使用SIMD实现矩阵乘法。_C_Makefile_下载.zip

#ifndef _MUTIPLY_H #define _MULTIPLY_H #include "list.h" void Strassen_save_result(int *res, int *m11, int *m12, int *m21, int *m22, int n); int *Strassen_helper(int *res, int *src1, int *src2, int n); int *Strassen_add(int *res, int *x, int *y, int n); int *Strassen_sub(int *res, int *x, int *y, int n); int *Strassen_mul(int *res, int *x, int *y, int n); extern algo_t *list; FUNC_IMPL(naive) { memset(dst, 0, sizeof(int) * src1_h * src2_w); for (int i = 0; i < src1_h; ++i) { for (int j = 0; j < src2_w; ++j) { for (int k = 0; k < src2_h; ++k) { dst[i * src2_w + j] += src1[i * src1_w + k] * src2[k * src2_w + j]; } } } } FUNC_IMPL(submatrix) { memset(dst, 0, sizeof(int) * src1_h * src2_w); for (int i = 0; i < src1_h; i += 4) { for (int j = 0; j < src2_w; j += 4) { for (int k = 0; k < src2_h; k += 4) { for (int i2 = 0; i2 < 4; ++i2) { for (int j2 = 0; j2 < 4; ++j2) { for (int k2 = 0; k2 < 4; ++k2) { dst[(i + i2) * src2_w + (j + j2)] += src1[(i + i2) * src1_w + (k + k2)] * src2[(k + k2) * src2_w + (j + j2)]; } } } } } } } FUNC_IMPL(sse) { for (int x = 0; x < src1_h; x += 4) { for (int y = 0; y < src2_w; y += 4) { __m128i des0 = _mm_setzero_si128 (); __m128i des1 = _mm_setzero_si128 (); __m128i des2 = _mm_setzero_si128 (); __m128i des3 = _mm_setzero_si128 (); for (int k = 0; k < src2_w; k += 4) { __m128i I0 = _mm_load_si128((__m128i *)(src1 + (x + 0) * src1_w + k)); __m128i I1 = _mm_load_si128((__m128i *)(src1 + (x + 1) * src1_w + k)); __m128i I2 = _mm_load_si128((__m128i *)(src1 + (x + 2) * src1_w + k)); __m128i I3 = _mm_load_si128((__m128i *)(src1 + (x + 3) * src1_w + k)); __m128i I4 = _mm_set_epi32 (src2[(k+3) * src2_w + y], src2[(k+2) * src2_w + y], src2[(k+1) * src2_w + y], src2[k * src2_w + y]); __m128i I5 = _mm_set_epi32 (src2[(k+3) * src2_w + (y+1)], src2[(k+2) * src2_w + (y+1)], src2[(k+1) * src2_w + (y+1)], src2[(k+0) * src2_w + (y+1)]); __m128i I6 = _mm_set_epi32 (src2[(k+3) * src2_w + (y+2)], src2[(k+2) * src2_w + (y+2)], src2[(k+1) * src2_w + (y+2)], src2[(k+0) * src2_w + (y+2)]); __m128i I7 = _mm_set_epi32 (src2[(k+3) * src2_w + (y+3)], src2[(k+2) * src2_w + (y+3)], src2[(k+1) * src2_w + (y+3)], src2[(k+0) * src2_w + (y+3)]); __m128i T0 = _mm_mullo_epi32(I0, I4); __m128i T1 = _mm_mullo_epi32(I0, I5); __m128i T2 = _mm_mullo_epi32(I0, I6); __m128i T3 = _mm_mullo_epi32(I0, I7); __m128i T4 = _mm_mullo_epi32(I1, I4); __m128i T5 = _mm_mullo_epi32(I1, I5); __m128i T6 = _mm_mullo_epi32(I1, I6); __m128i T7 = _mm_mullo_epi32(I1, I7); __m128i T8 = _mm_mullo_epi32(I2, I4); __m128i T9 = _mm_mullo_epi32(I2, I5); __m128i T10 = _mm_mullo_epi32(I2, I6); __m128i T11 = _mm_mullo_epi32(I2, I7); __m128i T12 = _mm_mullo_epi32(I3, I4); __m128i T13 = _mm_mullo_epi32(I3, I5); __m128i T14 = _mm_mullo_epi32(I3, I6); __m128i T15 = _mm_mullo_epi32(I3, I7); __m128i T16 = _mm_unpacklo_epi32(T0, T1); __m128i T17 = _mm_unpacklo_epi32(T2, T3); __m128i T18 = _mm_unpackhi_epi32(T0, T1); __m128i T19 = _mm_unpackhi_epi32(T2, T3); __m128i T20 = _mm_unpacklo_epi64(T16, T17); __m128i T21 = _mm_unpackhi_epi64(T16, T17); __m128i T22 = _mm_unpacklo_epi64(T18, T19); __m128i T23 = _mm_unpackhi_epi64(T18, T19); T20 = _mm_add_epi32(T20, T21); T20 = _mm_add_epi32(T20, T22); T20 = _mm_add_epi32(T20, T23); des0 = _mm_add_epi32(T20, des0); T16 = _mm_unpacklo_epi32(T4, T5); T17 = _mm_unpacklo_epi32(T6, T7); T18 = _mm_unpackhi_epi32(T4, T5); T19 = _mm_unpackhi_epi32(T6, T7); T20 = _mm_unpacklo_epi64(T16, T17); T21 = _mm_unpackhi_epi64(T16, T17); T22 = _mm_unpacklo_epi64(T18, T19); T23 = _mm_unpackhi_epi64(T18, T19); T20 = _mm_add_epi32(T20, T21); T20 = _mm_add_epi32(T20, T22); T20 = _mm_add_epi32(T20, T23); des1 = _mm_add_epi32(T20, des1); T16 = _mm_unpacklo_epi32(T8, T9); T17 = _mm_unpacklo_epi32(T10, T11); T18 = _mm_unpackhi_epi32(T8, T9); T19 = _mm_unpackhi_epi32(T10, T11); T20 = _mm_unpacklo_epi64(T16, T17); T21 = _mm_unpackhi_epi64(T16, T17); T22 = _mm_unpacklo_epi64(T18, T19); T23 = _mm_unpackhi_epi64(T18, T19); T20 = _mm_add_epi32(T20, T21); T20 = _mm_add_epi32(T20, T22); T20 = _mm_add_epi32(T20, T23); des2 = _mm_add_epi32(T20, des2); T16 = _mm_unpacklo_epi32(T12, T13); T17 = _mm_unpacklo_epi32(T14, T15); T18 = _mm_unpackhi_epi32(T12, T13); T19 = _mm_unpackhi_epi32(T14, T15); T20 = _mm_unpacklo_epi64(T16, T17); T21 = _mm_unpackhi_epi64(T16, T17); T22 = _mm_unpacklo_epi64(T18, T19); T23 = _mm_unpackhi_epi64(T18, T19); T20 = _mm_add_epi32(T20, T21); T20 = _mm_add_epi32(T20, T22); T20 = _mm_add_epi32(T20, T23); des3 = _mm_add_epi32(T20, des3); } _mm_store_si128((__m128i *)(dst + ((x + 0) * src2_w) + y), des0); _mm_store_si128((__m128i *)(dst + ((x + 1) * src2_w) + y), des1); _mm_store_si128((__m128i *)(dst + ((x + 2) * src2_w) + y), des2); _mm_store_si128((__m128i *)(dst + ((x + 3) * src2_w) + y), des3); } } } FUNC_IMPL(sse_prefetch) { for (int x = 0; x < src1_h; x += 4) { for (int y = 0; y < src2_w; y += 4) { __m128i des0 = _mm_setzero_si128 (); __m128i des1 = _mm_setzero_si128 (); __m128i des2 = _mm_setzero_si128 (); __m128i des3 = _mm_setzero_si128 (); for (int k = 0; k < src2_w; k += 4) { #define SSE_PFDIST 8 _mm_prefetch(src2 + (k + SSE_PFDIST + 0) * src2_w + y, _MM_HINT_T1); _mm_prefetch(src2 + (k + SSE_PFDIST + 1) * src2_w + y, _MM_HINT_T1); _mm_prefetch(src2 + (k + SSE_PFDIST + 2) * src2_w + y, _MM_HINT_T1); _mm_prefetch(src2 + (k + SSE_PFDIST + 3) * src2_w + y, _MM_HINT_T1); __m128i I0 = _mm_load_si128((__m128i *)(src1 + (x + 0) * src1_w + k)); __m128i I1 = _mm_load_si128((__m128i *)(src1 + (x + 1) * src1_w + k)); __m128i I2 = _mm_load_si128((__m128i *)(src1 + (x + 2) * src1_w + k)); __m128i I3 = _mm_load_si128((__m128i *)(src1 + (x + 3) * src1_w + k)); __m128i I4 = _mm_set_epi32 (src2[(k+3) * src2_w + y], src2[(k+2) * src2_w + y], src2[(k+1) * src2_w + y], src2[k *