T0b1/wasmtime
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

[AArch64] Port IaddPairwise to ISLE (#4201)

Browse Source 
Copyright (c) 2022, Arm Limited.
This commit is contained in:
Sam Parker
2022-06-06 15:37:13 +01:00
committed by GitHub
parent 7148882867
commit acfeda4d80
7 changed files with 150 additions and 50 deletions
Download Patch File Download Diff File Expand all files Collapse all files

50
cranelift/filetests/filetests/isa/aarch64/simd-pairwise-add.clif
View File

@@ -107,3 +107,53 @@ block0(v0: i8x16):
; addp v0.8h, v2.8h, v4.8h
; ret
function %fn9(i8x8, i8x8) -> i8x8 {
block0(v0: i8x8, v1: i8x8):
v2 = iadd_pairwise v0, v1
return v2
}
; block0:
; addp v0.8b, v0.8b, v1.8b
; ret
function %fn10(i8x16, i8x16) -> i8x16 {
block0(v0: i8x16, v1: i8x16):
v2 = iadd_pairwise v0, v1
return v2
}
; block0:
; addp v0.16b, v0.16b, v1.16b
; ret
function %fn11(i16x4, i16x4) -> i16x4 {
block0(v0: i16x4, v1: i16x4):
v2 = iadd_pairwise v0, v1
return v2
}
; block0:
; addp v0.4h, v0.4h, v1.4h
; ret
function %fn12(i16x8, i16x8) -> i16x8 {
block0(v0: i16x8, v1: i16x8):
v2 = iadd_pairwise v0, v1
return v2
}
; block0:
; addp v0.8h, v0.8h, v1.8h
; ret
function %fn14(i32x4, i32x4) -> i32x4 {
block0(v0: i32x4, v1: i32x4):
v2 = iadd_pairwise v0, v1
return v2
}
; block0:
; addp v0.4s, v0.4s, v1.4s
; ret

42
cranelift/filetests/filetests/runtests/simd-iaddpairwise.clif
View File

@@ -23,3 +23,45 @@ block0(v0: i32x4, v1: i32x4):
}
; run: %iaddp_i32x4([1 2 3 4], [5 6 7 8]) == [3 7 11 15]
; run: %iaddp_i32x4([4294967290 5 4294967290 5], [100 100 100 100]) == [4294967295 4294967295 200 200]
function %swiden_i8x16(i8x16) -> i16x8 {
block0(v0: i8x16):
v1 = swiden_low v0
v2 = swiden_high v0
v3 = iadd_pairwise v1, v2
return v3
}
; run: %swiden_i8x16([1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16]) == [3 7 11 15 19 23 27 31]
; run: %swiden_i8x16([-1 2 -3 4 -5 6 -7 8 -9 10 -11 12 -13 14 -15 16]) == [1 1 1 1 1 1 1 1]
; run: %swiden_i8x16([127 1 126 2 125 3 124 4 123 5 122 6 121 7 120 8]) == [128 128 128 128 128 128 128 128]
function %uwiden_i8x16(i8x16) -> i16x8 {
block0(v0: i8x16):
v1 = uwiden_low v0
v2 = uwiden_high v0
v3 = iadd_pairwise v1, v2
return v3
}
; run: %uwiden_i8x16([17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32]) == [35 39 43 47 51 55 59 63]
; run: %uwiden_i8x16([2 254 3 253 4 252 5 251 6 250 7 249 8 248 9 247]) == [256 256 256 256 256 256 256 256]
function %swiden_i16x8(i16x8) -> i32x4 {
block0(v0: i16x8):
v1 = swiden_low v0
v2 = swiden_high v0
v3 = iadd_pairwise v1, v2
return v3
}
; run: %swiden_i16x8([1 2 3 4 5 6 7 8]) == [3 7 11 15]
; run: %swiden_i16x8([32767 1 32766 3 32765 5 32764 8]) == [32768 32769 32770 32772]
; run: %swiden_i16x8([-32768 -1 32766 3 32765 5 -32764 -8]) == [-32769 32769 32770 -32772]
function %uwiden_i16x8(i16x8) -> i32x4 {
block0(v0: i16x8):
v1 = uwiden_low v0
v2 = uwiden_high v0
v3 = iadd_pairwise v1, v2
return v3
}
; run: %uwiden_i16x8([100 99 98 97 96 95 94 93]) == [199 195 191 187]
; run: %uwiden_i16x8([65535 1 65534 3 65533 5 65532 8]) == [65536 65537 65538 65540]