Skip to content

Commit 32ae4a5

Browse files
Martien de Jongmartien-de-jong
Martien de Jong
authored and
martien-de-jong
committed
[AIE] legalize saturating arith ops by lowering
1 parent 9d80537 commit 32ae4a5

File tree

2 files changed

+140
-2
lines changed

2 files changed

+140
-2
lines changed

‎llvm/lib/Target/AIE/AIELegalizerInfo.cpp‎

Lines changed: 3 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -294,8 +294,9 @@ AIELegalizerInfo::AIELegalizerInfo(const AIEBaseSubtarget &ST) {
294294
// FIXME: G_SADDE/G_SSUBE doesn't support lowering. To support this properly,
295295
// the action needs to be implemented
296296
// FIXME: AIE2 has ADC and SBC operations to read the carry.
297-
getActionDefinitionsBuilder(
298-
{G_UADDO, G_USUBO, G_UADDE, G_USUBE, G_SADDO, G_SSUBO, G_SADDE, G_SSUBE})
297+
getActionDefinitionsBuilder({G_UADDO, G_USUBO, G_UADDE, G_USUBE, G_SADDO,
298+
G_SSUBO, G_SADDE, G_SSUBE, G_UADDSAT, G_USUBSAT,
299+
G_SADDSAT, G_SSUBSAT})
299300
.lower();
300301

301302
getActionDefinitionsBuilder(G_MUL)

‎llvm/test/CodeGen/AIE/GlobalISel/legalize-satops.mir‎

Lines changed: 137 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,137 @@
1+
# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
2+
#
3+
# This file is licensed under the Apache License v2.0 with LLVM Exceptions.
4+
# See https://llvm.org/LICENSE.txt for license information.
5+
# SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6+
#
7+
# (c) Copyright 2024 Advanced Micro Devices, Inc. or its affiliates
8+
9+
# RUN: llc -mtriple aie2 -run-pass=legalizer %s -verify-machineinstrs -o - \
10+
# RUN: | FileCheck %s
11+
12+
# This test checks legalization of the saturated operations for ADD and SUB
13+
# in signed and unsigned variants
14+
15+
---
16+
name: uaddsat
17+
body: |
18+
; CHECK-LABEL: name: uaddsat
19+
; CHECK: bb.0:
20+
; CHECK-NEXT: successors: %bb.1(0x80000000)
21+
; CHECK-NEXT: liveins: $r0, $r1
22+
; CHECK-NEXT: {{ $}}
23+
; CHECK-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $r0
24+
; CHECK-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $r1
25+
; CHECK-NEXT: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[COPY]], [[COPY1]]
26+
; CHECK-NEXT: [[ICMP:%[0-9]+]]:_(s32) = G_ICMP intpred(ult), [[ADD]](s32), [[COPY1]]
27+
; CHECK-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 -1
28+
; CHECK-NEXT: [[SELECT:%[0-9]+]]:_(s32) = G_SELECT [[ICMP]](s32), [[C]], [[ADD]]
29+
; CHECK-NEXT: $r0 = COPY [[SELECT]](s32)
30+
; CHECK-NEXT: {{ $}}
31+
; CHECK-NEXT: bb.1:
32+
; CHECK-NEXT: PseudoRET implicit $lr, implicit $r0
33+
bb.1:
34+
liveins: $r0, $r1
35+
%3:_(s32) = COPY $r0
36+
%4:_(s32) = COPY $r1
37+
%5:_(s32) = G_UADDSAT %3, %4
38+
$r0 = COPY %5
39+
bb.2:
40+
PseudoRET implicit $lr, implicit $r0
41+
...
42+
43+
---
44+
name: usubsat
45+
body: |
46+
; CHECK-LABEL: name: usubsat
47+
; CHECK: bb.0:
48+
; CHECK-NEXT: successors: %bb.1(0x80000000)
49+
; CHECK-NEXT: liveins: $r0, $r1
50+
; CHECK-NEXT: {{ $}}
51+
; CHECK-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $r0
52+
; CHECK-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $r1
53+
; CHECK-NEXT: [[SUB:%[0-9]+]]:_(s32) = G_SUB [[COPY]], [[COPY1]]
54+
; CHECK-NEXT: [[ICMP:%[0-9]+]]:_(s32) = G_ICMP intpred(ult), [[COPY]](s32), [[COPY1]]
55+
; CHECK-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
56+
; CHECK-NEXT: [[SELECT:%[0-9]+]]:_(s32) = G_SELECT [[ICMP]](s32), [[C]], [[SUB]]
57+
; CHECK-NEXT: $r0 = COPY [[SELECT]](s32)
58+
; CHECK-NEXT: {{ $}}
59+
; CHECK-NEXT: bb.1:
60+
; CHECK-NEXT: PseudoRET implicit $lr, implicit $r0
61+
bb.1:
62+
liveins: $r0, $r1
63+
%3:_(s32) = COPY $r0
64+
%4:_(s32) = COPY $r1
65+
%5:_(s32) = G_USUBSAT %3, %4
66+
$r0 = COPY %5
67+
bb.2:
68+
PseudoRET implicit $lr, implicit $r0
69+
...
70+
71+
---
72+
name: saddsat
73+
body: |
74+
; CHECK-LABEL: name: saddsat
75+
; CHECK: bb.0:
76+
; CHECK-NEXT: successors: %bb.1(0x80000000)
77+
; CHECK-NEXT: liveins: $r0, $r1
78+
; CHECK-NEXT: {{ $}}
79+
; CHECK-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $r0
80+
; CHECK-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $r1
81+
; CHECK-NEXT: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[COPY]], [[COPY1]]
82+
; CHECK-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
83+
; CHECK-NEXT: [[ICMP:%[0-9]+]]:_(s32) = G_ICMP intpred(slt), [[ADD]](s32), [[COPY]]
84+
; CHECK-NEXT: [[ICMP1:%[0-9]+]]:_(s32) = G_ICMP intpred(slt), [[COPY1]](s32), [[C]]
85+
; CHECK-NEXT: [[XOR:%[0-9]+]]:_(s32) = G_XOR [[ICMP1]], [[ICMP]]
86+
; CHECK-NEXT: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 31
87+
; CHECK-NEXT: [[ASHR:%[0-9]+]]:_(s32) = G_ASHR [[ADD]], [[C1]](s32)
88+
; CHECK-NEXT: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 -2147483648
89+
; CHECK-NEXT: [[ADD1:%[0-9]+]]:_(s32) = G_ADD [[ASHR]], [[C2]]
90+
; CHECK-NEXT: [[SELECT:%[0-9]+]]:_(s32) = G_SELECT [[XOR]](s32), [[ADD1]], [[ADD]]
91+
; CHECK-NEXT: $r0 = COPY [[SELECT]](s32)
92+
; CHECK-NEXT: {{ $}}
93+
; CHECK-NEXT: bb.1:
94+
; CHECK-NEXT: PseudoRET implicit $lr, implicit $r0
95+
bb.1:
96+
liveins: $r0, $r1
97+
%3:_(s32) = COPY $r0
98+
%4:_(s32) = COPY $r1
99+
%5:_(s32) = G_SADDSAT %3, %4
100+
$r0 = COPY %5
101+
bb.2:
102+
PseudoRET implicit $lr, implicit $r0
103+
...
104+
105+
---
106+
name: ssubsat
107+
body: |
108+
; CHECK-LABEL: name: ssubsat
109+
; CHECK: bb.0:
110+
; CHECK-NEXT: successors: %bb.1(0x80000000)
111+
; CHECK-NEXT: liveins: $r0, $r1
112+
; CHECK-NEXT: {{ $}}
113+
; CHECK-NEXT: [[COPY:%[0-9]+]]:_(s32) = COPY $r0
114+
; CHECK-NEXT: [[COPY1:%[0-9]+]]:_(s32) = COPY $r1
115+
; CHECK-NEXT: [[SUB:%[0-9]+]]:_(s32) = G_SUB [[COPY]], [[COPY1]]
116+
; CHECK-NEXT: [[C:%[0-9]+]]:_(s32) = G_CONSTANT i32 0
117+
; CHECK-NEXT: [[ICMP:%[0-9]+]]:_(s32) = G_ICMP intpred(slt), [[SUB]](s32), [[COPY]]
118+
; CHECK-NEXT: [[ICMP1:%[0-9]+]]:_(s32) = G_ICMP intpred(sgt), [[COPY1]](s32), [[C]]
119+
; CHECK-NEXT: [[XOR:%[0-9]+]]:_(s32) = G_XOR [[ICMP1]], [[ICMP]]
120+
; CHECK-NEXT: [[C1:%[0-9]+]]:_(s32) = G_CONSTANT i32 31
121+
; CHECK-NEXT: [[ASHR:%[0-9]+]]:_(s32) = G_ASHR [[SUB]], [[C1]](s32)
122+
; CHECK-NEXT: [[C2:%[0-9]+]]:_(s32) = G_CONSTANT i32 -2147483648
123+
; CHECK-NEXT: [[ADD:%[0-9]+]]:_(s32) = G_ADD [[ASHR]], [[C2]]
124+
; CHECK-NEXT: [[SELECT:%[0-9]+]]:_(s32) = G_SELECT [[XOR]](s32), [[ADD]], [[SUB]]
125+
; CHECK-NEXT: $r0 = COPY [[SELECT]](s32)
126+
; CHECK-NEXT: {{ $}}
127+
; CHECK-NEXT: bb.1:
128+
; CHECK-NEXT: PseudoRET implicit $lr, implicit $r0
129+
bb.1:
130+
liveins: $r0, $r1
131+
%3:_(s32) = COPY $r0
132+
%4:_(s32) = COPY $r1
133+
%5:_(s32) = G_SSUBSAT %3, %4
134+
$r0 = COPY %5
135+
bb.2:
136+
PseudoRET implicit $lr, implicit $r0
137+
...

0 commit comments

Comments
 (0)