CUTLASS 3.5.1 (#1623)

* CUTLASS 3.5.1 * updates, optimizations, fixes
2024-07-29 08:46:24 -04:00
parent 56b46e2d13
commit be60a0b272
312 changed files with 19793 additions and 6775 deletions
@@ -143,16 +143,19 @@ void CutlassProfiler::enumerate_() {
 /// Profiles all operations
 int CutlassProfiler::profile_() {

-  int result = 0;
+  // Keep track of all device memory tensor in map
  DeviceContext device_context;
-  // For all profilers
+
+  int result = 0;
+  // For all profilers (e.g. gemm/sparse_gemm/conv2d...)
  for (auto & profiler : operation_profilers_) {

    if (options_.operation_kind == library::OperationKind::kInvalid ||
-      options_.operation_kind == profiler->kind()) {
+        options_.operation_kind == profiler->kind()) {

      result = profiler->profile_all(options_, library::Singleton::get().manifest, device_context);

+      // If some profile failed, terminate immediately
      if (result) {
        return result;
      }