[ hgemm/trivial ] Use aligned memory allocation in K1 transpose non_M…

…8_case

- Since K1 GEMM does not use data packing, I did not use aligned memory allocation.
- However, for SIMD situation, using such is more preferred.

**Self evaluation:**
1. Build test:     [X]Passed [ ]Failed [ ]Skipped
2. Run test:     [X]Passed [ ]Failed [ ]Skipped

Signed-off-by: skykongkong8 <[email protected]>

nntrainer/tensor/hgemm/hgemm.cpp

@@ -106,7 +106,7 @@ void hgemm_K1_transA(unsigned int M, unsigned int N, unsigned int K,
                      const __fp16 *A, unsigned int lda, const __fp16 *B,
                      unsigned int ldb, __fp16 *C, unsigned int ldc, float alpha,
                      float beta) {
-  __fp16 *A_T = new __fp16[M * K];
+  __fp16 *A_T = alignedMalloc(M * K);
 
   transpose_neon<__fp16>(K, M, A, M, A_T, K);
 
@@ -119,7 +119,7 @@ void hgemm_K1_transB(unsigned int M, unsigned int N, unsigned int K,
                      const __fp16 *A, unsigned int lda, const __fp16 *B,
                      unsigned int ldb, __fp16 *C, unsigned int ldc, float alpha,
                      float beta) {
-  __fp16 *B_T = new __fp16[K * N];
+  __fp16 *B_T = alignedMalloc(K * N);
 
   transpose_neon<__fp16>(N, K, B, K, B_T, N);
 
@@ -132,8 +132,8 @@ void hgemm_K1_transAB(unsigned int M, unsigned int N, unsigned int K,
                       const __fp16 *A, unsigned int lda, const __fp16 *B,
                       unsigned int ldb, __fp16 *C, unsigned int ldc,
                       float alpha, float beta) {
-  __fp16 *A_T = new __fp16[M * K];
-  __fp16 *B_T = new __fp16[K * N];
+  __fp16 *A_T = alignedMalloc(M * K);
+  __fp16 *B_T = alignedMalloc(K * N);
 
   transpose_neon<__fp16>(K, M, A, M, A_T, K);
   transpose_neon<__fp16>(N, K, B, K, B_T, N);