TheRock/compiler/amd-llvm/llvm/test/Transforms/LoopVectorize/cast-costs.ll

; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py UTC_ARGS: --filter "Cost.*(EMIT-SCALAR|CLONE).*(trunc|sext|zext)" --filter "Cost.*WIDEN-CAST" --filter-out-after "LV: Selecting VF" --version 6
; REQUIRES: asserts
; RUN: opt -p loop-vectorize -force-target-supports-scalable-vectors -S -disable-output -debug-only=loop-vectorize %s 2>&1 | FileCheck %s

define void @trunc_store(ptr %dst) {
; CHECK-LABEL: 'trunc_store'
; CHECK:  Cost of 0 for VF 2: EMIT-SCALAR vp<[[VP5:%[0-9]+]]> = trunc vp<[[VP4:%[0-9]+]]> to i8
; CHECK:  Cost of 0 for VF 4: EMIT-SCALAR vp<[[VP5]]> = trunc vp<[[VP4]]> to i8
;
entry:
  br label %loop

loop:
  %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
  %iv.trunc = trunc i64 %iv to i8
  %gep = getelementptr i8, ptr %dst, i64 %iv
  store i8 %iv.trunc, ptr %gep, align 1
  %iv.next = add i64 %iv, 1
  %ec = icmp eq i64 %iv.next, 7
  br i1 %ec, label %exit, label %loop

exit:
  ret void
}

define i32 @sext_reduc(i32 %x, i32 %y) {
; CHECK-LABEL: 'sext_reduc'
; CHECK:  Cost of 1 for VF 2: WIDEN-CAST vp<[[VP6:%[0-9]+]]> = trunc ir<%red.next> to i1
; CHECK:  Cost of 1 for VF 2: WIDEN-CAST vp<[[VP7:%[0-9]+]]> = sext vp<[[VP6]]> to i32
; CHECK:  Cost of 0 for VF 2: EMIT-SCALAR vp<[[VP10:%[0-9]+]]> = sext vp<[[VP9:%[0-9]+]]> to i32
; CHECK:  Cost of 1 for VF 4: WIDEN-CAST vp<[[VP6]]> = trunc ir<%red.next> to i1
; CHECK:  Cost of 1 for VF 4: WIDEN-CAST vp<[[VP7]]> = sext vp<[[VP6]]> to i32
; CHECK:  Cost of 0 for VF 4: EMIT-SCALAR vp<[[VP10]]> = sext vp<[[VP9]]> to i32
;
entry:
  br label %for.body

for.body:
  %iv = phi i32 [ %x, %entry ], [ %iv.next, %for.body ]
  %red = phi i32 [ %y, %entry ], [ %red.next, %for.body ]
  %t0 = and i32 %red, 1
  %red.next = add i32 %t0, -1
  %iv.next = add nsw i32 %iv, 1
  %cond = icmp sgt i32 %iv, 77
  br i1 %cond, label %for.end, label %for.body

for.end:
  ret i32 %red.next
}

define i8 @reduc_add_trunc(ptr %A) {
; CHECK-LABEL: 'reduc_add_trunc'
; CHECK:  Cost of 0 for VF 2: WIDEN-CAST ir<%zext> = zext ir<%load> to i32
; CHECK:  Cost of 1 for VF 2: WIDEN-CAST vp<[[VP7:%[0-9]+]]> = trunc ir<%red.next> to i8
; CHECK:  Cost of 1 for VF 2: WIDEN-CAST vp<[[VP8:%[0-9]+]]> = zext vp<[[VP7]]> to i32
; CHECK:  Cost of 0 for VF 2: EMIT-SCALAR vp<[[VP11:%[0-9]+]]> = zext vp<[[VP10:%[0-9]+]]> to i32
; CHECK:  Cost of 0 for VF 4: WIDEN-CAST ir<%zext> = zext ir<%load> to i32
; CHECK:  Cost of 1 for VF 4: WIDEN-CAST vp<[[VP7]]> = trunc ir<%red.next> to i8
; CHECK:  Cost of 1 for VF 4: WIDEN-CAST vp<[[VP8]]> = zext vp<[[VP7]]> to i32
; CHECK:  Cost of 0 for VF 4: EMIT-SCALAR vp<[[VP11]]> = zext vp<[[VP10]]> to i32
;
entry:
  br label %loop

loop:
  %iv = phi i32 [ %iv.next, %loop ], [ 0, %entry ]
  %red = phi i32 [ %red.next, %loop ], [ 255, %entry ]
  %red.trunc = and i32 %red, 255
  %gep = getelementptr inbounds i8, ptr %A, i32 %iv
  %load = load i8, ptr %gep, align 4
  %zext = zext i8 %load to i32
  %red.next = add i32 %red.trunc, %zext
  %iv.next = add i32 %iv, 1
  %exitcond = icmp eq i32 %iv.next, 256
  br i1 %exitcond, label %exit, label %loop

exit:
  %ret = trunc i32 %red.next to i8
  ret i8 %ret
}

define void @replicate_zext(ptr %A, i32 %n) {
; CHECK-LABEL: 'replicate_zext'
; CHECK:  Cost of 1 for VF vscale x 1: CLONE ir<%iv.ext> = zext vp<[[VP4:%[0-9]+]]>
;
entry:
  br label %loop

loop:
  %iv = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
  %iv.ext = zext i32 %iv to i64
  %gep.A = getelementptr inbounds i32, ptr %A, i64 %iv.ext
  store i32 0, ptr %gep.A, align 4
  %cmp = icmp ne i32 %iv, 1000
  %iv.next = add nsw i32 %iv, 1
  br i1 %cmp, label %loop, label %exit, !llvm.loop !0

exit:
  ret void
}

!0 = distinct !{!0, !1}
!1 = !{!"llvm.loop.vectorize.scalable.enable", i1 true}