Skip to content

Commit 7f49957

Browse files
artagnonartagnon
authored
SLP/RISCV: add test for vectorized ctpop, like in X86 (#65330)
Recently, 7f26c27 turned on SLP by default for RISC-V, and although there are quite a few tests for SLP under the X86/ target, it is unclear whether the same constructs would be vectorized on RISC-V. This patch takes a step in the direction of remedying this, by noticing that ctpop is often vectorized on RISC-V, and adding four tests for different integer widths.
1 parent 8dffb71 commit 7f49957

File tree

1 file changed

+110
-0
lines changed
  • llvm/test/Transforms/SLPVectorizer/RISCV

1 file changed

+110
-0
lines changed

‎llvm/test/Transforms/SLPVectorizer/RISCV/ctpop.ll

Lines changed: 110 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,110 @@
1+
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 2
2+
; RUN: opt < %s -passes=slp-vectorizer -S -mtriple=riscv32 -mattr=+m,+v | FileCheck %s
3+
; RUN: opt < %s -passes=slp-vectorizer -S -mtriple=riscv64 -mattr=+m,+v | FileCheck %s
4+
; RUN: opt < %s -passes=slp-vectorizer -S -mtriple=riscv32 -mattr=+v,+experimental-zvbb | FileCheck %s
5+
; RUN: opt < %s -passes=slp-vectorizer -S -mtriple=riscv64 -mattr=+v,+experimental-zvbb | FileCheck %s
6+
7+
define <4 x i8> @ctpop_v4i8(ptr %a) {
8+
; CHECK-LABEL: define <4 x i8> @ctpop_v4i8
9+
; CHECK-SAME: (ptr [[A:%.*]]) #[[ATTR0:[0-9]+]] {
10+
; CHECK-NEXT: entry:
11+
; CHECK-NEXT: [[TMP0:%.*]] = load <4 x i8>, ptr [[A]], align 4
12+
; CHECK-NEXT: [[TMP1:%.*]] = call <4 x i8> @llvm.ctpop.v4i8(<4 x i8> [[TMP0]])
13+
; CHECK-NEXT: ret <4 x i8> [[TMP1]]
14+
;
15+
entry:
16+
%0 = load <4 x i8>, ptr %a
17+
%vecext = extractelement <4 x i8> %0, i8 0
18+
%1 = call i8 @llvm.ctpop.i8(i8 %vecext)
19+
%vecins = insertelement <4 x i8> undef, i8 %1, i8 0
20+
%vecext.1 = extractelement <4 x i8> %0, i8 1
21+
%2 = call i8 @llvm.ctpop.i8(i8 %vecext.1)
22+
%vecins.1 = insertelement <4 x i8> %vecins, i8 %2, i8 1
23+
%vecext.2 = extractelement <4 x i8> %0, i8 2
24+
%3 = call i8 @llvm.ctpop.i8(i8 %vecext.2)
25+
%vecins.2 = insertelement <4 x i8> %vecins.1, i8 %3, i8 2
26+
%vecext.3 = extractelement <4 x i8> %0, i8 3
27+
%4 = call i8 @llvm.ctpop.i8(i8 %vecext.3)
28+
%vecins.3 = insertelement <4 x i8> %vecins.2, i8 %4, i8 3
29+
ret <4 x i8> %vecins.3
30+
}
31+
32+
define <4 x i16> @ctpop_v4i16(ptr %a) {
33+
; CHECK-LABEL: define <4 x i16> @ctpop_v4i16
34+
; CHECK-SAME: (ptr [[A:%.*]]) #[[ATTR0]] {
35+
; CHECK-NEXT: entry:
36+
; CHECK-NEXT: [[TMP0:%.*]] = load <4 x i16>, ptr [[A]], align 8
37+
; CHECK-NEXT: [[TMP1:%.*]] = call <4 x i16> @llvm.ctpop.v4i16(<4 x i16> [[TMP0]])
38+
; CHECK-NEXT: ret <4 x i16> [[TMP1]]
39+
;
40+
entry:
41+
%0 = load <4 x i16>, ptr %a
42+
%vecext = extractelement <4 x i16> %0, i16 0
43+
%1 = call i16 @llvm.ctpop.i16(i16 %vecext)
44+
%vecins = insertelement <4 x i16> undef, i16 %1, i16 0
45+
%vecext.1 = extractelement <4 x i16> %0, i16 1
46+
%2 = call i16 @llvm.ctpop.i16(i16 %vecext.1)
47+
%vecins.1 = insertelement <4 x i16> %vecins, i16 %2, i16 1
48+
%vecext.2 = extractelement <4 x i16> %0, i16 2
49+
%3 = call i16 @llvm.ctpop.i16(i16 %vecext.2)
50+
%vecins.2 = insertelement <4 x i16> %vecins.1, i16 %3, i16 2
51+
%vecext.3 = extractelement <4 x i16> %0, i16 3
52+
%4 = call i16 @llvm.ctpop.i16(i16 %vecext.3)
53+
%vecins.3 = insertelement <4 x i16> %vecins.2, i16 %4, i16 3
54+
ret <4 x i16> %vecins.3
55+
}
56+
57+
define <4 x i32> @ctpop_v4i32(ptr %a) {
58+
; CHECK-LABEL: define <4 x i32> @ctpop_v4i32
59+
; CHECK-SAME: (ptr [[A:%.*]]) #[[ATTR0]] {
60+
; CHECK-NEXT: entry:
61+
; CHECK-NEXT: [[TMP0:%.*]] = load <4 x i32>, ptr [[A]], align 16
62+
; CHECK-NEXT: [[TMP1:%.*]] = call <4 x i32> @llvm.ctpop.v4i32(<4 x i32> [[TMP0]])
63+
; CHECK-NEXT: ret <4 x i32> [[TMP1]]
64+
;
65+
entry:
66+
%0 = load <4 x i32>, ptr %a
67+
%vecext = extractelement <4 x i32> %0, i32 0
68+
%1 = call i32 @llvm.ctpop.i32(i32 %vecext)
69+
%vecins = insertelement <4 x i32> undef, i32 %1, i32 0
70+
%vecext.1 = extractelement <4 x i32> %0, i32 1
71+
%2 = call i32 @llvm.ctpop.i32(i32 %vecext.1)
72+
%vecins.1 = insertelement <4 x i32> %vecins, i32 %2, i32 1
73+
%vecext.2 = extractelement <4 x i32> %0, i32 2
74+
%3 = call i32 @llvm.ctpop.i32(i32 %vecext.2)
75+
%vecins.2 = insertelement <4 x i32> %vecins.1, i32 %3, i32 2
76+
%vecext.3 = extractelement <4 x i32> %0, i32 3
77+
%4 = call i32 @llvm.ctpop.i32(i32 %vecext.3)
78+
%vecins.3 = insertelement <4 x i32> %vecins.2, i32 %4, i32 3
79+
ret <4 x i32> %vecins.3
80+
}
81+
82+
define <4 x i64> @ctpop_v4i64(ptr %a) {
83+
; CHECK-LABEL: define <4 x i64> @ctpop_v4i64
84+
; CHECK-SAME: (ptr [[A:%.*]]) #[[ATTR0]] {
85+
; CHECK-NEXT: entry:
86+
; CHECK-NEXT: [[TMP0:%.*]] = load <4 x i64>, ptr [[A]], align 32
87+
; CHECK-NEXT: [[TMP1:%.*]] = call <4 x i64> @llvm.ctpop.v4i64(<4 x i64> [[TMP0]])
88+
; CHECK-NEXT: ret <4 x i64> [[TMP1]]
89+
;
90+
entry:
91+
%0 = load <4 x i64>, ptr %a
92+
%vecext = extractelement <4 x i64> %0, i32 0
93+
%1 = call i64 @llvm.ctpop.i64(i64 %vecext)
94+
%vecins = insertelement <4 x i64> undef, i64 %1, i64 0
95+
%vecext.1 = extractelement <4 x i64> %0, i32 1
96+
%2 = call i64 @llvm.ctpop.i64(i64 %vecext.1)
97+
%vecins.1 = insertelement <4 x i64> %vecins, i64 %2, i64 1
98+
%vecext.2 = extractelement <4 x i64> %0, i32 2
99+
%3 = call i64 @llvm.ctpop.i64(i64 %vecext.2)
100+
%vecins.2 = insertelement <4 x i64> %vecins.1, i64 %3, i64 2
101+
%vecext.3 = extractelement <4 x i64> %0, i32 3
102+
%4 = call i64 @llvm.ctpop.i64(i64 %vecext.3)
103+
%vecins.3 = insertelement <4 x i64> %vecins.2, i64 %4, i64 3
104+
ret <4 x i64> %vecins.3
105+
}
106+
107+
declare i8 @llvm.ctpop.i8(i8)
108+
declare i16 @llvm.ctpop.i16(i16)
109+
declare i32 @llvm.ctpop.i32(i32)
110+
declare i64 @llvm.ctpop.i64(i64)

0 commit comments

Comments
 (0)