leavelet/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 27 Packages Projects Releases Wiki Activity

cutlass 3.9 update (#2255)

Browse Source 
* cutlass 3.9 update

* rebase

* fixes out of shared memory for blockwise Blackwell

* doc format

* fix issue 2253

* disable host ref by default

* fix sm120 smem capacity

---------

Co-authored-by: yuzhai <yuzhai@nvidia.com>
Co-authored-by: Haicheng Wu <haichengw@nvidia.com>
This commit is contained in:
Yujia Zhai
2025-04-24 15:42:40 -04:00
committed by GitHub
co-authored by yuzhai Haicheng Wu
parent 8e345c5c5b
commit 331a1f5b3f
143 changed files with 18089 additions and 5935 deletions
Download Patch File Download Diff File Expand all files Collapse all files
include/cute/arch/config.hpp
+60 -2
View File
@@ -72,6 +72,27 @@
# define CUTE_ARCH_TCGEN05_F16BF16_MMA_SCALED_ENABLED
#endif
#if (defined(CUTLASS_ARCH_MMA_SM100F_ENABLED) || defined(CUTLASS_ARCH_MMA_SM101F_ENABLED))
# define CUTE_ARCH_TMA_SM90_ENABLED
# define CUTE_ARCH_DEVICE_MODIFIABLE_TMA_SM90_ENABLED
# define CUTE_ARCH_STSM_SM90_ENABLED
# define CUTE_ARCH_TCGEN05_TF32_MMA_ENABLED
# define CUTE_ARCH_TCGEN05_F16F32_MMA_ENABLED
# define CUTE_ARCH_TCGEN05_MXF8F6F4_MMA_ENABLED
# define CUTE_ARCH_TCGEN05_MXF4_MMA_ENABLED
# define CUTE_ARCH_TCGEN05_MXF4NVF4_MMA_ENABLED
#endif
#if defined(CUTLASS_ARCH_MMA_SM100F_ENABLED)
# define CUTE_ARCH_TCGEN05_F16BF16_MMA_SCALED_ENABLED
#endif
#if (defined(CUTLASS_ARCH_MMA_SM120F_ENABLED))
# define CUTE_ARCH_TMA_SM90_ENABLED
# define CUTE_ARCH_DEVICE_MODIFIABLE_TMA_SM90_ENABLED
# define CUTE_ARCH_STSM_SM90_ENABLED
#endif
#if (defined(CUTLASS_ARCH_MMA_SM100A_ENABLED) || defined(CUTLASS_ARCH_MMA_SM101A_ENABLED))
# define CUTE_ARCH_TCGEN05_S8_MMA_ENABLED
#endif
@@ -91,8 +112,11 @@
#endif
// {add, mul, fma}.f32x2 PTX
#if (defined(CUTLASS_ARCH_MMA_SM100A_ENABLED))
#define CUTE_ARCH_FLOAT2_MATH_ENABLED
#if defined(CUTLASS_ARCH_MMA_SM100_ENABLED) || defined(CUTLASS_ARCH_MMA_SM100A_ENABLED)
// Enable CuTe MMA Atoms
# define CUTE_ARCH_FFMA2_SM100_ENABLED
// Enable f32x2 PTX generation
# define CUTE_ARCH_FLOAT2_MATH_ENABLED
#endif
#if defined(CUTLASS_ARCH_MMA_SM120_ENABLED) || defined(CUTLASS_ARCH_MMA_SM120A_ENABLED)
@@ -109,3 +133,37 @@
# endif
#endif
#if defined(CUTLASS_ARCH_MMA_SM100F_ENABLED)
# define CUTE_ARCH_LDSM_SM100A_ENABLED
# define CUTE_ARCH_STSM_SM100A_ENABLED
# define CUTE_ARCH_TCGEN05_TMEM_ENABLED
# define CUTE_ARCH_TMA_SM100_ENABLED
# define CUTE_ARCH_FLOAT2_MATH_ENABLED
#endif
#if defined(CUTLASS_ARCH_MMA_SM101F_ENABLED)
# define CUTE_ARCH_LDSM_SM100A_ENABLED
# define CUTE_ARCH_STSM_SM100A_ENABLED
# define CUTE_ARCH_TCGEN05_TMEM_ENABLED
# define CUTE_ARCH_TMA_SM100_ENABLED
#endif
#if defined(CUTLASS_ARCH_MMA_SM120F_ENABLED)
# define CUTE_ARCH_LDSM_SM100A_ENABLED
# define CUTE_ARCH_STSM_SM100A_ENABLED
#endif
#if (defined(CUTLASS_ARCH_MMA_SM100A_ENABLED) || defined(CUTLASS_ARCH_MMA_SM100F_ENABLED) ||\
defined(CUTLASS_ARCH_MMA_SM101A_ENABLED) || defined(CUTLASS_ARCH_MMA_SM101F_ENABLED) ||\
defined(CUTLASS_ARCH_MMA_SM120A_ENABLED) || defined(CUTLASS_ARCH_MMA_SM120F_ENABLED))
# if (__CUDACC_VER_MAJOR__ > 12 || (__CUDACC_VER_MAJOR__ == 12 && __CUDACC_VER_MINOR__ >= 9))
# define CUTE_ARCH_LOAD256_SM100A_ENABLED
# define CUTE_ARCH_STORE256_SM100A_ENABLED
# endif
#endif
// {add, mul, fma}.f32x2 PTX
#if defined(CUTLASS_ARCH_MMA_SM100A_ENABLED) || defined(CUTLASS_ARCH_MMA_SM100F_ENABLED)
#define CUTE_ARCH_FLOAT2_MATH_ENABLED
#endif