[llvm] [AMDGPU] SIPeepholeSDWA: Handle V_CNDMASK_B32_e64 (PR #137930)

[Matt Arsenault via llvm-commits]

================
@@ -0,0 +1,43 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 5
+# RUN: llc %s -mtriple=amdgcn -mcpu=gfx1030 -run-pass=si-peephole-sdwa -o - | FileCheck %s
+
+# For conversion of_CNDMASK_B32_e64 to SDWA, the destination of V_CMP_O_F16_e64 must be
+# changed to vcc_lo first. This would introduce a vcc_hi use that requires special
+# handling in si-peephole-sdwa.
----------------
arsenm wrote:

It's not really special, wave32 handling is just bad 

https://github.com/llvm/llvm-project/pull/137930