📄 dct_sse2.asm
字号:
;/*****************************************************************************
; *
; * T264 AVC CODEC
; *
; * Copyright(C) 2004-2005 llcc <lcgate1@yahoo.com.cn>
; * 2004-2005 visionany <visionany@yahoo.com.cn>
; *
; * This program is free software ; you can redistribute it and/or modify
; * it under the terms of the GNU General Public License as published by
; * the Free Software Foundation ; either version 2 of the License, or
; * (at your option) any later version.
; *
; * This program is distributed in the hope that it will be useful,
; * but WITHOUT ANY WARRANTY ; without even the implied warranty of
; * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
; * GNU General Public License for more details.
; *
; * You should have received a copy of the GNU General Public License
; * along with this program ; if not, write to the Free Software
; * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
; *
; ****************************************************************************/
bits 32
; ideal from xvid
; modify by Thomascatlee@163.com
; for GCC
%macro cglobal 1
%ifdef NOPREFIX
global %1
%else
global _%1
%define %1 _%1
%endif
%endmacro
%macro cextern 1
%ifdef NOPREFIX
extern %1
%else
extern _%1
%define %1 _%1
%endif
%endmacro
; input 0 1 2 3, output 3 4 1 0
%macro transpose 5
movq %5, %1
punpckhwd %5, %2 ; mm4 = 8 4 7 3
punpcklwd %1, %2 ; mm0 = 6 2 5 1
movq %2, %3
punpckhwd %2, %4 ; mm1 = 16 12 15 11
punpcklwd %3, %4 ; mm2 = 14 10 13 9
movq %4, %5
punpckhdq %4, %2 ; mm3 = 16 12 8 4
punpckldq %5, %2 ; mm4 = 15 11 7 3
movq %2, %1
punpckhdq %2, %3 ; mm1 = 14 10 6 2
punpckldq %1, %3 ; mm0 = 13 9 5 1
%endmacro
%macro addsub 5
movq %5, %1
paddw %1, %4 ; %0 = s[0]
psubw %5, %4 ; %4 = s[3]
movq %4, %2
paddw %2, %3 ; %1 = s[1]
psubw %4, %3 ; %3 = s[2]
%endmacro
%macro addsub2 5
movq %5, %1 ; %5 = s[0]
paddw %1, %2 ; d[0] = s[0] + s[1]
psubw %5, %2 ; d[2] = tmp - s[1]
movq %2, %4 ; %2 = s[3]
paddw %2, %2 ; %2 = %2 + %2
paddw %2, %3 ; d[1] = %2 + s[2]
paddw %3, %3 ; s[2] = s[2]+ s[2]
psubw %4, %3 ; d[3] = s[3]- s[2]
%endmacro
; output 0 4 1 2
%macro idct_addsub2 5
movq %5, %1 ; %5 = d[0]
paddw %1, %3 ; s[0] = d[0] + d[2]
psubw %5, %3 ; s[1] = d[0] - d[2]
movq %3, %2 ; %3 = d[1]
psraw %2, 1 ; %2 = %2 / 2
psubw %2, %4 ; s[2] = %2 - d[3]
psraw %4, 1 ; d[3] = d[3] / 2
paddw %3, %4 ; s[3] = d[1] + d[3]
%endmacro
; %1 = mmx content, %2 = tmp mmx, %3 = zero mmx, %4 = xmm content, %5 = xmm tmp
%macro word2dw 5
movq %2, %1
punpcklwd %2, %3 ; dcba->0b0a
punpckhwd %1, %3 ; dcba->0d0c
movq2dq %4, %1 ; 00 00 0d 0c
pslldq %4, 8 ; 0d 0c 00 00
movq2dq %5, %2 ; 00 00 0b 0a
por %4, %5 ; 0d 0c 0b 0a
%endmacro
section .rodata data align=16
align 16
sse2_neg1 dw -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1
sse2_1 dw 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1
align 16
mmx1 dw 1, 1, 1, 1
align 16
mmx32 dw 32, 32, 32, 32
cextern quant
cextern dequant
align 16
section .text
;======================================================
;
; void
; dct4x4_mmx(int16_t* data)
;
;======================================================
align 16
cglobal dct4x4_mmx
dct4x4_mmx
mov eax, [esp + 4] ; data
movq mm0, [eax + 0]
movq mm1, [eax + 8]
movq mm2, [eax + 16]
movq mm3, [eax + 24]
transpose mm0, mm1, mm2, mm3, mm4 ; input 0 1 2 3, output 0 1 4 3
addsub mm0, mm1, mm4, mm3, mm2 ; input 0 1 2 3, output 0 1 3 4
; s[0] = mm0, s[1] = mm1, s[2] = mm3, s[3] = mm2
addsub2 mm0, mm1, mm3, mm2, mm4 ; input 0 1 2 3, output 0 1 4 3
transpose mm0, mm1, mm4, mm2, mm3 ; input 0 1 2 3, output 0 1 4 3
addsub mm0, mm1, mm3, mm2, mm4 ; input 0 1 2 3, output 0 1 3 4
; s[0] = mm0, s[1] = mm1, s[2] = mm2, s[3] = mm4
addsub2 mm0, mm1, mm2, mm4, mm3 ; input 0 1 2 3, output 0 1 4 3
movq [eax + 0], mm0
movq [eax + 8], mm1
movq [eax +16], mm3
movq [eax +24], mm4
ret
;======================================================
;
; void
; dct4x4dc_mmx(int16_t* data)
;
;======================================================
align 16
cglobal dct4x4dc_mmx
dct4x4dc_mmx
mov eax, [esp + 4] ; data
movq mm0, [eax + 0]
movq mm1, [eax + 8]
movq mm2, [eax + 16]
movq mm3, [eax + 24]
transpose mm0, mm1, mm2, mm3, mm4 ; input 0 1 2 3, output 0 1 4 3
addsub mm0, mm1, mm4, mm3, mm2 ; input 0 1 2 3, output 0 1 3 4
; s[0] = mm0, s[1] = mm1, s[2] = mm3, s[3] = mm2
addsub mm0, mm2, mm3, mm1, mm4 ; input 0 1 2 3, output 0 1 4 3
transpose mm0, mm2, mm4, mm1, mm3 ; input 0 1 2 3, output 0 1 4 3
addsub mm0, mm2, mm3, mm1, mm4 ; input 0 1 2 3, output 0 1 3 4
; s[0] = mm0, s[1] = mm2, s[2] = mm1, s[3] = mm4
addsub mm0, mm4, mm1, mm2, mm3 ; input 0 1 2 3, output 0 1 4 3
movq mm1, [mmx1]
paddw mm0, mm1
paddw mm4, mm1
paddw mm3, mm1
paddw mm2, mm1
psraw mm0, 1
psraw mm4, 1
psraw mm3, 1
psraw mm2, 1
movq [eax + 0], mm0
movq [eax + 8], mm4
movq [eax +16], mm3
movq [eax +24], mm2
ret
;======================================================
;
; void
; idct4x4_mmx(int16_t* data)
;
;======================================================
align 16
cglobal idct4x4_mmx
idct4x4_mmx
mov eax, [esp + 4] ; data
movq mm0, [eax + 0]
movq mm1, [eax + 8]
movq mm2, [eax + 16]
movq mm3, [eax + 24]
transpose mm0, mm1, mm2, mm3, mm4 ; input 0 1 2 3, output 0 1 4 3
idct_addsub2 mm0, mm1, mm4, mm3, mm2 ; input 0 1 2 3, output 0 4 1 2
; s[0] = mm0, s[1] = mm2, s[2] = mm1, s[3] = mm4
addsub mm0, mm2, mm1, mm4, mm3 ; input 0 1 2 3, output 0 1 3 4
transpose mm0, mm2, mm4, mm3, mm1 ; input 0 1 2 3, output 0 1 4 3
idct_addsub2 mm0, mm2, mm1, mm3, mm4 ; input 0 1 2 3, output 0 4 1 2
; s[0] = mm0, s[1] = mm4, s[2] = mm2, s[3] = mm1
addsub mm0, mm4, mm2, mm1, mm3 ; input 0 1 2 3, output 0 1 3 4
movq mm2, [mmx32]
paddw mm0, mm2
paddw mm4, mm2
paddw mm1, mm2
paddw mm3, mm2
psraw mm0, 6
psraw mm4, 6
psraw mm1, 6
psraw mm3, 6
movq [eax + 0], mm0
movq [eax + 8], mm4
movq [eax +16], mm1
movq [eax +24], mm3
ret
;======================================================
;
; void
; idct4x4dc_mmx(int16_t* data)
;
;======================================================
align 16
cglobal idct4x4dc_mmx
idct4x4dc_mmx
mov eax, [esp + 4] ; data
movq mm0, [eax + 0]
movq mm1, [eax + 8]
movq mm2, [eax + 16]
movq mm3, [eax + 24]
transpose mm0, mm1, mm2, mm3, mm4 ; input 0 1 2 3, output 0 1 4 3
addsub mm0, mm1, mm3, mm4, mm2 ; input 0 1 2 3, output 0 4 3 1
; s[0] = mm0, s[1] = mm2, s[2] = mm4, s[3] = mm1
addsub mm0, mm2, mm4, mm1, mm3 ; input 0 1 2 3, output 0 1 3 4
transpose mm0, mm2, mm1, mm3, mm4 ; input 0 1 2 3, output 0 1 4 3
addsub mm0, mm2, mm3, mm4, mm1 ; input 0 1 2 3, output 0 4 3 1
; s[0] = mm0, s[1] = mm1, s[2] = mm4, s[3] = mm2
addsub mm0, mm1, mm4, mm2, mm3 ; input 0 1 2 3, output 0 1 3 4
movq [eax + 0], mm0
movq [eax + 8], mm1
movq [eax +16], mm2
movq [eax +24], mm3
ret
;======================================================
;
; void
; quant4x4_sse2(int16_t* data, const int32_t Qp, int32_t is_intra)
;
;======================================================
align 16
cglobal quant4x4_sse2
quant4x4_sse2
push ebx
push esi
push edi
push ebp
mov edi, [esp + 4 + 16] ; data
mov eax, [esp + 8 + 16] ; qp
cdq
mov ebp, [esp + 12 + 16] ; is_intra
mov ebx, 6
idiv ebx
add eax, 15 ; qbits(eax) = 15 + qp / 6, mf_index(edx) = qp % 6
mov esi, edx
shl esi, 5
add esi, quant ; esi = quant[mf_index]
mov ecx, eax ; ecx = qbits
neg ebp
sbb ebp, ebp
and ebp, 0xfffffffd
add ebp, 6 ; is_intra(ecx) ? 3 : 6
mov eax, 1
shl eax, cl ; 1 << qbits
cdq
idiv ebp ; 1 << qbits / is_intra(ecx) ? 3 : 6
; eax = f, ecx = qbits, esi = quant[mf_index], edi = data
movd mm0, eax
movd mm1, ecx
pshufw mm0, mm0, 0x44
movq2dq xmm6, mm0
movq2dq xmm7, mm1
pshufd xmm6, xmm6, 0x44 ; f
pxor mm3, mm3
movdqa xmm0, [edi + 0] ; data
movdqa xmm1, [esi + 0] ; quant
; > 0
pxor xmm4, xmm4
movdqa xmm2, xmm0
pcmpgtw xmm0, xmm4
movdqa xmm4, xmm0
pand xmm0, xmm2
movdqa xmm3, xmm0
pmullw xmm0, xmm1 ; low part
pmulhw xmm3, xmm1 ; high part
movdqa xmm5, xmm0
punpcklwd xmm0, xmm3 ; low 4 - 32 bits
punpckhwd xmm5, xmm3 ; high 4 - 32 bits
movdqa xmm3, xmm4
punpcklwd xmm4, xmm4
pand xmm4, xmm6
paddd xmm0, xmm4 ; data * quant + f
psrad xmm0, xmm7 ; data * quant + f >> qbits
punpckhwd xmm3, xmm3
pand xmm3, xmm6
paddd xmm5, xmm3 ; data * quant + f
psrad xmm5, xmm7 ; data * quant + f >> qbits
packssdw xmm0, xmm5
; < 0
pxor xmm4, xmm4
movdqa xmm5, xmm2
pcmpgtw xmm4, xmm2
pand xmm5, xmm4
pmullw xmm5, [sse2_neg1]
movdqa xmm3, xmm5
pmullw xmm5, xmm1
pmulhw xmm3, xmm1
movdqa xmm1, xmm5
punpcklwd xmm5, xmm3
punpckhwd xmm1, xmm3
movdqa xmm3, xmm4
punpcklwd xmm4, xmm4
pand xmm4, xmm6
paddd xmm5, xmm4 ; data * quant - f
psrad xmm5, xmm7
punpckhwd xmm3, xmm3
pand xmm3, xmm6
paddd xmm1, xmm3
psrad xmm1, xmm7
packssdw xmm5, xmm1
pmullw xmm5, [sse2_neg1]
por xmm5, xmm0
movdqa [edi + 0], xmm5
movdqa xmm0, [edi + 16] ; data
movdqa xmm1, [esi + 16] ; quant
; > 0
pxor xmm4, xmm4
movdqa xmm2, xmm0
pcmpgtw xmm0, xmm4
movdqa xmm4, xmm0
pand xmm0, xmm2
movdqa xmm3, xmm0
pmullw xmm0, xmm1 ; low part
pmulhw xmm3, xmm1 ; high part
movdqa xmm5, xmm0
punpcklwd xmm0, xmm3 ; low 4 - 32 bits
punpckhwd xmm5, xmm3 ; high 4 - 32 bits
movdqa xmm3, xmm4
punpcklwd xmm4, xmm4
pand xmm4, xmm6
paddd xmm0, xmm4 ; data * quant + f
psrad xmm0, xmm7 ; data * quant + f >> qbits
punpckhwd xmm3, xmm3
pand xmm3, xmm6
paddd xmm5, xmm3 ; data * quant + f
psrad xmm5, xmm7 ; data * quant + f >> qbits
packssdw xmm0, xmm5
; < 0
pxor xmm4, xmm4
movdqa xmm5, xmm2
pcmpgtw xmm4, xmm2
pand xmm5, xmm4
pmullw xmm5, [sse2_neg1]
movdqa xmm3, xmm5
pmullw xmm5, xmm1
pmulhw xmm3, xmm1
movdqa xmm1, xmm5
punpcklwd xmm5, xmm3
punpckhwd xmm1, xmm3
movdqa xmm3, xmm4
punpcklwd xmm4, xmm4
pand xmm4, xmm6
paddd xmm5, xmm4 ; data * quant - f
psrad xmm5, xmm7
punpckhwd xmm3, xmm3
pand xmm3, xmm6
paddd xmm1, xmm3
psrad xmm1, xmm7
packssdw xmm5, xmm1
pmullw xmm5, [sse2_neg1]
por xmm5, xmm0
movdqa [edi + 16], xmm5
pop ebp
pop edi
pop esi
pop ebx
ret
;======================================================
;
; void
; iquant4x4_sse2(int16_t* data, const int32_t Qp)
;
;======================================================
align 16
cglobal iquant4x4_sse2
iquant4x4_sse2
mov eax, [esp + 8] ; qp
cdq
mov ecx, 6
idiv ecx ; qbits(eax) = qp / 6, mf_index(edx) = qp % 6
mov ecx, edx
shl ecx, 5
add ecx, dequant ; ecx = quant[mf_index]
mov edx, [esp + 4] ; data
; eax = qbits, ecx = quant[mf_index], edx = data
movdqa xmm6, [sse2_1]
movdqa xmm0, [edx + 0]
movdqa xmm2, [edx + 16]
movdqa xmm1, [ecx + 0]
movdqa xmm3, [ecx + 16]
pmullw xmm0, xmm1
pmullw xmm2, xmm3
movd xmm7, eax
psllw xmm6, xmm7 ; << qbits
pmullw xmm0, xmm6
pmullw xmm2, xmm6
movdqa [edx + 0], xmm0
movdqa [edx + 16], xmm2
ret ⌨️ 快捷键说明
复制代码
Ctrl + C
搜索代码
Ctrl + F
全屏模式
F11
切换主题
Ctrl + Shift + D
显示快捷键
?
增大字号
Ctrl + =
减小字号
Ctrl + -