[llvm-branch-commits] [llvm] AMDGPU/GlobalISel: Fix cs_chain use of constrainOperandRegClass (PR #212248)

[Petar Avramovic via llvm-branch-commits]

================
@@ -152,8 +152,9 @@ define amdgpu_cs_chain void @chain_call_exec_from_extract(<2 x i32> inreg %vec,
   ; GFX11-NEXT:   [[COPY8:%[0-9]+]]:_(i32) = COPY $vgpr11
   ; GFX11-NEXT:   [[C:%[0-9]+]]:_(i32) = G_CONSTANT i32 0
   ; GFX11-NEXT:   [[GV:%[0-9]+]]:_(p0) = G_GLOBAL_VALUE @callee
-  ; GFX11-NEXT:   [[EVEC:%[0-9]+]]:sreg_32(i32) = G_EXTRACT_VECTOR_ELT [[BUILD_VECTOR]](<2 x i32>), [[C]](i32)
----------------
petar-avramovic wrote:

this was the bug

https://github.com/llvm/llvm-project/pull/212248