Skip to content

Commit 4cc94b7

Browse files
davemgreendavemgreen
committed
[CostModel] Tests for showing the cost of intrinsics from the vectorizer. NFC
1 parent f959d81 commit 4cc94b7

File tree

3 files changed

+521
-0
lines changed

3 files changed

+521
-0
lines changed

‎llvm/test/Transforms/LoopVectorize/AArch64/intrinsiccost.ll

Lines changed: 211 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,211 @@
1+
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2+
; RUN: opt -loop-vectorize -instcombine -simplifycfg < %s -S -o - | FileCheck %s --check-prefix=CHECK
3+
; RUN: opt -loop-vectorize -debug-only=loop-vectorize -disable-output < %s 2>&1 | FileCheck %s --check-prefix=CHECK-COST
4+
; REQUIRES: asserts
5+
6+
target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128"
7+
target triple = "aarch64--linux-gnu"
8+
9+
; CHECK-COST-LABEL: sadd
10+
; CHECK-COST: Found an estimated cost of 10 for VF 1 For instruction: %1 = tail call i16 @llvm.sadd.sat.i16(i16 %0, i16 %offset)
11+
; CHECK-COST: Found an estimated cost of 26 for VF 2 For instruction: %1 = tail call i16 @llvm.sadd.sat.i16(i16 %0, i16 %offset)
12+
; CHECK-COST: Found an estimated cost of 58 for VF 4 For instruction: %1 = tail call i16 @llvm.sadd.sat.i16(i16 %0, i16 %offset)
13+
; CHECK-COST: Found an estimated cost of 122 for VF 8 For instruction: %1 = tail call i16 @llvm.sadd.sat.i16(i16 %0, i16 %offset)
14+
15+
define void @saddsat(i16* nocapture readonly %pSrc, i16 signext %offset, i16* nocapture noalias %pDst, i32 %blockSize) #0 {
16+
; CHECK-LABEL: @saddsat(
17+
; CHECK-NEXT: entry:
18+
; CHECK-NEXT: [[CMP_NOT6:%.*]] = icmp eq i32 [[BLOCKSIZE:%.*]], 0
19+
; CHECK-NEXT: br i1 [[CMP_NOT6]], label [[WHILE_END:%.*]], label [[WHILE_BODY_PREHEADER:%.*]]
20+
; CHECK: while.body.preheader:
21+
; CHECK-NEXT: [[TMP0:%.*]] = add i32 [[BLOCKSIZE]], -1
22+
; CHECK-NEXT: [[TMP1:%.*]] = zext i32 [[TMP0]] to i64
23+
; CHECK-NEXT: [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1
24+
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp eq i32 [[TMP0]], 0
25+
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]]
26+
; CHECK: vector.ph:
27+
; CHECK-NEXT: [[N_VEC:%.*]] = and i64 [[TMP2]], 8589934590
28+
; CHECK-NEXT: [[CAST_CRD:%.*]] = trunc i64 [[N_VEC]] to i32
29+
; CHECK-NEXT: [[IND_END:%.*]] = sub i32 [[BLOCKSIZE]], [[CAST_CRD]]
30+
; CHECK-NEXT: [[IND_END2:%.*]] = getelementptr i16, i16* [[PSRC:%.*]], i64 [[N_VEC]]
31+
; CHECK-NEXT: [[IND_END4:%.*]] = getelementptr i16, i16* [[PDST:%.*]], i64 [[N_VEC]]
32+
; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <2 x i16> poison, i16 [[OFFSET:%.*]], i32 0
33+
; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <2 x i16> [[BROADCAST_SPLATINSERT]], <2 x i16> poison, <2 x i32> zeroinitializer
34+
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
35+
; CHECK: vector.body:
36+
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
37+
; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr i16, i16* [[PSRC]], i64 [[INDEX]]
38+
; CHECK-NEXT: [[NEXT_GEP5:%.*]] = getelementptr i16, i16* [[PDST]], i64 [[INDEX]]
39+
; CHECK-NEXT: [[TMP3:%.*]] = bitcast i16* [[NEXT_GEP]] to <2 x i16>*
40+
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <2 x i16>, <2 x i16>* [[TMP3]], align 2
41+
; CHECK-NEXT: [[TMP4:%.*]] = call <2 x i16> @llvm.sadd.sat.v2i16(<2 x i16> [[WIDE_LOAD]], <2 x i16> [[BROADCAST_SPLAT]])
42+
; CHECK-NEXT: [[TMP5:%.*]] = bitcast i16* [[NEXT_GEP5]] to <2 x i16>*
43+
; CHECK-NEXT: store <2 x i16> [[TMP4]], <2 x i16>* [[TMP5]], align 2
44+
; CHECK-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 2
45+
; CHECK-NEXT: [[TMP6:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
46+
; CHECK-NEXT: br i1 [[TMP6]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], [[LOOP0:!llvm.loop !.*]]
47+
; CHECK: middle.block:
48+
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]]
49+
; CHECK-NEXT: br i1 [[CMP_N]], label [[WHILE_END]], label [[SCALAR_PH]]
50+
; CHECK: scalar.ph:
51+
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[BLOCKSIZE]], [[WHILE_BODY_PREHEADER]] ]
52+
; CHECK-NEXT: [[BC_RESUME_VAL1:%.*]] = phi i16* [ [[IND_END2]], [[MIDDLE_BLOCK]] ], [ [[PSRC]], [[WHILE_BODY_PREHEADER]] ]
53+
; CHECK-NEXT: [[BC_RESUME_VAL3:%.*]] = phi i16* [ [[IND_END4]], [[MIDDLE_BLOCK]] ], [ [[PDST]], [[WHILE_BODY_PREHEADER]] ]
54+
; CHECK-NEXT: br label [[WHILE_BODY:%.*]]
55+
; CHECK: while.body:
56+
; CHECK-NEXT: [[BLKCNT_09:%.*]] = phi i32 [ [[DEC:%.*]], [[WHILE_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ]
57+
; CHECK-NEXT: [[PSRC_ADDR_08:%.*]] = phi i16* [ [[INCDEC_PTR:%.*]], [[WHILE_BODY]] ], [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ]
58+
; CHECK-NEXT: [[PDST_ADDR_07:%.*]] = phi i16* [ [[INCDEC_PTR3:%.*]], [[WHILE_BODY]] ], [ [[BC_RESUME_VAL3]], [[SCALAR_PH]] ]
59+
; CHECK-NEXT: [[INCDEC_PTR]] = getelementptr inbounds i16, i16* [[PSRC_ADDR_08]], i64 1
60+
; CHECK-NEXT: [[TMP7:%.*]] = load i16, i16* [[PSRC_ADDR_08]], align 2
61+
; CHECK-NEXT: [[TMP8:%.*]] = tail call i16 @llvm.sadd.sat.i16(i16 [[TMP7]], i16 [[OFFSET]])
62+
; CHECK-NEXT: [[INCDEC_PTR3]] = getelementptr inbounds i16, i16* [[PDST_ADDR_07]], i64 1
63+
; CHECK-NEXT: store i16 [[TMP8]], i16* [[PDST_ADDR_07]], align 2
64+
; CHECK-NEXT: [[DEC]] = add i32 [[BLKCNT_09]], -1
65+
; CHECK-NEXT: [[CMP_NOT:%.*]] = icmp eq i32 [[DEC]], 0
66+
; CHECK-NEXT: br i1 [[CMP_NOT]], label [[WHILE_END]], label [[WHILE_BODY]], [[LOOP2:!llvm.loop !.*]]
67+
; CHECK: while.end:
68+
; CHECK-NEXT: ret void
69+
;
70+
entry:
71+
%cmp.not6 = icmp eq i32 %blockSize, 0
72+
br i1 %cmp.not6, label %while.end, label %while.body
73+
74+
while.body: ; preds = %entry, %while.body
75+
%blkCnt.09 = phi i32 [ %dec, %while.body ], [ %blockSize, %entry ]
76+
%pSrc.addr.08 = phi i16* [ %incdec.ptr, %while.body ], [ %pSrc, %entry ]
77+
%pDst.addr.07 = phi i16* [ %incdec.ptr3, %while.body ], [ %pDst, %entry ]
78+
%incdec.ptr = getelementptr inbounds i16, i16* %pSrc.addr.08, i32 1
79+
%0 = load i16, i16* %pSrc.addr.08, align 2
80+
%1 = tail call i16 @llvm.sadd.sat.i16(i16 %0, i16 %offset)
81+
%incdec.ptr3 = getelementptr inbounds i16, i16* %pDst.addr.07, i32 1
82+
store i16 %1, i16* %pDst.addr.07, align 2
83+
%dec = add i32 %blkCnt.09, -1
84+
%cmp.not = icmp eq i32 %dec, 0
85+
br i1 %cmp.not, label %while.end, label %while.body
86+
87+
while.end: ; preds = %while.body, %entry
88+
ret void
89+
}
90+
91+
; CHECK-COST-LABEL: umin
92+
; CHECK-COST: Found an estimated cost of 2 for VF 1 For instruction: %1 = tail call i8 @llvm.umin.i8(i8 %0, i8 %offset)
93+
; CHECK-COST: Found an estimated cost of 6 for VF 2 For instruction: %1 = tail call i8 @llvm.umin.i8(i8 %0, i8 %offset)
94+
; CHECK-COST: Found an estimated cost of 14 for VF 4 For instruction: %1 = tail call i8 @llvm.umin.i8(i8 %0, i8 %offset)
95+
; CHECK-COST: Found an estimated cost of 30 for VF 8 For instruction: %1 = tail call i8 @llvm.umin.i8(i8 %0, i8 %offset)
96+
; CHECK-COST: Found an estimated cost of 62 for VF 16 For instruction: %1 = tail call i8 @llvm.umin.i8(i8 %0, i8 %offset)
97+
98+
define void @umin(i8* nocapture readonly %pSrc, i8 signext %offset, i8* nocapture noalias %pDst, i32 %blockSize) #0 {
99+
; CHECK-LABEL: @umin(
100+
; CHECK-NEXT: entry:
101+
; CHECK-NEXT: [[CMP_NOT6:%.*]] = icmp eq i32 [[BLOCKSIZE:%.*]], 0
102+
; CHECK-NEXT: br i1 [[CMP_NOT6]], label [[WHILE_END:%.*]], label [[ITER_CHECK:%.*]]
103+
; CHECK: iter.check:
104+
; CHECK-NEXT: [[TMP0:%.*]] = add i32 [[BLOCKSIZE]], -1
105+
; CHECK-NEXT: [[TMP1:%.*]] = zext i32 [[TMP0]] to i64
106+
; CHECK-NEXT: [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1
107+
; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP0]], 7
108+
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH:%.*]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]]
109+
; CHECK: vector.main.loop.iter.check:
110+
; CHECK-NEXT: [[MIN_ITERS_CHECK1:%.*]] = icmp ult i32 [[TMP0]], 15
111+
; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK1]], label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]]
112+
; CHECK: vector.ph:
113+
; CHECK-NEXT: [[N_VEC:%.*]] = and i64 [[TMP2]], 8589934576
114+
; CHECK-NEXT: [[BROADCAST_SPLATINSERT:%.*]] = insertelement <16 x i8> poison, i8 [[OFFSET:%.*]], i32 0
115+
; CHECK-NEXT: [[BROADCAST_SPLAT:%.*]] = shufflevector <16 x i8> [[BROADCAST_SPLATINSERT]], <16 x i8> poison, <16 x i32> zeroinitializer
116+
; CHECK-NEXT: br label [[VECTOR_BODY:%.*]]
117+
; CHECK: vector.body:
118+
; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
119+
; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr i8, i8* [[PSRC:%.*]], i64 [[INDEX]]
120+
; CHECK-NEXT: [[NEXT_GEP2:%.*]] = getelementptr i8, i8* [[PDST:%.*]], i64 [[INDEX]]
121+
; CHECK-NEXT: [[TMP3:%.*]] = bitcast i8* [[NEXT_GEP]] to <16 x i8>*
122+
; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <16 x i8>, <16 x i8>* [[TMP3]], align 2
123+
; CHECK-NEXT: [[TMP4:%.*]] = call <16 x i8> @llvm.umin.v16i8(<16 x i8> [[WIDE_LOAD]], <16 x i8> [[BROADCAST_SPLAT]])
124+
; CHECK-NEXT: [[TMP5:%.*]] = bitcast i8* [[NEXT_GEP2]] to <16 x i8>*
125+
; CHECK-NEXT: store <16 x i8> [[TMP4]], <16 x i8>* [[TMP5]], align 2
126+
; CHECK-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 16
127+
; CHECK-NEXT: [[TMP6:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
128+
; CHECK-NEXT: br i1 [[TMP6]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], [[LOOP4:!llvm.loop !.*]]
129+
; CHECK: middle.block:
130+
; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]]
131+
; CHECK-NEXT: br i1 [[CMP_N]], label [[WHILE_END]], label [[VEC_EPILOG_ITER_CHECK:%.*]]
132+
; CHECK: vec.epilog.iter.check:
133+
; CHECK-NEXT: [[IND_END14:%.*]] = getelementptr i8, i8* [[PDST]], i64 [[N_VEC]]
134+
; CHECK-NEXT: [[IND_END11:%.*]] = getelementptr i8, i8* [[PSRC]], i64 [[N_VEC]]
135+
; CHECK-NEXT: [[CAST_CRD7:%.*]] = trunc i64 [[N_VEC]] to i32
136+
; CHECK-NEXT: [[IND_END8:%.*]] = sub i32 [[BLOCKSIZE]], [[CAST_CRD7]]
137+
; CHECK-NEXT: [[N_VEC_REMAINING:%.*]] = and i64 [[TMP2]], 8
138+
; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK_NOT_NOT:%.*]] = icmp eq i64 [[N_VEC_REMAINING]], 0
139+
; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK_NOT_NOT]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]]
140+
; CHECK: vec.epilog.ph:
141+
; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ]
142+
; CHECK-NEXT: [[TMP7:%.*]] = add i32 [[BLOCKSIZE]], -1
143+
; CHECK-NEXT: [[TMP8:%.*]] = zext i32 [[TMP7]] to i64
144+
; CHECK-NEXT: [[TMP9:%.*]] = add nuw nsw i64 [[TMP8]], 1
145+
; CHECK-NEXT: [[N_VEC4:%.*]] = and i64 [[TMP9]], 8589934584
146+
; CHECK-NEXT: [[CAST_CRD:%.*]] = trunc i64 [[N_VEC4]] to i32
147+
; CHECK-NEXT: [[IND_END:%.*]] = sub i32 [[BLOCKSIZE]], [[CAST_CRD]]
148+
; CHECK-NEXT: [[IND_END10:%.*]] = getelementptr i8, i8* [[PSRC]], i64 [[N_VEC4]]
149+
; CHECK-NEXT: [[IND_END13:%.*]] = getelementptr i8, i8* [[PDST]], i64 [[N_VEC4]]
150+
; CHECK-NEXT: [[BROADCAST_SPLATINSERT20:%.*]] = insertelement <8 x i8> poison, i8 [[OFFSET]], i32 0
151+
; CHECK-NEXT: [[BROADCAST_SPLAT21:%.*]] = shufflevector <8 x i8> [[BROADCAST_SPLATINSERT20]], <8 x i8> poison, <8 x i32> zeroinitializer
152+
; CHECK-NEXT: br label [[VEC_EPILOG_VECTOR_BODY:%.*]]
153+
; CHECK: vec.epilog.vector.body:
154+
; CHECK-NEXT: [[INDEX5:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT6:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ]
155+
; CHECK-NEXT: [[NEXT_GEP17:%.*]] = getelementptr i8, i8* [[PSRC]], i64 [[INDEX5]]
156+
; CHECK-NEXT: [[NEXT_GEP18:%.*]] = getelementptr i8, i8* [[PDST]], i64 [[INDEX5]]
157+
; CHECK-NEXT: [[TMP10:%.*]] = bitcast i8* [[NEXT_GEP17]] to <8 x i8>*
158+
; CHECK-NEXT: [[WIDE_LOAD19:%.*]] = load <8 x i8>, <8 x i8>* [[TMP10]], align 2
159+
; CHECK-NEXT: [[TMP11:%.*]] = call <8 x i8> @llvm.umin.v8i8(<8 x i8> [[WIDE_LOAD19]], <8 x i8> [[BROADCAST_SPLAT21]])
160+
; CHECK-NEXT: [[TMP12:%.*]] = bitcast i8* [[NEXT_GEP18]] to <8 x i8>*
161+
; CHECK-NEXT: store <8 x i8> [[TMP11]], <8 x i8>* [[TMP12]], align 2
162+
; CHECK-NEXT: [[INDEX_NEXT6]] = add i64 [[INDEX5]], 8
163+
; CHECK-NEXT: [[TMP13:%.*]] = icmp eq i64 [[INDEX_NEXT6]], [[N_VEC4]]
164+
; CHECK-NEXT: br i1 [[TMP13]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], [[LOOP5:!llvm.loop !.*]]
165+
; CHECK: vec.epilog.middle.block:
166+
; CHECK-NEXT: [[CMP_N15:%.*]] = icmp eq i64 [[TMP9]], [[N_VEC4]]
167+
; CHECK-NEXT: br i1 [[CMP_N15]], label [[WHILE_END]], label [[VEC_EPILOG_SCALAR_PH]]
168+
; CHECK: vec.epilog.scalar.ph:
169+
; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i32 [ [[IND_END]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[IND_END8]], [[VEC_EPILOG_ITER_CHECK]] ], [ [[BLOCKSIZE]], [[ITER_CHECK]] ]
170+
; CHECK-NEXT: [[BC_RESUME_VAL9:%.*]] = phi i8* [ [[IND_END10]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[IND_END11]], [[VEC_EPILOG_ITER_CHECK]] ], [ [[PSRC]], [[ITER_CHECK]] ]
171+
; CHECK-NEXT: [[BC_RESUME_VAL12:%.*]] = phi i8* [ [[IND_END13]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ [[IND_END14]], [[VEC_EPILOG_ITER_CHECK]] ], [ [[PDST]], [[ITER_CHECK]] ]
172+
; CHECK-NEXT: br label [[WHILE_BODY:%.*]]
173+
; CHECK: while.body:
174+
; CHECK-NEXT: [[BLKCNT_09:%.*]] = phi i32 [ [[DEC:%.*]], [[WHILE_BODY]] ], [ [[BC_RESUME_VAL]], [[VEC_EPILOG_SCALAR_PH]] ]
175+
; CHECK-NEXT: [[PSRC_ADDR_08:%.*]] = phi i8* [ [[INCDEC_PTR:%.*]], [[WHILE_BODY]] ], [ [[BC_RESUME_VAL9]], [[VEC_EPILOG_SCALAR_PH]] ]
176+
; CHECK-NEXT: [[PDST_ADDR_07:%.*]] = phi i8* [ [[INCDEC_PTR3:%.*]], [[WHILE_BODY]] ], [ [[BC_RESUME_VAL12]], [[VEC_EPILOG_SCALAR_PH]] ]
177+
; CHECK-NEXT: [[INCDEC_PTR]] = getelementptr inbounds i8, i8* [[PSRC_ADDR_08]], i64 1
178+
; CHECK-NEXT: [[TMP14:%.*]] = load i8, i8* [[PSRC_ADDR_08]], align 2
179+
; CHECK-NEXT: [[TMP15:%.*]] = tail call i8 @llvm.umin.i8(i8 [[TMP14]], i8 [[OFFSET]])
180+
; CHECK-NEXT: [[INCDEC_PTR3]] = getelementptr inbounds i8, i8* [[PDST_ADDR_07]], i64 1
181+
; CHECK-NEXT: store i8 [[TMP15]], i8* [[PDST_ADDR_07]], align 2
182+
; CHECK-NEXT: [[DEC]] = add i32 [[BLKCNT_09]], -1
183+
; CHECK-NEXT: [[CMP_NOT:%.*]] = icmp eq i32 [[DEC]], 0
184+
; CHECK-NEXT: br i1 [[CMP_NOT]], label [[WHILE_END]], label [[WHILE_BODY]], [[LOOP6:!llvm.loop !.*]]
185+
; CHECK: while.end:
186+
; CHECK-NEXT: ret void
187+
;
188+
entry:
189+
%cmp.not6 = icmp eq i32 %blockSize, 0
190+
br i1 %cmp.not6, label %while.end, label %while.body
191+
192+
while.body: ; preds = %entry, %while.body
193+
%blkCnt.09 = phi i32 [ %dec, %while.body ], [ %blockSize, %entry ]
194+
%pSrc.addr.08 = phi i8* [ %incdec.ptr, %while.body ], [ %pSrc, %entry ]
195+
%pDst.addr.07 = phi i8* [ %incdec.ptr3, %while.body ], [ %pDst, %entry ]
196+
%incdec.ptr = getelementptr inbounds i8, i8* %pSrc.addr.08, i32 1
197+
%0 = load i8, i8* %pSrc.addr.08, align 2
198+
%1 = tail call i8 @llvm.umin.i8(i8 %0, i8 %offset)
199+
%incdec.ptr3 = getelementptr inbounds i8, i8* %pDst.addr.07, i32 1
200+
store i8 %1, i8* %pDst.addr.07, align 2
201+
%dec = add i32 %blkCnt.09, -1
202+
%cmp.not = icmp eq i32 %dec, 0
203+
br i1 %cmp.not, label %while.end, label %while.body
204+
205+
while.end: ; preds = %while.body, %entry
206+
ret void
207+
}
208+
209+
declare i16 @llvm.sadd.sat.i16(i16, i16)
210+
declare i8 @llvm.umin.i8(i8, i8)
211+

‎llvm/test/Transforms/LoopVectorize/ARM/mve-saddsatcost.ll

Lines changed: 8 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1,9 +1,17 @@
11
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
22
; RUN: opt -loop-vectorize -instcombine -simplifycfg < %s -S -o - | FileCheck %s --check-prefix=CHECK
3+
; RUN: opt -loop-vectorize -debug-only=loop-vectorize -disable-output < %s 2>&1 | FileCheck %s --check-prefix=CHECK-COST
4+
; REQUIRES: asserts
35

46
target datalayout = "e-m:e-p:32:32-Fi8-i64:64-v128:64:128-a:0:32-n32-S64"
57
target triple = "thumbv8.1m.main-arm-none-eabi"
68

9+
; CHECK-COST-LABEL: arm_offset_q15
10+
; CHECK-COST: Found an estimated cost of 10 for VF 1 For instruction: %1 = tail call i16 @llvm.sadd.sat.i16(i16 %0, i16 %offset)
11+
; CHECK-COST: Found an estimated cost of 28 for VF 2 For instruction: %1 = tail call i16 @llvm.sadd.sat.i16(i16 %0, i16 %offset)
12+
; CHECK-COST: Found an estimated cost of 8 for VF 4 For instruction: %1 = tail call i16 @llvm.sadd.sat.i16(i16 %0, i16 %offset)
13+
; CHECK-COST: Found an estimated cost of 2 for VF 8 For instruction: %1 = tail call i16 @llvm.sadd.sat.i16(i16 %0, i16 %offset)
14+
715
define void @arm_offset_q15(i16* nocapture readonly %pSrc, i16 signext %offset, i16* nocapture noalias %pDst, i32 %blockSize) #0 {
816
; CHECK-LABEL: @arm_offset_q15(
917
; CHECK-NEXT: entry:

0 commit comments

Comments
 (0)