From c9a8c526acd185176e486bee4624039740f8c435 Mon Sep 17 00:00:00 2001 From: Andrew Senkevich Date: Thu, 18 Jun 2015 17:55:55 +0300 Subject: Vector sincos for x86_64 and tests. Here is implementation of vectorized sincos containing SSE, AVX, AVX2 and AVX512 versions according to Vector ABI . * NEWS: Mention addition of x86_64 vector sincos. * bits/libm-simd-decl-stubs.h: Added stubs for sincos. * math/math.h (__MATHDECL_VEC): New macro. * math/bits/mathcalls.h: Added sincos declaration with __MATHDECL_VEC. * math/gen-libm-have-vector-test.sh: Added generation of sincos wrapper declaration under condition. * math/test-vec-loop.h (TEST_VEC_LOOP): Refactored. * math/test-double-vlen2.h: Added wrapper for sincos tests, reflected TEST_VEC_LOOP change. * math/test-double-vlen4.h: Likewise. * math/test-double-vlen8.h: Likewise. * math/test-float-vlen16.h: Reflected TEST_VEC_LOOP change. * math/test-float-vlen4.h: Likewise. * math/test-float-vlen8.h: Likewise. * sysdeps/unix/sysv/linux/x86_64/libmvec.abilist: New symbols added. * sysdeps/x86/fpu/bits/math-vector.h: Added sincos SIMD declaration. * sysdeps/x86_64/fpu/Makefile (libmvec-support): Added new files. * sysdeps/x86_64/fpu/Versions: New versions added. * sysdeps/x86_64/fpu/libm-test-ulps: Regenerated. * sysdeps/x86_64/fpu/multiarch/Makefile (libmvec-sysdep_routines): Added build of SSE, AVX2 and AVX512 IFUNC versions. * sysdeps/x86_64/fpu/multiarch/svml_d_sincos2_core.S: New file. * sysdeps/x86_64/fpu/multiarch/svml_d_sincos2_core_sse4.S: New file. * sysdeps/x86_64/fpu/multiarch/svml_d_sincos4_core.S: New file. * sysdeps/x86_64/fpu/multiarch/svml_d_sincos4_core_avx2.S: New file. * sysdeps/x86_64/fpu/multiarch/svml_d_sincos8_core.S: New file. * sysdeps/x86_64/fpu/multiarch/svml_d_sincos8_core_avx512.S: New file. * sysdeps/x86_64/fpu/svml_d_sincos2_core.S: New file. * sysdeps/x86_64/fpu/svml_d_sincos4_core.S: New file. * sysdeps/x86_64/fpu/svml_d_sincos4_core_avx.S: New file. * sysdeps/x86_64/fpu/svml_d_sincos8_core.S: New file. * sysdeps/x86_64/fpu/svml_d_sincos_data.S: New file. * sysdeps/x86_64/fpu/svml_d_sincos_data.h: New file. * sysdeps/x86_64/fpu/svml_d_wrapper_impl.h: Added wrappers for sincos. * sysdeps/x86_64/fpu/test-double-vlen2-wrappers.c: Vector sincos tests. * sysdeps/x86_64/fpu/test-double-vlen2.c: Likewise. * sysdeps/x86_64/fpu/test-double-vlen4-avx2-wrappers.c: Likewise. * sysdeps/x86_64/fpu/test-double-vlen4-avx2.c: Likewise. * sysdeps/x86_64/fpu/test-double-vlen4-wrappers.c: Likewise. * sysdeps/x86_64/fpu/test-double-vlen4.c: Likewise. * sysdeps/x86_64/fpu/test-double-vlen8-wrappers.c: Likewise. * sysdeps/x86_64/fpu/test-double-vlen8.c: Likewise. --- math/bits/mathcalls.h | 4 ++-- math/gen-libm-have-vector-test.sh | 8 +++++++- math/math.h | 4 ++++ math/test-double-vlen2.h | 23 +++++++++++++++++++++-- math/test-double-vlen4.h | 23 +++++++++++++++++++++-- math/test-double-vlen8.h | 23 +++++++++++++++++++++-- math/test-float-vlen16.h | 6 ++++-- math/test-float-vlen4.h | 6 ++++-- math/test-float-vlen8.h | 6 ++++-- math/test-vec-loop.h | 8 ++++---- 10 files changed, 92 insertions(+), 19 deletions(-) (limited to 'math') diff --git a/math/bits/mathcalls.h b/math/bits/mathcalls.h index f297aa7..a881b7d 100644 --- a/math/bits/mathcalls.h +++ b/math/bits/mathcalls.h @@ -78,8 +78,8 @@ _Mdouble_END_NAMESPACE #ifdef __USE_GNU /* Cosine and sine of X. */ -__MATHDECL (void,sincos,, - (_Mdouble_ __x, _Mdouble_ *__sinx, _Mdouble_ *__cosx)); +__MATHDECL_VEC (void,sincos,, + (_Mdouble_ __x, _Mdouble_ *__sinx, _Mdouble_ *__cosx)); #endif #if defined __USE_XOPEN_EXTENDED || defined __USE_ISOC99 diff --git a/math/gen-libm-have-vector-test.sh b/math/gen-libm-have-vector-test.sh index 8c36078..529ea0c 100644 --- a/math/gen-libm-have-vector-test.sh +++ b/math/gen-libm-have-vector-test.sh @@ -33,7 +33,7 @@ print_defs() echo } -for func in $(cat libm-test.inc | grep ALL_RM_TEST | grep -v define | grep -v RUN_TEST_LOOP_ff_f | sed -r "s/.*\(//; s/,.*//" ); do +for func in $(cat libm-test.inc | grep ALL_RM_TEST | grep -v define | grep -v RUN_TEST_LOOP_ff_f | grep -v RUN_TEST_LOOP_fFF_11 | sed -r "s/.*\(//; s/,.*//" ); do print_defs ${func} print_defs ${func}f print_defs ${func}l @@ -45,6 +45,12 @@ for func in $(cat libm-test.inc | grep ALL_RM_TEST | grep RUN_TEST_LOOP_ff_f | s print_defs ${func}l "_ff" done +for func in $(cat libm-test.inc | grep ALL_RM_TEST | grep RUN_TEST_LOOP_fFF_11 | sed -r "s/.*\(//; s/,.*//" ); do + print_defs ${func} "_fFF" + print_defs ${func}f "_fFF" + print_defs ${func}l "_fFF" +done + # When all functions will use ALL_RM_TEST instead of using START directly, # this code can be removed. for func in $(grep 'START.*;$' libm-test.inc | sed -r "s/.*\(//; s/,.*//"); do diff --git a/math/math.h b/math/math.h index 7e959fc..c5115d7 100644 --- a/math/math.h +++ b/math/math.h @@ -58,6 +58,10 @@ __BEGIN_DECLS __SIMD_DECL (__MATH_PRECNAME (function, suffix)) \ __MATHCALL (function, suffix, args) +#define __MATHDECL_VEC(type, function,suffix, args) \ + __SIMD_DECL (__MATH_PRECNAME (function, suffix)) \ + __MATHDECL(type, function,suffix, args) + #define __MATHCALL(function,suffix, args) \ __MATHDECL (_Mdouble_,function,suffix, args) #define __MATHDECL(type, function,suffix, args) \ diff --git a/math/test-double-vlen2.h b/math/test-double-vlen2.h index 2e8415b..3f117e5 100644 --- a/math/test-double-vlen2.h +++ b/math/test-double-vlen2.h @@ -44,6 +44,7 @@ #define WRAPPER_DECL(function) extern FLOAT function (FLOAT); #define WRAPPER_DECL_ff(function) extern FLOAT function (FLOAT, FLOAT); +#define WRAPPER_DECL_fFF(function) extern void function (FLOAT, FLOAT *, FLOAT *); // Wrapper from scalar to vector function with vector length 2. #define VECTOR_WRAPPER(scalar_func, vector_func) \ @@ -54,7 +55,8 @@ FLOAT scalar_func (FLOAT x) \ VEC_TYPE mx; \ INIT_VEC_LOOP (mx, x, 2); \ VEC_TYPE mr = vector_func (mx); \ - TEST_VEC_LOOP (2); \ + TEST_VEC_LOOP (mr, 2); \ + return ((FLOAT) mr[0]); \ } // Wrapper from scalar 2 argument function to vector one. @@ -67,5 +69,22 @@ FLOAT scalar_func (FLOAT x, FLOAT y) \ INIT_VEC_LOOP (mx, x, 2); \ INIT_VEC_LOOP (my, y, 2); \ VEC_TYPE mr = vector_func (mx, my); \ - TEST_VEC_LOOP (2); \ + TEST_VEC_LOOP (mr, 2); \ + return ((FLOAT) mr[0]); \ +} + +// Wrapper from scalar 3 argument function to vector one. +#define VECTOR_WRAPPER_fFF(scalar_func, vector_func) \ +extern void vector_func (VEC_TYPE, VEC_TYPE *, VEC_TYPE *); \ +void scalar_func (FLOAT x, FLOAT * r, FLOAT * r1) \ +{ \ + int i; \ + VEC_TYPE mx, mr, mr1; \ + INIT_VEC_LOOP (mx, x, 2); \ + vector_func (mx, &mr, &mr1); \ + TEST_VEC_LOOP (mr, 2); \ + TEST_VEC_LOOP (mr1, 2); \ + *r = (FLOAT) mr[0]; \ + *r1 = (FLOAT) mr1[0]; \ + return; \ } diff --git a/math/test-double-vlen4.h b/math/test-double-vlen4.h index 4fca9d8..ffaba69 100644 --- a/math/test-double-vlen4.h +++ b/math/test-double-vlen4.h @@ -44,6 +44,7 @@ #define WRAPPER_DECL(function) extern FLOAT function (FLOAT); #define WRAPPER_DECL_ff(function) extern FLOAT function (FLOAT, FLOAT); +#define WRAPPER_DECL_fFF(function) extern void function (FLOAT, FLOAT *, FLOAT *); // Wrapper from scalar to vector function with vector length 4. #define VECTOR_WRAPPER(scalar_func, vector_func) \ @@ -54,7 +55,8 @@ FLOAT scalar_func (FLOAT x) \ VEC_TYPE mx; \ INIT_VEC_LOOP (mx, x, 4); \ VEC_TYPE mr = vector_func (mx); \ - TEST_VEC_LOOP (4); \ + TEST_VEC_LOOP (mr, 4); \ + return ((FLOAT) mr[0]); \ } // Wrapper from scalar 2 argument function to vector one. @@ -67,5 +69,22 @@ FLOAT scalar_func (FLOAT x, FLOAT y) \ INIT_VEC_LOOP (mx, x, 4); \ INIT_VEC_LOOP (my, y, 4); \ VEC_TYPE mr = vector_func (mx, my); \ - TEST_VEC_LOOP (4); \ + TEST_VEC_LOOP (mr, 4); \ + return ((FLOAT) mr[0]); \ +} + +// Wrapper from scalar 3 argument function to vector one. +#define VECTOR_WRAPPER_fFF(scalar_func, vector_func) \ +extern void vector_func (VEC_TYPE, VEC_TYPE *, VEC_TYPE *); \ +void scalar_func (FLOAT x, FLOAT * r, FLOAT * r1) \ +{ \ + int i; \ + VEC_TYPE mx, mr, mr1; \ + INIT_VEC_LOOP (mx, x, 4); \ + vector_func (mx, &mr, &mr1); \ + TEST_VEC_LOOP (mr, 4); \ + TEST_VEC_LOOP (mr1, 4); \ + *r = (FLOAT) mr[0]; \ + *r1 = (FLOAT) mr1[0]; \ + return; \ } diff --git a/math/test-double-vlen8.h b/math/test-double-vlen8.h index 6780e81..bdfccbb 100644 --- a/math/test-double-vlen8.h +++ b/math/test-double-vlen8.h @@ -44,6 +44,7 @@ #define WRAPPER_DECL(function) extern FLOAT function (FLOAT); #define WRAPPER_DECL_ff(function) extern FLOAT function (FLOAT, FLOAT); +#define WRAPPER_DECL_fFF(function) extern void function (FLOAT, FLOAT *, FLOAT *); // Wrapper from scalar to vector function with vector length 8. #define VECTOR_WRAPPER(scalar_func, vector_func) \ @@ -54,7 +55,8 @@ FLOAT scalar_func (FLOAT x) \ VEC_TYPE mx; \ INIT_VEC_LOOP (mx, x, 8); \ VEC_TYPE mr = vector_func (mx); \ - TEST_VEC_LOOP (8); \ + TEST_VEC_LOOP (mr, 8); \ + return ((FLOAT) mr[0]); \ } // Wrapper from scalar 2 argument function to vector one. @@ -67,5 +69,22 @@ FLOAT scalar_func (FLOAT x, FLOAT y) \ INIT_VEC_LOOP (mx, x, 8); \ INIT_VEC_LOOP (my, y, 8); \ VEC_TYPE mr = vector_func (mx, my); \ - TEST_VEC_LOOP (8); \ + TEST_VEC_LOOP (mr, 8); \ + return ((FLOAT) mr[0]); \ +} + +// Wrapper from scalar 3 argument function to vector one. +#define VECTOR_WRAPPER_fFF(scalar_func, vector_func) \ +extern void vector_func (VEC_TYPE, VEC_TYPE *, VEC_TYPE *); \ +void scalar_func (FLOAT x, FLOAT * r, FLOAT * r1) \ +{ \ + int i; \ + VEC_TYPE mx, mr, mr1; \ + INIT_VEC_LOOP (mx, x, 8); \ + vector_func (mx, &mr, &mr1); \ + TEST_VEC_LOOP (mr, 8); \ + TEST_VEC_LOOP (mr1, 8); \ + *r = (FLOAT) mr[0]; \ + *r1 = (FLOAT) mr1[0]; \ + return; \ } diff --git a/math/test-float-vlen16.h b/math/test-float-vlen16.h index 008e15e..802ae7b 100644 --- a/math/test-float-vlen16.h +++ b/math/test-float-vlen16.h @@ -54,7 +54,8 @@ FLOAT scalar_func (FLOAT x) \ VEC_TYPE mx; \ INIT_VEC_LOOP (mx, x, 16); \ VEC_TYPE mr = vector_func (mx); \ - TEST_VEC_LOOP (16); \ + TEST_VEC_LOOP (mr, 16); \ + return ((FLOAT) mr[0]); \ } // Wrapper from scalar 2 argument function to vector one. @@ -67,5 +68,6 @@ FLOAT scalar_func (FLOAT x, FLOAT y) \ INIT_VEC_LOOP (mx, x, 16); \ INIT_VEC_LOOP (my, y, 16); \ VEC_TYPE mr = vector_func (mx, my); \ - TEST_VEC_LOOP (16); \ + TEST_VEC_LOOP (mr, 16); \ + return ((FLOAT) mr[0]); \ } diff --git a/math/test-float-vlen4.h b/math/test-float-vlen4.h index eaf4b4b..f5e530b 100644 --- a/math/test-float-vlen4.h +++ b/math/test-float-vlen4.h @@ -54,7 +54,8 @@ FLOAT scalar_func (FLOAT x) \ VEC_TYPE mx; \ INIT_VEC_LOOP (mx, x, 4); \ VEC_TYPE mr = vector_func (mx); \ - TEST_VEC_LOOP (4); \ + TEST_VEC_LOOP (mr, 4); \ + return ((FLOAT) mr[0]); \ } // Wrapper from scalar 2 argument function to vector one. @@ -67,5 +68,6 @@ FLOAT scalar_func (FLOAT x, FLOAT y) \ INIT_VEC_LOOP (mx, x, 4); \ INIT_VEC_LOOP (my, y, 4); \ VEC_TYPE mr = vector_func (mx, my); \ - TEST_VEC_LOOP (4); \ + TEST_VEC_LOOP (mr, 4); \ + return ((FLOAT) mr[0]); \ } diff --git a/math/test-float-vlen8.h b/math/test-float-vlen8.h index 1a2eb52..697849f 100644 --- a/math/test-float-vlen8.h +++ b/math/test-float-vlen8.h @@ -54,7 +54,8 @@ FLOAT scalar_func (FLOAT x) \ VEC_TYPE mx; \ INIT_VEC_LOOP (mx, x, 8); \ VEC_TYPE mr = vector_func (mx); \ - TEST_VEC_LOOP (8); \ + TEST_VEC_LOOP (mr, 8); \ + return ((FLOAT) mr[0]); \ } // Wrapper from scalar 2 argument function to vector one. @@ -67,5 +68,6 @@ FLOAT scalar_func (FLOAT x, FLOAT y) \ INIT_VEC_LOOP (mx, x, 8); \ INIT_VEC_LOOP (my, y, 8); \ VEC_TYPE mr = vector_func (mx, my); \ - TEST_VEC_LOOP (8); \ + TEST_VEC_LOOP (mr, 8); \ + return ((FLOAT) mr[0]); \ } diff --git a/math/test-vec-loop.h b/math/test-vec-loop.h index 1a76c3e..66d7692 100644 --- a/math/test-vec-loop.h +++ b/math/test-vec-loop.h @@ -17,17 +17,17 @@ . */ /* This macro is used in VECTOR_WRAPPER macros for vector tests. */ -#define TEST_VEC_LOOP(len) \ +#define TEST_VEC_LOOP(vec, len) \ do \ { \ for (i = 1; i < len; i++) \ { \ - if ((FLOAT) mr[0] != (FLOAT) mr[i]) \ + if ((FLOAT) vec[0] != (FLOAT) vec[i]) \ { \ - return ((FLOAT) mr[0] + 0.1); \ + vec[0] = (FLOAT) vec[0] + 0.1; \ + break; \ } \ } \ - return ((FLOAT) mr[0]); \ } \ while (0) -- cgit v1.1