xref: /llvm-project/llvm/test/Transforms/InstCombine/gep-combine-loop-invariant.ll (revision ed1cb01baf1727ce2eb9c58deafe1a92d6fc65b7) 
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt < %s -instcombine -S -enable-new-pm=0 | FileCheck %s
3; RUN: opt < %s -passes='require<loops>,instcombine' -S | FileCheck %s
4target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
5target triple = "x86_64-unknown-linux-gnu"
6
7define i32 @foo(i8* nocapture readnone %match, i32 %cur_match, i32 %best_len, i32 %scan_end, i32* nocapture readonly %prev, i32 %limit, i32 %chain_length, i8* nocapture readonly %win, i32 %wmask) {
8; CHECK-LABEL: @foo(
9; CHECK-NEXT:  entry:
10; CHECK-NEXT:    [[IDX_EXT2:%.*]] = zext i32 [[CUR_MATCH:%.*]] to i64
11; CHECK-NEXT:    [[ADD_PTR4:%.*]] = getelementptr inbounds i8, i8* [[WIN:%.*]], i64 [[IDX_EXT2]]
12; CHECK-NEXT:    [[IDX_EXT1:%.*]] = zext i32 [[BEST_LEN:%.*]] to i64
13; CHECK-NEXT:    [[ADD_PTR25:%.*]] = getelementptr inbounds i8, i8* [[ADD_PTR4]], i64 [[IDX_EXT1]]
14; CHECK-NEXT:    [[ADD_PTR36:%.*]] = getelementptr inbounds i8, i8* [[ADD_PTR25]], i64 -1
15; CHECK-NEXT:    [[TMP0:%.*]] = bitcast i8* [[ADD_PTR36]] to i32*
16; CHECK-NEXT:    [[TMP1:%.*]] = load i32, i32* [[TMP0]], align 4
17; CHECK-NEXT:    [[CMP7:%.*]] = icmp eq i32 [[TMP1]], [[SCAN_END:%.*]]
18; CHECK-NEXT:    br i1 [[CMP7]], label [[DO_END:%.*]], label [[IF_THEN_LR_PH:%.*]]
19; CHECK:       if.then.lr.ph:
20; CHECK-NEXT:    br label [[IF_THEN:%.*]]
21; CHECK:       do.body:
22; CHECK-NEXT:    [[IDX_EXT:%.*]] = zext i32 [[TMP4:%.*]] to i64
23; CHECK-NEXT:    [[ADD_PTR:%.*]] = getelementptr inbounds i8, i8* [[WIN]], i64 [[IDX_EXT1]]
24; CHECK-NEXT:    [[ADD_PTR2:%.*]] = getelementptr inbounds i8, i8* [[ADD_PTR]], i64 -1
25; CHECK-NEXT:    [[ADD_PTR3:%.*]] = getelementptr inbounds i8, i8* [[ADD_PTR2]], i64 [[IDX_EXT]]
26; CHECK-NEXT:    [[TMP2:%.*]] = bitcast i8* [[ADD_PTR3]] to i32*
27; CHECK-NEXT:    [[TMP3:%.*]] = load i32, i32* [[TMP2]], align 4
28; CHECK-NEXT:    [[CMP:%.*]] = icmp eq i32 [[TMP3]], [[SCAN_END]]
29; CHECK-NEXT:    br i1 [[CMP]], label [[DO_END]], label [[IF_THEN]]
30; CHECK:       if.then:
31; CHECK-NEXT:    [[CUR_MATCH_ADDR_09:%.*]] = phi i32 [ [[CUR_MATCH]], [[IF_THEN_LR_PH]] ], [ [[TMP4]], [[DO_BODY:%.*]] ]
32; CHECK-NEXT:    [[CHAIN_LENGTH_ADDR_08:%.*]] = phi i32 [ [[CHAIN_LENGTH:%.*]], [[IF_THEN_LR_PH]] ], [ [[DEC:%.*]], [[DO_BODY]] ]
33; CHECK-NEXT:    [[AND:%.*]] = and i32 [[CUR_MATCH_ADDR_09]], [[WMASK:%.*]]
34; CHECK-NEXT:    [[IDXPROM:%.*]] = zext i32 [[AND]] to i64
35; CHECK-NEXT:    [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[PREV:%.*]], i64 [[IDXPROM]]
36; CHECK-NEXT:    [[TMP4]] = load i32, i32* [[ARRAYIDX]], align 4
37; CHECK-NEXT:    [[CMP4:%.*]] = icmp ugt i32 [[TMP4]], [[LIMIT:%.*]]
38; CHECK-NEXT:    br i1 [[CMP4]], label [[LAND_LHS_TRUE:%.*]], label [[DO_END]]
39; CHECK:       land.lhs.true:
40; CHECK-NEXT:    [[DEC]] = add i32 [[CHAIN_LENGTH_ADDR_08]], -1
41; CHECK-NEXT:    [[CMP5:%.*]] = icmp eq i32 [[DEC]], 0
42; CHECK-NEXT:    br i1 [[CMP5]], label [[DO_END]], label [[DO_BODY]]
43; CHECK:       do.end:
44; CHECK-NEXT:    [[CONT_0:%.*]] = phi i32 [ 1, [[ENTRY:%.*]] ], [ 0, [[IF_THEN]] ], [ 0, [[LAND_LHS_TRUE]] ], [ 1, [[DO_BODY]] ]
45; CHECK-NEXT:    ret i32 [[CONT_0]]
46;
47entry:
48  %idx.ext2 = zext i32 %cur_match to i64
49  %add.ptr4 = getelementptr inbounds i8, i8* %win, i64 %idx.ext2
50  %idx.ext1 = zext i32 %best_len to i64
51  %add.ptr25 = getelementptr inbounds i8, i8* %add.ptr4, i64 %idx.ext1
52  %add.ptr36 = getelementptr inbounds i8, i8* %add.ptr25, i64 -1
53  %0 = bitcast i8* %add.ptr36 to i32*
54  %1 = load i32, i32* %0, align 4
55  %cmp7 = icmp eq i32 %1, %scan_end
56  br i1 %cmp7, label %do.end, label %if.then.lr.ph
57
58if.then.lr.ph:                                    ; preds = %entry
59  br label %if.then
60
61do.body:                                          ; preds = %land.lhs.true
62  %chain_length.addr.0 = phi i32 [ %dec, %land.lhs.true ]
63  %cur_match.addr.0 = phi i32 [ %4, %land.lhs.true ]
64  %idx.ext = zext i32 %cur_match.addr.0 to i64
65  %add.ptr = getelementptr inbounds i8, i8* %win, i64 %idx.ext
66  %add.ptr2 = getelementptr inbounds i8, i8* %add.ptr, i64 %idx.ext1
67  %add.ptr3 = getelementptr inbounds i8, i8* %add.ptr2, i64 -1
68  %2 = bitcast i8* %add.ptr3 to i32*
69  %3 = load i32, i32* %2, align 4
70  %cmp = icmp eq i32 %3, %scan_end
71  br i1 %cmp, label %do.end, label %if.then
72
73if.then:                                          ; preds = %if.then.lr.ph, %do.body
74  %cur_match.addr.09 = phi i32 [ %cur_match, %if.then.lr.ph ], [ %cur_match.addr.0, %do.body ]
75  %chain_length.addr.08 = phi i32 [ %chain_length, %if.then.lr.ph ], [ %chain_length.addr.0, %do.body ]
76  %and = and i32 %cur_match.addr.09, %wmask
77  %idxprom = zext i32 %and to i64
78  %arrayidx = getelementptr inbounds i32, i32* %prev, i64 %idxprom
79  %4 = load i32, i32* %arrayidx, align 4
80  %cmp4 = icmp ugt i32 %4, %limit
81  br i1 %cmp4, label %land.lhs.true, label %do.end
82
83land.lhs.true:                                    ; preds = %if.then
84  %dec = add i32 %chain_length.addr.08, -1
85  %cmp5 = icmp eq i32 %dec, 0
86  br i1 %cmp5, label %do.end, label %do.body
87
88do.end:                                           ; preds = %do.body, %land.lhs.true, %if.then, %entry
89  %cont.0 = phi i32 [ 1, %entry ], [ 0, %if.then ], [ 0, %land.lhs.true ], [ 1, %do.body ]
90  ret i32 %cont.0
91}
92
93declare void @blackhole(<2 x i8*>)
94
95define void @PR37005(i8* %base, i8** %in) {
96; CHECK-LABEL: @PR37005(
97; CHECK-NEXT:  entry:
98; CHECK-NEXT:    br label [[LOOP:%.*]]
99; CHECK:       loop:
100; CHECK-NEXT:    [[E2:%.*]] = getelementptr inbounds i8*, i8** [[IN:%.*]], i64 undef
101; CHECK-NEXT:    [[E4:%.*]] = getelementptr inbounds i8*, i8** [[E2]], <2 x i64> <i64 0, i64 1>
102; CHECK-NEXT:    [[PI1:%.*]] = ptrtoint <2 x i8**> [[E4]] to <2 x i64>
103; CHECK-NEXT:    [[TMP0:%.*]] = lshr <2 x i64> [[PI1]], <i64 14, i64 14>
104; CHECK-NEXT:    [[SL1:%.*]] = and <2 x i64> [[TMP0]], <i64 1125899906842496, i64 1125899906842496>
105; CHECK-NEXT:    [[E51:%.*]] = getelementptr inbounds i8, i8* [[BASE:%.*]], i64 80
106; CHECK-NEXT:    [[E6:%.*]] = getelementptr inbounds i8, i8* [[E51]], <2 x i64> [[SL1]]
107; CHECK-NEXT:    call void @blackhole(<2 x i8*> [[E6]])
108; CHECK-NEXT:    br label [[LOOP]]
109;
110entry:
111  br label %loop
112
113loop:
114  %e1 = getelementptr inbounds i8*, i8** %in, i64 undef
115  %e2 = getelementptr inbounds i8*, i8** %e1, i64 6
116  %bc1 = bitcast i8** %e2 to <2 x i8*>*
117  %e3 = getelementptr inbounds <2 x i8*>, <2 x i8*>* %bc1, i64 0, i64 0
118  %e4 = getelementptr inbounds i8*, i8** %e3, <2 x i64> <i64 0, i64 1>
119  %pi1 = ptrtoint <2 x i8**> %e4 to <2 x i64>
120  %lr1 = lshr <2 x i64> %pi1, <i64 21, i64 21>
121  %sl1 = shl nuw nsw <2 x i64> %lr1, <i64 7, i64 7>
122  %e5 = getelementptr inbounds i8, i8* %base, <2 x i64> %sl1
123  %e6 = getelementptr inbounds i8, <2 x i8*> %e5, i64 80
124  call void @blackhole(<2 x i8*> %e6)
125  br label %loop
126}
127
128define void @PR37005_2(i8* %base, i8** %in) {
129; CHECK-LABEL: @PR37005_2(
130; CHECK-NEXT:  entry:
131; CHECK-NEXT:    br label [[LOOP:%.*]]
132; CHECK:       loop:
133; CHECK-NEXT:    [[E2:%.*]] = getelementptr inbounds i8*, i8** [[IN:%.*]], i64 undef
134; CHECK-NEXT:    [[PI1:%.*]] = ptrtoint i8** [[E2]] to i64
135; CHECK-NEXT:    [[TMP0:%.*]] = lshr i64 [[PI1]], 14
136; CHECK-NEXT:    [[SL1:%.*]] = and i64 [[TMP0]], 1125899906842496
137; CHECK-NEXT:    [[E51:%.*]] = getelementptr inbounds i8, i8* [[BASE:%.*]], <2 x i64> <i64 80, i64 60>
138; CHECK-NEXT:    [[E6:%.*]] = getelementptr inbounds i8, <2 x i8*> [[E51]], i64 [[SL1]]
139; CHECK-NEXT:    call void @blackhole(<2 x i8*> [[E6]])
140; CHECK-NEXT:    br label [[LOOP]]
141;
142entry:
143  br label %loop
144
145loop:
146  %e1 = getelementptr inbounds i8*, i8** %in, i64 undef
147  %e2 = getelementptr inbounds i8*, i8** %e1, i64 6
148  %pi1 = ptrtoint i8** %e2 to i64
149  %lr1 = lshr i64 %pi1, 21
150  %sl1 = shl nuw nsw i64 %lr1, 7
151  %e5 = getelementptr inbounds i8, i8* %base, i64 %sl1
152  %e6 = getelementptr inbounds i8, i8* %e5, <2 x i64> <i64 80, i64 60>
153  call void @blackhole(<2 x i8*> %e6)
154  br label %loop
155}
156
157define void @PR37005_3(<2 x i8*> %base, i8** %in) {
158; CHECK-LABEL: @PR37005_3(
159; CHECK-NEXT:  entry:
160; CHECK-NEXT:    br label [[LOOP:%.*]]
161; CHECK:       loop:
162; CHECK-NEXT:    [[E2:%.*]] = getelementptr inbounds i8*, i8** [[IN:%.*]], i64 undef
163; CHECK-NEXT:    [[E4:%.*]] = getelementptr inbounds i8*, i8** [[E2]], <2 x i64> <i64 0, i64 1>
164; CHECK-NEXT:    [[PI1:%.*]] = ptrtoint <2 x i8**> [[E4]] to <2 x i64>
165; CHECK-NEXT:    [[TMP0:%.*]] = lshr <2 x i64> [[PI1]], <i64 14, i64 14>
166; CHECK-NEXT:    [[SL1:%.*]] = and <2 x i64> [[TMP0]], <i64 1125899906842496, i64 1125899906842496>
167; CHECK-NEXT:    [[E5:%.*]] = getelementptr inbounds i8, <2 x i8*> [[BASE:%.*]], i64 80
168; CHECK-NEXT:    [[E6:%.*]] = getelementptr inbounds i8, <2 x i8*> [[E5]], <2 x i64> [[SL1]]
169; CHECK-NEXT:    call void @blackhole(<2 x i8*> [[E6]])
170; CHECK-NEXT:    br label [[LOOP]]
171;
172entry:
173  br label %loop
174
175loop:
176  %e1 = getelementptr inbounds i8*, i8** %in, i64 undef
177  %e2 = getelementptr inbounds i8*, i8** %e1, i64 6
178  %bc1 = bitcast i8** %e2 to <2 x i8*>*
179  %e3 = getelementptr inbounds <2 x i8*>, <2 x i8*>* %bc1, i64 0, i64 0
180  %e4 = getelementptr inbounds i8*, i8** %e3, <2 x i64> <i64 0, i64 1>
181  %pi1 = ptrtoint <2 x i8**> %e4 to <2 x i64>
182  %lr1 = lshr <2 x i64> %pi1, <i64 21, i64 21>
183  %sl1 = shl nuw nsw <2 x i64> %lr1, <i64 7, i64 7>
184  %e5 = getelementptr inbounds i8, <2 x i8*> %base, <2 x i64> %sl1
185  %e6 = getelementptr inbounds i8, <2 x i8*> %e5, i64 80
186  call void @blackhole(<2 x i8*> %e6)
187  br label %loop
188}
189
190; This would crash because we did not expect to be able to constant fold a GEP.
191
192define void @PR51485(<2 x i64> %v) {
193; CHECK-LABEL: @PR51485(
194; CHECK-NEXT:  entry:
195; CHECK-NEXT:    br label [[LOOP:%.*]]
196; CHECK:       loop:
197; CHECK-NEXT:    [[SL1:%.*]] = shl nuw nsw <2 x i64> [[V:%.*]], <i64 7, i64 7>
198; CHECK-NEXT:    [[E6:%.*]] = getelementptr inbounds i8, i8* getelementptr inbounds (i8, i8* bitcast (void (<2 x i64>)* @PR51485 to i8*), i64 80), <2 x i64> [[SL1]]
199; CHECK-NEXT:    call void @blackhole(<2 x i8*> [[E6]])
200; CHECK-NEXT:    br label [[LOOP]]
201;
202entry:
203  br label %loop
204
205loop:
206  %sl1 = shl nuw nsw <2 x i64> %v, <i64 7, i64 7>
207  %e5 = getelementptr inbounds i8, i8* bitcast (void (<2 x i64>)* @PR51485 to i8*), <2 x i64> %sl1
208  %e6 = getelementptr inbounds i8, <2 x i8*> %e5, i64 80
209  call void @blackhole(<2 x i8*> %e6)
210  br label %loop
211}
212
213; Avoid folding the GEP outside the loop to inside, and increasing loop
214; instruction count.
215define float @gep_cross_loop(i64* %_arg_, float* %_arg_3, float %_arg_8)
216; CHECK-LABEL: @gep_cross_loop(
217; CHECK-NEXT:  entry:
218; CHECK-NEXT:    [[TMP0:%.*]] = load i64, i64* [[_ARG_:%.*]], align 8
219; CHECK-NEXT:    [[ADD_PTR:%.*]] = getelementptr inbounds float, float* [[_ARG_3:%.*]], i64 [[TMP0]]
220; CHECK-NEXT:    br label [[FOR_COND_I:%.*]]
221; CHECK:       for.cond.i:
222; CHECK-NEXT:    [[IDX:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[ADD11_I:%.*]], [[FOR_BODY_I:%.*]] ]
223; CHECK-NEXT:    [[SUM:%.*]] = phi float [ 0.000000e+00, [[ENTRY]] ], [ [[ADD_I:%.*]], [[FOR_BODY_I]] ]
224; CHECK-NEXT:    [[CMP:%.*]] = icmp ult i64 [[IDX]], 17
225; CHECK-NEXT:    br i1 [[CMP]], label [[FOR_BODY_I]], label [[FOR_COND_I_I_I_PREHEADER:%.*]]
226; CHECK:       for.cond.i.i.i.preheader:
227; CHECK-NEXT:    ret float [[SUM]]
228; CHECK:       for.body.i:
229; CHECK-NEXT:    [[ARRAYIDX_I84_I:%.*]] = getelementptr inbounds float, float* [[ADD_PTR]], i64 [[IDX]]
230; CHECK-NEXT:    [[TMP1:%.*]] = load float, float* [[ARRAYIDX_I84_I]], align 4
231; CHECK-NEXT:    [[ADD_I]] = fadd fast float [[SUM]], [[TMP1]]
232; CHECK-NEXT:    [[ADD11_I]] = add nuw nsw i64 [[IDX]], 1
233; CHECK-NEXT:    br label [[FOR_COND_I]]
234;
235{
236entry:
237  %0 = load i64, i64* %_arg_, align 8
238  %add.ptr = getelementptr inbounds float, float* %_arg_3, i64 %0
239  br label %for.cond.i
240
241for.cond.i:                                       ; preds = %for.body.i, %entry
242  %idx = phi i64 [ 0, %entry ], [ %add11.i, %for.body.i ]
243  %sum = phi float [ 0.000000e+00, %entry ], [ %add.i, %for.body.i ]
244  %cmp = icmp ule i64 %idx, 16
245  br i1 %cmp, label %for.body.i, label %for.cond.i.i.i.preheader
246
247for.cond.i.i.i.preheader:                         ; preds = %for.cond.i
248  ret float %sum
249
250for.body.i:                                       ; preds = %for.cond.i
251  %arrayidx.i84.i = getelementptr inbounds float, float * %add.ptr, i64 %idx
252  %1 = load float, float* %arrayidx.i84.i, align 4
253  %add.i = fadd fast float %sum, %1
254  %add11.i = add nsw i64 %idx, 1
255  br label %for.cond.i
256}
257