mirror of
https://github.com/mozilla/gecko-dev.git
synced 2024-12-26 02:02:33 +00:00
2d0b90011e
--HG-- rename : jpeg/MOZCHANGES => media/libjpeg/MOZCHANGES rename : jpeg/Makefile.in => media/libjpeg/Makefile.in rename : jpeg/README => media/libjpeg/README rename : jpeg/README-turbo.txt => media/libjpeg/README-turbo.txt rename : jpeg/cderror.h => media/libjpeg/cderror.h rename : jpeg/cdjpeg.h => media/libjpeg/cdjpeg.h rename : jpeg/jaricom.c => media/libjpeg/jaricom.c rename : jpeg/jcapimin.c => media/libjpeg/jcapimin.c rename : jpeg/jcapistd.c => media/libjpeg/jcapistd.c rename : jpeg/jcarith.c => media/libjpeg/jcarith.c rename : jpeg/jccoefct.c => media/libjpeg/jccoefct.c rename : jpeg/jccolor.c => media/libjpeg/jccolor.c rename : jpeg/jcdctmgr.c => media/libjpeg/jcdctmgr.c rename : jpeg/jchuff.c => media/libjpeg/jchuff.c rename : jpeg/jchuff.h => media/libjpeg/jchuff.h rename : jpeg/jcinit.c => media/libjpeg/jcinit.c rename : jpeg/jcmainct.c => media/libjpeg/jcmainct.c rename : jpeg/jcmarker.c => media/libjpeg/jcmarker.c rename : jpeg/jcmaster.c => media/libjpeg/jcmaster.c rename : jpeg/jcomapi.c => media/libjpeg/jcomapi.c rename : jpeg/jconfig.h => media/libjpeg/jconfig.h rename : jpeg/jconfig.h.in => media/libjpeg/jconfig.h.in rename : jpeg/jcparam.c => media/libjpeg/jcparam.c rename : jpeg/jcphuff.c => media/libjpeg/jcphuff.c rename : jpeg/jcprepct.c => media/libjpeg/jcprepct.c rename : jpeg/jcsample.c => media/libjpeg/jcsample.c rename : jpeg/jdapimin.c => media/libjpeg/jdapimin.c rename : jpeg/jdapistd.c => media/libjpeg/jdapistd.c rename : jpeg/jdarith.c => media/libjpeg/jdarith.c rename : jpeg/jdatadst.c => media/libjpeg/jdatadst.c rename : jpeg/jdatasrc.c => media/libjpeg/jdatasrc.c rename : jpeg/jdcoefct.c => media/libjpeg/jdcoefct.c rename : jpeg/jdcolor.c => media/libjpeg/jdcolor.c rename : jpeg/jdct.h => media/libjpeg/jdct.h rename : jpeg/jddctmgr.c => media/libjpeg/jddctmgr.c rename : jpeg/jdhuff.c => media/libjpeg/jdhuff.c rename : jpeg/jdhuff.h => media/libjpeg/jdhuff.h rename : jpeg/jdinput.c => media/libjpeg/jdinput.c rename : jpeg/jdmainct.c => media/libjpeg/jdmainct.c rename : jpeg/jdmarker.c => media/libjpeg/jdmarker.c rename : jpeg/jdmaster.c => media/libjpeg/jdmaster.c rename : jpeg/jdmerge.c => media/libjpeg/jdmerge.c rename : jpeg/jdphuff.c => media/libjpeg/jdphuff.c rename : jpeg/jdpostct.c => media/libjpeg/jdpostct.c rename : jpeg/jdsample.c => media/libjpeg/jdsample.c rename : jpeg/jdtrans.c => media/libjpeg/jdtrans.c rename : jpeg/jerror.c => media/libjpeg/jerror.c rename : jpeg/jerror.h => media/libjpeg/jerror.h rename : jpeg/jfdctflt.c => media/libjpeg/jfdctflt.c rename : jpeg/jfdctfst.c => media/libjpeg/jfdctfst.c rename : jpeg/jfdctint.c => media/libjpeg/jfdctint.c rename : jpeg/jidctflt.c => media/libjpeg/jidctflt.c rename : jpeg/jidctfst.c => media/libjpeg/jidctfst.c rename : jpeg/jidctint.c => media/libjpeg/jidctint.c rename : jpeg/jidctred.c => media/libjpeg/jidctred.c rename : jpeg/jinclude.h => media/libjpeg/jinclude.h rename : jpeg/jmemmgr.c => media/libjpeg/jmemmgr.c rename : jpeg/jmemnobs.c => media/libjpeg/jmemnobs.c rename : jpeg/jmemsys.h => media/libjpeg/jmemsys.h rename : jpeg/jmorecfg.h => media/libjpeg/jmorecfg.h rename : jpeg/jpegcomp.h => media/libjpeg/jpegcomp.h rename : jpeg/jpegint.h => media/libjpeg/jpegint.h rename : jpeg/jpeglib.h => media/libjpeg/jpeglib.h rename : jpeg/jquant1.c => media/libjpeg/jquant1.c rename : jpeg/jquant2.c => media/libjpeg/jquant2.c rename : jpeg/jsimd.h => media/libjpeg/jsimd.h rename : jpeg/jsimd_none.c => media/libjpeg/jsimd_none.c rename : jpeg/jsimddct.h => media/libjpeg/jsimddct.h rename : jpeg/jutils.c => media/libjpeg/jutils.c rename : jpeg/jversion.h => media/libjpeg/jversion.h rename : jpeg/simd/Makefile.in => media/libjpeg/simd/Makefile.in rename : jpeg/simd/jcclrmmx.asm => media/libjpeg/simd/jcclrmmx.asm rename : jpeg/simd/jcclrss2-64.asm => media/libjpeg/simd/jcclrss2-64.asm rename : jpeg/simd/jcclrss2.asm => media/libjpeg/simd/jcclrss2.asm rename : jpeg/simd/jccolmmx.asm => media/libjpeg/simd/jccolmmx.asm rename : jpeg/simd/jccolss2-64.asm => media/libjpeg/simd/jccolss2-64.asm rename : jpeg/simd/jccolss2.asm => media/libjpeg/simd/jccolss2.asm rename : jpeg/simd/jcolsamp.inc => media/libjpeg/simd/jcolsamp.inc rename : jpeg/simd/jcqnt3dn.asm => media/libjpeg/simd/jcqnt3dn.asm rename : jpeg/simd/jcqntmmx.asm => media/libjpeg/simd/jcqntmmx.asm rename : jpeg/simd/jcqnts2f-64.asm => media/libjpeg/simd/jcqnts2f-64.asm rename : jpeg/simd/jcqnts2f.asm => media/libjpeg/simd/jcqnts2f.asm rename : jpeg/simd/jcqnts2i-64.asm => media/libjpeg/simd/jcqnts2i-64.asm rename : jpeg/simd/jcqnts2i.asm => media/libjpeg/simd/jcqnts2i.asm rename : jpeg/simd/jcqntsse.asm => media/libjpeg/simd/jcqntsse.asm rename : jpeg/simd/jcsammmx.asm => media/libjpeg/simd/jcsammmx.asm rename : jpeg/simd/jcsamss2-64.asm => media/libjpeg/simd/jcsamss2-64.asm rename : jpeg/simd/jcsamss2.asm => media/libjpeg/simd/jcsamss2.asm rename : jpeg/simd/jdclrmmx.asm => media/libjpeg/simd/jdclrmmx.asm rename : jpeg/simd/jdclrss2-64.asm => media/libjpeg/simd/jdclrss2-64.asm rename : jpeg/simd/jdclrss2.asm => media/libjpeg/simd/jdclrss2.asm rename : jpeg/simd/jdcolmmx.asm => media/libjpeg/simd/jdcolmmx.asm rename : jpeg/simd/jdcolss2-64.asm => media/libjpeg/simd/jdcolss2-64.asm rename : jpeg/simd/jdcolss2.asm => media/libjpeg/simd/jdcolss2.asm rename : jpeg/simd/jdct.inc => media/libjpeg/simd/jdct.inc rename : jpeg/simd/jdmermmx.asm => media/libjpeg/simd/jdmermmx.asm rename : jpeg/simd/jdmerss2-64.asm => media/libjpeg/simd/jdmerss2-64.asm rename : jpeg/simd/jdmerss2.asm => media/libjpeg/simd/jdmerss2.asm rename : jpeg/simd/jdmrgmmx.asm => media/libjpeg/simd/jdmrgmmx.asm rename : jpeg/simd/jdmrgss2-64.asm => media/libjpeg/simd/jdmrgss2-64.asm rename : jpeg/simd/jdmrgss2.asm => media/libjpeg/simd/jdmrgss2.asm rename : jpeg/simd/jdsammmx.asm => media/libjpeg/simd/jdsammmx.asm rename : jpeg/simd/jdsamss2-64.asm => media/libjpeg/simd/jdsamss2-64.asm rename : jpeg/simd/jdsamss2.asm => media/libjpeg/simd/jdsamss2.asm rename : jpeg/simd/jf3dnflt.asm => media/libjpeg/simd/jf3dnflt.asm rename : jpeg/simd/jfmmxfst.asm => media/libjpeg/simd/jfmmxfst.asm rename : jpeg/simd/jfmmxint.asm => media/libjpeg/simd/jfmmxint.asm rename : jpeg/simd/jfss2fst-64.asm => media/libjpeg/simd/jfss2fst-64.asm rename : jpeg/simd/jfss2fst.asm => media/libjpeg/simd/jfss2fst.asm rename : jpeg/simd/jfss2int-64.asm => media/libjpeg/simd/jfss2int-64.asm rename : jpeg/simd/jfss2int.asm => media/libjpeg/simd/jfss2int.asm rename : jpeg/simd/jfsseflt-64.asm => media/libjpeg/simd/jfsseflt-64.asm rename : jpeg/simd/jfsseflt.asm => media/libjpeg/simd/jfsseflt.asm rename : jpeg/simd/ji3dnflt.asm => media/libjpeg/simd/ji3dnflt.asm rename : jpeg/simd/jimmxfst.asm => media/libjpeg/simd/jimmxfst.asm rename : jpeg/simd/jimmxint.asm => media/libjpeg/simd/jimmxint.asm rename : jpeg/simd/jimmxred.asm => media/libjpeg/simd/jimmxred.asm rename : jpeg/simd/jiss2flt-64.asm => media/libjpeg/simd/jiss2flt-64.asm rename : jpeg/simd/jiss2flt.asm => media/libjpeg/simd/jiss2flt.asm rename : jpeg/simd/jiss2fst-64.asm => media/libjpeg/simd/jiss2fst-64.asm rename : jpeg/simd/jiss2fst.asm => media/libjpeg/simd/jiss2fst.asm rename : jpeg/simd/jiss2int-64.asm => media/libjpeg/simd/jiss2int-64.asm rename : jpeg/simd/jiss2int.asm => media/libjpeg/simd/jiss2int.asm rename : jpeg/simd/jiss2red-64.asm => media/libjpeg/simd/jiss2red-64.asm rename : jpeg/simd/jiss2red.asm => media/libjpeg/simd/jiss2red.asm rename : jpeg/simd/jisseflt.asm => media/libjpeg/simd/jisseflt.asm rename : jpeg/simd/jsimd.h => media/libjpeg/simd/jsimd.h rename : jpeg/simd/jsimd_i386.c => media/libjpeg/simd/jsimd_i386.c rename : jpeg/simd/jsimd_x86_64.c => media/libjpeg/simd/jsimd_x86_64.c rename : jpeg/simd/jsimdcfg.inc => media/libjpeg/simd/jsimdcfg.inc rename : jpeg/simd/jsimdcfg.inc.h => media/libjpeg/simd/jsimdcfg.inc.h rename : jpeg/simd/jsimdcpu.asm => media/libjpeg/simd/jsimdcpu.asm rename : jpeg/simd/jsimdext.inc => media/libjpeg/simd/jsimdext.inc rename : jpeg/transupp.h => media/libjpeg/transupp.h
453 lines
13 KiB
NASM
453 lines
13 KiB
NASM
;
|
|
; ji3dnflt.asm - floating-point IDCT (3DNow! & MMX)
|
|
;
|
|
; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
|
|
;
|
|
; Based on
|
|
; x86 SIMD extension for IJG JPEG library
|
|
; Copyright (C) 1999-2006, MIYASAKA Masaru.
|
|
; For conditions of distribution and use, see copyright notice in jsimdext.inc
|
|
;
|
|
; This file should be assembled with NASM (Netwide Assembler),
|
|
; can *not* be assembled with Microsoft's MASM or any compatible
|
|
; assembler (including Borland's Turbo Assembler).
|
|
; NASM is available from http://nasm.sourceforge.net/ or
|
|
; http://sourceforge.net/project/showfiles.php?group_id=6208
|
|
;
|
|
; This file contains a floating-point implementation of the inverse DCT
|
|
; (Discrete Cosine Transform). The following code is based directly on
|
|
; the IJG's original jidctflt.c; see the jidctflt.c for more details.
|
|
;
|
|
; [TAB8]
|
|
|
|
%include "jsimdext.inc"
|
|
%include "jdct.inc"
|
|
|
|
; --------------------------------------------------------------------------
|
|
SECTION SEG_CONST
|
|
|
|
alignz 16
|
|
global EXTN(jconst_idct_float_3dnow)
|
|
|
|
EXTN(jconst_idct_float_3dnow):
|
|
|
|
PD_1_414 times 2 dd 1.414213562373095048801689
|
|
PD_1_847 times 2 dd 1.847759065022573512256366
|
|
PD_1_082 times 2 dd 1.082392200292393968799446
|
|
PD_2_613 times 2 dd 2.613125929752753055713286
|
|
PD_RNDINT_MAGIC times 2 dd 100663296.0 ; (float)(0x00C00000 << 3)
|
|
PB_CENTERJSAMP times 8 db CENTERJSAMPLE
|
|
|
|
alignz 16
|
|
|
|
; --------------------------------------------------------------------------
|
|
SECTION SEG_TEXT
|
|
BITS 32
|
|
;
|
|
; Perform dequantization and inverse DCT on one block of coefficients.
|
|
;
|
|
; GLOBAL(void)
|
|
; jsimd_idct_float_3dnow (void * dct_table, JCOEFPTR coef_block,
|
|
; JSAMPARRAY output_buf, JDIMENSION output_col)
|
|
;
|
|
|
|
%define dct_table(b) (b)+8 ; void * dct_table
|
|
%define coef_block(b) (b)+12 ; JCOEFPTR coef_block
|
|
%define output_buf(b) (b)+16 ; JSAMPARRAY output_buf
|
|
%define output_col(b) (b)+20 ; JDIMENSION output_col
|
|
|
|
%define original_ebp ebp+0
|
|
%define wk(i) ebp-(WK_NUM-(i))*SIZEOF_MMWORD ; mmword wk[WK_NUM]
|
|
%define WK_NUM 2
|
|
%define workspace wk(0)-DCTSIZE2*SIZEOF_FAST_FLOAT
|
|
; FAST_FLOAT workspace[DCTSIZE2]
|
|
|
|
align 16
|
|
global EXTN(jsimd_idct_float_3dnow)
|
|
|
|
EXTN(jsimd_idct_float_3dnow):
|
|
push ebp
|
|
mov eax,esp ; eax = original ebp
|
|
sub esp, byte 4
|
|
and esp, byte (-SIZEOF_MMWORD) ; align to 64 bits
|
|
mov [esp],eax
|
|
mov ebp,esp ; ebp = aligned ebp
|
|
lea esp, [workspace]
|
|
push ebx
|
|
; push ecx ; need not be preserved
|
|
; push edx ; need not be preserved
|
|
push esi
|
|
push edi
|
|
|
|
get_GOT ebx ; get GOT address
|
|
|
|
; ---- Pass 1: process columns from input, store into work array.
|
|
|
|
; mov eax, [original_ebp]
|
|
mov edx, POINTER [dct_table(eax)] ; quantptr
|
|
mov esi, JCOEFPTR [coef_block(eax)] ; inptr
|
|
lea edi, [workspace] ; FAST_FLOAT * wsptr
|
|
mov ecx, DCTSIZE/2 ; ctr
|
|
alignx 16,7
|
|
.columnloop:
|
|
%ifndef NO_ZERO_COLUMN_TEST_FLOAT_3DNOW
|
|
mov eax, DWORD [DWBLOCK(1,0,esi,SIZEOF_JCOEF)]
|
|
or eax, DWORD [DWBLOCK(2,0,esi,SIZEOF_JCOEF)]
|
|
jnz short .columnDCT
|
|
|
|
pushpic ebx ; save GOT address
|
|
mov ebx, DWORD [DWBLOCK(3,0,esi,SIZEOF_JCOEF)]
|
|
mov eax, DWORD [DWBLOCK(4,0,esi,SIZEOF_JCOEF)]
|
|
or ebx, DWORD [DWBLOCK(5,0,esi,SIZEOF_JCOEF)]
|
|
or eax, DWORD [DWBLOCK(6,0,esi,SIZEOF_JCOEF)]
|
|
or ebx, DWORD [DWBLOCK(7,0,esi,SIZEOF_JCOEF)]
|
|
or eax,ebx
|
|
poppic ebx ; restore GOT address
|
|
jnz short .columnDCT
|
|
|
|
; -- AC terms all zero
|
|
|
|
movd mm0, DWORD [DWBLOCK(0,0,esi,SIZEOF_JCOEF)]
|
|
|
|
punpcklwd mm0,mm0
|
|
psrad mm0,(DWORD_BIT-WORD_BIT)
|
|
pi2fd mm0,mm0
|
|
|
|
pfmul mm0, MMWORD [MMBLOCK(0,0,edx,SIZEOF_FLOAT_MULT_TYPE)]
|
|
|
|
movq mm1,mm0
|
|
punpckldq mm0,mm0
|
|
punpckhdq mm1,mm1
|
|
|
|
movq MMWORD [MMBLOCK(0,0,edi,SIZEOF_FAST_FLOAT)], mm0
|
|
movq MMWORD [MMBLOCK(0,1,edi,SIZEOF_FAST_FLOAT)], mm0
|
|
movq MMWORD [MMBLOCK(0,2,edi,SIZEOF_FAST_FLOAT)], mm0
|
|
movq MMWORD [MMBLOCK(0,3,edi,SIZEOF_FAST_FLOAT)], mm0
|
|
movq MMWORD [MMBLOCK(1,0,edi,SIZEOF_FAST_FLOAT)], mm1
|
|
movq MMWORD [MMBLOCK(1,1,edi,SIZEOF_FAST_FLOAT)], mm1
|
|
movq MMWORD [MMBLOCK(1,2,edi,SIZEOF_FAST_FLOAT)], mm1
|
|
movq MMWORD [MMBLOCK(1,3,edi,SIZEOF_FAST_FLOAT)], mm1
|
|
jmp near .nextcolumn
|
|
alignx 16,7
|
|
%endif
|
|
.columnDCT:
|
|
|
|
; -- Even part
|
|
|
|
movd mm0, DWORD [DWBLOCK(0,0,esi,SIZEOF_JCOEF)]
|
|
movd mm1, DWORD [DWBLOCK(2,0,esi,SIZEOF_JCOEF)]
|
|
movd mm2, DWORD [DWBLOCK(4,0,esi,SIZEOF_JCOEF)]
|
|
movd mm3, DWORD [DWBLOCK(6,0,esi,SIZEOF_JCOEF)]
|
|
|
|
punpcklwd mm0,mm0
|
|
punpcklwd mm1,mm1
|
|
psrad mm0,(DWORD_BIT-WORD_BIT)
|
|
psrad mm1,(DWORD_BIT-WORD_BIT)
|
|
pi2fd mm0,mm0
|
|
pi2fd mm1,mm1
|
|
|
|
pfmul mm0, MMWORD [MMBLOCK(0,0,edx,SIZEOF_FLOAT_MULT_TYPE)]
|
|
pfmul mm1, MMWORD [MMBLOCK(2,0,edx,SIZEOF_FLOAT_MULT_TYPE)]
|
|
|
|
punpcklwd mm2,mm2
|
|
punpcklwd mm3,mm3
|
|
psrad mm2,(DWORD_BIT-WORD_BIT)
|
|
psrad mm3,(DWORD_BIT-WORD_BIT)
|
|
pi2fd mm2,mm2
|
|
pi2fd mm3,mm3
|
|
|
|
pfmul mm2, MMWORD [MMBLOCK(4,0,edx,SIZEOF_FLOAT_MULT_TYPE)]
|
|
pfmul mm3, MMWORD [MMBLOCK(6,0,edx,SIZEOF_FLOAT_MULT_TYPE)]
|
|
|
|
movq mm4,mm0
|
|
movq mm5,mm1
|
|
pfsub mm0,mm2 ; mm0=tmp11
|
|
pfsub mm1,mm3
|
|
pfadd mm4,mm2 ; mm4=tmp10
|
|
pfadd mm5,mm3 ; mm5=tmp13
|
|
|
|
pfmul mm1,[GOTOFF(ebx,PD_1_414)]
|
|
pfsub mm1,mm5 ; mm1=tmp12
|
|
|
|
movq mm6,mm4
|
|
movq mm7,mm0
|
|
pfsub mm4,mm5 ; mm4=tmp3
|
|
pfsub mm0,mm1 ; mm0=tmp2
|
|
pfadd mm6,mm5 ; mm6=tmp0
|
|
pfadd mm7,mm1 ; mm7=tmp1
|
|
|
|
movq MMWORD [wk(1)], mm4 ; tmp3
|
|
movq MMWORD [wk(0)], mm0 ; tmp2
|
|
|
|
; -- Odd part
|
|
|
|
movd mm2, DWORD [DWBLOCK(1,0,esi,SIZEOF_JCOEF)]
|
|
movd mm3, DWORD [DWBLOCK(3,0,esi,SIZEOF_JCOEF)]
|
|
movd mm5, DWORD [DWBLOCK(5,0,esi,SIZEOF_JCOEF)]
|
|
movd mm1, DWORD [DWBLOCK(7,0,esi,SIZEOF_JCOEF)]
|
|
|
|
punpcklwd mm2,mm2
|
|
punpcklwd mm3,mm3
|
|
psrad mm2,(DWORD_BIT-WORD_BIT)
|
|
psrad mm3,(DWORD_BIT-WORD_BIT)
|
|
pi2fd mm2,mm2
|
|
pi2fd mm3,mm3
|
|
|
|
pfmul mm2, MMWORD [MMBLOCK(1,0,edx,SIZEOF_FLOAT_MULT_TYPE)]
|
|
pfmul mm3, MMWORD [MMBLOCK(3,0,edx,SIZEOF_FLOAT_MULT_TYPE)]
|
|
|
|
punpcklwd mm5,mm5
|
|
punpcklwd mm1,mm1
|
|
psrad mm5,(DWORD_BIT-WORD_BIT)
|
|
psrad mm1,(DWORD_BIT-WORD_BIT)
|
|
pi2fd mm5,mm5
|
|
pi2fd mm1,mm1
|
|
|
|
pfmul mm5, MMWORD [MMBLOCK(5,0,edx,SIZEOF_FLOAT_MULT_TYPE)]
|
|
pfmul mm1, MMWORD [MMBLOCK(7,0,edx,SIZEOF_FLOAT_MULT_TYPE)]
|
|
|
|
movq mm4,mm2
|
|
movq mm0,mm5
|
|
pfadd mm2,mm1 ; mm2=z11
|
|
pfadd mm5,mm3 ; mm5=z13
|
|
pfsub mm4,mm1 ; mm4=z12
|
|
pfsub mm0,mm3 ; mm0=z10
|
|
|
|
movq mm1,mm2
|
|
pfsub mm2,mm5
|
|
pfadd mm1,mm5 ; mm1=tmp7
|
|
|
|
pfmul mm2,[GOTOFF(ebx,PD_1_414)] ; mm2=tmp11
|
|
|
|
movq mm3,mm0
|
|
pfadd mm0,mm4
|
|
pfmul mm0,[GOTOFF(ebx,PD_1_847)] ; mm0=z5
|
|
pfmul mm3,[GOTOFF(ebx,PD_2_613)] ; mm3=(z10 * 2.613125930)
|
|
pfmul mm4,[GOTOFF(ebx,PD_1_082)] ; mm4=(z12 * 1.082392200)
|
|
pfsubr mm3,mm0 ; mm3=tmp12
|
|
pfsub mm4,mm0 ; mm4=tmp10
|
|
|
|
; -- Final output stage
|
|
|
|
pfsub mm3,mm1 ; mm3=tmp6
|
|
movq mm5,mm6
|
|
movq mm0,mm7
|
|
pfadd mm6,mm1 ; mm6=data0=(00 01)
|
|
pfadd mm7,mm3 ; mm7=data1=(10 11)
|
|
pfsub mm5,mm1 ; mm5=data7=(70 71)
|
|
pfsub mm0,mm3 ; mm0=data6=(60 61)
|
|
pfsub mm2,mm3 ; mm2=tmp5
|
|
|
|
movq mm1,mm6 ; transpose coefficients
|
|
punpckldq mm6,mm7 ; mm6=(00 10)
|
|
punpckhdq mm1,mm7 ; mm1=(01 11)
|
|
movq mm3,mm0 ; transpose coefficients
|
|
punpckldq mm0,mm5 ; mm0=(60 70)
|
|
punpckhdq mm3,mm5 ; mm3=(61 71)
|
|
|
|
movq MMWORD [MMBLOCK(0,0,edi,SIZEOF_FAST_FLOAT)], mm6
|
|
movq MMWORD [MMBLOCK(1,0,edi,SIZEOF_FAST_FLOAT)], mm1
|
|
movq MMWORD [MMBLOCK(0,3,edi,SIZEOF_FAST_FLOAT)], mm0
|
|
movq MMWORD [MMBLOCK(1,3,edi,SIZEOF_FAST_FLOAT)], mm3
|
|
|
|
movq mm7, MMWORD [wk(0)] ; mm7=tmp2
|
|
movq mm5, MMWORD [wk(1)] ; mm5=tmp3
|
|
|
|
pfadd mm4,mm2 ; mm4=tmp4
|
|
movq mm6,mm7
|
|
movq mm1,mm5
|
|
pfadd mm7,mm2 ; mm7=data2=(20 21)
|
|
pfadd mm5,mm4 ; mm5=data4=(40 41)
|
|
pfsub mm6,mm2 ; mm6=data5=(50 51)
|
|
pfsub mm1,mm4 ; mm1=data3=(30 31)
|
|
|
|
movq mm0,mm7 ; transpose coefficients
|
|
punpckldq mm7,mm1 ; mm7=(20 30)
|
|
punpckhdq mm0,mm1 ; mm0=(21 31)
|
|
movq mm3,mm5 ; transpose coefficients
|
|
punpckldq mm5,mm6 ; mm5=(40 50)
|
|
punpckhdq mm3,mm6 ; mm3=(41 51)
|
|
|
|
movq MMWORD [MMBLOCK(0,1,edi,SIZEOF_FAST_FLOAT)], mm7
|
|
movq MMWORD [MMBLOCK(1,1,edi,SIZEOF_FAST_FLOAT)], mm0
|
|
movq MMWORD [MMBLOCK(0,2,edi,SIZEOF_FAST_FLOAT)], mm5
|
|
movq MMWORD [MMBLOCK(1,2,edi,SIZEOF_FAST_FLOAT)], mm3
|
|
|
|
.nextcolumn:
|
|
add esi, byte 2*SIZEOF_JCOEF ; coef_block
|
|
add edx, byte 2*SIZEOF_FLOAT_MULT_TYPE ; quantptr
|
|
add edi, byte 2*DCTSIZE*SIZEOF_FAST_FLOAT ; wsptr
|
|
dec ecx ; ctr
|
|
jnz near .columnloop
|
|
|
|
; -- Prefetch the next coefficient block
|
|
|
|
prefetch [esi + (DCTSIZE2-8)*SIZEOF_JCOEF + 0*32]
|
|
prefetch [esi + (DCTSIZE2-8)*SIZEOF_JCOEF + 1*32]
|
|
prefetch [esi + (DCTSIZE2-8)*SIZEOF_JCOEF + 2*32]
|
|
prefetch [esi + (DCTSIZE2-8)*SIZEOF_JCOEF + 3*32]
|
|
|
|
; ---- Pass 2: process rows from work array, store into output array.
|
|
|
|
mov eax, [original_ebp]
|
|
lea esi, [workspace] ; FAST_FLOAT * wsptr
|
|
mov edi, JSAMPARRAY [output_buf(eax)] ; (JSAMPROW *)
|
|
mov eax, JDIMENSION [output_col(eax)]
|
|
mov ecx, DCTSIZE/2 ; ctr
|
|
alignx 16,7
|
|
.rowloop:
|
|
|
|
; -- Even part
|
|
|
|
movq mm0, MMWORD [MMBLOCK(0,0,esi,SIZEOF_FAST_FLOAT)]
|
|
movq mm1, MMWORD [MMBLOCK(2,0,esi,SIZEOF_FAST_FLOAT)]
|
|
movq mm2, MMWORD [MMBLOCK(4,0,esi,SIZEOF_FAST_FLOAT)]
|
|
movq mm3, MMWORD [MMBLOCK(6,0,esi,SIZEOF_FAST_FLOAT)]
|
|
|
|
movq mm4,mm0
|
|
movq mm5,mm1
|
|
pfsub mm0,mm2 ; mm0=tmp11
|
|
pfsub mm1,mm3
|
|
pfadd mm4,mm2 ; mm4=tmp10
|
|
pfadd mm5,mm3 ; mm5=tmp13
|
|
|
|
pfmul mm1,[GOTOFF(ebx,PD_1_414)]
|
|
pfsub mm1,mm5 ; mm1=tmp12
|
|
|
|
movq mm6,mm4
|
|
movq mm7,mm0
|
|
pfsub mm4,mm5 ; mm4=tmp3
|
|
pfsub mm0,mm1 ; mm0=tmp2
|
|
pfadd mm6,mm5 ; mm6=tmp0
|
|
pfadd mm7,mm1 ; mm7=tmp1
|
|
|
|
movq MMWORD [wk(1)], mm4 ; tmp3
|
|
movq MMWORD [wk(0)], mm0 ; tmp2
|
|
|
|
; -- Odd part
|
|
|
|
movq mm2, MMWORD [MMBLOCK(1,0,esi,SIZEOF_FAST_FLOAT)]
|
|
movq mm3, MMWORD [MMBLOCK(3,0,esi,SIZEOF_FAST_FLOAT)]
|
|
movq mm5, MMWORD [MMBLOCK(5,0,esi,SIZEOF_FAST_FLOAT)]
|
|
movq mm1, MMWORD [MMBLOCK(7,0,esi,SIZEOF_FAST_FLOAT)]
|
|
|
|
movq mm4,mm2
|
|
movq mm0,mm5
|
|
pfadd mm2,mm1 ; mm2=z11
|
|
pfadd mm5,mm3 ; mm5=z13
|
|
pfsub mm4,mm1 ; mm4=z12
|
|
pfsub mm0,mm3 ; mm0=z10
|
|
|
|
movq mm1,mm2
|
|
pfsub mm2,mm5
|
|
pfadd mm1,mm5 ; mm1=tmp7
|
|
|
|
pfmul mm2,[GOTOFF(ebx,PD_1_414)] ; mm2=tmp11
|
|
|
|
movq mm3,mm0
|
|
pfadd mm0,mm4
|
|
pfmul mm0,[GOTOFF(ebx,PD_1_847)] ; mm0=z5
|
|
pfmul mm3,[GOTOFF(ebx,PD_2_613)] ; mm3=(z10 * 2.613125930)
|
|
pfmul mm4,[GOTOFF(ebx,PD_1_082)] ; mm4=(z12 * 1.082392200)
|
|
pfsubr mm3,mm0 ; mm3=tmp12
|
|
pfsub mm4,mm0 ; mm4=tmp10
|
|
|
|
; -- Final output stage
|
|
|
|
pfsub mm3,mm1 ; mm3=tmp6
|
|
movq mm5,mm6
|
|
movq mm0,mm7
|
|
pfadd mm6,mm1 ; mm6=data0=(00 10)
|
|
pfadd mm7,mm3 ; mm7=data1=(01 11)
|
|
pfsub mm5,mm1 ; mm5=data7=(07 17)
|
|
pfsub mm0,mm3 ; mm0=data6=(06 16)
|
|
pfsub mm2,mm3 ; mm2=tmp5
|
|
|
|
movq mm1,[GOTOFF(ebx,PD_RNDINT_MAGIC)] ; mm1=[PD_RNDINT_MAGIC]
|
|
pcmpeqd mm3,mm3
|
|
psrld mm3,WORD_BIT ; mm3={0xFFFF 0x0000 0xFFFF 0x0000}
|
|
|
|
pfadd mm6,mm1 ; mm6=roundint(data0/8)=(00 ** 10 **)
|
|
pfadd mm7,mm1 ; mm7=roundint(data1/8)=(01 ** 11 **)
|
|
pfadd mm0,mm1 ; mm0=roundint(data6/8)=(06 ** 16 **)
|
|
pfadd mm5,mm1 ; mm5=roundint(data7/8)=(07 ** 17 **)
|
|
|
|
pand mm6,mm3 ; mm6=(00 -- 10 --)
|
|
pslld mm7,WORD_BIT ; mm7=(-- 01 -- 11)
|
|
pand mm0,mm3 ; mm0=(06 -- 16 --)
|
|
pslld mm5,WORD_BIT ; mm5=(-- 07 -- 17)
|
|
por mm6,mm7 ; mm6=(00 01 10 11)
|
|
por mm0,mm5 ; mm0=(06 07 16 17)
|
|
|
|
movq mm1, MMWORD [wk(0)] ; mm1=tmp2
|
|
movq mm3, MMWORD [wk(1)] ; mm3=tmp3
|
|
|
|
pfadd mm4,mm2 ; mm4=tmp4
|
|
movq mm7,mm1
|
|
movq mm5,mm3
|
|
pfadd mm1,mm2 ; mm1=data2=(02 12)
|
|
pfadd mm3,mm4 ; mm3=data4=(04 14)
|
|
pfsub mm7,mm2 ; mm7=data5=(05 15)
|
|
pfsub mm5,mm4 ; mm5=data3=(03 13)
|
|
|
|
movq mm2,[GOTOFF(ebx,PD_RNDINT_MAGIC)] ; mm2=[PD_RNDINT_MAGIC]
|
|
pcmpeqd mm4,mm4
|
|
psrld mm4,WORD_BIT ; mm4={0xFFFF 0x0000 0xFFFF 0x0000}
|
|
|
|
pfadd mm3,mm2 ; mm3=roundint(data4/8)=(04 ** 14 **)
|
|
pfadd mm7,mm2 ; mm7=roundint(data5/8)=(05 ** 15 **)
|
|
pfadd mm1,mm2 ; mm1=roundint(data2/8)=(02 ** 12 **)
|
|
pfadd mm5,mm2 ; mm5=roundint(data3/8)=(03 ** 13 **)
|
|
|
|
pand mm3,mm4 ; mm3=(04 -- 14 --)
|
|
pslld mm7,WORD_BIT ; mm7=(-- 05 -- 15)
|
|
pand mm1,mm4 ; mm1=(02 -- 12 --)
|
|
pslld mm5,WORD_BIT ; mm5=(-- 03 -- 13)
|
|
por mm3,mm7 ; mm3=(04 05 14 15)
|
|
por mm1,mm5 ; mm1=(02 03 12 13)
|
|
|
|
movq mm2,[GOTOFF(ebx,PB_CENTERJSAMP)] ; mm2=[PB_CENTERJSAMP]
|
|
|
|
packsswb mm6,mm3 ; mm6=(00 01 10 11 04 05 14 15)
|
|
packsswb mm1,mm0 ; mm1=(02 03 12 13 06 07 16 17)
|
|
paddb mm6,mm2
|
|
paddb mm1,mm2
|
|
|
|
movq mm4,mm6 ; transpose coefficients(phase 2)
|
|
punpcklwd mm6,mm1 ; mm6=(00 01 02 03 10 11 12 13)
|
|
punpckhwd mm4,mm1 ; mm4=(04 05 06 07 14 15 16 17)
|
|
|
|
movq mm7,mm6 ; transpose coefficients(phase 3)
|
|
punpckldq mm6,mm4 ; mm6=(00 01 02 03 04 05 06 07)
|
|
punpckhdq mm7,mm4 ; mm7=(10 11 12 13 14 15 16 17)
|
|
|
|
pushpic ebx ; save GOT address
|
|
|
|
mov edx, JSAMPROW [edi+0*SIZEOF_JSAMPROW]
|
|
mov ebx, JSAMPROW [edi+1*SIZEOF_JSAMPROW]
|
|
movq MMWORD [edx+eax*SIZEOF_JSAMPLE], mm6
|
|
movq MMWORD [ebx+eax*SIZEOF_JSAMPLE], mm7
|
|
|
|
poppic ebx ; restore GOT address
|
|
|
|
add esi, byte 2*SIZEOF_FAST_FLOAT ; wsptr
|
|
add edi, byte 2*SIZEOF_JSAMPROW
|
|
dec ecx ; ctr
|
|
jnz near .rowloop
|
|
|
|
femms ; empty MMX/3DNow! state
|
|
|
|
pop edi
|
|
pop esi
|
|
; pop edx ; need not be preserved
|
|
; pop ecx ; need not be preserved
|
|
pop ebx
|
|
mov esp,ebp ; esp <- aligned ebp
|
|
pop esp ; esp <- original ebp
|
|
pop ebp
|
|
ret
|
|
|
|
; For some reason, the OS X linker does not honor the request to align the
|
|
; segment unless we do this.
|
|
align 16
|