1 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2 ; RUN: opt < %s -loop-vectorize -force-vector-width=4 -force-vector-interleave=1 -prefer-predicate-over-epilogue=predicate-else-scalar-epilogue -prefer-predicated-reduction-select -dce -instcombine -S | FileCheck %s
4 target datalayout = "e-m:e-p:32:32-Fi8-i64:64-v128:64:128-a:0:32-n32-S64"
6 define i32 @reduction_sum_single(i32* noalias nocapture %A) {
7 ; CHECK-LABEL: @reduction_sum_single(
9 ; CHECK: [[VEC_PHI:%.*]] = phi <4 x i32> [ zeroinitializer, %vector.ph ], [ [[TMP25:%.*]], %pred.load.continue6 ]
10 ; CHECK: [[TMP24:%.*]] = select <4 x i1> [[TMP0:%.*]], <4 x i32> [[TMP23:%.*]], <4 x i32> zeroinitializer
11 ; CHECK: [[TMP25]] = add <4 x i32> [[VEC_PHI]], [[TMP24]]
12 ; CHECK: middle.block:
13 ; CHECK: [[TMP27:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[TMP25]])
18 .lr.ph: ; preds = %entry, %.lr.ph
19 %indvars.iv = phi i32 [ %indvars.iv.next, %.lr.ph ], [ 0, %entry ]
20 %sum.02 = phi i32 [ %l7, %.lr.ph ], [ 0, %entry ]
21 %l2 = getelementptr inbounds i32, i32* %A, i32 %indvars.iv
22 %l3 = load i32, i32* %l2, align 4
23 %l7 = add i32 %sum.02, %l3
24 %indvars.iv.next = add i32 %indvars.iv, 1
25 %exitcond = icmp eq i32 %indvars.iv.next, 257
26 br i1 %exitcond, label %._crit_edge, label %.lr.ph
28 ._crit_edge: ; preds = %.lr.ph
29 %sum.0.lcssa = phi i32 [ %l7, %.lr.ph ]
33 define i32 @reduction_sum(i32* noalias nocapture %A, i32* noalias nocapture %B) {
34 ; CHECK-LABEL: @reduction_sum(
35 ; CHECK: [[VEC_PHI:%.*]] = phi <4 x i32> [ zeroinitializer, %vector.ph ], [ [[TMP47:%.*]], %pred.load.continue6 ]
36 ; CHECK: [[TMP44:%.*]] = add <4 x i32> [[VEC_PHI]], [[VEC_IND:%.*]]
37 ; CHECK: [[TMP45:%.*]] = add <4 x i32> [[TMP44]], [[TMP23:%.*]]
38 ; CHECK: [[TMP46:%.*]] = add <4 x i32> [[TMP45]], [[TMP43:%.*]]
39 ; CHECK: [[TMP47]] = select <4 x i1> [[TMP3:%.*]], <4 x i32> [[TMP46]], <4 x i32> [[VEC_PHI]]
40 ; CHECK: middle.block:
41 ; CHECK: [[TMP49:%.*]] = call i32 @llvm.vector.reduce.add.v4i32(<4 x i32> [[TMP47]])
46 .lr.ph: ; preds = %entry, %.lr.ph
47 %indvars.iv = phi i32 [ %indvars.iv.next, %.lr.ph ], [ 0, %entry ]
48 %sum.02 = phi i32 [ %l9, %.lr.ph ], [ 0, %entry ]
49 %l2 = getelementptr inbounds i32, i32* %A, i32 %indvars.iv
50 %l3 = load i32, i32* %l2, align 4
51 %l4 = getelementptr inbounds i32, i32* %B, i32 %indvars.iv
52 %l5 = load i32, i32* %l4, align 4
53 %l7 = add i32 %sum.02, %indvars.iv
54 %l8 = add i32 %l7, %l3
55 %l9 = add i32 %l8, %l5
56 %indvars.iv.next = add i32 %indvars.iv, 1
57 %exitcond = icmp eq i32 %indvars.iv.next, 257
58 br i1 %exitcond, label %._crit_edge, label %.lr.ph
60 ._crit_edge: ; preds = %.lr.ph
61 %sum.0.lcssa = phi i32 [ %l9, %.lr.ph ]
65 define i32 @reduction_prod(i32* noalias nocapture %A, i32* noalias nocapture %B) {
66 ; CHECK-LABEL: @reduction_prod(
68 ; CHECK: [[VEC_PHI:%.*]] = phi <4 x i32> [ <i32 1, i32 1, i32 1, i32 1>, %vector.ph ], [ [[TMP46:%.*]], %pred.load.continue6 ]
69 ; CHECK: [[TMP44:%.*]] = mul <4 x i32> [[VEC_PHI]], [[TMP23:%.*]]
70 ; CHECK: [[TMP45:%.*]] = mul <4 x i32> [[TMP44]], [[TMP43:%.*]]
71 ; CHECK: [[TMP46]] = select <4 x i1> [[TMP3:%.*]], <4 x i32> [[TMP45]], <4 x i32> [[VEC_PHI]]
72 ; CHECK: middle.block:
73 ; CHECK: [[TMP48:%.*]] = call i32 @llvm.vector.reduce.mul.v4i32(<4 x i32> [[TMP46]])
78 .lr.ph: ; preds = %entry, %.lr.ph
79 %indvars.iv = phi i32 [ %indvars.iv.next, %.lr.ph ], [ 0, %entry ]
80 %prod.02 = phi i32 [ %l9, %.lr.ph ], [ 1, %entry ]
81 %l2 = getelementptr inbounds i32, i32* %A, i32 %indvars.iv
82 %l3 = load i32, i32* %l2, align 4
83 %l4 = getelementptr inbounds i32, i32* %B, i32 %indvars.iv
84 %l5 = load i32, i32* %l4, align 4
85 %l8 = mul i32 %prod.02, %l3
86 %l9 = mul i32 %l8, %l5
87 %indvars.iv.next = add i32 %indvars.iv, 1
88 %exitcond = icmp eq i32 %indvars.iv.next, 257
89 br i1 %exitcond, label %._crit_edge, label %.lr.ph
91 ._crit_edge: ; preds = %.lr.ph
92 %prod.0.lcssa = phi i32 [ %l9, %.lr.ph ]
96 define i32 @reduction_and(i32* nocapture %A, i32* nocapture %B) {
97 ; CHECK-LABEL: @reduction_and(
99 ; CHECK: [[VEC_PHI:%.*]] = phi <4 x i32> [ <i32 -1, i32 -1, i32 -1, i32 -1>, %vector.ph ], [ [[TMP46:%.*]], %pred.load.continue6 ]
100 ; CHECK: [[TMP44:%.*]] = and <4 x i32> [[VEC_PHI]], [[TMP23:%.*]]
101 ; CHECK: [[TMP45:%.*]] = and <4 x i32> [[TMP44]], [[TMP43:%.*]]
102 ; CHECK: [[TMP46]] = select <4 x i1> [[TMP3:%.*]], <4 x i32> [[TMP45]], <4 x i32> [[VEC_PHI]]
103 ; CHECK: middle.block:
104 ; CHECK: [[TMP48:%.*]] = call i32 @llvm.vector.reduce.and.v4i32(<4 x i32> [[TMP46]])
109 for.body: ; preds = %entry, %for.body
110 %indvars.iv = phi i32 [ %indvars.iv.next, %for.body ], [ 0, %entry ]
111 %result.08 = phi i32 [ %and, %for.body ], [ -1, %entry ]
112 %arrayidx = getelementptr inbounds i32, i32* %A, i32 %indvars.iv
113 %l0 = load i32, i32* %arrayidx, align 4
114 %arrayidx2 = getelementptr inbounds i32, i32* %B, i32 %indvars.iv
115 %l1 = load i32, i32* %arrayidx2, align 4
116 %add = and i32 %result.08, %l0
117 %and = and i32 %add, %l1
118 %indvars.iv.next = add i32 %indvars.iv, 1
119 %exitcond = icmp eq i32 %indvars.iv.next, 257
120 br i1 %exitcond, label %for.end, label %for.body
122 for.end: ; preds = %for.body, %entry
123 %result.0.lcssa = phi i32 [ %and, %for.body ]
124 ret i32 %result.0.lcssa
127 define i32 @reduction_or(i32* nocapture %A, i32* nocapture %B) {
128 ; CHECK-LABEL: @reduction_or(
129 ; CHECK: vector.body:
130 ; CHECK: [[VEC_PHI:%.*]] = phi <4 x i32> [ zeroinitializer, %vector.ph ], [ [[TMP46:%.*]], %pred.load.continue6 ]
131 ; CHECK: [[TMP45:%.*]] = select <4 x i1> [[TMP3:%.*]], <4 x i32> [[TMP44:%.*]], <4 x i32> zeroinitializer
132 ; CHECK: [[TMP46]] = or <4 x i32> [[VEC_PHI]], [[TMP45]]
133 ; CHECK: middle.block:
134 ; CHECK: [[TMP48:%.*]] = call i32 @llvm.vector.reduce.or.v4i32(<4 x i32> [[TMP46]])
139 for.body: ; preds = %entry, %for.body
140 %indvars.iv = phi i32 [ %indvars.iv.next, %for.body ], [ 0, %entry ]
141 %result.08 = phi i32 [ %or, %for.body ], [ 0, %entry ]
142 %arrayidx = getelementptr inbounds i32, i32* %A, i32 %indvars.iv
143 %l0 = load i32, i32* %arrayidx, align 4
144 %arrayidx2 = getelementptr inbounds i32, i32* %B, i32 %indvars.iv
145 %l1 = load i32, i32* %arrayidx2, align 4
146 %add = add nsw i32 %l1, %l0
147 %or = or i32 %add, %result.08
148 %indvars.iv.next = add i32 %indvars.iv, 1
149 %exitcond = icmp eq i32 %indvars.iv.next, 257
150 br i1 %exitcond, label %for.end, label %for.body
152 for.end: ; preds = %for.body, %entry
153 %result.0.lcssa = phi i32 [ %or, %for.body ]
154 ret i32 %result.0.lcssa
157 define i32 @reduction_xor(i32* nocapture %A, i32* nocapture %B) {
158 ; CHECK-LABEL: @reduction_xor(
159 ; CHECK: vector.body:
160 ; CHECK: [[VEC_PHI:%.*]] = phi <4 x i32> [ zeroinitializer, %vector.ph ], [ [[TMP46:%.*]], %pred.load.continue6 ]
161 ; CHECK: [[TMP45:%.*]] = select <4 x i1> [[TMP3:%.*]], <4 x i32> [[TMP44:%.*]], <4 x i32> zeroinitializer
162 ; CHECK: [[TMP46]] = xor <4 x i32> [[VEC_PHI]], [[TMP45]]
163 ; CHECK: middle.block:
164 ; CHECK: [[TMP48:%.*]] = call i32 @llvm.vector.reduce.xor.v4i32(<4 x i32> [[TMP46]])
169 for.body: ; preds = %entry, %for.body
170 %indvars.iv = phi i32 [ %indvars.iv.next, %for.body ], [ 0, %entry ]
171 %result.08 = phi i32 [ %xor, %for.body ], [ 0, %entry ]
172 %arrayidx = getelementptr inbounds i32, i32* %A, i32 %indvars.iv
173 %l0 = load i32, i32* %arrayidx, align 4
174 %arrayidx2 = getelementptr inbounds i32, i32* %B, i32 %indvars.iv
175 %l1 = load i32, i32* %arrayidx2, align 4
176 %add = add nsw i32 %l1, %l0
177 %xor = xor i32 %add, %result.08
178 %indvars.iv.next = add i32 %indvars.iv, 1
179 %exitcond = icmp eq i32 %indvars.iv.next, 257
180 br i1 %exitcond, label %for.end, label %for.body
182 for.end: ; preds = %for.body, %entry
183 %result.0.lcssa = phi i32 [ %xor, %for.body ]
184 ret i32 %result.0.lcssa
187 define float @reduction_fadd(float* nocapture %A, float* nocapture %B) {
188 ; CHECK-LABEL: @reduction_fadd(
189 ; CHECK: vector.body:
190 ; CHECK: [[VEC_PHI:%.*]] = phi <4 x float> [ zeroinitializer, %vector.ph ], [ [[TMP46:%.*]], %pred.load.continue6 ]
191 ; CHECK: [[TMP44:%.*]] = fadd fast <4 x float> [[VEC_PHI]], [[TMP23:%.*]]
192 ; CHECK: [[TMP45:%.*]] = fadd fast <4 x float> [[TMP44]], [[TMP43:%.*]]
193 ; CHECK: [[TMP46]] = select <4 x i1> [[TMP3:%.*]], <4 x float> [[TMP45]], <4 x float> [[VEC_PHI]]
194 ; CHECK: middle.block:
195 ; CHECK: [[TMP48:%.*]] = call fast float @llvm.vector.reduce.fadd.v4f32(float -0.000000e+00, <4 x float> [[TMP46]])
200 for.body: ; preds = %entry, %for.body
201 %indvars.iv = phi i32 [ %indvars.iv.next, %for.body ], [ 0, %entry ]
202 %result.08 = phi float [ %fadd, %for.body ], [ 0.0, %entry ]
203 %arrayidx = getelementptr inbounds float, float* %A, i32 %indvars.iv
204 %l0 = load float, float* %arrayidx, align 4
205 %arrayidx2 = getelementptr inbounds float, float* %B, i32 %indvars.iv
206 %l1 = load float, float* %arrayidx2, align 4
207 %add = fadd fast float %result.08, %l0
208 %fadd = fadd fast float %add, %l1
209 %indvars.iv.next = add i32 %indvars.iv, 1
210 %exitcond = icmp eq i32 %indvars.iv.next, 257
211 br i1 %exitcond, label %for.end, label %for.body
213 for.end: ; preds = %for.body, %entry
214 %result.0.lcssa = phi float [ %fadd, %for.body ]
215 ret float %result.0.lcssa
218 define float @reduction_fmul(float* nocapture %A, float* nocapture %B) {
219 ; CHECK-LABEL: @reduction_fmul(
220 ; CHECK: vector.body:
221 ; CHECK: [[VEC_PHI:%.*]] = phi <4 x float> [ <float 0.000000e+00, float 1.000000e+00, float 1.000000e+00, float 1.000000e+00>, %vector.ph ], [ [[TMP46:%.*]], %pred.load.continue6 ]
222 ; CHECK: [[TMP44:%.*]] = fmul fast <4 x float> [[VEC_PHI]], [[TMP23:%.*]]
223 ; CHECK: [[TMP45:%.*]] = fmul fast <4 x float> [[TMP44]], [[TMP43:%.*]]
224 ; CHECK: [[TMP46]] = select <4 x i1> [[TMP3:%.*]], <4 x float> [[TMP45]], <4 x float> [[VEC_PHI]]
225 ; CHECK: middle.block:
226 ; CHECK: [[TMP48:%.*]] = call fast float @llvm.vector.reduce.fmul.v4f32(float 1.000000e+00, <4 x float> [[TMP46]])
231 for.body: ; preds = %entry, %for.body
232 %indvars.iv = phi i32 [ %indvars.iv.next, %for.body ], [ 0, %entry ]
233 %result.08 = phi float [ %fmul, %for.body ], [ 0.0, %entry ]
234 %arrayidx = getelementptr inbounds float, float* %A, i32 %indvars.iv
235 %l0 = load float, float* %arrayidx, align 4
236 %arrayidx2 = getelementptr inbounds float, float* %B, i32 %indvars.iv
237 %l1 = load float, float* %arrayidx2, align 4
238 %add = fmul fast float %result.08, %l0
239 %fmul = fmul fast float %add, %l1
240 %indvars.iv.next = add i32 %indvars.iv, 1
241 %exitcond = icmp eq i32 %indvars.iv.next, 257
242 br i1 %exitcond, label %for.end, label %for.body
244 for.end: ; preds = %for.body, %entry
245 %result.0.lcssa = phi float [ %fmul, %for.body ]
246 ret float %result.0.lcssa
249 define i32 @reduction_min(i32* nocapture %A, i32* nocapture %B) {
250 ; CHECK-LABEL: @reduction_min(
251 ; CHECK: vector.body:
252 ; CHECK: [[VEC_PHI:%.*]] = phi <4 x i32> [ <i32 1000, i32 1000, i32 1000, i32 1000>, %vector.ph ], [ [[TMP26:%.*]], %pred.load.continue6 ]
253 ; CHECK: [[TMP24:%.*]] = icmp slt <4 x i32> [[VEC_PHI]], [[TMP23:%.*]]
254 ; CHECK: [[TMP25:%.*]] = select <4 x i1> [[TMP24]], <4 x i32> [[VEC_PHI]], <4 x i32> [[TMP23]]
255 ; CHECK: [[TMP26]] = select <4 x i1> [[TMP0:%.*]], <4 x i32> [[TMP25]], <4 x i32> [[VEC_PHI]]
256 ; CHECK: middle.block:
257 ; CHECK: [[TMP28:%.*]] = call i32 @llvm.vector.reduce.smin.v4i32(<4 x i32> [[TMP26]])
262 for.body: ; preds = %entry, %for.body
263 %indvars.iv = phi i32 [ %indvars.iv.next, %for.body ], [ 0, %entry ]
264 %result.08 = phi i32 [ %v0, %for.body ], [ 1000, %entry ]
265 %arrayidx = getelementptr inbounds i32, i32* %A, i32 %indvars.iv
266 %l0 = load i32, i32* %arrayidx, align 4
267 %c0 = icmp slt i32 %result.08, %l0
268 %v0 = select i1 %c0, i32 %result.08, i32 %l0
269 %indvars.iv.next = add i32 %indvars.iv, 1
270 %exitcond = icmp eq i32 %indvars.iv.next, 257
271 br i1 %exitcond, label %for.end, label %for.body
273 for.end: ; preds = %for.body, %entry
274 %result.0.lcssa = phi i32 [ %v0, %for.body ]
275 ret i32 %result.0.lcssa
278 define i32 @reduction_max(i32* nocapture %A, i32* nocapture %B) {
279 ; CHECK-LABEL: @reduction_max(
280 ; CHECK: vector.body:
281 ; CHECK: [[VEC_PHI:%.*]] = phi <4 x i32> [ <i32 1000, i32 1000, i32 1000, i32 1000>, %vector.ph ], [ [[TMP26:%.*]], %pred.load.continue6 ]
282 ; CHECK: [[TMP24:%.*]] = icmp ugt <4 x i32> [[VEC_PHI]], [[TMP23:%.*]]
283 ; CHECK: [[TMP25:%.*]] = select <4 x i1> [[TMP24]], <4 x i32> [[VEC_PHI]], <4 x i32> [[TMP23]]
284 ; CHECK: [[TMP26]] = select <4 x i1> [[TMP0:%.*]], <4 x i32> [[TMP25]], <4 x i32> [[VEC_PHI]]
285 ; CHECK: middle.block:
286 ; CHECK: [[TMP28:%.*]] = call i32 @llvm.vector.reduce.umax.v4i32(<4 x i32> [[TMP26]])
291 for.body: ; preds = %entry, %for.body
292 %indvars.iv = phi i32 [ %indvars.iv.next, %for.body ], [ 0, %entry ]
293 %result.08 = phi i32 [ %v0, %for.body ], [ 1000, %entry ]
294 %arrayidx = getelementptr inbounds i32, i32* %A, i32 %indvars.iv
295 %l0 = load i32, i32* %arrayidx, align 4
296 %c0 = icmp ugt i32 %result.08, %l0
297 %v0 = select i1 %c0, i32 %result.08, i32 %l0
298 %indvars.iv.next = add i32 %indvars.iv, 1
299 %exitcond = icmp eq i32 %indvars.iv.next, 257
300 br i1 %exitcond, label %for.end, label %for.body
302 for.end: ; preds = %for.body, %entry
303 %result.0.lcssa = phi i32 [ %v0, %for.body ]
304 ret i32 %result.0.lcssa