leavelet/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 28 Packages Projects Releases Wiki Activity

v4.0 update. (#2371)

Browse Source
This commit is contained in:
Junkai-Wu
2025-06-06 02:39:20 -04:00
committed by GitHub
parent 2e2af190bd
commit 8bdbfca682
254 changed files with 29751 additions and 1980 deletions
Download Patch File Download Diff File Expand all files Collapse all files
examples/82_blackwell_distributed_gemm/82_blackwell_distributed_gemm.cu
+11 -11
View File
@@ -133,7 +133,7 @@ using TP = _8;
static constexpr int TP_ = TP{};
#if defined(CUTLASS_ARCH_MMA_SM100A_ENABLED) && \
(__CUDACC_VER_MAJOR__ > 12 || (__CUDACC_VER_MAJOR__ == 12 && __CUDACC_VER_MINOR__ >= 4))
(__CUDACC_VER_MAJOR__ > 12 || (__CUDACC_VER_MAJOR__ == 12 && __CUDACC_VER_MINOR__ >= 8))
// Distributed GEMM tiling/sharding schedule
// Choices:
@@ -254,7 +254,8 @@ HostTensorB tensor_B_arr[TP_];
HostTensorD tensor_C_arr[TP_];
HostTensorD tensor_D_arr[TP_];
#endif // (defined(CUTLASS_ARCH_MMA_SM100A_ENABLED) && (__CUDACC_VER_MAJOR__ >= 12) && (__CUDACC_VER_MINOR__ >= 4))
#endif // (defined(CUTLASS_ARCH_MMA_SM100A_ENABLED) &&
// (__CUDACC_VER_MAJOR__ > 12 || (__CUDACC_VER_MAJOR__ == 12 && __CUDACC_VER_MINOR__ >= 8))
/////////////////////////////////////////////////////////////////////////////////////////////////
/// Testbed utility types
@@ -347,7 +348,7 @@ struct Result {
};
#if defined(CUTLASS_ARCH_MMA_SM100A_ENABLED) && \
(__CUDACC_VER_MAJOR__ > 12 || (__CUDACC_VER_MAJOR__ == 12 && __CUDACC_VER_MINOR__ >= 4))
(__CUDACC_VER_MAJOR__ > 12 || (__CUDACC_VER_MAJOR__ == 12 && __CUDACC_VER_MINOR__ >= 8))
/////////////////////////////////////////////////////////////////////////////////////////////////
/// GEMM setup and evaluation
@@ -805,17 +806,16 @@ int run(Options &options) {
return 0;
}
#endif // (defined(CUTLASS_ARCH_MMA_SM100A_ENABLED) && (__CUDACC_VER_MAJOR__ >= 12) && (__CUDACC_VER_MINOR__ >= 4))
#endif // (defined(CUTLASS_ARCH_MMA_SM100A_ENABLED) &&
// (__CUDACC_VER_MAJOR__ > 12 || (__CUDACC_VER_MAJOR__ == 12 && __CUDACC_VER_MINOR__ >= 8))
///////////////////////////////////////////////////////////////////////////////////////////////////
int main(int argc, char const **args) {
// CUTLASS must be compiled with CUDA Toolkit 12.4 or newer to run this example
// and must have compute capability at least 90.
// Some necessary cuda graph APIs were only introduced in CUDA 12.4.
if (__CUDACC_VER_MAJOR__ < 12 || (__CUDACC_VER_MAJOR__ == 12 && __CUDACC_VER_MINOR__ < 4)) {
std::cerr << "This example requires CUDA 12.4 or newer." << std::endl;
// CUTLASS must be compiled with CUDA Toolkit 12.8 or newer to run Blackwell kernels.
if (__CUDACC_VER_MAJOR__ < 12 || (__CUDACC_VER_MAJOR__ == 12 && __CUDACC_VER_MINOR__ < 8)) {
std::cerr << "This example requires CUDA 12.8 or newer." << std::endl;
// Returning zero so this test passes on older Toolkits. Its actions are no-op.
return 0;
}
@@ -861,11 +861,11 @@ int main(int argc, char const **args) {
// Evaluate CUTLASS kernels
//
#if (defined(CUTLASS_ARCH_MMA_SM100A_ENABLED) && (__CUDACC_VER_MAJOR__ >= 12) && (__CUDACC_VER_MINOR__ >= 4))
#if (defined(CUTLASS_ARCH_MMA_SM100A_ENABLED) && (__CUDACC_VER_MAJOR__ > 12 || (__CUDACC_VER_MAJOR__ == 12 && __CUDACC_VER_MINOR__ >= 8)))
run(options);
#else
std::cerr
<< "This example must be compiled with `sm100a` and CUDA Toolkit 12.4 or later." << std::endl;
<< "This example must be compiled with `sm100a` and CUDA Toolkit 12.8 or later." << std::endl;
return 0;
#endif
examples/82_blackwell_distributed_gemm/REQUIREMENTS.md
+2 -2
View File
@@ -14,8 +14,8 @@ cmake $PATH -DCUTLASS_NVCC_ARCHS="100a" -DCUTLASS_ENABLE_GDC_FOR_SM100=1
### Minimum software
Like all other CUTLASS examples, the NVIDIA driver, runtime, and CUDA Toolkit are required.
This example specifically requires CUDA Toolkit 12.6 or newer, due to some of the necessary
CUDA graph APIs.
This example specifically requires CUDA Toolkit 12.8 or newer, since that is the first version
supporting the Blackwell architecture.
### Hardware / driver settings