D1.5: Remove duplicate tTMrO definition (keep unconditional one)

This commit is contained in:
2026-05-24 02:40:40 +00:00
parent 093de4e62f
commit 34db0bf802

View File

@@ -378,10 +378,6 @@ class FmhaKernel:
k2 = k_coord // 64
_sP_nostage[(m_coord, k0), 0, (k1, k2)] = rP_bf16[(j0, 0), j1, 0, 0]
cute.arch.fence_proxy("async.shared", space="cta")
# O rescale register tensor (defined unconditionally for CuTeDSL scoping)
tTMrO = cute.make_rmem_tensor(
(tTMEM_LOADcO.shape, 128 // corr_tile_size), self.acc_dtype
)
if kt > 0:
for i in range(n_corr_tiles):
tTMrO_i_ = tTMrO[None, i]