Skip to content

refactor! : Update default workspace size based on platforms. #668

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

Merged
merged 1 commit into from
Oct 18, 2021
Merged

refactor! : Update default workspace size based on platforms. #668

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
16 changes: 15 additions & 1 deletion core/compiler.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -347,6 +347,21 @@ torch::jit::script::Module CompileGraph(const torch::jit::script::Module& mod, C
if (cfg.partition_info.enabled) {
return CompileGraphWithFallback(mod, cfg);
}
auto device_spec = cfg.convert_info.engine_settings.device;

// GPU default WS size : 1 GB
// Set WS = 256 Mb for Jetson nano/TX1 like platforms whose compute capability is 5.X.
auto workspace_size = cfg.convert_info.engine_settings.workspace_size;
cudaDeviceProp device_prop;
cudaGetDeviceProperties(&device_prop, device_spec.gpu_id);
if (workspace_size == 0) {
if (device_prop.major < 6) {
cfg.convert_info.engine_settings.workspace_size = 256 * (1 << 20);
} else {
cfg.convert_info.engine_settings.workspace_size = 1 << 30;
}
}

// TODO: Should be doing a functional transform but need PR #31978
// [jit] More robust mangling
// torch::jit::script::Module new_mod = mod.clone();
Expand All @@ -357,7 +372,6 @@ torch::jit::script::Module CompileGraph(const torch::jit::script::Module& mod, C
if (method.name().compare("forward") == 0) {
auto engine = ConvertGraphToTRTEngine(mod, method.name(), cfg);
auto new_g = std::make_shared<torch::jit::Graph>();
auto device_spec = cfg.convert_info.engine_settings.device;
auto cuda_device = runtime::CudaDevice(device_spec.gpu_id, device_spec.device_type);
AddEngineToGraph(new_mod, new_g, engine, cuda_device);
auto new_method = new_mod._ivalue()->compilation_unit()->create_function(method.name(), new_g);
Expand Down
2 changes: 1 addition & 1 deletion core/conversion/conversionctx/ConversionCtx.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -58,7 +58,7 @@ ConversionCtx::ConversionCtx(BuilderSettings build_settings)
net = make_trt(
builder->createNetworkV2(1U << static_cast<uint32_t>(nvinfer1::NetworkDefinitionCreationFlag::kEXPLICIT_BATCH)));

LOG_DEBUG(build_settings);
LOG_INFO(settings);
cfg = make_trt(builder->createBuilderConfig());

for (auto p = settings.enabled_precisions.begin(); p != settings.enabled_precisions.end(); ++p) {
Expand Down