wip

Tencent · Oct 8, 2023 · 4cc7b2b · 4cc7b2b
1 parent 4198cb1
commit 4cc7b2b
Showing 1 changed file with 14 additions and 2 deletions.
diff --git a/src/layer/x86/convolution_3x3_winograd_int8.h b/src/layer/x86/convolution_3x3_winograd_int8.h
@@ -235,6 +235,7 @@ static void transpose_pack_B_tile_int8(const Mat& B, Mat& BT, int batch, int max
 
         int jj = 0;
 #if __SSE2__
+#if defined(__x86_64__) || defined(_M_X64)
 #if __AVX512F__
         for (; jj + 15 < max_jj; jj += 16)
         {
@@ -429,6 +430,7 @@ static void transpose_pack_B_tile_int8(const Mat& B, Mat& BT, int batch, int max
                 pp += 8;
             }
         }
+#endif // defined(__x86_64__) || defined(_M_X64)
         for (; jj + 3 < max_jj; jj += 4)
         {
             const short* p0 = B;
@@ -611,6 +613,7 @@ static void gemm_transB_packed_tile_int8(const Mat& AT_tile, const Mat& BT_tile,
             const short* pB = BT_tile.row<const short>(b);
 
             int jj = 0;
+#if defined(__x86_64__) || defined(_M_X64)
             for (; jj + 15 < max_jj; jj += 16)
             {
                 const short* pA = pAT;
@@ -1094,6 +1097,7 @@ static void gemm_transB_packed_tile_int8(const Mat& AT_tile, const Mat& BT_tile,
                 _mm512_store_si512((__m512i*)(outptr + 16 * 7), _sum7);
                 outptr += 16 * 8;
             }
+#endif // defined(__x86_64__) || defined(_M_X64)
             for (; jj + 3 < max_jj; jj += 4)
             {
                 const short* pA = pAT;
@@ -1323,6 +1327,7 @@ static void gemm_transB_packed_tile_int8(const Mat& AT_tile, const Mat& BT_tile,
             const short* pB = BT_tile.row<const short>(b);
 
             int jj = 0;
+#if defined(__x86_64__) || defined(_M_X64)
 #if __AVX512F__
             for (; jj + 15 < max_jj; jj += 16)
             {
@@ -1406,9 +1411,9 @@ static void gemm_transB_packed_tile_int8(const Mat& AT_tile, const Mat& BT_tile,
                     __m512i _pA00 = _mm512_inserti32x8(_mm512_castsi256_si512(_pA0), _pA0, 1);
                     __m512i _pA11 = _mm512_permutex_epi64(_pA00, _MM_SHUFFLE(1, 0, 3, 2));
 
-                    __m512i _pB1 = _mm512_shuffle_epi32(_pB0, _MM_SHUFFLE(0, 3, 2, 1));
+                    __m512i _pB1 = _mm512_shuffle_epi32(_pB0, _MM_PERM_ADCB);
                     __m512i _pB2 = _mm512_permutex_epi64(_pB0, _MM_SHUFFLE(2, 3, 0, 1));
-                    __m512i _pB3 = _mm512_shuffle_epi32(_pB0, _MM_SHUFFLE(2, 1, 0, 3));
+                    __m512i _pB3 = _mm512_shuffle_epi32(_pB0, _MM_PERM_CBAD);
 
                     __m512i _s0 = _mm512_mullo_epi32(_pA00, _pB0);
                     __m512i _s1 = _mm512_mullo_epi32(_pA00, _pB1);
@@ -1796,6 +1801,7 @@ static void gemm_transB_packed_tile_int8(const Mat& AT_tile, const Mat& BT_tile,
                 outptr += 8 * 8;
 #endif // __AVX512F__
             }
+#endif // defined(__x86_64__) || defined(_M_X64)
             for (; jj + 3 < max_jj; jj += 4)
             {
                 const short* pA = pAT;
@@ -2020,6 +2026,7 @@ static void gemm_transB_packed_tile_int8(const Mat& AT_tile, const Mat& BT_tile,
             const short* pB = BT_tile.row<const short>(b);
 
             int jj = 0;
+#if defined(__x86_64__) || defined(_M_X64)
 #if __AVX512F__
             for (; jj + 15 < max_jj; jj += 16)
             {
@@ -2414,6 +2421,7 @@ static void gemm_transB_packed_tile_int8(const Mat& AT_tile, const Mat& BT_tile,
                 outptr += 32;
 #endif // __AVX2__
             }
+#endif // defined(__x86_64__) || defined(_M_X64)
             for (; jj + 3 < max_jj; jj += 4)
             {
                 const short* pA = pAT;
@@ -2676,6 +2684,7 @@ static void gemm_transB_packed_tile_int8(const Mat& AT_tile, const Mat& BT_tile,
 
             int jj = 0;
 #if __SSE2__
+#if defined(__x86_64__) || defined(_M_X64)
 #if __AVX512F__
             for (; jj + 15 < max_jj; jj += 16)
             {
@@ -2875,6 +2884,7 @@ static void gemm_transB_packed_tile_int8(const Mat& AT_tile, const Mat& BT_tile,
                 outptr += 16;
 #endif // __AVX2__
             }
+#endif // defined(__x86_64__) || defined(_M_X64)
             for (; jj + 3 < max_jj; jj += 4)
             {
                 const short* pA = pAT;
@@ -3041,6 +3051,7 @@ static void gemm_transB_packed_tile_int8(const Mat& AT_tile, const Mat& BT_tile,
 
             int jj = 0;
 #if __SSE2__
+#if defined(__x86_64__) || defined(_M_X64)
 #if __AVX512F__
             for (; jj + 15 < max_jj; jj += 16)
             {
@@ -3136,6 +3147,7 @@ static void gemm_transB_packed_tile_int8(const Mat& AT_tile, const Mat& BT_tile,
                 _mm_storeu_si128((__m128i*)(outptr + 4), _sum1);
                 outptr += 8;
             }
+#endif // defined(__x86_64__) || defined(_M_X64)
             for (; jj + 3 < max_jj; jj += 4)
             {
                 const short* pA = pAT;