1 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2 ; RUN: opt < %s -passes=slp-vectorizer,dce -slp-threshold=-100 -S -mtriple=i386-apple-macosx10.8.0 -mcpu=corei7-avx | FileCheck %s
4 target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128-n8:16:32-S128"
5 target triple = "i386-apple-macosx10.9.0"
7 ;int foo(ptr A, int k) {
22 define i32 @foo(ptr nocapture %A, i32 %k) {
25 ; CHECK-NEXT: [[TOBOOL:%.*]] = icmp eq i32 [[K:%.*]], 0
26 ; CHECK-NEXT: br i1 [[TOBOOL]], label [[IF_ELSE:%.*]], label [[IF_END:%.*]]
28 ; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds double, ptr [[A:%.*]], i64 10
29 ; CHECK-NEXT: [[TMP0:%.*]] = load <2 x double>, ptr [[ARRAYIDX]], align 8
30 ; CHECK-NEXT: br label [[IF_END]]
32 ; CHECK-NEXT: [[TMP1:%.*]] = phi <2 x double> [ [[TMP0]], [[IF_ELSE]] ], [ <double 3.000000e+00, double 5.000000e+00>, [[ENTRY:%.*]] ]
33 ; CHECK-NEXT: store <2 x double> [[TMP1]], ptr [[A]], align 8
34 ; CHECK-NEXT: ret i32 undef
37 %tobool = icmp eq i32 %k, 0
38 br i1 %tobool, label %if.else, label %if.end
40 if.else: ; preds = %entry
41 %arrayidx = getelementptr inbounds double, ptr %A, i64 10
42 %0 = load double, ptr %arrayidx, align 8
43 %arrayidx1 = getelementptr inbounds double, ptr %A, i64 11
44 %1 = load double, ptr %arrayidx1, align 8
47 if.end: ; preds = %entry, %if.else
48 %A0.0 = phi double [ %0, %if.else ], [ 3.000000e+00, %entry ]
49 %A1.0 = phi double [ %1, %if.else ], [ 5.000000e+00, %entry ]
50 store double %A0.0, ptr %A, align 8
51 %arrayidx3 = getelementptr inbounds double, ptr %A, i64 1
52 store double %A1.0, ptr %arrayidx3, align 8
57 ;int foo(ptr restrict B, ptr restrict A, int n, int m) {
60 ; for (int i=0; i < 100; i++) {
73 define i32 @foo2(ptr noalias nocapture %B, ptr noalias nocapture %A, i32 %n, i32 %m) #0 {
76 ; CHECK-NEXT: [[TMP0:%.*]] = load <2 x double>, ptr [[A:%.*]], align 8
77 ; CHECK-NEXT: br label [[FOR_BODY:%.*]]
79 ; CHECK-NEXT: [[I_019:%.*]] = phi i32 [ 0, [[ENTRY:%.*]] ], [ [[INC:%.*]], [[FOR_BODY]] ]
80 ; CHECK-NEXT: [[TMP1:%.*]] = phi <2 x double> [ [[TMP0]], [[ENTRY]] ], [ [[TMP4:%.*]], [[FOR_BODY]] ]
81 ; CHECK-NEXT: [[TMP2:%.*]] = fadd <2 x double> [[TMP1]], <double 1.000000e+01, double 1.000000e+01>
82 ; CHECK-NEXT: [[TMP3:%.*]] = fmul <2 x double> [[TMP2]], <double 4.000000e+00, double 4.000000e+00>
83 ; CHECK-NEXT: [[TMP4]] = fadd <2 x double> [[TMP3]], <double 4.000000e+00, double 4.000000e+00>
84 ; CHECK-NEXT: [[INC]] = add nsw i32 [[I_019]], 1
85 ; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[INC]], 100
86 ; CHECK-NEXT: br i1 [[EXITCOND]], label [[FOR_END:%.*]], label [[FOR_BODY]]
88 ; CHECK-NEXT: store <2 x double> [[TMP4]], ptr [[B:%.*]], align 8
89 ; CHECK-NEXT: ret i32 0
92 %arrayidx = getelementptr inbounds double, ptr %A, i64 1
93 %0 = load double, ptr %arrayidx, align 8
94 %1 = load double, ptr %A, align 8
97 for.body: ; preds = %for.body, %entry
98 %i.019 = phi i32 [ 0, %entry ], [ %inc, %for.body ]
99 %G.018 = phi double [ %1, %entry ], [ %add5, %for.body ]
100 %R.017 = phi double [ %0, %entry ], [ %add4, %for.body ]
101 %add = fadd double %R.017, 1.000000e+01
102 %add2 = fadd double %G.018, 1.000000e+01
103 %mul = fmul double %add, 4.000000e+00
104 %mul3 = fmul double %add2, 4.000000e+00
105 %add4 = fadd double %mul, 4.000000e+00
106 %add5 = fadd double %mul3, 4.000000e+00
107 %inc = add nsw i32 %i.019, 1
108 %exitcond = icmp eq i32 %inc, 100
109 br i1 %exitcond, label %for.end, label %for.body
111 for.end: ; preds = %for.body
112 store double %add5, ptr %B, align 8
113 %arrayidx7 = getelementptr inbounds double, ptr %B, i64 1
114 store double %add4, ptr %arrayidx7, align 8
118 ; float foo3(ptr A) {
125 ; for (int i=0; i < 121; i+=3) {
136 define float @foo3(ptr nocapture readonly %A) #0 {
137 ; CHECK-LABEL: @foo3(
139 ; CHECK-NEXT: [[TMP0:%.*]] = load float, ptr [[A:%.*]], align 4
140 ; CHECK-NEXT: [[ARRAYIDX1:%.*]] = getelementptr inbounds float, ptr [[A]], i64 1
141 ; CHECK-NEXT: [[TMP1:%.*]] = load <4 x float>, ptr [[ARRAYIDX1]], align 4
142 ; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <4 x float> [[TMP1]], <4 x float> poison, <2 x i32> <i32 poison, i32 0>
143 ; CHECK-NEXT: [[TMP3:%.*]] = insertelement <2 x float> [[TMP2]], float [[TMP0]], i32 0
144 ; CHECK-NEXT: br label [[FOR_BODY:%.*]]
146 ; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ]
147 ; CHECK-NEXT: [[R_052:%.*]] = phi float [ [[TMP0]], [[ENTRY]] ], [ [[ADD6:%.*]], [[FOR_BODY]] ]
148 ; CHECK-NEXT: [[TMP4:%.*]] = phi <4 x float> [ [[TMP1]], [[ENTRY]] ], [ [[TMP13:%.*]], [[FOR_BODY]] ]
149 ; CHECK-NEXT: [[TMP5:%.*]] = phi <2 x float> [ [[TMP3]], [[ENTRY]] ], [ [[TMP9:%.*]], [[FOR_BODY]] ]
150 ; CHECK-NEXT: [[TMP6:%.*]] = extractelement <2 x float> [[TMP5]], i32 0
151 ; CHECK-NEXT: [[MUL:%.*]] = fmul float [[TMP6]], 7.000000e+00
152 ; CHECK-NEXT: [[ADD6]] = fadd float [[R_052]], [[MUL]]
153 ; CHECK-NEXT: [[TMP7:%.*]] = add nsw i64 [[INDVARS_IV]], 2
154 ; CHECK-NEXT: [[ARRAYIDX14:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[TMP7]]
155 ; CHECK-NEXT: [[TMP8:%.*]] = load float, ptr [[ARRAYIDX14]], align 4
156 ; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nuw nsw i64 [[INDVARS_IV]], 3
157 ; CHECK-NEXT: [[ARRAYIDX19:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[INDVARS_IV_NEXT]]
158 ; CHECK-NEXT: [[TMP9]] = load <2 x float>, ptr [[ARRAYIDX19]], align 4
159 ; CHECK-NEXT: [[TMP10:%.*]] = shufflevector <2 x float> [[TMP5]], <2 x float> [[TMP9]], <4 x i32> <i32 1, i32 poison, i32 2, i32 3>
160 ; CHECK-NEXT: [[TMP11:%.*]] = insertelement <4 x float> [[TMP10]], float [[TMP8]], i32 1
161 ; CHECK-NEXT: [[TMP12:%.*]] = fmul <4 x float> [[TMP11]], <float 8.000000e+00, float 9.000000e+00, float 1.000000e+01, float 1.100000e+01>
162 ; CHECK-NEXT: [[TMP13]] = fadd <4 x float> [[TMP4]], [[TMP12]]
163 ; CHECK-NEXT: [[TMP14:%.*]] = trunc i64 [[INDVARS_IV_NEXT]] to i32
164 ; CHECK-NEXT: [[CMP:%.*]] = icmp slt i32 [[TMP14]], 121
165 ; CHECK-NEXT: br i1 [[CMP]], label [[FOR_BODY]], label [[FOR_END:%.*]]
167 ; CHECK-NEXT: [[TMP15:%.*]] = extractelement <4 x float> [[TMP13]], i32 0
168 ; CHECK-NEXT: [[ADD28:%.*]] = fadd float [[ADD6]], [[TMP15]]
169 ; CHECK-NEXT: [[TMP16:%.*]] = extractelement <4 x float> [[TMP13]], i32 1
170 ; CHECK-NEXT: [[ADD29:%.*]] = fadd float [[ADD28]], [[TMP16]]
171 ; CHECK-NEXT: [[TMP17:%.*]] = extractelement <4 x float> [[TMP13]], i32 2
172 ; CHECK-NEXT: [[ADD30:%.*]] = fadd float [[ADD29]], [[TMP17]]
173 ; CHECK-NEXT: [[TMP18:%.*]] = extractelement <4 x float> [[TMP13]], i32 3
174 ; CHECK-NEXT: [[ADD31:%.*]] = fadd float [[ADD30]], [[TMP18]]
175 ; CHECK-NEXT: ret float [[ADD31]]
178 %0 = load float, ptr %A, align 4
179 %arrayidx1 = getelementptr inbounds float, ptr %A, i64 1
180 %1 = load float, ptr %arrayidx1, align 4
181 %arrayidx2 = getelementptr inbounds float, ptr %A, i64 2
182 %2 = load float, ptr %arrayidx2, align 4
183 %arrayidx3 = getelementptr inbounds float, ptr %A, i64 3
184 %3 = load float, ptr %arrayidx3, align 4
185 %arrayidx4 = getelementptr inbounds float, ptr %A, i64 4
186 %4 = load float, ptr %arrayidx4, align 4
189 for.body: ; preds = %for.body, %entry
190 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
191 %P.056 = phi float [ %4, %entry ], [ %add26, %for.body ]
192 %Y.055 = phi float [ %3, %entry ], [ %add21, %for.body ]
193 %B.054 = phi float [ %2, %entry ], [ %add16, %for.body ]
194 %G.053 = phi float [ %1, %entry ], [ %add11, %for.body ]
195 %R.052 = phi float [ %0, %entry ], [ %add6, %for.body ]
196 %5 = phi float [ %1, %entry ], [ %11, %for.body ]
197 %6 = phi float [ %0, %entry ], [ %9, %for.body ]
198 %mul = fmul float %6, 7.000000e+00
199 %add6 = fadd float %R.052, %mul
200 %mul10 = fmul float %5, 8.000000e+00
201 %add11 = fadd float %G.053, %mul10
202 %7 = add nsw i64 %indvars.iv, 2
203 %arrayidx14 = getelementptr inbounds float, ptr %A, i64 %7
204 %8 = load float, ptr %arrayidx14, align 4
205 %mul15 = fmul float %8, 9.000000e+00
206 %add16 = fadd float %B.054, %mul15
207 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 3
208 %arrayidx19 = getelementptr inbounds float, ptr %A, i64 %indvars.iv.next
209 %9 = load float, ptr %arrayidx19, align 4
210 %mul20 = fmul float %9, 1.000000e+01
211 %add21 = fadd float %Y.055, %mul20
212 %10 = add nsw i64 %indvars.iv, 4
213 %arrayidx24 = getelementptr inbounds float, ptr %A, i64 %10
214 %11 = load float, ptr %arrayidx24, align 4
215 %mul25 = fmul float %11, 1.100000e+01
216 %add26 = fadd float %P.056, %mul25
217 %12 = trunc i64 %indvars.iv.next to i32
218 %cmp = icmp slt i32 %12, 121
219 br i1 %cmp, label %for.body, label %for.end
221 for.end: ; preds = %for.body
222 %add28 = fadd float %add6, %add11
223 %add29 = fadd float %add28, %add16
224 %add30 = fadd float %add29, %add21
225 %add31 = fadd float %add30, %add26
229 ; Make sure the order of phi nodes of different types does not prevent
230 ; vectorization of same typed phi nodes.
231 define float @sort_phi_type(ptr nocapture readonly %A) {
232 ; CHECK-LABEL: @sort_phi_type(
234 ; CHECK-NEXT: br label [[FOR_BODY:%.*]]
236 ; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ]
237 ; CHECK-NEXT: [[TMP0:%.*]] = phi <4 x float> [ <float 1.000000e+01, float 1.000000e+01, float 1.000000e+01, float 1.000000e+01>, [[ENTRY]] ], [ [[TMP2:%.*]], [[FOR_BODY]] ]
238 ; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <4 x float> [[TMP0]], <4 x float> poison, <4 x i32> <i32 0, i32 1, i32 3, i32 2>
239 ; CHECK-NEXT: [[TMP2]] = fmul <4 x float> [[TMP1]], <float 8.000000e+00, float 9.000000e+00, float 1.000000e+02, float 1.110000e+02>
240 ; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add nsw i64 [[INDVARS_IV]], 4
241 ; CHECK-NEXT: [[CMP:%.*]] = icmp slt i64 [[INDVARS_IV_NEXT]], 128
242 ; CHECK-NEXT: br i1 [[CMP]], label [[FOR_BODY]], label [[FOR_END:%.*]]
244 ; CHECK-NEXT: [[TMP3:%.*]] = extractelement <4 x float> [[TMP2]], i32 0
245 ; CHECK-NEXT: [[TMP4:%.*]] = extractelement <4 x float> [[TMP2]], i32 1
246 ; CHECK-NEXT: [[ADD29:%.*]] = fadd float [[TMP3]], [[TMP4]]
247 ; CHECK-NEXT: [[TMP5:%.*]] = extractelement <4 x float> [[TMP2]], i32 2
248 ; CHECK-NEXT: [[ADD30:%.*]] = fadd float [[ADD29]], [[TMP5]]
249 ; CHECK-NEXT: [[TMP6:%.*]] = extractelement <4 x float> [[TMP2]], i32 3
250 ; CHECK-NEXT: [[ADD31:%.*]] = fadd float [[ADD30]], [[TMP6]]
251 ; CHECK-NEXT: ret float [[ADD31]]
256 for.body: ; preds = %for.body, %entry
257 %Y = phi float [ 1.000000e+01, %entry ], [ %mul10, %for.body ]
258 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
259 %B = phi float [ 1.000000e+01, %entry ], [ %mul15, %for.body ]
260 %G = phi float [ 1.000000e+01, %entry ], [ %mul20, %for.body ]
261 %R = phi float [ 1.000000e+01, %entry ], [ %mul25, %for.body ]
262 %mul10 = fmul float %Y, 8.000000e+00
263 %mul15 = fmul float %B, 9.000000e+00
264 %mul20 = fmul float %R, 10.000000e+01
265 %mul25 = fmul float %G, 11.100000e+01
266 %indvars.iv.next = add nsw i64 %indvars.iv, 4
267 %cmp = icmp slt i64 %indvars.iv.next, 128
268 br i1 %cmp, label %for.body, label %for.end
270 for.end: ; preds = %for.body
271 %add28 = fadd float 1.000000e+01, %mul10
272 %add29 = fadd float %mul10, %mul15
273 %add30 = fadd float %add29, %mul20
274 %add31 = fadd float %add30, %mul25
278 define void @test(ptr %i1, ptr %i2, ptr %o) {
279 ; CHECK-LABEL: @test(
281 ; CHECK-NEXT: [[I1_0:%.*]] = load x86_fp80, ptr [[I1:%.*]], align 16
282 ; CHECK-NEXT: [[I1_GEP1:%.*]] = getelementptr x86_fp80, ptr [[I1]], i64 1
283 ; CHECK-NEXT: [[I1_1:%.*]] = load x86_fp80, ptr [[I1_GEP1]], align 16
284 ; CHECK-NEXT: br i1 undef, label [[THEN:%.*]], label [[END:%.*]]
286 ; CHECK-NEXT: [[I2_0:%.*]] = load x86_fp80, ptr [[I2:%.*]], align 16
287 ; CHECK-NEXT: [[I2_GEP1:%.*]] = getelementptr inbounds x86_fp80, ptr [[I2]], i64 1
288 ; CHECK-NEXT: [[I2_1:%.*]] = load x86_fp80, ptr [[I2_GEP1]], align 16
289 ; CHECK-NEXT: br label [[END]]
291 ; CHECK-NEXT: [[PHI0:%.*]] = phi x86_fp80 [ [[I1_0]], [[ENTRY:%.*]] ], [ [[I2_0]], [[THEN]] ]
292 ; CHECK-NEXT: [[PHI1:%.*]] = phi x86_fp80 [ [[I1_1]], [[ENTRY]] ], [ [[I2_1]], [[THEN]] ]
293 ; CHECK-NEXT: store x86_fp80 [[PHI0]], ptr [[O:%.*]], align 16
294 ; CHECK-NEXT: [[O_GEP1:%.*]] = getelementptr inbounds x86_fp80, ptr [[O]], i64 1
295 ; CHECK-NEXT: store x86_fp80 [[PHI1]], ptr [[O_GEP1]], align 16
296 ; CHECK-NEXT: ret void
298 ; Test that we correctly recognize the discontiguous memory in arrays where the
299 ; size is less than the alignment, and through various different GEP formations.
300 ; We disable the vectorization of x86_fp80 for now.
303 %i1.0 = load x86_fp80, ptr %i1, align 16
304 %i1.gep1 = getelementptr x86_fp80, ptr %i1, i64 1
305 %i1.1 = load x86_fp80, ptr %i1.gep1, align 16
306 br i1 undef, label %then, label %end
309 %i2.0 = load x86_fp80, ptr %i2, align 16
310 %i2.gep1 = getelementptr inbounds x86_fp80, ptr %i2, i64 1
311 %i2.1 = load x86_fp80, ptr %i2.gep1, align 16
315 %phi0 = phi x86_fp80 [ %i1.0, %entry ], [ %i2.0, %then ]
316 %phi1 = phi x86_fp80 [ %i1.1, %entry ], [ %i2.1, %then ]
317 store x86_fp80 %phi0, ptr %o, align 16
318 %o.gep1 = getelementptr inbounds x86_fp80, ptr %o, i64 1
319 store x86_fp80 %phi1, ptr %o.gep1, align 16