Skip to content

Commit fcf02bc

Browse files
LiqinWengLiqinWeng
committed
[LV][EVL][Test] Prepare test for adding intrinsics of call Recipe
1 parent 24ac6cf commit fcf02bc

File tree

1 file changed

+223
-0
lines changed

1 file changed

+223
-0
lines changed

‎llvm/test/Transforms/LoopVectorize/RISCV/vplan-vp-call-intrinsics.ll‎

Lines changed: 223 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,223 @@
1+
; REQUIRES: asserts
2+
3+
; RUN: opt -passes=loop-vectorize -debug-only=loop-vectorize \
4+
; RUN: -force-tail-folding-style=data-with-evl \
5+
; RUN: -prefer-predicate-over-epilogue=predicate-dont-vectorize \
6+
; RUN: -mtriple=riscv64 -mattr=+v -riscv-v-vector-bits-max=128 -disable-output < %s 2>&1 | FileCheck --check-prefix=IF-EVL %s
7+
8+
define void @vp_smax(ptr noalias %a, ptr noalias %b, ptr noalias %c, i64 %N) {
9+
; IF-EVL: VPlan 'Initial VPlan for VF={vscale x 1,vscale x 2,vscale x 4},UF={1}' {
10+
; IF-EVL-NEXT: Live-in vp<[[VFUF:%[0-9]+]]> = VF * UF
11+
; IF-EVL-NEXT: Live-in vp<[[VTC:%[0-9]+]]> = vector-trip-count
12+
; IF-EVL-NEXT: Live-in ir<%N> = original trip-count
13+
14+
; IF-EVL: vector.ph:
15+
; IF-EVL-NEXT: Successor(s): vector loop
16+
17+
; IF-EVL: <x1> vector loop: {
18+
; IF-EVL-NEXT: vector.body:
19+
; IF-EVL-NEXT: EMIT vp<[[IV:%[0-9]+]]> = CANONICAL-INDUCTION
20+
; IF-EVL-NEXT: EXPLICIT-VECTOR-LENGTH-BASED-IV-PHI vp<[[EVL_PHI:%[0-9]+]]> = phi ir<0>, vp<[[IV_NEXT:%[0-9]+]]>
21+
; IF-EVL-NEXT: EMIT vp<[[AVL:%.+]]> = sub ir<%N>, vp<[[EVL_PHI]]>
22+
; IF-EVL-NEXT: EMIT vp<[[EVL:%.+]]> = EXPLICIT-VECTOR-LENGTH vp<[[AVL]]>
23+
; IF-EVL-NEXT: vp<[[ST:%[0-9]+]]> = SCALAR-STEPS vp<[[EVL_PHI]]>, ir<1>
24+
; IF-EVL-NEXT: CLONE ir<[[GEP1:%.+]]> = getelementptr inbounds ir<%b>, vp<[[ST]]>
25+
; IF-EVL-NEXT: vp<[[PTR1:%[0-9]+]]> = vector-pointer ir<[[GEP1]]>
26+
; IF-EVL-NEXT: WIDEN ir<[[LD1:%.+]]> = vp.load vp<[[PTR1]]>, vp<[[EVL]]>
27+
; IF-EVL-NEXT: CLONE ir<[[GEP2:%.+]]> = getelementptr inbounds ir<%c>, vp<[[ST]]>
28+
; IF-EVL-NEXT: vp<[[PTR2:%[0-9]+]]> = vector-pointer ir<[[GEP2]]>
29+
; IF-EVL-NEXT: WIDEN ir<[[LD2:%.+]]> = vp.load vp<[[PTR2]]>, vp<[[EVL]]>
30+
; IF-EVL-NEXT: WIDEN-INTRINSIC ir<[[SMAX:%.+]]> = call llvm.smax(ir<[[LD1]]>, ir<[[LD2]]>)
31+
; IF-EVL-NEXT: CLONE ir<[[GEP3:%.+]]> = getelementptr inbounds ir<%a>, vp<[[ST]]>
32+
; IF-EVL-NEXT: vp<[[PTR3:%[0-9]+]]> = vector-pointer ir<[[GEP3]]>
33+
; IF-EVL-NEXT: WIDEN vp.store vp<[[PTR3]]>, ir<[[SMAX]]>, vp<[[EVL]]>
34+
; IF-EVL-NEXT: SCALAR-CAST vp<[[CAST:%[0-9]+]]> = zext vp<[[EVL]]> to i64
35+
; IF-EVL-NEXT: EMIT vp<[[IV_NEXT]]> = add vp<[[CAST]]>, vp<[[EVL_PHI]]>
36+
; IF-EVL-NEXT: EMIT vp<[[IV_NEXT_EXIT:%[0-9]+]]> = add vp<[[IV]]>, vp<[[VFUF]]>
37+
; IF-EVL-NEXT: EMIT branch-on-count vp<[[IV_NEXT_EXIT]]>, vp<[[VTC]]>
38+
; IF-EVL-NEXT: No successors
39+
; IF-EVL-NEXT: }
40+
41+
entry:
42+
br label %for.body
43+
44+
for.body:
45+
%indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ]
46+
%arrayidx = getelementptr inbounds i32, ptr %b, i64 %indvars.iv
47+
%0 = load i32, ptr %arrayidx, align 4
48+
%arrayidx3 = getelementptr inbounds i32, ptr %c, i64 %indvars.iv
49+
%1 = load i32, ptr %arrayidx3, align 4
50+
%. = tail call i32 @llvm.smax.i32(i32 %0, i32 %1)
51+
%arrayidx11 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv
52+
store i32 %., ptr %arrayidx11, align 4
53+
%indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
54+
%exitcond.not = icmp eq i64 %indvars.iv.next, %N
55+
br i1 %exitcond.not, label %exit, label %for.body
56+
57+
exit:
58+
ret void
59+
}
60+
61+
define void @vp_smin(ptr noalias %a, ptr noalias %b, ptr noalias %c, i64 %N) {
62+
; IF-EVL: VPlan 'Initial VPlan for VF={vscale x 1,vscale x 2,vscale x 4},UF={1}' {
63+
; IF-EVL-NEXT: Live-in vp<[[VFUF:%[0-9]+]]> = VF * UF
64+
; IF-EVL-NEXT: Live-in vp<[[VTC:%[0-9]+]]> = vector-trip-count
65+
; IF-EVL-NEXT: Live-in ir<%N> = original trip-count
66+
67+
; IF-EVL: vector.ph:
68+
; IF-EVL-NEXT: Successor(s): vector loop
69+
70+
; IF-EVL: <x1> vector loop: {
71+
; IF-EVL-NEXT: vector.body:
72+
; IF-EVL-NEXT: EMIT vp<[[IV:%[0-9]+]]> = CANONICAL-INDUCTION
73+
; IF-EVL-NEXT: EXPLICIT-VECTOR-LENGTH-BASED-IV-PHI vp<[[EVL_PHI:%[0-9]+]]> = phi ir<0>, vp<[[IV_NEXT:%[0-9]+]]>
74+
; IF-EVL-NEXT: EMIT vp<[[AVL:%.+]]> = sub ir<%N>, vp<[[EVL_PHI]]>
75+
; IF-EVL-NEXT: EMIT vp<[[EVL:%.+]]> = EXPLICIT-VECTOR-LENGTH vp<[[AVL]]>
76+
; IF-EVL-NEXT: vp<[[ST:%[0-9]+]]> = SCALAR-STEPS vp<[[EVL_PHI]]>, ir<1>
77+
; IF-EVL-NEXT: CLONE ir<[[GEP1:%.+]]> = getelementptr inbounds ir<%b>, vp<[[ST]]>
78+
; IF-EVL-NEXT: vp<[[PTR1:%[0-9]+]]> = vector-pointer ir<[[GEP1]]>
79+
; IF-EVL-NEXT: WIDEN ir<[[LD1:%.+]]> = vp.load vp<[[PTR1]]>, vp<[[EVL]]>
80+
; IF-EVL-NEXT: CLONE ir<[[GEP2:%.+]]> = getelementptr inbounds ir<%c>, vp<[[ST]]>
81+
; IF-EVL-NEXT: vp<[[PTR2:%[0-9]+]]> = vector-pointer ir<[[GEP2]]>
82+
; IF-EVL-NEXT: WIDEN ir<[[LD2:%.+]]> = vp.load vp<[[PTR2]]>, vp<[[EVL]]>
83+
; IF-EVL-NEXT: WIDEN-INTRINSIC ir<[[SMIN:%.+]]> = call llvm.smin(ir<[[LD1]]>, ir<[[LD2]]>)
84+
; IF-EVL-NEXT: CLONE ir<[[GEP3:%.+]]> = getelementptr inbounds ir<%a>, vp<[[ST]]>
85+
; IF-EVL-NEXT: vp<[[PTR3:%[0-9]+]]> = vector-pointer ir<[[GEP3]]>
86+
; IF-EVL-NEXT: WIDEN vp.store vp<[[PTR3]]>, ir<[[SMIN]]>, vp<[[EVL]]>
87+
; IF-EVL-NEXT: SCALAR-CAST vp<[[CAST:%[0-9]+]]> = zext vp<[[EVL]]> to i64
88+
; IF-EVL-NEXT: EMIT vp<[[IV_NEXT]]> = add vp<[[CAST]]>, vp<[[EVL_PHI]]>
89+
; IF-EVL-NEXT: EMIT vp<[[IV_NEXT_EXIT:%[0-9]+]]> = add vp<[[IV]]>, vp<[[VFUF]]>
90+
; IF-EVL-NEXT: EMIT branch-on-count vp<[[IV_NEXT_EXIT]]>, vp<[[VTC]]>
91+
; IF-EVL-NEXT: No successors
92+
; IF-EVL-NEXT: }
93+
94+
entry:
95+
br label %for.body
96+
97+
for.body:
98+
%indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ]
99+
%arrayidx = getelementptr inbounds i32, ptr %b, i64 %indvars.iv
100+
%0 = load i32, ptr %arrayidx, align 4
101+
%arrayidx3 = getelementptr inbounds i32, ptr %c, i64 %indvars.iv
102+
%1 = load i32, ptr %arrayidx3, align 4
103+
%. = tail call i32 @llvm.smin.i32(i32 %0, i32 %1)
104+
%arrayidx11 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv
105+
store i32 %., ptr %arrayidx11, align 4
106+
%indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
107+
%exitcond.not = icmp eq i64 %indvars.iv.next, %N
108+
br i1 %exitcond.not, label %exit, label %for.body
109+
110+
exit:
111+
ret void
112+
}
113+
114+
define void @vp_umax(ptr noalias %a, ptr noalias %b, ptr noalias %c, i64 %N) {
115+
; IF-EVL: VPlan 'Initial VPlan for VF={vscale x 1,vscale x 2,vscale x 4},UF={1}' {
116+
; IF-EVL-NEXT: Live-in vp<[[VFUF:%[0-9]+]]> = VF * UF
117+
; IF-EVL-NEXT: Live-in vp<[[VTC:%[0-9]+]]> = vector-trip-count
118+
; IF-EVL-NEXT: Live-in ir<%N> = original trip-count
119+
120+
; IF-EVL: vector.ph:
121+
; IF-EVL-NEXT: Successor(s): vector loop
122+
123+
; IF-EVL: <x1> vector loop: {
124+
; IF-EVL-NEXT: vector.body:
125+
; IF-EVL-NEXT: EMIT vp<[[IV:%[0-9]+]]> = CANONICAL-INDUCTION
126+
; IF-EVL-NEXT: EXPLICIT-VECTOR-LENGTH-BASED-IV-PHI vp<[[EVL_PHI:%[0-9]+]]> = phi ir<0>, vp<[[IV_NEXT:%[0-9]+]]>
127+
; IF-EVL-NEXT: EMIT vp<[[AVL:%.+]]> = sub ir<%N>, vp<[[EVL_PHI]]>
128+
; IF-EVL-NEXT: EMIT vp<[[EVL:%.+]]> = EXPLICIT-VECTOR-LENGTH vp<[[AVL]]>
129+
; IF-EVL-NEXT: vp<[[ST:%[0-9]+]]> = SCALAR-STEPS vp<[[EVL_PHI]]>, ir<1>
130+
; IF-EVL-NEXT: CLONE ir<[[GEP1:%.+]]> = getelementptr inbounds ir<%b>, vp<[[ST]]>
131+
; IF-EVL-NEXT: vp<[[PTR1:%[0-9]+]]> = vector-pointer ir<[[GEP1]]>
132+
; IF-EVL-NEXT: WIDEN ir<[[LD1:%.+]]> = vp.load vp<[[PTR1]]>, vp<[[EVL]]>
133+
; IF-EVL-NEXT: CLONE ir<[[GEP2:%.+]]> = getelementptr inbounds ir<%c>, vp<[[ST]]>
134+
; IF-EVL-NEXT: vp<[[PTR2:%[0-9]+]]> = vector-pointer ir<[[GEP2]]>
135+
; IF-EVL-NEXT: WIDEN ir<[[LD2:%.+]]> = vp.load vp<[[PTR2]]>, vp<[[EVL]]>
136+
; IF-EVL-NEXT: WIDEN-INTRINSIC ir<[[UMAX:%.+]]> = call llvm.umax(ir<[[LD1]]>, ir<[[LD2]]>)
137+
; IF-EVL-NEXT: CLONE ir<[[GEP3:%.+]]> = getelementptr inbounds ir<%a>, vp<[[ST]]>
138+
; IF-EVL-NEXT: vp<[[PTR3:%[0-9]+]]> = vector-pointer ir<[[GEP3]]>
139+
; IF-EVL-NEXT: WIDEN vp.store vp<[[PTR3]]>, ir<[[UMAX]]>, vp<[[EVL]]>
140+
; IF-EVL-NEXT: SCALAR-CAST vp<[[CAST:%[0-9]+]]> = zext vp<[[EVL]]> to i64
141+
; IF-EVL-NEXT: EMIT vp<[[IV_NEXT]]> = add vp<[[CAST]]>, vp<[[EVL_PHI]]>
142+
; IF-EVL-NEXT: EMIT vp<[[IV_NEXT_EXIT:%[0-9]+]]> = add vp<[[IV]]>, vp<[[VFUF]]>
143+
; IF-EVL-NEXT: EMIT branch-on-count vp<[[IV_NEXT_EXIT]]>, vp<[[VTC]]>
144+
; IF-EVL-NEXT: No successors
145+
; IF-EVL-NEXT: }
146+
147+
entry:
148+
br label %for.body
149+
150+
for.body:
151+
%indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ]
152+
%arrayidx = getelementptr inbounds i32, ptr %b, i64 %indvars.iv
153+
%0 = load i32, ptr %arrayidx, align 4
154+
%arrayidx3 = getelementptr inbounds i32, ptr %c, i64 %indvars.iv
155+
%1 = load i32, ptr %arrayidx3, align 4
156+
%. = tail call i32 @llvm.umax.i32(i32 %0, i32 %1)
157+
%arrayidx11 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv
158+
store i32 %., ptr %arrayidx11, align 4
159+
%indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
160+
%exitcond.not = icmp eq i64 %indvars.iv.next, %N
161+
br i1 %exitcond.not, label %exit, label %for.body
162+
163+
exit:
164+
ret void
165+
}
166+
167+
define void @vp_umin(ptr noalias %a, ptr noalias %b, ptr noalias %c, i64 %N) {
168+
; IF-EVL: VPlan 'Initial VPlan for VF={vscale x 1,vscale x 2,vscale x 4},UF={1}' {
169+
; IF-EVL-NEXT: Live-in vp<[[VFUF:%[0-9]+]]> = VF * UF
170+
; IF-EVL-NEXT: Live-in vp<[[VTC:%[0-9]+]]> = vector-trip-count
171+
; IF-EVL-NEXT: Live-in ir<%N> = original trip-count
172+
173+
; IF-EVL: vector.ph:
174+
; IF-EVL-NEXT: Successor(s): vector loop
175+
176+
; IF-EVL: <x1> vector loop: {
177+
; IF-EVL-NEXT: vector.body:
178+
; IF-EVL-NEXT: EMIT vp<[[IV:%[0-9]+]]> = CANONICAL-INDUCTION
179+
; IF-EVL-NEXT: EXPLICIT-VECTOR-LENGTH-BASED-IV-PHI vp<[[EVL_PHI:%[0-9]+]]> = phi ir<0>, vp<[[IV_NEXT:%[0-9]+]]>
180+
; IF-EVL-NEXT: EMIT vp<[[AVL:%.+]]> = sub ir<%N>, vp<[[EVL_PHI]]>
181+
; IF-EVL-NEXT: EMIT vp<[[EVL:%.+]]> = EXPLICIT-VECTOR-LENGTH vp<[[AVL]]>
182+
; IF-EVL-NEXT: vp<[[ST:%[0-9]+]]> = SCALAR-STEPS vp<[[EVL_PHI]]>, ir<1>
183+
; IF-EVL-NEXT: CLONE ir<[[GEP1:%.+]]> = getelementptr inbounds ir<%b>, vp<[[ST]]>
184+
; IF-EVL-NEXT: vp<[[PTR1:%[0-9]+]]> = vector-pointer ir<[[GEP1]]>
185+
; IF-EVL-NEXT: WIDEN ir<[[LD1:%.+]]> = vp.load vp<[[PTR1]]>, vp<[[EVL]]>
186+
; IF-EVL-NEXT: CLONE ir<[[GEP2:%.+]]> = getelementptr inbounds ir<%c>, vp<[[ST]]>
187+
; IF-EVL-NEXT: vp<[[PTR2:%[0-9]+]]> = vector-pointer ir<[[GEP2]]>
188+
; IF-EVL-NEXT: WIDEN ir<[[LD2:%.+]]> = vp.load vp<[[PTR2]]>, vp<[[EVL]]>
189+
; IF-EVL-NEXT: WIDEN-INTRINSIC ir<[[UMIN:%.+]]> = call llvm.umin(ir<[[LD1]]>, ir<[[LD2]]>)
190+
; IF-EVL-NEXT: CLONE ir<[[GEP3:%.+]]> = getelementptr inbounds ir<%a>, vp<[[ST]]>
191+
; IF-EVL-NEXT: vp<[[PTR3:%[0-9]+]]> = vector-pointer ir<[[GEP3]]>
192+
; IF-EVL-NEXT: WIDEN vp.store vp<[[PTR3]]>, ir<[[UMIN]]>, vp<[[EVL]]>
193+
; IF-EVL-NEXT: SCALAR-CAST vp<[[CAST:%[0-9]+]]> = zext vp<[[EVL]]> to i64
194+
; IF-EVL-NEXT: EMIT vp<[[IV_NEXT]]> = add vp<[[CAST]]>, vp<[[EVL_PHI]]>
195+
; IF-EVL-NEXT: EMIT vp<[[IV_NEXT_EXIT:%[0-9]+]]> = add vp<[[IV]]>, vp<[[VFUF]]>
196+
; IF-EVL-NEXT: EMIT branch-on-count vp<[[IV_NEXT_EXIT]]>, vp<[[VTC]]>
197+
; IF-EVL-NEXT: No successors
198+
; IF-EVL-NEXT: }
199+
200+
entry:
201+
br label %for.body
202+
203+
for.body:
204+
%indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ]
205+
%arrayidx = getelementptr inbounds i32, ptr %b, i64 %indvars.iv
206+
%0 = load i32, ptr %arrayidx, align 4
207+
%arrayidx3 = getelementptr inbounds i32, ptr %c, i64 %indvars.iv
208+
%1 = load i32, ptr %arrayidx3, align 4
209+
%. = tail call i32 @llvm.umin.i32(i32 %0, i32 %1)
210+
%arrayidx11 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv
211+
store i32 %., ptr %arrayidx11, align 4
212+
%indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
213+
%exitcond.not = icmp eq i64 %indvars.iv.next, %N
214+
br i1 %exitcond.not, label %exit, label %for.body
215+
216+
exit:
217+
ret void
218+
}
219+
220+
declare i32 @llvm.smax.i32(i32, i32)
221+
declare i32 @llvm.smin.i32(i32, i32)
222+
declare i32 @llvm.umax.i32(i32, i32)
223+
declare i32 @llvm.umin.i32(i32, i32)

0 commit comments

Comments
 (0)