[KernelGen][MThreads] Add special_gammainc Moore Threads specialized operator

[Yukun-Cui]

[KernelGen][MThreads] Add special_gammainc Moore Threads specialized operator

Summary

Add a Moore Threads (MUSA) specialized Triton kernel for special_gammainc, overriding the generic
implementation via runtime.replace_customized_ops(). Implements the special gammainc operation.

Testing

• Reused the existing upstream accuracy tests tests/test_special_gammainc.py (-m special_gammainc)
• Validated against reference on the MUSA device; specialization confirmed active via the
  GEMS_MTHREADS SPECIAL_GAMMAINC debug log
• Falls back to the generic implementation for unsupported dtype/device/shape
  (fp64/int64 are not supported on Moore Threads hardware)

Performance

Compared against the generic FlagGems implementation on Moore Threads (MUSA).

special_gammainc

dtype	Size	Torch Latency (ms)	Gems Latency (ms)	Speedup	TFLOPS
float32	1073741824	5504.079590	796.210022	6.913x	0.003
float32	64, 64	0.364200	0.090720	4.015x	0.000
float32	4096, 4096	86.118317	12.576401	6.848x	0.003
float32	64, 512, 512	86.612679	12.582640	6.884x	0.003
float32	1024, 1024, 1024	5506.935059	796.191467	6.917x	0.003
float32	1024, 1	0.367960	0.094920	3.877x	0.000
float32	1024, 16	0.401440	0.087040	4.612x	0.000
float32	1024, 256	1.465340	0.227680	6.436x	0.002
float32	1024, 4096	21.792620	3.249160	6.707x	0.003
float32	1024, 65536	344.749237	49.897400	6.909x	0.003
float32	64, 64, 1	0.361640	0.094320	3.834x	0.000
float32	64, 64, 16	0.396440	0.097960	4.047x	0.001
float32	64, 64, 256	5.551840	0.903320	6.146x	0.002
float32	64, 64, 4096	86.310165	12.580560	6.861x	0.003
float32	64, 64, 65536	1377.455444	199.172760	6.916x	0.003

Operator	Arithmetic Mean Speedup
special_gammainc	5.86x

Files Changed

• src/flag_gems/runtime/backend/_mthreads/ops/special_gammainc.py: Moore Threads Triton kernel + fallback
• src/flag_gems/runtime/backend/_mthreads/ops/__init__.py: Register import and __all__

[Dongxu-H]

/test | special_gammainc:mthreads

[github-actions]

❌ On-demand test failed

Operator: special_gammainc
Runner: mthreads
Backend: mthreads-musa520

The test failed to complete. Check the workflow run for details.

[Dongxu-H]

/test | special_gammainc_moore_threads_specialized:mthreads

[github-actions]

❌ On-demand test failed

Operator: special_gammainc_moore_threads_specialized
Runner: mthreads
Backend: mthreads-musa520

The test failed to complete. Check the workflow run for details.

[Dongxu-H]

/test | special_gammainc:mthreads

[github-actions]

⚠️ Cannot proceed with operation

• reviewDecision: skip_reviews
• commitStatus: FAILURE

Reviews are not required for this operation but CI checks must be passing in order to continue

[Dongxu-H]

/test | special_gammainc:mthreads

[github-actions]

⚠️ Cannot proceed with operation

• reviewDecision: skip_reviews
• commitStatus: FAILURE

Reviews are not required for this operation but CI checks must be passing in order to continue

[Dongxu-H]

/test | special_gammainc:mthreads

[github-actions]

⚠️ Cannot proceed with operation

• reviewDecision: skip_reviews
• commitStatus: FAILURE

Reviews are not required for this operation but CI checks must be passing in order to continue

[Yukun-Cui]

2026.07.27已被移出待特化列表