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

2x faster h264_idct_add8_10.

Browse Source 
Signed-off-by: Ronald S. Bultje <rsbultje@gmail.com>
This commit is contained in:
Loren Merritt
2011-06-01 21:53:15 -04:00
committed by Ronald S. Bultje
parent cc9947ffbe
commit 994c3550ff
1 changed files with 21 additions and 34 deletions
Download Patch File Download Diff File Expand all files Collapse all files

55
libavcodec/x86/h264_idct_10bit.asm
View File

@@ -249,16 +249,17 @@ IDCT8_DC_ADD avx
jmp .skipadd%2 jmp .skipadd%2
%endmacro %endmacro
%assign last_block 16
%macro ADD16_OP_INTRA 3 %macro ADD16_OP_INTRA 3
cmp word [r4+%3], 0 cmp word [r4+%3], 0
jnz .ac%2 jnz .ac%2
mov r6d, dword [r2+ 0] mov r5d, dword [r2+ 0]
or r6d, dword [r2+64] or r5d, dword [r2+64]
jz .skipblock%2 jz .skipblock%2
mov r5d, dword [r1+(%2+0)*4] mov r5d, dword [r1+(%2+0)*4]
call idct_dc_add_%1 call idct_dc_add_%1
.skipblock%2: .skipblock%2:
%if %2<15 %if %2<last_block-2
add r2, 128 add r2, 128
%endif %endif
.skipadd%2: .skipadd%2:
@@ -302,47 +303,33 @@ INIT_AVX
IDCT_ADD16INTRA_10 avx IDCT_ADD16INTRA_10 avx
%endif %endif
%assign last_block 24
;----------------------------------------------------------------------------- ;-----------------------------------------------------------------------------
; h264_idct_add8(pixel **dst, const int *block_offset, dctcoef *block, int stride, const uint8_t nnzc[6*8]) ; h264_idct_add8(pixel **dst, const int *block_offset, dctcoef *block, int stride, const uint8_t nnzc[6*8])
;----------------------------------------------------------------------------- ;-----------------------------------------------------------------------------
%macro IDCT_ADD8 1 %macro IDCT_ADD8 1
cglobal h264_idct_add8_10_%1,5,7 cglobal h264_idct_add8_10_%1,5,7
mov r5, 16
add r2, 1024
%ifdef PIC
lea r11, [scan8_mem]
%endif
%ifdef ARCH_X86_64 %ifdef ARCH_X86_64
mov r10, r0 mov r10, r0
%endif %endif
.nextblock: add r2, 1024
movzx r6, byte [scan8+r5] mov r0, [r0]
movzx r6, byte [r4+r6] ADD16_OP_INTRA %1, 16, 1+1*8
or r6d, dword [r2] ADD16_OP_INTRA %1, 18, 1+2*8
test r6, r6
jz .skipblock
%ifdef ARCH_X86_64 %ifdef ARCH_X86_64
mov r0d, dword [r1+r5*4] mov r0, [r10+gprsize]
add r0, [r10]
%else %else
mov r0, r0m mov r0, r0m
mov r0, [r0] mov r0, [r0+gprsize]
add r0, dword [r1+r5*4]
%endif %endif
IDCT4_ADD_10 r0, r2, r3 ADD16_OP_INTRA %1, 20, 1+4*8
.skipblock: ADD16_OP_INTRA %1, 22, 1+5*8
inc r5
add r2, 64
test r5, 3
jnz .nextblock
%ifdef ARCH_X86_64
add r10, gprsize
%else
add r0mp, gprsize
%endif
test r5, 4
jnz .nextblock
REP_RET REP_RET
AC %1, 16
AC %1, 18
AC %1, 20
AC %1, 22
%endmacro ; IDCT_ADD8 %endmacro ; IDCT_ADD8
INIT_XMM INIT_XMM