Merge branch 'develop' into develop_python_develop

src/fpga/bias_scale.cpp

@@ -27,9 +27,6 @@ void align_element(float **data_in, int num_per_div_before_alignment, int num) {
       (num + num_per_div_before_alignment - 1) / num_per_div_before_alignment;
   int num_per_div_after_alignment =
       align_to_x(num_per_div_before_alignment, BS_NUM_ALIGNMENT);
-  if (num_per_div_before_alignment == num_per_div_after_alignment) {
-    return;
-  }
   int num_element =
       2 * div_num * num_per_div_after_alignment;  // including bias & scale
   float *ptr_aligned =

src/operators/math/gemm.cpp

@@ -3379,7 +3379,7 @@ void Gemm::SgemmWithBn_omp(int m, int n, int k, float alpha, const float *A,
     // 对 B 分块
     NC = L1 / (KC * sizeof(float));
     if (NC == 0) {
-      NC == NR;
+      NC = NR;
     } else {
       int nblock_num = (n + NC - 1) / NC;
       NC = (n + nblock_num - 1) / nblock_num;

src/operators/math/gemm.h

@@ -22,9 +22,11 @@ limitations under the License. */
 #define C(i, j) C[(i)*ldc + (j)]
 
 #if __aarch64__
+#define MR_INT8 4
 #define MR 6
 #define NR 16
 #else
+#define MR_INT8 4
 #define MR 6
 #define NR 8
 #endif
@@ -189,6 +191,8 @@ void PackMatrixB(int k, int n, int n_tail, const float *B, int ldb,
 
   // 8 bits function cluster begins
   // 8 bits int small block inner product
+  void AddDot4x8(int32_t k, const int8_t *a, const int8_t *b, int32_t *c,
+                 int32_t ldc);
   void AddDot6x8(int32_t k, const int8_t *a, const int8_t *b, int32_t *c,
                  int32_t ldc);
 
@@ -199,6 +203,8 @@ void PackMatrixB(int k, int n, int n_tail, const float *B, int ldb,
                            int8_t *bias);
 
   // 8 bits int pack function
+  void PackMatrixA_4r(int32_t m, int32_t k, int32_t m_tail, const int8_t *A,
+                      int32_t lda, int8_t *buffer);
   void PackMatrixA_6r(int32_t m, int32_t k, int32_t m_tail, const int8_t *A,
                       int32_t lda, int8_t *buffer);
   void PackMatrixB_8c(int32_t k, int32_t n, int32_t n_tail, const int8_t *B,

test/operators/test_mul_op.cpp

@@ -12,7 +12,6 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 
-#include <stdint-gcc.h>
 #include "../test_helper.h"
 #include "../test_include.h"
 #include "operators/mul_op.h"