Merge pull request !6887 from fuzhiye/tmptags/v1.1.0
| @@ -17,6 +17,7 @@ | |||||
| #include <string.h> | #include <string.h> | ||||
| #include "nnacl/fp16/pack_fp16.h" | #include "nnacl/fp16/pack_fp16.h" | ||||
| #include "nnacl/fp16/winograd_transform_fp16.h" | #include "nnacl/fp16/winograd_transform_fp16.h" | ||||
| #include "nnacl/fp16/matmul_fp16.h" | |||||
| #ifdef __cplusplus | #ifdef __cplusplus | ||||
| extern "C" { | extern "C" { | ||||
| @@ -122,7 +123,8 @@ void IndirectGemmFp16_16x8_c8(float16_t *output, float16_t *input, float16_t *we | |||||
| // fp16 convolution common (im2col+gemm) | // fp16 convolution common (im2col+gemm) | ||||
| void ConvFp16(float16_t *input_data, float16_t *packed_input, float16_t *packed_weight, float16_t *bias_data, | void ConvFp16(float16_t *input_data, float16_t *packed_input, float16_t *packed_weight, float16_t *bias_data, | ||||
| float16_t *tmp_out_block, float16_t *output_data, int task_id, ConvParameter *conv_param) { | |||||
| float16_t *col_major_input, float16_t *output_data, int task_id, ConvParameter *conv_param) { | |||||
| const int tile_n = 16; | |||||
| int kernel_h = conv_param->kernel_h_; | int kernel_h = conv_param->kernel_h_; | ||||
| int kernel_w = conv_param->kernel_w_; | int kernel_w = conv_param->kernel_w_; | ||||
| int in_batch = conv_param->input_batch_; | int in_batch = conv_param->input_batch_; | ||||
| @@ -132,203 +134,29 @@ void ConvFp16(float16_t *input_data, float16_t *packed_input, float16_t *packed_ | |||||
| int out_h = conv_param->output_h_; | int out_h = conv_param->output_h_; | ||||
| int out_w = conv_param->output_w_; | int out_w = conv_param->output_w_; | ||||
| int out_channel = conv_param->output_channel_; | int out_channel = conv_param->output_channel_; | ||||
| bool relu = conv_param->act_type_ == ActType_Relu; | |||||
| bool relu6 = conv_param->act_type_ == ActType_Relu6; | |||||
| int thread_count = conv_param->thread_num_; | int thread_count = conv_param->thread_num_; | ||||
| const int tile_n = 16; | |||||
| int output_count = out_h * out_w; | int output_count = out_h * out_w; | ||||
| int output_tile_count = UP_DIV(output_count, tile_n); | int output_tile_count = UP_DIV(output_count, tile_n); | ||||
| int channel_block = UP_DIV(in_channel, C4NUM); | |||||
| int kernel_plane = kernel_h * kernel_w; | int kernel_plane = kernel_h * kernel_w; | ||||
| int unit_size = kernel_plane * channel_block * C4NUM; | |||||
| // we accumulate 4 channels per time for input blocks | |||||
| int ic4 = UP_DIV(in_channel, C4NUM); | |||||
| int conv_depth = kernel_h * kernel_w; | |||||
| // bytes from one output's i-th channel to the next output's i-th channel | |||||
| // we write 32 bytes per st1 instruction, after which the pointer in register will step 32B forward | |||||
| int deep = kernel_plane * in_channel; | |||||
| for (int b = 0; b < in_batch; b++) { | for (int b = 0; b < in_batch; b++) { | ||||
| int in_batch_offset = b * ic4 * C4NUM * in_h * in_w; | |||||
| int in_batch_offset = b * in_channel * in_h * in_w; | |||||
| int out_batch_offset = b * out_channel * out_h * out_w; | int out_batch_offset = b * out_channel * out_h * out_w; | ||||
| for (int thread_id = task_id; thread_id < output_tile_count; thread_id += thread_count) { | for (int thread_id = task_id; thread_id < output_tile_count; thread_id += thread_count) { | ||||
| int start_index = thread_id * tile_n; | int start_index = thread_id * tile_n; | ||||
| int real_cal_num = (output_count - start_index) < tile_n ? (output_count - start_index) : tile_n; | int real_cal_num = (output_count - start_index) < tile_n ? (output_count - start_index) : tile_n; | ||||
| float16_t *gemm_input = (float16_t *)(packed_input + task_id * unit_size * tile_n); | |||||
| float16_t *gemm_input = packed_input + task_id * deep * tile_n; | |||||
| float16_t *col_major_gemm_input = col_major_input + task_id * deep * tile_n; | |||||
| size_t packed_input_size = deep * tile_n * sizeof(float16_t); | |||||
| memset(gemm_input, 0, packed_input_size); | |||||
| memset(col_major_gemm_input, 0, packed_input_size); | |||||
| Im2ColPackUnitFp16(input_data + in_batch_offset, conv_param, gemm_input, real_cal_num, start_index); | Im2ColPackUnitFp16(input_data + in_batch_offset, conv_param, gemm_input, real_cal_num, start_index); | ||||
| int out_offset = thread_id * tile_n * out_channel + out_batch_offset; | int out_offset = thread_id * tile_n * out_channel + out_batch_offset; | ||||
| if (real_cal_num == tile_n) { | |||||
| float16_t *gemm_output = output_data + out_offset; | |||||
| IndirectGemmFp16_16x8(gemm_output, gemm_input, packed_weight, bias_data, conv_depth, ic4, out_channel, | |||||
| out_channel * sizeof(float16_t), 0, 0, relu, relu6); | |||||
| } else { | |||||
| // res part | |||||
| float16_t *tmp_out_ptr = tmp_out_block + task_id * tile_n * out_channel; | |||||
| IndirectGemmFp16_16x8(tmp_out_ptr, gemm_input, packed_weight, bias_data, conv_depth, ic4, out_channel, | |||||
| out_channel * sizeof(float16_t), 0, 0, relu, relu6); | |||||
| memcpy(output_data + out_offset, tmp_out_ptr, real_cal_num * out_channel * sizeof(float16_t)); | |||||
| } | |||||
| } | |||||
| } | |||||
| } | |||||
| // fp16 conv3x3 | |||||
| void Conv3x3Fp16(float16_t *input_data, float16_t *transed_weight, const float16_t *bias_data, float16_t *output_data, | |||||
| float16_t *tile_buffer, float16_t *block_unit_buffer, float16_t *tmp_dst_buffer, float16_t *tmp_out, | |||||
| int task_id, ConvParameter *conv_param) { | |||||
| int thread_count = conv_param->thread_num_; | |||||
| const int tile_num = 16; | |||||
| const int output_unit = 4; | |||||
| const int k_plane = 36; | |||||
| int ic8 = UP_DIV(conv_param->input_channel_, C8NUM); | |||||
| int ic4 = ic8 * 2; | |||||
| int oc8 = UP_DIV(conv_param->output_channel_, C8NUM); | |||||
| int out_w_block = UP_DIV(conv_param->output_w_, C4NUM); | |||||
| int out_h_block = UP_DIV(conv_param->output_h_, C4NUM); | |||||
| int output_count = out_w_block * out_h_block; | |||||
| int output_tile_count = UP_DIV(output_count, tile_num); | |||||
| int tile_buffer_offset = tile_num * k_plane * ic4 * C4NUM; | |||||
| int block_unit_buffer_offset = k_plane * C8NUM; | |||||
| int tmp_dst_buffer_offset = tile_num * k_plane * oc8 * C8NUM; | |||||
| int input_batch = conv_param->input_batch_; | |||||
| for (int batch = 0; batch < input_batch; batch++) { | |||||
| int tmp_out_batch_offset = batch * oc8 * C8NUM * out_w_block * out_h_block * output_unit * output_unit; | |||||
| for (int thread_id = task_id; thread_id < output_tile_count; thread_id += thread_count) { | |||||
| int start_index = thread_id * tile_num; | |||||
| int real_cal_num = (output_count - start_index) < tile_num ? (output_count - start_index) : tile_num; | |||||
| Conv3x3Fp16InputTransform(input_data, tile_buffer + task_id * tile_buffer_offset, | |||||
| block_unit_buffer + task_id * block_unit_buffer_offset, start_index, real_cal_num, | |||||
| out_w_block, conv_param); | |||||
| IndirectGemmFp16_16x8(tmp_dst_buffer + task_id * tmp_dst_buffer_offset, | |||||
| tile_buffer + task_id * tile_buffer_offset, transed_weight, NULL, 36, ic4, oc8 * C8NUM, | |||||
| oc8 * C8NUM * 36 * sizeof(float16_t), 1, 1, 0, 0); | |||||
| Conv3x3Fp16OutputTransform(tmp_dst_buffer + task_id * tmp_dst_buffer_offset, tmp_out + tmp_out_batch_offset, | |||||
| bias_data, start_index, real_cal_num, out_w_block, conv_param); | |||||
| } | |||||
| } | |||||
| } | |||||
| void UnPack3x3OutputFp16(const float16_t *src, float16_t *dst, int batch, int height, int width, int channel) { | |||||
| int out_w_block = UP_DIV(width, C4NUM); | |||||
| int out_h_block = UP_DIV(height, C4NUM); | |||||
| int oc8 = UP_DIV(channel, C8NUM); | |||||
| for (int b = 0; b < batch; b++) { | |||||
| int tmp_out_batch_offset = b * oc8 * C8NUM * out_w_block * out_h_block * C4NUM * C4NUM; | |||||
| int ro_batch_size = b * channel * height * width; | |||||
| const float16_t *batch_tmp_out = src + tmp_out_batch_offset; | |||||
| float16_t *batch_out = dst + ro_batch_size; | |||||
| for (int h = 0; h < height; h++) { | |||||
| int src_h_offset = h * out_w_block * C4NUM * C8NUM; | |||||
| const int dst_h_offset = h * width * channel; | |||||
| for (int w = 0; w < width; w++) { | |||||
| int src_w_offset = src_h_offset + w * C8NUM; | |||||
| int dst_w_offset = dst_h_offset + w * channel; | |||||
| for (int c = 0; c < oc8 - 1; ++c) { | |||||
| int src_offset = c * C8NUM * out_w_block * out_h_block * C4NUM * C4NUM + src_w_offset; | |||||
| int dst_offset = dst_w_offset + c * C8NUM; | |||||
| vst1q_f16(batch_out + dst_offset, vld1q_f16(batch_tmp_out + src_offset)); | |||||
| } | |||||
| int c_res = channel - (oc8 - 1) * C8NUM; | |||||
| int src_c_res_offset = src_w_offset + (oc8 - 1) * C8NUM * out_w_block * out_h_block * C4NUM * C4NUM; | |||||
| int dst_c_res_offset = dst_w_offset + (oc8 - 1) * C8NUM; | |||||
| for (int c = 0; c < c_res; c++) { | |||||
| int src_offset = src_c_res_offset + c; | |||||
| int dst_offset = dst_c_res_offset + c; | |||||
| (batch_out + dst_offset)[0] = (batch_tmp_out + src_offset)[0]; | |||||
| } | |||||
| } | |||||
| } | |||||
| } | |||||
| } | |||||
| void UnPack3x3ReluOutputFp16(const float16_t *src, float16_t *dst, int batch, int height, int width, int channel) { | |||||
| int out_w_block = UP_DIV(width, C4NUM); | |||||
| int out_h_block = UP_DIV(height, C4NUM); | |||||
| int oc8 = UP_DIV(channel, C8NUM); | |||||
| for (int b = 0; b < batch; b++) { | |||||
| int tmp_out_batch_offset = b * oc8 * C8NUM * out_w_block * out_h_block * C4NUM * C4NUM; | |||||
| int ro_batch_size = b * channel * height * width; | |||||
| const float16_t *batch_tmp_out = src + tmp_out_batch_offset; | |||||
| float16_t *batch_out = dst + ro_batch_size; | |||||
| for (int h = 0; h < height; h++) { | |||||
| int src_h_offset = h * out_w_block * C4NUM * C8NUM; | |||||
| const int dst_h_offset = h * width * channel; | |||||
| for (int w = 0; w < width; w++) { | |||||
| int src_w_offset = src_h_offset + w * C8NUM; | |||||
| int dst_w_offset = dst_h_offset + w * channel; | |||||
| for (int c = 0; c < oc8 - 1; ++c) { | |||||
| int src_offset = c * C8NUM * out_w_block * out_h_block * C4NUM * C4NUM + src_w_offset; | |||||
| int dst_offset = dst_w_offset + c * C8NUM; | |||||
| float16x8_t input_ptr = vld1q_f16(batch_tmp_out + src_offset); | |||||
| float16x8_t zero = vdupq_n_f16(0); | |||||
| input_ptr = vmaxq_f16(zero, input_ptr); | |||||
| vst1q_f16(batch_out + dst_offset, input_ptr); | |||||
| } | |||||
| int c_res = channel - (oc8 - 1) * C8NUM; | |||||
| int src_c_res_offset = src_w_offset + (oc8 - 1) * C8NUM * out_w_block * out_h_block * C4NUM * C4NUM; | |||||
| int dst_c_res_offset = dst_w_offset + (oc8 - 1) * C8NUM; | |||||
| for (int c = 0; c < c_res; c++) { | |||||
| int src_offset = src_c_res_offset + c; | |||||
| int dst_offset = dst_c_res_offset + c; | |||||
| float16_t input_data = (batch_tmp_out + src_offset)[0]; | |||||
| input_data = input_data < 0 ? 0 : input_data; | |||||
| (batch_out + dst_offset)[0] = input_data; | |||||
| } | |||||
| } | |||||
| } | |||||
| } | |||||
| } | |||||
| void UnPack3x3Relu6OutputFp16(const float16_t *src, float16_t *dst, int batch, int height, int width, int channel) { | |||||
| int out_w_block = UP_DIV(width, C4NUM); | |||||
| int out_h_block = UP_DIV(height, C4NUM); | |||||
| int oc8 = UP_DIV(channel, C8NUM); | |||||
| for (int b = 0; b < batch; b++) { | |||||
| int tmp_out_batch_offset = b * oc8 * C8NUM * out_w_block * out_h_block * C4NUM * C4NUM; | |||||
| int ro_batch_size = b * channel * height * width; | |||||
| const float16_t *batch_tmp_out = src + tmp_out_batch_offset; | |||||
| float16_t *batch_out = dst + ro_batch_size; | |||||
| for (int h = 0; h < height; h++) { | |||||
| int src_h_offset = h * out_w_block * C4NUM * C8NUM; | |||||
| const int dst_h_offset = h * width * channel; | |||||
| for (int w = 0; w < width; w++) { | |||||
| int src_w_offset = src_h_offset + w * C8NUM; | |||||
| int dst_w_offset = dst_h_offset + w * channel; | |||||
| for (int c = 0; c < oc8 - 1; ++c) { | |||||
| int src_offset = c * C8NUM * out_w_block * out_h_block * C4NUM * C4NUM + src_w_offset; | |||||
| int dst_offset = dst_w_offset + c * C8NUM; | |||||
| float16x8_t input_ptr = vld1q_f16(batch_tmp_out + src_offset); | |||||
| float16x8_t zero = vdupq_n_f16(0); | |||||
| float16x8_t six = vdupq_n_f16(6); | |||||
| input_ptr = vmaxq_f16(zero, input_ptr); | |||||
| input_ptr = vminq_f16(six, input_ptr); | |||||
| vst1q_f16(batch_out + dst_offset, input_ptr); | |||||
| } | |||||
| int c_res = channel - (oc8 - 1) * C8NUM; | |||||
| int src_c_res_offset = src_w_offset + (oc8 - 1) * C8NUM * out_w_block * out_h_block * C4NUM * C4NUM; | |||||
| int dst_c_res_offset = dst_w_offset + (oc8 - 1) * C8NUM; | |||||
| for (int c = 0; c < c_res; c++) { | |||||
| int src_offset = src_c_res_offset + c; | |||||
| int dst_offset = dst_c_res_offset + c; | |||||
| float16_t input_data = (batch_tmp_out + src_offset)[0]; | |||||
| input_data = input_data < 0 ? 0 : input_data; | |||||
| input_data = input_data > 6 ? 6 : input_data; | |||||
| (batch_out + dst_offset)[0] = input_data; | |||||
| } | |||||
| } | |||||
| RowMajor2Col16MajorFp16Opt(gemm_input, col_major_gemm_input, tile_n, deep); | |||||
| MatMulFp16(col_major_gemm_input, packed_weight, output_data + out_offset, bias_data, conv_param->act_type_, deep, | |||||
| real_cal_num, out_channel, out_channel, OutType_Nhwc); | |||||
| } | } | ||||
| } | } | ||||
| } | } | ||||
| @@ -43,18 +43,7 @@ extern "C" { | |||||
| // fp16 convolution common (im2col+gemm) | // fp16 convolution common (im2col+gemm) | ||||
| void ConvFp16(float16_t *input_data, float16_t *packed_input, float16_t *packed_weight, float16_t *bias_data, | void ConvFp16(float16_t *input_data, float16_t *packed_input, float16_t *packed_weight, float16_t *bias_data, | ||||
| float16_t *tmp_out_block, float16_t *output_data, int task_id, ConvParameter *conv_param); | |||||
| // fp16 conv3x3 | |||||
| void Conv3x3Fp16(float16_t *input_data, float16_t *transed_weight, const float16_t *bias_data, float16_t *output_data, | |||||
| float16_t *tile_buffer, float16_t *block_unit_buffer, float16_t *tmp_dst_buffer, float16_t *tmp_out, | |||||
| int task_id, ConvParameter *conv_param); | |||||
| void UnPack3x3OutputFp16(const float16_t *src, float16_t *dst, int batch, int height, int width, int channel); | |||||
| void UnPack3x3ReluOutputFp16(const float16_t *src, float16_t *dst, int batch, int height, int width, int channel); | |||||
| void UnPack3x3Relu6OutputFp16(const float16_t *src, float16_t *dst, int batch, int height, int width, int channel); | |||||
| float16_t *col_major_input, float16_t *output_data, int task_id, ConvParameter *conv_param); | |||||
| // fp16 convolution winograd | // fp16 convolution winograd | ||||
| void ConvWinogardFp16(float16_t *input_data, float16_t *trans_weight, const float16_t *bias_data, | void ConvWinogardFp16(float16_t *input_data, float16_t *trans_weight, const float16_t *bias_data, | ||||
| @@ -23,6 +23,7 @@ void Im2ColPackUnitFp16(float16_t *input_data, ConvParameter *conv_param, float1 | |||||
| // input format : nhwc | // input format : nhwc | ||||
| int kernel_h = conv_param->kernel_h_; | int kernel_h = conv_param->kernel_h_; | ||||
| int kernel_w = conv_param->kernel_w_; | int kernel_w = conv_param->kernel_w_; | ||||
| int kernel_plane = kernel_h * kernel_w; | |||||
| int stride_h = conv_param->stride_h_; | int stride_h = conv_param->stride_h_; | ||||
| int stride_w = conv_param->stride_w_; | int stride_w = conv_param->stride_w_; | ||||
| int pad_h = conv_param->pad_u_; | int pad_h = conv_param->pad_u_; | ||||
| @@ -33,9 +34,6 @@ void Im2ColPackUnitFp16(float16_t *input_data, ConvParameter *conv_param, float1 | |||||
| int in_h = conv_param->input_h_; | int in_h = conv_param->input_h_; | ||||
| int in_w = conv_param->input_w_; | int in_w = conv_param->input_w_; | ||||
| int out_w = conv_param->output_w_; | int out_w = conv_param->output_w_; | ||||
| int ic4 = UP_DIV(in_channel, 4); | |||||
| int ic4_minus = in_channel / 4; | |||||
| memset(packed_input, 0, kernel_w * kernel_h * ic4 * C4NUM * 16 * sizeof(float16_t)); | |||||
| for (int i = 0; i < real_cal_num; i++) { | for (int i = 0; i < real_cal_num; i++) { | ||||
| int block_start = block_index + i; | int block_start = block_index + i; | ||||
| @@ -46,74 +44,25 @@ void Im2ColPackUnitFp16(float16_t *input_data, ConvParameter *conv_param, float1 | |||||
| int kh_e = MSMIN(kernel_h, UP_DIV(in_h - input_h, dilation_h)); | int kh_e = MSMIN(kernel_h, UP_DIV(in_h - input_h, dilation_h)); | ||||
| int kw_s = MSMAX(0, UP_DIV(-input_w, dilation_w)); | int kw_s = MSMAX(0, UP_DIV(-input_w, dilation_w)); | ||||
| int kw_e = MSMIN(kernel_w, UP_DIV(in_w - input_w, dilation_w)); | int kw_e = MSMIN(kernel_w, UP_DIV(in_w - input_w, dilation_w)); | ||||
| for (int j = kh_s; j < kh_e; j++) { | |||||
| int input_y_stride = j * dilation_h * in_w * in_channel + input_stride; | |||||
| for (int n = kw_s; n < kw_e; n++) { | |||||
| int input_x_stride = input_y_stride + n * dilation_w * in_channel; | |||||
| int input_plane_offset = (j * kernel_w + n) * 16 * C4NUM * ic4 + i * C4NUM; | |||||
| for (int m = 0; m < ic4_minus; m++) { | |||||
| int channel_block_stride = input_x_stride + m * C4NUM; | |||||
| int channel_block_offset = input_plane_offset + m * 16 * C4NUM; | |||||
| #ifdef ENABLE_ARM64 | |||||
| vst1_f16(packed_input + channel_block_offset, vld1_f16(input_data + channel_block_stride)); | |||||
| #else | |||||
| for (int l = 0; l < C4NUM; ++l) { | |||||
| (packed_input + channel_block_offset)[l] = (input_data + channel_block_stride)[l]; | |||||
| } | |||||
| #endif | |||||
| } // channel_block loop | |||||
| int ic_res = in_channel - ic4_minus * C4NUM; | |||||
| for (int l = 0; l < ic_res; ++l) { | |||||
| int channel_block_stride = input_x_stride + ic4_minus * C4NUM + l; | |||||
| int channel_block_offset = input_plane_offset + ic4_minus * 16 * C4NUM + l; | |||||
| packed_input[channel_block_offset] = input_data[channel_block_stride]; | |||||
| } | |||||
| } // kernel_w loop | |||||
| } // kernel_h loop | |||||
| } // tile num loop | |||||
| } | |||||
| void PackWeightFp16(float16_t *weight_data, ConvParameter *conv_param, float16_t *packed_weight) { | |||||
| // original weight format : ohwi | |||||
| const int tile_num = 8; | |||||
| const int inchannel_block = 4; | |||||
| int kernel_h = conv_param->kernel_h_; | |||||
| int kernel_w = conv_param->kernel_w_; | |||||
| int in_channel = conv_param->input_channel_; | |||||
| int out_channel = conv_param->output_channel_; | |||||
| int kernel_block = UP_DIV(out_channel, tile_num); | |||||
| int channel_block = UP_DIV(in_channel, inchannel_block); | |||||
| int kernel_plane = kernel_h * kernel_w; | |||||
| int pack_weight_size = kernel_block * channel_block * tile_num * inchannel_block * kernel_plane; | |||||
| int unit_size = tile_num * inchannel_block; | |||||
| int block_size = pack_weight_size / kernel_block; | |||||
| for (int m = 0; m < kernel_plane; m++) { | |||||
| int kernel_plane_stride = m * in_channel; | |||||
| int packed_kernel_plane_stride = m * unit_size * channel_block; | |||||
| for (int i = 0; i < channel_block; i++) { | |||||
| int channel_block_stride = kernel_plane_stride + i * inchannel_block; | |||||
| int packed_channel_block_size = packed_kernel_plane_stride + i * unit_size; | |||||
| int ic_remainder = in_channel - i * inchannel_block; | |||||
| int real_ic_num = ic_remainder < inchannel_block ? ic_remainder : inchannel_block; | |||||
| for (int h = 0; h < real_ic_num; h++) { | |||||
| int block_stride = channel_block_stride + h; | |||||
| int packed_block_stride = packed_channel_block_size + h * tile_num; | |||||
| for (int j = 0; j < kernel_block; j++) { | |||||
| int kernel_block_stride = block_stride + j * tile_num * kernel_plane * in_channel; | |||||
| int packed_kernel_block_size = packed_block_stride + j * block_size; | |||||
| int oc_remainder = out_channel - j * tile_num; | |||||
| int real_oc_num = oc_remainder < tile_num ? oc_remainder : tile_num; | |||||
| for (int k = 0; k < real_oc_num; k++) { | |||||
| float16_t *origin_data_ptr = weight_data + kernel_block_stride + k * kernel_plane * in_channel; | |||||
| float16_t *packed_data_ptr = packed_weight + packed_kernel_block_size + k; | |||||
| *packed_data_ptr = *origin_data_ptr; | |||||
| } | |||||
| } // kernel block loop | |||||
| } // inchannel block loop | |||||
| } // channel block loop | |||||
| } // kernel plane loop | |||||
| if (dilation_h == 1 && dilation_w == 1) { | |||||
| for (int j = kh_s; j < kh_e; j++) { | |||||
| int input_y_stride = j * in_w * in_channel + input_stride; | |||||
| int input_x_stride = input_y_stride + kw_s * in_channel; | |||||
| int input_plane_offset = (j * kernel_w + kw_s) * in_channel + i * in_channel * kernel_plane; | |||||
| memcpy(packed_input + input_plane_offset, input_data + input_x_stride, | |||||
| (kw_e - kw_s) * in_channel * sizeof(float16_t)); | |||||
| } // kernel_h loop | |||||
| } else { | |||||
| for (int j = kh_s; j < kh_e; j++) { | |||||
| int input_y_stride = j * dilation_h * in_w * in_channel + input_stride; | |||||
| for (int n = kw_s; n < kw_e; n++) { | |||||
| int input_x_stride = input_y_stride + n * dilation_w * in_channel; | |||||
| int input_plane_offset = (j * kernel_w + n) * in_channel + i * in_channel * kernel_plane; | |||||
| memcpy(packed_input + input_plane_offset, input_data + input_x_stride, in_channel * sizeof(float16_t)); | |||||
| } // kernel_w loop | |||||
| } // kernel_h loop | |||||
| } | |||||
| } // tile num loop | |||||
| } | } | ||||
| void PackWeightToC8Fp16(const float16_t *origin_weight_data, float16_t *packed_weight_data, ConvParameter *conv_param) { | void PackWeightToC8Fp16(const float16_t *origin_weight_data, float16_t *packed_weight_data, ConvParameter *conv_param) { | ||||
| @@ -29,8 +29,6 @@ extern "C" { | |||||
| void Im2ColPackUnitFp16(float16_t *input_data, ConvParameter *conv_param, float16_t *packed_input, int real_cal_num, | void Im2ColPackUnitFp16(float16_t *input_data, ConvParameter *conv_param, float16_t *packed_input, int real_cal_num, | ||||
| int block_index); | int block_index); | ||||
| void PackWeightFp16(float16_t *weight_data, ConvParameter *conv_param, float16_t *packed_weight); | |||||
| void PackWeightToC8Fp16(const float16_t *origin_weight_data, float16_t *packed_weight_data, ConvParameter *conv_param); | void PackWeightToC8Fp16(const float16_t *origin_weight_data, float16_t *packed_weight_data, ConvParameter *conv_param); | ||||
| void PackWeightToC4Fp16(const float16_t *origin_weight_data, float16_t *packed_weight_data, ConvParameter *conv_param); | void PackWeightToC4Fp16(const float16_t *origin_weight_data, float16_t *packed_weight_data, ConvParameter *conv_param); | ||||
| @@ -1,261 +0,0 @@ | |||||
| /** | |||||
| * Copyright 2020 Huawei Technologies Co., Ltd | |||||
| * | |||||
| * Licensed under the Apache License, Version 2.0 (the "License"); | |||||
| * you may not use this file except in compliance with the License. | |||||
| * You may obtain a copy of the License at | |||||
| * | |||||
| * http://www.apache.org/licenses/LICENSE-2.0 | |||||
| * | |||||
| * Unless required by applicable law or agreed to in writing, software | |||||
| * distributed under the License is distributed on an "AS IS" BASIS, | |||||
| * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | |||||
| * See the License for the specific language governing permissions and | |||||
| * limitations under the License. | |||||
| */ | |||||
| #include "src/runtime/kernel/arm/fp16/convolution_3x3_fp16.h" | |||||
| #include "nnacl/fp16/conv_fp16.h" | |||||
| #include "nnacl/fp16/cast_fp16.h" | |||||
| #include "nnacl/fp16/winograd_transform_fp16.h" | |||||
| #include "nnacl/fp16/pack_fp16.h" | |||||
| #include "src/runtime/kernel/arm/fp16/layout_transform_fp16.h" | |||||
| #include "schema/model_generated.h" | |||||
| #include "src/kernel_registry.h" | |||||
| #include "include/errorcode.h" | |||||
| #include "src/runtime/runtime_api.h" | |||||
| using mindspore::kernel::KERNEL_ARCH::kCPU; | |||||
| using mindspore::lite::KernelRegistrar; | |||||
| using mindspore::lite::RET_ERROR; | |||||
| using mindspore::lite::RET_OK; | |||||
| using mindspore::schema::PrimitiveType_Conv2D; | |||||
| namespace mindspore::kernel { | |||||
| void ProcessFilterFp16(float16_t *origin_weight, float16_t *dst_weight, ConvParameter *conv_param) { | |||||
| auto input_channel = conv_param->input_channel_; | |||||
| auto output_channel = conv_param->output_channel_; | |||||
| auto kernel_plane = conv_param->kernel_w_ * conv_param->kernel_h_; | |||||
| int iC8 = UP_DIV(input_channel, C8NUM); | |||||
| int oC8 = UP_DIV(output_channel, C8NUM); | |||||
| size_t tmp_size = oC8 * C8NUM * iC8 * C8NUM * kernel_plane * sizeof(float16_t); | |||||
| auto tmp_addr = reinterpret_cast<float16_t *>(malloc(tmp_size)); | |||||
| if (tmp_addr == nullptr) { | |||||
| MS_LOG(ERROR) << "malloc tmp_addr failed."; | |||||
| return; | |||||
| } | |||||
| memset(tmp_addr, 0, tmp_size); | |||||
| PackWeightToC4Fp16(origin_weight, tmp_addr, conv_param); | |||||
| Conv3x3Fp16FilterTransform(tmp_addr, dst_weight, iC8 * 2, output_channel, kernel_plane); | |||||
| free(tmp_addr); | |||||
| } | |||||
| int Convolution3x3FP16CPUKernel::InitWeightBias() { | |||||
| auto filter_tensor = in_tensors_.at(kWeightIndex); | |||||
| auto input_channel = filter_tensor->Channel(); | |||||
| auto output_channel = filter_tensor->Batch(); | |||||
| conv_param_->input_channel_ = input_channel; | |||||
| conv_param_->output_channel_ = output_channel; | |||||
| int iC8 = UP_DIV(input_channel, C8NUM); | |||||
| int oC8 = UP_DIV(output_channel, C8NUM); | |||||
| size_t transformed_size = iC8 * C8NUM * oC8 * C8NUM * 36 * sizeof(float16_t); | |||||
| transformed_filter_addr_ = reinterpret_cast<float16_t *>(malloc(transformed_size)); | |||||
| if (transformed_filter_addr_ == nullptr) { | |||||
| MS_LOG(ERROR) << "malloc transformed_filter_addr_ failed."; | |||||
| return RET_ERROR; | |||||
| } | |||||
| memset(transformed_filter_addr_, 0, transformed_size); | |||||
| auto ret = ConvolutionBaseFP16CPUKernel::GetExecuteFilter(); | |||||
| if (ret != RET_OK) { | |||||
| MS_LOG(ERROR) << "Get Execute filter failed."; | |||||
| return ret; | |||||
| } | |||||
| ProcessFilterFp16(execute_weight_, transformed_filter_addr_, conv_param_); | |||||
| size_t new_bias_size = oC8 * C8NUM * sizeof(float16_t); | |||||
| bias_data_ = malloc(new_bias_size); | |||||
| if (bias_data_ == nullptr) { | |||||
| MS_LOG(ERROR) << "malloc bias_data_ failed."; | |||||
| return RET_ERROR; | |||||
| } | |||||
| memset(bias_data_, 0, new_bias_size); | |||||
| auto fp16_bias_data = reinterpret_cast<float16_t *>(bias_data_); | |||||
| if (in_tensors_.size() == kInputSize2) { | |||||
| auto ori_bias_addr = reinterpret_cast<float *>(in_tensors_.at(kBiasIndex)->MutableData()); | |||||
| for (int i = 0; i < output_channel; ++i) { | |||||
| fp16_bias_data[i] = (float16_t)ori_bias_addr[i]; | |||||
| } | |||||
| } else { | |||||
| MS_ASSERT(inputs_.size() == kInputSize1); | |||||
| } | |||||
| return RET_OK; | |||||
| } | |||||
| int Convolution3x3FP16CPUKernel::InitTmpBuffer() { | |||||
| const int tile_num = 16; | |||||
| const int k_plane = 36; | |||||
| int oC8 = UP_DIV(conv_param_->output_channel_, C8NUM); | |||||
| int iC8 = UP_DIV(conv_param_->input_channel_, C8NUM); | |||||
| MS_ASSERT(ctx_->allocator != nullptr); | |||||
| size_t nhwc8_input_size = | |||||
| iC8 * C8NUM * conv_param_->input_batch_ * conv_param_->input_h_ * conv_param_->input_w_ * sizeof(float16_t); | |||||
| nhwc4_input_ = ctx_->allocator->Malloc(nhwc8_input_size); | |||||
| if (nhwc4_input_ == nullptr) { | |||||
| MS_LOG(ERROR) << "malloc nhwc4_input_ failed."; | |||||
| return RET_ERROR; | |||||
| } | |||||
| size_t tile_buffer_size = thread_count_ * tile_num * k_plane * iC8 * C8NUM * sizeof(float16_t); | |||||
| tile_buffer_ = reinterpret_cast<float16_t *>(ctx_->allocator->Malloc(tile_buffer_size)); | |||||
| if (tile_buffer_ == nullptr) { | |||||
| MS_LOG(ERROR) << "malloc tile_buffer_ failed."; | |||||
| return RET_ERROR; | |||||
| } | |||||
| size_t block_unit_buffer_size = thread_count_ * k_plane * C8NUM * sizeof(float16_t); | |||||
| block_unit_buffer_ = reinterpret_cast<float16_t *>(ctx_->allocator->Malloc(block_unit_buffer_size)); | |||||
| if (block_unit_buffer_ == nullptr) { | |||||
| MS_LOG(ERROR) << "malloc block_unit_buffer_ failed."; | |||||
| return RET_ERROR; | |||||
| } | |||||
| size_t tmp_dst_buffer_size = thread_count_ * tile_num * k_plane * oC8 * C8NUM * sizeof(float16_t); | |||||
| tmp_dst_buffer_ = reinterpret_cast<float16_t *>(ctx_->allocator->Malloc(tmp_dst_buffer_size)); | |||||
| if (tmp_dst_buffer_ == nullptr) { | |||||
| MS_LOG(ERROR) << "malloc tmp_dst_buffer_ failed."; | |||||
| return RET_ERROR; | |||||
| } | |||||
| int new_out_plane = UP_DIV(conv_param_->output_h_, C4NUM) * UP_DIV(conv_param_->output_w_, C4NUM) * C4NUM * C4NUM; | |||||
| size_t tmp_out_size = oC8 * C8NUM * conv_param_->output_batch_ * new_out_plane * sizeof(float16_t); | |||||
| tmp_out_ = reinterpret_cast<float16_t *>(ctx_->allocator->Malloc(tmp_out_size)); | |||||
| if (tmp_out_ == nullptr) { | |||||
| MS_LOG(ERROR) << "malloc tmp_out_ failed."; | |||||
| return RET_ERROR; | |||||
| } | |||||
| return RET_OK; | |||||
| } | |||||
| void Convolution3x3FP16CPUKernel::ConfigInputOutput() { | |||||
| auto input_tensor = in_tensors_.at(kInputIndex); | |||||
| auto input_format = input_tensor->GetFormat(); | |||||
| schema::Format execute_format = schema::Format::Format_NHWC4; | |||||
| convert_func_ = LayoutTransformFp16(input_format, execute_format); | |||||
| if (convert_func_ == nullptr) { | |||||
| MS_LOG(ERROR) << "layout convert func is nullptr."; | |||||
| return; | |||||
| } | |||||
| } | |||||
| int Convolution3x3FP16CPUKernel::Init() { | |||||
| auto ret = InitWeightBias(); | |||||
| if (ret != RET_OK) { | |||||
| MS_LOG(ERROR) << "Init weight bias failed."; | |||||
| return RET_ERROR; | |||||
| } | |||||
| if (!InferShapeDone()) { | |||||
| return RET_OK; | |||||
| } | |||||
| return ReSize(); | |||||
| } | |||||
| int Convolution3x3FP16CPUKernel::ReSize() { | |||||
| auto ret = ConvolutionBaseCPUKernel::CheckResizeValid(); | |||||
| if (ret != RET_OK) { | |||||
| MS_LOG(ERROR) << "Resize is invalid."; | |||||
| return ret; | |||||
| } | |||||
| ret = ConvolutionBaseCPUKernel::Init(); | |||||
| if (ret != RET_OK) { | |||||
| MS_LOG(ERROR) << "ConvolutionBase init failed."; | |||||
| return ret; | |||||
| } | |||||
| return RET_OK; | |||||
| } | |||||
| int Convolution3x3FP16CPUKernel::RunImpl(int task_id) { | |||||
| Conv3x3Fp16(reinterpret_cast<float16_t *>(nhwc4_input_), transformed_filter_addr_, | |||||
| reinterpret_cast<float16_t *>(bias_data_), execute_output_, tile_buffer_, block_unit_buffer_, | |||||
| tmp_dst_buffer_, tmp_out_, task_id, conv_param_); | |||||
| return RET_OK; | |||||
| } | |||||
| static int Convolution3x3Fp16Impl(void *cdata, int task_id) { | |||||
| auto conv = reinterpret_cast<Convolution3x3FP16CPUKernel *>(cdata); | |||||
| auto error_code = conv->RunImpl(task_id); | |||||
| if (error_code != RET_OK) { | |||||
| MS_LOG(ERROR) << "Convolution3x3 Fp16 Run error task_id[" << task_id << "] error_code[" << error_code << "]"; | |||||
| return RET_ERROR; | |||||
| } | |||||
| return RET_OK; | |||||
| } | |||||
| int Convolution3x3FP16CPUKernel::PostProcess() { | |||||
| auto act_type = conv_param_->act_type_; | |||||
| switch (act_type) { | |||||
| case ActType_No: | |||||
| UnPack3x3OutputFp16(tmp_out_, execute_output_, conv_param_->output_batch_, conv_param_->output_h_, | |||||
| conv_param_->output_w_, conv_param_->output_channel_); | |||||
| break; | |||||
| case ActType_Relu: | |||||
| UnPack3x3ReluOutputFp16(tmp_out_, execute_output_, conv_param_->output_batch_, conv_param_->output_h_, | |||||
| conv_param_->output_w_, conv_param_->output_channel_); | |||||
| break; | |||||
| case ActType_Relu6: | |||||
| UnPack3x3Relu6OutputFp16(tmp_out_, execute_output_, conv_param_->output_batch_, conv_param_->output_h_, | |||||
| conv_param_->output_w_, conv_param_->output_channel_); | |||||
| break; | |||||
| default: | |||||
| MS_LOG(ERROR) << "Unsupport activation type."; | |||||
| return RET_ERROR; | |||||
| } | |||||
| return RET_OK; | |||||
| } | |||||
| int Convolution3x3FP16CPUKernel::Run() { | |||||
| auto ret = Prepare(); | |||||
| if (ret != RET_OK) { | |||||
| MS_LOG(ERROR) << "Prepare failed."; | |||||
| return RET_ERROR; | |||||
| } | |||||
| ret = ConvolutionBaseFP16CPUKernel::GetExecuteTensor(); | |||||
| if (ret != RET_OK) { | |||||
| MS_LOG(ERROR) << "Get execute tensor failed."; | |||||
| return ret; | |||||
| } | |||||
| ret = InitTmpBuffer(); | |||||
| if (ret != RET_OK) { | |||||
| MS_LOG(ERROR) << "Init tmp buffer failed."; | |||||
| return RET_ERROR; | |||||
| } | |||||
| int in_batch = conv_param_->input_batch_; | |||||
| int in_h = conv_param_->input_h_; | |||||
| int in_w = conv_param_->input_w_; | |||||
| int in_channel = conv_param_->input_channel_; | |||||
| PackNHWCToNHWC8Fp16(reinterpret_cast<void *>(execute_input_), nhwc4_input_, in_batch, in_h * in_w, in_channel); | |||||
| int error_code = ParallelLaunch(this->context_->thread_pool_, Convolution3x3Fp16Impl, this, thread_count_); | |||||
| if (error_code != RET_OK) { | |||||
| MS_LOG(ERROR) << "conv3x3 fp16 error error_code[" << error_code << "]"; | |||||
| FreeTmpBuffer(); | |||||
| return RET_ERROR; | |||||
| } | |||||
| ret = PostProcess(); | |||||
| if (ret != RET_OK) { | |||||
| MS_LOG(ERROR) << "Post process failed."; | |||||
| return ret; | |||||
| } | |||||
| ConvolutionBaseFP16CPUKernel::IfCastOutput(); | |||||
| ConvolutionBaseFP16CPUKernel::FreeTmpBuffer(); | |||||
| FreeTmpBuffer(); | |||||
| return RET_OK; | |||||
| } | |||||
| } // namespace mindspore::kernel | |||||
| @@ -1,85 +0,0 @@ | |||||
| /** | |||||
| * Copyright 2020 Huawei Technologies Co., Ltd | |||||
| * | |||||
| * Licensed under the Apache License, Version 2.0 (the "License"); | |||||
| * you may not use this file except in compliance with the License. | |||||
| * You may obtain a copy of the License at | |||||
| * | |||||
| * http://www.apache.org/licenses/LICENSE-2.0 | |||||
| * | |||||
| * Unless required by applicable law or agreed to in writing, software | |||||
| * distributed under the License is distributed on an "AS IS" BASIS, | |||||
| * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | |||||
| * See the License for the specific language governing permissions and | |||||
| * limitations under the License. | |||||
| */ | |||||
| #ifndef MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_FP16_CONVOLUTION_3x3_FP16_H_ | |||||
| #define MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_FP16_CONVOLUTION_3x3_FP16_H_ | |||||
| #include <arm_neon.h> | |||||
| #include <vector> | |||||
| #include "src/lite_kernel.h" | |||||
| #include "src/runtime/kernel/arm/fp16/convolution_base_fp16.h" | |||||
| #include "nnacl/optimized_kernel.h" | |||||
| namespace mindspore::kernel { | |||||
| class Convolution3x3FP16CPUKernel : public ConvolutionBaseFP16CPUKernel { | |||||
| public: | |||||
| Convolution3x3FP16CPUKernel(OpParameter *parameter, const std::vector<lite::Tensor *> &inputs, | |||||
| const std::vector<lite::Tensor *> &outputs, const InnerContext *ctx, | |||||
| const mindspore::lite::PrimitiveC *primitive) | |||||
| : ConvolutionBaseFP16CPUKernel(parameter, inputs, outputs, ctx, primitive) {} | |||||
| ~Convolution3x3FP16CPUKernel() override { | |||||
| if (fp16_weight_ != nullptr) { | |||||
| free(fp16_weight_); | |||||
| fp16_weight_ = nullptr; | |||||
| } | |||||
| if (transformed_filter_addr_ != nullptr) { | |||||
| free(transformed_filter_addr_); | |||||
| transformed_filter_addr_ = nullptr; | |||||
| } | |||||
| } | |||||
| int Init() override; | |||||
| int ReSize() override; | |||||
| int Run() override; | |||||
| int RunImpl(int task_id); | |||||
| int InitWeightBias(); | |||||
| int InitTmpBuffer(); | |||||
| void ConfigInputOutput(); | |||||
| int PostProcess(); | |||||
| private: | |||||
| void FreeTmpBuffer() { | |||||
| if (nhwc4_input_ != nullptr) { | |||||
| ctx_->allocator->Free(nhwc4_input_); | |||||
| nhwc4_input_ = nullptr; | |||||
| } | |||||
| if (tile_buffer_ != nullptr) { | |||||
| ctx_->allocator->Free(tile_buffer_); | |||||
| tile_buffer_ = nullptr; | |||||
| } | |||||
| if (block_unit_buffer_ != nullptr) { | |||||
| ctx_->allocator->Free(block_unit_buffer_); | |||||
| block_unit_buffer_ = nullptr; | |||||
| } | |||||
| if (tmp_dst_buffer_ != nullptr) { | |||||
| ctx_->allocator->Free(tmp_dst_buffer_); | |||||
| tmp_dst_buffer_ = nullptr; | |||||
| } | |||||
| if (tmp_out_ != nullptr) { | |||||
| ctx_->allocator->Free(tmp_out_); | |||||
| tmp_out_ = nullptr; | |||||
| } | |||||
| } | |||||
| float16_t *transformed_filter_addr_ = nullptr; | |||||
| float16_t *tile_buffer_ = nullptr; | |||||
| float16_t *block_unit_buffer_ = nullptr; | |||||
| float16_t *tmp_dst_buffer_ = nullptr; | |||||
| float16_t *tmp_out_ = nullptr; | |||||
| }; | |||||
| void ProcessFilterFp16(float16_t *origin_weight, float16_t *dst_weight, ConvParameter *conv_param); | |||||
| } // namespace mindspore::kernel | |||||
| #endif // MINDSPORE_LITE_SRC_RUNTIME_KERNEL_ARM_FP16_CONVOLUTION_3x3_FP16_H_ | |||||
| @@ -17,7 +17,6 @@ | |||||
| #include "src/runtime/kernel/arm/fp16/convolution_fp16.h" | #include "src/runtime/kernel/arm/fp16/convolution_fp16.h" | ||||
| #include <vector> | #include <vector> | ||||
| #include "src/runtime/kernel/arm/fp16/convolution_winograd_fp16.h" | #include "src/runtime/kernel/arm/fp16/convolution_winograd_fp16.h" | ||||
| #include "src/runtime/kernel/arm/fp16/convolution_3x3_fp16.h" | |||||
| #include "src/runtime/kernel/arm/fp16/convolution_1x1_fp16.h" | #include "src/runtime/kernel/arm/fp16/convolution_1x1_fp16.h" | ||||
| #include "nnacl/fp16/conv_fp16.h" | #include "nnacl/fp16/conv_fp16.h" | ||||
| #include "nnacl/fp16/cast_fp16.h" | #include "nnacl/fp16/cast_fp16.h" | ||||
| @@ -45,9 +44,8 @@ int ConvolutionFP16CPUKernel::InitWeightBias() { | |||||
| conv_param_->input_channel_ = in_channel; | conv_param_->input_channel_ = in_channel; | ||||
| conv_param_->output_channel_ = out_channel; | conv_param_->output_channel_ = out_channel; | ||||
| int oc8 = UP_DIV(out_channel, C8NUM); | int oc8 = UP_DIV(out_channel, C8NUM); | ||||
| int ic4 = UP_DIV(in_channel, C4NUM); | |||||
| int kernel_plane = kernel_h * kernel_w; | int kernel_plane = kernel_h * kernel_w; | ||||
| int pack_weight_size = oc8 * ic4 * C8NUM * C4NUM * kernel_plane; | |||||
| int pack_weight_size = oc8 * C8NUM * in_channel * kernel_plane; | |||||
| // init weight | // init weight | ||||
| auto ret = ConvolutionBaseFP16CPUKernel::GetExecuteFilter(); | auto ret = ConvolutionBaseFP16CPUKernel::GetExecuteFilter(); | ||||
| @@ -61,7 +59,7 @@ int ConvolutionFP16CPUKernel::InitWeightBias() { | |||||
| return RET_ERROR; | return RET_ERROR; | ||||
| } | } | ||||
| memset(packed_weight_, 0, pack_weight_size * sizeof(float16_t)); | memset(packed_weight_, 0, pack_weight_size * sizeof(float16_t)); | ||||
| PackWeightFp16(execute_weight_, conv_param_, packed_weight_); | |||||
| RowMajor2Col8MajorFp16(execute_weight_, packed_weight_, out_channel, in_channel * kernel_plane, false); | |||||
| // init bias | // init bias | ||||
| bias_data_ = malloc(oc8 * C8NUM * sizeof(float16_t)); | bias_data_ = malloc(oc8 * C8NUM * sizeof(float16_t)); | ||||
| @@ -83,24 +81,20 @@ int ConvolutionFP16CPUKernel::InitWeightBias() { | |||||
| } | } | ||||
| int ConvolutionFP16CPUKernel::InitTmpBuffer() { | int ConvolutionFP16CPUKernel::InitTmpBuffer() { | ||||
| const int cal_num = 16; | |||||
| int in_channel = conv_param_->input_channel_; | int in_channel = conv_param_->input_channel_; | ||||
| int out_channel = conv_param_->output_channel_; | |||||
| int channel_block = UP_DIV(in_channel, C4NUM); | |||||
| int cal_num = 16; | |||||
| int kernel_plane = conv_param_->kernel_h_ * conv_param_->kernel_w_; | int kernel_plane = conv_param_->kernel_h_ * conv_param_->kernel_w_; | ||||
| int unit_size = kernel_plane * channel_block * C4NUM; | |||||
| int packed_input_size = thread_count_ * cal_num * unit_size; | |||||
| int unit_size = kernel_plane * in_channel * cal_num * thread_count_; | |||||
| packed_input_ = reinterpret_cast<float16_t *>(ctx_->allocator->Malloc(packed_input_size * sizeof(float16_t))); | |||||
| packed_input_ = reinterpret_cast<float16_t *>(ctx_->allocator->Malloc(unit_size * sizeof(float16_t))); | |||||
| if (packed_input_ == nullptr) { | if (packed_input_ == nullptr) { | ||||
| MS_LOG(ERROR) << "malloc packed_input_ failed."; | MS_LOG(ERROR) << "malloc packed_input_ failed."; | ||||
| return RET_ERROR; | return RET_ERROR; | ||||
| } | } | ||||
| tmp_output_block_ = | |||||
| reinterpret_cast<float16_t *>(ctx_->allocator->Malloc(thread_count_ * cal_num * out_channel * sizeof(float16_t))); | |||||
| if (tmp_output_block_ == nullptr) { | |||||
| MS_LOG(ERROR) << "malloc tmp_output_block_ failed."; | |||||
| col_major_input_ = reinterpret_cast<float16_t *>(ctx_->allocator->Malloc(unit_size * sizeof(float16_t))); | |||||
| if (col_major_input_ == nullptr) { | |||||
| MS_LOG(ERROR) << "malloc col_major_input_ failed."; | |||||
| return RET_ERROR; | return RET_ERROR; | ||||
| } | } | ||||
| return RET_OK; | return RET_OK; | ||||
| @@ -134,7 +128,7 @@ int ConvolutionFP16CPUKernel::ReSize() { | |||||
| } | } | ||||
| int ConvolutionFP16CPUKernel::RunImpl(int task_id) { | int ConvolutionFP16CPUKernel::RunImpl(int task_id) { | ||||
| ConvFp16(execute_input_, packed_input_, packed_weight_, reinterpret_cast<float16_t *>(bias_data_), tmp_output_block_, | |||||
| ConvFp16(execute_input_, packed_input_, packed_weight_, reinterpret_cast<float16_t *>(bias_data_), col_major_input_, | |||||
| execute_output_, task_id, conv_param_); | execute_output_, task_id, conv_param_); | ||||
| return RET_OK; | return RET_OK; | ||||
| } | } | ||||
| @@ -53,14 +53,14 @@ class ConvolutionFP16CPUKernel : public ConvolutionBaseFP16CPUKernel { | |||||
| ctx_->allocator->Free(packed_input_); | ctx_->allocator->Free(packed_input_); | ||||
| packed_input_ = nullptr; | packed_input_ = nullptr; | ||||
| } | } | ||||
| if (tmp_output_block_ != nullptr) { | |||||
| ctx_->allocator->Free(tmp_output_block_); | |||||
| tmp_output_block_ = nullptr; | |||||
| if (col_major_input_ != nullptr) { | |||||
| ctx_->allocator->Free(col_major_input_); | |||||
| col_major_input_ = nullptr; | |||||
| } | } | ||||
| } | } | ||||
| float16_t *packed_input_ = nullptr; | float16_t *packed_input_ = nullptr; | ||||
| float16_t *packed_weight_ = nullptr; | float16_t *packed_weight_ = nullptr; | ||||
| float16_t *tmp_output_block_ = nullptr; | |||||
| float16_t *col_major_input_ = nullptr; | |||||
| }; | }; | ||||
| } // namespace mindspore::kernel | } // namespace mindspore::kernel | ||||
| @@ -1,591 +0,0 @@ | |||||
| /** | |||||
| * Copyright 2020 Huawei Technologies Co., Ltd | |||||
| * | |||||
| * Licensed under the Apache License, Version 2.0 (the "License"); | |||||
| * you may not use this file except in compliance with the License. | |||||
| * You may obtain a copy of the License at | |||||
| * | |||||
| * http://www.apache.org/licenses/LICENSE-2.0 | |||||
| * | |||||
| * Unless required by applicable law or agreed to in writing, software | |||||
| * distributed under the License is distributed on an "AS IS" BASIS, | |||||
| * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. | |||||
| * See the License for the specific language governing permissions and | |||||
| * limitations under the License. | |||||
| */ | |||||
| #include <iostream> | |||||
| #include <memory> | |||||
| #include "src/common/log_adapter.h" | |||||
| #include "common/common_test.h" | |||||
| #include "mindspore/lite/src/common/utils.h" | |||||
| #include "src/common/file_utils.h" | |||||
| #include "mindspore/lite/src/runtime/kernel/arm/fp16/convolution_fp16.h" | |||||
| #include "mindspore/lite/src/runtime/kernel/arm/fp16/convolution_3x3_fp16.h" | |||||
| #include "nnacl/fp16/conv_fp16.h" | |||||
| namespace mindspore { | |||||
| class TestConvolutionFp16 : public mindspore::CommonTest { | |||||
| public: | |||||
| TestConvolutionFp16() {} | |||||
| }; | |||||
| void InitConvParamGroup1Fp16(ConvParameter *conv_param) { | |||||
| conv_param->input_batch_ = 1; | |||||
| conv_param->input_h_ = 28; | |||||
| conv_param->input_w_ = 28; | |||||
| conv_param->input_channel_ = 3; | |||||
| conv_param->output_batch_ = 1; | |||||
| conv_param->output_h_ = 28; | |||||
| conv_param->output_w_ = 28; | |||||
| conv_param->output_channel_ = 32; | |||||
| conv_param->kernel_h_ = 3; | |||||
| conv_param->kernel_w_ = 3; | |||||
| conv_param->stride_h_ = 1; | |||||
| conv_param->stride_w_ = 1; | |||||
| conv_param->dilation_h_ = 1; | |||||
| conv_param->dilation_w_ = 1; | |||||
| conv_param->pad_u_ = 1; | |||||
| conv_param->pad_l_ = 1; | |||||
| conv_param->thread_num_ = 1; | |||||
| } | |||||
| void InitConvParamGroup2Fp16(ConvParameter *conv_param) { | |||||
| conv_param->input_batch_ = 1; | |||||
| conv_param->input_h_ = 128; | |||||
| conv_param->input_w_ = 128; | |||||
| conv_param->input_channel_ = 32; | |||||
| conv_param->output_batch_ = 1; | |||||
| conv_param->output_h_ = 128; | |||||
| conv_param->output_w_ = 128; | |||||
| conv_param->output_channel_ = 32; | |||||
| conv_param->kernel_h_ = 3; | |||||
| conv_param->kernel_w_ = 3; | |||||
| conv_param->stride_h_ = 1; | |||||
| conv_param->stride_w_ = 1; | |||||
| conv_param->dilation_h_ = 1; | |||||
| conv_param->dilation_w_ = 1; | |||||
| conv_param->pad_u_ = 1; | |||||
| conv_param->pad_l_ = 1; | |||||
| conv_param->thread_num_ = 1; | |||||
| } | |||||
| TEST_F(TestConvolutionFp16, ConvTest1) { | |||||
| // prepare stage | |||||
| auto conv_param = new ConvParameter(); | |||||
| InitConvParamGroup1Fp16(conv_param); | |||||
| int tile_num = 16; | |||||
| int k_h = conv_param->kernel_h_; | |||||
| int k_w = conv_param->kernel_w_; | |||||
| int kernel_plane = k_h * k_w; | |||||
| int in_batch = conv_param->input_batch_; | |||||
| int in_channel = conv_param->input_channel_; | |||||
| int i_h = conv_param->input_h_; | |||||
| int i_w = conv_param->input_w_; | |||||
| int out_channel = conv_param->output_channel_; | |||||
| int ic4 = UP_DIV(in_channel, C4NUM); | |||||
| int oc8 = UP_DIV(out_channel, C8NUM); | |||||
| size_t weight_size; | |||||
| std::string weight_path = "./test_data/conv/convfp32_weight_32_3_3_3.bin"; | |||||
| auto weight_data = reinterpret_cast<float *>(mindspore::lite::ReadFile(weight_path.c_str(), &weight_size)); | |||||
| std::cout << "==============fp32 weight data===========" << std::endl; | |||||
| for (int i = 0; i < 20; i++) { | |||||
| std::cout << weight_data[i] << ", "; | |||||
| } | |||||
| std::cout << std::endl; | |||||
| std::cout << "weight data size: " << weight_size / sizeof(float) << std::endl; | |||||
| int weight_ele_size = weight_size / sizeof(float); | |||||
| auto fp16_weight_data = new float16_t[weight_ele_size]; | |||||
| for (int i = 0; i < weight_ele_size; i++) { | |||||
| fp16_weight_data[i] = static_cast<float16_t>(weight_data[i]); | |||||
| } | |||||
| std::cout << "==============fp16 weight data===========" << std::endl; | |||||
| for (int i = 0; i < 20; i++) { | |||||
| std::cout << fp16_weight_data[i] << ", "; | |||||
| } | |||||
| std::cout << std::endl; | |||||
| auto packed_weight = reinterpret_cast<float16_t *>(malloc(k_h * k_w * ic4 * C4NUM * oc8 * C8NUM * sizeof(float16_t))); | |||||
| PackWeightFp16(fp16_weight_data, conv_param, packed_weight); | |||||
| std::cout << "==============fp16 packed weight data===========" << std::endl; | |||||
| for (int i = 0; i < 20; i++) { | |||||
| std::cout << packed_weight[i] << ", "; | |||||
| } | |||||
| std::cout << std::endl; | |||||
| size_t input_size; | |||||
| std::string input_path = "./test_data/conv/convfp32_input_1_28_28_3.bin"; | |||||
| auto input_data = reinterpret_cast<float *>(mindspore::lite::ReadFile(input_path.c_str(), &input_size)); | |||||
| std::cout << "==============fp32 input data===========" << std::endl; | |||||
| for (int i = 0; i < 20; i++) { | |||||
| std::cout << input_data[i] << ", "; | |||||
| } | |||||
| std::cout << std::endl; | |||||
| int input_ele_size = input_size / sizeof(float); | |||||
| auto fp16_input_data = new float16_t[input_ele_size]; | |||||
| for (int i = 0; i < input_ele_size; i++) { | |||||
| fp16_input_data[i] = static_cast<float16_t>(input_data[i]); | |||||
| } | |||||
| auto nhwc4_input_data = reinterpret_cast<float16_t *>(malloc(i_h * i_w * ic4 * C4NUM * sizeof(float16_t))); | |||||
| PackNHWCToNHWC4Fp32(fp16_input_data, nhwc4_input_data, 1, i_h * i_w, in_channel); | |||||
| std::cout << "==============fp16 input data===========" << std::endl; | |||||
| for (int i = 0; i < 20; i++) { | |||||
| std::cout << fp16_input_data[i] << ", "; | |||||
| } | |||||
| std::cout << std::endl; | |||||
| int output_count = conv_param->output_h_ * conv_param->output_w_; | |||||
| int output_tile_count = UP_DIV(output_count, tile_num); | |||||
| int unit_size = kernel_plane * ic4 * C4NUM; | |||||
| int packed_input_size = output_tile_count * tile_num * unit_size; | |||||
| auto packed_input = reinterpret_cast<float16_t *>(malloc(in_batch * packed_input_size * sizeof(float16_t))); | |||||
| memset(packed_input, 0, in_batch * packed_input_size * sizeof(float16_t)); | |||||
| auto bias_data = reinterpret_cast<float16_t *>(malloc(conv_param->output_channel_ * sizeof(float16_t))); | |||||
| memset(bias_data, 0, conv_param->output_channel_ * sizeof(float16_t)); | |||||
| size_t output_data_size = | |||||
| conv_param->output_batch_ * conv_param->output_channel_ * conv_param->output_h_ * conv_param->output_w_; | |||||
| auto output_data = new float16_t[output_data_size]; | |||||
| auto tmp_output_block = reinterpret_cast<float16_t *>(malloc(tile_num * out_channel * sizeof(float16_t))); | |||||
| // runtime part | |||||
| printf("Calculating runtime cost...\n"); | |||||
| uint64_t time_avg = 0; | |||||
| // warmup | |||||
| for (int i = 0; i < 3; i++) { | |||||
| ConvFp16(nhwc4_input_data, packed_input, packed_weight, bias_data, tmp_output_block, output_data, 0, conv_param); | |||||
| } | |||||
| int loop_count = 100; | |||||
| auto time_start = mindspore::lite::GetTimeUs(); | |||||
| for (int i = 0; i < loop_count; i++) { | |||||
| ConvFp16(nhwc4_input_data, packed_input, packed_weight, bias_data, tmp_output_block, output_data, 0, conv_param); | |||||
| } | |||||
| auto time_end = mindspore::lite::GetTimeUs(); | |||||
| auto cost = time_end - time_start; | |||||
| time_avg = cost / loop_count; | |||||
| printf("single thread running time : %f ms\n", time_avg / 1000.0f); | |||||
| std::cout << "==============fp16 output data===========" << std::endl; | |||||
| for (int i = 0; i < 20; i++) { | |||||
| std::cout << output_data[i] << ", "; | |||||
| } | |||||
| std::cout << std::endl; | |||||
| auto fp32_output_data = new float[output_data_size]; | |||||
| for (int i = 0; i < output_data_size; i++) { | |||||
| fp32_output_data[i] = static_cast<float>(output_data[i]); | |||||
| } | |||||
| printf("==================output data=================\n"); | |||||
| for (int i = 0; i < 20; i++) { | |||||
| std::cout << fp32_output_data[i] << " ,"; | |||||
| } | |||||
| std::cout << std::endl; | |||||
| std::string output_path = "./test_data/conv/convfp32_out_1_28_28_32.bin"; | |||||
| lite::CompareOutput(fp32_output_data, output_path); | |||||
| free(nhwc4_input_data); | |||||
| free(packed_input); | |||||
| free(bias_data); | |||||
| free(packed_weight); | |||||
| free(tmp_output_block); | |||||
| delete conv_param; | |||||
| delete input_data; | |||||
| delete weight_data; | |||||
| delete[] fp16_weight_data; | |||||
| delete[] fp16_input_data; | |||||
| delete[] fp32_output_data; | |||||
| delete[] output_data; | |||||
| MS_LOG(INFO) << "TestConvolutionFp16 passed"; | |||||
| } | |||||
| TEST_F(TestConvolutionFp16, ConvTest2) { | |||||
| // prepare stage | |||||
| auto conv_param = new ConvParameter(); | |||||
| InitConvParamGroup2Fp16(conv_param); | |||||
| // parameter | |||||
| int tile_num = 16; | |||||
| int k_h = conv_param->kernel_h_; | |||||
| int k_w = conv_param->kernel_w_; | |||||
| int kernel_plane = k_h * k_w; | |||||
| int in_batch = conv_param->input_batch_; | |||||
| int in_channel = conv_param->input_channel_; | |||||
| int out_channel = conv_param->output_channel_; | |||||
| int ic4 = UP_DIV(in_channel, C4NUM); | |||||
| int oc8 = UP_DIV(out_channel, C8NUM); | |||||
| // weight | |||||
| size_t weight_size; | |||||
| std::string weight_path = "./test_data/conv/convfp32_weight_32_3_3_32.bin"; | |||||
| auto weight_data = reinterpret_cast<float *>(mindspore::lite::ReadFile(weight_path.c_str(), &weight_size)); | |||||
| int weight_ele_size = weight_size / sizeof(float); | |||||
| auto fp16_weight_data = new float16_t[weight_ele_size]; | |||||
| for (int i = 0; i < weight_ele_size; i++) { | |||||
| fp16_weight_data[i] = static_cast<float16_t>(weight_data[i]); | |||||
| } | |||||
| auto packed_weight = reinterpret_cast<float16_t *>(malloc(k_h * k_w * ic4 * C4NUM * oc8 * C8NUM * sizeof(float16_t))); | |||||
| PackWeightFp16(fp16_weight_data, conv_param, packed_weight); | |||||
| // input | |||||
| size_t input_size; | |||||
| std::string input_path = "./test_data/conv/convfp32_input_1_128_128_32.bin"; | |||||
| auto input_data = reinterpret_cast<float *>(mindspore::lite::ReadFile(input_path.c_str(), &input_size)); | |||||
| int input_ele_size = input_size / sizeof(float); | |||||
| auto fp16_input_data = new float16_t[input_ele_size]; | |||||
| for (int i = 0; i < input_ele_size; i++) { | |||||
| fp16_input_data[i] = static_cast<float16_t>(input_data[i]); | |||||
| } | |||||
| int output_count = conv_param->output_h_ * conv_param->output_w_; | |||||
| int output_tile_count = UP_DIV(output_count, tile_num); | |||||
| int unit_size = kernel_plane * ic4 * C4NUM; | |||||
| int packed_input_size = output_tile_count * tile_num * unit_size; | |||||
| auto packed_input = reinterpret_cast<float16_t *>(malloc(in_batch * packed_input_size * sizeof(float16_t))); | |||||
| memset(packed_input, 0, in_batch * packed_input_size * sizeof(float16_t)); | |||||
| // bias | |||||
| auto bias_data = reinterpret_cast<float16_t *>(malloc(conv_param->output_channel_ * sizeof(float16_t))); | |||||
| memset(bias_data, 0, conv_param->output_channel_ * sizeof(float16_t)); | |||||
| // output | |||||
| auto tmp_output_block = reinterpret_cast<float16_t *>(malloc(tile_num * out_channel * sizeof(float16_t))); | |||||
| size_t output_data_size = | |||||
| conv_param->output_batch_ * conv_param->output_channel_ * conv_param->output_h_ * conv_param->output_w_; | |||||
| auto output_data = new float16_t[output_data_size]; | |||||
| // runtime part | |||||
| printf("Calculating runtime cost...\n"); | |||||
| uint64_t time_avg = 0; | |||||
| // warmup | |||||
| for (int i = 0; i < 3; i++) { | |||||
| ConvFp16(fp16_input_data, packed_input, packed_weight, bias_data, tmp_output_block, output_data, 0, conv_param); | |||||
| } | |||||
| int loop_count = 100; | |||||
| auto time_start = mindspore::lite::GetTimeUs(); | |||||
| for (int i = 0; i < loop_count; i++) { | |||||
| ConvFp16(fp16_input_data, packed_input, packed_weight, bias_data, tmp_output_block, output_data, 0, conv_param); | |||||
| } | |||||
| auto time_end = mindspore::lite::GetTimeUs(); | |||||
| auto cost = time_end - time_start; | |||||
| time_avg = cost / loop_count; | |||||
| printf("single thread running time : %f ms\n", time_avg / 1000.0f); | |||||
| std::cout << "==============fp16 output data===========" << std::endl; | |||||
| for (int i = 0; i < 20; i++) { | |||||
| std::cout << output_data[i] << ", "; | |||||
| } | |||||
| std::cout << std::endl; | |||||
| auto fp32_output_data = new float[output_data_size]; | |||||
| for (int i = 0; i < output_data_size; i++) { | |||||
| fp32_output_data[i] = static_cast<float>(output_data[i]); | |||||
| } | |||||
| printf("==================output data=================\n"); | |||||
| for (int i = 0; i < 20; i++) { | |||||
| std::cout << fp32_output_data[i] << " ,"; | |||||
| } | |||||
| std::cout << std::endl; | |||||
| std::string output_path = "./test_data/conv/convfp32_out_1_128_128_32.bin"; | |||||
| lite::CompareOutput(fp32_output_data, output_path); | |||||
| free(packed_input); | |||||
| free(bias_data); | |||||
| free(packed_weight); | |||||
| free(tmp_output_block); | |||||
| delete conv_param; | |||||
| delete input_data; | |||||
| delete weight_data; | |||||
| delete[] fp16_weight_data; | |||||
| delete[] fp16_input_data; | |||||
| delete[] fp32_output_data; | |||||
| delete[] output_data; | |||||
| MS_LOG(INFO) << "TestConvolutionFp16 passed"; | |||||
| } | |||||
| TEST_F(TestConvolutionFp16, Conv3x3Test1) { | |||||
| auto conv_param = new ConvParameter(); | |||||
| InitConvParamGroup1Fp16(conv_param); | |||||
| int thread_count = 1; | |||||
| int tile_num = 16; | |||||
| int output_batch = conv_param->output_batch_; | |||||
| int output_h = conv_param->output_h_; | |||||
| int output_w = conv_param->output_w_; | |||||
| int ic4 = UP_DIV(conv_param->input_channel_, C4NUM); | |||||
| int oc8 = UP_DIV(conv_param->output_channel_, C8NUM); | |||||
| // tmp buffer | |||||
| int k_plane = 36; | |||||
| size_t tile_buffer_size = thread_count * tile_num * k_plane * ic4 * C4NUM * sizeof(float16_t); | |||||
| float16_t *tile_buffer = reinterpret_cast<float16_t *>(malloc(tile_buffer_size)); | |||||
| memset(tile_buffer, 0, tile_buffer_size); | |||||
| size_t block_unit_buffer_size = thread_count * k_plane * C4NUM * sizeof(float16_t); | |||||
| float16_t *block_unit_buffer = reinterpret_cast<float16_t *>(malloc(block_unit_buffer_size)); | |||||
| memset(block_unit_buffer, 0, block_unit_buffer_size); | |||||
| size_t tmp_dst_buffer_size = thread_count * tile_num * k_plane * oc8 * C8NUM * sizeof(float16_t); | |||||
| float16_t *tmp_dst_buffer = reinterpret_cast<float16_t *>(malloc(tmp_dst_buffer_size)); | |||||
| memset(tmp_dst_buffer, 0, tmp_dst_buffer_size); | |||||
| size_t tmp_out_size = oc8 * C8NUM * output_batch * output_h * output_w * tile_num * sizeof(float16_t); | |||||
| float16_t *tmp_out = reinterpret_cast<float16_t *>(malloc(tmp_out_size)); | |||||
| memset(tmp_out, 0, tmp_out_size); | |||||
| // weight | |||||
| size_t weight_size; | |||||
| std::string weight_path = "./test_data/conv/convfp32_weight_32_3_3_3.bin"; | |||||
| auto weight_data = reinterpret_cast<float *>(mindspore::lite::ReadFile(weight_path.c_str(), &weight_size)); | |||||
| std::cout << "==============fp32 weight data===========" << std::endl; | |||||
| for (int i = 0; i < 20; i++) { | |||||
| std::cout << weight_data[i] << ", "; | |||||
| } | |||||
| std::cout << std::endl; | |||||
| std::cout << "weight data size: " << weight_size / sizeof(float) << std::endl; | |||||
| int weight_ele_size = weight_size / sizeof(float); | |||||
| auto fp16_weight_data = new float16_t[weight_ele_size]; | |||||
| for (int i = 0; i < weight_ele_size; i++) { | |||||
| fp16_weight_data[i] = (float16_t)weight_data[i]; | |||||
| } | |||||
| std::cout << "==============fp16 weight data===========" << std::endl; | |||||
| for (int i = 0; i < 20; i++) { | |||||
| std::cout << fp16_weight_data[i] << ", "; | |||||
| } | |||||
| std::cout << std::endl; | |||||
| size_t transformed_size = ic4 * C4NUM * oc8 * C8NUM * 36; | |||||
| auto transformed_weight_data = new float16_t[transformed_size]; | |||||
| memset(transformed_weight_data, 0, transformed_size * sizeof(float16_t)); | |||||
| kernel::ProcessFilterFp16(fp16_weight_data, transformed_weight_data, conv_param); | |||||
| // bias | |||||
| auto bias_data = | |||||
| reinterpret_cast<float16_t *>(malloc(UP_DIV(conv_param->output_channel_, 8) * 8 * sizeof(float16_t))); | |||||
| memset(bias_data, 0, UP_DIV(conv_param->output_channel_, 8) * 8 * sizeof(float16_t)); | |||||
| // input | |||||
| size_t input_size; | |||||
| std::string input_path = "./test_data/conv/convfp32_input_1_28_28_3.bin"; | |||||
| auto input_data = reinterpret_cast<float *>(mindspore::lite::ReadFile(input_path.c_str(), &input_size)); | |||||
| std::cout << "==============fp32 input data===========" << std::endl; | |||||
| for (int i = 0; i < 20; i++) { | |||||
| std::cout << input_data[i] << ", "; | |||||
| } | |||||
| std::cout << std::endl; | |||||
| int input_ele_size = input_size / sizeof(float); | |||||
| auto fp16_input_data = new float16_t[input_ele_size]; | |||||
| for (int i = 0; i < input_ele_size; i++) { | |||||
| fp16_input_data[i] = static_cast<float16_t>(input_data[i]); | |||||
| } | |||||
| std::cout << "==============fp16 input data===========" << std::endl; | |||||
| for (int i = 0; i < 20; i++) { | |||||
| std::cout << fp16_input_data[i] << ", "; | |||||
| } | |||||
| std::cout << std::endl; | |||||
| // output | |||||
| size_t output_data_size = | |||||
| conv_param->output_batch_ * conv_param->output_channel_ * conv_param->output_h_ * conv_param->output_w_; | |||||
| auto output_data = new float16_t[output_data_size]; | |||||
| // runtime part | |||||
| printf("Calculating runtime cost...\n"); | |||||
| uint64_t time_avg = 0; | |||||
| // warmup | |||||
| for (int i = 0; i < 3; i++) { | |||||
| Conv3x3Fp16(fp16_input_data, transformed_weight_data, bias_data, output_data, tile_buffer, block_unit_buffer, | |||||
| tmp_dst_buffer, tmp_out, 0, conv_param); | |||||
| } | |||||
| int loop_count = 100; | |||||
| auto time_start = mindspore::lite::GetTimeUs(); | |||||
| for (int i = 0; i < loop_count; i++) { | |||||
| Conv3x3Fp16(fp16_input_data, transformed_weight_data, bias_data, output_data, tile_buffer, block_unit_buffer, | |||||
| tmp_dst_buffer, tmp_out, 0, conv_param); | |||||
| } | |||||
| auto time_end = mindspore::lite::GetTimeUs(); | |||||
| auto cost = time_end - time_start; | |||||
| time_avg = cost / loop_count; | |||||
| printf("single thread running time : %f ms\n", time_avg / 1000.0f); | |||||
| std::cout << "==============fp16 output data===========" << std::endl; | |||||
| for (int i = 0; i < 20; i++) { | |||||
| std::cout << output_data[i] << ", "; | |||||
| } | |||||
| std::cout << std::endl; | |||||
| auto fp32_output_data = new float[output_data_size]; | |||||
| for (int i = 0; i < output_data_size; i++) { | |||||
| fp32_output_data[i] = static_cast<float>(output_data[i]); | |||||
| } | |||||
| printf("==================output data=================\n"); | |||||
| for (int i = 0; i < 20; i++) { | |||||
| std::cout << fp32_output_data[i] << " ,"; | |||||
| } | |||||
| std::cout << std::endl; | |||||
| std::string output_path = "./test_data/conv/convfp32_out_1_28_28_32.bin"; | |||||
| lite::CompareOutput(fp32_output_data, output_path); | |||||
| free(bias_data); | |||||
| free(tile_buffer); | |||||
| free(block_unit_buffer); | |||||
| free(tmp_dst_buffer); | |||||
| free(tmp_out); | |||||
| delete input_data; | |||||
| delete weight_data; | |||||
| delete conv_param; | |||||
| delete[] fp16_weight_data; | |||||
| delete[] fp16_input_data; | |||||
| delete[] fp32_output_data; | |||||
| delete[] output_data; | |||||
| delete[] transformed_weight_data; | |||||
| MS_LOG(INFO) << "TestConvolutionFp16 Conv3x3 passed"; | |||||
| } | |||||
| TEST_F(TestConvolutionFp16, Conv3x3Test2) { | |||||
| auto conv_param = new ConvParameter(); | |||||
| InitConvParamGroup2Fp16(conv_param); | |||||
| int thread_count = 1; | |||||
| int tile_num = 16; | |||||
| int output_batch = conv_param->output_batch_; | |||||
| int output_h = conv_param->output_h_; | |||||
| int output_w = conv_param->output_w_; | |||||
| int ic4 = UP_DIV(conv_param->input_channel_, C4NUM); | |||||
| int oc8 = UP_DIV(conv_param->output_channel_, C8NUM); | |||||
| // tmp buffer | |||||
| int k_plane = 36; | |||||
| size_t tile_buffer_size = thread_count * tile_num * k_plane * ic4 * C4NUM * sizeof(float16_t); | |||||
| float16_t *tile_buffer = reinterpret_cast<float16_t *>(malloc(tile_buffer_size)); | |||||
| memset(tile_buffer, 0, tile_buffer_size); | |||||
| size_t block_unit_buffer_size = thread_count * k_plane * C4NUM * sizeof(float16_t); | |||||
| float16_t *block_unit_buffer = reinterpret_cast<float16_t *>(malloc(block_unit_buffer_size)); | |||||
| memset(block_unit_buffer, 0, block_unit_buffer_size); | |||||
| size_t tmp_dst_buffer_size = thread_count * tile_num * k_plane * oc8 * C8NUM * sizeof(float16_t); | |||||
| float16_t *tmp_dst_buffer = reinterpret_cast<float16_t *>(malloc(tmp_dst_buffer_size)); | |||||
| memset(tmp_dst_buffer, 0, tmp_dst_buffer_size); | |||||
| size_t tmp_out_size = oc8 * C8NUM * output_batch * output_h * output_w * tile_num * sizeof(float16_t); | |||||
| float16_t *tmp_out = reinterpret_cast<float16_t *>(malloc(tmp_out_size)); | |||||
| memset(tmp_out, 0, tmp_out_size); | |||||
| // weight | |||||
| size_t weight_size; | |||||
| std::string weight_path = "./test_data/conv/convfp32_weight_32_3_3_32.bin"; | |||||
| auto weight_data = reinterpret_cast<float *>(mindspore::lite::ReadFile(weight_path.c_str(), &weight_size)); | |||||
| int weight_ele_size = weight_size / sizeof(float); | |||||
| auto fp16_weight_data = new float16_t[weight_ele_size]; | |||||
| for (int i = 0; i < weight_ele_size; i++) { | |||||
| fp16_weight_data[i] = static_cast<float16_t>(weight_data[i]); | |||||
| } | |||||
| size_t transformed_size = ic4 * C4NUM * oc8 * C8NUM * 36; | |||||
| auto transformed_weight_data = new float16_t[transformed_size]; | |||||
| memset(transformed_weight_data, 0, transformed_size * sizeof(float16_t)); | |||||
| kernel::ProcessFilterFp16(fp16_weight_data, transformed_weight_data, conv_param); | |||||
| // bias | |||||
| auto bias_data = | |||||
| reinterpret_cast<float16_t *>(malloc(UP_DIV(conv_param->output_channel_, 8) * 8 * sizeof(float16_t))); | |||||
| memset(bias_data, 0, UP_DIV(conv_param->output_channel_, 8) * 8 * sizeof(float16_t)); | |||||
| // input | |||||
| size_t input_size; | |||||
| std::string input_path = "./test_data/conv/convfp32_input_1_128_128_32.bin"; | |||||
| auto input_data = reinterpret_cast<float *>(mindspore::lite::ReadFile(input_path.c_str(), &input_size)); | |||||
| int input_ele_size = input_size / sizeof(float); | |||||
| auto fp16_input_data = new float16_t[input_ele_size]; | |||||
| for (int i = 0; i < input_ele_size; i++) { | |||||
| fp16_input_data[i] = static_cast<float16_t>(input_data[i]); | |||||
| } | |||||
| // output | |||||
| size_t output_data_size = | |||||
| conv_param->output_batch_ * conv_param->output_channel_ * conv_param->output_h_ * conv_param->output_w_; | |||||
| auto output_data = new float16_t[output_data_size]; | |||||
| // runtime part | |||||
| printf("Calculating runtime cost...\n"); | |||||
| uint64_t time_avg = 0; | |||||
| // warmup | |||||
| for (int i = 0; i < 3; i++) { | |||||
| Conv3x3Fp16(fp16_input_data, transformed_weight_data, bias_data, output_data, tile_buffer, block_unit_buffer, | |||||
| tmp_dst_buffer, tmp_out, 0, conv_param); | |||||
| } | |||||
| int loop_count = 100; | |||||
| auto time_start = mindspore::lite::GetTimeUs(); | |||||
| for (int i = 0; i < loop_count; i++) { | |||||
| Conv3x3Fp16(fp16_input_data, transformed_weight_data, bias_data, output_data, tile_buffer, block_unit_buffer, | |||||
| tmp_dst_buffer, tmp_out, 0, conv_param); | |||||
| } | |||||
| auto time_end = mindspore::lite::GetTimeUs(); | |||||
| auto cost = time_end - time_start; | |||||
| time_avg = cost / loop_count; | |||||
| printf("single thread running time : %f ms\n", time_avg / 1000.0f); | |||||
| std::cout << "==============fp16 output data===========" << std::endl; | |||||
| for (int i = 0; i < 20; i++) { | |||||
| std::cout << output_data[i] << ", "; | |||||
| } | |||||
| std::cout << std::endl; | |||||
| auto fp32_output_data = new float[output_data_size]; | |||||
| for (int i = 0; i < output_data_size; i++) { | |||||
| fp32_output_data[i] = static_cast<float>(output_data[i]); | |||||
| } | |||||
| printf("==================output data=================\n"); | |||||
| for (int i = 0; i < 20; i++) { | |||||
| std::cout << fp32_output_data[i] << " ,"; | |||||
| } | |||||
| std::cout << std::endl; | |||||
| std::string output_path = "./test_data/conv/convfp32_out_1_128_128_32.bin"; | |||||
| lite::CompareOutput(fp32_output_data, output_path); | |||||
| free(bias_data); | |||||
| free(tile_buffer); | |||||
| free(block_unit_buffer); | |||||
| free(tmp_dst_buffer); | |||||
| free(tmp_out); | |||||
| delete input_data; | |||||
| delete weight_data; | |||||
| delete conv_param; | |||||
| delete[] fp16_weight_data; | |||||
| delete[] fp16_input_data; | |||||
| delete[] fp32_output_data; | |||||
| delete[] output_data; | |||||
| delete[] transformed_weight_data; | |||||
| MS_LOG(INFO) << "TestConvolutionFp16 Conv3x3 passed"; | |||||
| } | |||||
| } // namespace mindspore | |||||