[Bugfix][ROCm] Fix gemm_a4w4 call to use updated AITER API signature (#40754)

Signed-off-by: cheiluno <cheiluno@amd.com>
2026-04-29 09:04:53 +09:00
parent e68fa1b90a
commit d109eacd05
1 changed files with 2 additions and 11 deletions
@@ -96,21 +96,12 @@ try:
                    x_q = x
                    x_s = x_scales

-                # 32 alignment is enough for dim0 padding of output for
-                # gemm_a4w4 kernel
-                y = torch.empty(
-                    (M + 31) // 32 * 32,
-                    weight.shape[0],
-                    device=x_q.device,
-                    dtype=out_dtype,
-                )
-
-                gemm_a4w4(
+                y = gemm_a4w4(
                    x_q,
                    weight.view(x_q.dtype),
                    x_s,
                    weight_scale.view(x_s.dtype),
-                    y,
+                    dtype=out_dtype,
                    bpreshuffle=True,
                )
            return y[:M]