Skip to content

Commit 7bca4a2

Browse files
LebedevRILebedevRI
committed
[NFC][LoopVectorize] Autogenerate tests affected by isHighCostExpansionHelper() cost modelling (PR44668)
1 parent 9c801c4 commit 7bca4a2

File tree

1 file changed

+177
-21
lines changed

1 file changed

+177
-21
lines changed

‎llvm/test/Transforms/LoopVectorize/X86/float-induction-x86.ll

Lines changed: 177 additions & 21 deletions
Original file line numberDiff line numberDiff line change
@@ -1,3 +1,4 @@
1+
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
12
; RUN: opt < %s -O3 -simplifycfg -keep-loops=false -mcpu=core-avx2 -mtriple=x86_64-unknown-linux-gnu -S | FileCheck --check-prefix AUTO_VEC %s
23

34
; This test checks auto-vectorization with FP induction variable.
@@ -11,12 +12,58 @@
1112
; }
1213
;}
1314

14-
15-
; AUTO_VEC-LABEL: @fp_iv_loop1(
16-
; AUTO_VEC: vector.body
17-
; AUTO_VEC: store <8 x float>
18-
1915
define void @fp_iv_loop1(float* noalias nocapture %A, i32 %N) #0 {
16+
; AUTO_VEC-LABEL: @fp_iv_loop1(
17+
; AUTO_VEC-NEXT: entry:
18+
; AUTO_VEC-NEXT: [[CMP4:%.*]] = icmp sgt i32 [[N:%.*]], 0
19+
; AUTO_VEC-NEXT: br i1 [[CMP4]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_END:%.*]]
20+
; AUTO_VEC: for.body.preheader:
21+
; AUTO_VEC-NEXT: [[ZEXT:%.*]] = zext i32 [[N]] to i64
22+
; AUTO_VEC-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[N]], 32
23+
; AUTO_VEC-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[FOR_BODY:%.*]], label [[VECTOR_PH:%.*]]
24+
; AUTO_VEC: vector.ph:
25+
; AUTO_VEC-NEXT: [[N_VEC:%.*]] = and i64 [[ZEXT]], 4294967264
26+
; AUTO_VEC-NEXT: [[CAST_CRD:%.*]] = sitofp i64 [[N_VEC]] to float
27+
; AUTO_VEC-NEXT: [[TMP0:%.*]] = fmul fast float [[CAST_CRD]], 5.000000e-01
28+
; AUTO_VEC-NEXT: [[IND_END:%.*]] = fadd fast float [[TMP0]], 1.000000e+00
29+
; AUTO_VEC-NEXT: br label [[VECTOR_BODY:%.*]]
30+
; AUTO_VEC: vector.body:
31+
; AUTO_VEC-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
32+
; AUTO_VEC-NEXT: [[VEC_IND:%.*]] = phi <8 x float> [ <float 1.000000e+00, float 1.500000e+00, float 2.000000e+00, float 2.500000e+00, float 3.000000e+00, float 3.500000e+00, float 4.000000e+00, float 4.500000e+00>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
33+
; AUTO_VEC-NEXT: [[STEP_ADD:%.*]] = fadd fast <8 x float> [[VEC_IND]], <float 4.000000e+00, float 4.000000e+00, float 4.000000e+00, float 4.000000e+00, float 4.000000e+00, float 4.000000e+00, float 4.000000e+00, float 4.000000e+00>
34+
; AUTO_VEC-NEXT: [[STEP_ADD5:%.*]] = fadd fast <8 x float> [[VEC_IND]], <float 8.000000e+00, float 8.000000e+00, float 8.000000e+00, float 8.000000e+00, float 8.000000e+00, float 8.000000e+00, float 8.000000e+00, float 8.000000e+00>
35+
; AUTO_VEC-NEXT: [[STEP_ADD6:%.*]] = fadd fast <8 x float> [[VEC_IND]], <float 1.200000e+01, float 1.200000e+01, float 1.200000e+01, float 1.200000e+01, float 1.200000e+01, float 1.200000e+01, float 1.200000e+01, float 1.200000e+01>
36+
; AUTO_VEC-NEXT: [[TMP1:%.*]] = getelementptr inbounds float, float* [[A:%.*]], i64 [[INDEX]]
37+
; AUTO_VEC-NEXT: [[TMP2:%.*]] = bitcast float* [[TMP1]] to <8 x float>*
38+
; AUTO_VEC-NEXT: store <8 x float> [[VEC_IND]], <8 x float>* [[TMP2]], align 4
39+
; AUTO_VEC-NEXT: [[TMP3:%.*]] = getelementptr inbounds float, float* [[TMP1]], i64 8
40+
; AUTO_VEC-NEXT: [[TMP4:%.*]] = bitcast float* [[TMP3]] to <8 x float>*
41+
; AUTO_VEC-NEXT: store <8 x float> [[STEP_ADD]], <8 x float>* [[TMP4]], align 4
42+
; AUTO_VEC-NEXT: [[TMP5:%.*]] = getelementptr inbounds float, float* [[TMP1]], i64 16
43+
; AUTO_VEC-NEXT: [[TMP6:%.*]] = bitcast float* [[TMP5]] to <8 x float>*
44+
; AUTO_VEC-NEXT: store <8 x float> [[STEP_ADD5]], <8 x float>* [[TMP6]], align 4
45+
; AUTO_VEC-NEXT: [[TMP7:%.*]] = getelementptr inbounds float, float* [[TMP1]], i64 24
46+
; AUTO_VEC-NEXT: [[TMP8:%.*]] = bitcast float* [[TMP7]] to <8 x float>*
47+
; AUTO_VEC-NEXT: store <8 x float> [[STEP_ADD6]], <8 x float>* [[TMP8]], align 4
48+
; AUTO_VEC-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 32
49+
; AUTO_VEC-NEXT: [[VEC_IND_NEXT]] = fadd fast <8 x float> [[VEC_IND]], <float 1.600000e+01, float 1.600000e+01, float 1.600000e+01, float 1.600000e+01, float 1.600000e+01, float 1.600000e+01, float 1.600000e+01, float 1.600000e+01>
50+
; AUTO_VEC-NEXT: [[TMP9:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
51+
; AUTO_VEC-NEXT: br i1 [[TMP9]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop !0
52+
; AUTO_VEC: middle.block:
53+
; AUTO_VEC-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N_VEC]], [[ZEXT]]
54+
; AUTO_VEC-NEXT: br i1 [[CMP_N]], label [[FOR_END]], label [[FOR_BODY]]
55+
; AUTO_VEC: for.body:
56+
; AUTO_VEC-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ], [ 0, [[FOR_BODY_PREHEADER]] ], [ [[N_VEC]], [[MIDDLE_BLOCK]] ]
57+
; AUTO_VEC-NEXT: [[X_06:%.*]] = phi float [ [[CONV1:%.*]], [[FOR_BODY]] ], [ 1.000000e+00, [[FOR_BODY_PREHEADER]] ], [ [[IND_END]], [[MIDDLE_BLOCK]] ]
58+
; AUTO_VEC-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[INDVARS_IV]]
59+
; AUTO_VEC-NEXT: store float [[X_06]], float* [[ARRAYIDX]], align 4
60+
; AUTO_VEC-NEXT: [[CONV1]] = fadd float [[X_06]], 5.000000e-01
61+
; AUTO_VEC-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 1
62+
; AUTO_VEC-NEXT: [[TMP10:%.*]] = icmp eq i64 [[INDVARS_IV_NEXT]], [[ZEXT]]
63+
; AUTO_VEC-NEXT: br i1 [[TMP10]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop !2
64+
; AUTO_VEC: for.end:
65+
; AUTO_VEC-NEXT: ret void
66+
;
2067
entry:
2168
%cmp4 = icmp sgt i32 %N, 0
2269
br i1 %cmp4, label %for.body.preheader, label %for.end
@@ -52,11 +99,78 @@ for.end: ; preds = %for.end.loopexit, %
5299
; }
53100
;}
54101

55-
; AUTO_VEC-LABEL: @fp_iv_loop2(
56-
; AUTO_VEC-NOT: vector.body
57-
; AUTO_VEC-NOT: store <{{.*}} x float>
58-
59102
define void @fp_iv_loop2(float* noalias nocapture %A, i32 %N) #1 {
103+
; AUTO_VEC-LABEL: @fp_iv_loop2(
104+
; AUTO_VEC-NEXT: entry:
105+
; AUTO_VEC-NEXT: [[CMP4:%.*]] = icmp sgt i32 [[N:%.*]], 0
106+
; AUTO_VEC-NEXT: br i1 [[CMP4]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_END:%.*]]
107+
; AUTO_VEC: for.body.preheader:
108+
; AUTO_VEC-NEXT: [[ZEXT:%.*]] = zext i32 [[N]] to i64
109+
; AUTO_VEC-NEXT: [[TMP0:%.*]] = add nsw i64 [[ZEXT]], -1
110+
; AUTO_VEC-NEXT: [[XTRAITER:%.*]] = and i64 [[ZEXT]], 7
111+
; AUTO_VEC-NEXT: [[TMP1:%.*]] = icmp ult i64 [[TMP0]], 7
112+
; AUTO_VEC-NEXT: br i1 [[TMP1]], label [[FOR_END_LOOPEXIT_UNR_LCSSA:%.*]], label [[FOR_BODY_PREHEADER_NEW:%.*]]
113+
; AUTO_VEC: for.body.preheader.new:
114+
; AUTO_VEC-NEXT: [[UNROLL_ITER:%.*]] = sub nsw i64 [[ZEXT]], [[XTRAITER]]
115+
; AUTO_VEC-NEXT: br label [[FOR_BODY:%.*]]
116+
; AUTO_VEC: for.body:
117+
; AUTO_VEC-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ 0, [[FOR_BODY_PREHEADER_NEW]] ], [ [[INDVARS_IV_NEXT_7:%.*]], [[FOR_BODY]] ]
118+
; AUTO_VEC-NEXT: [[X_06:%.*]] = phi float [ 1.000000e+00, [[FOR_BODY_PREHEADER_NEW]] ], [ [[CONV1_7:%.*]], [[FOR_BODY]] ]
119+
; AUTO_VEC-NEXT: [[NITER:%.*]] = phi i64 [ [[UNROLL_ITER]], [[FOR_BODY_PREHEADER_NEW]] ], [ [[NITER_NSUB_7:%.*]], [[FOR_BODY]] ]
120+
; AUTO_VEC-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[A:%.*]], i64 [[INDVARS_IV]]
121+
; AUTO_VEC-NEXT: store float [[X_06]], float* [[ARRAYIDX]], align 4
122+
; AUTO_VEC-NEXT: [[CONV1:%.*]] = fadd float [[X_06]], 5.000000e-01
123+
; AUTO_VEC-NEXT: [[INDVARS_IV_NEXT:%.*]] = or i64 [[INDVARS_IV]], 1
124+
; AUTO_VEC-NEXT: [[ARRAYIDX_1:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[INDVARS_IV_NEXT]]
125+
; AUTO_VEC-NEXT: store float [[CONV1]], float* [[ARRAYIDX_1]], align 4
126+
; AUTO_VEC-NEXT: [[CONV1_1:%.*]] = fadd float [[CONV1]], 5.000000e-01
127+
; AUTO_VEC-NEXT: [[INDVARS_IV_NEXT_1:%.*]] = or i64 [[INDVARS_IV]], 2
128+
; AUTO_VEC-NEXT: [[ARRAYIDX_2:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[INDVARS_IV_NEXT_1]]
129+
; AUTO_VEC-NEXT: store float [[CONV1_1]], float* [[ARRAYIDX_2]], align 4
130+
; AUTO_VEC-NEXT: [[CONV1_2:%.*]] = fadd float [[CONV1_1]], 5.000000e-01
131+
; AUTO_VEC-NEXT: [[INDVARS_IV_NEXT_2:%.*]] = or i64 [[INDVARS_IV]], 3
132+
; AUTO_VEC-NEXT: [[ARRAYIDX_3:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[INDVARS_IV_NEXT_2]]
133+
; AUTO_VEC-NEXT: store float [[CONV1_2]], float* [[ARRAYIDX_3]], align 4
134+
; AUTO_VEC-NEXT: [[CONV1_3:%.*]] = fadd float [[CONV1_2]], 5.000000e-01
135+
; AUTO_VEC-NEXT: [[INDVARS_IV_NEXT_3:%.*]] = or i64 [[INDVARS_IV]], 4
136+
; AUTO_VEC-NEXT: [[ARRAYIDX_4:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[INDVARS_IV_NEXT_3]]
137+
; AUTO_VEC-NEXT: store float [[CONV1_3]], float* [[ARRAYIDX_4]], align 4
138+
; AUTO_VEC-NEXT: [[CONV1_4:%.*]] = fadd float [[CONV1_3]], 5.000000e-01
139+
; AUTO_VEC-NEXT: [[INDVARS_IV_NEXT_4:%.*]] = or i64 [[INDVARS_IV]], 5
140+
; AUTO_VEC-NEXT: [[ARRAYIDX_5:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[INDVARS_IV_NEXT_4]]
141+
; AUTO_VEC-NEXT: store float [[CONV1_4]], float* [[ARRAYIDX_5]], align 4
142+
; AUTO_VEC-NEXT: [[CONV1_5:%.*]] = fadd float [[CONV1_4]], 5.000000e-01
143+
; AUTO_VEC-NEXT: [[INDVARS_IV_NEXT_5:%.*]] = or i64 [[INDVARS_IV]], 6
144+
; AUTO_VEC-NEXT: [[ARRAYIDX_6:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[INDVARS_IV_NEXT_5]]
145+
; AUTO_VEC-NEXT: store float [[CONV1_5]], float* [[ARRAYIDX_6]], align 4
146+
; AUTO_VEC-NEXT: [[CONV1_6:%.*]] = fadd float [[CONV1_5]], 5.000000e-01
147+
; AUTO_VEC-NEXT: [[INDVARS_IV_NEXT_6:%.*]] = or i64 [[INDVARS_IV]], 7
148+
; AUTO_VEC-NEXT: [[ARRAYIDX_7:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[INDVARS_IV_NEXT_6]]
149+
; AUTO_VEC-NEXT: store float [[CONV1_6]], float* [[ARRAYIDX_7]], align 4
150+
; AUTO_VEC-NEXT: [[CONV1_7]] = fadd float [[CONV1_6]], 5.000000e-01
151+
; AUTO_VEC-NEXT: [[INDVARS_IV_NEXT_7]] = add nuw nsw i64 [[INDVARS_IV]], 8
152+
; AUTO_VEC-NEXT: [[NITER_NSUB_7]] = add i64 [[NITER]], -8
153+
; AUTO_VEC-NEXT: [[NITER_NCMP_7:%.*]] = icmp eq i64 [[NITER_NSUB_7]], 0
154+
; AUTO_VEC-NEXT: br i1 [[NITER_NCMP_7]], label [[FOR_END_LOOPEXIT_UNR_LCSSA]], label [[FOR_BODY]]
155+
; AUTO_VEC: for.end.loopexit.unr-lcssa:
156+
; AUTO_VEC-NEXT: [[INDVARS_IV_UNR:%.*]] = phi i64 [ 0, [[FOR_BODY_PREHEADER]] ], [ [[INDVARS_IV_NEXT_7]], [[FOR_BODY]] ]
157+
; AUTO_VEC-NEXT: [[X_06_UNR:%.*]] = phi float [ 1.000000e+00, [[FOR_BODY_PREHEADER]] ], [ [[CONV1_7]], [[FOR_BODY]] ]
158+
; AUTO_VEC-NEXT: [[LCMP_MOD:%.*]] = icmp eq i64 [[XTRAITER]], 0
159+
; AUTO_VEC-NEXT: br i1 [[LCMP_MOD]], label [[FOR_END]], label [[FOR_BODY_EPIL:%.*]]
160+
; AUTO_VEC: for.body.epil:
161+
; AUTO_VEC-NEXT: [[INDVARS_IV_EPIL:%.*]] = phi i64 [ [[INDVARS_IV_NEXT_EPIL:%.*]], [[FOR_BODY_EPIL]] ], [ [[INDVARS_IV_UNR]], [[FOR_END_LOOPEXIT_UNR_LCSSA]] ]
162+
; AUTO_VEC-NEXT: [[X_06_EPIL:%.*]] = phi float [ [[CONV1_EPIL:%.*]], [[FOR_BODY_EPIL]] ], [ [[X_06_UNR]], [[FOR_END_LOOPEXIT_UNR_LCSSA]] ]
163+
; AUTO_VEC-NEXT: [[EPIL_ITER:%.*]] = phi i64 [ [[EPIL_ITER_SUB:%.*]], [[FOR_BODY_EPIL]] ], [ [[XTRAITER]], [[FOR_END_LOOPEXIT_UNR_LCSSA]] ]
164+
; AUTO_VEC-NEXT: [[ARRAYIDX_EPIL:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[INDVARS_IV_EPIL]]
165+
; AUTO_VEC-NEXT: store float [[X_06_EPIL]], float* [[ARRAYIDX_EPIL]], align 4
166+
; AUTO_VEC-NEXT: [[CONV1_EPIL]] = fadd float [[X_06_EPIL]], 5.000000e-01
167+
; AUTO_VEC-NEXT: [[INDVARS_IV_NEXT_EPIL]] = add nuw nsw i64 [[INDVARS_IV_EPIL]], 1
168+
; AUTO_VEC-NEXT: [[EPIL_ITER_SUB]] = add i64 [[EPIL_ITER]], -1
169+
; AUTO_VEC-NEXT: [[EPIL_ITER_CMP:%.*]] = icmp eq i64 [[EPIL_ITER_SUB]], 0
170+
; AUTO_VEC-NEXT: br i1 [[EPIL_ITER_CMP]], label [[FOR_END]], label [[FOR_BODY_EPIL]], !llvm.loop !4
171+
; AUTO_VEC: for.end:
172+
; AUTO_VEC-NEXT: ret void
173+
;
60174
entry:
61175
%cmp4 = icmp sgt i32 %N, 0
62176
br i1 %cmp4, label %for.body.preheader, label %for.end
@@ -82,24 +196,59 @@ for.end: ; preds = %for.end.loopexit, %
82196
ret void
83197
}
84198

199+
define double @external_use_with_fast_math(double* %a, i64 %n) {
85200
; AUTO_VEC-LABEL: @external_use_with_fast_math(
86201
; AUTO_VEC-NEXT: entry:
87-
; AUTO_VEC-NEXT: [[TMP0:%.*]] = icmp sgt i64 %n, 1
88-
; AUTO_VEC-NEXT: [[SMAX:%.*]] = select i1 [[TMP0]], i64 %n, i64 1
89-
; AUTO_VEC: br i1 {{.*}}, label %for.body, label %vector.ph
202+
; AUTO_VEC-NEXT: [[TMP0:%.*]] = icmp sgt i64 [[N:%.*]], 1
203+
; AUTO_VEC-NEXT: [[SMAX:%.*]] = select i1 [[TMP0]], i64 [[N]], i64 1
204+
; AUTO_VEC-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[SMAX]], 16
205+
; AUTO_VEC-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[FOR_BODY:%.*]], label [[VECTOR_PH:%.*]]
90206
; AUTO_VEC: vector.ph:
91207
; AUTO_VEC-NEXT: [[N_VEC:%.*]] = and i64 [[SMAX]], 9223372036854775792
92-
; AUTO_VEC: br label %vector.body
208+
; AUTO_VEC-NEXT: [[CAST_CRD:%.*]] = sitofp i64 [[N_VEC]] to double
209+
; AUTO_VEC-NEXT: [[TMP1:%.*]] = fmul fast double [[CAST_CRD]], 3.000000e+00
210+
; AUTO_VEC-NEXT: br label [[VECTOR_BODY:%.*]]
211+
; AUTO_VEC: vector.body:
212+
; AUTO_VEC-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ]
213+
; AUTO_VEC-NEXT: [[VEC_IND:%.*]] = phi <4 x double> [ <double 0.000000e+00, double 3.000000e+00, double 6.000000e+00, double 9.000000e+00>, [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ]
214+
; AUTO_VEC-NEXT: [[STEP_ADD:%.*]] = fadd fast <4 x double> [[VEC_IND]], <double 1.200000e+01, double 1.200000e+01, double 1.200000e+01, double 1.200000e+01>
215+
; AUTO_VEC-NEXT: [[STEP_ADD5:%.*]] = fadd fast <4 x double> [[VEC_IND]], <double 2.400000e+01, double 2.400000e+01, double 2.400000e+01, double 2.400000e+01>
216+
; AUTO_VEC-NEXT: [[STEP_ADD6:%.*]] = fadd fast <4 x double> [[VEC_IND]], <double 3.600000e+01, double 3.600000e+01, double 3.600000e+01, double 3.600000e+01>
217+
; AUTO_VEC-NEXT: [[TMP2:%.*]] = getelementptr double, double* [[A:%.*]], i64 [[INDEX]]
218+
; AUTO_VEC-NEXT: [[TMP3:%.*]] = bitcast double* [[TMP2]] to <4 x double>*
219+
; AUTO_VEC-NEXT: store <4 x double> [[VEC_IND]], <4 x double>* [[TMP3]], align 8
220+
; AUTO_VEC-NEXT: [[TMP4:%.*]] = getelementptr double, double* [[TMP2]], i64 4
221+
; AUTO_VEC-NEXT: [[TMP5:%.*]] = bitcast double* [[TMP4]] to <4 x double>*
222+
; AUTO_VEC-NEXT: store <4 x double> [[STEP_ADD]], <4 x double>* [[TMP5]], align 8
223+
; AUTO_VEC-NEXT: [[TMP6:%.*]] = getelementptr double, double* [[TMP2]], i64 8
224+
; AUTO_VEC-NEXT: [[TMP7:%.*]] = bitcast double* [[TMP6]] to <4 x double>*
225+
; AUTO_VEC-NEXT: store <4 x double> [[STEP_ADD5]], <4 x double>* [[TMP7]], align 8
226+
; AUTO_VEC-NEXT: [[TMP8:%.*]] = getelementptr double, double* [[TMP2]], i64 12
227+
; AUTO_VEC-NEXT: [[TMP9:%.*]] = bitcast double* [[TMP8]] to <4 x double>*
228+
; AUTO_VEC-NEXT: store <4 x double> [[STEP_ADD6]], <4 x double>* [[TMP9]], align 8
229+
; AUTO_VEC-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 16
230+
; AUTO_VEC-NEXT: [[VEC_IND_NEXT]] = fadd fast <4 x double> [[VEC_IND]], <double 4.800000e+01, double 4.800000e+01, double 4.800000e+01, double 4.800000e+01>
231+
; AUTO_VEC-NEXT: [[TMP10:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]]
232+
; AUTO_VEC-NEXT: br i1 [[TMP10]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop !6
93233
; AUTO_VEC: middle.block:
94-
; AUTO_VEC: [[TMP11:%.*]] = add nsw i64 [[N_VEC]], -1
234+
; AUTO_VEC-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[SMAX]], [[N_VEC]]
235+
; AUTO_VEC-NEXT: [[TMP11:%.*]] = add nsw i64 [[N_VEC]], -1
95236
; AUTO_VEC-NEXT: [[CAST_CMO:%.*]] = sitofp i64 [[TMP11]] to double
96237
; AUTO_VEC-NEXT: [[TMP12:%.*]] = fmul fast double [[CAST_CMO]], 3.000000e+00
97-
; AUTO_VEC-NEXT: br i1 {{.*}}, label %for.end, label %for.body
238+
; AUTO_VEC-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[FOR_BODY]]
239+
; AUTO_VEC: for.body:
240+
; AUTO_VEC-NEXT: [[I:%.*]] = phi i64 [ [[I_NEXT:%.*]], [[FOR_BODY]] ], [ 0, [[ENTRY:%.*]] ], [ [[N_VEC]], [[MIDDLE_BLOCK]] ]
241+
; AUTO_VEC-NEXT: [[J:%.*]] = phi double [ [[J_NEXT:%.*]], [[FOR_BODY]] ], [ 0.000000e+00, [[ENTRY]] ], [ [[TMP1]], [[MIDDLE_BLOCK]] ]
242+
; AUTO_VEC-NEXT: [[TMP0:%.*]] = getelementptr double, double* [[A]], i64 [[I]]
243+
; AUTO_VEC-NEXT: store double [[J]], double* [[TMP0]], align 8
244+
; AUTO_VEC-NEXT: [[I_NEXT]] = add nuw nsw i64 [[I]], 1
245+
; AUTO_VEC-NEXT: [[J_NEXT]] = fadd fast double [[J]], 3.000000e+00
246+
; AUTO_VEC-NEXT: [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N]]
247+
; AUTO_VEC-NEXT: br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END]], !llvm.loop !7
98248
; AUTO_VEC: for.end:
99-
; AUTO_VEC-NEXT: [[J_LCSSA:%.*]] = phi double [ [[TMP12]], %middle.block ], [ %j, %for.body ]
249+
; AUTO_VEC-NEXT: [[J_LCSSA:%.*]] = phi double [ [[TMP12]], [[MIDDLE_BLOCK]] ], [ [[J]], [[FOR_BODY]] ]
100250
; AUTO_VEC-NEXT: ret double [[J_LCSSA]]
101251
;
102-
define double @external_use_with_fast_math(double* %a, i64 %n) {
103252
entry:
104253
br label %for.body
105254

@@ -118,15 +267,22 @@ for.end:
118267
ret double %tmp1
119268
}
120269

270+
define double @external_use_without_fast_math(double* %a, i64 %n) {
121271
; AUTO_VEC-LABEL: @external_use_without_fast_math(
272+
; AUTO_VEC-NEXT: entry:
273+
; AUTO_VEC-NEXT: br label [[FOR_BODY:%.*]]
122274
; AUTO_VEC: for.body:
123-
; AUTO_VEC: [[J:%.*]] = phi double [ 0.000000e+00, %entry ], [ [[J_NEXT:%.*]], %for.body ]
124-
; AUTO_VEC: [[J_NEXT]] = fadd double [[J]], 3.000000e+00
125-
; AUTO_VEC: br i1 {{.*}}, label %for.body, label %for.end
275+
; AUTO_VEC-NEXT: [[I:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[I_NEXT:%.*]], [[FOR_BODY]] ]
276+
; AUTO_VEC-NEXT: [[J:%.*]] = phi double [ 0.000000e+00, [[ENTRY]] ], [ [[J_NEXT:%.*]], [[FOR_BODY]] ]
277+
; AUTO_VEC-NEXT: [[TMP0:%.*]] = getelementptr double, double* [[A:%.*]], i64 [[I]]
278+
; AUTO_VEC-NEXT: store double [[J]], double* [[TMP0]], align 8
279+
; AUTO_VEC-NEXT: [[I_NEXT]] = add nuw nsw i64 [[I]], 1
280+
; AUTO_VEC-NEXT: [[J_NEXT]] = fadd double [[J]], 3.000000e+00
281+
; AUTO_VEC-NEXT: [[COND:%.*]] = icmp slt i64 [[I_NEXT]], [[N:%.*]]
282+
; AUTO_VEC-NEXT: br i1 [[COND]], label [[FOR_BODY]], label [[FOR_END:%.*]]
126283
; AUTO_VEC: for.end:
127284
; AUTO_VEC-NEXT: ret double [[J]]
128285
;
129-
define double @external_use_without_fast_math(double* %a, i64 %n) {
130286
entry:
131287
br label %for.body
132288

0 commit comments

Comments
 (0)