Skip to content

Commit 233c030

Browse files
fhahnfhahn
committed
[LV] Add extra tests for induction cost modeling.
1 parent 0e8b61f commit 233c030

File tree

2 files changed

+530
-0
lines changed

2 files changed

+530
-0
lines changed

‎llvm/test/Transforms/LoopVectorize/AArch64/induction-costs.ll

Lines changed: 299 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,299 @@
1+
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 4
2+
; RUN: opt -p loop-vectorize -mtriple=arm64-apple-macosx -S %s | FileCheck %s
3+
4+
target datalayout = "e-m:o-i64:64-i128:128-n32:64-S128"
5+
6+
define i32 @multi_exit_iv_uniform(i32 %a, i64 %N, ptr %dst) {
7+
; CHECK-LABEL: define i32 @multi_exit_iv_uniform(
8+
; CHECK-SAME: i32 [[A:%.*]], i64 [[N:%.*]], ptr [[DST:%.*]]) {
9+
; CHECK-NEXT: entry:
10+
; CHECK-NEXT: [[UMIN:%.*]] = call i64 @llvm.umin.i64(i64 [[N]], i64 2147483648)
11+
; CHECK-NEXT: [[TMP0:%.*]] = add nuw nsw i64 [[UMIN]], 1
12+
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ule i64 [[TMP0]], 4
13+
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
14+
; CHECK: vector.ph:
15+
; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP0]], 4
16+
; CHECK-NEXT: [[TMP1:%.*]] = icmp eq i64 [[N_MOD_VF]], 0
17+
; CHECK-NEXT: [[TMP2:%.*]] = select i1 [[TMP1]], i64 4, i64 [[N_MOD_VF]]
18+
; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP0]], [[TMP2]]
19+
; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i32> poison, i32 [[A]], i64 0
20+
; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i32> [[BROADCAST_SPLATINSERT]], <2 x i32> poison, <2 x i32> zeroinitializer
21+
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
22+
; CHECK: vector.body:
23+
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
24+
; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP10:%.*]], [[VECTOR_BODY]] ]
25+
; CHECK-NEXT: [[VEC_PHI1:%.*]] = phi <2 x i32> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP11:%.*]], [[VECTOR_BODY]] ]
26+
; CHECK-NEXT: [[TMP3:%.*]] = add i64 [[INDEX]], 0
27+
; CHECK-NEXT: [[TMP4:%.*]] = add i64 [[INDEX]], 2
28+
; CHECK-NEXT: [[TMP5:%.*]] = getelementptr i64, ptr [[DST]], i64 [[TMP3]]
29+
; CHECK-NEXT: [[TMP6:%.*]] = getelementptr i64, ptr [[DST]], i64 [[TMP4]]
30+
; CHECK-NEXT: [[TMP7:%.*]] = zext <2 x i32> [[BROADCAST_SPLAT]] to <2 x i64>
31+
; CHECK-NEXT: [[TMP8:%.*]] = getelementptr i64, ptr [[TMP5]], i32 0
32+
; CHECK-NEXT: [[TMP9:%.*]] = getelementptr i64, ptr [[TMP5]], i32 2
33+
; CHECK-NEXT: store <2 x i64> [[TMP7]], ptr [[TMP8]], align 8
34+
; CHECK-NEXT: store <2 x i64> [[TMP7]], ptr [[TMP9]], align 8
35+
; CHECK-NEXT: [[TMP10]] = add <2 x i32> [[VEC_PHI]], <i32 -1, i32 -1>
36+
; CHECK-NEXT: [[TMP11]] = add <2 x i32> [[VEC_PHI1]], <i32 -1, i32 -1>
37+
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
38+
; CHECK-NEXT: [[TMP12:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
39+
; CHECK-NEXT: br i1 [[TMP12]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]]
40+
; CHECK: middle.block:
41+
; CHECK-NEXT: [[BIN_RDX:%.*]] = add <2 x i32> [[TMP11]], [[TMP10]]
42+
; CHECK-NEXT: [[TMP13:%.*]] = call i32 @llvm.vector.reduce.add.v2i32(<2 x i32> [[BIN_RDX]])
43+
; CHECK-NEXT: br label [[SCALAR_PH]]
44+
; CHECK: scalar.ph:
45+
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
46+
; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi i32 [ 0, [[ENTRY]] ], [ [[TMP13]], [[MIDDLE_BLOCK]] ]
47+
; CHECK-NEXT: br label [[LOOP_HEADER:%.*]]
48+
; CHECK: loop.header:
49+
; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP_LATCH:%.*]] ]
50+
; CHECK-NEXT: [[IV_2:%.*]] = phi i32 [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[IV_2_NEXT:%.*]], [[LOOP_LATCH]] ]
51+
; CHECK-NEXT: [[C_1:%.*]] = icmp eq i64 [[IV]], [[N]]
52+
; CHECK-NEXT: br i1 [[C_1]], label [[EXIT_1:%.*]], label [[LOOP_LATCH]]
53+
; CHECK: loop.latch:
54+
; CHECK-NEXT: [[ARRAYIDX_I:%.*]] = getelementptr i64, ptr [[DST]], i64 [[IV]]
55+
; CHECK-NEXT: [[CONV7:%.*]] = zext i32 [[A]] to i64
56+
; CHECK-NEXT: store i64 [[CONV7]], ptr [[ARRAYIDX_I]], align 8
57+
; CHECK-NEXT: [[IV_NEXT]] = add i64 [[IV]], 1
58+
; CHECK-NEXT: [[IV_2_NEXT]] = add i32 [[IV_2]], -1
59+
; CHECK-NEXT: [[C_2:%.*]] = icmp eq i64 [[IV]], 2147483648
60+
; CHECK-NEXT: br i1 [[C_2]], label [[EXIT_2:%.*]], label [[LOOP_HEADER]], !llvm.loop [[LOOP3:![0-9]+]]
61+
; CHECK: exit.1:
62+
; CHECK-NEXT: ret i32 10
63+
; CHECK: exit.2:
64+
; CHECK-NEXT: [[IV_2_NEXT_LCSSA:%.*]] = phi i32 [ [[IV_2_NEXT]], [[LOOP_LATCH]] ]
65+
; CHECK-NEXT: ret i32 [[IV_2_NEXT_LCSSA]]
66+
;
67+
entry:
68+
br label %loop.header
69+
70+
loop.header:
71+
%iv = phi i64 [ 0, %entry ], [ %iv.next, %loop.latch ]
72+
%iv.2 = phi i32 [ 0, %entry ], [ %iv.2.next, %loop.latch ]
73+
%c.1 = icmp eq i64 %iv, %N
74+
br i1 %c.1, label %exit.1, label %loop.latch
75+
76+
loop.latch:
77+
%arrayidx.i = getelementptr i64, ptr %dst, i64 %iv
78+
%conv7 = zext i32 %a to i64
79+
store i64 %conv7, ptr %arrayidx.i, align 8
80+
%iv.next = add i64 %iv, 1
81+
%iv.2.next = add i32 %iv.2, -1
82+
%c.2 = icmp eq i64 %iv, 2147483648
83+
br i1 %c.2, label %exit.2, label %loop.header
84+
85+
exit.1:
86+
ret i32 10
87+
88+
exit.2:
89+
ret i32 %iv.2.next
90+
}
91+
92+
define i64 @pointer_induction_only(ptr %start, ptr %end) {
93+
; CHECK-LABEL: define i64 @pointer_induction_only(
94+
; CHECK-SAME: ptr [[START:%.*]], ptr [[END:%.*]]) {
95+
; CHECK-NEXT: entry:
96+
; CHECK-NEXT: [[START2:%.*]] = ptrtoint ptr [[START]] to i64
97+
; CHECK-NEXT: [[END1:%.*]] = ptrtoint ptr [[END]] to i64
98+
; CHECK-NEXT: [[TMP0:%.*]] = sub i64 [[END1]], [[START2]]
99+
; CHECK-NEXT: [[TMP1:%.*]] = lshr i64 [[TMP0]], 2
100+
; CHECK-NEXT: [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1
101+
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[TMP2]], 4
102+
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
103+
; CHECK: vector.ph:
104+
; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[TMP2]], 4
105+
; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[TMP2]], [[N_MOD_VF]]
106+
; CHECK-NEXT: [[TMP3:%.*]] = mul i64 [[N_VEC]], 4
107+
; CHECK-NEXT: [[IND_END:%.*]] = getelementptr i8, ptr [[START]], i64 [[TMP3]]
108+
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
109+
; CHECK: vector.body:
110+
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
111+
; CHECK-NEXT: [[VECTOR_RECUR:%.*]] = phi <2 x i64> [ <i64 poison, i64 0>, [[VECTOR_PH]] ], [ [[TMP9:%.*]], [[VECTOR_BODY]] ]
112+
; CHECK-NEXT: [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 4
113+
; CHECK-NEXT: [[TMP4:%.*]] = add i64 [[OFFSET_IDX]], 0
114+
; CHECK-NEXT: [[TMP5:%.*]] = add i64 [[OFFSET_IDX]], 8
115+
; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[START]], i64 [[TMP4]]
116+
; CHECK-NEXT: [[NEXT_GEP3:%.*]] = getelementptr i8, ptr [[START]], i64 [[TMP5]]
117+
; CHECK-NEXT: [[TMP6:%.*]] = getelementptr i32, ptr [[NEXT_GEP]], i32 0
118+
; CHECK-NEXT: [[TMP7:%.*]] = getelementptr i32, ptr [[NEXT_GEP]], i32 2
119+
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x i32>, ptr [[TMP6]], align 1
120+
; CHECK-NEXT: [[WIDE_LOAD4:%.*]] = load <2 x i32>, ptr [[TMP7]], align 1
121+
; CHECK-NEXT: [[TMP8:%.*]] = zext <2 x i32> [[WIDE_LOAD]] to <2 x i64>
122+
; CHECK-NEXT: [[TMP9]] = zext <2 x i32> [[WIDE_LOAD4]] to <2 x i64>
123+
; CHECK-NEXT: [[TMP10:%.*]] = shufflevector <2 x i64> [[VECTOR_RECUR]], <2 x i64> [[TMP8]], <2 x i32> <i32 1, i32 2>
124+
; CHECK-NEXT: [[TMP11:%.*]] = shufflevector <2 x i64> [[TMP8]], <2 x i64> [[TMP9]], <2 x i32> <i32 1, i32 2>
125+
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4
126+
; CHECK-NEXT: [[TMP12:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
127+
; CHECK-NEXT: br i1 [[TMP12]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]]
128+
; CHECK: middle.block:
129+
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]]
130+
; CHECK-NEXT: [[VECTOR_RECUR_EXTRACT:%.*]] = extractelement <2 x i64> [[TMP9]], i32 1
131+
; CHECK-NEXT: [[VECTOR_RECUR_EXTRACT_FOR_PHI:%.*]] = extractelement <2 x i64> [[TMP9]], i32 0
132+
; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]]
133+
; CHECK: scalar.ph:
134+
; CHECK-NEXT: [[SCALAR_RECUR_INIT:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[VECTOR_RECUR_EXTRACT]], [[MIDDLE_BLOCK]] ]
135+
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi ptr [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[START]], [[ENTRY]] ]
136+
; CHECK-NEXT: br label [[LOOP:%.*]]
137+
; CHECK: loop:
138+
; CHECK-NEXT: [[IV:%.*]] = phi ptr [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
139+
; CHECK-NEXT: [[SCALAR_RECUR:%.*]] = phi i64 [ [[SCALAR_RECUR_INIT]], [[SCALAR_PH]] ], [ [[RECUR_NEXT:%.*]], [[LOOP]] ]
140+
; CHECK-NEXT: [[L:%.*]] = load i32, ptr [[IV]], align 1
141+
; CHECK-NEXT: [[RECUR_NEXT]] = zext i32 [[L]] to i64
142+
; CHECK-NEXT: [[IV_NEXT]] = getelementptr inbounds i8, ptr [[IV]], i64 4
143+
; CHECK-NEXT: [[C:%.*]] = icmp eq ptr [[IV]], [[END]]
144+
; CHECK-NEXT: br i1 [[C]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP5:![0-9]+]]
145+
; CHECK: exit:
146+
; CHECK-NEXT: [[RECUR_LCSSA:%.*]] = phi i64 [ [[SCALAR_RECUR]], [[LOOP]] ], [ [[VECTOR_RECUR_EXTRACT_FOR_PHI]], [[MIDDLE_BLOCK]] ]
147+
; CHECK-NEXT: ret i64 [[RECUR_LCSSA]]
148+
;
149+
entry:
150+
br label %loop
151+
152+
loop:
153+
%iv = phi ptr [ %start, %entry ], [ %iv.next, %loop ]
154+
%recur = phi i64 [ 0, %entry ], [ %recur.next, %loop ]
155+
%l = load i32, ptr %iv, align 1
156+
%recur.next = zext i32 %l to i64
157+
%iv.next = getelementptr inbounds i8, ptr %iv, i64 4
158+
%c = icmp eq ptr %iv, %end
159+
br i1 %c, label %exit, label %loop
160+
161+
exit:
162+
ret i64 %recur
163+
}
164+
165+
166+
define i64 @int_and_pointer_iv(ptr %start, i32 %N) {
167+
; CHECK-LABEL: define i64 @int_and_pointer_iv(
168+
; CHECK-SAME: ptr [[START:%.*]], i32 [[N:%.*]]) {
169+
; CHECK-NEXT: entry:
170+
; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
171+
; CHECK: vector.ph:
172+
; CHECK-NEXT: [[IND_END:%.*]] = getelementptr i8, ptr [[START]], i64 4000
173+
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
174+
; CHECK: vector.body:
175+
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
176+
; CHECK-NEXT: [[VECTOR_RECUR:%.*]] = phi <4 x i64> [ <i64 poison, i64 poison, i64 poison, i64 0>, [[VECTOR_PH]] ], [ [[TMP5:%.*]], [[VECTOR_BODY]] ]
177+
; CHECK-NEXT: [[OFFSET_IDX:%.*]] = mul i64 [[INDEX]], 4
178+
; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[OFFSET_IDX]], 0
179+
; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[OFFSET_IDX]], 16
180+
; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, ptr [[START]], i64 [[TMP0]]
181+
; CHECK-NEXT: [[NEXT_GEP2:%.*]] = getelementptr i8, ptr [[START]], i64 [[TMP1]]
182+
; CHECK-NEXT: [[TMP2:%.*]] = getelementptr i32, ptr [[NEXT_GEP]], i32 0
183+
; CHECK-NEXT: [[TMP3:%.*]] = getelementptr i32, ptr [[NEXT_GEP]], i32 4
184+
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x i32>, ptr [[TMP2]], align 4
185+
; CHECK-NEXT: [[WIDE_LOAD3:%.*]] = load <4 x i32>, ptr [[TMP3]], align 4
186+
; CHECK-NEXT: [[TMP4:%.*]] = zext <4 x i32> [[WIDE_LOAD]] to <4 x i64>
187+
; CHECK-NEXT: [[TMP5]] = zext <4 x i32> [[WIDE_LOAD3]] to <4 x i64>
188+
; CHECK-NEXT: [[TMP6:%.*]] = shufflevector <4 x i64> [[VECTOR_RECUR]], <4 x i64> [[TMP4]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
189+
; CHECK-NEXT: [[TMP7:%.*]] = shufflevector <4 x i64> [[TMP4]], <4 x i64> [[TMP5]], <4 x i32> <i32 3, i32 4, i32 5, i32 6>
190+
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8
191+
; CHECK-NEXT: [[TMP8:%.*]] = icmp eq i64 [[INDEX_NEXT]], 1000
192+
; CHECK-NEXT: br i1 [[TMP8]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP6:![0-9]+]]
193+
; CHECK: middle.block:
194+
; CHECK-NEXT: [[VECTOR_RECUR_EXTRACT:%.*]] = extractelement <4 x i64> [[TMP5]], i32 3
195+
; CHECK-NEXT: [[VECTOR_RECUR_EXTRACT_FOR_PHI:%.*]] = extractelement <4 x i64> [[TMP5]], i32 2
196+
; CHECK-NEXT: br i1 true, label [[EXIT:%.*]], label [[SCALAR_PH]]
197+
; CHECK: scalar.ph:
198+
; CHECK-NEXT: [[SCALAR_RECUR_INIT:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[VECTOR_RECUR_EXTRACT]], [[MIDDLE_BLOCK]] ]
199+
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ 1000, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY]] ]
200+
; CHECK-NEXT: [[BC_RESUME_VAL1:%.*]] = phi ptr [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[START]], [[ENTRY]] ]
201+
; CHECK-NEXT: br label [[LOOP:%.*]]
202+
; CHECK: loop:
203+
; CHECK-NEXT: [[IV:%.*]] = phi i32 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
204+
; CHECK-NEXT: [[PTR_IV:%.*]] = phi ptr [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ], [ [[PTR_IV_NEXT:%.*]], [[LOOP]] ]
205+
; CHECK-NEXT: [[SCALAR_RECUR:%.*]] = phi i64 [ [[SCALAR_RECUR_INIT]], [[SCALAR_PH]] ], [ [[RECUR_NEXT:%.*]], [[LOOP]] ]
206+
; CHECK-NEXT: [[L:%.*]] = load i32, ptr [[PTR_IV]], align 4
207+
; CHECK-NEXT: [[RECUR_NEXT]] = zext i32 [[L]] to i64
208+
; CHECK-NEXT: [[PTR_IV_NEXT]] = getelementptr i8, ptr [[PTR_IV]], i64 4
209+
; CHECK-NEXT: [[IV_NEXT]] = add i32 [[IV]], 1
210+
; CHECK-NEXT: [[TOBOOL_NOT:%.*]] = icmp eq i32 [[IV_NEXT]], 1000
211+
; CHECK-NEXT: br i1 [[TOBOOL_NOT]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP7:![0-9]+]]
212+
; CHECK: exit:
213+
; CHECK-NEXT: [[RECUR_LCSSA:%.*]] = phi i64 [ [[SCALAR_RECUR]], [[LOOP]] ], [ [[VECTOR_RECUR_EXTRACT_FOR_PHI]], [[MIDDLE_BLOCK]] ]
214+
; CHECK-NEXT: ret i64 [[RECUR_LCSSA]]
215+
;
216+
entry:
217+
br label %loop
218+
219+
loop:
220+
%iv = phi i32 [ 0, %entry ], [ %iv.next, %loop ]
221+
%ptr.iv = phi ptr [ %start, %entry ], [ %ptr.iv.next, %loop ]
222+
%recur = phi i64 [ 0, %entry ], [ %recur.next, %loop ]
223+
%l = load i32, ptr %ptr.iv, align 4
224+
%recur.next = zext i32 %l to i64
225+
%ptr.iv.next = getelementptr i8, ptr %ptr.iv, i64 4
226+
%iv.next = add i32 %iv, 1
227+
%tobool.not = icmp eq i32 %iv.next, 1000
228+
br i1 %tobool.not, label %exit, label %loop
229+
230+
exit:
231+
ret i64 %recur
232+
}
233+
234+
define void @wide_truncated_iv(ptr %dst) {
235+
; CHECK-LABEL: define void @wide_truncated_iv(
236+
; CHECK-SAME: ptr [[DST:%.*]]) {
237+
; CHECK-NEXT: entry:
238+
; CHECK-NEXT: br i1 false, label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
239+
; CHECK: vector.ph:
240+
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
241+
; CHECK: vector.body:
242+
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
243+
; CHECK-NEXT: [[VEC_IND:%.*]] = phi <8 x i8> [ <i8 0, i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
244+
; CHECK-NEXT: [[STEP_ADD:%.*]] = add <8 x i8> [[VEC_IND]], <i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8>
245+
; CHECK-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0
246+
; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[INDEX]], 8
247+
; CHECK-NEXT: [[TMP2:%.*]] = getelementptr i8, ptr [[DST]], i64 [[TMP0]]
248+
; CHECK-NEXT: [[TMP3:%.*]] = getelementptr i8, ptr [[DST]], i64 [[TMP1]]
249+
; CHECK-NEXT: [[TMP4:%.*]] = getelementptr i8, ptr [[TMP2]], i32 0
250+
; CHECK-NEXT: [[TMP5:%.*]] = getelementptr i8, ptr [[TMP2]], i32 8
251+
; CHECK-NEXT: store <8 x i8> [[VEC_IND]], ptr [[TMP4]], align 1
252+
; CHECK-NEXT: store <8 x i8> [[STEP_ADD]], ptr [[TMP5]], align 1
253+
; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 16
254+
; CHECK-NEXT: [[VEC_IND_NEXT]] = add <8 x i8> [[STEP_ADD]], <i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8, i8 8>
255+
; CHECK-NEXT: [[TMP6:%.*]] = icmp eq i64 [[INDEX_NEXT]], 192
256+
; CHECK-NEXT: br i1 [[TMP6]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP8:![0-9]+]]
257+
; CHECK: middle.block:
258+
; CHECK-NEXT: br i1 false, label [[EXIT:%.*]], label [[SCALAR_PH]]
259+
; CHECK: scalar.ph:
260+
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ 192, [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ]
261+
; CHECK-NEXT: br label [[LOOP:%.*]]
262+
; CHECK: loop:
263+
; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[LOOP]] ]
264+
; CHECK-NEXT: [[TRUNC_IV:%.*]] = trunc i64 [[IV]] to i8
265+
; CHECK-NEXT: [[GEP:%.*]] = getelementptr i8, ptr [[DST]], i64 [[IV]]
266+
; CHECK-NEXT: store i8 [[TRUNC_IV]], ptr [[GEP]], align 1
267+
; CHECK-NEXT: [[IV_NEXT]] = add i64 [[IV]], 1
268+
; CHECK-NEXT: [[C:%.*]] = icmp eq i64 [[IV]], 200
269+
; CHECK-NEXT: br i1 [[C]], label [[EXIT]], label [[LOOP]], !llvm.loop [[LOOP9:![0-9]+]]
270+
; CHECK: exit:
271+
; CHECK-NEXT: ret void
272+
;
273+
entry:
274+
br label %loop
275+
276+
loop:
277+
%iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ]
278+
%trunc.iv = trunc i64 %iv to i8
279+
%gep = getelementptr i8, ptr %dst, i64 %iv
280+
store i8 %trunc.iv, ptr %gep, align 1
281+
%iv.next = add i64 %iv, 1
282+
%c = icmp eq i64 %iv, 200
283+
br i1 %c, label %exit, label %loop
284+
285+
exit:
286+
ret void
287+
}
288+
;.
289+
; CHECK: [[LOOP0]] = distinct !{[[LOOP0]], [[META1:![0-9]+]], [[META2:![0-9]+]]}
290+
; CHECK: [[META1]] = !{!"llvm.loop.isvectorized", i32 1}
291+
; CHECK: [[META2]] = !{!"llvm.loop.unroll.runtime.disable"}
292+
; CHECK: [[LOOP3]] = distinct !{[[LOOP3]], [[META2]], [[META1]]}
293+
; CHECK: [[LOOP4]] = distinct !{[[LOOP4]], [[META1]], [[META2]]}
294+
; CHECK: [[LOOP5]] = distinct !{[[LOOP5]], [[META2]], [[META1]]}
295+
; CHECK: [[LOOP6]] = distinct !{[[LOOP6]], [[META1]], [[META2]]}
296+
; CHECK: [[LOOP7]] = distinct !{[[LOOP7]], [[META2]], [[META1]]}
297+
; CHECK: [[LOOP8]] = distinct !{[[LOOP8]], [[META1]], [[META2]]}
298+
; CHECK: [[LOOP9]] = distinct !{[[LOOP9]], [[META2]], [[META1]]}
299+
;.

0 commit comments

Comments
 (0)