Skip to content

Commit 77d62e0

Browse files
ylzsxylzsx
committed
[LoongArch][NFC] Pre-commit tests for vector ceil,floor,trunc,roundeven
1 parent d162025 commit 77d62e0

File tree

2 files changed

+520
-0
lines changed

2 files changed

+520
-0
lines changed

‎llvm/test/CodeGen/LoongArch/lasx/fp-rounding.ll‎

Lines changed: 308 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,308 @@
1+
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
2+
; RUN: llc --mtriple=loongarch32 --mattr=+32s,+lasx < %s | FileCheck %s --check-prefix=CHECK
3+
; RUN: llc --mtriple=loongarch64 --mattr=+lasx < %s | FileCheck %s --check-prefix=CHECK
4+
5+
;; ceilf
6+
define void @ceil_v8f32(ptr %res, ptr %a0) nounwind {
7+
; CHECK-LABEL: ceil_v8f32:
8+
; CHECK: # %bb.0: # %entry
9+
; CHECK-NEXT: xvld $xr0, $a1, 0
10+
; CHECK-NEXT: xvpickve.w $xr1, $xr0, 5
11+
; CHECK-NEXT: vreplvei.w $vr1, $vr1, 0
12+
; CHECK-NEXT: vfrintrp.s $vr1, $vr1
13+
; CHECK-NEXT: xvpickve.w $xr2, $xr0, 4
14+
; CHECK-NEXT: vreplvei.w $vr2, $vr2, 0
15+
; CHECK-NEXT: vfrintrp.s $vr2, $vr2
16+
; CHECK-NEXT: vextrins.w $vr2, $vr1, 16
17+
; CHECK-NEXT: xvpickve.w $xr1, $xr0, 6
18+
; CHECK-NEXT: vreplvei.w $vr1, $vr1, 0
19+
; CHECK-NEXT: vfrintrp.s $vr1, $vr1
20+
; CHECK-NEXT: vextrins.w $vr2, $vr1, 32
21+
; CHECK-NEXT: xvpickve.w $xr1, $xr0, 7
22+
; CHECK-NEXT: vreplvei.w $vr1, $vr1, 0
23+
; CHECK-NEXT: vfrintrp.s $vr1, $vr1
24+
; CHECK-NEXT: vextrins.w $vr2, $vr1, 48
25+
; CHECK-NEXT: xvpickve.w $xr1, $xr0, 1
26+
; CHECK-NEXT: vreplvei.w $vr1, $vr1, 0
27+
; CHECK-NEXT: vfrintrp.s $vr1, $vr1
28+
; CHECK-NEXT: xvpickve.w $xr3, $xr0, 0
29+
; CHECK-NEXT: vreplvei.w $vr3, $vr3, 0
30+
; CHECK-NEXT: vfrintrp.s $vr3, $vr3
31+
; CHECK-NEXT: vextrins.w $vr3, $vr1, 16
32+
; CHECK-NEXT: xvpickve.w $xr1, $xr0, 2
33+
; CHECK-NEXT: vreplvei.w $vr1, $vr1, 0
34+
; CHECK-NEXT: vfrintrp.s $vr1, $vr1
35+
; CHECK-NEXT: vextrins.w $vr3, $vr1, 32
36+
; CHECK-NEXT: xvpickve.w $xr0, $xr0, 3
37+
; CHECK-NEXT: vreplvei.w $vr0, $vr0, 0
38+
; CHECK-NEXT: vfrintrp.s $vr0, $vr0
39+
; CHECK-NEXT: vextrins.w $vr3, $vr0, 48
40+
; CHECK-NEXT: xvpermi.q $xr3, $xr2, 2
41+
; CHECK-NEXT: xvst $xr3, $a0, 0
42+
; CHECK-NEXT: ret
43+
entry:
44+
%v0 = load <8 x float>, ptr %a0
45+
%r = call <8 x float> @llvm.ceil.v8f32(<8 x float> %v0)
46+
store <8 x float> %r, ptr %res
47+
ret void
48+
}
49+
50+
;; ceil
51+
define void @ceil_v4f64(ptr %res, ptr %a0) nounwind {
52+
; CHECK-LABEL: ceil_v4f64:
53+
; CHECK: # %bb.0: # %entry
54+
; CHECK-NEXT: xvld $xr0, $a1, 0
55+
; CHECK-NEXT: xvpickve.d $xr1, $xr0, 3
56+
; CHECK-NEXT: vreplvei.d $vr1, $vr1, 0
57+
; CHECK-NEXT: vfrintrp.d $vr1, $vr1
58+
; CHECK-NEXT: xvpickve.d $xr2, $xr0, 2
59+
; CHECK-NEXT: vreplvei.d $vr2, $vr2, 0
60+
; CHECK-NEXT: vfrintrp.d $vr2, $vr2
61+
; CHECK-NEXT: vextrins.d $vr2, $vr1, 16
62+
; CHECK-NEXT: xvpickve.d $xr1, $xr0, 1
63+
; CHECK-NEXT: vreplvei.d $vr1, $vr1, 0
64+
; CHECK-NEXT: vfrintrp.d $vr1, $vr1
65+
; CHECK-NEXT: xvpickve.d $xr0, $xr0, 0
66+
; CHECK-NEXT: vreplvei.d $vr0, $vr0, 0
67+
; CHECK-NEXT: vfrintrp.d $vr0, $vr0
68+
; CHECK-NEXT: vextrins.d $vr0, $vr1, 16
69+
; CHECK-NEXT: xvpermi.q $xr0, $xr2, 2
70+
; CHECK-NEXT: xvst $xr0, $a0, 0
71+
; CHECK-NEXT: ret
72+
entry:
73+
%v0 = load <4 x double>, ptr %a0
74+
%r = call <4 x double> @llvm.ceil.v4f64(<4 x double> %v0)
75+
store <4 x double> %r, ptr %res
76+
ret void
77+
}
78+
79+
;; floorf
80+
define void @floor_v8f32(ptr %res, ptr %a0) nounwind {
81+
; CHECK-LABEL: floor_v8f32:
82+
; CHECK: # %bb.0: # %entry
83+
; CHECK-NEXT: xvld $xr0, $a1, 0
84+
; CHECK-NEXT: xvpickve.w $xr1, $xr0, 5
85+
; CHECK-NEXT: vreplvei.w $vr1, $vr1, 0
86+
; CHECK-NEXT: vfrintrm.s $vr1, $vr1
87+
; CHECK-NEXT: xvpickve.w $xr2, $xr0, 4
88+
; CHECK-NEXT: vreplvei.w $vr2, $vr2, 0
89+
; CHECK-NEXT: vfrintrm.s $vr2, $vr2
90+
; CHECK-NEXT: vextrins.w $vr2, $vr1, 16
91+
; CHECK-NEXT: xvpickve.w $xr1, $xr0, 6
92+
; CHECK-NEXT: vreplvei.w $vr1, $vr1, 0
93+
; CHECK-NEXT: vfrintrm.s $vr1, $vr1
94+
; CHECK-NEXT: vextrins.w $vr2, $vr1, 32
95+
; CHECK-NEXT: xvpickve.w $xr1, $xr0, 7
96+
; CHECK-NEXT: vreplvei.w $vr1, $vr1, 0
97+
; CHECK-NEXT: vfrintrm.s $vr1, $vr1
98+
; CHECK-NEXT: vextrins.w $vr2, $vr1, 48
99+
; CHECK-NEXT: xvpickve.w $xr1, $xr0, 1
100+
; CHECK-NEXT: vreplvei.w $vr1, $vr1, 0
101+
; CHECK-NEXT: vfrintrm.s $vr1, $vr1
102+
; CHECK-NEXT: xvpickve.w $xr3, $xr0, 0
103+
; CHECK-NEXT: vreplvei.w $vr3, $vr3, 0
104+
; CHECK-NEXT: vfrintrm.s $vr3, $vr3
105+
; CHECK-NEXT: vextrins.w $vr3, $vr1, 16
106+
; CHECK-NEXT: xvpickve.w $xr1, $xr0, 2
107+
; CHECK-NEXT: vreplvei.w $vr1, $vr1, 0
108+
; CHECK-NEXT: vfrintrm.s $vr1, $vr1
109+
; CHECK-NEXT: vextrins.w $vr3, $vr1, 32
110+
; CHECK-NEXT: xvpickve.w $xr0, $xr0, 3
111+
; CHECK-NEXT: vreplvei.w $vr0, $vr0, 0
112+
; CHECK-NEXT: vfrintrm.s $vr0, $vr0
113+
; CHECK-NEXT: vextrins.w $vr3, $vr0, 48
114+
; CHECK-NEXT: xvpermi.q $xr3, $xr2, 2
115+
; CHECK-NEXT: xvst $xr3, $a0, 0
116+
; CHECK-NEXT: ret
117+
entry:
118+
%v0 = load <8 x float>, ptr %a0
119+
%r = call <8 x float> @llvm.floor.v8f32(<8 x float> %v0)
120+
store <8 x float> %r, ptr %res
121+
ret void
122+
}
123+
124+
;; floor
125+
define void @floor_v4f64(ptr %res, ptr %a0) nounwind {
126+
; CHECK-LABEL: floor_v4f64:
127+
; CHECK: # %bb.0: # %entry
128+
; CHECK-NEXT: xvld $xr0, $a1, 0
129+
; CHECK-NEXT: xvpickve.d $xr1, $xr0, 3
130+
; CHECK-NEXT: vreplvei.d $vr1, $vr1, 0
131+
; CHECK-NEXT: vfrintrm.d $vr1, $vr1
132+
; CHECK-NEXT: xvpickve.d $xr2, $xr0, 2
133+
; CHECK-NEXT: vreplvei.d $vr2, $vr2, 0
134+
; CHECK-NEXT: vfrintrm.d $vr2, $vr2
135+
; CHECK-NEXT: vextrins.d $vr2, $vr1, 16
136+
; CHECK-NEXT: xvpickve.d $xr1, $xr0, 1
137+
; CHECK-NEXT: vreplvei.d $vr1, $vr1, 0
138+
; CHECK-NEXT: vfrintrm.d $vr1, $vr1
139+
; CHECK-NEXT: xvpickve.d $xr0, $xr0, 0
140+
; CHECK-NEXT: vreplvei.d $vr0, $vr0, 0
141+
; CHECK-NEXT: vfrintrm.d $vr0, $vr0
142+
; CHECK-NEXT: vextrins.d $vr0, $vr1, 16
143+
; CHECK-NEXT: xvpermi.q $xr0, $xr2, 2
144+
; CHECK-NEXT: xvst $xr0, $a0, 0
145+
; CHECK-NEXT: ret
146+
entry:
147+
%v0 = load <4 x double>, ptr %a0
148+
%r = call <4 x double> @llvm.floor.v4f64(<4 x double> %v0)
149+
store <4 x double> %r, ptr %res
150+
ret void
151+
}
152+
153+
;; truncf
154+
define void @trunc_v8f32(ptr %res, ptr %a0) nounwind {
155+
; CHECK-LABEL: trunc_v8f32:
156+
; CHECK: # %bb.0: # %entry
157+
; CHECK-NEXT: xvld $xr0, $a1, 0
158+
; CHECK-NEXT: xvpickve.w $xr1, $xr0, 5
159+
; CHECK-NEXT: vreplvei.w $vr1, $vr1, 0
160+
; CHECK-NEXT: vfrintrz.s $vr1, $vr1
161+
; CHECK-NEXT: xvpickve.w $xr2, $xr0, 4
162+
; CHECK-NEXT: vreplvei.w $vr2, $vr2, 0
163+
; CHECK-NEXT: vfrintrz.s $vr2, $vr2
164+
; CHECK-NEXT: vextrins.w $vr2, $vr1, 16
165+
; CHECK-NEXT: xvpickve.w $xr1, $xr0, 6
166+
; CHECK-NEXT: vreplvei.w $vr1, $vr1, 0
167+
; CHECK-NEXT: vfrintrz.s $vr1, $vr1
168+
; CHECK-NEXT: vextrins.w $vr2, $vr1, 32
169+
; CHECK-NEXT: xvpickve.w $xr1, $xr0, 7
170+
; CHECK-NEXT: vreplvei.w $vr1, $vr1, 0
171+
; CHECK-NEXT: vfrintrz.s $vr1, $vr1
172+
; CHECK-NEXT: vextrins.w $vr2, $vr1, 48
173+
; CHECK-NEXT: xvpickve.w $xr1, $xr0, 1
174+
; CHECK-NEXT: vreplvei.w $vr1, $vr1, 0
175+
; CHECK-NEXT: vfrintrz.s $vr1, $vr1
176+
; CHECK-NEXT: xvpickve.w $xr3, $xr0, 0
177+
; CHECK-NEXT: vreplvei.w $vr3, $vr3, 0
178+
; CHECK-NEXT: vfrintrz.s $vr3, $vr3
179+
; CHECK-NEXT: vextrins.w $vr3, $vr1, 16
180+
; CHECK-NEXT: xvpickve.w $xr1, $xr0, 2
181+
; CHECK-NEXT: vreplvei.w $vr1, $vr1, 0
182+
; CHECK-NEXT: vfrintrz.s $vr1, $vr1
183+
; CHECK-NEXT: vextrins.w $vr3, $vr1, 32
184+
; CHECK-NEXT: xvpickve.w $xr0, $xr0, 3
185+
; CHECK-NEXT: vreplvei.w $vr0, $vr0, 0
186+
; CHECK-NEXT: vfrintrz.s $vr0, $vr0
187+
; CHECK-NEXT: vextrins.w $vr3, $vr0, 48
188+
; CHECK-NEXT: xvpermi.q $xr3, $xr2, 2
189+
; CHECK-NEXT: xvst $xr3, $a0, 0
190+
; CHECK-NEXT: ret
191+
entry:
192+
%v0 = load <8 x float>, ptr %a0
193+
%r = call <8 x float> @llvm.trunc.v8f32(<8 x float> %v0)
194+
store <8 x float> %r, ptr %res
195+
ret void
196+
}
197+
198+
;; trunc
199+
define void @trunc_v4f64(ptr %res, ptr %a0) nounwind {
200+
; CHECK-LABEL: trunc_v4f64:
201+
; CHECK: # %bb.0: # %entry
202+
; CHECK-NEXT: xvld $xr0, $a1, 0
203+
; CHECK-NEXT: xvpickve.d $xr1, $xr0, 3
204+
; CHECK-NEXT: vreplvei.d $vr1, $vr1, 0
205+
; CHECK-NEXT: vfrintrz.d $vr1, $vr1
206+
; CHECK-NEXT: xvpickve.d $xr2, $xr0, 2
207+
; CHECK-NEXT: vreplvei.d $vr2, $vr2, 0
208+
; CHECK-NEXT: vfrintrz.d $vr2, $vr2
209+
; CHECK-NEXT: vextrins.d $vr2, $vr1, 16
210+
; CHECK-NEXT: xvpickve.d $xr1, $xr0, 1
211+
; CHECK-NEXT: vreplvei.d $vr1, $vr1, 0
212+
; CHECK-NEXT: vfrintrz.d $vr1, $vr1
213+
; CHECK-NEXT: xvpickve.d $xr0, $xr0, 0
214+
; CHECK-NEXT: vreplvei.d $vr0, $vr0, 0
215+
; CHECK-NEXT: vfrintrz.d $vr0, $vr0
216+
; CHECK-NEXT: vextrins.d $vr0, $vr1, 16
217+
; CHECK-NEXT: xvpermi.q $xr0, $xr2, 2
218+
; CHECK-NEXT: xvst $xr0, $a0, 0
219+
; CHECK-NEXT: ret
220+
entry:
221+
%v0 = load <4 x double>, ptr %a0
222+
%r = call <4 x double> @llvm.trunc.v4f64(<4 x double> %v0)
223+
store <4 x double> %r, ptr %res
224+
ret void
225+
}
226+
227+
;; roundevenf
228+
define void @roundeven_v8f32(ptr %res, ptr %a0) nounwind {
229+
; CHECK-LABEL: roundeven_v8f32:
230+
; CHECK: # %bb.0: # %entry
231+
; CHECK-NEXT: xvld $xr0, $a1, 0
232+
; CHECK-NEXT: xvpickve.w $xr1, $xr0, 5
233+
; CHECK-NEXT: vreplvei.w $vr1, $vr1, 0
234+
; CHECK-NEXT: vfrintrne.s $vr1, $vr1
235+
; CHECK-NEXT: xvpickve.w $xr2, $xr0, 4
236+
; CHECK-NEXT: vreplvei.w $vr2, $vr2, 0
237+
; CHECK-NEXT: vfrintrne.s $vr2, $vr2
238+
; CHECK-NEXT: vextrins.w $vr2, $vr1, 16
239+
; CHECK-NEXT: xvpickve.w $xr1, $xr0, 6
240+
; CHECK-NEXT: vreplvei.w $vr1, $vr1, 0
241+
; CHECK-NEXT: vfrintrne.s $vr1, $vr1
242+
; CHECK-NEXT: vextrins.w $vr2, $vr1, 32
243+
; CHECK-NEXT: xvpickve.w $xr1, $xr0, 7
244+
; CHECK-NEXT: vreplvei.w $vr1, $vr1, 0
245+
; CHECK-NEXT: vfrintrne.s $vr1, $vr1
246+
; CHECK-NEXT: vextrins.w $vr2, $vr1, 48
247+
; CHECK-NEXT: xvpickve.w $xr1, $xr0, 1
248+
; CHECK-NEXT: vreplvei.w $vr1, $vr1, 0
249+
; CHECK-NEXT: vfrintrne.s $vr1, $vr1
250+
; CHECK-NEXT: xvpickve.w $xr3, $xr0, 0
251+
; CHECK-NEXT: vreplvei.w $vr3, $vr3, 0
252+
; CHECK-NEXT: vfrintrne.s $vr3, $vr3
253+
; CHECK-NEXT: vextrins.w $vr3, $vr1, 16
254+
; CHECK-NEXT: xvpickve.w $xr1, $xr0, 2
255+
; CHECK-NEXT: vreplvei.w $vr1, $vr1, 0
256+
; CHECK-NEXT: vfrintrne.s $vr1, $vr1
257+
; CHECK-NEXT: vextrins.w $vr3, $vr1, 32
258+
; CHECK-NEXT: xvpickve.w $xr0, $xr0, 3
259+
; CHECK-NEXT: vreplvei.w $vr0, $vr0, 0
260+
; CHECK-NEXT: vfrintrne.s $vr0, $vr0
261+
; CHECK-NEXT: vextrins.w $vr3, $vr0, 48
262+
; CHECK-NEXT: xvpermi.q $xr3, $xr2, 2
263+
; CHECK-NEXT: xvst $xr3, $a0, 0
264+
; CHECK-NEXT: ret
265+
entry:
266+
%v0 = load <8 x float>, ptr %a0
267+
%r = call <8 x float> @llvm.roundeven.v8f32(<8 x float> %v0)
268+
store <8 x float> %r, ptr %res
269+
ret void
270+
}
271+
272+
;; roundeven
273+
define void @roundeven_v4f64(ptr %res, ptr %a0) nounwind {
274+
; CHECK-LABEL: roundeven_v4f64:
275+
; CHECK: # %bb.0: # %entry
276+
; CHECK-NEXT: xvld $xr0, $a1, 0
277+
; CHECK-NEXT: xvpickve.d $xr1, $xr0, 3
278+
; CHECK-NEXT: vreplvei.d $vr1, $vr1, 0
279+
; CHECK-NEXT: vfrintrne.d $vr1, $vr1
280+
; CHECK-NEXT: xvpickve.d $xr2, $xr0, 2
281+
; CHECK-NEXT: vreplvei.d $vr2, $vr2, 0
282+
; CHECK-NEXT: vfrintrne.d $vr2, $vr2
283+
; CHECK-NEXT: vextrins.d $vr2, $vr1, 16
284+
; CHECK-NEXT: xvpickve.d $xr1, $xr0, 1
285+
; CHECK-NEXT: vreplvei.d $vr1, $vr1, 0
286+
; CHECK-NEXT: vfrintrne.d $vr1, $vr1
287+
; CHECK-NEXT: xvpickve.d $xr0, $xr0, 0
288+
; CHECK-NEXT: vreplvei.d $vr0, $vr0, 0
289+
; CHECK-NEXT: vfrintrne.d $vr0, $vr0
290+
; CHECK-NEXT: vextrins.d $vr0, $vr1, 16
291+
; CHECK-NEXT: xvpermi.q $xr0, $xr2, 2
292+
; CHECK-NEXT: xvst $xr0, $a0, 0
293+
; CHECK-NEXT: ret
294+
entry:
295+
%v0 = load <4 x double>, ptr %a0
296+
%r = call <4 x double> @llvm.roundeven.v4f64(<4 x double> %v0)
297+
store <4 x double> %r, ptr %res
298+
ret void
299+
}
300+
301+
declare <8 x float> @llvm.ceil.v8f32(<8 x float>)
302+
declare <4 x double> @llvm.ceil.v4f64(<4 x double>)
303+
declare <8 x float> @llvm.floor.v8f32(<8 x float>)
304+
declare <4 x double> @llvm.floor.v4f64(<4 x double>)
305+
declare <8 x float> @llvm.trunc.v8f32(<8 x float>)
306+
declare <4 x double> @llvm.trunc.v4f64(<4 x double>)
307+
declare <8 x float> @llvm.roundeven.v8f32(<8 x float>)
308+
declare <4 x double> @llvm.roundeven.v4f64(<4 x double>)

0 commit comments

Comments
 (0)