leavelet/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 1 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
2025-08-23 06:11:24 +08:00
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

2
tools/profiler/include/cutlass/profiler/block_scaled_gemm_operation_profiler.h
View File

@@ -180,6 +180,8 @@ public:
/// Buffer used for the cutlass reduction operations' host workspace
std::vector<uint8_t> reduction_host_workspace;
cudaStream_t stream;
};
protected:

15
tools/profiler/include/cutlass/profiler/blockwise_gemm_operation_profiler.h
View File

@@ -100,6 +100,13 @@ public:
cutlass::library::RasterOrder raster_order{cutlass::library::RasterOrder::kHeuristic};
int swizzle_size{1};
/// For profiling purposes
std::vector<gemm::GemmCoord> problem_sizes;
std::vector<std::array<int64_t, 3>> leading_dims;
std::vector<std::array<int64_t, 3>> preferred_clusters;
std::vector<std::array<int64_t, 3>> fallback_clusters;
std::vector<cutlass::library::RasterOrder> raster_orders;
std::vector<int> swizzle_sizes;
cutlass::library::RuntimeDatatype runtime_input_datatype_a{};
cutlass::library::RuntimeDatatype runtime_input_datatype_b{};
@@ -122,6 +129,14 @@ public:
ProblemSpace const &problem_space,
ProblemSpace::Problem const &problem);
int64_t bytes_with_problem_shape(
library::BlockwiseGemmDescription const &operation_desc,
gemm::GemmCoord const &problem_shape) const;
int64_t flops_with_problem_shape(
library::BlockwiseGemmDescription const &operation_desc,
gemm::GemmCoord const &problem_shape) const;
/// Total number of bytes loaded
int64_t bytes(library::BlockwiseGemmDescription const &operation_desc) const;

2
tools/profiler/include/cutlass/profiler/grouped_gemm_operation_profiler.h
View File

@@ -176,6 +176,8 @@ public:
std::vector<uint8_t> host_workspace;
DeviceAllocation device_workspace;
cudaStream_t stream;
};
private:

4
tools/profiler/include/cutlass/profiler/options.h
View File

@@ -346,6 +346,10 @@ public:
/// Vector of operation name substrings
std::vector<std::string> operation_names;
/// Map of problems to run for each operation
/// [operation_name] -> vector of problems, each problem specified as a vector of [argument name] -> [argument value]
std::unordered_map<std::string, std::vector<CommandLine>> operation_problems;
/// Vector of operation name substrings
std::vector<std::string> excluded_operation_names;