generic-library/isa-l
1
0
Fork 0
mirror of https://github.com/intel/isa-l.git synced 2024-12-12 17:33:50 +01:00
Code Issues Projects Releases Wiki Activity

igzip: Move code in igzip_stateless to hide latencies more in ivybridge.

Browse Source 
Signed-off-by: Roy Oursler <roy.j.oursler@intel.com>
Reviewed-by: Greg Tucker <greg.b.tucker@intel.com>
This commit is contained in:
Roy Oursler 2016-06-20 11:37:15 -07:00 committed by Greg Tucker
parent cf30138c7b
commit eb1b7788d0
1 changed files with 15 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

27
igzip/igzip_stateless.asm
View File

@ -236,34 +236,30 @@ loop2:
inc f_i inc f_i
MOVQ tmp6, xdata
shr tmp5, 16
mov tmp8, tmp5
compute_hash tmp6, tmp5
mov dist2 %+ w, f_i %+ w mov dist2 %+ w, f_i %+ w
sub dist2 %+ w, word [stream + _internal_state_head + 2 * hash2] sub dist2 %+ w, word [stream + _internal_state_head + 2 * hash2]
mov [stream + _internal_state_head + 2 * hash2], f_i %+ w mov [stream + _internal_state_head + 2 * hash2], f_i %+ w
dec dist2 dec dist2
MOVQ tmp8, xdata
shr tmp8, 16
mov tmp6, tmp8
compute_hash tmp2, tmp8
; if ((dist-1) < (D-1)) { ; if ((dist-1) < (D-1)) {
cmp dist %+ d, (D-1) cmp dist %+ d, (D-1)
cmovae dist, tmp3 cmovae dist, tmp3
add dist, 1 add dist, 1
neg dist neg dist
shr tmp8, 8
compute_hash tmp2, tmp8
cmp dist2 %+ d, (D-1) cmp dist2 %+ d, (D-1)
cmovae dist2, tmp3 cmovae dist2, tmp3
add dist2, 1 add dist2, 1
neg dist2 neg dist2
shr tmp6, 8
compute_hash tmp3, tmp6
MOVD xhash, tmp2 %+ d
PINSRD xhash, tmp3 %+ d, 1
PAND xhash, xhash, xmask
MARK __stateless_compare_ %+ ARCH MARK __stateless_compare_ %+ ARCH
;; Check for long len/dist match (>7) with first literal ;; Check for long len/dist match (>7) with first literal
MOVQ len, xdata MOVQ len, xdata
@ -272,6 +268,10 @@ MARK __stateless_compare_ %+ ARCH
xor len, [tmp1 + dist] xor len, [tmp1 + dist]
jz compare_loop jz compare_loop
MOVD xhash, tmp6 %+ d
PINSRD xhash, tmp2 %+ d, 1
PAND xhash, xhash, xmask
;; Check for len/dist match (>7) with second literal ;; Check for len/dist match (>7) with second literal
MOVQ len2, xdata MOVQ len2, xdata
xor len2, [tmp1 + dist2 + 1] xor len2, [tmp1 + dist2 + 1]
@ -605,6 +605,9 @@ end:
MARK __stateless_compare_loops_ %+ ARCH MARK __stateless_compare_loops_ %+ ARCH
compare_loop: compare_loop:
MOVD xhash, tmp6 %+ d
PINSRD xhash, tmp2 %+ d, 1
PAND xhash, xhash, xmask
lea tmp2, [tmp1 + dist] lea tmp2, [tmp1 + dist]
%if (COMPARE_TYPE == 1) %if (COMPARE_TYPE == 1)
compare250 tmp1, tmp2, len, tmp3 compare250 tmp1, tmp2, len, tmp3