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

media/libtremor/lib/asm_arm.h@b8a032363ba2 (annotated)

media/libtremor/lib/asm_arm.h

Thu, 22 Jan 2015 13:21:57 +0100

author

Michael Schloh von Bennewitz <michael@schloh.com>

date

Thu, 22 Jan 2015 13:21:57 +0100

branch

TOR_BUG_9701

changeset 15

b8a032363ba2

permissions

-rw-r--r--

Incorporate requested changes from Mozilla in review:
https://bugzilla.mozilla.org/show_bug.cgi?id=1123480#c6

michael@0	1	/********************************************************************
michael@0	2	* *
michael@0	3	* THIS FILE IS PART OF THE OggVorbis 'TREMOR' CODEC SOURCE CODE. *
michael@0	4	* *
michael@0	5	* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
michael@0	6	* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
michael@0	7	* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
michael@0	8	* *
michael@0	9	* THE OggVorbis 'TREMOR' SOURCE CODE IS (C) COPYRIGHT 1994-2002 *
michael@0	10	* BY THE Xiph.Org FOUNDATION http://www.xiph.org/ *
michael@0	11	* *
michael@0	12	********************************************************************
michael@0	13
michael@0	14	function: arm7 and later wide math functions
michael@0	15
michael@0	16	********************************************************************/
michael@0	17
michael@0	18	#ifdef _ARM_ASSEM_
michael@0	19
michael@0	20	#if !defined(_V_WIDE_MATH) && !defined(_LOW_ACCURACY_)
michael@0	21	#define _V_WIDE_MATH
michael@0	22
michael@0	23	static inline ogg_int32_t MULT32(ogg_int32_t x, ogg_int32_t y) {
michael@0	24	int lo,hi;
michael@0	25	asm volatile("smull\t%0, %1, %2, %3"
michael@0	26	: "=&r"(lo),"=&r"(hi)
michael@0	27	: "%r"(x),"r"(y)
michael@0	28	: "cc");
michael@0	29	return(hi);
michael@0	30	}
michael@0	31
michael@0	32	static inline ogg_int32_t MULT31(ogg_int32_t x, ogg_int32_t y) {
michael@0	33	return MULT32(x,y)<<1;
michael@0	34	}
michael@0	35
michael@0	36	static inline ogg_int32_t MULT31_SHIFT15(ogg_int32_t x, ogg_int32_t y) {
michael@0	37	int lo,hi;
michael@0	38	asm volatile("smull %0, %1, %2, %3\n\t"
michael@0	39	"movs %0, %0, lsr #15\n\t"
michael@0	40	"adc %1, %0, %1, lsl #17\n\t"
michael@0	41	: "=&r"(lo),"=&r"(hi)
michael@0	42	: "%r"(x),"r"(y)
michael@0	43	: "cc");
michael@0	44	return(hi);
michael@0	45	}
michael@0	46
michael@0	47	#define MB() asm volatile ("" : : : "memory")
michael@0	48
michael@0	49	static inline void XPROD32(ogg_int32_t a, ogg_int32_t b,
michael@0	50	ogg_int32_t t, ogg_int32_t v,
michael@0	51	ogg_int32_t *x, ogg_int32_t *y)
michael@0	52	{
michael@0	53	int x1, y1, l;
michael@0	54	asm( "smull %0, %1, %4, %6\n\t"
michael@0	55	"smlal %0, %1, %5, %7\n\t"
michael@0	56	"rsb %3, %4, #0\n\t"
michael@0	57	"smull %0, %2, %5, %6\n\t"
michael@0	58	"smlal %0, %2, %3, %7"
michael@0	59	: "=&r" (l), "=&r" (x1), "=&r" (y1), "=r" (a)
michael@0	60	: "3" (a), "r" (b), "r" (t), "r" (v)
michael@0	61	: "cc" );
michael@0	62	*x = x1;
michael@0	63	MB();
michael@0	64	*y = y1;
michael@0	65	}
michael@0	66
michael@0	67	static inline void XPROD31(ogg_int32_t a, ogg_int32_t b,
michael@0	68	ogg_int32_t t, ogg_int32_t v,
michael@0	69	ogg_int32_t *x, ogg_int32_t *y)
michael@0	70	{
michael@0	71	int x1, y1, l;
michael@0	72	asm( "smull %0, %1, %4, %6\n\t"
michael@0	73	"smlal %0, %1, %5, %7\n\t"
michael@0	74	"rsb %3, %4, #0\n\t"
michael@0	75	"smull %0, %2, %5, %6\n\t"
michael@0	76	"smlal %0, %2, %3, %7"
michael@0	77	: "=&r" (l), "=&r" (x1), "=&r" (y1), "=r" (a)
michael@0	78	: "3" (a), "r" (b), "r" (t), "r" (v)
michael@0	79	: "cc" );
michael@0	80	*x = x1 << 1;
michael@0	81	MB();
michael@0	82	*y = y1 << 1;
michael@0	83	}
michael@0	84
michael@0	85	static inline void XNPROD31(ogg_int32_t a, ogg_int32_t b,
michael@0	86	ogg_int32_t t, ogg_int32_t v,
michael@0	87	ogg_int32_t *x, ogg_int32_t *y)
michael@0	88	{
michael@0	89	int x1, y1, l;
michael@0	90	asm( "rsb %2, %4, #0\n\t"
michael@0	91	"smull %0, %1, %3, %5\n\t"
michael@0	92	"smlal %0, %1, %2, %6\n\t"
michael@0	93	"smull %0, %2, %4, %5\n\t"
michael@0	94	"smlal %0, %2, %3, %6"
michael@0	95	: "=&r" (l), "=&r" (x1), "=&r" (y1)
michael@0	96	: "r" (a), "r" (b), "r" (t), "r" (v)
michael@0	97	: "cc" );
michael@0	98	*x = x1 << 1;
michael@0	99	MB();
michael@0	100	*y = y1 << 1;
michael@0	101	}
michael@0	102
michael@0	103	#endif
michael@0	104
michael@0	105	#ifndef _V_CLIP_MATH
michael@0	106	#define _V_CLIP_MATH
michael@0	107
michael@0	108	static inline ogg_int32_t CLIP_TO_15(ogg_int32_t x) {
michael@0	109	int tmp;
michael@0	110	asm volatile("subs %1, %0, #32768\n\t"
michael@0	111	"movpl %0, #0x7f00\n\t"
michael@0	112	"orrpl %0, %0, #0xff\n"
michael@0	113	"adds %1, %0, #32768\n\t"
michael@0	114	"movmi %0, #0x8000"
michael@0	115	: "+r"(x),"=r"(tmp)
michael@0	116	:
michael@0	117	: "cc");
michael@0	118	return(x);
michael@0	119	}
michael@0	120
michael@0	121	#endif
michael@0	122
michael@0	123	#ifndef _V_LSP_MATH_ASM
michael@0	124	#define _V_LSP_MATH_ASM
michael@0	125
michael@0	126	static inline void lsp_loop_asm(ogg_uint32_t *qip,ogg_uint32_t *pip,
michael@0	127	ogg_int32_t *qexpp,
michael@0	128	ogg_int32_t *ilsp,ogg_int32_t wi,
michael@0	129	ogg_int32_t m){
michael@0	130
michael@0	131	ogg_uint32_t qi=*qip,pi=*pip;
michael@0	132	ogg_int32_t qexp=*qexpp;
michael@0	133
michael@0	134	asm("mov r0,%3;"
michael@0	135	"movs r1,%5,asr#1;"
michael@0	136	"add r0,r0,r1,lsl#3;"
michael@0	137	"beq 2f;\n"
michael@0	138	"1:"
michael@0	139
michael@0	140	"ldmdb r0!,{r1,r3};"
michael@0	141	"subs r1,r1,%4;" //ilsp[j]-wi
michael@0	142	"rsbmi r1,r1,#0;" //labs(ilsp[j]-wi)
michael@0	143	"umull %0,r2,r1,%0;" //qi*=labs(ilsp[j]-wi)
michael@0	144
michael@0	145	"subs r1,r3,%4;" //ilsp[j+1]-wi
michael@0	146	"rsbmi r1,r1,#0;" //labs(ilsp[j+1]-wi)
michael@0	147	"umull %1,r3,r1,%1;" //pi*=labs(ilsp[j+1]-wi)
michael@0	148
michael@0	149	"cmn r2,r3;" // shift down 16?
michael@0	150	"beq 0f;"
michael@0	151	"add %2,%2,#16;"
michael@0	152	"mov %0,%0,lsr #16;"
michael@0	153	"orr %0,%0,r2,lsl #16;"
michael@0	154	"mov %1,%1,lsr #16;"
michael@0	155	"orr %1,%1,r3,lsl #16;"
michael@0	156	"0:"
michael@0	157	"cmp r0,%3;\n"
michael@0	158	"bhi 1b;\n"
michael@0	159
michael@0	160	"2:"
michael@0	161	// odd filter assymetry
michael@0	162	"ands r0,%5,#1;\n"
michael@0	163	"beq 3f;\n"
michael@0	164	"add r0,%3,%5,lsl#2;\n"
michael@0	165
michael@0	166	"ldr r1,[r0,#-4];\n"
michael@0	167	"mov r0,#0x4000;\n"
michael@0	168
michael@0	169	"subs r1,r1,%4;\n" //ilsp[j]-wi
michael@0	170	"rsbmi r1,r1,#0;\n" //labs(ilsp[j]-wi)
michael@0	171	"umull %0,r2,r1,%0;\n" //qi*=labs(ilsp[j]-wi)
michael@0	172	"umull %1,r3,r0,%1;\n" //pi*=labs(ilsp[j+1]-wi)
michael@0	173
michael@0	174	"cmn r2,r3;\n" // shift down 16?
michael@0	175	"beq 3f;\n"
michael@0	176	"add %2,%2,#16;\n"
michael@0	177	"mov %0,%0,lsr #16;\n"
michael@0	178	"orr %0,%0,r2,lsl #16;\n"
michael@0	179	"mov %1,%1,lsr #16;\n"
michael@0	180	"orr %1,%1,r3,lsl #16;\n"
michael@0	181
michael@0	182	//qi=(pi>>shift)*labs(ilsp[j]-wi);
michael@0	183	//pi=(qi>>shift)*labs(ilsp[j+1]-wi);
michael@0	184	//qexp+=shift;
michael@0	185
michael@0	186	//}
michael@0	187
michael@0	188	/* normalize to max 16 sig figs */
michael@0	189	"3:"
michael@0	190	"mov r2,#0;"
michael@0	191	"orr r1,%0,%1;"
michael@0	192	"tst r1,#0xff000000;"
michael@0	193	"addne r2,r2,#8;"
michael@0	194	"movne r1,r1,lsr #8;"
michael@0	195	"tst r1,#0x00f00000;"
michael@0	196	"addne r2,r2,#4;"
michael@0	197	"movne r1,r1,lsr #4;"
michael@0	198	"tst r1,#0x000c0000;"
michael@0	199	"addne r2,r2,#2;"
michael@0	200	"movne r1,r1,lsr #2;"
michael@0	201	"tst r1,#0x00020000;"
michael@0	202	"addne r2,r2,#1;"
michael@0	203	"movne r1,r1,lsr #1;"
michael@0	204	"tst r1,#0x00010000;"
michael@0	205	"addne r2,r2,#1;"
michael@0	206	"mov %0,%0,lsr r2;"
michael@0	207	"mov %1,%1,lsr r2;"
michael@0	208	"add %2,%2,r2;"
michael@0	209
michael@0	210	: "+r"(qi),"+r"(pi),"+r"(qexp)
michael@0	211	: "r"(ilsp),"r"(wi),"r"(m)
michael@0	212	: "r0","r1","r2","r3","cc");
michael@0	213
michael@0	214	*qip=qi;
michael@0	215	*pip=pi;
michael@0	216	*qexpp=qexp;
michael@0	217	}
michael@0	218
michael@0	219	static inline void lsp_norm_asm(ogg_uint32_t *qip,ogg_int32_t *qexpp){
michael@0	220
michael@0	221	ogg_uint32_t qi=*qip;
michael@0	222	ogg_int32_t qexp=*qexpp;
michael@0	223
michael@0	224	asm("tst %0,#0x0000ff00;"
michael@0	225	"moveq %0,%0,lsl #8;"
michael@0	226	"subeq %1,%1,#8;"
michael@0	227	"tst %0,#0x0000f000;"
michael@0	228	"moveq %0,%0,lsl #4;"
michael@0	229	"subeq %1,%1,#4;"
michael@0	230	"tst %0,#0x0000c000;"
michael@0	231	"moveq %0,%0,lsl #2;"
michael@0	232	"subeq %1,%1,#2;"
michael@0	233	"tst %0,#0x00008000;"
michael@0	234	"moveq %0,%0,lsl #1;"
michael@0	235	"subeq %1,%1,#1;"
michael@0	236	: "+r"(qi),"+r"(qexp)
michael@0	237	:
michael@0	238	: "cc");
michael@0	239	*qip=qi;
michael@0	240	*qexpp=qexp;
michael@0	241	}
michael@0	242
michael@0	243	#endif
michael@0	244	#endif
michael@0	245