diff options
Diffstat (limited to 'renderscript/clang-include')
26 files changed, 3168 insertions, 222 deletions
diff --git a/renderscript/clang-include/CMakeLists.txt b/renderscript/clang-include/CMakeLists.txt index 1faf92f..6e9cc68 100644 --- a/renderscript/clang-include/CMakeLists.txt +++ b/renderscript/clang-include/CMakeLists.txt @@ -1,16 +1,24 @@ set(files altivec.h + ammintrin.h avxintrin.h + avx2intrin.h + bmiintrin.h + bmi2intrin.h emmintrin.h float.h + fma4intrin.h + fmaintrin.h immintrin.h iso646.h limits.h + lzcntintrin.h mm3dnow.h mmintrin.h mm_malloc.h nmmintrin.h pmmintrin.h + popcntintrin.h smmintrin.h stdalign.h stdarg.h @@ -23,6 +31,10 @@ set(files wmmintrin.h x86intrin.h xmmintrin.h + xopintrin.h + cpuid.h + unwind.h + module.map ) set(output_dir ${LLVM_BINARY_DIR}/lib/clang/${CLANG_VERSION}/include) diff --git a/renderscript/clang-include/LICENSE.TXT b/renderscript/clang-include/LICENSE.TXT index 91895eb..6c224f8 100644 --- a/renderscript/clang-include/LICENSE.TXT +++ b/renderscript/clang-include/LICENSE.TXT @@ -4,7 +4,7 @@ LLVM Release License University of Illinois/NCSA Open Source License -Copyright (c) 2007-2011 University of Illinois at Urbana-Champaign. +Copyright (c) 2007-2012 University of Illinois at Urbana-Champaign. All rights reserved. Developed by: diff --git a/renderscript/clang-include/ammintrin.h b/renderscript/clang-include/ammintrin.h new file mode 100644 index 0000000..d87b9cd --- /dev/null +++ b/renderscript/clang-include/ammintrin.h @@ -0,0 +1,68 @@ +/*===---- ammintrin.h - SSE4a intrinsics -----------------------------------=== + * + * Permission is hereby granted, free of charge, to any person obtaining a copy + * of this software and associated documentation files (the "Software"), to deal + * in the Software without restriction, including without limitation the rights + * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell + * copies of the Software, and to permit persons to whom the Software is + * furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in + * all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE + * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN + * THE SOFTWARE. + * + *===-----------------------------------------------------------------------=== + */ + +#ifndef __AMMINTRIN_H +#define __AMMINTRIN_H + +#ifndef __SSE4A__ +#error "SSE4A instruction set not enabled" +#else + +#include <pmmintrin.h> + +#define _mm_extracti_si64(x, len, idx) \ + ((__m128i)__builtin_ia32_extrqi((__v2di)(__m128i)(x), \ + (char)(len), (char)(idx))) + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_extract_si64(__m128i __x, __m128i __y) +{ + return (__m128i)__builtin_ia32_extrq((__v2di)__x, (__v16qi)__y); +} + +#define _mm_inserti_si64(x, y, len, idx) \ + ((__m128i)__builtin_ia32_insertqi((__v2di)(__m128i)(x), \ + (__v2di)(__m128i)(y), \ + (char)(len), (char)(idx))) + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_insert_si64(__m128i __x, __m128i __y) +{ + return (__m128i)__builtin_ia32_insertq((__v2di)__x, (__v2di)__y); +} + +static __inline__ void __attribute__((__always_inline__, __nodebug__)) +_mm_stream_sd(double *__p, __m128d __a) +{ + __builtin_ia32_movntsd(__p, (__v2df)__a); +} + +static __inline__ void __attribute__((__always_inline__, __nodebug__)) +_mm_stream_ss(float *__p, __m128 __a) +{ + __builtin_ia32_movntss(__p, (__v4sf)__a); +} + +#endif /* __SSE4A__ */ + +#endif /* __AMMINTRIN_H */ diff --git a/renderscript/clang-include/avx2intrin.h b/renderscript/clang-include/avx2intrin.h new file mode 100644 index 0000000..2c53aed --- /dev/null +++ b/renderscript/clang-include/avx2intrin.h @@ -0,0 +1,1201 @@ +/*===---- avx2intrin.h - AVX2 intrinsics -----------------------------------=== + * + * Permission is hereby granted, free of charge, to any person obtaining a copy + * of this software and associated documentation files (the "Software"), to deal + * in the Software without restriction, including without limitation the rights + * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell + * copies of the Software, and to permit persons to whom the Software is + * furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in + * all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE + * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN + * THE SOFTWARE. + * + *===-----------------------------------------------------------------------=== + */ + +#ifndef __IMMINTRIN_H +#error "Never use <avx2intrin.h> directly; include <immintrin.h> instead." +#endif + +/* SSE4 Multiple Packed Sums of Absolute Difference. */ +#define _mm256_mpsadbw_epu8(X, Y, M) __builtin_ia32_mpsadbw256((X), (Y), (M)) + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_abs_epi8(__m256i a) +{ + return (__m256i)__builtin_ia32_pabsb256((__v32qi)a); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_abs_epi16(__m256i a) +{ + return (__m256i)__builtin_ia32_pabsw256((__v16hi)a); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_abs_epi32(__m256i a) +{ + return (__m256i)__builtin_ia32_pabsd256((__v8si)a); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_packs_epi16(__m256i a, __m256i b) +{ + return (__m256i)__builtin_ia32_packsswb256((__v16hi)a, (__v16hi)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_packs_epi32(__m256i a, __m256i b) +{ + return (__m256i)__builtin_ia32_packssdw256((__v8si)a, (__v8si)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_packus_epi16(__m256i a, __m256i b) +{ + return (__m256i)__builtin_ia32_packuswb256((__v16hi)a, (__v16hi)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_packus_epi32(__m256i __V1, __m256i __V2) +{ + return (__m256i) __builtin_ia32_packusdw256((__v8si)__V1, (__v8si)__V2); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_add_epi8(__m256i a, __m256i b) +{ + return (__m256i)((__v32qi)a + (__v32qi)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_add_epi16(__m256i a, __m256i b) +{ + return (__m256i)((__v16hi)a + (__v16hi)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_add_epi32(__m256i a, __m256i b) +{ + return (__m256i)((__v8si)a + (__v8si)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_add_epi64(__m256i a, __m256i b) +{ + return a + b; +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_adds_epi8(__m256i a, __m256i b) +{ + return (__m256i)__builtin_ia32_paddsb256((__v32qi)a, (__v32qi)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_adds_epi16(__m256i a, __m256i b) +{ + return (__m256i)__builtin_ia32_paddsw256((__v16hi)a, (__v16hi)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_adds_epu8(__m256i a, __m256i b) +{ + return (__m256i)__builtin_ia32_paddusb256((__v32qi)a, (__v32qi)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_adds_epu16(__m256i a, __m256i b) +{ + return (__m256i)__builtin_ia32_paddusw256((__v16hi)a, (__v16hi)b); +} + +#define _mm256_alignr_epi8(a, b, n) __extension__ ({ \ + __m256i __a = (a); \ + __m256i __b = (b); \ + (__m256i)__builtin_ia32_palignr256((__v32qi)__a, (__v32qi)__b, (n)); }) + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_and_si256(__m256i a, __m256i b) +{ + return a & b; +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_andnot_si256(__m256i a, __m256i b) +{ + return ~a & b; +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_avg_epu8(__m256i a, __m256i b) +{ + return (__m256i)__builtin_ia32_pavgb256((__v32qi)a, (__v32qi)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_avg_epu16(__m256i a, __m256i b) +{ + return (__m256i)__builtin_ia32_pavgw256((__v16hi)a, (__v16hi)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_blendv_epi8(__m256i __V1, __m256i __V2, __m256i __M) +{ + return (__m256i)__builtin_ia32_pblendvb256((__v32qi)__V1, (__v32qi)__V2, + (__v32qi)__M); +} + +#define _mm256_blend_epi16(V1, V2, M) __extension__ ({ \ + __m256i __V1 = (V1); \ + __m256i __V2 = (V2); \ + (__m256i)__builtin_ia32_pblendw256((__v16hi)__V1, (__v16hi)__V2, (M)); }) + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_cmpeq_epi8(__m256i a, __m256i b) +{ + return (__m256i)((__v32qi)a == (__v32qi)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_cmpeq_epi16(__m256i a, __m256i b) +{ + return (__m256i)((__v16hi)a == (__v16hi)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_cmpeq_epi32(__m256i a, __m256i b) +{ + return (__m256i)((__v8si)a == (__v8si)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_cmpeq_epi64(__m256i a, __m256i b) +{ + return (__m256i)(a == b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_cmpgt_epi8(__m256i a, __m256i b) +{ + return (__m256i)((__v32qi)a > (__v32qi)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_cmpgt_epi16(__m256i a, __m256i b) +{ + return (__m256i)((__v16hi)a > (__v16hi)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_cmpgt_epi32(__m256i a, __m256i b) +{ + return (__m256i)((__v8si)a > (__v8si)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_cmpgt_epi64(__m256i a, __m256i b) +{ + return (__m256i)(a > b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_hadd_epi16(__m256i a, __m256i b) +{ + return (__m256i)__builtin_ia32_phaddw256((__v16hi)a, (__v16hi)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_hadd_epi32(__m256i a, __m256i b) +{ + return (__m256i)__builtin_ia32_phaddd256((__v8si)a, (__v8si)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_hadds_epi16(__m256i a, __m256i b) +{ + return (__m256i)__builtin_ia32_phaddsw256((__v16hi)a, (__v16hi)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_hsub_epi16(__m256i a, __m256i b) +{ + return (__m256i)__builtin_ia32_phsubw256((__v16hi)a, (__v16hi)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_hsub_epi32(__m256i a, __m256i b) +{ + return (__m256i)__builtin_ia32_phsubd256((__v8si)a, (__v8si)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_hsubs_epi16(__m256i a, __m256i b) +{ + return (__m256i)__builtin_ia32_phsubsw256((__v16hi)a, (__v16hi)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_maddubs_epi16(__m256i a, __m256i b) +{ + return (__m256i)__builtin_ia32_pmaddubsw256((__v32qi)a, (__v32qi)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_madd_epi16(__m256i a, __m256i b) +{ + return (__m256i)__builtin_ia32_pmaddwd256((__v16hi)a, (__v16hi)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_max_epi8(__m256i a, __m256i b) +{ + return (__m256i)__builtin_ia32_pmaxsb256((__v32qi)a, (__v32qi)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_max_epi16(__m256i a, __m256i b) +{ + return (__m256i)__builtin_ia32_pmaxsw256((__v16hi)a, (__v16hi)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_max_epi32(__m256i a, __m256i b) +{ + return (__m256i)__builtin_ia32_pmaxsd256((__v8si)a, (__v8si)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_max_epu8(__m256i a, __m256i b) +{ + return (__m256i)__builtin_ia32_pmaxub256((__v32qi)a, (__v32qi)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_max_epu16(__m256i a, __m256i b) +{ + return (__m256i)__builtin_ia32_pmaxuw256((__v16hi)a, (__v16hi)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_max_epu32(__m256i a, __m256i b) +{ + return (__m256i)__builtin_ia32_pmaxud256((__v8si)a, (__v8si)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_min_epi8(__m256i a, __m256i b) +{ + return (__m256i)__builtin_ia32_pminsb256((__v32qi)a, (__v32qi)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_min_epi16(__m256i a, __m256i b) +{ + return (__m256i)__builtin_ia32_pminsw256((__v16hi)a, (__v16hi)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_min_epi32(__m256i a, __m256i b) +{ + return (__m256i)__builtin_ia32_pminsd256((__v8si)a, (__v8si)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_min_epu8(__m256i a, __m256i b) +{ + return (__m256i)__builtin_ia32_pminub256((__v32qi)a, (__v32qi)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_min_epu16(__m256i a, __m256i b) +{ + return (__m256i)__builtin_ia32_pminuw256 ((__v16hi)a, (__v16hi)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_min_epu32(__m256i a, __m256i b) +{ + return (__m256i)__builtin_ia32_pminud256((__v8si)a, (__v8si)b); +} + +static __inline__ int __attribute__((__always_inline__, __nodebug__)) +_mm256_movemask_epi8(__m256i a) +{ + return __builtin_ia32_pmovmskb256((__v32qi)a); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_cvtepi8_epi16(__m128i __V) +{ + return (__m256i)__builtin_ia32_pmovsxbw256((__v16qi)__V); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_cvtepi8_epi32(__m128i __V) +{ + return (__m256i)__builtin_ia32_pmovsxbd256((__v16qi)__V); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_cvtepi8_epi64(__m128i __V) +{ + return (__m256i)__builtin_ia32_pmovsxbq256((__v16qi)__V); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_cvtepi16_epi32(__m128i __V) +{ + return (__m256i)__builtin_ia32_pmovsxwd256((__v8hi)__V); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_cvtepi16_epi64(__m128i __V) +{ + return (__m256i)__builtin_ia32_pmovsxwq256((__v8hi)__V); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_cvtepi32_epi64(__m128i __V) +{ + return (__m256i)__builtin_ia32_pmovsxdq256((__v4si)__V); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_cvtepu8_epi16(__m128i __V) +{ + return (__m256i)__builtin_ia32_pmovzxbw256((__v16qi)__V); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_cvtepu8_epi32(__m128i __V) +{ + return (__m256i)__builtin_ia32_pmovzxbd256((__v16qi)__V); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_cvtepu8_epi64(__m128i __V) +{ + return (__m256i)__builtin_ia32_pmovzxbq256((__v16qi)__V); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_cvtepu16_epi32(__m128i __V) +{ + return (__m256i)__builtin_ia32_pmovzxwd256((__v8hi)__V); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_cvtepu16_epi64(__m128i __V) +{ + return (__m256i)__builtin_ia32_pmovzxwq256((__v8hi)__V); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_cvtepu32_epi64(__m128i __V) +{ + return (__m256i)__builtin_ia32_pmovzxdq256((__v4si)__V); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_mul_epi32(__m256i a, __m256i b) +{ + return (__m256i)__builtin_ia32_pmuldq256((__v8si)a, (__v8si)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_mulhrs_epi16(__m256i a, __m256i b) +{ + return (__m256i)__builtin_ia32_pmulhrsw256((__v16hi)a, (__v16hi)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_mulhi_epu16(__m256i a, __m256i b) +{ + return (__m256i)__builtin_ia32_pmulhuw256((__v16hi)a, (__v16hi)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_mulhi_epi16(__m256i a, __m256i b) +{ + return (__m256i)__builtin_ia32_pmulhw256((__v16hi)a, (__v16hi)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_mullo_epi16(__m256i a, __m256i b) +{ + return (__m256i)((__v16hi)a * (__v16hi)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_mullo_epi32 (__m256i a, __m256i b) +{ + return (__m256i)((__v8si)a * (__v8si)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_mul_epu32(__m256i a, __m256i b) +{ + return __builtin_ia32_pmuludq256((__v8si)a, (__v8si)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_or_si256(__m256i a, __m256i b) +{ + return a | b; +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_sad_epu8(__m256i a, __m256i b) +{ + return __builtin_ia32_psadbw256((__v32qi)a, (__v32qi)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_shuffle_epi8(__m256i a, __m256i b) +{ + return (__m256i)__builtin_ia32_pshufb256((__v32qi)a, (__v32qi)b); +} + +#define _mm256_shuffle_epi32(a, imm) __extension__ ({ \ + __m256i __a = (a); \ + (__m256i)__builtin_shufflevector((__v8si)__a, (__v8si)_mm256_set1_epi32(0), \ + (imm) & 0x3, ((imm) & 0xc) >> 2, \ + ((imm) & 0x30) >> 4, ((imm) & 0xc0) >> 6, \ + 4 + (((imm) & 0x03) >> 0), \ + 4 + (((imm) & 0x0c) >> 2), \ + 4 + (((imm) & 0x30) >> 4), \ + 4 + (((imm) & 0xc0) >> 6)); }) + +#define _mm256_shufflehi_epi16(a, imm) __extension__ ({ \ + __m256i __a = (a); \ + (__m256i)__builtin_shufflevector((__v16hi)__a, (__v16hi)_mm256_set1_epi16(0), \ + 0, 1, 2, 3, \ + 4 + (((imm) & 0x03) >> 0), \ + 4 + (((imm) & 0x0c) >> 2), \ + 4 + (((imm) & 0x30) >> 4), \ + 4 + (((imm) & 0xc0) >> 6), \ + 8, 9, 10, 11, \ + 12 + (((imm) & 0x03) >> 0), \ + 12 + (((imm) & 0x0c) >> 2), \ + 12 + (((imm) & 0x30) >> 4), \ + 12 + (((imm) & 0xc0) >> 6)); }) + +#define _mm256_shufflelo_epi16(a, imm) __extension__ ({ \ + __m256i __a = (a); \ + (__m256i)__builtin_shufflevector((__v16hi)__a, (__v16hi)_mm256_set1_epi16(0), \ + (imm) & 0x3,((imm) & 0xc) >> 2, \ + ((imm) & 0x30) >> 4, ((imm) & 0xc0) >> 6, \ + 4, 5, 6, 7, \ + 8 + (((imm) & 0x03) >> 0), \ + 8 + (((imm) & 0x0c) >> 2), \ + 8 + (((imm) & 0x30) >> 4), \ + 8 + (((imm) & 0xc0) >> 6), \ + 12, 13, 14, 15); }) + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_sign_epi8(__m256i a, __m256i b) +{ + return (__m256i)__builtin_ia32_psignb256((__v32qi)a, (__v32qi)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_sign_epi16(__m256i a, __m256i b) +{ + return (__m256i)__builtin_ia32_psignw256((__v16hi)a, (__v16hi)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_sign_epi32(__m256i a, __m256i b) +{ + return (__m256i)__builtin_ia32_psignd256((__v8si)a, (__v8si)b); +} + +#define _mm256_slli_si256(a, count) __extension__ ({ \ + __m256i __a = (a); \ + (__m256i)__builtin_ia32_pslldqi256(__a, (count)*8); }) + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_slli_epi16(__m256i a, int count) +{ + return (__m256i)__builtin_ia32_psllwi256((__v16hi)a, count); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_sll_epi16(__m256i a, __m128i count) +{ + return (__m256i)__builtin_ia32_psllw256((__v16hi)a, (__v8hi)count); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_slli_epi32(__m256i a, int count) +{ + return (__m256i)__builtin_ia32_pslldi256((__v8si)a, count); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_sll_epi32(__m256i a, __m128i count) +{ + return (__m256i)__builtin_ia32_pslld256((__v8si)a, (__v4si)count); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_slli_epi64(__m256i a, int count) +{ + return __builtin_ia32_psllqi256(a, count); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_sll_epi64(__m256i a, __m128i count) +{ + return __builtin_ia32_psllq256(a, count); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_srai_epi16(__m256i a, int count) +{ + return (__m256i)__builtin_ia32_psrawi256((__v16hi)a, count); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_sra_epi16(__m256i a, __m128i count) +{ + return (__m256i)__builtin_ia32_psraw256((__v16hi)a, (__v8hi)count); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_srai_epi32(__m256i a, int count) +{ + return (__m256i)__builtin_ia32_psradi256((__v8si)a, count); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_sra_epi32(__m256i a, __m128i count) +{ + return (__m256i)__builtin_ia32_psrad256((__v8si)a, (__v4si)count); +} + +#define _mm256_srli_si256(a, count) __extension__ ({ \ + __m256i __a = (a); \ + (__m256i)__builtin_ia32_psrldqi256(__a, (count)*8); }) + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_srli_epi16(__m256i a, int count) +{ + return (__m256i)__builtin_ia32_psrlwi256((__v16hi)a, count); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_srl_epi16(__m256i a, __m128i count) +{ + return (__m256i)__builtin_ia32_psrlw256((__v16hi)a, (__v8hi)count); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_srli_epi32(__m256i a, int count) +{ + return (__m256i)__builtin_ia32_psrldi256((__v8si)a, count); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_srl_epi32(__m256i a, __m128i count) +{ + return (__m256i)__builtin_ia32_psrld256((__v8si)a, (__v4si)count); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_srli_epi64(__m256i a, int count) +{ + return __builtin_ia32_psrlqi256(a, count); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_srl_epi64(__m256i a, __m128i count) +{ + return __builtin_ia32_psrlq256(a, count); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_sub_epi8(__m256i a, __m256i b) +{ + return (__m256i)((__v32qi)a - (__v32qi)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_sub_epi16(__m256i a, __m256i b) +{ + return (__m256i)((__v16hi)a - (__v16hi)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_sub_epi32(__m256i a, __m256i b) +{ + return (__m256i)((__v8si)a - (__v8si)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_sub_epi64(__m256i a, __m256i b) +{ + return a - b; +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_subs_epi8(__m256i a, __m256i b) +{ + return (__m256i)__builtin_ia32_psubsb256((__v32qi)a, (__v32qi)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_subs_epi16(__m256i a, __m256i b) +{ + return (__m256i)__builtin_ia32_psubsw256((__v16hi)a, (__v16hi)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_subs_epu8(__m256i a, __m256i b) +{ + return (__m256i)__builtin_ia32_psubusb256((__v32qi)a, (__v32qi)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_subs_epu16(__m256i a, __m256i b) +{ + return (__m256i)__builtin_ia32_psubusw256((__v16hi)a, (__v16hi)b); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_unpackhi_epi8(__m256i a, __m256i b) +{ + return (__m256i)__builtin_shufflevector((__v32qi)a, (__v32qi)b, 8, 32+8, 9, 32+9, 10, 32+10, 11, 32+11, 12, 32+12, 13, 32+13, 14, 32+14, 15, 32+15, 24, 32+24, 25, 32+25, 26, 32+26, 27, 32+27, 28, 32+28, 29, 32+29, 30, 32+30, 31, 32+31); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_unpackhi_epi16(__m256i a, __m256i b) +{ + return (__m256i)__builtin_shufflevector((__v16hi)a, (__v16hi)b, 4, 16+4, 5, 16+5, 6, 16+6, 7, 16+7, 12, 16+12, 13, 16+13, 14, 16+14, 15, 16+15); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_unpackhi_epi32(__m256i a, __m256i b) +{ + return (__m256i)__builtin_shufflevector((__v8si)a, (__v8si)b, 2, 8+2, 3, 8+3, 6, 8+6, 7, 8+7); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_unpackhi_epi64(__m256i a, __m256i b) +{ + return (__m256i)__builtin_shufflevector(a, b, 1, 4+1, 3, 4+3); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_unpacklo_epi8(__m256i a, __m256i b) +{ + return (__m256i)__builtin_shufflevector((__v32qi)a, (__v32qi)b, 0, 32+0, 1, 32+1, 2, 32+2, 3, 32+3, 4, 32+4, 5, 32+5, 6, 32+6, 7, 32+7, 16, 32+16, 17, 32+17, 18, 32+18, 19, 32+19, 20, 32+20, 21, 32+21, 22, 32+22, 23, 32+23); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_unpacklo_epi16(__m256i a, __m256i b) +{ + return (__m256i)__builtin_shufflevector((__v16hi)a, (__v16hi)b, 0, 16+0, 1, 16+1, 2, 16+2, 3, 16+3, 8, 16+8, 9, 16+9, 10, 16+10, 11, 16+11); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_unpacklo_epi32(__m256i a, __m256i b) +{ + return (__m256i)__builtin_shufflevector((__v8si)a, (__v8si)b, 0, 8+0, 1, 8+1, 4, 8+4, 5, 8+5); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_unpacklo_epi64(__m256i a, __m256i b) +{ + return (__m256i)__builtin_shufflevector(a, b, 0, 4+0, 2, 4+2); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_xor_si256(__m256i a, __m256i b) +{ + return a ^ b; +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_stream_load_si256(__m256i *__V) +{ + return (__m256i)__builtin_ia32_movntdqa256((__v4di *)__V); +} + +static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) +_mm_broadcastss_ps(__m128 __X) +{ + return (__m128)__builtin_ia32_vbroadcastss_ps((__v4sf)__X); +} + +static __inline__ __m256 __attribute__((__always_inline__, __nodebug__)) +_mm256_broadcastss_ps(__m128 __X) +{ + return (__m256)__builtin_ia32_vbroadcastss_ps256((__v4sf)__X); +} + +static __inline__ __m256d __attribute__((__always_inline__, __nodebug__)) +_mm256_broadcastsd_pd(__m128d __X) +{ + return (__m256d)__builtin_ia32_vbroadcastsd_pd256((__v2df)__X); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm_broadcastsi128_si256(__m128i const *a) +{ + return (__m256i)__builtin_ia32_vbroadcastsi256(a); +} + +#define _mm_blend_epi32(V1, V2, M) __extension__ ({ \ + __m128i __V1 = (V1); \ + __m128i __V2 = (V2); \ + (__m128i)__builtin_ia32_pblendd128((__v4si)__V1, (__v4si)__V2, (M)); }) + +#define _mm256_blend_epi32(V1, V2, M) __extension__ ({ \ + __m256i __V1 = (V1); \ + __m256i __V2 = (V2); \ + (__m256i)__builtin_ia32_pblendd256((__v8si)__V1, (__v8si)__V2, (M)); }) + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_broadcastb_epi8(__m128i __X) +{ + return (__m256i)__builtin_ia32_pbroadcastb256((__v16qi)__X); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_broadcastw_epi16(__m128i __X) +{ + return (__m256i)__builtin_ia32_pbroadcastw256((__v8hi)__X); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_broadcastd_epi32(__m128i __X) +{ + return (__m256i)__builtin_ia32_pbroadcastd256((__v4si)__X); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_broadcastq_epi64(__m128i __X) +{ + return (__m256i)__builtin_ia32_pbroadcastq256(__X); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_broadcastb_epi8(__m128i __X) +{ + return (__m128i)__builtin_ia32_pbroadcastb128((__v16qi)__X); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_broadcastw_epi16(__m128i __X) +{ + return (__m128i)__builtin_ia32_pbroadcastw128((__v8hi)__X); +} + + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_broadcastd_epi32(__m128i __X) +{ + return (__m128i)__builtin_ia32_pbroadcastd128((__v4si)__X); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_broadcastq_epi64(__m128i __X) +{ + return (__m128i)__builtin_ia32_pbroadcastq128(__X); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_permutevar8x32_epi32(__m256i a, __m256i b) +{ + return (__m256i)__builtin_ia32_permvarsi256((__v8si)a, (__v8si)b); +} + +#define _mm256_permute4x64_pd(V, M) __extension__ ({ \ + __m256d __V = (V); \ + (__m256d)__builtin_shufflevector((__v4df)__V, (__v4df) _mm256_setzero_pd(), \ + (M) & 0x3, ((M) & 0xc) >> 2, \ + ((M) & 0x30) >> 4, ((M) & 0xc0) >> 6); }) + +static __inline__ __m256 __attribute__((__always_inline__, __nodebug__)) +_mm256_permutevar8x32_ps(__m256 a, __m256 b) +{ + return (__m256)__builtin_ia32_permvarsf256((__v8sf)a, (__v8sf)b); +} + +#define _mm256_permute4x64_epi64(V, M) __extension__ ({ \ + __m256i __V = (V); \ + (__m256i)__builtin_shufflevector((__v4di)__V, (__v4di) _mm256_setzero_si256(), \ + (M) & 0x3, ((M) & 0xc) >> 2, \ + ((M) & 0x30) >> 4, ((M) & 0xc0) >> 6); }) + +#define _mm256_permute2x128_si256(V1, V2, M) __extension__ ({ \ + __m256i __V1 = (V1); \ + __m256i __V2 = (V2); \ + (__m256i)__builtin_ia32_permti256(__V1, __V2, (M)); }) + +#define _mm256_extracti128_si256(A, O) __extension__ ({ \ + __m256i __A = (A); \ + (__m128i)__builtin_ia32_extract128i256(__A, (O)); }) + +#define _mm256_inserti128_si256(V1, V2, O) __extension__ ({ \ + __m256i __V1 = (V1); \ + __m128i __V2 = (V2); \ + (__m256i)__builtin_ia32_insert128i256(__V1, __V2, (O)); }) + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_maskload_epi32(int const *__X, __m256i __M) +{ + return (__m256i)__builtin_ia32_maskloadd256((const __v8si *)__X, (__v8si)__M); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_maskload_epi64(long long const *__X, __m256i __M) +{ + return (__m256i)__builtin_ia32_maskloadq256((const __v4di *)__X, __M); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_maskload_epi32(int const *__X, __m128i __M) +{ + return (__m128i)__builtin_ia32_maskloadd((const __v4si *)__X, (__v4si)__M); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_maskload_epi64(long long const *__X, __m128i __M) +{ + return (__m128i)__builtin_ia32_maskloadq((const __v2di *)__X, (__v2di)__M); +} + +static __inline__ void __attribute__((__always_inline__, __nodebug__)) +_mm256_maskstore_epi32(int *__X, __m256i __M, __m256i __Y) +{ + __builtin_ia32_maskstored256((__v8si *)__X, (__v8si)__M, (__v8si)__Y); +} + +static __inline__ void __attribute__((__always_inline__, __nodebug__)) +_mm256_maskstore_epi64(long long *__X, __m256i __M, __m256i __Y) +{ + __builtin_ia32_maskstoreq256((__v4di *)__X, __M, __Y); +} + +static __inline__ void __attribute__((__always_inline__, __nodebug__)) +_mm_maskstore_epi32(int *__X, __m128i __M, __m128i __Y) +{ + __builtin_ia32_maskstored((__v4si *)__X, (__v4si)__M, (__v4si)__Y); +} + +static __inline__ void __attribute__((__always_inline__, __nodebug__)) +_mm_maskstore_epi64(long long *__X, __m128i __M, __m128i __Y) +{ + __builtin_ia32_maskstoreq(( __v2di *)__X, __M, __Y); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_sllv_epi32(__m256i __X, __m256i __Y) +{ + return (__m256i)__builtin_ia32_psllv8si((__v8si)__X, (__v8si)__Y); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_sllv_epi32(__m128i __X, __m128i __Y) +{ + return (__m128i)__builtin_ia32_psllv4si((__v4si)__X, (__v4si)__Y); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_sllv_epi64(__m256i __X, __m256i __Y) +{ + return (__m256i)__builtin_ia32_psllv4di(__X, __Y); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_sllv_epi64(__m128i __X, __m128i __Y) +{ + return (__m128i)__builtin_ia32_psllv2di(__X, __Y); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_srav_epi32(__m256i __X, __m256i __Y) +{ + return (__m256i)__builtin_ia32_psrav8si((__v8si)__X, (__v8si)__Y); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_srav_epi32(__m128i __X, __m128i __Y) +{ + return (__m128i)__builtin_ia32_psrav4si((__v4si)__X, (__v4si)__Y); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_srlv_epi32(__m256i __X, __m256i __Y) +{ + return (__m256i)__builtin_ia32_psrlv8si((__v8si)__X, (__v8si)__Y); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_srlv_epi32(__m128i __X, __m128i __Y) +{ + return (__m128i)__builtin_ia32_psrlv4si((__v4si)__X, (__v4si)__Y); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_srlv_epi64(__m256i __X, __m256i __Y) +{ + return (__m256i)__builtin_ia32_psrlv4di(__X, __Y); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_srlv_epi64(__m128i __X, __m128i __Y) +{ + return (__m128i)__builtin_ia32_psrlv2di(__X, __Y); +} + +#define _mm_mask_i32gather_pd(a, m, i, mask, s) __extension__ ({ \ + __m128d __a = (a); \ + double const *__m = (m); \ + __m128i __i = (i); \ + __m128d __mask = (mask); \ + (__m128d)__builtin_ia32_gatherd_pd((__v2df)__a, (const __v2df *)__m, \ + (__v4si)__i, (__v2df)__mask, (s)); }) + +#define _mm256_mask_i32gather_pd(a, m, i, mask, s) __extension__ ({ \ + __m256d __a = (a); \ + double const *__m = (m); \ + __m128i __i = (i); \ + __m256d __mask = (mask); \ + (__m256d)__builtin_ia32_gatherd_pd256((__v4df)__a, (const __v4df *)__m, \ + (__v4si)__i, (__v4df)__mask, (s)); }) + +#define _mm_mask_i64gather_pd(a, m, i, mask, s) __extension__ ({ \ + __m128d __a = (a); \ + double const *__m = (m); \ + __m128i __i = (i); \ + __m128d __mask = (mask); \ + (__m128d)__builtin_ia32_gatherq_pd((__v2df)__a, (const __v2df *)__m, \ + (__v2di)__i, (__v2df)__mask, (s)); }) + +#define _mm256_mask_i64gather_pd(a, m, i, mask, s) __extension__ ({ \ + __m256d __a = (a); \ + double const *__m = (m); \ + __m256i __i = (i); \ + __m256d __mask = (mask); \ + (__m256d)__builtin_ia32_gatherq_pd256((__v4df)__a, (const __v4df *)__m, \ + (__v4di)__i, (__v4df)__mask, (s)); }) + +#define _mm_mask_i32gather_ps(a, m, i, mask, s) __extension__ ({ \ + __m128 __a = (a); \ + float const *__m = (m); \ + __m128i __i = (i); \ + __m128 __mask = (mask); \ + (__m128)__builtin_ia32_gatherd_ps((__v4sf)__a, (const __v4sf *)__m, \ + (__v4si)__i, (__v4sf)__mask, (s)); }) + +#define _mm256_mask_i32gather_ps(a, m, i, mask, s) __extension__ ({ \ + __m256 __a = (a); \ + float const *__m = (m); \ + __m256i __i = (i); \ + __m256 __mask = (mask); \ + (__m256)__builtin_ia32_gatherd_ps256((__v8sf)__a, (const __v8sf *)__m, \ + (__v8si)__i, (__v8sf)__mask, (s)); }) + +#define _mm_mask_i64gather_ps(a, m, i, mask, s) __extension__ ({ \ + __m128 __a = (a); \ + float const *__m = (m); \ + __m128i __i = (i); \ + __m128 __mask = (mask); \ + (__m128)__builtin_ia32_gatherq_ps((__v4sf)__a, (const __v4sf *)__m, \ + (__v2di)__i, (__v4sf)__mask, (s)); }) + +#define _mm256_mask_i64gather_ps(a, m, i, mask, s) __extension__ ({ \ + __m128 __a = (a); \ + float const *__m = (m); \ + __m256i __i = (i); \ + __m128 __mask = (mask); \ + (__m128)__builtin_ia32_gatherq_ps256((__v4sf)__a, (const __v4sf *)__m, \ + (__v4di)__i, (__v4sf)__mask, (s)); }) + +#define _mm_mask_i32gather_epi32(a, m, i, mask, s) __extension__ ({ \ + __m128i __a = (a); \ + int const *__m = (m); \ + __m128i __i = (i); \ + __m128i __mask = (mask); \ + (__m128i)__builtin_ia32_gatherd_d((__v4si)__a, (const __v4si *)__m, \ + (__v4si)__i, (__v4si)__mask, (s)); }) + +#define _mm256_mask_i32gather_epi32(a, m, i, mask, s) __extension__ ({ \ + __m256i __a = (a); \ + int const *__m = (m); \ + __m256i __i = (i); \ + __m256i __mask = (mask); \ + (__m256i)__builtin_ia32_gatherd_d256((__v8si)__a, (const __v8si *)__m, \ + (__v8si)__i, (__v8si)__mask, (s)); }) + +#define _mm_mask_i64gather_epi32(a, m, i, mask, s) __extension__ ({ \ + __m128i __a = (a); \ + int const *__m = (m); \ + __m128i __i = (i); \ + __m128i __mask = (mask); \ + (__m128i)__builtin_ia32_gatherq_d((__v4si)__a, (const __v4si *)__m, \ + (__v2di)__i, (__v4si)__mask, (s)); }) + +#define _mm256_mask_i64gather_epi32(a, m, i, mask, s) __extension__ ({ \ + __m128i __a = (a); \ + int const *__m = (m); \ + __m256i __i = (i); \ + __m128i __mask = (mask); \ + (__m128i)__builtin_ia32_gatherq_d256((__v4si)__a, (const __v4si *)__m, \ + (__v4di)__i, (__v4si)__mask, (s)); }) + +#define _mm_mask_i32gather_epi64(a, m, i, mask, s) __extension__ ({ \ + __m128i __a = (a); \ + int const *__m = (m); \ + __m128i __i = (i); \ + __m128i __mask = (mask); \ + (__m128i)__builtin_ia32_gatherd_q((__v2di)__a, (const __v2di *)__m, \ + (__v4si)__i, (__v2di)__mask, (s)); }) + +#define _mm256_mask_i32gather_epi64(a, m, i, mask, s) __extension__ ({ \ + __m256i __a = (a); \ + int const *__m = (m); \ + __m128i __i = (i); \ + __m256i __mask = (mask); \ + (__m256i)__builtin_ia32_gatherd_q256((__v4di)__a, (const __v4di *)__m, \ + (__v4si)__i, (__v4di)__mask, (s)); }) + +#define _mm_mask_i64gather_epi64(a, m, i, mask, s) __extension__ ({ \ + __m128i __a = (a); \ + int const *__m = (m); \ + __m128i __i = (i); \ + __m128i __mask = (mask); \ + (__m128i)__builtin_ia32_gatherq_q((__v2di)__a, (const __v2di *)__m, \ + (__v2di)__i, (__v2di)__mask, (s)); }) + +#define _mm256_mask_i64gather_epi64(a, m, i, mask, s) __extension__ ({ \ + __m256i __a = (a); \ + int const *__m = (m); \ + __m256i __i = (i); \ + __m256i __mask = (mask); \ + (__m256i)__builtin_ia32_gatherq_q256((__v4di)__a, (const __v4di *)__m, \ + (__v4di)__i, (__v4di)__mask, (s)); }) + +#define _mm_i32gather_pd(m, i, s) __extension__ ({ \ + double const *__m = (m); \ + __m128i __i = (i); \ + (__m128d)__builtin_ia32_gatherd_pd((__v2df)_mm_setzero_pd(), \ + (const __v2df *)__m, (__v4si)__i, \ + (__v2df)_mm_set1_pd((double)(long long int)-1), (s)); }) + +#define _mm256_i32gather_pd(m, i, s) __extension__ ({ \ + double const *__m = (m); \ + __m128i __i = (i); \ + (__m256d)__builtin_ia32_gatherd_pd256((__v4df)_mm256_setzero_pd(), \ + (const __v4df *)__m, (__v4si)__i, \ + (__v4df)_mm256_set1_pd((double)(long long int)-1), (s)); }) + +#define _mm_i64gather_pd(m, i, s) __extension__ ({ \ + double const *__m = (m); \ + __m128i __i = (i); \ + (__m128d)__builtin_ia32_gatherq_pd((__v2df)_mm_setzero_pd(), \ + (const __v2df *)__m, (__v2di)__i, \ + (__v2df)_mm_set1_pd((double)(long long int)-1), (s)); }) + +#define _mm256_i64gather_pd(m, i, s) __extension__ ({ \ + double const *__m = (m); \ + __m256i __i = (i); \ + (__m256d)__builtin_ia32_gatherq_pd256((__v4df)_mm256_setzero_pd(), \ + (const __v4df *)__m, (__v4di)__i, \ + (__v4df)_mm256_set1_pd((double)(long long int)-1), (s)); }) + +#define _mm_i32gather_ps(m, i, s) __extension__ ({ \ + float const *__m = (m); \ + __m128i __i = (i); \ + (__m128)__builtin_ia32_gatherd_ps((__v4sf)_mm_setzero_ps(), \ + (const __v4sf *)__m, (__v4si)__i, \ + (__v4sf)_mm_set1_ps((float)(int)-1), (s)); }) + +#define _mm256_i32gather_ps(m, i, s) __extension__ ({ \ + float const *__m = (m); \ + __m256i __i = (i); \ + (__m256)__builtin_ia32_gatherd_ps256((__v8sf)_mm256_setzero_ps(), \ + (const __v8sf *)__m, (__v8si)__i, \ + (__v8sf)_mm256_set1_ps((float)(int)-1), (s)); }) + +#define _mm_i64gather_ps(m, i, s) __extension__ ({ \ + float const *__m = (m); \ + __m128i __i = (i); \ + (__m128)__builtin_ia32_gatherq_ps((__v4sf)_mm_setzero_ps(), \ + (const __v4sf *)__m, (__v2di)__i, \ + (__v4sf)_mm_set1_ps((float)(int)-1), (s)); }) + +#define _mm256_i64gather_ps(m, i, s) __extension__ ({ \ + float const *__m = (m); \ + __m256i __i = (i); \ + (__m128)__builtin_ia32_gatherq_ps256((__v4sf)_mm_setzero_ps(), \ + (const __v4sf *)__m, (__v4di)__i, \ + (__v4sf)_mm_set1_ps((float)(int)-1), (s)); }) + +#define _mm_i32gather_epi32(m, i, s) __extension__ ({ \ + int const *__m = (m); \ + __m128i __i = (i); \ + (__m128i)__builtin_ia32_gatherd_d((__v4si)_mm_setzero_si128(), \ + (const __v4si *)__m, (__v4si)__i, \ + (__v4si)_mm_set1_epi32(-1), (s)); }) + +#define _mm256_i32gather_epi32(m, i, s) __extension__ ({ \ + int const *__m = (m); \ + __m256i __i = (i); \ + (__m256i)__builtin_ia32_gatherd_d256((__v8si)_mm256_setzero_si256(), \ + (const __v8si *)__m, (__v8si)__i, \ + (__v8si)_mm256_set1_epi32(-1), (s)); }) + +#define _mm_i64gather_epi32(m, i, s) __extension__ ({ \ + int const *__m = (m); \ + __m128i __i = (i); \ + (__m128i)__builtin_ia32_gatherq_d((__v4si)_mm_setzero_si128(), \ + (const __v4si *)__m, (__v2di)__i, \ + (__v4si)_mm_set1_epi32(-1), (s)); }) + +#define _mm256_i64gather_epi32(m, i, s) __extension__ ({ \ + int const *__m = (m); \ + __m256i __i = (i); \ + (__m128i)__builtin_ia32_gatherq_d256((__v4si)_mm_setzero_si128(), \ + (const __v4si *)__m, (__v4di)__i, \ + (__v4si)_mm_set1_epi32(-1), (s)); }) + +#define _mm_i32gather_epi64(m, i, s) __extension__ ({ \ + int const *__m = (m); \ + __m128i __i = (i); \ + (__m128i)__builtin_ia32_gatherd_q((__v2di)_mm_setzero_si128(), \ + (const __v2di *)__m, (__v4si)__i, \ + (__v2di)_mm_set1_epi64x(-1), (s)); }) + +#define _mm256_i32gather_epi64(m, i, s) __extension__ ({ \ + int const *__m = (m); \ + __m128i __i = (i); \ + (__m256i)__builtin_ia32_gatherd_q256((__v4di)_mm256_setzero_si256(), \ + (const __v4di *)__m, (__v4si)__i, \ + (__v4di)_mm256_set1_epi64x(-1), (s)); }) + +#define _mm_i64gather_epi64(m, i, s) __extension__ ({ \ + int const *__m = (m); \ + __m128i __i = (i); \ + (__m128i)__builtin_ia32_gatherq_q((__v2di)_mm_setzero_si128(), \ + (const __v2di *)__m, (__v2di)__i, \ + (__v2di)_mm_set1_epi64x(-1), (s)); }) + +#define _mm256_i64gather_epi64(m, i, s) __extension__ ({ \ + int const *__m = (m); \ + __m256i __i = (i); \ + (__m256i)__builtin_ia32_gatherq_q256((__v4di)_mm256_setzero_si256(), \ + (const __v4di *)__m, (__v4di)__i, \ + (__v4di)_mm256_set1_epi64x(-1), (s)); }) diff --git a/renderscript/clang-include/avxintrin.h b/renderscript/clang-include/avxintrin.h index 0a0d2e4..ee7f835 100644 --- a/renderscript/clang-include/avxintrin.h +++ b/renderscript/clang-include/avxintrin.h @@ -145,17 +145,13 @@ _mm256_rcp_ps(__m256 a) return (__m256)__builtin_ia32_rcpps256((__v8sf)a); } -static __inline __m256d __attribute__((__always_inline__, __nodebug__)) -_mm256_round_pd(__m256d v, const int m) -{ - return (__m256d)__builtin_ia32_roundpd256((__v4df)v, m); -} +#define _mm256_round_pd(V, M) __extension__ ({ \ + __m256d __V = (V); \ + (__m256d)__builtin_ia32_roundpd256((__v4df)__V, (M)); }) -static __inline __m256 __attribute__((__always_inline__, __nodebug__)) -_mm256_round_ps(__m256 v, const int m) -{ - return (__m256)__builtin_ia32_roundps256((__v8sf)v, m); -} +#define _mm256_round_ps(V, M) __extension__ ({ \ + __m256 __V = (V); \ + (__m256)__builtin_ia32_roundps256((__v8sf)__V, (M)); }) #define _mm256_ceil_pd(V) _mm256_round_pd((V), _MM_FROUND_CEIL) #define _mm256_floor_pd(V) _mm256_round_pd((V), _MM_FROUND_FLOOR) @@ -262,60 +258,59 @@ _mm256_permutevar_ps(__m256 a, __m256i c) (__v8si)c); } -static __inline __m128d __attribute__((__always_inline__, __nodebug__)) -_mm_permute_pd(__m128d a, const int c) -{ - return (__m128d)__builtin_ia32_vpermilpd((__v2df)a, c); -} - -static __inline __m256d __attribute__((__always_inline__, __nodebug__)) -_mm256_permute_pd(__m256d a, const int c) -{ - return (__m256d)__builtin_ia32_vpermilpd256((__v4df)a, c); -} - -static __inline __m128 __attribute__((__always_inline__, __nodebug__)) -_mm_permute_ps(__m128 a, const int c) -{ - return (__m128)__builtin_ia32_vpermilps((__v4sf)a, c); -} - -static __inline __m256 __attribute__((__always_inline__, __nodebug__)) -_mm256_permute_ps(__m256 a, const int c) -{ - return (__m256)__builtin_ia32_vpermilps256((__v8sf)a, c); -} - -static __inline __m256d __attribute__((__always_inline__, __nodebug__)) -_mm256_permute2f128_pd(__m256d a, __m256d b, const int c) -{ - return (__m256d)__builtin_ia32_vperm2f128_pd256((__v4df)a, (__v4df)b, c); -} - -static __inline __m256 __attribute__((__always_inline__, __nodebug__)) -_mm256_permute2f128_ps(__m256 a, __m256 b, const int c) -{ - return (__m256)__builtin_ia32_vperm2f128_ps256((__v8sf)a, (__v8sf)b, c); -} - -static __inline __m256i __attribute__((__always_inline__, __nodebug__)) -_mm256_permute2f128_si256(__m256i a, __m256i b, const int c) -{ - return (__m256i)__builtin_ia32_vperm2f128_si256((__v8si)a, (__v8si)b, c); -} +#define _mm_permute_pd(A, C) __extension__ ({ \ + __m128d __A = (A); \ + (__m128d)__builtin_shufflevector((__v2df)__A, (__v2df) _mm_setzero_pd(), \ + (C) & 0x1, ((C) & 0x2) >> 1); }) + +#define _mm256_permute_pd(A, C) __extension__ ({ \ + __m256d __A = (A); \ + (__m256d)__builtin_shufflevector((__v4df)__A, (__v4df) _mm256_setzero_pd(), \ + (C) & 0x1, ((C) & 0x2) >> 1, \ + 2 + (((C) & 0x4) >> 2), \ + 2 + (((C) & 0x8) >> 3)); }) + +#define _mm_permute_ps(A, C) __extension__ ({ \ + __m128 __A = (A); \ + (__m128)__builtin_shufflevector((__v4sf)__A, (__v4sf) _mm_setzero_ps(), \ + (C) & 0x3, ((C) & 0xc) >> 2, \ + ((C) & 0x30) >> 4, ((C) & 0xc0) >> 6); }) + +#define _mm256_permute_ps(A, C) __extension__ ({ \ + __m256 __A = (A); \ + (__m256)__builtin_shufflevector((__v8sf)__A, (__v8sf) _mm256_setzero_ps(), \ + (C) & 0x3, ((C) & 0xc) >> 2, \ + ((C) & 0x30) >> 4, ((C) & 0xc0) >> 6, \ + 4 + (((C) & 0x03) >> 0), \ + 4 + (((C) & 0x0c) >> 2), \ + 4 + (((C) & 0x30) >> 4), \ + 4 + (((C) & 0xc0) >> 6)); }) + +#define _mm256_permute2f128_pd(V1, V2, M) __extension__ ({ \ + __m256d __V1 = (V1); \ + __m256d __V2 = (V2); \ + (__m256d)__builtin_ia32_vperm2f128_pd256((__v4df)__V1, (__v4df)__V2, (M)); }) + +#define _mm256_permute2f128_ps(V1, V2, M) __extension__ ({ \ + __m256 __V1 = (V1); \ + __m256 __V2 = (V2); \ + (__m256)__builtin_ia32_vperm2f128_ps256((__v8sf)__V1, (__v8sf)__V2, (M)); }) + +#define _mm256_permute2f128_si256(V1, V2, M) __extension__ ({ \ + __m256i __V1 = (V1); \ + __m256i __V2 = (V2); \ + (__m256i)__builtin_ia32_vperm2f128_si256((__v8si)__V1, (__v8si)__V2, (M)); }) /* Vector Blend */ -static __inline __m256d __attribute__((__always_inline__, __nodebug__)) -_mm256_blend_pd(__m256d a, __m256d b, const int c) -{ - return (__m256d)__builtin_ia32_blendpd256((__v4df)a, (__v4df)b, c); -} +#define _mm256_blend_pd(V1, V2, M) __extension__ ({ \ + __m256d __V1 = (V1); \ + __m256d __V2 = (V2); \ + (__m256d)__builtin_ia32_blendpd256((__v4df)__V1, (__v4df)__V2, (M)); }) -static __inline __m256 __attribute__((__always_inline__, __nodebug__)) -_mm256_blend_ps(__m256 a, __m256 b, const int c) -{ - return (__m256)__builtin_ia32_blendps256((__v8sf)a, (__v8sf)b, c); -} +#define _mm256_blend_ps(V1, V2, M) __extension__ ({ \ + __m256 __V1 = (V1); \ + __m256 __V2 = (V2); \ + (__m256)__builtin_ia32_blendps256((__v8sf)__V1, (__v8sf)__V2, (M)); }) static __inline __m256d __attribute__((__always_inline__, __nodebug__)) _mm256_blendv_pd(__m256d a, __m256d b, __m256d c) @@ -330,26 +325,29 @@ _mm256_blendv_ps(__m256 a, __m256 b, __m256 c) } /* Vector Dot Product */ -static __inline __m256 __attribute__((__always_inline__, __nodebug__)) -_mm256_dp_ps(__m256 a, __m256 b, const int c) -{ - return (__m256)__builtin_ia32_dpps256((__v8sf)a, (__v8sf)b, c); -} +#define _mm256_dp_ps(V1, V2, M) __extension__ ({ \ + __m256 __V1 = (V1); \ + __m256 __V2 = (V2); \ + (__m256)__builtin_ia32_dpps256((__v8sf)__V1, (__v8sf)__V2, (M)); }) /* Vector shuffle */ -#define _mm256_shuffle_ps(a, b, mask) \ - (__builtin_shufflevector((__v8sf)(a), (__v8sf)(b), \ +#define _mm256_shuffle_ps(a, b, mask) __extension__ ({ \ + __m256 __a = (a); \ + __m256 __b = (b); \ + (__m256)__builtin_shufflevector((__v8sf)__a, (__v8sf)__b, \ (mask) & 0x3, ((mask) & 0xc) >> 2, \ (((mask) & 0x30) >> 4) + 8, (((mask) & 0xc0) >> 6) + 8, \ ((mask) & 0x3) + 4, (((mask) & 0xc) >> 2) + 4, \ - (((mask) & 0x30) >> 4) + 12, (((mask) & 0xc0) >> 6) + 12)) + (((mask) & 0x30) >> 4) + 12, (((mask) & 0xc0) >> 6) + 12); }) -#define _mm256_shuffle_pd(a, b, mask) \ - (__builtin_shufflevector((__v4df)(a), (__v4df)(b), \ +#define _mm256_shuffle_pd(a, b, mask) __extension__ ({ \ + __m256d __a = (a); \ + __m256d __b = (b); \ + (__m256d)__builtin_shufflevector((__v4df)__a, (__v4df)__b, \ (mask) & 0x1, \ (((mask) & 0x2) >> 1) + 4, \ (((mask) & 0x4) >> 2) + 2, \ - (((mask) & 0x8) >> 3) + 6)) + (((mask) & 0x8) >> 3) + 6); }) /* Compare */ #define _CMP_EQ_OQ 0x00 /* Equal (ordered, non-signaling) */ @@ -385,42 +383,48 @@ _mm256_dp_ps(__m256 a, __m256 b, const int c) #define _CMP_GT_OQ 0x1e /* Greater-than (ordered, non-signaling) */ #define _CMP_TRUE_US 0x1f /* True (unordered, signaling) */ -#define _mm_cmp_pd(a, b, c) \ - (__m128d)__builtin_ia32_cmppd((__v2df)(a), (__v2df)(b), (c)) +#define _mm_cmp_pd(a, b, c) __extension__ ({ \ + __m128d __a = (a); \ + __m128d __b = (b); \ + (__m128d)__builtin_ia32_cmppd((__v2df)__a, (__v2df)__b, (c)); }) -#define _mm_cmp_ps(a, b, c) \ - (__m128)__builtin_ia32_cmpps((__v4sf)(a), (__v4sf)(b), (c)) +#define _mm_cmp_ps(a, b, c) __extension__ ({ \ + __m128 __a = (a); \ + __m128 __b = (b); \ + (__m128)__builtin_ia32_cmpps((__v4sf)__a, (__v4sf)__b, (c)); }) -#define _mm256_cmp_pd(a, b, c) \ - (__m256d)__builtin_ia32_cmppd256((__v4df)(a), (__v4df)(b), (c)) +#define _mm256_cmp_pd(a, b, c) __extension__ ({ \ + __m256d __a = (a); \ + __m256d __b = (b); \ + (__m256d)__builtin_ia32_cmppd256((__v4df)__a, (__v4df)__b, (c)); }) -#define _mm256_cmp_ps(a, b, c) \ - (__m256)__builtin_ia32_cmpps256((__v8sf)(a), (__v8sf)(b), (c)) +#define _mm256_cmp_ps(a, b, c) __extension__ ({ \ + __m256 __a = (a); \ + __m256 __b = (b); \ + (__m256)__builtin_ia32_cmpps256((__v8sf)__a, (__v8sf)__b, (c)); }) -#define _mm_cmp_sd(a, b, c) \ - (__m128d)__builtin_ia32_cmpsd((__v2df)(a), (__v2df)(b), (c)) +#define _mm_cmp_sd(a, b, c) __extension__ ({ \ + __m128d __a = (a); \ + __m128d __b = (b); \ + (__m128d)__builtin_ia32_cmpsd((__v2df)__a, (__v2df)__b, (c)); }) -#define _mm_cmp_ss(a, b, c) \ - (__m128)__builtin_ia32_cmpss((__v4sf)(a), (__v4sf)(b), (c)) +#define _mm_cmp_ss(a, b, c) __extension__ ({ \ + __m128 __a = (a); \ + __m128 __b = (b); \ + (__m128)__builtin_ia32_cmpss((__v4sf)__a, (__v4sf)__b, (c)); }) /* Vector extract */ -static __inline __m128d __attribute__((__always_inline__, __nodebug__)) -_mm256_extractf128_pd(__m256d a, const int o) -{ - return (__m128d)__builtin_ia32_vextractf128_pd256((__v4df)a, o); -} +#define _mm256_extractf128_pd(A, O) __extension__ ({ \ + __m256d __A = (A); \ + (__m128d)__builtin_ia32_vextractf128_pd256((__v4df)__A, (O)); }) -static __inline __m128 __attribute__((__always_inline__, __nodebug__)) -_mm256_extractf128_ps(__m256 a, const int o) -{ - return (__m128)__builtin_ia32_vextractf128_ps256((__v8sf)a, o); -} +#define _mm256_extractf128_ps(A, O) __extension__ ({ \ + __m256 __A = (A); \ + (__m128)__builtin_ia32_vextractf128_ps256((__v8sf)__A, (O)); }) -static __inline __m128i __attribute__((__always_inline__, __nodebug__)) -_mm256_extractf128_si256(__m256i a, const int o) -{ - return (__m128i)__builtin_ia32_vextractf128_si256((__v8si)a, o); -} +#define _mm256_extractf128_si256(A, O) __extension__ ({ \ + __m256i __A = (A); \ + (__m128i)__builtin_ia32_vextractf128_si256((__v8si)__A, (O)); }) static __inline int __attribute__((__always_inline__, __nodebug__)) _mm256_extract_epi32(__m256i a, int const imm) @@ -453,23 +457,20 @@ _mm256_extract_epi64(__m256i a, const int imm) #endif /* Vector insert */ -static __inline __m256d __attribute__((__always_inline__, __nodebug__)) -_mm256_insertf128_pd(__m256d a, __m128d b, const int o) -{ - return (__m256d)__builtin_ia32_vinsertf128_pd256((__v4df)a, (__v2df)b, o); -} +#define _mm256_insertf128_pd(V1, V2, O) __extension__ ({ \ + __m256d __V1 = (V1); \ + __m128d __V2 = (V2); \ + (__m256d)__builtin_ia32_vinsertf128_pd256((__v4df)__V1, (__v2df)__V2, (O)); }) -static __inline __m256 __attribute__((__always_inline__, __nodebug__)) -_mm256_insertf128_ps(__m256 a, __m128 b, const int o) -{ - return (__m256)__builtin_ia32_vinsertf128_ps256((__v8sf)a, (__v4sf)b, o); -} +#define _mm256_insertf128_ps(V1, V2, O) __extension__ ({ \ + __m256 __V1 = (V1); \ + __m128 __V2 = (V2); \ + (__m256)__builtin_ia32_vinsertf128_ps256((__v8sf)__V1, (__v4sf)__V2, (O)); }) -static __inline __m256i __attribute__((__always_inline__, __nodebug__)) -_mm256_insertf128_si256(__m256i a, __m128i b, const int o) -{ - return (__m256i)__builtin_ia32_vinsertf128_si256((__v8si)a, (__v4si)b, o); -} +#define _mm256_insertf128_si256(V1, V2, O) __extension__ ({ \ + __m256i __V1 = (V1); \ + __m128i __V2 = (V2); \ + (__m256i)__builtin_ia32_vinsertf128_si256((__v8si)__V1, (__v4si)__V2, (O)); }) static __inline __m256i __attribute__((__always_inline__, __nodebug__)) _mm256_insert_epi32(__m256i a, int b, int const imm) @@ -762,13 +763,19 @@ _mm256_load_ps(float const *p) static __inline __m256d __attribute__((__always_inline__, __nodebug__)) _mm256_loadu_pd(double const *p) { - return (__m256d)__builtin_ia32_loadupd256(p); + struct __loadu_pd { + __m256d v; + } __attribute__((packed, may_alias)); + return ((struct __loadu_pd*)p)->v; } static __inline __m256 __attribute__((__always_inline__, __nodebug__)) _mm256_loadu_ps(float const *p) { - return (__m256)__builtin_ia32_loadups256(p); + struct __loadu_ps { + __m256 v; + } __attribute__((packed, may_alias)); + return ((struct __loadu_ps*)p)->v; } static __inline __m256i __attribute__((__always_inline__, __nodebug__)) @@ -780,7 +787,10 @@ _mm256_load_si256(__m256i const *p) static __inline __m256i __attribute__((__always_inline__, __nodebug__)) _mm256_loadu_si256(__m256i const *p) { - return (__m256i)__builtin_ia32_loaddqu256((char const *)p); + struct __loadu_si256 { + __m256i v; + } __attribute__((packed, may_alias)); + return ((struct __loadu_si256*)p)->v; } static __inline __m256i __attribute__((__always_inline__, __nodebug__)) @@ -1136,3 +1146,70 @@ _mm256_castsi128_si256(__m128i in) __m128i zero = _mm_setzero_si128(); return __builtin_shufflevector(in, zero, 0, 1, 2, 2); } + +/* SIMD load ops (unaligned) */ +static __inline __m256 __attribute__((__always_inline__, __nodebug__)) +_mm256_loadu2_m128(float const *addr_hi, float const *addr_lo) +{ + struct __loadu_ps { + __m128 v; + } __attribute__((__packed__, __may_alias__)); + + __m256 v256 = _mm256_castps128_ps256(((struct __loadu_ps*)addr_lo)->v); + return _mm256_insertf128_ps(v256, ((struct __loadu_ps*)addr_hi)->v, 1); +} + +static __inline __m256d __attribute__((__always_inline__, __nodebug__)) +_mm256_loadu2_m128d(double const *addr_hi, double const *addr_lo) +{ + struct __loadu_pd { + __m128d v; + } __attribute__((__packed__, __may_alias__)); + + __m256d v256 = _mm256_castpd128_pd256(((struct __loadu_pd*)addr_lo)->v); + return _mm256_insertf128_pd(v256, ((struct __loadu_pd*)addr_hi)->v, 1); +} + +static __inline __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_loadu2_m128i(__m128i const *addr_hi, __m128i const *addr_lo) +{ + struct __loadu_si128 { + __m128i v; + } __attribute__((packed, may_alias)); + __m256i v256 = _mm256_castsi128_si256(((struct __loadu_si128*)addr_lo)->v); + return _mm256_insertf128_si256(v256, ((struct __loadu_si128*)addr_hi)->v, 1); +} + +/* SIMD store ops (unaligned) */ +static __inline void __attribute__((__always_inline__, __nodebug__)) +_mm256_storeu2_m128(float *addr_hi, float *addr_lo, __m256 a) +{ + __m128 v128; + + v128 = _mm256_castps256_ps128(a); + __builtin_ia32_storeups(addr_lo, v128); + v128 = _mm256_extractf128_ps(a, 1); + __builtin_ia32_storeups(addr_hi, v128); +} + +static __inline void __attribute__((__always_inline__, __nodebug__)) +_mm256_storeu2_m128d(double *addr_hi, double *addr_lo, __m256d a) +{ + __m128d v128; + + v128 = _mm256_castpd256_pd128(a); + __builtin_ia32_storeupd(addr_lo, v128); + v128 = _mm256_extractf128_pd(a, 1); + __builtin_ia32_storeupd(addr_hi, v128); +} + +static __inline void __attribute__((__always_inline__, __nodebug__)) +_mm256_storeu2_m128i(__m128i *addr_hi, __m128i *addr_lo, __m256i a) +{ + __m128i v128; + + v128 = _mm256_castsi256_si128(a); + __builtin_ia32_storedqu((char *)addr_lo, (__v16qi)v128); + v128 = _mm256_extractf128_si256(a, 1); + __builtin_ia32_storedqu((char *)addr_hi, (__v16qi)v128); +} diff --git a/renderscript/clang-include/bmi2intrin.h b/renderscript/clang-include/bmi2intrin.h new file mode 100644 index 0000000..c60b0c4 --- /dev/null +++ b/renderscript/clang-include/bmi2intrin.h @@ -0,0 +1,75 @@ +/*===---- bmi2intrin.h - BMI2 intrinsics -----------------------------------=== + * + * Permission is hereby granted, free of charge, to any person obtaining a copy + * of this software and associated documentation files (the "Software"), to deal + * in the Software without restriction, including without limitation the rights + * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell + * copies of the Software, and to permit persons to whom the Software is + * furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in + * all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE + * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN + * THE SOFTWARE. + * + *===-----------------------------------------------------------------------=== + */ + +#if !defined __X86INTRIN_H && !defined __IMMINTRIN_H +#error "Never use <bmi2intrin.h> directly; include <x86intrin.h> instead." +#endif + +#ifndef __BMI2__ +# error "BMI2 instruction set not enabled" +#endif /* __BMI2__ */ + +#ifndef __BMI2INTRIN_H +#define __BMI2INTRIN_H + +static __inline__ unsigned int __attribute__((__always_inline__, __nodebug__)) +_bzhi_u32(unsigned int __X, unsigned int __Y) +{ + return __builtin_ia32_bzhi_si(__X, __Y); +} + +static __inline__ unsigned int __attribute__((__always_inline__, __nodebug__)) +_pdep_u32(unsigned int __X, unsigned int __Y) +{ + return __builtin_ia32_pdep_si(__X, __Y); +} + +static __inline__ unsigned int __attribute__((__always_inline__, __nodebug__)) +_pext_u32(unsigned int __X, unsigned int __Y) +{ + return __builtin_ia32_pext_si(__X, __Y); +} + +#ifdef __x86_64__ + +static __inline__ unsigned long long __attribute__((__always_inline__, __nodebug__)) +_bzhi_u64(unsigned long long __X, unsigned long long __Y) +{ + return __builtin_ia32_bzhi_di(__X, __Y); +} + +static __inline__ unsigned long long __attribute__((__always_inline__, __nodebug__)) +_pdep_u64(unsigned long long __X, unsigned long long __Y) +{ + return __builtin_ia32_pdep_di(__X, __Y); +} + +static __inline__ unsigned long long __attribute__((__always_inline__, __nodebug__)) +_pext_u64(unsigned long long __X, unsigned long long __Y) +{ + return __builtin_ia32_pext_di(__X, __Y); +} + +#endif /* !__x86_64__ */ + +#endif /* __BMI2INTRIN_H */ diff --git a/renderscript/clang-include/bmiintrin.h b/renderscript/clang-include/bmiintrin.h new file mode 100644 index 0000000..8cb00f5 --- /dev/null +++ b/renderscript/clang-include/bmiintrin.h @@ -0,0 +1,115 @@ +/*===---- bmiintrin.h - BMI intrinsics -------------------------------------=== + * + * Permission is hereby granted, free of charge, to any person obtaining a copy + * of this software and associated documentation files (the "Software"), to deal + * in the Software without restriction, including without limitation the rights + * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell + * copies of the Software, and to permit persons to whom the Software is + * furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in + * all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE + * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN + * THE SOFTWARE. + * + *===-----------------------------------------------------------------------=== + */ + +#if !defined __X86INTRIN_H && !defined __IMMINTRIN_H +#error "Never use <bmiintrin.h> directly; include <x86intrin.h> instead." +#endif + +#ifndef __BMI__ +# error "BMI instruction set not enabled" +#endif /* __BMI__ */ + +#ifndef __BMIINTRIN_H +#define __BMIINTRIN_H + +static __inline__ unsigned short __attribute__((__always_inline__, __nodebug__)) +__tzcnt_u16(unsigned short __X) +{ + return __builtin_ctzs(__X); +} + +static __inline__ unsigned int __attribute__((__always_inline__, __nodebug__)) +__andn_u32(unsigned int __X, unsigned int __Y) +{ + return ~__X & __Y; +} + +static __inline__ unsigned int __attribute__((__always_inline__, __nodebug__)) +__bextr_u32(unsigned int __X, unsigned int __Y) +{ + return __builtin_ia32_bextr_u32(__X, __Y); +} + +static __inline__ unsigned int __attribute__((__always_inline__, __nodebug__)) +__blsi_u32(unsigned int __X) +{ + return __X & -__X; +} + +static __inline__ unsigned int __attribute__((__always_inline__, __nodebug__)) +__blsmsk_u32(unsigned int __X) +{ + return __X ^ (__X - 1); +} + +static __inline__ unsigned int __attribute__((__always_inline__, __nodebug__)) +__blsr_u32(unsigned int __X) +{ + return __X & (__X - 1); +} + +static __inline__ unsigned int __attribute__((__always_inline__, __nodebug__)) +__tzcnt_u32(unsigned int __X) +{ + return __builtin_ctz(__X); +} + +#ifdef __x86_64__ +static __inline__ unsigned long long __attribute__((__always_inline__, __nodebug__)) +__andn_u64 (unsigned long long __X, unsigned long long __Y) +{ + return ~__X & __Y; +} + +static __inline__ unsigned long long __attribute__((__always_inline__, __nodebug__)) +__bextr_u64(unsigned long long __X, unsigned long long __Y) +{ + return __builtin_ia32_bextr_u64(__X, __Y); +} + +static __inline__ unsigned long long __attribute__((__always_inline__, __nodebug__)) +__blsi_u64(unsigned long long __X) +{ + return __X & -__X; +} + +static __inline__ unsigned long long __attribute__((__always_inline__, __nodebug__)) +__blsmsk_u64(unsigned long long __X) +{ + return __X ^ (__X - 1); +} + +static __inline__ unsigned long long __attribute__((__always_inline__, __nodebug__)) +__blsr_u64(unsigned long long __X) +{ + return __X & (__X - 1); +} + +static __inline__ unsigned long long __attribute__((__always_inline__, __nodebug__)) +__tzcnt_u64(unsigned long long __X) +{ + return __builtin_ctzll(__X); +} +#endif + +#endif /* __BMIINTRIN_H */ diff --git a/renderscript/clang-include/cpuid.h b/renderscript/clang-include/cpuid.h new file mode 100644 index 0000000..05c293f --- /dev/null +++ b/renderscript/clang-include/cpuid.h @@ -0,0 +1,33 @@ +/*===---- cpuid.h - X86 cpu model detection --------------------------------=== + * + * Permission is hereby granted, free of charge, to any person obtaining a copy + * of this software and associated documentation files (the "Software"), to deal + * in the Software without restriction, including without limitation the rights + * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell + * copies of the Software, and to permit persons to whom the Software is + * furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in + * all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE + * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN + * THE SOFTWARE. + * + *===-----------------------------------------------------------------------=== + */ + +#if !(__x86_64__ || __i386__) +#error this header is for x86 only +#endif + +static inline int __get_cpuid (unsigned int level, unsigned int *eax, + unsigned int *ebx, unsigned int *ecx, + unsigned int *edx) { + asm("cpuid" : "=a"(*eax), "=b" (*ebx), "=c"(*ecx), "=d"(*edx) : "0"(level)); + return 1; +} diff --git a/renderscript/clang-include/emmintrin.h b/renderscript/clang-include/emmintrin.h index 903cfde..91395ed 100644 --- a/renderscript/clang-include/emmintrin.h +++ b/renderscript/clang-include/emmintrin.h @@ -821,8 +821,9 @@ _mm_xor_si128(__m128i a, __m128i b) return a ^ b; } -#define _mm_slli_si128(VEC, IMM) \ - ((__m128i)__builtin_ia32_pslldqi128((__m128i)(VEC), (IMM)*8)) +#define _mm_slli_si128(a, count) __extension__ ({ \ + __m128i __a = (a); \ + (__m128i)__builtin_ia32_pslldqi128(__a, (count)*8); }) static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) _mm_slli_epi16(__m128i a, int count) @@ -885,8 +886,9 @@ _mm_sra_epi32(__m128i a, __m128i count) } -#define _mm_srli_si128(VEC, IMM) \ - ((__m128i)__builtin_ia32_psrldqi128((__m128i)(VEC), (IMM)*8)) +#define _mm_srli_si128(a, count) __extension__ ({ \ + __m128i __a = (a); \ + (__m128i)__builtin_ia32_psrldqi128(__a, (count)*8); }) static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) _mm_srli_epi16(__m128i a, int count) @@ -945,7 +947,10 @@ _mm_cmpeq_epi32(__m128i a, __m128i b) static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) _mm_cmpgt_epi8(__m128i a, __m128i b) { - return (__m128i)((__v16qi)a > (__v16qi)b); + /* This function always performs a signed comparison, but __v16qi is a char + which may be signed or unsigned. */ + typedef signed char __v16qs __attribute__((__vector_size__(16))); + return (__m128i)((__v16qs)a > (__v16qs)b); } static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) @@ -1181,7 +1186,10 @@ _mm_maskmoveu_si128(__m128i d, __m128i n, char *p) static __inline__ void __attribute__((__always_inline__, __nodebug__)) _mm_storel_epi64(__m128i *p, __m128i a) { - __builtin_ia32_storelv4si((__v2si *)p, a); + struct __mm_storel_epi64_struct { + long long u; + } __attribute__((__packed__, __may_alias__)); + ((struct __mm_storel_epi64_struct*)p)->u = a[0]; } static __inline__ void __attribute__((__always_inline__, __nodebug__)) @@ -1259,23 +1267,27 @@ _mm_movemask_epi8(__m128i a) return __builtin_ia32_pmovmskb128((__v16qi)a); } -#define _mm_shuffle_epi32(a, imm) \ - ((__m128i)__builtin_shufflevector((__v4si)(a), (__v4si) _mm_set1_epi32(0), \ - (imm) & 0x3, ((imm) & 0xc) >> 2, \ - ((imm) & 0x30) >> 4, ((imm) & 0xc0) >> 6)) +#define _mm_shuffle_epi32(a, imm) __extension__ ({ \ + __m128i __a = (a); \ + (__m128i)__builtin_shufflevector((__v4si)__a, (__v4si) _mm_set1_epi32(0), \ + (imm) & 0x3, ((imm) & 0xc) >> 2, \ + ((imm) & 0x30) >> 4, ((imm) & 0xc0) >> 6); }) +#define _mm_shufflelo_epi16(a, imm) __extension__ ({ \ + __m128i __a = (a); \ + (__m128i)__builtin_shufflevector((__v8hi)__a, (__v8hi) _mm_set1_epi16(0), \ + (imm) & 0x3, ((imm) & 0xc) >> 2, \ + ((imm) & 0x30) >> 4, ((imm) & 0xc0) >> 6, \ + 4, 5, 6, 7); }) -#define _mm_shufflelo_epi16(a, imm) \ - ((__m128i)__builtin_shufflevector((__v8hi)(a), (__v8hi) _mm_set1_epi16(0), \ - (imm) & 0x3, ((imm) & 0xc) >> 2, \ - ((imm) & 0x30) >> 4, ((imm) & 0xc0) >> 6, \ - 4, 5, 6, 7)) -#define _mm_shufflehi_epi16(a, imm) \ - ((__m128i)__builtin_shufflevector((__v8hi)(a), (__v8hi) _mm_set1_epi16(0), 0, 1, 2, 3, \ - 4 + (((imm) & 0x03) >> 0), \ - 4 + (((imm) & 0x0c) >> 2), \ - 4 + (((imm) & 0x30) >> 4), \ - 4 + (((imm) & 0xc0) >> 6))) +#define _mm_shufflehi_epi16(a, imm) __extension__ ({ \ + __m128i __a = (a); \ + (__m128i)__builtin_shufflevector((__v8hi)__a, (__v8hi) _mm_set1_epi16(0), \ + 0, 1, 2, 3, \ + 4 + (((imm) & 0x03) >> 0), \ + 4 + (((imm) & 0x0c) >> 2), \ + 4 + (((imm) & 0x30) >> 4), \ + 4 + (((imm) & 0xc0) >> 6)); }) static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) _mm_unpackhi_epi8(__m128i a, __m128i b) @@ -1361,9 +1373,10 @@ _mm_movemask_pd(__m128d a) return __builtin_ia32_movmskpd(a); } -#define _mm_shuffle_pd(a, b, i) \ - (__builtin_shufflevector((__m128d)(a), (__m128d)(b), (i) & 1, \ - (((i) & 2) >> 1) + 2)) +#define _mm_shuffle_pd(a, b, i) __extension__ ({ \ + __m128d __a = (a); \ + __m128d __b = (b); \ + __builtin_shufflevector(__a, __b, (i) & 1, (((i) & 2) >> 1) + 2); }) static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) _mm_castpd_ps(__m128d in) diff --git a/renderscript/clang-include/float.h b/renderscript/clang-include/float.h index b7cb73a..2cb13d3 100644 --- a/renderscript/clang-include/float.h +++ b/renderscript/clang-include/float.h @@ -28,7 +28,7 @@ * additional definitions provided for Windows. * For more details see http://msdn.microsoft.com/en-us/library/y0ybw9fy.aspx */ -#if defined(__MINGW32__) && \ +#if (defined(__MINGW32__) || defined(_MSC_VER)) && \ defined(__has_include_next) && __has_include_next(<float.h>) # include_next <float.h> @@ -64,6 +64,11 @@ # undef FLT_MIN # undef DBL_MIN # undef LDBL_MIN +# if __STDC_VERSION__ >= 201112L || !defined(__STRICT_ANSI__) +# undef FLT_TRUE_MIN +# undef DBL_TRUE_MIN +# undef LDBL_TRUE_MIN +# endif #endif /* Characteristics of floating point types, C99 5.2.4.2.2 */ @@ -110,4 +115,10 @@ #define DBL_MIN __DBL_MIN__ #define LDBL_MIN __LDBL_MIN__ +#if __STDC_VERSION__ >= 201112L || !defined(__STRICT_ANSI__) +# define FLT_TRUE_MIN __FLT_DENORM_MIN__ +# define DBL_TRUE_MIN __DBL_DENORM_MIN__ +# define LDBL_TRUE_MIN __LDBL_DENORM_MIN__ +#endif + #endif /* __FLOAT_H */ diff --git a/renderscript/clang-include/fma4intrin.h b/renderscript/clang-include/fma4intrin.h new file mode 100644 index 0000000..c30920d --- /dev/null +++ b/renderscript/clang-include/fma4intrin.h @@ -0,0 +1,231 @@ +/*===---- fma4intrin.h - FMA4 intrinsics -----------------------------------=== + * + * Permission is hereby granted, free of charge, to any person obtaining a copy + * of this software and associated documentation files (the "Software"), to deal + * in the Software without restriction, including without limitation the rights + * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell + * copies of the Software, and to permit persons to whom the Software is + * furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in + * all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE + * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN + * THE SOFTWARE. + * + *===-----------------------------------------------------------------------=== + */ + +#ifndef __X86INTRIN_H +#error "Never use <fma4intrin.h> directly; include <x86intrin.h> instead." +#endif + +#ifndef __FMA4INTRIN_H +#define __FMA4INTRIN_H + +#ifndef __FMA4__ +# error "FMA4 instruction set is not enabled" +#else + +#include <pmmintrin.h> + +static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) +_mm_macc_ps(__m128 __A, __m128 __B, __m128 __C) +{ + return (__m128)__builtin_ia32_vfmaddps(__A, __B, __C); +} + +static __inline__ __m128d __attribute__((__always_inline__, __nodebug__)) +_mm_macc_pd(__m128d __A, __m128d __B, __m128d __C) +{ + return (__m128d)__builtin_ia32_vfmaddpd(__A, __B, __C); +} + +static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) +_mm_macc_ss(__m128 __A, __m128 __B, __m128 __C) +{ + return (__m128)__builtin_ia32_vfmaddss(__A, __B, __C); +} + +static __inline__ __m128d __attribute__((__always_inline__, __nodebug__)) +_mm_macc_sd(__m128d __A, __m128d __B, __m128d __C) +{ + return (__m128d)__builtin_ia32_vfmaddsd(__A, __B, __C); +} + +static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) +_mm_msub_ps(__m128 __A, __m128 __B, __m128 __C) +{ + return (__m128)__builtin_ia32_vfmsubps(__A, __B, __C); +} + +static __inline__ __m128d __attribute__((__always_inline__, __nodebug__)) +_mm_msub_pd(__m128d __A, __m128d __B, __m128d __C) +{ + return (__m128d)__builtin_ia32_vfmsubpd(__A, __B, __C); +} + +static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) +_mm_msub_ss(__m128 __A, __m128 __B, __m128 __C) +{ + return (__m128)__builtin_ia32_vfmsubss(__A, __B, __C); +} + +static __inline__ __m128d __attribute__((__always_inline__, __nodebug__)) +_mm_msub_sd(__m128d __A, __m128d __B, __m128d __C) +{ + return (__m128d)__builtin_ia32_vfmsubsd(__A, __B, __C); +} + +static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) +_mm_nmacc_ps(__m128 __A, __m128 __B, __m128 __C) +{ + return (__m128)__builtin_ia32_vfnmaddps(__A, __B, __C); +} + +static __inline__ __m128d __attribute__((__always_inline__, __nodebug__)) +_mm_nmacc_pd(__m128d __A, __m128d __B, __m128d __C) +{ + return (__m128d)__builtin_ia32_vfnmaddpd(__A, __B, __C); +} + +static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) +_mm_nmacc_ss(__m128 __A, __m128 __B, __m128 __C) +{ + return (__m128)__builtin_ia32_vfnmaddss(__A, __B, __C); +} + +static __inline__ __m128d __attribute__((__always_inline__, __nodebug__)) +_mm_nmacc_sd(__m128d __A, __m128d __B, __m128d __C) +{ + return (__m128d)__builtin_ia32_vfnmaddsd(__A, __B, __C); +} + +static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) +_mm_nmsub_ps(__m128 __A, __m128 __B, __m128 __C) +{ + return (__m128)__builtin_ia32_vfnmsubps(__A, __B, __C); +} + +static __inline__ __m128d __attribute__((__always_inline__, __nodebug__)) +_mm_nmsub_pd(__m128d __A, __m128d __B, __m128d __C) +{ + return (__m128d)__builtin_ia32_vfnmsubpd(__A, __B, __C); +} + +static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) +_mm_nmsub_ss(__m128 __A, __m128 __B, __m128 __C) +{ + return (__m128)__builtin_ia32_vfnmsubss(__A, __B, __C); +} + +static __inline__ __m128d __attribute__((__always_inline__, __nodebug__)) +_mm_nmsub_sd(__m128d __A, __m128d __B, __m128d __C) +{ + return (__m128d)__builtin_ia32_vfnmsubsd(__A, __B, __C); +} + +static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) +_mm_maddsub_ps(__m128 __A, __m128 __B, __m128 __C) +{ + return (__m128)__builtin_ia32_vfmaddsubps(__A, __B, __C); +} + +static __inline__ __m128d __attribute__((__always_inline__, __nodebug__)) +_mm_maddsub_pd(__m128d __A, __m128d __B, __m128d __C) +{ + return (__m128d)__builtin_ia32_vfmaddsubpd(__A, __B, __C); +} + +static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) +_mm_msubadd_ps(__m128 __A, __m128 __B, __m128 __C) +{ + return (__m128)__builtin_ia32_vfmsubaddps(__A, __B, __C); +} + +static __inline__ __m128d __attribute__((__always_inline__, __nodebug__)) +_mm_msubadd_pd(__m128d __A, __m128d __B, __m128d __C) +{ + return (__m128d)__builtin_ia32_vfmsubaddpd(__A, __B, __C); +} + +static __inline__ __m256 __attribute__((__always_inline__, __nodebug__)) +_mm256_macc_ps(__m256 __A, __m256 __B, __m256 __C) +{ + return (__m256)__builtin_ia32_vfmaddps256(__A, __B, __C); +} + +static __inline__ __m256d __attribute__((__always_inline__, __nodebug__)) +_mm256_macc_pd(__m256d __A, __m256d __B, __m256d __C) +{ + return (__m256d)__builtin_ia32_vfmaddpd256(__A, __B, __C); +} + +static __inline__ __m256 __attribute__((__always_inline__, __nodebug__)) +_mm256_msub_ps(__m256 __A, __m256 __B, __m256 __C) +{ + return (__m256)__builtin_ia32_vfmsubps256(__A, __B, __C); +} + +static __inline__ __m256d __attribute__((__always_inline__, __nodebug__)) +_mm256_msub_pd(__m256d __A, __m256d __B, __m256d __C) +{ + return (__m256d)__builtin_ia32_vfmsubpd256(__A, __B, __C); +} + +static __inline__ __m256 __attribute__((__always_inline__, __nodebug__)) +_mm256_nmacc_ps(__m256 __A, __m256 __B, __m256 __C) +{ + return (__m256)__builtin_ia32_vfnmaddps256(__A, __B, __C); +} + +static __inline__ __m256d __attribute__((__always_inline__, __nodebug__)) +_mm256_nmacc_pd(__m256d __A, __m256d __B, __m256d __C) +{ + return (__m256d)__builtin_ia32_vfnmaddpd256(__A, __B, __C); +} + +static __inline__ __m256 __attribute__((__always_inline__, __nodebug__)) +_mm256_nmsub_ps(__m256 __A, __m256 __B, __m256 __C) +{ + return (__m256)__builtin_ia32_vfnmsubps256(__A, __B, __C); +} + +static __inline__ __m256d __attribute__((__always_inline__, __nodebug__)) +_mm256_nmsub_pd(__m256d __A, __m256d __B, __m256d __C) +{ + return (__m256d)__builtin_ia32_vfnmsubpd256(__A, __B, __C); +} + +static __inline__ __m256 __attribute__((__always_inline__, __nodebug__)) +_mm256_maddsub_ps(__m256 __A, __m256 __B, __m256 __C) +{ + return (__m256)__builtin_ia32_vfmaddsubps256(__A, __B, __C); +} + +static __inline__ __m256d __attribute__((__always_inline__, __nodebug__)) +_mm256_maddsub_pd(__m256d __A, __m256d __B, __m256d __C) +{ + return (__m256d)__builtin_ia32_vfmaddsubpd256(__A, __B, __C); +} + +static __inline__ __m256 __attribute__((__always_inline__, __nodebug__)) +_mm256_msubadd_ps(__m256 __A, __m256 __B, __m256 __C) +{ + return (__m256)__builtin_ia32_vfmsubaddps256(__A, __B, __C); +} + +static __inline__ __m256d __attribute__((__always_inline__, __nodebug__)) +_mm256_msubadd_pd(__m256d __A, __m256d __B, __m256d __C) +{ + return (__m256d)__builtin_ia32_vfmsubaddpd256(__A, __B, __C); +} + +#endif /* __FMA4__ */ + +#endif /* __FMA4INTRIN_H */ diff --git a/renderscript/clang-include/fmaintrin.h b/renderscript/clang-include/fmaintrin.h new file mode 100644 index 0000000..6bfd5a8 --- /dev/null +++ b/renderscript/clang-include/fmaintrin.h @@ -0,0 +1,229 @@ +/*===---- fma4intrin.h - FMA4 intrinsics -----------------------------------=== + * + * Permission is hereby granted, free of charge, to any person obtaining a copy + * of this software and associated documentation files (the "Software"), to deal + * in the Software without restriction, including without limitation the rights + * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell + * copies of the Software, and to permit persons to whom the Software is + * furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in + * all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE + * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN + * THE SOFTWARE. + * + *===-----------------------------------------------------------------------=== + */ + +#ifndef __IMMINTRIN_H +#error "Never use <fmaintrin.h> directly; include <immintrin.h> instead." +#endif + +#ifndef __FMAINTRIN_H +#define __FMAINTRIN_H + +#ifndef __FMA__ +# error "FMA instruction set is not enabled" +#else + +static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) +_mm_fmadd_ps(__m128 __A, __m128 __B, __m128 __C) +{ + return (__m128)__builtin_ia32_vfmaddps(__A, __B, __C); +} + +static __inline__ __m128d __attribute__((__always_inline__, __nodebug__)) +_mm_fmadd_pd(__m128d __A, __m128d __B, __m128d __C) +{ + return (__m128d)__builtin_ia32_vfmaddpd(__A, __B, __C); +} + +static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) +_mm_fmadd_ss(__m128 __A, __m128 __B, __m128 __C) +{ + return (__m128)__builtin_ia32_vfmaddss(__A, __B, __C); +} + +static __inline__ __m128d __attribute__((__always_inline__, __nodebug__)) +_mm_fmadd_sd(__m128d __A, __m128d __B, __m128d __C) +{ + return (__m128d)__builtin_ia32_vfmaddsd(__A, __B, __C); +} + +static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) +_mm_fmsub_ps(__m128 __A, __m128 __B, __m128 __C) +{ + return (__m128)__builtin_ia32_vfmsubps(__A, __B, __C); +} + +static __inline__ __m128d __attribute__((__always_inline__, __nodebug__)) +_mm_fmsub_pd(__m128d __A, __m128d __B, __m128d __C) +{ + return (__m128d)__builtin_ia32_vfmsubpd(__A, __B, __C); +} + +static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) +_mm_fmsub_ss(__m128 __A, __m128 __B, __m128 __C) +{ + return (__m128)__builtin_ia32_vfmsubss(__A, __B, __C); +} + +static __inline__ __m128d __attribute__((__always_inline__, __nodebug__)) +_mm_fmsub_sd(__m128d __A, __m128d __B, __m128d __C) +{ + return (__m128d)__builtin_ia32_vfmsubsd(__A, __B, __C); +} + +static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) +_mm_fnmadd_ps(__m128 __A, __m128 __B, __m128 __C) +{ + return (__m128)__builtin_ia32_vfnmaddps(__A, __B, __C); +} + +static __inline__ __m128d __attribute__((__always_inline__, __nodebug__)) +_mm_fnmadd_pd(__m128d __A, __m128d __B, __m128d __C) +{ + return (__m128d)__builtin_ia32_vfnmaddpd(__A, __B, __C); +} + +static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) +_mm_fnmadd_ss(__m128 __A, __m128 __B, __m128 __C) +{ + return (__m128)__builtin_ia32_vfnmaddss(__A, __B, __C); +} + +static __inline__ __m128d __attribute__((__always_inline__, __nodebug__)) +_mm_fnmadd_sd(__m128d __A, __m128d __B, __m128d __C) +{ + return (__m128d)__builtin_ia32_vfnmaddsd(__A, __B, __C); +} + +static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) +_mm_fnmsub_ps(__m128 __A, __m128 __B, __m128 __C) +{ + return (__m128)__builtin_ia32_vfnmsubps(__A, __B, __C); +} + +static __inline__ __m128d __attribute__((__always_inline__, __nodebug__)) +_mm_fnmsub_pd(__m128d __A, __m128d __B, __m128d __C) +{ + return (__m128d)__builtin_ia32_vfnmsubpd(__A, __B, __C); +} + +static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) +_mm_fnmsub_ss(__m128 __A, __m128 __B, __m128 __C) +{ + return (__m128)__builtin_ia32_vfnmsubss(__A, __B, __C); +} + +static __inline__ __m128d __attribute__((__always_inline__, __nodebug__)) +_mm_fnmsub_sd(__m128d __A, __m128d __B, __m128d __C) +{ + return (__m128d)__builtin_ia32_vfnmsubsd(__A, __B, __C); +} + +static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) +_mm_fmaddsub_ps(__m128 __A, __m128 __B, __m128 __C) +{ + return (__m128)__builtin_ia32_vfmaddsubps(__A, __B, __C); +} + +static __inline__ __m128d __attribute__((__always_inline__, __nodebug__)) +_mm_fmaddsub_pd(__m128d __A, __m128d __B, __m128d __C) +{ + return (__m128d)__builtin_ia32_vfmaddsubpd(__A, __B, __C); +} + +static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) +_mm_fmsubadd_ps(__m128 __A, __m128 __B, __m128 __C) +{ + return (__m128)__builtin_ia32_vfmsubaddps(__A, __B, __C); +} + +static __inline__ __m128d __attribute__((__always_inline__, __nodebug__)) +_mm_fmsubadd_pd(__m128d __A, __m128d __B, __m128d __C) +{ + return (__m128d)__builtin_ia32_vfmsubaddpd(__A, __B, __C); +} + +static __inline__ __m256 __attribute__((__always_inline__, __nodebug__)) +_mm256_fmadd_ps(__m256 __A, __m256 __B, __m256 __C) +{ + return (__m256)__builtin_ia32_vfmaddps256(__A, __B, __C); +} + +static __inline__ __m256d __attribute__((__always_inline__, __nodebug__)) +_mm256_fmadd_pd(__m256d __A, __m256d __B, __m256d __C) +{ + return (__m256d)__builtin_ia32_vfmaddpd256(__A, __B, __C); +} + +static __inline__ __m256 __attribute__((__always_inline__, __nodebug__)) +_mm256_fmsub_ps(__m256 __A, __m256 __B, __m256 __C) +{ + return (__m256)__builtin_ia32_vfmsubps256(__A, __B, __C); +} + +static __inline__ __m256d __attribute__((__always_inline__, __nodebug__)) +_mm256_fmsub_pd(__m256d __A, __m256d __B, __m256d __C) +{ + return (__m256d)__builtin_ia32_vfmsubpd256(__A, __B, __C); +} + +static __inline__ __m256 __attribute__((__always_inline__, __nodebug__)) +_mm256_fnmadd_ps(__m256 __A, __m256 __B, __m256 __C) +{ + return (__m256)__builtin_ia32_vfnmaddps256(__A, __B, __C); +} + +static __inline__ __m256d __attribute__((__always_inline__, __nodebug__)) +_mm256_fnmadd_pd(__m256d __A, __m256d __B, __m256d __C) +{ + return (__m256d)__builtin_ia32_vfnmaddpd256(__A, __B, __C); +} + +static __inline__ __m256 __attribute__((__always_inline__, __nodebug__)) +_mm256_fnmsub_ps(__m256 __A, __m256 __B, __m256 __C) +{ + return (__m256)__builtin_ia32_vfnmsubps256(__A, __B, __C); +} + +static __inline__ __m256d __attribute__((__always_inline__, __nodebug__)) +_mm256_fnmsub_pd(__m256d __A, __m256d __B, __m256d __C) +{ + return (__m256d)__builtin_ia32_vfnmsubpd256(__A, __B, __C); +} + +static __inline__ __m256 __attribute__((__always_inline__, __nodebug__)) +_mm256_fmaddsub_ps(__m256 __A, __m256 __B, __m256 __C) +{ + return (__m256)__builtin_ia32_vfmaddsubps256(__A, __B, __C); +} + +static __inline__ __m256d __attribute__((__always_inline__, __nodebug__)) +_mm256_fmaddsub_pd(__m256d __A, __m256d __B, __m256d __C) +{ + return (__m256d)__builtin_ia32_vfmaddsubpd256(__A, __B, __C); +} + +static __inline__ __m256 __attribute__((__always_inline__, __nodebug__)) +_mm256_fmsubadd_ps(__m256 __A, __m256 __B, __m256 __C) +{ + return (__m256)__builtin_ia32_vfmsubaddps256(__A, __B, __C); +} + +static __inline__ __m256d __attribute__((__always_inline__, __nodebug__)) +_mm256_fmsubadd_pd(__m256d __A, __m256d __B, __m256d __C) +{ + return (__m256d)__builtin_ia32_vfmsubaddpd256(__A, __B, __C); +} + +#endif /* __FMA__ */ + +#endif /* __FMAINTRIN_H */ diff --git a/renderscript/clang-include/immintrin.h b/renderscript/clang-include/immintrin.h index a19deaa..15b65f3 100644 --- a/renderscript/clang-include/immintrin.h +++ b/renderscript/clang-include/immintrin.h @@ -48,7 +48,7 @@ #include <smmintrin.h> #endif -#if defined (__AES__) || defined (__PCLMUL__) +#if defined (__AES__) #include <wmmintrin.h> #endif @@ -56,4 +56,46 @@ #include <avxintrin.h> #endif +#ifdef __AVX2__ +#include <avx2intrin.h> +#endif + +#ifdef __BMI__ +#include <bmiintrin.h> +#endif + +#ifdef __BMI2__ +#include <bmi2intrin.h> +#endif + +#ifdef __LZCNT__ +#include <lzcntintrin.h> +#endif + +#ifdef __FMA__ +#include <fmaintrin.h> +#endif + +#ifdef __RDRND__ +static __inline__ int __attribute__((__always_inline__, __nodebug__)) +_rdrand16_step(unsigned short *__p) +{ + return __builtin_ia32_rdrand16_step(__p); +} + +static __inline__ int __attribute__((__always_inline__, __nodebug__)) +_rdrand32_step(unsigned int *__p) +{ + return __builtin_ia32_rdrand32_step(__p); +} + +#ifdef __x86_64__ +static __inline__ int __attribute__((__always_inline__, __nodebug__)) +_rdrand64_step(unsigned long long *__p) +{ + return __builtin_ia32_rdrand64_step(__p); +} +#endif +#endif /* __RDRND__ */ + #endif /* __IMMINTRIN_H */ diff --git a/renderscript/clang-include/lzcntintrin.h b/renderscript/clang-include/lzcntintrin.h new file mode 100644 index 0000000..62ab5ca --- /dev/null +++ b/renderscript/clang-include/lzcntintrin.h @@ -0,0 +1,55 @@ +/*===---- lzcntintrin.h - LZCNT intrinsics ---------------------------------=== + * + * Permission is hereby granted, free of charge, to any person obtaining a copy + * of this software and associated documentation files (the "Software"), to deal + * in the Software without restriction, including without limitation the rights + * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell + * copies of the Software, and to permit persons to whom the Software is + * furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in + * all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE + * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN + * THE SOFTWARE. + * + *===-----------------------------------------------------------------------=== + */ + +#if !defined __X86INTRIN_H && !defined __IMMINTRIN_H +#error "Never use <lzcntintrin.h> directly; include <x86intrin.h> instead." +#endif + +#ifndef __LZCNT__ +# error "LZCNT instruction is not enabled" +#endif /* __LZCNT__ */ + +#ifndef __LZCNTINTRIN_H +#define __LZCNTINTRIN_H + +static __inline__ unsigned short __attribute__((__always_inline__, __nodebug__)) +__lzcnt16(unsigned short __X) +{ + return __builtin_clzs(__X); +} + +static __inline__ unsigned int __attribute__((__always_inline__, __nodebug__)) +__lzcnt32(unsigned int __X) +{ + return __builtin_clz(__X); +} + +#ifdef __x86_64__ +static __inline__ unsigned long long __attribute__((__always_inline__, __nodebug__)) +__lzcnt64(unsigned long long __X) +{ + return __builtin_clzll(__X); +} +#endif + +#endif /* __LZCNTINTRIN_H */ diff --git a/renderscript/clang-include/mm3dnow.h b/renderscript/clang-include/mm3dnow.h index 2f456ad..d5236f8 100644 --- a/renderscript/clang-include/mm3dnow.h +++ b/renderscript/clang-include/mm3dnow.h @@ -105,7 +105,7 @@ _m_pfrsqrt(__m64 __m) { static __inline__ __m64 __attribute__((__always_inline__, __nodebug__)) _m_pfrsqrtit1(__m64 __m1, __m64 __m2) { - return (__m64)__builtin_ia32_pfrsqrtit1((__v2sf)__m1, (__v2sf)__m2); + return (__m64)__builtin_ia32_pfrsqit1((__v2sf)__m1, (__v2sf)__m2); } static __inline__ __m64 __attribute__((__always_inline__, __nodebug__)) diff --git a/renderscript/clang-include/module.map b/renderscript/clang-include/module.map new file mode 100644 index 0000000..418ba50 --- /dev/null +++ b/renderscript/clang-include/module.map @@ -0,0 +1,108 @@ +module _Builtin_intrinsics [system] { + explicit module altivec { + requires altivec + header "altivec.h" + } + + explicit module intel { + requires x86 + export * + + header "immintrin.h" + header "x86intrin.h" + + explicit module mm_malloc { + header "mm_malloc.h" + export * // note: for <stdlib.h> dependency + } + + explicit module cpuid { + header "cpuid.h" + } + + explicit module mmx { + requires mmx + header "mmintrin.h" + } + + explicit module sse { + requires sse + export mmx + export * // note: for hackish <emmintrin.h> dependency + header "xmmintrin.h" + } + + explicit module sse2 { + requires sse2 + export sse + header "emmintrin.h" + } + + explicit module sse3 { + requires sse3 + export sse2 + header "pmmintrin.h" + } + + explicit module ssse3 { + requires ssse3 + export sse3 + header "tmmintrin.h" + } + + explicit module sse4_1 { + requires sse41 + export ssse3 + header "smmintrin.h" + } + + explicit module sse4_2 { + requires sse42 + export sse4_1 + header "nmmintrin.h" + } + + explicit module avx { + requires avx + export sse4_2 + header "avxintrin.h" + } + + explicit module avx2 { + requires avx2 + export avx + header "avx2intrin.h" + } + + explicit module bmi { + requires bmi + header "bmiintrin.h" + } + + explicit module bmi2 { + requires bmi2 + header "bmi2intrin.h" + } + + explicit module fma4 { + requires fma4 + export sse3 + header "fma4intrin.h" + } + + explicit module lzcnt { + requires lzcnt + header "lzcntintrin.h" + } + + explicit module popcnt { + requires popcnt + header "popcntintrin.h" + } + + explicit module mm3dnow { + requires mm3dnow + header "mm3dnow.h" + } + } +} diff --git a/renderscript/clang-include/popcntintrin.h b/renderscript/clang-include/popcntintrin.h new file mode 100644 index 0000000..d439daa --- /dev/null +++ b/renderscript/clang-include/popcntintrin.h @@ -0,0 +1,45 @@ +/*===---- popcntintrin.h - POPCNT intrinsics -------------------------------=== + * + * Permission is hereby granted, free of charge, to any person obtaining a copy + * of this software and associated documentation files (the "Software"), to deal + * in the Software without restriction, including without limitation the rights + * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell + * copies of the Software, and to permit persons to whom the Software is + * furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in + * all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE + * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN + * THE SOFTWARE. + * + *===-----------------------------------------------------------------------=== + */ + +#ifndef __POPCNT__ +#error "POPCNT instruction set not enabled" +#endif + +#ifndef _POPCNTINTRIN_H +#define _POPCNTINTRIN_H + +static __inline__ int __attribute__((__always_inline__, __nodebug__)) +_mm_popcnt_u32(unsigned int __A) +{ + return __builtin_popcount(__A); +} + +#ifdef __x86_64__ +static __inline__ long long __attribute__((__always_inline__, __nodebug__)) +_mm_popcnt_u64(unsigned long long __A) +{ + return __builtin_popcountll(__A); +} +#endif /* __x86_64__ */ + +#endif /* _POPCNTINTRIN_H */ diff --git a/renderscript/clang-include/smmintrin.h b/renderscript/clang-include/smmintrin.h index 2b8b321..2fab50e 100644 --- a/renderscript/clang-include/smmintrin.h +++ b/renderscript/clang-include/smmintrin.h @@ -57,23 +57,34 @@ #define _mm_floor_ss(X, Y) _mm_round_ss((X), (Y), _MM_FROUND_FLOOR) #define _mm_floor_sd(X, Y) _mm_round_sd((X), (Y), _MM_FROUND_FLOOR) -#define _mm_round_ps(X, Y) __builtin_ia32_roundps((X), (Y)) -#define _mm_round_ss(X, Y, M) __builtin_ia32_roundss((X), (Y), (M)) -#define _mm_round_pd(X, M) __builtin_ia32_roundpd((X), (M)) -#define _mm_round_sd(X, Y, M) __builtin_ia32_roundsd((X), (Y), (M)) +#define _mm_round_ps(X, M) __extension__ ({ \ + __m128 __X = (X); \ + (__m128) __builtin_ia32_roundps((__v4sf)__X, (M)); }) + +#define _mm_round_ss(X, Y, M) __extension__ ({ \ + __m128 __X = (X); \ + __m128 __Y = (Y); \ + (__m128) __builtin_ia32_roundss((__v4sf)__X, (__v4sf)__Y, (M)); }) + +#define _mm_round_pd(X, M) __extension__ ({ \ + __m128d __X = (X); \ + (__m128d) __builtin_ia32_roundpd((__v2df)__X, (M)); }) + +#define _mm_round_sd(X, Y, M) __extension__ ({ \ + __m128d __X = (X); \ + __m128d __Y = (Y); \ + (__m128d) __builtin_ia32_roundsd((__v2df)__X, (__v2df)__Y, (M)); }) /* SSE4 Packed Blending Intrinsics. */ -static __inline__ __m128d __attribute__((__always_inline__, __nodebug__)) -_mm_blend_pd (__m128d __V1, __m128d __V2, const int __M) -{ - return (__m128d) __builtin_ia32_blendpd ((__v2df)__V1, (__v2df)__V2, __M); -} +#define _mm_blend_pd(V1, V2, M) __extension__ ({ \ + __m128d __V1 = (V1); \ + __m128d __V2 = (V2); \ + (__m128d) __builtin_ia32_blendpd ((__v2df)__V1, (__v2df)__V2, (M)); }) -static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) -_mm_blend_ps (__m128 __V1, __m128 __V2, const int __M) -{ - return (__m128) __builtin_ia32_blendps ((__v4sf)__V1, (__v4sf)__V2, __M); -} +#define _mm_blend_ps(V1, V2, M) __extension__ ({ \ + __m128 __V1 = (V1); \ + __m128 __V2 = (V2); \ + (__m128) __builtin_ia32_blendps ((__v4sf)__V1, (__v4sf)__V2, (M)); }) static __inline__ __m128d __attribute__((__always_inline__, __nodebug__)) _mm_blendv_pd (__m128d __V1, __m128d __V2, __m128d __M) @@ -96,11 +107,10 @@ _mm_blendv_epi8 (__m128i __V1, __m128i __V2, __m128i __M) (__v16qi)__M); } -static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) -_mm_blend_epi16 (__m128i __V1, __m128i __V2, const int __M) -{ - return (__m128i) __builtin_ia32_pblendw128 ((__v8hi)__V1, (__v8hi)__V2, __M); -} +#define _mm_blend_epi16(V1, V2, M) __extension__ ({ \ + __m128i __V1 = (V1); \ + __m128i __V2 = (V2); \ + (__m128i) __builtin_ia32_pblendw128 ((__v8hi)__V1, (__v8hi)__V2, (M)); }) /* SSE4 Dword Multiply Instructions. */ static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) @@ -116,8 +126,15 @@ _mm_mul_epi32 (__m128i __V1, __m128i __V2) } /* SSE4 Floating Point Dot Product Instructions. */ -#define _mm_dp_ps(X, Y, M) __builtin_ia32_dpps ((X), (Y), (M)) -#define _mm_dp_pd(X, Y, M) __builtin_ia32_dppd ((X), (Y), (M)) +#define _mm_dp_ps(X, Y, M) __extension__ ({ \ + __m128 __X = (X); \ + __m128 __Y = (Y); \ + (__m128) __builtin_ia32_dpps((__v4sf)__X, (__v4sf)__Y, (M)); }) + +#define _mm_dp_pd(X, Y, M) __extension__ ({\ + __m128d __X = (X); \ + __m128d __Y = (Y); \ + (__m128d) __builtin_ia32_dppd((__v2df)__X, (__v2df)__Y, (M)); }) /* SSE4 Streaming Load Hint Instruction. */ static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) @@ -198,14 +215,14 @@ _mm_max_epu32 (__m128i __V1, __m128i __V2) /* Insert int into packed integer array at index. */ #define _mm_insert_epi8(X, I, N) (__extension__ ({ __v16qi __a = (__v16qi)(X); \ - __a[N] = I; \ + __a[(N)] = (I); \ __a;})) #define _mm_insert_epi32(X, I, N) (__extension__ ({ __v4si __a = (__v4si)(X); \ - __a[N] = I; \ + __a[(N)] = (I); \ __a;})) #ifdef __x86_64__ #define _mm_insert_epi64(X, I, N) (__extension__ ({ __v2di __a = (__v2di)(X); \ - __a[N] = I; \ + __a[(N)] = (I); \ __a;})) #endif /* __x86_64__ */ @@ -213,12 +230,12 @@ _mm_max_epu32 (__m128i __V1, __m128i __V2) * as a zero extended value, so it is unsigned. */ #define _mm_extract_epi8(X, N) (__extension__ ({ __v16qi __a = (__v16qi)(X); \ - (unsigned char)__a[N];})) + (unsigned char)__a[(N)];})) #define _mm_extract_epi32(X, N) (__extension__ ({ __v4si __a = (__v4si)(X); \ - (unsigned)__a[N];})) + (unsigned)__a[(N)];})) #ifdef __x86_64__ #define _mm_extract_epi64(X, N) (__extension__ ({ __v2di __a = (__v2di)(X); \ - __a[N];})) + __a[(N)];})) #endif /* __x86_64 */ /* SSE4 128-bit Packed Integer Comparisons. */ @@ -242,13 +259,13 @@ _mm_testnzc_si128(__m128i __M, __m128i __V) #define _mm_test_all_ones(V) _mm_testc_si128((V), _mm_cmpeq_epi32((V), (V))) #define _mm_test_mix_ones_zeros(M, V) _mm_testnzc_si128((M), (V)) -#define _mm_test_all_zeros(M, V) _mm_testz_si128 ((V), (V)) +#define _mm_test_all_zeros(M, V) _mm_testz_si128 ((M), (V)) /* SSE4 64-bit Packed Integer Comparisons. */ static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) _mm_cmpeq_epi64(__m128i __V1, __m128i __V2) { - return (__m128i) __builtin_ia32_pcmpeqq((__v2di)__V1, (__v2di)__V2); + return (__m128i)((__v2di)__V1 == (__v2di)__V2); } /* SSE4 Packed Integer Sign-Extension. */ @@ -333,7 +350,16 @@ _mm_packus_epi32(__m128i __V1, __m128i __V2) } /* SSE4 Multiple Packed Sums of Absolute Difference. */ -#define _mm_mpsadbw_epu8(X, Y, M) __builtin_ia32_mpsadbw128((X), (Y), (M)) +#define _mm_mpsadbw_epu8(X, Y, M) __extension__ ({ \ + __m128i __X = (X); \ + __m128i __Y = (Y); \ + (__m128i) __builtin_ia32_mpsadbw128((__v16qi)__X, (__v16qi)__Y, (M)); }) + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_minpos_epu16(__m128i __V) +{ + return (__m128i) __builtin_ia32_phminposuw128((__v8hi)__V); +} /* These definitions are normally in nmmintrin.h, but gcc puts them in here so we'll do the same. */ @@ -371,20 +397,20 @@ _mm_packus_epi32(__m128i __V1, __m128i __V2) #define _mm_cmpestrm(A, LA, B, LB, M) \ __builtin_ia32_pcmpestrm128((A), (LA), (B), (LB), (M)) -#define _mm_cmpestri(X, LX, Y, LY, M) \ +#define _mm_cmpestri(A, LA, B, LB, M) \ __builtin_ia32_pcmpestri128((A), (LA), (B), (LB), (M)) /* SSE4.2 Packed Comparison Intrinsics and EFlag Reading. */ -#define _mm_cmpistra(A, LA, B, LB, M) \ - __builtin_ia32_pcmpistria128((A), (LA), (B), (LB), (M)) -#define _mm_cmpistrc(A, LA, B, LB, M) \ - __builtin_ia32_pcmpistric128((A), (LA), (B), (LB), (M)) -#define _mm_cmpistro(A, LA, B, LB, M) \ - __builtin_ia32_pcmpistrio128((A), (LA), (B), (LB), (M)) -#define _mm_cmpistrs(A, LA, B, LB, M) \ - __builtin_ia32_pcmpistris128((A), (LA), (B), (LB), (M)) -#define _mm_cmpistrz(A, LA, B, LB, M) \ - __builtin_ia32_pcmpistriz128((A), (LA), (B), (LB), (M)) +#define _mm_cmpistra(A, B, M) \ + __builtin_ia32_pcmpistria128((A), (B), (M)) +#define _mm_cmpistrc(A, B, M) \ + __builtin_ia32_pcmpistric128((A), (B), (M)) +#define _mm_cmpistro(A, B, M) \ + __builtin_ia32_pcmpistrio128((A), (B), (M)) +#define _mm_cmpistrs(A, B, M) \ + __builtin_ia32_pcmpistris128((A), (B), (M)) +#define _mm_cmpistrz(A, B, M) \ + __builtin_ia32_pcmpistriz128((A), (B), (M)) #define _mm_cmpestra(A, LA, B, LB, M) \ __builtin_ia32_pcmpestria128((A), (LA), (B), (LB), (M)) @@ -401,7 +427,7 @@ _mm_packus_epi32(__m128i __V1, __m128i __V2) static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) _mm_cmpgt_epi64(__m128i __V1, __m128i __V2) { - return __builtin_ia32_pcmpgtq((__v2di)__V1, (__v2di)__V2); + return (__m128i)((__v2di)__V1 > (__v2di)__V2); } /* SSE4.2 Accumulate CRC32. */ @@ -431,20 +457,9 @@ _mm_crc32_u64(unsigned long long __C, unsigned long long __D) } #endif /* __x86_64__ */ -/* SSE4.2 Population Count. */ -static __inline__ int __attribute__((__always_inline__, __nodebug__)) -_mm_popcnt_u32(unsigned int __A) -{ - return __builtin_popcount(__A); -} - -#ifdef __x86_64__ -static __inline__ long long __attribute__((__always_inline__, __nodebug__)) -_mm_popcnt_u64(unsigned long long __A) -{ - return __builtin_popcountll(__A); -} -#endif /* __x86_64__ */ +#ifdef __POPCNT__ +#include <popcntintrin.h> +#endif #endif /* __SSE4_2__ */ #endif /* __SSE4_1__ */ diff --git a/renderscript/clang-include/stddef.h b/renderscript/clang-include/stddef.h index 9e87ee8..eb919b5 100644 --- a/renderscript/clang-include/stddef.h +++ b/renderscript/clang-include/stddef.h @@ -43,10 +43,20 @@ typedef __WCHAR_TYPE__ wchar_t; #undef NULL #ifdef __cplusplus -#undef __null // VC++ hack. -#define NULL __null +# if !defined(__MINGW32__) && !defined(_MSC_VER) +# define NULL __null +# else +# define NULL 0 +# endif #else -#define NULL ((void*)0) +# define NULL ((void*)0) +#endif + +#ifdef __cplusplus +#if defined(_MSC_EXTENSIONS) && defined(_NATIVE_NULLPTR_SUPPORTED) +namespace std { typedef decltype(nullptr) nullptr_t; } +using ::std::nullptr_t; +#endif #endif #define offsetof(t, d) __builtin_offsetof(t, d) diff --git a/renderscript/clang-include/tgmath.h b/renderscript/clang-include/tgmath.h index 1b0b9d2..4fa1cf7 100644 --- a/renderscript/clang-include/tgmath.h +++ b/renderscript/clang-include/tgmath.h @@ -540,15 +540,15 @@ static long double _TG_ATTRS __tg_fabs(long double __x) {return fabsl(__x);} -static float _Complex +static float _TG_ATTRS __tg_fabs(float _Complex __x) {return cabsf(__x);} -static double _Complex +static double _TG_ATTRS __tg_fabs(double _Complex __x) {return cabs(__x);} -static long double _Complex +static long double _TG_ATTRS __tg_fabs(long double _Complex __x) {return cabsl(__x);} @@ -976,6 +976,23 @@ static long double #undef log2 #define log2(__x) __tg_log2(__tg_promote1((__x))(__x)) +// logb + +static float + _TG_ATTRS + __tg_logb(float __x) {return logbf(__x);} + +static double + _TG_ATTRS + __tg_logb(double __x) {return logb(__x);} + +static long double + _TG_ATTRS + __tg_logb(long double __x) {return logbl(__x);} + +#undef logb +#define logb(__x) __tg_logb(__tg_promote1((__x))(__x)) + // lrint static long diff --git a/renderscript/clang-include/tmmintrin.h b/renderscript/clang-include/tmmintrin.h index 07fea1c..a62c6cc 100644 --- a/renderscript/clang-include/tmmintrin.h +++ b/renderscript/clang-include/tmmintrin.h @@ -66,8 +66,15 @@ _mm_abs_epi32(__m128i a) return (__m128i)__builtin_ia32_pabsd128((__v4si)a); } -#define _mm_alignr_epi8(a, b, n) (__builtin_ia32_palignr128((a), (b), (n))) -#define _mm_alignr_pi8(a, b, n) (__builtin_ia32_palignr((a), (b), (n))) +#define _mm_alignr_epi8(a, b, n) __extension__ ({ \ + __m128i __a = (a); \ + __m128i __b = (b); \ + (__m128i)__builtin_ia32_palignr128((__v16qi)__a, (__v16qi)__b, (n)); }) + +#define _mm_alignr_pi8(a, b, n) __extension__ ({ \ + __m64 __a = (a); \ + __m64 __b = (b); \ + (__m64)__builtin_ia32_palignr((__v8qi)__a, (__v8qi)__b, (n)); }) static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) _mm_hadd_epi16(__m128i a, __m128i b) diff --git a/renderscript/clang-include/unwind.h b/renderscript/clang-include/unwind.h new file mode 100644 index 0000000..a065920 --- /dev/null +++ b/renderscript/clang-include/unwind.h @@ -0,0 +1,124 @@ +/*===---- unwind.h - Stack unwinding ----------------------------------------=== + * + * Permission is hereby granted, free of charge, to any person obtaining a copy + * of this software and associated documentation files (the "Software"), to deal + * in the Software without restriction, including without limitation the rights + * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell + * copies of the Software, and to permit persons to whom the Software is + * furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in + * all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE + * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN + * THE SOFTWARE. + * + *===-----------------------------------------------------------------------=== + */ + +/* See "Data Definitions for libgcc_s" in the Linux Standard Base.*/ + +#if __has_include_next(<unwind.h>) +/* Darwin and libunwind provide an unwind.h. If that's available, use + * it. libunwind wraps some of its definitions in #ifdef _GNU_SOURCE, + * so define that around the include.*/ +# ifndef _GNU_SOURCE +# define _SHOULD_UNDEFINE_GNU_SOURCE +# define _GNU_SOURCE +# endif +// libunwind's unwind.h reflects the current visibility. However, Mozilla +// builds with -fvisibility=hidden and relies on gcc's unwind.h to reset the +// visibility to default and export its contents. gcc also allows users to +// override its override by #defining HIDE_EXPORTS (but note, this only obeys +// the user's -fvisibility setting; it doesn't hide any exports on its own). We +// imitate gcc's header here: +# ifdef HIDE_EXPORTS +# include_next <unwind.h> +# else +# pragma GCC visibility push(default) +# include_next <unwind.h> +# pragma GCC visibility pop +# endif +# ifdef _SHOULD_UNDEFINE_GNU_SOURCE +# undef _GNU_SOURCE +# undef _SHOULD_UNDEFINE_GNU_SOURCE +# endif +#else + +#include <stdint.h> + +#ifdef __cplusplus +extern "C" { +#endif + +/* It is a bit strange for a header to play with the visibility of the + symbols it declares, but this matches gcc's behavior and some programs + depend on it */ +#pragma GCC visibility push(default) + +struct _Unwind_Context; +typedef enum { + _URC_NO_REASON = 0, + _URC_FOREIGN_EXCEPTION_CAUGHT = 1, + + _URC_FATAL_PHASE2_ERROR = 2, + _URC_FATAL_PHASE1_ERROR = 3, + _URC_NORMAL_STOP = 4, + + _URC_END_OF_STACK = 5, + _URC_HANDLER_FOUND = 6, + _URC_INSTALL_CONTEXT = 7, + _URC_CONTINUE_UNWIND = 8 +} _Unwind_Reason_Code; + + +#ifdef __arm__ + +typedef enum { + _UVRSC_CORE = 0, /* integer register */ + _UVRSC_VFP = 1, /* vfp */ + _UVRSC_WMMXD = 3, /* Intel WMMX data register */ + _UVRSC_WMMXC = 4 /* Intel WMMX control register */ +} _Unwind_VRS_RegClass; + +typedef enum { + _UVRSD_UINT32 = 0, + _UVRSD_VFPX = 1, + _UVRSD_UINT64 = 3, + _UVRSD_FLOAT = 4, + _UVRSD_DOUBLE = 5 +} _Unwind_VRS_DataRepresentation; + +typedef enum { + _UVRSR_OK = 0, + _UVRSR_NOT_IMPLEMENTED = 1, + _UVRSR_FAILED = 2 +} _Unwind_VRS_Result; + +_Unwind_VRS_Result _Unwind_VRS_Get(_Unwind_Context *context, + _Unwind_VRS_RegClass regclass, + uint32_t regno, + _Unwind_VRS_DataRepresentation representation, + void *valuep); + +#else + +uintptr_t _Unwind_GetIP(struct _Unwind_Context* context); + +#endif + +typedef _Unwind_Reason_Code (*_Unwind_Trace_Fn)(struct _Unwind_Context*, void*); +_Unwind_Reason_Code _Unwind_Backtrace(_Unwind_Trace_Fn, void*); + +#pragma GCC visibility pop + +#ifdef __cplusplus +} +#endif + +#endif diff --git a/renderscript/clang-include/wmmintrin.h b/renderscript/clang-include/wmmintrin.h index 6b2e468..dca896f 100644 --- a/renderscript/clang-include/wmmintrin.h +++ b/renderscript/clang-include/wmmintrin.h @@ -24,11 +24,13 @@ #ifndef _WMMINTRIN_H #define _WMMINTRIN_H -#if !defined (__AES__) -# error "AES instructions not enabled" +#include <emmintrin.h> + +#if !defined (__AES__) && !defined (__PCLMUL__) +# error "AES/PCLMUL instructions not enabled" #else -#include <smmintrin.h> +#ifdef __AES__ static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) _mm_aesenc_si128(__m128i __V, __m128i __R) @@ -64,4 +66,14 @@ _mm_aesimc_si128(__m128i __V) __builtin_ia32_aeskeygenassist128((C), (R)) #endif /* __AES__ */ + +#ifdef __PCLMUL__ + +#define _mm_clmulepi64_si128(__X, __Y, __I) \ + ((__m128i)__builtin_ia32_pclmulqdq128((__v2di)(__m128i)(__X), \ + (__v2di)(__m128i)(__Y), (char)(__I))) + +#endif /* __PCLMUL__ */ + +#endif /* __AES__ || __PCLMUL__ */ #endif /* _WMMINTRIN_H */ diff --git a/renderscript/clang-include/x86intrin.h b/renderscript/clang-include/x86intrin.h index e5e7a6a..556cd01 100644 --- a/renderscript/clang-include/x86intrin.h +++ b/renderscript/clang-include/x86intrin.h @@ -26,6 +26,38 @@ #include <immintrin.h> -// FIXME: SSE4A, 3dNOW, FMA4, XOP, LWP, ABM, POPCNT +#ifdef __3dNOW__ +#include <mm3dnow.h> +#endif + +#ifdef __BMI__ +#include <bmiintrin.h> +#endif + +#ifdef __BMI2__ +#include <bmi2intrin.h> +#endif + +#ifdef __LZCNT__ +#include <lzcntintrin.h> +#endif + +#ifdef __POPCNT__ +#include <popcntintrin.h> +#endif + +#ifdef __SSE4A__ +#include <ammintrin.h> +#endif + +#ifdef __FMA4__ +#include <fma4intrin.h> +#endif + +#ifdef __XOP__ +#include <xopintrin.h> +#endif + +// FIXME: LWP #endif /* __X86INTRIN_H */ diff --git a/renderscript/clang-include/xmmintrin.h b/renderscript/clang-include/xmmintrin.h index a0bc0bb..e616157 100644 --- a/renderscript/clang-include/xmmintrin.h +++ b/renderscript/clang-include/xmmintrin.h @@ -664,7 +664,7 @@ _mm_storer_ps(float *p, __m128 a) /* FIXME: We have to #define this because "sel" must be a constant integer, and Sema doesn't do any form of constant propagation yet. */ -#define _mm_prefetch(a, sel) (__builtin_prefetch((void *)(a), 0, sel)) +#define _mm_prefetch(a, sel) (__builtin_prefetch((void *)(a), 0, (sel))) static __inline__ void __attribute__((__always_inline__, __nodebug__)) _mm_stream_pi(__m64 *p, __m64 a) @@ -735,8 +735,9 @@ _mm_mulhi_pu16(__m64 a, __m64 b) return (__m64)__builtin_ia32_pmulhuw((__v4hi)a, (__v4hi)b); } -#define _mm_shuffle_pi16(a, n) \ - ((__m64)__builtin_ia32_pshufw(a, n)) +#define _mm_shuffle_pi16(a, n) __extension__ ({ \ + __m64 __a = (a); \ + (__m64)__builtin_ia32_pshufw((__v4hi)__a, (n)); }) static __inline__ void __attribute__((__always_inline__, __nodebug__)) _mm_maskmove_si64(__m64 d, __m64 n, char *p) @@ -774,11 +775,13 @@ _mm_setcsr(unsigned int i) __builtin_ia32_ldmxcsr(i); } -#define _mm_shuffle_ps(a, b, mask) \ - (__builtin_shufflevector((__v4sf)(a), (__v4sf)(b), \ - (mask) & 0x3, ((mask) & 0xc) >> 2, \ - (((mask) & 0x30) >> 4) + 4, \ - (((mask) & 0xc0) >> 6) + 4)) +#define _mm_shuffle_ps(a, b, mask) __extension__ ({ \ + __m128 __a = (a); \ + __m128 __b = (b); \ + (__m128)__builtin_shufflevector((__v4sf)__a, (__v4sf)__b, \ + (mask) & 0x3, ((mask) & 0xc) >> 2, \ + (((mask) & 0x30) >> 4) + 4, \ + (((mask) & 0xc0) >> 6) + 4); }) static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) _mm_unpackhi_ps(__m128 a, __m128 b) @@ -935,7 +938,7 @@ _mm_movemask_ps(__m128 a) #define _MM_FLUSH_ZERO_MASK (0x8000) #define _MM_FLUSH_ZERO_ON (0x8000) -#define _MM_FLUSH_ZERO_OFF (0x8000) +#define _MM_FLUSH_ZERO_OFF (0x0000) #define _MM_GET_EXCEPTION_MASK() (_mm_getcsr() & _MM_MASK_MASK) #define _MM_GET_EXCEPTION_STATE() (_mm_getcsr() & _MM_EXCEPT_MASK) diff --git a/renderscript/clang-include/xopintrin.h b/renderscript/clang-include/xopintrin.h new file mode 100644 index 0000000..d107be4 --- /dev/null +++ b/renderscript/clang-include/xopintrin.h @@ -0,0 +1,411 @@ +/*===---- xopintrin.h - FMA4 intrinsics ------------------------------------=== + * + * Permission is hereby granted, free of charge, to any person obtaining a copy + * of this software and associated documentation files (the "Software"), to deal + * in the Software without restriction, including without limitation the rights + * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell + * copies of the Software, and to permit persons to whom the Software is + * furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice shall be included in + * all copies or substantial portions of the Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE + * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, + * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN + * THE SOFTWARE. + * + *===-----------------------------------------------------------------------=== + */ + +#ifndef __X86INTRIN_H +#error "Never use <fma4intrin.h> directly; include <x86intrin.h> instead." +#endif + +#ifndef __XOPINTRIN_H +#define __XOPINTRIN_H + +#ifndef __XOP__ +# error "XOP instruction set is not enabled" +#else + +#include <fma4intrin.h> + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_maccs_epi16(__m128i __A, __m128i __B, __m128i __C) +{ + return (__m128i)__builtin_ia32_vpmacssww((__v8hi)__A, (__v8hi)__B, (__v8hi)__C); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_macc_epi16(__m128i __A, __m128i __B, __m128i __C) +{ + return (__m128i)__builtin_ia32_vpmacsww((__v8hi)__A, (__v8hi)__B, (__v8hi)__C); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_maccsd_epi16(__m128i __A, __m128i __B, __m128i __C) +{ + return (__m128i)__builtin_ia32_vpmacsswd((__v8hi)__A, (__v8hi)__B, (__v4si)__C); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_maccd_epi16(__m128i __A, __m128i __B, __m128i __C) +{ + return (__m128i)__builtin_ia32_vpmacswd((__v8hi)__A, (__v8hi)__B, (__v4si)__C); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_maccs_epi32(__m128i __A, __m128i __B, __m128i __C) +{ + return (__m128i)__builtin_ia32_vpmacssdd((__v4si)__A, (__v4si)__B, (__v4si)__C); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_macc_epi32(__m128i __A, __m128i __B, __m128i __C) +{ + return (__m128i)__builtin_ia32_vpmacsdd((__v4si)__A, (__v4si)__B, (__v4si)__C); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_maccslo_epi32(__m128i __A, __m128i __B, __m128i __C) +{ + return (__m128i)__builtin_ia32_vpmacssdql((__v4si)__A, (__v4si)__B, (__v2di)__C); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_macclo_epi32(__m128i __A, __m128i __B, __m128i __C) +{ + return (__m128i)__builtin_ia32_vpmacsdql((__v4si)__A, (__v4si)__B, (__v2di)__C); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_maccshi_epi32(__m128i __A, __m128i __B, __m128i __C) +{ + return (__m128i)__builtin_ia32_vpmacssdqh((__v4si)__A, (__v4si)__B, (__v2di)__C); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_macchi_epi32(__m128i __A, __m128i __B, __m128i __C) +{ + return (__m128i)__builtin_ia32_vpmacsdqh((__v4si)__A, (__v4si)__B, (__v2di)__C); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_maddsd_epi16(__m128i __A, __m128i __B, __m128i __C) +{ + return (__m128i)__builtin_ia32_vpmadcsswd((__v8hi)__A, (__v8hi)__B, (__v4si)__C); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_maddd_epi16(__m128i __A, __m128i __B, __m128i __C) +{ + return (__m128i)__builtin_ia32_vpmadcswd((__v8hi)__A, (__v8hi)__B, (__v4si)__C); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_haddw_epi8(__m128i __A) +{ + return (__m128i)__builtin_ia32_vphaddbw((__v16qi)__A); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_haddd_epi8(__m128i __A) +{ + return (__m128i)__builtin_ia32_vphaddbd((__v16qi)__A); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_haddq_epi8(__m128i __A) +{ + return (__m128i)__builtin_ia32_vphaddbq((__v16qi)__A); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_haddd_epi16(__m128i __A) +{ + return (__m128i)__builtin_ia32_vphaddwd((__v8hi)__A); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_haddq_epi16(__m128i __A) +{ + return (__m128i)__builtin_ia32_vphaddwq((__v8hi)__A); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_haddq_epi32(__m128i __A) +{ + return (__m128i)__builtin_ia32_vphadddq((__v4si)__A); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_haddw_epu8(__m128i __A) +{ + return (__m128i)__builtin_ia32_vphaddubw((__v16qi)__A); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_haddd_epu8(__m128i __A) +{ + return (__m128i)__builtin_ia32_vphaddubd((__v16qi)__A); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_haddq_epu8(__m128i __A) +{ + return (__m128i)__builtin_ia32_vphaddubq((__v16qi)__A); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_haddd_epu16(__m128i __A) +{ + return (__m128i)__builtin_ia32_vphadduwd((__v8hi)__A); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_haddq_epu16(__m128i __A) +{ + return (__m128i)__builtin_ia32_vphadduwq((__v8hi)__A); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_haddq_epu32(__m128i __A) +{ + return (__m128i)__builtin_ia32_vphaddudq((__v4si)__A); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_hsubw_epi8(__m128i __A) +{ + return (__m128i)__builtin_ia32_vphsubbw((__v16qi)__A); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_hsubd_epi16(__m128i __A) +{ + return (__m128i)__builtin_ia32_vphsubwd((__v8hi)__A); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_hsubq_epi32(__m128i __A) +{ + return (__m128i)__builtin_ia32_vphsubdq((__v4si)__A); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_cmov_si128(__m128i __A, __m128i __B, __m128i __C) +{ + return (__m128i)__builtin_ia32_vpcmov(__A, __B, __C); +} + +static __inline__ __m256i __attribute__((__always_inline__, __nodebug__)) +_mm256_cmov_si256(__m256i __A, __m256i __B, __m256i __C) +{ + return (__m256i)__builtin_ia32_vpcmov_256(__A, __B, __C); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_perm_epi8(__m128i __A, __m128i __B, __m128i __C) +{ + return (__m128i)__builtin_ia32_vpperm((__v16qi)__A, (__v16qi)__B, (__v16qi)__C); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_rot_epi8(__m128i __A, __m128i __B) +{ + return (__m128i)__builtin_ia32_vprotb((__v16qi)__A, (__v16qi)__B); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_rot_epi16(__m128i __A, __m128i __B) +{ + return (__m128i)__builtin_ia32_vprotw((__v8hi)__A, (__v8hi)__B); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_rot_epi32(__m128i __A, __m128i __B) +{ + return (__m128i)__builtin_ia32_vprotd((__v4si)__A, (__v4si)__B); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_rot_epi64(__m128i __A, __m128i __B) +{ + return (__m128i)__builtin_ia32_vprotq((__v2di)__A, (__v2di)__B); +} + +#define _mm_roti_epi8(A, N) __extension__ ({ \ + __m128i __A = (A); \ + (__m128i)__builtin_ia32_vprotbi((__v16qi)__A, (N)); }) + +#define _mm_roti_epi16(A, N) __extension__ ({ \ + __m128i __A = (A); \ + (__m128i)__builtin_ia32_vprotwi((__v8hi)__A, (N)); }) + +#define _mm_roti_epi32(A, N) __extension__ ({ \ + __m128i __A = (A); \ + (__m128i)__builtin_ia32_vprotdi((__v4si)__A, (N)); }) + +#define _mm_roti_epi64(A, N) __extension__ ({ \ + __m128i __A = (A); \ + (__m128i)__builtin_ia32_vprotqi((__v2di)__A, (N)); }) + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_shl_epi8(__m128i __A, __m128i __B) +{ + return (__m128i)__builtin_ia32_vpshlb((__v16qi)__A, (__v16qi)__B); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_shl_epi16(__m128i __A, __m128i __B) +{ + return (__m128i)__builtin_ia32_vpshlw((__v8hi)__A, (__v8hi)__B); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_shl_epi32(__m128i __A, __m128i __B) +{ + return (__m128i)__builtin_ia32_vpshld((__v4si)__A, (__v4si)__B); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_shl_epi64(__m128i __A, __m128i __B) +{ + return (__m128i)__builtin_ia32_vpshlq((__v2di)__A, (__v2di)__B); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_sha_epi8(__m128i __A, __m128i __B) +{ + return (__m128i)__builtin_ia32_vpshab((__v16qi)__A, (__v16qi)__B); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_sha_epi16(__m128i __A, __m128i __B) +{ + return (__m128i)__builtin_ia32_vpshaw((__v8hi)__A, (__v8hi)__B); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_sha_epi32(__m128i __A, __m128i __B) +{ + return (__m128i)__builtin_ia32_vpshad((__v4si)__A, (__v4si)__B); +} + +static __inline__ __m128i __attribute__((__always_inline__, __nodebug__)) +_mm_sha_epi64(__m128i __A, __m128i __B) +{ + return (__m128i)__builtin_ia32_vpshaq((__v2di)__A, (__v2di)__B); +} + +#define _mm_com_epu8(A, B, N) __extension__ ({ \ + __m128i __A = (A); \ + __m128i __B = (B); \ + (__m128i)__builtin_ia32_vpcomub((__v16qi)__A, (__v16qi)__B, (N)); }) + +#define _mm_com_epu16(A, B, N) __extension__ ({ \ + __m128i __A = (A); \ + __m128i __B = (B); \ + (__m128i)__builtin_ia32_vpcomuw((__v8hi)__A, (__v8hi)__B, (N)); }) + +#define _mm_com_epu32(A, B, N) __extension__ ({ \ + __m128i __A = (A); \ + __m128i __B = (B); \ + (__m128i)__builtin_ia32_vpcomud((__v4si)__A, (__v4si)__B, (N)); }) + +#define _mm_com_epu64(A, B, N) __extension__ ({ \ + __m128i __A = (A); \ + __m128i __B = (B); \ + (__m128i)__builtin_ia32_vpcomuq((__v2di)__A, (__v2di)__B, (N)); }) + +#define _mm_com_epi8(A, B, N) __extension__ ({ \ + __m128i __A = (A); \ + __m128i __B = (B); \ + (__m128i)__builtin_ia32_vpcomb((__v16qi)__A, (__v16qi)__B, (N)); }) + +#define _mm_com_epi16(A, B, N) __extension__ ({ \ + __m128i __A = (A); \ + __m128i __B = (B); \ + (__m128i)__builtin_ia32_vpcomw((__v8hi)__A, (__v8hi)__B, (N)); }) + +#define _mm_com_epi32(A, B, N) __extension__ ({ \ + __m128i __A = (A); \ + __m128i __B = (B); \ + (__m128i)__builtin_ia32_vpcomd((__v4si)__A, (__v4si)__B, (N)); }) + +#define _mm_com_epi64(A, B, N) __extension__ ({ \ + __m128i __A = (A); \ + __m128i __B = (B); \ + (__m128i)__builtin_ia32_vpcomq((__v2di)__A, (__v2di)__B, (N)); }) + +#define _mm_permute2_pd(X, Y, C, I) __extension__ ({ \ + __m128d __X = (X); \ + __m128d __Y = (Y); \ + __m128i __C = (C); \ + (__m128d)__builtin_ia32_vpermil2pd((__v2df)__X, (__v2df)__Y, \ + (__v2di)__C, (I)); }) + +#define _mm256_permute2_pd(X, Y, C, I) __extension__ ({ \ + __m256d __X = (X); \ + __m256d __Y = (Y); \ + __m256i __C = (C); \ + (__m256d)__builtin_ia32_vpermil2pd256((__v4df)__X, (__v4df)__Y, \ + (__v4di)__C, (I)); }) + +#define _mm_permute2_ps(X, Y, C, I) __extension__ ({ \ + __m128 __X = (X); \ + __m128 __Y = (Y); \ + __m128i __C = (C); \ + (__m128)__builtin_ia32_vpermil2ps((__v4sf)__X, (__v4sf)__Y, \ + (__v4si)__C, (I)); }) + +#define _mm256_permute2_ps(X, Y, C, I) __extension__ ({ \ + __m256 __X = (X); \ + __m256 __Y = (Y); \ + __m256i __C = (C); \ + (__m256)__builtin_ia32_vpermil2ps256((__v8sf)__X, (__v8sf)__Y, \ + (__v8si)__C, (I)); }) + +static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) +_mm_frcz_ss(__m128 __A) +{ + return (__m128)__builtin_ia32_vfrczss((__v4sf)__A); +} + +static __inline__ __m128d __attribute__((__always_inline__, __nodebug__)) +_mm_frcz_sd(__m128d __A) +{ + return (__m128d)__builtin_ia32_vfrczsd((__v2df)__A); +} + +static __inline__ __m128 __attribute__((__always_inline__, __nodebug__)) +_mm_frcz_ps(__m128 __A) +{ + return (__m128)__builtin_ia32_vfrczps((__v4sf)__A); +} + +static __inline__ __m128d __attribute__((__always_inline__, __nodebug__)) +_mm_frcz_pd(__m128d __A) +{ + return (__m128d)__builtin_ia32_vfrczpd((__v2df)__A); +} + +static __inline__ __m256 __attribute__((__always_inline__, __nodebug__)) +_mm256_frcz_ps(__m256 __A) +{ + return (__m256)__builtin_ia32_vfrczps256((__v8sf)__A); +} + +static __inline__ __m256d __attribute__((__always_inline__, __nodebug__)) +_mm256_frcz_pd(__m256d __A) +{ + return (__m256d)__builtin_ia32_vfrczpd256((__v4df)__A); +} + +#endif /* __XOP__ */ + +#endif /* __XOPINTRIN_H */ |