[NewIR]refactor pd op to kernel pass #56984
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
你的PR提交成功，感谢你对开源项目的贡献! |
|
❌ The PR is not created using PR's template. You can refer to this Demo. |
… refactor_pd_op_to_kernel_pass
phlrain
changed the title
… refactor_pd_op_to_kernel_pass
… refactor_pd_op_to_kernel_pass
… refactor_pd_op_to_kernel_pass
… refactor_pd_op_to_kernel_pass
zhangbo9674
approved these changes
Sep 12, 2023
Aurelius84
reviewed
Sep 12, 2023
| @@ -255,7 +101,7 @@ bool NeedFallBackCpu(const pir::Operation* op, | |||
|
|
|||
| phi::Backend GetDstBackend(const std::string& op_name, | |||
| phi::Place place, | |||
| OpYamlInfoParser* op_yaml_info_parser, | |||
| const OpYamlInfoParser* op_yaml_info_parser, | |||
There was a problem hiding this comment.
phi::Backend GetDstBackend(const std::string& op_name,
const phi::Place& place,
const OpYamlInfoParser* op_yaml_info_parser,
const phi::Backend& kernel_def_backend,
size_t input_index) {对于const的变量，相对于指针类型，我们更推荐引用，应为引用永远是「合法的」。
| std::stringstream ss; | ||
| prog->Print(ss); | ||
| VLOG(2) << "Program after lowering to kernel pass : " << ss.str(); | ||
| if (op->name() == "pd_op.load_combine") { |
There was a problem hiding this comment.
对于op name 我们更推荐使用 paddle::dialect::LoadCombineOp::name()。其实更好的方式是: op->isa<LoadCombineOp>()
| const pir::Value cur_in, | ||
| const std::unordered_map<pir::Value, pir::OpResult>& map_value_pair, | ||
| const int index, | ||
| const std::string op_name) { |
There was a problem hiding this comment.
Suggested change
| const std::string op_name) { | |
| const std::string& op_name) { |
| std::unordered_map<pir::Value, pir::OpResult>* map_value_pair) { | ||
| std::vector<pir::OpResult> vec_inputs; | ||
| std::vector<pir::Type> op_output_types; | ||
| if (op_item->name() == "builtin.combine") { |
|
|
||
| if (NeedFallBackFromGPUDNN2GPU(op_item, kernel_key)) { | ||
| kernel_key.set_backend(phi::Backend::GPU); | ||
| pir::OpInfo op_info = ctx->GetRegisteredOpInfo(op_item->name()); |
There was a problem hiding this comment.
Suggested change
| pir::OpInfo op_info = ctx->GetRegisteredOpInfo(op_item->name()); | |
| pir::OpInfo& op_info = ctx->GetRegisteredOpInfo(op_item->name()); |
| std::vector<pir::Type> op_output_types; | ||
| auto phi_kernel = phi::KernelFactory::Instance().SelectKernelWithGPUDNN( | ||
| kernel_fn_str, kernel_key); | ||
| auto args_def = phi_kernel.args_def(); |
There was a problem hiding this comment.
Suggested change
| auto args_def = phi_kernel.args_def(); | |
| auto& args_def = phi_kernel.args_def(); |
| auto phi_kernel = phi::KernelFactory::Instance().SelectKernelWithGPUDNN( | ||
| kernel_fn_str, kernel_key); | ||
| auto args_def = phi_kernel.args_def(); | ||
| auto output_defs = args_def.output_defs(); |
There was a problem hiding this comment.
Suggested change
| auto output_defs = args_def.output_defs(); | |
| auto& output_defs = args_def.output_defs(); |
| pir::Operation* op_item, | ||
| const std::string& kernel_fn_str, | ||
| const phi::KernelKey& kernel_key, | ||
| const phi::Place place, |
There was a problem hiding this comment.
Suggested change
| const phi::Place place, | |
| const phi::Place& place, |
|
|
||
| // get input args def type | ||
| auto args_def = kernel.args_def(); | ||
| auto input_defs = args_def.input_defs(); |
There was a problem hiding this comment.
Suggested change
| auto input_defs = args_def.input_defs(); | |
| auto& input_defs = args_def.input_defs(); |
| std::unordered_map<pir::Operation*, pir::Operation*>* map_op_pair, | ||
| std::unordered_map<pir::Value, pir::OpResult>* map_value_pair) { | ||
| bool feed_op_add_shadow_feed = | ||
| (op_item->name() == "pd_op.feed") && platform::is_gpu_place(place); |
danleifeng
pushed a commit
to danleifeng/Paddle
that referenced
this pull request
Nov 14, 2023
* refactor pd op to kernel pass * fix bug * update * add code * fix merge bug * fix bug * update * fix bug * polish code
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
PR types
Others
PR changes
Others
Description
对于pd op to kernel pass 做代码结构的调整，进行模块化划分，方便控制流的接入
Pcard-67164