1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt -O2                   -S -mattr=avx < %s | FileCheck %s
3; RUN: opt -passes='default<O2>' -S -mattr=avx < %s | FileCheck %s
4
5target triple = "x86_64--"
6target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
7
8define i32 @ext_ext_or_reduction_v4i32(<4 x i32> %x, <4 x i32> %y) {
9; CHECK-LABEL: @ext_ext_or_reduction_v4i32(
10; CHECK-NEXT:    [[Z:%.*]] = and <4 x i32> [[Y:%.*]], [[X:%.*]]
11; CHECK-NEXT:    [[TMP1:%.*]] = call i32 @llvm.vector.reduce.or.v4i32(<4 x i32> [[Z]])
12; CHECK-NEXT:    ret i32 [[TMP1]]
13;
14  %z = and <4 x i32> %x, %y
15  %z0 = extractelement <4 x i32> %z, i32 0
16  %z1 = extractelement <4 x i32> %z, i32 1
17  %z01 = or i32 %z0, %z1
18  %z2 = extractelement <4 x i32> %z, i32 2
19  %z012 = or i32 %z01, %z2
20  %z3 = extractelement <4 x i32> %z, i32 3
21  %z0123 = or i32 %z3, %z012
22  ret i32 %z0123
23}
24
25define i32 @ext_ext_partial_add_reduction_v4i32(<4 x i32> %x) {
26; CHECK-LABEL: @ext_ext_partial_add_reduction_v4i32(
27; CHECK-NEXT:    [[SHIFT:%.*]] = shufflevector <4 x i32> [[X:%.*]], <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
28; CHECK-NEXT:    [[TMP1:%.*]] = add <4 x i32> [[SHIFT]], [[X]]
29; CHECK-NEXT:    [[SHIFT1:%.*]] = shufflevector <4 x i32> [[X]], <4 x i32> undef, <4 x i32> <i32 2, i32 undef, i32 undef, i32 undef>
30; CHECK-NEXT:    [[TMP2:%.*]] = add <4 x i32> [[TMP1]], [[SHIFT1]]
31; CHECK-NEXT:    [[X210:%.*]] = extractelement <4 x i32> [[TMP2]], i64 0
32; CHECK-NEXT:    ret i32 [[X210]]
33;
34  %x0 = extractelement <4 x i32> %x, i32 0
35  %x1 = extractelement <4 x i32> %x, i32 1
36  %x10 = add i32 %x1, %x0
37  %x2 = extractelement <4 x i32> %x, i32 2
38  %x210 = add i32 %x2, %x10
39  ret i32 %x210
40}
41
42define i32 @ext_ext_partial_add_reduction_and_extra_add_v4i32(<4 x i32> %x, <4 x i32> %y) {
43; CHECK-LABEL: @ext_ext_partial_add_reduction_and_extra_add_v4i32(
44; CHECK-NEXT:    [[SHIFT:%.*]] = shufflevector <4 x i32> [[X:%.*]], <4 x i32> undef, <4 x i32> <i32 2, i32 undef, i32 undef, i32 undef>
45; CHECK-NEXT:    [[TMP1:%.*]] = add <4 x i32> [[SHIFT]], [[Y:%.*]]
46; CHECK-NEXT:    [[SHIFT1:%.*]] = shufflevector <4 x i32> [[Y]], <4 x i32> undef, <4 x i32> <i32 1, i32 undef, i32 undef, i32 undef>
47; CHECK-NEXT:    [[TMP2:%.*]] = add <4 x i32> [[TMP1]], [[SHIFT1]]
48; CHECK-NEXT:    [[SHIFT2:%.*]] = shufflevector <4 x i32> [[Y]], <4 x i32> undef, <4 x i32> <i32 2, i32 undef, i32 undef, i32 undef>
49; CHECK-NEXT:    [[TMP3:%.*]] = add <4 x i32> [[TMP2]], [[SHIFT2]]
50; CHECK-NEXT:    [[X2Y210:%.*]] = extractelement <4 x i32> [[TMP3]], i32 0
51; CHECK-NEXT:    ret i32 [[X2Y210]]
52;
53  %y0 = extractelement <4 x i32> %y, i32 0
54  %y1 = extractelement <4 x i32> %y, i32 1
55  %y10 = add i32 %y1, %y0
56  %y2 = extractelement <4 x i32> %y, i32 2
57  %y210 = add i32 %y2, %y10
58  %x2 = extractelement <4 x i32> %x, i32 2
59  %x2y210 = add i32 %x2, %y210
60  ret i32 %x2y210
61}
62
63; PR43953 - https://bugs.llvm.org/show_bug.cgi?id=43953
64; We want to end up with a single reduction on the next 4 tests.
65
66define i32 @TestVectorsEqual(i32* noalias %Vec0, i32* noalias %Vec1, i32 %Tolerance) {
67; CHECK-LABEL: @TestVectorsEqual(
68; CHECK-NEXT:  entry:
69; CHECK-NEXT:    [[TMP0:%.*]] = bitcast i32* [[VEC0:%.*]] to <4 x i32>*
70; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x i32>, <4 x i32>* [[TMP0]], align 4
71; CHECK-NEXT:    [[TMP2:%.*]] = bitcast i32* [[VEC1:%.*]] to <4 x i32>*
72; CHECK-NEXT:    [[TMP3:%.*]] = load <4 x i32>, <4 x i32>* [[TMP2]], align 4
73; CHECK-NEXT:    [[TMP4:%.*]] = sub nsw <4 x i32> [[TMP1]], [[TMP3]]
74; CHECK-NEXT:    [[TMP5:%.*]] = icmp slt <4 x i32> [[TMP4]], zeroinitializer
75; CHECK-NEXT:    [[TMP6:%.*]] = sub nsw <4 x i32> zeroinitializer, [[TMP4]]
76; CHECK-NEXT:    [[TMP7:%.*]] = select <4 x i1> [[TMP5]], <4 x i32> [[TMP6]], <4 x i32> [[TMP4]]
77; CHECK-NEXT:    [[TMP8:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[TMP7]])
78; CHECK-NEXT:    [[CMP5_NOT:%.*]] = icmp sle i32 [[TMP8]], [[TOLERANCE:%.*]]
79; CHECK-NEXT:    [[COND6:%.*]] = zext i1 [[CMP5_NOT]] to i32
80; CHECK-NEXT:    ret i32 [[COND6]]
81;
82entry:
83  br label %for.cond
84
85for.cond:
86  %sum.0 = phi i32 [ 0, %entry ], [ %add, %for.inc ]
87  %Component.0 = phi i32 [ 0, %entry ], [ %inc, %for.inc ]
88  %cmp = icmp slt i32 %Component.0, 4
89  br i1 %cmp, label %for.body, label %for.cond.cleanup
90
91for.cond.cleanup:
92  br label %for.end
93
94for.body:
95  %idxprom = sext i32 %Component.0 to i64
96  %arrayidx = getelementptr inbounds i32, i32* %Vec0, i64 %idxprom
97  %0 = load i32, i32* %arrayidx, align 4
98  %idxprom1 = sext i32 %Component.0 to i64
99  %arrayidx2 = getelementptr inbounds i32, i32* %Vec1, i64 %idxprom1
100  %1 = load i32, i32* %arrayidx2, align 4
101  %sub = sub nsw i32 %0, %1
102  %cmp3 = icmp sge i32 %sub, 0
103  br i1 %cmp3, label %cond.true, label %cond.false
104
105cond.true:
106  br label %cond.end
107
108cond.false:
109  %sub4 = sub nsw i32 0, %sub
110  br label %cond.end
111
112cond.end:
113  %cond = phi i32 [ %sub, %cond.true ], [ %sub4, %cond.false ]
114  %add = add nsw i32 %sum.0, %cond
115  br label %for.inc
116
117for.inc:
118  %inc = add nsw i32 %Component.0, 1
119  br label %for.cond
120
121for.end:
122  %cmp5 = icmp sle i32 %sum.0, %Tolerance
123  %2 = zext i1 %cmp5 to i64
124  %cond6 = select i1 %cmp5, i32 1, i32 0
125  ret i32 %cond6
126}
127
128define i32 @TestVectorsEqual_alt(i32* noalias %Vec0, i32* noalias %Vec1, i32 %Tolerance) {
129; CHECK-LABEL: @TestVectorsEqual_alt(
130; CHECK-NEXT:  entry:
131; CHECK-NEXT:    [[TMP0:%.*]] = bitcast i32* [[VEC0:%.*]] to <4 x i32>*
132; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x i32>, <4 x i32>* [[TMP0]], align 4
133; CHECK-NEXT:    [[TMP2:%.*]] = bitcast i32* [[VEC1:%.*]] to <4 x i32>*
134; CHECK-NEXT:    [[TMP3:%.*]] = load <4 x i32>, <4 x i32>* [[TMP2]], align 4
135; CHECK-NEXT:    [[TMP4:%.*]] = sub <4 x i32> [[TMP1]], [[TMP3]]
136; CHECK-NEXT:    [[TMP5:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[TMP4]])
137; CHECK-NEXT:    [[CMP3_NOT:%.*]] = icmp ule i32 [[TMP5]], [[TOLERANCE:%.*]]
138; CHECK-NEXT:    [[COND:%.*]] = zext i1 [[CMP3_NOT]] to i32
139; CHECK-NEXT:    ret i32 [[COND]]
140;
141entry:
142  br label %for.cond
143
144for.cond:
145  %sum.0 = phi i32 [ 0, %entry ], [ %add, %for.inc ]
146  %Component.0 = phi i32 [ 0, %entry ], [ %inc, %for.inc ]
147  %cmp = icmp slt i32 %Component.0, 4
148  br i1 %cmp, label %for.body, label %for.cond.cleanup
149
150for.cond.cleanup:
151  br label %for.end
152
153for.body:
154  %idxprom = sext i32 %Component.0 to i64
155  %arrayidx = getelementptr inbounds i32, i32* %Vec0, i64 %idxprom
156  %0 = load i32, i32* %arrayidx, align 4
157  %idxprom1 = sext i32 %Component.0 to i64
158  %arrayidx2 = getelementptr inbounds i32, i32* %Vec1, i64 %idxprom1
159  %1 = load i32, i32* %arrayidx2, align 4
160  %sub = sub i32 %0, %1
161  %add = add i32 %sum.0, %sub
162  br label %for.inc
163
164for.inc:
165  %inc = add nsw i32 %Component.0, 1
166  br label %for.cond
167
168for.end:
169  %cmp3 = icmp ule i32 %sum.0, %Tolerance
170  %2 = zext i1 %cmp3 to i64
171  %cond = select i1 %cmp3, i32 1, i32 0
172  ret i32 %cond
173}
174
175define i32 @TestVectorsEqualFP(float* noalias %Vec0, float* noalias %Vec1, float %Tolerance) {
176; CHECK-LABEL: @TestVectorsEqualFP(
177; CHECK-NEXT:  entry:
178; CHECK-NEXT:    [[TMP0:%.*]] = bitcast float* [[VEC0:%.*]] to <4 x float>*
179; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x float>, <4 x float>* [[TMP0]], align 4
180; CHECK-NEXT:    [[TMP2:%.*]] = bitcast float* [[VEC1:%.*]] to <4 x float>*
181; CHECK-NEXT:    [[TMP3:%.*]] = load <4 x float>, <4 x float>* [[TMP2]], align 4
182; CHECK-NEXT:    [[TMP4:%.*]] = fsub fast <4 x float> [[TMP1]], [[TMP3]]
183; CHECK-NEXT:    [[TMP5:%.*]] = call fast <4 x float> @llvm.fabs.v4f32(<4 x float> [[TMP4]])
184; CHECK-NEXT:    [[TMP6:%.*]] = call fast float @llvm.vector.reduce.fadd.v4f32(float -0.000000e+00, <4 x float> [[TMP5]])
185; CHECK-NEXT:    [[CMP4:%.*]] = fcmp fast ole float [[TMP6]], [[TOLERANCE:%.*]]
186; CHECK-NEXT:    [[COND5:%.*]] = zext i1 [[CMP4]] to i32
187; CHECK-NEXT:    ret i32 [[COND5]]
188;
189entry:
190  br label %for.cond
191
192for.cond:
193  %sum.0 = phi float [ 0.000000e+00, %entry ], [ %add, %for.inc ]
194  %Component.0 = phi i32 [ 0, %entry ], [ %inc, %for.inc ]
195  %cmp = icmp slt i32 %Component.0, 4
196  br i1 %cmp, label %for.body, label %for.cond.cleanup
197
198for.cond.cleanup:
199  br label %for.end
200
201for.body:
202  %idxprom = sext i32 %Component.0 to i64
203  %arrayidx = getelementptr inbounds float, float* %Vec0, i64 %idxprom
204  %0 = load float, float* %arrayidx, align 4
205  %idxprom1 = sext i32 %Component.0 to i64
206  %arrayidx2 = getelementptr inbounds float, float* %Vec1, i64 %idxprom1
207  %1 = load float, float* %arrayidx2, align 4
208  %sub = fsub fast float %0, %1
209  %cmp3 = fcmp fast oge float %sub, 0.000000e+00
210  br i1 %cmp3, label %cond.true, label %cond.false
211
212cond.true:
213  br label %cond.end
214
215cond.false:
216  %fneg = fneg fast float %sub
217  br label %cond.end
218
219cond.end:
220  %cond = phi fast float [ %sub, %cond.true ], [ %fneg, %cond.false ]
221  %add = fadd fast float %sum.0, %cond
222  br label %for.inc
223
224for.inc:
225  %inc = add nsw i32 %Component.0, 1
226  br label %for.cond
227
228for.end:
229  %cmp4 = fcmp fast ole float %sum.0, %Tolerance
230  %2 = zext i1 %cmp4 to i64
231  %cond5 = select i1 %cmp4, i32 1, i32 0
232  ret i32 %cond5
233}
234
235define i32 @TestVectorsEqualFP_alt(float* noalias %Vec0, float* noalias %Vec1, float %Tolerance) {
236; CHECK-LABEL: @TestVectorsEqualFP_alt(
237; CHECK-NEXT:  entry:
238; CHECK-NEXT:    [[TMP0:%.*]] = bitcast float* [[VEC0:%.*]] to <4 x float>*
239; CHECK-NEXT:    [[TMP1:%.*]] = load <4 x float>, <4 x float>* [[TMP0]], align 4
240; CHECK-NEXT:    [[TMP2:%.*]] = bitcast float* [[VEC1:%.*]] to <4 x float>*
241; CHECK-NEXT:    [[TMP3:%.*]] = load <4 x float>, <4 x float>* [[TMP2]], align 4
242; CHECK-NEXT:    [[TMP4:%.*]] = fsub fast <4 x float> [[TMP1]], [[TMP3]]
243; CHECK-NEXT:    [[TMP5:%.*]] = call fast float @llvm.vector.reduce.fadd.v4f32(float -0.000000e+00, <4 x float> [[TMP4]])
244; CHECK-NEXT:    [[CMP3:%.*]] = fcmp fast ole float [[TMP5]], [[TOLERANCE:%.*]]
245; CHECK-NEXT:    [[COND:%.*]] = zext i1 [[CMP3]] to i32
246; CHECK-NEXT:    ret i32 [[COND]]
247;
248entry:
249  br label %for.cond
250
251for.cond:
252  %sum.0 = phi float [ 0.000000e+00, %entry ], [ %add, %for.inc ]
253  %Component.0 = phi i32 [ 0, %entry ], [ %inc, %for.inc ]
254  %cmp = icmp slt i32 %Component.0, 4
255  br i1 %cmp, label %for.body, label %for.cond.cleanup
256
257for.cond.cleanup:
258  br label %for.end
259
260for.body:
261  %idxprom = sext i32 %Component.0 to i64
262  %arrayidx = getelementptr inbounds float, float* %Vec0, i64 %idxprom
263  %0 = load float, float* %arrayidx, align 4
264  %idxprom1 = sext i32 %Component.0 to i64
265  %arrayidx2 = getelementptr inbounds float, float* %Vec1, i64 %idxprom1
266  %1 = load float, float* %arrayidx2, align 4
267  %sub = fsub fast float %0, %1
268  %add = fadd fast float %sum.0, %sub
269  br label %for.inc
270
271for.inc:
272  %inc = add nsw i32 %Component.0, 1
273  br label %for.cond
274
275for.end:
276  %cmp3 = fcmp fast ole float %sum.0, %Tolerance
277  %2 = zext i1 %cmp3 to i64
278  %cond = select i1 %cmp3, i32 1, i32 0
279  ret i32 %cond
280}
281
282; PR43745 - https://bugs.llvm.org/show_bug.cgi?id=43745
283
284define i1 @cmp_lt_gt(double %a, double %b, double %c) {
285; CHECK-LABEL: @cmp_lt_gt(
286; CHECK-NEXT:  entry:
287; CHECK-NEXT:    [[FNEG:%.*]] = fneg double [[B:%.*]]
288; CHECK-NEXT:    [[MUL:%.*]] = fmul double [[A:%.*]], 2.000000e+00
289; CHECK-NEXT:    [[TMP0:%.*]] = insertelement <2 x double> undef, double [[C:%.*]], i32 0
290; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <2 x double> [[TMP0]], double [[FNEG]], i32 1
291; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <2 x double> undef, double [[B]], i32 0
292; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <2 x double> [[TMP2]], double [[C]], i32 1
293; CHECK-NEXT:    [[TMP4:%.*]] = fsub <2 x double> [[TMP1]], [[TMP3]]
294; CHECK-NEXT:    [[TMP5:%.*]] = insertelement <2 x double> undef, double [[MUL]], i32 0
295; CHECK-NEXT:    [[TMP6:%.*]] = shufflevector <2 x double> [[TMP5]], <2 x double> undef, <2 x i32> zeroinitializer
296; CHECK-NEXT:    [[TMP7:%.*]] = fdiv <2 x double> [[TMP4]], [[TMP6]]
297; CHECK-NEXT:    [[TMP8:%.*]] = fcmp olt <2 x double> [[TMP7]], <double 0x3EB0C6F7A0B5ED8D, double 0x3EB0C6F7A0B5ED8D>
298; CHECK-NEXT:    [[SHIFT:%.*]] = shufflevector <2 x i1> [[TMP8]], <2 x i1> undef, <2 x i32> <i32 1, i32 undef>
299; CHECK-NEXT:    [[TMP9:%.*]] = and <2 x i1> [[TMP8]], [[SHIFT]]
300; CHECK-NEXT:    [[OR_COND:%.*]] = extractelement <2 x i1> [[TMP9]], i64 0
301; CHECK-NEXT:    br i1 [[OR_COND]], label [[CLEANUP:%.*]], label [[LOR_LHS_FALSE:%.*]]
302; CHECK:       lor.lhs.false:
303; CHECK-NEXT:    [[TMP10:%.*]] = fcmp ule <2 x double> [[TMP7]], <double 1.000000e+00, double 1.000000e+00>
304; CHECK-NEXT:    [[SHIFT2:%.*]] = shufflevector <2 x i1> [[TMP10]], <2 x i1> undef, <2 x i32> <i32 1, i32 undef>
305; CHECK-NEXT:    [[TMP11:%.*]] = or <2 x i1> [[TMP10]], [[SHIFT2]]
306; CHECK-NEXT:    [[NOT_OR_COND1:%.*]] = extractelement <2 x i1> [[TMP11]], i32 0
307; CHECK-NEXT:    ret i1 [[NOT_OR_COND1]]
308; CHECK:       cleanup:
309; CHECK-NEXT:    ret i1 false
310;
311entry:
312  %fneg = fneg double %b
313  %add = fadd double %fneg, %c
314  %mul = fmul double 2.0, %a
315  %div = fdiv double %add, %mul
316  %fneg1 = fneg double %b
317  %sub = fsub double %fneg1, %c
318  %mul2 = fmul double 2.0, %a
319  %div3 = fdiv double %sub, %mul2
320  %cmp = fcmp olt double %div, 0x3EB0C6F7A0B5ED8D
321  br i1 %cmp, label %land.lhs.true, label %lor.lhs.false
322
323land.lhs.true:
324  %cmp4 = fcmp olt double %div3, 0x3EB0C6F7A0B5ED8D
325  br i1 %cmp4, label %if.then, label %lor.lhs.false
326
327lor.lhs.false:
328  %cmp5 = fcmp ogt double %div, 1.0
329  br i1 %cmp5, label %land.lhs.true6, label %if.end
330
331land.lhs.true6:
332  %cmp7 = fcmp ogt double %div3, 1.0
333  br i1 %cmp7, label %if.then, label %if.end
334
335if.then:
336  br label %cleanup
337
338if.end:
339  br label %cleanup
340
341cleanup:
342  %retval.0 = phi i1 [ false, %if.then ], [ true, %if.end ]
343  ret i1 %retval.0
344}
345