!5411 optimize pooling_relu(max_mode)

Merge pull request !5411 from songhonglei413/roi

mindspore/lite/nnacl/fp32/pooling.c

@@ -330,15 +330,14 @@ void MaxPoolingRelu(const float *input_ptr, float *output_ptr, PoolingParameter
   int out_tile_count = UP_DIV(out_plane, TILE_NUM);
   int thread_num = pooling_param->thread_num_;
   int c4 = UP_DIV(channel, C4NUM);
-  // input channel is equal to output channel
 
 #ifdef ENABLE_NEON
   float32x4_t zeros = vdupq_n_f32(0);
 #endif
 
   for (int batch = 0; batch < output_batch; batch++) {
-    int in_batch_offset = batch * in_h * in_w * channel;
-    int out_batch_offset = batch * output_h * output_w * channel;
+    const float *src_b_ptr = input_ptr + batch * in_h * in_w * channel;
+    float *dst_b_ptr = output_ptr + batch * output_h * output_w * channel;
     for (int thread_id = task_id; thread_id < out_tile_count; thread_id += thread_num) {
       int cal_start_index = thread_id * TILE_NUM;
       int real_cal_num = (out_plane - cal_start_index) > TILE_NUM ? TILE_NUM : (out_plane - cal_start_index);
@@ -348,10 +347,18 @@ void MaxPoolingRelu(const float *input_ptr, float *output_ptr, PoolingParameter
         int out_h_index = index / output_w;
         int in_w_index = out_w_index * stride_w - pad_w;
         int in_h_index = out_h_index * stride_h - pad_h;
-        int out_plane_offset = out_batch_offset + index * channel;
-        for (int j = 0; j < c4 - 1; j++) {
-          int in_channel_offset = in_batch_offset + j * C4NUM;
-          int out_channel_offset = out_plane_offset + j * C4NUM;
+
+        const float *src_plane_ptr = src_b_ptr;
+        float *dst_plane_ptr = dst_b_ptr + index * channel;
+
+        int real_win_h_start = MSMAX(0, -in_h_index);
+        int real_win_h_end = MSMIN(win_h, in_h - in_h_index);
+        int resl_win_w_start = MSMAX(0, -in_w_index);
+        int real_win_w_end = MSMIN(win_w, in_w - in_w_index);
+
+        for (int ci = 0; ci < c4 - 1; ci++) {
+          const float *src_c_ptr = src_plane_ptr + ci * C4NUM;
+          float *dst_c_ptr = dst_plane_ptr + ci * C4NUM;
 #ifdef ENABLE_NEON
           float32x4_t tmp_max = vdupq_n_f32(-FLT_MAX);
 #else
@@ -360,28 +367,24 @@ void MaxPoolingRelu(const float *input_ptr, float *output_ptr, PoolingParameter
           float tmp_max3 = -FLT_MAX;
           float tmp_max4 = -FLT_MAX;
 #endif
-          for (int h = 0; h < win_h; h++) {
-            for (int w = 0; w < win_w; w++) {
-              if ((in_h_index + h) < 0 || (in_h_index + h) >= in_h || (in_w_index + w) < 0 ||
-                  (in_w_index + w) >= in_w) {
-                continue;
-              } else {
-                int in_offset = in_channel_offset + ((in_h_index + h) * in_w + in_w_index + w) * channel;
+
+          for (int kh = real_win_h_start; kh < real_win_h_end; kh++) {
+            for (int kw = resl_win_w_start; kw < real_win_w_end; kw++) {
+              const float *src_win_ptr = src_c_ptr + ((in_h_index + kh) * in_w + in_w_index + kw) * channel;
 #ifdef ENABLE_NEON
-                tmp_max = vmaxq_f32(tmp_max, vld1q_f32(input_ptr + in_offset));
+              tmp_max = vmaxq_f32(tmp_max, vld1q_f32(src_win_ptr));
 #else
-                tmp_max1 = fmax(tmp_max1, *(input_ptr + in_offset));
-                tmp_max2 = fmax(tmp_max2, *(input_ptr + in_offset + 1));
-                tmp_max3 = fmax(tmp_max3, *(input_ptr + in_offset + 2));
-                tmp_max4 = fmax(tmp_max4, *(input_ptr + in_offset + 3));
+              tmp_max1 = fmax(tmp_max1, src_win_ptr[0]);
+              tmp_max2 = fmax(tmp_max2, src_win_ptr[1]);
+              tmp_max3 = fmax(tmp_max3, src_win_ptr[2]);
+              tmp_max4 = fmax(tmp_max4, src_win_ptr[3]);
 
 #endif
-              }
             }  // win_w loop
           }    // win_h loop
 #ifdef ENABLE_NEON
           tmp_max = vmaxq_f32(tmp_max, zeros);
-          vst1q_f32(output_ptr + out_channel_offset, tmp_max);
+          vst1q_f32(dst_c_ptr, tmp_max);
 #else
           // relu:
           tmp_max1 = fmax(tmp_max1, 0);
@@ -389,30 +392,25 @@ void MaxPoolingRelu(const float *input_ptr, float *output_ptr, PoolingParameter
           tmp_max3 = fmax(tmp_max3, 0);
           tmp_max4 = fmax(tmp_max4, 0);
 
-          *(output_ptr + out_channel_offset) = tmp_max1;
-          *(output_ptr + out_channel_offset + 1) = tmp_max2;
-          *(output_ptr + out_channel_offset + 2) = tmp_max3;
-          *(output_ptr + out_channel_offset + 3) = tmp_max4;
+          dst_c_ptr[0] = tmp_max1;
+          dst_c_ptr[1] = tmp_max2;
+          dst_c_ptr[2] = tmp_max3;
+          dst_c_ptr[3] = tmp_max4;
 #endif
         }  // ic4-1 loop
         int channel_s = (c4 - 1) * C4NUM;
-        for (int k = channel_s; k < channel; k++) {
-          int in_channel_offset = in_batch_offset + k;
-          int out_channel_offset = out_plane_offset + k;
+        for (int ci = channel_s; ci < channel; ci++) {
+          float *dst_c_ptr = dst_plane_ptr + ci;
+          const float *src_c_ptr = src_plane_ptr + ci;
           float tmp_max = -FLT_MAX;
-          for (int h = 0; h < win_h; h++) {
-            for (int w = 0; w < win_w; w++) {
-              if ((in_h_index + h) < 0 || (in_h_index + h) >= in_h || (in_w_index + w) < 0 ||
-                  (in_w_index + w) >= in_w) {
-                continue;
-              } else {
-                int in_offset = in_channel_offset + ((in_h_index + h) * in_w + in_w_index + w) * channel;
-                tmp_max = fmax(tmp_max, *(input_ptr + in_offset));
-                tmp_max = fmax(tmp_max, 0);
-              }
+
+          for (int kh = real_win_h_start; kh < real_win_h_end; kh++) {
+            for (int kw = resl_win_w_start; kw < real_win_w_end; kw++) {
+              const float *src_win_ptr = src_c_ptr + ((in_h_index + kh) * in_w + in_w_index + kw) * channel;
+              tmp_max = fmax(tmp_max, src_win_ptr[0]);
             }  // win_w loop
           }    // win_h loop
-          *(output_ptr + out_channel_offset) = tmp_max;
+          dst_c_ptr[0] = tmp_max;
         }  // channel_res loop
       }    // real_cal_num loop
     }      // out_plane loop
@@ -557,7 +555,6 @@ void MaxPoolingRelu6(const float *input_ptr, float *output_ptr, PoolingParameter
   int out_tile_count = UP_DIV(out_plane, TILE_NUM);
   int thread_num = pooling_param->thread_num_;
   int c4 = UP_DIV(channel, C4NUM);
-  // input channel is equal to output channel
 
 #ifdef ENABLE_NEON
   float32x4_t zeros = vdupq_n_f32(0);
@@ -565,8 +562,8 @@ void MaxPoolingRelu6(const float *input_ptr, float *output_ptr, PoolingParameter
 #endif
 
   for (int batch = 0; batch < output_batch; batch++) {
-    int in_batch_offset = batch * in_h * in_w * channel;
-    int out_batch_offset = batch * output_h * output_w * channel;
+    const float *src_b_ptr = input_ptr + batch * in_h * in_w * channel;
+    float *dst_b_ptr = output_ptr + batch * output_h * output_w * channel;
     for (int thread_id = task_id; thread_id < out_tile_count; thread_id += thread_num) {
       int cal_start_index = thread_id * TILE_NUM;
       int real_cal_num = (out_plane - cal_start_index) > TILE_NUM ? TILE_NUM : (out_plane - cal_start_index);
@@ -576,10 +573,18 @@ void MaxPoolingRelu6(const float *input_ptr, float *output_ptr, PoolingParameter
         int out_h_index = index / output_w;
         int in_w_index = out_w_index * stride_w - pad_w;
         int in_h_index = out_h_index * stride_h - pad_h;
-        int out_plane_offset = out_batch_offset + index * channel;
-        for (int j = 0; j < c4 - 1; j++) {
-          int in_channel_offset = in_batch_offset + j * C4NUM;
-          int out_channel_offset = out_plane_offset + j * C4NUM;
+
+        const float *src_plane_ptr = src_b_ptr;
+        float *dst_plane_ptr = dst_b_ptr + index * channel;
+
+        int real_win_h_start = MSMAX(0, -in_h_index);
+        int real_win_h_end = MSMIN(win_h, in_h - in_h_index);
+        int resl_win_w_start = MSMAX(0, -in_w_index);
+        int real_win_w_end = MSMIN(win_w, in_w - in_w_index);
+
+        for (int ci = 0; ci < c4 - 1; ci++) {
+          const float *src_c_ptr = src_plane_ptr + ci * C4NUM;
+          float *dst_c_ptr = dst_plane_ptr + ci * C4NUM;
 #ifdef ENABLE_NEON
           float32x4_t tmp_max = vdupq_n_f32(-FLT_MAX);
 #else
@@ -588,30 +593,27 @@ void MaxPoolingRelu6(const float *input_ptr, float *output_ptr, PoolingParameter
           float tmp_max3 = -FLT_MAX;
           float tmp_max4 = -FLT_MAX;
 #endif
-          for (int h = 0; h < win_h; h++) {
-            for (int w = 0; w < win_w; w++) {
-              if ((in_h_index + h) < 0 || (in_h_index + h) >= in_h || (in_w_index + w) < 0 ||
-                  (in_w_index + w) >= in_w) {
-                continue;
-              } else {
-                int in_offset = in_channel_offset + ((in_h_index + h) * in_w + in_w_index + w) * channel;
+
+          for (int kh = real_win_h_start; kh < real_win_h_end; kh++) {
+            for (int kw = resl_win_w_start; kw < real_win_w_end; kw++) {
+              const float *src_win_ptr = src_c_ptr + ((in_h_index + kh) * in_w + in_w_index + kw) * channel;
 #ifdef ENABLE_NEON
-                tmp_max = vmaxq_f32(tmp_max, vld1q_f32(input_ptr + in_offset));
+              tmp_max = vmaxq_f32(tmp_max, vld1q_f32(src_win_ptr));
 #else
-                tmp_max1 = fmax(tmp_max1, *(input_ptr + in_offset));
-                tmp_max2 = fmax(tmp_max2, *(input_ptr + in_offset + 1));
-                tmp_max3 = fmax(tmp_max3, *(input_ptr + in_offset + 2));
-                tmp_max4 = fmax(tmp_max4, *(input_ptr + in_offset + 3));
+              tmp_max1 = fmax(tmp_max1, src_win_ptr[0]);
+              tmp_max2 = fmax(tmp_max2, src_win_ptr[1]);
+              tmp_max3 = fmax(tmp_max3, src_win_ptr[2]);
+              tmp_max4 = fmax(tmp_max4, src_win_ptr[3]);
 
 #endif
-              }
             }  // win_w loop
           }    // win_h loop
 #ifdef ENABLE_NEON
           tmp_max = vmaxq_f32(tmp_max, zeros);
           tmp_max = vminq_f32(tmp_max, bounds);
-          vst1q_f32(output_ptr + out_channel_offset, tmp_max);
+          vst1q_f32(dst_c_ptr, tmp_max);
 #else
+          // relu:
           tmp_max1 = fmax(tmp_max1, 0);
           tmp_max2 = fmax(tmp_max2, 0);
           tmp_max3 = fmax(tmp_max3, 0);
@@ -621,31 +623,25 @@ void MaxPoolingRelu6(const float *input_ptr, float *output_ptr, PoolingParameter
           tmp_max3 = fmin(tmp_max3, 6);
           tmp_max4 = fmin(tmp_max4, 6);
 
-          *(output_ptr + out_channel_offset) = tmp_max1;
-          *(output_ptr + out_channel_offset + 1) = tmp_max2;
-          *(output_ptr + out_channel_offset + 2) = tmp_max3;
-          *(output_ptr + out_channel_offset + 3) = tmp_max4;
+          dst_c_ptr[0] = tmp_max1;
+          dst_c_ptr[1] = tmp_max2;
+          dst_c_ptr[2] = tmp_max3;
+          dst_c_ptr[3] = tmp_max4;
 #endif
         }  // ic4-1 loop
         int channel_s = (c4 - 1) * C4NUM;
-        for (int k = channel_s; k < channel; k++) {
-          int in_channel_offset = in_batch_offset + k;
-          int out_channel_offset = out_plane_offset + k;
+        for (int ci = channel_s; ci < channel; ci++) {
+          float *dst_c_ptr = dst_plane_ptr + ci;
+          const float *src_c_ptr = src_plane_ptr + ci;
           float tmp_max = -FLT_MAX;
-          for (int h = 0; h < win_h; h++) {
-            for (int w = 0; w < win_w; w++) {
-              if ((in_h_index + h) < 0 || (in_h_index + h) >= in_h || (in_w_index + w) < 0 ||
-                  (in_w_index + w) >= in_w) {
-                continue;
-              } else {
-                int in_offset = in_channel_offset + ((in_h_index + h) * in_w + in_w_index + w) * channel;
-                tmp_max = fmax(tmp_max, *(input_ptr + in_offset));
-                tmp_max = fmax(tmp_max, 0);
-                tmp_max = fmin(tmp_max, 6);
-              }
+
+          for (int kh = real_win_h_start; kh < real_win_h_end; kh++) {
+            for (int kw = resl_win_w_start; kw < real_win_w_end; kw++) {
+              const float *src_win_ptr = src_c_ptr + ((in_h_index + kh) * in_w + in_w_index + kw) * channel;
+              tmp_max = fmax(tmp_max, src_win_ptr[0]);
             }  // win_w loop
           }    // win_h loop
-          *(output_ptr + out_channel_offset) = tmp_max;
+          dst_c_ptr[0] = tmp_max;
         }  // channel_res loop
       }    // real_cal_num loop
     }      // out_plane loop