--- mem_transfer_mmx.asm 2002/03/08 02:45:01 1.1 +++ mem_transfer_mmx.asm 2004/03/22 22:36:24 1.10 @@ -1,473 +1,461 @@ -;/************************************************************************** +;/**************************************************************************** ; * -; * XVID MPEG-4 VIDEO CODEC -; * mmx 8bit<->16bit transfers +; * XVID MPEG-4 VIDEO CODEC +; * - 8<->16 bit transfer functions - ; * -; * This program is an implementation of a part of one or more MPEG-4 -; * Video tools as specified in ISO/IEC 14496-2 standard. Those intending -; * to use this software module in hardware or software products are -; * advised that its use may infringe existing patents or copyrights, and -; * any such use would be at such party's own risk. The original -; * developer of this software module and his/her company, and subsequent -; * editors and their companies, will have no liability for use of this -; * software or modifications or derivatives thereof. +; * Copyright (C) 2001 Peter Ross +; * 2001 Michael Militzer +; * 2002 Pascal Massimino ; * -; * 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 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. +; * 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., 675 Mass Ave, Cambridge, MA 02139, USA. +; * 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 ; * -; *************************************************************************/ - -;/************************************************************************** -; * -; * History: +; * $Id: mem_transfer_mmx.asm,v 1.10 2004/03/22 22:36:24 edgomez Exp $ ; * -; * 07.01.2002 merge functions from compensate_mmx; rename functions -; * 07.11.2001 initial version; (c)2001 peter ross -; * -; *************************************************************************/ - +; ***************************************************************************/ -bits 32 +BITS 32 -%macro cglobal 1 +%macro cglobal 1 %ifdef PREFIX - global _%1 + global _%1 %define %1 _%1 %else global %1 %endif %endmacro +;============================================================================= +; Read only data +;============================================================================= + +%ifdef FORMAT_COFF +SECTION .rodata data +%else +SECTION .rodata data align=16 +%endif + +ALIGN 16 +mmx_one: + dw 1, 1, 1, 1 + +;============================================================================= +; Code +;============================================================================= -section .text +SECTION .text +cglobal transfer_8to16copy_mmx +cglobal transfer_16to8copy_mmx +cglobal transfer_8to16sub_mmx +cglobal transfer_8to16subro_mmx +cglobal transfer_8to16sub2_mmx +cglobal transfer_8to16sub2_xmm +cglobal transfer_16to8add_mmx +cglobal transfer8x8_copy_mmx -;=========================================================================== +;----------------------------------------------------------------------------- ; ; void transfer_8to16copy_mmx(int16_t * const dst, ; const uint8_t * const src, ; uint32_t stride); ; -;=========================================================================== - -align 16 -cglobal transfer_8to16copy_mmx -transfer_8to16copy_mmx - - push esi - push edi - - mov edi, [esp + 8 + 4] ; dst - mov esi, [esp + 8 + 8] ; src - mov ecx, [esp + 8 + 12] ; stride +;----------------------------------------------------------------------------- - pxor mm7, mm7 ; mm7 = zero - - mov eax, 8 - -.loop - movq mm0, [esi] - movq mm1, mm0 - punpcklbw mm0, mm7 ; mm01 = unpack([src]) - punpckhbw mm1, mm7 - - movq [edi], mm0 ; [dst] = mm01 - movq [edi + 8], mm1 - - add edi, 16 - add esi, ecx - dec eax - jnz .loop - - pop edi - pop esi - - ret +%macro COPY_8_TO_16 1 + movq mm0, [eax] + movq mm1, [eax+edx] + movq mm2, mm0 + movq mm3, mm1 + punpcklbw mm0, mm7 + movq [ecx+%1*32], mm0 + punpcklbw mm1, mm7 + movq [ecx+%1*32+16], mm1 + punpckhbw mm2, mm7 + punpckhbw mm3, mm7 + lea eax, [eax+2*edx] + movq [ecx+%1*32+8], mm2 + movq [ecx+%1*32+24], mm3 +%endmacro +ALIGN 16 +transfer_8to16copy_mmx: + mov ecx, [esp+ 4] ; Dst + mov eax, [esp+ 8] ; Src + mov edx, [esp+12] ; Stride + pxor mm7, mm7 + + COPY_8_TO_16 0 + COPY_8_TO_16 1 + COPY_8_TO_16 2 + COPY_8_TO_16 3 + ret -;=========================================================================== +;----------------------------------------------------------------------------- ; ; void transfer_16to8copy_mmx(uint8_t * const dst, ; const int16_t * const src, ; uint32_t stride); ; -;=========================================================================== - -align 16 -cglobal transfer_16to8copy_mmx -transfer_16to8copy_mmx +;----------------------------------------------------------------------------- - push esi - push edi - - mov edi, [esp + 8 + 4] ; dst - mov esi, [esp + 8 + 8] ; src - mov ecx, [esp + 8 + 12] ; stride - - mov eax, 8 - -.loop - movq mm0, [esi] - packuswb mm0, [esi + 8] ; mm0 = pack([src]) - - movq [edi], mm0 ; [dst] = mm0 - - add esi, 16 - add edi, ecx - dec eax - jnz .loop - - pop edi - pop esi +%macro COPY_16_TO_8 1 + movq mm0, [eax+%1*32] + movq mm1, [eax+%1*32+8] + packuswb mm0, mm1 + movq [ecx], mm0 + movq mm2, [eax+%1*32+16] + movq mm3, [eax+%1*32+24] + packuswb mm2, mm3 + movq [ecx+edx], mm2 +%endmacro - ret +ALIGN 16 +transfer_16to8copy_mmx: + mov ecx, [esp+ 4] ; Dst + mov eax, [esp+ 8] ; Src + mov edx, [esp+12] ; Stride + + COPY_16_TO_8 0 + lea ecx,[ecx+2*edx] + COPY_16_TO_8 1 + lea ecx,[ecx+2*edx] + COPY_16_TO_8 2 + lea ecx,[ecx+2*edx] + COPY_16_TO_8 3 + ret -;=========================================================================== +;----------------------------------------------------------------------------- ; ; void transfer_8to16sub_mmx(int16_t * const dct, ; uint8_t * const cur, ; const uint8_t * const ref, ; const uint32_t stride); ; -;=========================================================================== -;/************************************************************************** -; * -; * History: -; * -; * 27.12.2001 renamed from 'compensate' to 'transfer_8to16sub' -; * 02.12.2001 loop unrolled, code runs 10% faster now (Isibaar) -; * 30.11.2001 16 pixels are processed per iteration (Isibaar) -; * 30.11.2001 .text missing -; * 06.11.2001 inital version; (c)2001 peter ross -; * -; *************************************************************************/ - -align 16 -cglobal transfer_8to16sub_mmx -transfer_8to16sub_mmx - push esi - push edi - push ebx - - mov edi, [esp + 12 + 4] ; dct [out] - mov edx, [esp + 12 + 8] ; cur [in/out] - mov esi, [esp + 12 + 12] ; ref [in] - mov ecx, [esp + 12 + 16] ; stride [in] - - mov eax, edx ; cur -> eax - mov ebx, esi ; ref -> ebx - add eax, ecx ; cur + stride - add ebx, ecx ; ref + stride - - shl ecx, 1 - - pxor mm7, mm7 ; mm7 = zero - - movq mm0, [edx] ; mm01 = [cur] - movq mm1, mm0 - - punpcklbw mm0, mm7 - punpckhbw mm1, mm7 - - movq mm4, [eax] - movq mm5, mm4 - - punpcklbw mm4, mm7 - punpckhbw mm5, mm7 - - movq mm2, [esi] ; mm23 = [ref] - movq mm3, mm2 - - movq mm6, [ebx] - - movq [edx], mm2 ; [cur] = [ref] - movq [eax], mm6 - - punpcklbw mm2, mm7 - punpckhbw mm3, mm7 - - psubsw mm0, mm2 ; mm01 -= mm23 - - movq mm2, mm6 - - punpcklbw mm2, mm7 - punpckhbw mm6, mm7 - - psubsw mm1, mm3 - - psubsw mm4, mm2 - psubsw mm5, mm6 - - movq [edi], mm0 ; dct[] = mm01 - movq [edi + 8], mm1 - movq [edi + 16], mm4 - movq [edi + 24], mm5 - - add edx, ecx - add esi, ecx - add eax, ecx - add ebx, ecx - - movq mm0, [edx] ; mm01 = [cur] - movq mm1, mm0 - - punpcklbw mm0, mm7 - punpckhbw mm1, mm7 - - movq mm4, [eax] - movq mm5, mm4 - - punpcklbw mm4, mm7 - punpckhbw mm5, mm7 - - movq mm2, [esi] ; mm23 = [ref] - movq mm3, mm2 - - movq mm6, [ebx] - - movq [edx], mm2 ; [cur] = [ref] - movq [eax], mm6 +;----------------------------------------------------------------------------- - punpcklbw mm2, mm7 - punpckhbw mm3, mm7 - - psubsw mm0, mm2 ; mm01 -= mm23 - - movq mm2, mm6 - - punpcklbw mm2, mm7 - punpckhbw mm6, mm7 - - psubsw mm1, mm3 - - psubsw mm4, mm2 - psubsw mm5, mm6 - - movq [edi + 32], mm0 ; dct[] = mm01 - movq [edi + 40], mm1 - movq [edi + 48], mm4 - movq [edi + 56], mm5 - - add edx, ecx - add esi, ecx - add eax, ecx - add ebx, ecx - - movq mm0, [edx] ; mm01 = [cur] - movq mm1, mm0 - - punpcklbw mm0, mm7 - punpckhbw mm1, mm7 - - movq mm4, [eax] - movq mm5, mm4 - - punpcklbw mm4, mm7 - punpckhbw mm5, mm7 - - movq mm2, [esi] ; mm23 = [ref] - movq mm3, mm2 - - movq mm6, [ebx] - - movq [edx], mm2 ; [cur] = [ref] - movq [eax], mm6 - - punpcklbw mm2, mm7 - punpckhbw mm3, mm7 - - psubsw mm0, mm2 ; mm01 -= mm23 - - movq mm2, mm6 - - punpcklbw mm2, mm7 - punpckhbw mm6, mm7 - - psubsw mm1, mm3 - - psubsw mm4, mm2 - psubsw mm5, mm6 - - movq [edi + 64], mm0 ; dct[] = mm01 - movq [edi + 72], mm1 - movq [edi + 80], mm4 - movq [edi + 88], mm5 - - add edx, ecx - add esi, ecx - add eax, ecx - add ebx, ecx - - movq mm0, [edx] ; mm01 = [cur] - movq mm1, mm0 - - punpcklbw mm0, mm7 - punpckhbw mm1, mm7 - - movq mm4, [eax] - movq mm5, mm4 - - punpcklbw mm4, mm7 - punpckhbw mm5, mm7 - - movq mm2, [esi] ; mm23 = [ref] - movq mm3, mm2 - - movq mm6, [ebx] - - movq [edx], mm2 ; [cur] = [ref] - movq [eax], mm6 - - punpcklbw mm2, mm7 - punpckhbw mm3, mm7 - - psubsw mm0, mm2 ; mm01 -= mm23 - - movq mm2, mm6 - - punpcklbw mm2, mm7 - punpckhbw mm6, mm7 +; when second argument == 1, reference (ebx) block is to current (eax) +%macro COPY_8_TO_16_SUB 2 + movq mm0, [eax] ; cur + movq mm2, [eax+edx] + movq mm1, mm0 + movq mm3, mm2 + + punpcklbw mm0, mm7 + punpcklbw mm2, mm7 + movq mm4, [ebx] ; ref + punpckhbw mm1, mm7 + punpckhbw mm3, mm7 + movq mm5, [ebx+edx] ; ref + + movq mm6, mm4 +%if %2 == 1 + movq [eax], mm4 + movq [eax+edx], mm5 +%endif + punpcklbw mm4, mm7 + punpckhbw mm6, mm7 + psubsw mm0, mm4 + psubsw mm1, mm6 + movq mm6, mm5 + punpcklbw mm5, mm7 + punpckhbw mm6, mm7 + psubsw mm2, mm5 + lea eax, [eax+2*edx] + psubsw mm3, mm6 + lea ebx,[ebx+2*edx] + + movq [ecx+%1*32+ 0], mm0 ; dst + movq [ecx+%1*32+ 8], mm1 + movq [ecx+%1*32+16], mm2 + movq [ecx+%1*32+24], mm3 +%endmacro - psubsw mm1, mm3 +ALIGN 16 +transfer_8to16sub_mmx: + mov ecx, [esp + 4] ; Dst + mov eax, [esp + 8] ; Cur + push ebx + mov ebx, [esp+4+12] ; Ref + mov edx, [esp+4+16] ; Stride + pxor mm7, mm7 + + COPY_8_TO_16_SUB 0, 1 + COPY_8_TO_16_SUB 1, 1 + COPY_8_TO_16_SUB 2, 1 + COPY_8_TO_16_SUB 3, 1 + + pop ebx + ret + + +ALIGN 16 +transfer_8to16subro_mmx: + mov ecx, [esp + 4] ; Dst + mov eax, [esp + 8] ; Cur + push ebx + mov ebx, [esp+4+12] ; Ref + mov edx, [esp+4+16] ; Stride + pxor mm7, mm7 + + COPY_8_TO_16_SUB 0, 0 + COPY_8_TO_16_SUB 1, 0 + COPY_8_TO_16_SUB 2, 0 + COPY_8_TO_16_SUB 3, 0 - psubsw mm4, mm2 - psubsw mm5, mm6 + pop ebx + ret - movq [edi + 96], mm0 ; dct[] = mm01 - movq [edi + 104], mm1 - movq [edi + 112], mm4 - movq [edi + 120], mm5 - pop ebx - pop edi - pop esi +;----------------------------------------------------------------------------- +; +; void transfer_8to16sub2_mmx(int16_t * const dct, +; uint8_t * const cur, +; const uint8_t * ref1, +; const uint8_t * ref2, +; const uint32_t stride) +; +;----------------------------------------------------------------------------- + +%macro COPY_8_TO_16_SUB2_MMX 1 + movq mm0, [eax] ; cur + movq mm2, [eax+edx] + + ; mm4 <- (ref1+ref2+1) / 2 + movq mm4, [ebx] ; ref1 + movq mm1, [esi] ; ref2 + movq mm6, mm4 + movq mm3, mm1 + punpcklbw mm4, mm7 + punpcklbw mm1, mm7 + punpckhbw mm6, mm7 + punpckhbw mm3, mm7 + paddusw mm4, mm1 + paddusw mm6, mm3 + paddusw mm4, [mmx_one] + paddusw mm6, [mmx_one] + psrlw mm4, 1 + psrlw mm6, 1 + packuswb mm4, mm6 + movq [eax], mm4 + + ; mm5 <- (ref1+ref2+1) / 2 + movq mm5, [ebx+edx] ; ref1 + movq mm1, [esi+edx] ; ref2 + movq mm6, mm5 + movq mm3, mm1 + punpcklbw mm5, mm7 + punpcklbw mm1, mm7 + punpckhbw mm6, mm7 + punpckhbw mm3, mm7 + paddusw mm5, mm1 + paddusw mm6, mm3 + paddusw mm5, [mmx_one] + paddusw mm6, [mmx_one] + lea esi, [esi+2*edx] + psrlw mm5, 1 + psrlw mm6, 1 + packuswb mm5, mm6 + movq [eax+edx], mm5 + + movq mm1, mm0 + movq mm3, mm2 + punpcklbw mm0, mm7 + punpcklbw mm2, mm7 + punpckhbw mm1, mm7 + punpckhbw mm3, mm7 + + movq mm6, mm4 + punpcklbw mm4, mm7 + punpckhbw mm6, mm7 + psubsw mm0, mm4 + psubsw mm1, mm6 + movq mm6, mm5 + punpcklbw mm5, mm7 + punpckhbw mm6, mm7 + psubsw mm2, mm5 + lea eax, [eax+2*edx] + psubsw mm3, mm6 + lea ebx, [ebx+2*edx] + + movq [ecx+%1*32+ 0], mm0 ; dst + movq [ecx+%1*32+ 8], mm1 + movq [ecx+%1*32+16], mm2 + movq [ecx+%1*32+24], mm3 +%endmacro - ret +ALIGN 16 +transfer_8to16sub2_mmx: + mov ecx, [esp + 4] ; Dst + mov eax, [esp + 8] ; Cur + push ebx + mov ebx, [esp+4+12] ; Ref1 + push esi + mov esi, [esp+8+16] ; Ref2 + mov edx, [esp+8+20] ; Stride + pxor mm7, mm7 + + COPY_8_TO_16_SUB2_MMX 0 + COPY_8_TO_16_SUB2_MMX 1 + COPY_8_TO_16_SUB2_MMX 2 + COPY_8_TO_16_SUB2_MMX 3 + + pop esi + pop ebx + ret +;----------------------------------------------------------------------------- +; +; void transfer_8to16sub2_xmm(int16_t * const dct, +; uint8_t * const cur, +; const uint8_t * ref1, +; const uint8_t * ref2, +; const uint32_t stride) +; +;----------------------------------------------------------------------------- + +%macro COPY_8_TO_16_SUB2_SSE 1 + movq mm0, [eax] ; cur + movq mm2, [eax+edx] + movq mm1, mm0 + movq mm3, mm2 + + punpcklbw mm0, mm7 + punpcklbw mm2, mm7 + movq mm4, [ebx] ; ref1 + pavgb mm4, [esi] ; ref2 + movq [eax], mm4 + punpckhbw mm1, mm7 + punpckhbw mm3, mm7 + movq mm5, [ebx+edx] ; ref + pavgb mm5, [esi+edx] ; ref2 + movq [eax+edx], mm5 + + movq mm6, mm4 + punpcklbw mm4, mm7 + punpckhbw mm6, mm7 + psubsw mm0, mm4 + psubsw mm1, mm6 + lea esi, [esi+2*edx] + movq mm6, mm5 + punpcklbw mm5, mm7 + punpckhbw mm6, mm7 + psubsw mm2, mm5 + lea eax, [eax+2*edx] + psubsw mm3, mm6 + lea ebx, [ebx+2*edx] + + movq [ecx+%1*32+ 0], mm0 ; dst + movq [ecx+%1*32+ 8], mm1 + movq [ecx+%1*32+16], mm2 + movq [ecx+%1*32+24], mm3 +%endmacro +ALIGN 16 +transfer_8to16sub2_xmm: + mov ecx, [esp + 4] ; Dst + mov eax, [esp + 8] ; Cur + push ebx + mov ebx, [esp+4+12] ; Ref1 + push esi + mov esi, [esp+8+16] ; Ref2 + mov edx, [esp+8+20] ; Stride + pxor mm7, mm7 + + COPY_8_TO_16_SUB2_SSE 0 + COPY_8_TO_16_SUB2_SSE 1 + COPY_8_TO_16_SUB2_SSE 2 + COPY_8_TO_16_SUB2_SSE 3 + + pop esi + pop ebx + ret -;=========================================================================== +;----------------------------------------------------------------------------- ; ; void transfer_16to8add_mmx(uint8_t * const dst, ; const int16_t * const src, ; uint32_t stride); ; -;=========================================================================== - -align 16 -cglobal transfer_16to8add_mmx -transfer_16to8add_mmx - - push esi - push edi +;----------------------------------------------------------------------------- - mov edi, [esp + 8 + 4] ; dst - mov esi, [esp + 8 + 8] ; src - mov ecx, [esp + 8 + 12] ; stride - - pxor mm7, mm7 - - mov eax, 8 - -.loop - movq mm0, [edi] - movq mm1, mm0 - punpcklbw mm0, mm7 ; mm23 = unpack([dst]) - punpckhbw mm1, mm7 - - movq mm2, [esi] ; mm01 = [src] - movq mm3, [esi + 8] - - paddsw mm0, mm2 ; mm01 += mm23 - paddsw mm1, mm3 - - packuswb mm0, mm1 ; [dst] = pack(mm01) - movq [edi], mm0 - - add esi, 16 - add edi, ecx - dec eax - jnz .loop - - pop edi - pop esi +%macro COPY_16_TO_8_ADD 1 + movq mm0, [ecx] + movq mm2, [ecx+edx] + movq mm1, mm0 + movq mm3, mm2 + punpcklbw mm0, mm7 + punpcklbw mm2, mm7 + punpckhbw mm1, mm7 + punpckhbw mm3, mm7 + paddsw mm0, [eax+%1*32+ 0] + paddsw mm1, [eax+%1*32+ 8] + paddsw mm2, [eax+%1*32+16] + paddsw mm3, [eax+%1*32+24] + packuswb mm0, mm1 + movq [ecx], mm0 + packuswb mm2, mm3 + movq [ecx+edx], mm2 +%endmacro - ret +ALIGN 16 +transfer_16to8add_mmx: + mov ecx, [esp+ 4] ; Dst + mov eax, [esp+ 8] ; Src + mov edx, [esp+12] ; Stride + pxor mm7, mm7 + + COPY_16_TO_8_ADD 0 + lea ecx,[ecx+2*edx] + COPY_16_TO_8_ADD 1 + lea ecx,[ecx+2*edx] + COPY_16_TO_8_ADD 2 + lea ecx,[ecx+2*edx] + COPY_16_TO_8_ADD 3 + ret -;=========================================================================== +;----------------------------------------------------------------------------- ; ; void transfer8x8_copy_mmx(uint8_t * const dst, ; const uint8_t * const src, ; const uint32_t stride); ; ; -;=========================================================================== +;----------------------------------------------------------------------------- -align 16 -cglobal transfer8x8_copy_mmx -transfer8x8_copy_mmx - push esi - push edi - - mov edi, [esp + 8 + 4] ; dst [out] - mov esi, [esp + 8 + 8] ; src [in] - mov eax, [esp + 8 + 12] ; stride [in] - - movq mm0, [esi] - movq mm1, [esi+eax] - movq [edi], mm0 - movq [edi+eax], mm1 - - add esi, eax - add edi, eax - add esi, eax - add edi, eax - - movq mm0, [esi] - movq mm1, [esi+eax] - movq [edi], mm0 - movq [edi+eax], mm1 - - add esi, eax - add edi, eax - add esi, eax - add edi, eax - - movq mm0, [esi] - movq mm1, [esi+eax] - movq [edi], mm0 - movq [edi+eax], mm1 - - add esi, eax - add edi, eax - add esi, eax - add edi, eax - - movq mm0, [esi] - movq mm1, [esi+eax] - movq [edi], mm0 - movq [edi+eax], mm1 - - add esi, eax - add edi, eax - add esi, eax - add edi, eax - - pop edi - pop esi +%macro COPY_8_TO_8 0 + movq mm0, [eax] + movq mm1, [eax+edx] + movq [ecx], mm0 + lea eax, [eax+2*edx] + movq [ecx+edx], mm1 +%endmacro - ret +ALIGN 16 +transfer8x8_copy_mmx: + mov ecx, [esp+ 4] ; Dst + mov eax, [esp+ 8] ; Src + mov edx, [esp+12] ; Stride + + COPY_8_TO_8 + lea ecx,[ecx+2*edx] + COPY_8_TO_8 + lea ecx,[ecx+2*edx] + COPY_8_TO_8 + lea ecx,[ecx+2*edx] + COPY_8_TO_8 + ret