xref: /llvm-project/llvm/test/CodeGen/PowerPC/cse-despite-rounding-mode.ll (revision 427fb35192f1f7bb694a5910b05abc5925a798b2) 
1; The non-strictfp version of test/CodeGen/PowerPC/respect-rounding-mode.ll
2; Without strictfp, CSE should be free to eliminate the repeated multiply
3; and conversion instructions.
4; RUN: llc -verify-machineinstrs --mtriple powerpc64le-unknown-linux-gnu \
5; RUN:   -mcpu=pwr8 -ppc-asm-full-reg-names < %s | grep 'xvrdpic' | count 2
6; RUN: llc -verify-machineinstrs --mtriple powerpc-unknown-linux-gnu \
7; RUN:   -mcpu=pwr9 -ppc-asm-full-reg-names < %s | grep 'xvrdpic' | count 2
8; RUN: llc -verify-machineinstrs --mtriple powerpc64le-unknown-linux-gnu \
9; RUN:   -mcpu=pwr10 -ppc-asm-full-reg-names < %s | grep 'xvrdpic' | count 2
10
11; RUN: llc -verify-machineinstrs --mtriple powerpc64le-unknown-linux-gnu \
12; RUN:   -mcpu=pwr8 -ppc-asm-full-reg-names < %s | grep 'xvmuldp' | count 2
13; RUN: llc -verify-machineinstrs --mtriple powerpc-unknown-linux-gnu \
14; RUN:   -mcpu=pwr9 -ppc-asm-full-reg-names < %s | grep 'xvmuldp' | count 2
15; RUN: llc -verify-machineinstrs --mtriple powerpc64le-unknown-linux-gnu \
16; RUN:   -mcpu=pwr10 -ppc-asm-full-reg-names < %s | grep 'xvmuldp' | count 2
17@IndirectCallPtr = dso_local local_unnamed_addr global ptr null, align 8
18
19define dso_local signext i32 @func1() local_unnamed_addr #0 {
20entry:
21  tail call void @directCall() #0
22  %0 = tail call <2 x double> @llvm.experimental.constrained.rint.v2f64(<2 x double> <double -9.990000e+01, double 9.990000e+01>, metadata !"round.dynamic", metadata !"fpexcept.ignore") #0
23  %vecext = extractelement <2 x double> %0, i32 0
24  %sub = tail call double @llvm.experimental.constrained.fsub.f64(double %vecext, double -9.900000e+01, metadata !"round.dynamic", metadata !"fpexcept.ignore") #0
25  %conv = tail call i32 @llvm.experimental.constrained.fptosi.i32.f64(double %sub, metadata !"fpexcept.ignore") #0
26  tail call void @directCall() #0
27  %1 = tail call <2 x double> @llvm.experimental.constrained.rint.v2f64(<2 x double> <double -9.990000e+01, double 9.990000e+01>, metadata !"round.dynamic", metadata !"fpexcept.ignore") #0
28  %vecext3 = extractelement <2 x double> %1, i32 1
29  %cmp = tail call i1 @llvm.experimental.constrained.fcmp.f64(double %vecext3, double 9.900000e+01, metadata !"une", metadata !"fpexcept.ignore") #0
30  br i1 %cmp, label %if.then, label %if.end
31
32if.then:                                          ; preds = %entry
33  tail call void @exit(i32 signext 2) #0
34  unreachable
35
36if.end:                                           ; preds = %entry
37  ret i32 %conv
38}
39
40declare void @directCall(...) local_unnamed_addr
41
42declare double @llvm.experimental.constrained.fsub.f64(double, double, metadata, metadata)
43
44declare i32 @llvm.experimental.constrained.fptosi.i32.f64(double, metadata)
45
46declare i1 @llvm.experimental.constrained.fcmp.f64(double, double, metadata, metadata)
47
48declare void @exit(i32 signext) local_unnamed_addr
49
50define dso_local signext i32 @func2() local_unnamed_addr #0 {
51entry:
52  %call = tail call <2 x double> @getvector1() #0
53  %call1 = tail call <2 x double> @getvector2() #0
54  tail call void @directCall() #0
55  %mul = tail call <2 x double> @llvm.experimental.constrained.fmul.v2f64(<2 x double> %call, <2 x double> %call1, metadata !"round.dynamic", metadata !"fpexcept.ignore") #0
56  %vecext = extractelement <2 x double> %mul, i32 0
57  %cmp = tail call i1 @llvm.experimental.constrained.fcmp.f64(double %vecext, double 4.000000e+00, metadata !"oeq", metadata !"fpexcept.ignore") #0
58  br i1 %cmp, label %cleanup, label %if.end
59
60if.end:                                           ; preds = %entry
61  tail call void @directCall() #0
62  %mul10 = tail call <2 x double> @llvm.experimental.constrained.fmul.v2f64(<2 x double> %call, <2 x double> %call1, metadata !"round.dynamic", metadata !"fpexcept.ignore") #0
63  %0 = tail call i32 @llvm.ppc.vsx.xvcmpeqdp.p(i32 2, <2 x double> %mul, <2 x double> %mul10) #0
64  br label %cleanup
65
66cleanup:                                          ; preds = %entry, %if.end
67  %retval.0 = phi i32 [ %0, %if.end ], [ 11, %entry ]
68  ret i32 %retval.0
69}
70
71declare <2 x double> @getvector1(...) local_unnamed_addr
72
73declare <2 x double> @getvector2(...) local_unnamed_addr
74
75declare <2 x double> @llvm.experimental.constrained.fmul.v2f64(<2 x double>, <2 x double>, metadata, metadata)
76
77declare i32 @llvm.ppc.vsx.xvcmpeqdp.p(i32, <2 x double>, <2 x double>)
78
79define dso_local signext i32 @func3() local_unnamed_addr #0 {
80entry:
81  %0 = load ptr, ptr @IndirectCallPtr, align 8
82  tail call void %0() #0
83  %1 = tail call <2 x double> @llvm.experimental.constrained.rint.v2f64(<2 x double> <double -9.990000e+01, double 9.990000e+01>, metadata !"round.dynamic", metadata !"fpexcept.ignore") #0
84  %vecext = extractelement <2 x double> %1, i32 0
85  %sub = tail call double @llvm.experimental.constrained.fsub.f64(double %vecext, double -9.900000e+01, metadata !"round.dynamic", metadata !"fpexcept.ignore") #0
86  %conv = tail call i32 @llvm.experimental.constrained.fptosi.i32.f64(double %sub, metadata !"fpexcept.ignore") #0
87  %2 = load ptr, ptr @IndirectCallPtr, align 8
88  tail call void %2() #0
89  %3 = tail call <2 x double> @llvm.experimental.constrained.rint.v2f64(<2 x double> <double -9.990000e+01, double 9.990000e+01>, metadata !"round.dynamic", metadata !"fpexcept.ignore") #0
90  %vecext4 = extractelement <2 x double> %3, i32 1
91  %cmp = tail call i1 @llvm.experimental.constrained.fcmp.f64(double %vecext4, double 9.900000e+01, metadata !"une", metadata !"fpexcept.ignore") #0
92  br i1 %cmp, label %if.then, label %if.end
93
94if.then:                                          ; preds = %entry
95  tail call void @exit(i32 signext 2) #0
96  unreachable
97
98if.end:                                           ; preds = %entry
99  ret i32 %conv
100}
101
102define dso_local signext i32 @func4() local_unnamed_addr #0 {
103entry:
104  %call = tail call <2 x double> @getvector1() #0
105  %call1 = tail call <2 x double> @getvector2() #0
106  %0 = load ptr, ptr @IndirectCallPtr, align 8
107  tail call void %0() #0
108  %mul = tail call <2 x double> @llvm.experimental.constrained.fmul.v2f64(<2 x double> %call, <2 x double> %call1, metadata !"round.dynamic", metadata !"fpexcept.ignore") #0
109  %vecext = extractelement <2 x double> %mul, i32 0
110  %cmp = tail call i1 @llvm.experimental.constrained.fcmp.f64(double %vecext, double 4.000000e+00, metadata !"oeq", metadata !"fpexcept.ignore") #0
111  br i1 %cmp, label %cleanup, label %if.end
112
113if.end:                                           ; preds = %entry
114  %1 = load ptr, ptr @IndirectCallPtr, align 8
115  tail call void %1() #0
116  %mul11 = tail call <2 x double> @llvm.experimental.constrained.fmul.v2f64(<2 x double> %call, <2 x double> %call1, metadata !"round.dynamic", metadata !"fpexcept.ignore") #0
117  %2 = tail call i32 @llvm.ppc.vsx.xvcmpeqdp.p(i32 2, <2 x double> %mul, <2 x double> %mul11) #0
118  br label %cleanup
119
120cleanup:                                          ; preds = %entry, %if.end
121  %retval.0 = phi i32 [ %2, %if.end ], [ 11, %entry ]
122  ret i32 %retval.0
123}
124
125declare <2 x double> @llvm.experimental.constrained.rint.v2f64(<2 x double>, metadata, metadata)
126
127attributes #0 = { nounwind }
128