Skip to content

Fix warp_size in triton kernel for AMD GPUs #476

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 3 commits into
base: main
Choose a base branch
from
Open

Fix warp_size in triton kernel for AMD GPUs #476

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
41a849f
Fix warp_size in triton kernel for AMD GPUs
divakar-amd Nov 20, 2025
467979c
fix pre-commit
divakar-amd Nov 22, 2025
eafd4db
handle AMD Navi GPUs
divakar-amd Nov 23, 2025
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
9 changes: 8 additions & 1 deletion python/xgrammar/kernels/apply_token_bitmask_inplace_triton.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -86,6 +86,13 @@ def apply_token_bitmask_inplace_triton(
NUM_SMS = torch.cuda.get_device_properties("cuda").multi_processor_count
BLOCK_SIZE = 4096

arch = torch.cuda.get_device_properties(0).gcnArchName
if torch.version.hip is not None and "gfx1" not in arch:
# For AMD GPUs (non-Navi)
WARP_SIZE = 64
else:
WARP_SIZE = 32

assert bitmask.dtype == torch.int32, "bitmask must be of type int32"

detected_vocab_size = min(logits.shape[-1], bitmask.shape[-1] * 32)
Expand Down Expand Up @@ -114,6 +121,6 @@ def apply_token_bitmask_inplace_triton(
bitmask.stride()[0],
NUM_SMS,
BLOCK_SIZE,
num_warps=BLOCK_SIZE // 32 // (16 // logits.element_size()),
num_warps=BLOCK_SIZE // WARP_SIZE // (16 // logits.element_size()),
num_stages=3,
)
Loading