Skip to content

[ROCm] triton: skip dequantize_nvfp4 #47

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
Apophis3158 wants to merge 1 commit into
base: main
Choose a base branch
from
Open

[ROCm] triton: skip dequantize_nvfp4 #47

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
38 changes: 22 additions & 16 deletions comfy_kitchen/backends/triton/__init__.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -43,6 +43,27 @@ def _build_constraints() -> dict:
triton_devices = frozenset({"cuda", "xpu"})
standard_floats = frozenset({torch.float32, torch.float16, torch.bfloat16})

# Skip on ROCm — the AMD backend has no `r` register and would crash
# with "error: couldn't allocate input reg for constraint 'r'".
sm_specific = {} if getattr(torch.version, "hip", None) else {
# Uses inline PTX: cvt.rn.f16x2.e2m1x2 (SM100/Blackwell instruction)
"dequantize_nvfp4": FunctionConstraints(
params={
"qx": ParamConstraint(
dtypes=frozenset({torch.uint8}),
shape_rules=(ExactDims(2),),
),
"per_tensor_scale": ParamConstraint(dtypes=frozenset({torch.float32})),
"block_scales": ParamConstraint(
dtypes=frozenset({torch.float8_e4m3fn}),
),
"output_type": ParamConstraint(dtypes=standard_floats),
},
default_devices=cuda_devices,
min_compute_capability=(10, 0), # SM100 required for cvt.rn.f16x2.e2m1x2
),
}

return {
"quantize_per_tensor_fp8": FunctionConstraints(
params={
Expand Down Expand Up @@ -83,22 +104,7 @@ def _build_constraints() -> dict:
},
default_devices=cuda_devices,
),
# Uses inline PTX: cvt.rn.f16x2.e2m1x2 (SM100/Blackwell instruction)
"dequantize_nvfp4": FunctionConstraints(
params={
"qx": ParamConstraint(
dtypes=frozenset({torch.uint8}),
shape_rules=(ExactDims(2),),
),
"per_tensor_scale": ParamConstraint(dtypes=frozenset({torch.float32})),
"block_scales": ParamConstraint(
dtypes=frozenset({torch.float8_e4m3fn}),
),
"output_type": ParamConstraint(dtypes=standard_floats),
},
default_devices=cuda_devices,
min_compute_capability=(10, 0), # SM100 required for cvt.rn.f16x2.e2m1x2
),
**sm_specific,
"apply_rope1": FunctionConstraints(
params={
"x": ParamConstraint(dtypes=standard_floats),
Expand Down