v4.2 release. (#2587)

* Fix default cluster callback values to 1 to avoid profiler failure when these values are not set in command line.

* v4.2 release.
This commit is contained in:
Junkai-Wu
2025-08-22 18:11:24 -04:00
committed by GitHub
parent 11cad1f67b
commit a49a78ffef
351 changed files with 28182 additions and 2032 deletions
+60 -14
View File
@@ -54,13 +54,15 @@
////////////////////////////////////////////////////////////////////////////////////////////////////
#if (defined(CUTLASS_ARCH_MMA_SM100A_ENABLED) || defined(CUTLASS_ARCH_MMA_SM101A_ENABLED) ||\
defined(CUTLASS_ARCH_MMA_SM120A_ENABLED))
defined(CUTLASS_ARCH_MMA_SM103A_ENABLED) || defined(CUTLASS_ARCH_MMA_SM120A_ENABLED) ||\
defined(CUTLASS_ARCH_MMA_SM121A_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))
#if (defined(CUTLASS_ARCH_MMA_SM100A_ENABLED) || defined(CUTLASS_ARCH_MMA_SM101A_ENABLED) ||\
defined(CUTLASS_ARCH_MMA_SM103A_ENABLED))
# define CUTE_ARCH_TCGEN05_TF32_MMA_ENABLED
# define CUTE_ARCH_TCGEN05_F16F32_MMA_ENABLED
# define CUTE_ARCH_TCGEN05_MXF8F6F4_MMA_ENABLED
@@ -68,11 +70,12 @@
# define CUTE_ARCH_TCGEN05_MXF4NVF4_MMA_ENABLED
#endif
#if defined(CUTLASS_ARCH_MMA_SM100A_ENABLED)
#if defined(CUTLASS_ARCH_MMA_SM100A_ENABLED) || defined(CUTLASS_ARCH_MMA_SM103A_ENABLED)
# define CUTE_ARCH_TCGEN05_F16BF16_MMA_SCALED_ENABLED
#endif
#if (defined(CUTLASS_ARCH_MMA_SM100F_ENABLED) || defined(CUTLASS_ARCH_MMA_SM101F_ENABLED))
#if (defined(CUTLASS_ARCH_MMA_SM100F_ENABLED) || defined(CUTLASS_ARCH_MMA_SM101F_ENABLED) ||\
defined(CUTLASS_ARCH_MMA_SM103F_ENABLED))
# define CUTE_ARCH_TMA_SM90_ENABLED
# define CUTE_ARCH_DEVICE_MODIFIABLE_TMA_SM90_ENABLED
# define CUTE_ARCH_STSM_SM90_ENABLED
@@ -83,32 +86,59 @@
# define CUTE_ARCH_TCGEN05_MXF4NVF4_MMA_ENABLED
#endif
#if defined(CUTLASS_ARCH_MMA_SM100F_ENABLED)
#if defined(CUTLASS_ARCH_MMA_SM100F_ENABLED) || defined(CUTLASS_ARCH_MMA_SM103F_ENABLED)
# define CUTE_ARCH_TCGEN05_F16BF16_MMA_SCALED_ENABLED
#endif
#if (defined(CUTLASS_ARCH_MMA_SM120F_ENABLED))
#if (defined(CUTLASS_ARCH_MMA_SM120F_ENABLED) || defined(CUTLASS_ARCH_MMA_SM121F_ENABLED))
# define CUTE_ARCH_TMA_SM90_ENABLED
# define CUTE_ARCH_DEVICE_MODIFIABLE_TMA_SM90_ENABLED
# define CUTE_ARCH_STSM_SM90_ENABLED
#endif
// SM110 specific configs
#if (defined(CUTLASS_ARCH_MMA_SM110A_ENABLED) || defined(CUTLASS_ARCH_MMA_SM110F_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
# define CUTE_ARCH_TCGEN05_S8_MMA_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_LOAD256_SM100A_ENABLED
# define CUTE_ARCH_STORE256_SM100A_ENABLED
# define CUTE_ARCH_FLOAT2_MATH_ENABLED
#endif
#if (defined(CUTLASS_ARCH_MMA_SM110A_ENABLED))
# define CUTE_ARCH_TCGEN05_S8_MMA_ENABLED
#endif
#if (defined(CUTLASS_ARCH_MMA_SM100A_ENABLED) || defined(CUTLASS_ARCH_MMA_SM101A_ENABLED))
# define CUTE_ARCH_TCGEN05_S8_MMA_ENABLED
#endif
#if (defined(CUTLASS_ARCH_MMA_SM100A_ENABLED) || defined(CUTLASS_ARCH_MMA_SM101A_ENABLED) ||\
defined(CUTLASS_ARCH_MMA_SM120A_ENABLED))
defined(CUTLASS_ARCH_MMA_SM103A_ENABLED) || defined(CUTLASS_ARCH_MMA_SM120A_ENABLED) ||\
defined(CUTLASS_ARCH_MMA_SM120A_ENABLED) || defined(CUTLASS_ARCH_MMA_SM121A_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_SM101A_ENABLED))
#if (defined(CUTLASS_ARCH_MMA_SM100A_ENABLED) || defined(CUTLASS_ARCH_MMA_SM101A_ENABLED) ||\
defined(CUTLASS_ARCH_MMA_SM103A_ENABLED))
# define CUTE_ARCH_TCGEN05_TMEM_ENABLED
#endif
#if (defined(CUTLASS_ARCH_MMA_SM100A_ENABLED) || defined(CUTLASS_ARCH_MMA_SM101A_ENABLED))
#if (defined(CUTLASS_ARCH_MMA_SM100A_ENABLED) || defined(CUTLASS_ARCH_MMA_SM101A_ENABLED) ||\
defined(CUTLASS_ARCH_MMA_SM103A_ENABLED))
# define CUTE_ARCH_TMA_SM100_ENABLED
#endif
@@ -120,12 +150,13 @@
# define CUTE_ARCH_FLOAT2_MATH_ENABLED
#endif
#if defined(CUTLASS_ARCH_MMA_SM120_ENABLED) || defined(CUTLASS_ARCH_MMA_SM120A_ENABLED)
#if (defined(CUTLASS_ARCH_MMA_SM120_ENABLED) || defined(CUTLASS_ARCH_MMA_SM120A_ENABLED) ||\
defined(CUTLASS_ARCH_MMA_SM121_ENABLED) || defined(CUTLASS_ARCH_MMA_SM121A_ENABLED))
# define CUTE_ARCH_MMA_SM120_ENABLED
# define CUTE_ARCH_TMA_SM120_ENABLED
#endif
#if defined(CUTLASS_ARCH_MMA_SM120_ENABLED) || defined(CUTLASS_ARCH_MMA_SM120A_ENABLED)
#if (defined(CUTLASS_ARCH_MMA_SM120_ENABLED) || defined(CUTLASS_ARCH_MMA_SM120A_ENABLED))
# if (__CUDACC_VER_MAJOR__ > 12 || (__CUDACC_VER_MAJOR__ == 12 && __CUDACC_VER_MINOR__ >= 8))
# define CUTE_ARCH_F8F6F4_MMA_ENABLED
# define CUTE_ARCH_MXF8F6F4_MMA_ENABLED
@@ -134,7 +165,16 @@
# endif
#endif
#if defined(CUTLASS_ARCH_MMA_SM100F_ENABLED)
#if (defined(CUTLASS_ARCH_MMA_SM121_ENABLED) || defined(CUTLASS_ARCH_MMA_SM121A_ENABLED))
# if (__CUDACC_VER_MAJOR__ > 12 || (__CUDACC_VER_MAJOR__ == 12 && __CUDACC_VER_MINOR__ >= 9))
# define CUTE_ARCH_F8F6F4_MMA_ENABLED
# define CUTE_ARCH_MXF8F6F4_MMA_ENABLED
# define CUTE_ARCH_MXF4NVF4_2X_UE8M0_MMA_ENABLED
# define CUTE_ARCH_MXF4NVF4_4X_UE4M3_MMA_ENABLED
# endif
#endif
#if defined(CUTLASS_ARCH_MMA_SM100F_ENABLED) || defined(CUTLASS_ARCH_MMA_SM103F_ENABLED)
# define CUTE_ARCH_LDSM_SM100A_ENABLED
# define CUTE_ARCH_STSM_SM100A_ENABLED
# define CUTE_ARCH_TCGEN05_TMEM_ENABLED
@@ -149,14 +189,16 @@
# define CUTE_ARCH_TMA_SM100_ENABLED
#endif
#if defined(CUTLASS_ARCH_MMA_SM120F_ENABLED)
#if (defined(CUTLASS_ARCH_MMA_SM120F_ENABLED) || defined(CUTLASS_ARCH_MMA_SM121F_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))
defined(CUTLASS_ARCH_MMA_SM103A_ENABLED) || defined(CUTLASS_ARCH_MMA_SM103F_ENABLED) ||\
defined(CUTLASS_ARCH_MMA_SM120A_ENABLED) || defined(CUTLASS_ARCH_MMA_SM120F_ENABLED) ||\
defined(CUTLASS_ARCH_MMA_SM121A_ENABLED) || defined(CUTLASS_ARCH_MMA_SM121F_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
@@ -168,3 +210,7 @@
#define CUTE_ARCH_FLOAT2_MATH_ENABLED
#endif
#if defined(CUTLASS_ARCH_MMA_SM103_ENABLED) || defined(CUTLASS_ARCH_MMA_SM100F_ENABLED)
# define CUTE_ARCH_TCGEN05_MXF4NVF4_MMA_ULTRA_ENABLED
#endif