highperfocused/ffmpeg
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

x86/hevc_deblock: use constants instead of generating values at runtime

Browse Source 
Signed-off-by: James Almer <jamrial@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
This commit is contained in:
James Almer 2014-05-19 17:39:02 -03:00 committed by Michael Niedermayer
parent 41b928c5fa
commit d43c303038
1 changed files with 15 additions and 32 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

47
libavcodec/x86/hevc_deblock.asm
View File

@ -27,6 +27,12 @@
SECTION_RODATA SECTION_RODATA
pw_pixel_max: times 8 dw ((1 << 10)-1) pw_pixel_max: times 8 dw ((1 << 10)-1)
pw_m1: times 8 dw -1
pw_m2: times 8 dw -2
pd_1 : times 4 dd 1
cextern pw_4
cextern pw_8
SECTION .text SECTION .text
INIT_XMM sse2 INIT_XMM sse2
@ -318,14 +324,10 @@ ALIGN 16
movd m7, [r2]; tc1 movd m7, [r2]; tc1
punpcklwd m7, m7 punpcklwd m7, m7
shufps m6, m7, 0; tc0, tc1 shufps m6, m7, 0; tc0, tc1
pcmpeqw m7, m7; set all bits to 1 pmullw m4, m6, [pw_m1]; -tc0, -tc1
pxor m4, m6, m7; flip all bits of first reg
psrlw m7, 15; 1 in every cell
paddw m4, m7; -tc0, -tc1
;end tc calculations ;end tc calculations
psllw m7, 2; 4 in every cell paddw m5, [pw_4]; +4
paddw m5, m7; +4
psraw m5, 3; >> 3 psraw m5, 3; >> 3
psllw m4, %1-8; << (BIT_DEPTH - 8) psllw m4, %1-8; << (BIT_DEPTH - 8)
@ -414,9 +416,7 @@ ALIGN 16
shl r2, 1 shl r2, 1
or r13, r2 or r13, r2
pcmpeqd m15, m15; set all bits to 1 pcmpeqd m11, [pd_1]; filtering mask
psrld m15, 31; set to 32bit 1
pcmpeqd m11, m15; filtering mask
;decide between strong and weak filtering ;decide between strong and weak filtering
;tc25 calculations ;tc25 calculations
@ -469,13 +469,8 @@ ALIGN 16
shr r2, 1; shr r2, 1;
and r14, r2; strong mask, bits 2 and 0 and r14, r2; strong mask, bits 2 and 0
pcmpeqw m13, m13; set all bits to 1 pmullw m14, m9, [pw_m2]; -tc * 2
pxor m14, m9, m13; invert bits
psrlw m13, 15; 1 in every cell
paddw m14, m13; -tc
psllw m9, 1; tc * 2 psllw m9, 1; tc * 2
psllw m14, 1; -tc * 2
and r14, 5; 0b101 and r14, 5; 0b101
mov r2, r14; strong mask mov r2, r14; strong mask
@ -488,12 +483,9 @@ ALIGN 16
jz .weakfilter jz .weakfilter
shufps m10, m12, 0 shufps m10, m12, 0
pcmpeqd m10, [pd_1]; strong mask
pcmpeqd m12, m12; set all bits to 1 mova m13, [pw_4]; 4 in every cell
psrld m12, 31; set to 32bit 1
pcmpeqd m10, m12; strong mask
psllw m13, 2; 4 in every cell
pand m11, m10; combine filtering mask and strong mask pand m11, m10; combine filtering mask and strong mask
paddw m12, m2, m3; p1 + p0 paddw m12, m2, m3; p1 + p0
paddw m12, m4; p1 + p0 + q0 paddw m12, m4; p1 + p0 + q0
@ -583,10 +575,7 @@ ALIGN 16
and r14, 1 and r14, 1
movd m11, r14d movd m11, r14d
shufps m11, m12, 0 shufps m11, m12, 0
pcmpeqd m11, [pd_1]; filtering mask
pcmpeqd m12, m12; set all bits to 1
psrld m12, 31; set to 32bit 1
pcmpeqd m11, m12; filtering mask
mov r13, r11; beta0 mov r13, r11; beta0
shr r13, 1; shr r13, 1;
@ -598,10 +587,7 @@ ALIGN 16
add r12, r13 add r12, r13
shr r12, 3; ((beta1+(beta1>>1))>>3)) shr r12, 3; ((beta1+(beta1>>1))>>3))
pcmpeqw m13, m13; set all bits to 1 mova m13, [pw_8]
psrlw m13, 15; 1 in every cell
psllw m13, 3; 8 in every cell
psubw m12, m4, m3 ; q0 - p0 psubw m12, m4, m3 ; q0 - p0
psllw m10, m12, 3; 8 * (q0 - p0) psllw m10, m12, 3; 8 * (q0 - p0)
paddw m12, m10 ; 9 * (q0 - p0) paddw m12, m10 ; 9 * (q0 - p0)
@ -626,11 +612,8 @@ ALIGN 16
pmaxsw m12, m14 pmaxsw m12, m14
pminsw m12, m9; av_clip(delta0, -tc, tc) pminsw m12, m9; av_clip(delta0, -tc, tc)
pcmpeqw m13, m13; set all bits to 1
psraw m9, 1; tc -> tc / 2 psraw m9, 1; tc -> tc / 2
pxor m14, m9, m13; complement -tc pmullw m14, m9, [pw_m1]; -tc / 2
psrlw m13, 15; set all cells to 1
paddw m14, m13; add 1, -tc / 2
pavgw m15, m1, m3; (p2 + p0 + 1) >> 1 pavgw m15, m1, m3; (p2 + p0 + 1) >> 1
psubw m15, m2; ((p2 + p0 + 1) >> 1) - p1 psubw m15, m2; ((p2 + p0 + 1) >> 1) - p1