|
|
;-------------------------------------------------------------------------; INTEL Corporation Proprietary Information;; This listing is supplied under the terms of a license; agreement with INTEL Corporation and may not be copied; nor disclosed except in accordance with the terms of; that agreement.;; Copyright (c) 1996 Intel Corporation.; All Rights Reserved.;;-------------------------------------------------------------------------
; $Header: S:\h26x\src\dec\cxm12321.asv 1.4 24 May 1996 10:30:20 AGUPTA2 $; $Log: S:\h26x\src\dec\cxm12321.asv $;// ;// Rev 1.4 24 May 1996 10:30:20 AGUPTA2;// Cosmetic changes to adhere to a common coding convention in all;// MMX color convertor files.;// ;// ;// Rev 1.3 11 Apr 1996 09:51:14 RMCKENZX;// Changed return to pop the stack.;// ;// Rev 1.2 09 Apr 1996 17:15:30 RMCKENZX;// Optimized.;// ;// Rev 1.1 09 Apr 1996 09:50:32 RMCKENZX;// Added aspect correction, fixed wrap-around, changed calling sequence.;// ;// Rev 1.0 06 Apr 1996 17:06:06 RMCKENZX;// Initial revision.;;-------------------------------------------------------------------------;; +---------- Color convertor.; |+--------- For both H261 and H263.; ||+-------- MMx Version.; |||++------ Convert from YUV12.; |||||++---- Convert to RGB32.; |||||||+--- Zoom by one, i.e. non-zoom.; ||||||||; cxm12321 -- This function performs YUV12-to-RGB32 color conversion for H26x.; It handles the format in which the low order byte is B, the; second byte is G, and the third byte is R, and the high order; byte is 0.;; The YUV12 input is planar, 8 bits per pel. The Y plane may have; a pitch of up to 768. It may have a width less than or equal; to the pitch. It must be DWORD aligned, and preferably QWORD; aligned. Pitch and Width must be a multiple of 8. The U; and V planes may have a different pitch than the Y plane, subject; to the same limitations.;OPTION CASEMAP:NONEOPTION PROLOGUE:NoneOPTION EPILOGUE:ReturnAndRelieveEpilogueMacro
.586.xlistinclude iammx.incinclude memmodel.inc.list
MMXCODE1 SEGMENT PARA USE32 PUBLIC 'CODE'MMXCODE1 ENDS
MMXDATA1 SEGMENT PARA USE32 PUBLIC 'DATA'MMXDATA1 ENDS
MMXDATA1 SEGMENTALIGN 8;; constants for direct RGB calculation: 4x10.6 values; chroma constants are multiplied by 64 (6 fraction bits) and 255/224 (scale).; luma constant is 64 * (255/219) = 74.55055, so it is dithered.; PUBLIC C VtR, VtG, UtG, UtB, Ymul0, Ymul1, Ysub, UVsubVtR DWORD 00660066h, 00660066h ; 1.402 -> 102.14571VtG DWORD 0ffccffcch, 0ffccffcch ; -.71414 -> -52.03020UtG DWORD 0ffe7ffe7h, 0ffe7ffe7h ; -.34414 -> -25.07306UtB DWORD 00810081h, 00810081h ; 1.772 -> 129.10286Ymul0 DWORD 004a004bh, 004a004bh ; 74.55055Ymul1 DWORD 004b004ah, 004b004ah ; 74.55055Ysub DWORD 00100010h, 00100010h ; bias for yUVsub DWORD 00800080h, 00800080h ; bias for uvMMXDATA1 ENDS
MMXCODE1 SEGMENT
MMX_YUV12ToRGB32 PROC DIST LANG PUBLIC, AYPlane: DWORD, AVPlane: DWORD, AUPlane: DWORD, AFrameWidth: DWORD, AFrameHeight: DWORD, AYPitch: DWORD, AVPitch: DWORD, AAspectAdjustmentCnt: DWORD, AColorConvertedFrame: DWORD, ADCIOffset: DWORD, ACCOffsetToLine0: DWORD, ACCOPitch: DWORD, ACCType: DWORD
LocalSize = 20h ; for 7 local variablesRegisterSize = 10h ; for the 4 push/popsStashSize = 1200h ; 768 (max width) * 6 LocalFrameSize = LocalSize + StashSizeFrameAdjustOne = 800hFrameAdjustTwo = LocalFrameSize - FrameAdjustOneargument_base EQU ebp + RegisterSize local_base EQU espstash_base EQU esp + LocalSize
; Arguments:YPlane EQU argument_base + 04hVPlane EQU argument_base + 08hUPlane EQU argument_base + 0chFrameWidth EQU argument_base + 10hFrameHeight EQU argument_base + 14hLumaPitch EQU argument_base + 18hChromaPitch EQU argument_base + 1chAspectAdjustmentCount EQU argument_base + 20hColorConvertedFrame EQU argument_base + 24hDCIOffset EQU argument_base + 28hCCOffsetToLine0 EQU argument_base + 2chCCOPitch EQU argument_base + 30h
; Locals (on local stack frame)localAspectCount EQU local_base + 00hlocalAspectAdjustment EQU local_base + 04hlocalWidth EQU local_base + 08hlocalYPitch EQU local_base + 0chlocalUVPitch EQU local_base + 10hlocalOutPitch EQU local_base + 14hlocalStashEsp EQU local_base + 18h
; symbolic register names for shuffle segmentsmmx_zero EQU mm0 ; mmx_zero push esi push edi push ebp push ebx mov ebp, esp and esp, -32 ; align to cache-line size ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ; Initialize: 'x'=live, '-'=dead, 'o'=live(ALU op) ; esi eax ebx ecx edx edi ebp pxor mmx_zero, mmx_zero ; sub esp, FrameAdjustOne ; mov edi, [CCOPitch] ; x mov ecx, [ChromaPitch] ; x | mov ebx, [esp] ; | - | | sub esp, FrameAdjustTwo ; | | | mov eax, [LumaPitch] ; x | | nop mov [localStashEsp], ebp ; | | | mov [localOutPitch], edi ; | | - mov [localUVPitch], ecx ; | - mov [localYPitch], eax ; - mov eax, [AspectAdjustmentCount] ; x mov edi, [ColorConvertedFrame] ; | x mov [localAspectCount], eax ; | mov esi, [FrameWidth] ; x | mov ebx, [DCIOffset] ; | | x | mov edx, [CCOffsetToLine0] ; | | | x | add edi, ebx ; | | - | o add edi, edx ; | | - o mov [localAspectAdjustment], eax ; | - | mov eax, [YPlane] ; | x | lea edi, [edi+4*esi] ; | | o RGB plane base mov ecx, [UPlane] ; | | x | mov edx, [VPlane] ; | | | x | mov ebx, [FrameHeight] ; | | x | | | Outer loop control sar esi, 1 ; o | | | | | xor ebp, ebp ; | | | | | | + add ecx, esi ; + | | o | | | U plane base add edx, esi ; + | | | o | | V plane base lea eax, [eax+2*esi] ; + o | | | | | Y plane base sub ebp, esi ; - | | | | | o Inner loop control mov [localWidth], ebp ; + | | | | | | xor esi, esi ; x | | | | | | Stash pointer ; v v v v v v v ; esi eax ebx ecx esi edi ebp
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ; ; ALGORITHM: ; The following outer loop (do_two_lines) does two lines of Y (sharing ; one line of UV) per iteration. It contains two inner loops. ; ; The first inner loop (do_next_even_line) does 8 pels of the even line ; per iteration and stashes the chroma contribution on the stack. ; ; The second inner loop (do_next_odd_line) reads the stashed chroma and ; does 8 pels of the odd line per iteration. ; ; Aspect Adjustment is accomplished by skipping the second inner loop ; if needed. ; ; CORE REGISTERS: ; (all registers are pre-loaded): ; eax Y plane base address. ; ebx outer loop control. Starts at Height, runs down to 0. ; ecx U plane base address. ; edx V plane base address. ; esi stash pointer. ; edi output RGB plane base address. ; ebp inner loop control. Starts at -Width/2, runs up to 0. ; ; All plane base addresses are previously biased by Width (y plane), ; Width/2 (uv plane), or 4*Width (rgb plane) and are used in conjunction ; with the inner loop control, ebp. The base addresses are updated after ; the first inner loop (Y/U/V/RGB), and after the second inner loop (Y/RGB). ; ; The stash pointer is referenced in chromaC (with esp). It is updated ; inside each inner loop and reset to 0 after each inner loop. ; ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ; start outer loop ; start first inner loop ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;do_two_lines:do_next_even_line: movd mm3, [ecx+ebp] ; ...3.... xxxxxxxx U76 U54 U32 U10 ; movd mm4, [edx+ebp] ; ...34... xxxxxxxx V76 V54 V32 V10 punpcklbw mm3, mmx_zero ; ...34... .U76 .U54 .U32 .U10 psubw mm3, UVsub ; ...34... unbias U (sub 128) punpcklbw mm4, mmx_zero ; ...34... .V76 .V54 .V32 .V10 psubw mm4, UVsub ; ...34... unbias V (sub 128) movq mm1, mm3 ; .1.34... .U76 .U54 .U32 .U10 pmullw mm3, UtG ; .1.34... .G76 .G54 .G32 .G10 (from U) movq mm2, mm4 ; .1234... .V76 .V54 .V32 .V10 pmullw mm4, VtG ; .1234... .G76 .G54 .G32 .G10 (from V) ; movq mm6, [eax+2*ebp] ; .123..6. Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 ; movq mm7, mm6 ; .123..67 Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 punpcklbw mm6, mmx_zero ; .123..67 ..Y3 ..Y2 ..Y1 ..Y0 psubw mm6, Ysub ; .123..67 unbias Y (sub 16) & clip at 0 punpckhbw mm7, mmx_zero ; .123..67 ..Y7 ..Y6 ..Y5 ..Y4 psubw mm7, Ysub ; .123..67 unbias Y (sub 16) & clip at 0 paddsw mm3, mm4 ; .123..67 .G76 .G54 .G32 .G10 (from chroma) pmullw mm6, Ymul0 ; .123..67 RGB3 RGB2 RGB1 RGB0 (from luma) movq mm5, mm3 ; .123.567 .G76 .G54 .G32 .G10 (from chroma) pmullw mm7, Ymul0 ; .123.567 RGB7 RGB6 RGB5 RGB4 (from luma) punpcklwd mm3, mm3 ; .123.567 ..G3 ..G2 ..G1 ..G0 (from chroma) pmullw mm1, UtB ; .123.567 .B76 .B54 .B32 .B10 (from U) punpckhwd mm5, mm5 ; .123.567 ..G7 ..G6 ..G5 ..G4 (from chroma) movq [stash_base+esi+00h], mm3 ; .123.567 stash low green from chroma paddsw mm3, mm6 ; .123.567 ..G3 ..G2 ..G1 ..G0 (scaled total) movq [stash_base+esi+08h], mm5 ; .123.567 stash high green from chroma paddsw mm5, mm7 ; .123.567 ..G7 ..G6 ..G5 ..G4 (scaled total) movq mm4, mm1 ; .1234567 .B76 .B54 .B32 .B10 (from U) psraw mm3, 6 ; .1234567 ..G3 ..G2 ..G1 ..G0 (total) pmullw mm2, VtR ; .1234567 .R76 .R54 .R32 .R10 (from V) psraw mm5, 6 ; .1234567 ..G7 ..G6 ..G5 ..G4 (total) packuswb mm3, mm5 ; .1234.67 G7 G6 G5 G4 G3 G2 G1 G0 movq mm5, mm2 ; .1234567 .R76 .R54 .R32 .R10 (from V) ; -------- green done --------
punpcklwd mm1, mm1 ; .1234567 ..B3 ..B2 ..B1 ..B0 (from U) ; punpckhwd mm4, mm4 ; .1234567 ..B7 ..B6 ..B5 ..B4 (from U) ; movq [stash_base+esi+10h], mm1 ; .1234567 stash low blue from chroma punpcklwd mm2, mm2 ; .1234567 ..R3 ..R2 ..R1 ..R0 (from V) movq [stash_base+esi+18h], mm4 ; .1234567 stash high blue from chroma punpckhwd mm5, mm5 ; .1234567 ..R7 ..R6 ..R5 ..R4 (from V) paddsw mm1, mm6 ; .1234567 ..B3 ..B2 ..B1 ..B0 (scaled total) paddsw mm4, mm7 ; .1234567 ..B7 ..B6 ..B5 ..B4 (scaled total) movq [stash_base+esi+20h], mm2 ; .1234567 stash low red from chroma psraw mm1, 6 ; .1234567 ..B3 ..B2 ..B1 ..B0 (total) movq [stash_base+esi+28h], mm5 ; .1234567 stash high red from chroma psraw mm4, 6 ; .1234567 ..B7 ..B6 ..B5 ..B4 (total) paddsw mm2, mm6 ; .12345.7 ..R3 ..R2 ..R1 ..R0 (total scaled) packuswb mm1, mm4 ; .123.5.7 B7 B6 B5 B4 B3 B2 B1 B0 ; -------- blue done -------- paddsw mm5, mm7 ; .123.5.. ..R7 ..R6 ..R5 ..R4 (total scaled) psraw mm2, 6 ; .123.5.. ..R3 ..R2 ..R1 ..R0 (total) psraw mm5, 6 ; .123.5.. ..R7 ..R6 ..R5 ..R4 (total) ; packuswb mm2, mm5 ; .123.... R7 R6 R5 R4 R3 R2 R1 R0 ; ; -------- red done -------- ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ; shuffle up the results: ; red = mm2 ; green = mm4 ; blue = mm1 ; into red-green-blue order and store ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; movq mm5, mm1 ; .123.5.. blue copy punpcklbw mm1, mm3 ; .123.5.. G3 B3 G2 B2 G1 B1 G0 B0 movq mm4, mm2 ; .12345.. red copy punpcklbw mm2, mmx_zero ; .12345.. -- R3 -- R2 -- R1 -- R0 movq mm6, mm1 ; .123456. G3 B3 G2 B2 G1 B1 G0 B0 punpcklwd mm1, mm2 ; .123456. R1 G1 B1 -- R0 G0 B0 punpckhwd mm6, mm2 ; .1.3456. -- R3 G3 B3 -- R2 G2 B2 ; movq [edi+8*ebp+00], mm1 ; ...3456. write first two pels punpckhbw mm5, mm3 ; ....456. G7 B7 G6 B6 G5 B5 G4 B4 movq [edi+8*ebp+08], mm6 ; ....45.. write second two pels punpckhbw mm4, mmx_zero ; ....45.. -- R7 -- R6 -- R5 -- R4 movq mm7, mm5 ; ....45.7 G7 B7 G6 B6 G5 B5 G4 B4 punpcklwd mm5, mm4 ; ....45.7 -- R5 G5 B5 -- R4 G4 B4 punpckhwd mm7, mm4 ; .....5.7 -- R7 G7 B7 -- R6 G6 B6 add esi, 30h ; increment stash pointer movq [edi+8*ebp+16], mm5 ; .......7 write third two pels ; movq [edi+8*ebp+24], mm7 ; ........ write fourth two pels ; add ebp, 4 ; increment loop control jl do_next_even_line ; back up if not done ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ; end do next even line loop ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
; add pitches to base plane addresses and check aspect mov ebp, [localOutPitch] mov esi, [localUVPitch] add edi, ebp ; update RGB plane base address add edx, esi ; update V plane base address add ecx, esi ; update U plane base address mov esi, [localYPitch] add eax, esi ; update Y plane base address mov ebp, [localAspectCount] sub ebp, 2 jle skip_odd_line mov [localAspectCount], ebp ; store aspect count mov ebp, [localWidth] ; load inner loop control xor esi, esi ; reset stash pointer ; movq mm7, Ymul1 ; pre-load Y scaling factor to mm7 ; ; ; start odd line loop ;do_next_odd_line: movq mm3, [eax+2*ebp] ; ...3.... Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 ; movq mm4, mm3 ; ...34... Y7 Y6 Y5 Y4 Y3 Y2 Y1 Y0 punpcklbw mm3, mmx_zero ; ...34... ..Y3 ..Y2 ..Y1 ..Y0 psubw mm3, Ysub ; ...34... unbias Y punpckhbw mm4, mmx_zero ; ...34... ..Y7 ..Y6 ..Y5 ..Y4 psubw mm4, Ysub ; ...34... unbias Y pmullw mm3, mm7 ; ...34... RGB3 RGB2 RGB1 RGB0 (from luma) movq mm2, [stash_base+esi+20h] ; ..234... ..R3 ..R2 ..R1 ..R0 (from V) pmullw mm4, mm7 ; ...34... RGB7 RGB6 RGB5 RGB4 (from luma) movq mm5, [stash_base+esi+28h] ; ..2345.. ..R7 ..R6 ..R5 ..R4 (from V) paddsw mm2, mm3 ; ..2345.. ..R3 ..R2 ..R1 ..R0 (scaled total) movq mm1, [stash_base+esi+10h] ; .12345.. ..B3 ..B2 ..B1 ..B0 (from U) paddsw mm5, mm4 ; .12345.. ..R7 ..R6 ..R5 ..R4 (scaled total) movq mm6, [stash_base+esi+18h] ; .123456. ..B7 ..B6 ..B5 ..B4 (from U) psraw mm2, 6 ; .123456. ..R3 ..R2 ..R1 ..R0 (total) paddsw mm1, mm3 ; .123456. ..B3 ..B2 ..B1 ..B0 (scaled total) psraw mm5, 6 ; .123456. ..R7 ..R6 ..R5 ..R4 (total) paddsw mm6, mm4 ; .123456. ..B7 ..B6 ..B5 ..B4 (scaled total) packuswb mm2, mm5 ; .1234.6. R7 R6 R5 R4 R3 R2 R1 R0 ; -------- red done -------- paddsw mm3, [stash_base+esi+00h] ; .1234.6. ..G3 ..G2 ..G1 ..G0 (scaled total) psraw mm1, 6 ; .1234.6. ..B3 ..B2 ..B1 ..B0 (total) paddsw mm4, [stash_base+esi+08h] ; .1234.6. ..G7 ..G6 ..G5 ..G4 (scaled total) psraw mm6, 6 ; .1234.6. ..B7 ..B6 ..B5 ..B4 (total) packuswb mm1, mm6 ; .1234... B7 B6 B5 B4 B3 B2 B1 B0 ; ; -------- blue done -------- psraw mm3, 6 ; .1234... ..G3 ..G2 ..G1 ..G0 (total) ; psraw mm4, 6 ; .1234... ..G7 ..G6 ..G5 ..G4 (total) ; packuswb mm3, mm4 ; .123.... G7 G6 G5 G4 G3 G2 G1 G0 ; ; -------- green done -------- ; ; shuffle up the results: ; red = mm2 ; green = mm3 ; blue = mm1 ; into red-green-blue order and store ; movq mm5, mm1 ; .123.5.. blue copy punpcklbw mm1, mm3 ; .123.5.. G3 B3 G2 B2 G1 B1 G0 B0 movq mm4, mm2 ; .12345.. red copy punpcklbw mm2, mmx_zero ; .12345.. -- R3 -- R2 -- R1 -- R0 movq mm6, mm1 ; .123456. G3 B3 G2 B2 G1 B1 G0 B0 punpcklwd mm1, mm2 ; .123456. R1 G1 B1 -- R0 G0 B0 punpckhwd mm6, mm2 ; .1.3456. -- R3 G3 B3 -- R2 G2 B2 ; movq [edi+8*ebp+00], mm1 ; ...3456. write first two pels punpckhbw mm5, mm3 ; ....456. G7 B7 G6 B6 G5 B5 G4 B4 movq [edi+8*ebp+08], mm6 ; ....45.. write second two pels punpckhbw mm4, mmx_zero ; ....45.. -- R7 -- R6 -- R5 -- R4 movq mm1, mm5 ; .1..45.. G7 B7 G6 B6 G5 B5 G4 B4 punpcklwd mm5, mm4 ; .1..45.. -- R5 G5 B5 -- R4 G4 B4 punpckhwd mm1, mm4 ; .1...5.. -- R7 G7 B7 -- R6 G6 B6 add esi, 30h ; increment stash pointer movq [edi+8*ebp+16], mm5 ; .1...... write third two pels ; movq [edi+8*ebp+24], mm1 ; ........ write fourth two pels ; add ebp, 4 ; increment loop control jl do_next_odd_line ; back up if not done ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ; end do next odd line loop ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; mov ebp, [localYPitch] mov esi, [localOutPitch] add eax, ebp ; update Y plane base address add edi, esi ; update RGB plane base address mov ebp, [localWidth] ; load inner loop control xor esi, esi ; reset stash pointer sub ebx, 2 ; decrement outer loop control jg do_two_lines ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;; ; end do two lines loop ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
finish: mov esp, [localStashEsp] ; pop ebx pop ebp pop edi pop esi ret 52
skip_odd_line: add eax, esi ; update Y plane base address mov esi, [localAspectAdjustment] add ebp, esi ; reset aspect adjustment count xor esi, esi ; reset stash pointer mov [localAspectCount], ebp ; store aspect count mov ebp, [localWidth] ; load inner loop control sub ebx, 2 ; decrement outer loop control jg do_two_lines ; back up if not done; else go home mov esp, [localStashEsp] ; pop ebx pop ebp pop edi pop esi retMMX_YUV12ToRGB32 ENDP
MMXCODE1 ENDS
END
|
