move macro to inline function for check_out_of_range_for_image2d

mace/kernels/opencl/cl/activation.cl

@@ -31,8 +31,7 @@ __kernel void activation(KERNEL_ERROR_PARAMS
   DATA_TYPE4 out = do_activation(in, relux_max_limit);
 #endif
 
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(output, pos, hb, kernel_error);
-#endif
+
   WRITE_IMAGET(output, (int2)(pos, hb), out);
 }

mace/kernels/opencl/cl/addn.cl

@@ -32,9 +32,7 @@ __kernel void addn(KERNEL_ERROR_PARAMS
   out = out + in3;
 #endif
 
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(output, w, hb, kernel_error);
-#endif
   WRITE_IMAGET(output, (int2)(w, hb), out);
 }
 

mace/kernels/opencl/cl/batch_norm.cl

@@ -49,8 +49,7 @@ __kernel void batch_norm(KERNEL_ERROR_PARAMS
   out = do_activation(out, relux_max_limit);
 #endif
 
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(output, pos, hb, kernel_error);
-#endif
+
   WRITE_IMAGET(output, (int2)(pos, hb), out);
 }

mace/kernels/opencl/cl/bias_add.cl

@@ -24,8 +24,7 @@ __kernel void bias_add(KERNEL_ERROR_PARAMS
   DATA_TYPE4 bias_value = READ_IMAGET(bias, SAMPLER, (int2)(ch_blk, 0));
   DATA_TYPE4 out = in + bias_value;
 
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(output, pos, hb, kernel_error);
-#endif
+
   WRITE_IMAGET(output, (int2)(pos, hb), out);
 }

mace/kernels/opencl/cl/buffer_to_image.cl

@@ -50,9 +50,7 @@ __kernel void filter_buffer_to_image(KERNEL_ERROR_PARAMS
   }
 
   int2 coord = (int2)(w, h);
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(output, w, h, kernel_error);
-#endif
   WRITE_IMAGET(output, coord, values);
 }
 
@@ -160,9 +158,7 @@ __kernel void dw_filter_buffer_to_image(KERNEL_ERROR_PARAMS
   }
 
   int2 coord = (int2)(w, h);
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(output, w, h, kernel_error);
-#endif
   WRITE_IMAGET(output, coord, values);
 }
 
@@ -205,9 +201,7 @@ __kernel void in_out_buffer_to_image(KERNEL_ERROR_PARAMS
     values = vload4(0, input + offset);
   }
   int2 coord = (int2)(w, h);
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(output, w, h, kernel_error);
-#endif
   WRITE_IMAGET(output, coord, values);
 }
 
@@ -284,9 +278,7 @@ __kernel void arg_buffer_to_image(KERNEL_ERROR_PARAMS
     values = vload4(0, input + offset);
   }
   int2 coord = (int2)(w, h);
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(output, w, h, kernel_error);
-#endif
   WRITE_IMAGET(output, coord, values);
 }
 
@@ -364,9 +356,7 @@ __kernel void in_out_height_buffer_to_image(KERNEL_ERROR_PARAMS
       values.x = *(input + offset);
   }
   int2 coord = (int2)(w, h);
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(output, w, h, kernel_error);
-#endif
   WRITE_IMAGET(output, coord, values);
 }
 
@@ -448,9 +438,7 @@ __kernel void in_out_width_buffer_to_image(KERNEL_ERROR_PARAMS
       values.x = *(input + offset);
   }
   int2 coord = (int2)(w, h);
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(output, w, h, kernel_error);
-#endif
   WRITE_IMAGET(output, coord, values);
 }
 
@@ -525,9 +513,7 @@ __kernel void winograd_filter_buffer_to_image(KERNEL_ERROR_PARAMS
 
   int2 coord = (int2)(w, h);
 
-#ifdef OUT_OF_RANGE_CHECK
-    check_out_of_range_for_image2d(output, coord.x, coord.y + out_channels * 15, kernel_error);
-#endif
+  check_out_of_range_for_image2d(output, coord.x, coord.y + out_channels * 15, kernel_error);
 
 #pragma unroll
   for (short i = 0; i < 4; ++i) {

mace/kernels/opencl/cl/channel_shuffle.cl

@@ -51,9 +51,7 @@ __kernel void channel_shuffle(KERNEL_ERROR_PARAMS
 
     int out_x = mad24(mad24(group_chan_blk_idx, groups, g_blk), width, width_idx);
 
-#ifdef OUT_OF_RANGE_CHECK
-  check_out_of_range_for_image2d(output, out_x + groups_blks_width * 3, hb_idx, kernel_error);
-#endif
+    check_out_of_range_for_image2d(output, out_x + groups_blks_width * 3, hb_idx, kernel_error);
 
     WRITE_IMAGET(output, (int2)(out_x, hb_idx), out_chan_data0);
     out_x += groups_blks_width;

mace/kernels/opencl/cl/common.h

@@ -78,10 +78,12 @@ inline void check_out_of_range_for_image2d(__write_only image2d_t image,
                                            __private const int x,
                                            __private const int y,
                                            global char *kernel_error) {
+#ifdef OUT_OF_RANGE_CHECK
   int2 image_dim = get_image_dim(image);
   if (x >= image_dim.x || y >= image_dim.y) {
     *kernel_error = '1';
   }
+#endif
 }
 
 #endif  // MACE_KERNELS_OPENCL_CL_COMMON_H_

mace/kernels/opencl/cl/concat.cl

@@ -82,9 +82,8 @@ __kernel void concat_channel(KERNEL_ERROR_PARAMS
 
   const int pos = mad24(chan_blk_idx, width, width_idx);
 
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(output, pos, hb_idx, kernel_error);
-#endif
+
   WRITE_IMAGET(output, (int2)(pos, hb_idx), data);
 }
 
@@ -115,9 +114,8 @@ __kernel void concat_channel_multi(KERNEL_ERROR_PARAMS
 
   const int pos = mad24(chan_blk_idx + chan_blk_offset, width, width_idx);
 
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(output, pos, hb_idx, kernel_error);
-#endif
+
   WRITE_IMAGET(output, (int2)(pos, hb_idx), data);
 }
 

mace/kernels/opencl/cl/conv_2d.cl

@@ -129,30 +129,22 @@ __kernel void conv_2d(KERNEL_ERROR_PARAMS
   const int out_x_base = mul24(out_ch_blk, out_width);
 
   int w = out_w_blk;
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(output, out_x_base + w, out_hb, kernel_error);
-#endif
   WRITE_IMAGET(output, (int2)(out_x_base + w, out_hb), out0);
 
   w += out_w_blks;
   if (w >= out_width) return;
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(output, out_x_base + w, out_hb, kernel_error);
-#endif
   WRITE_IMAGET(output, (int2)(out_x_base + w, out_hb), out1);
 
   w += out_w_blks;
   if (w >= out_width) return;
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(output, out_x_base + w, out_hb, kernel_error);
-#endif
   WRITE_IMAGET(output, (int2)(out_x_base + w, out_hb), out2);
 
   w += out_w_blks;
   if (w >= out_width) return;
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(output, out_x_base + w, out_hb, kernel_error);
-#endif
   WRITE_IMAGET(output, (int2)(out_x_base + w, out_hb), out3);
 
 }

mace/kernels/opencl/cl/conv_2d_1x1.cl

@@ -105,29 +105,21 @@ __kernel void conv_2d_1x1(KERNEL_ERROR_PARAMS
 
   const int out_x_base = mul24(out_ch_blk, width);
   int out_x_idx = out_w_blk;
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(output, out_x_base + out_x_idx, out_hb, kernel_error);
-#endif
   WRITE_IMAGET(output, (int2)(out_x_base + out_x_idx, out_hb), out0);
 
   out_x_idx += out_w_blks;
   if (out_x_idx >= width) return;
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(output, out_x_base + out_x_idx, out_hb, kernel_error);
-#endif
   WRITE_IMAGET(output, (int2)(out_x_base + out_x_idx, out_hb), out1);
 
   out_x_idx += out_w_blks;
   if (out_x_idx >= width) return;
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(output, out_x_base + out_x_idx, out_hb, kernel_error);
-#endif
   WRITE_IMAGET(output, (int2)(out_x_base + out_x_idx, out_hb), out2);
 
   out_x_idx += out_w_blks;
   if (out_x_idx >= width) return;
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(output, out_x_base + out_x_idx, out_hb, kernel_error);
-#endif
   WRITE_IMAGET(output, (int2)(out_x_base + out_x_idx, out_hb), out3);
 }

mace/kernels/opencl/cl/conv_2d_3x3.cl

@@ -136,45 +136,35 @@ __kernel void conv_2d_3x3(KERNEL_ERROR_PARAMS
 
   const int out_x_base = mul24(out_ch_blk, out_width);
   int w = out_w_blk;
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(output, out_x_base + w, out_hb, kernel_error);
-#endif
   WRITE_IMAGET(output,
                (int2)(out_x_base + w, out_hb),
                out0);
 
   w += out_w_blks;
   if (w >= out_width) return;
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(output, out_x_base + w, out_hb, kernel_error);
-#endif
   WRITE_IMAGET(output,
                (int2)(out_x_base + w, out_hb),
                out1);
 
   w += out_w_blks;
   if (w >= out_width) return;
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(output, out_x_base + w, out_hb, kernel_error);
-#endif
   WRITE_IMAGET(output,
                (int2)(out_x_base + w, out_hb),
                out2);
 
   w += out_w_blks;
   if (w >= out_width) return;
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(output, out_x_base + w, out_hb, kernel_error);
-#endif
   WRITE_IMAGET(output,
                (int2)(out_x_base + w, out_hb),
                out3);
 
   w += out_w_blks;
   if (w >= out_width) return;
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(output, out_x_base + w, out_hb, kernel_error);
-#endif
   WRITE_IMAGET(output,
                (int2)(out_x_base + w, out_hb),
                out4);

mace/kernels/opencl/cl/cwise.cl

@@ -44,8 +44,7 @@ __kernel void cwise(KERNEL_ERROR_PARAMS
   out.w = fabs(in0.w);
 #endif
 
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(output, w, hb, kernel_error);
-#endif
+
   WRITE_IMAGET(output, (int2)(w, hb), out);
 }

mace/kernels/opencl/cl/depth_to_space.cl

@@ -37,9 +37,8 @@ __kernel void depth_to_space(KERNEL_ERROR_PARAMS
   const int in_pos = mad24(in_d, input_width, in_w);
   DATA_TYPE4 in_data = READ_IMAGET(input, SAMPLER, (int2)(in_pos, in_hb));
 
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(output, out_pos, out_hb, kernel_error);
-#endif
+
   WRITE_IMAGET(output, (int2)(out_pos, out_hb), in_data);
 }
 
@@ -80,8 +79,7 @@ __kernel void space_to_depth(KERNEL_ERROR_PARAMS
   const int out_pos = mad24(out_d, output_width, out_w);
   DATA_TYPE4 in_data = READ_IMAGET(input, SAMPLER, (int2)(in_pos, hb));
 
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(output, out_pos, out_hb, kernel_error);
-#endif
+
   WRITE_IMAGET(output, (int2)(out_pos, out_hb), in_data);
 }

mace/kernels/opencl/cl/depthwise_conv2d.cl

@@ -123,30 +123,22 @@ __kernel void depthwise_conv2d(KERNEL_ERROR_PARAMS
 
   const short out_x_base = mul24(out_ch_blk, out_width);
   short w = out_w_blk;
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(output, out_x_base + w, out_hb, kernel_error);
-#endif
   WRITE_IMAGET(output, (int2)(out_x_base + w, out_hb), out0);
 
   w += out_w_blks;
   if (w >= out_width) return;
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(output, out_x_base + w, out_hb, kernel_error);
-#endif
   WRITE_IMAGET(output, (int2)(out_x_base + w, out_hb), out1);
 
   w += out_w_blks;
   if (w >= out_width) return;
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(output, out_x_base + w, out_hb, kernel_error);
-#endif
   WRITE_IMAGET(output, (int2)(out_x_base + w, out_hb), out2);
 
   w += out_w_blks;
   if (w >= out_width) return;
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(output, out_x_base + w, out_hb, kernel_error);
-#endif
   WRITE_IMAGET(output, (int2)(out_x_base + w, out_hb), out3);
 }
 
@@ -261,29 +253,21 @@ __kernel void depthwise_conv2d_s1(KERNEL_ERROR_PARAMS
 
   const short out_x_base = mul24(out_ch_blk, out_width);
   short w = out_w_blk;
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(output, out_x_base + w, out_hb, kernel_error);
-#endif
   WRITE_IMAGET(output, (int2)(out_x_base + w, out_hb), out0);
 
   w += 1;
   if (w >= out_width) return;
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(output, out_x_base + w, out_hb, kernel_error);
-#endif
   WRITE_IMAGET(output, (int2)(out_x_base + w, out_hb), out1);
 
   w += 1;
   if (w >= out_width) return;
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(output, out_x_base + w, out_hb, kernel_error);
-#endif
   WRITE_IMAGET(output, (int2)(out_x_base + w, out_hb), out2);
 
   w += 1;
   if (w >= out_width) return;
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(output, out_x_base + w, out_hb, kernel_error);
-#endif
   WRITE_IMAGET(output, (int2)(out_x_base + w, out_hb), out3);
 }

mace/kernels/opencl/cl/eltwise.cl

@@ -37,8 +37,7 @@ __kernel void eltwise(KERNEL_ERROR_PARAMS
   out = in0 - in1;
 #endif
 
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(output, w, hb, kernel_error);
-#endif
+
   WRITE_IMAGET(output, (int2)(w, hb), out);
 }

mace/kernels/opencl/cl/fully_connected.cl

@@ -60,9 +60,8 @@ __kernel void fully_connected(KERNEL_ERROR_PARAMS
   result = do_activation(result, relux_max_limit);
 #endif
 
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(output, out_blk_idx, batch_idx, kernel_error);
-#endif
+
   WRITE_IMAGET(output, (int2)(out_blk_idx, batch_idx), result);
 }
 
@@ -154,9 +153,8 @@ __kernel void fully_connected_width(KERNEL_ERROR_PARAMS
     result = do_activation(result, relux_max_limit);
 #endif
 
-#ifdef OUT_OF_RANGE_CHECK
     check_out_of_range_for_image2d(output, out_blk_idx, batch_idx, kernel_error);
-#endif
+
     WRITE_IMAGET(output, (int2)(out_blk_idx, batch_idx), result);
   }
 }

mace/kernels/opencl/cl/matmul.cl

@@ -48,26 +48,18 @@ __kernel void matmul(KERNEL_ERROR_PARAMS
     c3 += (DATA_TYPE4)(dot(a0, b3), dot(a1, b3), dot(a2, b3), dot(a3, b3));
   }
 
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(C, gx, gy, kernel_error);
-#endif
   WRITE_IMAGET(C, (int2)(gx, gy), c0);
 
   if ((gx + 1) >= N) return;
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(C, gx + 1, gy, kernel_error);
-#endif
   WRITE_IMAGET(C, (int2)(gx + 1, gy), c1);
 
   if ((gx + 2) >= N) return;
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(C, gx + 2, gy, kernel_error);
-#endif
   WRITE_IMAGET(C, (int2)(gx + 2, gy), c2);
 
   if ((gx + 3) >= N) return;
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(C, gx + 3, gy, kernel_error);
-#endif
   WRITE_IMAGET(C, (int2)(gx + 3, gy), c3);
 }

mace/kernels/opencl/cl/pooling.cl

@@ -96,8 +96,6 @@ __kernel void pooling(KERNEL_ERROR_PARAMS
 #endif
 
   const int pos = mad24(out_chan_idx, out_width, out_width_idx);
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(output, pos, out_hb_idx, kernel_error);
-#endif
   WRITE_IMAGET(output, (int2)(pos, out_hb_idx), res);
 }

mace/kernels/opencl/cl/resize_bilinear.cl

@@ -58,9 +58,8 @@ __kernel void resize_bilinear_nocache(KERNEL_ERROR_PARAMS
   const int out_w_offset = mul24(ch_blk, out_width);
   const int out_h_offset = mul24(b, out_height);
 
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(output, out_w_offset + w, out_h_offset + h, kernel_error);
-#endif
+
   WRITE_IMAGET(output, (int2)(out_w_offset + w, out_h_offset + h), out);
 }
 

mace/kernels/opencl/cl/slice.cl

@@ -24,9 +24,6 @@ __kernel void slice(KERNEL_ERROR_PARAMS
                                              width, width_idx), hb_idx));
 
   const int pos = mad24(chan_blk_idx, width, width_idx);
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(output, pos, hb_idx, kernel_error);
-#endif
-  WRITE_IMAGET(output,
-               (int2)(pos, hb_idx), data);
+  WRITE_IMAGET(output, (int2)(pos, hb_idx), data);
 }

mace/kernels/opencl/cl/softmax.cl

@@ -85,8 +85,7 @@ __kernel void softmax(KERNEL_ERROR_PARAMS
       data = native_exp(data) / sum;
   }
 
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(output, pos, hb_idx, kernel_error);
-#endif
+
   WRITE_IMAGET(output, (int2)(pos, hb_idx), data);
 }

mace/kernels/opencl/cl/space_to_batch.cl

@@ -46,9 +46,8 @@ __kernel void space_to_batch(KERNEL_ERROR_PARAMS
 
   int2 batch_coord = (int2)(mul24(chan_idx, batch_width) + batch_w_idx, batch_hb_idx);
 
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(batch_data, batch_coord.x, batch_coord.y, kernel_error);
-#endif
+
   WRITE_IMAGET(batch_data, batch_coord, value);
 }
 
@@ -94,9 +93,8 @@ __kernel void batch_to_space(KERNEL_ERROR_PARAMS
     int2 space_coord = (int2)(mul24(chan_idx, space_width) + space_w_idx,
                               space_b_idx * space_height + space_h_idx);
 
-#ifdef OUT_OF_RANGE_CHECK
     check_out_of_range_for_image2d(space_data, space_coord.x, space_coord.y, kernel_error);
-#endif
+
     WRITE_IMAGET(space_data, space_coord, value);
   }
 }

mace/kernels/opencl/cl/winograd_transform.cl

@@ -94,9 +94,8 @@ __kernel void winograd_transform_2x2(KERNEL_ERROR_PARAMS
   input3[2] = tv3[2] - tv3[1];
   input3[3] = tv3[1] - tv3[3];
 
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(output, out_width_idx, chan_blk_idx + chan_blk_idx * 15, kernel_error);
-#endif
+
 #pragma unroll
   for (short i = 0; i < 4; ++i) {
     WRITE_IMAGET(output, (int2)(out_width_idx, chan_blk_idx), input0[i]);
@@ -213,30 +212,23 @@ __kernel void winograd_inverse_transform_2x2(KERNEL_ERROR_PARAMS
   in1[1] = do_activation(in1[1], relux_max_limit);
 #endif
 
-#ifdef OUT_OF_RANGE_CHECK
   check_out_of_range_for_image2d(output, coord_x, coord_y, kernel_error);
-#endif
+
   WRITE_IMAGET(output, (int2)(coord_x, coord_y), in0[0]);
 
   t = 0;
   if (out_width_idx + 1 < out_width) {
-#ifdef OUT_OF_RANGE_CHECK
     check_out_of_range_for_image2d(output, coord_x + 1, coord_y, kernel_error);
-#endif
     WRITE_IMAGET(output, (int2)(coord_x + 1, coord_y), in0[1]);
     t += 1;
   }
   if (out_height_idx + 1 < out_height) {
-#ifdef OUT_OF_RANGE_CHECK
     check_out_of_range_for_image2d(output, coord_x, coord_y + 1, kernel_error);
-#endif
     WRITE_IMAGET(output, (int2)(coord_x, coord_y + 1), in1[0]);
     t += 1;
   }
   if (t == 2) {
-#ifdef OUT_OF_RANGE_CHECK
     check_out_of_range_for_image2d(output, coord_x + 1, coord_y + 1, kernel_error);
-#endif
     WRITE_IMAGET(output, (int2)(coord_x + 1, coord_y + 1), in1[1]);
   }