From 53a6ec5944132000e2c6779c04d722b3b2d2501c Mon Sep 17 00:00:00 2001 From: Xinghang Zhou Date: Tue, 14 Nov 2017 15:14:25 +0800 Subject: APPBROWSER-304,342: Add exclude padding support for OpenGL ES implementation and implement MaxPool operators Change-Id: Ie6ba36ff114feec2a21739dba11bbb60b76af443 Reviewed-on: https://eu-gerrit-1.euhpc.arm.com/113697 Tested-by: Jenkins Reviewed-by: Stephen Li Reviewed-by: Pablo Tello Reviewed-by: Anthony Barbier --- src/core/GLES_COMPUTE/cs_shaders/fill_border.cs | 46 +++++- src/core/GLES_COMPUTE/cs_shaders/pooling_layer.cs | 174 ++++++++++++++++------ 2 files changed, 168 insertions(+), 52 deletions(-) (limited to 'src/core/GLES_COMPUTE/cs_shaders') diff --git a/src/core/GLES_COMPUTE/cs_shaders/fill_border.cs b/src/core/GLES_COMPUTE/cs_shaders/fill_border.cs index c64572b061..4f87b9242a 100644 --- a/src/core/GLES_COMPUTE/cs_shaders/fill_border.cs +++ b/src/core/GLES_COMPUTE/cs_shaders/fill_border.cs @@ -132,7 +132,7 @@ void main() ImageIterator buf_iter = CONVERT_TENSOR3D_TO_IMAGE_ITERATOR_NO_STEP(buf_attrs, buf_shift); // Update pointer to point to the starting point of the valid region - TENSOR_ITERATOR_ADVANCE_IN_BYTES(buf_iter, uint(start_pos_y) * buf_attrs.stride_y + uint(start_pos_x) * buf_attrs.stride_x); + TENSOR_ITERATOR_ADVANCE_IN_BYTES(buf_iter, start_pos_y * int(buf_attrs.stride_y) + start_pos_x * int(buf_attrs.stride_x)); int total_width = BORDER_SIZE_LEFT + int(width) + BORDER_SIZE_RIGHT; int gid0 = int(gl_GlobalInvocationID.x); @@ -158,12 +158,29 @@ void main() { if(pos % 2 == 0) { - STORE_PACK2_HALF(buf_ptr, offset, left_val.xx); + if(BORDER_SIZE_LEFT % 2 == 0) + { + STORE_PACK2_HALF(buf_ptr, offset, left_val.xx); + } + else + { + STORE_PACK2_HALF(buf_ptr, offset, left_val.yy); + } + i++; } } } // Handle right border - vec2 right_val = LOAD_UNPACK2_HALF(buf_ptr, IMAGE_OFFSET(buf_iter, int(width) - 1, gidH)); + vec2 right_val_origin = LOAD_UNPACK2_HALF(buf_ptr, IMAGE_OFFSET(buf_iter, int(width) - 1, gidH)); + vec2 right_val; + if((((BORDER_SIZE_LEFT + int(width)) % 2)) == 1) + { + right_val = vec2(right_val_origin.x, right_val_origin.x); + } + else + { + right_val = vec2(right_val_origin.y, right_val_origin.y); + } for(int i = 0; i < BORDER_SIZE_RIGHT; ++i) { uint offset = IMAGE_OFFSET(buf_iter, int(width) + i, gidH); @@ -173,7 +190,8 @@ void main() { if(pos % 2 == 0) { - STORE_PACK2_HALF(buf_ptr, offset, right_val.yy); + STORE_PACK2_HALF(buf_ptr, offset, right_val); + i++; } else { @@ -184,7 +202,8 @@ void main() { if(pos % 2 == 0) { - STORE_PACK2_HALF(buf_ptr, offset, right_val.yy); + STORE_PACK2_HALF(buf_ptr, offset, right_val); + i++; } } } @@ -208,7 +227,14 @@ void main() { if(gidW == (int(width) - 1)) { - STORE_PACK2_HALF(buf_ptr, offset, top_val.xx); + if(((BORDER_SIZE_LEFT + int(width)) % 2 == 1)) + { + STORE_PACK2_HALF(buf_ptr, offset, top_val.xx); + } + else + { + STORE_PACK2_HALF(buf_ptr, offset, top_val.yy); + } } else { @@ -229,6 +255,10 @@ void main() { STORE_PACK2_HALF(buf_ptr, offset, top_val.yy); } + else + { + STORE_PACK2_HALF(buf_ptr, offset, top_val.xx); + } } else { @@ -268,6 +298,10 @@ void main() { STORE_PACK2_HALF(buf_ptr, offset, bottom_val.yy); } + else + { + STORE_PACK2_HALF(buf_ptr, offset, bottom_val.xx); + } } else { diff --git a/src/core/GLES_COMPUTE/cs_shaders/pooling_layer.cs b/src/core/GLES_COMPUTE/cs_shaders/pooling_layer.cs index 401b002111..64767a7ef1 100644 --- a/src/core/GLES_COMPUTE/cs_shaders/pooling_layer.cs +++ b/src/core/GLES_COMPUTE/cs_shaders/pooling_layer.cs @@ -259,7 +259,7 @@ layout(std140) uniform shader_params POOL_OP(data001.xyzw, data001.xyzw, data201.xyzw); \ POOL_OP(data010.xyzw, data010.xyzw, data21.xyzw); \ POOL_OP(res.xyzw, vec4(data000.xw, data001.z, data010.y), vec4(data000.y, data001.xw, data010.z)); \ - POOL_OP(res.xyzw, res.xyzw, vec4(data000.z, data001.y data010.xw)) + POOL_OP(res.xyzw, res.xyzw, vec4(data000.z, data001.y, data010.xw)) float calculate_max(const int pool_size, Tensor3D src, const int upper_bound_w, const int upper_bound_h, const int pad_x, const int pad_y, const int stride_x, const int stride_y) { @@ -271,11 +271,11 @@ float calculate_max(const int pool_size, Tensor3D src, const int upper_bound_w, float data_max; data_max = LOAD4(src, tensor3D_offset(src, 0, 0, 0)); - for(int i = 0; (start_x + i) < end_x; ++i) + for(int i = 0; (start_y + i) < end_y; ++i) { - for(int j = 0; (start_y + j) < end_y; ++j) + for(int j = 0; (start_x + j) < end_x; ++j) { - float data = LOAD4(src, tensor3D_offset(src, i, j, 0)); + float data = LOAD4(src, tensor3D_offset(src, j, i, 0)); POOL_OP_float(data_max, data_max, data); } } @@ -308,6 +308,11 @@ float calculate_avg(const int pool_size, Tensor3D src, const int upper_bound_w, } } +#if defined(EXCLUDE_PADDING) + start_x = max(0, start_x); + start_y = max(0, start_y); +#endif /* defined(EXCLUDE_PADDING) */ + return data_total / float((end_y - start_y) * (end_x - start_x)); } @@ -460,6 +465,10 @@ void main(void) int start_y = int(gl_GlobalInvocationID.y) * STRIDE_Y - PAD_Y; ivec4 end_x = min((start_x + (ivec4(3))), (ivec4(MAX_WIDTH))); int end_y = min((start_y + 3), MAX_HEIGHT); +#if defined(EXCLUDE_PADDING) + start_x = max(ivec4(0), start_x); + start_y = max(0, start_y); +#endif /* defined(EXCLUDE_PADDING) */ res *= (vec4((1.f)) / vec4((ivec4(end_y - start_y)) * (end_x - start_x))); #endif /*POOL_AVG*/ @@ -606,12 +615,16 @@ void main(void) #if defined(POOL_AVG) || defined(POOL_L2) { // Divide by pool region in case of average pooling - int start_x = int(gl_GlobalInvocationID.x) * STRIDE_X - PAD_X; - int start_y = int(gl_GlobalInvocationID.y) * STRIDE_Y - PAD_Y; - int end_x = int(min(start_x + POOL_SIZE, MAX_WIDTH)); - int end_y = int(min(start_y + POOL_SIZE, MAX_HEIGHT)); - float res1 = float((end_y - start_y) * (end_x - start_x)); - res = DIV_OP(res, res1); + int start_x = int(gl_GlobalInvocationID.x) * STRIDE_X - PAD_X; + int start_y = int(gl_GlobalInvocationID.y) * STRIDE_Y - PAD_Y; + int end_x = int(min(start_x + POOL_SIZE, MAX_WIDTH)); + int end_y = int(min(start_y + POOL_SIZE, MAX_HEIGHT)); +#if defined(EXCLUDE_PADDING) + start_x = max(0, start_x); + start_y = max(0, start_y); +#endif /* defined(EXCLUDE_PADDING) */ + float res1 = float((end_y - start_y) * (end_x - start_x)); + res = DIV_OP(res, res1); } #endif /* defined(POOL_AVG) || defined(POOL_L2) */ @@ -867,7 +880,7 @@ layout(std140) uniform shader_params POOL_OP(data001.xyzw, data001.xyzw, data201.xyzw); \ POOL_OP(data010.xyzw, data010.xyzw, data21.xyzw); \ POOL_OP(res.xyzw, vec4(data000.xw, data001.z, data010.y), vec4(data000.y, data001.xw, data010.z)); \ - POOL_OP(res.xyzw, res.xyzw, vec4(data000.z, data001.y data010.xw)) + POOL_OP(res.xyzw, res.xyzw, vec4(data000.z, data001.y, data010.xw)) vec2 load_and_unpack(Tensor3D src, uint offset) { @@ -970,7 +983,7 @@ vec2 calculate_max(const int pool_size, Tensor3D src, const int upper_bound_w, c vec2 calculate_avg(const int pool_size, Tensor3D src, const int upper_bound_w, const int upper_bound_h, const int pad_x, const int pad_y, const int stride_x, const int stride_y) { - int start_x1 = int(gl_GlobalInvocationID.x) * stride_x - pad_x; + int start_x1 = (2 * int(gl_GlobalInvocationID.x)) * stride_x - pad_x; int start_y1 = int(gl_GlobalInvocationID.y) * stride_y - pad_y; int end_x1 = int(min(start_x1 + pool_size, upper_bound_w)); int end_y1 = int(min(start_y1 + pool_size, upper_bound_h)); @@ -1002,11 +1015,11 @@ vec2 calculate_avg(const int pool_size, Tensor3D src, const int upper_bound_w, c } //Calculate sum2 - if((start_x2 + j) < end_x2 && end_x1 < upper_bound_w) + if((start_x2 + j) < end_x2 && end_x1 <= upper_bound_w) { if((stride_x % 2) == 0) { - vec2 data2 = load_and_unpack(src, (tensor3D_offset_fp16(src, (j + stride_x + 1), i, 0) >> uint(2))); + vec2 data2 = load_and_unpack(src, (tensor3D_offset_fp16(src, (j + stride_x), i, 0) >> uint(2))); #if defined(POOL_L2) // Raise to power of 2 for L2 Pooling data2 = POW2_OP(data2, 2); @@ -1040,6 +1053,13 @@ vec2 calculate_avg(const int pool_size, Tensor3D src, const int upper_bound_w, c } } } +#if defined(EXCLUDE_PADDING) + start_x1 = max(0, start_x1); + start_y1 = max(0, start_y1); + start_x2 = max(0, start_x2); + start_y2 = max(0, start_y2); +#endif /* defined(EXCLUDE_PADDING) */ + //Calculate average vec2 data_avg; data_avg.x = data_total1 / float((end_y1 - start_y1) * (end_x1 - start_x1)); @@ -1203,6 +1223,10 @@ void main(void) int start_y = int(gl_GlobalInvocationID.y) * STRIDE_Y - PAD_Y; ivec4 end_x = min((start_x + (ivec4(3))), (ivec4(MAX_WIDTH))); int end_y = min((start_y + 3), MAX_HEIGHT); +#if defined(EXCLUDE_PADDING) + start_x = max(ivec4(0), start_x); + start_y = max(0, start_y); +#endif /* defined(EXCLUDE_PADDING) */ res *= (vec4((1.f)) / vec4((ivec4(end_y - start_y)) * (end_x - start_x))); #endif /*POOL_AVG*/ @@ -1354,44 +1378,96 @@ void main(void) } } - for(int y = STRIDE_X; y < int(POOL_SIZE + STRIDE_X); y++) + for(int y = 0; y < int(POOL_SIZE); y++) { - int x1 = STRIDE_X; - for(; x1 <= (int(POOL_SIZE + STRIDE_X) - 8); x1 += 8) + if((STRIDE_X % 2) == 0) { - vec4 data2; - vec4 data3; - LOAD4_fp16(data2, src, (tensor3D_offset_fp16(src, x1, y, 0) >> uint(2))); - LOAD4_fp16(data3, src, (tensor3D_offset_fp16(src, x1, y, 0) >> uint(2)) + uint(2)); + int x1 = STRIDE_X; + for(; x1 <= (int(POOL_SIZE + STRIDE_X) - 8); x1 += 8) + { + vec4 data2; + vec4 data3; + LOAD4_fp16(data2, src, (tensor3D_offset_fp16(src, x1, y, 0) >> uint(2))); + LOAD4_fp16(data3, src, (tensor3D_offset_fp16(src, x1, y, 0) >> uint(2)) + uint(2)); #if defined(POOL_L2) - // Raise to power of 2 for L2 Pooling - data2 *= data2; - data3 *= data3; + // Raise to power of 2 for L2 Pooling + data2 *= data2; + data3 *= data3; #endif /* defined(POOL_L2) */ - POOL_OP(vdata01, vdata01, data2); - POOL_OP(vdata11, vdata11, data3); - } + POOL_OP(vdata01, vdata01, data2); + POOL_OP(vdata11, vdata11, data3); + } - // Leftover - for(; x1 < int(POOL_SIZE + STRIDE_X); x1 = x1 + 2) + // Leftover + for(; x1 < int(POOL_SIZE + STRIDE_X); x1 = x1 + 2) + { + vec2 data4middle; + data4middle = load_and_unpack(src, (tensor3D_offset_fp16(src, x1, y, 0) >> uint(2))); +#if defined(POOL_L2) + // Raise to power of 2 for L2 Pooling + data4middle *= data4middle; +#endif /* defined(POOL_L2) */ + if((x1 + 1) >= int(POOL_SIZE + STRIDE_X)) + { + POOL_OP_float(sdata.y, sdata.y, data4middle.x); + } + else + { + float data4; + POOL_OP_float(data4, data4middle.x, data4middle.y); + POOL_OP_float(sdata.y, sdata.y, data4); + } + } + } + else { - vec2 data4middle; - data4middle = load_and_unpack(src, (tensor3D_offset_fp16(src, x1, y, 0) >> uint(2))); + vec2 dataorigin2; + dataorigin2 = load_and_unpack(src, (tensor3D_offset_fp16(src, (STRIDE_X - 1), y, 0) >> uint(2))); #if defined(POOL_L2) // Raise to power of 2 for L2 Pooling - data4middle *= data4middle; + dataorigin2.y *= dataorigin2.y; #endif /* defined(POOL_L2) */ - if((x1 + 1) >= int(POOL_SIZE + STRIDE_X)) + POOL_OP_float(sdata.y, sdata.y, dataorigin2.y); + + int x1 = STRIDE_X + 1; + for(; x1 <= (int(POOL_SIZE + STRIDE_X) - 8); x1 += 8) { - POOL_OP_float(sdata.y, sdata.y, data4middle.x); + vec4 data2; + vec4 data3; + LOAD4_fp16(data2, src, (tensor3D_offset_fp16(src, x1, y, 0) >> uint(2))); + LOAD4_fp16(data3, src, (tensor3D_offset_fp16(src, x1, y, 0) >> uint(2)) + uint(2)); + +#if defined(POOL_L2) + // Raise to power of 2 for L2 Pooling + data2 *= data2; + data3 *= data3; +#endif /* defined(POOL_L2) */ + + POOL_OP(vdata01, vdata01, data2); + POOL_OP(vdata11, vdata11, data3); } - else + + // Leftover + for(; x1 < int(POOL_SIZE + STRIDE_X); x1 = x1 + 2) { - float data4; - POOL_OP_float(data4, data4middle.x, data4middle.y); - POOL_OP_float(sdata.y, sdata.y, data4); + vec2 data4middle; + data4middle = load_and_unpack(src, (tensor3D_offset_fp16(src, x1, y, 0) >> uint(2))); +#if defined(POOL_L2) + // Raise to power of 2 for L2 Pooling + data4middle *= data4middle; +#endif /* defined(POOL_L2) */ + if((x1 + 1) >= int(POOL_SIZE + STRIDE_X)) + { + POOL_OP_float(sdata.y, sdata.y, data4middle.x); + } + else + { + float data4; + POOL_OP_float(data4, data4middle.x, data4middle.y); + POOL_OP_float(sdata.y, sdata.y, data4); + } } } } @@ -1414,14 +1490,20 @@ void main(void) #if defined(POOL_AVG) || defined(POOL_L2) { // Divide by pool region in case of average pooling - int start_x1 = int(gl_GlobalInvocationID.x) * STRIDE_X - PAD_X; - int start_y1 = int(gl_GlobalInvocationID.y) * STRIDE_Y - PAD_Y; - int end_x1 = int(min(start_x1 + POOL_SIZE, MAX_WIDTH)); - int end_y1 = int(min(start_y1 + POOL_SIZE, MAX_HEIGHT)); - int start_x2 = start_x1 + STRIDE_X; - int start_y2 = start_y1; - int end_x2 = int(min(start_x2 + POOL_SIZE, MAX_WIDTH)); - int end_y2 = int(min(start_y2 + POOL_SIZE, MAX_HEIGHT)); + int start_x1 = (2 * int(gl_GlobalInvocationID.x)) * STRIDE_X - PAD_X; + int start_y1 = int(gl_GlobalInvocationID.y) * STRIDE_Y - PAD_Y; + int end_x1 = int(min(start_x1 + POOL_SIZE, MAX_WIDTH)); + int end_y1 = int(min(start_y1 + POOL_SIZE, MAX_HEIGHT)); + int start_x2 = start_x1 + STRIDE_X; + int start_y2 = start_y1; + int end_x2 = int(min(start_x2 + POOL_SIZE, MAX_WIDTH)); + int end_y2 = int(min(start_y2 + POOL_SIZE, MAX_HEIGHT)); +#if defined(EXCLUDE_PADDING) + start_x1 = max(0, start_x1); + start_y1 = max(0, start_y1); + start_x2 = max(0, start_x2); + start_y2 = max(0, start_y2); +#endif /* defined(EXCLUDE_PADDING) */ vec2 res1; res1.x = float((end_y1 - start_y1) * (end_x1 - start_x1)); res1.y = float((end_y2 - start_y2) * (end_x2 - start_x2)); -- cgit v1.2.1