• file
• revisions
• annotate
• diff
• comparison
• raw

media/libtheora/lib/x86_vc/mmxloop.h@6474c204b198 (annotated)

media/libtheora/lib/x86_vc/mmxloop.h

Wed, 31 Dec 2014 06:09:35 +0100

author

Michael Schloh von Bennewitz <michael@schloh.com>

date

Wed, 31 Dec 2014 06:09:35 +0100

changeset 0

6474c204b198

permissions

-rw-r--r--

Cloned upstream origin tor-browser at tor-browser-31.3.0esr-4.5-1-build1
revision ID fc1c9ff7c1b2defdbc039f12214767608f46423f for hacking purpose.

michael@0	1	#if !defined(_x86_vc_mmxloop_H)
michael@0	2	# define _x86_vc_mmxloop_H (1)
michael@0	3	# include <stddef.h>
michael@0	4	# include "x86int.h"
michael@0	5
michael@0	6	#if defined(OC_X86_ASM)
michael@0	7
michael@0	8	/*On entry, mm0={a0,...,a7}, mm1={b0,...,b7}, mm2={c0,...,c7}, mm3={d0,...d7}.
michael@0	9	On exit, mm1={b0+lflim(R_0,L),...,b7+lflim(R_7,L)} and
michael@0	10	mm2={c0-lflim(R_0,L),...,c7-lflim(R_7,L)}; mm0 and mm3 are clobbered.*/
michael@0	11	#define OC_LOOP_FILTER8_MMX __asm{ \
michael@0	12	/*mm7=0*/ \
michael@0	13	__asm pxor mm7,mm7 \
michael@0	14	/*mm6:mm0={a0,...,a7}*/ \
michael@0	15	__asm movq mm6,mm0 \
michael@0	16	__asm punpcklbw mm0,mm7 \
michael@0	17	__asm punpckhbw mm6,mm7 \
michael@0	18	/*mm3:mm5={d0,...,d7}*/ \
michael@0	19	__asm movq mm5,mm3 \
michael@0	20	__asm punpcklbw mm3,mm7 \
michael@0	21	__asm punpckhbw mm5,mm7 \
michael@0	22	/*mm6:mm0={a0-d0,...,a7-d7}*/ \
michael@0	23	__asm psubw mm0,mm3 \
michael@0	24	__asm psubw mm6,mm5 \
michael@0	25	/*mm3:mm1={b0,...,b7}*/ \
michael@0	26	__asm movq mm3,mm1 \
michael@0	27	__asm punpcklbw mm1,mm7 \
michael@0	28	__asm movq mm4,mm2 \
michael@0	29	__asm punpckhbw mm3,mm7 \
michael@0	30	/*mm5:mm4={c0,...,c7}*/ \
michael@0	31	__asm movq mm5,mm2 \
michael@0	32	__asm punpcklbw mm4,mm7 \
michael@0	33	__asm punpckhbw mm5,mm7 \
michael@0	34	/*mm7={3}x4 \
michael@0	35	mm5:mm4={c0-b0,...,c7-b7}*/ \
michael@0	36	__asm pcmpeqw mm7,mm7 \
michael@0	37	__asm psubw mm4,mm1 \
michael@0	38	__asm psrlw mm7,14 \
michael@0	39	__asm psubw mm5,mm3 \
michael@0	40	/*Scale by 3.*/ \
michael@0	41	__asm pmullw mm4,mm7 \
michael@0	42	__asm pmullw mm5,mm7 \
michael@0	43	/*mm7={4}x4 \
michael@0	44	mm5:mm4=f={a0-d0+3*(c0-b0),...,a7-d7+3*(c7-b7)}*/ \
michael@0	45	__asm psrlw mm7,1 \
michael@0	46	__asm paddw mm4,mm0 \
michael@0	47	__asm psllw mm7,2 \
michael@0	48	__asm movq mm0,[LL] \
michael@0	49	__asm paddw mm5,mm6 \
michael@0	50	/*R_i has the range [-127,128], so we compute -R_i instead. \
michael@0	51	mm4=-R_i=-(f+4>>3)=0xFF^(f-4>>3)*/ \
michael@0	52	__asm psubw mm4,mm7 \
michael@0	53	__asm psubw mm5,mm7 \
michael@0	54	__asm psraw mm4,3 \
michael@0	55	__asm psraw mm5,3 \
michael@0	56	__asm pcmpeqb mm7,mm7 \
michael@0	57	__asm packsswb mm4,mm5 \
michael@0	58	__asm pxor mm6,mm6 \
michael@0	59	__asm pxor mm4,mm7 \
michael@0	60	__asm packuswb mm1,mm3 \
michael@0	61	/*Now compute lflim of -mm4 cf. Section 7.10 of the sepc.*/ \
michael@0	62	/*There's no unsigned byte+signed byte with unsigned saturation op code, so \
michael@0	63	we have to split things by sign (the other option is to work in 16 bits, \
michael@0	64	but working in 8 bits gives much better parallelism). \
michael@0	65	We compute abs(R_i), but save a mask of which terms were negative in mm6. \
michael@0	66	Then we compute mm4=abs(lflim(R_i,L))=min(abs(R_i),max(2*L-abs(R_i),0)). \
michael@0	67	Finally, we split mm4 into positive and negative pieces using the mask in \
michael@0	68	mm6, and add and subtract them as appropriate.*/ \
michael@0	69	/*mm4=abs(-R_i)*/ \
michael@0	70	/*mm7=255-2*L*/ \
michael@0	71	__asm pcmpgtb mm6,mm4 \
michael@0	72	__asm psubb mm7,mm0 \
michael@0	73	__asm pxor mm4,mm6 \
michael@0	74	__asm psubb mm7,mm0 \
michael@0	75	__asm psubb mm4,mm6 \
michael@0	76	/*mm7=255-max(2*L-abs(R_i),0)*/ \
michael@0	77	__asm paddusb mm7,mm4 \
michael@0	78	/*mm4=min(abs(R_i),max(2*L-abs(R_i),0))*/ \
michael@0	79	__asm paddusb mm4,mm7 \
michael@0	80	__asm psubusb mm4,mm7 \
michael@0	81	/*Now split mm4 by the original sign of -R_i.*/ \
michael@0	82	__asm movq mm5,mm4 \
michael@0	83	__asm pand mm4,mm6 \
michael@0	84	__asm pandn mm6,mm5 \
michael@0	85	/*mm1={b0+lflim(R_0,L),...,b7+lflim(R_7,L)}*/ \
michael@0	86	/*mm2={c0-lflim(R_0,L),...,c7-lflim(R_7,L)}*/ \
michael@0	87	__asm paddusb mm1,mm4 \
michael@0	88	__asm psubusb mm2,mm4 \
michael@0	89	__asm psubusb mm1,mm6 \
michael@0	90	__asm paddusb mm2,mm6 \
michael@0	91	}
michael@0	92
michael@0	93	#define OC_LOOP_FILTER_V_MMX(_pix,_ystride,_ll) \
michael@0	94	do{ \
michael@0	95	/*Used local variable pix__ in order to fix compilation errors like: \
michael@0	96	"error C2425: 'SHL' : non-constant expression in 'second operand'".*/ \
michael@0	97	unsigned char *pix__; \
michael@0	98	unsigned char *ll__; \
michael@0	99	ll__=(_ll); \
michael@0	100	pix__=(_pix); \
michael@0	101	__asm mov YSTRIDE,_ystride \
michael@0	102	__asm mov LL,ll__ \
michael@0	103	__asm mov PIX,pix__ \
michael@0	104	__asm sub PIX,YSTRIDE \
michael@0	105	__asm sub PIX,YSTRIDE \
michael@0	106	/*mm0={a0,...,a7}*/ \
michael@0	107	__asm movq mm0,[PIX] \
michael@0	108	/*ystride3=_ystride*3*/ \
michael@0	109	__asm lea YSTRIDE3,[YSTRIDE+YSTRIDE*2] \
michael@0	110	/*mm3={d0,...,d7}*/ \
michael@0	111	__asm movq mm3,[PIX+YSTRIDE3] \
michael@0	112	/*mm1={b0,...,b7}*/ \
michael@0	113	__asm movq mm1,[PIX+YSTRIDE] \
michael@0	114	/*mm2={c0,...,c7}*/ \
michael@0	115	__asm movq mm2,[PIX+YSTRIDE*2] \
michael@0	116	OC_LOOP_FILTER8_MMX \
michael@0	117	/*Write it back out.*/ \
michael@0	118	__asm movq [PIX+YSTRIDE],mm1 \
michael@0	119	__asm movq [PIX+YSTRIDE*2],mm2 \
michael@0	120	} \
michael@0	121	while(0)
michael@0	122
michael@0	123	#define OC_LOOP_FILTER_H_MMX(_pix,_ystride,_ll) \
michael@0	124	do{ \
michael@0	125	/*Used local variable ll__ in order to fix compilation errors like: \
michael@0	126	"error C2443: operand size conflict".*/ \
michael@0	127	unsigned char *ll__; \
michael@0	128	unsigned char *pix__; \
michael@0	129	ll__=(_ll); \
michael@0	130	pix__=(_pix)-2; \
michael@0	131	__asm mov PIX,pix__ \
michael@0	132	__asm mov YSTRIDE,_ystride \
michael@0	133	__asm mov LL,ll__ \
michael@0	134	/*x x x x d0 c0 b0 a0*/ \
michael@0	135	__asm movd mm0,[PIX] \
michael@0	136	/*x x x x d1 c1 b1 a1*/ \
michael@0	137	__asm movd mm1,[PIX+YSTRIDE] \
michael@0	138	/*ystride3=_ystride*3*/ \
michael@0	139	__asm lea YSTRIDE3,[YSTRIDE+YSTRIDE*2] \
michael@0	140	/*x x x x d2 c2 b2 a2*/ \
michael@0	141	__asm movd mm2,[PIX+YSTRIDE*2] \
michael@0	142	/*x x x x d3 c3 b3 a3*/ \
michael@0	143	__asm lea D,[PIX+YSTRIDE*4] \
michael@0	144	__asm movd mm3,[PIX+YSTRIDE3] \
michael@0	145	/*x x x x d4 c4 b4 a4*/ \
michael@0	146	__asm movd mm4,[D] \
michael@0	147	/*x x x x d5 c5 b5 a5*/ \
michael@0	148	__asm movd mm5,[D+YSTRIDE] \
michael@0	149	/*x x x x d6 c6 b6 a6*/ \
michael@0	150	__asm movd mm6,[D+YSTRIDE*2] \
michael@0	151	/*x x x x d7 c7 b7 a7*/ \
michael@0	152	__asm movd mm7,[D+YSTRIDE3] \
michael@0	153	/*mm0=d1 d0 c1 c0 b1 b0 a1 a0*/ \
michael@0	154	__asm punpcklbw mm0,mm1 \
michael@0	155	/*mm2=d3 d2 c3 c2 b3 b2 a3 a2*/ \
michael@0	156	__asm punpcklbw mm2,mm3 \
michael@0	157	/*mm3=d1 d0 c1 c0 b1 b0 a1 a0*/ \
michael@0	158	__asm movq mm3,mm0 \
michael@0	159	/*mm0=b3 b2 b1 b0 a3 a2 a1 a0*/ \
michael@0	160	__asm punpcklwd mm0,mm2 \
michael@0	161	/*mm3=d3 d2 d1 d0 c3 c2 c1 c0*/ \
michael@0	162	__asm punpckhwd mm3,mm2 \
michael@0	163	/*mm1=b3 b2 b1 b0 a3 a2 a1 a0*/ \
michael@0	164	__asm movq mm1,mm0 \
michael@0	165	/*mm4=d5 d4 c5 c4 b5 b4 a5 a4*/ \
michael@0	166	__asm punpcklbw mm4,mm5 \
michael@0	167	/*mm6=d7 d6 c7 c6 b7 b6 a7 a6*/ \
michael@0	168	__asm punpcklbw mm6,mm7 \
michael@0	169	/*mm5=d5 d4 c5 c4 b5 b4 a5 a4*/ \
michael@0	170	__asm movq mm5,mm4 \
michael@0	171	/*mm4=b7 b6 b5 b4 a7 a6 a5 a4*/ \
michael@0	172	__asm punpcklwd mm4,mm6 \
michael@0	173	/*mm5=d7 d6 d5 d4 c7 c6 c5 c4*/ \
michael@0	174	__asm punpckhwd mm5,mm6 \
michael@0	175	/*mm2=d3 d2 d1 d0 c3 c2 c1 c0*/ \
michael@0	176	__asm movq mm2,mm3 \
michael@0	177	/*mm0=a7 a6 a5 a4 a3 a2 a1 a0*/ \
michael@0	178	__asm punpckldq mm0,mm4 \
michael@0	179	/*mm1=b7 b6 b5 b4 b3 b2 b1 b0*/ \
michael@0	180	__asm punpckhdq mm1,mm4 \
michael@0	181	/*mm2=c7 c6 c5 c4 c3 c2 c1 c0*/ \
michael@0	182	__asm punpckldq mm2,mm5 \
michael@0	183	/*mm3=d7 d6 d5 d4 d3 d2 d1 d0*/ \
michael@0	184	__asm punpckhdq mm3,mm5 \
michael@0	185	OC_LOOP_FILTER8_MMX \
michael@0	186	/*mm2={b0+R_0'',...,b7+R_7''}*/ \
michael@0	187	__asm movq mm0,mm1 \
michael@0	188	/*mm1={b0+R_0'',c0-R_0'',...,b3+R_3'',c3-R_3''}*/ \
michael@0	189	__asm punpcklbw mm1,mm2 \
michael@0	190	/*mm2={b4+R_4'',c4-R_4'',...,b7+R_7'',c7-R_7''}*/ \
michael@0	191	__asm punpckhbw mm0,mm2 \
michael@0	192	/*[d]=c1 b1 c0 b0*/ \
michael@0	193	__asm movd D,mm1 \
michael@0	194	__asm mov [PIX+1],D_WORD \
michael@0	195	__asm psrlq mm1,32 \
michael@0	196	__asm shr D,16 \
michael@0	197	__asm mov [PIX+YSTRIDE+1],D_WORD \
michael@0	198	/*[d]=c3 b3 c2 b2*/ \
michael@0	199	__asm movd D,mm1 \
michael@0	200	__asm mov [PIX+YSTRIDE*2+1],D_WORD \
michael@0	201	__asm shr D,16 \
michael@0	202	__asm mov [PIX+YSTRIDE3+1],D_WORD \
michael@0	203	__asm lea PIX,[PIX+YSTRIDE*4] \
michael@0	204	/*[d]=c5 b5 c4 b4*/ \
michael@0	205	__asm movd D,mm0 \
michael@0	206	__asm mov [PIX+1],D_WORD \
michael@0	207	__asm psrlq mm0,32 \
michael@0	208	__asm shr D,16 \
michael@0	209	__asm mov [PIX+YSTRIDE+1],D_WORD \
michael@0	210	/*[d]=c7 b7 c6 b6*/ \
michael@0	211	__asm movd D,mm0 \
michael@0	212	__asm mov [PIX+YSTRIDE*2+1],D_WORD \
michael@0	213	__asm shr D,16 \
michael@0	214	__asm mov [PIX+YSTRIDE3+1],D_WORD \
michael@0	215	} \
michael@0	216	while(0)
michael@0	217
michael@0	218	# endif
michael@0	219	#endif