Skip to content

P
Paddle-Lite

PaddlePaddle / Paddle-Lite

通知 338
Star 4
Fork 1
P
Paddle-Lite
提交 9348cb41 编写于 作者: C chenjiaoAngel
浏览文件
操作

fix format

上级 9a09cf28
隐藏空白更改
内联 并排
Showing with 60 addition and 200 deletion
lite/backends/arm/math/conv5x5s1_depthwise_fp32.cc
浏览文件 @ 9348cb41
......@@ -1211,13 +1211,11 @@ inline void compute_all_padding_pre(float* dout,
int wout,
int pad_left,
int pad_right,
int pad_left_new,
int pad_right_new,
int cnt,
int remain,
int num) {
int tmp_index = num - 1;
for (int i = pad_left_new; i > 0; i--) {
for (int i = pad_left; i > 0; i--) {
float sum = compute_one_data_pre(din_ptr_arr[num], weights[4], bias[0], weights[6][0], 4 - i);
for (int k = 0; k < num; k++) {
sum += compute_one_data_pre(din_ptr_arr[tmp_index - k], weights[3 - k], 0.f, weights[5][3 - k], 4 - i);
......@@ -1429,7 +1427,7 @@ inline void compute_all_padding_pre(float* dout,
*dout++ = sum;
}
// right
for (int i = 0; i < pad_right_new; i++) {
for (int i = 0; i < pad_right; i++) {
float sum = compute_one_data_post(din_ptr_arr[num], weights[4], bias[0], weights[4][3 - i], 3 - i);
din_ptr_arr[num]++;
for (int k = 0; k < num; k++) {
......@@ -1447,14 +1445,12 @@ inline void compute_all_padding_mid(float* dout,
int wout,
int pad_left,
int pad_right,
int pad_left_new,
int pad_right_new,
int cnt,
int remain,
int num) {
// left
int tmp = num - 1;
for (int i = pad_left_new; i > 0; i--) {
for (int i = pad_left; i > 0; i--) {
float sum = compute_one_data_pre(din_ptr_arr[num], weights[num], bias[0], weights[6][0], 4 - i);
for (int k = 0; k < num; k++) {
sum += compute_one_data_pre(din_ptr_arr[tmp - k], weights[tmp - k], 0.f, weights[5][tmp - k], 4 - i);
......@@ -1531,7 +1527,7 @@ inline void compute_all_padding_mid(float* dout,
*dout++ = sum;
}
// right
for (int i = 0; i < pad_right_new; i++) {
for (int i = 0; i < pad_right; i++) {
float sum = compute_one_data_post(din_ptr_arr[num], weights[num], bias[0], weights[num][3 - i], 3 - i);
din_ptr_arr[num]++;
for (int k = 0; k < num; k++) {
......@@ -1550,15 +1546,13 @@ inline void compute_all_padding_mid_out2(float* dout0,
int wout,
int pad_left,
int pad_right,
int pad_left_new,
int pad_right_new,
int cnt,
int remain,
int num) {
int tmp1 = num + 1;
int tmp = num - 1;
// left
for (int i = pad_left_new; i > 0; i--) {
for (int i = pad_left; i > 0; i--) {
float sum = compute_one_data_pre(din_ptr_arr[num], weights[num], bias[0], weights[6][0], 4 - i);
float sum1 = compute_one_data_pre(din_ptr_arr[tmp1], weights[num], bias[0], weights[6][0], 4 - i);
for (int k = 0; k < num; k++) {
......@@ -1647,7 +1641,7 @@ inline void compute_all_padding_mid_out2(float* dout0,
*dout1++ = sum1;
}
// right
for (int i = 0; i < pad_right_new; i++) {
for (int i = 0; i < pad_right; i++) {
float sum = compute_one_data_post(din_ptr_arr[num], weights[num], bias[0], weights[num][3 - i], 3 - i);
float sum1 = compute_one_data_post(din_ptr_arr[tmp1], weights[num], bias[0], weights[num][3 - i], 3 - i);
din_ptr_arr[tmp1]++;
......@@ -1660,11 +1654,6 @@ inline void compute_all_padding_mid_out2(float* dout0,
*dout0++ = sum;
*dout1++ = sum1;
}
for (int w = pad_right; w > 4; w--) {
*dout0++ = bias[0];
*dout1++ = bias[0];
}
}
inline void compute_all_padding_post(float* dout,
const float** din_ptr_arr,
......@@ -1674,17 +1663,12 @@ inline void compute_all_padding_post(float* dout,
int wout,
int pad_left,
int pad_right,
int pad_left_new,
int pad_right_new,
int cnt,
int remain,
int num) {
// left
/* for (int w = pad_left; w > 4; w--) {
*dout++ = bias[0];
}*/
int tmp = num - 1;
for (int i = pad_left_new; i > 0; i--) {
for (int i = pad_left; i > 0; i--) {
float sum = compute_one_data_pre(din_ptr_arr[3], weights[num], bias[0], weights[5][num], 4 - i);
for (int k = 0; k < num; k++) {
sum += compute_one_data_pre(din_ptr_arr[2 - k], weights[tmp - k], 0.f, weights[5][tmp - k], 4 - i);
......@@ -1894,7 +1878,7 @@ inline void compute_all_padding_post(float* dout,
}
// right
for (int i = 0; i < pad_right_new; i++) {
for (int i = 0; i < pad_right; i++) {
float sum = compute_one_data_post(din_ptr_arr[3], weights[num], bias[0], weights[num][3 - i], 3 - i);
din_ptr_arr[3]++;
for (int k = 0; k < num; k++) {
......@@ -1927,10 +1911,6 @@ void conv_depthwise_5x5s1_bias(float* dout,
int out_size = wout * hout;
int cnt = loop_w >> 2;
int remain = loop_w & 3;
int pad_left_new = pad_left > 4 ? 4 : pad_left;
int pad_right_new = pad_right > 4 ? 4 : pad_right;
int pad_top_new = pad_top > 4 ? 4 : pad_top;
int pad_bottom_new = pad_bottom > 4 ? 4 : pad_bottom;
int in_channel_size = chin * in_size;
int out_channel_size = chin * out_size;
int weights_size = 25;
......@@ -1968,9 +1948,9 @@ void conv_depthwise_5x5s1_bias(float* dout,
const float* din_ptr_arr[] = {din_ptr0, din_ptr1, din_ptr2, din_ptr3, din_ptr4, din_ptr5};
float32x4_t weights_vec[] = {wr0, wr1, wr2, wr3, wr4, wr5, wr6};
// top_h
for (int h = pad_top_new; h > 0; h--) {
for (int h = pad_top; h > 0; h--) {
compute_all_padding_pre(dout_ptr0, din_ptr_arr, vbias, weights_vec, win, wout, pad_left,
pad_right, pad_left_new, pad_right_new, cnt, remain, 4 - h);
pad_right, cnt, remain, 4 - h);
dout_ptr0 += wout;
din_ptr_arr[0] = din_ptr0;
din_ptr_arr[1] = din_ptr1;
......@@ -1982,7 +1962,7 @@ void conv_depthwise_5x5s1_bias(float* dout,
// mid_h
for (int h = 0; h < loop_h - 1; h += 2) {
compute_all_padding_mid_out2(dout_ptr0, dout_ptr1, din_ptr_arr, vbias, weights_vec, win, wout, pad_left,
pad_right, pad_left_new, pad_right_new, cnt, remain, 4);
pad_right, cnt, remain, 4);
dout_ptr0 += num_out;
dout_ptr1 += num_out;
din_ptr0 = din_ptr2;
......@@ -2000,7 +1980,7 @@ void conv_depthwise_5x5s1_bias(float* dout,
}
if (loop_h % 2 != 0) {
compute_all_padding_mid(dout_ptr0, din_ptr_arr, vbias, weights_vec, win, wout, pad_left,
pad_right, pad_left_new, pad_right_new, cnt, remain, 4);
pad_right, cnt, remain, 4);
dout_ptr0 = dout_ptr1;
din_ptr0 = din_ptr1;
din_ptr1 = din_ptr2;
......@@ -2014,9 +1994,9 @@ void conv_depthwise_5x5s1_bias(float* dout,
din_ptr_arr[4] = din_ptr4;
}
// bottom
for (int h = 0; h < pad_bottom_new; h++) {
for (int h = 0; h < pad_bottom; h++) {
compute_all_padding_post(dout_ptr0, din_ptr_arr, vbias, weights_vec, win, wout, pad_left,
pad_right, pad_left_new, pad_right_new, cnt, remain, 3 - h);
pad_right, cnt, remain, 3 - h);
dout_ptr0 += wout;
din_ptr_arr[0] = din_ptr0;
din_ptr_arr[1] = din_ptr1;
......@@ -2037,17 +2017,11 @@ inline void compute_all_padding_pre_relu(float* dout,
int wout,
int pad_left,
int pad_right,
int pad_left_new,
int pad_right_new,
int cnt,
int remain,
int num) {
int tmp_index = num - 1;
// left
/* for (int w = pad_left; w > 4; w--) {
*dout++ = bias[0] > 0.f ? bias[0] : 0.f;
}*/
for (int i = pad_left_new; i > 0; i--) {
for (int i = pad_left; i > 0; i--) {
float sum = compute_one_data_pre(din_ptr_arr[num], weights[4], bias[0], weights[6][0], 4 - i);
for (int k = 0; k < num; k++) {
sum += compute_one_data_pre(din_ptr_arr[tmp_index - k], weights[3 - k], 0.f, weights[5][3 - k], 4 - i);
......@@ -2267,7 +2241,7 @@ inline void compute_all_padding_pre_relu(float* dout,
*dout++ = sum > 0.f ? sum : 0.f;
}
// right
for (int i = 0; i < pad_right_new; i++) {
for (int i = 0; i < pad_right; i++) {
float sum = compute_one_data_post(din_ptr_arr[num], weights[4], bias[0], weights[4][3 - i], 3 - i);
din_ptr_arr[num]++;
for (int k = 0; k < num; k++) {
......@@ -2276,10 +2250,6 @@ inline void compute_all_padding_pre_relu(float* dout,
}
*dout++ = sum > 0.f ? sum : 0.f;
}
/*
for (int w = pad_right; w > 4; w--) {
*dout++ = bias[0] > 0.f ? bias[0] : 0.f;
}*/
}
inline void compute_all_padding_mid_relu(float* dout,
const float** din_ptr_arr,
......@@ -2290,17 +2260,11 @@ inline void compute_all_padding_mid_relu(float* dout,
int wout,
int pad_left,
int pad_right,
int pad_left_new,
int pad_right_new,
int cnt,
int remain,
int num) {
// left
/* for (int w = pad_left; w > 4; w--) {
*dout++ = bias[0] > 0.f ? bias[0] : 0.f;
}*/
int tmp = num - 1;
for (int i = pad_left_new; i > 0; i--) {
for (int i = pad_left; i > 0; i--) {
float sum = compute_one_data_pre(din_ptr_arr[num], weights[num], bias[0], weights[6][0], 4 - i);
for (int k = 0; k < num; k++) {
sum += compute_one_data_pre(din_ptr_arr[tmp - k], weights[tmp - k], 0.f, weights[5][tmp - k], 4 - i);
......@@ -2378,7 +2342,7 @@ inline void compute_all_padding_mid_relu(float* dout,
*dout++ = sum > 0.f ? sum : 0.f;
}
// right
for (int i = 0; i < pad_right_new; i++) {
for (int i = 0; i < pad_right; i++) {
float sum = compute_one_data_post(din_ptr_arr[num], weights[num], bias[0], weights[num][3 - i], 3 - i);
din_ptr_arr[num]++;
for (int k = 0; k < num; k++) {
......@@ -2387,10 +2351,6 @@ inline void compute_all_padding_mid_relu(float* dout,
}
*dout++ = sum > 0.f ? sum : 0.f;
}
/*
for (int w = pad_right; w > 4; w--) {
*dout++ = bias[0] > 0.f ? bias[0] : 0.f;
}*/
}
inline void compute_all_padding_mid_relu_out2(float* dout0,
float* dout1,
......@@ -2402,19 +2362,13 @@ inline void compute_all_padding_mid_relu_out2(float* dout0,
int wout,
int pad_left,
int pad_right,
int pad_left_new,
int pad_right_new,
int cnt,
int remain,
int num) {
// left
for (int w = pad_left; w > 4; w--) {
*dout0++ = bias[0] > 0.f ? bias[0] : 0.f;
*dout1++ = bias[0] > 0.f ? bias[0] : 0.f;
}
int tmp = num - 1;
int tmp1 = num + 1;
for (int i = pad_left_new; i > 0; i--) {
for (int i = pad_left; i > 0; i--) {
float sum = compute_one_data_pre(din_ptr_arr[num], weights[num], bias[0], weights[6][0], 4 - i);
float sum1 = compute_one_data_pre(din_ptr_arr[tmp1], weights[num], bias[0], weights[6][0], 4 - i);
for (int k = 0; k < num; k++) {
......@@ -2505,7 +2459,7 @@ inline void compute_all_padding_mid_relu_out2(float* dout0,
*dout1++ = sum1 > 0.f ? sum1 : 0.f;
}
// right
for (int i = 0; i < pad_right_new; i++) {
for (int i = 0; i < pad_right; i++) {
float sum = compute_one_data_post(din_ptr_arr[num], weights[num], bias[0], weights[num][3 - i], 3 - i);
float sum1 = compute_one_data_post(din_ptr_arr[tmp1], weights[num], bias[0], weights[num][3 - i], 3 - i);
din_ptr_arr[tmp1]++;
......@@ -2518,10 +2472,6 @@ inline void compute_all_padding_mid_relu_out2(float* dout0,
*dout0++ = sum > 0.f ? sum : 0.f;
*dout1++ = sum1 > 0.f ? sum1 : 0.f;
}
for (int w = pad_right; w > 4; w--) {
*dout0++ = bias[0] > 0.f ? bias[0] : 0.f;
*dout1++ = bias[0] > 0.f ? bias[0] : 0.f;
}
}
inline void compute_all_padding_post_relu(float* dout,
const float** din_ptr_arr,
......@@ -2532,17 +2482,12 @@ inline void compute_all_padding_post_relu(float* dout,
int wout,
int pad_left,
int pad_right,
int pad_left_new,
int pad_right_new,
int cnt,
int remain,
int num) {
// left
/*for (int w = pad_left; w > 4; w--) {
*dout++ = bias[0] > 0.f ? bias[0] : 0.f;
}*/
int tmp = num - 1;
for (int i = pad_left_new; i > 0; i--) {
for (int i = pad_left; i > 0; i--) {
float sum = compute_one_data_pre(din_ptr_arr[3], weights[num], bias[0], weights[5][num], 4 - i);
for (int k = 0; k < num; k++) {
sum += compute_one_data_pre(din_ptr_arr[2 - k], weights[tmp - k], 0.f, weights[5][tmp - k], 4 - i);
......@@ -2760,7 +2705,7 @@ inline void compute_all_padding_post_relu(float* dout,
}
// right
for (int i = 0; i < pad_right_new; i++) {
for (int i = 0; i < pad_right; i++) {
float sum = compute_one_data_post(din_ptr_arr[3], weights[num], bias[0], weights[num][3 - i], 3 - i);
din_ptr_arr[3]++;
for (int k = 0; k < num; k++) {
......@@ -2769,10 +2714,6 @@ inline void compute_all_padding_post_relu(float* dout,
}
*dout++ = sum > 0.f ? sum : 0.f;
}
/*
for (int w = pad_right; w > 4; w--) {
*dout++ = bias[0] > 0.f ? bias[0] : 0.f;
}*/
}
void conv_depthwise_5x5s1_bias_relu(float* dout,
......@@ -2797,10 +2738,6 @@ void conv_depthwise_5x5s1_bias_relu(float* dout,
int out_size = wout * hout;
int cnt = loop_w >> 2;
int remain = loop_w & 3;
int pad_left_new = pad_left > 4 ? 4 : pad_left;
int pad_right_new = pad_right > 4 ? 4 : pad_right;
int pad_top_new = pad_top > 4 ? 4 : pad_top;
int pad_bottom_new = pad_bottom > 4 ? 4 : pad_bottom;
int in_channel_size = chin * in_size;
int out_channel_size = chin * out_size;
int weights_size = 25;
......@@ -2839,9 +2776,9 @@ void conv_depthwise_5x5s1_bias_relu(float* dout,
const float* din_ptr_arr[] = {din_ptr0, din_ptr1, din_ptr2, din_ptr3, din_ptr4, din_ptr5};
float32x4_t weights_vec[] = {wr0, wr1, wr2, wr3, wr4, wr5, wr6};
// top_h
for (int h = pad_top_new; h > 0; h--) {
for (int h = pad_top; h > 0; h--) {
compute_all_padding_pre_relu(dout_ptr0, din_ptr_arr, vbias, weights_vec, vzero, win, wout, pad_left,
pad_right, pad_left_new, pad_right_new, cnt, remain, 4 - h);
pad_right, cnt, remain, 4 - h);
dout_ptr0 += wout;
din_ptr_arr[0] = din_ptr0;
din_ptr_arr[1] = din_ptr1;
......@@ -2853,7 +2790,7 @@ void conv_depthwise_5x5s1_bias_relu(float* dout,
// mid_h
for (int h = 0; h < loop_h - 1; h += 2) {
compute_all_padding_mid_relu_out2(dout_ptr0, dout_ptr1, din_ptr_arr, vbias, weights_vec, vzero, win, wout, pad_left,
pad_right, pad_left_new, pad_right_new, cnt, remain, 4);
pad_right, cnt, remain, 4);
dout_ptr0 += num_out;
dout_ptr1 += num_out;
din_ptr0 = din_ptr2;
......@@ -2871,7 +2808,7 @@ void conv_depthwise_5x5s1_bias_relu(float* dout,
}
if (loop_h % 2 != 0) {
compute_all_padding_mid_relu(dout_ptr0, din_ptr_arr, vbias, weights_vec, vzero, win, wout, pad_left,
pad_right, pad_left_new, pad_right_new, cnt, remain, 4);
pad_right, cnt, remain, 4);
dout_ptr0 = dout_ptr1;
din_ptr0 = din_ptr1;
din_ptr1 = din_ptr2;
......@@ -2885,9 +2822,9 @@ void conv_depthwise_5x5s1_bias_relu(float* dout,
din_ptr_arr[4] = din_ptr4;
}
// bottom
for (int h = 0; h < pad_bottom_new; h++) {
for (int h = 0; h < pad_bottom; h++) {
compute_all_padding_post_relu(dout_ptr0, din_ptr_arr, vbias, weights_vec, vzero, win, wout, pad_left,
pad_right, pad_left_new, pad_right_new, cnt, remain, 3 - h);
pad_right, cnt, remain, 3 - h);
dout_ptr0 += wout;
din_ptr_arr[0] = din_ptr0;
din_ptr_arr[1] = din_ptr1;
......@@ -2909,8 +2846,6 @@ inline void compute_all_padding_pre_relu6(float* dout,
int wout,
int pad_left,
int pad_right,
int pad_left_new,
int pad_right_new,
int cnt,
int remain,
int num) {
......@@ -2919,10 +2854,7 @@ inline void compute_all_padding_pre_relu6(float* dout,
#endif
int tmp_index = num - 1;
// left
for (int w = pad_left; w > 4; w--) {
*dout++ = bias[0] > 0.f ? (bias[0] < six[0] ? bias[0] : six[0]) : 0.f;
}
for (int i = pad_left_new; i > 0; i--) {
for (int i = pad_left; i > 0; i--) {
float sum = compute_one_data_pre(din_ptr_arr[num], weights[4], bias[0], weights[6][0], 4 - i);
for (int k = 0; k < num; k++) {
sum += compute_one_data_pre(din_ptr_arr[tmp_index - k], weights[3 - k], 0.f, weights[5][3 - k], 4 - i);
......@@ -3150,7 +3082,7 @@ inline void compute_all_padding_pre_relu6(float* dout,
*dout++ = sum > 0.f ? (sum < six[0] ? sum : six[0]) : 0.f;
}
// right
for (int i = 0; i < pad_right_new; i++) {
for (int i = 0; i < pad_right; i++) {
float sum = compute_one_data_post(din_ptr_arr[num], weights[4], bias[0], weights[4][3 - i], 3 - i);
din_ptr_arr[num]++;
for (int k = 0; k < num; k++) {
......@@ -3159,10 +3091,6 @@ inline void compute_all_padding_pre_relu6(float* dout,
}
*dout++ = sum > 0.f ? (sum < six[0] ? sum : six[0]) : 0.f;
}
for (int w = pad_right; w > 4; w--) {
*dout++ = bias[0] > 0.f ? (bias[0] < six[0] ? bias[0] : six[0]) : 0.f;
}
}
inline void compute_all_padding_mid_relu6(float* dout,
const float** din_ptr_arr,
......@@ -3174,8 +3102,6 @@ inline void compute_all_padding_mid_relu6(float* dout,
int wout,
int pad_left,
int pad_right,
int pad_left_new,
int pad_right_new,
int cnt,
int remain,
int num) {
......@@ -3183,11 +3109,8 @@ inline void compute_all_padding_mid_relu6(float* dout,
float32x4_t vsix = vld1q_f32(six);
#endif
// left
for (int w = pad_left; w > 4; w--) {
*dout++ = bias[0] > 0.f ? (bias[0] < six[0] ? bias[0] : six[0]) : 0.f;
}
int tmp = num - 1;
for (int i = pad_left_new; i > 0; i--) {
for (int i = pad_left; i > 0; i--) {
float sum = compute_one_data_pre(din_ptr_arr[num], weights[num], bias[0], weights[6][0], 4 - i);
for (int k = 0; k < num; k++) {
sum += compute_one_data_pre(din_ptr_arr[tmp - k], weights[tmp - k], 0.f, weights[5][tmp - k], 4 - i);
......@@ -3267,7 +3190,7 @@ inline void compute_all_padding_mid_relu6(float* dout,
*dout++ = sum > 0.f ? (sum < six[0] ? sum : six[0]) : 0.f;
}
// right
for (int i = 0; i < pad_right_new; i++) {
for (int i = 0; i < pad_right; i++) {
float sum = compute_one_data_post(din_ptr_arr[num], weights[num], bias[0], weights[num][3 - i], 3 - i);
din_ptr_arr[num]++;
for (int k = 0; k < num; k++) {
......@@ -3276,9 +3199,6 @@ inline void compute_all_padding_mid_relu6(float* dout,
}
*dout++ = sum > 0.f ? (sum < six[0] ? sum : six[0]) : 0.f;
}
for (int w = pad_right; w > 4; w--) {
*dout++ = bias[0] > 0.f ? (bias[0] < six[0] ? bias[0] : six[0]) : 0.f;
}
}
inline void compute_all_padding_mid_relu6_out2(float* dout0,
......@@ -3292,8 +3212,6 @@ inline void compute_all_padding_mid_relu6_out2(float* dout0,
int wout,
int pad_left,
int pad_right,
int pad_left_new,
int pad_right_new,
int cnt,
int remain,
int num) {
......@@ -3301,13 +3219,9 @@ inline void compute_all_padding_mid_relu6_out2(float* dout0,
float32x4_t vsix = vld1q_f32(six);
#endif
// left
for (int w = pad_left; w > 4; w--) {
*dout0++ = bias[0] > 0.f ? (bias[0] < six[0] ? bias[0] : six[0]) : 0.f;
*dout1++ = bias[0] > 0.f ? (bias[0] < six[0] ? bias[0] : six[0]) : 0.f;
}
int tmp = num - 1;
int tmp1 = num + 1;
for (int i = pad_left_new; i > 0; i--) {
for (int i = pad_left; i > 0; i--) {
float sum = compute_one_data_pre(din_ptr_arr[num], weights[num], bias[0], weights[6][0], 4 - i);
float sum1 = compute_one_data_pre(din_ptr_arr[tmp1], weights[num], bias[0], weights[6][0], 4 - i);
for (int k = 0; k < num; k++) {
......@@ -3399,7 +3313,7 @@ inline void compute_all_padding_mid_relu6_out2(float* dout0,
*dout1++ = sum1 > 0.f ? (sum1 < six[0] ? sum1 : six[0]) : 0.f;
}
// right
for (int i = 0; i < pad_right_new; i++) {
for (int i = 0; i < pad_right; i++) {
float sum = compute_one_data_post(din_ptr_arr[num], weights[num], bias[0], weights[num][3 - i], 3 - i);
float sum1 = compute_one_data_post(din_ptr_arr[tmp1], weights[num], bias[0], weights[num][3 - i], 3 - i);
din_ptr_arr[tmp1]++;
......@@ -3412,10 +3326,6 @@ inline void compute_all_padding_mid_relu6_out2(float* dout0,
*dout0++ = sum > 0.f ? (sum < six[0] ? sum : six[0]) : 0.f;
*dout1++ = sum1 > 0.f ? (sum1 < six[0] ? sum1 : six[0]) : 0.f;
}
for (int w = pad_right; w > 4; w--) {
*dout0++ = bias[0] > 0.f ? (bias[0] < six[0] ? bias[0] : six[0]) : 0.f;
*dout1++ = bias[0] > 0.f ? (bias[0] < six[0] ? bias[0] : six[0]) : 0.f;
}
}
inline void compute_all_padding_post_relu6(float* dout,
const float** din_ptr_arr,
......@@ -3427,8 +3337,6 @@ inline void compute_all_padding_post_relu6(float* dout,
int wout,
int pad_left,
int pad_right,
int pad_left_new,
int pad_right_new,
int cnt,
int remain,
int num) {
......@@ -3436,11 +3344,8 @@ inline void compute_all_padding_post_relu6(float* dout,
float32x4_t vsix = vld1q_f32(six);
#endif
// left
for (int w = pad_left; w > 4; w--) {
*dout++ = bias[0] > 0.f ? (bias[0] < six[0] ? bias[0] : six[0]) : 0.f;
}
int tmp = num - 1;
for (int i = pad_left_new; i > 0; i--) {
for (int i = pad_left; i > 0; i--) {
float sum = compute_one_data_pre(din_ptr_arr[3], weights[num], bias[0], weights[5][num], 4 - i);
for (int k = 0; k < num; k++) {
sum += compute_one_data_pre(din_ptr_arr[2 - k], weights[tmp - k], 0.f, weights[5][tmp - k], 4 - i);
......@@ -3666,7 +3571,7 @@ inline void compute_all_padding_post_relu6(float* dout,
}
// right
for (int i = 0; i < pad_right_new; i++) {
for (int i = 0; i < pad_right; i++) {
float sum = compute_one_data_post(din_ptr_arr[3], weights[num], bias[0], weights[num][3 - i], 3 - i);
din_ptr_arr[3]++;
for (int k = 0; k < num; k++) {
......@@ -3675,9 +3580,6 @@ inline void compute_all_padding_post_relu6(float* dout,
}
*dout++ = sum > 0.f ? (sum < six[0] ? sum : six[0]) : 0.f;
}
for (int w = pad_right; w > 4; w--) {
*dout++ = bias[0] > 0.f ? (bias[0] < six[0] ? bias[0] : six[0]) : 0.f;
}
}
void conv_depthwise_5x5s1_bias_relu6(float* dout,
......@@ -3703,10 +3605,6 @@ void conv_depthwise_5x5s1_bias_relu6(float* dout,
int out_size = wout * hout;
int cnt = loop_w >> 2;
int remain = loop_w & 3;
int pad_left_new = pad_left > 4 ? 4 : pad_left;
int pad_right_new = pad_right > 4 ? 4 : pad_right;
int pad_top_new = pad_top > 4 ? 4 : pad_top;
int pad_bottom_new = pad_bottom > 4 ? 4 : pad_bottom;
int in_channel_size = chin * in_size;
int out_channel_size = chin * out_size;
int weights_size = 25;
......@@ -3745,10 +3643,9 @@ void conv_depthwise_5x5s1_bias_relu6(float* dout,
const float* din_ptr_arr[] = {din_ptr0, din_ptr1, din_ptr2, din_ptr3, din_ptr4, din_ptr5};
float32x4_t weights_vec[] = {wr0, wr1, wr2, wr3, wr4, wr5, wr6};
// top_h
for (int h = pad_top_new; h > 0; h--) {
for (int h = pad_top; h > 0; h--) {
compute_all_padding_pre_relu6(dout_ptr0, din_ptr_arr, vbias, six, weights_vec, vzero,
win, wout, pad_left, pad_right, pad_left_new,
pad_right_new, cnt, remain, 4 - h);
win, wout, pad_left, pad_right, cnt, remain, 4 - h);
dout_ptr0 += wout;
din_ptr_arr[0] = din_ptr0;
din_ptr_arr[1] = din_ptr1;
......@@ -3760,8 +3657,7 @@ void conv_depthwise_5x5s1_bias_relu6(float* dout,
// mid_h
for (int h = 0; h < loop_h - 1; h += 2) {
compute_all_padding_mid_relu6_out2(dout_ptr0, dout_ptr1, din_ptr_arr, vbias, six, weights_vec, vzero,
win, wout, pad_left, pad_right, pad_left_new,
pad_right_new, cnt, remain, 4);
win, wout, pad_left, pad_right, cnt, remain, 4);
dout_ptr0 += num_out;
dout_ptr1 += num_out;
din_ptr0 = din_ptr2;
......@@ -3779,8 +3675,7 @@ void conv_depthwise_5x5s1_bias_relu6(float* dout,
}
if (loop_h % 2 != 0) {
compute_all_padding_mid_relu6(dout_ptr0, din_ptr_arr, vbias, six, weights_vec, vzero,
win, wout, pad_left, pad_right, pad_left_new,
pad_right_new, cnt, remain, 4);
win, wout, pad_left, pad_right, cnt, remain, 4);
dout_ptr0 = dout_ptr1;
din_ptr0 = din_ptr1;
din_ptr1 = din_ptr2;
......@@ -3794,10 +3689,9 @@ void conv_depthwise_5x5s1_bias_relu6(float* dout,
din_ptr_arr[4] = din_ptr4;
}
// bottom
for (int h = 0; h < pad_bottom_new; h++) {
for (int h = 0; h < pad_bottom; h++) {
compute_all_padding_post_relu6(dout_ptr0, din_ptr_arr, vbias, six, weights_vec, vzero,
win, wout, pad_left, pad_right, pad_left_new,
pad_right_new, cnt, remain, 3 - h);
win, wout, pad_left, pad_right, cnt, remain, 3 - h);
dout_ptr0 += wout;
din_ptr_arr[0] = din_ptr0;
din_ptr_arr[1] = din_ptr1;
......@@ -3819,8 +3713,6 @@ inline void compute_all_padding_pre_leakyRelu(float* dout,
int wout,
int pad_left,
int pad_right,
int pad_left_new,
int pad_right_new,
int cnt,
int remain,
int num) {
......@@ -3829,10 +3721,7 @@ inline void compute_all_padding_pre_leakyRelu(float* dout,
#endif
int tmp_index = num - 1;
// left
for (int w = pad_left; w > 4; w--) {
*dout++ = bias[0] > 0.f ? bias[0] : bias[0] * scale[0];
}
for (int i = pad_left_new; i > 0; i--) {
for (int i = pad_left; i > 0; i--) {
float sum = compute_one_data_pre(din_ptr_arr[num], weights[4], bias[0], weights[6][0], 4 - i);
for (int k = 0; k < num; k++) {
sum += compute_one_data_pre(din_ptr_arr[tmp_index - k], weights[3 - k], 0.f, weights[5][3 - k], 4 - i);
......@@ -4068,7 +3957,7 @@ inline void compute_all_padding_pre_leakyRelu(float* dout,
*dout++ = sum > 0.f ? sum : sum * scale[0];
}
// right
for (int i = 0; i < pad_right_new; i++) {
for (int i = 0; i < pad_right; i++) {
float sum = compute_one_data_post(din_ptr_arr[num], weights[4], bias[0], weights[4][3 - i], 3 - i);
din_ptr_arr[num]++;
for (int k = 0; k < num; k++) {
......@@ -4092,8 +3981,6 @@ inline void compute_all_padding_mid_leakyRelu(float* dout,
int wout,
int pad_left,
int pad_right,
int pad_left_new,
int pad_right_new,
int cnt,
int remain,
int num) {
......@@ -4105,7 +3992,7 @@ inline void compute_all_padding_mid_leakyRelu(float* dout,
*dout++ = bias[0] > 0.f ? bias[0] : bias[0] * scale[0];
}
int tmp = num - 1;
for (int i = pad_left_new; i > 0; i--) {
for (int i = pad_left; i > 0; i--) {
float sum = compute_one_data_pre(din_ptr_arr[num], weights[num], bias[0], weights[6][0], 4 - i);
for (int k = 0; k < num; k++) {
sum += compute_one_data_pre(din_ptr_arr[tmp - k], weights[tmp - k], 0.f, weights[5][tmp - k], 4 - i);
......@@ -4188,7 +4075,7 @@ inline void compute_all_padding_mid_leakyRelu(float* dout,
}
// right
for (int i = 0; i < pad_right_new; i++) {
for (int i = 0; i < pad_right; i++) {
float sum = compute_one_data_post(din_ptr_arr[num], weights[num], bias[0], weights[num][3 - i], 3 - i);
din_ptr_arr[num]++;
for (int k = 0; k < num; k++) {
......@@ -4197,9 +4084,6 @@ inline void compute_all_padding_mid_leakyRelu(float* dout,
}
*dout++ = sum > 0.f ? sum : sum * scale[0];
}
for (int w = pad_right; w > 4; w--) {
*dout++ = bias[0] > 0.f ? bias[0] : bias[0] * scale[0];
}
}
inline void compute_all_padding_mid_leakyRelu_out2(float* dout0,
float* dout1,
......@@ -4212,8 +4096,6 @@ inline void compute_all_padding_mid_leakyRelu_out2(float* dout0,
int wout,
int pad_left,
int pad_right,
int pad_left_new,
int pad_right_new,
int cnt,
int remain,
int num) {
......@@ -4221,13 +4103,9 @@ inline void compute_all_padding_mid_leakyRelu_out2(float* dout0,
float32x4_t vscale = vld1q_f32(scale);
#endif
// left
for (int w = pad_left; w > 4; w--) {
*dout0++ = bias[0] > 0.f ? bias[0] : bias[0] * scale[0];
*dout1++ = bias[0] > 0.f ? bias[0] : bias[0] * scale[0];
}
int tmp = num - 1;
int tmp1 = num + 1;
for (int i = pad_left_new; i > 0; i--) {
for (int i = pad_left; i > 0; i--) {
float sum = compute_one_data_pre(din_ptr_arr[num], weights[num], bias[0], weights[6][0], 4 - i);
float sum1 = compute_one_data_pre(din_ptr_arr[tmp1], weights[num], bias[0], weights[6][0], 4 - i);
for (int k = 0; k < num; k++) {
......@@ -4322,7 +4200,7 @@ inline void compute_all_padding_mid_leakyRelu_out2(float* dout0,
*dout1++ = sum1 > 0.f ? sum1 : sum1 * scale[0];
}
// right
for (int i = 0; i < pad_right_new; i++) {
for (int i = 0; i < pad_right; i++) {
float sum = compute_one_data_post(din_ptr_arr[num], weights[num], bias[0], weights[num][3 - i], 3 - i);
float sum1 = compute_one_data_post(din_ptr_arr[tmp1], weights[num], bias[0], weights[num][3 - i], 3 - i);
din_ptr_arr[tmp1]++;
......@@ -4335,10 +4213,6 @@ inline void compute_all_padding_mid_leakyRelu_out2(float* dout0,
*dout0++ = sum > 0.f ? sum : sum * scale[0];
*dout1++ = sum1 > 0.f ? sum1 : sum1 * scale[0];
}
for (int w = pad_right; w > 4; w--) {
*dout0++ = bias[0] > 0.f ? bias[0] : bias[0] * scale[0];
*dout1++ = bias[0] > 0.f ? bias[0] : bias[0] * scale[0];
}
}
inline void compute_all_padding_post_leakyRelu(float* dout,
const float** din_ptr_arr,
......@@ -4350,8 +4224,6 @@ inline void compute_all_padding_post_leakyRelu(float* dout,
int wout,
int pad_left,
int pad_right,
int pad_left_new,
int pad_right_new,
int cnt,
int remain,
int num) {
......@@ -4359,11 +4231,8 @@ inline void compute_all_padding_post_leakyRelu(float* dout,
float32x4_t vscale = vld1q_f32(scale);
#endif
// left
for (int w = pad_left; w > 4; w--) {
*dout++ = bias[0] > 0.f ? bias[0] : bias[0] * scale[0];
}
int tmp = num - 1;
for (int i = pad_left_new; i > 0; i--) {
for (int i = pad_left; i > 0; i--) {
float sum = compute_one_data_pre(din_ptr_arr[3], weights[num], bias[0], weights[5][num], 4 - i);
for (int k = 0; k < num; k++) {
sum += compute_one_data_pre(din_ptr_arr[2 - k], weights[tmp - k], 0.f, weights[5][tmp - k], 4 - i);
......@@ -4597,7 +4466,7 @@ inline void compute_all_padding_post_leakyRelu(float* dout,
}
// right
for (int i = 0; i < pad_right_new; i++) {
for (int i = 0; i < pad_right; i++) {
float sum = compute_one_data_post(din_ptr_arr[3], weights[num], bias[0], weights[num][3 - i], 3 - i);
din_ptr_arr[3]++;
for (int k = 0; k < num; k++) {
......@@ -4606,9 +4475,6 @@ inline void compute_all_padding_post_leakyRelu(float* dout,
}
*dout++ = sum > 0.f ? sum : sum * scale[0];
}
for (int w = pad_right; w > 4; w--) {
*dout++ = bias[0] > 0.f ? bias[0] : bias[0] * scale[0];
}
}
void conv_depthwise_5x5s1_bias_leakyRelu(float* dout,
......@@ -4634,10 +4500,6 @@ void conv_depthwise_5x5s1_bias_leakyRelu(float* dout,
int out_size = wout * hout;
int cnt = loop_w >> 2;
int remain = loop_w & 3;
int pad_left_new = pad_left > 4 ? 4 : pad_left;
int pad_right_new = pad_right > 4 ? 4 : pad_right;
int pad_top_new = pad_top > 4 ? 4 : pad_top;
int pad_bottom_new = pad_bottom > 4 ? 4 : pad_bottom;
int in_channel_size = chin * in_size;
int out_channel_size = chin * out_size;
int weights_size = 25;
......@@ -4676,10 +4538,9 @@ void conv_depthwise_5x5s1_bias_leakyRelu(float* dout,
const float* din_ptr_arr[] = {din_ptr0, din_ptr1, din_ptr2, din_ptr3, din_ptr4, din_ptr5};
float32x4_t weights_vec[] = {wr0, wr1, wr2, wr3, wr4, wr5, wr6};
// top_h
for (int h = pad_top_new; h > 0; h--) {
for (int h = pad_top; h > 0; h--) {
compute_all_padding_pre_leakyRelu(dout_ptr0, din_ptr_arr, vbias, scale, weights_vec, vzero,
win, wout, pad_left, pad_right, pad_left_new,
pad_right_new, cnt, remain, 4 - h);
win, wout, pad_left, pad_right, cnt, remain, 4 - h);
dout_ptr0 += wout;
din_ptr_arr[0] = din_ptr0;
din_ptr_arr[1] = din_ptr1;
......@@ -4691,8 +4552,7 @@ void conv_depthwise_5x5s1_bias_leakyRelu(float* dout,
// mid_h
for (int h = 0; h < loop_h - 1; h += 2) {
compute_all_padding_mid_leakyRelu_out2(dout_ptr0, dout_ptr1, din_ptr_arr, vbias, scale, weights_vec, vzero,
win, wout, pad_left, pad_right, pad_left_new,
pad_right_new, cnt, remain, 4);
win, wout, pad_left, pad_right, cnt, remain, 4);
dout_ptr0 += num_out;
dout_ptr1 += num_out;
din_ptr0 = din_ptr2;
......@@ -4710,8 +4570,7 @@ void conv_depthwise_5x5s1_bias_leakyRelu(float* dout,
}
if (loop_h % 2 != 0) {
compute_all_padding_mid_leakyRelu(dout_ptr0, din_ptr_arr, vbias, scale, weights_vec, vzero,
win, wout, pad_left, pad_right, pad_left_new,
pad_right_new, cnt, remain, 4);
win, wout, pad_left, pad_right, cnt, remain, 4);
dout_ptr0 = dout_ptr1;
din_ptr0 = din_ptr1;
din_ptr1 = din_ptr2;
......@@ -4725,10 +4584,9 @@ void conv_depthwise_5x5s1_bias_leakyRelu(float* dout,
din_ptr_arr[4] = din_ptr4;
}
// bottom
for (int h = 0; h < pad_bottom_new; h++) {
for (int h = 0; h < pad_bottom; h++) {
compute_all_padding_post_leakyRelu(dout_ptr0, din_ptr_arr, vbias, scale, weights_vec, vzero,
win, wout, pad_left, pad_right, pad_left_new,
pad_right_new, cnt, remain, 3 - h);
win, wout, pad_left, pad_right, cnt, remain, 3 - h);
dout_ptr0 += wout;
din_ptr_arr[0] = din_ptr0;
din_ptr_arr[1] = din_ptr1;
......
lite/backends/arm/math/conv_impl.cc
浏览文件 @ 9348cb41
......@@ -736,6 +736,7 @@ void conv_depthwise_5x5_fp32(const void* din,
bool flag_relu = param.fuse_relu;
bool flag_bias = param.bias != nullptr;
bool ch_four = ch_in > 4 * w_in;
bool pads_five = (pad_h < 5) || (pad_w < 5);
ctx->ExtendWorkspace((w_in + w_out) * sizeof(float));
bool flag_act = act_param.has_active;
if (stride == 2) {
......@@ -754,7 +755,7 @@ void conv_depthwise_5x5_fp32(const void* din,
act_param,
ctx);
} else if (stride == 1) {
if (ch_four || h_in < 5 || w_in < 5) {
if (ch_four || !pads_five || h_in < 5 || w_in < 5) {
conv_depthwise_5x5s1_fp32(reinterpret_cast<float*>(dout),
reinterpret_cast<const float*>(din),
reinterpret_cast<const float*>(weights),
......
lite/kernels/arm/conv_depthwise.cc
浏览文件 @ 9348cb41
......@@ -59,8 +59,9 @@ void DepthwiseConv<PRECISION(kFloat), PRECISION(kFloat)>::PrepareForRun() {
#endif
} else if (kw == 5) {
// VLOG(5) << "invoke 5x5 dw conv fp32";
bool pads_five = (paddings[0] < 5) || (paddings[2] < 5);
auto strides = param.strides;
if (ch_four && win >= kw && hin >= kw && (strides[0] == 1 && strides[1] == 1)) {
if (ch_four && pads_five && win >= kw && hin >= kw && (strides[0] == 1 && strides[1] == 1)) {
flag_trans_weights_ = false;
impl_ = lite::arm::math::conv_depthwise_5x5_fp32;
#ifdef LITE_WITH_PROFILE
......
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