Skip to content

Use CUDA driver APIs to avoid scheduling too large blocks #690

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
wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Use CUDA driver APIs to avoid scheduling too large blocks #690

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
5 changes: 3 additions & 2 deletions makefile
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -172,8 +172,9 @@ update-examples/%: examples/%.dx build
$(dex) script --allow-errors $< > $<.tmp
mv $<.tmp $<

run-gpu-tests: export DEX_ALLOC_CONTRACTIONS=0
run-gpu-tests: tests/gpu-tests.dx build
gpu-tests: run-gpu-tests/gpu-tests

run-gpu-tests/%: tests/%.dx build
misc/check-quine $< $(dex) --backend llvm-cuda script --allow-errors

update-gpu-tests: export DEX_ALLOW_CONTRACTIONS=0
Expand Down
16 changes: 11 additions & 5 deletions src/lib/dexrt.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -234,17 +234,23 @@ void dex_cuMemcpyHtoD(int64_t bytes, char* device_ptr, char* host_ptr) {
CHECK(cuMemcpyHtoD, reinterpret_cast<CUdeviceptr>(device_ptr), host_ptr, bytes);
}

void dex_queryParallelismCUDA(const char* kernel_func, int64_t iters,
void dex_queryParallelismCUDA(char* kernel_func, int64_t iters,
int32_t* numWorkgroups, int32_t* workgroupSize) {
if (iters == 0) {
*numWorkgroups = 0;
*workgroupSize = 0;
return;
}
// TODO: Use the occupancy calculator, or at least use a fixed number of blocks?
const int64_t fixedWgSize = 1024;
*workgroupSize = fixedWgSize;
*numWorkgroups = std::min((iters + fixedWgSize - 1) / fixedWgSize, fixedWgSize);
int min_grid_size_for_max_occupancy;
int block_size_32;
CUfunction kernel = reinterpret_cast<CUfunction>(kernel_func);
CHECK(cuOccupancyMaxPotentialBlockSize,
&min_grid_size_for_max_occupancy,
&block_size_32,
kernel, nullptr, 0, 0);
int64_t block_size = block_size_32;
*workgroupSize = block_size;
*numWorkgroups = (iters + block_size - 1) / block_size;
}

void dex_loadKernelCUDA(const char* kernel_text, char** module_storage, char** kernel_storage) {
Expand Down