gecko-dev/media/libjpeg/simd/jiss2fst-64.asm
Bobby Holley 2d0b90011e Bug 584894 - Move libjpeg to media/libjpeg. r=khuey
--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
2011-10-17 21:25:53 -07:00

493 lines
16 KiB
NASM

;
; jiss2fst-64.asm - fast integer IDCT (64-bit SSE2)
;
; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
; Copyright 2009 D. R. Commander
;
; 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/projecpt/showfiles.php?group_id=6208
;
; This file contains a fast, not so accurate integer implementation of
; the inverse DCT (Discrete Cosine Transform). The following code is
; based directly on the IJG's original jidctfst.c; see the jidctfst.c
; for more details.
;
; [TAB8]
%include "jsimdext.inc"
%include "jdct.inc"
; --------------------------------------------------------------------------
%define CONST_BITS 8 ; 14 is also OK.
%define PASS1_BITS 2
%if IFAST_SCALE_BITS != PASS1_BITS
%error "'IFAST_SCALE_BITS' must be equal to 'PASS1_BITS'."
%endif
%if CONST_BITS == 8
F_1_082 equ 277 ; FIX(1.082392200)
F_1_414 equ 362 ; FIX(1.414213562)
F_1_847 equ 473 ; FIX(1.847759065)
F_2_613 equ 669 ; FIX(2.613125930)
F_1_613 equ (F_2_613 - 256) ; FIX(2.613125930) - FIX(1)
%else
; NASM cannot do compile-time arithmetic on floating-point constants.
%define DESCALE(x,n) (((x)+(1<<((n)-1)))>>(n))
F_1_082 equ DESCALE(1162209775,30-CONST_BITS) ; FIX(1.082392200)
F_1_414 equ DESCALE(1518500249,30-CONST_BITS) ; FIX(1.414213562)
F_1_847 equ DESCALE(1984016188,30-CONST_BITS) ; FIX(1.847759065)
F_2_613 equ DESCALE(2805822602,30-CONST_BITS) ; FIX(2.613125930)
F_1_613 equ (F_2_613 - (1 << CONST_BITS)) ; FIX(2.613125930) - FIX(1)
%endif
; --------------------------------------------------------------------------
SECTION SEG_CONST
; PRE_MULTIPLY_SCALE_BITS <= 2 (to avoid overflow)
; CONST_BITS + CONST_SHIFT + PRE_MULTIPLY_SCALE_BITS == 16 (for pmulhw)
%define PRE_MULTIPLY_SCALE_BITS 2
%define CONST_SHIFT (16 - PRE_MULTIPLY_SCALE_BITS - CONST_BITS)
alignz 16
global EXTN(jconst_idct_ifast_sse2)
EXTN(jconst_idct_ifast_sse2):
PW_F1414 times 8 dw F_1_414 << CONST_SHIFT
PW_F1847 times 8 dw F_1_847 << CONST_SHIFT
PW_MF1613 times 8 dw -F_1_613 << CONST_SHIFT
PW_F1082 times 8 dw F_1_082 << CONST_SHIFT
PB_CENTERJSAMP times 16 db CENTERJSAMPLE
alignz 16
; --------------------------------------------------------------------------
SECTION SEG_TEXT
BITS 64
;
; Perform dequantization and inverse DCT on one block of coefficients.
;
; GLOBAL(void)
; jsimd_idct_ifast_sse2 (void * dct_table, JCOEFPTR coef_block,
; JSAMPARRAY output_buf, JDIMENSION output_col)
;
; r10 = jpeg_component_info * compptr
; r11 = JCOEFPTR coef_block
; r12 = JSAMPARRAY output_buf
; r13 = JDIMENSION output_col
%define original_rbp rbp+0
%define wk(i) rbp-(WK_NUM-(i))*SIZEOF_XMMWORD ; xmmword wk[WK_NUM]
%define WK_NUM 2
align 16
global EXTN(jsimd_idct_ifast_sse2)
EXTN(jsimd_idct_ifast_sse2):
push rbp
mov rax,rsp ; rax = original rbp
sub rsp, byte 4
and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits
mov [rsp],rax
mov rbp,rsp ; rbp = aligned rbp
lea rsp, [wk(0)]
collect_args
; ---- Pass 1: process columns from input.
mov rdx, r10 ; quantptr
mov rsi, r11 ; inptr
%ifndef NO_ZERO_COLUMN_TEST_IFAST_SSE2
mov eax, DWORD [DWBLOCK(1,0,rsi,SIZEOF_JCOEF)]
or eax, DWORD [DWBLOCK(2,0,rsi,SIZEOF_JCOEF)]
jnz near .columnDCT
movdqa xmm0, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_JCOEF)]
movdqa xmm1, XMMWORD [XMMBLOCK(2,0,rsi,SIZEOF_JCOEF)]
por xmm0, XMMWORD [XMMBLOCK(3,0,rsi,SIZEOF_JCOEF)]
por xmm1, XMMWORD [XMMBLOCK(4,0,rsi,SIZEOF_JCOEF)]
por xmm0, XMMWORD [XMMBLOCK(5,0,rsi,SIZEOF_JCOEF)]
por xmm1, XMMWORD [XMMBLOCK(6,0,rsi,SIZEOF_JCOEF)]
por xmm0, XMMWORD [XMMBLOCK(7,0,rsi,SIZEOF_JCOEF)]
por xmm1,xmm0
packsswb xmm1,xmm1
packsswb xmm1,xmm1
movd eax,xmm1
test rax,rax
jnz short .columnDCT
; -- AC terms all zero
movdqa xmm0, XMMWORD [XMMBLOCK(0,0,rsi,SIZEOF_JCOEF)]
pmullw xmm0, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_ISLOW_MULT_TYPE)]
movdqa xmm7,xmm0 ; xmm0=in0=(00 01 02 03 04 05 06 07)
punpcklwd xmm0,xmm0 ; xmm0=(00 00 01 01 02 02 03 03)
punpckhwd xmm7,xmm7 ; xmm7=(04 04 05 05 06 06 07 07)
pshufd xmm6,xmm0,0x00 ; xmm6=col0=(00 00 00 00 00 00 00 00)
pshufd xmm2,xmm0,0x55 ; xmm2=col1=(01 01 01 01 01 01 01 01)
pshufd xmm5,xmm0,0xAA ; xmm5=col2=(02 02 02 02 02 02 02 02)
pshufd xmm0,xmm0,0xFF ; xmm0=col3=(03 03 03 03 03 03 03 03)
pshufd xmm1,xmm7,0x00 ; xmm1=col4=(04 04 04 04 04 04 04 04)
pshufd xmm4,xmm7,0x55 ; xmm4=col5=(05 05 05 05 05 05 05 05)
pshufd xmm3,xmm7,0xAA ; xmm3=col6=(06 06 06 06 06 06 06 06)
pshufd xmm7,xmm7,0xFF ; xmm7=col7=(07 07 07 07 07 07 07 07)
movdqa XMMWORD [wk(0)], xmm2 ; wk(0)=col1
movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=col3
jmp near .column_end
%endif
.columnDCT:
; -- Even part
movdqa xmm0, XMMWORD [XMMBLOCK(0,0,rsi,SIZEOF_JCOEF)]
movdqa xmm1, XMMWORD [XMMBLOCK(2,0,rsi,SIZEOF_JCOEF)]
pmullw xmm0, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_IFAST_MULT_TYPE)]
pmullw xmm1, XMMWORD [XMMBLOCK(2,0,rdx,SIZEOF_IFAST_MULT_TYPE)]
movdqa xmm2, XMMWORD [XMMBLOCK(4,0,rsi,SIZEOF_JCOEF)]
movdqa xmm3, XMMWORD [XMMBLOCK(6,0,rsi,SIZEOF_JCOEF)]
pmullw xmm2, XMMWORD [XMMBLOCK(4,0,rdx,SIZEOF_IFAST_MULT_TYPE)]
pmullw xmm3, XMMWORD [XMMBLOCK(6,0,rdx,SIZEOF_IFAST_MULT_TYPE)]
movdqa xmm4,xmm0
movdqa xmm5,xmm1
psubw xmm0,xmm2 ; xmm0=tmp11
psubw xmm1,xmm3
paddw xmm4,xmm2 ; xmm4=tmp10
paddw xmm5,xmm3 ; xmm5=tmp13
psllw xmm1,PRE_MULTIPLY_SCALE_BITS
pmulhw xmm1,[rel PW_F1414]
psubw xmm1,xmm5 ; xmm1=tmp12
movdqa xmm6,xmm4
movdqa xmm7,xmm0
psubw xmm4,xmm5 ; xmm4=tmp3
psubw xmm0,xmm1 ; xmm0=tmp2
paddw xmm6,xmm5 ; xmm6=tmp0
paddw xmm7,xmm1 ; xmm7=tmp1
movdqa XMMWORD [wk(1)], xmm4 ; wk(1)=tmp3
movdqa XMMWORD [wk(0)], xmm0 ; wk(0)=tmp2
; -- Odd part
movdqa xmm2, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_JCOEF)]
movdqa xmm3, XMMWORD [XMMBLOCK(3,0,rsi,SIZEOF_JCOEF)]
pmullw xmm2, XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_IFAST_MULT_TYPE)]
pmullw xmm3, XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_IFAST_MULT_TYPE)]
movdqa xmm5, XMMWORD [XMMBLOCK(5,0,rsi,SIZEOF_JCOEF)]
movdqa xmm1, XMMWORD [XMMBLOCK(7,0,rsi,SIZEOF_JCOEF)]
pmullw xmm5, XMMWORD [XMMBLOCK(5,0,rdx,SIZEOF_IFAST_MULT_TYPE)]
pmullw xmm1, XMMWORD [XMMBLOCK(7,0,rdx,SIZEOF_IFAST_MULT_TYPE)]
movdqa xmm4,xmm2
movdqa xmm0,xmm5
psubw xmm2,xmm1 ; xmm2=z12
psubw xmm5,xmm3 ; xmm5=z10
paddw xmm4,xmm1 ; xmm4=z11
paddw xmm0,xmm3 ; xmm0=z13
movdqa xmm1,xmm5 ; xmm1=z10(unscaled)
psllw xmm2,PRE_MULTIPLY_SCALE_BITS
psllw xmm5,PRE_MULTIPLY_SCALE_BITS
movdqa xmm3,xmm4
psubw xmm4,xmm0
paddw xmm3,xmm0 ; xmm3=tmp7
psllw xmm4,PRE_MULTIPLY_SCALE_BITS
pmulhw xmm4,[rel PW_F1414] ; xmm4=tmp11
; To avoid overflow...
;
; (Original)
; tmp12 = -2.613125930 * z10 + z5;
;
; (This implementation)
; tmp12 = (-1.613125930 - 1) * z10 + z5;
; = -1.613125930 * z10 - z10 + z5;
movdqa xmm0,xmm5
paddw xmm5,xmm2
pmulhw xmm5,[rel PW_F1847] ; xmm5=z5
pmulhw xmm0,[rel PW_MF1613]
pmulhw xmm2,[rel PW_F1082]
psubw xmm0,xmm1
psubw xmm2,xmm5 ; xmm2=tmp10
paddw xmm0,xmm5 ; xmm0=tmp12
; -- Final output stage
psubw xmm0,xmm3 ; xmm0=tmp6
movdqa xmm1,xmm6
movdqa xmm5,xmm7
paddw xmm6,xmm3 ; xmm6=data0=(00 01 02 03 04 05 06 07)
paddw xmm7,xmm0 ; xmm7=data1=(10 11 12 13 14 15 16 17)
psubw xmm1,xmm3 ; xmm1=data7=(70 71 72 73 74 75 76 77)
psubw xmm5,xmm0 ; xmm5=data6=(60 61 62 63 64 65 66 67)
psubw xmm4,xmm0 ; xmm4=tmp5
movdqa xmm3,xmm6 ; transpose coefficients(phase 1)
punpcklwd xmm6,xmm7 ; xmm6=(00 10 01 11 02 12 03 13)
punpckhwd xmm3,xmm7 ; xmm3=(04 14 05 15 06 16 07 17)
movdqa xmm0,xmm5 ; transpose coefficients(phase 1)
punpcklwd xmm5,xmm1 ; xmm5=(60 70 61 71 62 72 63 73)
punpckhwd xmm0,xmm1 ; xmm0=(64 74 65 75 66 76 67 77)
movdqa xmm7, XMMWORD [wk(0)] ; xmm7=tmp2
movdqa xmm1, XMMWORD [wk(1)] ; xmm1=tmp3
movdqa XMMWORD [wk(0)], xmm5 ; wk(0)=(60 70 61 71 62 72 63 73)
movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=(64 74 65 75 66 76 67 77)
paddw xmm2,xmm4 ; xmm2=tmp4
movdqa xmm5,xmm7
movdqa xmm0,xmm1
paddw xmm7,xmm4 ; xmm7=data2=(20 21 22 23 24 25 26 27)
paddw xmm1,xmm2 ; xmm1=data4=(40 41 42 43 44 45 46 47)
psubw xmm5,xmm4 ; xmm5=data5=(50 51 52 53 54 55 56 57)
psubw xmm0,xmm2 ; xmm0=data3=(30 31 32 33 34 35 36 37)
movdqa xmm4,xmm7 ; transpose coefficients(phase 1)
punpcklwd xmm7,xmm0 ; xmm7=(20 30 21 31 22 32 23 33)
punpckhwd xmm4,xmm0 ; xmm4=(24 34 25 35 26 36 27 37)
movdqa xmm2,xmm1 ; transpose coefficients(phase 1)
punpcklwd xmm1,xmm5 ; xmm1=(40 50 41 51 42 52 43 53)
punpckhwd xmm2,xmm5 ; xmm2=(44 54 45 55 46 56 47 57)
movdqa xmm0,xmm3 ; transpose coefficients(phase 2)
punpckldq xmm3,xmm4 ; xmm3=(04 14 24 34 05 15 25 35)
punpckhdq xmm0,xmm4 ; xmm0=(06 16 26 36 07 17 27 37)
movdqa xmm5,xmm6 ; transpose coefficients(phase 2)
punpckldq xmm6,xmm7 ; xmm6=(00 10 20 30 01 11 21 31)
punpckhdq xmm5,xmm7 ; xmm5=(02 12 22 32 03 13 23 33)
movdqa xmm4, XMMWORD [wk(0)] ; xmm4=(60 70 61 71 62 72 63 73)
movdqa xmm7, XMMWORD [wk(1)] ; xmm7=(64 74 65 75 66 76 67 77)
movdqa XMMWORD [wk(0)], xmm3 ; wk(0)=(04 14 24 34 05 15 25 35)
movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=(06 16 26 36 07 17 27 37)
movdqa xmm3,xmm1 ; transpose coefficients(phase 2)
punpckldq xmm1,xmm4 ; xmm1=(40 50 60 70 41 51 61 71)
punpckhdq xmm3,xmm4 ; xmm3=(42 52 62 72 43 53 63 73)
movdqa xmm0,xmm2 ; transpose coefficients(phase 2)
punpckldq xmm2,xmm7 ; xmm2=(44 54 64 74 45 55 65 75)
punpckhdq xmm0,xmm7 ; xmm0=(46 56 66 76 47 57 67 77)
movdqa xmm4,xmm6 ; transpose coefficients(phase 3)
punpcklqdq xmm6,xmm1 ; xmm6=col0=(00 10 20 30 40 50 60 70)
punpckhqdq xmm4,xmm1 ; xmm4=col1=(01 11 21 31 41 51 61 71)
movdqa xmm7,xmm5 ; transpose coefficients(phase 3)
punpcklqdq xmm5,xmm3 ; xmm5=col2=(02 12 22 32 42 52 62 72)
punpckhqdq xmm7,xmm3 ; xmm7=col3=(03 13 23 33 43 53 63 73)
movdqa xmm1, XMMWORD [wk(0)] ; xmm1=(04 14 24 34 05 15 25 35)
movdqa xmm3, XMMWORD [wk(1)] ; xmm3=(06 16 26 36 07 17 27 37)
movdqa XMMWORD [wk(0)], xmm4 ; wk(0)=col1
movdqa XMMWORD [wk(1)], xmm7 ; wk(1)=col3
movdqa xmm4,xmm1 ; transpose coefficients(phase 3)
punpcklqdq xmm1,xmm2 ; xmm1=col4=(04 14 24 34 44 54 64 74)
punpckhqdq xmm4,xmm2 ; xmm4=col5=(05 15 25 35 45 55 65 75)
movdqa xmm7,xmm3 ; transpose coefficients(phase 3)
punpcklqdq xmm3,xmm0 ; xmm3=col6=(06 16 26 36 46 56 66 76)
punpckhqdq xmm7,xmm0 ; xmm7=col7=(07 17 27 37 47 57 67 77)
.column_end:
; -- Prefetch the next coefficient block
prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 0*32]
prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 1*32]
prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 2*32]
prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 3*32]
; ---- Pass 2: process rows from work array, store into output array.
mov rax, [original_rbp]
mov rdi, r12 ; (JSAMPROW *)
mov rax, r13
; -- Even part
; xmm6=col0, xmm5=col2, xmm1=col4, xmm3=col6
movdqa xmm2,xmm6
movdqa xmm0,xmm5
psubw xmm6,xmm1 ; xmm6=tmp11
psubw xmm5,xmm3
paddw xmm2,xmm1 ; xmm2=tmp10
paddw xmm0,xmm3 ; xmm0=tmp13
psllw xmm5,PRE_MULTIPLY_SCALE_BITS
pmulhw xmm5,[rel PW_F1414]
psubw xmm5,xmm0 ; xmm5=tmp12
movdqa xmm1,xmm2
movdqa xmm3,xmm6
psubw xmm2,xmm0 ; xmm2=tmp3
psubw xmm6,xmm5 ; xmm6=tmp2
paddw xmm1,xmm0 ; xmm1=tmp0
paddw xmm3,xmm5 ; xmm3=tmp1
movdqa xmm0, XMMWORD [wk(0)] ; xmm0=col1
movdqa xmm5, XMMWORD [wk(1)] ; xmm5=col3
movdqa XMMWORD [wk(0)], xmm2 ; wk(0)=tmp3
movdqa XMMWORD [wk(1)], xmm6 ; wk(1)=tmp2
; -- Odd part
; xmm0=col1, xmm5=col3, xmm4=col5, xmm7=col7
movdqa xmm2,xmm0
movdqa xmm6,xmm4
psubw xmm0,xmm7 ; xmm0=z12
psubw xmm4,xmm5 ; xmm4=z10
paddw xmm2,xmm7 ; xmm2=z11
paddw xmm6,xmm5 ; xmm6=z13
movdqa xmm7,xmm4 ; xmm7=z10(unscaled)
psllw xmm0,PRE_MULTIPLY_SCALE_BITS
psllw xmm4,PRE_MULTIPLY_SCALE_BITS
movdqa xmm5,xmm2
psubw xmm2,xmm6
paddw xmm5,xmm6 ; xmm5=tmp7
psllw xmm2,PRE_MULTIPLY_SCALE_BITS
pmulhw xmm2,[rel PW_F1414] ; xmm2=tmp11
; To avoid overflow...
;
; (Original)
; tmp12 = -2.613125930 * z10 + z5;
;
; (This implementation)
; tmp12 = (-1.613125930 - 1) * z10 + z5;
; = -1.613125930 * z10 - z10 + z5;
movdqa xmm6,xmm4
paddw xmm4,xmm0
pmulhw xmm4,[rel PW_F1847] ; xmm4=z5
pmulhw xmm6,[rel PW_MF1613]
pmulhw xmm0,[rel PW_F1082]
psubw xmm6,xmm7
psubw xmm0,xmm4 ; xmm0=tmp10
paddw xmm6,xmm4 ; xmm6=tmp12
; -- Final output stage
psubw xmm6,xmm5 ; xmm6=tmp6
movdqa xmm7,xmm1
movdqa xmm4,xmm3
paddw xmm1,xmm5 ; xmm1=data0=(00 10 20 30 40 50 60 70)
paddw xmm3,xmm6 ; xmm3=data1=(01 11 21 31 41 51 61 71)
psraw xmm1,(PASS1_BITS+3) ; descale
psraw xmm3,(PASS1_BITS+3) ; descale
psubw xmm7,xmm5 ; xmm7=data7=(07 17 27 37 47 57 67 77)
psubw xmm4,xmm6 ; xmm4=data6=(06 16 26 36 46 56 66 76)
psraw xmm7,(PASS1_BITS+3) ; descale
psraw xmm4,(PASS1_BITS+3) ; descale
psubw xmm2,xmm6 ; xmm2=tmp5
packsswb xmm1,xmm4 ; xmm1=(00 10 20 30 40 50 60 70 06 16 26 36 46 56 66 76)
packsswb xmm3,xmm7 ; xmm3=(01 11 21 31 41 51 61 71 07 17 27 37 47 57 67 77)
movdqa xmm5, XMMWORD [wk(1)] ; xmm5=tmp2
movdqa xmm6, XMMWORD [wk(0)] ; xmm6=tmp3
paddw xmm0,xmm2 ; xmm0=tmp4
movdqa xmm4,xmm5
movdqa xmm7,xmm6
paddw xmm5,xmm2 ; xmm5=data2=(02 12 22 32 42 52 62 72)
paddw xmm6,xmm0 ; xmm6=data4=(04 14 24 34 44 54 64 74)
psraw xmm5,(PASS1_BITS+3) ; descale
psraw xmm6,(PASS1_BITS+3) ; descale
psubw xmm4,xmm2 ; xmm4=data5=(05 15 25 35 45 55 65 75)
psubw xmm7,xmm0 ; xmm7=data3=(03 13 23 33 43 53 63 73)
psraw xmm4,(PASS1_BITS+3) ; descale
psraw xmm7,(PASS1_BITS+3) ; descale
movdqa xmm2,[rel PB_CENTERJSAMP] ; xmm2=[rel PB_CENTERJSAMP]
packsswb xmm5,xmm6 ; xmm5=(02 12 22 32 42 52 62 72 04 14 24 34 44 54 64 74)
packsswb xmm7,xmm4 ; xmm7=(03 13 23 33 43 53 63 73 05 15 25 35 45 55 65 75)
paddb xmm1,xmm2
paddb xmm3,xmm2
paddb xmm5,xmm2
paddb xmm7,xmm2
movdqa xmm0,xmm1 ; transpose coefficients(phase 1)
punpcklbw xmm1,xmm3 ; xmm1=(00 01 10 11 20 21 30 31 40 41 50 51 60 61 70 71)
punpckhbw xmm0,xmm3 ; xmm0=(06 07 16 17 26 27 36 37 46 47 56 57 66 67 76 77)
movdqa xmm6,xmm5 ; transpose coefficients(phase 1)
punpcklbw xmm5,xmm7 ; xmm5=(02 03 12 13 22 23 32 33 42 43 52 53 62 63 72 73)
punpckhbw xmm6,xmm7 ; xmm6=(04 05 14 15 24 25 34 35 44 45 54 55 64 65 74 75)
movdqa xmm4,xmm1 ; transpose coefficients(phase 2)
punpcklwd xmm1,xmm5 ; xmm1=(00 01 02 03 10 11 12 13 20 21 22 23 30 31 32 33)
punpckhwd xmm4,xmm5 ; xmm4=(40 41 42 43 50 51 52 53 60 61 62 63 70 71 72 73)
movdqa xmm2,xmm6 ; transpose coefficients(phase 2)
punpcklwd xmm6,xmm0 ; xmm6=(04 05 06 07 14 15 16 17 24 25 26 27 34 35 36 37)
punpckhwd xmm2,xmm0 ; xmm2=(44 45 46 47 54 55 56 57 64 65 66 67 74 75 76 77)
movdqa xmm3,xmm1 ; transpose coefficients(phase 3)
punpckldq xmm1,xmm6 ; xmm1=(00 01 02 03 04 05 06 07 10 11 12 13 14 15 16 17)
punpckhdq xmm3,xmm6 ; xmm3=(20 21 22 23 24 25 26 27 30 31 32 33 34 35 36 37)
movdqa xmm7,xmm4 ; transpose coefficients(phase 3)
punpckldq xmm4,xmm2 ; xmm4=(40 41 42 43 44 45 46 47 50 51 52 53 54 55 56 57)
punpckhdq xmm7,xmm2 ; xmm7=(60 61 62 63 64 65 66 67 70 71 72 73 74 75 76 77)
pshufd xmm5,xmm1,0x4E ; xmm5=(10 11 12 13 14 15 16 17 00 01 02 03 04 05 06 07)
pshufd xmm0,xmm3,0x4E ; xmm0=(30 31 32 33 34 35 36 37 20 21 22 23 24 25 26 27)
pshufd xmm6,xmm4,0x4E ; xmm6=(50 51 52 53 54 55 56 57 40 41 42 43 44 45 46 47)
pshufd xmm2,xmm7,0x4E ; xmm2=(70 71 72 73 74 75 76 77 60 61 62 63 64 65 66 67)
mov rdx, JSAMPROW [rdi+0*SIZEOF_JSAMPROW]
mov rsi, JSAMPROW [rdi+2*SIZEOF_JSAMPROW]
movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm1
movq XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE], xmm3
mov rdx, JSAMPROW [rdi+4*SIZEOF_JSAMPROW]
mov rsi, JSAMPROW [rdi+6*SIZEOF_JSAMPROW]
movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm4
movq XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE], xmm7
mov rdx, JSAMPROW [rdi+1*SIZEOF_JSAMPROW]
mov rsi, JSAMPROW [rdi+3*SIZEOF_JSAMPROW]
movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm5
movq XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE], xmm0
mov rdx, JSAMPROW [rdi+5*SIZEOF_JSAMPROW]
mov rsi, JSAMPROW [rdi+7*SIZEOF_JSAMPROW]
movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm6
movq XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE], xmm2
uncollect_args
mov rsp,rbp ; rsp <- aligned rbp
pop rsp ; rsp <- original rbp
pop rbp
ret
ret
; For some reason, the OS X linker does not honor the request to align the
; segment unless we do this.
align 16