1 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2 ; RUN: opt < %s -passes=slp-vectorizer -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-macosx10.8.0"
14 define i32 @test(ptr nocapture %G) {
17 ; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds double, ptr [[G:%.*]], i64 5
18 ; CHECK-NEXT: [[TMP0:%.*]] = load <2 x double>, ptr [[ARRAYIDX]], align 8
19 ; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <2 x double> [[TMP0]], <2 x double> poison, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
20 ; CHECK-NEXT: [[TMP2:%.*]] = fmul <4 x double> [[TMP1]], <double 4.000000e+00, double 3.000000e+00, double 4.000000e+00, double 4.000000e+00>
21 ; CHECK-NEXT: [[TMP3:%.*]] = fadd <4 x double> [[TMP2]], <double 1.000000e+00, double 6.000000e+00, double 7.000000e+00, double 8.000000e+00>
22 ; CHECK-NEXT: store <4 x double> [[TMP3]], ptr [[G]], align 8
23 ; CHECK-NEXT: ret i32 undef
26 %arrayidx = getelementptr inbounds double, ptr %G, i64 5
27 %0 = load double, ptr %arrayidx, align 8
28 %mul = fmul double %0, 4.000000e+00
29 %add = fadd double %mul, 1.000000e+00
30 store double %add, ptr %G, align 8
31 %arrayidx2 = getelementptr inbounds double, ptr %G, i64 6
32 %1 = load double, ptr %arrayidx2, align 8
33 %mul3 = fmul double %1, 3.000000e+00
34 %add4 = fadd double %mul3, 6.000000e+00
35 %arrayidx5 = getelementptr inbounds double, ptr %G, i64 1
36 store double %add4, ptr %arrayidx5, align 8
37 %add8 = fadd double %mul, 7.000000e+00
38 %arrayidx9 = getelementptr inbounds double, ptr %G, i64 2
39 store double %add8, ptr %arrayidx9, align 8
40 %mul11 = fmul double %1, 4.000000e+00
41 %add12 = fadd double %mul11, 8.000000e+00
42 %arrayidx13 = getelementptr inbounds double, ptr %G, i64 3
43 store double %add12, ptr %arrayidx13, align 8
47 ;int foo(ptr A, int n) {
48 ; A[0] = Aptr 7.9 * n + 6.0;
49 ; A[1] = Aptr 7.7 * n + 2.0;
50 ; A[2] = Aptr 7.6 * n + 3.0;
51 ; A[3] = Aptr 7.4 * n + 4.0;
54 define i32 @foo(ptr nocapture %A, i32 %n) {
57 ; CHECK-NEXT: [[CONV:%.*]] = sitofp i32 [[N:%.*]] to double
58 ; CHECK-NEXT: [[TMP0:%.*]] = load <4 x double>, ptr [[A:%.*]], align 8
59 ; CHECK-NEXT: [[TMP1:%.*]] = fmul <4 x double> [[TMP0]], <double 7.900000e+00, double 7.700000e+00, double 7.600000e+00, double 7.400000e+00>
60 ; CHECK-NEXT: [[TMP2:%.*]] = insertelement <4 x double> poison, double [[CONV]], i32 0
61 ; CHECK-NEXT: [[TMP3:%.*]] = shufflevector <4 x double> [[TMP2]], <4 x double> poison, <4 x i32> zeroinitializer
62 ; CHECK-NEXT: [[TMP4:%.*]] = fmul <4 x double> [[TMP3]], [[TMP1]]
63 ; CHECK-NEXT: [[TMP5:%.*]] = fadd <4 x double> [[TMP4]], <double 6.000000e+00, double 2.000000e+00, double 3.000000e+00, double 4.000000e+00>
64 ; CHECK-NEXT: store <4 x double> [[TMP5]], ptr [[A]], align 8
65 ; CHECK-NEXT: ret i32 undef
68 %0 = load double, ptr %A, align 8
69 %mul = fmul double %0, 7.900000e+00
70 %conv = sitofp i32 %n to double
71 %mul1 = fmul double %conv, %mul
72 %add = fadd double %mul1, 6.000000e+00
73 store double %add, ptr %A, align 8
74 %arrayidx3 = getelementptr inbounds double, ptr %A, i64 1
75 %1 = load double, ptr %arrayidx3, align 8
76 %mul4 = fmul double %1, 7.700000e+00
77 %mul6 = fmul double %conv, %mul4
78 %add7 = fadd double %mul6, 2.000000e+00
79 store double %add7, ptr %arrayidx3, align 8
80 %arrayidx9 = getelementptr inbounds double, ptr %A, i64 2
81 %2 = load double, ptr %arrayidx9, align 8
82 %mul10 = fmul double %2, 7.600000e+00
83 %mul12 = fmul double %conv, %mul10
84 %add13 = fadd double %mul12, 3.000000e+00
85 store double %add13, ptr %arrayidx9, align 8
86 %arrayidx15 = getelementptr inbounds double, ptr %A, i64 3
87 %3 = load double, ptr %arrayidx15, align 8
88 %mul16 = fmul double %3, 7.400000e+00
89 %mul18 = fmul double %conv, %mul16
90 %add19 = fadd double %mul18, 4.000000e+00
91 store double %add19, ptr %arrayidx15, align 8
95 ; int test2(ptr G, int k) {
105 ; We can't merge the gather sequences because one does not dominate the other.
107 define i32 @test2(ptr nocapture %G, i32 %k) {
108 ; CHECK-LABEL: @test2(
109 ; CHECK-NEXT: [[TMP1:%.*]] = icmp eq i32 [[K:%.*]], 0
110 ; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds double, ptr [[G:%.*]], i64 5
111 ; CHECK-NEXT: [[TMP3:%.*]] = load double, ptr [[TMP2]], align 8
112 ; CHECK-NEXT: [[TMP4:%.*]] = fmul double [[TMP3]], 4.000000e+00
113 ; CHECK-NEXT: br i1 [[TMP1]], label [[TMP12:%.*]], label [[TMP5:%.*]]
115 ; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds double, ptr [[G]], i64 6
116 ; CHECK-NEXT: [[TMP7:%.*]] = load double, ptr [[TMP6]], align 8
117 ; CHECK-NEXT: [[TMP8:%.*]] = fmul double [[TMP7]], 3.000000e+00
118 ; CHECK-NEXT: [[TMP9:%.*]] = insertelement <2 x double> poison, double [[TMP4]], i32 0
119 ; CHECK-NEXT: [[TMP10:%.*]] = insertelement <2 x double> [[TMP9]], double [[TMP8]], i32 1
120 ; CHECK-NEXT: [[TMP11:%.*]] = fadd <2 x double> [[TMP10]], <double 1.000000e+00, double 6.000000e+00>
121 ; CHECK-NEXT: store <2 x double> [[TMP11]], ptr [[G]], align 8
122 ; CHECK-NEXT: br label [[TMP20:%.*]]
124 ; CHECK-NEXT: [[TMP13:%.*]] = getelementptr inbounds double, ptr [[G]], i64 2
125 ; CHECK-NEXT: [[TMP14:%.*]] = getelementptr inbounds double, ptr [[G]], i64 6
126 ; CHECK-NEXT: [[TMP15:%.*]] = load double, ptr [[TMP14]], align 8
127 ; CHECK-NEXT: [[TMP16:%.*]] = fmul double [[TMP15]], 3.000000e+00
128 ; CHECK-NEXT: [[TMP17:%.*]] = insertelement <2 x double> poison, double [[TMP4]], i32 0
129 ; CHECK-NEXT: [[TMP18:%.*]] = insertelement <2 x double> [[TMP17]], double [[TMP16]], i32 1
130 ; CHECK-NEXT: [[TMP19:%.*]] = fadd <2 x double> [[TMP18]], <double 7.000000e+00, double 8.000000e+00>
131 ; CHECK-NEXT: store <2 x double> [[TMP19]], ptr [[TMP13]], align 8
132 ; CHECK-NEXT: br label [[TMP20]]
134 ; CHECK-NEXT: ret i32 undef
136 %1 = icmp eq i32 %k, 0
137 %2 = getelementptr inbounds double, ptr %G, i64 5
138 %3 = load double, ptr %2, align 8
139 %4 = fmul double %3, 4.000000e+00
140 br i1 %1, label %12, label %5
142 ; <label>:5 ; preds = %0
143 %6 = fadd double %4, 1.000000e+00
144 store double %6, ptr %G, align 8
145 %7 = getelementptr inbounds double, ptr %G, i64 6
146 %8 = load double, ptr %7, align 8
147 %9 = fmul double %8, 3.000000e+00
148 %10 = fadd double %9, 6.000000e+00
149 %11 = getelementptr inbounds double, ptr %G, i64 1
150 store double %10, ptr %11, align 8
153 ; <label>:12 ; preds = %0
154 %13 = fadd double %4, 7.000000e+00
155 %14 = getelementptr inbounds double, ptr %G, i64 2
156 store double %13, ptr %14, align 8
157 %15 = getelementptr inbounds double, ptr %G, i64 6
158 %16 = load double, ptr %15, align 8
159 %17 = fmul double %16, 3.000000e+00
160 %18 = fadd double %17, 8.000000e+00
161 %19 = getelementptr inbounds double, ptr %G, i64 3
162 store double %18, ptr %19, align 8
165 ; <label>:20 ; preds = %12, %5
170 ;int foo(ptr A, int n) {
171 ; A[0] = Aptr 7.9 * n + 6.0;
172 ; A[1] = Aptr 7.9 * n + 6.0;
173 ; A[2] = Aptr 7.9 * n + 6.0;
174 ; A[3] = Aptr 7.9 * n + 6.0;
177 define i32 @foo4(ptr nocapture %A, i32 %n) {
178 ; CHECK-LABEL: @foo4(
180 ; CHECK-NEXT: [[CONV:%.*]] = sitofp i32 [[N:%.*]] to double
181 ; CHECK-NEXT: [[TMP0:%.*]] = load <4 x double>, ptr [[A:%.*]], align 8
182 ; CHECK-NEXT: [[TMP1:%.*]] = fmul <4 x double> [[TMP0]], <double 7.900000e+00, double 7.900000e+00, double 7.900000e+00, double 7.900000e+00>
183 ; CHECK-NEXT: [[TMP2:%.*]] = insertelement <4 x double> poison, double [[CONV]], i32 0
184 ; CHECK-NEXT: [[TMP3:%.*]] = shufflevector <4 x double> [[TMP2]], <4 x double> poison, <4 x i32> zeroinitializer
185 ; CHECK-NEXT: [[TMP4:%.*]] = fmul <4 x double> [[TMP3]], [[TMP1]]
186 ; CHECK-NEXT: [[TMP5:%.*]] = fadd <4 x double> [[TMP4]], <double 6.000000e+00, double 6.000000e+00, double 6.000000e+00, double 6.000000e+00>
187 ; CHECK-NEXT: store <4 x double> [[TMP5]], ptr [[A]], align 8
188 ; CHECK-NEXT: ret i32 undef
191 %0 = load double, ptr %A, align 8
192 %mul = fmul double %0, 7.900000e+00
193 %conv = sitofp i32 %n to double
194 %mul1 = fmul double %conv, %mul
195 %add = fadd double %mul1, 6.000000e+00
196 store double %add, ptr %A, align 8
197 %arrayidx3 = getelementptr inbounds double, ptr %A, i64 1
198 %1 = load double, ptr %arrayidx3, align 8
199 %mul4 = fmul double %1, 7.900000e+00
200 %mul6 = fmul double %conv, %mul4
201 %add7 = fadd double %mul6, 6.000000e+00
202 store double %add7, ptr %arrayidx3, align 8
203 %arrayidx9 = getelementptr inbounds double, ptr %A, i64 2
204 %2 = load double, ptr %arrayidx9, align 8
205 %mul10 = fmul double %2, 7.900000e+00
206 %mul12 = fmul double %conv, %mul10
207 %add13 = fadd double %mul12, 6.000000e+00
208 store double %add13, ptr %arrayidx9, align 8
209 %arrayidx15 = getelementptr inbounds double, ptr %A, i64 3
210 %3 = load double, ptr %arrayidx15, align 8
211 %mul16 = fmul double %3, 7.900000e+00
212 %mul18 = fmul double %conv, %mul16
213 %add19 = fadd double %mul18, 6.000000e+00
214 store double %add19, ptr %arrayidx15, align 8
218 ;int partial_mrg(ptr A, int n) {
221 ; if (n < 4) return 0;
226 define i32 @partial_mrg(ptr nocapture %A, i32 %n) {
227 ; CHECK-LABEL: @partial_mrg(
229 ; CHECK-NEXT: [[CONV:%.*]] = sitofp i32 [[N:%.*]] to double
230 ; CHECK-NEXT: [[TMP0:%.*]] = load <2 x double>, ptr [[A:%.*]], align 8
231 ; CHECK-NEXT: [[TMP1:%.*]] = insertelement <2 x double> poison, double [[CONV]], i32 0
232 ; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <2 x double> [[TMP1]], <2 x double> poison, <2 x i32> zeroinitializer
233 ; CHECK-NEXT: [[TMP3:%.*]] = fmul <2 x double> [[TMP2]], [[TMP0]]
234 ; CHECK-NEXT: store <2 x double> [[TMP3]], ptr [[A]], align 8
235 ; CHECK-NEXT: [[CMP:%.*]] = icmp slt i32 [[N]], 4
236 ; CHECK-NEXT: br i1 [[CMP]], label [[RETURN:%.*]], label [[IF_END:%.*]]
238 ; CHECK-NEXT: [[ARRAYIDX7:%.*]] = getelementptr inbounds double, ptr [[A]], i64 2
239 ; CHECK-NEXT: [[ADD:%.*]] = add nsw i32 [[N]], 4
240 ; CHECK-NEXT: [[CONV12:%.*]] = sitofp i32 [[ADD]] to double
241 ; CHECK-NEXT: [[TMP4:%.*]] = load <2 x double>, ptr [[ARRAYIDX7]], align 8
242 ; CHECK-NEXT: [[TMP5:%.*]] = insertelement <2 x double> [[TMP1]], double [[CONV12]], i32 1
243 ; CHECK-NEXT: [[TMP6:%.*]] = fmul <2 x double> [[TMP5]], [[TMP4]]
244 ; CHECK-NEXT: store <2 x double> [[TMP6]], ptr [[ARRAYIDX7]], align 8
245 ; CHECK-NEXT: br label [[RETURN]]
247 ; CHECK-NEXT: ret i32 0
250 %0 = load double, ptr %A, align 8
251 %conv = sitofp i32 %n to double
252 %mul = fmul double %conv, %0
253 store double %mul, ptr %A, align 8
254 %arrayidx2 = getelementptr inbounds double, ptr %A, i64 1
255 %1 = load double, ptr %arrayidx2, align 8
256 %mul4 = fmul double %conv, %1
257 store double %mul4, ptr %arrayidx2, align 8
258 %cmp = icmp slt i32 %n, 4
259 br i1 %cmp, label %return, label %if.end
261 if.end: ; preds = %entry
262 %arrayidx7 = getelementptr inbounds double, ptr %A, i64 2
263 %2 = load double, ptr %arrayidx7, align 8
264 %mul9 = fmul double %conv, %2
265 store double %mul9, ptr %arrayidx7, align 8
266 %arrayidx11 = getelementptr inbounds double, ptr %A, i64 3
267 %3 = load double, ptr %arrayidx11, align 8
268 %add = add nsw i32 %n, 4
269 %conv12 = sitofp i32 %add to double
270 %mul13 = fmul double %conv12, %3
271 store double %mul13, ptr %arrayidx11, align 8
274 return: ; preds = %entry, %if.end
278 %class.B.53.55 = type { %class.A.52.54, double }
279 %class.A.52.54 = type { double, double, double }
281 @a = external global double, align 8
283 define void @PR19646(ptr %this) {
284 ; CHECK-LABEL: @PR19646(
286 ; CHECK-NEXT: br i1 undef, label [[IF_END13:%.*]], label [[IF_END13]]
288 ; CHECK-NEXT: [[DOTIN:%.*]] = getelementptr inbounds [[CLASS_B_53_55:%.*]], ptr [[THIS:%.*]], i64 0, i32 0, i32 1
289 ; CHECK-NEXT: [[TMP0:%.*]] = load double, ptr [[DOTIN]], align 8
290 ; CHECK-NEXT: [[ADD:%.*]] = fadd double undef, 0.000000e+00
291 ; CHECK-NEXT: [[ADD6:%.*]] = fadd double [[ADD]], [[TMP0]]
292 ; CHECK-NEXT: [[TMP1:%.*]] = load double, ptr @a, align 8
293 ; CHECK-NEXT: [[ADD8:%.*]] = fadd double [[TMP1]], 0.000000e+00
294 ; CHECK-NEXT: [[_DY:%.*]] = getelementptr inbounds [[CLASS_B_53_55]], ptr [[THIS]], i64 0, i32 0, i32 2
295 ; CHECK-NEXT: [[TMP2:%.*]] = load double, ptr [[_DY]], align 8
296 ; CHECK-NEXT: [[ADD10:%.*]] = fadd double [[ADD8]], [[TMP2]]
297 ; CHECK-NEXT: br i1 undef, label [[IF_THEN12:%.*]], label [[IF_END13]]
299 ; CHECK-NEXT: [[TMP3:%.*]] = load double, ptr undef, align 8
300 ; CHECK-NEXT: br label [[IF_END13]]
302 ; CHECK-NEXT: [[X_1:%.*]] = phi double [ 0.000000e+00, [[IF_THEN12]] ], [ [[ADD6]], [[SW_EPILOG7:%.*]] ], [ undef, [[ENTRY:%.*]] ], [ undef, [[ENTRY]] ]
303 ; CHECK-NEXT: [[B_0:%.*]] = phi double [ [[TMP3]], [[IF_THEN12]] ], [ [[ADD10]], [[SW_EPILOG7]] ], [ undef, [[ENTRY]] ], [ undef, [[ENTRY]] ]
304 ; CHECK-NEXT: unreachable
307 br i1 undef, label %if.end13, label %if.end13
309 sw.epilog7: ; No predecessors!
310 %.in = getelementptr inbounds %class.B.53.55, ptr %this, i64 0, i32 0, i32 1
311 %0 = load double, ptr %.in, align 8
312 %add = fadd double undef, 0.000000e+00
313 %add6 = fadd double %add, %0
314 %1 = load double, ptr @a, align 8
315 %add8 = fadd double %1, 0.000000e+00
316 %_dy = getelementptr inbounds %class.B.53.55, ptr %this, i64 0, i32 0, i32 2
317 %2 = load double, ptr %_dy, align 8
318 %add10 = fadd double %add8, %2
319 br i1 undef, label %if.then12, label %if.end13
321 if.then12: ; preds = %sw.epilog7
322 %3 = load double, ptr undef, align 8
325 if.end13: ; preds = %if.then12, %sw.epilog7, %entry
326 %x.1 = phi double [ 0.000000e+00, %if.then12 ], [ %add6, %sw.epilog7 ], [ undef, %entry ], [ undef, %entry ]
327 %b.0 = phi double [ %3, %if.then12 ], [ %add10, %sw.epilog7 ], [ undef, %entry], [ undef, %entry ]
331 define void @cse_for_hoisted_instructions_in_preheader(ptr %dst, i32 %a, i1 %c) {
332 ; CHECK-LABEL: @cse_for_hoisted_instructions_in_preheader(
334 ; CHECK-NEXT: [[TMP0:%.*]] = insertelement <2 x i32> poison, i32 [[A:%.*]], i32 0
335 ; CHECK-NEXT: [[TMP1:%.*]] = shufflevector <2 x i32> [[TMP0]], <2 x i32> poison, <2 x i32> zeroinitializer
336 ; CHECK-NEXT: br label [[LOOP:%.*]]
338 ; CHECK-NEXT: [[TMP2:%.*]] = or <2 x i32> <i32 22, i32 22>, [[TMP1]]
339 ; CHECK-NEXT: [[TMP3:%.*]] = or <2 x i32> [[TMP2]], <i32 3, i32 3>
340 ; CHECK-NEXT: store <2 x i32> [[TMP3]], ptr [[DST:%.*]], align 4
341 ; CHECK-NEXT: [[OR_2:%.*]] = or i32 [[A]], 3
342 ; CHECK-NEXT: [[GEP_2:%.*]] = getelementptr inbounds i32, ptr [[DST]], i64 10
343 ; CHECK-NEXT: store i32 [[OR_2]], ptr [[GEP_2]], align 4
344 ; CHECK-NEXT: [[OR_3:%.*]] = or i32 [[A]], 3
345 ; CHECK-NEXT: [[GEP_3:%.*]] = getelementptr inbounds i32, ptr [[DST]], i64 11
346 ; CHECK-NEXT: store i32 [[OR_3]], ptr [[GEP_3]], align 4
347 ; CHECK-NEXT: br i1 [[C:%.*]], label [[LOOP]], label [[EXIT:%.*]]
349 ; CHECK-NEXT: ret void
355 %or.a = or i32 22, %a
356 %or.0 = or i32 %or.a, 3
357 store i32 %or.0, ptr %dst
358 %or.a.2 = or i32 22, %a
359 %or.1 = or i32 %or.a.2, 3
360 %gep.1 = getelementptr inbounds i32, ptr %dst, i64 1
361 store i32 %or.1, ptr %gep.1
363 %gep.2 = getelementptr inbounds i32, ptr %dst, i64 10
364 store i32 %or.2, ptr %gep.2
366 %gep.3 = getelementptr inbounds i32, ptr %dst, i64 11
367 store i32 %or.3, ptr %gep.3
368 br i1 %c, label %loop, label %exit