); +} +#endif // F7MOD_log10_ + + +#ifdef F7MOD_logx_ + +#define ARRAY_NAME coeff_artanh +#include "libf7-array.def" +#undef ARRAY_NAME + +// Compute P * ln(A) if P != NULL and ln(A), otherwise. +// P is a LD-address if USE_LD and a LPM-address if USE_LPM. +// Assumption is that P > 0. + +F7_WEAK +void f7_logx (f7_t *cc, const f7_t *aa, const f7_t *p) +{ + uint8_t a_class = f7_classify (aa); + + if (f7_class_nan (a_class) || f7_class_sign (a_class)) + return f7_set_nan (cc); + + if (f7_class_inf (a_class)) + return f7_set_inf (cc, 0); + + if (f7_class_zero (a_class)) + return f7_set_inf (cc, 1); + + f7_t *yy = cc; + f7_t xx7, *xx = &xx7; + f7_t tmp7, *tmp = &tmp7; + + // Y in [1, 2] = A * 2 ^ (-a_expo). + int16_t a_expo = aa->expo; + f7_copy (yy, aa); + yy->expo = 0; + + // Y in [1 / sqrt2, sqrt2]. + + if (f7_abscmp_msb_ge (yy, F7_(const_sqrt2_msb), F7_(const_sqrt2_expo))) + { + yy->expo = -1; + a_expo = add_ssat16 (a_expo, 1); + } + + const f7_t *one = F7_U16_ADDR (1, & tmp7); + + // X := (Y - 1) / (Y + 1), |X| <= (sqrt2 - 1) / (sqrt2 + 1) ~ 0.172. + f7_sub (xx, yy, one); + f7_Iadd (yy, one); + f7_Idiv (xx, yy); + + // Y := X^2, |Y| < 0.03. + f7_square (yy, xx); + + // Y := artanh (X^2) / X + f7_horner (yy, yy, n_coeff_artanh, coeff_artanh, tmp); + + // C = X * Y = ln A - a_expo * ln2. + f7_mul (cc, xx, yy); + + // X := a_expo * ln2. + f7_set_s16 (xx, a_expo); + f7_Imul (xx, F7_CONST_ADDR (ln2, & tmp7)); + + // C = ln A. + f7_Iadd (cc, xx); + + if (p && USE_LPM) + f7_Imul (cc, f7_copy_P (tmp, p)); + if (p && USE_LD) + f7_Imul (cc, p); +} +#endif // F7MOD_logx_ + + +#ifdef F7MOD_exp_ + +#define ARRAY_NAME coeff_exp +#include "libf7-array.def" +#undef ARRAY_NAME + +#define STATIC static +#include "libf7-constdef.h" // ln2_low +#undef STATIC + +F7_WEAK +void f7_exp (f7_t *cc, const f7_t *aa) +{ + uint8_t a_class = f7_classify (aa); + + if (f7_class_nan (a_class)) + return f7_set_nan (cc); + + /* The maximal exponent of 2 for a double is 1023, hence we may limit + to |A| < 1023 * ln2 ~ 709. We limit to 1024 ~ 1.99 * 2^9 */ + + if (f7_class_inf (a_class) + || (f7_class_nonzero (a_class) && aa->expo >= 9)) + { + if (f7_class_sign (a_class)) + return f7_clr (cc); + else + return f7_set_inf (cc, 0); + } + + f7_t const *cst; + f7_t qq7, *qq = &qq7; + + F7_PCONST (cst, ln2); + + // We limited |A| to 1024 and are now dividing by ln2, hence Q will + // be at most 1024 / ln2 ~ 1477 and fit into 11 bits. We will + // round Q anyway, hence only request 11 bits from the division and + // one additional bit that might be needed to normalize the quotient. + f7_divx (qq, aa, cst, 1 + 11); + + // Use the smallest (by absolute value) remainder system. + f7_round (qq, qq); + int16_t q = f7_get_s16 (qq); + + // Reducing A mod ln2 gives |C| <= ln2 / 2, C = -A mod ln2. + f7_msub (cc, qq, cst, aa); + + // Corrigendum: We added Q * ln2; now add Q times the low part of ln2 + // for better precision. Due to |C| < |A| this is not a no-op in general. + const f7_t *yy = F7_CONST_ADDR (ln2_low, &_var_for_ln2); + f7_madd (cc, qq, yy, cc); + + // Because we computed C = -A mod ... + cc->flags ^= F7_FLAG_sign; + + // Reduce further to |C| < ln2 / 8 which is the range of our MiniMax poly. + const uint8_t MAX_LN2_RED = 3; + int8_t scal2 = 0; + + while (f7_abscmp_msb_ge (cc, F7_(const_ln2_msb), + F7_(const_ln2_expo) - MAX_LN2_RED)) + { + scal2++; + cc->expo--; + } + + f7_horner (cc, cc, n_coeff_exp, coeff_exp, qq); + + while (--scal2 >= 0) + f7_Isquare (cc); + + f7_Ildexp (cc, q); +} +#endif // F7MOD_exp_ + + +#ifdef F7MOD_pow10_ +F7_WEAK +void f7_pow10 (f7_t *cc, const f7_t *aa) +{ + const f7_t *p_ln10; + F7_PCONST (p_ln10, ln10); + f7_mul (cc, aa, p_ln10); + f7_exp (cc, cc); +} +ALIAS (f7_pow10, f7_exp10) +#endif // F7MOD_pow10_ + + +#ifdef F7MOD_cbrt_ +F7_WEAK +void f7_cbrt (f7_t *cc, const f7_t *aa) +{ + f7_copy (cc, aa); + const f7_t *p_1_3; + uint8_t c_flags = cc->flags; + cc->flags &= ~F7_FLAG_sign; + f7_log (cc, cc); + F7_PCONST (p_1_3, 1_3); + f7_Imul (cc, p_1_3); + f7_exp (cc, cc); + + if (c_flags & F7_FLAG_sign) + cc->flags |= F7_FLAG_sign; +} +#endif // F7MOD_cbrt_ + + +#ifdef F7MOD_pow_ +F7_WEAK +void f7_pow (f7_t *cc, const f7_t *aa, const f7_t *bb) +{ +#if 0 + f7_t slots[cc == bb]; + f7_t *yy = cc == bb ? slots : cc; +#else + f7_t yy7, *yy = &yy7; +#endif + f7_log (yy, aa); + f7_Imul (yy, bb); + f7_exp (cc, yy); +} +#endif // F7MOD_pow_ + + +#ifdef F7MOD_powi_ +F7_WEAK +void f7_powi (f7_t *cc, const f7_t *aa, int ii) +{ + uint16_t u16 = ii; + f7_t xx27, *xx2 = &xx27; + + if (ii < 0) + u16 = -u16; + + f7_copy (xx2, aa); + + f7_set_u16 (cc, 1); + + while (1) + { + if (u16 & 1) + f7_Imul (cc, xx2); + + if (! f7_is_nonzero (cc)) + break; + + u16 >>= 1; + if (u16 == 0) + break; + f7_Isquare (xx2); + } + + if (ii < 0) + f7_div1 (xx2, aa); +} +#endif // F7MOD_powi_ + + +#ifdef F7MOD_sincos_ + +#define ARRAY_NAME coeff_sin + #define FOR_SIN + #include "libf7-array.def" + #undef FOR_SIN +#undef ARRAY_NAME + +#define ARRAY_NAME coeff_cos + #define FOR_COS + #include "libf7-array.def" + #undef FOR_COS +#undef ARRAY_NAME + +#define STATIC static +#include "libf7-constdef.h" // pi_low +#undef STATIC + +typedef union +{ + struct + { + bool neg_sin : 1; // Must be bit F7_FLAGNO_sign. + bool neg_cos : 1; + bool do_sin: 1; + bool do_cos: 1; + bool swap_sincos : 1; + uint8_t res : 3; + }; + uint8_t bits; +} sincos_t; + + +F7_WEAK +void f7_sincos (f7_t *ss, f7_t *cc, const f7_t *aa) +{ + uint8_t a_class = f7_classify (aa); + + sincos_t sc = { .bits = a_class & F7_FLAG_sign }; + if (ss != NULL) sc.do_sin = 1; + if (cc != NULL) sc.do_cos = 1; + + if (f7_class_nan (a_class) || f7_class_inf (a_class)) + { + if (sc.do_sin) f7_set_nan (ss); + if (sc.do_cos) f7_set_nan (cc); + return; + } + + f7_t pi7, *pi = &pi7; + f7_t xx7, *xx = &xx7; + f7_t yy7, *yy = &yy7; + f7_t *hh = sc.do_sin ? ss : cc; + + // X = |A| + f7_copy (xx, aa); + xx->flags = 0; + + // Y is how often we subtract PI from X. + f7_clr (yy); + f7_const (pi, pi); + + if (f7_abscmp_msb_ge (xx, F7_(const_pi_msb), F7_(const_pi_expo) + 1)) + { + pi->expo = 1 + F7_(const_pi_expo); // 2*pi + + // Y = X / 2pi. + f7_div (yy, xx, pi); + + // The integral part of |A| / pi mod 2 is bit 55 - x_expo. + if (yy->expo >= F7_MANT_BITS && !f7_is_zero (yy)) + { + // Too big for sensible calculation: Should this be NaN instead? + if (sc.do_sin) f7_clr (ss); + if (sc.do_cos) f7_clr (cc); + return; + } + + // X -= 2pi * [ X / 2pi ] + f7_floor (yy, yy); + + f7_msub (xx, yy, pi, xx); + xx->flags ^= F7_FLAG_sign; + + // We divided by 2pi, but Y should count times we subtracted pi. + yy->expo++; + } + + pi->expo = F7_(const_pi_expo); // pi + f7_sub (hh, xx, pi); + if (!f7_signbit (hh)) + { + // H = X - pi >= 0 => X >= pi + // sin(x) = -sin(x - pi) + // cos(x) = -cos(x - pi) + f7_copy (xx, hh); + // Y: We subtracted pi one more time. + f7_Iadd (yy, f7_set_u16 (hh, 1)); + sc.neg_sin ^= 1; + sc.neg_cos ^= 1; + } + + pi->expo = F7_(const_pi_expo) - 1; // pi/2 + if (f7_gt (xx, pi)) + { + // x > pi/2 + // sin(x) = sin(pi - x) + // cos(x) = -cos(pi - x) + pi->expo = F7_(const_pi_expo); // pi + f7_IRsub (xx, pi); + // Y: We subtracted pi one more time (and then negated). + f7_Iadd (yy, f7_set_u16 (hh, 1)); + yy->flags ^= F7_FLAG_sign; + sc.neg_cos ^= 1; + } + + pi->expo = F7_(const_pi_expo) - 2; // pi/4 + if (f7_gt (xx, pi)) + { + // x > pi/4 + // sin(x) = cos(pi/2 - x) + // cos(x) = sin(pi/2 - x) + pi->expo = F7_(const_pi_expo) - 1; // pi/2 + f7_IRsub (xx, pi); + // Y: We subtracted pi/2 one more time (and then negated). + f7_Iadd (yy, f7_set_1pow2 (hh, -1, 0)); + yy->flags ^= F7_FLAG_sign; + sc.swap_sincos = 1; + } + + if (!f7_is0 (yy)) + { + // Y counts how often we subtracted pi from X in order to + // get 0 <= X < pi/4 as small as possible (Y is 0 mod 0.5). + // Now also subtract the low part of pi: + // f7_const_pi_low = pi - f7_const_pi in order to get more precise + // results in the cases where the final result is close to 0. + const f7_t *pi_low = F7_CONST_ADDR (pi_low, pi); + //f7_const (pi, pi_low); + f7_Imul (yy, pi_low); + f7_Isub (xx, yy); + } + + // X in [0, pi/4]. + // X^2 in [0, pi^2/16] ~ [0, 0.6169] + + f7_square (yy, xx); + + f7_t *x_sin = xx; + f7_t *x_cos = yy; + + if ((sc.do_sin && !sc.swap_sincos) + || (sc.do_cos && sc.swap_sincos)) + { + f7_horner (hh, yy, n_coeff_sin, coeff_sin, NULL); + f7_mul (x_sin, hh, xx); + } + + if ((sc.do_cos && !sc.swap_sincos) + || (sc.do_sin && sc.swap_sincos)) + { + f7_horner (x_cos, yy, n_coeff_cos, coeff_cos, hh); + } + + if (sc.swap_sincos) + { + x_sin = yy; + x_cos = xx; + } + + if (sc.do_sin) + { + x_sin->flags ^= sc.bits; + x_sin->flags &= F7_FLAG_sign; + f7_copy (ss, x_sin); + } + + if (sc.do_cos) + { + x_cos->flags = sc.neg_cos; + f7_copy (cc, x_cos); + } +} +#endif // F7MOD_sincos_ + +#ifdef F7MOD_sin_ +F7_WEAK +void f7_sin (f7_t *ss, const f7_t *aa) +{ + f7_sincos (ss, NULL, aa); +} +#endif // F7MOD_sin_ + +#ifdef F7MOD_cos_ +F7_WEAK +void f7_cos (f7_t *cc, const f7_t *aa) +{ + f7_sincos (NULL, cc, aa); +} +#endif // F7MOD_cos_ + + +#ifdef F7MOD_tan_ +F7_WEAK +void f7_tan (f7_t *tt, const f7_t *aa) +{ + f7_t xcos; + f7_sincos (tt, & xcos, aa); + f7_Idiv (tt, & xcos); +} +#endif // F7MOD_tan_ + + +#ifdef F7MOD_cotan_ +F7_WEAK +void f7_cotan (f7_t *ct, const f7_t *aa) +{ + f7_t xcos; + f7_sincos (ct, & xcos, aa); + f7_div (ct, & xcos, ct); +} +#endif // F7MOD_cotan_ + + +#ifdef F7MOD_sinhcosh_ +F7_WEAK +void f7_sinhcosh (f7_t *cc, const f7_t *aa, bool do_sinh) +{ + f7_t xx7, *xx = &xx7; + // C = exp(A) + f7_exp (cc, aa); + // X = exp(-A) + f7_div (xx, f7_set_u16 (xx, 1), cc); + // sinh(A) = (exp(A) - exp(-A)) / 2 + // cosh(A) = (exp(A) + exp(-A)) / 2 + f7_addsub (cc, cc, xx, do_sinh); + cc->expo--; +} +#endif // F7MOD_sinhcosh_ + + +#ifdef F7MOD_sinh_ +F7_WEAK +void f7_sinh (f7_t *cc, const f7_t *aa) +{ + f7_sinhcosh (cc, aa, true); +} +#endif // F7MOD_sinh_ + + +#ifdef F7MOD_cosh_ +F7_WEAK +void f7_cosh (f7_t *cc, const f7_t *aa) +{ + f7_sinhcosh (cc, aa, false); +} +#endif // F7MOD_cosh_ + + +#ifdef F7MOD_tanh_ +F7_WEAK +void f7_tanh (f7_t *cc, const f7_t *aa) +{ + // tanh(A) = (exp(2A) - 1) / (exp(2A) + 1) + f7_t xx7, *xx = &xx7; + F7_PCONST_U16 (const f7_t *one, 1); + // C = 2A + f7_copy (cc, aa); + cc->expo++; + // C = exp(2A) + f7_exp (cc, cc); + // X = exp(2A) + 1 + f7_add (xx, cc, one); + // C = exp(2A) - 1 + f7_Isub (cc, one); + // C = tanh(A) + f7_Idiv (cc, xx); +} +#endif // F7MOD_tanh_ + + +#ifdef F7MOD_atan_ + +#define MINIMAX_6_6_IN_0_1 + +#define ARRAY_NAME coeff_atan_zahler +#define FOR_NUMERATOR +#include "libf7-array.def" +#undef FOR_NUMERATOR +#undef ARRAY_NAME + +#define ARRAY_NAME coeff_atan_nenner +#define FOR_DENOMINATOR +#include "libf7-array.def" +#undef FOR_DENOMINATOR +#undef ARRAY_NAME + +#include "libf7-constdef.h" + +F7_WEAK +void f7_atan (f7_t *cc, const f7_t *aa) +{ + uint8_t a_class = f7_classify (aa); + uint8_t flags = a_class & F7_FLAG_sign; + + if (f7_class_nan (a_class)) + return f7_set_nan (cc); + + f7_t yy7, *yy = &yy7; + f7_t zz7, *zz = &zz7; + + if (f7_class_inf (a_class)) + { + f7_set_u16 (cc, 0); + goto do_Inf; + } + + // C = |A| + f7_copy (cc, aa); + cc->flags = 0; + + if (!f7_class_zero (a_class) && cc->expo >= 0) + { + // C >= 1: use atan (x) + atan (1/x) = pi / 2 to reduce to [0, 1]. + flags |= 1 << 1; + f7_div (cc, f7_set_u16 (yy, 1), cc); + } +#if !defined (MINIMAX_6_6_IN_0_1) + const uint8_t const_a_msb = 0x89; + const int16_t const_a_expo = -2; + if (f7_abscmp_msb_ge (cc, const_a_msb, const_a_expo)) + { + // We have C in [0, 1] and we want to use argument reduction by means + // of addition theorem atan(x) - atan(y) = atan((x - y) / (1 + xy)). + // We want to split [0, 1] into [0, a] u [a, 1] in such a way that + // the upper interval will be mapped to [-a, a]. The system is easy + // to solve and yiels + // y = 1 / sqrt (3) ~ 0.57735 atan(y) = pi / 6 + // a = (1 - y) / (1 + y) ~ 0.26795 ~ 0x0.8930A2F * 2^-1. + flags |= 1 << 2; + // C <- (C - Y) / (1 + C * Y) in [-a, a] + const f7_t *cst = F7_CONST_ADDR (1_sqrt3, zz); + f7_mul (yy, cc, cst); + f7_Isub (cc, cst); + f7_Iadd (yy, F7_U16_ADDR (1, zz)); + f7_Idiv (cc, yy); + } +#endif + // C <- C * p(C^2) / q(C^2) + f7_square (yy, cc); + f7_horner (zz, yy, n_coeff_atan_zahler, coeff_atan_zahler, NULL); + f7_Imul (zz, cc); + f7_horner (cc, yy, n_coeff_atan_nenner, coeff_atan_nenner, NULL); + f7_div (cc, zz, cc); + +#if !defined (MINIMAX_6_6_IN_0_1) + if (flags & (1 << 2)) + f7_Iadd (cc, F7_CONST_ADDR (pi_6, yy)); +#endif + + if (flags & (1 << 1)) + { + do_Inf:; + // Y = pi / 2 + f7_const (yy, pi); + yy->expo = F7_(const_pi_expo) - 1; + f7_IRsub (cc, yy); + } + + cc->flags = a_class & F7_FLAG_sign; +} +#undef MINIMAX_6_6_IN_0_1 +#endif // F7MOD_atan_ + + +#ifdef F7MOD_asinacos_ + +#define ARRAY_NAME coeff_func_a_zahler +#define FOR_NUMERATOR +#include "libf7-array.def" +#undef FOR_NUMERATOR +#undef ARRAY_NAME + +#define ARRAY_NAME coeff_func_a_nenner +#define FOR_DENOMINATOR +#include "libf7-array.def" +#undef FOR_DENOMINATOR +#undef ARRAY_NAME + +typedef union +{ + struct + { + bool sign : 1; // Must be bit F7_FLAGNO_sign. + bool do_acos : 1; // From caller. + bool have_acos : 1; // What we compute from rational approx p/q. + uint8_t res : 5; + }; + uint8_t bits; +} asinacos_t; + +F7_WEAK +void f7_asinacos (f7_t *cc, const f7_t *aa, uint8_t what) +{ + f7_t xx7, *xx = &xx7; + f7_t xx27, *xx2 = &xx27; + + asinacos_t flags = { .bits = what | f7_signbit (aa) }; + + f7_abs (xx, aa); + + int8_t cmp = f7_cmp (xx, f7_set_u16 (cc, 1)); + + if (cmp == INT8_MIN + || cmp > 0) + { + return f7_set_nan (cc); + } + + if (xx->expo <= -2 || f7_is_zero (xx)) + { + // |A| < 1/2: asin(x) = x * a(2*x^2) + f7_square (xx2, xx); + xx2->expo ++; + } + else + { + // |A| > 1/2: acos (1-x) = sqrt(2*x) * a(x) + // C is 1 from above. + f7_IRsub (xx, cc); + f7_copy (xx2, xx); + flags.have_acos = 1; + } + + // MiniMax [5/4] numerator. + f7_horner (cc, xx2, n_coeff_func_a_zahler, coeff_func_a_zahler, NULL); + + if (flags.have_acos) + { + xx->expo ++; + f7_Isqrt (xx); + } + f7_Imul (cc, xx); + + // MiniMax [5/4] denominator. + f7_horner (xx, xx2, n_coeff_func_a_nenner, coeff_func_a_nenner, NULL); + + f7_Idiv (cc, xx); + + /* + With the current value of C, we have: + + | | do_asin | do_acos + | C | A <= 0 | A >= 0 | A <= 0 | A >= 0 + ----------+------------+-----------+----------+----------+---------- + have_asin | asin (|A|) | -C | C | pi/2 + C | pi/2 - C + have_acos | acos (|A|) | -pi/2 + C | pi/2 - C | pi - C | C + + Result = n_pi2 * pi/2 + C * (c_sign ? -1 : 1) + Result (A, do_asin) = asin (A) + Result (A, do_acos) = acos (A) + + with + c_sign = do_acos ^ have_acos ^ a_sign + n_pi2 = do_acos + have_acos * (a_sign ^ do_acos) ? (-1 : 1) + n_pi2 in { -1, 0, 1, 2 } + */ + + // All what matters for c_sign is bit 0. + uint8_t c_sign = flags.bits; + int8_t n_pi2 = flags.do_acos; + c_sign ^= flags.do_acos; + if (flags.have_acos) + { + n_pi2++; + __asm ("" : "+r" (n_pi2)); + if (c_sign & 1) // c_sign & 1 = a_sign ^ do_acos + n_pi2 -= 2; + c_sign++; + } + + cc->flags = c_sign & F7_FLAG_sign; + + if (n_pi2) + { + f7_const (xx, pi); + if (n_pi2 < 0) + xx->sign = 1; + if (n_pi2 != 2) + xx->expo = F7_(const_pi_expo) - 1; + + f7_Iadd (cc, xx); + } +} +#endif // F7MOD_asinacos_ + + +#ifdef F7MOD_asin_ +F7_WEAK +void f7_asin (f7_t *cc, const f7_t *aa) +{ + f7_asinacos (cc, aa, 0); +} +#endif // F7MOD_asin_ + + +#ifdef F7MOD_acos_ +F7_WEAK +void f7_acos (f7_t *cc, const f7_t *aa) +{ + f7_asinacos (cc, aa, 1 << 1); +} +#endif // F7MOD_acos_ + + +#ifndef IN_LIBGCC2 + +#ifdef F7MOD_put_C_ + +#include <stdio.h> +#include <avr/pgmspace.h> + +static F7_INLINE +uint8_t f7_hex_digit (uint8_t nibble) +{ + nibble = (uint8_t) (nibble + '0'); + if (nibble > '9') + nibble = (uint8_t) (nibble + ('a' - '0' - 10)); + return nibble; +} + +static void f7_put_hex2 (uint8_t x, FILE *stream) +{ + putc ('0', stream); + if (x) + { + putc ('x', stream); + putc (f7_hex_digit (x >> 4), stream); + putc (f7_hex_digit (x & 0xf), stream); + } +} + +#define XPUT(str) \ + fputs_P (PSTR (str), stream) + +// Write to STREAM a line that is appropriate for usage in libf7-const.def. + +F7_WEAK +void f7_put_CDEF (const char *name, const f7_t *aa, FILE *stream) +{ + char buf[7]; + XPUT ("F7_CONST_DEF ("); + fputs (name, stream); + XPUT (",\t"); + uint8_t a_class = f7_classify (aa); + if (! f7_class_nonzero (a_class)) + { + f7_put_hex2 (a_class & F7_FLAGS, stream); + XPUT (",\t0,0,0,0,0,0,0,\t0)"); + return; + } + putc ('0' + (a_class & F7_FLAGS), stream); + XPUT (",\t"); + + for (uint8_t i = 0; i < F7_MANT_BYTES; i++) + { + f7_put_hex2 (aa->mant[F7_MANT_BYTES-1 - i], stream); + putc (',', stream); + } + putc ('\t', stream); + + itoa (aa->expo, buf, 10); + fputs (buf, stream); + XPUT (")"); +} + +void f7_put_C (const f7_t *aa, FILE *stream) +{ + char buf[7]; + + uint8_t a_class = f7_classify (aa); + if (f7_class_nan (a_class)) + { + XPUT ("{ .is_nan = 1 }"); + return; + } + bool sign = a_class & F7_FLAG_sign; + + if (f7_class_inf (a_class)) + { + XPUT ("{ .is_nan = 1, .sign = "); + putc ('0' + sign, stream); + XPUT (" }"); + return; + } + + XPUT ("{ .sign = "); + putc ('0' + sign, stream); + + XPUT (", .mant = { "); + for (uint8_t i = 0; i < F7_MANT_BYTES; i++) + { + f7_put_hex2 (aa->mant[F7_MANT_BYTES-1 - i], stream); + if (i != F7_MANT_BYTES - 1) + putc (',', stream); + } + + XPUT (" }, .expo = "); + itoa (aa->expo, buf, 10); + fputs (buf, stream); + XPUT (" }"); +} +#endif //F7MOD_put_C_ + + +#ifdef F7MOD_dump_ + +#include <avr/pgmspace.h> + +#ifndef AVRTEST_H + +#include <stdio.h> + +static void LOG_PSTR (const char *str) +{ + fputs_P (str, stdout); +} + +static void LOG_PFMT_U16 (const char *fmt, uint16_t x) +{ + printf_P (fmt, x); +} + +static void LOG_PFMT_FLOAT (const char *fmt, float x) +{ + printf_P (fmt, x); +} + +#define LOG_X8(X) LOG_PFMT_U16 (PSTR (" 0x%02x "), (uint8_t)(X)) +#define LOG_PFMT_S16(FMT, X) LOG_PFMT_U16 (FMT, (unsigned)(X)) +#define LOG_PFMT_ADDR(FMT, X) LOG_PFMT_U16 (FMT, (unsigned)(X)) + +#endif // AVRTEST_H + +static void dump_byte (uint8_t b) +{ + LOG_PSTR (PSTR (" ")); + for (uint8_t i = 0; i < 8; i++) + { + LOG_PSTR ((b & 0x80) ? PSTR ("1") : PSTR ("0")); + b = (uint8_t) (b << 1); + } +} + +void f7_dump_mant (const f7_t *aa) +{ + LOG_PSTR (PSTR ("\tmant =")); + for (int i = F7_MANT_BYTES - 1; i >= 0; i--) + LOG_X8 (aa->mant[i]); + LOG_PSTR (PSTR ("\n\t =")); + + for (int i = F7_MANT_BYTES - 1; i >= 0; i--) + dump_byte (aa->mant[i]); + LOG_PSTR (PSTR ("\n")); +} + +void f7_dump (const f7_t *aa) +{ + LOG_PFMT_ADDR (PSTR ("\n0x%04x\tflags = "), aa); + dump_byte (aa->flags); + uint8_t a_class = f7_classify (aa); + LOG_PSTR (PSTR (" = ")); + LOG_PSTR (f7_class_sign (a_class) ? PSTR ("-") : PSTR ("+")); + if (f7_class_inf (a_class)) LOG_PSTR (PSTR ("Inf ")); + if (f7_class_nan (a_class)) LOG_PSTR (PSTR ("NaN ")); + if (f7_class_zero (a_class)) LOG_PSTR (PSTR ("0 ")); + if (f7_class_number (a_class)) LOG_PSTR (PSTR ("Number ")); + + LOG_PFMT_FLOAT (PSTR (" = %.10g\n"), f7_get_float (aa)); + LOG_PFMT_S16 (PSTR ("\texpo = %d\n"), aa->expo); + + f7_dump_mant (aa); +} +#endif // F7MOD_dump_ + +#endif // ! libgcc + +#endif // !AVR_TINY Property changes on: libgcc/config/avr/libf7/libf7.c ___________________________________________________________________ Added: svn:eol-style ## -0,0 +1 ## +native \ No newline at end of property Added: svn:mime-type ## -0,0 +1 ## +text/plain \ No newline at end of property Index: libgcc/config/avr/libf7/libf7.h =================================================================== --- libgcc/config/avr/libf7/libf7.h (nonexistent) +++ libgcc/config/avr/libf7/libf7.h (working copy) @@ -0,0 +1,687 @@ +/* Copyright (C) 2019 Free Software Foundation, Inc. + + This file is part of LIBF7, which is part of GCC. + + GCC is free software; you can redistribute it and/or modify it under + the terms of the GNU General Public License as published by the Free + Software Foundation; either version 3, or (at your option) any later + version. + + GCC is distributed in the hope that it will be useful, but WITHOUT ANY + WARRANTY; without even the implied warranty of MERCHANTABILITY or + FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License + for more details. + + Under Section 7 of GPL version 3, you are granted additional + permissions described in the GCC Runtime Library Exception, version + 3.1, as published by the Free Software Foundation. + + You should have received a copy of the GNU General Public License and + a copy of the GCC Runtime Library Exception along with this program; + see the files COPYING3 and COPYING.RUNTIME respectively. If not, see + <http://www.gnu.org/licenses/>. */ + +#ifndef LIBF7_H +#define LIBF7_H +#define IN_LIBF7_H + +#include "f7-renames.h" + +#define F7_MANT_BYTES 7 +#define F7_MANT_BITS (8 * F7_MANT_BYTES) + +/* Using the following GCC features: + -- Unnamed structs / unions (GNU-C) + -- Fixed-point types (GNU-C) + -- Inline asm + -- Setting assembler names by means of __asm (GNU-C). + -- Attributes: alias, always_inline, const, noinline, unused, + progmem, pure, weak, warning + -- GCC built-ins: __builtin_abort, __builtin_constant_p + -- AVR built-ins: __builtin_avr_bitsr, __builtin_avr_rbits +*/ + +/* We have 2 kinds of flags: + + A) The flags that are stored in f7_t.flags: + -- f7_t.is_nan (NaN) + -- f7_t.is_inf (+Inf or -Inf) + -- f7_t.sign (negative or -Inf). + + B) The flags that are returned by f7_classify(). This are the + flags from A) together with + -- _zero: indicate that a number is zero. +*/ + +#define F7_FLAGNO_sign 0 +#define F7_FLAGNO_zero 1 +#define F7_FLAGNO_nan 2 +#define F7_FLAGNO_inf 7 + +#define F7_HAVE_Inf 1 + +// Flags that might be set by f7_classify(). +#define F7_FLAG_sign (1 << F7_FLAGNO_sign) +#define F7_FLAG_zero (1 << F7_FLAGNO_zero) +#define F7_FLAG_nan (1 << F7_FLAGNO_nan) +#define F7_FLAG_inf (F7_HAVE_Inf << F7_FLAGNO_inf) + +// Flags that might be set in f7_t.flags. +#define F7_FLAGS (F7_FLAG_inf | F7_FLAG_nan | F7_FLAG_sign) + +#if !defined __ASSEMBLER__ + +#ifndef IN_LIBGCC2 +#include <stdint.h> +#include <stdbool.h> +#include <stdlib.h> +#include <stdio.h> +#else +/* Do not assume that we have std headers when we build libgcc. */ + +typedef __UINT64_TYPE__ uint64_t; +typedef __UINT32_TYPE__ uint32_t; +typedef __UINT16_TYPE__ uint16_t; +typedef __UINT8_TYPE__ uint8_t; +typedef __INT64_TYPE__ int64_t; +typedef __INT32_TYPE__ int32_t; +typedef __INT16_TYPE__ int16_t; +typedef __INT8_TYPE__ int8_t; +typedef _Bool bool; +#define false 0 +#define true 1 +#define INT8_MIN (-1 - __INT8_MAX__) +#define INT16_MAX __INT16_MAX__ +#define NULL ((void*) 0) +#endif /* IN_LIBGCC2 */ + +#include "asm-defs.h" + +#ifdef __cplusplus +extern "C" { +#define _Static_assert(X, Y) static_assert (X) +#endif // C++ + +#define F7_INLINE inline __attribute__((__always_inline__)) +#define F7_NOINLINE __attribute__((__noinline__)) +#define F7_WEAK __attribute__((__weak__)) +#define F7_PURE __attribute__((__pure__)) +#define F7_UNUSED __attribute__((__unused__)) +#define F7_CONST __attribute__((__const__)) + +#define F7_STRINGY2(X) #X +#define F7_STRINGY(X) F7_STRINGY2(X) +#define F7ASM(X) __asm (F7_STRINGY2(X)) + +typedef struct f7_t +{ + union + { + struct + { + uint8_t sign :1; + uint8_t reserved1 :1; + uint8_t is_nan :1; + uint8_t reserved2 :4; + uint8_t is_inf :1; + }; + uint8_t flags; + }; + + uint8_t mant[7]; + int16_t expo; +} f7_t; + +typedef uint64_t f7_double_t; + +#define F7_MANT_HI4(X) \ + (*(uint32_t*) & (X)->mant[F7_MANT_BYTES - 4]) + +#define F7_MANT_CONST_HI4(X) \ + (*(const uint32_t*) & (X)->mant[F7_MANT_BYTES - 4]) + +#define F7_MANT_HI2(X) \ + (*(uint16_t*) & (X)->mant[F7_MANT_BYTES - 2]) + +static F7_INLINE F7_PURE +uint8_t f7_classify (const f7_t *aa) +{ + extern void f7_classify_asm (void); + register uint8_t rclass __asm ("r24"); + __asm ("%~call %x[f]" + : "=r" (rclass) + : [f] "i" (f7_classify_asm), "z" (aa)); + return rclass; +} + + +// +Inf or -Inf +static F7_INLINE +bool f7_class_inf (uint8_t c) +{ +#if defined (F7_HAVE_Inf) && F7_HAVE_Inf == 1 + return c >= F7_FLAG_inf; +#elif defined (F7_HAVE_Inf) && F7_HAVE_Inf == 0 + (void) c; + return false; +#else +#error macro F7_HAVE_Inf must be defined to 0 or to 1. +#endif // Have Inf +} + +static F7_INLINE +bool f7_is_inf (const f7_t *aa) +{ + return f7_class_inf (aa->flags); +} + +// Not-a-Number (NaN). +static F7_INLINE +bool f7_class_nan (uint8_t c) +{ + return c & F7_FLAG_nan; +} + +static F7_INLINE +bool f7_is_nan (const f7_t *aa) +{ + return f7_class_nan (aa->flags); +} + +// Some number +static F7_INLINE +bool f7_class_number (uint8_t c) +{ + return c <= (F7_FLAG_sign | F7_FLAG_zero); +} + +static F7_INLINE +bool f7_is_number (const f7_t *aa) +{ + return f7_class_number (f7_classify (aa)); +} + +// Zero +static F7_INLINE +bool f7_class_zero (uint8_t c) +{ + return c & F7_FLAG_zero; +} + +static F7_INLINE +bool f7_is_zero (const f7_t *aa) +{ + return f7_class_zero (f7_classify (aa)); +} + +// A non-zero number. +static F7_INLINE +bool f7_class_nonzero (uint8_t c) +{ + return c <= F7_FLAG_sign; +} + +static F7_INLINE +bool f7_is_nonzero (const f7_t *aa) +{ + return f7_class_nonzero (f7_classify (aa)); +} + +static F7_INLINE +bool f7_class_sign (uint8_t c) +{ + return c & F7_FLAG_sign; +} + +static F7_INLINE +bool f7_signbit (const f7_t *aa) +{ + return aa->flags & F7_FLAG_sign; +} + +static F7_INLINE +void f7_set_sign (f7_t *cc, bool sign) +{ + _Static_assert (F7_FLAGNO_sign == 0, ""); + cc->flags &= ~F7_FLAG_sign; + cc->flags |= sign; +} + +static F7_INLINE +void f7_set_nan (f7_t *cc) +{ + cc->flags = F7_FLAG_nan; +} + +static F7_INLINE +void f7_clr (f7_t *cc) +{ + extern void f7_clr_asm (void); + __asm ("%~call %x[f]" + : + : [f] "i" (f7_clr_asm), "z" (cc) + : "memory"); +} + +static F7_INLINE +f7_t* f7_copy (f7_t *cc, const f7_t *aa) +{ + extern void f7_copy_asm (void); + __asm ("%~call %x[f]" + : + : [f] "i" (f7_copy_asm), "z" (cc), "x" (aa) + : "memory"); + return cc; +} + +static F7_INLINE +f7_t* f7_copy_P (f7_t *cc, const f7_t *aa) +{ + extern void f7_copy_P_asm (void); + __asm ("%~call %x[f]" + : + : [f] "i" (f7_copy_P_asm), "x" (cc), "z" (aa) + : "memory"); + return cc; +} + +static F7_INLINE +void f7_copy_mant (f7_t *cc, const f7_t *aa) +{ + extern void f7_copy_mant_asm (void); + __asm ("%~call %x[f]" + : + : [f] "i" (f7_copy_mant_asm), "z" (cc), "x" (aa) + : "memory"); +} + +static F7_INLINE +void f7_set_inf (f7_t *cc, bool sign) +{ +#if F7_HAVE_Inf == 1 + cc->flags = F7_FLAG_inf | sign; +#else + (void) sign; + cc->flags = F7_FLAG_nan; +#endif // Have Inf +} + + +static F7_INLINE +bool f7_msbit (const f7_t *aa) +{ + return aa->mant[F7_MANT_BYTES - 1] & 0x80; +} + +// Quick test against 0 if A is known to be a number (neither NaN nor Inf). +static F7_INLINE +bool f7_is0 (const f7_t *aa) +{ + return 0 == f7_msbit (aa); +} + + +static F7_INLINE +int8_t f7_cmp_mant (const f7_t *aa, const f7_t *bb) +{ + extern void f7_cmp_mant_asm (void); + register int8_t r24 __asm ("r24"); + __asm ("%~call %x[f] ;; %1 %3" + : "=r" (r24) + : [f] "i" (f7_cmp_mant_asm), "x" (aa), "z" (bb)); + return r24; +} + +static F7_INLINE +bool f7_store_expo (f7_t *cc, int16_t expo) +{ + extern void f7_store_expo_asm (void); + register bool r24 __asm ("r24"); + register int16_t rexpo __asm ("r24") = expo; + __asm ("%~call %x[f] ;; %0 %2 %3" + : "=r" (r24) + : [f] "i" (f7_store_expo_asm), "z" (cc), "r" (rexpo)); + return r24; +} + +static F7_INLINE +f7_t* f7_abs (f7_t *cc, const f7_t *aa) +{ + f7_copy (cc, aa); + f7_set_sign (cc, 0); + + return cc; +} + + +F7_PURE extern int8_t f7_cmp (const f7_t*, const f7_t*); +F7_PURE extern bool f7_lt_impl (const f7_t*, const f7_t*); +F7_PURE extern bool f7_le_impl (const f7_t*, const f7_t*); +F7_PURE extern bool f7_gt_impl (const f7_t*, const f7_t*); +F7_PURE extern bool f7_ge_impl (const f7_t*, const f7_t*); +F7_PURE extern bool f7_ne_impl (const f7_t*, const f7_t*); +F7_PURE extern bool f7_eq_impl (const f7_t*, const f7_t*); +F7_PURE extern bool f7_unord_impl (const f7_t*, const f7_t*); + +static F7_INLINE +bool f7_lt (const f7_t *aa, const f7_t *bb) +{ + return 2 & f7_cmp (aa, bb); +} + +static F7_INLINE +bool f7_gt (const f7_t *aa, const f7_t *bb) +{ + return 1 == f7_cmp (aa, bb); +} + +static F7_INLINE +bool f7_le (const f7_t *aa, const f7_t *bb) +{ + int8_t c = f7_cmp (aa, bb); + return (uint8_t) (c + 1) <= 1; +} + +static F7_INLINE +bool f7_ge (const f7_t *aa, const f7_t *bb) +{ + return f7_cmp (aa, bb) >= 0; +} + +static F7_INLINE +bool f7_unordered (const f7_t *aa, const f7_t *bb) +{ + return INT8_MIN == f7_cmp (aa, bb); +} + +static F7_INLINE +bool f7_ordered (const f7_t *aa, const f7_t *bb) +{ + return INT8_MIN != f7_cmp (aa, bb); +} + +static F7_INLINE +bool f7_eq (const f7_t *aa, const f7_t *bb) +{ + return 0 == f7_cmp (aa, bb); +} + +static F7_INLINE +bool f7_ne (const f7_t *aa, const f7_t *bb) +{ + return 1 & f7_cmp (aa, bb); +} + +extern void f7_clr (f7_t*); + +__attribute__((warning ("foo_u16"))) void foo_u16 (void); +__attribute__((warning ("foo_s16"))) void foo_s16 (void); + +extern f7_t* f7_set_s16_impl (f7_t*, int16_t); +extern f7_t* f7_set_u16_impl (f7_t*, uint16_t); + +static F7_INLINE +f7_t* f7_set_u16_worker (f7_t *cc, uint16_t u16) +{ + if (__builtin_constant_p (u16)) + { + if (u16 == 0) + return cc; + + uint8_t off = __builtin_clz (u16); + if (15 - off) + * (uint8_t*) & cc->expo = (uint8_t) (15 - off); + u16 <<= off; + if (u16 & 0xff) + cc->mant[5] = (uint8_t) u16; + if (u16 & 0xff00) + cc->mant[6] = (uint8_t) (u16 >> 8); + + return cc; + } + else + { + foo_u16(); + __builtin_abort(); + return NULL; + } +} + +static F7_INLINE +f7_t* f7_set_u16 (f7_t *cc, uint16_t u16) +{ + if (__builtin_constant_p (u16)) + { + f7_clr (cc); + return f7_set_u16_worker (cc, u16); + } + + return f7_set_u16_impl (cc, u16); +} + +static F7_INLINE +f7_t* f7_set_s16 (f7_t *cc, int16_t s16) +{ + if (__builtin_constant_p (s16)) + { + f7_clr (cc); + + uint16_t u16 = (uint16_t) s16; + + if (s16 < 0) + { + u16 = -u16; + cc->flags = F7_FLAG_sign; + } + + return f7_set_u16_worker (cc, u16); + } + + return f7_set_s16_impl (cc, s16); +} + +static F7_INLINE +void f7_set_eps (f7_t *cc, uint8_t eps, bool sign) +{ + cc = f7_set_u16 (cc, 1); + if (!__builtin_constant_p (sign) || sign) + cc->flags = sign; + cc->mant[0] = eps; +} + +static F7_INLINE +f7_t* f7_set_1pow2 (f7_t *cc, int16_t expo, bool sign) +{ + cc = f7_set_u16 (cc, 1); + cc->expo = expo; + if (!__builtin_constant_p (sign) || sign) + cc->flags = sign; + return cc; +} + +static F7_INLINE +f7_t* f7_set_u64 (f7_t *cc, uint64_t u64) +{ + extern f7_t* f7_set_u64_asm (uint64_t, f7_t*); + return f7_set_u64_asm (u64, cc); +} + +static F7_INLINE +f7_t* f7_set_s64 (f7_t *cc, int64_t s64) +{ + extern f7_t* f7_set_s64_asm (int64_t, f7_t*); + return f7_set_s64_asm (s64, cc); +} + +extern void f7_set_double_impl (f7_double_t, f7_t*); +static F7_INLINE +void f7_set_double (f7_t *cc, f7_double_t val64) +{ + f7_set_double_impl (val64, cc); +} + +extern f7_t* f7_init_impl (uint64_t, uint8_t, f7_t*, int16_t); + +static F7_INLINE +f7_t* f7_init (f7_t *cc, uint8_t flags, uint64_t mant, int16_t expo) +{ + return f7_init_impl (mant, flags, cc, expo); +} + +extern f7_t* f7_set_s32 (f7_t*, int32_t); +extern f7_t* f7_set_u16 (f7_t*, uint16_t); +extern f7_t* f7_set_u32 (f7_t*, uint32_t); +extern void f7_set_float (f7_t*, float); +extern void f7_set_pdouble (f7_t*, const f7_double_t*); + +F7_PURE extern int16_t f7_get_s16 (const f7_t*); +F7_PURE extern int32_t f7_get_s32 (const f7_t*); +F7_PURE extern int64_t f7_get_s64 (const f7_t*); +F7_PURE extern uint16_t f7_get_u16 (const f7_t*); +F7_PURE extern uint32_t f7_get_u32 (const f7_t*); +F7_PURE extern uint64_t f7_get_u64 (const f7_t*); +F7_PURE extern float f7_get_float (const f7_t*); +F7_PURE extern f7_double_t f7_get_double (const f7_t*); + +#if USE_LPM == 1 + #define F7_PGMSPACE __attribute__((__progmem__)) + #define f7_copy_flash f7_copy_P + + #define f7_const(X, NAME) \ + f7_copy_P ((X), & F7_(const_ ## NAME ## _P)) + + #define F7_CONST_DEF(NAME, FLAGS, M0, M1, M2, M3, M4, M5, M6, EXPO) \ + extern const f7_t F7_(const_ ## NAME ## _P); + #include "libf7-const.def" + #undef F7_CONST_DEF +#else + #define F7_PGMSPACE // Empty + #define f7_copy_flash f7_copy + + #define f7_const(X, NAME) \ + f7_copy ((X), & F7_(const_ ## NAME)) + + #define F7_CONST_DEF(NAME, FLAGS, M0, M1, M2, M3, M4, M5, M6, EXPO) \ + extern const f7_t F7_(const_ ## NAME); + #include "libf7-const.def" + #undef F7_CONST_DEF +#endif // USE_LPM + + +// Basic floating point arithmetic: +// double output <=> f7_t* +// double input <=> const f7_t* +extern f7_t* f7_neg (f7_t*, const f7_t*); +extern void f7_add (f7_t*, const f7_t*, const f7_t*); +extern void f7_sub (f7_t*, const f7_t*, const f7_t*); +extern void f7_mul (f7_t*, const f7_t*, const f7_t*); +extern void f7_div (f7_t*, const f7_t*, const f7_t*); + +// Analogies of functions from math.h: +// double output <=> f7_t* +// double input <=> const f7_t* +extern void f7_fabs (f7_t*, const f7_t*); +extern void f7_fmod (f7_t*, const f7_t*, const f7_t*); +extern void f7_frexp (f7_t*, const f7_t*, int*); +extern void f7_exp (f7_t*, const f7_t*); +extern void f7_log (f7_t*, const f7_t*); +extern void f7_pow (f7_t*, const f7_t*, const f7_t*); +extern void f7_sqrt (f7_t*, const f7_t*); +extern void f7_cbrt (f7_t*, const f7_t*); +extern void f7_hypot (f7_t*, const f7_t*, const f7_t*); +extern f7_t* f7_ldexp (f7_t*, const f7_t*, int); +extern f7_t* f7_fmax (f7_t*, const f7_t*, const f7_t*); +extern f7_t* f7_fmin (f7_t*, const f7_t*, const f7_t*); +extern f7_t* f7_trunc (f7_t*, const f7_t*); +extern f7_t* f7_floor (f7_t*, const f7_t*); +extern void f7_ceil (f7_t*, const f7_t*); +extern void f7_round (f7_t*, const f7_t*); +extern void f7_sin (f7_t*, const f7_t*); +extern void f7_cos (f7_t*, const f7_t*); +extern void f7_tan (f7_t*, const f7_t*); +extern void f7_atan (f7_t*, const f7_t*); +extern void f7_asin (f7_t*, const f7_t*); +extern void f7_acos (f7_t*, const f7_t*); +extern void f7_tanh (f7_t*, const f7_t*); +extern void f7_sinh (f7_t*, const f7_t*); +extern void f7_cosh (f7_t*, const f7_t*); +extern void f7_log2 (f7_t*, const f7_t*); +extern void f7_log10 (f7_t*, const f7_t*); +extern void f7_exp10 (f7_t*, const f7_t*); +extern void f7_pow10 (f7_t*, const f7_t*); + +// Just prototypes, not implemented yet. +extern void f7_atan2 (f7_t*, const f7_t*, const f7_t*); +extern long f7_lrint (const f7_t*); +extern long f7_lround (const f7_t*); + +// Helper functions, aliases, convenience. +extern void f7_div1 (f7_t*, const f7_t*); +extern void f7_square (f7_t*, const f7_t*); + +extern void f7_powi (f7_t*, const f7_t*, int); +extern f7_t* f7_max (f7_t*, const f7_t*, const f7_t*); +extern f7_t* f7_min (f7_t*, const f7_t*, const f7_t*); +extern f7_t* f7_truncx (f7_t*, const f7_t*, bool); +extern void f7_cotan (f7_t*, const f7_t*); +extern void f7_sincos (f7_t*, f7_t*, const f7_t*); +extern void f7_asinacos (f7_t*, const f7_t*, uint8_t); +extern void f7_sinhcosh (f7_t*, const f7_t*, bool); + +extern void f7_horner (f7_t*, const f7_t*, uint8_t, const f7_t *coeff, f7_t*); +extern void f7_mul_noround (f7_t*, const f7_t*, const f7_t*); +extern void f7_clr_mant_lsbs (f7_t*, const f7_t*, uint8_t) F7ASM(f7_clr_mant_lsbs_asm); + +F7_PURE extern int8_t f7_cmp_unordered (const f7_t*, const f7_t*, bool); +F7_PURE extern int8_t f7_cmp_abs (const f7_t*, const f7_t*); + +F7_PURE extern bool f7_abscmp_msb_ge (const f7_t*, uint8_t msb, int16_t expo); +extern void f7_addsub (f7_t*, const f7_t*, const f7_t*, bool neg_b); +extern void f7_madd_msub (f7_t*, const f7_t*, const f7_t*, const f7_t*, bool); +extern void f7_madd (f7_t*, const f7_t*, const f7_t*, const f7_t*); +extern void f7_msub (f7_t*, const f7_t*, const f7_t*, const f7_t*); +extern uint8_t f7_mulx (f7_t*, const f7_t*, const f7_t*, bool); +extern void f7_divx (f7_t*, const f7_t*, const f7_t*, uint8_t); +extern void f7_logx (f7_t*, const f7_t*, const f7_t*); +extern f7_t* f7_minmax (f7_t*, const f7_t*, const f7_t*, bool); + +// Idem: +// f7_Ifunc (y) = f7_func (y, y) +// f7_Ifunc (y, x) = f7_func (y, y, x) +extern void f7_Iadd (f7_t*, const f7_t*); +extern void f7_Isub (f7_t*, const f7_t*); +extern void f7_Imul (f7_t*, const f7_t*); +extern void f7_Idiv (f7_t*, const f7_t*); +extern void f7_IRsub (f7_t*, const f7_t*); +extern void f7_Ineg (f7_t*); +extern void f7_Isqrt (f7_t*); +extern void f7_Isquare (f7_t*); +extern f7_t* f7_Ildexp (f7_t*, int); + +// Protoypes for some functions from libf7-asm.sx. +F7_CONST extern uint16_t f7_sqrt16_round (uint16_t) F7ASM(f7_sqrt16_round_asm); +F7_CONST extern uint8_t f7_sqrt16_floor (uint16_t) F7ASM(f7_sqrt16_floor_asm); +extern void f7_addsub_mant_scaled_asm (f7_t*, const f7_t*, const f7_t*, uint8_t); +extern uint8_t f7_mul_mant_asm (f7_t*, const f7_t*, const f7_t*, uint8_t); +extern void f7_sqrt_approx_asm (f7_t*, const f7_t*); +extern uint64_t f7_lshrdi3 (uint64_t, uint8_t) F7ASM(f7_lshrdi3_asm); +extern uint64_t f7_ashldi3 (uint64_t, uint8_t) F7ASM(f7_ashldi3_asm); +// Normalize a non-Inf, non-NaN value. Sets .sign to 0. +extern f7_t* f7_normalize_asm (f7_t*); + +// Dumping. +#ifndef IN_LIBGCC2 +extern void f7_dump (const f7_t*); +extern void f7_dump_mant (const f7_t*); +extern void f7_put_C (const f7_t*, FILE*); +extern void f7_put_CDEF (const char *name, const f7_t*, FILE*); +#endif /* IN_LIBGCC2 */ + +#ifdef __cplusplus +} // extern "C" +#include "libf7-class.h" +#endif // C++ + +#endif /* __ASSEMBLER__ */ +#undef IN_LIBF7_H +#endif /* LIBF7_H */ Property changes on: libgcc/config/avr/libf7/libf7.h ___________________________________________________________________ Added: svn:eol-style ## -0,0 +1 ## +native \ No newline at end of property Added: svn:mime-type ## -0,0 +1 ## +text/plain \ No newline at end of property Index: libgcc/config/avr/libf7/t-libf7 =================================================================== --- libgcc/config/avr/libf7/t-libf7 (nonexistent) +++ libgcc/config/avr/libf7/t-libf7 (working copy) @@ -0,0 +1,159 @@ +# Used except --with-libf7=no + +avrsrc := $(srcdir)/config/avr + +libf7 := $(avrsrc)/libf7 + +F7_PREFIX = __f7_ + +include $(libf7)/libf7-common.mk + +LIBF7_DF_CONV += floatundidf floatdidf # floatunsidf floatsidf + +# Wrappers like f7_lt_impl for f7_lt etc. because the latter is inline. +LIBF7_DF_CMP += lt le gt ge ne eq unord + +F7_C_PARTS += $(LIBF7_DF_CONV) $(LIBF7_DF_CMP) + +# -mcall-prologues +CALL_PROLOGUES += $(LIBF7_DF_CONV) + +# -Wno-missing-prototypes +NO_PROTO += $(LIBF7_DF_CONV) + +F7F += le_impl lt_impl gt_impl ge_impl ne_impl eq_impl unord_impl + + +$(libf7)/f7-renames.h: $(libf7)/f7renames.sh $(libf7)/libf7-common.mk + $< head $(F7_PREFIX) t-libf7 > $@ + $< c $(F7_PREFIX) $(F7F) >> $@ + $< cst $(F7_PREFIX) $(F7F_cst) >> $@ + $< asm $(F7_PREFIX) $(F7F_asm) >> $@ + $< tail $(F7_PREFIX) >> $@ + +# The right-hand sides like g_ddd come from libf7-common.mk. +# The _m_ wraps are added by t-libf7-math + +# __adddf3, ... +F7_ASM_WRAPS_g_ddd += $(g_ddd) + +# __ltdf2, ... +F7_ASM_WRAPS_g_xdd_cmp += $(g_xdd_cmp) + +# __floatsidf, ... +F7_ASM_WRAPS_g_dx += $(g_dx) + +# __fixdfsi, ... +F7_ASM_WRAPS_g_xd += $(g_xd) + +$(libf7)/f7-wraps.h: $(libf7)/f7wraps.sh \ + $(libf7)/libf7-common.mk $(libf7)/t-libf7-math + $< header "WITH_LIBF7_MATH_FUNCTIONS=$(WITH_LIBF7_MATH_FUNCTIONS)" "WITH_LIBF7_MATH_SYMBOLS=$(WITH_LIBF7_MATH_SYMBOLS)" > $@ + $< ddd_libgcc $(F7_ASM_WRAPS_g_ddd) >> $@ + $< xdd_libgcc_cmp $(F7_ASM_WRAPS_g_xdd_cmp) >> $@ + $< xd_libgcc $(F7_ASM_WRAPS_g_xd) >> $@ + $< dx_libgcc $(F7_ASM_WRAPS_g_dx) >> $@ + $< ddd_math $(F7_ASM_WRAPS_m_ddd) >> $@ + $< ddx_math $(F7_ASM_WRAPS_m_ddx) >> $@ + $< dd_math $(F7_ASM_WRAPS_m_dd) >> $@ + $< xd_math $(F7_ASM_WRAPS_m_xd) >> $@ + +F7_ASM_WRAPS += $(F7_ASM_WRAPS_g_xd) +F7_ASM_WRAPS += $(F7_ASM_WRAPS_g_dx) +F7_ASM_WRAPS += $(F7_ASM_WRAPS_g_ddd) +F7_ASM_WRAPS += $(F7_ASM_WRAPS_g_xdd_cmp) +F7_ASM_WRAPS += $(F7_ASM_WRAPS_m_ddd) +F7_ASM_WRAPS += $(F7_ASM_WRAPS_m_ddx) +F7_ASM_WRAPS += $(F7_ASM_WRAPS_m_dd) +F7_ASM_WRAPS += $(F7_ASM_WRAPS_m_xd) + +F7_ASM_PARTS += $(patsubst %, D_%, $(F7_ASM_WRAPS)) + +# Options +F7_FLAGS += -I $(libf7) -save-temps=obj + +# t-avr::HOST_LIBGCC2_CFLAGS sets -mcall-prologues which will inhibits +# tail-call optimizations. The user could get it with -mrelax, but we +# just switch it off here and then explicitly on again for the +# CALL_PROLOGUES modules. + +F7_C_FLAGS += $(F7_FLAGS) \ + -dp -g0 \ + -mno-call-prologues \ + -fno-lto -Os \ + -fdata-sections -ffunction-sections \ + -fno-reorder-blocks \ + -fno-tree-loop-optimize \ + -fno-tree-loop-im -fno-move-loop-invariants + +F7_ASM_FLAGS += $(F7_FLAGS) + +$(patsubst %, f7_c_%.o, $(CALL_PROLOGUES)) \ + : F7_C_FLAGS += -mcall-prologues + +$(patsubst %, f7_c_%.o, $(STRICT_X)) \ + : F7_C_FLAGS += -mstrict-X + +$(patsubst %, f7_c_%.o, $(NO_PROTO)) \ + : F7_C_FLAGS += -Wno-missing-prototypes + +# Depends will be worked out by the libgcc build system. + +F7_C_OBJECTS = $(patsubst %, f7_c_%$(objext), $(F7_C_PARTS)) +F7_ASM_OBJECTS = $(patsubst %, f7_asm_%$(objext), $(F7_ASM_PARTS)) + +$(F7_ASM_OBJECTS) $(F7_C_OBJECTS) : $(libf7)/t-libf7 +$(F7_ASM_OBJECTS) $(F7_C_OBJECTS) : $(libf7)/t-libf7-math +$(F7_ASM_OBJECTS) $(F7_C_OBJECTS) : $(libf7)/t-libf7-math-symbols + +.PHONY: log_vars + +all: log_vars + +log_vars: + $(info # libf7: WITH_LIBF7_MATH_FUNCTIONS = $(WITH_LIBF7_MATH_FUNCTIONS)) + $(info # libf7: WITH_LIBF7_MATH_SYMBOLS = $(WITH_LIBF7_MATH_SYMBOLS)) + $(info # libf7: F7_C_PARTS = $(F7_C_PARTS)) + $(info # libf7: F7_C_OBJECTS = $(F7_C_OBJECTS)) + $(info # libf7: F7_ASM_PARTS = $(F7_ASM_PARTS)) + $(info # libf7: F7_ASM_OBJECTS = $(F7_ASM_OBJECTS)) + +# Build the libf7 C objects and add them to libgcc.a. + +f7_parts := $(F7_C_PARTS) + +iter-items := $(f7_parts) +iter-labels := $(f7_parts) + +include $(srcdir)/empty.mk $(patsubst %,$(libf7)/libf7-c-object.mk,$(iter-items)) + +libgcc-objects += $(patsubst %,f7_c_%$(objext),$(F7_C_PARTS)) + +# Build the libf7 ASM objects and add them to libgcc.a. + +f7_parts := $(F7_ASM_PARTS) + +iter-items := $(f7_parts) +iter-labels := $(f7_parts) + +include $(srcdir)/empty.mk $(patsubst %,$(libf7)/libf7-asm-object.mk,$(iter-items)) + +libgcc-objects += $(patsubst %,f7_asm_%$(objext),$(F7_ASM_PARTS)) + +.PHONY: clean-f7 + +clean: clean-f7 + +clean-f7: + rm -f $(wildcard f7_*.i f7_*.s f7_*.o) + +# Get rid if any DFmode remains. + +LIB2FUNCS_EXCLUDE += \ + _sf_to_df \ + _fixdfdi \ + _fixunsdfsi \ + _floatundidf \ + _fixunsdfdi \ + _floatdidf \ + _powidf2 Property changes on: libgcc/config/avr/libf7/t-libf7 ___________________________________________________________________ Added: svn:eol-style ## -0,0 +1 ## +native \ No newline at end of property Added: svn:mime-type ## -0,0 +1 ## +text/plain \ No newline at end of property Index: libgcc/config/avr/libf7/t-libf7-math =================================================================== --- libgcc/config/avr/libf7/t-libf7-math (nonexistent) +++ libgcc/config/avr/libf7/t-libf7-math (working copy) @@ -0,0 +1,21 @@ +# Triggered by --with-libf7=math or --with-libf7=math-symbols +# +# We provide weak double wrappers for functions specified in math.h, +# but with __ prepended to the symbol name used for the double function. +# For example we provide double __sin (double) but neither sin nor sinl. +# To get weak symbols according to math.h, t-libf7-math-symbols has to +# be used which is triggered by --with-libf7=math-symbols. + +WITH_LIBF7_MATH_FUNCTIONS = 1 + +# __sin, ... +F7_ASM_WRAPS_m_dd += $(m_dd) + +# __pow, __fmin, ... +F7_ASM_WRAPS_m_ddd += $(m_ddd) + +# __ldexp, ... +F7_ASM_WRAPS_m_ddx += $(m_ddx) + +# __lrint, ... +F7_ASM_WRAPS_m_xd += $(m_xd) Property changes on: libgcc/config/avr/libf7/t-libf7-math ___________________________________________________________________ Added: svn:eol-style ## -0,0 +1 ## +native \ No newline at end of property Added: svn:mime-type ## -0,0 +1 ## +text/plain \ No newline at end of property Index: libgcc/config/avr/libf7/t-libf7-math-symbols =================================================================== --- libgcc/config/avr/libf7/t-libf7-math-symbols (nonexistent) +++ libgcc/config/avr/libf7/t-libf7-math-symbols (working copy) @@ -0,0 +1,11 @@ +# Triggered by --with-libf7=math-symbols +# +# We have at least one module in libgcc that depends on __SIZEOF_DOUBLE__ +# or __SIZEOF_LONG_DOUBLE__ which means that t-avrlibc must not copy +# double32/64 or long-double32/64 variants from the vanilla one. +# This occurs when some module(s) define(s) weak aliases for functions +# that usually live in libm. + +WITH_LIBF7_MATH_SYMBOLS = 1 + +F7_FLAGS += -DWITH_LIBF7_MATH_SYMBOLS Property changes on: libgcc/config/avr/libf7/t-libf7-math-symbols ___________________________________________________________________ Added: svn:eol-style ## -0,0 +1 ## +native \ No newline at end of property Added: svn:mime-type ## -0,0 +1 ## +text/plain \ No newline at end of property ^ permalink raw reply [flat|nested] 22+ messages in thread
* Re: [patch,avr, 3/3] Support 64-bit (long) double: libf7. 2019-12-16 17:09 ` [patch,avr, 3/3] Support 64-bit (long) double: libf7 Georg-Johann Lay 2019-12-28 12:02 ` [PING^1][patch,avr, " Georg-Johann Lay 2020-01-06 13:09 ` [Ping^2][patch,avr, " Georg-Johann Lay @ 2020-01-06 16:26 ` Jeff Law 2 siblings, 0 replies; 22+ messages in thread From: Jeff Law @ 2020-01-06 16:26 UTC (permalink / raw) To: Georg-Johann Lay, GCC Patches On Mon, 2019-12-16 at 17:49 +0100, Georg-Johann Lay wrote: > Am 16.12.19 um 17:40 schrieb Georg-Johann Lay: > Patch 3/3 is the actual libf7 implementation. A great deal of which is > assembly, together with C + inline assembly for higher routines. > > Johann > > libgcc/config/avr/libf7/ > * t-libf7: New file. > * t-libf7-math: New file. > * t-libf7-math-symbols: New file. > * libf7-common.mk: New file. > * libf7-asm-object.mk: New file. > * libf7-c-object.mk: New file. > * asm-defs.h: New file. > * libf7.h: New file. > * libf7.c: New file. > * libf7-asm.sx: New file. > * libf7-array.def: New file. > * libf7-const.def: New file. > * libf7-constdef.h: New file. > * f7renames.sh: New script. > * f7wraps.sh: New script. > * f7-renames.h: New generated file. > * f7-wraps.h: New generated file. Same with this one was well. libf7->libf77. jeff ^ permalink raw reply [flat|nested] 22+ messages in thread
* [PING^1][patch,avr, 0/3] Support 64-bit (long) double. 2019-12-16 16:43 [patch,avr, 0/3] Support 64-bit (long) double Georg-Johann Lay ` (2 preceding siblings ...) 2019-12-16 17:09 ` [patch,avr, 3/3] Support 64-bit (long) double: libf7 Georg-Johann Lay @ 2019-12-28 11:59 ` Georg-Johann Lay 2019-12-28 11:59 ` Georg-Johann Lay 2020-01-06 13:08 ` [Ping^2][patch,avr, " Georg-Johann Lay 5 siblings, 0 replies; 22+ messages in thread From: Georg-Johann Lay @ 2019-12-28 11:59 UTC (permalink / raw) To: GCC Patches Ping #1 > Now that the avr backend can support 64-bit floats by means of > configure-options --with-double= and --with-long-double=, this patch > series adds some routines to support it. > > It's an ad-hoc, avr-specific implementation in assembly and GNU-C which > is added as a new subfolder in libgcc/config/avr/libf7. > > Patch 1/3 is the GCC changes: Documentation and new avr-specific > configure options: > > --with-libf7 selects to which level double support from libf7 is added > to libgcc. > > --with-double-comparison select what FLOAT_LIB_COMPARE_RETURNS_BOOL > returns. > > I wrote the libf7 code from scratch and put it under GPL v3 + library > exception, so it should be no problem to have it as part of libgcc. > > Patch 2/3 is the libgcc additions: > > --with-libf7 selects which makefile-snips from libf7 to use. > > Patch 3/3 is the actual libf7 implementation. A great deal of which is > assembly, together with C + inline assembly for higher routines. > > Ok for trunk? > > Johann > ^ permalink raw reply [flat|nested] 22+ messages in thread
* [PING^1][patch,avr, 0/3] Support 64-bit (long) double. 2019-12-16 16:43 [patch,avr, 0/3] Support 64-bit (long) double Georg-Johann Lay ` (3 preceding siblings ...) 2019-12-28 11:59 ` [PING^1][patch,avr, 0/3] Support 64-bit (long) double Georg-Johann Lay @ 2019-12-28 11:59 ` Georg-Johann Lay 2020-01-06 13:08 ` [Ping^2][patch,avr, " Georg-Johann Lay 5 siblings, 0 replies; 22+ messages in thread From: Georg-Johann Lay @ 2019-12-28 11:59 UTC (permalink / raw) To: GCC Patches; +Cc: Denis Chertykov Ping #1 > Now that the avr backend can support 64-bit floats by means of > configure-options --with-double= and --with-long-double=, this patch > series adds some routines to support it. > > It's an ad-hoc, avr-specific implementation in assembly and GNU-C which > is added as a new subfolder in libgcc/config/avr/libf7. > > Patch 1/3 is the GCC changes: Documentation and new avr-specific > configure options: > > --with-libf7 selects to which level double support from libf7 is added > to libgcc. > > --with-double-comparison select what FLOAT_LIB_COMPARE_RETURNS_BOOL > returns. > > I wrote the libf7 code from scratch and put it under GPL v3 + library > exception, so it should be no problem to have it as part of libgcc. > > Patch 2/3 is the libgcc additions: > > --with-libf7 selects which makefile-snips from libf7 to use. > > Patch 3/3 is the actual libf7 implementation. A great deal of which is > assembly, together with C + inline assembly for higher routines. > > Ok for trunk? > > Johann > ^ permalink raw reply [flat|nested] 22+ messages in thread
* [Ping^2][patch,avr, 0/3] Support 64-bit (long) double. 2019-12-16 16:43 [patch,avr, 0/3] Support 64-bit (long) double Georg-Johann Lay ` (4 preceding siblings ...) 2019-12-28 11:59 ` Georg-Johann Lay @ 2020-01-06 13:08 ` Georg-Johann Lay 5 siblings, 0 replies; 22+ messages in thread From: Georg-Johann Lay @ 2020-01-06 13:08 UTC (permalink / raw) To: GCC Patches; +Cc: Denis Chertykov Ping #2 Georg-Johann Lay schrieb: > Now that the avr backend can support 64-bit floats by means of > configure-options --with-double= and --with-long-double=, this patch > series adds some routines to support it. > > It's an ad-hoc, avr-specific implementation in assembly and GNU-C which > is added as a new subfolder in libgcc/config/avr/libf7. > > Patch 1/3 is the GCC changes: Documentation and new avr-specific > configure options: > > --with-libf7 selects to which level double support from libf7 is added > to libgcc. > > --with-double-comparison select what FLOAT_LIB_COMPARE_RETURNS_BOOL > returns. > > I wrote the libf7 code from scratch and put it under GPL v3 + library > exception, so it should be no problem to have it as part of libgcc. > > Patch 2/3 is the libgcc additions: > > --with-libf7 selects which makefile-snips from libf7 to use. > > Patch 3/3 is the actual libf7 implementation. A great deal of which is > assembly, together with C + inline assembly for higher routines. > > Ok for trunk? > > Johann > ^ permalink raw reply [flat|nested] 22+ messages in thread
end of thread, other threads:[~2020-12-18 16:53 UTC | newest] Thread overview: 22+ messages (download: mbox.gz / follow: Atom feed) -- links below jump to the message on this page -- 2019-12-16 16:43 [patch,avr, 0/3] Support 64-bit (long) double Georg-Johann Lay 2019-12-16 16:46 ` [patch,avr, 1/3] Support 64-bit (long) double: The gcc part Georg-Johann Lay 2019-12-28 13:20 ` [PING^1][patch,avr, " Georg-Johann Lay 2020-01-06 13:08 ` [Ping^2][patch,avr, " Georg-Johann Lay 2020-01-06 16:25 ` [patch,avr, " Jeff Law 2020-01-06 17:10 ` Georg-Johann Lay 2020-01-06 17:16 ` Jeff Law 2020-12-15 7:40 ` AVR maintainership (was: [patch,avr, 1/3] Support 64-bit (long) double: The gcc part.) Gerald Pfeifer 2020-12-15 14:30 ` AVR maintainership (was: [patch, avr, " Segher Boessenkool 2020-12-15 20:42 ` AVR maintainership (was: [patch,avr, " Jeff Law 2020-12-18 16:53 ` AVR maintainership Georg-Johann Lay 2019-12-16 16:49 ` [patch,avr, 2/3] Support 64-bit (long) double: The libgcc changes Georg-Johann Lay 2019-12-28 12:00 ` [PING^1][patch,avr, " Georg-Johann Lay 2020-01-06 13:08 ` [Ping^2][patch,avr, " Georg-Johann Lay 2020-01-06 16:26 ` [patch,avr, " Jeff Law 2019-12-16 17:09 ` [patch,avr, 3/3] Support 64-bit (long) double: libf7 Georg-Johann Lay 2019-12-28 12:02 ` [PING^1][patch,avr, " Georg-Johann Lay 2020-01-06 13:09 ` [Ping^2][patch,avr, " Georg-Johann Lay 2020-01-06 16:26 ` [patch,avr, " Jeff Law 2019-12-28 11:59 ` [PING^1][patch,avr, 0/3] Support 64-bit (long) double Georg-Johann Lay 2019-12-28 11:59 ` Georg-Johann Lay 2020-01-06 13:08 ` [Ping^2][patch,avr, " Georg-Johann Lay
This is a public inbox, see mirroring instructions for how to clone and mirror all data and code used for this inbox; as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).