[llvm] [AMDGPU][SIPreEmitPeephole] pre-commit tests: mustRetainExeczBranch: use a cost model (PR #109816)

[Juan Manuel Martinez Caamaño via llvm-commits]

================
@@ -0,0 +1,265 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc -mtriple=amdgcn -mcpu=gfx1010 < %s | FileCheck -check-prefixes=GFX10,GFX1010 %s
+; RUN: llc -mtriple=amdgcn -mcpu=gfx1030 < %s | FileCheck -check-prefixes=GFX10,GFX1030 %s
+
+define void @convergent_cmp_no_metadata(i32 noundef inreg %value, ptr addrspace(8) nocapture writeonly inreg %res, i32 noundef inreg %v_offset, i32 noundef inreg %0, i32 noundef inreg %flag) {
----------------
jmmartinez wrote:

Isn't "convergent" when the register has the same value across the wave ?

I've created the "divergent" test cases after the "convergent" ones using a call to `llvm.amdgcn.workitem.id.x` as one of the operands of the comparison.

https://github.com/llvm/llvm-project/pull/109816