================
@@ -166,3 +166,156 @@ body: |
S_ENDPGM 0, implicit $vcc
...
+
+---
+# SCC is live *into* the scalar frame index user: S_CMP_EQ_U32 produces it and
+# S_CSELECT_B32 reads it. With no SGPR free to hold the V_READFIRSTLANE_B32
+# result, scavenging one emergency-spills, and an SGPR spill flips EXEC with
+# S_NOT_B64. Those EXEC flips land between the S_CMP_EQ_U32 and the
----------------
PankajDwivedi-25 wrote:
One case looks like it got dropped, though. In #224080 the nonzero-offset test
ended with `S_CBRANCH_SCC1 %bb.2, implicit $scc`
so SCC was live *across* the frame index user. Here that line is gone and the
test is the dead-after variant instead, which is a different scenario rather
than a rename?
https://github.com/llvm/llvm-project/pull/225196
_______________________________________________
llvm-branch-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-branch-commits
