!7230 [MSLITE][Develop] optimize arm cpu arithmetic: remove redundant code

Merge pull request !7230 from yangruoqi713/lite
4 years ago · b80fa15918
parent 2801429db0 43d0020564
commit b80fa15918
3 changed files with 1061 additions and 2116 deletions
--- a/mindspore/lite/nnacl/fp16/arithmetic_fp16.c
+++ b/mindspore/lite/nnacl/fp16/arithmetic_fp16.c
--- a/mindspore/lite/nnacl/fp32/arithmetic.c
+++ b/mindspore/lite/nnacl/fp32/arithmetic.c
--- a/mindspore/lite/nnacl/fp32/lstm.c
+++ b/mindspore/lite/nnacl/fp32/lstm.c
@ -42,7 +42,7 @@ void MatMulAcc(float *output, const float *input, const float *weight, int rows,
      int index = 0;
 #ifdef ENABLE_ARM
      float32x4_t out = vdupq_n_f32(0.0f);
-      for (; index < inner_size - 4; index += 4) {
+      for (; index <= inner_size - 4; index += 4) {
        float32x4_t in_0 = vld1q_f32(input_col + index);
        float32x4_t in_1 = vld1q_f32(weight_col + index);
        out = vmlaq_f32(out, in_1, in_0);
@ -66,7 +66,7 @@ void MatMulAcc(float *output, const float *input, const float *weight, int rows,
 void ElementMulAcc(const float *input0, const float *input1, float *output, int element_size) {
  int index = 0;
 #ifdef ENABLE_ARM
-  for (; index < element_size - 4; index += 4) {
+  for (; index <= element_size - 4; index += 4) {
    float32x4_t in_0 = vld1q_f32(input0 + index);
    float32x4_t in_1 = vld1q_f32(input1 + index);
    float32x4_t out = vld1q_f32(output + index);