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

media/libyuv/source/rotate_mips.cc@b8a032363ba2 (annotated)

media/libyuv/source/rotate_mips.cc

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	* Copyright 2011 The LibYuv Project Authors. All rights reserved.
michael@0	3	*
michael@0	4	* Use of this source code is governed by a BSD-style license
michael@0	5	* that can be found in the LICENSE file in the root of the source
michael@0	6	* tree. An additional intellectual property rights grant can be found
michael@0	7	* in the file PATENTS. All contributing project authors may
michael@0	8	* be found in the AUTHORS file in the root of the source tree.
michael@0	9	*/
michael@0	10
michael@0	11	#include "libyuv/row.h"
michael@0	12
michael@0	13	#include "libyuv/basic_types.h"
michael@0	14
michael@0	15	#ifdef __cplusplus
michael@0	16	namespace libyuv {
michael@0	17	extern "C" {
michael@0	18	#endif
michael@0	19
michael@0	20	#if !defined(LIBYUV_DISABLE_MIPS) && \
michael@0	21	defined(__mips_dsp) && (__mips_dsp_rev >= 2)
michael@0	22
michael@0	23	void TransposeWx8_MIPS_DSPR2(const uint8* src, int src_stride,
michael@0	24	uint8* dst, int dst_stride,
michael@0	25	int width) {
michael@0	26	__asm__ __volatile__ (
michael@0	27	".set push \n"
michael@0	28	".set noreorder \n"
michael@0	29	"sll $t2, %[src_stride], 0x1 \n" // src_stride x 2
michael@0	30	"sll $t4, %[src_stride], 0x2 \n" // src_stride x 4
michael@0	31	"sll $t9, %[src_stride], 0x3 \n" // src_stride x 8
michael@0	32	"addu $t3, $t2, %[src_stride] \n"
michael@0	33	"addu $t5, $t4, %[src_stride] \n"
michael@0	34	"addu $t6, $t2, $t4 \n"
michael@0	35	"andi $t0, %[dst], 0x3 \n"
michael@0	36	"andi $t1, %[dst_stride], 0x3 \n"
michael@0	37	"or $t0, $t0, $t1 \n"
michael@0	38	"bnez $t0, 11f \n"
michael@0	39	" subu $t7, $t9, %[src_stride] \n"
michael@0	40	//dst + dst_stride word aligned
michael@0	41	"1: \n"
michael@0	42	"lbu $t0, 0(%[src]) \n"
michael@0	43	"lbux $t1, %[src_stride](%[src]) \n"
michael@0	44	"lbux $t8, $t2(%[src]) \n"
michael@0	45	"lbux $t9, $t3(%[src]) \n"
michael@0	46	"sll $t1, $t1, 16 \n"
michael@0	47	"sll $t9, $t9, 16 \n"
michael@0	48	"or $t0, $t0, $t1 \n"
michael@0	49	"or $t8, $t8, $t9 \n"
michael@0	50	"precr.qb.ph $s0, $t8, $t0 \n"
michael@0	51	"lbux $t0, $t4(%[src]) \n"
michael@0	52	"lbux $t1, $t5(%[src]) \n"
michael@0	53	"lbux $t8, $t6(%[src]) \n"
michael@0	54	"lbux $t9, $t7(%[src]) \n"
michael@0	55	"sll $t1, $t1, 16 \n"
michael@0	56	"sll $t9, $t9, 16 \n"
michael@0	57	"or $t0, $t0, $t1 \n"
michael@0	58	"or $t8, $t8, $t9 \n"
michael@0	59	"precr.qb.ph $s1, $t8, $t0 \n"
michael@0	60	"sw $s0, 0(%[dst]) \n"
michael@0	61	"addiu %[width], -1 \n"
michael@0	62	"addiu %[src], 1 \n"
michael@0	63	"sw $s1, 4(%[dst]) \n"
michael@0	64	"bnez %[width], 1b \n"
michael@0	65	" addu %[dst], %[dst], %[dst_stride] \n"
michael@0	66	"b 2f \n"
michael@0	67	//dst + dst_stride unaligned
michael@0	68	"11: \n"
michael@0	69	"lbu $t0, 0(%[src]) \n"
michael@0	70	"lbux $t1, %[src_stride](%[src]) \n"
michael@0	71	"lbux $t8, $t2(%[src]) \n"
michael@0	72	"lbux $t9, $t3(%[src]) \n"
michael@0	73	"sll $t1, $t1, 16 \n"
michael@0	74	"sll $t9, $t9, 16 \n"
michael@0	75	"or $t0, $t0, $t1 \n"
michael@0	76	"or $t8, $t8, $t9 \n"
michael@0	77	"precr.qb.ph $s0, $t8, $t0 \n"
michael@0	78	"lbux $t0, $t4(%[src]) \n"
michael@0	79	"lbux $t1, $t5(%[src]) \n"
michael@0	80	"lbux $t8, $t6(%[src]) \n"
michael@0	81	"lbux $t9, $t7(%[src]) \n"
michael@0	82	"sll $t1, $t1, 16 \n"
michael@0	83	"sll $t9, $t9, 16 \n"
michael@0	84	"or $t0, $t0, $t1 \n"
michael@0	85	"or $t8, $t8, $t9 \n"
michael@0	86	"precr.qb.ph $s1, $t8, $t0 \n"
michael@0	87	"swr $s0, 0(%[dst]) \n"
michael@0	88	"swl $s0, 3(%[dst]) \n"
michael@0	89	"addiu %[width], -1 \n"
michael@0	90	"addiu %[src], 1 \n"
michael@0	91	"swr $s1, 4(%[dst]) \n"
michael@0	92	"swl $s1, 7(%[dst]) \n"
michael@0	93	"bnez %[width], 11b \n"
michael@0	94	"addu %[dst], %[dst], %[dst_stride] \n"
michael@0	95	"2: \n"
michael@0	96	".set pop \n"
michael@0	97	:[src] "+r" (src),
michael@0	98	[dst] "+r" (dst),
michael@0	99	[width] "+r" (width)
michael@0	100	:[src_stride] "r" (src_stride),
michael@0	101	[dst_stride] "r" (dst_stride)
michael@0	102	: "t0", "t1", "t2", "t3", "t4", "t5",
michael@0	103	"t6", "t7", "t8", "t9",
michael@0	104	"s0", "s1"
michael@0	105	);
michael@0	106	}
michael@0	107
michael@0	108	void TransposeWx8_FAST_MIPS_DSPR2(const uint8* src, int src_stride,
michael@0	109	uint8* dst, int dst_stride,
michael@0	110	int width) {
michael@0	111	__asm__ __volatile__ (
michael@0	112	".set noat \n"
michael@0	113	".set push \n"
michael@0	114	".set noreorder \n"
michael@0	115	"beqz %[width], 2f \n"
michael@0	116	" sll $t2, %[src_stride], 0x1 \n" // src_stride x 2
michael@0	117	"sll $t4, %[src_stride], 0x2 \n" // src_stride x 4
michael@0	118	"sll $t9, %[src_stride], 0x3 \n" // src_stride x 8
michael@0	119	"addu $t3, $t2, %[src_stride] \n"
michael@0	120	"addu $t5, $t4, %[src_stride] \n"
michael@0	121	"addu $t6, $t2, $t4 \n"
michael@0	122
michael@0	123	"srl $AT, %[width], 0x2 \n"
michael@0	124	"andi $t0, %[dst], 0x3 \n"
michael@0	125	"andi $t1, %[dst_stride], 0x3 \n"
michael@0	126	"or $t0, $t0, $t1 \n"
michael@0	127	"bnez $t0, 11f \n"
michael@0	128	" subu $t7, $t9, %[src_stride] \n"
michael@0	129	//dst + dst_stride word aligned
michael@0	130	"1: \n"
michael@0	131	"lw $t0, 0(%[src]) \n"
michael@0	132	"lwx $t1, %[src_stride](%[src]) \n"
michael@0	133	"lwx $t8, $t2(%[src]) \n"
michael@0	134	"lwx $t9, $t3(%[src]) \n"
michael@0	135
michael@0	136	// t0 = | 30 | 20 | 10 | 00 |
michael@0	137	// t1 = | 31 | 21 | 11 | 01 |
michael@0	138	// t8 = | 32 | 22 | 12 | 02 |
michael@0	139	// t9 = | 33 | 23 | 13 | 03 |
michael@0	140
michael@0	141	"precr.qb.ph $s0, $t1, $t0 \n"
michael@0	142	"precr.qb.ph $s1, $t9, $t8 \n"
michael@0	143	"precrq.qb.ph $s2, $t1, $t0 \n"
michael@0	144	"precrq.qb.ph $s3, $t9, $t8 \n"
michael@0	145
michael@0	146	// s0 = | 21 | 01 | 20 | 00 |
michael@0	147	// s1 = | 23 | 03 | 22 | 02 |
michael@0	148	// s2 = | 31 | 11 | 30 | 10 |
michael@0	149	// s3 = | 33 | 13 | 32 | 12 |
michael@0	150
michael@0	151	"precr.qb.ph $s4, $s1, $s0 \n"
michael@0	152	"precrq.qb.ph $s5, $s1, $s0 \n"
michael@0	153	"precr.qb.ph $s6, $s3, $s2 \n"
michael@0	154	"precrq.qb.ph $s7, $s3, $s2 \n"
michael@0	155
michael@0	156	// s4 = | 03 | 02 | 01 | 00 |
michael@0	157	// s5 = | 23 | 22 | 21 | 20 |
michael@0	158	// s6 = | 13 | 12 | 11 | 10 |
michael@0	159	// s7 = | 33 | 32 | 31 | 30 |
michael@0	160
michael@0	161	"lwx $t0, $t4(%[src]) \n"
michael@0	162	"lwx $t1, $t5(%[src]) \n"
michael@0	163	"lwx $t8, $t6(%[src]) \n"
michael@0	164	"lwx $t9, $t7(%[src]) \n"
michael@0	165
michael@0	166	// t0 = | 34 | 24 | 14 | 04 |
michael@0	167	// t1 = | 35 | 25 | 15 | 05 |
michael@0	168	// t8 = | 36 | 26 | 16 | 06 |
michael@0	169	// t9 = | 37 | 27 | 17 | 07 |
michael@0	170
michael@0	171	"precr.qb.ph $s0, $t1, $t0 \n"
michael@0	172	"precr.qb.ph $s1, $t9, $t8 \n"
michael@0	173	"precrq.qb.ph $s2, $t1, $t0 \n"
michael@0	174	"precrq.qb.ph $s3, $t9, $t8 \n"
michael@0	175
michael@0	176	// s0 = | 25 | 05 | 24 | 04 |
michael@0	177	// s1 = | 27 | 07 | 26 | 06 |
michael@0	178	// s2 = | 35 | 15 | 34 | 14 |
michael@0	179	// s3 = | 37 | 17 | 36 | 16 |
michael@0	180
michael@0	181	"precr.qb.ph $t0, $s1, $s0 \n"
michael@0	182	"precrq.qb.ph $t1, $s1, $s0 \n"
michael@0	183	"precr.qb.ph $t8, $s3, $s2 \n"
michael@0	184	"precrq.qb.ph $t9, $s3, $s2 \n"
michael@0	185
michael@0	186	// t0 = | 07 | 06 | 05 | 04 |
michael@0	187	// t1 = | 27 | 26 | 25 | 24 |
michael@0	188	// t8 = | 17 | 16 | 15 | 14 |
michael@0	189	// t9 = | 37 | 36 | 35 | 34 |
michael@0	190
michael@0	191	"addu $s0, %[dst], %[dst_stride] \n"
michael@0	192	"addu $s1, $s0, %[dst_stride] \n"
michael@0	193	"addu $s2, $s1, %[dst_stride] \n"
michael@0	194
michael@0	195	"sw $s4, 0(%[dst]) \n"
michael@0	196	"sw $t0, 4(%[dst]) \n"
michael@0	197	"sw $s6, 0($s0) \n"
michael@0	198	"sw $t8, 4($s0) \n"
michael@0	199	"sw $s5, 0($s1) \n"
michael@0	200	"sw $t1, 4($s1) \n"
michael@0	201	"sw $s7, 0($s2) \n"
michael@0	202	"sw $t9, 4($s2) \n"
michael@0	203
michael@0	204	"addiu $AT, -1 \n"
michael@0	205	"addiu %[src], 4 \n"
michael@0	206
michael@0	207	"bnez $AT, 1b \n"
michael@0	208	" addu %[dst], $s2, %[dst_stride] \n"
michael@0	209	"b 2f \n"
michael@0	210	//dst + dst_stride unaligned
michael@0	211	"11: \n"
michael@0	212	"lw $t0, 0(%[src]) \n"
michael@0	213	"lwx $t1, %[src_stride](%[src]) \n"
michael@0	214	"lwx $t8, $t2(%[src]) \n"
michael@0	215	"lwx $t9, $t3(%[src]) \n"
michael@0	216
michael@0	217	// t0 = | 30 | 20 | 10 | 00 |
michael@0	218	// t1 = | 31 | 21 | 11 | 01 |
michael@0	219	// t8 = | 32 | 22 | 12 | 02 |
michael@0	220	// t9 = | 33 | 23 | 13 | 03 |
michael@0	221
michael@0	222	"precr.qb.ph $s0, $t1, $t0 \n"
michael@0	223	"precr.qb.ph $s1, $t9, $t8 \n"
michael@0	224	"precrq.qb.ph $s2, $t1, $t0 \n"
michael@0	225	"precrq.qb.ph $s3, $t9, $t8 \n"
michael@0	226
michael@0	227	// s0 = | 21 | 01 | 20 | 00 |
michael@0	228	// s1 = | 23 | 03 | 22 | 02 |
michael@0	229	// s2 = | 31 | 11 | 30 | 10 |
michael@0	230	// s3 = | 33 | 13 | 32 | 12 |
michael@0	231
michael@0	232	"precr.qb.ph $s4, $s1, $s0 \n"
michael@0	233	"precrq.qb.ph $s5, $s1, $s0 \n"
michael@0	234	"precr.qb.ph $s6, $s3, $s2 \n"
michael@0	235	"precrq.qb.ph $s7, $s3, $s2 \n"
michael@0	236
michael@0	237	// s4 = | 03 | 02 | 01 | 00 |
michael@0	238	// s5 = | 23 | 22 | 21 | 20 |
michael@0	239	// s6 = | 13 | 12 | 11 | 10 |
michael@0	240	// s7 = | 33 | 32 | 31 | 30 |
michael@0	241
michael@0	242	"lwx $t0, $t4(%[src]) \n"
michael@0	243	"lwx $t1, $t5(%[src]) \n"
michael@0	244	"lwx $t8, $t6(%[src]) \n"
michael@0	245	"lwx $t9, $t7(%[src]) \n"
michael@0	246
michael@0	247	// t0 = | 34 | 24 | 14 | 04 |
michael@0	248	// t1 = | 35 | 25 | 15 | 05 |
michael@0	249	// t8 = | 36 | 26 | 16 | 06 |
michael@0	250	// t9 = | 37 | 27 | 17 | 07 |
michael@0	251
michael@0	252	"precr.qb.ph $s0, $t1, $t0 \n"
michael@0	253	"precr.qb.ph $s1, $t9, $t8 \n"
michael@0	254	"precrq.qb.ph $s2, $t1, $t0 \n"
michael@0	255	"precrq.qb.ph $s3, $t9, $t8 \n"
michael@0	256
michael@0	257	// s0 = | 25 | 05 | 24 | 04 |
michael@0	258	// s1 = | 27 | 07 | 26 | 06 |
michael@0	259	// s2 = | 35 | 15 | 34 | 14 |
michael@0	260	// s3 = | 37 | 17 | 36 | 16 |
michael@0	261
michael@0	262	"precr.qb.ph $t0, $s1, $s0 \n"
michael@0	263	"precrq.qb.ph $t1, $s1, $s0 \n"
michael@0	264	"precr.qb.ph $t8, $s3, $s2 \n"
michael@0	265	"precrq.qb.ph $t9, $s3, $s2 \n"
michael@0	266
michael@0	267	// t0 = | 07 | 06 | 05 | 04 |
michael@0	268	// t1 = | 27 | 26 | 25 | 24 |
michael@0	269	// t8 = | 17 | 16 | 15 | 14 |
michael@0	270	// t9 = | 37 | 36 | 35 | 34 |
michael@0	271
michael@0	272	"addu $s0, %[dst], %[dst_stride] \n"
michael@0	273	"addu $s1, $s0, %[dst_stride] \n"
michael@0	274	"addu $s2, $s1, %[dst_stride] \n"
michael@0	275
michael@0	276	"swr $s4, 0(%[dst]) \n"
michael@0	277	"swl $s4, 3(%[dst]) \n"
michael@0	278	"swr $t0, 4(%[dst]) \n"
michael@0	279	"swl $t0, 7(%[dst]) \n"
michael@0	280	"swr $s6, 0($s0) \n"
michael@0	281	"swl $s6, 3($s0) \n"
michael@0	282	"swr $t8, 4($s0) \n"
michael@0	283	"swl $t8, 7($s0) \n"
michael@0	284	"swr $s5, 0($s1) \n"
michael@0	285	"swl $s5, 3($s1) \n"
michael@0	286	"swr $t1, 4($s1) \n"
michael@0	287	"swl $t1, 7($s1) \n"
michael@0	288	"swr $s7, 0($s2) \n"
michael@0	289	"swl $s7, 3($s2) \n"
michael@0	290	"swr $t9, 4($s2) \n"
michael@0	291	"swl $t9, 7($s2) \n"
michael@0	292
michael@0	293	"addiu $AT, -1 \n"
michael@0	294	"addiu %[src], 4 \n"
michael@0	295
michael@0	296	"bnez $AT, 11b \n"
michael@0	297	" addu %[dst], $s2, %[dst_stride] \n"
michael@0	298	"2: \n"
michael@0	299	".set pop \n"
michael@0	300	".set at \n"
michael@0	301	:[src] "+r" (src),
michael@0	302	[dst] "+r" (dst),
michael@0	303	[width] "+r" (width)
michael@0	304	:[src_stride] "r" (src_stride),
michael@0	305	[dst_stride] "r" (dst_stride)
michael@0	306	: "t0", "t1", "t2", "t3", "t4", "t5",
michael@0	307	"t6", "t7", "t8", "t9",
michael@0	308	"s0", "s1", "s2", "s3", "s4",
michael@0	309	"s5", "s6", "s7"
michael@0	310	);
michael@0	311	}
michael@0	312
michael@0	313	void TransposeUVWx8_MIPS_DSPR2(const uint8* src, int src_stride,
michael@0	314	uint8* dst_a, int dst_stride_a,
michael@0	315	uint8* dst_b, int dst_stride_b,
michael@0	316	int width) {
michael@0	317	__asm__ __volatile__ (
michael@0	318	".set push \n"
michael@0	319	".set noreorder \n"
michael@0	320	"beqz %[width], 2f \n"
michael@0	321	" sll $t2, %[src_stride], 0x1 \n" // src_stride x 2
michael@0	322	"sll $t4, %[src_stride], 0x2 \n" // src_stride x 4
michael@0	323	"sll $t9, %[src_stride], 0x3 \n" // src_stride x 8
michael@0	324	"addu $t3, $t2, %[src_stride] \n"
michael@0	325	"addu $t5, $t4, %[src_stride] \n"
michael@0	326	"addu $t6, $t2, $t4 \n"
michael@0	327	"subu $t7, $t9, %[src_stride] \n"
michael@0	328	"srl $t1, %[width], 1 \n"
michael@0	329
michael@0	330	// check word aligment for dst_a, dst_b, dst_stride_a and dst_stride_b
michael@0	331	"andi $t0, %[dst_a], 0x3 \n"
michael@0	332	"andi $t8, %[dst_b], 0x3 \n"
michael@0	333	"or $t0, $t0, $t8 \n"
michael@0	334	"andi $t8, %[dst_stride_a], 0x3 \n"
michael@0	335	"andi $s5, %[dst_stride_b], 0x3 \n"
michael@0	336	"or $t8, $t8, $s5 \n"
michael@0	337	"or $t0, $t0, $t8 \n"
michael@0	338	"bnez $t0, 11f \n"
michael@0	339	" nop \n"
michael@0	340	// dst + dst_stride word aligned (both, a & b dst addresses)
michael@0	341	"1: \n"
michael@0	342	"lw $t0, 0(%[src]) \n" // |B0|A0|b0|a0|
michael@0	343	"lwx $t8, %[src_stride](%[src]) \n" // |B1|A1|b1|a1|
michael@0	344	"addu $s5, %[dst_a], %[dst_stride_a] \n"
michael@0	345	"lwx $t9, $t2(%[src]) \n" // |B2|A2|b2|a2|
michael@0	346	"lwx $s0, $t3(%[src]) \n" // |B3|A3|b3|a3|
michael@0	347	"addu $s6, %[dst_b], %[dst_stride_b] \n"
michael@0	348
michael@0	349	"precrq.ph.w $s1, $t8, $t0 \n" // |B1|A1|B0|A0|
michael@0	350	"precrq.ph.w $s2, $s0, $t9 \n" // |B3|A3|B2|A2|
michael@0	351	"precr.qb.ph $s3, $s2, $s1 \n" // |A3|A2|A1|A0|
michael@0	352	"precrq.qb.ph $s4, $s2, $s1 \n" // |B3|B2|B1|B0|
michael@0	353
michael@0	354	"sll $t0, $t0, 16 \n"
michael@0	355	"packrl.ph $s1, $t8, $t0 \n" // |b1|a1|b0|a0|
michael@0	356	"sll $t9, $t9, 16 \n"
michael@0	357	"packrl.ph $s2, $s0, $t9 \n" // |b3|a3|b2|a2|
michael@0	358
michael@0	359	"sw $s3, 0($s5) \n"
michael@0	360	"sw $s4, 0($s6) \n"
michael@0	361
michael@0	362	"precr.qb.ph $s3, $s2, $s1 \n" // |a3|a2|a1|a0|
michael@0	363	"precrq.qb.ph $s4, $s2, $s1 \n" // |b3|b2|b1|b0|
michael@0	364
michael@0	365	"lwx $t0, $t4(%[src]) \n" // |B4|A4|b4|a4|
michael@0	366	"lwx $t8, $t5(%[src]) \n" // |B5|A5|b5|a5|
michael@0	367	"lwx $t9, $t6(%[src]) \n" // |B6|A6|b6|a6|
michael@0	368	"lwx $s0, $t7(%[src]) \n" // |B7|A7|b7|a7|
michael@0	369	"sw $s3, 0(%[dst_a]) \n"
michael@0	370	"sw $s4, 0(%[dst_b]) \n"
michael@0	371
michael@0	372	"precrq.ph.w $s1, $t8, $t0 \n" // |B5|A5|B4|A4|
michael@0	373	"precrq.ph.w $s2, $s0, $t9 \n" // |B6|A6|B7|A7|
michael@0	374	"precr.qb.ph $s3, $s2, $s1 \n" // |A7|A6|A5|A4|
michael@0	375	"precrq.qb.ph $s4, $s2, $s1 \n" // |B7|B6|B5|B4|
michael@0	376
michael@0	377	"sll $t0, $t0, 16 \n"
michael@0	378	"packrl.ph $s1, $t8, $t0 \n" // |b5|a5|b4|a4|
michael@0	379	"sll $t9, $t9, 16 \n"
michael@0	380	"packrl.ph $s2, $s0, $t9 \n" // |b7|a7|b6|a6|
michael@0	381	"sw $s3, 4($s5) \n"
michael@0	382	"sw $s4, 4($s6) \n"
michael@0	383
michael@0	384	"precr.qb.ph $s3, $s2, $s1 \n" // |a7|a6|a5|a4|
michael@0	385	"precrq.qb.ph $s4, $s2, $s1 \n" // |b7|b6|b5|b4|
michael@0	386
michael@0	387	"addiu %[src], 4 \n"
michael@0	388	"addiu $t1, -1 \n"
michael@0	389	"sll $t0, %[dst_stride_a], 1 \n"
michael@0	390	"sll $t8, %[dst_stride_b], 1 \n"
michael@0	391	"sw $s3, 4(%[dst_a]) \n"
michael@0	392	"sw $s4, 4(%[dst_b]) \n"
michael@0	393	"addu %[dst_a], %[dst_a], $t0 \n"
michael@0	394	"bnez $t1, 1b \n"
michael@0	395	" addu %[dst_b], %[dst_b], $t8 \n"
michael@0	396	"b 2f \n"
michael@0	397	" nop \n"
michael@0	398
michael@0	399	// dst_a or dst_b or dst_stride_a or dst_stride_b not word aligned
michael@0	400	"11: \n"
michael@0	401	"lw $t0, 0(%[src]) \n" // |B0|A0|b0|a0|
michael@0	402	"lwx $t8, %[src_stride](%[src]) \n" // |B1|A1|b1|a1|
michael@0	403	"addu $s5, %[dst_a], %[dst_stride_a] \n"
michael@0	404	"lwx $t9, $t2(%[src]) \n" // |B2|A2|b2|a2|
michael@0	405	"lwx $s0, $t3(%[src]) \n" // |B3|A3|b3|a3|
michael@0	406	"addu $s6, %[dst_b], %[dst_stride_b] \n"
michael@0	407
michael@0	408	"precrq.ph.w $s1, $t8, $t0 \n" // |B1|A1|B0|A0|
michael@0	409	"precrq.ph.w $s2, $s0, $t9 \n" // |B3|A3|B2|A2|
michael@0	410	"precr.qb.ph $s3, $s2, $s1 \n" // |A3|A2|A1|A0|
michael@0	411	"precrq.qb.ph $s4, $s2, $s1 \n" // |B3|B2|B1|B0|
michael@0	412
michael@0	413	"sll $t0, $t0, 16 \n"
michael@0	414	"packrl.ph $s1, $t8, $t0 \n" // |b1|a1|b0|a0|
michael@0	415	"sll $t9, $t9, 16 \n"
michael@0	416	"packrl.ph $s2, $s0, $t9 \n" // |b3|a3|b2|a2|
michael@0	417
michael@0	418	"swr $s3, 0($s5) \n"
michael@0	419	"swl $s3, 3($s5) \n"
michael@0	420	"swr $s4, 0($s6) \n"
michael@0	421	"swl $s4, 3($s6) \n"
michael@0	422
michael@0	423	"precr.qb.ph $s3, $s2, $s1 \n" // |a3|a2|a1|a0|
michael@0	424	"precrq.qb.ph $s4, $s2, $s1 \n" // |b3|b2|b1|b0|
michael@0	425
michael@0	426	"lwx $t0, $t4(%[src]) \n" // |B4|A4|b4|a4|
michael@0	427	"lwx $t8, $t5(%[src]) \n" // |B5|A5|b5|a5|
michael@0	428	"lwx $t9, $t6(%[src]) \n" // |B6|A6|b6|a6|
michael@0	429	"lwx $s0, $t7(%[src]) \n" // |B7|A7|b7|a7|
michael@0	430	"swr $s3, 0(%[dst_a]) \n"
michael@0	431	"swl $s3, 3(%[dst_a]) \n"
michael@0	432	"swr $s4, 0(%[dst_b]) \n"
michael@0	433	"swl $s4, 3(%[dst_b]) \n"
michael@0	434
michael@0	435	"precrq.ph.w $s1, $t8, $t0 \n" // |B5|A5|B4|A4|
michael@0	436	"precrq.ph.w $s2, $s0, $t9 \n" // |B6|A6|B7|A7|
michael@0	437	"precr.qb.ph $s3, $s2, $s1 \n" // |A7|A6|A5|A4|
michael@0	438	"precrq.qb.ph $s4, $s2, $s1 \n" // |B7|B6|B5|B4|
michael@0	439
michael@0	440	"sll $t0, $t0, 16 \n"
michael@0	441	"packrl.ph $s1, $t8, $t0 \n" // |b5|a5|b4|a4|
michael@0	442	"sll $t9, $t9, 16 \n"
michael@0	443	"packrl.ph $s2, $s0, $t9 \n" // |b7|a7|b6|a6|
michael@0	444
michael@0	445	"swr $s3, 4($s5) \n"
michael@0	446	"swl $s3, 7($s5) \n"
michael@0	447	"swr $s4, 4($s6) \n"
michael@0	448	"swl $s4, 7($s6) \n"
michael@0	449
michael@0	450	"precr.qb.ph $s3, $s2, $s1 \n" // |a7|a6|a5|a4|
michael@0	451	"precrq.qb.ph $s4, $s2, $s1 \n" // |b7|b6|b5|b4|
michael@0	452
michael@0	453	"addiu %[src], 4 \n"
michael@0	454	"addiu $t1, -1 \n"
michael@0	455	"sll $t0, %[dst_stride_a], 1 \n"
michael@0	456	"sll $t8, %[dst_stride_b], 1 \n"
michael@0	457	"swr $s3, 4(%[dst_a]) \n"
michael@0	458	"swl $s3, 7(%[dst_a]) \n"
michael@0	459	"swr $s4, 4(%[dst_b]) \n"
michael@0	460	"swl $s4, 7(%[dst_b]) \n"
michael@0	461	"addu %[dst_a], %[dst_a], $t0 \n"
michael@0	462	"bnez $t1, 11b \n"
michael@0	463	" addu %[dst_b], %[dst_b], $t8 \n"
michael@0	464
michael@0	465	"2: \n"
michael@0	466	".set pop \n"
michael@0	467	: [src] "+r" (src),
michael@0	468	[dst_a] "+r" (dst_a),
michael@0	469	[dst_b] "+r" (dst_b),
michael@0	470	[width] "+r" (width),
michael@0	471	[src_stride] "+r" (src_stride)
michael@0	472	: [dst_stride_a] "r" (dst_stride_a),
michael@0	473	[dst_stride_b] "r" (dst_stride_b)
michael@0	474	: "t0", "t1", "t2", "t3", "t4", "t5",
michael@0	475	"t6", "t7", "t8", "t9",
michael@0	476	"s0", "s1", "s2", "s3",
michael@0	477	"s4", "s5", "s6"
michael@0	478	);
michael@0	479	}
michael@0	480
michael@0	481	#endif // defined(__mips_dsp) && (__mips_dsp_rev >= 2)
michael@0	482
michael@0	483	#ifdef __cplusplus
michael@0	484	} // extern "C"
michael@0	485	} // namespace libyuv
michael@0	486	#endif