xref: /llvm-project/llvm/test/CodeGen/X86/unused_stackslots.ll (revision eecb99c5f66c8491766628a2925587e20f3b1dbd)
1; PR26374: Check no stack slots are allocated for vregs which have no real reference.
2; RUN: llc < %s | FileCheck %s
3target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
4target triple = "x86_64-unknown-linux-gnu"
5
6%struct.ImageParameters = type { i32, i32, [0 x [16 x i16]] }
7%struct.InputParameters = type { i32, i32 }
8
9@c = common global ptr null, align 8
10@a = common global ptr null, align 8
11@d = common global [6 x i32] zeroinitializer, align 16
12@b = common global ptr null, align 8
13@e = common global [4 x i32] zeroinitializer, align 16
14
15; It is not easy to check there is no unused holes in stack allocated for spills,
16; so simply check the size of stack allocated cannot exceed 350.
17; (408 is used before the fix for PR26374. 344 is used after the fix).
18;
19; CHECK-LABEL: @fn
20; CHECK: subq {{\$3[0-4][0-9]}}, %rsp
21
22; Function Attrs: nounwind uwtable
23define i32 @fn() #0 {
24entry:
25  %n = alloca [8 x [8 x i32]], align 16
26  call void @llvm.lifetime.start.p0(i64 256, ptr %n) #3
27  %arraydecay.1 = getelementptr inbounds [8 x [8 x i32]], ptr %n, i64 0, i64 1, i64 0
28  %arraydecay.2 = getelementptr inbounds [8 x [8 x i32]], ptr %n, i64 0, i64 2, i64 0
29  %arraydecay.3 = getelementptr inbounds [8 x [8 x i32]], ptr %n, i64 0, i64 3, i64 0
30  %arraydecay.4 = getelementptr inbounds [8 x [8 x i32]], ptr %n, i64 0, i64 4, i64 0
31  %arraydecay.5 = getelementptr inbounds [8 x [8 x i32]], ptr %n, i64 0, i64 5, i64 0
32  %arraydecay.6 = getelementptr inbounds [8 x [8 x i32]], ptr %n, i64 0, i64 6, i64 0
33  %arraydecay.7 = getelementptr inbounds [8 x [8 x i32]], ptr %n, i64 0, i64 7, i64 0
34  br label %for.body
35
36for.body:                                         ; preds = %for.inc73, %entry
37  %q.0131 = phi i32 [ 0, %entry ], [ %inc74, %for.inc73 ]
38  %m.0130 = phi i32 [ 0, %entry ], [ %m.4, %for.inc73 ]
39  %div = sdiv i32 %q.0131, 2
40  %shl = shl i32 %div, 3
41  %rem = srem i32 %q.0131, 2
42  %shl1 = shl nsw i32 %rem, 3
43  %tmp9 = sext i32 %shl1 to i64
44  %tmp10 = sext i32 %shl to i64
45  %tmp11 = or disjoint i32 %shl1, 4
46  %tmp12 = sext i32 %tmp11 to i64
47  %tmp13 = or disjoint i32 %shl, 4
48  %tmp14 = sext i32 %tmp13 to i64
49  br label %for.body4
50
51for.body4:                                        ; preds = %for.inc48, %for.body
52  %indvars.iv148 = phi i64 [ %tmp10, %for.body ], [ %indvars.iv.next149, %for.inc48 ]
53  %m.1126 = phi i32 [ %m.0130, %for.body ], [ %m.3.lcssa, %for.inc48 ]
54  %tmp15 = load ptr, ptr @c, align 8
55  %opix_y = getelementptr inbounds %struct.ImageParameters, ptr %tmp15, i64 0, i32 1
56  %tmp16 = load i32, ptr %opix_y, align 4
57  %tmp17 = trunc i64 %indvars.iv148 to i32
58  %add5 = add nsw i32 %tmp16, %tmp17
59  %tmp18 = sub nuw nsw i64 %indvars.iv148, %tmp10
60  %tmp19 = sext i32 %add5 to i64
61  %tmp20 = add nsw i64 %tmp19, 1
62  %tmp21 = or disjoint i64 %indvars.iv148, 1
63  %tmp22 = or disjoint i64 %tmp18, 1
64  %tmp23 = add nsw i64 %tmp19, 2
65  %tmp24 = or disjoint i64 %indvars.iv148, 2
66  %tmp25 = or disjoint i64 %tmp18, 2
67  %tmp26 = add nsw i64 %tmp19, 3
68  %tmp27 = or disjoint i64 %indvars.iv148, 3
69  %tmp28 = or disjoint i64 %tmp18, 3
70  br label %for.body9
71
72for.body9:                                        ; preds = %for.inc45.for.body9_crit_edge, %for.body4
73  %tmp29 = phi ptr [ %tmp15, %for.body4 ], [ %.pre, %for.inc45.for.body9_crit_edge ]
74  %indvars.iv145 = phi i64 [ %tmp9, %for.body4 ], [ %indvars.iv.next146, %for.inc45.for.body9_crit_edge ]
75  %m.2124 = phi i32 [ %m.1126, %for.body4 ], [ %m.3, %for.inc45.for.body9_crit_edge ]
76  %tmp30 = load i32, ptr %tmp29, align 4
77  %tmp31 = trunc i64 %indvars.iv145 to i32
78  %add10 = add nsw i32 %tmp30, %tmp31
79  tail call void @LumaPrediction4x4(i32 %tmp31, i32 %tmp17, i32 0, i32 0, i32 0, i16 signext 0, i16 signext 0) #3
80  %tmp32 = load ptr, ptr @a, align 8
81  %tmp33 = load ptr, ptr @c, align 8
82  %tmp34 = sub nuw nsw i64 %indvars.iv145, %tmp9
83  %tmp35 = sext i32 %add10 to i64
84  br label %for.cond14.preheader
85
86for.cond14.preheader:                             ; preds = %for.body9
87  %arrayidx = getelementptr inbounds ptr, ptr %tmp32, i64 %tmp19
88  %tmp36 = load ptr, ptr %arrayidx, align 8
89  %arrayidx20 = getelementptr inbounds i16, ptr %tmp36, i64 %tmp35
90  %arrayidx26 = getelementptr inbounds %struct.ImageParameters, ptr %tmp33, i64 0, i32 2, i64 %indvars.iv148, i64 %indvars.iv145
91  %arrayidx35 = getelementptr inbounds [8 x [8 x i32]], ptr %n, i64 0, i64 %tmp18, i64 %tmp34
92  %tmp38 = load <4 x i16>, ptr %arrayidx20, align 2
93  %tmp39 = zext <4 x i16> %tmp38 to <4 x i32>
94  %tmp41 = load <4 x i16>, ptr %arrayidx26, align 2
95  %tmp42 = zext <4 x i16> %tmp41 to <4 x i32>
96  %tmp43 = sub nsw <4 x i32> %tmp39, %tmp42
97  store <4 x i32> %tmp43, ptr %arrayidx35, align 16
98  store <4 x i32> %tmp43, ptr @d, align 16
99  %arrayidx.1 = getelementptr inbounds ptr, ptr %tmp32, i64 %tmp20
100  %tmp45 = load ptr, ptr %arrayidx.1, align 8
101  %arrayidx20.1 = getelementptr inbounds i16, ptr %tmp45, i64 %tmp35
102  %arrayidx26.1 = getelementptr inbounds %struct.ImageParameters, ptr %tmp33, i64 0, i32 2, i64 %tmp21, i64 %indvars.iv145
103  %arrayidx35.1 = getelementptr inbounds [8 x [8 x i32]], ptr %n, i64 0, i64 %tmp22, i64 %tmp34
104  %tmp47 = load <4 x i16>, ptr %arrayidx20.1, align 2
105  %tmp48 = zext <4 x i16> %tmp47 to <4 x i32>
106  %tmp50 = load <4 x i16>, ptr %arrayidx26.1, align 2
107  %tmp51 = zext <4 x i16> %tmp50 to <4 x i32>
108  %tmp52 = sub nsw <4 x i32> %tmp48, %tmp51
109  store <4 x i32> %tmp52, ptr %arrayidx35.1, align 16
110  store <4 x i32> %tmp52, ptr getelementptr inbounds ([6 x i32], ptr @d, i64 0, i64 4), align 16
111  %arrayidx.2 = getelementptr inbounds ptr, ptr %tmp32, i64 %tmp23
112  %tmp54 = load ptr, ptr %arrayidx.2, align 8
113  %arrayidx20.2 = getelementptr inbounds i16, ptr %tmp54, i64 %tmp35
114  %arrayidx26.2 = getelementptr inbounds %struct.ImageParameters, ptr %tmp33, i64 0, i32 2, i64 %tmp24, i64 %indvars.iv145
115  %arrayidx35.2 = getelementptr inbounds [8 x [8 x i32]], ptr %n, i64 0, i64 %tmp25, i64 %tmp34
116  %tmp56 = load <4 x i16>, ptr %arrayidx20.2, align 2
117  %tmp57 = zext <4 x i16> %tmp56 to <4 x i32>
118  %tmp59 = load <4 x i16>, ptr %arrayidx26.2, align 2
119  %tmp60 = zext <4 x i16> %tmp59 to <4 x i32>
120  %tmp61 = sub nsw <4 x i32> %tmp57, %tmp60
121  store <4 x i32> %tmp61, ptr %arrayidx35.2, align 16
122  store <4 x i32> %tmp61, ptr getelementptr ([6 x i32], ptr @d, i64 1, i64 2), align 16
123  %arrayidx.3 = getelementptr inbounds ptr, ptr %tmp32, i64 %tmp26
124  %tmp63 = load ptr, ptr %arrayidx.3, align 8
125  %arrayidx20.3 = getelementptr inbounds i16, ptr %tmp63, i64 %tmp35
126  %arrayidx26.3 = getelementptr inbounds %struct.ImageParameters, ptr %tmp33, i64 0, i32 2, i64 %tmp27, i64 %indvars.iv145
127  %arrayidx35.3 = getelementptr inbounds [8 x [8 x i32]], ptr %n, i64 0, i64 %tmp28, i64 %tmp34
128  %tmp65 = load <4 x i16>, ptr %arrayidx20.3, align 2
129  %tmp66 = zext <4 x i16> %tmp65 to <4 x i32>
130  %tmp68 = load <4 x i16>, ptr %arrayidx26.3, align 2
131  %tmp69 = zext <4 x i16> %tmp68 to <4 x i32>
132  %tmp70 = sub nsw <4 x i32> %tmp66, %tmp69
133  store <4 x i32> %tmp70, ptr %arrayidx35.3, align 16
134  store <4 x i32> %tmp70, ptr getelementptr ([6 x i32], ptr @d, i64 2, i64 0), align 16
135  %tmp72 = load ptr, ptr @b, align 8
136  %tmp73 = load i32, ptr %tmp72, align 4
137  %cmp42 = icmp eq i32 %tmp73, 0
138  br i1 %cmp42, label %land.lhs.true, label %if.then
139
140land.lhs.true:                                    ; preds = %for.cond14.preheader
141  %Transform8x8Mode = getelementptr inbounds %struct.InputParameters, ptr %tmp72, i64 0, i32 1
142  %tmp74 = load i32, ptr %Transform8x8Mode, align 4
143  %tobool = icmp eq i32 %tmp74, 0
144  br i1 %tobool, label %if.then, label %for.inc45
145
146if.then:                                          ; preds = %land.lhs.true, %for.cond14.preheader
147  %call = tail call i32 @distortion4x4(ptr nonnull @d) #3
148  %add44 = add nsw i32 %call, %m.2124
149  br label %for.inc45
150
151for.inc45:                                        ; preds = %if.then, %land.lhs.true
152  %m.3 = phi i32 [ %m.2124, %land.lhs.true ], [ %add44, %if.then ]
153  %cmp8 = icmp slt i64 %indvars.iv145, %tmp12
154  br i1 %cmp8, label %for.inc45.for.body9_crit_edge, label %for.inc48
155
156for.inc45.for.body9_crit_edge:                    ; preds = %for.inc45
157  %indvars.iv.next146 = add nsw i64 %indvars.iv145, 4
158  %.pre = load ptr, ptr @c, align 8
159  br label %for.body9
160
161for.inc48:                                        ; preds = %for.inc45
162  %m.3.lcssa = phi i32 [ %m.3, %for.inc45 ]
163  %indvars.iv.next149 = add nsw i64 %indvars.iv148, 4
164  %cmp3 = icmp slt i64 %indvars.iv148, %tmp14
165  br i1 %cmp3, label %for.body4, label %for.end50
166
167for.end50:                                        ; preds = %for.inc48
168  %m.3.lcssa.lcssa = phi i32 [ %m.3.lcssa, %for.inc48 ]
169  %tmp75 = load ptr, ptr @b, align 8
170  %tmp76 = load i32, ptr %tmp75, align 4
171  %cmp52 = icmp eq i32 %tmp76, 0
172  br i1 %cmp52, label %land.lhs.true54, label %for.inc73
173
174land.lhs.true54:                                  ; preds = %for.end50
175  %Transform8x8Mode55 = getelementptr inbounds %struct.InputParameters, ptr %tmp75, i64 0, i32 1
176  %tmp77 = load i32, ptr %Transform8x8Mode55, align 4
177  %tobool56 = icmp eq i32 %tmp77, 0
178  br i1 %tobool56, label %for.inc73, label %for.body61.preheader
179
180for.body61.preheader:                             ; preds = %land.lhs.true54
181  call void @llvm.memcpy.p0.p0.i64(ptr align 16 nonnull getelementptr ([4 x i32], ptr @e, i64 4, i64 0), ptr align 16 %n, i64 32, i1 false)
182  call void @llvm.memcpy.p0.p0.i64(ptr align 16 nonnull getelementptr ([4 x i32], ptr @e, i64 6, i64 0), ptr align 16 %arraydecay.1, i64 32, i1 false)
183  call void @llvm.memcpy.p0.p0.i64(ptr align 16 nonnull getelementptr ([4 x i32], ptr @e, i64 8, i64 0), ptr align 16 %arraydecay.2, i64 32, i1 false)
184  call void @llvm.memcpy.p0.p0.i64(ptr align 16 nonnull getelementptr ([4 x i32], ptr @e, i64 10, i64 0), ptr align 16 %arraydecay.3, i64 32, i1 false)
185  call void @llvm.memcpy.p0.p0.i64(ptr align 16 nonnull getelementptr ([4 x i32], ptr @e, i64 12, i64 0), ptr align 16 %arraydecay.4, i64 32, i1 false)
186  call void @llvm.memcpy.p0.p0.i64(ptr align 16 nonnull getelementptr ([4 x i32], ptr @e, i64 14, i64 0), ptr align 16 %arraydecay.5, i64 32, i1 false)
187  call void @llvm.memcpy.p0.p0.i64(ptr align 16 nonnull getelementptr ([4 x i32], ptr @e, i64 16, i64 0), ptr align 16 %arraydecay.6, i64 32, i1 false)
188  call void @llvm.memcpy.p0.p0.i64(ptr align 16 nonnull getelementptr ([4 x i32], ptr @e, i64 18, i64 0), ptr align 16 %arraydecay.7, i64 32, i1 false)
189  %call70 = tail call i32 @distortion4x4(ptr nonnull @e) #3
190  %add71 = add nsw i32 %call70, %m.3.lcssa.lcssa
191  br label %for.inc73
192
193for.inc73:                                        ; preds = %for.body61.preheader, %land.lhs.true54, %for.end50
194  %m.4 = phi i32 [ %add71, %for.body61.preheader ], [ %m.3.lcssa.lcssa, %land.lhs.true54 ], [ %m.3.lcssa.lcssa, %for.end50 ]
195  %inc74 = add nuw nsw i32 %q.0131, 1
196  %exitcond156 = icmp eq i32 %inc74, 4
197  br i1 %exitcond156, label %for.end75, label %for.body
198
199for.end75:                                        ; preds = %for.inc73
200  %m.4.lcssa = phi i32 [ %m.4, %for.inc73 ]
201  call void @llvm.lifetime.end.p0(i64 256, ptr %n) #3
202  ret i32 %m.4.lcssa
203}
204
205; Function Attrs: argmemonly nounwind
206declare void @llvm.lifetime.start.p0(i64, ptr nocapture) #1
207
208declare void @LumaPrediction4x4(i32, i32, i32, i32, i32, i16 signext, i16 signext) #2
209
210declare i32 @distortion4x4(ptr) #2
211
212; Function Attrs: argmemonly nounwind
213declare void @llvm.memcpy.p0.p0.i64(ptr nocapture, ptr nocapture readonly, i64, i1) #1
214
215; Function Attrs: argmemonly nounwind
216declare void @llvm.lifetime.end.p0(i64, ptr nocapture) #1
217
218attributes #0 = { nounwind uwtable "disable-tail-calls"="false" "less-precise-fpmad"="false" "frame-pointer"="none" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+fxsr,+mmx,+sse,+sse2" "unsafe-fp-math"="false" "use-soft-float"="false" }
219attributes #1 = { argmemonly nounwind }
220attributes #2 = { "disable-tail-calls"="false" "less-precise-fpmad"="false" "frame-pointer"="none" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+fxsr,+mmx,+sse,+sse2" "unsafe-fp-math"="false" "use-soft-float"="false" }
221attributes #3 = { nounwind }
222
223