CUTLASS 3.0.0 (#786)

* CUTLASS 3.0.0
2023-01-23 20:55:28 -05:00
parent 66d9cddc83
commit 277bd6e537
377 changed files with 76396 additions and 1186 deletions
@@ -525,7 +525,7 @@ public:

    // Broadcast the warp_id computed by lane 0 to ensure dependent code
    // is compiled as warp-uniform.
-    int warp_idx = __shfl_sync(0xffffffff, threadIdx.x / 32, 0);
+    int warp_idx = canonical_warp_idx();

    int lane_idx = threadIdx.x % 32;