Skip to content

Commit 4a7c0b8

Browse files
fhahnfhahn
committed
[LV] Add X86-specific induction step tests.
Adds additional test coverage for induction codegen.
1 parent da2a9ed commit 4a7c0b8

File tree

1 file changed

+155
-0
lines changed

1 file changed

+155
-0
lines changed

‎llvm/test/Transforms/LoopVectorize/X86/induction-step.ll

Lines changed: 155 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,155 @@
1+
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2+
; RUN: opt -passes=loop-vectorize -force-vector-width=4 -S %s | FileCheck %s
3+
4+
target triple = "x86_64-unknown-linux-gnu"
5+
6+
define i16 @wide_add_induction_step_live_in(ptr %dst, i64 %N, i16 %off) {
7+
; CHECK-LABEL: @wide_add_induction_step_live_in(
8+
; CHECK-NEXT: entry:
9+
; CHECK-NEXT: [[O_1:%.*]] = add i16 [[OFF:%.*]], 2
10+
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], 8
11+
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
12+
; CHECK: vector.ph:
13+
; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], 8
14+
; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]]
15+
; CHECK-NEXT: [[DOTCAST:%.*]] = trunc i64 [[N_VEC]] to i16
16+
; CHECK-NEXT: [[TMP0:%.*]] = mul i16 [[DOTCAST]], [[O_1]]
17+
; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i16> poison, i16 [[O_1]], i64 0
18+
; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i16> [[BROADCAST_SPLATINSERT]], <4 x i16> poison, <4 x i32> zeroinitializer
19+
; CHECK-NEXT: [[TMP1:%.*]] = mul <4 x i16> splat (i16 4), [[BROADCAST_SPLAT]]
20+
; CHECK-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <4 x i16> poison, i16 [[O_1]], i64 0
21+
; CHECK-NEXT: [[DOTSPLAT:%.*]] = shufflevector <4 x i16> [[DOTSPLATINSERT]], <4 x i16> poison, <4 x i32> zeroinitializer
22+
; CHECK-NEXT: [[TMP2:%.*]] = mul <4 x i16> <i16 0, i16 1, i16 2, i16 3>, [[DOTSPLAT]]
23+
; CHECK-NEXT: [[INDUCTION:%.*]] = add <4 x i16> zeroinitializer, [[TMP2]]
24+
; CHECK-NEXT: [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <4 x i16> poison, i16 [[O_1]], i64 0
25+
; CHECK-NEXT: [[BROADCAST_SPLAT2:%.*]] = shufflevector <4 x i16> [[BROADCAST_SPLATINSERT1]], <4 x i16> poison, <4 x i32> zeroinitializer
26+
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
27+
; CHECK: vector.body:
28+
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
29+
; CHECK-NEXT: [[VEC_IND:%.*]] = phi <4 x i16> [ [[INDUCTION]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
30+
; CHECK-NEXT: [[STEP_ADD:%.*]] = add <4 x i16> [[VEC_IND]], [[TMP1]]
31+
; CHECK-NEXT: [[TMP3:%.*]] = add i64 [[INDEX]], 0
32+
; CHECK-NEXT: [[TMP4:%.*]] = add <4 x i16> [[VEC_IND]], [[BROADCAST_SPLAT2]]
33+
; CHECK-NEXT: [[TMP9:%.*]] = add <4 x i16> [[STEP_ADD]], [[BROADCAST_SPLAT2]]
34+
; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds i16, ptr [[DST:%.*]], i64 [[TMP3]]
35+
; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds i16, ptr [[TMP5]], i32 0
36+
; CHECK-NEXT: [[TMP8:%.*]] = getelementptr inbounds i16, ptr [[TMP5]], i32 4
37+
; CHECK-NEXT: store <4 x i16> [[TMP4]], ptr [[TMP6]], align 2
38+
; CHECK-NEXT: store <4 x i16> [[TMP9]], ptr [[TMP8]], align 2
39+
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8
40+
; CHECK-NEXT: [[VEC_IND_NEXT]] = add <4 x i16> [[STEP_ADD]], [[TMP1]]
41+
; CHECK-NEXT: [[TMP7:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
42+
; CHECK-NEXT: br i1 [[TMP7]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
43+
; CHECK: middle.block:
44+
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
45+
; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
46+
; CHECK: scalar.ph:
47+
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
48+
; CHECK-NEXT: [[BC_RESUME_VAL1:%.*]] = phi i16 [ [[TMP0]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ]
49+
; CHECK-NEXT: br label [[LOOP:%.*]]
50+
; CHECK: loop:
51+
; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
52+
; CHECK-NEXT: [[IV_2:%.*]] = phi i16 [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ], [ [[ADD:%.*]], [[LOOP]] ]
53+
; CHECK-NEXT: [[ADD]] = add i16 [[IV_2]], [[O_1]]
54+
; CHECK-NEXT: [[GEP_DST:%.*]] = getelementptr inbounds i16, ptr [[DST]], i64 [[IV]]
55+
; CHECK-NEXT: store i16 [[ADD]], ptr [[GEP_DST]], align 2
56+
; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1
57+
; CHECK-NEXT: [[EC:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]]
58+
; CHECK-NEXT: br i1 [[EC]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP3:![0-9]+]]
59+
; CHECK: exit:
60+
; CHECK-NEXT: [[ADD_LCSSA:%.*]] = phi i16 [ [[ADD]], [[LOOP]] ], [ [[TMP0]], [[MIDDLE_BLOCK]] ]
61+
; CHECK-NEXT: ret i16 [[ADD_LCSSA]]
62+
;
63+
entry:
64+
%o.1 = add i16 %off, 2
65+
br label %loop
66+
67+
loop:
68+
%iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
69+
%iv.2 = phi i16 [ 0, %entry ], [ %add, %loop ]
70+
%add = add i16 %iv.2, %o.1
71+
%gep.dst = getelementptr inbounds i16, ptr %dst, i64 %iv
72+
store i16 %add, ptr %gep.dst, align 2
73+
%iv.next = add nuw nsw i64 %iv, 1
74+
%ec = icmp eq i64 %iv.next, %N
75+
br i1 %ec , label %exit, label %loop
76+
77+
exit:
78+
ret i16 %add
79+
}
80+
81+
define i16 @wide_sub_induction_step_live_in(ptr %dst, i64 %N, i16 %off) {
82+
; CHECK-LABEL: @wide_sub_induction_step_live_in(
83+
; CHECK-NEXT: entry:
84+
; CHECK-NEXT: [[O_1:%.*]] = add i16 [[OFF:%.*]], 2
85+
; CHECK-NEXT: [[TMP0:%.*]] = sub i16 -2, [[OFF]]
86+
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], 8
87+
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
88+
; CHECK: vector.ph:
89+
; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], 8
90+
; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]]
91+
; CHECK-NEXT: [[DOTCAST:%.*]] = trunc i64 [[N_VEC]] to i16
92+
; CHECK-NEXT: [[TMP1:%.*]] = mul i16 [[DOTCAST]], [[TMP0]]
93+
; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <4 x i16> poison, i16 [[TMP0]], i64 0
94+
; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <4 x i16> [[BROADCAST_SPLATINSERT]], <4 x i16> poison, <4 x i32> zeroinitializer
95+
; CHECK-NEXT: [[TMP2:%.*]] = mul <4 x i16> splat (i16 4), [[BROADCAST_SPLAT]]
96+
; CHECK-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <4 x i16> poison, i16 [[TMP0]], i64 0
97+
; CHECK-NEXT: [[DOTSPLAT:%.*]] = shufflevector <4 x i16> [[DOTSPLATINSERT]], <4 x i16> poison, <4 x i32> zeroinitializer
98+
; CHECK-NEXT: [[TMP3:%.*]] = mul <4 x i16> <i16 0, i16 1, i16 2, i16 3>, [[DOTSPLAT]]
99+
; CHECK-NEXT: [[INDUCTION:%.*]] = add <4 x i16> zeroinitializer, [[TMP3]]
100+
; CHECK-NEXT: [[BROADCAST_SPLATINSERT1:%.*]] = insertelement <4 x i16> poison, i16 [[O_1]], i64 0
101+
; CHECK-NEXT: [[BROADCAST_SPLAT2:%.*]] = shufflevector <4 x i16> [[BROADCAST_SPLATINSERT1]], <4 x i16> poison, <4 x i32> zeroinitializer
102+
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
103+
; CHECK: vector.body:
104+
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
105+
; CHECK-NEXT: [[VEC_IND:%.*]] = phi <4 x i16> [ [[INDUCTION]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
106+
; CHECK-NEXT: [[STEP_ADD:%.*]] = add <4 x i16> [[VEC_IND]], [[TMP2]]
107+
; CHECK-NEXT: [[TMP4:%.*]] = add i64 [[INDEX]], 0
108+
; CHECK-NEXT: [[TMP5:%.*]] = sub <4 x i16> [[VEC_IND]], [[BROADCAST_SPLAT2]]
109+
; CHECK-NEXT: [[TMP10:%.*]] = sub <4 x i16> [[STEP_ADD]], [[BROADCAST_SPLAT2]]
110+
; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds i16, ptr [[DST:%.*]], i64 [[TMP4]]
111+
; CHECK-NEXT: [[TMP7:%.*]] = getelementptr inbounds i16, ptr [[TMP6]], i32 0
112+
; CHECK-NEXT: [[TMP9:%.*]] = getelementptr inbounds i16, ptr [[TMP6]], i32 4
113+
; CHECK-NEXT: store <4 x i16> [[TMP5]], ptr [[TMP7]], align 2
114+
; CHECK-NEXT: store <4 x i16> [[TMP10]], ptr [[TMP9]], align 2
115+
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8
116+
; CHECK-NEXT: [[VEC_IND_NEXT]] = add <4 x i16> [[STEP_ADD]], [[TMP2]]
117+
; CHECK-NEXT: [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
118+
; CHECK-NEXT: br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
119+
; CHECK: middle.block:
120+
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]]
121+
; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
122+
; CHECK: scalar.ph:
123+
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
124+
; CHECK-NEXT: [[BC_RESUME_VAL3:%.*]] = phi i16 [ [[TMP1]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ]
125+
; CHECK-NEXT: br label [[LOOP:%.*]]
126+
; CHECK: loop:
127+
; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
128+
; CHECK-NEXT: [[IV_2:%.*]] = phi i16 [ [[BC_RESUME_VAL3]], [[SCALAR_PH]] ], [ [[SUB:%.*]], [[LOOP]] ]
129+
; CHECK-NEXT: [[SUB]] = sub i16 [[IV_2]], [[O_1]]
130+
; CHECK-NEXT: [[GEP_DST:%.*]] = getelementptr inbounds i16, ptr [[DST]], i64 [[IV]]
131+
; CHECK-NEXT: store i16 [[SUB]], ptr [[GEP_DST]], align 2
132+
; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1
133+
; CHECK-NEXT: [[EC:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]]
134+
; CHECK-NEXT: br i1 [[EC]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP5:![0-9]+]]
135+
; CHECK: exit:
136+
; CHECK-NEXT: [[SUB_LCSSA:%.*]] = phi i16 [ [[SUB]], [[LOOP]] ], [ [[TMP1]], [[MIDDLE_BLOCK]] ]
137+
; CHECK-NEXT: ret i16 [[SUB_LCSSA]]
138+
;
139+
entry:
140+
%o.1 = add i16 %off, 2
141+
br label %loop
142+
143+
loop:
144+
%iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
145+
%iv.2 = phi i16 [ 0, %entry ], [ %sub, %loop ]
146+
%sub = sub i16 %iv.2, %o.1
147+
%gep.dst = getelementptr inbounds i16, ptr %dst, i64 %iv
148+
store i16 %sub, ptr %gep.dst, align 2
149+
%iv.next = add nuw nsw i64 %iv, 1
150+
%ec = icmp eq i64 %iv.next, %N
151+
br i1 %ec , label %exit, label %loop
152+
153+
exit:
154+
ret i16 %sub
155+
}

0 commit comments

Comments
 (0)