Skip to content

Commit bdd8d1b

Browse files
bongjunjbongjunj
authored
[Cranelift] folds two shifts over add|sub|and (#11640)
* [Cranelift] add opts and tests * [Cranelift] add opts and tests * [Cranelift] add opts and tests * [Cranelift] update tests
1 parent 078364f commit bdd8d1b

4 files changed

Lines changed: 197 additions & 0 deletions

File tree

‎cranelift/codegen/src/opts/shifts.isle‎

Lines changed: 5 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -305,3 +305,8 @@
305305
(rule (simplify (rotl ty x (iconst_u kty k)))
306306
(if-let false (u64_eq k (u64_and k (ty_shift_mask ty))))
307307
(rotl ty x (iconst_u kty (u64_and k (ty_shift_mask ty)))))
308+
309+
(rule (simplify (band ty (ishl ty x z) (ishl ty y z))) (ishl ty (band ty x y) z))
310+
(rule (simplify (isub ty (ishl ty x z) (ishl ty y z))) (ishl ty (isub ty x y) z))
311+
(rule (simplify (iadd ty (ishl ty x z) (ishl ty y z))) (ishl ty (iadd ty x y) z))
312+

‎cranelift/filetests/filetests/egraph/fold-add-shifts.clif‎

Lines changed: 64 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,64 @@
1+
test optimize precise-output
2+
set opt_level=speed
3+
target x86_64
4+
5+
function %test1(i8, i8, i8) -> i8 {
6+
block0(v0: i8, v1: i8, v3: i8):
7+
v2 = ishl v0, v1
8+
v4 = ishl v3, v1
9+
v5 = iadd v2, v4
10+
return v5
11+
}
12+
13+
; function %test1(i8, i8, i8) -> i8 fast {
14+
; block0(v0: i8, v1: i8, v3: i8):
15+
; v6 = iadd v0, v3
16+
; v7 = ishl v6, v1
17+
; return v7
18+
; }
19+
20+
function %test2(i16, i16, i16) -> i16 {
21+
block0(v0: i16, v1: i16, v3: i16):
22+
v2 = ishl v0, v1
23+
v4 = ishl v3, v1
24+
v5 = iadd v2, v4
25+
return v5
26+
}
27+
28+
; function %test2(i16, i16, i16) -> i16 fast {
29+
; block0(v0: i16, v1: i16, v3: i16):
30+
; v6 = iadd v0, v3
31+
; v7 = ishl v6, v1
32+
; return v7
33+
; }
34+
35+
function %test3(i32, i32, i32) -> i32 {
36+
block0(v0: i32, v1: i32, v3: i32):
37+
v2 = ishl v0, v1
38+
v4 = ishl v3, v1
39+
v5 = iadd v2, v4
40+
return v5
41+
}
42+
43+
; function %test3(i32, i32, i32) -> i32 fast {
44+
; block0(v0: i32, v1: i32, v3: i32):
45+
; v6 = iadd v0, v3
46+
; v7 = ishl v6, v1
47+
; return v7
48+
; }
49+
50+
function %test4(i64, i64, i64) -> i64 {
51+
block0(v0: i64, v1: i64, v3: i64):
52+
v2 = ishl v0, v1
53+
v4 = ishl v3, v1
54+
v5 = iadd v2, v4
55+
return v5
56+
}
57+
58+
; function %test4(i64, i64, i64) -> i64 fast {
59+
; block0(v0: i64, v1: i64, v3: i64):
60+
; v6 = iadd v0, v3
61+
; v7 = ishl v6, v1
62+
; return v7
63+
; }
64+

‎cranelift/filetests/filetests/egraph/fold-and-shifts.clif‎

Lines changed: 64 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,64 @@
1+
test optimize precise-output
2+
set opt_level=speed
3+
target x86_64
4+
5+
function %test1(i8, i8, i8) -> i8 {
6+
block0(v0: i8, v1: i8, v3: i8):
7+
v2 = ishl v0, v1
8+
v4 = ishl v3, v1
9+
v5 = band v2, v4
10+
return v5
11+
}
12+
13+
; function %test1(i8, i8, i8) -> i8 fast {
14+
; block0(v0: i8, v1: i8, v3: i8):
15+
; v6 = band v0, v3
16+
; v7 = ishl v6, v1
17+
; return v7
18+
; }
19+
20+
function %test2(i16, i16, i16) -> i16 {
21+
block0(v0: i16, v1: i16, v3: i16):
22+
v2 = ishl v0, v1
23+
v4 = ishl v3, v1
24+
v5 = band v2, v4
25+
return v5
26+
}
27+
28+
; function %test2(i16, i16, i16) -> i16 fast {
29+
; block0(v0: i16, v1: i16, v3: i16):
30+
; v6 = band v0, v3
31+
; v7 = ishl v6, v1
32+
; return v7
33+
; }
34+
35+
function %test3(i32, i32, i32) -> i32 {
36+
block0(v0: i32, v1: i32, v3: i32):
37+
v2 = ishl v0, v1
38+
v4 = ishl v3, v1
39+
v5 = band v2, v4
40+
return v5
41+
}
42+
43+
; function %test3(i32, i32, i32) -> i32 fast {
44+
; block0(v0: i32, v1: i32, v3: i32):
45+
; v6 = band v0, v3
46+
; v7 = ishl v6, v1
47+
; return v7
48+
; }
49+
50+
function %test4(i64, i64, i64) -> i64 {
51+
block0(v0: i64, v1: i64, v3: i64):
52+
v2 = ishl v0, v1
53+
v4 = ishl v3, v1
54+
v5 = band v2, v4
55+
return v5
56+
}
57+
58+
; function %test4(i64, i64, i64) -> i64 fast {
59+
; block0(v0: i64, v1: i64, v3: i64):
60+
; v6 = band v0, v3
61+
; v7 = ishl v6, v1
62+
; return v7
63+
; }
64+

‎cranelift/filetests/filetests/egraph/fold-sub-shifts.clif‎

Lines changed: 64 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,64 @@
1+
test optimize precise-output
2+
set opt_level=speed
3+
target x86_64
4+
5+
function %test1(i8, i8, i8) -> i8 {
6+
block0(v0: i8, v1: i8, v3: i8):
7+
v2 = ishl v0, v1
8+
v4 = ishl v3, v1
9+
v5 = isub v2, v4
10+
return v5
11+
}
12+
13+
; function %test1(i8, i8, i8) -> i8 fast {
14+
; block0(v0: i8, v1: i8, v3: i8):
15+
; v6 = isub v0, v3
16+
; v7 = ishl v6, v1
17+
; return v7
18+
; }
19+
20+
function %test2(i16, i16, i16) -> i16 {
21+
block0(v0: i16, v1: i16, v3: i16):
22+
v2 = ishl v0, v1
23+
v4 = ishl v3, v1
24+
v5 = isub v2, v4
25+
return v5
26+
}
27+
28+
; function %test2(i16, i16, i16) -> i16 fast {
29+
; block0(v0: i16, v1: i16, v3: i16):
30+
; v6 = isub v0, v3
31+
; v7 = ishl v6, v1
32+
; return v7
33+
; }
34+
35+
function %test3(i32, i32, i32) -> i32 {
36+
block0(v0: i32, v1: i32, v3: i32):
37+
v2 = ishl v0, v1
38+
v4 = ishl v3, v1
39+
v5 = isub v2, v4
40+
return v5
41+
}
42+
43+
; function %test3(i32, i32, i32) -> i32 fast {
44+
; block0(v0: i32, v1: i32, v3: i32):
45+
; v6 = isub v0, v3
46+
; v7 = ishl v6, v1
47+
; return v7
48+
; }
49+
50+
function %test4(i64, i64, i64) -> i64 {
51+
block0(v0: i64, v1: i64, v3: i64):
52+
v2 = ishl v0, v1
53+
v4 = ishl v3, v1
54+
v5 = isub v2, v4
55+
return v5
56+
}
57+
58+
; function %test4(i64, i64, i64) -> i64 fast {
59+
; block0(v0: i64, v1: i64, v3: i64):
60+
; v6 = isub v0, v3
61+
; v7 = ishl v6, v1
62+
; return v7
63+
; }
64+

0 commit comments

Comments
 (0)