vdec_idctmmx.s

vlc stand 0.1.99 ist sehr einfach

        psubsw %mm4, %mm0		/* tm10; free mm4 */
/* moved from next block */
        pmulhw 8*5(%esi), %mm1		/* V5 */
        movq %mm6, 8*4(%esi)		/* tm4; free mm6 */
        movq %mm0, 8*10(%esi)		/* tm10; free mm0 */
/* column 1: even part
 * use V5, V13, V1, V9 to produce V56..V59
 */
/* moved to prev block:
 *	movq 8*5(%ecx), %mm1
 *	pmulhw 8*5(%esi), %mm1		 V5
 */
        movq 8*13(%ecx), %mm7
        psllw $1, %mm1			/* t128=t130 */
        pmulhw 8*13(%esi), %mm7		/* V13 */
        movq %mm1, %mm2			/* duplicate t128=t130 */
        movq 8(%ecx), %mm3
        pmulhw 8(%esi), %mm3		/* V1 */
        movq 8*9(%ecx), %mm5
        psubsw %mm7, %mm1		/* V50 */
        pmulhw 8*9(%esi), %mm5		/* V9 */
        paddsw %mm7, %mm2		/* V51 */
        pmulhw x5a825a825a825a82, %mm1	/* 23170 ->V52 */
        movq %mm2, %mm6			/* duplicate V51 */
        psraw $1, %mm2			/* t138=t144 */
        movq %mm3, %mm4			/* duplicate V1 */
        psraw $2, %mm6			/* t136 */
        paddsw %mm5, %mm3		/* V53 */
        psubsw %mm5, %mm4		/* V54 ;mm5 free */
        movq %mm3, %mm7			/* duplicate V53 */
/* moved from next block */
        movq 8*11(%ecx), %mm0
        psraw $1, %mm4			/* t140=t142 */
        psubsw %mm6, %mm1		/* V55 ; mm6 free */
        paddsw %mm2, %mm3		/* V56 */
        movq %mm4, %mm5			/* duplicate t140=t142 */
        paddsw %mm1, %mm4		/* V57 */
        movq %mm3, 8*5(%esi)		/* V56 */
        psubsw %mm1, %mm5		/* V58; mm1 free */
        movq %mm4, 8*13(%esi)		/* V57 */
        psubsw %mm2, %mm7		/* V59; mm2 free */
        movq %mm5, 8*9(%esi)		/* V58 */
/* keep mm7 alive all along the next block
 *	movq %mm7, 8(%esi)		V59
 * moved above
 *	movq 8*11(%ecx), %mm0
 */
        pmulhw 8*11(%esi), %mm0		/* V11 */
        movq 8*7(%ecx), %mm6
        pmulhw 8*7(%esi), %mm6		/* V7 */
        movq 8*15(%ecx), %mm4
        movq %mm0, %mm3			/* duplicate V11 */
        pmulhw 8*15(%esi), %mm4		/* V15 */
        movq 8*3(%ecx), %mm5
        psllw $1, %mm6			/* t146=t152 */
        pmulhw 8*3(%esi), %mm5		/* V3 */
        paddsw %mm6, %mm0		/* V63 */
/* note that V15 computation has a correction step:
 * this is a 'magic' constant that rebiases the results to be closer to the
 * expected result.  this magic constant can be refined to reduce the error
 * even more by doing the correction step in a later stage when the number
 * is actually multiplied by 16
 */
        paddw x0005000200010001, %mm4
        psubsw %mm6, %mm3		/* V60 ; free mm6 */
        psraw $1, %mm0			/* t154=t156 */
        movq %mm3, %mm1			/* duplicate V60 */
        pmulhw x539f539f539f539f, %mm1	/* V67 */
        movq %mm5, %mm6			/* duplicate V3 */
        psraw $2, %mm4			/* t148=t150 */
        paddsw %mm4, %mm5		/* V61 */
        psubsw %mm4, %mm6		/* V62 ; free mm4 */
        movq %mm5, %mm4			/* duplicate V61 */
        psllw $1, %mm1			/* t169 */
        paddsw %mm0, %mm5		/* V65 -> result */
        psubsw %mm0, %mm4		/* V64 ; free mm0 */
        pmulhw x5a825a825a825a82, %mm4	/* V68 */
        psraw $1, %mm3			/* t158 */
        psubsw %mm6, %mm3		/* V66 */
        movq %mm5, %mm2			/* duplicate V65 */
        pmulhw x61f861f861f861f8, %mm3	/* V70 */
        psllw $1, %mm6			/* t165 */
        pmulhw x4546454645464546, %mm6	/* V69 */
        psraw $1, %mm2			/* t172 */
/* moved from next block */
        movq 8*5(%esi), %mm0		/* V56 */
        psllw $1, %mm4			/* t174 */
/* moved from next block */
        psraw $1, %mm0			/* t177=t188 */
        nop
        psubsw %mm3, %mm6		/* V72 */
        psubsw %mm1, %mm3		/* V71 ; free mm1 */
        psubsw %mm2, %mm6		/* V73 ; free mm2 */
/* moved from next block */
        psraw $1, %mm5			/* t178=t189 */
        psubsw %mm6, %mm4		/* V74 */
/* moved from next block */
        movq %mm0, %mm1			/* duplicate t177=t188 */
        paddsw %mm4, %mm3		/* V75 */
/* moved from next block */
        paddsw %mm5, %mm0		/* tm1 */
/* location
 *  5 - V56
 * 13 - V57
 *  9 - V58
 *  X - V59, mm7
 *  X - V65, mm5
 *  X - V73, mm6
 *  X - V74, mm4
 *  X - V75, mm3
 * free mm0, mm1 & mm2
 * moved above
 *	movq 8*5(%esi), %mm0		V56
 *	psllw $1, %mm0			t177=t188 ! new !!
 *	psllw $1, %mm5			t178=t189 ! new !!
 *	movq %mm0, %mm1			duplicate t177=t188
 *	paddsw %mm5, %mm0		tm1
 */
        movq 8*13(%esi), %mm2		/* V57 */
        psubsw %mm5, %mm1		/* tm15; free mm5 */
        movq %mm0, 8(%esi)		/* tm1; free mm0 */
        psraw $1, %mm7			/* t182=t184 ! new !! */
/* save the store as used directly in the transpose
 *	movq %mm1, 120(%esi)		tm15; free mm1
 */
        movq %mm7, %mm5			/* duplicate t182=t184 */
        psubsw %mm3, %mm7		/* tm7 */
        paddsw %mm3, %mm5		/* tm9; free mm3 */
        movq 8*9(%esi), %mm0		/* V58 */
        movq %mm2, %mm3			/* duplicate V57 */
        movq %mm7, 8*7(%esi)		/* tm7; free mm7 */
        psubsw %mm6, %mm3		/* tm13 */
        paddsw %mm6, %mm2		/* tm3 ; free mm6 */
/* moved up from the transpose */
        movq %mm3, %mm7
/* moved up from the transpose */
        punpcklwd %mm1, %mm3
        movq %mm0, %mm6			/* duplicate V58 */
        movq %mm2, 8*3(%esi)		/* tm3; free mm2 */
        paddsw %mm4, %mm0		/* tm5 */
        psubsw %mm4, %mm6		/* tm11; free mm4 */
/* moved up from the transpose */
        punpckhwd %mm1, %mm7
        movq %mm0, 8*5(%esi)		/* tm5; free mm0 */
/* moved up from the transpose */
        movq %mm5, %mm2
/* transpose - M4 part
 *  ---------       ---------
 * | M1 | M2 |     | M1'| M3'|
 *  ---------  -->  ---------
 * | M3 | M4 |     | M2'| M4'|
 *  ---------       ---------
 * Two alternatives: use full mmword approach so the following code can be
 * scheduled before the transpose is done without stores, or use the faster
 * half mmword stores (when possible)
 */
        movd %mm3, 8*9+4(%esi)		/* MS part of tmt9 */
        punpcklwd %mm6, %mm5
        movd %mm7, 8*13+4(%esi)		/* MS part of tmt13 */
        punpckhwd %mm6, %mm2
        movd %mm5, 8*9(%esi)		/* LS part of tmt9 */
        punpckhdq %mm3, %mm5		/* free mm3 */
        movd %mm2, 8*13(%esi)		/* LS part of tmt13 */
        punpckhdq %mm7, %mm2		/* free mm7 */
/* moved up from the M3 transpose */
        movq 8*8(%esi), %mm0
/* moved up from the M3 transpose */
        movq 8*10(%esi), %mm1
/* moved up from the M3 transpose */
        movq %mm0, %mm3
/* shuffle the rest of the data, and write it with 2 mmword writes */
        movq %mm5, 8*11(%esi)		/* tmt11 */
/* moved up from the M3 transpose */
        punpcklwd %mm1, %mm0
        movq %mm2, 8*15(%esi)		/* tmt15 */
/* moved up from the M3 transpose */
        punpckhwd %mm1, %mm3
/* transpose - M3 part
 * moved up to previous code section
 *	movq 8*8(%esi), %mm0
 *	movq 8*10(%esi), %mm1
 *	movq %mm0, %mm3
 *	punpcklwd %mm1, %mm0
 *	punpckhwd %mm1, %mm3
 */
        movq 8*12(%esi), %mm6
        movq 8*14(%esi), %mm4
        movq %mm6, %mm2
/* shuffle the data and write the lower parts of the transposed in 4 dwords */
        punpcklwd %mm4, %mm6
        movq %mm0, %mm1
        punpckhdq %mm6, %mm1
        movq %mm3, %mm7
        punpckhwd %mm4, %mm2		/* free mm4 */
        punpckldq %mm6, %mm0		/* free mm6 */
/* moved from next block */
        movq 8*13(%esi), %mm4		/* tmt13 */
        punpckldq %mm2, %mm3
        punpckhdq %mm2, %mm7		/* free mm2 */
/* moved from next block */
        movq %mm3, %mm5			/* duplicate tmt5 */
/* column 1: even part (after transpose)
* moved above
*	movq %mm3, %mm5			duplicate tmt5
*	movq 8*13(%esi), %mm4		tmt13
*/
        psubsw %mm4, %mm3		/* V134 */
        pmulhw x5a825a825a825a82, %mm3	/* 23170 ->V136 */
        movq 8*9(%esi), %mm6		/* tmt9 */
        paddsw %mm4, %mm5		/* V135 ; mm4 free */
        movq %mm0, %mm4			/* duplicate tmt1 */
        paddsw %mm6, %mm0		/* V137 */
        psubsw %mm6, %mm4		/* V138 ; mm6 free */
        psllw $2, %mm3			/* t290 */
        psubsw %mm5, %mm3		/* V139 */
        movq %mm0, %mm6			/* duplicate V137 */
        paddsw %mm5, %mm0		/* V140 */
        movq %mm4, %mm2			/* duplicate V138 */
        paddsw %mm3, %mm2		/* V141 */
        psubsw %mm3, %mm4		/* V142 ; mm3 free */
        movq %mm0, 8*9(%esi)		/* V140 */
        psubsw %mm5, %mm6		/* V143 ; mm5 free */
/* moved from next block */
        movq 8*11(%esi), %mm0		/* tmt11 */
        movq %mm2, 8*13(%esi)		/* V141 */
/* moved from next block */
        movq %mm0, %mm2			/* duplicate tmt11 */
/* column 1: odd part (after transpose) */
/* moved up to the prev block
 *	movq 8*11(%esi), %mm0		tmt11
 *	movq %mm0, %mm2			duplicate tmt11
 */
        movq 8*15(%esi), %mm5		/* tmt15 */
        psubsw %mm7, %mm0		/* V144 */
        movq %mm0, %mm3			/* duplicate V144 */
        paddsw %mm7, %mm2		/* V147 ; free mm7 */
        pmulhw x539f539f539f539f, %mm0	/* 21407-> V151 */
        movq %mm1, %mm7			/* duplicate tmt3 */
        paddsw %mm5, %mm7		/* V145 */
        psubsw %mm5, %mm1		/* V146 ; free mm5 */
        psubsw %mm1, %mm3		/* V150 */
        movq %mm7, %mm5			/* duplicate V145 */
        pmulhw x4546454645464546, %mm1	/* 17734-> V153 */
        psubsw %mm2, %mm5		/* V148 */
        pmulhw x61f861f861f861f8, %mm3	/* 25080-> V154 */
        psllw $2, %mm0			/* t311 */
        pmulhw x5a825a825a825a82, %mm5	/* 23170-> V152 */
        paddsw %mm2, %mm7		/* V149 ; free mm2 */
        psllw $1, %mm1			/* t313 */
        nop	/* without the nop - freeze here for one clock */
        movq %mm3, %mm2			/* duplicate V154 */
        psubsw %mm0, %mm3		/* V155 ; free mm0 */
        psubsw %mm2, %mm1		/* V156 ; free mm2 */
/* moved from the next block */
        movq %mm6, %mm2			/* duplicate V143 */
/* moved from the next block */
        movq 8*13(%esi), %mm0		/* V141 */
        psllw $1, %mm1			/* t315 */
        psubsw %mm7, %mm1		/* V157 (keep V149) */
        psllw $2, %mm5			/* t317 */
        psubsw %mm1, %mm5		/* V158 */
        psllw $1, %mm3			/* t319 */
        paddsw %mm5, %mm3		/* V159 */
/* column 1: output butterfly (after transform)
 * moved to the prev block
 *	movq %mm6, %mm2			duplicate V143
 *	movq 8*13(%esi), %mm0		V141
 */
        psubsw %mm3, %mm2		/* V163 */
        paddsw %mm3, %mm6		/* V164 ; free mm3 */
        movq %mm4, %mm3			/* duplicate V142 */
        psubsw %mm5, %mm4		/* V165 ; free mm5 */
        movq %mm2, scratch7		/* out7 */
        psraw $4, %mm6
        psraw $4, %mm4
        paddsw %mm5, %mm3		/* V162 */