feat(fp4-gemm): portable SIMT NVFP4 GEMM fallback for SM110 (Thor) - #6
Closed
shideqin wants to merge 1 commit into
Closed
feat(fp4-gemm): portable SIMT NVFP4 GEMM fallback for SM110 (Thor)#6shideqin wants to merge 1 commit into
shideqin wants to merge 1 commit into
Conversation
Add pure-SIMT FMA references for the block-scaled NVFP4 fused-epilogue GEMMs (linear/residual/bias-GELU/fp4-out/stream-K) and route sm_110a to them; SM120 keeps the CUTLASS paths. Record the SM110 Thor validation and benchmark (the production GEMM uses the CUTLASS sm110 dispatch and is 42-94x faster than the dequantized PyTorch reference).
Author
|
Closed as superseded: |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Add pure-SIMT FMA references (
portable_fp4_gemm_simt.cu) for the block-scaled NVFP4 fused-epilogue GEMMs (linear/residual/bias-GELU/fp4-out/stream-K) and routesm_110ato them intorch_binding.cpp; SM120 keeps the CUTLASS paths.Validation (real hardware, NVIDIA Thor sm_110a)
nvfp4_gemm_bf16) uses the dedicated CUTLASSsm110_gemm_dispatchon SM110 and is 42-94x faster than the dequantized PyTorch reference with bitwise-equal output (max_abs=0.0). SIMT remains the fallback for fused-epilogue ops without an SM110 CUTLASS path.