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: More optimizations by speeding up rarely taken branch

Browse Source 
For some reason optimizing the rarely taken branch speeds up the program.

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 2016-07-11 09:55:52 -07:00 committed by Greg Tucker
parent 84ffaead82
commit 40b5104397
1 changed files with 10 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

18
igzip/igzip_inflate.asm
View File

@ -221,6 +221,9 @@ extern rfc1951_lookup_table
%define %%next_bits2 %6 %define %%next_bits2 %6
;; Save length associated with symbol ;; Save length associated with symbol
mov %%next_bits2, %%read_in
shr %%next_bits2, DECODE_LOOKUP_SIZE
mov rcx, %%next_sym mov rcx, %%next_sym
shr rcx, 9 shr rcx, 9
@ -230,22 +233,20 @@ extern rfc1951_lookup_table
jl %%end jl %%end
;; Extract the 15-DECODE_LOOKUP_SIZE bits beyond the first DECODE_LOOKUP_SIZE bits. ;; Extract the 15-DECODE_LOOKUP_SIZE bits beyond the first DECODE_LOOKUP_SIZE bits.
lea %%next_sym, [%%state + 2 * %%next_sym]
sub rcx, 0x40 + DECODE_LOOKUP_SIZE
%ifdef USE_HSWNI %ifdef USE_HSWNI
and rcx, 0x1F bzhi %%next_bits2, %%next_bits2, rcx
bzhi %%next_bits2, %%read_in, rcx
%else %else
;; Decode next_sym using hint ;; Decode next_sym using hint
mov %%next_bits2, %%read_in
neg rcx neg rcx
shl %%next_bits2, cl shl %%next_bits2, cl
shr %%next_bits2, cl shr %%next_bits2, cl
%endif %endif
shr %%next_bits2, DECODE_LOOKUP_SIZE
add %%next_bits2, %%next_sym
;; Lookup actual next symbol ;; Lookup actual next symbol
movzx %%next_sym, word [%%state + %%state_offset + 2 * %%next_bits2 + 2 * ((1 << DECODE_LOOKUP_SIZE) - 0x8000)] movzx %%next_sym, word [%%next_sym + %%state_offset + 2 * %%next_bits2 + 2 * ((1 << DECODE_LOOKUP_SIZE) - 0x8000)]
;; Save length associated with symbol ;; Save length associated with symbol
mov rcx, %%next_sym mov rcx, %%next_sym
@ -312,7 +313,8 @@ loop_block:
and tmp3, (1 << DECODE_LOOKUP_SIZE) - 1 and tmp3, (1 << DECODE_LOOKUP_SIZE) - 1
;; Start reloading read_in ;; Start reloading read_in
SHLX tmp1, [next_in], read_in_length mov tmp1, [next_in]
SHLX tmp1, tmp1, read_in_length
or read_in, tmp1 or read_in, tmp1
;; Specutively load data associated with length symbol ;; Specutively load data associated with length symbol