videolan/ffmpeg
1
Fork 0
mirror of https://git.videolan.org/git/ffmpeg.git synced 2024-09-14 02:53:37 +02:00
Code Releases Activity

x86/dsputilenc: implement XOP version of pix_sum16

Browse Source 
SSE2: 137 cycles
XOP:   87 cycles
Signed-off-by: James Almer <jamrial@gmail.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
This commit is contained in:
James Almer 2014-05-29 00:57:39 -03:00 committed by Michael Niedermayer
parent 232959f184
commit 05de4d3011
2 changed files with 29 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

29
libavcodec/x86/dsputilenc.asm
View File

@ -446,13 +446,24 @@ cglobal diff_pixels, 4, 5, 5
; int ff_pix_sum16_mmx(uint8_t *pix, int line_size)
; %1 = number of xmm registers used
; %2 = number of loops
%macro PIX_SUM16 2
cglobal pix_sum16, 2, 3, %1
; %3 = number of GPRs used
%macro PIX_SUM16 4
cglobal pix_sum16, 2, %3, %1
movsxdifnidn r1, r1d
mov r2, %2
%if cpuflag(xop)
lea r3, [r1*3]
%else
pxor m5, m5
%endif
pxor m4, m4
.loop:
%if cpuflag(xop)
vphaddubq m0, [r0]
vphaddubq m1, [r0+r1]
vphaddubq m2, [r0+r1*2]
vphaddubq m3, [r0+r3]
%else
mova m0, [r0]
%if mmsize == 8
mova m1, [r0+8]
@ -463,6 +474,7 @@ cglobal pix_sum16, 2, 3, %1
punpcklbw m0, m5
punpckhbw m3, m1, m5
punpcklbw m1, m5
%endif ; cpuflag(xop)
paddw m1, m0
paddw m3, m2
paddw m3, m1
@ -470,19 +482,26 @@ cglobal pix_sum16, 2, 3, %1
%if mmsize == 8
add r0, r1
%else
lea r0, [r0+r1*2]
lea r0, [r0+r1*%4]
%endif
dec r2
jne .loop
%if cpuflag(xop)
pshufd m0, m4, q0032
paddd m4, m0
%else
HADDW m4, m5
%endif
movd eax, m4
RET
%endmacro
INIT_MMX mmx
PIX_SUM16 0, 16
PIX_SUM16 0, 16, 3, 0
INIT_XMM sse2
PIX_SUM16 6, 8
PIX_SUM16 6, 8, 3, 2
INIT_XMM xop
PIX_SUM16 5, 4, 4, 4
; int ff_pix_norm1_mmx(uint8_t *pix, int line_size)
; %1 = number of xmm registers used

5
libavcodec/x86/dsputilenc_mmx.c
View File

@ -39,6 +39,7 @@ void ff_diff_pixels_sse2(int16_t *block, const uint8_t *s1, const uint8_t *s2,
int stride);
int ff_pix_sum16_mmx(uint8_t *pix, int line_size);
int ff_pix_sum16_sse2(uint8_t *pix, int line_size);
int ff_pix_sum16_xop(uint8_t *pix, int line_size);
int ff_pix_norm1_mmx(uint8_t *pix, int line_size);
int ff_pix_norm1_sse2(uint8_t *pix, int line_size);
int ff_sum_abs_dctelem_mmx(int16_t *block);
@ -925,5 +926,9 @@ av_cold void ff_dsputilenc_init_mmx(DSPContext *c, AVCodecContext *avctx,
#endif
}
if (EXTERNAL_XOP(cpu_flags)) {
c->pix_sum = ff_pix_sum16_xop;
}
ff_dsputil_init_pix_mmx(c, avctx);
}