1*f4a2713aSLionel Sambuctarget datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128"
2*f4a2713aSLionel Sambuc; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -mcpu=corei7 -bb-vectorize -bb-vectorize-req-chain-depth=3 -instcombine -gvn -S | FileCheck %s
3*f4a2713aSLionel Sambuc
4*f4a2713aSLionel Sambucdeclare double @llvm.fma.f64(double, double, double)
5*f4a2713aSLionel Sambucdeclare double @llvm.fmuladd.f64(double, double, double)
6*f4a2713aSLionel Sambucdeclare double @llvm.cos.f64(double)
7*f4a2713aSLionel Sambucdeclare double @llvm.powi.f64(double, i32)
8*f4a2713aSLionel Sambuc
9*f4a2713aSLionel Sambuc; Basic depth-3 chain with fma
10*f4a2713aSLionel Sambucdefine double @test1(double %A1, double %A2, double %B1, double %B2, double %C1, double %C2) {
11*f4a2713aSLionel Sambuc	%X1 = fsub double %A1, %B1
12*f4a2713aSLionel Sambuc	%X2 = fsub double %A2, %B2
13*f4a2713aSLionel Sambuc	%Y1 = call double @llvm.fma.f64(double %X1, double %A1, double %C1)
14*f4a2713aSLionel Sambuc	%Y2 = call double @llvm.fma.f64(double %X2, double %A2, double %C2)
15*f4a2713aSLionel Sambuc	%Z1 = fadd double %Y1, %B1
16*f4a2713aSLionel Sambuc	%Z2 = fadd double %Y2, %B2
17*f4a2713aSLionel Sambuc	%R  = fmul double %Z1, %Z2
18*f4a2713aSLionel Sambuc	ret double %R
19*f4a2713aSLionel Sambuc; CHECK-LABEL: @test1(
20*f4a2713aSLionel Sambuc; CHECK: ret double %R
21*f4a2713aSLionel Sambuc}
22*f4a2713aSLionel Sambuc
23*f4a2713aSLionel Sambuc; Basic depth-3 chain with fmuladd
24*f4a2713aSLionel Sambucdefine double @test1a(double %A1, double %A2, double %B1, double %B2, double %C1, double %C2) {
25*f4a2713aSLionel Sambuc	%X1 = fsub double %A1, %B1
26*f4a2713aSLionel Sambuc	%X2 = fsub double %A2, %B2
27*f4a2713aSLionel Sambuc	%Y1 = call double @llvm.fmuladd.f64(double %X1, double %A1, double %C1)
28*f4a2713aSLionel Sambuc	%Y2 = call double @llvm.fmuladd.f64(double %X2, double %A2, double %C2)
29*f4a2713aSLionel Sambuc	%Z1 = fadd double %Y1, %B1
30*f4a2713aSLionel Sambuc	%Z2 = fadd double %Y2, %B2
31*f4a2713aSLionel Sambuc	%R  = fmul double %Z1, %Z2
32*f4a2713aSLionel Sambuc	ret double %R
33*f4a2713aSLionel Sambuc; CHECK-LABEL: @test1a(
34*f4a2713aSLionel Sambuc; CHECK: ret double %R
35*f4a2713aSLionel Sambuc}
36*f4a2713aSLionel Sambuc
37*f4a2713aSLionel Sambuc; Basic depth-3 chain with cos
38*f4a2713aSLionel Sambucdefine double @test2(double %A1, double %A2, double %B1, double %B2) {
39*f4a2713aSLionel Sambuc	%X1 = fsub double %A1, %B1
40*f4a2713aSLionel Sambuc	%X2 = fsub double %A2, %B2
41*f4a2713aSLionel Sambuc	%Y1 = call double @llvm.cos.f64(double %X1)
42*f4a2713aSLionel Sambuc	%Y2 = call double @llvm.cos.f64(double %X2)
43*f4a2713aSLionel Sambuc	%Z1 = fadd double %Y1, %B1
44*f4a2713aSLionel Sambuc	%Z2 = fadd double %Y2, %B2
45*f4a2713aSLionel Sambuc	%R  = fmul double %Z1, %Z2
46*f4a2713aSLionel Sambuc	ret double %R
47*f4a2713aSLionel Sambuc; CHECK-LABEL: @test2(
48*f4a2713aSLionel Sambuc; CHECK: ret double %R
49*f4a2713aSLionel Sambuc}
50*f4a2713aSLionel Sambuc
51*f4a2713aSLionel Sambuc; Basic depth-3 chain with powi
52*f4a2713aSLionel Sambucdefine double @test3(double %A1, double %A2, double %B1, double %B2, i32 %P) {
53*f4a2713aSLionel Sambuc	%X1 = fsub double %A1, %B1
54*f4a2713aSLionel Sambuc	%X2 = fsub double %A2, %B2
55*f4a2713aSLionel Sambuc	%Y1 = call double @llvm.powi.f64(double %X1, i32 %P)
56*f4a2713aSLionel Sambuc	%Y2 = call double @llvm.powi.f64(double %X2, i32 %P)
57*f4a2713aSLionel Sambuc	%Z1 = fadd double %Y1, %B1
58*f4a2713aSLionel Sambuc	%Z2 = fadd double %Y2, %B2
59*f4a2713aSLionel Sambuc	%R  = fmul double %Z1, %Z2
60*f4a2713aSLionel Sambuc	ret double %R
61*f4a2713aSLionel Sambuc; CHECK-LABEL: @test3(
62*f4a2713aSLionel Sambuc; CHECK: ret double %R
63*f4a2713aSLionel Sambuc}
64*f4a2713aSLionel Sambuc
65*f4a2713aSLionel Sambuc; Basic depth-3 chain with powi (different powers: should not vectorize)
66*f4a2713aSLionel Sambucdefine double @test4(double %A1, double %A2, double %B1, double %B2, i32 %P) {
67*f4a2713aSLionel Sambuc	%X1 = fsub double %A1, %B1
68*f4a2713aSLionel Sambuc	%X2 = fsub double %A2, %B2
69*f4a2713aSLionel Sambuc        %P2 = add i32 %P, 1
70*f4a2713aSLionel Sambuc	%Y1 = call double @llvm.powi.f64(double %X1, i32 %P)
71*f4a2713aSLionel Sambuc	%Y2 = call double @llvm.powi.f64(double %X2, i32 %P2)
72*f4a2713aSLionel Sambuc	%Z1 = fadd double %Y1, %B1
73*f4a2713aSLionel Sambuc	%Z2 = fadd double %Y2, %B2
74*f4a2713aSLionel Sambuc	%R  = fmul double %Z1, %Z2
75*f4a2713aSLionel Sambuc	ret double %R
76*f4a2713aSLionel Sambuc; CHECK-LABEL: @test4(
77*f4a2713aSLionel Sambuc; CHECK: ret double %R
78*f4a2713aSLionel Sambuc}
79*f4a2713aSLionel Sambuc
80