* [PATCH] Add _Float<N>/_Float<N>X rounding built-ins & improve gimple optimization of _Float<N>/_Float<N>X built-in functions
@ 2017-11-17 5:33 Michael Meissner
2017-11-17 14:26 ` Segher Boessenkool
0 siblings, 1 reply; 12+ messages in thread
From: Michael Meissner @ 2017-11-17 5:33 UTC (permalink / raw)
To: GCC Patches, Segher Boessenkool, David Edelsohn, Bill Schmidt,
Joseph Myers
Cc: Richard Earnshaw, Richard Biener, Jeff Law, Jason Merrill,
David S. Miller, Bernd Schmidt, Ian Lance Taylor, Jim Wilson
[-- Attachment #1: Type: text/plain, Size: 4267 bytes --]
This patch is an enhancement of a previous page that never got approved.
https://gcc.gnu.org/ml/gcc-patches/2017-10/threads.html#02124
In the original patch, I added support to the machine independent
infrastructure to support the rounding built-in functions for _Float<N> and
_Float<N>X types (i.e. roundf128, ceilf128, etc.). I also added PowerPC ISA
3.0 support to generate these built-in functions.
In addition to the previous changes, this patch now adds more optimizations for
_Float<N> and _Float<N>X types in match.pd.
I modified the gencfn generation to dump out <BUILTIN>_ALL case statements and
match.pd operators in addition to <BUILTIN> that puts out the normal float,
double, and long double variants, and <BUILTIN>_FN that puts out the _Float<N>
and _Float<N>X variants. The <BUILTIN>_ALL version puts out the built-in
functions that include float, double, long double, and all of the _Float<N> and
_Float<N>X versions.
I also modified match.pd where possible to use COPYSIGN_ALL, SQRT_ALL,
FMIN_ALL, FMAX_ALL, TRUNC_ALL, FLOOR_ALL, CEIL_ALL, ROUND_ALL, NEARBYINT_ALL,
and RINT_ALL instead of the normal operator lists. Sometimes this wasn't
possible (for example, we don't currently have _Float<N> pow support, so we
can't optimize powf128 (x, 0.5q) into sqrtf128 (x).
I added PowerPC tests for many of the optimizations when the compiler is
targeting ISA 3.0 (power9).
I checked this on an x86-64 system and little endian power8 system, doing a
bootstrap build and running make check. There were no regresions. Can I check
this patch into the trunk?
[gcc]
2017-11-16 Michael Meissner <meissner@linux.vnet.ibm.com>
* builtins.def: (_Float<N> and _Float<N>X BUILT_IN_CEIL): Add
_Float<N> and _Float<N>X variants for rounding built-in
functions.
(_Float<N> and _Float<N>X BUILT_IN_FLOOR): Likewise.
(_Float<N> and _Float<N>X BUILT_IN_NEARBYINT): Likewise.
(_Float<N> and _Float<N>X BUILT_IN_RINT): Likewise.
(_Float<N> and _Float<N>X BUILT_IN_ROUND): Likewise.
(_Float<N> and _Float<N>X BUILT_IN_TRUNC): Likewise.
* builtins.c (mathfn_built_in_2): Likewise.
* internal-fn.def (CEIL): Likewise.
(FLOOR): Likewise.
(NEARBYINT): Likewise.
(RINT): Likewise.
(ROUND): Likewise.
(TRUNC): Likewise.
* convert.c (convert_to_integer_1): Likewise.
* fold-const.c (tree_call_nonnegative_warnv_p): Likewise.
(integer_valued_real_call_p): Likewise.
* fold-const-call.c (fold_const_call_ss): Likewise.
* gencfn-macros.c (print_case_cfn): Change CFN and operator
printers to take a const char * suffix instead of a bool.
(print_define_operator_list): Likewise.
(fltall_suffixes): New list of suffixes, that include the
traditional suffixes as well as all of the _Float<N> and
_Float<N>X suffixes.
(main): For _Float<N> and _Float<N>X functions, emit both
<name>_FN and <name>_ALL variants. The <macro>_FN variant only
has the _Float<N> and _Float<N>X case names or operators. The
<name>_ALL variant has both the traditional and the
_Float<N>/_Float<N>X case names or operators.
* match.pd (COPYSIGN optimizations): Provide optimizations for
_Float<N> and _Float<N>X types where possible.
(MIN/MAX optimizations): Likewise.
(sqrt optimizations): Likewise.
(rounding optimizations): Likewise.
* config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
128-bit round to integer instructions.
(ceil<mode>2): Likewise.
(btrunc<mode>2): Likewise.
(round<mode>2): Likewise.
[gcc/c]
2017-11-16 Michael Meissner <meissner@linux.vnet.ibm.com>
* c-decl.c (header_for_builtin_fn): Add integer rounding _Float<N>
and _Float<N>X built-in functions.
[gcc/testsuite]
2017-11-16 Michael Meissner <meissner@linux.vnet.ibm.com>
* gcc.target/powerpc/float128-hw2.c: Add tests for ceilf128,
floorf128, truncf128, and roundf128.
* gcc.target/powerpc/float128-hw5.c: New tests for _Float128
optimizations added in match.pd.
* gcc.target/powerpc/float128-hw6.c: Likewise.
* gcc.target/powerpc/float128-hw7.c: Likewise.
* gcc.target/powerpc/float128-hw8.c: Likewise.
* gcc.target/powerpc/float128-hw9.c: Likewise.
* gcc.target/powerpc/float128-hw10.c: Likewise.
--
Michael Meissner, IBM
IBM, M/S 2506R, 550 King Street, Littleton, MA 01460-6245, USA
email: meissner@linux.vnet.ibm.com, phone: +1 (978) 899-4797
[-- Attachment #2: ieee128-patch60b --]
[-- Type: text/plain, Size: 32276 bytes --]
Index: gcc/builtins.def
===================================================================
--- gcc/builtins.def (revision 254846)
+++ gcc/builtins.def (working copy)
@@ -343,6 +343,9 @@ DEF_C99_BUILTIN (BUILT_IN_CBRTL,
DEF_LIB_BUILTIN (BUILT_IN_CEIL, "ceil", BT_FN_DOUBLE_DOUBLE, ATTR_CONST_NOTHROW_LEAF_LIST)
DEF_C99_C90RES_BUILTIN (BUILT_IN_CEILF, "ceilf", BT_FN_FLOAT_FLOAT, ATTR_CONST_NOTHROW_LEAF_LIST)
DEF_C99_C90RES_BUILTIN (BUILT_IN_CEILL, "ceill", BT_FN_LONGDOUBLE_LONGDOUBLE, ATTR_CONST_NOTHROW_LEAF_LIST)
+#define CEIL_TYPE(F) BT_FN_##F##_##F
+DEF_EXT_LIB_FLOATN_NX_BUILTINS (BUILT_IN_CEIL, "ceil", CEIL_TYPE, ATTR_CONST_NOTHROW_LEAF_LIST)
+#undef CEIL_TYPE
DEF_C99_BUILTIN (BUILT_IN_COPYSIGN, "copysign", BT_FN_DOUBLE_DOUBLE_DOUBLE, ATTR_CONST_NOTHROW_LEAF_LIST)
DEF_C99_BUILTIN (BUILT_IN_COPYSIGNF, "copysignf", BT_FN_FLOAT_FLOAT_FLOAT, ATTR_CONST_NOTHROW_LEAF_LIST)
DEF_C99_BUILTIN (BUILT_IN_COPYSIGNL, "copysignl", BT_FN_LONGDOUBLE_LONGDOUBLE_LONGDOUBLE, ATTR_CONST_NOTHROW_LEAF_LIST)
@@ -402,6 +405,9 @@ DEF_C99_BUILTIN (BUILT_IN_FEUPDAT
DEF_LIB_BUILTIN (BUILT_IN_FLOOR, "floor", BT_FN_DOUBLE_DOUBLE, ATTR_CONST_NOTHROW_LEAF_LIST)
DEF_C99_C90RES_BUILTIN (BUILT_IN_FLOORF, "floorf", BT_FN_FLOAT_FLOAT, ATTR_CONST_NOTHROW_LEAF_LIST)
DEF_C99_C90RES_BUILTIN (BUILT_IN_FLOORL, "floorl", BT_FN_LONGDOUBLE_LONGDOUBLE, ATTR_CONST_NOTHROW_LEAF_LIST)
+#define FLOOR_TYPE(F) BT_FN_##F##_##F
+DEF_EXT_LIB_FLOATN_NX_BUILTINS (BUILT_IN_FLOOR, "floor", FLOOR_TYPE, ATTR_CONST_NOTHROW_LEAF_LIST)
+#undef FLOOR_TYPE
DEF_C99_BUILTIN (BUILT_IN_FMA, "fma", BT_FN_DOUBLE_DOUBLE_DOUBLE_DOUBLE, ATTR_MATHFN_FPROUNDING)
DEF_C99_BUILTIN (BUILT_IN_FMAF, "fmaf", BT_FN_FLOAT_FLOAT_FLOAT_FLOAT, ATTR_MATHFN_FPROUNDING)
DEF_C99_BUILTIN (BUILT_IN_FMAL, "fmal", BT_FN_LONGDOUBLE_LONGDOUBLE_LONGDOUBLE_LONGDOUBLE, ATTR_MATHFN_FPROUNDING)
@@ -539,6 +545,9 @@ DEF_GCC_FLOATN_NX_BUILTINS (BUILT_IN_NAN
DEF_C99_BUILTIN (BUILT_IN_NEARBYINT, "nearbyint", BT_FN_DOUBLE_DOUBLE, ATTR_CONST_NOTHROW_LEAF_LIST)
DEF_C99_BUILTIN (BUILT_IN_NEARBYINTF, "nearbyintf", BT_FN_FLOAT_FLOAT, ATTR_CONST_NOTHROW_LEAF_LIST)
DEF_C99_BUILTIN (BUILT_IN_NEARBYINTL, "nearbyintl", BT_FN_LONGDOUBLE_LONGDOUBLE, ATTR_CONST_NOTHROW_LEAF_LIST)
+#define NEARBYINT_TYPE(F) BT_FN_##F##_##F
+DEF_EXT_LIB_FLOATN_NX_BUILTINS (BUILT_IN_NEARBYINT, "nearbyint", NEARBYINT_TYPE, ATTR_CONST_NOTHROW_LEAF_LIST)
+#undef NEARBYINT_TYPE
DEF_C99_BUILTIN (BUILT_IN_NEXTAFTER, "nextafter", BT_FN_DOUBLE_DOUBLE_DOUBLE, ATTR_MATHFN_FPROUNDING_ERRNO)
DEF_C99_BUILTIN (BUILT_IN_NEXTAFTERF, "nextafterf", BT_FN_FLOAT_FLOAT_FLOAT, ATTR_MATHFN_FPROUNDING_ERRNO)
DEF_C99_BUILTIN (BUILT_IN_NEXTAFTERL, "nextafterl", BT_FN_LONGDOUBLE_LONGDOUBLE_LONGDOUBLE, ATTR_MATHFN_FPROUNDING_ERRNO)
@@ -563,9 +572,15 @@ DEF_C99_BUILTIN (BUILT_IN_REMQUOL
DEF_C99_BUILTIN (BUILT_IN_RINT, "rint", BT_FN_DOUBLE_DOUBLE, ATTR_MATHFN_FPROUNDING)
DEF_C99_BUILTIN (BUILT_IN_RINTF, "rintf", BT_FN_FLOAT_FLOAT, ATTR_MATHFN_FPROUNDING)
DEF_C99_BUILTIN (BUILT_IN_RINTL, "rintl", BT_FN_LONGDOUBLE_LONGDOUBLE, ATTR_MATHFN_FPROUNDING)
+#define RINT_TYPE(F) BT_FN_##F##_##F
+DEF_EXT_LIB_FLOATN_NX_BUILTINS (BUILT_IN_RINT, "rint", RINT_TYPE, ATTR_CONST_NOTHROW_LEAF_LIST)
+#undef RINT_TYPE
DEF_C99_BUILTIN (BUILT_IN_ROUND, "round", BT_FN_DOUBLE_DOUBLE, ATTR_CONST_NOTHROW_LEAF_LIST)
DEF_C99_BUILTIN (BUILT_IN_ROUNDF, "roundf", BT_FN_FLOAT_FLOAT, ATTR_CONST_NOTHROW_LEAF_LIST)
DEF_C99_BUILTIN (BUILT_IN_ROUNDL, "roundl", BT_FN_LONGDOUBLE_LONGDOUBLE, ATTR_CONST_NOTHROW_LEAF_LIST)
+#define ROUND_TYPE(F) BT_FN_##F##_##F
+DEF_EXT_LIB_FLOATN_NX_BUILTINS (BUILT_IN_ROUND, "round", ROUND_TYPE, ATTR_CONST_NOTHROW_LEAF_LIST)
+#undef ROUND_TYPE
DEF_EXT_LIB_BUILTIN (BUILT_IN_SCALB, "scalb", BT_FN_DOUBLE_DOUBLE_DOUBLE, ATTR_MATHFN_FPROUNDING_ERRNO)
DEF_EXT_LIB_BUILTIN (BUILT_IN_SCALBF, "scalbf", BT_FN_FLOAT_FLOAT_FLOAT, ATTR_MATHFN_FPROUNDING_ERRNO)
DEF_EXT_LIB_BUILTIN (BUILT_IN_SCALBL, "scalbl", BT_FN_LONGDOUBLE_LONGDOUBLE_LONGDOUBLE, ATTR_MATHFN_FPROUNDING_ERRNO)
@@ -611,6 +626,9 @@ DEF_C99_BUILTIN (BUILT_IN_TGAMMAL
DEF_C99_BUILTIN (BUILT_IN_TRUNC, "trunc", BT_FN_DOUBLE_DOUBLE, ATTR_CONST_NOTHROW_LEAF_LIST)
DEF_C99_BUILTIN (BUILT_IN_TRUNCF, "truncf", BT_FN_FLOAT_FLOAT, ATTR_CONST_NOTHROW_LEAF_LIST)
DEF_C99_BUILTIN (BUILT_IN_TRUNCL, "truncl", BT_FN_LONGDOUBLE_LONGDOUBLE, ATTR_CONST_NOTHROW_LEAF_LIST)
+#define TRUNC_TYPE(F) BT_FN_##F##_##F
+DEF_EXT_LIB_FLOATN_NX_BUILTINS (BUILT_IN_TRUNC, "trunc", TRUNC_TYPE, ATTR_CONST_NOTHROW_LEAF_LIST)
+#undef TRUNC_TYPE
DEF_EXT_LIB_BUILTIN (BUILT_IN_Y0, "y0", BT_FN_DOUBLE_DOUBLE, ATTR_MATHFN_FPROUNDING_ERRNO)
DEF_EXT_LIB_BUILTIN (BUILT_IN_Y0F, "y0f", BT_FN_FLOAT_FLOAT, ATTR_MATHFN_FPROUNDING_ERRNO)
DEF_EXT_LIB_BUILTIN (BUILT_IN_Y0L, "y0l", BT_FN_LONGDOUBLE_LONGDOUBLE, ATTR_MATHFN_FPROUNDING_ERRNO)
Index: gcc/builtins.c
===================================================================
--- gcc/builtins.c (revision 254846)
+++ gcc/builtins.c (working copy)
@@ -1872,7 +1872,7 @@ mathfn_built_in_2 (tree type, combined_f
CASE_MATHFN (ATAN2)
CASE_MATHFN (ATANH)
CASE_MATHFN (CBRT)
- CASE_MATHFN (CEIL)
+ CASE_MATHFN_FLOATN (CEIL)
CASE_MATHFN (CEXPI)
CASE_MATHFN_FLOATN (COPYSIGN)
CASE_MATHFN (COS)
@@ -1886,7 +1886,7 @@ mathfn_built_in_2 (tree type, combined_f
CASE_MATHFN (EXPM1)
CASE_MATHFN (FABS)
CASE_MATHFN (FDIM)
- CASE_MATHFN (FLOOR)
+ CASE_MATHFN_FLOATN (FLOOR)
CASE_MATHFN_FLOATN (FMA)
CASE_MATHFN_FLOATN (FMAX)
CASE_MATHFN_FLOATN (FMIN)
@@ -1925,7 +1925,7 @@ mathfn_built_in_2 (tree type, combined_f
CASE_MATHFN (MODF)
CASE_MATHFN (NAN)
CASE_MATHFN (NANS)
- CASE_MATHFN (NEARBYINT)
+ CASE_MATHFN_FLOATN (NEARBYINT)
CASE_MATHFN (NEXTAFTER)
CASE_MATHFN (NEXTTOWARD)
CASE_MATHFN (POW)
@@ -1933,8 +1933,8 @@ mathfn_built_in_2 (tree type, combined_f
CASE_MATHFN (POW10)
CASE_MATHFN (REMAINDER)
CASE_MATHFN (REMQUO)
- CASE_MATHFN (RINT)
- CASE_MATHFN (ROUND)
+ CASE_MATHFN_FLOATN (RINT)
+ CASE_MATHFN_FLOATN (ROUND)
CASE_MATHFN (SCALB)
CASE_MATHFN (SCALBLN)
CASE_MATHFN (SCALBN)
@@ -1947,7 +1947,7 @@ mathfn_built_in_2 (tree type, combined_f
CASE_MATHFN (TAN)
CASE_MATHFN (TANH)
CASE_MATHFN (TGAMMA)
- CASE_MATHFN (TRUNC)
+ CASE_MATHFN_FLOATN (TRUNC)
CASE_MATHFN (Y0)
CASE_MATHFN (Y1)
CASE_MATHFN (YN)
Index: gcc/internal-fn.def
===================================================================
--- gcc/internal-fn.def (revision 254846)
+++ gcc/internal-fn.def (working copy)
@@ -118,12 +118,12 @@ DEF_INTERNAL_FLT_FLOATN_FN (SQRT, ECF_CO
DEF_INTERNAL_FLT_FN (TAN, ECF_CONST, tan, unary)
/* FP rounding. */
-DEF_INTERNAL_FLT_FN (CEIL, ECF_CONST, ceil, unary)
-DEF_INTERNAL_FLT_FN (FLOOR, ECF_CONST, floor, unary)
-DEF_INTERNAL_FLT_FN (NEARBYINT, ECF_CONST, nearbyint, unary)
-DEF_INTERNAL_FLT_FN (RINT, ECF_CONST, rint, unary)
-DEF_INTERNAL_FLT_FN (ROUND, ECF_CONST, round, unary)
-DEF_INTERNAL_FLT_FN (TRUNC, ECF_CONST, btrunc, unary)
+DEF_INTERNAL_FLT_FLOATN_FN (CEIL, ECF_CONST, ceil, unary)
+DEF_INTERNAL_FLT_FLOATN_FN (FLOOR, ECF_CONST, floor, unary)
+DEF_INTERNAL_FLT_FLOATN_FN (NEARBYINT, ECF_CONST, nearbyint, unary)
+DEF_INTERNAL_FLT_FLOATN_FN (RINT, ECF_CONST, rint, unary)
+DEF_INTERNAL_FLT_FLOATN_FN (ROUND, ECF_CONST, round, unary)
+DEF_INTERNAL_FLT_FLOATN_FN (TRUNC, ECF_CONST, btrunc, unary)
/* Binary math functions. */
DEF_INTERNAL_FLT_FN (ATAN2, ECF_CONST, atan2, binary)
Index: gcc/convert.c
===================================================================
--- gcc/convert.c (revision 254846)
+++ gcc/convert.c (working copy)
@@ -554,6 +554,7 @@ convert_to_integer_1 (tree type, tree ex
switch (fcode)
{
CASE_FLT_FN (BUILT_IN_CEIL):
+ CASE_FLT_FN_FLOATN_NX (BUILT_IN_CEIL):
/* Only convert in ISO C99 mode. */
if (!targetm.libc_has_function (function_c99_misc))
break;
@@ -570,6 +571,7 @@ convert_to_integer_1 (tree type, tree ex
break;
CASE_FLT_FN (BUILT_IN_FLOOR):
+ CASE_FLT_FN_FLOATN_NX (BUILT_IN_FLOOR):
/* Only convert in ISO C99 mode. */
if (!targetm.libc_has_function (function_c99_misc))
break;
@@ -586,6 +588,7 @@ convert_to_integer_1 (tree type, tree ex
break;
CASE_FLT_FN (BUILT_IN_ROUND):
+ CASE_FLT_FN_FLOATN_NX (BUILT_IN_ROUND):
/* Only convert in ISO C99 mode and with -fno-math-errno. */
if (!targetm.libc_has_function (function_c99_misc) || flag_errno_math)
break;
@@ -602,11 +605,13 @@ convert_to_integer_1 (tree type, tree ex
break;
CASE_FLT_FN (BUILT_IN_NEARBYINT):
+ CASE_FLT_FN_FLOATN_NX (BUILT_IN_NEARBYINT):
/* Only convert nearbyint* if we can ignore math exceptions. */
if (flag_trapping_math)
break;
gcc_fallthrough ();
CASE_FLT_FN (BUILT_IN_RINT):
+ CASE_FLT_FN_FLOATN_NX (BUILT_IN_RINT):
/* Only convert in ISO C99 mode and with -fno-math-errno. */
if (!targetm.libc_has_function (function_c99_misc) || flag_errno_math)
break;
@@ -623,6 +628,7 @@ convert_to_integer_1 (tree type, tree ex
break;
CASE_FLT_FN (BUILT_IN_TRUNC):
+ CASE_FLT_FN_FLOATN_NX (BUILT_IN_TRUNC):
return convert_to_integer_1 (type, CALL_EXPR_ARG (s_expr, 0), dofold);
default:
Index: gcc/fold-const.c
===================================================================
--- gcc/fold-const.c (revision 254846)
+++ gcc/fold-const.c (working copy)
@@ -12784,9 +12784,11 @@ tree_call_nonnegative_warnv_p (tree type
CASE_CFN_ATANH:
CASE_CFN_CBRT:
CASE_CFN_CEIL:
+ CASE_CFN_CEIL_FN:
CASE_CFN_ERF:
CASE_CFN_EXPM1:
CASE_CFN_FLOOR:
+ CASE_CFN_FLOOR_FN:
CASE_CFN_FMOD:
CASE_CFN_FREXP:
CASE_CFN_ICEIL:
@@ -12804,8 +12806,11 @@ tree_call_nonnegative_warnv_p (tree type
CASE_CFN_LROUND:
CASE_CFN_MODF:
CASE_CFN_NEARBYINT:
+ CASE_CFN_NEARBYINT_FN:
CASE_CFN_RINT:
+ CASE_CFN_RINT_FN:
CASE_CFN_ROUND:
+ CASE_CFN_ROUND_FN:
CASE_CFN_SCALB:
CASE_CFN_SCALBLN:
CASE_CFN_SCALBN:
@@ -12814,6 +12819,7 @@ tree_call_nonnegative_warnv_p (tree type
CASE_CFN_SINH:
CASE_CFN_TANH:
CASE_CFN_TRUNC:
+ CASE_CFN_TRUNC_FN:
/* True if the 1st argument is nonnegative. */
return RECURSE (arg0);
@@ -13319,11 +13325,17 @@ integer_valued_real_call_p (combined_fn
switch (fn)
{
CASE_CFN_CEIL:
+ CASE_CFN_CEIL_FN:
CASE_CFN_FLOOR:
+ CASE_CFN_FLOOR_FN:
CASE_CFN_NEARBYINT:
+ CASE_CFN_NEARBYINT_FN:
CASE_CFN_RINT:
+ CASE_CFN_RINT_FN:
CASE_CFN_ROUND:
+ CASE_CFN_ROUND_FN:
CASE_CFN_TRUNC:
+ CASE_CFN_TRUNC_FN:
return true;
CASE_CFN_FMIN:
Index: gcc/fold-const-call.c
===================================================================
--- gcc/fold-const-call.c (revision 254846)
+++ gcc/fold-const-call.c (working copy)
@@ -699,6 +699,7 @@ fold_const_call_ss (real_value *result,
&& do_mpfr_arg1 (result, mpfr_y1, arg, format));
CASE_CFN_FLOOR:
+ CASE_CFN_FLOOR_FN:
if (!REAL_VALUE_ISNAN (*arg) || !flag_errno_math)
{
real_floor (result, format, arg);
@@ -707,6 +708,7 @@ fold_const_call_ss (real_value *result,
return false;
CASE_CFN_CEIL:
+ CASE_CFN_CEIL_FN:
if (!REAL_VALUE_ISNAN (*arg) || !flag_errno_math)
{
real_ceil (result, format, arg);
@@ -715,10 +717,12 @@ fold_const_call_ss (real_value *result,
return false;
CASE_CFN_TRUNC:
+ CASE_CFN_TRUNC_FN:
real_trunc (result, format, arg);
return true;
CASE_CFN_ROUND:
+ CASE_CFN_ROUND_FN:
if (!REAL_VALUE_ISNAN (*arg) || !flag_errno_math)
{
real_round (result, format, arg);
Index: gcc/gencfn-macros.c
===================================================================
--- gcc/gencfn-macros.c (revision 254846)
+++ gcc/gencfn-macros.c (working copy)
@@ -94,13 +94,15 @@ is_group (string_set *builtins, const ch
/* Print a macro for all combined functions related to NAME, with the
null-terminated list of suffixes in SUFFIXES. INTERNAL_P says whether
- CFN_<NAME> also exists. */
+ CFN_<NAME> also exists. FLOATN_P is a suffix to the operator name, blank
+ for normal operators, "_FN" for _Float<N>/_Float<N>X operators only, and
+ "_ALL" for both the traditional operators and the _Float<N>/_Float<N>X
+ operators. */
static void
print_case_cfn (const char *name, bool internal_p,
- const char *const *suffixes, bool floatn_p)
+ const char *const *suffixes, const char *floatn)
{
- const char *floatn = (floatn_p) ? "_FN" : "";
printf ("#define CASE_CFN_%s%s", name, floatn);
if (internal_p)
printf (" \\\n case CFN_%s%s", name, floatn);
@@ -110,15 +112,18 @@ print_case_cfn (const char *name, bool i
printf ("\n");
}
-/* Print an operator list for all combined functions related to NAME,
- with the null-terminated list of suffixes in SUFFIXES. INTERNAL_P
- says whether CFN_<NAME> also exists. */
+/* Print an operator list for all combined functions related to NAME, with the
+ null-terminated list of suffixes in SUFFIXES. INTERNAL_P says whether
+ CFN_<NAME> also exists. FLOATN_P is a suffix to the operator name, blank
+ for normal operators, "_FN" for _Float<N>/_Float<N>X operators only, and
+ "_ALL" for both the traditional operators and the _Float<N>/_Float<N>X
+ operators. */
static void
print_define_operator_list (const char *name, bool internal_p,
- const char *const *suffixes, bool floatn_p)
+ const char *const *suffixes,
+ const char *floatn)
{
- const char *floatn = (floatn_p) ? "_FN" : "";
printf ("(define_operator_list %s%s\n", name, floatn);
for (unsigned int i = 0; suffixes[i]; ++i)
printf (" BUILT_IN_%s%s\n", name, suffixes[i]);
@@ -152,6 +157,9 @@ const char *const internal_fn_int_names[
static const char *const flt_suffixes[] = { "F", "", "L", NULL };
static const char *const fltfn_suffixes[] = { "F16", "F32", "F64", "F128",
"F32X", "F64X", "F128X", NULL };
+static const char *const fltall_suffixes[] = { "F", "", "L", "F16", "F32",
+ "F64", "F128", "F32X", "F64X",
+ "F128X", NULL };
static const char *const int_suffixes[] = { "", "L", "LL", "IMAX", NULL };
static const char *const *const suffix_lists[] = {
@@ -212,22 +220,31 @@ main (int argc, char **argv)
bool internal_p = internal_fns.contains (root);
if (type == 'c')
- print_case_cfn (root, internal_p, suffix, false);
+ print_case_cfn (root, internal_p, suffix, "");
else
- print_define_operator_list (root, internal_p,
- suffix, false);
+ print_define_operator_list (root, internal_p, suffix, "");
/* Support the _Float<N> and _Float<N>X math functions if
- they exist. We put these out as a separate CFN macro,
- so code can add support or not as needed. */
+ they exist. We put these out as a separate CFN or
+ operator macro, so code can add support or not as
+ needed. We also put out a combined CFN or operator
+ macro that includes both the traditional names and the
+ _Float<N> and _Float<N>X versions. */
if (suffix == flt_suffixes
&& is_group (&builtins, root, fltfn_suffixes))
{
if (type == 'c')
- print_case_cfn (root, false, fltfn_suffixes, true);
+ {
+ print_case_cfn (root, false, fltfn_suffixes, "_FN");
+ print_case_cfn (root, false, fltall_suffixes, "_ALL");
+ }
else
- print_define_operator_list (root, false, fltfn_suffixes,
- true);
+ {
+ print_define_operator_list (root, false,
+ fltfn_suffixes, "_FN");
+ print_define_operator_list (root, internal_p,
+ fltall_suffixes, "_ALL");
+ }
}
}
}
Index: gcc/match.pd
===================================================================
--- gcc/match.pd (revision 254846)
+++ gcc/match.pd (working copy)
@@ -191,21 +191,21 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
/* Transform X * copysign (1.0, X) into abs(X). */
(simplify
- (mult:c @0 (COPYSIGN real_onep @0))
+ (mult:c @0 (COPYSIGN_ALL real_onep @0))
(if (!HONOR_NANS (type) && !HONOR_SIGNED_ZEROS (type))
(abs @0)))
/* Transform X * copysign (1.0, -X) into -abs(X). */
(simplify
- (mult:c @0 (COPYSIGN real_onep (negate @0)))
+ (mult:c @0 (COPYSIGN_ALL real_onep (negate @0)))
(if (!HONOR_NANS (type) && !HONOR_SIGNED_ZEROS (type))
(negate (abs @0))))
/* Transform copysign (CST, X) into copysign (ABS(CST), X). */
(simplify
- (COPYSIGN REAL_CST@0 @1)
+ (COPYSIGN_ALL REAL_CST@0 @1)
(if (REAL_VALUE_NEGATIVE (TREE_REAL_CST (@0)))
- (COPYSIGN (negate @0) @1)))
+ (COPYSIGN_ALL (negate @0) @1)))
/* X * 1, X / 1 -> X. */
(for op (mult trunc_div ceil_div floor_div round_div exact_div)
@@ -534,7 +534,7 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
(hypots @0 (op @1))
(hypots @0 @1)))
/* copysign(-x, y) and copysign(abs(x), y) -> copysign(x, y). */
- (for copysigns (COPYSIGN)
+ (for copysigns (COPYSIGN_ALL)
(simplify
(copysigns (op @0) @1)
(copysigns @0 @1))))
@@ -579,7 +579,7 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
(hypots @0 @1)))
/* copysign(x, CST) -> [-]abs (x). */
-(for copysigns (COPYSIGN)
+(for copysigns (COPYSIGN_ALL)
(simplify
(copysigns @0 REAL_CST@1)
(if (REAL_VALUE_NEGATIVE (TREE_REAL_CST (@1)))
@@ -587,13 +587,13 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
(abs @0))))
/* copysign(copysign(x, y), z) -> copysign(x, z). */
-(for copysigns (COPYSIGN)
+(for copysigns (COPYSIGN_ALL)
(simplify
(copysigns (copysigns @0 @1) @2)
(copysigns @0 @2)))
/* copysign(x,y)*copysign(x,y) -> x*x. */
-(for copysigns (COPYSIGN)
+(for copysigns (COPYSIGN_ALL)
(simplify
(mult (copysigns@2 @0 @1) @2)
(mult @0 @0)))
@@ -1802,7 +1802,7 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
/* Simplifications of MIN_EXPR, MAX_EXPR, fmin() and fmax(). */
-(for minmax (min max FMIN FMIN_FN FMAX FMAX_FN)
+(for minmax (min max FMIN_ALL FMAX_ALL)
(simplify
(minmax @0 @0)
@0))
@@ -1880,7 +1880,7 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
&& TYPE_PRECISION (TREE_TYPE (@0)) > TYPE_PRECISION (type))
(minmax @1 (convert @2)))))
-(for minmax (FMIN FMIN_FN FMAX FMAX_FN)
+(for minmax (FMIN_ALL FMAX_ALL)
/* If either argument is NaN, return the other one. Avoid the
transformation if we get (and honor) a signalling NaN. */
(simplify
@@ -1895,20 +1895,14 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
worry about it either. */
(if (flag_finite_math_only)
(simplify
- (FMIN @0 @1)
+ (FMIN_ALL @0 @1)
(min @0 @1))
(simplify
- (FMIN_FN @0 @1)
- (min @0 @1))
- (simplify
- (FMAX @0 @1)
- (max @0 @1))
- (simplify
- (FMAX_FN @0 @1)
+ (FMAX_ALL @0 @1)
(max @0 @1)))
/* min (-A, -B) -> -max (A, B) */
-(for minmax (min max FMIN FMIN_FN FMAX FMAX_FN)
- maxmin (max min FMAX FMAX_FN FMIN FMAX_FN)
+(for minmax (min max FMIN_ALL FMAX_ALL)
+ maxmin (max min FMAX_ALL FMIN_ALL)
(simplify
(minmax (negate:s@2 @0) (negate:s@3 @1))
(if (FLOAT_TYPE_P (TREE_TYPE (@0))
@@ -3697,7 +3691,7 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
(if (flag_unsafe_math_optimizations)
/* Simplify sqrt(x) * sqrt(x) -> x. */
(simplify
- (mult (SQRT@1 @0) @1)
+ (mult (SQRT_ALL@1 @0) @1)
(if (!HONOR_SNANS (type))
@0))
@@ -3850,12 +3844,12 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
(abs @0))
/* trunc(trunc(x)) -> trunc(x), etc. */
-(for fns (TRUNC FLOOR CEIL ROUND NEARBYINT RINT)
+(for fns (TRUNC_ALL FLOOR_ALL CEIL_ALL ROUND_ALL NEARBYINT_ALL RINT_ALL)
(simplify
(fns (fns @0))
(fns @0)))
/* f(x) -> x if x is integer valued and f does nothing for such values. */
-(for fns (TRUNC FLOOR CEIL ROUND NEARBYINT RINT)
+(for fns (TRUNC_ALL FLOOR_ALL CEIL_ALL ROUND_ALL NEARBYINT_ALL RINT_ALL)
(simplify
(fns integer_valued_real_p@0)
@0))
@@ -3872,12 +3866,12 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
(simplify
/* copysign(x,x) -> x. */
- (COPYSIGN @0 @0)
+ (COPYSIGN_ALL @0 @0)
@0)
(simplify
/* copysign(x,y) -> fabs(x) if y is nonnegative. */
- (COPYSIGN @0 tree_expr_nonnegative_p@1)
+ (COPYSIGN_ALL @0 tree_expr_nonnegative_p@1)
(abs @0))
(for scale (LDEXP SCALBN SCALBLN)
@@ -4028,8 +4022,8 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
(if (canonicalize_math_p ())
/* floor(x) -> trunc(x) if x is nonnegative. */
- (for floors (FLOOR)
- truncs (TRUNC)
+ (for floors (FLOOR_ALL)
+ truncs (TRUNC_ALL)
(simplify
(floors tree_expr_nonnegative_p@0)
(truncs @0))))
Index: gcc/config/rs6000/rs6000.md
===================================================================
--- gcc/config/rs6000/rs6000.md (revision 254846)
+++ gcc/config/rs6000/rs6000.md (working copy)
@@ -14733,6 +14733,47 @@ (define_insn_and_split "floatuns<QHI:mod
(set_attr "type" "vecfloat")
(set_attr "size" "128")])
+;; IEEE 128-bit round to integer built-in functions
+(define_insn "floor<mode>2"
+ [(set (match_operand:IEEE128 0 "altivec_register_operand" "=v")
+ (unspec:IEEE128
+ [(match_operand:IEEE128 1 "altivec_register_operand" "v")]
+ UNSPEC_FRIM))]
+ "TARGET_FLOAT128_HW && FLOAT128_IEEE_P (<MODE>mode)"
+ "xsrqpi 1,%0,%1,3"
+ [(set_attr "type" "vecfloat")
+ (set_attr "size" "128")])
+
+(define_insn "ceil<mode>2"
+ [(set (match_operand:IEEE128 0 "altivec_register_operand" "=v")
+ (unspec:IEEE128
+ [(match_operand:IEEE128 1 "altivec_register_operand" "v")]
+ UNSPEC_FRIP))]
+ "TARGET_FLOAT128_HW && FLOAT128_IEEE_P (<MODE>mode)"
+ "xsrqpi 1,%0,%1,2"
+ [(set_attr "type" "vecfloat")
+ (set_attr "size" "128")])
+
+(define_insn "btrunc<mode>2"
+ [(set (match_operand:IEEE128 0 "altivec_register_operand" "=v")
+ (unspec:IEEE128
+ [(match_operand:IEEE128 1 "altivec_register_operand" "v")]
+ UNSPEC_FRIZ))]
+ "TARGET_FLOAT128_HW && FLOAT128_IEEE_P (<MODE>mode)"
+ "xsrqpi 1,%0,%1,0"
+ [(set_attr "type" "vecfloat")
+ (set_attr "size" "128")])
+
+(define_insn "round<mode>2"
+ [(set (match_operand:IEEE128 0 "altivec_register_operand" "=v")
+ (unspec:IEEE128
+ [(match_operand:IEEE128 1 "altivec_register_operand" "v")]
+ UNSPEC_FRIN))]
+ "TARGET_FLOAT128_HW && FLOAT128_IEEE_P (<MODE>mode)"
+ "xsrqpi 0,%0,%1,3"
+ [(set_attr "type" "vecfloat")
+ (set_attr "size" "128")])
+
;; IEEE 128-bit instructions with round to odd semantics
(define_insn "add<mode>3_odd"
[(set (match_operand:IEEE128 0 "altivec_register_operand" "=v")
Index: gcc/c/c-decl.c
===================================================================
--- gcc/c/c-decl.c (revision 254846)
+++ gcc/c/c-decl.c (working copy)
@@ -3162,6 +3162,7 @@ header_for_builtin_fn (enum built_in_fun
CASE_FLT_FN (BUILT_IN_ATAN2):
CASE_FLT_FN (BUILT_IN_CBRT):
CASE_FLT_FN (BUILT_IN_CEIL):
+ CASE_FLT_FN_FLOATN_NX (BUILT_IN_CEIL):
CASE_FLT_FN (BUILT_IN_COPYSIGN):
CASE_FLT_FN_FLOATN_NX (BUILT_IN_COPYSIGN):
CASE_FLT_FN (BUILT_IN_COS):
@@ -3175,6 +3176,7 @@ header_for_builtin_fn (enum built_in_fun
CASE_FLT_FN_FLOATN_NX (BUILT_IN_FABS):
CASE_FLT_FN (BUILT_IN_FDIM):
CASE_FLT_FN (BUILT_IN_FLOOR):
+ CASE_FLT_FN_FLOATN_NX (BUILT_IN_FLOOR):
CASE_FLT_FN (BUILT_IN_FMA):
CASE_FLT_FN_FLOATN_NX (BUILT_IN_FMA):
CASE_FLT_FN (BUILT_IN_FMAX):
@@ -3199,13 +3201,16 @@ header_for_builtin_fn (enum built_in_fun
CASE_FLT_FN (BUILT_IN_MODF):
CASE_FLT_FN (BUILT_IN_NAN):
CASE_FLT_FN (BUILT_IN_NEARBYINT):
+ CASE_FLT_FN_FLOATN_NX (BUILT_IN_NEARBYINT):
CASE_FLT_FN (BUILT_IN_NEXTAFTER):
CASE_FLT_FN (BUILT_IN_NEXTTOWARD):
CASE_FLT_FN (BUILT_IN_POW):
CASE_FLT_FN (BUILT_IN_REMAINDER):
CASE_FLT_FN (BUILT_IN_REMQUO):
CASE_FLT_FN (BUILT_IN_RINT):
+ CASE_FLT_FN_FLOATN_NX (BUILT_IN_RINT):
CASE_FLT_FN (BUILT_IN_ROUND):
+ CASE_FLT_FN_FLOATN_NX (BUILT_IN_ROUND):
CASE_FLT_FN (BUILT_IN_SCALBLN):
CASE_FLT_FN (BUILT_IN_SCALBN):
CASE_FLT_FN (BUILT_IN_SIN):
@@ -3217,6 +3222,7 @@ header_for_builtin_fn (enum built_in_fun
CASE_FLT_FN (BUILT_IN_TANH):
CASE_FLT_FN (BUILT_IN_TGAMMA):
CASE_FLT_FN (BUILT_IN_TRUNC):
+ CASE_FLT_FN_FLOATN_NX (BUILT_IN_TRUNC):
case BUILT_IN_ISINF:
case BUILT_IN_ISNAN:
return "<math.h>";
Index: gcc/testsuite/gcc.target/powerpc/float128-hw2.c
===================================================================
--- gcc/testsuite/gcc.target/powerpc/float128-hw2.c (revision 254846)
+++ gcc/testsuite/gcc.target/powerpc/float128-hw2.c (working copy)
@@ -14,6 +14,10 @@
extern _Float128 copysignf128 (_Float128, _Float128);
extern _Float128 sqrtf128 (_Float128);
extern _Float128 fmaf128 (_Float128, _Float128, _Float128);
+extern _Float128 ceilf128 (_Float128);
+extern _Float128 floorf128 (_Float128);
+extern _Float128 truncf128 (_Float128);
+extern _Float128 roundf128 (_Float128);
_Float128
do_copysign (_Float128 a, _Float128 b)
@@ -51,10 +55,35 @@ do_nfms (_Float128 a, _Float128 b, _Floa
return -fmaf128 (a, b, -c);
}
+_Float128
+do_ceil (_Float128 a)
+{
+ return ceilf128 (a);
+}
+
+_Float128
+do_floor (_Float128 a)
+{
+ return floorf128 (a);
+}
+
+_Float128
+do_trunc (_Float128 a)
+{
+ return truncf128 (a);
+}
+
+_Float128
+do_round (_Float128 a)
+{
+ return roundf128 (a);
+}
+
/* { dg-final { scan-assembler {\mxscpsgnqp\M} } } */
/* { dg-final { scan-assembler {\mxssqrtqp\M} } } */
/* { dg-final { scan-assembler {\mxsmaddqp\M} } } */
/* { dg-final { scan-assembler {\mxsmsubqp\M} } } */
/* { dg-final { scan-assembler {\mxsnmaddqp\M} } } */
/* { dg-final { scan-assembler {\mxsnmsubqp\M} } } */
+/* { dg-final { scan-assembler {\mxsrqpi\M} } } */
/* { dg-final { scan-assembler-not {\mbl\M} } } */
Index: gcc/testsuite/gcc.target/powerpc/float128-hw5.c
===================================================================
--- gcc/testsuite/gcc.target/powerpc/float128-hw5.c (revision 0)
+++ gcc/testsuite/gcc.target/powerpc/float128-hw5.c (revision 0)
@@ -0,0 +1,33 @@
+/* { dg-do compile { target { powerpc*-*-* && lp64 } } } */
+/* { dg-require-effective-target powerpc_p9vector_ok } */
+/* { dg-options "-mpower9-vector -O2 -ffast-math" } */
+
+extern _Float128 copysignf128 (_Float128, _Float128);
+
+/* Check copysign optimizations that are done for double are also done for
+ _Float128. */
+
+_Float128
+x_times_cs_one_negx (_Float128 x)
+{
+ return x * copysignf128 (1.0Q, -x); /* XSNABSQP */
+}
+
+_Float128
+x_times_cs_one_x (_Float128 x)
+{
+ return x * copysignf128 (1.0Q, x); /* XSABSQP */
+}
+
+_Float128
+cs_x_x (_Float128 x)
+{
+ return copysignf128 (x, x); /* no operation. */
+}
+
+/* { dg-final { scan-assembler-times {\mxsabsqp\M} 1 } } */
+/* { dg-final { scan-assembler-times {\mxsnabsqp\M} 1 } } */
+/* { dg-final { scan-assembler-not {\mxscpsgnqp\M} } } */
+/* { dg-final { scan-assembler-not {\mlxvx\M} } } */
+/* { dg-final { scan-assembler-not {\mlxv\M} } } */
+/* { dg-final { scan-assembler-not {\mbl\M} } } */
Index: gcc/testsuite/gcc.target/powerpc/float128-hw6.c
===================================================================
--- gcc/testsuite/gcc.target/powerpc/float128-hw6.c (revision 0)
+++ gcc/testsuite/gcc.target/powerpc/float128-hw6.c (revision 0)
@@ -0,0 +1,26 @@
+/* { dg-do compile { target { powerpc*-*-* && lp64 } } } */
+/* { dg-require-effective-target powerpc_p9vector_ok } */
+/* { dg-options "-mpower9-vector -O2" } */
+
+extern _Float128 fabsf128 (_Float128);
+extern _Float128 copysignf128 (_Float128, _Float128);
+
+/* Check copysign optimizations that are done for double are also done for
+ _Float128. */
+
+_Float128
+cs_negx_y (_Float128 x, _Float128 y)
+{
+ return copysignf128 (-x, y); /* eliminate negation. */
+}
+
+_Float128
+cs_absx_y (_Float128 x, _Float128 y)
+{
+ return copysignf128 (fabsf128 (x), y); /* eliminate fabsf128. */
+}
+
+/* { dg-final { scan-assembler-times {\mxscpsgnqp\M} 2 } } */
+/* { dg-final { scan-assembler-not {\mxsnegqp\M} } } */
+/* { dg-final { scan-assembler-not {\mxsabsqp\M} } } */
+/* { dg-final { scan-assembler-not {\mbl\M} } } */
Index: gcc/testsuite/gcc.target/powerpc/float128-hw7.c
===================================================================
--- gcc/testsuite/gcc.target/powerpc/float128-hw7.c (revision 0)
+++ gcc/testsuite/gcc.target/powerpc/float128-hw7.c (revision 0)
@@ -0,0 +1,27 @@
+/* { dg-do compile { target { powerpc*-*-* && lp64 } } } */
+/* { dg-require-effective-target powerpc_p9vector_ok } */
+/* { dg-options "-mpower9-vector -O2" } */
+
+extern _Float128 fabsf128 (_Float128);
+extern _Float128 copysignf128 (_Float128, _Float128);
+
+/* Check copysign optimizations that are done for double are also done for
+ _Float128. */
+
+_Float128
+cs_x_pos1 (_Float128 x)
+{
+ return copysignf128 (x, 1.0Q); /* XSABSQP. */
+}
+
+_Float128 cs_x_neg2 (_Float128 x)
+{
+ return copysignf128 (x, -2.0Q); /* XSNABSQP. */
+}
+
+/* { dg-final { scan-assembler-times {\mxsabsqp\M} 1 } } */
+/* { dg-final { scan-assembler-not {\mxsnabsqp\M} 1 } } */
+/* { dg-final { scan-assembler-not {\mxscpsgnqp\M} } } */
+/* { dg-final { scan-assembler-not {\mlxvx\M} } } */
+/* { dg-final { scan-assembler-not {\mlxv\M} } } */
+/* { dg-final { scan-assembler-not {\mbl\M} } } */
Index: gcc/testsuite/gcc.target/powerpc/float128-hw8.c
===================================================================
--- gcc/testsuite/gcc.target/powerpc/float128-hw8.c (revision 0)
+++ gcc/testsuite/gcc.target/powerpc/float128-hw8.c (revision 0)
@@ -0,0 +1,24 @@
+/* { dg-do compile { target { powerpc*-*-* && lp64 } } } */
+/* { dg-require-effective-target powerpc_p9vector_ok } */
+/* { dg-options "-mpower9-vector -O2" } */
+
+extern _Float128 fminf128 (_Float128, _Float128);
+extern _Float128 fmaxf128 (_Float128, _Float128);
+
+/* Check min/max optimizations that are done for double are also done for
+ _Float128. */
+
+_Float128
+min_x_x (_Float128 x)
+{
+ return fminf128 (x, x);
+}
+
+_Float128
+max_x_x (_Float128 x)
+{
+ return fmaxf128 (x, x);
+}
+
+/* { dg-final { scan-assembler-not {\mxscmpuqp\M} } } */
+/* { dg-final { scan-assembler-not {\mbl\M} } } */
Index: gcc/testsuite/gcc.target/powerpc/float128-hw9.c
===================================================================
--- gcc/testsuite/gcc.target/powerpc/float128-hw9.c (revision 0)
+++ gcc/testsuite/gcc.target/powerpc/float128-hw9.c (revision 0)
@@ -0,0 +1,17 @@
+/* { dg-do compile { target { powerpc*-*-* && lp64 } } } */
+/* { dg-require-effective-target powerpc_p9vector_ok } */
+/* { dg-options "-mpower9-vector -O2 -ffast-math" } */
+
+extern _Float128 sqrtf128 (_Float128);
+
+/* Check sqrt optimizations that are done for double are also done for
+ _Float128. */
+
+_Float128
+sqrt_x_times_sqrt_x (_Float128 x)
+{
+ return sqrtf128 (x) * sqrtf128 (x);
+}
+
+/* { dg-final { scan-assembler-not {\mxssqrtqp\M} } } */
+/* { dg-final { scan-assembler-not {\mbl\M} } } */
Index: gcc/testsuite/gcc.target/powerpc/float128-hw10.c
===================================================================
--- gcc/testsuite/gcc.target/powerpc/float128-hw10.c (revision 0)
+++ gcc/testsuite/gcc.target/powerpc/float128-hw10.c (revision 0)
@@ -0,0 +1,38 @@
+/* { dg-do compile { target { powerpc*-*-* && lp64 } } } */
+/* { dg-require-effective-target powerpc_p9vector_ok } */
+/* { dg-options "-mpower9-vector -O2" } */
+
+extern _Float128 floorf128 (_Float128);
+extern _Float128 ceilf128 (_Float128);
+extern _Float128 roundf128 (_Float128);
+extern _Float128 truncf128 (_Float128);
+
+/* Check rounding optimizations that are done for double are also done for
+ _Float128. */
+
+_Float128
+floor_floor_x (_Float128 x)
+{
+ return floorf128 (floorf128 (x));
+}
+
+_Float128
+ceil_ceil_x (_Float128 x)
+{
+ return ceilf128 (ceilf128 (x));
+}
+
+_Float128
+trunc_trunc_x (_Float128 x)
+{
+ return truncf128 (truncf128 (x));
+}
+
+_Float128
+round_round_x (_Float128 x)
+{
+ return roundf128 (roundf128 (x));
+}
+
+/* { dg-final { scan-assembler-times {\mxsrqpi\M} 4 } } */
+/* { dg-final { scan-assembler-not {\mbl\M} } } */
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH] Add _Float<N>/_Float<N>X rounding built-ins & improve gimple optimization of _Float<N>/_Float<N>X built-in functions
2017-11-17 5:33 [PATCH] Add _Float<N>/_Float<N>X rounding built-ins & improve gimple optimization of _Float<N>/_Float<N>X built-in functions Michael Meissner
@ 2017-11-17 14:26 ` Segher Boessenkool
2017-11-17 21:41 ` Michael Meissner
2017-11-18 1:33 ` Michael Meissner
0 siblings, 2 replies; 12+ messages in thread
From: Segher Boessenkool @ 2017-11-17 14:26 UTC (permalink / raw)
To: Michael Meissner, GCC Patches, David Edelsohn, Bill Schmidt,
Joseph Myers, Richard Earnshaw, Richard Biener, Jeff Law,
Jason Merrill, David S. Miller, Bernd Schmidt, Ian Lance Taylor,
Jim Wilson
Hi!
On Fri, Nov 17, 2017 at 12:04:45AM -0500, Michael Meissner wrote:
> This patch is an enhancement of a previous page that never got approved.
> https://gcc.gnu.org/ml/gcc-patches/2017-10/threads.html#02124
>
> In the original patch, I added support to the machine independent
> infrastructure to support the rounding built-in functions for _Float<N> and
> _Float<N>X types (i.e. roundf128, ceilf128, etc.). I also added PowerPC ISA
> 3.0 support to generate these built-in functions.
>
> In addition to the previous changes, this patch now adds more optimizations for
> _Float<N> and _Float<N>X types in match.pd.
> +(define_insn "btrunc<mode>2"
> + [(set (match_operand:IEEE128 0 "altivec_register_operand" "=v")
> + (unspec:IEEE128
> + [(match_operand:IEEE128 1 "altivec_register_operand" "v")]
> + UNSPEC_FRIZ))]
> + "TARGET_FLOAT128_HW && FLOAT128_IEEE_P (<MODE>mode)"
> + "xsrqpi 1,%0,%1,0"
> + [(set_attr "type" "vecfloat")
> + (set_attr "size" "128")])
Is this one correct? Truncate is RMC=1, not RMC=0, I think?
The rest of the rs6000 part looks fine. Thanks!
Segher
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH] Add _Float<N>/_Float<N>X rounding built-ins & improve gimple optimization of _Float<N>/_Float<N>X built-in functions
2017-11-17 14:26 ` Segher Boessenkool
@ 2017-11-17 21:41 ` Michael Meissner
2017-11-18 1:33 ` Michael Meissner
1 sibling, 0 replies; 12+ messages in thread
From: Michael Meissner @ 2017-11-17 21:41 UTC (permalink / raw)
To: Segher Boessenkool
Cc: Michael Meissner, GCC Patches, David Edelsohn, Bill Schmidt,
Joseph Myers, Richard Earnshaw, Richard Biener, Jeff Law,
Jason Merrill, David S. Miller, Bernd Schmidt, Ian Lance Taylor,
Jim Wilson
On Fri, Nov 17, 2017 at 08:06:09AM -0600, Segher Boessenkool wrote:
> Hi!
>
> On Fri, Nov 17, 2017 at 12:04:45AM -0500, Michael Meissner wrote:
> > This patch is an enhancement of a previous page that never got approved.
> > https://gcc.gnu.org/ml/gcc-patches/2017-10/threads.html#02124
> >
> > In the original patch, I added support to the machine independent
> > infrastructure to support the rounding built-in functions for _Float<N> and
> > _Float<N>X types (i.e. roundf128, ceilf128, etc.). I also added PowerPC ISA
> > 3.0 support to generate these built-in functions.
> >
> > In addition to the previous changes, this patch now adds more optimizations for
> > _Float<N> and _Float<N>X types in match.pd.
>
> > +(define_insn "btrunc<mode>2"
> > + [(set (match_operand:IEEE128 0 "altivec_register_operand" "=v")
> > + (unspec:IEEE128
> > + [(match_operand:IEEE128 1 "altivec_register_operand" "v")]
> > + UNSPEC_FRIZ))]
> > + "TARGET_FLOAT128_HW && FLOAT128_IEEE_P (<MODE>mode)"
> > + "xsrqpi 1,%0,%1,0"
> > + [(set_attr "type" "vecfloat")
> > + (set_attr "size" "128")])
>
> Is this one correct? Truncate is RMC=1, not RMC=0, I think?
Good catch. I just tried it, and you were right.
> The rest of the rs6000 part looks fine. Thanks!
I will re-submit a patch once the bootstraps finish. Just to be sure, I'm
adding a new test that tests it at runtime.
--
Michael Meissner, IBM
IBM, M/S 2506R, 550 King Street, Littleton, MA 01460-6245, USA
email: meissner@linux.vnet.ibm.com, phone: +1 (978) 899-4797
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH] Add _Float<N>/_Float<N>X rounding built-ins & improve gimple optimization of _Float<N>/_Float<N>X built-in functions
2017-11-17 14:26 ` Segher Boessenkool
2017-11-17 21:41 ` Michael Meissner
@ 2017-11-18 1:33 ` Michael Meissner
2017-11-20 17:39 ` Segher Boessenkool
2017-12-21 18:16 ` Joseph Myers
1 sibling, 2 replies; 12+ messages in thread
From: Michael Meissner @ 2017-11-18 1:33 UTC (permalink / raw)
To: Segher Boessenkool
Cc: Michael Meissner, GCC Patches, David Edelsohn, Bill Schmidt,
Joseph Myers
[-- Attachment #1: Type: text/plain, Size: 4317 bytes --]
On Fri, Nov 17, 2017 at 08:06:09AM -0600, Segher Boessenkool wrote:
> Hi!
>
> On Fri, Nov 17, 2017 at 12:04:45AM -0500, Michael Meissner wrote:
> > This patch is an enhancement of a previous page that never got approved.
> > https://gcc.gnu.org/ml/gcc-patches/2017-10/threads.html#02124
> >
> > In the original patch, I added support to the machine independent
> > infrastructure to support the rounding built-in functions for _Float<N> and
> > _Float<N>X types (i.e. roundf128, ceilf128, etc.). I also added PowerPC ISA
> > 3.0 support to generate these built-in functions.
> >
> > In addition to the previous changes, this patch now adds more optimizations for
> > _Float<N> and _Float<N>X types in match.pd.
>
> > +(define_insn "btrunc<mode>2"
> > + [(set (match_operand:IEEE128 0 "altivec_register_operand" "=v")
> > + (unspec:IEEE128
> > + [(match_operand:IEEE128 1 "altivec_register_operand" "v")]
> > + UNSPEC_FRIZ))]
> > + "TARGET_FLOAT128_HW && FLOAT128_IEEE_P (<MODE>mode)"
> > + "xsrqpi 1,%0,%1,0"
> > + [(set_attr "type" "vecfloat")
> > + (set_attr "size" "128")])
>
> Is this one correct? Truncate is RMC=1, not RMC=0, I think?
>
> The rest of the rs6000 part looks fine. Thanks!
Here is the fixed patch. It fixes the btrunc<mode>2 insn to use the correct
XSRPQI variant for truncf128. I added the float128-hw11.c test as a runtime
test to make sure round, trunc, ceil, and floor return the correct values. The
machine independent portions are the same.
Assuming the machine independent versions are approved, can I check in the
PowerPC bits?
[gcc]
2017-11-17 Michael Meissner <meissner@linux.vnet.ibm.com>
* builtins.def: (_Float<N> and _Float<N>X BUILT_IN_CEIL): Add
_Float<N> and _Float<N>X variants for rounding built-in
functions.
(_Float<N> and _Float<N>X BUILT_IN_FLOOR): Likewise.
(_Float<N> and _Float<N>X BUILT_IN_NEARBYINT): Likewise.
(_Float<N> and _Float<N>X BUILT_IN_RINT): Likewise.
(_Float<N> and _Float<N>X BUILT_IN_ROUND): Likewise.
(_Float<N> and _Float<N>X BUILT_IN_TRUNC): Likewise.
* builtins.c (mathfn_built_in_2): Likewise.
* internal-fn.def (CEIL): Likewise.
(FLOOR): Likewise.
(NEARBYINT): Likewise.
(RINT): Likewise.
(ROUND): Likewise.
(TRUNC): Likewise.
* convert.c (convert_to_integer_1): Likewise.
* fold-const.c (tree_call_nonnegative_warnv_p): Likewise.
(integer_valued_real_call_p): Likewise.
* fold-const-call.c (fold_const_call_ss): Likewise.
* gencfn-macros.c (print_case_cfn): Change CFN and operator
printers to take a const char * suffix instead of a bool.
(print_define_operator_list): Likewise.
(fltall_suffixes): New list of suffixes, that include the
traditional suffixes as well as all of the _Float<N> and
_Float<N>X suffixes.
(main): For _Float<N> and _Float<N>X functions, emit both
<name>_FN and <name>_ALL variants. The <macro>_FN variant only
has the _Float<N> and _Float<N>X case names or operators. The
<name>_ALL variant has both the traditional and the
_Float<N>/_Float<N>X case names or operators.
* match.pd (COPYSIGN optimizations): Provide optimizations for
_Float<N> and _Float<N>X types where possible.
(MIN/MAX optimizations): Likewise.
(sqrt optimizations): Likewise.
(rounding optimizations): Likewise.
* config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
128-bit round to integer instructions.
(ceil<mode>2): Likewise.
(btrunc<mode>2): Likewise.
(round<mode>2): Likewise.
[gcc/c]
2017-11-17 Michael Meissner <meissner@linux.vnet.ibm.com>
* c-decl.c (header_for_builtin_fn): Add integer rounding _Float<N>
and _Float<N>X built-in functions.
[gcc/testsuite]
2017-11-17 Michael Meissner <meissner@linux.vnet.ibm.com>
* gcc.target/powerpc/float128-hw2.c: Add tests for ceilf128,
floorf128, truncf128, and roundf128.
* gcc.target/powerpc/float128-hw5.c: New tests for _Float128
optimizations added in match.pd.
* gcc.target/powerpc/float128-hw6.c: Likewise.
* gcc.target/powerpc/float128-hw7.c: Likewise.
* gcc.target/powerpc/float128-hw8.c: Likewise.
* gcc.target/powerpc/float128-hw9.c: Likewise.
* gcc.target/powerpc/float128-hw10.c: Likewise.
* gcc.target/powerpc/float128-hw11.c: Likewise.
--
Michael Meissner, IBM
IBM, M/S 2506R, 550 King Street, Littleton, MA 01460-6245, USA
email: meissner@linux.vnet.ibm.com, phone: +1 (978) 899-4797
[-- Attachment #2: ieee128-patch61b --]
[-- Type: text/plain, Size: 33985 bytes --]
Index: gcc/builtins.def
===================================================================
--- gcc/builtins.def (revision 254846)
+++ gcc/builtins.def (working copy)
@@ -343,6 +343,9 @@ DEF_C99_BUILTIN (BUILT_IN_CBRTL,
DEF_LIB_BUILTIN (BUILT_IN_CEIL, "ceil", BT_FN_DOUBLE_DOUBLE, ATTR_CONST_NOTHROW_LEAF_LIST)
DEF_C99_C90RES_BUILTIN (BUILT_IN_CEILF, "ceilf", BT_FN_FLOAT_FLOAT, ATTR_CONST_NOTHROW_LEAF_LIST)
DEF_C99_C90RES_BUILTIN (BUILT_IN_CEILL, "ceill", BT_FN_LONGDOUBLE_LONGDOUBLE, ATTR_CONST_NOTHROW_LEAF_LIST)
+#define CEIL_TYPE(F) BT_FN_##F##_##F
+DEF_EXT_LIB_FLOATN_NX_BUILTINS (BUILT_IN_CEIL, "ceil", CEIL_TYPE, ATTR_CONST_NOTHROW_LEAF_LIST)
+#undef CEIL_TYPE
DEF_C99_BUILTIN (BUILT_IN_COPYSIGN, "copysign", BT_FN_DOUBLE_DOUBLE_DOUBLE, ATTR_CONST_NOTHROW_LEAF_LIST)
DEF_C99_BUILTIN (BUILT_IN_COPYSIGNF, "copysignf", BT_FN_FLOAT_FLOAT_FLOAT, ATTR_CONST_NOTHROW_LEAF_LIST)
DEF_C99_BUILTIN (BUILT_IN_COPYSIGNL, "copysignl", BT_FN_LONGDOUBLE_LONGDOUBLE_LONGDOUBLE, ATTR_CONST_NOTHROW_LEAF_LIST)
@@ -402,6 +405,9 @@ DEF_C99_BUILTIN (BUILT_IN_FEUPDAT
DEF_LIB_BUILTIN (BUILT_IN_FLOOR, "floor", BT_FN_DOUBLE_DOUBLE, ATTR_CONST_NOTHROW_LEAF_LIST)
DEF_C99_C90RES_BUILTIN (BUILT_IN_FLOORF, "floorf", BT_FN_FLOAT_FLOAT, ATTR_CONST_NOTHROW_LEAF_LIST)
DEF_C99_C90RES_BUILTIN (BUILT_IN_FLOORL, "floorl", BT_FN_LONGDOUBLE_LONGDOUBLE, ATTR_CONST_NOTHROW_LEAF_LIST)
+#define FLOOR_TYPE(F) BT_FN_##F##_##F
+DEF_EXT_LIB_FLOATN_NX_BUILTINS (BUILT_IN_FLOOR, "floor", FLOOR_TYPE, ATTR_CONST_NOTHROW_LEAF_LIST)
+#undef FLOOR_TYPE
DEF_C99_BUILTIN (BUILT_IN_FMA, "fma", BT_FN_DOUBLE_DOUBLE_DOUBLE_DOUBLE, ATTR_MATHFN_FPROUNDING)
DEF_C99_BUILTIN (BUILT_IN_FMAF, "fmaf", BT_FN_FLOAT_FLOAT_FLOAT_FLOAT, ATTR_MATHFN_FPROUNDING)
DEF_C99_BUILTIN (BUILT_IN_FMAL, "fmal", BT_FN_LONGDOUBLE_LONGDOUBLE_LONGDOUBLE_LONGDOUBLE, ATTR_MATHFN_FPROUNDING)
@@ -539,6 +545,9 @@ DEF_GCC_FLOATN_NX_BUILTINS (BUILT_IN_NAN
DEF_C99_BUILTIN (BUILT_IN_NEARBYINT, "nearbyint", BT_FN_DOUBLE_DOUBLE, ATTR_CONST_NOTHROW_LEAF_LIST)
DEF_C99_BUILTIN (BUILT_IN_NEARBYINTF, "nearbyintf", BT_FN_FLOAT_FLOAT, ATTR_CONST_NOTHROW_LEAF_LIST)
DEF_C99_BUILTIN (BUILT_IN_NEARBYINTL, "nearbyintl", BT_FN_LONGDOUBLE_LONGDOUBLE, ATTR_CONST_NOTHROW_LEAF_LIST)
+#define NEARBYINT_TYPE(F) BT_FN_##F##_##F
+DEF_EXT_LIB_FLOATN_NX_BUILTINS (BUILT_IN_NEARBYINT, "nearbyint", NEARBYINT_TYPE, ATTR_CONST_NOTHROW_LEAF_LIST)
+#undef NEARBYINT_TYPE
DEF_C99_BUILTIN (BUILT_IN_NEXTAFTER, "nextafter", BT_FN_DOUBLE_DOUBLE_DOUBLE, ATTR_MATHFN_FPROUNDING_ERRNO)
DEF_C99_BUILTIN (BUILT_IN_NEXTAFTERF, "nextafterf", BT_FN_FLOAT_FLOAT_FLOAT, ATTR_MATHFN_FPROUNDING_ERRNO)
DEF_C99_BUILTIN (BUILT_IN_NEXTAFTERL, "nextafterl", BT_FN_LONGDOUBLE_LONGDOUBLE_LONGDOUBLE, ATTR_MATHFN_FPROUNDING_ERRNO)
@@ -563,9 +572,15 @@ DEF_C99_BUILTIN (BUILT_IN_REMQUOL
DEF_C99_BUILTIN (BUILT_IN_RINT, "rint", BT_FN_DOUBLE_DOUBLE, ATTR_MATHFN_FPROUNDING)
DEF_C99_BUILTIN (BUILT_IN_RINTF, "rintf", BT_FN_FLOAT_FLOAT, ATTR_MATHFN_FPROUNDING)
DEF_C99_BUILTIN (BUILT_IN_RINTL, "rintl", BT_FN_LONGDOUBLE_LONGDOUBLE, ATTR_MATHFN_FPROUNDING)
+#define RINT_TYPE(F) BT_FN_##F##_##F
+DEF_EXT_LIB_FLOATN_NX_BUILTINS (BUILT_IN_RINT, "rint", RINT_TYPE, ATTR_CONST_NOTHROW_LEAF_LIST)
+#undef RINT_TYPE
DEF_C99_BUILTIN (BUILT_IN_ROUND, "round", BT_FN_DOUBLE_DOUBLE, ATTR_CONST_NOTHROW_LEAF_LIST)
DEF_C99_BUILTIN (BUILT_IN_ROUNDF, "roundf", BT_FN_FLOAT_FLOAT, ATTR_CONST_NOTHROW_LEAF_LIST)
DEF_C99_BUILTIN (BUILT_IN_ROUNDL, "roundl", BT_FN_LONGDOUBLE_LONGDOUBLE, ATTR_CONST_NOTHROW_LEAF_LIST)
+#define ROUND_TYPE(F) BT_FN_##F##_##F
+DEF_EXT_LIB_FLOATN_NX_BUILTINS (BUILT_IN_ROUND, "round", ROUND_TYPE, ATTR_CONST_NOTHROW_LEAF_LIST)
+#undef ROUND_TYPE
DEF_EXT_LIB_BUILTIN (BUILT_IN_SCALB, "scalb", BT_FN_DOUBLE_DOUBLE_DOUBLE, ATTR_MATHFN_FPROUNDING_ERRNO)
DEF_EXT_LIB_BUILTIN (BUILT_IN_SCALBF, "scalbf", BT_FN_FLOAT_FLOAT_FLOAT, ATTR_MATHFN_FPROUNDING_ERRNO)
DEF_EXT_LIB_BUILTIN (BUILT_IN_SCALBL, "scalbl", BT_FN_LONGDOUBLE_LONGDOUBLE_LONGDOUBLE, ATTR_MATHFN_FPROUNDING_ERRNO)
@@ -611,6 +626,9 @@ DEF_C99_BUILTIN (BUILT_IN_TGAMMAL
DEF_C99_BUILTIN (BUILT_IN_TRUNC, "trunc", BT_FN_DOUBLE_DOUBLE, ATTR_CONST_NOTHROW_LEAF_LIST)
DEF_C99_BUILTIN (BUILT_IN_TRUNCF, "truncf", BT_FN_FLOAT_FLOAT, ATTR_CONST_NOTHROW_LEAF_LIST)
DEF_C99_BUILTIN (BUILT_IN_TRUNCL, "truncl", BT_FN_LONGDOUBLE_LONGDOUBLE, ATTR_CONST_NOTHROW_LEAF_LIST)
+#define TRUNC_TYPE(F) BT_FN_##F##_##F
+DEF_EXT_LIB_FLOATN_NX_BUILTINS (BUILT_IN_TRUNC, "trunc", TRUNC_TYPE, ATTR_CONST_NOTHROW_LEAF_LIST)
+#undef TRUNC_TYPE
DEF_EXT_LIB_BUILTIN (BUILT_IN_Y0, "y0", BT_FN_DOUBLE_DOUBLE, ATTR_MATHFN_FPROUNDING_ERRNO)
DEF_EXT_LIB_BUILTIN (BUILT_IN_Y0F, "y0f", BT_FN_FLOAT_FLOAT, ATTR_MATHFN_FPROUNDING_ERRNO)
DEF_EXT_LIB_BUILTIN (BUILT_IN_Y0L, "y0l", BT_FN_LONGDOUBLE_LONGDOUBLE, ATTR_MATHFN_FPROUNDING_ERRNO)
Index: gcc/builtins.c
===================================================================
--- gcc/builtins.c (revision 254846)
+++ gcc/builtins.c (working copy)
@@ -1872,7 +1872,7 @@ mathfn_built_in_2 (tree type, combined_f
CASE_MATHFN (ATAN2)
CASE_MATHFN (ATANH)
CASE_MATHFN (CBRT)
- CASE_MATHFN (CEIL)
+ CASE_MATHFN_FLOATN (CEIL)
CASE_MATHFN (CEXPI)
CASE_MATHFN_FLOATN (COPYSIGN)
CASE_MATHFN (COS)
@@ -1886,7 +1886,7 @@ mathfn_built_in_2 (tree type, combined_f
CASE_MATHFN (EXPM1)
CASE_MATHFN (FABS)
CASE_MATHFN (FDIM)
- CASE_MATHFN (FLOOR)
+ CASE_MATHFN_FLOATN (FLOOR)
CASE_MATHFN_FLOATN (FMA)
CASE_MATHFN_FLOATN (FMAX)
CASE_MATHFN_FLOATN (FMIN)
@@ -1925,7 +1925,7 @@ mathfn_built_in_2 (tree type, combined_f
CASE_MATHFN (MODF)
CASE_MATHFN (NAN)
CASE_MATHFN (NANS)
- CASE_MATHFN (NEARBYINT)
+ CASE_MATHFN_FLOATN (NEARBYINT)
CASE_MATHFN (NEXTAFTER)
CASE_MATHFN (NEXTTOWARD)
CASE_MATHFN (POW)
@@ -1933,8 +1933,8 @@ mathfn_built_in_2 (tree type, combined_f
CASE_MATHFN (POW10)
CASE_MATHFN (REMAINDER)
CASE_MATHFN (REMQUO)
- CASE_MATHFN (RINT)
- CASE_MATHFN (ROUND)
+ CASE_MATHFN_FLOATN (RINT)
+ CASE_MATHFN_FLOATN (ROUND)
CASE_MATHFN (SCALB)
CASE_MATHFN (SCALBLN)
CASE_MATHFN (SCALBN)
@@ -1947,7 +1947,7 @@ mathfn_built_in_2 (tree type, combined_f
CASE_MATHFN (TAN)
CASE_MATHFN (TANH)
CASE_MATHFN (TGAMMA)
- CASE_MATHFN (TRUNC)
+ CASE_MATHFN_FLOATN (TRUNC)
CASE_MATHFN (Y0)
CASE_MATHFN (Y1)
CASE_MATHFN (YN)
Index: gcc/internal-fn.def
===================================================================
--- gcc/internal-fn.def (revision 254846)
+++ gcc/internal-fn.def (working copy)
@@ -118,12 +118,12 @@ DEF_INTERNAL_FLT_FLOATN_FN (SQRT, ECF_CO
DEF_INTERNAL_FLT_FN (TAN, ECF_CONST, tan, unary)
/* FP rounding. */
-DEF_INTERNAL_FLT_FN (CEIL, ECF_CONST, ceil, unary)
-DEF_INTERNAL_FLT_FN (FLOOR, ECF_CONST, floor, unary)
-DEF_INTERNAL_FLT_FN (NEARBYINT, ECF_CONST, nearbyint, unary)
-DEF_INTERNAL_FLT_FN (RINT, ECF_CONST, rint, unary)
-DEF_INTERNAL_FLT_FN (ROUND, ECF_CONST, round, unary)
-DEF_INTERNAL_FLT_FN (TRUNC, ECF_CONST, btrunc, unary)
+DEF_INTERNAL_FLT_FLOATN_FN (CEIL, ECF_CONST, ceil, unary)
+DEF_INTERNAL_FLT_FLOATN_FN (FLOOR, ECF_CONST, floor, unary)
+DEF_INTERNAL_FLT_FLOATN_FN (NEARBYINT, ECF_CONST, nearbyint, unary)
+DEF_INTERNAL_FLT_FLOATN_FN (RINT, ECF_CONST, rint, unary)
+DEF_INTERNAL_FLT_FLOATN_FN (ROUND, ECF_CONST, round, unary)
+DEF_INTERNAL_FLT_FLOATN_FN (TRUNC, ECF_CONST, btrunc, unary)
/* Binary math functions. */
DEF_INTERNAL_FLT_FN (ATAN2, ECF_CONST, atan2, binary)
Index: gcc/convert.c
===================================================================
--- gcc/convert.c (revision 254846)
+++ gcc/convert.c (working copy)
@@ -554,6 +554,7 @@ convert_to_integer_1 (tree type, tree ex
switch (fcode)
{
CASE_FLT_FN (BUILT_IN_CEIL):
+ CASE_FLT_FN_FLOATN_NX (BUILT_IN_CEIL):
/* Only convert in ISO C99 mode. */
if (!targetm.libc_has_function (function_c99_misc))
break;
@@ -570,6 +571,7 @@ convert_to_integer_1 (tree type, tree ex
break;
CASE_FLT_FN (BUILT_IN_FLOOR):
+ CASE_FLT_FN_FLOATN_NX (BUILT_IN_FLOOR):
/* Only convert in ISO C99 mode. */
if (!targetm.libc_has_function (function_c99_misc))
break;
@@ -586,6 +588,7 @@ convert_to_integer_1 (tree type, tree ex
break;
CASE_FLT_FN (BUILT_IN_ROUND):
+ CASE_FLT_FN_FLOATN_NX (BUILT_IN_ROUND):
/* Only convert in ISO C99 mode and with -fno-math-errno. */
if (!targetm.libc_has_function (function_c99_misc) || flag_errno_math)
break;
@@ -602,11 +605,13 @@ convert_to_integer_1 (tree type, tree ex
break;
CASE_FLT_FN (BUILT_IN_NEARBYINT):
+ CASE_FLT_FN_FLOATN_NX (BUILT_IN_NEARBYINT):
/* Only convert nearbyint* if we can ignore math exceptions. */
if (flag_trapping_math)
break;
gcc_fallthrough ();
CASE_FLT_FN (BUILT_IN_RINT):
+ CASE_FLT_FN_FLOATN_NX (BUILT_IN_RINT):
/* Only convert in ISO C99 mode and with -fno-math-errno. */
if (!targetm.libc_has_function (function_c99_misc) || flag_errno_math)
break;
@@ -623,6 +628,7 @@ convert_to_integer_1 (tree type, tree ex
break;
CASE_FLT_FN (BUILT_IN_TRUNC):
+ CASE_FLT_FN_FLOATN_NX (BUILT_IN_TRUNC):
return convert_to_integer_1 (type, CALL_EXPR_ARG (s_expr, 0), dofold);
default:
Index: gcc/fold-const.c
===================================================================
--- gcc/fold-const.c (revision 254846)
+++ gcc/fold-const.c (working copy)
@@ -12784,9 +12784,11 @@ tree_call_nonnegative_warnv_p (tree type
CASE_CFN_ATANH:
CASE_CFN_CBRT:
CASE_CFN_CEIL:
+ CASE_CFN_CEIL_FN:
CASE_CFN_ERF:
CASE_CFN_EXPM1:
CASE_CFN_FLOOR:
+ CASE_CFN_FLOOR_FN:
CASE_CFN_FMOD:
CASE_CFN_FREXP:
CASE_CFN_ICEIL:
@@ -12804,8 +12806,11 @@ tree_call_nonnegative_warnv_p (tree type
CASE_CFN_LROUND:
CASE_CFN_MODF:
CASE_CFN_NEARBYINT:
+ CASE_CFN_NEARBYINT_FN:
CASE_CFN_RINT:
+ CASE_CFN_RINT_FN:
CASE_CFN_ROUND:
+ CASE_CFN_ROUND_FN:
CASE_CFN_SCALB:
CASE_CFN_SCALBLN:
CASE_CFN_SCALBN:
@@ -12814,6 +12819,7 @@ tree_call_nonnegative_warnv_p (tree type
CASE_CFN_SINH:
CASE_CFN_TANH:
CASE_CFN_TRUNC:
+ CASE_CFN_TRUNC_FN:
/* True if the 1st argument is nonnegative. */
return RECURSE (arg0);
@@ -13319,11 +13325,17 @@ integer_valued_real_call_p (combined_fn
switch (fn)
{
CASE_CFN_CEIL:
+ CASE_CFN_CEIL_FN:
CASE_CFN_FLOOR:
+ CASE_CFN_FLOOR_FN:
CASE_CFN_NEARBYINT:
+ CASE_CFN_NEARBYINT_FN:
CASE_CFN_RINT:
+ CASE_CFN_RINT_FN:
CASE_CFN_ROUND:
+ CASE_CFN_ROUND_FN:
CASE_CFN_TRUNC:
+ CASE_CFN_TRUNC_FN:
return true;
CASE_CFN_FMIN:
Index: gcc/fold-const-call.c
===================================================================
--- gcc/fold-const-call.c (revision 254846)
+++ gcc/fold-const-call.c (working copy)
@@ -699,6 +699,7 @@ fold_const_call_ss (real_value *result,
&& do_mpfr_arg1 (result, mpfr_y1, arg, format));
CASE_CFN_FLOOR:
+ CASE_CFN_FLOOR_FN:
if (!REAL_VALUE_ISNAN (*arg) || !flag_errno_math)
{
real_floor (result, format, arg);
@@ -707,6 +708,7 @@ fold_const_call_ss (real_value *result,
return false;
CASE_CFN_CEIL:
+ CASE_CFN_CEIL_FN:
if (!REAL_VALUE_ISNAN (*arg) || !flag_errno_math)
{
real_ceil (result, format, arg);
@@ -715,10 +717,12 @@ fold_const_call_ss (real_value *result,
return false;
CASE_CFN_TRUNC:
+ CASE_CFN_TRUNC_FN:
real_trunc (result, format, arg);
return true;
CASE_CFN_ROUND:
+ CASE_CFN_ROUND_FN:
if (!REAL_VALUE_ISNAN (*arg) || !flag_errno_math)
{
real_round (result, format, arg);
Index: gcc/gencfn-macros.c
===================================================================
--- gcc/gencfn-macros.c (revision 254846)
+++ gcc/gencfn-macros.c (working copy)
@@ -94,13 +94,15 @@ is_group (string_set *builtins, const ch
/* Print a macro for all combined functions related to NAME, with the
null-terminated list of suffixes in SUFFIXES. INTERNAL_P says whether
- CFN_<NAME> also exists. */
+ CFN_<NAME> also exists. FLOATN_P is a suffix to the operator name, blank
+ for normal operators, "_FN" for _Float<N>/_Float<N>X operators only, and
+ "_ALL" for both the traditional operators and the _Float<N>/_Float<N>X
+ operators. */
static void
print_case_cfn (const char *name, bool internal_p,
- const char *const *suffixes, bool floatn_p)
+ const char *const *suffixes, const char *floatn)
{
- const char *floatn = (floatn_p) ? "_FN" : "";
printf ("#define CASE_CFN_%s%s", name, floatn);
if (internal_p)
printf (" \\\n case CFN_%s%s", name, floatn);
@@ -110,15 +112,18 @@ print_case_cfn (const char *name, bool i
printf ("\n");
}
-/* Print an operator list for all combined functions related to NAME,
- with the null-terminated list of suffixes in SUFFIXES. INTERNAL_P
- says whether CFN_<NAME> also exists. */
+/* Print an operator list for all combined functions related to NAME, with the
+ null-terminated list of suffixes in SUFFIXES. INTERNAL_P says whether
+ CFN_<NAME> also exists. FLOATN_P is a suffix to the operator name, blank
+ for normal operators, "_FN" for _Float<N>/_Float<N>X operators only, and
+ "_ALL" for both the traditional operators and the _Float<N>/_Float<N>X
+ operators. */
static void
print_define_operator_list (const char *name, bool internal_p,
- const char *const *suffixes, bool floatn_p)
+ const char *const *suffixes,
+ const char *floatn)
{
- const char *floatn = (floatn_p) ? "_FN" : "";
printf ("(define_operator_list %s%s\n", name, floatn);
for (unsigned int i = 0; suffixes[i]; ++i)
printf (" BUILT_IN_%s%s\n", name, suffixes[i]);
@@ -152,6 +157,9 @@ const char *const internal_fn_int_names[
static const char *const flt_suffixes[] = { "F", "", "L", NULL };
static const char *const fltfn_suffixes[] = { "F16", "F32", "F64", "F128",
"F32X", "F64X", "F128X", NULL };
+static const char *const fltall_suffixes[] = { "F", "", "L", "F16", "F32",
+ "F64", "F128", "F32X", "F64X",
+ "F128X", NULL };
static const char *const int_suffixes[] = { "", "L", "LL", "IMAX", NULL };
static const char *const *const suffix_lists[] = {
@@ -212,22 +220,31 @@ main (int argc, char **argv)
bool internal_p = internal_fns.contains (root);
if (type == 'c')
- print_case_cfn (root, internal_p, suffix, false);
+ print_case_cfn (root, internal_p, suffix, "");
else
- print_define_operator_list (root, internal_p,
- suffix, false);
+ print_define_operator_list (root, internal_p, suffix, "");
/* Support the _Float<N> and _Float<N>X math functions if
- they exist. We put these out as a separate CFN macro,
- so code can add support or not as needed. */
+ they exist. We put these out as a separate CFN or
+ operator macro, so code can add support or not as
+ needed. We also put out a combined CFN or operator
+ macro that includes both the traditional names and the
+ _Float<N> and _Float<N>X versions. */
if (suffix == flt_suffixes
&& is_group (&builtins, root, fltfn_suffixes))
{
if (type == 'c')
- print_case_cfn (root, false, fltfn_suffixes, true);
+ {
+ print_case_cfn (root, false, fltfn_suffixes, "_FN");
+ print_case_cfn (root, false, fltall_suffixes, "_ALL");
+ }
else
- print_define_operator_list (root, false, fltfn_suffixes,
- true);
+ {
+ print_define_operator_list (root, false,
+ fltfn_suffixes, "_FN");
+ print_define_operator_list (root, internal_p,
+ fltall_suffixes, "_ALL");
+ }
}
}
}
Index: gcc/match.pd
===================================================================
--- gcc/match.pd (revision 254846)
+++ gcc/match.pd (working copy)
@@ -191,21 +191,21 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
/* Transform X * copysign (1.0, X) into abs(X). */
(simplify
- (mult:c @0 (COPYSIGN real_onep @0))
+ (mult:c @0 (COPYSIGN_ALL real_onep @0))
(if (!HONOR_NANS (type) && !HONOR_SIGNED_ZEROS (type))
(abs @0)))
/* Transform X * copysign (1.0, -X) into -abs(X). */
(simplify
- (mult:c @0 (COPYSIGN real_onep (negate @0)))
+ (mult:c @0 (COPYSIGN_ALL real_onep (negate @0)))
(if (!HONOR_NANS (type) && !HONOR_SIGNED_ZEROS (type))
(negate (abs @0))))
/* Transform copysign (CST, X) into copysign (ABS(CST), X). */
(simplify
- (COPYSIGN REAL_CST@0 @1)
+ (COPYSIGN_ALL REAL_CST@0 @1)
(if (REAL_VALUE_NEGATIVE (TREE_REAL_CST (@0)))
- (COPYSIGN (negate @0) @1)))
+ (COPYSIGN_ALL (negate @0) @1)))
/* X * 1, X / 1 -> X. */
(for op (mult trunc_div ceil_div floor_div round_div exact_div)
@@ -534,7 +534,7 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
(hypots @0 (op @1))
(hypots @0 @1)))
/* copysign(-x, y) and copysign(abs(x), y) -> copysign(x, y). */
- (for copysigns (COPYSIGN)
+ (for copysigns (COPYSIGN_ALL)
(simplify
(copysigns (op @0) @1)
(copysigns @0 @1))))
@@ -579,7 +579,7 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
(hypots @0 @1)))
/* copysign(x, CST) -> [-]abs (x). */
-(for copysigns (COPYSIGN)
+(for copysigns (COPYSIGN_ALL)
(simplify
(copysigns @0 REAL_CST@1)
(if (REAL_VALUE_NEGATIVE (TREE_REAL_CST (@1)))
@@ -587,13 +587,13 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
(abs @0))))
/* copysign(copysign(x, y), z) -> copysign(x, z). */
-(for copysigns (COPYSIGN)
+(for copysigns (COPYSIGN_ALL)
(simplify
(copysigns (copysigns @0 @1) @2)
(copysigns @0 @2)))
/* copysign(x,y)*copysign(x,y) -> x*x. */
-(for copysigns (COPYSIGN)
+(for copysigns (COPYSIGN_ALL)
(simplify
(mult (copysigns@2 @0 @1) @2)
(mult @0 @0)))
@@ -1802,7 +1802,7 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
/* Simplifications of MIN_EXPR, MAX_EXPR, fmin() and fmax(). */
-(for minmax (min max FMIN FMIN_FN FMAX FMAX_FN)
+(for minmax (min max FMIN_ALL FMAX_ALL)
(simplify
(minmax @0 @0)
@0))
@@ -1880,7 +1880,7 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
&& TYPE_PRECISION (TREE_TYPE (@0)) > TYPE_PRECISION (type))
(minmax @1 (convert @2)))))
-(for minmax (FMIN FMIN_FN FMAX FMAX_FN)
+(for minmax (FMIN_ALL FMAX_ALL)
/* If either argument is NaN, return the other one. Avoid the
transformation if we get (and honor) a signalling NaN. */
(simplify
@@ -1895,20 +1895,14 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
worry about it either. */
(if (flag_finite_math_only)
(simplify
- (FMIN @0 @1)
+ (FMIN_ALL @0 @1)
(min @0 @1))
(simplify
- (FMIN_FN @0 @1)
- (min @0 @1))
- (simplify
- (FMAX @0 @1)
- (max @0 @1))
- (simplify
- (FMAX_FN @0 @1)
+ (FMAX_ALL @0 @1)
(max @0 @1)))
/* min (-A, -B) -> -max (A, B) */
-(for minmax (min max FMIN FMIN_FN FMAX FMAX_FN)
- maxmin (max min FMAX FMAX_FN FMIN FMAX_FN)
+(for minmax (min max FMIN_ALL FMAX_ALL)
+ maxmin (max min FMAX_ALL FMIN_ALL)
(simplify
(minmax (negate:s@2 @0) (negate:s@3 @1))
(if (FLOAT_TYPE_P (TREE_TYPE (@0))
@@ -3697,7 +3691,7 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
(if (flag_unsafe_math_optimizations)
/* Simplify sqrt(x) * sqrt(x) -> x. */
(simplify
- (mult (SQRT@1 @0) @1)
+ (mult (SQRT_ALL@1 @0) @1)
(if (!HONOR_SNANS (type))
@0))
@@ -3850,12 +3844,12 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
(abs @0))
/* trunc(trunc(x)) -> trunc(x), etc. */
-(for fns (TRUNC FLOOR CEIL ROUND NEARBYINT RINT)
+(for fns (TRUNC_ALL FLOOR_ALL CEIL_ALL ROUND_ALL NEARBYINT_ALL RINT_ALL)
(simplify
(fns (fns @0))
(fns @0)))
/* f(x) -> x if x is integer valued and f does nothing for such values. */
-(for fns (TRUNC FLOOR CEIL ROUND NEARBYINT RINT)
+(for fns (TRUNC_ALL FLOOR_ALL CEIL_ALL ROUND_ALL NEARBYINT_ALL RINT_ALL)
(simplify
(fns integer_valued_real_p@0)
@0))
@@ -3872,12 +3866,12 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
(simplify
/* copysign(x,x) -> x. */
- (COPYSIGN @0 @0)
+ (COPYSIGN_ALL @0 @0)
@0)
(simplify
/* copysign(x,y) -> fabs(x) if y is nonnegative. */
- (COPYSIGN @0 tree_expr_nonnegative_p@1)
+ (COPYSIGN_ALL @0 tree_expr_nonnegative_p@1)
(abs @0))
(for scale (LDEXP SCALBN SCALBLN)
@@ -4028,8 +4022,8 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
(if (canonicalize_math_p ())
/* floor(x) -> trunc(x) if x is nonnegative. */
- (for floors (FLOOR)
- truncs (TRUNC)
+ (for floors (FLOOR_ALL)
+ truncs (TRUNC_ALL)
(simplify
(floors tree_expr_nonnegative_p@0)
(truncs @0))))
Index: gcc/config/rs6000/rs6000.md
===================================================================
--- gcc/config/rs6000/rs6000.md (revision 254846)
+++ gcc/config/rs6000/rs6000.md (working copy)
@@ -14733,6 +14733,47 @@ (define_insn_and_split "floatuns<QHI:mod
(set_attr "type" "vecfloat")
(set_attr "size" "128")])
+;; IEEE 128-bit round to integer built-in functions
+(define_insn "floor<mode>2"
+ [(set (match_operand:IEEE128 0 "altivec_register_operand" "=v")
+ (unspec:IEEE128
+ [(match_operand:IEEE128 1 "altivec_register_operand" "v")]
+ UNSPEC_FRIM))]
+ "TARGET_FLOAT128_HW && FLOAT128_IEEE_P (<MODE>mode)"
+ "xsrqpi 1,%0,%1,3"
+ [(set_attr "type" "vecfloat")
+ (set_attr "size" "128")])
+
+(define_insn "ceil<mode>2"
+ [(set (match_operand:IEEE128 0 "altivec_register_operand" "=v")
+ (unspec:IEEE128
+ [(match_operand:IEEE128 1 "altivec_register_operand" "v")]
+ UNSPEC_FRIP))]
+ "TARGET_FLOAT128_HW && FLOAT128_IEEE_P (<MODE>mode)"
+ "xsrqpi 1,%0,%1,2"
+ [(set_attr "type" "vecfloat")
+ (set_attr "size" "128")])
+
+(define_insn "btrunc<mode>2"
+ [(set (match_operand:IEEE128 0 "altivec_register_operand" "=v")
+ (unspec:IEEE128
+ [(match_operand:IEEE128 1 "altivec_register_operand" "v")]
+ UNSPEC_FRIZ))]
+ "TARGET_FLOAT128_HW && FLOAT128_IEEE_P (<MODE>mode)"
+ "xsrqpi 1,%0,%1,1"
+ [(set_attr "type" "vecfloat")
+ (set_attr "size" "128")])
+
+(define_insn "round<mode>2"
+ [(set (match_operand:IEEE128 0 "altivec_register_operand" "=v")
+ (unspec:IEEE128
+ [(match_operand:IEEE128 1 "altivec_register_operand" "v")]
+ UNSPEC_FRIN))]
+ "TARGET_FLOAT128_HW && FLOAT128_IEEE_P (<MODE>mode)"
+ "xsrqpi 0,%0,%1,3"
+ [(set_attr "type" "vecfloat")
+ (set_attr "size" "128")])
+
;; IEEE 128-bit instructions with round to odd semantics
(define_insn "add<mode>3_odd"
[(set (match_operand:IEEE128 0 "altivec_register_operand" "=v")
Index: gcc/c/c-decl.c
===================================================================
--- gcc/c/c-decl.c (revision 254846)
+++ gcc/c/c-decl.c (working copy)
@@ -3162,6 +3162,7 @@ header_for_builtin_fn (enum built_in_fun
CASE_FLT_FN (BUILT_IN_ATAN2):
CASE_FLT_FN (BUILT_IN_CBRT):
CASE_FLT_FN (BUILT_IN_CEIL):
+ CASE_FLT_FN_FLOATN_NX (BUILT_IN_CEIL):
CASE_FLT_FN (BUILT_IN_COPYSIGN):
CASE_FLT_FN_FLOATN_NX (BUILT_IN_COPYSIGN):
CASE_FLT_FN (BUILT_IN_COS):
@@ -3175,6 +3176,7 @@ header_for_builtin_fn (enum built_in_fun
CASE_FLT_FN_FLOATN_NX (BUILT_IN_FABS):
CASE_FLT_FN (BUILT_IN_FDIM):
CASE_FLT_FN (BUILT_IN_FLOOR):
+ CASE_FLT_FN_FLOATN_NX (BUILT_IN_FLOOR):
CASE_FLT_FN (BUILT_IN_FMA):
CASE_FLT_FN_FLOATN_NX (BUILT_IN_FMA):
CASE_FLT_FN (BUILT_IN_FMAX):
@@ -3199,13 +3201,16 @@ header_for_builtin_fn (enum built_in_fun
CASE_FLT_FN (BUILT_IN_MODF):
CASE_FLT_FN (BUILT_IN_NAN):
CASE_FLT_FN (BUILT_IN_NEARBYINT):
+ CASE_FLT_FN_FLOATN_NX (BUILT_IN_NEARBYINT):
CASE_FLT_FN (BUILT_IN_NEXTAFTER):
CASE_FLT_FN (BUILT_IN_NEXTTOWARD):
CASE_FLT_FN (BUILT_IN_POW):
CASE_FLT_FN (BUILT_IN_REMAINDER):
CASE_FLT_FN (BUILT_IN_REMQUO):
CASE_FLT_FN (BUILT_IN_RINT):
+ CASE_FLT_FN_FLOATN_NX (BUILT_IN_RINT):
CASE_FLT_FN (BUILT_IN_ROUND):
+ CASE_FLT_FN_FLOATN_NX (BUILT_IN_ROUND):
CASE_FLT_FN (BUILT_IN_SCALBLN):
CASE_FLT_FN (BUILT_IN_SCALBN):
CASE_FLT_FN (BUILT_IN_SIN):
@@ -3217,6 +3222,7 @@ header_for_builtin_fn (enum built_in_fun
CASE_FLT_FN (BUILT_IN_TANH):
CASE_FLT_FN (BUILT_IN_TGAMMA):
CASE_FLT_FN (BUILT_IN_TRUNC):
+ CASE_FLT_FN_FLOATN_NX (BUILT_IN_TRUNC):
case BUILT_IN_ISINF:
case BUILT_IN_ISNAN:
return "<math.h>";
Index: gcc/testsuite/gcc.target/powerpc/float128-hw2.c
===================================================================
--- gcc/testsuite/gcc.target/powerpc/float128-hw2.c (revision 254846)
+++ gcc/testsuite/gcc.target/powerpc/float128-hw2.c (working copy)
@@ -14,6 +14,10 @@
extern _Float128 copysignf128 (_Float128, _Float128);
extern _Float128 sqrtf128 (_Float128);
extern _Float128 fmaf128 (_Float128, _Float128, _Float128);
+extern _Float128 ceilf128 (_Float128);
+extern _Float128 floorf128 (_Float128);
+extern _Float128 truncf128 (_Float128);
+extern _Float128 roundf128 (_Float128);
_Float128
do_copysign (_Float128 a, _Float128 b)
@@ -51,10 +55,35 @@ do_nfms (_Float128 a, _Float128 b, _Floa
return -fmaf128 (a, b, -c);
}
+_Float128
+do_ceil (_Float128 a)
+{
+ return ceilf128 (a);
+}
+
+_Float128
+do_floor (_Float128 a)
+{
+ return floorf128 (a);
+}
+
+_Float128
+do_trunc (_Float128 a)
+{
+ return truncf128 (a);
+}
+
+_Float128
+do_round (_Float128 a)
+{
+ return roundf128 (a);
+}
+
/* { dg-final { scan-assembler {\mxscpsgnqp\M} } } */
/* { dg-final { scan-assembler {\mxssqrtqp\M} } } */
/* { dg-final { scan-assembler {\mxsmaddqp\M} } } */
/* { dg-final { scan-assembler {\mxsmsubqp\M} } } */
/* { dg-final { scan-assembler {\mxsnmaddqp\M} } } */
/* { dg-final { scan-assembler {\mxsnmsubqp\M} } } */
+/* { dg-final { scan-assembler {\mxsrqpi\M} } } */
/* { dg-final { scan-assembler-not {\mbl\M} } } */
Index: gcc/testsuite/gcc.target/powerpc/float128-hw5.c
===================================================================
--- gcc/testsuite/gcc.target/powerpc/float128-hw5.c (revision 0)
+++ gcc/testsuite/gcc.target/powerpc/float128-hw5.c (revision 0)
@@ -0,0 +1,33 @@
+/* { dg-do compile { target { powerpc*-*-* && lp64 } } } */
+/* { dg-require-effective-target powerpc_p9vector_ok } */
+/* { dg-options "-mpower9-vector -O2 -ffast-math" } */
+
+extern _Float128 copysignf128 (_Float128, _Float128);
+
+/* Check copysign optimizations that are done for double are also done for
+ _Float128. */
+
+_Float128
+x_times_cs_one_negx (_Float128 x)
+{
+ return x * copysignf128 (1.0Q, -x); /* XSNABSQP */
+}
+
+_Float128
+x_times_cs_one_x (_Float128 x)
+{
+ return x * copysignf128 (1.0Q, x); /* XSABSQP */
+}
+
+_Float128
+cs_x_x (_Float128 x)
+{
+ return copysignf128 (x, x); /* no operation. */
+}
+
+/* { dg-final { scan-assembler-times {\mxsabsqp\M} 1 } } */
+/* { dg-final { scan-assembler-times {\mxsnabsqp\M} 1 } } */
+/* { dg-final { scan-assembler-not {\mxscpsgnqp\M} } } */
+/* { dg-final { scan-assembler-not {\mlxvx\M} } } */
+/* { dg-final { scan-assembler-not {\mlxv\M} } } */
+/* { dg-final { scan-assembler-not {\mbl\M} } } */
Index: gcc/testsuite/gcc.target/powerpc/float128-hw6.c
===================================================================
--- gcc/testsuite/gcc.target/powerpc/float128-hw6.c (revision 0)
+++ gcc/testsuite/gcc.target/powerpc/float128-hw6.c (revision 0)
@@ -0,0 +1,26 @@
+/* { dg-do compile { target { powerpc*-*-* && lp64 } } } */
+/* { dg-require-effective-target powerpc_p9vector_ok } */
+/* { dg-options "-mpower9-vector -O2" } */
+
+extern _Float128 fabsf128 (_Float128);
+extern _Float128 copysignf128 (_Float128, _Float128);
+
+/* Check copysign optimizations that are done for double are also done for
+ _Float128. */
+
+_Float128
+cs_negx_y (_Float128 x, _Float128 y)
+{
+ return copysignf128 (-x, y); /* eliminate negation. */
+}
+
+_Float128
+cs_absx_y (_Float128 x, _Float128 y)
+{
+ return copysignf128 (fabsf128 (x), y); /* eliminate fabsf128. */
+}
+
+/* { dg-final { scan-assembler-times {\mxscpsgnqp\M} 2 } } */
+/* { dg-final { scan-assembler-not {\mxsnegqp\M} } } */
+/* { dg-final { scan-assembler-not {\mxsabsqp\M} } } */
+/* { dg-final { scan-assembler-not {\mbl\M} } } */
Index: gcc/testsuite/gcc.target/powerpc/float128-hw7.c
===================================================================
--- gcc/testsuite/gcc.target/powerpc/float128-hw7.c (revision 0)
+++ gcc/testsuite/gcc.target/powerpc/float128-hw7.c (revision 0)
@@ -0,0 +1,27 @@
+/* { dg-do compile { target { powerpc*-*-* && lp64 } } } */
+/* { dg-require-effective-target powerpc_p9vector_ok } */
+/* { dg-options "-mpower9-vector -O2" } */
+
+extern _Float128 fabsf128 (_Float128);
+extern _Float128 copysignf128 (_Float128, _Float128);
+
+/* Check copysign optimizations that are done for double are also done for
+ _Float128. */
+
+_Float128
+cs_x_pos1 (_Float128 x)
+{
+ return copysignf128 (x, 1.0Q); /* XSABSQP. */
+}
+
+_Float128 cs_x_neg2 (_Float128 x)
+{
+ return copysignf128 (x, -2.0Q); /* XSNABSQP. */
+}
+
+/* { dg-final { scan-assembler-times {\mxsabsqp\M} 1 } } */
+/* { dg-final { scan-assembler-not {\mxsnabsqp\M} 1 } } */
+/* { dg-final { scan-assembler-not {\mxscpsgnqp\M} } } */
+/* { dg-final { scan-assembler-not {\mlxvx\M} } } */
+/* { dg-final { scan-assembler-not {\mlxv\M} } } */
+/* { dg-final { scan-assembler-not {\mbl\M} } } */
Index: gcc/testsuite/gcc.target/powerpc/float128-hw8.c
===================================================================
--- gcc/testsuite/gcc.target/powerpc/float128-hw8.c (revision 0)
+++ gcc/testsuite/gcc.target/powerpc/float128-hw8.c (revision 0)
@@ -0,0 +1,24 @@
+/* { dg-do compile { target { powerpc*-*-* && lp64 } } } */
+/* { dg-require-effective-target powerpc_p9vector_ok } */
+/* { dg-options "-mpower9-vector -O2" } */
+
+extern _Float128 fminf128 (_Float128, _Float128);
+extern _Float128 fmaxf128 (_Float128, _Float128);
+
+/* Check min/max optimizations that are done for double are also done for
+ _Float128. */
+
+_Float128
+min_x_x (_Float128 x)
+{
+ return fminf128 (x, x);
+}
+
+_Float128
+max_x_x (_Float128 x)
+{
+ return fmaxf128 (x, x);
+}
+
+/* { dg-final { scan-assembler-not {\mxscmpuqp\M} } } */
+/* { dg-final { scan-assembler-not {\mbl\M} } } */
Index: gcc/testsuite/gcc.target/powerpc/float128-hw9.c
===================================================================
--- gcc/testsuite/gcc.target/powerpc/float128-hw9.c (revision 0)
+++ gcc/testsuite/gcc.target/powerpc/float128-hw9.c (revision 0)
@@ -0,0 +1,17 @@
+/* { dg-do compile { target { powerpc*-*-* && lp64 } } } */
+/* { dg-require-effective-target powerpc_p9vector_ok } */
+/* { dg-options "-mpower9-vector -O2 -ffast-math" } */
+
+extern _Float128 sqrtf128 (_Float128);
+
+/* Check sqrt optimizations that are done for double are also done for
+ _Float128. */
+
+_Float128
+sqrt_x_times_sqrt_x (_Float128 x)
+{
+ return sqrtf128 (x) * sqrtf128 (x);
+}
+
+/* { dg-final { scan-assembler-not {\mxssqrtqp\M} } } */
+/* { dg-final { scan-assembler-not {\mbl\M} } } */
Index: gcc/testsuite/gcc.target/powerpc/float128-hw10.c
===================================================================
--- gcc/testsuite/gcc.target/powerpc/float128-hw10.c (revision 0)
+++ gcc/testsuite/gcc.target/powerpc/float128-hw10.c (revision 0)
@@ -0,0 +1,38 @@
+/* { dg-do compile { target { powerpc*-*-* && lp64 } } } */
+/* { dg-require-effective-target powerpc_p9vector_ok } */
+/* { dg-options "-mpower9-vector -O2" } */
+
+extern _Float128 floorf128 (_Float128);
+extern _Float128 ceilf128 (_Float128);
+extern _Float128 roundf128 (_Float128);
+extern _Float128 truncf128 (_Float128);
+
+/* Check rounding optimizations that are done for double are also done for
+ _Float128. */
+
+_Float128
+floor_floor_x (_Float128 x)
+{
+ return floorf128 (floorf128 (x));
+}
+
+_Float128
+ceil_ceil_x (_Float128 x)
+{
+ return ceilf128 (ceilf128 (x));
+}
+
+_Float128
+trunc_trunc_x (_Float128 x)
+{
+ return truncf128 (truncf128 (x));
+}
+
+_Float128
+round_round_x (_Float128 x)
+{
+ return roundf128 (roundf128 (x));
+}
+
+/* { dg-final { scan-assembler-times {\mxsrqpi\M} 4 } } */
+/* { dg-final { scan-assembler-not {\mbl\M} } } */
Index: gcc/testsuite/gcc.target/powerpc/float128-hw11.c
===================================================================
--- gcc/testsuite/gcc.target/powerpc/float128-hw11.c (revision 0)
+++ gcc/testsuite/gcc.target/powerpc/float128-hw11.c (revision 0)
@@ -0,0 +1,59 @@
+/* { dg-do run { target { powerpc*-*-* && lp64 } } } */
+/* { dg-require-effective-target powerpc_p9vector_ok } */
+/* { dg-options "-mpower9-vector -O2" } */
+/* { dg-options "-mvsx -O2" } */
+
+#define __STDC_WANT_IEC_60559_TYPES_EXT__ 1
+#define __STDC_WANT_IEC_60559_FUNCS_EXT__ 1
+
+#include <math.h>
+#include <stdlib.h>
+#include <stddef.h>
+
+extern _Float128 roundf128 (_Float128);
+extern _Float128 floorf128 (_Float128);
+extern _Float128 ceilf128 (_Float128);
+extern _Float128 truncf128 (_Float128);
+
+static const struct {
+ _Float128 value;
+ _Float128 exp_round;
+ _Float128 exp_floor;
+ _Float128 exp_ceil;
+ _Float128 exp_trunc;
+} a[] = {
+ { -2.0Q, -2.0Q, -2.0Q, -2.0Q, -2.0Q },
+ { -1.7Q, -2.0Q, -2.0Q, -1.0Q, -1.0Q },
+ { -1.5Q, -2.0Q, -2.0Q, -1.0Q, -1.0Q },
+ { -1.3Q, -1.0Q, -2.0Q, -1.0Q, -1.0Q },
+ { +0.0Q, +0.0Q, +0.0Q, +0.0Q, +0.0Q },
+ { +1.3Q, +1.0Q, +1.0Q, +2.0Q, +1.0Q },
+ { +1.5Q, +2.0Q, +1.0Q, +2.0Q, +1.0Q },
+ { +1.7Q, +2.0Q, +1.0Q, +2.0Q, +1.0Q },
+ { +2.0Q, +2.0Q, +2.0Q, +2.0Q, +2.0Q }
+};
+
+int
+main (void)
+{
+ size_t i;
+ _Float128 v;
+
+ for (i = 0; i < sizeof (a) / sizeof (a[0]); i++)
+ {
+ v = a[i].value;
+ if (roundf128 (v) != a[i].exp_round)
+ abort ();
+
+ if (floorf128 (v) != a[i].exp_floor)
+ abort ();
+
+ if (ceilf128 (v) != a[i].exp_ceil)
+ abort ();
+
+ if (truncf128 (v) != a[i].exp_trunc)
+ abort ();
+ }
+
+ return 0;
+}
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH] Add _Float<N>/_Float<N>X rounding built-ins & improve gimple optimization of _Float<N>/_Float<N>X built-in functions
2017-11-18 1:33 ` Michael Meissner
@ 2017-11-20 17:39 ` Segher Boessenkool
2017-11-20 18:28 ` Michael Meissner
2017-12-21 18:16 ` Joseph Myers
1 sibling, 1 reply; 12+ messages in thread
From: Segher Boessenkool @ 2017-11-20 17:39 UTC (permalink / raw)
To: Michael Meissner, GCC Patches, David Edelsohn, Bill Schmidt,
Joseph Myers
Hi!
On Fri, Nov 17, 2017 at 07:35:05PM -0500, Michael Meissner wrote:
> Here is the fixed patch. It fixes the btrunc<mode>2 insn to use the correct
> XSRPQI variant for truncf128. I added the float128-hw11.c test as a runtime
> test to make sure round, trunc, ceil, and floor return the correct values. The
> machine independent portions are the same.
>
> Assuming the machine independent versions are approved, can I check in the
> PowerPC bits?
The rs6000 parts are okay, with a trivial fix:
> --- gcc/testsuite/gcc.target/powerpc/float128-hw11.c (revision 0)
> +++ gcc/testsuite/gcc.target/powerpc/float128-hw11.c (revision 0)
> @@ -0,0 +1,59 @@
> +/* { dg-do run { target { powerpc*-*-* && lp64 } } } */
> +/* { dg-require-effective-target powerpc_p9vector_ok } */
> +/* { dg-options "-mpower9-vector -O2" } */
> +/* { dg-options "-mvsx -O2" } */
This last line should not be there.
Thanks,
Segher
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH] Add _Float<N>/_Float<N>X rounding built-ins & improve gimple optimization of _Float<N>/_Float<N>X built-in functions
2017-11-20 17:39 ` Segher Boessenkool
@ 2017-11-20 18:28 ` Michael Meissner
0 siblings, 0 replies; 12+ messages in thread
From: Michael Meissner @ 2017-11-20 18:28 UTC (permalink / raw)
To: Segher Boessenkool
Cc: Michael Meissner, GCC Patches, David Edelsohn, Bill Schmidt,
Joseph Myers
On Mon, Nov 20, 2017 at 11:32:08AM -0600, Segher Boessenkool wrote:
> Hi!
>
> On Fri, Nov 17, 2017 at 07:35:05PM -0500, Michael Meissner wrote:
> > Here is the fixed patch. It fixes the btrunc<mode>2 insn to use the correct
> > XSRPQI variant for truncf128. I added the float128-hw11.c test as a runtime
> > test to make sure round, trunc, ceil, and floor return the correct values. The
> > machine independent portions are the same.
> >
> > Assuming the machine independent versions are approved, can I check in the
> > PowerPC bits?
>
> The rs6000 parts are okay, with a trivial fix:
>
> > --- gcc/testsuite/gcc.target/powerpc/float128-hw11.c (revision 0)
> > +++ gcc/testsuite/gcc.target/powerpc/float128-hw11.c (revision 0)
> > @@ -0,0 +1,59 @@
> > +/* { dg-do run { target { powerpc*-*-* && lp64 } } } */
> > +/* { dg-require-effective-target powerpc_p9vector_ok } */
> > +/* { dg-options "-mpower9-vector -O2" } */
> > +/* { dg-options "-mvsx -O2" } */
>
> This last line should not be there.
Ok. I will also substitute p9vector_hw for powerpc_p9vector_ok, since the
former says we have power9 hardware to run tests, and the later says the
compiler supports compiling for a power9 target. Sorry about that.
--
Michael Meissner, IBM
IBM, M/S 2506R, 550 King Street, Littleton, MA 01460-6245, USA
email: meissner@linux.vnet.ibm.com, phone: +1 (978) 899-4797
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH] Add _Float<N>/_Float<N>X rounding built-ins & improve gimple optimization of _Float<N>/_Float<N>X built-in functions
2017-11-18 1:33 ` Michael Meissner
2017-11-20 17:39 ` Segher Boessenkool
@ 2017-12-21 18:16 ` Joseph Myers
2017-12-21 19:03 ` Segher Boessenkool
2017-12-28 21:21 ` [PATCH] Add _Float<N>/_Float<N>X rounding built-ins & improve gimple optimization of _Float<N>/_Float<N>X built-in functions Michael Meissner
1 sibling, 2 replies; 12+ messages in thread
From: Joseph Myers @ 2017-12-21 18:16 UTC (permalink / raw)
To: Michael Meissner
Cc: Segher Boessenkool, GCC Patches, David Edelsohn, Bill Schmidt
On Fri, 17 Nov 2017, Michael Meissner wrote:
> Here is the fixed patch. It fixes the btrunc<mode>2 insn to use the correct
> XSRPQI variant for truncf128. I added the float128-hw11.c test as a runtime
> test to make sure round, trunc, ceil, and floor return the correct values. The
> machine independent portions are the same.
The architecture-independent changes are OK. However, I have a comment on
the target parts:
> +(define_insn "round<mode>2"
> + [(set (match_operand:IEEE128 0 "altivec_register_operand" "=v")
> + (unspec:IEEE128
> + [(match_operand:IEEE128 1 "altivec_register_operand" "v")]
> + UNSPEC_FRIN))]
> + "TARGET_FLOAT128_HW && FLOAT128_IEEE_P (<MODE>mode)"
> + "xsrqpi 0,%0,%1,3"
> + [(set_attr "type" "vecfloat")
> + (set_attr "size" "128")])
My reading of Power ISA 3.0B documentation is that 0,%0,%1,3 means round
in the mode specified by FPSCR and you need 0,%0,%1,0 for
round-to-nearest-away semantics which are what the round<mode>2
instruction has (i.e., what you've written here is actually correct for
nearbyint<mode>2, and would be rint<mode>2 if xsrqpix were used instead).
Note that the testcase
> Index: gcc/testsuite/gcc.target/powerpc/float128-hw11.c
> +static const struct {
> + _Float128 value;
> + _Float128 exp_round;
> + _Float128 exp_floor;
> + _Float128 exp_ceil;
> + _Float128 exp_trunc;
> +} a[] = {
> + { -2.0Q, -2.0Q, -2.0Q, -2.0Q, -2.0Q },
> + { -1.7Q, -2.0Q, -2.0Q, -1.0Q, -1.0Q },
> + { -1.5Q, -2.0Q, -2.0Q, -1.0Q, -1.0Q },
> + { -1.3Q, -1.0Q, -2.0Q, -1.0Q, -1.0Q },
> + { +0.0Q, +0.0Q, +0.0Q, +0.0Q, +0.0Q },
> + { +1.3Q, +1.0Q, +1.0Q, +2.0Q, +1.0Q },
> + { +1.5Q, +2.0Q, +1.0Q, +2.0Q, +1.0Q },
> + { +1.7Q, +2.0Q, +1.0Q, +2.0Q, +1.0Q },
> + { +2.0Q, +2.0Q, +2.0Q, +2.0Q, +2.0Q }
> +};
has only -1.5Q and +1.5Q as half-way inputs, and both of those inputs have
the same results for round-to-nearest-even and round-to-nearest-away, and
the test doesn't test changing the rounding mode from the default
round-to-nearest-even. So getting this case wrong would not have shown up
with a test failure; you'd need to add a further test input such as 2.5
for which round-to-nearest-even and round-to-nearest-away differ, or test
in multiple rounding modes to verify that the results of these built-in
functions do not depend on the rounding mode, or both.
(Of course xsrqpi can also implement roundevenf128 using 1,%0,%1,0 but GCC
doesn't currently have any built-in function or machine description
support for roundeven for any type.)
--
Joseph S. Myers
joseph@codesourcery.com
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH] Add _Float<N>/_Float<N>X rounding built-ins & improve gimple optimization of _Float<N>/_Float<N>X built-in functions
2017-12-21 18:16 ` Joseph Myers
@ 2017-12-21 19:03 ` Segher Boessenkool
2017-12-29 5:36 ` Michael Meissner
2017-12-28 21:21 ` [PATCH] Add _Float<N>/_Float<N>X rounding built-ins & improve gimple optimization of _Float<N>/_Float<N>X built-in functions Michael Meissner
1 sibling, 1 reply; 12+ messages in thread
From: Segher Boessenkool @ 2017-12-21 19:03 UTC (permalink / raw)
To: Joseph Myers; +Cc: Michael Meissner, GCC Patches, David Edelsohn, Bill Schmidt
On Thu, Dec 21, 2017 at 06:16:16PM +0000, Joseph Myers wrote:
> On Fri, 17 Nov 2017, Michael Meissner wrote:
> The architecture-independent changes are OK. However, I have a comment on
> the target parts:
>
> > +(define_insn "round<mode>2"
> > + [(set (match_operand:IEEE128 0 "altivec_register_operand" "=v")
> > + (unspec:IEEE128
> > + [(match_operand:IEEE128 1 "altivec_register_operand" "v")]
> > + UNSPEC_FRIN))]
> > + "TARGET_FLOAT128_HW && FLOAT128_IEEE_P (<MODE>mode)"
> > + "xsrqpi 0,%0,%1,3"
> > + [(set_attr "type" "vecfloat")
> > + (set_attr "size" "128")])
>
> My reading of Power ISA 3.0B documentation is that 0,%0,%1,3 means round
> in the mode specified by FPSCR and you need 0,%0,%1,0 for
> round-to-nearest-away semantics which are what the round<mode>2
> instruction has (i.e., what you've written here is actually correct for
> nearbyint<mode>2, and would be rint<mode>2 if xsrqpix were used instead).
Ah yes, the roundM2 insn is round-away-from-zero, so you are right.
Tricky, from the name I assumed it would be "current rounding mode" :-/
Not that "frin" would make sense if that were true.
Thanks! And thanks for all the reviews in general.
Segher
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH] Add _Float<N>/_Float<N>X rounding built-ins & improve gimple optimization of _Float<N>/_Float<N>X built-in functions
2017-12-21 18:16 ` Joseph Myers
2017-12-21 19:03 ` Segher Boessenkool
@ 2017-12-28 21:21 ` Michael Meissner
1 sibling, 0 replies; 12+ messages in thread
From: Michael Meissner @ 2017-12-28 21:21 UTC (permalink / raw)
To: Joseph Myers
Cc: Michael Meissner, Segher Boessenkool, GCC Patches,
David Edelsohn, Bill Schmidt
On Thu, Dec 21, 2017 at 06:16:16PM +0000, Joseph Myers wrote:
> On Fri, 17 Nov 2017, Michael Meissner wrote:
>
> > Here is the fixed patch. It fixes the btrunc<mode>2 insn to use the correct
> > XSRPQI variant for truncf128. I added the float128-hw11.c test as a runtime
> > test to make sure round, trunc, ceil, and floor return the correct values. The
> > machine independent portions are the same.
>
> The architecture-independent changes are OK. However, I have a comment on
> the target parts:
Ok, I have committed the machine independent patches, and I will revise the
machine dependent patches, and add more tests. Thanks for the review.
--
Michael Meissner, IBM
IBM, M/S 2506R, 550 King Street, Littleton, MA 01460-6245, USA
email: meissner@linux.vnet.ibm.com, phone: +1 (978) 899-4797
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [PATCH] Add _Float<N>/_Float<N>X rounding built-ins & improve gimple optimization of _Float<N>/_Float<N>X built-in functions
2017-12-21 19:03 ` Segher Boessenkool
@ 2017-12-29 5:36 ` Michael Meissner
2018-01-12 20:23 ` [committed] Fix float128-hw7.c testcase Jakub Jelinek
0 siblings, 1 reply; 12+ messages in thread
From: Michael Meissner @ 2017-12-29 5:36 UTC (permalink / raw)
To: Segher Boessenkool
Cc: Joseph Myers, Michael Meissner, GCC Patches, David Edelsohn,
Bill Schmidt
[-- Attachment #1: Type: text/plain, Size: 2476 bytes --]
On Thu, Dec 21, 2017 at 01:03:26PM -0600, Segher Boessenkool wrote:
> On Thu, Dec 21, 2017 at 06:16:16PM +0000, Joseph Myers wrote:
> > On Fri, 17 Nov 2017, Michael Meissner wrote:
> > The architecture-independent changes are OK. However, I have a comment on
> > the target parts:
> >
> > > +(define_insn "round<mode>2"
> > > + [(set (match_operand:IEEE128 0 "altivec_register_operand" "=v")
> > > + (unspec:IEEE128
> > > + [(match_operand:IEEE128 1 "altivec_register_operand" "v")]
> > > + UNSPEC_FRIN))]
> > > + "TARGET_FLOAT128_HW && FLOAT128_IEEE_P (<MODE>mode)"
> > > + "xsrqpi 0,%0,%1,3"
> > > + [(set_attr "type" "vecfloat")
> > > + (set_attr "size" "128")])
> >
> > My reading of Power ISA 3.0B documentation is that 0,%0,%1,3 means round
> > in the mode specified by FPSCR and you need 0,%0,%1,0 for
> > round-to-nearest-away semantics which are what the round<mode>2
> > instruction has (i.e., what you've written here is actually correct for
> > nearbyint<mode>2, and would be rint<mode>2 if xsrqpix were used instead).
>
> Ah yes, the roundM2 insn is round-away-from-zero, so you are right.
> Tricky, from the name I assumed it would be "current rounding mode" :-/
> Not that "frin" would make sense if that were true.
>
> Thanks! And thanks for all the reviews in general.
Here is the corrected rs6000 part of the patch. I added more round tests and I
checked it on a power9 prototype machine. Roundf128 now produces the correct
answer. Can I check this into the trunk?
[gcc]
2017-12-29 Michael Meissner <meissner@linux.vnet.ibm.com>
* config/rs6000/rs6000.md (floor<mode>2): Add support for IEEE
128-bit round to integer instructions.
(ceil<mode>2): Likewise.
(btrunc<mode>2): Likewise.
(round<mode>2): Likewise.
[gcc/testsuite]
2017-12-29 Michael Meissner <meissner@linux.vnet.ibm.com>
* gcc.target/powerpc/float128-hw2.c: Add tests for ceilf128,
floorf128, truncf128, and roundf128.
* gcc.target/powerpc/float128-hw5.c: New tests for _Float128
optimizations added in match.pd.
* gcc.target/powerpc/float128-hw6.c: Likewise.
* gcc.target/powerpc/float128-hw7.c: Likewise.
* gcc.target/powerpc/float128-hw8.c: Likewise.
* gcc.target/powerpc/float128-hw9.c: Likewise.
* gcc.target/powerpc/float128-hw10.c: Likewise.
* gcc.target/powerpc/float128-hw11.c: Likewise.
--
Michael Meissner, IBM
IBM, M/S 2506R, 550 King Street, Littleton, MA 01460-6245, USA
email: meissner@linux.vnet.ibm.com, phone: +1 (978) 899-4797
[-- Attachment #2: ieee128-patch71b --]
[-- Type: text/plain, Size: 11790 bytes --]
Index: gcc/config/rs6000/rs6000.md
===================================================================
--- gcc/config/rs6000/rs6000.md (revision 256026)
+++ gcc/config/rs6000/rs6000.md (working copy)
@@ -14777,6 +14777,47 @@ (define_insn_and_split "floatuns<QHI:mod
(set_attr "type" "vecfloat")
(set_attr "size" "128")])
+;; IEEE 128-bit round to integer built-in functions
+(define_insn "floor<mode>2"
+ [(set (match_operand:IEEE128 0 "altivec_register_operand" "=v")
+ (unspec:IEEE128
+ [(match_operand:IEEE128 1 "altivec_register_operand" "v")]
+ UNSPEC_FRIM))]
+ "TARGET_FLOAT128_HW && FLOAT128_IEEE_P (<MODE>mode)"
+ "xsrqpi 1,%0,%1,3"
+ [(set_attr "type" "vecfloat")
+ (set_attr "size" "128")])
+
+(define_insn "ceil<mode>2"
+ [(set (match_operand:IEEE128 0 "altivec_register_operand" "=v")
+ (unspec:IEEE128
+ [(match_operand:IEEE128 1 "altivec_register_operand" "v")]
+ UNSPEC_FRIP))]
+ "TARGET_FLOAT128_HW && FLOAT128_IEEE_P (<MODE>mode)"
+ "xsrqpi 1,%0,%1,2"
+ [(set_attr "type" "vecfloat")
+ (set_attr "size" "128")])
+
+(define_insn "btrunc<mode>2"
+ [(set (match_operand:IEEE128 0 "altivec_register_operand" "=v")
+ (unspec:IEEE128
+ [(match_operand:IEEE128 1 "altivec_register_operand" "v")]
+ UNSPEC_FRIZ))]
+ "TARGET_FLOAT128_HW && FLOAT128_IEEE_P (<MODE>mode)"
+ "xsrqpi 1,%0,%1,1"
+ [(set_attr "type" "vecfloat")
+ (set_attr "size" "128")])
+
+(define_insn "round<mode>2"
+ [(set (match_operand:IEEE128 0 "altivec_register_operand" "=v")
+ (unspec:IEEE128
+ [(match_operand:IEEE128 1 "altivec_register_operand" "v")]
+ UNSPEC_FRIN))]
+ "TARGET_FLOAT128_HW && FLOAT128_IEEE_P (<MODE>mode)"
+ "xsrqpi 0,%0,%1,0"
+ [(set_attr "type" "vecfloat")
+ (set_attr "size" "128")])
+
;; IEEE 128-bit instructions with round to odd semantics
(define_insn "add<mode>3_odd"
[(set (match_operand:IEEE128 0 "altivec_register_operand" "=v")
Index: gcc/testsuite/gcc.target/powerpc/float128-hw2.c
===================================================================
--- gcc/testsuite/gcc.target/powerpc/float128-hw2.c (revision 256026)
+++ gcc/testsuite/gcc.target/powerpc/float128-hw2.c (working copy)
@@ -14,6 +14,10 @@
extern _Float128 copysignf128 (_Float128, _Float128);
extern _Float128 sqrtf128 (_Float128);
extern _Float128 fmaf128 (_Float128, _Float128, _Float128);
+extern _Float128 ceilf128 (_Float128);
+extern _Float128 floorf128 (_Float128);
+extern _Float128 truncf128 (_Float128);
+extern _Float128 roundf128 (_Float128);
_Float128
do_copysign (_Float128 a, _Float128 b)
@@ -51,10 +55,35 @@ do_nfms (_Float128 a, _Float128 b, _Floa
return -fmaf128 (a, b, -c);
}
+_Float128
+do_ceil (_Float128 a)
+{
+ return ceilf128 (a);
+}
+
+_Float128
+do_floor (_Float128 a)
+{
+ return floorf128 (a);
+}
+
+_Float128
+do_trunc (_Float128 a)
+{
+ return truncf128 (a);
+}
+
+_Float128
+do_round (_Float128 a)
+{
+ return roundf128 (a);
+}
+
/* { dg-final { scan-assembler {\mxscpsgnqp\M} } } */
/* { dg-final { scan-assembler {\mxssqrtqp\M} } } */
/* { dg-final { scan-assembler {\mxsmaddqp\M} } } */
/* { dg-final { scan-assembler {\mxsmsubqp\M} } } */
/* { dg-final { scan-assembler {\mxsnmaddqp\M} } } */
/* { dg-final { scan-assembler {\mxsnmsubqp\M} } } */
+/* { dg-final { scan-assembler {\mxsrqpi\M} } } */
/* { dg-final { scan-assembler-not {\mbl\M} } } */
Index: gcc/testsuite/gcc.target/powerpc/float128-hw5.c
===================================================================
--- gcc/testsuite/gcc.target/powerpc/float128-hw5.c (nonexistent)
+++ gcc/testsuite/gcc.target/powerpc/float128-hw5.c (working copy)
@@ -0,0 +1,33 @@
+/* { dg-do compile { target { powerpc*-*-* && lp64 } } } */
+/* { dg-require-effective-target powerpc_p9vector_ok } */
+/* { dg-options "-mpower9-vector -O2 -ffast-math" } */
+
+extern _Float128 copysignf128 (_Float128, _Float128);
+
+/* Check copysign optimizations that are done for double are also done for
+ _Float128. */
+
+_Float128
+x_times_cs_one_negx (_Float128 x)
+{
+ return x * copysignf128 (1.0Q, -x); /* XSNABSQP */
+}
+
+_Float128
+x_times_cs_one_x (_Float128 x)
+{
+ return x * copysignf128 (1.0Q, x); /* XSABSQP */
+}
+
+_Float128
+cs_x_x (_Float128 x)
+{
+ return copysignf128 (x, x); /* no operation. */
+}
+
+/* { dg-final { scan-assembler-times {\mxsabsqp\M} 1 } } */
+/* { dg-final { scan-assembler-times {\mxsnabsqp\M} 1 } } */
+/* { dg-final { scan-assembler-not {\mxscpsgnqp\M} } } */
+/* { dg-final { scan-assembler-not {\mlxvx\M} } } */
+/* { dg-final { scan-assembler-not {\mlxv\M} } } */
+/* { dg-final { scan-assembler-not {\mbl\M} } } */
Index: gcc/testsuite/gcc.target/powerpc/float128-hw6.c
===================================================================
--- gcc/testsuite/gcc.target/powerpc/float128-hw6.c (nonexistent)
+++ gcc/testsuite/gcc.target/powerpc/float128-hw6.c (working copy)
@@ -0,0 +1,26 @@
+/* { dg-do compile { target { powerpc*-*-* && lp64 } } } */
+/* { dg-require-effective-target powerpc_p9vector_ok } */
+/* { dg-options "-mpower9-vector -O2" } */
+
+extern _Float128 fabsf128 (_Float128);
+extern _Float128 copysignf128 (_Float128, _Float128);
+
+/* Check copysign optimizations that are done for double are also done for
+ _Float128. */
+
+_Float128
+cs_negx_y (_Float128 x, _Float128 y)
+{
+ return copysignf128 (-x, y); /* eliminate negation. */
+}
+
+_Float128
+cs_absx_y (_Float128 x, _Float128 y)
+{
+ return copysignf128 (fabsf128 (x), y); /* eliminate fabsf128. */
+}
+
+/* { dg-final { scan-assembler-times {\mxscpsgnqp\M} 2 } } */
+/* { dg-final { scan-assembler-not {\mxsnegqp\M} } } */
+/* { dg-final { scan-assembler-not {\mxsabsqp\M} } } */
+/* { dg-final { scan-assembler-not {\mbl\M} } } */
Index: gcc/testsuite/gcc.target/powerpc/float128-hw7.c
===================================================================
--- gcc/testsuite/gcc.target/powerpc/float128-hw7.c (nonexistent)
+++ gcc/testsuite/gcc.target/powerpc/float128-hw7.c (working copy)
@@ -0,0 +1,27 @@
+/* { dg-do compile { target { powerpc*-*-* && lp64 } } } */
+/* { dg-require-effective-target powerpc_p9vector_ok } */
+/* { dg-options "-mpower9-vector -O2" } */
+
+extern _Float128 fabsf128 (_Float128);
+extern _Float128 copysignf128 (_Float128, _Float128);
+
+/* Check copysign optimizations that are done for double are also done for
+ _Float128. */
+
+_Float128
+cs_x_pos1 (_Float128 x)
+{
+ return copysignf128 (x, 1.0Q); /* XSABSQP. */
+}
+
+_Float128 cs_x_neg2 (_Float128 x)
+{
+ return copysignf128 (x, -2.0Q); /* XSNABSQP. */
+}
+
+/* { dg-final { scan-assembler-times {\mxsabsqp\M} 1 } } */
+/* { dg-final { scan-assembler-not {\mxsnabsqp\M} 1 } } */
+/* { dg-final { scan-assembler-not {\mxscpsgnqp\M} } } */
+/* { dg-final { scan-assembler-not {\mlxvx\M} } } */
+/* { dg-final { scan-assembler-not {\mlxv\M} } } */
+/* { dg-final { scan-assembler-not {\mbl\M} } } */
Index: gcc/testsuite/gcc.target/powerpc/float128-hw8.c
===================================================================
--- gcc/testsuite/gcc.target/powerpc/float128-hw8.c (nonexistent)
+++ gcc/testsuite/gcc.target/powerpc/float128-hw8.c (working copy)
@@ -0,0 +1,24 @@
+/* { dg-do compile { target { powerpc*-*-* && lp64 } } } */
+/* { dg-require-effective-target powerpc_p9vector_ok } */
+/* { dg-options "-mpower9-vector -O2" } */
+
+extern _Float128 fminf128 (_Float128, _Float128);
+extern _Float128 fmaxf128 (_Float128, _Float128);
+
+/* Check min/max optimizations that are done for double are also done for
+ _Float128. */
+
+_Float128
+min_x_x (_Float128 x)
+{
+ return fminf128 (x, x);
+}
+
+_Float128
+max_x_x (_Float128 x)
+{
+ return fmaxf128 (x, x);
+}
+
+/* { dg-final { scan-assembler-not {\mxscmpuqp\M} } } */
+/* { dg-final { scan-assembler-not {\mbl\M} } } */
Index: gcc/testsuite/gcc.target/powerpc/float128-hw9.c
===================================================================
--- gcc/testsuite/gcc.target/powerpc/float128-hw9.c (nonexistent)
+++ gcc/testsuite/gcc.target/powerpc/float128-hw9.c (working copy)
@@ -0,0 +1,17 @@
+/* { dg-do compile { target { powerpc*-*-* && lp64 } } } */
+/* { dg-require-effective-target powerpc_p9vector_ok } */
+/* { dg-options "-mpower9-vector -O2 -ffast-math" } */
+
+extern _Float128 sqrtf128 (_Float128);
+
+/* Check sqrt optimizations that are done for double are also done for
+ _Float128. */
+
+_Float128
+sqrt_x_times_sqrt_x (_Float128 x)
+{
+ return sqrtf128 (x) * sqrtf128 (x);
+}
+
+/* { dg-final { scan-assembler-not {\mxssqrtqp\M} } } */
+/* { dg-final { scan-assembler-not {\mbl\M} } } */
Index: gcc/testsuite/gcc.target/powerpc/float128-hw10.c
===================================================================
--- gcc/testsuite/gcc.target/powerpc/float128-hw10.c (nonexistent)
+++ gcc/testsuite/gcc.target/powerpc/float128-hw10.c (working copy)
@@ -0,0 +1,38 @@
+/* { dg-do compile { target { powerpc*-*-* && lp64 } } } */
+/* { dg-require-effective-target powerpc_p9vector_ok } */
+/* { dg-options "-mpower9-vector -O2" } */
+
+extern _Float128 floorf128 (_Float128);
+extern _Float128 ceilf128 (_Float128);
+extern _Float128 roundf128 (_Float128);
+extern _Float128 truncf128 (_Float128);
+
+/* Check rounding optimizations that are done for double are also done for
+ _Float128. */
+
+_Float128
+floor_floor_x (_Float128 x)
+{
+ return floorf128 (floorf128 (x));
+}
+
+_Float128
+ceil_ceil_x (_Float128 x)
+{
+ return ceilf128 (ceilf128 (x));
+}
+
+_Float128
+trunc_trunc_x (_Float128 x)
+{
+ return truncf128 (truncf128 (x));
+}
+
+_Float128
+round_round_x (_Float128 x)
+{
+ return roundf128 (roundf128 (x));
+}
+
+/* { dg-final { scan-assembler-times {\mxsrqpi\M} 4 } } */
+/* { dg-final { scan-assembler-not {\mbl\M} } } */
Index: gcc/testsuite/gcc.target/powerpc/float128-hw11.c
===================================================================
--- gcc/testsuite/gcc.target/powerpc/float128-hw11.c (nonexistent)
+++ gcc/testsuite/gcc.target/powerpc/float128-hw11.c (working copy)
@@ -0,0 +1,85 @@
+/* { dg-do run { target { powerpc*-*-* && lp64 } } } */
+/* { dg-require-effective-target p9vector_hw } */
+/* { dg-options "-mpower9-vector -O2" } */
+
+#define __STDC_WANT_IEC_60559_TYPES_EXT__ 1
+#define __STDC_WANT_IEC_60559_FUNCS_EXT__ 1
+
+#include <math.h>
+#include <stdlib.h>
+#include <stddef.h>
+
+#ifdef DEBUG
+#include <stdio.h>
+#define FAIL(T,G,E) printf ("%s: %g %g\n", (T), (G), (E))
+#else
+#define FAIL(T,G,E) abort ()
+#endif
+
+extern _Float128 roundf128 (_Float128);
+extern _Float128 floorf128 (_Float128);
+extern _Float128 ceilf128 (_Float128);
+extern _Float128 truncf128 (_Float128);
+
+static const _Float128 values[] = {
+ -3.7f128,
+ -3.5f128,
+ -3.3f128,
+ -3.0f128,
+ -2.7f128,
+ -2.5f128,
+ -2.3f128,
+ -2.0f128,
+ -1.7f128,
+ -1.5f128,
+ -1.3f128,
+ -0.7f128,
+ -0.5f128,
+ -0.3f128,
+ +0.0f128,
+ +0.3f128,
+ +0.5f128,
+ +0.7f128,
+ +1.0f128,
+ +1.3f128,
+ +1.5f128,
+ +1.7f128,
+ +2.0f128,
+ +2.3f128,
+ +2.5f128,
+ +2.7f128,
+ +3.0f128,
+ +3.3f128,
+ +3.5f128,
+ +3.7f128,
+};
+
+__attribute__((__noinline__))
+void
+do_test (const char *test, _Float128 got, double expected)
+{
+ double got2 = (double) got;
+ if (got2 != expected)
+ FAIL (test, got2, expected);
+}
+
+int
+main (void)
+{
+ size_t i;
+ _Float128 v;
+ double v2;
+
+ for (i = 0; i < sizeof (values) / sizeof (values[0]); i++)
+ {
+ v = values[i];
+ v2 = (double) v;
+
+ do_test ("roundf128", roundf128 (v), round (v2));
+ do_test ("floorf128", floorf128 (v), floor (v2));
+ do_test ("ceilf128", ceilf128 (v), ceil (v2));
+ do_test ("truncf128", truncf128 (v), trunc (v2));
+ }
+
+ return 0;
+}
^ permalink raw reply [flat|nested] 12+ messages in thread
* [committed] Fix float128-hw7.c testcase
2017-12-29 5:36 ` Michael Meissner
@ 2018-01-12 20:23 ` Jakub Jelinek
2018-01-12 21:29 ` Segher Boessenkool
0 siblings, 1 reply; 12+ messages in thread
From: Jakub Jelinek @ 2018-01-12 20:23 UTC (permalink / raw)
To: Michael Meissner, Segher Boessenkool, Joseph Myers, GCC Patches,
David Edelsohn, Bill Schmidt
On Fri, Dec 29, 2017 at 12:35:03AM -0500, Michael Meissner wrote:
> * gcc.target/powerpc/float128-hw7.c: Likewise.
> --- gcc/testsuite/gcc.target/powerpc/float128-hw7.c (nonexistent)
> +++ gcc/testsuite/gcc.target/powerpc/float128-hw7.c (working copy)
> @@ -0,0 +1,27 @@
> +/* { dg-do compile { target { powerpc*-*-* && lp64 } } } */
> +/* { dg-require-effective-target powerpc_p9vector_ok } */
> +/* { dg-options "-mpower9-vector -O2" } */
> +
> +extern _Float128 fabsf128 (_Float128);
> +extern _Float128 copysignf128 (_Float128, _Float128);
> +
> +/* Check copysign optimizations that are done for double are also done for
> + _Float128. */
> +
> +_Float128
> +cs_x_pos1 (_Float128 x)
> +{
> + return copysignf128 (x, 1.0Q); /* XSABSQP. */
> +}
> +
> +_Float128 cs_x_neg2 (_Float128 x)
> +{
> + return copysignf128 (x, -2.0Q); /* XSNABSQP. */
> +}
> +
> +/* { dg-final { scan-assembler-times {\mxsabsqp\M} 1 } } */
> +/* { dg-final { scan-assembler-not {\mxsnabsqp\M} 1 } } */
> +/* { dg-final { scan-assembler-not {\mxscpsgnqp\M} } } */
> +/* { dg-final { scan-assembler-not {\mlxvx\M} } } */
> +/* { dg-final { scan-assembler-not {\mlxv\M} } } */
> +/* { dg-final { scan-assembler-not {\mbl\M} } } */
This testcase fails with a syntax error, scan-assembler-not with
2 operands would need to have target or xfail as the second operand, not 1.
Removing the 1 doesn't work, because xsnabsqp instruction is actually
emitted. Given that the comments on the functions explicitly mention
that it expects one XSABSQP and one XSNABSQP instruction, I've changed
it to scan-assembler-times instead, tested on powerpc64le-linux and
committed as obvious to trunk.
2018-01-12 Jakub Jelinek <jakub@redhat.com>
* gcc.target/powerpc/float128-hw7.c: Use scan-assembler-times
instead of scan-assembler-not for xsnabsqp.
--- gcc/testsuite/gcc.target/powerpc/float128-hw7.c.jj 2018-01-12 18:14:53.714091910 +0000
+++ gcc/testsuite/gcc.target/powerpc/float128-hw7.c 2018-01-12 20:13:43.706918034 +0000
@@ -20,7 +20,7 @@ _Float128 cs_x_neg2 (_Float128 x)
}
/* { dg-final { scan-assembler-times {\mxsabsqp\M} 1 } } */
-/* { dg-final { scan-assembler-not {\mxsnabsqp\M} 1 } } */
+/* { dg-final { scan-assembler-times {\mxsnabsqp\M} 1 } } */
/* { dg-final { scan-assembler-not {\mxscpsgnqp\M} } } */
/* { dg-final { scan-assembler-not {\mlxvx\M} } } */
/* { dg-final { scan-assembler-not {\mlxv\M} } } */
Jakub
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [committed] Fix float128-hw7.c testcase
2018-01-12 20:23 ` [committed] Fix float128-hw7.c testcase Jakub Jelinek
@ 2018-01-12 21:29 ` Segher Boessenkool
0 siblings, 0 replies; 12+ messages in thread
From: Segher Boessenkool @ 2018-01-12 21:29 UTC (permalink / raw)
To: Jakub Jelinek
Cc: Michael Meissner, Joseph Myers, GCC Patches, David Edelsohn,
Bill Schmidt
On Fri, Jan 12, 2018 at 09:20:50PM +0100, Jakub Jelinek wrote:
> On Fri, Dec 29, 2017 at 12:35:03AM -0500, Michael Meissner wrote:
> > +/* { dg-final { scan-assembler-times {\mxsabsqp\M} 1 } } */
> > +/* { dg-final { scan-assembler-not {\mxsnabsqp\M} 1 } } */
> > +/* { dg-final { scan-assembler-not {\mxscpsgnqp\M} } } */
> > +/* { dg-final { scan-assembler-not {\mlxvx\M} } } */
> > +/* { dg-final { scan-assembler-not {\mlxv\M} } } */
> > +/* { dg-final { scan-assembler-not {\mbl\M} } } */
>
> This testcase fails with a syntax error, scan-assembler-not with
> 2 operands would need to have target or xfail as the second operand, not 1.
> Removing the 1 doesn't work, because xsnabsqp instruction is actually
> emitted. Given that the comments on the functions explicitly mention
> that it expects one XSABSQP and one XSNABSQP instruction, I've changed
> it to scan-assembler-times instead, tested on powerpc64le-linux and
> committed as obvious to trunk.
Thanks!
Segher
^ permalink raw reply [flat|nested] 12+ messages in thread
end of thread, other threads:[~2018-01-12 21:26 UTC | newest]
Thread overview: 12+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2017-11-17 5:33 [PATCH] Add _Float<N>/_Float<N>X rounding built-ins & improve gimple optimization of _Float<N>/_Float<N>X built-in functions Michael Meissner
2017-11-17 14:26 ` Segher Boessenkool
2017-11-17 21:41 ` Michael Meissner
2017-11-18 1:33 ` Michael Meissner
2017-11-20 17:39 ` Segher Boessenkool
2017-11-20 18:28 ` Michael Meissner
2017-12-21 18:16 ` Joseph Myers
2017-12-21 19:03 ` Segher Boessenkool
2017-12-29 5:36 ` Michael Meissner
2018-01-12 20:23 ` [committed] Fix float128-hw7.c testcase Jakub Jelinek
2018-01-12 21:29 ` Segher Boessenkool
2017-12-28 21:21 ` [PATCH] Add _Float<N>/_Float<N>X rounding built-ins & improve gimple optimization of _Float<N>/_Float<N>X built-in functions Michael Meissner
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).