leavelet/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 24 Packages Projects Releases Wiki Activity

v4.2 release. (#2587)

Browse Source 
* 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
Download Patch File Download Diff File Expand all files Collapse all files
examples/80_blackwell_geforce_sparse_gemm/80a_blackwell_geforce_mxfp8_bf16_sparse_gemm.cu
+19 -10
View File
@@ -78,7 +78,7 @@
#include "helper.h"
using namespace cute;
#if defined(CUTLASS_ARCH_MMA_SM120_SUPPORTED)
#if defined(CUTLASS_ARCH_MMA_SM120_SUPPORTED) || defined(CUTLASS_ARCH_MMA_SM121_SUPPORTED)
/////////////////////////////////////////////////////////////////////////////////////////////////
/// GEMM kernel configurations
/////////////////////////////////////////////////////////////////////////////////////////////////
@@ -248,7 +248,7 @@ struct Result
avg_runtime_ms(avg_runtime_ms), gflops(gflops), status(status), error(error), passed(false)
{}
};
#if defined(CUTLASS_ARCH_MMA_SM120_SUPPORTED)
#if defined(CUTLASS_ARCH_MMA_SM120_SUPPORTED) || defined(CUTLASS_ARCH_MMA_SM121_SUPPORTED)
/////////////////////////////////////////////////////////////////////////////////////////////////
/// GEMM setup and evaluation
/////////////////////////////////////////////////////////////////////////////////////////////////
@@ -507,25 +507,34 @@ int run(Options &options)
}
return 0;
}
#endif // defined(CUTLASS_ARCH_MMA_SM120_SUPPORTED)
#endif // defined(CUTLASS_ARCH_MMA_SM120_SUPPORTED) || defined(CUTLASS_ARCH_MMA_SM121_SUPPORTED)
///////////////////////////////////////////////////////////////////////////////////////////////////
int main(int argc, char const **args) {
// CUTLASS must be compiled with CUDA 12.8 or higher Toolkit to run this example
// and must have compute capability at least 120.
// CUTLASS must be compiled with CUDA 12.8 or higher Toolkit for SM120 support,
// or CUDA 12.9 or higher for SM121 support.
// Must have compute capability at least 120.
#if defined(CUTLASS_ARCH_MMA_SM120_SUPPORTED)
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;
std::cerr << "This example requires CUDA 12.8 or newer for SM120 support." << std::endl;
// Returning zero so this test passes on older Toolkits. Its actions are no-op.
return 0;
}
#elif defined(CUTLASS_ARCH_MMA_SM121_SUPPORTED)
if (__CUDACC_VER_MAJOR__ < 12 || (__CUDACC_VER_MAJOR__ == 12 && __CUDACC_VER_MINOR__ < 9)) {
std::cerr << "This example requires CUDA 12.9 or newer for SM121 support." << std::endl;
// Returning zero so this test passes on older Toolkits. Its actions are no-op.
return 0;
}
#endif
cudaDeviceProp props;
int current_device_id;
CUDA_CHECK(cudaGetDevice(&current_device_id));
CUDA_CHECK(cudaGetDeviceProperties(&props, current_device_id));
if (!(props.major == 12 && props.minor == 0)) {
std::cerr << "This example requires a GPU of NVIDIA's Blackwell architecture (compute capability 120)." << std::endl;
if (!(props.major == 12 && (props.minor == 0 || props.minor == 1))) {
std::cerr << "This example requires a GPU of NVIDIA's Blackwell architecture (compute capability 120 or 121)." << std::endl;
return 0;
}
//
@@ -540,9 +549,9 @@ int main(int argc, char const **args) {
//
// Evaluate CUTLASS kernels
//
#if defined(CUTLASS_ARCH_MMA_SM120_SUPPORTED)
#if defined(CUTLASS_ARCH_MMA_SM120_SUPPORTED) || defined(CUTLASS_ARCH_MMA_SM121_SUPPORTED)
run<Gemm>(options);
#endif // defined(CUTLASS_ARCH_MMA_SM120_SUPPORTED)
#endif // defined(CUTLASS_ARCH_MMA_SM120_SUPPORTED) || defined(CUTLASS_ARCH_MMA_SM121_SUPPORTED)
return 0;
}
/////////////////////////////////////////////////////////////////////////////////////////////////
examples/80_blackwell_geforce_sparse_gemm/80b_blackwell_geforce_nvfp4_nvfp4_sparse_gemm.cu
+20 -11
View File
@@ -78,7 +78,7 @@
#include "helper.h"
using namespace cute;
#if defined(CUTLASS_ARCH_MMA_SM120_SUPPORTED)
#if defined(CUTLASS_ARCH_MMA_SM120_SUPPORTED) || defined(CUTLASS_ARCH_MMA_SM121_SUPPORTED)
/////////////////////////////////////////////////////////////////////////////////////////////////
/// GEMM kernel configurations
/////////////////////////////////////////////////////////////////////////////////////////////////
@@ -183,7 +183,7 @@ cutlass::HostTensor<outputScaleFactor, cutlass::layout::PackedVectorLayout> bloc
cutlass::HostTensor<ElementD, cutlass::layout::PackedVectorLayout> block_reference_D;
cutlass::HostTensor<outputScaleFactor, cutlass::layout::PackedVectorLayout> block_reference_SFD;
cutlass::HostTensor<ElementCompute, cutlass::layout::PackedVectorLayout> block_Normconst;
#endif // defined(CUTLASS_ARCH_MMA_SM120_SUPPORTED)
#endif // defined(CUTLASS_ARCH_MMA_SM120_SUPPORTED) || defined(CUTLASS_ARCH_MMA_SM121_SUPPORTED)
template <typename T>
auto make_iterator(T* ptr) {
return cute::recast_ptr<T>(ptr);
@@ -259,7 +259,7 @@ struct Result
avg_runtime_ms(avg_runtime_ms), gflops(gflops), status(status), error(error), passed(false)
{}
};
#if defined(CUTLASS_ARCH_MMA_SM120_SUPPORTED)
#if defined(CUTLASS_ARCH_MMA_SM120_SUPPORTED) || defined(CUTLASS_ARCH_MMA_SM121_SUPPORTED)
/////////////////////////////////////////////////////////////////////////////////////////////////
/// GEMM setup and evaluation
/////////////////////////////////////////////////////////////////////////////////////////////////
@@ -531,25 +531,34 @@ int run(Options &options)
}
return 0;
}
#endif // defined(CUTLASS_ARCH_MMA_SM120_SUPPORTED)
#endif // defined(CUTLASS_ARCH_MMA_SM120_SUPPORTED) || defined(CUTLASS_ARCH_MMA_SM121_SUPPORTED)
///////////////////////////////////////////////////////////////////////////////////////////////////
int main(int argc, char const **args) {
// CUTLASS must be compiled with CUDA 12.8 or higher Toolkit to run this example
// and must have compute capability at least 120.
// CUTLASS must be compiled with CUDA 12.8 or higher Toolkit for SM120 support,
// or CUDA 12.9 or higher for SM121 support.
// Must have compute capability at least 120.
#if defined(CUTLASS_ARCH_MMA_SM120_SUPPORTED)
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;
std::cerr << "This example requires CUDA 12.8 or newer for SM120 support." << std::endl;
// Returning zero so this test passes on older Toolkits. Its actions are no-op.
return 0;
}
#elif defined(CUTLASS_ARCH_MMA_SM121_SUPPORTED)
if (__CUDACC_VER_MAJOR__ < 12 || (__CUDACC_VER_MAJOR__ == 12 && __CUDACC_VER_MINOR__ < 9)) {
std::cerr << "This example requires CUDA 12.9 or newer for SM121 support." << std::endl;
// Returning zero so this test passes on older Toolkits. Its actions are no-op.
return 0;
}
#endif
cudaDeviceProp props;
int current_device_id;
CUDA_CHECK(cudaGetDevice(&current_device_id));
CUDA_CHECK(cudaGetDeviceProperties(&props, current_device_id));
if (!(props.major == 12 && props.minor == 0)) {
std::cerr << "This example requires a GPU of NVIDIA's Blackwell architecture (compute capability 120)." << std::endl;
if (!(props.major == 12 && (props.minor == 0 || props.minor == 1))) {
std::cerr << "This example requires a GPU of NVIDIA's Blackwell architecture (compute capability 120 or 121)." << std::endl;
return 0;
}
//
@@ -564,9 +573,9 @@ int main(int argc, char const **args) {
//
// Evaluate CUTLASS kernels
//
#if defined(CUTLASS_ARCH_MMA_SM120_SUPPORTED)
#if defined(CUTLASS_ARCH_MMA_SM120_SUPPORTED) || defined(CUTLASS_ARCH_MMA_SM121_SUPPORTED)
run<Gemm>(options);
#endif // defined(CUTLASS_ARCH_MMA_SM120_SUPPORTED)
#endif // defined(CUTLASS_ARCH_MMA_SM120_SUPPORTED) || defined(CUTLASS_ARCH_MMA_SM121_SUPPORTED)
return 0;
}
/////////////////////////////////////////////////////////////////////////////////////////////////
examples/80_blackwell_geforce_sparse_gemm/CMakeLists.txt
+1 -1
View File
@@ -27,7 +27,7 @@
# OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
if (CUTLASS_NVCC_ARCHS MATCHES 120a)
if (CUTLASS_NVCC_ARCHS MATCHES "120a|121a")
cutlass_example_add_executable(
80a_blackwell_geforce_mxfp8_bf16_sparse_gemm
80a_blackwell_geforce_mxfp8_bf16_sparse_gemm.cu