leavelet/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

update 3.8 v2 (#2112)

Browse Source 
* update 3.8 v2

* update 3.8

---------

Co-authored-by: yuzhai <yuzhai@nvidia.com>
This commit is contained in:
Yujia Zhai
2025-02-19 19:03:14 -08:00
committed by GitHub
parent e9627ce55b
commit b84e9802d8
166 changed files with 3986 additions and 4037 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
CHANGELOG.md
View File

@@ -32,7 +32,9 @@
* CUTLASS library and profiler integration for block scaled data types for kernel emission, profiling, and verification.
- Support for preferred and fallback cluster shapes via profiler command line arguments parsing to set dynamic cluster shapes.
- Support for dynamic datatypes by parsing profiler via profiler command line arguments parsing to set dynamic datatype setting in TCGen05 MMA instruction descriptors.
- Support for mixed input GEMM kernels on Hopper in the profiler.
* New CUTLASS profiler flag `use-cuda-graphs` to reduce overheads when benchmarking launch-bound kernels.
* A new 3.x version of grouped GEMM to the CUTLASS library and generates kernels for Hopper and Blackwell. Now grouped GEMM support is enabled in the CUTLASS profiler (`./cutlass_profiler --operation=GroupedGemm --help` for details).
* Set of examples that demonstrate the usage of the 3.x API for targeting Blackwell SM100 architecture:
- [Basic FP16 and FP8 GEMMs with minimal changes from Hopper examples](./examples/70_blackwell_gemm/), demonstrating ease of migration for off the shelf kernels using the 3.x collective builder API.
- GEMM with [opt-in collective builder schedules showcasing available recipes](./examples/71_blackwell_gemm_with_collective_builder/71_blackwell_gemm_with_collective_builder.cu) for Blackwell.
@@ -45,12 +47,16 @@
- Grouped GEMM for [vanilla FP8 data inputs](./examples/75_blackwell_grouped_gemm/75_blackwell_grouped_gemm.cu) and [NVFP4 block scaled inputs](./examples/75_blackwell_grouped_gemm/75_blackwell_grouped_gemm_block_scaled.cu).
- Convolution kernels for [fprop](./examples/76_blackwell_conv/76_blackwell_conv_fprop.cu), [dgrad](./examples/76_blackwell_conv/76_blackwell_conv_dgrad.cu), and [wgrad](./examples/76_blackwell_conv/76_blackwell_conv_wgrad.cu).
- [Fused multi-head attention fprop kernel](./examples/77_blackwell_fmha/77_blackwell_fmha.cu) supporting fp16/bf16/fp8 data types across head dims of 32,64, and 128.
- A new BF16x9 GEMM [kernel](./examples/78_blackwell_emulated_bf16x9_gemm/78_blackwell_emulated_bf16x9_gemm.cu) that emulates FP32 GEMM (SGEMM) using BF16 operations.
* Set of examples that demonstrate the usage of the 3.x API for targeting Hopper architecture:
- A set of new [Hopper grouped GEMM kernels](./examples/69_hopper_mixed_dtype_grouped_gemm/) that support mixed A and B datatypes.
- A new [Hopper FP8 GEMM with groupwise scaling](./examples/67_hopper_fp8_warp_specialized_gemm_with_blockwise_scaling/67_hopper_fp8_warp_specialized_gemm_with_groupwise_scaling.cu).
* Documentation updates:
- [Quickstart - instantiating a Blackwell block-scaled GEMM](./media/docs/quickstart.md#instantiating-a-blackwell-gemm-kernel).
- Detailed [Blackwell block-scaled GEMM functionality documentation](./media/docs/blackwell_functionality.md)
- A new [functionality documentation](./media/docs/functionality.md) specifically for 3.x API comprehensively documenting all supported kernel types, data types, kernel features, minimum CUDA tookit support etc for 3.x supported architectures.
- Updates to [compatibility](./README.md#compatibility) section regarding supported compilers, operating systems, CUDA Toolkits, Hardware Architectures, and [Target Architecture](./README.md#Target-Architecture).
- Support grouped GEMM in the CUTLASS profiler (`./cutlass_profiler --operation=GroupedGemm --help` for details).
- Updates to [profiler documentation](./media/docs/profiler.md) for testing mixed input GEMM kernels on Hopper.
## [3.7.0](https://github.com/NVIDIA/cutlass/releases/tag/v3.7.0) (2025-01-11)
- [Hopper blockwise scaling FP8 GEMM](./examples/67_hopper_fp8_warp_specialized_gemm_with_blockwise_scaling/67_hopper_fp8_warp_specialized_gemm_with_blockwise_scaling.cu) uses 2D scaling tensor, assigning one value per threadblock. This allows a finer-grained scaling to be applied for each output tile per gemm-k iteration. The operands and scaling tensors are loaded from global memory to shared memory using TMA and cp_async, respectively. The scaling is applied inside the mainloop. Details with figures are [here](https://github.com/NVIDIA/cutlass/pull/1932#issue-2645398439).