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
5 target triple = "x86_64--"
6 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
8 define 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]]
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
25 define 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> poison, <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> poison, <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]]
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
42 define 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> poison, <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> poison, <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> poison, <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]]
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
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.
66 define i32 @TestVectorsEqual(i32* noalias %Vec0, i32* noalias %Vec1, i32 %Tolerance) {
67 ; CHECK-LABEL: @TestVectorsEqual(
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:%.*]] = call <4 x i32> @llvm.abs.v4i32(<4 x i32> [[TMP4]], i1 true)
75 ; CHECK-NEXT: [[TMP6:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[TMP5]])
76 ; CHECK-NEXT: [[CMP5_NOT:%.*]] = icmp sle i32 [[TMP6]], [[TOLERANCE:%.*]]
77 ; CHECK-NEXT: [[COND6:%.*]] = zext i1 [[CMP5_NOT]] to i32
78 ; CHECK-NEXT: ret i32 [[COND6]]
84 %sum.0 = phi i32 [ 0, %entry ], [ %add, %for.inc ]
85 %Component.0 = phi i32 [ 0, %entry ], [ %inc, %for.inc ]
86 %cmp = icmp slt i32 %Component.0, 4
87 br i1 %cmp, label %for.body, label %for.cond.cleanup
93 %idxprom = sext i32 %Component.0 to i64
94 %arrayidx = getelementptr inbounds i32, i32* %Vec0, i64 %idxprom
95 %0 = load i32, i32* %arrayidx, align 4
96 %idxprom1 = sext i32 %Component.0 to i64
97 %arrayidx2 = getelementptr inbounds i32, i32* %Vec1, i64 %idxprom1
98 %1 = load i32, i32* %arrayidx2, align 4
99 %sub = sub nsw i32 %0, %1
100 %cmp3 = icmp sge i32 %sub, 0
101 br i1 %cmp3, label %cond.true, label %cond.false
107 %sub4 = sub nsw i32 0, %sub
111 %cond = phi i32 [ %sub, %cond.true ], [ %sub4, %cond.false ]
112 %add = add nsw i32 %sum.0, %cond
116 %inc = add nsw i32 %Component.0, 1
120 %cmp5 = icmp sle i32 %sum.0, %Tolerance
121 %2 = zext i1 %cmp5 to i64
122 %cond6 = select i1 %cmp5, i32 1, i32 0
126 define i32 @TestVectorsEqual_alt(i32* noalias %Vec0, i32* noalias %Vec1, i32 %Tolerance) {
127 ; CHECK-LABEL: @TestVectorsEqual_alt(
129 ; CHECK-NEXT: [[TMP0:%.*]] = bitcast i32* [[VEC0:%.*]] to <4 x i32>*
130 ; CHECK-NEXT: [[TMP1:%.*]] = load <4 x i32>, <4 x i32>* [[TMP0]], align 4
131 ; CHECK-NEXT: [[TMP2:%.*]] = bitcast i32* [[VEC1:%.*]] to <4 x i32>*
132 ; CHECK-NEXT: [[TMP3:%.*]] = load <4 x i32>, <4 x i32>* [[TMP2]], align 4
133 ; CHECK-NEXT: [[TMP4:%.*]] = sub <4 x i32> [[TMP1]], [[TMP3]]
134 ; CHECK-NEXT: [[TMP5:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[TMP4]])
135 ; CHECK-NEXT: [[CMP3_NOT:%.*]] = icmp ule i32 [[TMP5]], [[TOLERANCE:%.*]]
136 ; CHECK-NEXT: [[COND:%.*]] = zext i1 [[CMP3_NOT]] to i32
137 ; CHECK-NEXT: ret i32 [[COND]]
143 %sum.0 = phi i32 [ 0, %entry ], [ %add, %for.inc ]
144 %Component.0 = phi i32 [ 0, %entry ], [ %inc, %for.inc ]
145 %cmp = icmp slt i32 %Component.0, 4
146 br i1 %cmp, label %for.body, label %for.cond.cleanup
152 %idxprom = sext i32 %Component.0 to i64
153 %arrayidx = getelementptr inbounds i32, i32* %Vec0, i64 %idxprom
154 %0 = load i32, i32* %arrayidx, align 4
155 %idxprom1 = sext i32 %Component.0 to i64
156 %arrayidx2 = getelementptr inbounds i32, i32* %Vec1, i64 %idxprom1
157 %1 = load i32, i32* %arrayidx2, align 4
158 %sub = sub i32 %0, %1
159 %add = add i32 %sum.0, %sub
163 %inc = add nsw i32 %Component.0, 1
167 %cmp3 = icmp ule i32 %sum.0, %Tolerance
168 %2 = zext i1 %cmp3 to i64
169 %cond = select i1 %cmp3, i32 1, i32 0
173 define i32 @TestVectorsEqualFP(float* noalias %Vec0, float* noalias %Vec1, float %Tolerance) {
174 ; CHECK-LABEL: @TestVectorsEqualFP(
176 ; CHECK-NEXT: [[TMP0:%.*]] = bitcast float* [[VEC0:%.*]] to <4 x float>*
177 ; CHECK-NEXT: [[TMP1:%.*]] = load <4 x float>, <4 x float>* [[TMP0]], align 4
178 ; CHECK-NEXT: [[TMP2:%.*]] = bitcast float* [[VEC1:%.*]] to <4 x float>*
179 ; CHECK-NEXT: [[TMP3:%.*]] = load <4 x float>, <4 x float>* [[TMP2]], align 4
180 ; CHECK-NEXT: [[TMP4:%.*]] = fsub fast <4 x float> [[TMP1]], [[TMP3]]
181 ; CHECK-NEXT: [[TMP5:%.*]] = call fast <4 x float> @llvm.fabs.v4f32(<4 x float> [[TMP4]])
182 ; CHECK-NEXT: [[TMP6:%.*]] = call fast float @llvm.vector.reduce.fadd.v4f32(float -0.000000e+00, <4 x float> [[TMP5]])
183 ; CHECK-NEXT: [[CMP4:%.*]] = fcmp fast ole float [[TMP6]], [[TOLERANCE:%.*]]
184 ; CHECK-NEXT: [[COND5:%.*]] = zext i1 [[CMP4]] to i32
185 ; CHECK-NEXT: ret i32 [[COND5]]
191 %sum.0 = phi float [ 0.000000e+00, %entry ], [ %add, %for.inc ]
192 %Component.0 = phi i32 [ 0, %entry ], [ %inc, %for.inc ]
193 %cmp = icmp slt i32 %Component.0, 4
194 br i1 %cmp, label %for.body, label %for.cond.cleanup
200 %idxprom = sext i32 %Component.0 to i64
201 %arrayidx = getelementptr inbounds float, float* %Vec0, i64 %idxprom
202 %0 = load float, float* %arrayidx, align 4
203 %idxprom1 = sext i32 %Component.0 to i64
204 %arrayidx2 = getelementptr inbounds float, float* %Vec1, i64 %idxprom1
205 %1 = load float, float* %arrayidx2, align 4
206 %sub = fsub fast float %0, %1
207 %cmp3 = fcmp fast oge float %sub, 0.000000e+00
208 br i1 %cmp3, label %cond.true, label %cond.false
214 %fneg = fneg fast float %sub
218 %cond = phi fast float [ %sub, %cond.true ], [ %fneg, %cond.false ]
219 %add = fadd fast float %sum.0, %cond
223 %inc = add nsw i32 %Component.0, 1
227 %cmp4 = fcmp fast ole float %sum.0, %Tolerance
228 %2 = zext i1 %cmp4 to i64
229 %cond5 = select i1 %cmp4, i32 1, i32 0
233 define i32 @TestVectorsEqualFP_alt(float* noalias %Vec0, float* noalias %Vec1, float %Tolerance) {
234 ; CHECK-LABEL: @TestVectorsEqualFP_alt(
236 ; CHECK-NEXT: [[TMP0:%.*]] = bitcast float* [[VEC0:%.*]] to <4 x float>*
237 ; CHECK-NEXT: [[TMP1:%.*]] = load <4 x float>, <4 x float>* [[TMP0]], align 4
238 ; CHECK-NEXT: [[TMP2:%.*]] = bitcast float* [[VEC1:%.*]] to <4 x float>*
239 ; CHECK-NEXT: [[TMP3:%.*]] = load <4 x float>, <4 x float>* [[TMP2]], align 4
240 ; CHECK-NEXT: [[TMP4:%.*]] = fsub fast <4 x float> [[TMP1]], [[TMP3]]
241 ; CHECK-NEXT: [[TMP5:%.*]] = call fast float @llvm.vector.reduce.fadd.v4f32(float -0.000000e+00, <4 x float> [[TMP4]])
242 ; CHECK-NEXT: [[CMP3:%.*]] = fcmp fast ole float [[TMP5]], [[TOLERANCE:%.*]]
243 ; CHECK-NEXT: [[COND:%.*]] = zext i1 [[CMP3]] to i32
244 ; CHECK-NEXT: ret i32 [[COND]]
250 %sum.0 = phi float [ 0.000000e+00, %entry ], [ %add, %for.inc ]
251 %Component.0 = phi i32 [ 0, %entry ], [ %inc, %for.inc ]
252 %cmp = icmp slt i32 %Component.0, 4
253 br i1 %cmp, label %for.body, label %for.cond.cleanup
259 %idxprom = sext i32 %Component.0 to i64
260 %arrayidx = getelementptr inbounds float, float* %Vec0, i64 %idxprom
261 %0 = load float, float* %arrayidx, align 4
262 %idxprom1 = sext i32 %Component.0 to i64
263 %arrayidx2 = getelementptr inbounds float, float* %Vec1, i64 %idxprom1
264 %1 = load float, float* %arrayidx2, align 4
265 %sub = fsub fast float %0, %1
266 %add = fadd fast float %sum.0, %sub
270 %inc = add nsw i32 %Component.0, 1
274 %cmp3 = fcmp fast ole float %sum.0, %Tolerance
275 %2 = zext i1 %cmp3 to i64
276 %cond = select i1 %cmp3, i32 1, i32 0
280 ; PR43745 - https://bugs.llvm.org/show_bug.cgi?id=43745
282 ; FIXME: this should be vectorized
283 define i1 @cmp_lt_gt(double %a, double %b, double %c) {
284 ; CHECK-LABEL: @cmp_lt_gt(
286 ; CHECK-NEXT: [[FNEG:%.*]] = fneg double [[B:%.*]]
287 ; CHECK-NEXT: [[ADD:%.*]] = fsub double [[C:%.*]], [[B]]
288 ; CHECK-NEXT: [[MUL:%.*]] = fmul double [[A:%.*]], 2.000000e+00
289 ; CHECK-NEXT: [[DIV:%.*]] = fdiv double [[ADD]], [[MUL]]
290 ; CHECK-NEXT: [[SUB:%.*]] = fsub double [[FNEG]], [[C]]
291 ; CHECK-NEXT: [[DIV3:%.*]] = fdiv double [[SUB]], [[MUL]]
292 ; CHECK-NEXT: [[CMP:%.*]] = fcmp olt double [[DIV]], 0x3EB0C6F7A0B5ED8D
293 ; CHECK-NEXT: [[CMP4:%.*]] = fcmp olt double [[DIV3]], 0x3EB0C6F7A0B5ED8D
294 ; CHECK-NEXT: [[OR_COND:%.*]] = select i1 [[CMP]], i1 [[CMP4]], i1 false
295 ; CHECK-NEXT: br i1 [[OR_COND]], label [[CLEANUP:%.*]], label [[LOR_LHS_FALSE:%.*]]
296 ; CHECK: lor.lhs.false:
297 ; CHECK-NEXT: [[CMP5:%.*]] = fcmp ule double [[DIV]], 1.000000e+00
298 ; CHECK-NEXT: [[CMP7:%.*]] = fcmp ule double [[DIV3]], 1.000000e+00
299 ; CHECK-NEXT: [[OR_COND1:%.*]] = select i1 [[CMP5]], i1 true, i1 [[CMP7]]
300 ; CHECK-NEXT: br label [[CLEANUP]]
302 ; CHECK-NEXT: [[RETVAL_0:%.*]] = phi i1 [ false, [[ENTRY:%.*]] ], [ [[OR_COND1]], [[LOR_LHS_FALSE]] ]
303 ; CHECK-NEXT: ret i1 [[RETVAL_0]]
306 %fneg = fneg double %b
307 %add = fadd double %fneg, %c
308 %mul = fmul double 2.0, %a
309 %div = fdiv double %add, %mul
310 %fneg1 = fneg double %b
311 %sub = fsub double %fneg1, %c
312 %mul2 = fmul double 2.0, %a
313 %div3 = fdiv double %sub, %mul2
314 %cmp = fcmp olt double %div, 0x3EB0C6F7A0B5ED8D
315 br i1 %cmp, label %land.lhs.true, label %lor.lhs.false
318 %cmp4 = fcmp olt double %div3, 0x3EB0C6F7A0B5ED8D
319 br i1 %cmp4, label %if.then, label %lor.lhs.false
322 %cmp5 = fcmp ogt double %div, 1.0
323 br i1 %cmp5, label %land.lhs.true6, label %if.end
326 %cmp7 = fcmp ogt double %div3, 1.0
327 br i1 %cmp7, label %if.then, label %if.end
336 %retval.0 = phi i1 [ false, %if.then ], [ true, %if.end ]