From: Ilya Tocar <tocarip.intel@gmail.com>
To: Uros Bizjak <ubizjak@gmail.com>
Cc: gcc-patches@gcc.gnu.org
Subject: Re: [PATCH, i386, testsuite] FMA intrinsics
Date: Mon, 22 Aug 2011 17:32:00 -0000 [thread overview]
Message-ID: <CAFFGCRB-auRYMdLGPzSdb9ftGd++vtgeWzOG2xARQwe4vij9iQ@mail.gmail.com> (raw)
In-Reply-To: <CAFULd4ao3DakdR+JdSW5kj_OhDD6Xd5DoEjH_LuKmvgp6JfnQA@mail.gmail.com>
[-- Attachment #1: Type: text/plain, Size: 5062 bytes --]
2011/8/20 Uros Bizjak <ubizjak@gmail.com>:
> Hello!
>
>> This patch adds intrinsics for FMA instruction set along with tests for them.
>> Bootstraps and passes make check (including make check on simulator
>> for new runtime tests).
>
> ? ? ? ? ? ? ? * config/i386/fmaintrin.h: New.
>
> It is not included in the patch.
Sorry about that
>
> ? ? ? ? ? ? ? * config.gcc: Add fmaintrin.h.
> ? ? ? ? ? ? ? * config/i386/i386.c
> ? ? ? ? ? ? ? * <ix86_builtins> (IX86_BUILTIN_VFMADDSS3): New.
> ? ? ? ? ? ? ? (IX86_BUILTIN_VFMADDSD3): Likewise.
> ? ? ? ? ? ? ? (X86_BUILTIN_VFNMADDSS3): Likewise.
> ? ? ? ? ? ? ? (X86_BUILTIN_VFNMADDSD3): Likewise.
> ? ? ? ? ? ? ? (X86_BUILTIN_VFMSUBSS3): Likewise.
> ? ? ? ? ? ? ? (X86_BUILTIN_VFMSUBSD3): Likewise.
> ? ? ? ? ? ? ? (X86_BUILTIN_VFNMSUBSS3): Likewise.
> ? ? ? ? ? ? ? (X86_BUILTIN_VFNMSUBSD3): Likewise.
> ? ? ? ? ? ? ? (X86_BUILTIN_VFMSUBPS): Likewise.
> ? ? ? ? ? ? ? (X86_BUILTIN_VFMSUBPD): Likewise.
> ? ? ? ? ? ? ? (X86_BUILTIN_VFMSUBPS256): Likewise.
> ? ? ? ? ? ? ? (X86_BUILTIN_VFMSUBPD256): Likewise.
> ? ? ? ? ? ? ? (X86_BUILTIN_VFNMADDPS): Likewise.
> ? ? ? ? ? ? ? (X86_BUILTIN_VFNMADDPD): Likewise.
> ? ? ? ? ? ? ? (X86_BUILTIN_VFNMADDPS256): Likewise.
> ? ? ? ? ? ? ? (X86_BUILTIN_VFNMADDPD256): Likewise.
> ? ? ? ? ? ? ? (X86_BUILTIN_VFNMSUBPS): Likewise.
> ? ? ? ? ? ? ? (X86_BUILTIN_VFNMSUBPD): Likewise.
> ? ? ? ? ? ? ? (X86_BUILTIN_VFNMSUBPS256): Likewise.
> ? ? ? ? ? ? ? (X86_BUILTIN_VFNMSUBPD256): Likewise.
> ? ? ? ? ? ? ? (X86_BUILTIN_VFMSUBADDPS): Likewise.
> ? ? ? ? ? ? ? (X86_BUILTIN_VFMSUBADDPD): Likewise.
> ? ? ? ? ? ? ? (X86_BUILTIN_VFMSUBADDPS256): Likewise.
> ? ? ? ? ? ? ? (X86_BUILTIN_VFMSUBADDPD256): Likewise.
>
> You don't need to add "negated" versions, one FMA builtin per mode is
> enough, please see existing FMA4 descriptions. Just put unary minus
> sign in the intrinsics header for "negated" operand and let GCC do its
> job. Please see existing FMA4 intrinsics header.
>
Actually i tried that.But in such case when i compile(FMA4 example)
#include <x86intrin.h>
extern __m128 a,b,c;
void foo(){
a = _mm_nmsub_ps(a,b,c);
}
with -S -O0 -mfma4
The asm have
vxorps %xmm1, %xmm0, %xmm0
vmovaps -16(%rbp), %xmm1
vmovaps .LC0(%rip), %xmm2
vxorps %xmm2, %xmm1, %xmm1
vfmaddps %xmm0, -32(%rbp), %xmm1, %xmm0
So vfmaddps of negated values is generated instead of vfnmsubps.
I think it is bad that intrinsic for instruction can generate code
without this instruction.
So to make sure that exact instruction is always generated i
introduced additional expands and builtins.
Is it wrong?
> ? ? ? ? ? ? ? * config/i386/sse.md (fmai_fnmadd_<mode>): New.
> ? ? ? ? ? ? ? (fmai_fmsub_<mode>): Likewise.
> ? ? ? ? ? ? ? (fmai_fnmsub_<mode>): Likewise.
> ? ? ? ? ? ? ? (fmai_fmadd_s_<mode>): Likewise.
> ? ? ? ? ? ? ? (fmai_vmfmadd_s_<mode>): Likewise.
> ? ? ? ? ? ? ? (fmai_vmfmsub_s_<mode>): Likewise.
> ? ? ? ? ? ? ? (fmai_vmfnmadd_s_<mode>): Likewise.
> ? ? ? ? ? ? ? (fmai_vmfnmsub_s_<mode>): Likewise.
> ? ? ? ? ? ? ? (*fmai_fmadd_s_<mode>): Likewise.
> ? ? ? ? ? ? ? (*fmai_fmsub_s_<mode>): Likewise.
> ? ? ? ? ? ? ? (*fmai_fnmadd_s_<mode>): Likewise.
> ? ? ? ? ? ? ? (*fmai_fnmsub_s_<mode>): Likewise.
> ? ? ? ? ? ? ? (fmsubadd_<mode>): Likewise.
>
> Also here. All your FMAMODE patterns should be expanded through
> existing "fma4i_fmadd_<mode>" expander (you can rename it to
> "fmai_fmadd..." to make its name more generic). This includes new
> "fmsubadd_<mode>" pattern that should be expanded through existing
> "fmaddsub_<mode>" expander.
>
See above explanation why i included new expands.
_s_ is removed
> vec_merge scalar versions also need only one expander, again follow
> existing FMA4 version. Also, there is no need to include "_s_" in the
> name. We know that these are scalar versions.
>
> ? ? ? ? ? ? ? * gcc.target/i386/fma-check.h: New.
> ? ? ? ? ? ? ? * gcc.target/i386/fma-256-fmaddXX.c: New testcase.
> ? ? ? ? ? ? ? * gcc.target/i386/fma-256-fmaddsubXX.c: Likewise.
> ? ? ? ? ? ? ? * gcc.target/i386/fma-256-fmsubXX.c: Likewise.
> ? ? ? ? ? ? ? * gcc.target/i386/fma-256-fmsubaddXX.c: Likewise.
> ? ? ? ? ? ? ? * gcc.target/i386/fma-256-fnmaddXX.c: Likewise.
> ? ? ? ? ? ? ? * gcc.target/i386/fma-256-fnmsubXX.c: Likewise.
> ? ? ? ? ? ? ? * gcc.target/i386/fma-fmaddXX.c: Likewise.
> ? ? ? ? ? ? ? * gcc.target/i386/fma-fmaddsubXX.c: Likewise.
> ? ? ? ? ? ? ? * gcc.target/i386/fma-fmsubXX.c: Likewise.
> ? ? ? ? ? ? ? * gcc.target/i386/fma-fmsubaddXX.c: Likewise.
> ? ? ? ? ? ? ? * gcc.target/i386/fma-fnmaddXX.c: Likewise.
> ? ? ? ? ? ? ? * gcc.target/i386/fma-fnmsubXX.c: Likewise.
> ? ? ? ? ? ? ? * gcc.target/i386/fma-compile.c: Likewise.
> ? ? ? ? ? ? ? * gcc.target/i386/i386.exp (check_effective_target_fma): New.
>
> Is there a reason that all runtime tests are compiled with -O0 except
> that there are some existing FMA tests in the testsuite using -O0?
> Usually, these kind of tests are compiled using -O2, so optimizations
> are applied also to the builtins.
Changed to O2.
>
> Uros.
>
[-- Attachment #2: patch --]
[-- Type: application/octet-stream, Size: 50042 bytes --]
diff --git a/gcc/config.gcc b/gcc/config.gcc
index ec13d93..3879a2a 100644
--- a/gcc/config.gcc
+++ b/gcc/config.gcc
@@ -352,7 +352,7 @@ i[34567]86-*-*)
nmmintrin.h bmmintrin.h fma4intrin.h wmmintrin.h
immintrin.h x86intrin.h avxintrin.h xopintrin.h
ia32intrin.h cross-stdarg.h lwpintrin.h popcntintrin.h
- lzcntintrin.h bmiintrin.h tbmintrin.h"
+ lzcntintrin.h bmiintrin.h tbmintrin.h fmaintrin.h"
;;
x86_64-*-*)
cpu_type=i386
@@ -364,7 +364,7 @@ x86_64-*-*)
nmmintrin.h bmmintrin.h fma4intrin.h wmmintrin.h
immintrin.h x86intrin.h avxintrin.h xopintrin.h
ia32intrin.h cross-stdarg.h lwpintrin.h popcntintrin.h
- lzcntintrin.h bmiintrin.h tbmintrin.h"
+ lzcntintrin.h bmiintrin.h tbmintrin.h fmaintrin.h"
need_64bit_hwint=yes
;;
ia64-*-*)
diff --git a/gcc/config/i386/fmaintrin.h b/gcc/config/i386/fmaintrin.h
new file mode 100644
index 0000000..b607114
--- /dev/null
+++ b/gcc/config/i386/fmaintrin.h
@@ -0,0 +1,233 @@
+/* Copyright (C) 2007, 2008, 2009, 2010, 2011 Free Software Foundation, Inc.
+
+ This file is part of GCC.
+
+ GCC is free software; you can redistribute it and/or modify
+ it under the terms of the GNU General Public License as published by
+ the Free Software Foundation; either version 3, or (at your option)
+ any later version.
+
+ GCC is distributed in the hope that it will be useful,
+ but WITHOUT ANY WARRANTY; without even the implied warranty of
+ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ GNU General Public License for more details.
+
+ Under Section 7 of GPL version 3, you are granted additional
+ permissions described in the GCC Runtime Library Exception, version
+ 3.1, as published by the Free Software Foundation.
+
+ You should have received a copy of the GNU General Public License and
+ a copy of the GCC Runtime Library Exception along with this program;
+ see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
+ <http://www.gnu.org/licenses/>. */
+
+#ifndef _X86INTRIN_H_INCLUDED
+# error "Never use <fmaintrin.h> directly; include <x86intrin.h> instead."
+#endif
+
+#ifndef _FMAINTRIN_H_INCLUDED
+#define _FMAINTRIN_H_INCLUDED
+
+#ifndef __FMA__
+# error "FMA instruction set not enabled"
+#else
+
+#include <immintrin.h>
+
+
+
+extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm_fmadd_pd (__m128d __A, __m128d __B, __m128d __C)
+{
+ return (__m128d)__builtin_ia32_vfmaddpd ((__v2df)__A, (__v2df)__B, (__v2df)__C);
+}
+
+extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm256_fmadd_pd (__m256d __A, __m256d __B, __m256d __C)
+{
+ return (__m256d)__builtin_ia32_vfmaddpd256 ((__v4df)__A, (__v4df)__B, (__v4df)__C);
+}
+
+extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm_fmadd_ps (__m128 __A, __m128 __B, __m128 __C)
+{
+ return (__m128)__builtin_ia32_vfmaddps ((__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
+}
+
+extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm256_fmadd_ps (__m256 __A, __m256 __B, __m256 __C)
+{
+ return (__m256)__builtin_ia32_vfmaddps256 ((__v8sf)__A, (__v8sf)__B, (__v8sf)__C);
+}
+
+extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm_fmadd_sd (__m128d __A, __m128d __B, __m128d __C)
+{
+ return (__m128d) __builtin_ia32_vfmaddsd3 ((__v2df)__A, (__v2df)__B, (__v2df)__C);
+}
+
+extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm_fmadd_ss (__m128 __A, __m128 __B, __m128 __C)
+{
+ return (__m128) __builtin_ia32_vfmaddss3 ((__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
+}
+
+extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm_fmsub_pd (__m128d __A, __m128d __B, __m128d __C)
+{
+ return (__m128d)__builtin_ia32_vfmsubpd ((__v2df)__A, (__v2df)__B, (__v2df)__C);
+}
+
+extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm256_fmsub_pd (__m256d __A, __m256d __B, __m256d __C)
+{
+ return (__m256d)__builtin_ia32_vfmsubpd256 ((__v4df)__A, (__v4df)__B, (__v4df)__C);
+}
+
+extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm_fmsub_ps (__m128 __A, __m128 __B, __m128 __C)
+{
+ return (__m128)__builtin_ia32_vfmsubps ((__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
+}
+
+extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm256_fmsub_ps (__m256 __A, __m256 __B, __m256 __C)
+{
+ return (__m256)__builtin_ia32_vfmsubps256 ((__v8sf)__A, (__v8sf)__B, (__v8sf)__C);
+}
+
+extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm_fmsub_sd (__m128d __A, __m128d __B, __m128d __C)
+{
+ return (__m128d)__builtin_ia32_vfmsubsd3 ((__v2df)__A, (__v2df)__B, (__v2df)__C);
+}
+
+extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm_fmsub_ss (__m128 __A, __m128 __B, __m128 __C)
+{
+ return (__m128)__builtin_ia32_vfmsubss3 ((__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
+}
+
+extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm_fnmadd_pd (__m128d __A, __m128d __B, __m128d __C)
+{
+ return (__m128d)__builtin_ia32_vfnmaddpd ((__v2df)__A, (__v2df)__B, (__v2df)__C);
+}
+
+extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm256_fnmadd_pd (__m256d __A, __m256d __B, __m256d __C)
+{
+ return (__m256d)__builtin_ia32_vfnmaddpd256 ((__v4df)__A, (__v4df)__B, (__v4df)__C);
+}
+
+extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm_fnmadd_ps (__m128 __A, __m128 __B, __m128 __C)
+{
+ return (__m128)__builtin_ia32_vfnmaddps ((__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
+}
+
+extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm256_fnmadd_ps (__m256 __A, __m256 __B, __m256 __C)
+{
+ return (__m256)__builtin_ia32_vfnmaddps256 ((__v8sf)__A, (__v8sf)__B, (__v8sf)__C);
+}
+
+extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm_fnmadd_sd (__m128d __A, __m128d __B, __m128d __C)
+{
+ return (__m128d)__builtin_ia32_vfnmaddsd3 ((__v2df)__A, (__v2df)__B, (__v2df)__C);
+}
+
+extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm_fnmadd_ss (__m128 __A, __m128 __B, __m128 __C)
+{
+ return (__m128)__builtin_ia32_vfnmaddss3 ((__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
+}
+
+extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm_fnmsub_pd (__m128d __A, __m128d __B, __m128d __C)
+{
+ return (__m128d)__builtin_ia32_vfnmsubpd ((__v2df)__A, (__v2df)__B, (__v2df)__C);
+}
+
+extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm256_fnmsub_pd (__m256d __A, __m256d __B, __m256d __C)
+{
+ return (__m256d)__builtin_ia32_vfnmsubpd256 ((__v4df)__A, (__v4df)__B, (__v4df)__C);
+}
+
+extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm_fnmsub_ps (__m128 __A, __m128 __B, __m128 __C)
+{
+ return (__m128)__builtin_ia32_vfnmsubps ((__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
+}
+
+extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm256_fnmsub_ps (__m256 __A, __m256 __B, __m256 __C)
+{
+ return (__m256)__builtin_ia32_vfnmsubps256 ((__v8sf)__A, (__v8sf)__B, (__v8sf)__C);
+}
+
+extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm_fnmsub_sd (__m128d __A, __m128d __B, __m128d __C)
+{
+ return (__m128d)__builtin_ia32_vfnmsubsd3 ((__v2df)__A, (__v2df)__B, (__v2df)__C);
+}
+
+extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm_fnmsub_ss (__m128 __A, __m128 __B, __m128 __C)
+{
+ return (__m128)__builtin_ia32_vfnmsubss3 ((__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
+}
+
+extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm_fmaddsub_pd (__m128d __A, __m128d __B, __m128d __C)
+{
+ return (__m128d)__builtin_ia32_vfmaddsubpd ((__v2df)__A, (__v2df)__B, (__v2df)__C);
+}
+
+extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm256_fmaddsub_pd (__m256d __A, __m256d __B, __m256d __C)
+{
+ return (__m256d)__builtin_ia32_vfmaddsubpd256 ((__v4df)__A, (__v4df)__B, (__v4df)__C);
+}
+
+extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm_fmaddsub_ps (__m128 __A, __m128 __B, __m128 __C)
+{
+ return (__m128)__builtin_ia32_vfmaddsubps ((__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
+}
+
+extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm256_fmaddsub_ps (__m256 __A, __m256 __B, __m256 __C)
+{
+ return (__m256)__builtin_ia32_vfmaddsubps256 ((__v8sf)__A, (__v8sf)__B, (__v8sf)__C);
+}
+
+extern __inline __m128d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm_fmsubadd_pd (__m128d __A, __m128d __B, __m128d __C)
+{
+ return (__m128d)__builtin_ia32_vfmsubaddpd ((__v2df)__A, (__v2df)__B, (__v2df)__C);
+}
+
+extern __inline __m256d __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm256_fmsubadd_pd (__m256d __A, __m256d __B, __m256d __C)
+{
+ return (__m256d)__builtin_ia32_vfmsubaddpd256 ((__v4df)__A, (__v4df)__B, (__v4df)__C);
+}
+
+extern __inline __m128 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm_fmsubadd_ps (__m128 __A, __m128 __B, __m128 __C)
+{
+ return (__m128)__builtin_ia32_vfmsubaddps ((__v4sf)__A, (__v4sf)__B, (__v4sf)__C);
+}
+
+extern __inline __m256 __attribute__((__gnu_inline__, __always_inline__, __artificial__))
+_mm256_fmsubadd_ps (__m256 __A, __m256 __B, __m256 __C)
+{
+ return (__m256)__builtin_ia32_vfmsubaddps256 ((__v8sf)__A, (__v8sf)__B, (__v8sf)__C);
+}
+
+#endif
+
+#endif
diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c
index fe6ccbe..c806f72 100644
--- a/gcc/config/i386/i386.c
+++ b/gcc/config/i386/i386.c
@@ -23888,7 +23888,7 @@ enum ix86_builtins
IX86_BUILTIN_VEC_PERM_V4DF,
IX86_BUILTIN_VEC_PERM_V8SF,
- /* FMA4 and XOP instructions. */
+ /* FMA4 instructions. */
IX86_BUILTIN_VFMADDSS,
IX86_BUILTIN_VFMADDSD,
IX86_BUILTIN_VFMADDPS,
@@ -23900,6 +23900,33 @@ enum ix86_builtins
IX86_BUILTIN_VFMADDSUBPS256,
IX86_BUILTIN_VFMADDSUBPD256,
+ /* FMA3 instructions. */
+ IX86_BUILTIN_VFMADDSS3,
+ IX86_BUILTIN_VFMADDSD3,
+ IX86_BUILTIN_VFNMADDSS3,
+ IX86_BUILTIN_VFNMADDSD3,
+ IX86_BUILTIN_VFMSUBSS3,
+ IX86_BUILTIN_VFMSUBSD3,
+ IX86_BUILTIN_VFNMSUBSS3,
+ IX86_BUILTIN_VFNMSUBSD3,
+ IX86_BUILTIN_VFMSUBPS,
+ IX86_BUILTIN_VFMSUBPD,
+ IX86_BUILTIN_VFMSUBPS256,
+ IX86_BUILTIN_VFMSUBPD256,
+ IX86_BUILTIN_VFNMADDPS,
+ IX86_BUILTIN_VFNMADDPD,
+ IX86_BUILTIN_VFNMADDPS256,
+ IX86_BUILTIN_VFNMADDPD256,
+ IX86_BUILTIN_VFNMSUBPS,
+ IX86_BUILTIN_VFNMSUBPD,
+ IX86_BUILTIN_VFNMSUBPS256,
+ IX86_BUILTIN_VFNMSUBPD256,
+ IX86_BUILTIN_VFMSUBADDPS,
+ IX86_BUILTIN_VFMSUBADDPD,
+ IX86_BUILTIN_VFMSUBADDPS256,
+ IX86_BUILTIN_VFMSUBADDPD256,
+
+ /* XOP instructions. */
IX86_BUILTIN_VPCMOV,
IX86_BUILTIN_VPCMOV_V2DI,
IX86_BUILTIN_VPCMOV_V4SI,
@@ -25100,6 +25127,31 @@ static const struct builtin_description bdesc_multi_arg[] =
"__builtin_ia32_vfmaddsd", IX86_BUILTIN_VFMADDSD,
UNKNOWN, (int)MULTI_ARG_3_DF },
+ { OPTION_MASK_ISA_FMA, CODE_FOR_fmai_vmfmadd_v4sf,
+ "__builtin_ia32_vfmaddss3", IX86_BUILTIN_VFMADDSS3,
+ UNKNOWN, (int)MULTI_ARG_3_SF },
+ { OPTION_MASK_ISA_FMA, CODE_FOR_fmai_vmfmadd_v2df,
+ "__builtin_ia32_vfmaddsd3", IX86_BUILTIN_VFMADDSD3,
+ UNKNOWN, (int)MULTI_ARG_3_DF },
+ { OPTION_MASK_ISA_FMA, CODE_FOR_fmai_vmfnmadd_v4sf,
+ "__builtin_ia32_vfnmaddss3", IX86_BUILTIN_VFNMADDSS3,
+ UNKNOWN, (int)MULTI_ARG_3_SF },
+ { OPTION_MASK_ISA_FMA, CODE_FOR_fmai_vmfnmadd_v2df,
+ "__builtin_ia32_vfnmaddsd3", IX86_BUILTIN_VFNMADDSD3,
+ UNKNOWN, (int)MULTI_ARG_3_DF },
+ { OPTION_MASK_ISA_FMA, CODE_FOR_fmai_vmfmsub_v4sf,
+ "__builtin_ia32_vfmsubss3", IX86_BUILTIN_VFMSUBSS3,
+ UNKNOWN, (int)MULTI_ARG_3_SF },
+ { OPTION_MASK_ISA_FMA, CODE_FOR_fmai_vmfmsub_v2df,
+ "__builtin_ia32_vfmsubsd3", IX86_BUILTIN_VFMSUBSD3,
+ UNKNOWN, (int)MULTI_ARG_3_DF },
+ { OPTION_MASK_ISA_FMA, CODE_FOR_fmai_vmfnmsub_v4sf,
+ "__builtin_ia32_vfnmsubss3", IX86_BUILTIN_VFNMSUBSS3,
+ UNKNOWN, (int)MULTI_ARG_3_SF },
+ { OPTION_MASK_ISA_FMA, CODE_FOR_fmai_vmfnmsub_v2df,
+ "__builtin_ia32_vfnmsubsd3", IX86_BUILTIN_VFNMSUBSD3,
+ UNKNOWN, (int)MULTI_ARG_3_DF },
+
{ OPTION_MASK_ISA_FMA | OPTION_MASK_ISA_FMA4, CODE_FOR_fma4i_fmadd_v4sf,
"__builtin_ia32_vfmaddps", IX86_BUILTIN_VFMADDPS,
UNKNOWN, (int)MULTI_ARG_3_SF },
@@ -25113,6 +25165,45 @@ static const struct builtin_description bdesc_multi_arg[] =
"__builtin_ia32_vfmaddpd256", IX86_BUILTIN_VFMADDPD256,
UNKNOWN, (int)MULTI_ARG_3_DF2 },
+ { OPTION_MASK_ISA_FMA, CODE_FOR_fmai_fmsub_v4sf,
+ "__builtin_ia32_vfmsubps", IX86_BUILTIN_VFMSUBPS,
+ UNKNOWN, (int)MULTI_ARG_3_SF },
+ { OPTION_MASK_ISA_FMA, CODE_FOR_fmai_fmsub_v2df,
+ "__builtin_ia32_vfmsubpd", IX86_BUILTIN_VFMSUBPD,
+ UNKNOWN, (int)MULTI_ARG_3_DF },
+ { OPTION_MASK_ISA_FMA, CODE_FOR_fmai_fmsub_v8sf,
+ "__builtin_ia32_vfmsubps256", IX86_BUILTIN_VFMSUBPS256,
+ UNKNOWN, (int)MULTI_ARG_3_SF2 },
+ { OPTION_MASK_ISA_FMA, CODE_FOR_fmai_fmsub_v4df,
+ "__builtin_ia32_vfmsubpd256", IX86_BUILTIN_VFMSUBPD256,
+ UNKNOWN, (int)MULTI_ARG_3_DF2 },
+ { OPTION_MASK_ISA_FMA, CODE_FOR_fmai_fnmadd_v4sf,
+ "__builtin_ia32_vfnmaddps", IX86_BUILTIN_VFNMADDPS,
+ UNKNOWN, (int)MULTI_ARG_3_SF },
+ { OPTION_MASK_ISA_FMA, CODE_FOR_fmai_fnmadd_v2df,
+ "__builtin_ia32_vfnmaddpd", IX86_BUILTIN_VFNMADDPD,
+ UNKNOWN, (int)MULTI_ARG_3_DF },
+ { OPTION_MASK_ISA_FMA, CODE_FOR_fmai_fnmadd_v8sf,
+ "__builtin_ia32_vfnmaddps256", IX86_BUILTIN_VFNMADDPS256,
+ UNKNOWN, (int)MULTI_ARG_3_SF2 },
+ { OPTION_MASK_ISA_FMA, CODE_FOR_fmai_fnmadd_v4df,
+ "__builtin_ia32_vfnmaddpd256", IX86_BUILTIN_VFNMADDPD256,
+ UNKNOWN, (int)MULTI_ARG_3_DF2 },
+ { OPTION_MASK_ISA_FMA, CODE_FOR_fmai_fnmsub_v4sf,
+ "__builtin_ia32_vfnmsubps", IX86_BUILTIN_VFNMSUBPS,
+ UNKNOWN, (int)MULTI_ARG_3_SF },
+ { OPTION_MASK_ISA_FMA, CODE_FOR_fmai_fnmsub_v2df,
+ "__builtin_ia32_vfnmsubpd", IX86_BUILTIN_VFNMSUBPD,
+ UNKNOWN, (int)MULTI_ARG_3_DF },
+ { OPTION_MASK_ISA_FMA, CODE_FOR_fmai_fnmsub_v8sf,
+ "__builtin_ia32_vfnmsubps256", IX86_BUILTIN_VFNMSUBPS256,
+ UNKNOWN, (int)MULTI_ARG_3_SF2 },
+ { OPTION_MASK_ISA_FMA, CODE_FOR_fmai_fnmsub_v4df,
+ "__builtin_ia32_vfnmsubpd256", IX86_BUILTIN_VFNMSUBPD256,
+ UNKNOWN, (int)MULTI_ARG_3_DF2 },
+
+
+
{ OPTION_MASK_ISA_FMA | OPTION_MASK_ISA_FMA4, CODE_FOR_fmaddsub_v4sf,
"__builtin_ia32_vfmaddsubps", IX86_BUILTIN_VFMADDSUBPS,
UNKNOWN, (int)MULTI_ARG_3_SF },
@@ -25125,6 +25216,18 @@ static const struct builtin_description bdesc_multi_arg[] =
{ OPTION_MASK_ISA_FMA | OPTION_MASK_ISA_FMA4, CODE_FOR_fmaddsub_v4df,
"__builtin_ia32_vfmaddsubpd256", IX86_BUILTIN_VFMADDSUBPD256,
UNKNOWN, (int)MULTI_ARG_3_DF2 },
+ { OPTION_MASK_ISA_FMA, CODE_FOR_fmsubadd_v4sf,
+ "__builtin_ia32_vfmsubaddps", IX86_BUILTIN_VFMSUBADDPS,
+ UNKNOWN, (int)MULTI_ARG_3_SF },
+ { OPTION_MASK_ISA_FMA, CODE_FOR_fmsubadd_v2df,
+ "__builtin_ia32_vfmsubaddpd", IX86_BUILTIN_VFMSUBADDPD,
+ UNKNOWN, (int)MULTI_ARG_3_DF },
+ { OPTION_MASK_ISA_FMA, CODE_FOR_fmsubadd_v8sf,
+ "__builtin_ia32_vfmsubaddps256", IX86_BUILTIN_VFMSUBADDPS256,
+ UNKNOWN, (int)MULTI_ARG_3_SF2 },
+ { OPTION_MASK_ISA_FMA, CODE_FOR_fmsubadd_v4df,
+ "__builtin_ia32_vfmsubaddpd256", IX86_BUILTIN_VFMSUBADDPD256,
+ UNKNOWN, (int)MULTI_ARG_3_DF2 },
{ OPTION_MASK_ISA_XOP, CODE_FOR_xop_pcmov_v2di, "__builtin_ia32_vpcmov", IX86_BUILTIN_VPCMOV, UNKNOWN, (int)MULTI_ARG_3_DI },
{ OPTION_MASK_ISA_XOP, CODE_FOR_xop_pcmov_v2di, "__builtin_ia32_vpcmov_v2di", IX86_BUILTIN_VPCMOV_V2DI, UNKNOWN, (int)MULTI_ARG_3_DI },
diff --git a/gcc/config/i386/sse.md b/gcc/config/i386/sse.md
index e9f6c3d..5004606 100644
--- a/gcc/config/i386/sse.md
+++ b/gcc/config/i386/sse.md
@@ -1528,6 +1528,44 @@
(match_operand:FMAMODE 3 "nonimmediate_operand")))]
"TARGET_FMA || TARGET_FMA4")
+
+(define_expand "fmai_fnmadd_<mode>"
+ [(set (match_operand:FMAMODE 0 "register_operand")
+ (fma:FMAMODE
+ (neg:FMAMODE
+ (match_operand:FMAMODE 1 "nonimmediate_operand"))
+ (match_operand:FMAMODE 2 "nonimmediate_operand")
+ (match_operand:FMAMODE 3 "nonimmediate_operand")))]
+ "TARGET_FMA")
+
+(define_expand "fmai_fmsub_<mode>"
+ [(set (match_operand:FMAMODE 0 "register_operand")
+ (fma:FMAMODE
+ (match_operand:FMAMODE 1 "nonimmediate_operand")
+ (match_operand:FMAMODE 2 "nonimmediate_operand")
+ (neg:FMAMODE
+ (match_operand:FMAMODE 3 "nonimmediate_operand"))))]
+ "TARGET_FMA")
+
+(define_expand "fmai_fnmsub_<mode>"
+ [(set (match_operand:FMAMODE 0 "register_operand")
+ (fma:FMAMODE
+ (neg:FMAMODE
+ (match_operand:FMAMODE 1 "nonimmediate_operand"))
+ (match_operand:FMAMODE 2 "nonimmediate_operand")
+ (neg:FMAMODE
+ (match_operand:FMAMODE 3 "nonimmediate_operand"))))]
+ "TARGET_FMA")
+
+(define_expand "fmai_fmadd_<mode>"
+ [(set (match_operand:VF_128 0 "register_operand")
+ (fma:VF_128
+ (match_operand:VF_128 1 "nonimmediate_operand")
+ (match_operand:VF_128 2 "nonimmediate_operand")
+ (match_operand:VF_128 3 "nonimmediate_operand")))]
+ "TARGET_FMA")
+
+
(define_insn "*fma4i_fmadd_<mode>"
[(set (match_operand:FMAMODE 0 "register_operand" "=x,x")
(fma:FMAMODE
@@ -1593,6 +1631,126 @@
operands[4] = CONST0_RTX (<MODE>mode);
})
+(define_expand "fmai_vmfmadd_<mode>"
+ [(set (match_operand:VF_128 0 "register_operand")
+ (vec_merge:VF_128
+ (fma:VF_128
+ (match_operand:VF_128 1 "nonimmediate_operand")
+ (match_operand:VF_128 2 "nonimmediate_operand")
+ (match_operand:VF_128 3 "nonimmediate_operand"))
+ (match_dup 0)
+ (const_int 1)))]
+ "TARGET_FMA")
+
+(define_expand "fmai_vmfmsub_<mode>"
+ [(set (match_operand:VF_128 0 "register_operand")
+ (vec_merge:VF_128
+ (fma:VF_128
+ (match_operand:VF_128 1 "nonimmediate_operand")
+ (match_operand:VF_128 2 "nonimmediate_operand")
+ (neg:VF_128
+ (match_operand:VF_128 3 "nonimmediate_operand")))
+ (match_dup 0)
+ (const_int 1)))]
+ "TARGET_FMA")
+
+(define_expand "fmai_vmfnmadd_<mode>"
+ [(set (match_operand:VF_128 0 "register_operand")
+ (vec_merge:VF_128
+ (fma:VF_128
+ (neg:VF_128
+ (match_operand:VF_128 1 "nonimmediate_operand"))
+ (match_operand:VF_128 2 "nonimmediate_operand")
+ (match_operand:VF_128 3 "nonimmediate_operand"))
+ (match_dup 0)
+ (const_int 1)))]
+ "TARGET_FMA")
+
+(define_expand "fmai_vmfnmsub_<mode>"
+ [(set (match_operand:VF_128 0 "register_operand")
+ (vec_merge:VF_128
+ (fma:VF_128
+ (neg:VF_128
+ (match_operand:VF_128 1 "nonimmediate_operand"))
+ (match_operand:VF_128 2 "nonimmediate_operand")
+ (neg:VF_128
+ (match_operand:VF_128 3 "nonimmediate_operand")))
+ (match_dup 0)
+ (const_int 1)))]
+ "TARGET_FMA")
+
+(define_insn "*fmai_fmadd_<mode>"
+ [(set (match_operand:VF_128 0 "register_operand" "=x,x,x")
+ (vec_merge:VF_128
+ (fma:VF_128
+ (match_operand:VF_128 1 "nonimmediate_operand" "%0, 0,x")
+ (match_operand:VF_128 2 "nonimmediate_operand" "xm, x,xm")
+ (match_operand:VF_128 3 "nonimmediate_operand" " x,xm,0"))
+ (match_dup 0)
+ (const_int 1)))]
+ "TARGET_FMA"
+ "@
+ vfmadd132<ssescalarmodesuffix>\t{%2, %3, %0|%0, %3, %2}
+ vfmadd213<ssescalarmodesuffix>\t{%3, %2, %0|%0, %2, %3}
+ vfmadd231<ssescalarmodesuffix>\t{%2, %1, %0|%0, %1, %2}"
+ [(set_attr "type" "ssemuladd")
+ (set_attr "mode" "<MODE>")])
+
+(define_insn "*fmai_fmsub_<mode>"
+ [(set (match_operand:VF_128 0 "register_operand" "=x,x,x")
+ (vec_merge:VF_128
+ (fma:VF_128
+ (match_operand:VF_128 1 "nonimmediate_operand" "%0, 0,x")
+ (match_operand:VF_128 2 "nonimmediate_operand" "xm, x,xm")
+ (neg:VF_128
+ (match_operand:VF_128 3 "nonimmediate_operand" " x,xm,0")))
+ (match_dup 0)
+ (const_int 1)))]
+ "TARGET_FMA"
+ "@
+ vfmsub132<ssescalarmodesuffix>\t{%2, %3, %0|%0, %3, %2}
+ vfmsub213<ssescalarmodesuffix>\t{%3, %2, %0|%0, %2, %3}
+ vfmsub231<ssescalarmodesuffix>\t{%2, %1, %0|%0, %1, %2}"
+ [(set_attr "type" "ssemuladd")
+ (set_attr "mode" "<MODE>")])
+
+(define_insn "*fmai_fnmadd_<mode>"
+ [(set (match_operand:VF_128 0 "register_operand" "=x,x,x")
+ (vec_merge:VF_128
+ (fma:VF_128
+ (neg:VF_128
+ (match_operand:VF_128 1 "nonimmediate_operand" "%0, 0,x"))
+ (match_operand:VF_128 2 "nonimmediate_operand" "xm, x,xm")
+ (match_operand:VF_128 3 "nonimmediate_operand" " x,xm,0"))
+ (match_dup 0)
+ (const_int 1)))]
+ "TARGET_FMA"
+ "@
+ vfnmadd132<ssescalarmodesuffix>\t{%2, %3, %0|%0, %3, %2}
+ vfnmadd213<ssescalarmodesuffix>\t{%3, %2, %0|%0, %2, %3}
+ vfnmadd231<ssescalarmodesuffix>\t{%2, %1, %0|%0, %1, %2}"
+ [(set_attr "type" "ssemuladd")
+ (set_attr "mode" "<MODE>")])
+
+(define_insn "*fmai_fnmsub_<mode>"
+ [(set (match_operand:VF_128 0 "register_operand" "=x,x,x")
+ (vec_merge:VF_128
+ (fma:VF_128
+ (neg:VF_128
+ (match_operand:VF_128 1 "nonimmediate_operand" "%0, 0,x"))
+ (match_operand:VF_128 2 "nonimmediate_operand" "xm, x,xm")
+ (neg:VF_128
+ (match_operand:VF_128 3 "nonimmediate_operand" " x,xm,0")))
+ (match_dup 0)
+ (const_int 1)))]
+ "TARGET_FMA"
+ "@
+ vfnmsub132<ssescalarmodesuffix>\t{%2, %3, %0|%0, %3, %2}
+ vfnmsub213<ssescalarmodesuffix>\t{%3, %2, %0|%0, %2, %3}
+ vfnmsub231<ssescalarmodesuffix>\t{%2, %1, %0|%0, %1, %2}"
+ [(set_attr "type" "ssemuladd")
+ (set_attr "mode" "<MODE>")])
+
(define_insn "*fma4i_vmfmadd_<mode>"
[(set (match_operand:VF_128 0 "register_operand" "=x,x")
(vec_merge:VF_128
@@ -1677,6 +1835,16 @@
UNSPEC_FMADDSUB))]
"TARGET_FMA || TARGET_FMA4")
+(define_expand "fmsubadd_<mode>"
+ [(set (match_operand:VF 0 "register_operand")
+ (unspec:VF
+ [(match_operand:VF 1 "nonimmediate_operand")
+ (match_operand:VF 2 "nonimmediate_operand")
+ (neg:VF
+ (match_operand:VF 3 "nonimmediate_operand"))]
+ UNSPEC_FMADDSUB))]
+ "TARGET_FMA")
+
(define_insn "*fma4_fmaddsub_<mode>"
[(set (match_operand:VF 0 "register_operand" "=x,x")
(unspec:VF
diff --git a/gcc/config/i386/x86intrin.h b/gcc/config/i386/x86intrin.h
index 88456f9..546b8a8 100644
--- a/gcc/config/i386/x86intrin.h
+++ b/gcc/config/i386/x86intrin.h
@@ -93,4 +93,8 @@
#include <popcntintrin.h>
#endif
+#ifdef __FMA__
+#include <fmaintrin.h>
+#endif
+
#endif /* _X86INTRIN_H_INCLUDED */
diff --git a/gcc/testsuite/gcc.target/i386/fma-256-fmaddXX.c b/gcc/testsuite/gcc.target/i386/fma-256-fmaddXX.c
new file mode 100644
index 0000000..dcabe43
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/fma-256-fmaddXX.c
@@ -0,0 +1,61 @@
+/* { dg-do run } */
+/* { dg-require-effective-target fma } */
+/* { dg-options "-O2 -mfma" } */
+
+#include "fma-check.h"
+
+#include <x86intrin.h>
+#include "m256-check.h"
+
+void
+check_mm256_fmadd_pd (__m256d __A, __m256d __B, __m256d __C)
+{
+ union256d a,b,c,e;
+ a.x = __A;
+ b.x = __B;
+ c.x = __C;
+ double d[4];
+ int i;
+ e.x = _mm256_fmadd_pd (__A, __B, __C);
+ for (i=0; i < 4;i++)
+ {
+ d[i] = a.a[i]*b.a[i]+c.a[i];
+ }
+ if (check_union256d (e, d))
+ abort ();
+}
+
+void
+check_mm256_fmadd_ps (__m256 __A, __m256 __B, __m256 __C)
+{
+ union256 a,b,c,e;
+ a.x = __A;
+ b.x = __B;
+ c.x = __C;
+ float d[8];
+ int i;
+ e.x = _mm256_fmadd_ps (__A, __B, __C);
+ for (i=0; i < 8;i++)
+ {
+ d[i] = a.a[i]*b.a[i]+c.a[i];
+ }
+ if (check_union256 (e, d))
+ abort ();
+}
+
+static void
+fma_test (void)
+{
+ union256 c[3];
+ union256d d[3];
+ int i,j;
+ for (i = 0;i < 3;i++)
+ {
+ for (j = 0;j < 4;j++)
+ c[i].a[j] = i*j + 3.5;
+ for (j = 0;j < 4;j++)
+ d[i].a[j] = i*j + 3.5;
+ }
+ check_mm256_fmadd_pd (d[0].x, d[1].x, d[2].x);
+ check_mm256_fmadd_ps (c[0].x, c[1].x, c[2].x);
+}
diff --git a/gcc/testsuite/gcc.target/i386/fma-256-fmaddsubXX.c b/gcc/testsuite/gcc.target/i386/fma-256-fmaddsubXX.c
new file mode 100644
index 0000000..5843c21
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/fma-256-fmaddsubXX.c
@@ -0,0 +1,61 @@
+/* { dg-do run } */
+/* { dg-require-effective-target fma } */
+/* { dg-options "-O2 -mfma" } */
+
+#include "fma-check.h"
+
+#include <x86intrin.h>
+#include "m256-check.h"
+
+void
+check_mm256_fmaddsub_ps (__m256 __A, __m256 __B, __m256 __C)
+{
+ union256 a,b,c,e;
+ a.x = __A;
+ b.x = __B;
+ c.x = __C;
+ float d[8];
+ int i;
+ e.x = _mm256_fmaddsub_ps (__A, __B, __C);
+ for (i=0; i < 8;i++)
+ {
+ d[i] = a.a[i]*b.a[i] + (i%2 == 1 ? c.a[i] : -c.a[i]);
+ }
+ if (check_union256 (e, d))
+ abort ();
+}
+
+void
+check_mm256_fmaddsub_pd (__m256d __A, __m256d __B, __m256d __C)
+{
+ union256d a,b,c,e;
+ a.x = __A;
+ b.x = __B;
+ c.x = __C;
+ double d[4];
+ int i;
+ e.x = _mm256_fmaddsub_pd (__A, __B, __C);
+ for (i=0; i < 4;i++)
+ {
+ d[i] = a.a[i]*b.a[i] + (i%2 == 1 ? c.a[i] : -c.a[i]);
+ }
+ if (check_union256d (e, d))
+ abort ();
+}
+
+static void
+fma_test (void)
+{
+ union256 c[3];
+ union256d d[3];
+ int i,j;
+ for (i = 0;i < 3;i++)
+ {
+ for (j = 0;j < 4;j++)
+ c[i].a[j] = i*j + 3.5;
+ for (j = 0;j < 4;j++)
+ d[i].a[j] = i*j + 3.5;
+ }
+ check_mm256_fmaddsub_pd (b[0].x, b[1].x, b[2].x);
+ check_mm256_fmaddsub_ps (a[0].x, a[1].x, a[2].x);
+}
diff --git a/gcc/testsuite/gcc.target/i386/fma-256-fmsubXX.c b/gcc/testsuite/gcc.target/i386/fma-256-fmsubXX.c
new file mode 100644
index 0000000..8673a78
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/fma-256-fmsubXX.c
@@ -0,0 +1,63 @@
+/* { dg-do run } */
+/* { dg-require-effective-target fma } */
+/* { dg-options "-O2 -mfma" } */
+
+#include "fma-check.h"
+
+#include <x86intrin.h>
+#include "m256-check.h"
+
+
+void
+check_mm256_fmsub_pd (__m256d __A, __m256d __B, __m256d __C)
+{
+ union256d a,b,c,e;
+ a.x = __A;
+ b.x = __B;
+ c.x = __C;
+ double d[4];
+ int i;
+ e.x = _mm256_fmsub_pd (__A, __B, __C);
+ for (i=0; i < 4;i++)
+ {
+ d[i] = a.a[i]*b.a[i]-c.a[i];
+ }
+ if (check_union256d (e, d))
+ abort ();
+}
+
+void
+check_mm256_fmsub_ps (__m256 __A, __m256 __B, __m256 __C)
+{
+ union256 a,b,c,e;
+ a.x = __A;
+ b.x = __B;
+ c.x = __C;
+ float d[8];
+ int i;
+ e.x = _mm256_fmsub_ps (__A, __B, __C);
+ for (i=0; i < 8;i++)
+ {
+ d[i] = a.a[i]*b.a[i]-c.a[i];
+ }
+ if (check_union256 (e, d))
+ abort ();
+}
+
+void
+static void
+fma_test (void)
+{
+ union256 c[3];
+ union256d d[3];
+ int i,j;
+ for (i = 0;i < 3;i++)
+ {
+ for (j = 0;j < 4;j++)
+ c[i].a[j] = i*j + 3.5;
+ for (j = 0;j < 4;j++)
+ d[i].a[j] = i*j + 3.5;
+ }
+ check_mm256_fmsub_pd (d[0].x, d[1].x, d[2].x);
+ check_mm256_fmsub_ps (c[0].x, c[1].x, c[2].x);
+}
diff --git a/gcc/testsuite/gcc.target/i386/fma-256-fmsubaddXX.c b/gcc/testsuite/gcc.target/i386/fma-256-fmsubaddXX.c
new file mode 100644
index 0000000..c5e47ae
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/fma-256-fmsubaddXX.c
@@ -0,0 +1,61 @@
+/* { dg-do run } */
+/* { dg-require-effective-target fma } */
+/* { dg-options "-O2 -mfma" } */
+
+#include "fma-check.h"
+
+#include <x86intrin.h>
+#include "m256-check.h"
+
+void
+check_mm256_fmsubadd_ps (__m256 __A, __m256 __B, __m256 __C)
+{
+ union256 a,b,c,e;
+ a.x = __A;
+ b.x = __B;
+ c.x = __C;
+ float d[8];
+ int i;
+ e.x = _mm256_fmsubadd_ps (__A, __B, __C);
+ for (i=0; i < 8;i++)
+ {
+ d[i] = a.a[i]*b.a[i] + (i%2 == 1 ? -c.a[i] : c.a[i]);
+ }
+ if (check_union256 (e, d))
+ abort ();
+}
+
+void
+check_mm256_fmsubadd_pd (__m256d __A, __m256d __B, __m256d __C)
+{
+ union256d a,b,c,e;
+ a.x = __A;
+ b.x = __B;
+ c.x = __C;
+ double d[4];
+ int i;
+ e.x = _mm256_fmsubadd_pd (__A, __B, __C);
+ for (i=0; i < 4;i++)
+ {
+ d[i] = a.a[i]*b.a[i] + (i%2 == 1 ? -c.a[i] : c.a[i]);
+ }
+ if (check_union256d (e, d))
+ abort ();
+}
+
+static void
+fma_test (void)
+{
+ union256 c[3];
+ union256d d[3];
+ int i,j;
+ for (i = 0;i < 3;i++)
+ {
+ for (j = 0;j < 4;j++)
+ c[i].a[j] = i*j + 3.5;
+ for (j = 0;j < 4;j++)
+ d[i].a[j] = i*j + 3.5;
+ }
+ check_mm256_fmsubadd_pd (b[0].x, b[1].x, b[2].x);
+ check_mm256_fmsubadd_ps (a[0].x, a[1].x, a[2].x);
+}
diff --git a/gcc/testsuite/gcc.target/i386/fma-256-fnmaddXX.c b/gcc/testsuite/gcc.target/i386/fma-256-fnmaddXX.c
new file mode 100644
index 0000000..ed3a24f
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/fma-256-fnmaddXX.c
@@ -0,0 +1,63 @@
+/* { dg-do run } */
+/* { dg-require-effective-target fma } */
+/* { dg-options "-O2 -mfma" } */
+
+#include "fma-check.h"
+
+#include <x86intrin.h>
+#include "m256-check.h"
+
+void
+check_mm256_fnmadd_pd (__m256d __A, __m256d __B, __m256d __C)
+{
+ union256d a,b,c,e;
+ a.x = __A;
+ b.x = __B;
+ c.x = __C;
+ double d[4];
+ int i;
+ e.x = _mm256_fnmadd_pd (__A, __B, __C);
+ for (i=0; i < 4;i++)
+ {
+ d[i] = -a.a[i]*b.a[i]+c.a[i];
+ }
+ if (check_union256d (e, d))
+ abort ();
+}
+
+void
+check_mm256_fnmadd_ps (__m256 __A, __m256 __B, __m256 __C)
+{
+ union256 a,b,c,e;
+ a.x = __A;
+ b.x = __B;
+ c.x = __C;
+ float d[8];
+ int i;
+ e.x = _mm256_fnmadd_ps (__A, __B, __C);
+ for (i=0; i < 8;i++)
+ {
+ d[i] = -a.a[i]*b.a[i]+c.a[i];
+ }
+ if (check_union256 (e, d))
+ abort ();
+}
+
+static void
+fma_test (void)
+{
+ union256 c[3];
+ union256d d[3];
+ int i,j;
+ for (i = 0;i < 3;i++)
+ {
+ for (j = 0;j < 4;j++)
+ c[i].a[j] = i*j + 3.5;
+ for (j = 0;j < 4;j++)
+ d[i].a[j] = i*j + 3.5;
+ }
+ check_mm256_fnmadd_pd (d[0].x, d[1].x, d[2].x);
+ check_mm256_fnmadd_ps (c[0].x, c[1].x, c[2].x);
+}
+
+
diff --git a/gcc/testsuite/gcc.target/i386/fma-256-fnmsubXX.c b/gcc/testsuite/gcc.target/i386/fma-256-fnmsubXX.c
new file mode 100644
index 0000000..a39c282
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/fma-256-fnmsubXX.c
@@ -0,0 +1,62 @@
+/* { dg-do run } */
+/* { dg-require-effective-target fma } */
+/* { dg-options "-O2 -mfma" } */
+
+#include "fma-check.h"
+
+#include <x86intrin.h>
+#include "m256-check.h"
+
+
+void
+check_mm256_fnmsub_pd (__m256d __A, __m256d __B, __m256d __C)
+{
+ union256d a,b,c,e;
+ a.x = __A;
+ b.x = __B;
+ c.x = __C;
+ double d[4];
+ int i;
+ e.x = _mm256_fnmsub_pd (__A, __B, __C);
+ for (i=0; i < 4;i++)
+ {
+ d[i] = -a.a[i]*b.a[i]-c.a[i];
+ }
+ if (check_union256d (e, d))
+ abort ();
+}
+
+void
+check_mm256_fnmsub_ps (__m256 __A, __m256 __B, __m256 __C)
+{
+ union256 a,b,c,e;
+ a.x = __A;
+ b.x = __B;
+ c.x = __C;
+ float d[8];
+ int i;
+ e.x = _mm256_fnmsub_ps (__A, __B, __C);
+ for (i=0; i < 8;i++)
+ {
+ d[i] = -a.a[i]*b.a[i]-c.a[i];
+ }
+ if (check_union256 (e, d))
+ abort ();
+}
+
+static void
+fma_test (void)
+{
+ union256 c[3];
+ union256d d[3];
+ int i,j;
+ for (i = 0;i < 3;i++)
+ {
+ for (j = 0;j < 4;j++)
+ c[i].a[j] = i*j + 3.5;
+ for (j = 0;j < 4;j++)
+ d[i].a[j] = i*j + 3.5;
+ }
+ check_mm256_fnmsub_pd (d[0].x, d[1].x, d[2].x);
+ check_mm256_fnmsub_ps (c[0].x, c[1].x, c[2].x);
+}
diff --git a/gcc/testsuite/gcc.target/i386/fma-check.h b/gcc/testsuite/gcc.target/i386/fma-check.h
new file mode 100644
index 0000000..db90c49
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/fma-check.h
@@ -0,0 +1,27 @@
+#include <stdlib.h>
+
+#include "cpuid.h"
+
+static void fma_test (void);
+
+static void
+__attribute__ ((noinline))
+do_test (void)
+{
+ fma_test ();
+}
+
+int
+main ()
+{
+ unsigned int eax, ebx, ecx, edx;
+
+ if (!__get_cpuid (1, &eax, &ebx, &ecx, &edx))
+ return 0;
+
+ /* Run FMA test only if host has FMA support. */
+ if (ecx & bit_FMA)
+ do_test ();
+
+ exit (0);
+}
diff --git a/gcc/testsuite/gcc.target/i386/fma-compile.c b/gcc/testsuite/gcc.target/i386/fma-compile.c
new file mode 100644
index 0000000..fe145b5
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/fma-compile.c
@@ -0,0 +1,221 @@
+/* Test that the compiler properly generates floating point multiply
+ and add instructions FMA systems. */
+
+/* { dg-do compile } */
+/* { dg-options "-O2 -mfma" } */
+
+#include <x86intrin.h>
+
+__m128d
+check_mm_fmadd_pd (__m128d a, __m128d b,__m128d c)
+{
+ return _mm_fmadd_pd (a, b, c);
+}
+
+__m256d
+check_mm256_fmadd_pd (__m256d a, __m256d b,__m256d c)
+{
+ return _mm256_fmadd_pd (a, b, c);
+}
+
+__m128
+check_mm_fmadd_ps (__m128 a, __m128 b,__m128 c)
+{
+ return _mm_fmadd_ps (a, b, c);
+}
+
+__m256
+check_mm256_fmadd_ps (__m256 a, __m256 b,__m256 c)
+{
+ return _mm256_fmadd_ps (a, b, c);
+}
+
+__m128d
+check_mm_fmadd_sd (__m128d a, __m128d b,__m128d c)
+{
+ return _mm_fmadd_sd (a, b, c);
+}
+
+__m128
+check_mm_fmadd_ss (__m128 a, __m128 b,__m128 c)
+{
+ return _mm_fmadd_ss (a, b, c);
+}
+
+__m128d
+check_mm_fmsub_pd (__m128d a, __m128d b,__m128d c)
+{
+ return _mm_fmsub_pd (a, b, c);
+}
+
+__m256d
+check_mm256_fmsub_pd (__m256d a, __m256d b,__m256d c)
+{
+ return _mm256_fmsub_pd (a, b, c);
+}
+
+__m128
+check_mm_fmsub_ps (__m128 a, __m128 b,__m128 c)
+{
+ return _mm_fmsub_ps (a, b, c);
+}
+
+__m256
+check_mm256_fmsub_ps (__m256 a, __m256 b,__m256 c)
+{
+ return _mm256_fmsub_ps (a, b, c);
+}
+
+__m128d
+check_mm_fmsub_sd (__m128d a, __m128d b,__m128d c)
+{
+ return _mm_fmsub_sd (a, b, c);
+}
+
+__m128
+check_mm_fmsub_ss (__m128 a, __m128 b,__m128 c)
+{
+ return _mm_fmsub_ss (a, b, c);
+}
+
+__m128d
+check_mm_fnmadd_pd (__m128d a, __m128d b,__m128d c)
+{
+ return _mm_fnmadd_pd (a, b, c);
+}
+
+__m256d
+check_mm256_fnmadd_pd (__m256d a, __m256d b,__m256d c)
+{
+ return _mm256_fnmadd_pd (a, b, c);
+}
+
+__m128
+check_mm_fnmadd_ps (__m128 a, __m128 b,__m128 c)
+{
+ return _mm_fnmadd_ps (a, b, c);
+}
+
+__m256
+check_mm256_fnmadd_ps (__m256 a, __m256 b,__m256 c)
+{
+ return _mm256_fnmadd_ps (a, b, c);
+}
+
+__m128d
+check_mm_fnmadd_sd (__m128d a, __m128d b,__m128d c)
+{
+ return _mm_fnmadd_sd (a, b, c);
+}
+
+__m128
+check_mm_fnmadd_ss (__m128 a, __m128 b,__m128 c)
+{
+ return _mm_fnmadd_ss (a, b, c);
+}
+
+__m128d
+check_mm_fnmsub_pd (__m128d a, __m128d b,__m128d c)
+{
+ return _mm_fnmsub_pd (a, b, c);
+}
+
+__m256d
+check_mm256_fnmsub_pd (__m256d a, __m256d b,__m256d c)
+{
+ return _mm256_fnmsub_pd (a, b, c);
+}
+
+__m128
+check_mm_fnmsub_ps (__m128 a, __m128 b,__m128 c)
+{
+ return _mm_fnmsub_ps (a, b, c);
+}
+
+__m256
+check_mm256_fnmsub_ps (__m256 a, __m256 b,__m256 c)
+{
+ return _mm256_fnmsub_ps (a, b, c);
+}
+
+__m128d
+check_mm_fnmsub_sd (__m128d a, __m128d b,__m128d c)
+{
+ return _mm_fnmsub_sd (a, b, c);
+}
+
+__m128
+check_mm_fnmsub_ss (__m128 a, __m128 b,__m128 c)
+{
+ return _mm_fnmsub_ss (a, b, c);
+}
+
+__m128d
+check_mm_fmaddsub_pd (__m128d a, __m128d b,__m128d c)
+{
+ return _mm_fmaddsub_pd (a, b, c);
+}
+
+__m256d
+check_mm256_fmaddsub_pd (__m256d a, __m256d b,__m256d c)
+{
+ return _mm256_fmaddsub_pd (a, b, c);
+}
+
+__m128
+check_mm_fmaddsub_ps (__m128 a, __m128 b,__m128 c)
+{
+ return _mm_fmaddsub_ps (a, b, c);
+}
+
+__m256
+check_mm256_fmaddsub_ps (__m256 a, __m256 b,__m256 c)
+{
+ return _mm256_fmaddsub_ps (a, b, c);
+}
+
+__m128d
+check_mm_fmsubadd_pd (__m128d a, __m128d b,__m128d c)
+{
+ return _mm_fmsubadd_pd (a, b, c);
+}
+
+__m256d
+check_mm256_fmsubadd_pd (__m256d a, __m256d b,__m256d c)
+{
+ return _mm256_fmsubadd_pd (a, b, c);
+}
+
+__m128
+check_mm_fmsubadd_ps (__m128 a, __m128 b,__m128 c)
+{
+ return _mm_fmsubadd_ps (a, b, c);
+}
+
+__m256
+check_mm256_fmsubadd_ps (__m256 a, __m256 b,__m256 c)
+{
+ return _mm256_fmsubadd_ps (a, b, c);
+}
+
+
+/* { dg-final { scan-assembler-times "vfmadd[^s]..ps" 2 } } */
+/* { dg-final { scan-assembler-times "vfmsub[^s]..ps" 2 } } */
+/* { dg-final { scan-assembler-times "vfnmadd...ps" 2 } } */
+/* { dg-final { scan-assembler-times "vfnmsub...ps" 2 } } */
+/* { dg-final { scan-assembler-times "vfmaddsub...ps" 2 } } */
+/* { dg-final { scan-assembler-times "vfmsubadd...ps" 2 } } */
+/* { dg-final { scan-assembler-times "vfmadd[^s]..pd" 2 } } */
+/* { dg-final { scan-assembler-times "vfmsub[^s]..pd" 2 } } */
+/* { dg-final { scan-assembler-times "vfnmadd...pd" 2 } } */
+/* { dg-final { scan-assembler-times "vfnmsub...pd" 2 } } */
+/* { dg-final { scan-assembler-times "vfmaddsub...pd" 2 } } */
+/* { dg-final { scan-assembler-times "vfmsubadd...pd" 2 } } */
+/* { dg-final { scan-assembler-times "vfmadd[^s]..ss" 1 } } */
+/* { dg-final { scan-assembler-times "vfmsub[^s]..ss" 1 } } */
+/* { dg-final { scan-assembler-times "vfnmadd...ss" 1 } } */
+/* { dg-final { scan-assembler-times "vfnmsub...ss" 1 } } */
+/* { dg-final { scan-assembler-times "vfmadd[^s]..sd" 1 } } */
+/* { dg-final { scan-assembler-times "vfmsub[^s]..sd" 1 } } */
+/* { dg-final { scan-assembler-times "vfnmadd...sd" 1 } } */
+/* { dg-final { scan-assembler-times "vfnmsub...sd" 1 } } */
diff --git a/gcc/testsuite/gcc.target/i386/fma-fmaddXX.c b/gcc/testsuite/gcc.target/i386/fma-fmaddXX.c
new file mode 100644
index 0000000..c4e9062
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/fma-fmaddXX.c
@@ -0,0 +1,102 @@
+/* { dg-do run } */
+/* { dg-require-effective-target fma } */
+/* { dg-options "-O2 -mfma" } */
+
+#include "fma-check.h"
+
+#include <x86intrin.h>
+#include "m256-check.h"
+
+void
+check_mm_fmadd_pd (__m128d __A, __m128d __B, __m128d __C)
+{
+ union128d a,b,c,e;
+ a.x = __A;
+ b.x = __B;
+ c.x = __C;
+ double d[2];
+ int i;
+ e.x = _mm_fmadd_pd (__A, __B, __C);
+ for (i=0; i < 2;i++)
+ {
+ d[i] = a.a[i]*b.a[i]+c.a[i];
+ }
+
+ if (check_union128d (e, d))
+ abort ();
+}
+
+void
+check_mm_fmadd_ps (__m128 __A, __m128 __B, __m128 __C)
+{
+ union128 a,b,c,e;
+ a.x = __A;
+ b.x = __B;
+ c.x = __C;
+ float d[4];
+ int i;
+ e.x = _mm_fmadd_ps (__A, __B, __C);
+ for (i=0; i < 4;i++)
+ {
+ d[i] = a.a[i]*b.a[i]+c.a[i];
+ }
+ if (check_union128 (e, d))
+ abort ();
+}
+
+void
+check_mm_fmadd_sd (__m128d __A, __m128d __B, __m128d __C)
+{
+ union128d a,b,c,e;
+ a.x = __A;
+ b.x = __B;
+ c.x = __C;
+ double d[2];
+ int i;
+ e.x = _mm_fmadd_sd (__A, __B, __C);
+ for (i=1; i < 2;i++)
+ {
+ d[i] = a.a[i];
+ }
+ d[0] = a.a[0]*b.a[0]+c.a[0];
+ if (check_union128d (e, d))
+ abort ();
+}
+
+void
+check_mm_fmadd_ss (__m128 __A, __m128 __B, __m128 __C)
+{
+ union128 a,b,c,e;
+ a.x = __A;
+ b.x = __B;
+ c.x = __C;
+ float d[4];
+ int i;
+ e.x = _mm_fmadd_ss (__A, __B, __C);
+ for (i=1; i < 4;i++)
+ {
+ d[i] = a.a[i];
+ }
+ d[0] = a.a[0]*b.a[0]+c.a[0];
+ if (check_union128 (e, d))
+ abort ();
+}
+
+static void
+fma_test (void)
+{
+ union128 a[3];
+ union128d b[3];
+ int i,j;
+ for (i = 0;i < 3;i++)
+ {
+ for (j = 0;j < 2;j++)
+ a[i].a[j] = i*j + 3.5;
+ for (j = 0;j < 2;j++)
+ b[i].a[j] = i*j + 3.5;
+ }
+ check_mm_fmadd_pd (b[0].x, b[1].x, b[2].x);
+ check_mm_fmadd_sd (b[0].x, b[1].x, b[2].x);
+ check_mm_fmadd_ps (a[0].x, a[1].x, a[2].x);
+ check_mm_fmadd_ss (a[0].x, a[1].x, a[2].x);
+}
diff --git a/gcc/testsuite/gcc.target/i386/fma-fmaddsubXX.c b/gcc/testsuite/gcc.target/i386/fma-fmaddsubXX.c
new file mode 100644
index 0000000..17dd026
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/fma-fmaddsubXX.c
@@ -0,0 +1,61 @@
+/* { dg-do run } */
+/* { dg-require-effective-target fma } */
+/* { dg-options "-O2 -mfma" } */
+
+#include "fma-check.h"
+
+#include <x86intrin.h>
+#include "m256-check.h"
+
+void
+check_mm_fmaddsub_ps (__m128 __A, __m128 __B, __m128 __C)
+{
+ union128 a,b,c,e;
+ a.x = __A;
+ b.x = __B;
+ c.x = __C;
+ float d[4];
+ int i;
+ e.x = _mm_fmaddsub_ps (__A, __B, __C);
+ for (i=0; i < 4;i++)
+ {
+ d[i] = a.a[i]*b.a[i] + (i%2 == 1 ? c.a[i] : -c.a[i]);
+ }
+ if (check_union128 (e, d))
+ abort ();
+}
+
+void
+check_mm_fmaddsub_pd (__m128d __A, __m128d __B, __m128d __C)
+{
+ union128d a,b,c,e;
+ a.x = __A;
+ b.x = __B;
+ c.x = __C;
+ double d[2];
+ int i;
+ e.x = _mm_fmaddsub_pd (__A, __B, __C);
+ for (i=0; i < 2;i++)
+ {
+ d[i] = a.a[i]*b.a[i] + (i%2 == 1 ? c.a[i] : -c.a[i]);
+ }
+ if (check_union128d (e, d))
+ abort ();
+}
+
+static void
+fma_test (void)
+{
+ union128 a[3];
+ union128d b[3];
+ int i,j;
+ for (i = 0;i < 3;i++)
+ {
+ for (j = 0;j < 2;j++)
+ a[i].a[j] = i*j + 3.5;
+ for (j = 0;j < 2;j++)
+ b[i].a[j] = i*j + 3.5;
+ }
+ check_mm_fmaddsub_pd (b[0].x, b[1].x, b[2].x);
+ check_mm_fmaddsub_ps (a[0].x, a[1].x, a[2].x);
+}
diff --git a/gcc/testsuite/gcc.target/i386/fma-fmsubXX.c b/gcc/testsuite/gcc.target/i386/fma-fmsubXX.c
new file mode 100644
index 0000000..13e6d51
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/fma-fmsubXX.c
@@ -0,0 +1,101 @@
+/* { dg-do run } */
+/* { dg-require-effective-target fma } */
+/* { dg-options "-O2 -mfma" } */
+
+#include "fma-check.h"
+
+#include <x86intrin.h>
+#include "m256-check.h"
+
+void
+check_mm_fmsub_pd (__m128d __A, __m128d __B, __m128d __C)
+{
+ union128d a,b,c,e;
+ a.x = __A;
+ b.x = __B;
+ c.x = __C;
+ double d[2];
+ int i;
+ e.x = _mm_fmsub_pd (__A, __B, __C);
+ for (i=0; i < 2;i++)
+ {
+ d[i] = a.a[i]*b.a[i]-c.a[i];
+ }
+ if (check_union128d (e, d))
+ abort ();
+}
+
+void
+check_mm_fmsub_ps (__m128 __A, __m128 __B, __m128 __C)
+{
+ union128 a,b,c,e;
+ a.x = __A;
+ b.x = __B;
+ c.x = __C;
+ float d[4];
+ int i;
+ e.x = _mm_fmsub_ps (__A, __B, __C);
+ for (i=0; i < 4;i++)
+ {
+ d[i] = a.a[i]*b.a[i]-c.a[i];
+ }
+ if (check_union128 (e, d))
+ abort ();
+}
+
+void
+check_mm_fmsub_sd (__m128d __A, __m128d __B, __m128d __C)
+{
+ union128d a,b,c,e;
+ a.x = __A;
+ b.x = __B;
+ c.x = __C;
+ double d[2];
+ int i;
+ e.x = _mm_fmsub_sd (__A, __B, __C);
+ for (i=1; i < 2;i++)
+ {
+ d[i] = a.a[i];
+ }
+ d[0] = a.a[0]*b.a[0]-c.a[0];
+ if (check_union128d (e, d))
+ abort ();
+}
+
+void
+check_mm_fmsub_ss (__m128 __A, __m128 __B, __m128 __C)
+{
+ union128 a,b,c,e;
+ a.x = __A;
+ b.x = __B;
+ c.x = __C;
+ float d[4];
+ int i;
+ e.x = _mm_fmsub_ss (__A, __B, __C);
+ for (i=1; i < 4;i++)
+ {
+ d[i] = a.a[i];
+ }
+ d[0] = a.a[0]*b.a[0]-c.a[0];
+ if (check_union128 (e, d))
+ abort ();
+}
+
+static void
+fma_test (void)
+{
+ union128 a[3];
+ union128d b[3];
+ int i,j;
+ for (i = 0;i < 3;i++)
+ {
+ for (j = 0;j < 2;j++)
+ a[i].a[j] = i*j + 3.5;
+ for (j = 0;j < 2;j++)
+ b[i].a[j] = i*j + 3.5;
+ }
+ check_mm_fmsub_pd (b[0].x, b[1].x, b[2].x);
+ check_mm_fmsub_sd (b[0].x, b[1].x, b[2].x);
+ check_mm_fmsub_ps (a[0].x, a[1].x, a[2].x);
+ check_mm_fmsub_ss (a[0].x, a[1].x, a[2].x);
+}
diff --git a/gcc/testsuite/gcc.target/i386/fma-fmsubaddXX.c b/gcc/testsuite/gcc.target/i386/fma-fmsubaddXX.c
new file mode 100644
index 0000000..dddee41
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/fma-fmsubaddXX.c
@@ -0,0 +1,61 @@
+/* { dg-do run } */
+/* { dg-require-effective-target fma } */
+/* { dg-options "-O2 -mfma" } */
+
+#include "fma-check.h"
+
+#include <x86intrin.h>
+#include "m256-check.h"
+
+void
+check_mm_fmsubadd_ps (__m128 __A, __m128 __B, __m128 __C)
+{
+ union128 a,b,c,e;
+ a.x = __A;
+ b.x = __B;
+ c.x = __C;
+ float d[4];
+ int i;
+ e.x = _mm_fmsubadd_ps (__A, __B, __C);
+ for (i=0; i < 4;i++)
+ {
+ d[i] = a.a[i]*b.a[i] + (i%2 == 1 ? -c.a[i] : c.a[i]);
+ }
+ if (check_union128 (e, d))
+ abort ();
+}
+
+void
+check_mm_fmsubadd_pd (__m128d __A, __m128d __B, __m128d __C)
+{
+ union128d a,b,c,e;
+ a.x = __A;
+ b.x = __B;
+ c.x = __C;
+ double d[2];
+ int i;
+ e.x = _mm_fmsubadd_pd (__A, __B, __C);
+ for (i=0; i < 2;i++)
+ {
+ d[i] = a.a[i]*b.a[i] + (i%2 == 1 ? -c.a[i] : c.a[i]);
+ }
+ if (check_union128d (e, d))
+ abort ();
+}
+
+static void
+fma_test (void)
+{
+ union128 a[3];
+ union128d b[3];
+ int i,j;
+ for (i = 0;i < 3;i++)
+ {
+ for (j = 0;j < 2;j++)
+ a[i].a[j] = i*j + 3.5;
+ for (j = 0;j < 2;j++)
+ b[i].a[j] = i*j + 3.5;
+ }
+ check_mm_fmsubadd_pd (b[0].x, b[1].x, b[2].x);
+ check_mm_fmsubadd_ps (a[0].x, a[1].x, a[2].x);
+}
diff --git a/gcc/testsuite/gcc.target/i386/fma-fnmaddXX.c b/gcc/testsuite/gcc.target/i386/fma-fnmaddXX.c
new file mode 100644
index 0000000..20edd4c
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/fma-fnmaddXX.c
@@ -0,0 +1,101 @@
+/* { dg-do run } */
+/* { dg-require-effective-target fma } */
+/* { dg-options "-O2 -mfma" } */
+
+#include "fma-check.h"
+
+#include <x86intrin.h>
+#include "m256-check.h"
+
+void
+check_mm_fnmadd_ps (__m128 __A, __m128 __B, __m128 __C)
+{
+ union128 a,b,c,e;
+ a.x = __A;
+ b.x = __B;
+ c.x = __C;
+ float d[4];
+ int i;
+ e.x = _mm_fnmadd_ps (__A, __B, __C);
+ for (i=0; i < 4;i++)
+ {
+ d[i] = -a.a[i]*b.a[i]+c.a[i];
+ }
+ if (check_union128 (e, d))
+ abort ();
+}
+
+void
+check_mm_fnmadd_pd (__m128d __A, __m128d __B, __m128d __C)
+{
+ union128d a,b,c,e;
+ a.x = __A;
+ b.x = __B;
+ c.x = __C;
+ double d[2];
+ int i;
+ e.x = _mm_fnmadd_pd (__A, __B, __C);
+ for (i=0; i < 2;i++)
+ {
+ d[i] = -a.a[i]*b.a[i]+c.a[i];
+ }
+ if (check_union128d (e, d))
+ abort ();
+}
+
+void
+check_mm_fnmadd_sd (__m128d __A, __m128d __B, __m128d __C)
+{
+ union128d a,b,c,e;
+ a.x = __A;
+ b.x = __B;
+ c.x = __C;
+ double d[2];
+ int i;
+ e.x = _mm_fnmadd_sd (__A, __B, __C);
+ for (i=1; i < 2;i++)
+ {
+ d[i] = a.a[i];
+ }
+ d[0] = -a.a[0]*b.a[0]+c.a[0];
+ if (check_union128d (e, d))
+ abort ();
+}
+
+void
+check_mm_fnmadd_ss (__m128 __A, __m128 __B, __m128 __C)
+{
+ union128 a,b,c,e;
+ a.x = __A;
+ b.x = __B;
+ c.x = __C;
+ float d[4];
+ int i;
+ e.x = _mm_fnmadd_ss (__A, __B, __C);
+ for (i=1; i < 4;i++)
+ {
+ d[i] = a.a[i];
+ }
+ d[0] = -a.a[0]*b.a[0]+c.a[0];
+ if (check_union128 (e, d))
+ abort ();
+}
+
+static void
+fma_test (void)
+{
+ union128 a[3];
+ union128d b[3];
+ int i,j;
+ for (i = 0;i < 3;i++)
+ {
+ for (j = 0;j < 2;j++)
+ a[i].a[j] = i*j + 3.5;
+ for (j = 0;j < 2;j++)
+ b[i].a[j] = i*j + 3.5;
+ }
+ check_mm_fnmadd_pd (b[0].x, b[1].x, b[2].x);
+ check_mm_fnmadd_sd (b[0].x, b[1].x, b[2].x);
+ check_mm_fnmadd_ps (a[0].x, a[1].x, a[2].x);
+ check_mm_fnmadd_ss (a[0].x, a[1].x, a[2].x);
+}
diff --git a/gcc/testsuite/gcc.target/i386/fma-fnmsubXX.c b/gcc/testsuite/gcc.target/i386/fma-fnmsubXX.c
new file mode 100644
index 0000000..52b2b61
--- /dev/null
+++ b/gcc/testsuite/gcc.target/i386/fma-fnmsubXX.c
@@ -0,0 +1,102 @@
+/* { dg-do run } */
+/* { dg-require-effective-target fma } */
+/* { dg-options "-O2 -mfma" } */
+
+#include "fma-check.h"
+
+#include <x86intrin.h>
+#include "m256-check.h"
+
+void
+check_mm_fnmsub_sd (__m128d __A, __m128d __B, __m128d __C)
+{
+ union128d a,b,c,e;
+ a.x = __A;
+ b.x = __B;
+ c.x = __C;
+ double d[2];
+ int i;
+ e.x = _mm_fnmsub_sd (__A, __B, __C);
+ for (i=1; i < 2;i++)
+ {
+ d[i] = a.a[i];
+ }
+ d[0] = -a.a[0]*b.a[0]-c.a[0];
+ if (check_union128d (e, d))
+ abort ();
+}
+
+void
+check_mm_fnmsub_ss (__m128 __A, __m128 __B, __m128 __C)
+{
+ union128 a,b,c,e;
+ a.x = __A;
+ b.x = __B;
+ c.x = __C;
+ float d[4];
+ int i;
+ e.x = _mm_fnmsub_ss (__A, __B, __C);
+ for (i=1; i < 4;i++)
+ {
+ d[i] = a.a[i];
+ }
+ d[0] = -a.a[0]*b.a[0]-c.a[0];
+ if (check_union128 (e, d))
+ abort ();
+}
+
+void
+check_mm_fnmsub_ps (__m128 __A, __m128 __B, __m128 __C)
+{
+ union128 a,b,c,e;
+ a.x = __A;
+ b.x = __B;
+ c.x = __C;
+ float d[4];
+ int i;
+ e.x = _mm_fnmsub_ps (__A, __B, __C);
+ for (i=0; i < 4;i++)
+ {
+ d[i] = -a.a[i]*b.a[i]-c.a[i];
+ }
+ if (check_union128 (e, d))
+ abort ();
+}
+
+void
+check_mm_fnmsub_pd (__m128d __A, __m128d __B, __m128d __C)
+{
+ union128d a,b,c,e;
+ a.x = __A;
+ b.x = __B;
+ c.x = __C;
+ double d[2];
+ int i;
+ e.x = _mm_fnmsub_pd (__A, __B, __C);
+ for (i=0; i < 2;i++)
+ {
+ d[i] = -a.a[i]*b.a[i]-c.a[i];
+ }
+ if (check_union128d (e, d))
+ abort ();
+}
+
+void
+static void
+fma_test (void)
+{
+ union128 a[3];
+ union128d b[3];
+ int i,j;
+ for (i = 0;i < 3;i++)
+ {
+ for (j = 0;j < 2;j++)
+ a[i].a[j] = i*j + 3.5;
+ for (j = 0;j < 2;j++)
+ b[i].a[j] = i*j + 3.5;
+ }
+ check_mm_fnmsub_pd (b[0].x, b[1].x, b[2].x);
+ check_mm_fnmsub_sd (b[0].x, b[1].x, b[2].x);
+ check_mm_fnmsub_ps (a[0].x, a[1].x, a[2].x);
+ check_mm_fnmsub_ss (a[0].x, a[1].x, a[2].x);
+}
diff --git a/gcc/testsuite/gcc.target/i386/i386.exp b/gcc/testsuite/gcc.target/i386/i386.exp
index 167b79b..8ebe57d 100644
--- a/gcc/testsuite/gcc.target/i386/i386.exp
+++ b/gcc/testsuite/gcc.target/i386/i386.exp
@@ -172,6 +172,20 @@ proc check_effective_target_fma4 { } {
} "-O2 -mfma4" ]
}
+# Return 1 if fma instructions can be compiled.
+proc check_effective_target_fma { } {
+ return [check_no_compiler_messages fma object {
+ typedef float __m128 __attribute__ ((__vector_size__ (16)));
+ typedef float __v4sf __attribute__ ((__vector_size__ (16)));
+ __m128 _mm_macc_ps(__m128 __A, __m128 __B, __m128 __C)
+ {
+ return (__m128) __builtin_ia32_vfmsubps ((__v4sf)__A,
+ (__v4sf)__B,
+ (__v4sf)__C);
+ }
+ } "-O2 -mfma" ]
+}
+
# Return 1 if xop instructions can be compiled.
proc check_effective_target_xop { } {
return [check_no_compiler_messages xop object {
next prev parent reply other threads:[~2011-08-22 16:25 UTC|newest]
Thread overview: 27+ messages / expand[flat|nested] mbox.gz Atom feed top
2011-08-20 11:23 Uros Bizjak
2011-08-22 17:32 ` Ilya Tocar [this message]
2011-08-22 20:41 ` Uros Bizjak
2011-08-23 14:55 ` Ilya Tocar
2011-08-23 15:33 ` Uros Bizjak
2011-08-24 10:06 ` Ilya Tocar
2011-08-24 10:12 ` Jakub Jelinek
2011-08-24 11:12 ` Ilya Tocar
2011-08-24 13:31 ` Uros Bizjak
2011-08-24 13:52 ` Ilya Tocar
2011-08-24 22:00 ` Uros Bizjak
2011-08-25 10:15 ` Ilya Tocar
2011-08-25 10:46 ` Uros Bizjak
2011-08-25 11:46 ` Ilya Tocar
2011-08-25 11:46 ` Jakub Jelinek
2011-08-25 11:49 ` Ilya Tocar
2011-08-26 11:03 ` Ilya Tocar
2011-08-26 14:07 ` H.J. Lu
2011-08-26 15:47 ` Ilya Tocar
2011-08-26 17:02 ` H.J. Lu
2011-08-26 17:06 ` H.J. Lu
2011-08-30 12:05 ` Ilya Tocar
2011-08-30 14:02 ` H.J. Lu
2011-08-30 14:10 ` Ilya Tocar
2011-08-30 14:32 ` Ilya Tocar
2011-08-30 15:25 ` H.J. Lu
-- strict thread matches above, loose matches on Subject: below --
2011-08-17 14:08 [PATCH, i386, testsuite]FMA intrinsics Ilya Tocar
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=CAFFGCRB-auRYMdLGPzSdb9ftGd++vtgeWzOG2xARQwe4vij9iQ@mail.gmail.com \
--to=tocarip.intel@gmail.com \
--cc=gcc-patches@gcc.gnu.org \
--cc=ubizjak@gmail.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
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).