| /third_party/mindspore/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/int8/ |
| D | deconv_int8.c | 26 int oc4 = UP_DIV(output_channel, C4NUM); in DeConvPostInt8C4() local
38 for (int c = 0; c < oc4; c++) { in DeConvPostInt8C4()
95 int oc4 = UP_ROUND(output_channel, C4NUM); in DeConvWeightTransInt8() local
102 …int dst_index = hw * ic16 * oc4 + oc4div * ic16 * C4NUM + ic16div * C16NUM * C4NUM + oc4mod * C16N… in DeConvWeightTransInt8()
|
| D | conv3x3_int8.c | 753 const int oc4 = UP_DIV(output_channel, C4NUM); in Conv3x3Int8OutputTransform() local
761 int src_tile_offset = i * oc4 * C4NUM * input_unit * input_unit; in Conv3x3Int8OutputTransform()
764 for (int j = 0; j < oc4; j++) { in Conv3x3Int8OutputTransform()
827 int oc4 = UP_DIV(oc, C4NUM); in Conv3x3Int8Gemm() local
829 IndirectGemmInt16to32_8x4(dst, src, weight, 16, ic8, oc4, (size_t)oc4 * 4 * 16 * sizeof(int32_t)); in Conv3x3Int8Gemm()
832 for (int c = 0; c < oc4; c++) { in Conv3x3Int8Gemm()
837 int dst_tile_offset = dst_oc_offset + n * oc4 * C4NUM * input_unit_square; in Conv3x3Int8Gemm()
879 int oc4 = UP_DIV(conv_param->output_channel_, C4NUM); in Conv3x3Int8() local
882 int tmp_dst_buffer_offset = TILE_NUM * 16 * oc4 * C4NUM; in Conv3x3Int8()
886 int tmp_out_batch_offset = batch * oc4 * C4NUM * conv_param->output_w_ * conv_param->output_h_; in Conv3x3Int8()
|
| D | common_func_int8.h | 42 size_t oc4, size_t offset);
|
| /third_party/mindspore/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/fp32/ |
| D | winograd_transform.c | 106 int oc4 = UP_DIV(output_channel, C4NUM); in WinogradOutputNHWCTransform() local
128 for (int j = 0; j < oc4; j++) { in WinogradOutputNHWCTransform()
166 int oc4 = UP_DIV(output_channel, C4NUM); in WinogradOutputNC4HW4Transform() local
198 for (int j = 0; j < oc4; j++) { in WinogradOutputNC4HW4Transform()
|
| D | deconv_winograd_fp32.h | 37 …mulFp32(float *dst, const float *src, const float *weight, size_t ic4, size_t cal_num, size_t oc4);
|
| D | deconv_winograd_fp32.c | 195 …ulFp32(float *dst, const float *src, const float *weight, size_t cal_num, size_t ic4, size_t oc4) { in TiledC4MatmulFp32() argument
198 for (dz = 0; dz < oc4; ++dz) { in TiledC4MatmulFp32()
|
| /third_party/mindspore/mindspore/lite/micro/coder/opcoders/nnacl/int8/ |
| D | conv2d_3x3_int8_coder.cc | 81 int oc4 = UP_DIV(conv_param_->output_channel_, C4NUM); in InitTmpBuffer() local
99 tmp_dst_buffer_size_ = thread_num_ * TILE_NUM * 16 * oc4 * C4NUM * sizeof(int32_t); in InitTmpBuffer()
103 tmp_out_size_ = oc4 * C4NUM * output_batch * output_w * output_h * sizeof(uint8_t); in InitTmpBuffer()
|
| D | deconvolution_int8_coder.cc | 58 int oc4 = UP_DIV(conv_param_->output_channel_, C4NUM); in InitParam() local
59 thread_count_ = MSMIN(conv_param_->op_parameter_.thread_num_, oc4); in InitParam()
61 thread_stride_ = UP_DIV(oc4, thread_count_); in InitParam()
|
| /third_party/mindspore/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/intrinsics/avx/ |
| D | TiledC4MatMulFp32.c | 31 …lFp32(float *dst, const float *src, const float *weight, size_t dst_step, size_t ic4, size_t oc4) { in TiledC4MatmulFp32() argument
32 for (int oc = 0; oc < oc4; oc++) { in TiledC4MatmulFp32()
|
| /third_party/mindspore/mindspore/lite/src/runtime/kernel/arm/int8/ |
| D | convolution_3x3_int8.cc | 133 int oc4 = UP_DIV(conv_param_->output_channel_, C4NUM); in InitTmpBuffer() local
162 …size_t tmp_dst_buffer_size = thread_count_ * TILE_NUM * kUnitBufferMultipler * oc4 * C4NUM * sizeo… in InitTmpBuffer()
169 size_t tmp_out_size = oc4 * C4NUM * output_batch * output_w * output_h * sizeof(uint8_t); in InitTmpBuffer()
|
| D | deconvolution_int8.cc | 136 int oc4 = UP_DIV(conv_param_->output_channel_, C4NUM); in InitParam() local
137 thread_count_ = MSMIN(op_parameter_->thread_num_, oc4); in InitParam()
139 thread_stride_ = UP_DIV(oc4, thread_count_); in InitParam()
|
| /third_party/mindspore/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/fp16/ |
| D | deconv_winograd_fp16.h | 36 size_t oc4);
|
| D | deconv_winograd_fp16.c | 489 size_t oc4) { in TiledC4MatmulFp16() argument
492 for (dz = 0; dz < oc4; ++dz) { in TiledC4MatmulFp16()
|
| /third_party/mindspore/mindspore/lite/micro/coder/opcoders/nnacl/fp32/ |
| D | convolution_winograd_fp32_coder.cc | 129 int oc4 = UP_DIV(out_channel, C4NUM); in InitWeightBias() local
157 int new_bias_ele_num = oc4 * C4NUM; in InitWeightBias()
|
| /third_party/mindspore/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/intrinsics/sse/ |
| D | TiledC4MatMulFp32.c | 37 …ulFp32(float *dst, const float *src, const float *weight, size_t cal_num, size_t ic4, size_t oc4) { in TiledC4MatmulFp32() argument
39 for (int i = 0; i < oc4; ++i) { in TiledC4MatmulFp32()
|