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

Add x86 SIMD instructions for min and max

Browse Source 
Only the I8, I16, and I32 versions are included since Cranelift lacks support for AVX.
This commit is contained in:
Andrew Brown
2019-10-25 10:12:35 -07:00
parent f053595748
commit 0ab5760fd7
5 changed files with 194 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

27
cranelift/filetests/filetests/isa/x86/simd-comparison-binemit.clif
View File

@@ -25,3 +25,30 @@ ebb0(v0: i64x2 [%xmm0], v1: i64x2 [%xmm7]):
[-, %xmm0] v2 = icmp sgt v0, v1 ; bin: 66 0f 38 37 c7
return v2
}
function %min_max_i8x16(i8x16, i8x16) {
ebb0(v0: i8x16 [%xmm3], v1: i8x16 [%xmm1]):
[-, %xmm3] v2 = x86_pmaxs v0, v1 ; bin: 66 0f 38 3c d9
[-, %xmm3] v3 = x86_pmaxu v0, v1 ; bin: 66 0f de d9
[-, %xmm3] v4 = x86_pmins v0, v1 ; bin: 66 0f 38 38 d9
[-, %xmm3] v5 = x86_pminu v0, v1 ; bin: 66 0f da d9
return
}
function %min_max_i16x8(i16x8, i16x8) {
ebb0(v0: i16x8 [%xmm2], v1: i16x8 [%xmm5]):
[-, %xmm2] v2 = x86_pmaxs v0, v1 ; bin: 66 0f ee d5
[-, %xmm2] v3 = x86_pmaxu v0, v1 ; bin: 66 0f 38 3e d5
[-, %xmm2] v4 = x86_pmins v0, v1 ; bin: 66 0f ea d5
[-, %xmm2] v5 = x86_pminu v0, v1 ; bin: 66 0f 38 3a d5
return
}
function %min_max_i32x4(i32x4, i32x4) {
ebb0(v0: i32x4 [%xmm2], v1: i32x4 [%xmm4]):
[-, %xmm2] v2 = x86_pmaxs v0, v1 ; bin: 66 0f 38 3d d4
[-, %xmm2] v3 = x86_pmaxu v0, v1 ; bin: 66 0f 38 3f d4
[-, %xmm2] v4 = x86_pmins v0, v1 ; bin: 66 0f 38 39 d4
[-, %xmm2] v5 = x86_pminu v0, v1 ; bin: 66 0f 38 3b d4
return
}

41
cranelift/filetests/filetests/isa/x86/simd-comparison-run.clif
View File

@@ -66,3 +66,44 @@ ebb0:
return v8
}
; run
function %maxs_i8x16() -> b1 {
ebb0:
v0 = vconst.i8x16 [1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1] ; 1 will be greater than -1 == 0xff with
; signed max
v1 = vconst.i8x16 [0xff 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1]
v2 = x86_pmaxs v0, v1
v8 = vall_true v2
return v8
}
; run
function %maxu_i16x8() -> b1 {
ebb0:
v0 = vconst.i16x8 [0 1 1 1 1 1 1 1]
v1 = vconst.i16x8 [-1 1 1 1 1 1 1 1] ; -1 == 0xff will be greater with unsigned max
v2 = x86_pmaxu v0, v1
v8 = vall_true v2
return v8
}
; run
function %mins_i32x4() -> b1 {
ebb0:
v0 = vconst.i32x4 [0 1 1 1]
v1 = vconst.i32x4 [-1 1 1 1] ; -1 == 0xff will be less with signed min
v2 = x86_pmins v0, v1
v8 = vall_true v2
return v8
}
; run
function %minu_i8x16() -> b1 {
ebb0:
v0 = vconst.i8x16 [1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1] ; 1 < 2 with unsiged min
v1 = vconst.i8x16 [2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2]
v2 = x86_pminu v0, v1
v8 = vall_true v2
return v8
}
; run