* [PATCH 6/N v2] [x86_64] Vectorized math functions
@ 2014-12-02 16:50 Andrew Senkevich
2014-12-03 18:05 ` Joseph Myers
0 siblings, 1 reply; 5+ messages in thread
From: Andrew Senkevich @ 2014-12-02 16:50 UTC (permalink / raw)
To: libc-alpha
[-- Attachment #1: Type: text/plain, Size: 1263 bytes --]
Patch adds installation of libm.so as linker script (only in case of
libmvec.so build).
2014-12-02 Andrew Senkevich <andrew.n.senkevich@gmail.com>
* math/Makefile: Added rule for libm.so installation.
diff --git a/math/Makefile b/math/Makefile
index fbd2ded..a77687e 100644
--- a/math/Makefile
+++ b/math/Makefile
@@ -86,6 +86,22 @@ generated += $(foreach s,.c .S l.c l.S f.c
f.S,$(calls:s_%=m_%$s))
routines = $(calls) $(calls:=f) $(long-c-$(long-double-fcts))
long-c-yes = $(calls:=l)
+ifeq ($(build-mathvec),yes)
+# We need to install libm.so as linker script
+# for more comfortable use of vector math library.
+install-lib-ldscripts := libm.so
+libmvec-so-ver = libmvec.so$(libmvec.so-version)
+install_subdir: $(inst_libdir)/libm.so
+$(inst_libdir)/libm.so: $(common-objpfx)format.lds $(libm) \
+ $(common-objpfx)mathvec/$(libmvec-so-ver) \
+ $(+force)
+ (echo '/* GNU ld script'; echo '*/';\
+ cat $<; \
+ echo 'GROUP ( $(slibdir)/libm.so$(libm.so-version) ' \
+ 'AS_NEEDED ( $(slibdir)/libmvec.so$(libmvec.so-version) ) )' \
+ ) > $@
+endif
+
# Rules for the test suite.
tests = test-matherr test-fenv atest-exp atest-sincos atest-exp2 basic-test \
test-misc test-fpucw test-fpucw-ieee tst-definitions test-tgmath \
--
WBR,
Andrew
[-- Attachment #2: libmvec_6_of_N_v4.patch --]
[-- Type: application/octet-stream, Size: 1035 bytes --]
diff --git a/math/Makefile b/math/Makefile
index fbd2ded..a77687e 100644
--- a/math/Makefile
+++ b/math/Makefile
@@ -86,6 +86,22 @@ generated += $(foreach s,.c .S l.c l.S f.c f.S,$(calls:s_%=m_%$s))
routines = $(calls) $(calls:=f) $(long-c-$(long-double-fcts))
long-c-yes = $(calls:=l)
+ifeq ($(build-mathvec),yes)
+# We need to install libm.so as linker script
+# for more comfortable use of vector math library.
+install-lib-ldscripts := libm.so
+libmvec-so-ver = libmvec.so$(libmvec.so-version)
+install_subdir: $(inst_libdir)/libm.so
+$(inst_libdir)/libm.so: $(common-objpfx)format.lds $(libm) \
+ $(common-objpfx)mathvec/$(libmvec-so-ver) \
+ $(+force)
+ (echo '/* GNU ld script'; echo '*/';\
+ cat $<; \
+ echo 'GROUP ( $(slibdir)/libm.so$(libm.so-version) ' \
+ 'AS_NEEDED ( $(slibdir)/libmvec.so$(libmvec.so-version) ) )' \
+ ) > $@
+endif
+
# Rules for the test suite.
tests = test-matherr test-fenv atest-exp atest-sincos atest-exp2 basic-test \
test-misc test-fpucw test-fpucw-ieee tst-definitions test-tgmath \
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH 6/N v2] [x86_64] Vectorized math functions
2014-12-02 16:50 [PATCH 6/N v2] [x86_64] Vectorized math functions Andrew Senkevich
@ 2014-12-03 18:05 ` Joseph Myers
0 siblings, 0 replies; 5+ messages in thread
From: Joseph Myers @ 2014-12-03 18:05 UTC (permalink / raw)
To: Andrew Senkevich; +Cc: libc-alpha
On Tue, 2 Dec 2014, Andrew Senkevich wrote:
> +ifeq ($(build-mathvec),yes)
> +# We need to install libm.so as linker script
> +# for more comfortable use of vector math library.
> +install-lib-ldscripts := libm.so
> +libmvec-so-ver = libmvec.so$(libmvec.so-version)
> +install_subdir: $(inst_libdir)/libm.so
> +$(inst_libdir)/libm.so: $(common-objpfx)format.lds $(libm) \
> + $(common-objpfx)mathvec/$(libmvec-so-ver) \
> + $(+force)
> + (echo '/* GNU ld script'; echo '*/';\
> + cat $<; \
> + echo 'GROUP ( $(slibdir)/libm.so$(libm.so-version) ' \
> + 'AS_NEEDED ( $(slibdir)/libmvec.so$(libmvec.so-version) ) )' \
Watch indentation. Commands inside () should be indented one space after
the initial TAB so they all line up; the 'AS_NEEDED...' argument to echo
should be indented to line up with the 'GROUP...' argument.
I think patches 5 and 6 in the series should be combined (that is,
building and installing the library should be in the same patch).
--
Joseph S. Myers
joseph@codesourcery.com
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH 6/N v2] [x86_64] Vectorized math functions
2014-11-24 19:00 Andrew Senkevich
2014-11-25 17:39 ` Andrew Senkevich
@ 2014-11-26 17:56 ` Joseph Myers
1 sibling, 0 replies; 5+ messages in thread
From: Joseph Myers @ 2014-11-26 17:56 UTC (permalink / raw)
To: Andrew Senkevich; +Cc: libc-alpha
On Mon, 24 Nov 2014, Andrew Senkevich wrote:
> +#define CONCAT(a,b) __CONCAT (a,b)
> +
> +#define WRAPPER_NAME(function) CONCAT (function,VEC_SUFF)
Watch out for spaces after commas. Lots of places in this patch need
them.
> diff --git a/sysdeps/x86_64/fpu/libm-test-ulps
> b/sysdeps/x86_64/fpu/libm-test-ulps
> index 36e1b76..0a0906e 100644
> --- a/sysdeps/x86_64/fpu/libm-test-ulps
> +++ b/sysdeps/x86_64/fpu/libm-test-ulps
> @@ -905,6 +905,15 @@ idouble: 1
> ildouble: 2
> ldouble: 2
>
> +Function: "cos_vlen2":
> +double: 1
> +
> +Function: "cos_vlen4_avx":
> +double: 1
> +
> +Function: "cos_vlen4_avx2":
> +double: 1
> +
> Function: "cosh":
> double: 1
> float: 1
> @@ -1961,6 +1970,15 @@ ifloat: 3
> ildouble: 4
> ldouble: 4
>
> +Function: "vlen2_cos":
> +double: 1
> +
> +Function: "vlen4_avx2_cos":
> +double: 1
> +
> +Function: "vlen4_avx_cos":
> +double: 1
> +
> Function: "y0":
> double: 2
> float: 1
You shouldn't have this second part of the change, now suffixes are being
used.
> +// Wrapper from scalar to vector function implemented in SSE4.
> +#define VECTOR_WRAPPER(scalar_func,vector_func) \
> +extern __m128d vector_func ( __m128d);\
> +FLOAT scalar_func (FLOAT x)\
> +{\
> + int i;\
> + __m128d mx = _mm_set1_pd (x);\
> + __m128d mr = vector_func (mx);\
> + for (i=1;i<2;i++)\
> + {\
> + if (((FLOAT*)&mr)[0]!=((FLOAT*)&mr)[i])\
> + {\
> + return ((FLOAT*)&mr)[0]+0.1;\
> + }\
> + }\
> + return ((FLOAT*)&mr)[0];\
> +}
Watch out for formatting. Spaces around binary operators, before '*',
after the ')' of a cast, after ';' in for (). No space after '(' in the
extern declaration.
The loop and return look like something architecture-independent and
vector-length-independent that should be factored out accordingly into
something taking the vector length (= loop length) as an argument.
--
Joseph S. Myers
joseph@codesourcery.com
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH 6/N v2] [x86_64] Vectorized math functions
2014-11-24 19:00 Andrew Senkevich
@ 2014-11-25 17:39 ` Andrew Senkevich
2014-11-26 17:56 ` Joseph Myers
1 sibling, 0 replies; 5+ messages in thread
From: Andrew Senkevich @ 2014-11-25 17:39 UTC (permalink / raw)
To: libc-alpha
[-- Attachment #1: Type: text/plain, Size: 12966 bytes --]
2014-11-24 21:59 GMT+03:00 Andrew Senkevich <andrew.n.senkevich@gmail.com>:
> This modified according updates in 1st, 2nd and 3rd patches version of
> 6th patch (sent in
> https://sourceware.org/ml/libc-alpha/2014-11/msg00603.html).
> Patches 4 and 5 are unchanged.
>
> It is addition of tests for vectorized cos, which are
> implemented as wrappers to vector versions from scalar ones.
[Patch 6/N v2] doesn't contain changes of libm-test.inc, now added.
ChangeLog
2014-11-25 Andrew Senkevich <andrew.n.senkevich@gmail.com>
* math/libm-test.inc: Added INIT_ARCH_EXT.
* math/Makefile: Added rules for libmvec tests.
* sysdeps/x86_64/fpu/Makefile: Likewise.
* sysdeps/x86_64/fpu/libm-test-ulps: Regenerated.
* math/test-double-vlen2.h: New file.
* math/test-double-vlen4.h: New file.
* sysdeps/x86_64/fpu/test-double-vlen2.c: New file.
* sysdeps/x86_64/fpu/test-double-vlen4.c: New file.
* sysdeps/x86_64/fpu/test-double-vlen4-avx2.c: New file.
diff --git a/math/Makefile b/math/Makefile
index 8a89308..a2240d0 100644
--- a/math/Makefile
+++ b/math/Makefile
@@ -97,8 +97,9 @@ tests-static = test-fpucw-static test-fpucw-ieee-static
test-longdouble-yes = test-ldouble test-ildoubl
ifneq (no,$(PERL))
+libm-vec-tests = $(addprefix test-,$(libmvec-tests))
libm-tests = test-float test-double $(test-longdouble-$(long-double-fcts)) \
- test-ifloat test-idouble
+ test-ifloat test-idouble $(libm-vec-tests)
libm-tests.o = $(addsuffix .o,$(libm-tests))
tests += $(libm-tests)
@@ -125,8 +126,22 @@ $(objpfx)test-double.o: $(objpfx)libm-test.stmp
$(objpfx)test-idouble.o: $(objpfx)libm-test.stmp
$(objpfx)test-ldouble.o: $(objpfx)libm-test.stmp
$(objpfx)test-ildoubl.o: $(objpfx)libm-test.stmp
+
+$(objpfx)test-double-vlen2.o: $(objpfx)libm-test.stmp
+$(objpfx)test-double-vlen4.o: $(objpfx)libm-test.stmp
+
+$(objpfx)test-double-vlen2: $(common-objpfx)mathvec/libmvec.so \
+ $(objpfx)init-arch.o
+$(objpfx)test-double-vlen4: $(common-objpfx)mathvec/libmvec.so \
+ $(objpfx)init-arch.o
endif
+CFLAGS-test-double-vlen2.c = -fno-inline -ffloat-store -fno-builtin
-frounding-math \
+ -D__FAST_MATH__ -DTEST_FAST_MATH -D_OPENMP=201307 \
+ -Wno-unknown-pragmas
+CFLAGS-test-double-vlen4.c = -fno-inline -ffloat-store -fno-builtin
-frounding-math \
+ -D__FAST_MATH__ -DTEST_FAST_MATH -D_OPENMP=201307 \
+ -Wno-unknown-pragmas $(arch-ext-cflags)
CFLAGS-test-float.c = -fno-inline -ffloat-store -fno-builtin -frounding-math
CFLAGS-test-double.c = -fno-inline -ffloat-store -fno-builtin -frounding-math
CFLAGS-test-ldouble.c = -fno-inline -ffloat-store -fno-builtin -frounding-math
diff --git a/math/libm-test.inc b/math/libm-test.inc
index 7a91a2b..f7f8b57 100644
--- a/math/libm-test.inc
+++ b/math/libm-test.inc
@@ -9839,6 +9839,8 @@ main (int argc, char **argv)
initialize ();
printf (TEST_MSG);
+ INIT_ARCH_EXT
+
check_ulp ();
/* Keep the tests a wee bit ordered (according to ISO C99). */
diff --git a/math/test-double-vlen2.h b/math/test-double-vlen2.h
new file mode 100644
index 0000000..e896659
--- /dev/null
+++ b/math/test-double-vlen2.h
@@ -0,0 +1,42 @@
+/* Copyright (C) 2014 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+#define FLOAT double
+#define FUNC(function) function
+#define TEST_MSG "testing double vector math (without inline functions)\n"
+#define MATHCONST(x) x
+#define CHOOSE(Clongdouble,Cdouble,Cfloat,Cinlinelongdouble,Cinlinedouble,Cinlinefloat)
Cdouble
+#define PRINTF_EXPR "e"
+#define PRINTF_XEXPR "a"
+#define PRINTF_NEXPR "f"
+#define TEST_DOUBLE 1
+#define TEST_MATHVEC 1
+
+#ifndef __NO_MATH_INLINES
+# define __NO_MATH_INLINES
+#endif
+
+#define EXCEPTION_TESTS_double 0
+#define ROUNDING_TESTS_double(MODE) ((MODE) == FE_TONEAREST)
+
+#define VEC_SUFF _vlen2
+
+#define CONCAT(a,b) __CONCAT (a,b)
+
+#define WRAPPER_NAME(function) CONCAT (function,VEC_SUFF)
+
+#define FUNC_TEST(function) function ## _VEC_SUFF
diff --git a/math/test-double-vlen4.h b/math/test-double-vlen4.h
new file mode 100644
index 0000000..d7c8c24
--- /dev/null
+++ b/math/test-double-vlen4.h
@@ -0,0 +1,40 @@
+/* Copyright (C) 2014 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+#define FLOAT double
+#define FUNC(function) function
+#define TEST_MSG "testing double vector math (without inline functions)\n"
+#define MATHCONST(x) x
+#define CHOOSE(Clongdouble,Cdouble,Cfloat,Cinlinelongdouble,Cinlinedouble,Cinlinefloat)
Cdouble
+#define PRINTF_EXPR "e"
+#define PRINTF_XEXPR "a"
+#define PRINTF_NEXPR "f"
+#define TEST_DOUBLE 1
+#define TEST_MATHVEC 1
+
+#ifndef __NO_MATH_INLINES
+# define __NO_MATH_INLINES
+#endif
+
+#define EXCEPTION_TESTS_double 0
+#define ROUNDING_TESTS_double(MODE) ((MODE) == FE_TONEAREST)
+
+#define CONCAT(a,b) __CONCAT (a,b)
+
+#define WRAPPER_NAME(function) CONCAT (function,VEC_SUFF)
+
+#define FUNC_TEST(function) function ## _VEC_SUFF
diff --git a/sysdeps/x86_64/fpu/Makefile b/sysdeps/x86_64/fpu/Makefile
index 25fe0d4..e6add6d 100644
--- a/sysdeps/x86_64/fpu/Makefile
+++ b/sysdeps/x86_64/fpu/Makefile
@@ -2,3 +2,20 @@ ifeq ($(subdir),mathvec)
libmvec-support += svml_d_cos2_core svml_d_cos4_core_avx \
svml_d_cos4_core_avx2 svml_d_cos_data
endif
+
+# Rules for libmvec tests
+ifeq ($(subdir),math)
+ifeq ($(build-mathvec),yes)
+libmvec-tests += double-vlen2 double-vlen4 double-vlen4-avx2
+arch-ext-cflags = -mavx
+
+$(objpfx)test-double-vlen4-avx2.o: $(objpfx)libm-test.stmp
+
+$(objpfx)test-double-vlen4-avx2: $(common-objpfx)mathvec/libmvec.so \
+ $(objpfx)init-arch.o
+
+CFLAGS-test-double-vlen4-avx2.c = -fno-inline -ffloat-store
-fno-builtin -frounding-math \
+ -D__FAST_MATH__ -DTEST_FAST_MATH -D_OPENMP=201307 \
+ -Wno-unknown-pragmas -mavx2
+endif
+endif
diff --git a/sysdeps/x86_64/fpu/libm-test-ulps
b/sysdeps/x86_64/fpu/libm-test-ulps
index 36e1b76..0a0906e 100644
--- a/sysdeps/x86_64/fpu/libm-test-ulps
+++ b/sysdeps/x86_64/fpu/libm-test-ulps
@@ -905,6 +905,15 @@ idouble: 1
ildouble: 2
ldouble: 2
+Function: "cos_vlen2":
+double: 1
+
+Function: "cos_vlen4_avx":
+double: 1
+
+Function: "cos_vlen4_avx2":
+double: 1
+
Function: "cosh":
double: 1
float: 1
@@ -1961,6 +1970,15 @@ ifloat: 3
ildouble: 4
ldouble: 4
+Function: "vlen2_cos":
+double: 1
+
+Function: "vlen4_avx2_cos":
+double: 1
+
+Function: "vlen4_avx_cos":
+double: 1
+
Function: "y0":
double: 2
float: 1
diff --git a/sysdeps/x86_64/fpu/test-double-vlen2.c
b/sysdeps/x86_64/fpu/test-double-vlen2.c
new file mode 100644
index 0000000..674c5de
--- /dev/null
+++ b/sysdeps/x86_64/fpu/test-double-vlen2.c
@@ -0,0 +1,44 @@
+/* Tests for SSE4 ISA versions of vector math functions.
+ Copyright (C) 2014 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+#include "test-double-vlen2.h"
+#include <immintrin.h>
+
+// Wrapper from scalar to vector function implemented in SSE4.
+#define VECTOR_WRAPPER(scalar_func,vector_func) \
+extern __m128d vector_func ( __m128d);\
+FLOAT scalar_func (FLOAT x)\
+{\
+ int i;\
+ __m128d mx = _mm_set1_pd (x);\
+ __m128d mr = vector_func (mx);\
+ for (i=1;i<2;i++)\
+ {\
+ if (((FLOAT*)&mr)[0]!=((FLOAT*)&mr)[i])\
+ {\
+ return ((FLOAT*)&mr)[0]+0.1;\
+ }\
+ }\
+ return ((FLOAT*)&mr)[0];\
+}
+
+VECTOR_WRAPPER (WRAPPER_NAME (cos),_ZGVbN2v_cos)
+
+#define TEST_VECTOR_cos 1
+
+#include "libm-test.c"
diff --git a/sysdeps/x86_64/fpu/test-double-vlen4-avx2.c
b/sysdeps/x86_64/fpu/test-double-vlen4-avx2.c
new file mode 100644
index 0000000..17ef946
--- /dev/null
+++ b/sysdeps/x86_64/fpu/test-double-vlen4-avx2.c
@@ -0,0 +1,48 @@
+/* Tests for AVX2 ISA versions of vector math functions.
+ Copyright (C) 2014 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+#include "test-double-vlen4.h"
+#include <immintrin.h>
+
+// Wrapper from scalar to vector function implemented in AVX2.
+#define VECTOR_WRAPPER(scalar_func,vector_func) \
+extern __m256d vector_func (__m256d);\
+FLOAT scalar_func (FLOAT x)\
+{\
+ int i;\
+ __m256d mx = _mm256_set1_pd (x);\
+ __m256d mr = vector_func (mx);\
+ for (i=1;i<4;i++)\
+ {\
+ if (((FLOAT*)&mr)[0]!=((FLOAT*)&mr)[i])\
+ {\
+ return ((FLOAT*)&mr)[0]+0.1;\
+ }\
+ }\
+ return ((FLOAT*)&mr)[0];\
+}
+
+#define VEC_SUFF _vlen4_avx2
+
+VECTOR_WRAPPER (WRAPPER_NAME (cos),_ZGVdN4v_cos)
+
+#define TEST_VECTOR_cos 1
+
+#define REQUIRE_AVX2
+
+#include "libm-test.c"
diff --git a/sysdeps/x86_64/fpu/test-double-vlen4.c
b/sysdeps/x86_64/fpu/test-double-vlen4.c
new file mode 100644
index 0000000..8377ac8
--- /dev/null
+++ b/sysdeps/x86_64/fpu/test-double-vlen4.c
@@ -0,0 +1,46 @@
+/* Tests for AVX ISA versions of vector math functions.
+ Copyright (C) 2014 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+#include "test-double-vlen4.h"
+#include <immintrin.h>
+
+// Wrapper from scalar to vector function implemented in AVX.
+#define VECTOR_WRAPPER(scalar_func,vector_func) \
+extern __m256d vector_func (__m256d);\
+FLOAT scalar_func (FLOAT x)\
+{\
+ int i;\
+ __m256d mx = _mm256_set1_pd (x);\
+ __m256d mr = vector_func (mx);\
+ for (i=1;i<4;i++)\
+ {\
+ if (((FLOAT*)&mr)[0]!=((FLOAT*)&mr)[i])\
+ {\
+ return ((FLOAT*)&mr)[0]+0.1;\
+ }\
+ }\
+ return ((FLOAT*)&mr)[0];\
+}
+
+#define VEC_SUFF _vlen4_avx
+
+VECTOR_WRAPPER (WRAPPER_NAME (cos),_ZGVcN4v_cos)
+
+#define TEST_VECTOR_cos 1
+
+#include "libm-test.c"
--
WBR,
Andrew
[-- Attachment #2: libmvec_6_of_N_v3.patch --]
[-- Type: application/octet-stream, Size: 11966 bytes --]
diff --git a/math/Makefile b/math/Makefile
index 8a89308..a2240d0 100644
--- a/math/Makefile
+++ b/math/Makefile
@@ -97,8 +97,9 @@ tests-static = test-fpucw-static test-fpucw-ieee-static
test-longdouble-yes = test-ldouble test-ildoubl
ifneq (no,$(PERL))
+libm-vec-tests = $(addprefix test-,$(libmvec-tests))
libm-tests = test-float test-double $(test-longdouble-$(long-double-fcts)) \
- test-ifloat test-idouble
+ test-ifloat test-idouble $(libm-vec-tests)
libm-tests.o = $(addsuffix .o,$(libm-tests))
tests += $(libm-tests)
@@ -125,8 +126,22 @@ $(objpfx)test-double.o: $(objpfx)libm-test.stmp
$(objpfx)test-idouble.o: $(objpfx)libm-test.stmp
$(objpfx)test-ldouble.o: $(objpfx)libm-test.stmp
$(objpfx)test-ildoubl.o: $(objpfx)libm-test.stmp
+
+$(objpfx)test-double-vlen2.o: $(objpfx)libm-test.stmp
+$(objpfx)test-double-vlen4.o: $(objpfx)libm-test.stmp
+
+$(objpfx)test-double-vlen2: $(common-objpfx)mathvec/libmvec.so \
+ $(objpfx)init-arch.o
+$(objpfx)test-double-vlen4: $(common-objpfx)mathvec/libmvec.so \
+ $(objpfx)init-arch.o
endif
+CFLAGS-test-double-vlen2.c = -fno-inline -ffloat-store -fno-builtin -frounding-math \
+ -D__FAST_MATH__ -DTEST_FAST_MATH -D_OPENMP=201307 \
+ -Wno-unknown-pragmas
+CFLAGS-test-double-vlen4.c = -fno-inline -ffloat-store -fno-builtin -frounding-math \
+ -D__FAST_MATH__ -DTEST_FAST_MATH -D_OPENMP=201307 \
+ -Wno-unknown-pragmas $(arch-ext-cflags)
CFLAGS-test-float.c = -fno-inline -ffloat-store -fno-builtin -frounding-math
CFLAGS-test-double.c = -fno-inline -ffloat-store -fno-builtin -frounding-math
CFLAGS-test-ldouble.c = -fno-inline -ffloat-store -fno-builtin -frounding-math
diff --git a/math/libm-test.inc b/math/libm-test.inc
index 7a91a2b..f7f8b57 100644
--- a/math/libm-test.inc
+++ b/math/libm-test.inc
@@ -9839,6 +9839,8 @@ main (int argc, char **argv)
initialize ();
printf (TEST_MSG);
+ INIT_ARCH_EXT
+
check_ulp ();
/* Keep the tests a wee bit ordered (according to ISO C99). */
diff --git a/math/test-double-vlen2.h b/math/test-double-vlen2.h
new file mode 100644
index 0000000..e896659
--- /dev/null
+++ b/math/test-double-vlen2.h
@@ -0,0 +1,42 @@
+/* Copyright (C) 2014 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+#define FLOAT double
+#define FUNC(function) function
+#define TEST_MSG "testing double vector math (without inline functions)\n"
+#define MATHCONST(x) x
+#define CHOOSE(Clongdouble,Cdouble,Cfloat,Cinlinelongdouble,Cinlinedouble,Cinlinefloat) Cdouble
+#define PRINTF_EXPR "e"
+#define PRINTF_XEXPR "a"
+#define PRINTF_NEXPR "f"
+#define TEST_DOUBLE 1
+#define TEST_MATHVEC 1
+
+#ifndef __NO_MATH_INLINES
+# define __NO_MATH_INLINES
+#endif
+
+#define EXCEPTION_TESTS_double 0
+#define ROUNDING_TESTS_double(MODE) ((MODE) == FE_TONEAREST)
+
+#define VEC_SUFF _vlen2
+
+#define CONCAT(a,b) __CONCAT (a,b)
+
+#define WRAPPER_NAME(function) CONCAT (function,VEC_SUFF)
+
+#define FUNC_TEST(function) function ## _VEC_SUFF
diff --git a/math/test-double-vlen4.h b/math/test-double-vlen4.h
new file mode 100644
index 0000000..d7c8c24
--- /dev/null
+++ b/math/test-double-vlen4.h
@@ -0,0 +1,40 @@
+/* Copyright (C) 2014 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+#define FLOAT double
+#define FUNC(function) function
+#define TEST_MSG "testing double vector math (without inline functions)\n"
+#define MATHCONST(x) x
+#define CHOOSE(Clongdouble,Cdouble,Cfloat,Cinlinelongdouble,Cinlinedouble,Cinlinefloat) Cdouble
+#define PRINTF_EXPR "e"
+#define PRINTF_XEXPR "a"
+#define PRINTF_NEXPR "f"
+#define TEST_DOUBLE 1
+#define TEST_MATHVEC 1
+
+#ifndef __NO_MATH_INLINES
+# define __NO_MATH_INLINES
+#endif
+
+#define EXCEPTION_TESTS_double 0
+#define ROUNDING_TESTS_double(MODE) ((MODE) == FE_TONEAREST)
+
+#define CONCAT(a,b) __CONCAT (a,b)
+
+#define WRAPPER_NAME(function) CONCAT (function,VEC_SUFF)
+
+#define FUNC_TEST(function) function ## _VEC_SUFF
diff --git a/sysdeps/x86_64/fpu/Makefile b/sysdeps/x86_64/fpu/Makefile
index 25fe0d4..e6add6d 100644
--- a/sysdeps/x86_64/fpu/Makefile
+++ b/sysdeps/x86_64/fpu/Makefile
@@ -2,3 +2,20 @@ ifeq ($(subdir),mathvec)
libmvec-support += svml_d_cos2_core svml_d_cos4_core_avx \
svml_d_cos4_core_avx2 svml_d_cos_data
endif
+
+# Rules for libmvec tests
+ifeq ($(subdir),math)
+ifeq ($(build-mathvec),yes)
+libmvec-tests += double-vlen2 double-vlen4 double-vlen4-avx2
+arch-ext-cflags = -mavx
+
+$(objpfx)test-double-vlen4-avx2.o: $(objpfx)libm-test.stmp
+
+$(objpfx)test-double-vlen4-avx2: $(common-objpfx)mathvec/libmvec.so \
+ $(objpfx)init-arch.o
+
+CFLAGS-test-double-vlen4-avx2.c = -fno-inline -ffloat-store -fno-builtin -frounding-math \
+ -D__FAST_MATH__ -DTEST_FAST_MATH -D_OPENMP=201307 \
+ -Wno-unknown-pragmas -mavx2
+endif
+endif
diff --git a/sysdeps/x86_64/fpu/libm-test-ulps b/sysdeps/x86_64/fpu/libm-test-ulps
index 36e1b76..0a0906e 100644
--- a/sysdeps/x86_64/fpu/libm-test-ulps
+++ b/sysdeps/x86_64/fpu/libm-test-ulps
@@ -905,6 +905,15 @@ idouble: 1
ildouble: 2
ldouble: 2
+Function: "cos_vlen2":
+double: 1
+
+Function: "cos_vlen4_avx":
+double: 1
+
+Function: "cos_vlen4_avx2":
+double: 1
+
Function: "cosh":
double: 1
float: 1
@@ -1961,6 +1970,15 @@ ifloat: 3
ildouble: 4
ldouble: 4
+Function: "vlen2_cos":
+double: 1
+
+Function: "vlen4_avx2_cos":
+double: 1
+
+Function: "vlen4_avx_cos":
+double: 1
+
Function: "y0":
double: 2
float: 1
diff --git a/sysdeps/x86_64/fpu/test-double-vlen2.c b/sysdeps/x86_64/fpu/test-double-vlen2.c
new file mode 100644
index 0000000..674c5de
--- /dev/null
+++ b/sysdeps/x86_64/fpu/test-double-vlen2.c
@@ -0,0 +1,44 @@
+/* Tests for SSE4 ISA versions of vector math functions.
+ Copyright (C) 2014 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+#include "test-double-vlen2.h"
+#include <immintrin.h>
+
+// Wrapper from scalar to vector function implemented in SSE4.
+#define VECTOR_WRAPPER(scalar_func,vector_func) \
+extern __m128d vector_func ( __m128d);\
+FLOAT scalar_func (FLOAT x)\
+{\
+ int i;\
+ __m128d mx = _mm_set1_pd (x);\
+ __m128d mr = vector_func (mx);\
+ for (i=1;i<2;i++)\
+ {\
+ if (((FLOAT*)&mr)[0]!=((FLOAT*)&mr)[i])\
+ {\
+ return ((FLOAT*)&mr)[0]+0.1;\
+ }\
+ }\
+ return ((FLOAT*)&mr)[0];\
+}
+
+VECTOR_WRAPPER (WRAPPER_NAME (cos),_ZGVbN2v_cos)
+
+#define TEST_VECTOR_cos 1
+
+#include "libm-test.c"
diff --git a/sysdeps/x86_64/fpu/test-double-vlen4-avx2.c b/sysdeps/x86_64/fpu/test-double-vlen4-avx2.c
new file mode 100644
index 0000000..17ef946
--- /dev/null
+++ b/sysdeps/x86_64/fpu/test-double-vlen4-avx2.c
@@ -0,0 +1,48 @@
+/* Tests for AVX2 ISA versions of vector math functions.
+ Copyright (C) 2014 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+#include "test-double-vlen4.h"
+#include <immintrin.h>
+
+// Wrapper from scalar to vector function implemented in AVX2.
+#define VECTOR_WRAPPER(scalar_func,vector_func) \
+extern __m256d vector_func (__m256d);\
+FLOAT scalar_func (FLOAT x)\
+{\
+ int i;\
+ __m256d mx = _mm256_set1_pd (x);\
+ __m256d mr = vector_func (mx);\
+ for (i=1;i<4;i++)\
+ {\
+ if (((FLOAT*)&mr)[0]!=((FLOAT*)&mr)[i])\
+ {\
+ return ((FLOAT*)&mr)[0]+0.1;\
+ }\
+ }\
+ return ((FLOAT*)&mr)[0];\
+}
+
+#define VEC_SUFF _vlen4_avx2
+
+VECTOR_WRAPPER (WRAPPER_NAME (cos),_ZGVdN4v_cos)
+
+#define TEST_VECTOR_cos 1
+
+#define REQUIRE_AVX2
+
+#include "libm-test.c"
diff --git a/sysdeps/x86_64/fpu/test-double-vlen4.c b/sysdeps/x86_64/fpu/test-double-vlen4.c
new file mode 100644
index 0000000..8377ac8
--- /dev/null
+++ b/sysdeps/x86_64/fpu/test-double-vlen4.c
@@ -0,0 +1,46 @@
+/* Tests for AVX ISA versions of vector math functions.
+ Copyright (C) 2014 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+#include "test-double-vlen4.h"
+#include <immintrin.h>
+
+// Wrapper from scalar to vector function implemented in AVX.
+#define VECTOR_WRAPPER(scalar_func,vector_func) \
+extern __m256d vector_func (__m256d);\
+FLOAT scalar_func (FLOAT x)\
+{\
+ int i;\
+ __m256d mx = _mm256_set1_pd (x);\
+ __m256d mr = vector_func (mx);\
+ for (i=1;i<4;i++)\
+ {\
+ if (((FLOAT*)&mr)[0]!=((FLOAT*)&mr)[i])\
+ {\
+ return ((FLOAT*)&mr)[0]+0.1;\
+ }\
+ }\
+ return ((FLOAT*)&mr)[0];\
+}
+
+#define VEC_SUFF _vlen4_avx
+
+VECTOR_WRAPPER (WRAPPER_NAME (cos),_ZGVcN4v_cos)
+
+#define TEST_VECTOR_cos 1
+
+#include "libm-test.c"
^ permalink raw reply [flat|nested] 5+ messages in thread
* [PATCH 6/N v2] [x86_64] Vectorized math functions
@ 2014-11-24 19:00 Andrew Senkevich
2014-11-25 17:39 ` Andrew Senkevich
2014-11-26 17:56 ` Joseph Myers
0 siblings, 2 replies; 5+ messages in thread
From: Andrew Senkevich @ 2014-11-24 19:00 UTC (permalink / raw)
To: libc-alpha
[-- Attachment #1: Type: text/plain, Size: 12428 bytes --]
This modified according updates in 1st, 2nd and 3rd patches version of
6th patch (sent in
https://sourceware.org/ml/libc-alpha/2014-11/msg00603.html).
Patches 4 and 5 are unchanged.
It is addition of tests for vectorized cos, which are
implemented as wrappers to vector versions from scalar ones.
ChangeLog
2014-11-24 Andrew Senkevich <andrew.n.senkevich@gmail.com>
* math/Makefile: Added rules for libmvec tests.
* sysdeps/x86_64/fpu/Makefile: Likewise.
* sysdeps/x86_64/fpu/libm-test-ulps: Regenerated.
* math/test-double-vlen2.h: New file.
* math/test-double-vlen4.h: New file.
* sysdeps/x86_64/fpu/test-double-vlen2.c: New file.
* sysdeps/x86_64/fpu/test-double-vlen4.c: New file.
* sysdeps/x86_64/fpu/test-double-vlen4-avx2.c: New file.
diff --git a/math/Makefile b/math/Makefile
index 8a89308..a2240d0 100644
--- a/math/Makefile
+++ b/math/Makefile
@@ -97,8 +97,9 @@ tests-static = test-fpucw-static test-fpucw-ieee-static
test-longdouble-yes = test-ldouble test-ildoubl
ifneq (no,$(PERL))
+libm-vec-tests = $(addprefix test-,$(libmvec-tests))
libm-tests = test-float test-double $(test-longdouble-$(long-double-fcts)) \
- test-ifloat test-idouble
+ test-ifloat test-idouble $(libm-vec-tests)
libm-tests.o = $(addsuffix .o,$(libm-tests))
tests += $(libm-tests)
@@ -125,8 +126,22 @@ $(objpfx)test-double.o: $(objpfx)libm-test.stmp
$(objpfx)test-idouble.o: $(objpfx)libm-test.stmp
$(objpfx)test-ldouble.o: $(objpfx)libm-test.stmp
$(objpfx)test-ildoubl.o: $(objpfx)libm-test.stmp
+
+$(objpfx)test-double-vlen2.o: $(objpfx)libm-test.stmp
+$(objpfx)test-double-vlen4.o: $(objpfx)libm-test.stmp
+
+$(objpfx)test-double-vlen2: $(common-objpfx)mathvec/libmvec.so \
+ $(objpfx)init-arch.o
+$(objpfx)test-double-vlen4: $(common-objpfx)mathvec/libmvec.so \
+ $(objpfx)init-arch.o
endif
+CFLAGS-test-double-vlen2.c = -fno-inline -ffloat-store -fno-builtin
-frounding-math \
+ -D__FAST_MATH__ -DTEST_FAST_MATH -D_OPENMP=201307 \
+ -Wno-unknown-pragmas
+CFLAGS-test-double-vlen4.c = -fno-inline -ffloat-store -fno-builtin
-frounding-math \
+ -D__FAST_MATH__ -DTEST_FAST_MATH -D_OPENMP=201307 \
+ -Wno-unknown-pragmas $(arch-ext-cflags)
CFLAGS-test-float.c = -fno-inline -ffloat-store -fno-builtin -frounding-math
CFLAGS-test-double.c = -fno-inline -ffloat-store -fno-builtin -frounding-math
CFLAGS-test-ldouble.c = -fno-inline -ffloat-store -fno-builtin -frounding-math
diff --git a/math/test-double-vlen2.h b/math/test-double-vlen2.h
new file mode 100644
index 0000000..6ce47eb
--- /dev/null
+++ b/math/test-double-vlen2.h
@@ -0,0 +1,42 @@
+/* Copyright (C) 2014 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+#define FLOAT double
+#define FUNC(function) function
+#define TEST_MSG "testing double vector math (without inline functions)\n"
+#define MATHCONST(x) x
+#define CHOOSE(Clongdouble,Cdouble,Cfloat,Cinlinelongdouble,Cinlinedouble,Cinlinefloat)
Cdouble
+#define PRINTF_EXPR "e"
+#define PRINTF_XEXPR "a"
+#define PRINTF_NEXPR "f"
+#define TEST_DOUBLE 1
+#define TEST_MATHVEC 1
+
+#ifndef __NO_MATH_INLINES
+# define __NO_MATH_INLINES
+#endif
+
+#define EXCEPTION_TESTS_double 0
+#define ROUNDING_TESTS_double(MODE) ((MODE) == FE_TONEAREST)
+
+#define VEC_SUFF _vlen2
+
+#define CONCAT(a,b) __CONCAT (a,b)
+
+#define WRAPPER_NAME(function) CONCAT (function,VEC_SUFF)
+
+#define FUNC_TEST(function) function ## _VEC_SUFF
diff --git a/math/test-double-vlen4.h b/math/test-double-vlen4.h
new file mode 100644
index 0000000..4e26f97
--- /dev/null
+++ b/math/test-double-vlen4.h
@@ -0,0 +1,40 @@
+/* Copyright (C) 2014 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+#define FLOAT double
+#define FUNC(function) function
+#define TEST_MSG "testing double vector math (without inline functions)\n"
+#define MATHCONST(x) x
+#define CHOOSE(Clongdouble,Cdouble,Cfloat,Cinlinelongdouble,Cinlinedouble,Cinlinefloat)
Cdouble
+#define PRINTF_EXPR "e"
+#define PRINTF_XEXPR "a"
+#define PRINTF_NEXPR "f"
+#define TEST_DOUBLE 1
+#define TEST_MATHVEC 1
+
+#ifndef __NO_MATH_INLINES
+# define __NO_MATH_INLINES
+#endif
+
+#define EXCEPTION_TESTS_double 0
+#define ROUNDING_TESTS_double(MODE) ((MODE) == FE_TONEAREST)
+
+#define CONCAT(a,b) __CONCAT (a,b)
+
+#define WRAPPER_NAME(function) CONCAT (function,VEC_SUFF)
+
+#define FUNC_TEST(function) function ## _VEC_SUFF
diff --git a/sysdeps/x86_64/fpu/Makefile b/sysdeps/x86_64/fpu/Makefile
index 25fe0d4..e6add6d 100644
--- a/sysdeps/x86_64/fpu/Makefile
+++ b/sysdeps/x86_64/fpu/Makefile
@@ -2,3 +2,20 @@ ifeq ($(subdir),mathvec)
libmvec-support += svml_d_cos2_core svml_d_cos4_core_avx \
svml_d_cos4_core_avx2 svml_d_cos_data
endif
+
+# Rules for libmvec tests
+ifeq ($(subdir),math)
+ifeq ($(build-mathvec),yes)
+libmvec-tests += double-vlen2 double-vlen4 double-vlen4-avx2
+arch-ext-cflags = -mavx
+
+$(objpfx)test-double-vlen4-avx2.o: $(objpfx)libm-test.stmp
+
+$(objpfx)test-double-vlen4-avx2: $(common-objpfx)mathvec/libmvec.so \
+ $(objpfx)init-arch.o
+
+CFLAGS-test-double-vlen4-avx2.c = -fno-inline -ffloat-store
-fno-builtin -frounding-math \
+ -D__FAST_MATH__ -DTEST_FAST_MATH -D_OPENMP=201307 \
+ -Wno-unknown-pragmas -mavx2
+endif
+endif
diff --git a/sysdeps/x86_64/fpu/libm-test-ulps
b/sysdeps/x86_64/fpu/libm-test-ulps
index 36e1b76..0a0906e 100644
--- a/sysdeps/x86_64/fpu/libm-test-ulps
+++ b/sysdeps/x86_64/fpu/libm-test-ulps
@@ -905,6 +905,15 @@ idouble: 1
ildouble: 2
ldouble: 2
+Function: "cos_vlen2":
+double: 1
+
+Function: "cos_vlen4_avx":
+double: 1
+
+Function: "cos_vlen4_avx2":
+double: 1
+
Function: "cosh":
double: 1
float: 1
@@ -1961,6 +1970,15 @@ ifloat: 3
ildouble: 4
ldouble: 4
+Function: "vlen2_cos":
+double: 1
+
+Function: "vlen4_avx2_cos":
+double: 1
+
+Function: "vlen4_avx_cos":
+double: 1
+
Function: "y0":
double: 2
float: 1
diff --git a/sysdeps/x86_64/fpu/test-double-vlen2.c
b/sysdeps/x86_64/fpu/test-double-vlen2.c
new file mode 100644
index 0000000..674c5de
--- /dev/null
+++ b/sysdeps/x86_64/fpu/test-double-vlen2.c
@@ -0,0 +1,44 @@
+/* Tests for SSE4 ISA versions of vector math functions.
+ Copyright (C) 2014 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+#include "test-double-vlen2.h"
+#include <immintrin.h>
+
+// Wrapper from scalar to vector function implemented in SSE4.
+#define VECTOR_WRAPPER(scalar_func,vector_func) \
+extern __m128d vector_func ( __m128d);\
+FLOAT scalar_func (FLOAT x)\
+{\
+ int i;\
+ __m128d mx = _mm_set1_pd (x);\
+ __m128d mr = vector_func (mx);\
+ for (i=1;i<2;i++)\
+ {\
+ if (((FLOAT*)&mr)[0]!=((FLOAT*)&mr)[i])\
+ {\
+ return ((FLOAT*)&mr)[0]+0.1;\
+ }\
+ }\
+ return ((FLOAT*)&mr)[0];\
+}
+
+VECTOR_WRAPPER (WRAPPER_NAME (cos),_ZGVbN2v_cos)
+
+#define TEST_VECTOR_cos 1
+
+#include "libm-test.c"
diff --git a/sysdeps/x86_64/fpu/test-double-vlen4-avx2.c
b/sysdeps/x86_64/fpu/test-double-vlen4-avx2.c
new file mode 100644
index 0000000..15b7930
--- /dev/null
+++ b/sysdeps/x86_64/fpu/test-double-vlen4-avx2.c
@@ -0,0 +1,48 @@
+/* Tests for AVX2 ISA versions of vector math functions.
+ Copyright (C) 2014 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+#include "test-double-vlen4.h"
+#include <immintrin.h>
+
+// Wrapper from scalar to vector function implemented in AVX2.
+#define VECTOR_WRAPPER(scalar_func,vector_func) \
+extern __m256d vector_func (__m256d);\
+FLOAT scalar_func (FLOAT x)\
+{\
+ int i;\
+ __m256d mx = _mm256_set1_pd (x);\
+ __m256d mr = vector_func (mx);\
+ for (i=1;i<4;i++)\
+ {\
+ if (((FLOAT*)&mr)[0]!=((FLOAT*)&mr)[i])\
+ {\
+ return ((FLOAT*)&mr)[0]+0.1;\
+ }\
+ }\
+ return ((FLOAT*)&mr)[0];\
+}
+
+#define VEC_SUFF _vlen4_avx2
+
+VECTOR_WRAPPER (WRAPPER_NAME (cos),_ZGVdN4v_cos)
+
+#define TEST_VECTOR_cos 1
+
+#define REQUIRE_AVX2
+
+#include "libm-test.c"
diff --git a/sysdeps/x86_64/fpu/test-double-vlen4.c
b/sysdeps/x86_64/fpu/test-double-vlen4.c
new file mode 100644
index 0000000..5f68af5
--- /dev/null
+++ b/sysdeps/x86_64/fpu/test-double-vlen4.c
@@ -0,0 +1,46 @@
+/* Tests for AVX ISA versions of vector math functions.
+ Copyright (C) 2014 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+#include "test-double-vlen4.h"
+#include <immintrin.h>
+
+// Wrapper from scalar to vector function implemented in AVX.
+#define VECTOR_WRAPPER(scalar_func,vector_func) \
+extern __m256d vector_func (__m256d);\
+FLOAT scalar_func (FLOAT x)\
+{\
+ int i;\
+ __m256d mx = _mm256_set1_pd (x);\
+ __m256d mr = vector_func (mx);\
+ for (i=1;i<4;i++)\
+ {\
+ if (((FLOAT*)&mr)[0]!=((FLOAT*)&mr)[i])\
+ {\
+ return ((FLOAT*)&mr)[0]+0.1;\
+ }\
+ }\
+ return ((FLOAT*)&mr)[0];\
+}
+
+#define VEC_SUFF _vlen4_avx
+
+VECTOR_WRAPPER (WRAPPER_NAME (cos),_ZGVcN4v_cos)
+
+#define TEST_VECTOR_cos 1
+
+#include "libm-test.c"
--
WBR,
Andrew
[-- Attachment #2: libmvec_6_of_N_v2.patch --]
[-- Type: application/octet-stream, Size: 11633 bytes --]
diff --git a/math/Makefile b/math/Makefile
index 8a89308..a2240d0 100644
--- a/math/Makefile
+++ b/math/Makefile
@@ -97,8 +97,9 @@ tests-static = test-fpucw-static test-fpucw-ieee-static
test-longdouble-yes = test-ldouble test-ildoubl
ifneq (no,$(PERL))
+libm-vec-tests = $(addprefix test-,$(libmvec-tests))
libm-tests = test-float test-double $(test-longdouble-$(long-double-fcts)) \
- test-ifloat test-idouble
+ test-ifloat test-idouble $(libm-vec-tests)
libm-tests.o = $(addsuffix .o,$(libm-tests))
tests += $(libm-tests)
@@ -125,8 +126,22 @@ $(objpfx)test-double.o: $(objpfx)libm-test.stmp
$(objpfx)test-idouble.o: $(objpfx)libm-test.stmp
$(objpfx)test-ldouble.o: $(objpfx)libm-test.stmp
$(objpfx)test-ildoubl.o: $(objpfx)libm-test.stmp
+
+$(objpfx)test-double-vlen2.o: $(objpfx)libm-test.stmp
+$(objpfx)test-double-vlen4.o: $(objpfx)libm-test.stmp
+
+$(objpfx)test-double-vlen2: $(common-objpfx)mathvec/libmvec.so \
+ $(objpfx)init-arch.o
+$(objpfx)test-double-vlen4: $(common-objpfx)mathvec/libmvec.so \
+ $(objpfx)init-arch.o
endif
+CFLAGS-test-double-vlen2.c = -fno-inline -ffloat-store -fno-builtin -frounding-math \
+ -D__FAST_MATH__ -DTEST_FAST_MATH -D_OPENMP=201307 \
+ -Wno-unknown-pragmas
+CFLAGS-test-double-vlen4.c = -fno-inline -ffloat-store -fno-builtin -frounding-math \
+ -D__FAST_MATH__ -DTEST_FAST_MATH -D_OPENMP=201307 \
+ -Wno-unknown-pragmas $(arch-ext-cflags)
CFLAGS-test-float.c = -fno-inline -ffloat-store -fno-builtin -frounding-math
CFLAGS-test-double.c = -fno-inline -ffloat-store -fno-builtin -frounding-math
CFLAGS-test-ldouble.c = -fno-inline -ffloat-store -fno-builtin -frounding-math
diff --git a/math/test-double-vlen2.h b/math/test-double-vlen2.h
new file mode 100644
index 0000000..6ce47eb
--- /dev/null
+++ b/math/test-double-vlen2.h
@@ -0,0 +1,42 @@
+/* Copyright (C) 2014 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+#define FLOAT double
+#define FUNC(function) function
+#define TEST_MSG "testing double vector math (without inline functions)\n"
+#define MATHCONST(x) x
+#define CHOOSE(Clongdouble,Cdouble,Cfloat,Cinlinelongdouble,Cinlinedouble,Cinlinefloat) Cdouble
+#define PRINTF_EXPR "e"
+#define PRINTF_XEXPR "a"
+#define PRINTF_NEXPR "f"
+#define TEST_DOUBLE 1
+#define TEST_MATHVEC 1
+
+#ifndef __NO_MATH_INLINES
+# define __NO_MATH_INLINES
+#endif
+
+#define EXCEPTION_TESTS_double 0
+#define ROUNDING_TESTS_double(MODE) ((MODE) == FE_TONEAREST)
+
+#define VEC_SUFF _vlen2
+
+#define CONCAT(a,b) __CONCAT (a,b)
+
+#define WRAPPER_NAME(function) CONCAT (function,VEC_SUFF)
+
+#define FUNC_TEST(function) function ## _VEC_SUFF
diff --git a/math/test-double-vlen4.h b/math/test-double-vlen4.h
new file mode 100644
index 0000000..4e26f97
--- /dev/null
+++ b/math/test-double-vlen4.h
@@ -0,0 +1,40 @@
+/* Copyright (C) 2014 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+#define FLOAT double
+#define FUNC(function) function
+#define TEST_MSG "testing double vector math (without inline functions)\n"
+#define MATHCONST(x) x
+#define CHOOSE(Clongdouble,Cdouble,Cfloat,Cinlinelongdouble,Cinlinedouble,Cinlinefloat) Cdouble
+#define PRINTF_EXPR "e"
+#define PRINTF_XEXPR "a"
+#define PRINTF_NEXPR "f"
+#define TEST_DOUBLE 1
+#define TEST_MATHVEC 1
+
+#ifndef __NO_MATH_INLINES
+# define __NO_MATH_INLINES
+#endif
+
+#define EXCEPTION_TESTS_double 0
+#define ROUNDING_TESTS_double(MODE) ((MODE) == FE_TONEAREST)
+
+#define CONCAT(a,b) __CONCAT (a,b)
+
+#define WRAPPER_NAME(function) CONCAT (function,VEC_SUFF)
+
+#define FUNC_TEST(function) function ## _VEC_SUFF
diff --git a/sysdeps/x86_64/fpu/Makefile b/sysdeps/x86_64/fpu/Makefile
index 25fe0d4..e6add6d 100644
--- a/sysdeps/x86_64/fpu/Makefile
+++ b/sysdeps/x86_64/fpu/Makefile
@@ -2,3 +2,20 @@ ifeq ($(subdir),mathvec)
libmvec-support += svml_d_cos2_core svml_d_cos4_core_avx \
svml_d_cos4_core_avx2 svml_d_cos_data
endif
+
+# Rules for libmvec tests
+ifeq ($(subdir),math)
+ifeq ($(build-mathvec),yes)
+libmvec-tests += double-vlen2 double-vlen4 double-vlen4-avx2
+arch-ext-cflags = -mavx
+
+$(objpfx)test-double-vlen4-avx2.o: $(objpfx)libm-test.stmp
+
+$(objpfx)test-double-vlen4-avx2: $(common-objpfx)mathvec/libmvec.so \
+ $(objpfx)init-arch.o
+
+CFLAGS-test-double-vlen4-avx2.c = -fno-inline -ffloat-store -fno-builtin -frounding-math \
+ -D__FAST_MATH__ -DTEST_FAST_MATH -D_OPENMP=201307 \
+ -Wno-unknown-pragmas -mavx2
+endif
+endif
diff --git a/sysdeps/x86_64/fpu/libm-test-ulps b/sysdeps/x86_64/fpu/libm-test-ulps
index 36e1b76..0a0906e 100644
--- a/sysdeps/x86_64/fpu/libm-test-ulps
+++ b/sysdeps/x86_64/fpu/libm-test-ulps
@@ -905,6 +905,15 @@ idouble: 1
ildouble: 2
ldouble: 2
+Function: "cos_vlen2":
+double: 1
+
+Function: "cos_vlen4_avx":
+double: 1
+
+Function: "cos_vlen4_avx2":
+double: 1
+
Function: "cosh":
double: 1
float: 1
@@ -1961,6 +1970,15 @@ ifloat: 3
ildouble: 4
ldouble: 4
+Function: "vlen2_cos":
+double: 1
+
+Function: "vlen4_avx2_cos":
+double: 1
+
+Function: "vlen4_avx_cos":
+double: 1
+
Function: "y0":
double: 2
float: 1
diff --git a/sysdeps/x86_64/fpu/test-double-vlen2.c b/sysdeps/x86_64/fpu/test-double-vlen2.c
new file mode 100644
index 0000000..674c5de
--- /dev/null
+++ b/sysdeps/x86_64/fpu/test-double-vlen2.c
@@ -0,0 +1,44 @@
+/* Tests for SSE4 ISA versions of vector math functions.
+ Copyright (C) 2014 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+#include "test-double-vlen2.h"
+#include <immintrin.h>
+
+// Wrapper from scalar to vector function implemented in SSE4.
+#define VECTOR_WRAPPER(scalar_func,vector_func) \
+extern __m128d vector_func ( __m128d);\
+FLOAT scalar_func (FLOAT x)\
+{\
+ int i;\
+ __m128d mx = _mm_set1_pd (x);\
+ __m128d mr = vector_func (mx);\
+ for (i=1;i<2;i++)\
+ {\
+ if (((FLOAT*)&mr)[0]!=((FLOAT*)&mr)[i])\
+ {\
+ return ((FLOAT*)&mr)[0]+0.1;\
+ }\
+ }\
+ return ((FLOAT*)&mr)[0];\
+}
+
+VECTOR_WRAPPER (WRAPPER_NAME (cos),_ZGVbN2v_cos)
+
+#define TEST_VECTOR_cos 1
+
+#include "libm-test.c"
diff --git a/sysdeps/x86_64/fpu/test-double-vlen4-avx2.c b/sysdeps/x86_64/fpu/test-double-vlen4-avx2.c
new file mode 100644
index 0000000..15b7930
--- /dev/null
+++ b/sysdeps/x86_64/fpu/test-double-vlen4-avx2.c
@@ -0,0 +1,48 @@
+/* Tests for AVX2 ISA versions of vector math functions.
+ Copyright (C) 2014 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+#include "test-double-vlen4.h"
+#include <immintrin.h>
+
+// Wrapper from scalar to vector function implemented in AVX2.
+#define VECTOR_WRAPPER(scalar_func,vector_func) \
+extern __m256d vector_func (__m256d);\
+FLOAT scalar_func (FLOAT x)\
+{\
+ int i;\
+ __m256d mx = _mm256_set1_pd (x);\
+ __m256d mr = vector_func (mx);\
+ for (i=1;i<4;i++)\
+ {\
+ if (((FLOAT*)&mr)[0]!=((FLOAT*)&mr)[i])\
+ {\
+ return ((FLOAT*)&mr)[0]+0.1;\
+ }\
+ }\
+ return ((FLOAT*)&mr)[0];\
+}
+
+#define VEC_SUFF _vlen4_avx2
+
+VECTOR_WRAPPER (WRAPPER_NAME (cos),_ZGVdN4v_cos)
+
+#define TEST_VECTOR_cos 1
+
+#define REQUIRE_AVX2
+
+#include "libm-test.c"
diff --git a/sysdeps/x86_64/fpu/test-double-vlen4.c b/sysdeps/x86_64/fpu/test-double-vlen4.c
new file mode 100644
index 0000000..5f68af5
--- /dev/null
+++ b/sysdeps/x86_64/fpu/test-double-vlen4.c
@@ -0,0 +1,46 @@
+/* Tests for AVX ISA versions of vector math functions.
+ Copyright (C) 2014 Free Software Foundation, Inc.
+ This file is part of the GNU C Library.
+
+ The GNU C Library is free software; you can redistribute it and/or
+ modify it under the terms of the GNU Lesser General Public
+ License as published by the Free Software Foundation; either
+ version 2.1 of the License, or (at your option) any later version.
+
+ The GNU C Library is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ Lesser General Public License for more details.
+
+ You should have received a copy of the GNU Lesser General Public
+ License along with the GNU C Library; if not, see
+ <http://www.gnu.org/licenses/>. */
+
+#include "test-double-vlen4.h"
+#include <immintrin.h>
+
+// Wrapper from scalar to vector function implemented in AVX.
+#define VECTOR_WRAPPER(scalar_func,vector_func) \
+extern __m256d vector_func (__m256d);\
+FLOAT scalar_func (FLOAT x)\
+{\
+ int i;\
+ __m256d mx = _mm256_set1_pd (x);\
+ __m256d mr = vector_func (mx);\
+ for (i=1;i<4;i++)\
+ {\
+ if (((FLOAT*)&mr)[0]!=((FLOAT*)&mr)[i])\
+ {\
+ return ((FLOAT*)&mr)[0]+0.1;\
+ }\
+ }\
+ return ((FLOAT*)&mr)[0];\
+}
+
+#define VEC_SUFF _vlen4_avx
+
+VECTOR_WRAPPER (WRAPPER_NAME (cos),_ZGVcN4v_cos)
+
+#define TEST_VECTOR_cos 1
+
+#include "libm-test.c"
^ permalink raw reply [flat|nested] 5+ messages in thread
end of thread, other threads:[~2014-12-03 18:05 UTC | newest]
Thread overview: 5+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2014-12-02 16:50 [PATCH 6/N v2] [x86_64] Vectorized math functions Andrew Senkevich
2014-12-03 18:05 ` Joseph Myers
-- strict thread matches above, loose matches on Subject: below --
2014-11-24 19:00 Andrew Senkevich
2014-11-25 17:39 ` Andrew Senkevich
2014-11-26 17:56 ` Joseph Myers
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).