|
|
#pragma once
|
|
|
#if defined(__GNUC__) && (defined(__x86_64__) || defined(__i386__))
|
|
|
|
|
|
#include <x86intrin.h>
|
|
|
#elif defined(__clang__) && (defined(__ARM_NEON__) || defined(__aarch64__))
|
|
|
|
|
|
#include <arm_neon.h>
|
|
|
#if defined(__ARM_FEATURE_SVE)
|
|
|
|
|
|
#include <arm_sve.h>
|
|
|
#endif
|
|
|
#elif defined(_MSC_VER)
|
|
|
|
|
|
#include <intrin.h>
|
|
|
#if _MSC_VER <= 1900
|
|
|
#define _mm256_extract_epi64(X, Y) \
|
|
|
(_mm_extract_epi64(_mm256_extractf128_si256(X, Y >> 1), Y % 2))
|
|
|
#define _mm256_extract_epi32(X, Y) \
|
|
|
(_mm_extract_epi32(_mm256_extractf128_si256(X, Y >> 2), Y % 4))
|
|
|
#define _mm256_extract_epi16(X, Y) \
|
|
|
(_mm_extract_epi16(_mm256_extractf128_si256(X, Y >> 3), Y % 8))
|
|
|
#define _mm256_extract_epi8(X, Y) \
|
|
|
(_mm_extract_epi8(_mm256_extractf128_si256(X, Y >> 4), Y % 16))
|
|
|
#endif
|
|
|
#elif defined(__GNUC__) && (defined(__ARM_NEON__) || defined(__aarch64__))
|
|
|
|
|
|
#include <arm_neon.h>
|
|
|
#if defined(__ARM_FEATURE_SVE)
|
|
|
|
|
|
#include <arm_sve.h>
|
|
|
#endif
|
|
|
#if defined(MISSING_ARM_VLD1)
|
|
|
#include <ATen/cpu/vec/vec256/missing_vld1_neon.h>
|
|
|
#elif defined(MISSING_ARM_VST1)
|
|
|
#include <ATen/cpu/vec/vec256/missing_vst1_neon.h>
|
|
|
#endif
|
|
|
#elif defined(__GNUC__) && defined(__IWMMXT__)
|
|
|
|
|
|
#include <mmintrin.h>
|
|
|
#elif defined(__s390x__)
|
|
|
|
|
|
|
|
|
#elif (defined(__GNUC__) || defined(__xlC__)) && \
|
|
|
(defined(__VEC__) || defined(__ALTIVEC__))
|
|
|
|
|
|
#include <altivec.h>
|
|
|
|
|
|
|
|
|
#undef bool
|
|
|
#undef vector
|
|
|
#undef pixel
|
|
|
#elif defined(__GNUC__) && defined(__SPE__)
|
|
|
|
|
|
#include <spe.h>
|
|
|
#endif
|
|
|
|