Skip to content

do not modify output at same cases #5

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
Xinyu302 wants to merge 4 commits into
base: auto_mixed_precision
Choose a base branch
from
Open

do not modify output at same cases #5

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
6893257
do not modify output at same cases
Xinyu302 Jan 16, 2024
af97319
auto_mixed_precision ca run
Xinyu302 Jan 20, 2024
909b6fc
Merge remote-tracking branch 'upstream/develop' into auto_mixed_preci…
Xinyu302 Jan 20, 2024
bbe99e9
delete useless code
Xinyu302 Jan 20, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 8 additions & 0 deletions paddle/fluid/framework/ir/auto_mixed_precision_pass.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -26,6 +26,8 @@
#include "paddle/phi/backends/device_manager.h"
#endif

PHI_DECLARE_bool(enable_pir_in_executor);

namespace paddle {
namespace framework {
namespace ir {
Expand All @@ -40,6 +42,9 @@ bool PhiKernelSupportPrecision(
phi::DataType data_type,
phi::DataLayout layout = phi::DataLayout::ALL_LAYOUT) {
const auto& kernels = phi::KernelFactory::Instance().kernels();
// for (auto [k, v] : kernels) {
// LOG(INFO) << "kernel name " << k << std::endl;
// }
if (kernels.count(op_type) == 0) {
return false;
}
Expand Down Expand Up @@ -270,6 +275,9 @@ void AutoMixedPrecisionPass::Init(Graph* graph) const {
}

void AutoMixedPrecisionPass::ApplyImpl(Graph* graph) const {
if (FLAGS_enable_pir_in_executor) {
return;
}
PADDLE_ENFORCE_NOT_NULL(graph,
platform::errors::PreconditionNotMet(
"During the auto_mixed_precision_pass, the graph "
Expand Down
19 changes: 16 additions & 3 deletions paddle/fluid/inference/api/analysis_predictor.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -105,6 +105,7 @@
#endif

#include "paddle/fluid/ir_adaptor/translator/translate.h"
#include "paddle/fluid/pir/transforms/auto_mixed_precision_pass.h"
#include "paddle/fluid/pir/transforms/constant_folding_pass.h"
#include "paddle/fluid/pir/transforms/dead_code_elimination_pass.h"
#include "paddle/fluid/pir/transforms/fusion/conv2d_add_act_fuse_pass.h"
Expand Down Expand Up @@ -811,6 +812,20 @@ bool AnalysisPredictor::PrepareExecutor() {
gpu_pm.AddPass(::pir::CreateMatmulScaleFusePass());
//----------------------------------------------------------------------------------------------//

//----------------------------------------------------------------------------------------------//
// Functional pass
// Do auto mixed precision pass first, so do not need to handle
// shadowoutput.
auto auto_mixed_precision_pass = ::pir::CreateAutoMixedPrecisionPass();
auto_mixed_precision_pass->SetNotOwned(pir::kPlaceAttr, &place_);
phi::DataType data_type =
ConvertPrecision(config_.mixed_precision_mode_);
auto_mixed_precision_pass->SetNotOwned("__mixed_precision_mode__",
&data_type);
gpu_pm.AddPass(std::move(auto_mixed_precision_pass));
gpu_pm.AddPass(::pir::CreateIdentityOpCleanPass());
//----------------------------------------------------------------------------------------------//

//----------------------------------------------------------------------------------------------//
// Basic pass required by the framework
auto params_sync_among_devices_pass =
Expand All @@ -831,9 +846,7 @@ bool AnalysisPredictor::PrepareExecutor() {
if (!config_.glog_info_disabled()) {
gpu_pm.EnablePrintStatistics();
}
if (config_.ir_debug_) {
gpu_pm.EnableIRPrinting();
}
gpu_pm.EnableIRPrinting();
gpu_pm.Run(pir_program_.get());
} else {
::pir::PassManager cpu_pm(::pir::IrContext::Instance(), 2);
Expand Down
Loading