Skip to content

Commit b3f8263

Browse files
chen.qiangerekon
chen.qian
authored and
gerekon
committed
[Test] add biquad test case for CustomLICM Pass
1 parent d0c3899 commit b3f8263

File tree

1 file changed

+169
-0
lines changed
  • llvm/test/CodeGen/RISCV/RISCVCustomLICM

1 file changed

+169
-0
lines changed

‎llvm/test/CodeGen/RISCV/RISCVCustomLICM/biquad.ll

Lines changed: 169 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,169 @@
1+
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 4
2+
; RUN: opt -S -mtriple=riscv32-esp-unknown-elf -passes=riscv-custom-licm -riscv-custom-licm=false < %s | FileCheck %s
3+
define dso_local noundef i32 @dsps_biquad_f32_ansi(ptr nocapture noundef readonly %input, ptr nocapture noundef writeonly %output, i32 noundef %len, ptr nocapture noundef readonly %coef, ptr nocapture noundef %w) local_unnamed_addr {
4+
; CHECK-LABEL: define dso_local noundef i32 @dsps_biquad_f32_ansi(
5+
; CHECK-SAME: ptr nocapture noundef readonly [[INPUT:%.*]], ptr nocapture noundef writeonly [[OUTPUT:%.*]], i32 noundef [[LEN:%.*]], ptr nocapture noundef readonly [[COEF:%.*]], ptr nocapture noundef [[W:%.*]]) local_unnamed_addr {
6+
; CHECK-NEXT: entry:
7+
; CHECK-NEXT: [[TMP0:%.*]] = icmp sgt i32 [[LEN]], 2
8+
; CHECK-NEXT: br i1 [[TMP0]], label [[FOR_COND_PREHEADER:%.*]], label [[FOR_BODY_LR_PH_CLONE:%.*]]
9+
; CHECK: for.cond.preheader:
10+
; CHECK-NEXT: [[CMP30:%.*]] = icmp sgt i32 [[LEN]], 0
11+
; CHECK-NEXT: br i1 [[CMP30]], label [[FOR_BODY_LR_PH:%.*]], label [[IF_END:%.*]]
12+
; CHECK: for.body.lr.ph:
13+
; CHECK-NEXT: [[ARRAYIDX1:%.*]] = getelementptr inbounds float, ptr [[COEF]], i32 3
14+
; CHECK-NEXT: [[ARRAYIDX3:%.*]] = getelementptr inbounds float, ptr [[COEF]], i32 4
15+
; CHECK-NEXT: [[ARRAYIDX4:%.*]] = getelementptr inbounds float, ptr [[W]], i32 1
16+
; CHECK-NEXT: [[ARRAYIDX7:%.*]] = getelementptr inbounds float, ptr [[COEF]], i32 1
17+
; CHECK-NEXT: [[ARRAYIDX10:%.*]] = getelementptr inbounds float, ptr [[COEF]], i32 2
18+
; CHECK-NEXT: [[DOTPRE:%.*]] = load float, ptr [[W]], align 4
19+
; CHECK-NEXT: [[DOTPRE32:%.*]] = load float, ptr [[ARRAYIDX4]], align 4
20+
; CHECK-NEXT: br label [[FOR_BODY:%.*]]
21+
; CHECK: if.end:
22+
; CHECK-NEXT: ret i32 0
23+
; CHECK: for.body:
24+
; CHECK-NEXT: [[TMP1:%.*]] = phi float [ [[DOTPRE32]], [[FOR_BODY_LR_PH]] ], [ [[TMP13:%.*]], [[FOR_BODY]] ]
25+
; CHECK-NEXT: [[TMP2:%.*]] = phi float [ [[DOTPRE]], [[FOR_BODY_LR_PH]] ], [ [[TMP7:%.*]], [[FOR_BODY]] ]
26+
; CHECK-NEXT: [[I_031:%.*]] = phi i32 [ 0, [[FOR_BODY_LR_PH]] ], [ [[INC:%.*]], [[FOR_BODY]] ]
27+
; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds float, ptr [[INPUT]], i32 [[I_031]]
28+
; CHECK-NEXT: [[TMP3:%.*]] = load float, ptr [[ARRAYIDX]], align 4
29+
; CHECK-NEXT: [[TMP4:%.*]] = load float, ptr [[ARRAYIDX1]], align 4
30+
; CHECK-NEXT: [[NEG:%.*]] = fneg float [[TMP4]]
31+
; CHECK-NEXT: [[TMP5:%.*]] = tail call float @llvm.fmuladd.f32(float [[NEG]], float [[TMP2]], float [[TMP3]])
32+
; CHECK-NEXT: [[TMP6:%.*]] = load float, ptr [[ARRAYIDX3]], align 4
33+
; CHECK-NEXT: [[NEG5:%.*]] = fneg float [[TMP6]]
34+
; CHECK-NEXT: [[TMP7]] = tail call float @llvm.fmuladd.f32(float [[NEG5]], float [[TMP1]], float [[TMP5]])
35+
; CHECK-NEXT: [[TMP8:%.*]] = load float, ptr [[COEF]], align 4
36+
; CHECK-NEXT: [[TMP9:%.*]] = load float, ptr [[ARRAYIDX7]], align 4
37+
; CHECK-NEXT: [[MUL9:%.*]] = fmul float [[TMP2]], [[TMP9]]
38+
; CHECK-NEXT: [[TMP10:%.*]] = tail call float @llvm.fmuladd.f32(float [[TMP8]], float [[TMP7]], float [[MUL9]])
39+
; CHECK-NEXT: [[TMP11:%.*]] = load float, ptr [[ARRAYIDX10]], align 4
40+
; CHECK-NEXT: [[TMP12:%.*]] = tail call float @llvm.fmuladd.f32(float [[TMP11]], float [[TMP1]], float [[TMP10]])
41+
; CHECK-NEXT: [[ARRAYIDX12:%.*]] = getelementptr inbounds float, ptr [[OUTPUT]], i32 [[I_031]]
42+
; CHECK-NEXT: store float [[TMP12]], ptr [[ARRAYIDX12]], align 4
43+
; CHECK-NEXT: [[TMP13]] = load float, ptr [[W]], align 4
44+
; CHECK-NEXT: store float [[TMP13]], ptr [[ARRAYIDX4]], align 4
45+
; CHECK-NEXT: store float [[TMP7]], ptr [[W]], align 4
46+
; CHECK-NEXT: [[INC]] = add nuw nsw i32 [[I_031]], 1
47+
; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i32 [[INC]], [[LEN]]
48+
; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label [[IF_END]], label [[FOR_BODY]]
49+
; CHECK: for.body.lr.ph.clone:
50+
; CHECK-NEXT: [[ARRAYIDX1_CLONE:%.*]] = getelementptr inbounds float, ptr [[COEF]], i32 3
51+
; CHECK-NEXT: [[ARRAYIDX3_CLONE:%.*]] = getelementptr inbounds float, ptr [[COEF]], i32 4
52+
; CHECK-NEXT: [[ARRAYIDX4_CLONE:%.*]] = getelementptr inbounds float, ptr [[W]], i32 1
53+
; CHECK-NEXT: [[ARRAYIDX7_CLONE:%.*]] = getelementptr inbounds float, ptr [[COEF]], i32 1
54+
; CHECK-NEXT: [[ARRAYIDX10_CLONE:%.*]] = getelementptr inbounds float, ptr [[COEF]], i32 2
55+
; CHECK-NEXT: [[DOTPRE_CLONE:%.*]] = load float, ptr [[W]], align 4
56+
; CHECK-NEXT: [[DOTPRE32_CLONE:%.*]] = load float, ptr [[ARRAYIDX4_CLONE]], align 4
57+
; CHECK-NEXT: br label [[FOR_BODY_CLONE:%.*]]
58+
; CHECK: for.body.clone:
59+
; CHECK-NEXT: [[TMP14:%.*]] = phi float [ [[DOTPRE32_CLONE]], [[FOR_BODY_LR_PH_CLONE]] ], [ [[TMP26:%.*]], [[FOR_BODY_CLONE]] ]
60+
; CHECK-NEXT: [[TMP15:%.*]] = phi float [ [[DOTPRE_CLONE]], [[FOR_BODY_LR_PH_CLONE]] ], [ [[TMP20:%.*]], [[FOR_BODY_CLONE]] ]
61+
; CHECK-NEXT: [[I_031_CLONE:%.*]] = phi i32 [ 0, [[FOR_BODY_LR_PH_CLONE]] ], [ [[INC_CLONE:%.*]], [[FOR_BODY_CLONE]] ]
62+
; CHECK-NEXT: [[ARRAYIDX_CLONE:%.*]] = getelementptr inbounds float, ptr [[INPUT]], i32 [[I_031_CLONE]]
63+
; CHECK-NEXT: [[TMP16:%.*]] = load float, ptr [[ARRAYIDX_CLONE]], align 4
64+
; CHECK-NEXT: [[TMP17:%.*]] = load float, ptr [[ARRAYIDX1_CLONE]], align 4
65+
; CHECK-NEXT: [[NEG_CLONE:%.*]] = fneg float [[TMP17]]
66+
; CHECK-NEXT: [[TMP18:%.*]] = tail call float @llvm.fmuladd.f32(float [[NEG_CLONE]], float [[TMP15]], float [[TMP16]])
67+
; CHECK-NEXT: [[TMP19:%.*]] = load float, ptr [[ARRAYIDX3_CLONE]], align 4
68+
; CHECK-NEXT: [[NEG5_CLONE:%.*]] = fneg float [[TMP19]]
69+
; CHECK-NEXT: [[TMP20]] = tail call float @llvm.fmuladd.f32(float [[NEG5_CLONE]], float [[TMP14]], float [[TMP18]])
70+
; CHECK-NEXT: [[TMP21:%.*]] = load float, ptr [[COEF]], align 4
71+
; CHECK-NEXT: [[TMP22:%.*]] = load float, ptr [[ARRAYIDX7_CLONE]], align 4
72+
; CHECK-NEXT: [[MUL9_CLONE:%.*]] = fmul float [[TMP15]], [[TMP22]]
73+
; CHECK-NEXT: [[TMP23:%.*]] = tail call float @llvm.fmuladd.f32(float [[TMP21]], float [[TMP20]], float [[MUL9_CLONE]])
74+
; CHECK-NEXT: [[TMP24:%.*]] = load float, ptr [[ARRAYIDX10_CLONE]], align 4
75+
; CHECK-NEXT: [[TMP25:%.*]] = tail call float @llvm.fmuladd.f32(float [[TMP24]], float [[TMP14]], float [[TMP23]])
76+
; CHECK-NEXT: [[ARRAYIDX12_CLONE:%.*]] = getelementptr inbounds float, ptr [[OUTPUT]], i32 [[I_031_CLONE]]
77+
; CHECK-NEXT: store float [[TMP25]], ptr [[ARRAYIDX12_CLONE]], align 4
78+
; CHECK-NEXT: [[TMP26]] = load float, ptr [[W]], align 4
79+
; CHECK-NEXT: store float [[TMP26]], ptr [[ARRAYIDX4_CLONE]], align 4
80+
; CHECK-NEXT: store float [[TMP20]], ptr [[W]], align 4
81+
; CHECK-NEXT: [[INC_CLONE]] = add nuw nsw i32 [[I_031_CLONE]], 1
82+
; CHECK-NEXT: [[EXITCOND_NOT_CLONE:%.*]] = icmp eq i32 [[INC_CLONE]], [[LEN]]
83+
; CHECK-NEXT: br i1 [[EXITCOND_NOT_CLONE]], label [[IF_END]], label [[FOR_BODY_CLONE]]
84+
;
85+
entry:
86+
%0 = icmp sgt i32 %len, 2
87+
br i1 %0, label %for.cond.preheader, label %for.body.lr.ph.clone
88+
89+
for.cond.preheader: ; preds = %entry
90+
%cmp30 = icmp sgt i32 %len, 0
91+
br i1 %cmp30, label %for.body.lr.ph, label %if.end
92+
93+
for.body.lr.ph: ; preds = %for.cond.preheader
94+
%arrayidx1 = getelementptr inbounds float, ptr %coef, i32 3
95+
%arrayidx3 = getelementptr inbounds float, ptr %coef, i32 4
96+
%arrayidx4 = getelementptr inbounds float, ptr %w, i32 1
97+
%arrayidx7 = getelementptr inbounds float, ptr %coef, i32 1
98+
%arrayidx10 = getelementptr inbounds float, ptr %coef, i32 2
99+
%.pre = load float, ptr %w, align 4
100+
%.pre32 = load float, ptr %arrayidx4, align 4
101+
br label %for.body
102+
103+
if.end: ; preds = %for.body.clone, %for.body, %for.cond.preheader
104+
ret i32 0
105+
106+
for.body: ; preds = %for.body, %for.body.lr.ph
107+
%1 = phi float [ %.pre32, %for.body.lr.ph ], [ %13, %for.body ]
108+
%2 = phi float [ %.pre, %for.body.lr.ph ], [ %7, %for.body ]
109+
%i.031 = phi i32 [ 0, %for.body.lr.ph ], [ %inc, %for.body ]
110+
%arrayidx = getelementptr inbounds float, ptr %input, i32 %i.031
111+
%3 = load float, ptr %arrayidx, align 4
112+
%4 = load float, ptr %arrayidx1, align 4
113+
%neg = fneg float %4
114+
%5 = tail call float @llvm.fmuladd.f32(float %neg, float %2, float %3)
115+
%6 = load float, ptr %arrayidx3, align 4
116+
%neg5 = fneg float %6
117+
%7 = tail call float @llvm.fmuladd.f32(float %neg5, float %1, float %5)
118+
%8 = load float, ptr %coef, align 4
119+
%9 = load float, ptr %arrayidx7, align 4
120+
%mul9 = fmul float %2, %9
121+
%10 = tail call float @llvm.fmuladd.f32(float %8, float %7, float %mul9)
122+
%11 = load float, ptr %arrayidx10, align 4
123+
%12 = tail call float @llvm.fmuladd.f32(float %11, float %1, float %10)
124+
%arrayidx12 = getelementptr inbounds float, ptr %output, i32 %i.031
125+
store float %12, ptr %arrayidx12, align 4
126+
%13 = load float, ptr %w, align 4
127+
store float %13, ptr %arrayidx4, align 4
128+
store float %7, ptr %w, align 4
129+
%inc = add nuw nsw i32 %i.031, 1
130+
%exitcond.not = icmp eq i32 %inc, %len
131+
br i1 %exitcond.not, label %if.end, label %for.body
132+
133+
for.body.lr.ph.clone: ; preds = %entry
134+
%arrayidx1.clone = getelementptr inbounds float, ptr %coef, i32 3
135+
%arrayidx3.clone = getelementptr inbounds float, ptr %coef, i32 4
136+
%arrayidx4.clone = getelementptr inbounds float, ptr %w, i32 1
137+
%arrayidx7.clone = getelementptr inbounds float, ptr %coef, i32 1
138+
%arrayidx10.clone = getelementptr inbounds float, ptr %coef, i32 2
139+
%.pre.clone = load float, ptr %w, align 4
140+
%.pre32.clone = load float, ptr %arrayidx4.clone, align 4
141+
br label %for.body.clone
142+
143+
for.body.clone: ; preds = %for.body.clone, %for.body.lr.ph.clone
144+
%14 = phi float [ %.pre32.clone, %for.body.lr.ph.clone ], [ %26, %for.body.clone ]
145+
%15 = phi float [ %.pre.clone, %for.body.lr.ph.clone ], [ %20, %for.body.clone ]
146+
%i.031.clone = phi i32 [ 0, %for.body.lr.ph.clone ], [ %inc.clone, %for.body.clone ]
147+
%arrayidx.clone = getelementptr inbounds float, ptr %input, i32 %i.031.clone
148+
%16 = load float, ptr %arrayidx.clone, align 4
149+
%17 = load float, ptr %arrayidx1.clone, align 4
150+
%neg.clone = fneg float %17
151+
%18 = tail call float @llvm.fmuladd.f32(float %neg.clone, float %15, float %16)
152+
%19 = load float, ptr %arrayidx3.clone, align 4
153+
%neg5.clone = fneg float %19
154+
%20 = tail call float @llvm.fmuladd.f32(float %neg5.clone, float %14, float %18)
155+
%21 = load float, ptr %coef, align 4
156+
%22 = load float, ptr %arrayidx7.clone, align 4
157+
%mul9.clone = fmul float %15, %22
158+
%23 = tail call float @llvm.fmuladd.f32(float %21, float %20, float %mul9.clone)
159+
%24 = load float, ptr %arrayidx10.clone, align 4
160+
%25 = tail call float @llvm.fmuladd.f32(float %24, float %14, float %23)
161+
%arrayidx12.clone = getelementptr inbounds float, ptr %output, i32 %i.031.clone
162+
store float %25, ptr %arrayidx12.clone, align 4
163+
%26 = load float, ptr %w, align 4
164+
store float %26, ptr %arrayidx4.clone, align 4
165+
store float %20, ptr %w, align 4
166+
%inc.clone = add nuw nsw i32 %i.031.clone, 1
167+
%exitcond.not.clone = icmp eq i32 %inc.clone, %len
168+
br i1 %exitcond.not.clone, label %if.end, label %for.body.clone
169+
}

0 commit comments

Comments
 (0)