xref: /llvm-project/llvm/test/Transforms/LoopVectorize/X86/pr48340.ll (revision 38fffa630ee80163dc65e759392ad29798905679) 
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt -passes=loop-vectorize --force-vector-width=4 --force-vector-interleave=0 -S -o - < %s | FileCheck %s
3
4target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
5target triple = "x86_64-unknown-linux-gnu"
6
7%0 = type { i32 }
8%1 = type { i64 }
9
10define ptr @foo(ptr %p, ptr %p.last) unnamed_addr #0 {
11; CHECK-LABEL: @foo(
12; CHECK-NEXT:  entry:
13; CHECK-NEXT:    [[P3:%.*]] = ptrtoint ptr [[P:%.*]] to i64
14; CHECK-NEXT:    [[P_LAST1:%.*]] = ptrtoint ptr [[P_LAST:%.*]] to i64
15; CHECK-NEXT:    [[TMP0:%.*]] = add i64 [[P_LAST1]], -1024
16; CHECK-NEXT:    [[TMP1:%.*]] = sub i64 [[TMP0]], [[P3]]
17; CHECK-NEXT:    [[TMP2:%.*]] = lshr i64 [[TMP1]], 10
18; CHECK-NEXT:    [[TMP3:%.*]] = add nuw nsw i64 [[TMP2]], 1
19; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP3]], 16
20; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
21; CHECK:       vector.ph:
22; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[TMP3]], 16
23; CHECK-NEXT:    [[N_VEC:%.*]] = sub i64 [[TMP3]], [[N_MOD_VF]]
24; CHECK-NEXT:    [[TMP4:%.*]] = mul i64 [[N_VEC]], 1024
25; CHECK-NEXT:    [[IND_END:%.*]] = getelementptr i8, ptr [[P]], i64 [[TMP4]]
26; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
27; CHECK:       vector.body:
28; CHECK-NEXT:    [[POINTER_PHI:%.*]] = phi ptr [ [[P]], [[VECTOR_PH]] ], [ [[PTR_IND:%.*]], [[VECTOR_BODY]] ]
29; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
30; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr i8, ptr [[POINTER_PHI]], <4 x i64> <i64 0, i64 1024, i64 2048, i64 3072>
31; CHECK-NEXT:    [[TMP8:%.*]] = getelementptr i8, ptr [[POINTER_PHI]], <4 x i64> <i64 12288, i64 13312, i64 14336, i64 15360>
32; CHECK-NEXT:    [[WIDE_MASKED_GATHER6:%.*]] = call <4 x ptr> @llvm.masked.gather.v4p0.v4p0(<4 x ptr> [[TMP8]], i32 8, <4 x i1> splat (i1 true), <4 x ptr> poison)
33; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
34; CHECK-NEXT:    [[PTR_IND]] = getelementptr i8, ptr [[POINTER_PHI]], i64 16384
35; CHECK-NEXT:    [[TMP9:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
36; CHECK-NEXT:    br i1 [[TMP9]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
37; CHECK:       middle.block:
38; CHECK-NEXT:    [[TMP10:%.*]] = extractelement <4 x ptr> [[WIDE_MASKED_GATHER6]], i32 3
39; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[TMP3]], [[N_VEC]]
40; CHECK-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
41; CHECK:       scalar.ph:
42; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi ptr [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[P]], [[ENTRY:%.*]] ]
43; CHECK-NEXT:    br label [[LOOP:%.*]]
44; CHECK:       loop:
45; CHECK-NEXT:    [[P2:%.*]] = phi ptr [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[P_INC:%.*]], [[LOOP]] ]
46; CHECK-NEXT:    [[P_INC]] = getelementptr inbounds i64, ptr [[P2]], i64 128
47; CHECK-NEXT:    [[V:%.*]] = load ptr, ptr [[P2]], align 8
48; CHECK-NEXT:    [[B:%.*]] = icmp eq ptr [[P_INC]], [[P_LAST]]
49; CHECK-NEXT:    br i1 [[B]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP3:![0-9]+]]
50; CHECK:       exit:
51; CHECK-NEXT:    [[V_LCSSA:%.*]] = phi ptr [ [[V]], [[LOOP]] ], [ [[TMP10]], [[MIDDLE_BLOCK]] ]
52; CHECK-NEXT:    ret ptr [[V_LCSSA]]
53;
54entry:
55  br label %loop
56
57loop:
58  %p2 = phi ptr [ %p, %entry ], [ %p.inc, %loop ]
59  %p.inc = getelementptr inbounds i64, ptr %p2, i64 128
60  %v = load ptr, ptr %p2, align 8
61  %b = icmp eq ptr %p.inc, %p.last
62  br i1 %b, label %exit, label %loop
63
64exit:
65  ret ptr %v
66}
67
68define ptr @bar(ptr %p, ptr %p.last) unnamed_addr #0 {
69; CHECK-LABEL: @bar(
70; CHECK-NEXT:  entry:
71; CHECK-NEXT:    [[P3:%.*]] = ptrtoint ptr [[P:%.*]] to i64
72; CHECK-NEXT:    [[P_LAST1:%.*]] = ptrtoint ptr [[P_LAST:%.*]] to i64
73; CHECK-NEXT:    [[TMP0:%.*]] = add i64 [[P_LAST1]], -1024
74; CHECK-NEXT:    [[TMP1:%.*]] = sub i64 [[TMP0]], [[P3]]
75; CHECK-NEXT:    [[TMP2:%.*]] = lshr i64 [[TMP1]], 10
76; CHECK-NEXT:    [[TMP3:%.*]] = add nuw nsw i64 [[TMP2]], 1
77; CHECK-NEXT:    [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP3]], 16
78; CHECK-NEXT:    br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
79; CHECK:       vector.ph:
80; CHECK-NEXT:    [[N_MOD_VF:%.*]] = urem i64 [[TMP3]], 16
81; CHECK-NEXT:    [[N_VEC:%.*]] = sub i64 [[TMP3]], [[N_MOD_VF]]
82; CHECK-NEXT:    [[TMP4:%.*]] = mul i64 [[N_VEC]], 1024
83; CHECK-NEXT:    [[IND_END:%.*]] = getelementptr i8, ptr [[P]], i64 [[TMP4]]
84; CHECK-NEXT:    br label [[VECTOR_BODY:%.*]]
85; CHECK:       vector.body:
86; CHECK-NEXT:    [[POINTER_PHI:%.*]] = phi ptr [ [[P]], [[VECTOR_PH]] ], [ [[PTR_IND:%.*]], [[VECTOR_BODY]] ]
87; CHECK-NEXT:    [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
88; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr i8, ptr [[POINTER_PHI]], <4 x i64> <i64 0, i64 1024, i64 2048, i64 3072>
89; CHECK-NEXT:    [[TMP8:%.*]] = getelementptr i8, ptr [[POINTER_PHI]], <4 x i64> <i64 12288, i64 13312, i64 14336, i64 15360>
90; CHECK-NEXT:    [[WIDE_MASKED_GATHER6:%.*]] = call <4 x ptr> @llvm.masked.gather.v4p0.v4p0(<4 x ptr> [[TMP8]], i32 8, <4 x i1> splat (i1 true), <4 x ptr> poison)
91; CHECK-NEXT:    [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
92; CHECK-NEXT:    [[PTR_IND]] = getelementptr i8, ptr [[POINTER_PHI]], i64 16384
93; CHECK-NEXT:    [[TMP9:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
94; CHECK-NEXT:    br i1 [[TMP9]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
95; CHECK:       middle.block:
96; CHECK-NEXT:    [[TMP10:%.*]] = extractelement <4 x ptr> [[WIDE_MASKED_GATHER6]], i32 3
97; CHECK-NEXT:    [[CMP_N:%.*]] = icmp eq i64 [[TMP3]], [[N_VEC]]
98; CHECK-NEXT:    br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
99; CHECK:       scalar.ph:
100; CHECK-NEXT:    [[BC_RESUME_VAL:%.*]] = phi ptr [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[P]], [[ENTRY:%.*]] ]
101; CHECK-NEXT:    br label [[LOOP:%.*]]
102; CHECK:       loop:
103; CHECK-NEXT:    [[P2:%.*]] = phi ptr [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[P_INC:%.*]], [[LOOP]] ]
104; CHECK-NEXT:    [[P_INC]] = getelementptr inbounds i64, ptr [[P2]], i64 128
105; CHECK-NEXT:    [[V:%.*]] = load ptr, ptr [[P2]], align 8
106; CHECK-NEXT:    [[B:%.*]] = icmp eq ptr [[P_INC]], [[P_LAST]]
107; CHECK-NEXT:    br i1 [[B]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP5:![0-9]+]]
108; CHECK:       exit:
109; CHECK-NEXT:    [[V_LCSSA:%.*]] = phi ptr [ [[V]], [[LOOP]] ], [ [[TMP10]], [[MIDDLE_BLOCK]] ]
110; CHECK-NEXT:    ret ptr [[V_LCSSA]]
111;
112entry:
113  br label %loop
114
115loop:
116  %p2 = phi ptr [ %p, %entry ], [ %p.inc, %loop ]
117  %p.inc = getelementptr inbounds i64, ptr %p2, i64 128
118  %v = load ptr, ptr %p2, align 8
119  %b = icmp eq ptr %p.inc, %p.last
120  br i1 %b, label %exit, label %loop
121
122exit:
123  ret ptr %v
124}
125
126attributes #0 = { "target-cpu"="skylake" }
127
128