1 ; RUN: llc -mtriple=x86_64-pc-linux -x86-cmov-converter=true -verify-machineinstrs < %s | FileCheck %s
3 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
4 ;; This test checks that x86-cmov-converter optimization does not transform CMOV
5 ;; instruction when the gain (in cycles) of converting to branch is less than
6 ;; a fix threshold (measured for "-x86-cmov-converter-threshold=4").
8 ;; Test was created using the following command line:
9 ;; > clang -S -O2 -m64 -fno-vectorize -fno-unroll-loops -emit-llvm foo.c -o -
11 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
12 ;;int bar(int *a, int *b, int n) {
14 ;; for (int i = 0; i < n; ++i) {
15 ;; int x = a[i] * a[i+1] * a[i+2];
16 ;; int y = b[i] * b[i+1];
17 ;; sum += y > x ? x : 0;
21 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
22 ;; Adding a test to the above function shows code with CMOV is 25% faster than
23 ;; the code with branch.
24 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
32 ;; for (int i=0; i< N; ++i) {
37 ;; for (int i=0; i< N*10; ++i)
38 ;; sum += bar(A, B, N);
41 ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
45 define i32 @bar(i32* nocapture readonly %a, i32* nocapture readonly %b, i32 %n) #0 {
47 %cmp30 = icmp sgt i32 %n, 0
48 br i1 %cmp30, label %for.body.preheader, label %for.cond.cleanup
50 for.body.preheader: ; preds = %entry
51 %.pre = load i32, i32* %a, align 4
52 %arrayidx2.phi.trans.insert = getelementptr inbounds i32, i32* %a, i64 1
53 %.pre34 = load i32, i32* %arrayidx2.phi.trans.insert, align 4
54 %.pre35 = load i32, i32* %b, align 4
55 %wide.trip.count = zext i32 %n to i64
58 for.cond.cleanup: ; preds = %for.body, %entry
59 %sum.0.lcssa = phi i32 [ 0, %entry ], [ %add14, %for.body ]
62 for.body: ; preds = %for.body, %for.body.preheader
63 %0 = phi i32 [ %.pre35, %for.body.preheader ], [ %5, %for.body ]
64 %1 = phi i32 [ %.pre34, %for.body.preheader ], [ %4, %for.body ]
65 %2 = phi i32 [ %.pre, %for.body.preheader ], [ %1, %for.body ]
66 %indvars.iv = phi i64 [ 0, %for.body.preheader ], [ %indvars.iv.next, %for.body ]
67 %sum.032 = phi i32 [ 0, %for.body.preheader ], [ %add14, %for.body ]
68 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
69 %mul = mul nsw i32 %1, %2
70 %3 = add nuw nsw i64 %indvars.iv, 2
71 %arrayidx5 = getelementptr inbounds i32, i32* %a, i64 %3
72 %4 = load i32, i32* %arrayidx5, align 4
73 %mul6 = mul nsw i32 %mul, %4
74 %arrayidx11 = getelementptr inbounds i32, i32* %b, i64 %indvars.iv.next
75 %5 = load i32, i32* %arrayidx11, align 4
76 %mul12 = mul nsw i32 %5, %0
77 %cmp13 = icmp sgt i32 %mul12, %mul6
78 %cond = select i1 %cmp13, i32 %mul6, i32 0
79 %add14 = add nsw i32 %cond, %sum.032
80 %exitcond = icmp eq i64 %indvars.iv.next, %wide.trip.count
81 br i1 %exitcond, label %for.cond.cleanup, label %for.body
84 attributes #0 = {"target-cpu"="skylake"}
86 !llvm.module.flags = !{!0, !1}
89 !0 = !{i32 1, !"wchar_size", i32 2}
90 !1 = !{i32 7, !"PIC Level", i32 2}
91 !2 = !{!"clang version 5.0.0 (trunk)"}