x86 saturating multiply sum defs

This commit is contained in:
dzaima 2025-03-24 00:30:00 +02:00
parent 44ee417d40
commit 8beff254e2
2 changed files with 5 additions and 1 deletions

View File

@ -181,7 +181,7 @@ def {
all_hom,any_hom,blend_hom,hom_to_int,store_masked_hom,store_blended_hom,
all_top,any_top,blend_top,top_to_int,store_masked_top,store_blended_top,
load_expand_bits,make,mask_to_hom,mulw_split,mulh,narrow,narrow_trunc,narrow_pair,
pair,pdep,pext,rbit,sel,shuf_ind,reverse_units,broadcast_sel,absdiff_sum,mul_sum,
pair,pdep,pext,rbit,sel,shuf_ind,reverse_units,broadcast_sel,absdiff_sum,mul_sum,mul_sum_sat,
unord,unzip,vfold,vec_select,vec_shuffle,widen,widen_upper,multishift,
addp,addpw,addpwa,
}

View File

@ -84,6 +84,10 @@ def absdiff_sum{8, a:V=[k](u8), b:V if has_bw{V}} = {
def mul_sum{2, a:V=[k](i16), b:V if has_bw{V}} = {
emit{[k/2]i32, intrin{V, 'madd_epi16'}, a, b}
}
def mul_sum_sat{2, a:V=[k](i8), b:[k](u8) if has_bw{V}} = mul_sum_sat{2, b, a}
def mul_sum_sat{2, a:V=[k](u8), b:[k](i8) if has_bw{V}} = {
emit{[k/2]i16, intrin{V, 'maddubs_epi16'}, a, b}
}
def x86_low_elts{n, x:V=[k]E} = extract{x86_vec_low{n,E}, x, 0}