Putting skip back till Rocm triton bug is fixed.

umechand-amd · umechand-amd · commit 8d5f5c764bff · 2026-03-27T04:25:46.000Z
diff --git a/test/test_examples.py b/test/test_examples.py
@@ -1617,7 +1617,7 @@ def test_exp_bwd(self):
     @xfailIfCute(
         "CuTe squeeze-and-excitation forward still exceeds thread-block limits"
     )
-    @skipIfRocm("failure on rocm")
+    @skipIfRocm("Triton ROCm store-forwarding bug: stale global memory reads")
     @skipIfCudaSharedMemoryLessThan(
         131072, reason="block sizes exceed device shared memory limit"
     )

Original file line number	Diff line number	Diff line change
`@@ -1617,7 +1617,7 @@ def test_exp_bwd(self):`
`1617`	`1617`	`@xfailIfCute(`
`1618`	`1618`	`"CuTe squeeze-and-excitation forward still exceeds thread-block limits"`
`1619`	`1619`	`)`
`1620`		`- @skipIfRocm("failure on rocm")`
	`1620`	`+ @skipIfRocm("Triton ROCm store-forwarding bug: stale global memory reads")`
`1621`	`1621`	`@skipIfCudaSharedMemoryLessThan(`
`1622`	`1622`	`131072, reason="block sizes exceed device shared memory limit"`
`1623`	`1623`	`)`