xref: /llvm-project/llvm/test/Transforms/SLPVectorizer/X86/arith-max-cost.ll (revision 3be72f402925b99adbec4a2ee5bacdf76ba6c8d1)
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt < %s -passes=slp-vectorizer -mcpu=corei7-avx -mtriple=x86_64-unknown-linux -S | FileCheck %s
3
4; This test checks whether the cost of the vector max intrinsic is calculated
5; correctly. A max vector intrinsic combines the select and icmp instructions.
6; This maps to a single PMAX instruction in x86.
7define void @smax_intrinsic_cost(i64 %arg0, i64 %arg1) {
8; CHECK-LABEL: @smax_intrinsic_cost(
9; CHECK-NEXT:    [[ICMP0:%.*]] = icmp sgt i64 [[ARG0:%.*]], 123
10; CHECK-NEXT:    [[ICMP1:%.*]] = icmp sgt i64 [[ARG1:%.*]], 456
11; CHECK-NEXT:    [[SELECT0:%.*]] = select i1 [[ICMP0]], i64 [[ARG0]], i64 123
12; CHECK-NEXT:    [[SELECT1:%.*]] = select i1 [[ICMP1]], i64 [[ARG1]], i64 456
13; CHECK-NEXT:    [[ROOT:%.*]] = icmp sle i64 [[SELECT0]], [[SELECT1]]
14; CHECK-NEXT:    ret void
15;
16  %icmp0 = icmp sgt i64 %arg0, 123
17  %icmp1 = icmp sgt i64 %arg1, 456
18  %select0 = select i1 %icmp0, i64 %arg0, i64 123
19  %select1 = select i1 %icmp1, i64 %arg1, i64 456
20  %root = icmp sle i64 %select0, %select1
21  ret void
22}
23
24
25define void @umax_intrinsic_cost(i64 %arg0, i64 %arg1) {
26; CHECK-LABEL: @umax_intrinsic_cost(
27; CHECK-NEXT:    [[ICMP0:%.*]] = icmp ugt i64 [[ARG0:%.*]], 123
28; CHECK-NEXT:    [[ICMP1:%.*]] = icmp ugt i64 [[ARG1:%.*]], 456
29; CHECK-NEXT:    [[SELECT0:%.*]] = select i1 [[ICMP0]], i64 [[ARG0]], i64 123
30; CHECK-NEXT:    [[SELECT1:%.*]] = select i1 [[ICMP1]], i64 [[ARG1]], i64 456
31; CHECK-NEXT:    [[ROOT:%.*]] = icmp sle i64 [[SELECT0]], [[SELECT1]]
32; CHECK-NEXT:    ret void
33;
34  %icmp0 = icmp ugt i64 %arg0, 123
35  %icmp1 = icmp ugt i64 %arg1, 456
36  %select0 = select i1 %icmp0, i64 %arg0, i64 123
37  %select1 = select i1 %icmp1, i64 %arg1, i64 456
38  %root = icmp sle i64 %select0, %select1
39  ret void
40}
41