1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
|
; REQUIRES: asserts
; RUN: opt -S -passes=loop-vectorize -force-vector-width=8 -force-vector-interleave=1 -enable-interleaved-mem-accesses -debug-only=loop-vectorize,vectorutils -disable-output < %s 2>&1 | FileCheck %s -check-prefix=STRIDED_UNMASKED
; RUN: opt -S -passes=loop-vectorize -force-vector-width=8 -force-vector-interleave=1 -enable-interleaved-mem-accesses -enable-masked-interleaved-mem-accesses -debug-only=loop-vectorize,vectorutils -disable-output < %s 2>&1 | FileCheck %s -check-prefix=STRIDED_MASKED
target datalayout = "e-m:e-p:32:32-f64:32:64-f80:32-n8:16:32-S128"
; We test here that the loop-vectorizer forms an interleave-groups from
; predicated memory accesses only if they are both in the same (predicated)
; block (first scenario below).
; If the accesses are not in the same predicated block, an interleave-group
; is not formed (scenarios 2,3 below).
; Scenario 1: Check the case where it is legal to create masked interleave-
; groups. Altogether two groups are created (one for loads and one for stores)
; when masked-interleaved-acceses are enabled. When masked-interleaved-acceses
; are disabled we do not create any interleave-group.
;
; void masked_strided1(const unsigned char* restrict p,
; unsigned char* restrict q,
; unsigned char guard) {
; for(ix=0; ix < 1024; ++ix) {
; if (ix > guard) {
; char left = p[2*ix];
; char right = p[2*ix + 1];
; char max = max(left, right);
; q[2*ix] = max;
; q[2*ix+1] = 0 - max;
; }
; }
;}
; STRIDED_UNMASKED: LV: Checking a loop in 'masked_strided1'
; STRIDED_UNMASKED: LV: Analyzing interleaved accesses...
; STRIDED_UNMASKED-NOT: LV: Creating an interleave group
; STRIDED_MASKED: LV: Checking a loop in 'masked_strided1'
; STRIDED_MASKED: LV: Analyzing interleaved accesses...
; STRIDED_MASKED-NEXT: LV: Creating an interleave group with: store i8 %{{.*}}, ptr %{{.*}}, align 1
; STRIDED_MASKED-NEXT: LV: Inserted: store i8 %{{.*}}, ptr %{{.*}}, align 1
; STRIDED_MASKED-NEXT: into the interleave group with store i8 %{{.*}}, ptr %{{.*}}, align 1
; STRIDED_MASKED-NEXT: LV: Creating an interleave group with: %{{.*}} = load i8, ptr %{{.*}}, align 1
; STRIDED_MASKED-NEXT: LV: Inserted: %{{.*}} = load i8, ptr %{{.*}}, align 1
; STRIDED_MASKED-NEXT: into the interleave group with %{{.*}} = load i8, ptr %{{.*}}, align 1
; Scenario 2: Check the case where it is illegal to create a masked interleave-
; group because the first access is predicated, and the second isn't.
; We therefore create a separate interleave-group with gaps for each of the
; stores (if masked-interleaved-accesses are enabled).
; If masked-interleaved-accesses is not enabled we create only one interleave
; group of stores (for the non-predicated store) and it is later invalidated
; due to gaps.
;
; void masked_strided2(const unsigned char* restrict p,
; unsigned char* restrict q,
; unsigned char guard1,
; unsigned char guard2) {
; for(ix=0; ix < 1024; ++ix) {
; if (ix > guard1) {
; q[2*ix] = 1;
; }
; q[2*ix+1] = 2;
; }
;}
; STRIDED_UNMASKED: LV: Checking a loop in 'masked_strided2'
; STRIDED_UNMASKED: LV: Analyzing interleaved accesses...
; STRIDED_UNMASKED-NEXT: LV: Creating an interleave group with: store i8 1, ptr %{{.*}}, align 1
; STRIDED_UNMASKED-NEXT: LV: Invalidate candidate interleaved store group due to gaps.
; STRIDED_UNMASKED-NOT: LV: Creating an interleave group
; STRIDED_MASKED: LV: Checking a loop in 'masked_strided2'
; STRIDED_MASKED: LV: Analyzing interleaved accesses...
; STRIDED_MASKED-NEXT: LV: Creating an interleave group with: store i8 2, ptr %{{.*}}, align 1
; STRIDED_MASKED-NEXT: LV: Creating an interleave group with: store i8 1, ptr %{{.*}}, align 1
; STRIDED_MASKED-NOT: LV: Invalidate candidate interleaved store group due to gaps.
; Scenario 3: Check the case where it is illegal to create a masked interleave-
; group because the two accesses are in separate predicated blocks.
; We therefore create a separate interleave-group with gaps for each of the accesses,
; If masked-interleaved-accesses is not enabled we don't create any interleave
; group because all accesses are predicated.
;
; void masked_strided3(const unsigned char* restrict p,
; unsigned char* restrict q,
; unsigned char guard1,
; unsigned char guard2) {
; for(ix=0; ix < 1024; ++ix) {
; if (ix > guard1) {
; q[2*ix] = 1;
; }
; if (ix > guard2) {
; q[2*ix+1] = 2;
; }
; }
;}
; STRIDED_UNMASKED: LV: Checking a loop in 'masked_strided3'
; STRIDED_UNMASKED: LV: Analyzing interleaved accesses...
; STRIDED_UNMASKED-NOT: LV: Creating an interleave group
; STRIDED_MASKED: LV: Checking a loop in 'masked_strided3'
; STRIDED_MASKED: LV: Analyzing interleaved accesses...
; STRIDED_MASKED-NEXT: LV: Creating an interleave group with: store i8 2, ptr %{{.*}}, align 1
; STRIDED_MASKED-NEXT: LV: Creating an interleave group with: store i8 1, ptr %{{.*}}, align 1
; STRIDED_MASKED-NOT: LV: Invalidate candidate interleaved store group due to gaps.
; ModuleID = 'test.c'
source_filename = "test.c"
target datalayout = "e-m:e-p:32:32-f64:32:64-f80:32-n8:16:32-S128"
target triple = "i386-unknown-linux-gnu"
define dso_local void @masked_strided1(ptr noalias nocapture readonly %p, ptr noalias nocapture %q, i8 zeroext %guard) local_unnamed_addr #0 {
entry:
%conv = zext i8 %guard to i32
br label %for.body
for.body:
%ix.024 = phi i32 [ 0, %entry ], [ %inc, %for.inc ]
%cmp1 = icmp ugt i32 %ix.024, %conv
br i1 %cmp1, label %if.then, label %for.inc
if.then:
%mul = shl nuw nsw i32 %ix.024, 1
%arrayidx = getelementptr inbounds i8, ptr %p, i32 %mul
%0 = load i8, ptr %arrayidx, align 1
%add = or disjoint i32 %mul, 1
%arrayidx4 = getelementptr inbounds i8, ptr %p, i32 %add
%1 = load i8, ptr %arrayidx4, align 1
%cmp.i = icmp slt i8 %0, %1
%spec.select.i = select i1 %cmp.i, i8 %1, i8 %0
%arrayidx6 = getelementptr inbounds i8, ptr %q, i32 %mul
store i8 %spec.select.i, ptr %arrayidx6, align 1
%sub = sub i8 0, %spec.select.i
%arrayidx11 = getelementptr inbounds i8, ptr %q, i32 %add
store i8 %sub, ptr %arrayidx11, align 1
br label %for.inc
for.inc:
%inc = add nuw nsw i32 %ix.024, 1
%exitcond = icmp eq i32 %inc, 1024
br i1 %exitcond, label %for.end, label %for.body
for.end:
ret void
}
define dso_local void @masked_strided2(ptr noalias nocapture readnone %p, ptr noalias nocapture %q, i8 zeroext %guard) local_unnamed_addr #0 {
entry:
%conv = zext i8 %guard to i32
br label %for.body
for.body:
%ix.012 = phi i32 [ 0, %entry ], [ %inc, %for.inc ]
%mul = shl nuw nsw i32 %ix.012, 1
%arrayidx = getelementptr inbounds i8, ptr %q, i32 %mul
store i8 1, ptr %arrayidx, align 1
%cmp1 = icmp ugt i32 %ix.012, %conv
br i1 %cmp1, label %if.then, label %for.inc
if.then:
%add = or disjoint i32 %mul, 1
%arrayidx3 = getelementptr inbounds i8, ptr %q, i32 %add
store i8 2, ptr %arrayidx3, align 1
br label %for.inc
for.inc:
%inc = add nuw nsw i32 %ix.012, 1
%exitcond = icmp eq i32 %inc, 1024
br i1 %exitcond, label %for.end, label %for.body
for.end:
ret void
}
define dso_local void @masked_strided3(ptr noalias nocapture readnone %p, ptr noalias nocapture %q, i8 zeroext %guard1, i8 zeroext %guard2) local_unnamed_addr #0 {
entry:
%conv = zext i8 %guard1 to i32
%conv3 = zext i8 %guard2 to i32
br label %for.body
for.body:
%ix.018 = phi i32 [ 0, %entry ], [ %inc, %for.inc ]
%mul = shl nuw nsw i32 %ix.018, 1
%cmp1 = icmp ugt i32 %ix.018, %conv
br i1 %cmp1, label %if.then, label %if.end
if.then:
%arrayidx = getelementptr inbounds i8, ptr %q, i32 %mul
store i8 1, ptr %arrayidx, align 1
br label %if.end
if.end:
%cmp4 = icmp ugt i32 %ix.018, %conv3
br i1 %cmp4, label %if.then6, label %for.inc
if.then6:
%add = or disjoint i32 %mul, 1
%arrayidx7 = getelementptr inbounds i8, ptr %q, i32 %add
store i8 2, ptr %arrayidx7, align 1
br label %for.inc
for.inc:
%inc = add nuw nsw i32 %ix.018, 1
%exitcond = icmp eq i32 %inc, 1024
br i1 %exitcond, label %for.end, label %for.body
for.end:
ret void
}
attributes #0 = { "target-features"="+fxsr,+mmx,+sse,+sse2,+x87" }
|
