/third_party/mindspore/mindspore/lite/src/runtime/kernel/arm/fp32/ |
D | activation_fp32.cc | 53 int ActivationCPUKernel::DoActivation(int task_id) { in DoActivation() argument 61 int count = MSMIN(stride, length - stride * task_id); in DoActivation() 65 if (INT_MUL_OVERFLOW(stride, task_id)) { in DoActivation() 72 ret = Fp32Relu(input_addr + stride * task_id, count, output_addr + stride * task_id); in DoActivation() 74 ret = Fp32Relu6(input_addr + stride * task_id, count, output_addr + stride * task_id); in DoActivation() 76 ret = LRelu(input_addr + stride * task_id, count, output_addr + stride * task_id, alpha_); in DoActivation() 78 ret = Sigmoid(input_addr + stride * task_id, count, output_addr + stride * task_id); in DoActivation() 80 ret = Tanh(input_addr + stride * task_id, count, output_addr + stride * task_id); in DoActivation() 82 ret = Swish(input_addr + stride * task_id, count, output_addr + stride * task_id); in DoActivation() 84 ret = HSwish(input_addr + stride * task_id, count, output_addr + stride * task_id); in DoActivation() [all …]
|
D | glu_fp32.cc | 99 int GluCPUKernel::Split(int task_id) { in Split() argument 101 MS_CHECK_INT_MUL_NOT_OVERFLOW(task_id, thread_n_stride_, RET_ERROR); in Split() 102 int num_unit_thread = MSMIN(thread_n_stride_, num_unit_ - task_id * thread_n_stride_); in Split() 106 int thread_offset = task_id * thread_n_stride_; in Split() 111 MS_LOG(ERROR) << "Split error task_id[" << task_id << "] error_code[" << ret << "]"; in Split() 117 int GluCPUKernel::Sigmoid(int task_id) { in Sigmoid() argument 123 MS_CHECK_INT_MUL_NOT_OVERFLOW(stride, task_id, RET_ERROR); in Sigmoid() 124 int count = MSMIN(stride, length - stride * task_id); in Sigmoid() 128 return ::Sigmoid(input_addr + stride * task_id, count, output_addr + stride * task_id); in Sigmoid() 131 int GluCPUKernel::Mul(int task_id) { in Mul() argument [all …]
|
D | l2_norm_fp32.cc | 92 int L2NormCPUKernel::CalcSquareSum(int task_id) { in CalcSquareSum() argument 94 if (INT_MUL_OVERFLOW(task_id, unit)) { in CalcSquareSum() 98 int begin = task_id * unit; in CalcSquareSum() 100 return CalcThreadSquareSum(input_ptr_, tmp_sum_ + task_id, begin, end); in CalcSquareSum() 103 int L2NormCPUKernel::DivSqrtSum(int task_id) { in DivSqrtSum() argument 105 if (INT_MUL_OVERFLOW(task_id, unit)) { in DivSqrtSum() 109 int begin = task_id * unit; in DivSqrtSum() 114 int L2NormCPUKernel::CalcL2NormTrailingAxis(int task_id) { in CalcL2NormTrailingAxis() argument 122 if (INT_MUL_OVERFLOW(task_id, unit)) { in CalcL2NormTrailingAxis() 126 int begin = task_id * unit; in CalcL2NormTrailingAxis() [all …]
|
D | local_response_norm_fp32.cc | 38 int LocalResponseNormCPUKernel::DoLocalResponseNorm(int task_id) { in DoLocalResponseNorm() argument 55 MS_CHECK_INT_MUL_NOT_OVERFLOW(stride, task_id, RET_ERROR); in DoLocalResponseNorm() 56 int count = MSMIN(stride, outer_size - stride * task_id); in DoLocalResponseNorm() 58 input_ptr += stride * task_id * channel; in DoLocalResponseNorm() 59 output_ptr += stride * task_id * channel; in DoLocalResponseNorm() 64 …MS_LOG(ERROR) << "DoLocalResponseNorm error task_id[" << task_id << "] error_code[" << error_code … in DoLocalResponseNorm() 70 int LocalResponseNormRun(void *cdata, int task_id, float lhs_scale, float rhs_scale) { in LocalResponseNormRun() argument 72 auto error_code = lrn->DoLocalResponseNorm(task_id); in LocalResponseNormRun() 74 …MS_LOG(ERROR) << "LocalResponseNormRun error task_id[" << task_id << "] error_code[" << error_code… in LocalResponseNormRun()
|
D | cumsum_fp32.cc | 29 int CumsumLaunch(void *cdata, int task_id, float lhs_scale, float rhs_scale) { in CumsumLaunch() argument 39 ret = kernel->DoCumsum(task_id); in CumsumLaunch() 41 ret = kernel->DoCumsumInt(task_id); in CumsumLaunch() 90 int CumSumCPUKernel::DoCumsum(int task_id) { in DoCumsum() argument 96 float *input = input_data + task_id * unit_ * axis_dim_ * in_dim_; in DoCumsum() 97 int out_dim = MSMIN(out_dim_ - unit_ * task_id, unit_); in DoCumsum() 98 float *output = output_data + task_id * unit_ * axis_dim_ * in_dim_; in DoCumsum() 107 int CumSumCPUKernel::DoCumsumInt(int task_id) { in DoCumsumInt() argument 113 int *input = input_data + task_id * unit_ * axis_dim_ * in_dim_; in DoCumsumInt() 114 int out_dim = MSMIN(out_dim_ - unit_ * task_id, unit_); in DoCumsumInt() [all …]
|
D | convolution_1x1_fp32.cc | 145 int Convolution1x1CPUKernel::DoConv1x1(int task_id) { in DoConv1x1() argument 146 int res_stride = matmul_param_->col_ - task_id * thread_stride_; in DoConv1x1() 152 …ias_data_ == nullptr) ? nullptr : reinterpret_cast<float *>(bias_data_) + thread_stride_ * task_id; in DoConv1x1() 154 …MatMulOpt(pack_input_, reinterpret_cast<float *>(packed_weight_) + task_id * thread_stride_ * matm… in DoConv1x1() 155 … output_ptr_ + task_id * thread_stride_, bias, matmul_param_->act_type_, matmul_param_->deep_, in DoConv1x1() 158 …MatMulOpt(pack_input_, reinterpret_cast<float *>(packed_weight_) + task_id * thread_stride_ * matm… in DoConv1x1() 159 … output_ptr_ + task_id * thread_stride_ * matmul_param_->row_, bias, matmul_param_->act_type_, in DoConv1x1() 165 int Convolution1x1Run(void *cdata, int task_id, float lhs_scale, float rhs_scale) { in Convolution1x1Run() argument 167 auto error_code = conv1x1->DoConv1x1(task_id); in Convolution1x1Run() 169 …MS_LOG(ERROR) << "Convolution1x1Run error task_id[" << task_id << "] error_code[" << error_code <<… in Convolution1x1Run() [all …]
|
D | power_fp32.cc | 36 int PowerImpl(void *cdata, int task_id, float lhs_scale, float rhs_scale) { in PowerImpl() argument 40 auto ret = kernel->RunImpl(task_id); in PowerImpl() 57 int PowerCPUKernel::RunImpl(int task_id) { in RunImpl() argument 64 int len = MSMIN(stride, size - stride * task_id); in RunImpl() 79 cur_exp = exp_addr + stride * task_id; in RunImpl() 83 …Power(x_addr + stride * task_id, cur_exp, output_addr + stride * task_id, len, scale_, shift_, bro… in RunImpl() 85 …MS_LOG(ERROR) << "PowerCPUKernel RunImpl error task_id[" << task_id << "] error_code[" << error_co… in RunImpl()
|
D | arithmetic_compare_fp32.cc | 65 int ArithmeticCompareCPUKernel::DoArithmetic(int task_id) { in DoArithmetic() argument 70 int count = MSMIN(stride, element_num - stride * task_id); in DoArithmetic() 83 int out_count = MSMIN(stride, outside_ - stride * task_id); in DoArithmetic() 84 int out_thread_stride = stride * task_id; in DoArithmetic() 97 error_code = func_fp32_(reinterpret_cast<float *>(input0_ptr_) + stride * task_id, in DoArithmetic() 98 reinterpret_cast<float *>(input1_ptr_) + stride * task_id, in DoArithmetic() 99 … reinterpret_cast<uint8_t *>(out_tensors_[0]->data()) + stride * task_id, count); in DoArithmetic() 101 error_code = func_int32_(reinterpret_cast<int *>(input0_ptr_) + stride * task_id, in DoArithmetic() 102 reinterpret_cast<int *>(input1_ptr_) + stride * task_id, in DoArithmetic() 103 … reinterpret_cast<uint8_t *>(out_tensors_[0]->data()) + stride * task_id, count); in DoArithmetic()
|
/third_party/mindspore/mindspore/lite/src/runtime/kernel/arm/fp16/ |
D | activation_fp16.cc | 45 int ActivationFp16CPUKernel::DoActivation(int task_id) { in DoActivation() argument 49 int count = MSMIN(stride, length - stride * task_id); in DoActivation() 53 if (INT_MUL_OVERFLOW(stride, task_id)) { in DoActivation() 59 error_code = ReluFp16(fp16_input_ + stride * task_id, fp16_output_ + stride * task_id, count); in DoActivation() 61 error_code = Relu6Fp16(fp16_input_ + stride * task_id, fp16_output_ + stride * task_id, count); in DoActivation() 63 …error_code = LReluFp16(fp16_input_ + stride * task_id, fp16_output_ + stride * task_id, count, alp… in DoActivation() 65 … error_code = SigmoidFp16(fp16_input_ + stride * task_id, fp16_output_ + stride * task_id, count); in DoActivation() 67 error_code = TanhFp16(fp16_input_ + stride * task_id, fp16_output_ + stride * task_id, count); in DoActivation() 69 error_code = HSwishFp16(fp16_input_ + stride * task_id, fp16_output_ + stride * task_id, count); in DoActivation() 71 error_code = SwishFp16(fp16_input_ + stride * task_id, fp16_output_ + stride * task_id, count); in DoActivation() [all …]
|
D | convolution_1x1_fp16.cc | 199 int Convolution1x1FP16CPUKernel::RunOc(int task_id) { in RunOc() argument 200 int cur_stride = matmul_param_->col_ - task_id * thread_stride_; in RunOc() 206 …data_ == nullptr) ? nullptr : reinterpret_cast<float16_t *>(bias_data_) + thread_stride_ * task_id; in RunOc() 210 … reinterpret_cast<float16_t *>(bias_data_), output_ptr_, task_id, in RunOc() 215 … reinterpret_cast<float16_t *>(packed_weight_) + task_id * thread_stride_ * matmul_param_->deep_, in RunOc() 216 … output_ptr_ + task_id * thread_stride_, bias, matmul_param_->act_type_, matmul_param_->deep_, in RunOc() 220 … reinterpret_cast<float16_t *>(packed_weight_) + task_id * thread_stride_ * matmul_param_->deep_, in RunOc() 221 … output_ptr_ + task_id * thread_stride_, bias, matmul_param_->act_type_, matmul_param_->deep_, in RunOc() 228 int Convolution1x1FP16CPUKernel::RunHw(int task_id) { in RunHw() argument 231 … reinterpret_cast<float16_t *>(bias_data_), output_ptr_, task_id, in RunHw() [all …]
|
/third_party/mindspore/mindspore/lite/micro/coder/wrapper/fp32/ |
D | deconvolution_fp32_wrapper.c | 23 const ConvParameter *conv_param, int task_id) { in DoDeconvFp32() argument 26 int res_stride = UP_DIV(conv_param->output_channel_, C8NUM) - task_id * thread_stride; in DoDeconvFp32() 29 res_stride = conv_param->output_channel_ - task_id * thread_stride * C8NUM; in DoDeconvFp32() 39 …float *tmp_buffer = tmp_ori_buffer + task_id * thread_stride * C8NUM * kernel_plane * matmul_param… in DoDeconvFp32() 40 …MatMulOpt(packed_input, packed_weight + task_id * thread_stride * C8NUM * kernel_plane * matmul_pa… in DoDeconvFp32() 44 …float *tmp_buffer = tmp_ori_buffer + task_id * thread_stride * C8NUM * kernel_plane * matmul_param… in DoDeconvFp32() 45 …MatMulOpt(packed_input, packed_weight + task_id * thread_stride * C8NUM * kernel_plane * matmul_pa… in DoDeconvFp32() 50 DeConvPostFp32C8(tmp_buffer, packed_output + task_id * thread_stride * C8NUM * output_plane, in DoDeconvFp32() 51 … packed_bias + thread_stride * task_id * C8NUM, output + task_id * thread_stride * C8NUM, oc_res, in DoDeconvFp32() 56 int DeConvFp32Run(void *cdata, int task_id, float lhs_scale, float rhs_scale) { in DeConvFp32Run() argument [all …]
|
/third_party/mindspore/mindspore/ccsrc/backend/kernel_compiler/tbe/ |
D | tbe_kernel_parallel_build.cc | 87 auto task_id = ParallelBuildManager::StartCompileOp(kernel_json); in TbeOpParallelBuild() local 88 build_manger->SaveTaskInfo(task_id, anf_node, json_name, input_size_list, output_size_list); in TbeOpParallelBuild() 91 int task_id = -1; in TbeOpParallelBuild() local 94 auto ret = ParallelBuildManager::WaitOne(&task_id, &task_result, &build_result); in TbeOpParallelBuild() 96 MS_EXCEPTION(ArgumentError) << "Build Failed. wait one ret:" << ret << ", task id:" << task_id in TbeOpParallelBuild() 97 … << " trace: " << trace::DumpSourceLines(build_manger->GetAnfNodeByTaskID(task_id)); in TbeOpParallelBuild() 101 …MS_EXCEPTION(ArgumentError) << "task compile Failed, task id:" << task_id << ", cause:" << task_re… in TbeOpParallelBuild() 102 … << " trace: " << trace::DumpSourceLines(build_manger->GetAnfNodeByTaskID(task_id)); in TbeOpParallelBuild() 104 (void)build_manger->TaskFinishProcess(task_id, build_result); in TbeOpParallelBuild() 111 void ParallelBuildManager::SavePreBuildTaskInfo(int32_t task_id, const AnfNodePtr &anf_node, in SavePreBuildTaskInfo() argument [all …]
|
/third_party/mindspore/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/fp16/ |
D | crop_fp16.c | 23 void Fp16Crop(const float16_t *input, float16_t *output, int task_id, const CropParameter *para) { in Fp16Crop() argument 27 Fp16Crop1D(input, output, task_id, para); in Fp16Crop() 30 Fp16Crop2D(input, output, task_id, para); in Fp16Crop() 33 Fp16Crop3D(input, output, task_id, para); in Fp16Crop() 36 Fp16Crop4D(input, output, task_id, para); in Fp16Crop() 43 void Fp16Crop1D(const float16_t *input, float16_t *output, int task_id, const CropParameter *para) { in Fp16Crop1D() argument 50 int n = task_id * task_id_stride; in Fp16Crop1D() 56 int64_t out_dist_stride = MSMIN(out_batch - task_id * task_id_stride, task_id_stride); in Fp16Crop1D() 60 void Fp16Crop2D(const float16_t *input, float16_t *output, int task_id, const CropParameter *para) { in Fp16Crop2D() argument 71 int h = task_id * task_id_stride; in Fp16Crop2D() [all …]
|
D | conv_fp16.c | 24 … const float16_t *bias_data, float16_t *col_major_input, float16_t *output_data, int task_id, in ConvFp16() argument 35 int start_block = block_per_thread * task_id; in ConvFp16() 43 packed_input += task_id * deep * tile_n; in ConvFp16() 44 col_major_input += task_id * deep * tile_n; in ConvFp16() 66 … const float16_t *bias_data, float16_t *col_major_input, float16_t *output_data, int task_id, in ConvOutNc8hw8Fp16() argument 78 int start_block = block_per_thread * task_id; in ConvOutNc8hw8Fp16() 85 packed_input += deep * tile_n * task_id; in ConvOutNc8hw8Fp16() 86 col_major_input += deep * tile_n * task_id; in ConvOutNc8hw8Fp16() 113 const float16_t *bias, float16_t *output, int task_id, in Conv1x1OutNc8hw8MultiThreadByInputFp16() argument 126 int input_start_block = block_per_thread * task_id; in Conv1x1OutNc8hw8MultiThreadByInputFp16() [all …]
|
/third_party/mindspore/mindspore/lite/micro/coder/wrapper/int8/ |
D | add_int8_wrapper.c | 20 int AddBroadcastInt8Run(void *cdata, int task_id, float lhs_scale, float rhs_scale) { in AddBroadcastInt8Run() argument 23 int real_out_count = MSMIN(stride, args->out_size_ - stride * task_id); in AddBroadcastInt8Run() 32 cur_in0 = args->input0_data_ + task_id * stride * args->in_size_ + i * args->in_size_; in AddBroadcastInt8Run() 34 cur_out = args->output_data_ + task_id * stride * args->in_size_ + i * args->in_size_; in AddBroadcastInt8Run() 37 cur_in1 = args->input1_data_ + task_id * stride * args->in_size_ + i * args->in_size_; in AddBroadcastInt8Run() 38 cur_out = args->output_data_ + task_id * stride * args->in_size_ + i * args->in_size_; in AddBroadcastInt8Run() 45 int AddInt8Run(void *cdata, int task_id, float lhs_scale, float rhs_scale) { in AddInt8Run() argument 49 int rest_count = args->elements_num_ - task_id * stride; in AddInt8Run() 54 int8_t *cur_in0 = args->input0_data_ + stride * task_id; in AddInt8Run() 55 int8_t *cur_in1 = args->input1_data_ + stride * task_id; in AddInt8Run() [all …]
|
D | conv1x1_run_int8_wrapper.c | 33 int OcOptPre(void *cdata, int task_id, float lhs_scale, float rhs_scale) { in OcOptPre() argument 36 int res_stride = args->matmul_param_->row_ - task_id * args->thread_stride_hw_ * C4NUM; in OcOptPre() 41 …int8_t *hw_in = args->input_ptr_ + task_id * args->thread_stride_hw_ * C4NUM * args->conv_param_->… in OcOptPre() 42 …int8_t *hw_packed_in = args->packed_input_ + task_id * args->thread_stride_hw_ * C4NUM * args->mat… in OcOptPre() 43 int32_t *hw_input_sum = args->input_sum_ + task_id * args->thread_stride_hw_ * C4NUM; in OcOptPre() 54 int RunArm64OptOc(void *cdata, int task_id, float lhs_scale, float rhs_scale) { in RunArm64OptOc() argument 57 int cur_stride = task_id * stride; in RunArm64OptOc() 80 int RunArmOc(void *cdata, int task_id, float lhs_scale, float rhs_scale) { in RunArmOc() argument 88 int cur_stride = task_id * stride; in RunArmOc() 111 int RunArm64OptHw(void *cdata, int task_id, float lhs_scale, float rhs_scale) { in RunArm64OptHw() argument [all …]
|
/third_party/mindspore/mindspore/lite/src/runtime/kernel/arm/int8/ |
D | convolution_1x1_int8.cc | 63 int Convolution1x1Int8HwRun(void *cdata, int task_id, float lhs_scale, float rhs_scale) { in Convolution1x1Int8HwRun() argument 65 auto error_code = conv->HwRun(task_id); in Convolution1x1Int8HwRun() 67 …MS_LOG(ERROR) << "conv1x1 Int8 Run error task_id[" << task_id << "] error_code[" << error_code << … in Convolution1x1Int8HwRun() 73 int Convolution1x1Int8OcRun(void *cdata, int task_id, float lhs_scale, float rhs_scale) { in Convolution1x1Int8OcRun() argument 75 auto error_code = conv->OcRun(task_id); in Convolution1x1Int8OcRun() 77 …MS_LOG(ERROR) << "conv1x1 Int8 Run error task_id[" << task_id << "] error_code[" << error_code << … in Convolution1x1Int8OcRun() 83 int Convolution1x1Int8OcOptPre(void *cdata, int task_id, float lhs_scale, float rhs_scale) { in Convolution1x1Int8OcOptPre() argument 85 auto error_code = conv->OcOptPre(task_id); in Convolution1x1Int8OcOptPre() 87 …MS_LOG(ERROR) << "conv1x1 Int8 Run error task_id[" << task_id << "] error_code[" << error_code << … in Convolution1x1Int8OcOptPre() 93 int Convolution1x1Int8CPUKernel::OcRun(int task_id) { in OcRun() argument [all …]
|
D | div_int8.cc | 74 int DivInt8CPUKernel::DoExecute(int task_id) { in DoExecute() argument 82 int count = MSMIN(stride, element_num - stride * task_id); in DoExecute() 86 …ret = DivInt8(tile0_data_ + task_id * count, tile1_data_ + task_id * count, output_data_ + task_id… in DoExecute() 89 …ret = DivInt8(input0_data_ + task_id * count, input1_data_ + task_id * count, output_data_ + task_… in DoExecute() 99 int DivInt8Run(void *cdata, int task_id, float lhs_scale, float rhs_scale) { in DivInt8Run() argument 101 auto ret = div_kernel->DoExecute(task_id); in DivInt8Run() 103 MS_LOG(ERROR) << "DivInt8 DoExecute error task_id[" << task_id << "] error_code[" << ret << "]"; in DivInt8Run()
|
D | convolution_1x1_int8.h | 48 int OcRun(int task_id); 49 int HwRun(int task_id); 50 int OcOptPre(int task_id) const; 53 int RunArmOc(int task_id); 54 int RunArm64OptOc(int task_id); 55 int RunArmHw(int task_id); 56 int RunArm64OptHw(int task_id);
|
/third_party/mindspore/mindspore/_extends/parallel_compile/tbe_compiler/ |
D | tbe_process.py | 330 task_id = self.__next_task_id 337 self.__task_info[task_id] = op_json 347 self.__running_tasks.append((task_id, task_future)) 349 log.info("start_compile_op: task id: {} op json:\n {}".format(task_id, op_json)) 363 self.__all_tune_tasks.append(task_id) 364 self.__running_tune_tasks.append(task_id) 367 ret, job_type, compile_info = self.__tuner.rl_tune(task_id, op_json) 373 res = task_id, "Success", compile_info 375 self.__running_tune_tasks.remove(task_id) 378 res = task_id, "Fail", compile_info [all …]
|
/third_party/mindspore/mindspore/lite/src/runtime/kernel/arm/fp16_grad/ |
D | pooling_fp16_grad.cc | 75 int PoolingGradCPUKernelFp16::DoExecute(int task_id) { in DoExecute() argument 83 int count = MSMIN(stride, pool_param->output_batch_ - stride * task_id); in DoExecute() 87 …std::fill(output_ptr + task_id * stride * in_batch_size, output_ptr + ((task_id * stride) + count)… in DoExecute() 91 …MaxPoolingFp16Grad(input_ptr + task_id * stride * in_batch_size, dy_ptr + task_id * stride * out_b… in DoExecute() 92 output_ptr + task_id * stride * in_batch_size, count, pool_param); in DoExecute() 95 …AvgPoolingFp16Grad(input_ptr + task_id * stride * out_batch_size, output_ptr + task_id * stride * … in DoExecute() 102 int PoolingFp16GradImpl(void *cdata, int task_id, float lhs_scale, float rhs_scale) { in PoolingFp16GradImpl() argument 105 auto error_code = pooling->DoExecute(task_id); in PoolingFp16GradImpl() 107 … MS_LOG(ERROR) << "Pooling Run error task_id[" << task_id << "] error_code[" << error_code << "]"; in PoolingFp16GradImpl()
|
/third_party/mindspore/mindspore/lite/src/runtime/kernel/arm/fp32_grad/ |
D | pooling_grad.cc | 76 int PoolingGradCPUKernel::Execute(int task_id) { in Execute() argument 84 int count = MSMIN(stride, pool_param->output_batch_ - stride * task_id); in Execute() 88 …std::fill(output_ptr + task_id * stride * in_batch_size, output_ptr + ((task_id * stride) + count)… in Execute() 92 …MaxPoolingGrad(input_ptr + task_id * stride * in_batch_size, dy_ptr + task_id * stride * out_batch… in Execute() 93 output_ptr + task_id * stride * in_batch_size, count, pool_param); in Execute() 96 …AvgPoolingGrad(input_ptr + task_id * stride * out_batch_size, output_ptr + task_id * stride * in_b… in Execute() 103 int PoolingGradImpl(void *cdata, int task_id, float lhs_scale, float rhs_scale) { in PoolingGradImpl() argument 106 auto error_code = pooling->Execute(task_id); in PoolingGradImpl() 108 … MS_LOG(ERROR) << "Pooling Run error task_id[" << task_id << "] error_code[" << error_code << "]"; in PoolingGradImpl()
|
D | sgd.cc | 71 int SgdCPUKernel::Execute(int task_id) { in Execute() argument 84 int count = MSMIN(stride, length - stride * task_id); in Execute() 86 int start = stride * task_id; in Execute() 95 int SgdCPUKernel::ExecuteInit(int task_id) { in ExecuteInit() argument 110 int count = MSMIN(stride, length - stride * task_id); in ExecuteInit() 112 int start = stride * task_id; in ExecuteInit() 121 int SgdRun(void *cdata, int task_id, float lhs_scale, float rhs_scale) { in SgdRun() argument 126 error_code = sgd_kernel->ExecuteVirtualBatch(task_id); in SgdRun() 128 error_code = sgd_kernel->ExecuteVirtualBatch(task_id); in SgdRun() 130 error_code = sgd_kernel->Execute(task_id); in SgdRun() [all …]
|
/third_party/mindspore/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/base/ |
D | transpose_base.h | 45 const TransposeParameter *transpose_param, int task_id, int thread_num); 47 const TransposeParameter *transpose_param, int task_id, int thread_num); 49 const TransposeParameter *transpose_param, int task_id, int thread_num); 51 const TransposeParameter *transpose_param, int task_id, int thread_num); 53 const TransposeParameter *transpose_param, int task_id, int thread_num); 55 const TransposeParameter *transpose_param, int task_id, int thread_num); 57 const TransposeParameter *transpose_param, int task_id, int thread_num); 59 const TransposeParameter *transpose_param, int task_id, int thread_num);
|
/third_party/mindspore/mindspore/ccsrc/backend/kernel_compiler/cpu/nnacl/int8/ |
D | crop_int8.c | 21 void Int8Crop(const int8_t *input, int8_t *output, int task_id, const CropParameter *para) { in Int8Crop() argument 25 Int8Crop1D(input, output, task_id, para); in Int8Crop() 28 Int8Crop2D(input, output, task_id, para); in Int8Crop() 31 Int8Crop3D(input, output, task_id, para); in Int8Crop() 34 Int8Crop4D(input, output, task_id, para); in Int8Crop() 39 void Int8Crop1D(const int8_t *input, int8_t *output, int task_id, const CropParameter *para) { in Int8Crop1D() argument 54 int n = task_id * task_id_stride; in Int8Crop1D() 60 int64_t out_dist_stride = MSMIN(out_batch - task_id * task_id_stride, task_id_stride); in Int8Crop1D() 78 void Int8Crop2D(const int8_t *input, int8_t *output, int task_id, const CropParameter *para) { in Int8Crop2D() argument 96 int h = task_id * task_id_stride; in Int8Crop2D() [all …]
|