| 1 | /*************************************************************************** |
|---|
| 2 | * Copyright (C) 2005 by Prakash Punnoor * |
|---|
| 3 | * prakash@punnoor.de * |
|---|
| 4 | * * |
|---|
| 5 | * This program is free software; you can redistribute it and/or modify * |
|---|
| 6 | * it under the terms of the GNU Library General Public License as * |
|---|
| 7 | * published by the Free Software Foundation; either version 2 of the * |
|---|
| 8 | * License, or (at your option) any later version. * |
|---|
| 9 | * * |
|---|
| 10 | * This program is distributed in the hope that it will be useful, * |
|---|
| 11 | * but WITHOUT ANY WARRANTY; without even the implied warranty of * |
|---|
| 12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the * |
|---|
| 13 | * GNU General Public License for more details. * |
|---|
| 14 | * * |
|---|
| 15 | * You should have received a copy of the GNU Library General Public * |
|---|
| 16 | * License along with this program; if not, write to the * |
|---|
| 17 | * Free Software Foundation, Inc., * |
|---|
| 18 | * 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. * |
|---|
| 19 | ***************************************************************************/ |
|---|
| 20 | #ifndef _X86_SIMD_SUPPORT_H_ |
|---|
| 21 | #define _X86_SIMD_SUPPORT_H_ |
|---|
| 22 | |
|---|
| 23 | #ifdef __MMX__ |
|---|
| 24 | /* |
|---|
| 25 | * We use built-ins for gcc instead of Intel/MSVC style intrinsics |
|---|
| 26 | * as (older) gccs are slower with them |
|---|
| 27 | */ |
|---|
| 28 | #if __GNUC__ && !__INTEL_COMPILER |
|---|
| 29 | #if __GNUC__ < 4 |
|---|
| 30 | typedef short v4hi __attribute__ ((__mode__(__V4HI__))); |
|---|
| 31 | typedef int v2si __attribute__ ((__mode__(__V2SI__))); |
|---|
| 32 | typedef int di __attribute__ ((__mode__(__DI__))); |
|---|
| 33 | #else |
|---|
| 34 | typedef short v4hi __attribute__ ((vector_size (8))); |
|---|
| 35 | typedef int v2si __attribute__ ((vector_size (8))); |
|---|
| 36 | typedef int di __attribute__ ((vector_size (8))); |
|---|
| 37 | #endif |
|---|
| 38 | |
|---|
| 39 | #define ALIGN16(x) x __attribute__((aligned(16))) |
|---|
| 40 | typedef unsigned long aint; |
|---|
| 41 | |
|---|
| 42 | #else /* NO GCC */ |
|---|
| 43 | #include <mmintrin.h> |
|---|
| 44 | typedef __m64 v4hi; |
|---|
| 45 | typedef __m64 v2si; |
|---|
| 46 | typedef __m64 di; |
|---|
| 47 | |
|---|
| 48 | #define __builtin_ia32_pand(X,Y) _mm_and_si64(X,Y) |
|---|
| 49 | #define __builtin_ia32_pcmpeqw(X,Y) _mm_cmpeq_pi16(X,Y) |
|---|
| 50 | #define __builtin_ia32_packssdw(X,Y) _mm_packs_pi32(X,Y) |
|---|
| 51 | #define __builtin_ia32_punpcklwd(X,Y) _mm_unpacklo_pi16(X,Y) |
|---|
| 52 | #define __builtin_ia32_punpckhwd(X,Y) _mm_unpackhi_pi16(X,Y) |
|---|
| 53 | #define __builtin_ia32_paddd(X,Y) _mm_add_pi32(X,Y) |
|---|
| 54 | #define __builtin_ia32_paddsw(X,Y) _mm_adds_pi16(X,Y) |
|---|
| 55 | #define __builtin_ia32_pmulhw(X,Y) _mm_mulhi_pi16(X,Y) |
|---|
| 56 | #define __builtin_ia32_psllw(X,Y) _mm_slli_pi16(X,Y) |
|---|
| 57 | #define __builtin_ia32_emms() _mm_empty() |
|---|
| 58 | |
|---|
| 59 | #define ALIGN16(x) __declspec(align(16)) x |
|---|
| 60 | |
|---|
| 61 | /* FIXME: msvc++'s long in x86_64 isn't 8bytes? */ |
|---|
| 62 | typedef unsigned long aint; |
|---|
| 63 | #endif /* __GNUC__ */ |
|---|
| 64 | |
|---|
| 65 | #define MMX_ALIGN 8 |
|---|
| 66 | #endif /* __MMX__ */ |
|---|
| 67 | |
|---|
| 68 | #endif /* _X86_SIMD_SUPPORT_H_ */ |
|---|
| 69 | |
|---|
