Fix bug for compute MatMul E=1 for multi-thread

xiaying · xiaying · commit d766c6d7c099 · 2021-01-11T14:04:25.000+08:00
diff --git a/source/backend/cpu/x86_x64/avx/CommonOptFunction.cpp b/source/backend/cpu/x86_x64/avx/CommonOptFunction.cpp
@@ -747,7 +747,7 @@ void _AVX_MNNComputeMatMulForE_1(const float* A, const float* B, float* C, const
             }
             _mm256_storeu_ps(C + 8 * y, sumValue);
         }
-        for (int y=hR; y<h; y+=numberThread) {
+        for (int y = hR + tId; y<h; y+=numberThread) {
             auto bs = B + y;
             float sumValue = 0.0f;
             if (biasPtr != nullptr) {
@@ -802,7 +802,7 @@ void _AVX_MNNComputeMatMulForE_1FMA(const float* A, const float* B, float* C, co
             }
             _mm256_storeu_ps(C + 8 * y, sumValue);
         }
-        for (int y=hR; y<h; y+=numberThread) {
+        for (int y= hR + tId; y<h; y+=numberThread) {
             auto bs = B + y;
             float sumValue = 0.0f;
             if (biasPtr != nullptr) {
diff --git a/source/backend/cpu/x86_x64/sse/CommonOptFunction.cpp b/source/backend/cpu/x86_x64/sse/CommonOptFunction.cpp
@@ -670,7 +670,7 @@ void _SSE_MNNComputeMatMulForE_1(const float* A, const float* B, float* C, const
             }
             _mm_storeu_ps(C + 4 * y, sumValue);
         }
-        for (int y=hR; y<h; y+=numberThread) {
+        for (int y=hR + tId; y<h; y+=numberThread) {
             auto bs = B + y;
             float sumValue = 0.0f;
             if (biasPtr != nullptr) {

Original file line number	Diff line number	Diff line change
`@@ -747,7 +747,7 @@ void _AVX_MNNComputeMatMulForE_1(const float* A, const float* B, float* C, const`
`747`	`747`	`}`
`748`	`748`	`_mm256_storeu_ps(C + 8 * y, sumValue);`
`749`	`749`	`}`
`750`		`- for (int y=hR; y<h; y+=numberThread) {`
	`750`	`+ for (int y = hR + tId; y<h; y+=numberThread) {`
`751`	`751`	`auto bs = B + y;`
`752`	`752`	`float sumValue = 0.0f;`
`753`	`753`	`if (biasPtr != nullptr) {`
`@@ -802,7 +802,7 @@ void _AVX_MNNComputeMatMulForE_1FMA(const float* A, const float* B, float* C, co`
`802`	`802`	`}`
`803`	`803`	`_mm256_storeu_ps(C + 8 * y, sumValue);`
`804`	`804`	`}`
`805`		`- for (int y=hR; y<h; y+=numberThread) {`
	`805`	`+ for (int y= hR + tId; y<h; y+=numberThread) {`
`806`	`806`	`auto bs = B + y;`
`807`	`807`	`float sumValue = 0.0f;`
`808`	`808`	`if (biasPtr != nullptr) {`
Original file line number	Diff line number	Diff line change
`@@ -670,7 +670,7 @@ void _SSE_MNNComputeMatMulForE_1(const float* A, const float* B, float* C, const`
`670`	`670`	`}`
`671`	`671`	`_mm_storeu_ps(C + 4 * y, sumValue);`
`672`	`672`	`}`
`673`		`- for (int y=hR; y<h; y+=numberThread) {`
	`673`	`+ for (int y=hR + tId; y<h; y+=numberThread) {`
`674`	`674`	`auto bs = B + y;`
`675`	`675`	`float sumValue = 0.0f;`
`676`	`676`	`if (biasPtr != nullptr) {`