generic-library/ffmpeg
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

ac3dsp: do not use pshufb in ac3_extract_exponents_ssse3()

Browse Source 
We need to do unsigned saturation in order to cover the corner case when the
absolute coefficient value is 16777215 (the maximum value).

Fixes Bug #216
This commit is contained in:
Justin Ruggles 2012-02-09 13:00:30 -05:00
parent e6d9fa66f1
commit d483bb58c3
1 changed files with 4 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

13
libavcodec/x86/ac3dsp.asm
View File

@ -35,7 +35,6 @@ pw_bap_mul2: dw 5, 7, 0, 7, 5, 7, 0, 7
; used in ff_ac3_extract_exponents() ; used in ff_ac3_extract_exponents()
pd_1: times 4 dd 1 pd_1: times 4 dd 1
pd_151: times 4 dd 151 pd_151: times 4 dd 151
pb_shuf_4dwb: db 0, 4, 8, 12
SECTION .text SECTION .text
@ -404,15 +403,12 @@ cglobal ac3_extract_exponents_3dnow, 3,3,0, exp, coef, len
%endif %endif
%macro AC3_EXTRACT_EXPONENTS 1 %macro AC3_EXTRACT_EXPONENTS 1
cglobal ac3_extract_exponents_%1, 3,3,5, exp, coef, len cglobal ac3_extract_exponents_%1, 3,3,4, exp, coef, len
add expq, lenq add expq, lenq
lea coefq, [coefq+4*lenq] lea coefq, [coefq+4*lenq]
neg lenq neg lenq
mova m2, [pd_1] mova m2, [pd_1]
mova m3, [pd_151] mova m3, [pd_151]
%ifidn %1, ssse3 ;
movd m4, [pb_shuf_4dwb]
%endif
.loop: .loop:
; move 4 32-bit coefs to xmm0 ; move 4 32-bit coefs to xmm0
mova m0, [coefq+4*lenq] mova m0, [coefq+4*lenq]
@ -426,12 +422,11 @@ cglobal ac3_extract_exponents_%1, 3,3,5, exp, coef, len
mova m0, m3 mova m0, m3
psubd m0, m1 psubd m0, m1
; move the lowest byte in each of 4 dwords to the low dword ; move the lowest byte in each of 4 dwords to the low dword
%ifidn %1, ssse3 ; NOTE: We cannot just extract the low bytes with pshufb because the dword
pshufb m0, m4 ; result for 16777215 is -1 due to float inaccuracy. Using packuswb
%else ; clips this to 0, which is the correct exponent.
packssdw m0, m0 packssdw m0, m0
packuswb m0, m0 packuswb m0, m0
%endif
movd [expq+lenq], m0 movd [expq+lenq], m0
add lenq, 4 add lenq, 4