1 ; RUN: llc < %s -mtriple=x86_64-linux -mattr=+sse2 -mcpu=nehalem | FileCheck %s
2 ; RUN: llc < %s -mtriple=x86_64-win32 -mattr=+sse2 -mcpu=nehalem | FileCheck %s
4 define double @t1(float* nocapture %x) nounwind readonly ssp {
7 ; CHECK: movss ([[A0:%rdi|%rcx]]), %xmm0
8 ; CHECK: cvtss2sd %xmm0, %xmm0
10 %0 = load float* %x, align 4
11 %1 = fpext float %0 to double
15 define float @t2(double* nocapture %x) nounwind readonly ssp optsize {
18 ; CHECK: cvtsd2ss ([[A0]]), %xmm0
19 %0 = load double* %x, align 8
20 %1 = fptrunc double %0 to float
24 define float @squirtf(float* %x) nounwind {
26 ; CHECK-LABEL: squirtf:
27 ; CHECK: movss ([[A0]]), %xmm0
28 ; CHECK: sqrtss %xmm0, %xmm0
30 %t = call float @llvm.sqrt.f32(float %z)
34 define double @squirt(double* %x) nounwind {
36 ; CHECK-LABEL: squirt:
37 ; CHECK: movsd ([[A0]]), %xmm0
38 ; CHECK: sqrtsd %xmm0, %xmm0
40 %t = call double @llvm.sqrt.f64(double %z)
44 define float @squirtf_size(float* %x) nounwind optsize {
46 ; CHECK-LABEL: squirtf_size:
47 ; CHECK: sqrtss ([[A0]]), %xmm0
49 %t = call float @llvm.sqrt.f32(float %z)
53 define double @squirt_size(double* %x) nounwind optsize {
55 ; CHECK-LABEL: squirt_size:
56 ; CHECK: sqrtsd ([[A0]]), %xmm0
58 %t = call double @llvm.sqrt.f64(double %z)
62 declare float @llvm.sqrt.f32(float)
63 declare double @llvm.sqrt.f64(double)
65 ; CHECK-LABEL: loopdep1
68 ; This loop contains two cvtsi2ss instructions that update the same xmm
69 ; register. Verify that the execution dependency fix pass breaks those
70 ; dependencies by inserting xorps instructions.
72 ; If the register allocator chooses different registers for the two cvtsi2ss
73 ; instructions, they are still dependent on themselves.
74 ; CHECK: xorps [[XMM1:%xmm[0-9]+]]
76 ; CHECK: cvtsi2ssl %{{.*}}, [[XMM1]]
77 ; CHECK: xorps [[XMM2:%xmm[0-9]+]]
79 ; CHECK: cvtsi2ssl %{{.*}}, [[XMM2]]
81 define float @loopdep1(i32 %m) nounwind uwtable readnone ssp {
83 %tobool3 = icmp eq i32 %m, 0
84 br i1 %tobool3, label %for.end, label %for.body
86 for.body: ; preds = %entry, %for.body
87 %m.addr.07 = phi i32 [ %dec, %for.body ], [ %m, %entry ]
88 %s1.06 = phi float [ %add, %for.body ], [ 0.000000e+00, %entry ]
89 %s2.05 = phi float [ %add2, %for.body ], [ 0.000000e+00, %entry ]
90 %n.04 = phi i32 [ %inc, %for.body ], [ 1, %entry ]
91 %conv = sitofp i32 %n.04 to float
92 %add = fadd float %s1.06, %conv
93 %conv1 = sitofp i32 %m.addr.07 to float
94 %add2 = fadd float %s2.05, %conv1
95 %inc = add nsw i32 %n.04, 1
96 %dec = add nsw i32 %m.addr.07, -1
97 %tobool = icmp eq i32 %dec, 0
98 br i1 %tobool, label %for.end, label %for.body
100 for.end: ; preds = %for.body, %entry
101 %s1.0.lcssa = phi float [ 0.000000e+00, %entry ], [ %add, %for.body ]
102 %s2.0.lcssa = phi float [ 0.000000e+00, %entry ], [ %add2, %for.body ]
103 %sub = fsub float %s1.0.lcssa, %s2.0.lcssa
107 ; This loop contains a cvtsi2sd instruction that has a loop-carried
108 ; false dependency on an xmm that is modified by other scalar instructions
109 ; that follow it in the loop. Additionally, the source of convert is a
110 ; memory operand. Verify the execution dependency fix pass breaks this
111 ; dependency by inserting a xor before the convert.
112 @x = common global [1024 x double] zeroinitializer, align 16
113 @y = common global [1024 x double] zeroinitializer, align 16
114 @z = common global [1024 x double] zeroinitializer, align 16
115 @w = common global [1024 x double] zeroinitializer, align 16
116 @v = common global [1024 x i32] zeroinitializer, align 16
118 define void @loopdep2() {
120 br label %for.cond1.preheader
122 for.cond1.preheader: ; preds = %for.inc14, %entry
123 %i.025 = phi i32 [ 0, %entry ], [ %inc15, %for.inc14 ]
127 %indvars.iv = phi i64 [ 0, %for.cond1.preheader ], [ %indvars.iv.next, %for.body3 ]
128 %arrayidx = getelementptr inbounds [1024 x i32]* @v, i64 0, i64 %indvars.iv
129 %0 = load i32* %arrayidx, align 4
130 %conv = sitofp i32 %0 to double
131 %arrayidx5 = getelementptr inbounds [1024 x double]* @x, i64 0, i64 %indvars.iv
132 %1 = load double* %arrayidx5, align 8
133 %mul = fmul double %conv, %1
134 %arrayidx7 = getelementptr inbounds [1024 x double]* @y, i64 0, i64 %indvars.iv
135 %2 = load double* %arrayidx7, align 8
136 %mul8 = fmul double %mul, %2
137 %arrayidx10 = getelementptr inbounds [1024 x double]* @z, i64 0, i64 %indvars.iv
138 %3 = load double* %arrayidx10, align 8
139 %mul11 = fmul double %mul8, %3
140 %arrayidx13 = getelementptr inbounds [1024 x double]* @w, i64 0, i64 %indvars.iv
141 store double %mul11, double* %arrayidx13, align 8
142 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
143 %exitcond = icmp eq i64 %indvars.iv.next, 1024
144 br i1 %exitcond, label %for.inc14, label %for.body3
146 for.inc14: ; preds = %for.body3
147 %inc15 = add nsw i32 %i.025, 1
148 %exitcond26 = icmp eq i32 %inc15, 100000
149 br i1 %exitcond26, label %for.end16, label %for.cond1.preheader
151 for.end16: ; preds = %for.inc14
154 ;CHECK-LABEL:@loopdep2
155 ;CHECK: xorps [[XMM0:%xmm[0-9]+]], [[XMM0]]
156 ;CHECK-NEXT: cvtsi2sdl {{.*}}, [[XMM0]]
157 ;CHECK-NEXT: mulsd {{.*}}, [[XMM0]]
158 ;CHECK-NEXT: mulsd {{.*}}, [[XMM0]]
159 ;CHECK-NEXT: mulsd {{.*}}, [[XMM0]]
160 ;CHECK-NEXT: movsd [[XMM0]],