Skip to content

P
Paddle-Lite

PaddlePaddle / Paddle-Lite

通知 338
Star 4
Fork 1
P
Paddle-Lite
提交 ec335c5b 编写于 作者: C chenjiaoAngel
浏览文件
操作

fix format, test=develop

上级 54651d90
隐藏空白更改
内联 并排
Showing with 50 addition and 104 deletion
lite/backends/arm/math/conv5x5s1_depthwise_fp32.cc
浏览文件 @ ec335c5b
...@@ -1681,16 +1681,10 @@ inline void compute_all_padding_mid_out2(float* dout0, ...@@ -1681,16 +1681,10 @@ inline void compute_all_padding_mid_out2(float* dout0,
din_ptr_arr[tmp1], weights[num], bias[0], weights[6][0], 4); din_ptr_arr[tmp1], weights[num], bias[0], weights[6][0], 4);
din_ptr_arr[tmp1]++; din_ptr_arr[tmp1]++;
for (int i = 0; i < num; i++) { for (int i = 0; i < num; i++) {
sum += compute_one_data_post(din_ptr_arr[tmp - i], sum += compute_one_data_post(
weights[tmp - i], din_ptr_arr[tmp - i], weights[tmp - i], 0.f, weights[5][tmp - i], 4);
0.f, sum1 += compute_one_data_post(
weights[5][tmp - i], din_ptr_arr[num - i], weights[tmp - i], 0.f, weights[5][tmp - i], 4);
4);
sum1 += compute_one_data_post(din_ptr_arr[num - i],
weights[tmp - i],
0.f,
weights[5][tmp - i],
4);
din_ptr_arr[num - i]++; din_ptr_arr[num - i]++;
} }
din_ptr_arr[0]++; din_ptr_arr[0]++;
...@@ -1947,11 +1941,8 @@ inline void compute_all_padding_post(float* dout, ...@@ -1947,11 +1941,8 @@ inline void compute_all_padding_post(float* dout,
din_ptr_arr[3], weights[num], bias[0], weights[5][num], 4); din_ptr_arr[3], weights[num], bias[0], weights[5][num], 4);
din_ptr_arr[3]++; din_ptr_arr[3]++;
for (int i = 0; i < num; i++) { for (int i = 0; i < num; i++) {
sum += compute_one_data_post(din_ptr_arr[2 - i], sum += compute_one_data_post(
weights[tmp - i], din_ptr_arr[2 - i], weights[tmp - i], 0.f, weights[5][tmp - i], 4);
0.f,
weights[5][tmp - i],
4);
din_ptr_arr[2 - i]++; din_ptr_arr[2 - i]++;
} }
*dout++ = sum; *dout++ = sum;
...@@ -2366,12 +2357,9 @@ inline void compute_all_padding_pre_relu(float* dout, ...@@ -2366,12 +2357,9 @@ inline void compute_all_padding_pre_relu(float* dout,
din_ptr_arr[num], weights[4], bias[0], weights[6][0], 4); din_ptr_arr[num], weights[4], bias[0], weights[6][0], 4);
din_ptr_arr[num]++; din_ptr_arr[num]++;
for (int i = 0; i < num; i++) { for (int i = 0; i < num; i++) {
sum += compute_one_data_post(din_ptr_arr[tmp_index - i], sum += compute_one_data_post(
weights[3 - i], din_ptr_arr[tmp_index - i], weights[3 - i], 0.f, weights[5][3 - i], 4);
0.f, din_ptr_arr[tmp_index - i]++;
weights[5][3 - i],
4);
din_ptr_arr[tmp_index - i]++;
} }
*dout++ = sum > 0.f ? sum : 0.f; *dout++ = sum > 0.f ? sum : 0.f;
} }
...@@ -2484,11 +2472,8 @@ inline void compute_all_padding_mid_relu(float* dout, ...@@ -2484,11 +2472,8 @@ inline void compute_all_padding_mid_relu(float* dout,
din_ptr_arr[num], weights[num], bias[0], weights[6][0], 4); din_ptr_arr[num], weights[num], bias[0], weights[6][0], 4);
din_ptr_arr[num]++; din_ptr_arr[num]++;
for (int i = 0; i < num; i++) { for (int i = 0; i < num; i++) {
sum += compute_one_data_post(din_ptr_arr[tmp - i], sum += compute_one_data_post(
weights[tmp - i], din_ptr_arr[tmp - i], weights[tmp - i], 0.f, weights[5][tmp - i], 4);
0.f,
weights[5][tmp - i],
4);
din_ptr_arr[tmp - i]++; din_ptr_arr[tmp - i]++;
} }
*dout++ = sum > 0.f ? sum : 0.f; *dout++ = sum > 0.f ? sum : 0.f;
...@@ -2620,16 +2605,10 @@ inline void compute_all_padding_mid_relu_out2(float* dout0, ...@@ -2620,16 +2605,10 @@ inline void compute_all_padding_mid_relu_out2(float* dout0,
din_ptr_arr[tmp1], weights[num], bias[0], weights[6][0], 4); din_ptr_arr[tmp1], weights[num], bias[0], weights[6][0], 4);
din_ptr_arr[tmp1]++; din_ptr_arr[tmp1]++;
for (int i = 0; i < num; i++) { for (int i = 0; i < num; i++) {
sum += compute_one_data_post(din_ptr_arr[tmp - i], sum += compute_one_data_post(
weights[tmp - i], din_ptr_arr[tmp - i], weights[tmp - i], 0.f, weights[5][tmp - i], 4);
0.f, sum1 += compute_one_data_post(
weights[5][tmp - i], din_ptr_arr[num - i], weights[tmp - i], 0.f, weights[5][tmp - i], 4);
4);
sum1 += compute_one_data_post(din_ptr_arr[num - i],
weights[tmp - i],
0.f,
weights[5][tmp - i],
4);
din_ptr_arr[num - i]++; din_ptr_arr[num - i]++;
} }
din_ptr_arr[0]++; din_ptr_arr[0]++;
...@@ -2894,11 +2873,8 @@ inline void compute_all_padding_post_relu(float* dout, ...@@ -2894,11 +2873,8 @@ inline void compute_all_padding_post_relu(float* dout,
din_ptr_arr[3], weights[num], bias[0], weights[5][num], 4); din_ptr_arr[3], weights[num], bias[0], weights[5][num], 4);
din_ptr_arr[3]++; din_ptr_arr[3]++;
for (int i = 0; i < num; i++) { for (int i = 0; i < num; i++) {
sum += compute_one_data_post(din_ptr_arr[2 - i], sum += compute_one_data_post(
weights[tmp - i], din_ptr_arr[2 - i], weights[tmp - i], 0.f, weights[5][tmp - i], 4);
0.f,
weights[5][tmp - i],
4);
din_ptr_arr[2 - i]++; din_ptr_arr[2 - i]++;
} }
*dout++ = sum > 0.f ? sum : 0.f; *dout++ = sum > 0.f ? sum : 0.f;
...@@ -3331,11 +3307,8 @@ inline void compute_all_padding_pre_relu6(float* dout, ...@@ -3331,11 +3307,8 @@ inline void compute_all_padding_pre_relu6(float* dout,
din_ptr_arr[num], weights[4], bias[0], weights[6][0], 4); din_ptr_arr[num], weights[4], bias[0], weights[6][0], 4);
din_ptr_arr[num]++; din_ptr_arr[num]++;
for (int i = 0; i < num; i++) { for (int i = 0; i < num; i++) {
sum += compute_one_data_post(din_ptr_arr[tmp_index - i], sum += compute_one_data_post(
weights[3 - i], din_ptr_arr[tmp_index - i], weights[3 - i], 0.f, weights[5][3 - i], 4);
0.f,
weights[5][3 - i],
4);
din_ptr_arr[tmp_index - i]++; din_ptr_arr[tmp_index - i]++;
} }
*dout++ = sum > 0.f ? (sum < six[0] ? sum : six[0]) : 0.f; *dout++ = sum > 0.f ? (sum < six[0] ? sum : six[0]) : 0.f;
...@@ -3456,11 +3429,8 @@ inline void compute_all_padding_mid_relu6(float* dout, ...@@ -3456,11 +3429,8 @@ inline void compute_all_padding_mid_relu6(float* dout,
din_ptr_arr[num], weights[num], bias[0], weights[6][0], 4); din_ptr_arr[num], weights[num], bias[0], weights[6][0], 4);
din_ptr_arr[num]++; din_ptr_arr[num]++;
for (int i = 0; i < num; i++) { for (int i = 0; i < num; i++) {
sum += compute_one_data_post(din_ptr_arr[tmp - i], sum += compute_one_data_post(
weights[tmp - i], din_ptr_arr[tmp - i], weights[tmp - i], 0.f, weights[5][tmp - i], 4);
0.f,
weights[5][tmp - i],
4);
din_ptr_arr[tmp - i]++; din_ptr_arr[tmp - i]++;
} }
*dout++ = sum > 0.f ? (sum < six[0] ? sum : six[0]) : 0.f; *dout++ = sum > 0.f ? (sum < six[0] ? sum : six[0]) : 0.f;
...@@ -3599,16 +3569,10 @@ inline void compute_all_padding_mid_relu6_out2(float* dout0, ...@@ -3599,16 +3569,10 @@ inline void compute_all_padding_mid_relu6_out2(float* dout0,
din_ptr_arr[tmp1], weights[num], bias[0], weights[6][0], 4); din_ptr_arr[tmp1], weights[num], bias[0], weights[6][0], 4);
din_ptr_arr[tmp1]++; din_ptr_arr[tmp1]++;
for (int i = 0; i < num; i++) { for (int i = 0; i < num; i++) {
sum += compute_one_data_post(din_ptr_arr[tmp - i], sum += compute_one_data_post(
weights[tmp - i], din_ptr_arr[tmp - i], weights[tmp - i], 0.f, weights[5][tmp - i], 4);
0.f, sum1 += compute_one_data_post(
weights[5][tmp - i], din_ptr_arr[num - i], weights[tmp - i], 0.f, weights[5][tmp - i], 4);
4);
sum1 += compute_one_data_post(din_ptr_arr[num - i],
weights[tmp - i],
0.f,
weights[5][tmp - i],
4);
din_ptr_arr[num - i]++; din_ptr_arr[num - i]++;
} }
din_ptr_arr[0]++; din_ptr_arr[0]++;
...@@ -3885,11 +3849,8 @@ inline void compute_all_padding_post_relu6(float* dout, ...@@ -3885,11 +3849,8 @@ inline void compute_all_padding_post_relu6(float* dout,
din_ptr_arr[3], weights[num], bias[0], weights[5][num], 4); din_ptr_arr[3], weights[num], bias[0], weights[5][num], 4);
din_ptr_arr[3]++; din_ptr_arr[3]++;
for (int i = 0; i < num; i++) { for (int i = 0; i < num; i++) {
sum += compute_one_data_post(din_ptr_arr[2 - i], sum += compute_one_data_post(
weights[tmp - i], din_ptr_arr[2 - i], weights[tmp - i], 0.f, weights[5][tmp - i], 4);
0.f,
weights[5][tmp - i],
4);
din_ptr_arr[2 - i]++; din_ptr_arr[2 - i]++;
} }
*dout++ = sum > 0.f ? (sum < six[0] ? sum : six[0]) : 0.f; *dout++ = sum > 0.f ? (sum < six[0] ? sum : six[0]) : 0.f;
...@@ -4335,11 +4296,8 @@ inline void compute_all_padding_pre_leakyRelu(float* dout, ...@@ -4335,11 +4296,8 @@ inline void compute_all_padding_pre_leakyRelu(float* dout,
din_ptr_arr[num], weights[4], bias[0], weights[6][0], 4); din_ptr_arr[num], weights[4], bias[0], weights[6][0], 4);
din_ptr_arr[num]++; din_ptr_arr[num]++;
for (int i = 0; i < num; i++) { for (int i = 0; i < num; i++) {
sum += compute_one_data_post(din_ptr_arr[tmp_index - i], sum += compute_one_data_post(
weights[3 - i], din_ptr_arr[tmp_index - i], weights[3 - i], 0.f, weights[5][3 - i], 4);
0.f,
weights[5][3 - i],
4);
din_ptr_arr[tmp_index - i]++; din_ptr_arr[tmp_index - i]++;
} }
*dout++ = sum > 0.f ? sum : sum * scale[0]; *dout++ = sum > 0.f ? sum : sum * scale[0];
...@@ -4466,11 +4424,8 @@ inline void compute_all_padding_mid_leakyRelu(float* dout, ...@@ -4466,11 +4424,8 @@ inline void compute_all_padding_mid_leakyRelu(float* dout,
din_ptr_arr[num], weights[num], bias[0], weights[6][0], 4); din_ptr_arr[num], weights[num], bias[0], weights[6][0], 4);
din_ptr_arr[num]++; din_ptr_arr[num]++;
for (int i = 0; i < num; i++) { for (int i = 0; i < num; i++) {
sum += compute_one_data_post(din_ptr_arr[tmp - i], sum += compute_one_data_post(
weights[tmp - i], din_ptr_arr[tmp - i], weights[tmp - i], 0.f, weights[5][tmp - i], 4);
0.f,
weights[5][tmp - i],
4);
din_ptr_arr[tmp - i]++; din_ptr_arr[tmp - i]++;
} }
*dout++ = sum > 0.f ? sum : sum * scale[0]; *dout++ = sum > 0.f ? sum : sum * scale[0];
...@@ -4493,19 +4448,19 @@ inline void compute_all_padding_mid_leakyRelu(float* dout, ...@@ -4493,19 +4448,19 @@ inline void compute_all_padding_mid_leakyRelu(float* dout,
} }
} }
inline void compute_all_padding_mid_leakyRelu_out2(float* dout0, inline void compute_all_padding_mid_leakyRelu_out2(float* dout0,
float* dout1, float* dout1,
const float** din_ptr_arr, const float** din_ptr_arr,
const float* bias, const float* bias,
const float* scale, const float* scale,
float32x4_t* weights, float32x4_t* weights,
float32x4_t vzero, float32x4_t vzero,
int win, int win,
int wout, int wout,
int pad_left, int pad_left,
int pad_right, int pad_right,
int cnt, int cnt,
int remain, int remain,
int num) { int num) {
#ifdef __aarch64__ #ifdef __aarch64__
float32x4_t vscale = vld1q_f32(scale); float32x4_t vscale = vld1q_f32(scale);
#endif #endif
...@@ -4615,16 +4570,10 @@ inline void compute_all_padding_mid_leakyRelu_out2(float* dout0, ...@@ -4615,16 +4570,10 @@ inline void compute_all_padding_mid_leakyRelu_out2(float* dout0,
din_ptr_arr[tmp1], weights[num], bias[0], weights[6][0], 4); din_ptr_arr[tmp1], weights[num], bias[0], weights[6][0], 4);
din_ptr_arr[tmp1]++; din_ptr_arr[tmp1]++;
for (int i = 0; i < num; i++) { for (int i = 0; i < num; i++) {
sum += compute_one_data_post(din_ptr_arr[tmp - i], sum += compute_one_data_post(
weights[tmp - i], din_ptr_arr[tmp - i], weights[tmp - i], 0.f, weights[5][tmp - i], 4);
0.f, sum1 += compute_one_data_post(
weights[5][tmp - i], din_ptr_arr[num - i], weights[tmp - i], 0.f, weights[5][tmp - i], 4);
4);
sum1 += compute_one_data_post(din_ptr_arr[num - i],
weights[tmp - i],
0.f,
weights[5][tmp - i],
4);
din_ptr_arr[num - i]++; din_ptr_arr[num - i]++;
} }
din_ptr_arr[0]++; din_ptr_arr[0]++;
...@@ -4909,11 +4858,8 @@ inline void compute_all_padding_post_leakyRelu(float* dout, ...@@ -4909,11 +4858,8 @@ inline void compute_all_padding_post_leakyRelu(float* dout,
din_ptr_arr[3], weights[num], bias[0], weights[5][num], 4); din_ptr_arr[3], weights[num], bias[0], weights[5][num], 4);
din_ptr_arr[3]++; din_ptr_arr[3]++;
for (int i = 0; i < num; i++) { for (int i = 0; i < num; i++) {
sum += compute_one_data_post(din_ptr_arr[2 - i], sum += compute_one_data_post(
weights[tmp - i], din_ptr_arr[2 - i], weights[tmp - i], 0.f, weights[5][tmp - i], 4);
0.f,
weights[5][tmp - i],
4);
din_ptr_arr[2 - i]++; din_ptr_arr[2 - i]++;
} }
*dout++ = sum > 0.f ? sum : sum * scale[0]; *dout++ = sum > 0.f ? sum : sum * scale[0];
......
Markdown is supported
0% 或 .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑！
想要评论请 注册 或