Replace IsRunMKLDNNKernel IsRunONEDNNKernel [fluid_ops]

paddle/fluid/framework/infershape_utils.cc

@@ -556,7 +556,7 @@ CompatInferMetaContext BuildInferMetaContext(InferShapeContext* ctx,
 
  // 2. build infermeta context
  CompatInferMetaContext infer_meta_context(
- {ctx->IsRuntime(), ctx->IsRunMKLDNNKernel()});
+ {ctx->IsRuntime(), ctx->IsRunONEDNNKernel()});
 
  const auto& input_names = signature.input_names;
  const auto& attr_names = signature.attr_names;

paddle/fluid/framework/op_desc.cc

@@ -301,7 +301,7 @@ class CompileTimeInferShapeContext : public InferShapeContext {
 
  bool IsRuntime() const override;
 
- bool IsRunMKLDNNKernel() const override;
+ bool IsRunONEDNNKernel() const override;
 
  proto::VarType::Type GetInputVarType(const std::string &name) const override {
  return GetVarType(Inputs(name).at(0));
@@ -1349,7 +1349,7 @@ void CompileTimeInferShapeContext::SetRepeatedDims(
 
 bool CompileTimeInferShapeContext::IsRuntime() const { return false; }
 
-bool CompileTimeInferShapeContext::IsRunMKLDNNKernel() const { return false; }
+bool CompileTimeInferShapeContext::IsRunONEDNNKernel() const { return false; }
 
 proto::VarType::Type CompileTimeInferShapeContext::GetVarType(
  const std::string &name) const {

paddle/fluid/framework/operator.cc

@@ -526,7 +526,7 @@ void RuntimeInferShapeContext::SetLoDLevel(const std::string& out,
 
 bool RuntimeInferShapeContext::IsRuntime() const { return true; }
 
-bool RuntimeInferShapeContext::IsRunMKLDNNKernel() const {
+bool RuntimeInferShapeContext::IsRunONEDNNKernel() const {
  try {
  auto& op_with_kernel = dynamic_cast<const OperatorWithKernel&>(op_);
  return ((op_with_kernel.kernel_type()) &&
@@ -1441,7 +1441,7 @@ bool OperatorWithKernel::SupportCustomDevice() const {
 #endif
 }
 
-bool OperatorWithKernel::SupportsMKLDNN(const phi::DataType data_type) const {
+bool OperatorWithKernel::SupportsONEDNN(const phi::DataType data_type) const {
  auto phi_kernels = phi::KernelFactory::Instance().SelectKernelMap(
  phi::TransToPhiKernelName(type_));
  auto has_phi_kernel =
@@ -1578,7 +1578,7 @@ bool OperatorWithKernel::SupportsKernelType(
 // 3. Whether onednn kernel can be used.
 #ifdef PADDLE_WITH_DNNL
  if (!this->DnnFallback() && !paddle::platform::in_onednn_white_list(type_) &&
- this->CanMKLDNNBeUsed(exe_ctx, kernel_type.data_type_)) {
+ this->CanONEDNNBeUsed(exe_ctx, kernel_type.data_type_)) {
  auto tmp_kernel_type = kernel_type;
  tmp_kernel_type.library_type_ = framework::LibraryType::kMKLDNN;
  tmp_kernel_type.data_layout_ = framework::DataLayout::ONEDNN;
@@ -1597,15 +1597,15 @@ bool OperatorWithKernel::SupportsKernelType(
  return kernel_iter != kernels.end();
 }
 
-bool OperatorWithKernel::CanMKLDNNBeUsed(const framework::ExecutionContext& ctx,
+bool OperatorWithKernel::CanONEDNNBeUsed(const framework::ExecutionContext& ctx,
  phi::DataType data_type) const {
  return ctx.HasAttr("use_mkldnn") && ctx.Attr<bool>("use_mkldnn") &&
- phi::is_cpu_place(ctx.GetPlace()) && this->SupportsMKLDNN(data_type);
+ phi::is_cpu_place(ctx.GetPlace()) && this->SupportsONEDNN(data_type);
 }
 
-bool OperatorWithKernel::CanMKLDNNBeUsed(const framework::ExecutionContext& ctx,
+bool OperatorWithKernel::CanONEDNNBeUsed(const framework::ExecutionContext& ctx,
  proto::VarType::Type data_type) const {
- return this->CanMKLDNNBeUsed(ctx, phi::TransToPhiDataType(data_type));
+ return this->CanONEDNNBeUsed(ctx, phi::TransToPhiDataType(data_type));
 }
 
 bool OperatorWithKernel::CanCUDNNBeUsed(const framework::ExecutionContext& ctx,
@@ -1853,14 +1853,14 @@ void OperatorWithKernel::RunImpl(const Scope& scope,
 #ifdef PADDLE_WITH_DNNL
  if (!this->DnnFallback() &&
  !paddle::platform::in_onednn_white_list(type_) &&
- this->CanMKLDNNBeUsed(exe_ctx, kernel_type_->data_type_)) {
+ this->CanONEDNNBeUsed(exe_ctx, kernel_type_->data_type_)) {
  kernel_type_->library_type_ = framework::LibraryType::kMKLDNN;
  kernel_type_->data_layout_ = framework::DataLayout::ONEDNN;
  } else if (phi::is_cpu_place(kernel_type_->place_) &&
  kernel_type_->data_type_ ==
  proto::VarType::Type::VarType_Type_BF16 &&
  !this->SupportsCPUBF16() &&
- this->SupportsMKLDNN(phi::DataType::BFLOAT16)) {
+ this->SupportsONEDNN(phi::DataType::BFLOAT16)) {
  kernel_type_->library_type_ = framework::LibraryType::kMKLDNN;
  kernel_type_->data_layout_ = framework::DataLayout::ONEDNN;
  }
@@ -2172,14 +2172,14 @@ OpKernelType OperatorWithKernel::InnerGetExpectedKernelType(
 // 3. Whether onednn kernel can be used.
 #ifdef PADDLE_WITH_DNNL
  if (!this->DnnFallback() && !paddle::platform::in_onednn_white_list(type_) &&
- this->CanMKLDNNBeUsed(ctx, expected_kernel_key.data_type_)) {
+ this->CanONEDNNBeUsed(ctx, expected_kernel_key.data_type_)) {
  expected_kernel_key.library_type_ = framework::LibraryType::kMKLDNN;
  expected_kernel_key.data_layout_ = framework::DataLayout::ONEDNN;
  } else if (phi::is_cpu_place(expected_kernel_key.place_) &&
  expected_kernel_key.data_type_ ==
  proto::VarType::Type::VarType_Type_BF16 &&
  !this->SupportsCPUBF16() &&
- this->SupportsMKLDNN(phi::DataType::BFLOAT16)) {
+ this->SupportsONEDNN(phi::DataType::BFLOAT16)) {
  expected_kernel_key.library_type_ = framework::LibraryType::kMKLDNN;
  expected_kernel_key.data_layout_ = framework::DataLayout::ONEDNN;
  }

paddle/fluid/framework/operator.h

@@ -199,7 +199,7 @@ class RuntimeInferShapeContext : public InferShapeContext {
 
  bool IsRuntime() const override;
 
- bool IsRunMKLDNNKernel() const override;
+ bool IsRunONEDNNKernel() const override;
 
  // TODO(paddle-dev): Can this be template?
  paddle::small_vector<InferShapeVarPtr, phi::kInputSmallVectorSize>
@@ -768,7 +768,7 @@ class OperatorWithKernel : public OperatorBase {
 
  bool SupportCustomDevice() const override;
 
- bool SupportsMKLDNN(phi::DataType data_type) const;
+ bool SupportsONEDNN(phi::DataType data_type) const;
 
  bool SupportsCUDNN(phi::DataType data_type) const;
 
@@ -777,10 +777,10 @@ class OperatorWithKernel : public OperatorBase {
 
  bool SupportsCPUBF16() const;
 
- bool CanMKLDNNBeUsed(const framework::ExecutionContext& ctx,
+ bool CanONEDNNBeUsed(const framework::ExecutionContext& ctx,
  phi::DataType data_type) const;
 
- bool CanMKLDNNBeUsed(const framework::ExecutionContext& ctx,
+ bool CanONEDNNBeUsed(const framework::ExecutionContext& ctx,
  proto::VarType::Type data_type) const;
 
  bool CanCUDNNBeUsed(const framework::ExecutionContext& ctx,

paddle/fluid/framework/shape_inference.h

@@ -114,7 +114,7 @@ class InferShapeContext {
 
  virtual bool IsRuntime() const = 0;
 
- virtual bool IsRunMKLDNNKernel() const = 0;
+ virtual bool IsRunONEDNNKernel() const = 0;
 
  virtual paddle::small_vector<InferShapeVarPtr, phi::kInputSmallVectorSize>
  GetInputVarPtrs(const std::string &name) const = 0;

paddle/fluid/imperative/infer_shape_context.h

@@ -250,7 +250,7 @@ class DygraphInferShapeContext : public framework::InferShapeContext {
 
  bool IsRuntime() const override { return true; }
 
- bool IsRunMKLDNNKernel() const override {
+ bool IsRunONEDNNKernel() const override {
  return (op_kernel_key_ &&
  (op_kernel_key_->layout() == phi::DataLayout::ONEDNN));
  }

paddle/fluid/imperative/prepared_operator.cc

@@ -206,7 +206,7 @@ PreparedOp PrepareImpl(
 // 3. Whether onednn kernel can be used.
 #ifdef PADDLE_WITH_DNNL
  if (!op.DnnFallback() && !paddle::platform::in_onednn_white_list(op.Type()) &&
- op.CanMKLDNNBeUsed(dygraph_exe_ctx, expected_kernel_key.dtype())) {
+ op.CanONEDNNBeUsed(dygraph_exe_ctx, expected_kernel_key.dtype())) {
  expected_kernel_key.set_backend(phi::Backend::ONEDNN);
  expected_kernel_key.set_layout(phi::DataLayout::ONEDNN);
  }

paddle/fluid/operators/batch_norm_op.cc

@@ -107,10 +107,10 @@ void BatchNormOp::InferShape(framework::InferShapeContext *ctx) const {
  "= [%s], the dimension of input X = [%d]",
  x_dims,
  x_dims.size()));
- VLOG(4) << ctx->IsRunMKLDNNKernel();
+ VLOG(4) << ctx->IsRunONEDNNKernel();
  VLOG(4) << data_layout;
  const int64_t C =
- ((ctx->IsRunMKLDNNKernel() == true) || (data_layout == DataLayout::kNCHW)
+ ((ctx->IsRunONEDNNKernel() == true) || (data_layout == DataLayout::kNCHW)
  ? x_dims[1]
  : x_dims[x_dims.size() - 1]);
 
@@ -370,7 +370,7 @@ void BatchNormGradOp::InferShape(framework::InferShapeContext *ctx) const {
  common::StringToDataLayout(ctx->Attrs().Get<std::string>("data_layout"));
 
  const int C = static_cast<int>(
- ((ctx->IsRunMKLDNNKernel() == true) || (data_layout == DataLayout::kNCHW)
+ ((ctx->IsRunONEDNNKernel() == true) || (data_layout == DataLayout::kNCHW)
  ? x_dims[1]
  : x_dims[x_dims.size() - 1]));
 
@@ -511,7 +511,7 @@ void BatchNormDoubleGradOp::InferShape(
  const DataLayout data_layout =
  common::StringToDataLayout(ctx->Attrs().Get<std::string>("data_layout"));
  const int C = static_cast<int>(
- ((ctx->IsRunMKLDNNKernel() == true) || (data_layout == DataLayout::kNCHW)
+ ((ctx->IsRunONEDNNKernel() == true) || (data_layout == DataLayout::kNCHW)
  ? x_dims[1]
  : x_dims[x_dims.size() - 1]));
 

paddle/fluid/operators/elementwise/elementwise_op.h

@@ -113,7 +113,7 @@ class ElementwiseOp : public framework::OperatorWithKernel {
  // Broadcasting of dims has to be done on Paddle shapes (NHWC)
  // if model is using NHWC and any of shapes in at least 3D
  bool should_rotate =
- ctx->IsRunMKLDNNKernel() &&
+ ctx->IsRunONEDNNKernel() &&
  (phi::OneDNNContext::tls().get_cur_paddle_data_layout() ==
  phi::DataLayout::kNHWC) &&
  (x_dims.size() >= 3 || y_dims.size() >= 3);

paddle/fluid/operators/generator/get_expected_kernel_func.cc

@@ -324,7 +324,7 @@ phi::KernelKey GetPad3dExpectedKernelType(
  auto input_data_type = op_ptr->IndicateVarDataType(ctx, "X");
 #ifdef PADDLE_WITH_DNNL
  // only constant mode and non-blocked layouts are supported for oneDNN
- if (op_ptr->CanMKLDNNBeUsed(ctx, input_data_type) &&
+ if (op_ptr->CanONEDNNBeUsed(ctx, input_data_type) &&
  ctx.Attr<std::string>("mode") == "constant" &&
  ctx.Input<phi::DenseTensor>("X")->mem_desc().get_inner_nblks() == 0) {
  return phi::KernelKey(phi::Backend::ONEDNN,

paddle/fluid/operators/matmul_op.cc

@@ -82,7 +82,7 @@ class MatMulOp : public framework::OperatorWithKernel {
  // For NHWC execution output shape needs to be
  // computed like instead x*y we are to do y*x
  bool channelwise_onednn =
- context->IsRunMKLDNNKernel() &&
+ context->IsRunONEDNNKernel() &&
  (phi::OneDNNContext::tls().get_cur_paddle_data_layout() ==
  phi::DataLayout::kNHWC);
  if (channelwise_onednn) {

paddle/fluid/operators/slice_op.cc

@@ -157,7 +157,7 @@ class SliceOp : public framework::OperatorWithKernel {
  auto vec_dims = common::vectorize(in_tensor.dims());
  bool all_zero_dims = std::all_of(
  vec_dims.cbegin(), vec_dims.cend(), [](int64_t i) { return i == 0; });
- if (!all_zero_dims && this->CanMKLDNNBeUsed(ctx, input_data_type)) {
+ if (!all_zero_dims && this->CanONEDNNBeUsed(ctx, input_data_type)) {
  // OneDNN uses blocking format, which cannot be always supported with
  // reorders, because if blocked dimension is not divisible by 8 or
  // 16(depending on which blocking format is used) submemory cannot be
@@ -333,7 +333,7 @@ class SliceOpGrad : public framework::OperatorWithKernel {
  ctx, framework::GradVarName("Out"));
 
 #ifdef PADDLE_WITH_DNNL
- if (this->CanMKLDNNBeUsed(ctx, input_data_type)) {
+ if (this->CanONEDNNBeUsed(ctx, input_data_type)) {
  // OneDNN uses blocking format, which cannot be always supported with
  // reorders, because if blocked dimension is not divisible by 8 or
  // 16(depending on which blocking format is used) submemory cannot be

paddle/fluid/operators/split_op.cc

@@ -118,7 +118,7 @@ class SplitOp : public framework::OperatorWithKernel {
  framework::OperatorWithKernel::IndicateVarDataType(ctx, "X");
 
 #ifdef PADDLE_WITH_DNNL
- if (this->CanMKLDNNBeUsed(ctx, input_data_type)) {
+ if (this->CanONEDNNBeUsed(ctx, input_data_type)) {
  // OneDNN uses blocking format, which cannot be always supported with
  // reorders, because if blocked dimension is not divisible by 8 or
  // 16(depending on which blocking format is used) submemory cannot be

paddle/fluid/pir/transforms/pd_op_to_kernel_pass.cc

@@ -1057,7 +1057,7 @@ std::string GetKernelName(const OpYamlInfoParser* op_info_parser,
 }
 
 #ifdef PADDLE_WITH_DNNL
-bool SupportsMKLDNN(const std::string& kernel_name,
+bool SupportsONEDNN(const std::string& kernel_name,
  const phi::DataType data_type) {
  auto phi_kernels =
  phi::KernelFactory::Instance().SelectKernelMap(kernel_name);
@@ -1441,7 +1441,7 @@ phi::KernelKey GetKernelKey(
  elems.erase("");
 
  if (op->HasTrait<OneDNNTrait>() && res.backend() == phi::Backend::CPU &&
- SupportsMKLDNN(kernel_fn_str, res.dtype()) &&
+ SupportsONEDNN(kernel_fn_str, res.dtype()) &&
  elems.count(op->name().substr(
  strlen(OneDNNOperatorDialect::name()) + 1,
  op->name().size() - strlen(OneDNNOperatorDialect::name()) - 1)) ==
@@ -3613,7 +3613,7 @@ void ProcessBlock(
  if (kernel_key.dtype() == phi::DataType::BFLOAT16 &&
  kernel_key.backend() == phi::Backend::CPU &&
  !op_item->HasTrait<OneDNNTrait>() && !SupportsCPUBF16(kernel_name) &&
- SupportsMKLDNN(kernel_name, phi::DataType::BFLOAT16)) {
+ SupportsONEDNN(kernel_name, phi::DataType::BFLOAT16)) {
  auto op_item_inner = PdOp2OneDNNOp(op_item, block, ctx);
  if (op_item_inner != op_item) {
  op_item = op_item_inner;
@@ -3623,7 +3623,7 @@ void ProcessBlock(
  }
  } else if (FLAGS_use_mkldnn && kernel_key.backend() == phi::Backend::CPU &&
  !op_item->HasTrait<OneDNNTrait>() &&
- SupportsMKLDNN(kernel_name, kernel_key.dtype())) {
+ SupportsONEDNN(kernel_name, kernel_key.dtype())) {
  // Support FLAGS_use_mkldnn
  auto op_item_inner = PdOp2OneDNNOp(op_item, block, ctx);
  if (op_item_inner != op_item) {