generic-library/isa-l
1
0
Fork 0
mirror of https://github.com/intel/isa-l.git synced 2025-01-19 04:26:08 +01:00
Code Issues Projects Releases Wiki Activity

igzip: Fix to run on SSE

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 2016-07-26 11:08:30 -07:00 committed by Greg Tucker
parent a1981da94e
commit 6b2bedfe28
1 changed files with 5 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
igzip/igzip_decode_block_stateless.asm
View File

@ -398,7 +398,7 @@ decode_len_dist:
;; ;; Check if a valid look back distances was decoded
cmp copy_start, [rsp + start_out_mem_offset]
jl invalid_look_back_distance
vmovdqu xmm1, [copy_start]
MOVDQU xmm1, [copy_start]
;; Set tmp2 to be the minimum of COPY_SIZE and repeat_length
;; This is to decrease use of small_byte_copy branch
@ -412,23 +412,23 @@ decode_len_dist:
large_byte_copy:
;; Copy length distance pair when memory overlap is not an issue
vmovdqu [copy_start + look_back_dist2], xmm1
MOVDQU [copy_start + look_back_dist2], xmm1
sub repeat_length, COPY_SIZE
jle loop_block
add copy_start, COPY_SIZE
vmovdqu xmm1, [copy_start]
MOVDQU xmm1, [copy_start]
jmp large_byte_copy
small_byte_copy_pre:
;; Copy length distance pair when source and destination overlap
add repeat_length, look_back_dist2
small_byte_copy:
vmovdqu [copy_start + look_back_dist2], xmm1
MOVDQU [copy_start + look_back_dist2], xmm1
shl look_back_dist2, 1
vmovdqu xmm1, [copy_start]
MOVDQU xmm1, [copy_start]
cmp look_back_dist2, COPY_SIZE
jl small_byte_copy