fix: cute.copy(dst, src) order

2026-05-23 03:51:00 +00:00
parent 6996abcef3
commit 373c395810
1 changed files with 1 additions and 1 deletions
--- a/dsv4/kernels/attention/fmha.py
+++ b/dsv4/kernels/attention/fmha.py
@@ -250,7 +250,7 @@ class FmhaKernel:
                for j in cutlass.range(cute.size(rP_bf16_reg), vectorize=True):
                    # TODO: proper element mapping from QK→PV partition
                    rP_bf16_reg[j] = BFloat16(0.0)
-                cute.copy(rP_bf16_reg, tCrP_smem)
+                cute.copy(tCrP_smem, rP_bf16_reg)
                cute.arch.fence_proxy("async.shared", space="cta")

                si_handle.release()