blas_neon: fix compiler errors in aarch64/Linux

With stricter compilers, fp16 codes are not compilable.
To enable testing in non-android, fix type mismatches.

Signed-off-by: MyungJoo Ham <[email protected]>

nntrainer/tensor/blas_neon.cpp

@@ -13,6 +13,7 @@
  */
 
 #include <blas_neon.h>
+#include <memory>
 #include <nntrainer_error.h>
 namespace nntrainer::neon {
 
@@ -339,7 +340,7 @@ void scopy_neon_int4_to_fp32(const unsigned int N, const uint8_t *X, float *Y) {
 
   // processing remaining batch of 8
   for (; (N - idx) >= 8; idx += 8) {
-    int8x8_t batch = vld1_u8(&X[idx]);
+    uint8x8_t batch = vld1_u8(&X[idx]);
 
     unsigned int i = 0;
     for (; i < 8; ++i) {
@@ -444,7 +445,7 @@ void sgemv_neon_fp16(const __fp16 *A, const __fp16 *X, __fp16 *Y, uint32_t rows,
                      uint32_t cols, float alpha, float beta) {
   const int batch = 0;
   const __fp16 *__restrict x;
-  float Y32[rows];
+  float *Y32 = new float[rows];
 
   unsigned int idx = 0;
 
@@ -688,20 +689,21 @@ void sgemv_neon_fp16(const __fp16 *A, const __fp16 *X, __fp16 *Y, uint32_t rows,
 
       float32x4_t y0 = vmulq_f32(wvec0_3, x0_3);
 
-      for (int k = 0; k < cols - idx; ++k) {
+      for (unsigned int k = 0; k < cols - idx; ++k) {
         Y32[j] += y0[k];
       }
     }
   }
 
   scopy_neon_fp32_to_fp16(rows, Y32, Y);
+  delete[] Y32;
   return;
 }
 
 void sgemv_transpose_neon_fp16(const __fp16 *A, const __fp16 *X, __fp16 *Y,
                                uint32_t rows, uint32_t cols, float alpha,
                                float beta) {
-  float Y32[cols];
+  float *Y32 = new float[cols];
   const int batch = 20;
   unsigned int idx = 0;
   for (; cols - idx >= 8; idx += 8) {
@@ -1092,7 +1094,7 @@ void sgemv_transpose_neon_fp16(const __fp16 *A, const __fp16 *X, __fp16 *Y,
       if (cols != idx) {
         float y0_3[4];
         float wvec0_3[4];
-        for (int j = 0; j < cols - idx; ++j) {
+        for (unsigned int j = 0; j < cols - idx; ++j) {
           y0_3[j] = Y32[idx + j];
           wvec0_3[j] = A[i * cols + idx + j];
         }
@@ -1111,6 +1113,7 @@ void sgemv_transpose_neon_fp16(const __fp16 *A, const __fp16 *X, __fp16 *Y,
     }
   }
   scopy_neon_fp32_to_fp16(cols, Y32, Y);
+  delete[] Y32;
   return;
 }
 
@@ -1350,7 +1353,7 @@ void scopy_neon_int4_to_fp16(const unsigned int N, const uint8_t *X,
 
   // processing remaining batch of 8
   for (; (N - idx) >= 8; idx += 8) {
-    int8x8_t batch = vld1_u8(&X[idx]);
+    uint8x8_t batch = vld1_u8(&X[idx]);
 
     for (int i = 0; i < 8; ++i) {
       low0 = batch[i] >> 4;
@@ -1448,7 +1451,7 @@ void scopy_neon_int8_to_fp32(const unsigned int N, const uint8_t *X, float *Y) {
 }
 
 void scopy_neon_fp16_to_fp32(const unsigned int N, const __fp16 *X, float *Y) {
-  int idx = 0;
+  unsigned int idx = 0;
 
   for (; N - idx >= 8; idx += 8) {
     float32x4_t y1 = vcvt_f32_f16(vld1_f16(&X[idx]));
@@ -1470,7 +1473,7 @@ void scopy_neon_fp16_to_fp32(const unsigned int N, const __fp16 *X, float *Y) {
 }
 
 void scopy_neon_fp32_to_fp16(const unsigned int N, const float *X, __fp16 *Y) {
-  int idx = 0;
+  unsigned int idx = 0;
 
   for (; N - idx >= 8; idx += 8) {
     float32x4_t x1 = vld1q_f32(&X[idx]);
@@ -1556,14 +1559,14 @@ void sgemm_neon_fp16(const __fp16 *A, const __fp16 *B, __fp16 *C, uint32_t M,
   // performing beta*C
   unsigned int idx = 0;
   unsigned int size = M * N;
-  for (; idx < (size - idx) >= 8; idx += 8) {
+  for (; idx < (size - idx) && (size - idx) >= 8; idx += 8) {
     float16x8_t c = vmulq_n_f16(vld1q_f16(&C[idx]), static_cast<__fp16>(beta));
 
     vst1q_f32(&C32[idx], vcvt_f32_f16(vget_low_f16(c)));
     vst1q_f32(&C32[idx + 4], vcvt_f32_f16(vget_high_f16(c)));
   }
   // remaining 4
-  for (; idx < (size - idx) >= 4; idx += 4) {
+  for (; idx < (size - idx) && (size - idx) >= 4; idx += 4) {
     float16x4_t c = vmul_n_f16(vld1_f16(&C[idx]), static_cast<__fp16>(beta));
 
     vst1q_f32(&C32[idx], vcvt_f32_f16(c));
@@ -2025,7 +2028,7 @@ void sgemm_neon_fp16_transAB(const __fp16 *A, const __fp16 *B, float *C,
 void elementwise_vector_multiplication_neon_fp16(const unsigned int N,
                                                  const __fp16 *X,
                                                  const __fp16 *Y, __fp16 *Z) {
-  int i = 0;
+  unsigned int i = 0;
   for (; N - i >= 8; i += 8) {
     float16x8_t x0_7 = vld1q_f16(&X[i]);
     float16x8_t y0_7 = vld1q_f16(&Y[i]);
@@ -2042,7 +2045,7 @@ void elementwise_vector_multiplication_neon_fp16(const unsigned int N,
 void elementwise_vector_addition_neon_fp16(const unsigned int N,
                                            const __fp16 *X, const __fp16 *Y,
                                            __fp16 *Z) {
-  int i = 0;
+  unsigned int i = 0;
   for (; N - i >= 8; i += 8) {
     float16x8_t x0_7 = vld1q_f16(&X[i]);
     float16x8_t y0_7 = vld1q_f16(&Y[i]);