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
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
|
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=armv6t2-eabi %s -o - | FileCheck %s --check-prefix=CHECK
; Checks SSAT is still generated when loop unrolling is on
define void @ssat_unroll(ptr %pSrcA, ptr %pSrcB, ptr %pDst, i32 %blockSize) {
; CHECK-LABEL: ssat_unroll:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: cmp r3, #0
; CHECK-NEXT: bxeq lr
; CHECK-NEXT: .LBB0_1: @ %while.body.preheader
; CHECK-NEXT: .save {r11, lr}
; CHECK-NEXT: push {r11, lr}
; CHECK-NEXT: sub r12, r3, #1
; CHECK-NEXT: tst r3, #1
; CHECK-NEXT: beq .LBB0_3
; CHECK-NEXT: @ %bb.2: @ %while.body.prol.preheader
; CHECK-NEXT: ldrsh lr, [r0], #2
; CHECK-NEXT: ldrsh r3, [r1], #2
; CHECK-NEXT: smulbb r3, r3, lr
; CHECK-NEXT: ssat r3, #16, r3, asr #14
; CHECK-NEXT: strh r3, [r2], #2
; CHECK-NEXT: mov r3, r12
; CHECK-NEXT: .LBB0_3: @ %while.body.prol.loopexit
; CHECK-NEXT: cmp r12, #0
; CHECK-NEXT: beq .LBB0_5
; CHECK-NEXT: .LBB0_4: @ %while.body
; CHECK-NEXT: @ =>This Inner Loop Header: Depth=1
; CHECK-NEXT: ldrsh r12, [r0]
; CHECK-NEXT: subs r3, r3, #2
; CHECK-NEXT: ldrsh lr, [r1]
; CHECK-NEXT: smulbb r12, lr, r12
; CHECK-NEXT: ssat r12, #16, r12, asr #14
; CHECK-NEXT: strh r12, [r2]
; CHECK-NEXT: ldrsh r12, [r0, #2]
; CHECK-NEXT: add r0, r0, #4
; CHECK-NEXT: ldrsh lr, [r1, #2]
; CHECK-NEXT: add r1, r1, #4
; CHECK-NEXT: smulbb r12, lr, r12
; CHECK-NEXT: ssat r12, #16, r12, asr #14
; CHECK-NEXT: strh r12, [r2, #2]
; CHECK-NEXT: add r2, r2, #4
; CHECK-NEXT: bne .LBB0_4
; CHECK-NEXT: .LBB0_5:
; CHECK-NEXT: pop {r11, lr}
; CHECK-NEXT: bx lr
entry:
%cmp.not7 = icmp eq i32 %blockSize, 0
br i1 %cmp.not7, label %while.end, label %while.body.preheader
while.body.preheader: ; preds = %entry
%0 = add i32 %blockSize, -1
%xtraiter = and i32 %blockSize, 1
%lcmp.mod.not = icmp eq i32 %xtraiter, 0
br i1 %lcmp.mod.not, label %while.body.prol.loopexit, label %while.body.prol.preheader
while.body.prol.preheader: ; preds = %while.body.preheader
%incdec.ptr.prol = getelementptr inbounds i16, ptr %pSrcA, i32 1
%1 = load i16, ptr %pSrcA
%conv.prol = sext i16 %1 to i32
%incdec.ptr1.prol = getelementptr inbounds i16, ptr %pSrcB, i32 1
%2 = load i16, ptr %pSrcB
%conv2.prol = sext i16 %2 to i32
%mul.prol = mul nsw i32 %conv2.prol, %conv.prol
%shr.prol = ashr i32 %mul.prol, 14
%3 = icmp sgt i32 %shr.prol, -32768
%4 = select i1 %3, i32 %shr.prol, i32 -32768
%5 = icmp slt i32 %4, 32767
%spec.select.i.prol = select i1 %5, i32 %4, i32 32767
%conv3.prol = trunc i32 %spec.select.i.prol to i16
%incdec.ptr4.prol = getelementptr inbounds i16, ptr %pDst, i32 1
store i16 %conv3.prol, ptr %pDst
br label %while.body.prol.loopexit
while.body.prol.loopexit: ; preds = %while.body.prol.preheader, %while.body.preheader
%blkCnt.011.unr = phi i32 [ %blockSize, %while.body.preheader ], [ %0, %while.body.prol.preheader ]
%pSrcA.addr.010.unr = phi ptr [ %pSrcA, %while.body.preheader ], [ %incdec.ptr.prol, %while.body.prol.preheader ]
%pDst.addr.09.unr = phi ptr [ %pDst, %while.body.preheader ], [ %incdec.ptr4.prol, %while.body.prol.preheader ]
%pSrcB.addr.08.unr = phi ptr [ %pSrcB, %while.body.preheader ], [ %incdec.ptr1.prol, %while.body.prol.preheader ]
%6 = icmp eq i32 %0, 0
br i1 %6, label %while.end, label %while.body
while.body: ; preds = %while.body.prol.loopexit, %while.body
%blkCnt.011 = phi i32 [ %dec.1, %while.body ], [ %blkCnt.011.unr, %while.body.prol.loopexit ]
%pSrcA.addr.010 = phi ptr [ %incdec.ptr.1, %while.body ], [ %pSrcA.addr.010.unr, %while.body.prol.loopexit ]
%pDst.addr.09 = phi ptr [ %incdec.ptr4.1, %while.body ], [ %pDst.addr.09.unr, %while.body.prol.loopexit ]
%pSrcB.addr.08 = phi ptr [ %incdec.ptr1.1, %while.body ], [ %pSrcB.addr.08.unr, %while.body.prol.loopexit ]
%incdec.ptr = getelementptr inbounds i16, ptr %pSrcA.addr.010, i32 1
%7 = load i16, ptr %pSrcA.addr.010
%conv = sext i16 %7 to i32
%incdec.ptr1 = getelementptr inbounds i16, ptr %pSrcB.addr.08, i32 1
%8 = load i16, ptr %pSrcB.addr.08
%conv2 = sext i16 %8 to i32
%mul = mul nsw i32 %conv2, %conv
%shr = ashr i32 %mul, 14
%9 = icmp sgt i32 %shr, -32768
%10 = select i1 %9, i32 %shr, i32 -32768
%11 = icmp slt i32 %10, 32767
%spec.select.i = select i1 %11, i32 %10, i32 32767
%conv3 = trunc i32 %spec.select.i to i16
%incdec.ptr4 = getelementptr inbounds i16, ptr %pDst.addr.09, i32 1
store i16 %conv3, ptr %pDst.addr.09
%incdec.ptr.1 = getelementptr inbounds i16, ptr %pSrcA.addr.010, i32 2
%12 = load i16, ptr %incdec.ptr
%conv.1 = sext i16 %12 to i32
%incdec.ptr1.1 = getelementptr inbounds i16, ptr %pSrcB.addr.08, i32 2
%13 = load i16, ptr %incdec.ptr1
%conv2.1 = sext i16 %13 to i32
%mul.1 = mul nsw i32 %conv2.1, %conv.1
%shr.1 = ashr i32 %mul.1, 14
%14 = icmp sgt i32 %shr.1, -32768
%15 = select i1 %14, i32 %shr.1, i32 -32768
%16 = icmp slt i32 %15, 32767
%spec.select.i.1 = select i1 %16, i32 %15, i32 32767
%conv3.1 = trunc i32 %spec.select.i.1 to i16
%incdec.ptr4.1 = getelementptr inbounds i16, ptr %pDst.addr.09, i32 2
store i16 %conv3.1, ptr %incdec.ptr4
%dec.1 = add i32 %blkCnt.011, -2
%cmp.not.1 = icmp eq i32 %dec.1, 0
br i1 %cmp.not.1, label %while.end, label %while.body
while.end: ; preds = %while.body, %while.body.prol.loopexit, %entry
ret void
}
define void @ssat_unroll_minmax(ptr nocapture readonly %pSrcA, ptr nocapture readonly %pSrcB, ptr nocapture writeonly %pDst, i32 %blockSize) {
; CHECK-LABEL: ssat_unroll_minmax:
; CHECK: @ %bb.0: @ %entry
; CHECK-NEXT: cmp r3, #0
; CHECK-NEXT: bxeq lr
; CHECK-NEXT: .LBB1_1: @ %while.body.preheader
; CHECK-NEXT: .save {r11, lr}
; CHECK-NEXT: push {r11, lr}
; CHECK-NEXT: sub r12, r3, #1
; CHECK-NEXT: tst r3, #1
; CHECK-NEXT: beq .LBB1_3
; CHECK-NEXT: @ %bb.2: @ %while.body.prol.preheader
; CHECK-NEXT: ldrsh lr, [r0], #2
; CHECK-NEXT: ldrsh r3, [r1], #2
; CHECK-NEXT: smulbb r3, r3, lr
; CHECK-NEXT: ssat r3, #16, r3, asr #14
; CHECK-NEXT: strh r3, [r2], #2
; CHECK-NEXT: mov r3, r12
; CHECK-NEXT: .LBB1_3: @ %while.body.prol.loopexit
; CHECK-NEXT: cmp r12, #0
; CHECK-NEXT: beq .LBB1_5
; CHECK-NEXT: .LBB1_4: @ %while.body
; CHECK-NEXT: @ =>This Inner Loop Header: Depth=1
; CHECK-NEXT: ldrsh r12, [r0]
; CHECK-NEXT: subs r3, r3, #2
; CHECK-NEXT: ldrsh lr, [r1]
; CHECK-NEXT: smulbb r12, lr, r12
; CHECK-NEXT: ssat r12, #16, r12, asr #14
; CHECK-NEXT: strh r12, [r2]
; CHECK-NEXT: ldrsh r12, [r0, #2]
; CHECK-NEXT: add r0, r0, #4
; CHECK-NEXT: ldrsh lr, [r1, #2]
; CHECK-NEXT: add r1, r1, #4
; CHECK-NEXT: smulbb r12, lr, r12
; CHECK-NEXT: ssat r12, #16, r12, asr #14
; CHECK-NEXT: strh r12, [r2, #2]
; CHECK-NEXT: add r2, r2, #4
; CHECK-NEXT: bne .LBB1_4
; CHECK-NEXT: .LBB1_5:
; CHECK-NEXT: pop {r11, lr}
; CHECK-NEXT: bx lr
entry:
%cmp.not7 = icmp eq i32 %blockSize, 0
br i1 %cmp.not7, label %while.end, label %while.body.preheader
while.body.preheader: ; preds = %entry
%0 = add i32 %blockSize, -1
%xtraiter = and i32 %blockSize, 1
%lcmp.mod.not = icmp eq i32 %xtraiter, 0
br i1 %lcmp.mod.not, label %while.body.prol.loopexit, label %while.body.prol.preheader
while.body.prol.preheader: ; preds = %while.body.preheader
%incdec.ptr.prol = getelementptr inbounds i16, ptr %pSrcA, i64 1
%1 = load i16, ptr %pSrcA, align 2
%conv.prol = sext i16 %1 to i32
%incdec.ptr1.prol = getelementptr inbounds i16, ptr %pSrcB, i64 1
%2 = load i16, ptr %pSrcB, align 2
%conv2.prol = sext i16 %2 to i32
%mul.prol = mul nsw i32 %conv2.prol, %conv.prol
%shr.prol = ashr i32 %mul.prol, 14
%3 = call i32 @llvm.smax.i32(i32 %shr.prol, i32 -32768)
%4 = call i32 @llvm.smin.i32(i32 %3, i32 32767)
%conv3.prol = trunc i32 %4 to i16
%incdec.ptr4.prol = getelementptr inbounds i16, ptr %pDst, i64 1
store i16 %conv3.prol, ptr %pDst, align 2
br label %while.body.prol.loopexit
while.body.prol.loopexit: ; preds = %while.body.prol.preheader, %while.body.preheader
%blkCnt.011.unr = phi i32 [ %blockSize, %while.body.preheader ], [ %0, %while.body.prol.preheader ]
%pSrcA.addr.010.unr = phi ptr [ %pSrcA, %while.body.preheader ], [ %incdec.ptr.prol, %while.body.prol.preheader ]
%pDst.addr.09.unr = phi ptr [ %pDst, %while.body.preheader ], [ %incdec.ptr4.prol, %while.body.prol.preheader ]
%pSrcB.addr.08.unr = phi ptr [ %pSrcB, %while.body.preheader ], [ %incdec.ptr1.prol, %while.body.prol.preheader ]
%5 = icmp eq i32 %0, 0
br i1 %5, label %while.end, label %while.body
while.body: ; preds = %while.body.prol.loopexit, %while.body
%blkCnt.011 = phi i32 [ %dec.1, %while.body ], [ %blkCnt.011.unr, %while.body.prol.loopexit ]
%pSrcA.addr.010 = phi ptr [ %incdec.ptr.1, %while.body ], [ %pSrcA.addr.010.unr, %while.body.prol.loopexit ]
%pDst.addr.09 = phi ptr [ %incdec.ptr4.1, %while.body ], [ %pDst.addr.09.unr, %while.body.prol.loopexit ]
%pSrcB.addr.08 = phi ptr [ %incdec.ptr1.1, %while.body ], [ %pSrcB.addr.08.unr, %while.body.prol.loopexit ]
%incdec.ptr = getelementptr inbounds i16, ptr %pSrcA.addr.010, i64 1
%6 = load i16, ptr %pSrcA.addr.010, align 2
%conv = sext i16 %6 to i32
%incdec.ptr1 = getelementptr inbounds i16, ptr %pSrcB.addr.08, i64 1
%7 = load i16, ptr %pSrcB.addr.08, align 2
%conv2 = sext i16 %7 to i32
%mul = mul nsw i32 %conv2, %conv
%shr = ashr i32 %mul, 14
%8 = call i32 @llvm.smax.i32(i32 %shr, i32 -32768)
%9 = call i32 @llvm.smin.i32(i32 %8, i32 32767)
%conv3 = trunc i32 %9 to i16
%incdec.ptr4 = getelementptr inbounds i16, ptr %pDst.addr.09, i64 1
store i16 %conv3, ptr %pDst.addr.09, align 2
%incdec.ptr.1 = getelementptr inbounds i16, ptr %pSrcA.addr.010, i64 2
%10 = load i16, ptr %incdec.ptr, align 2
%conv.1 = sext i16 %10 to i32
%incdec.ptr1.1 = getelementptr inbounds i16, ptr %pSrcB.addr.08, i64 2
%11 = load i16, ptr %incdec.ptr1, align 2
%conv2.1 = sext i16 %11 to i32
%mul.1 = mul nsw i32 %conv2.1, %conv.1
%shr.1 = ashr i32 %mul.1, 14
%12 = call i32 @llvm.smax.i32(i32 %shr.1, i32 -32768)
%13 = call i32 @llvm.smin.i32(i32 %12, i32 32767)
%conv3.1 = trunc i32 %13 to i16
%incdec.ptr4.1 = getelementptr inbounds i16, ptr %pDst.addr.09, i64 2
store i16 %conv3.1, ptr %incdec.ptr4, align 2
%dec.1 = add i32 %blkCnt.011, -2
%cmp.not.1 = icmp eq i32 %dec.1, 0
br i1 %cmp.not.1, label %while.end, label %while.body
while.end: ; preds = %while.body, %while.body.prol.loopexit, %entry
ret void
}
declare i32 @llvm.smax.i32(i32, i32) #1
declare i32 @llvm.smin.i32(i32, i32) #1
|
