llvm/test/Transforms/SimplifyCFG/AMDGPU/speculate-block.ll

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6
; RUN: opt -S -passes=simplifycfg -mtriple=amdgpu9.0a-amd-amdhsa < %s | FileCheck %s

; Test that SimplifyCFG speculatively executes blocks on AMDGPU even for vector
; types. The AMDGPU getCmpSelInstrCost override returns a low cost for selects,
; allowing the speculation regardless of vector width.

define amdgpu_kernel void @ham(i1 %arg3, i1 %icmp84) {
; CHECK-LABEL: define amdgpu_kernel void @ham(
; CHECK-SAME: i1 [[ARG3:%.*]], i1 [[ICMP84:%.*]]) {
; CHECK-NEXT:  [[ENTRY:.*:]]
; CHECK-NEXT:    [[SPEC_SELECT:%.*]] = select i1 [[ICMP84]], <4 x float> zeroinitializer, <4 x float> splat (float 1.000000e+00)
; CHECK-NEXT:    [[SPEC_SELECT1:%.*]] = select i1 [[ARG3]], <4 x float> [[SPEC_SELECT]], <4 x float> zeroinitializer
; CHECK-NEXT:    ret void
;
entry:
  br i1 %arg3, label %then, label %merge

then:                                             ; preds = %entry
  %icmp841 = icmp sge i32 0, 0
  br i1 %icmp84, label %then.if, label %merge

then.if:                                          ; preds = %then
  br label %merge

merge:                                            ; preds = %then.if, %then, %entry
  %phi92 = phi <4 x float> [ zeroinitializer, %then.if ], [ splat (float 1.000000e+00), %then ], [ zeroinitializer, %entry ]
  ret void
}