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

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

media/libyuv/source/scale_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 2012 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/basic_types.h"
michael@0	12	#include "libyuv/row.h"
michael@0	13
michael@0	14	#ifdef __cplusplus
michael@0	15	namespace libyuv {
michael@0	16	extern "C" {
michael@0	17	#endif
michael@0	18
michael@0	19	// This module is for GCC MIPS DSPR2
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 ScaleRowDown2_MIPS_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
michael@0	24	uint8* dst, int dst_width) {
michael@0	25	__asm__ __volatile__(
michael@0	26	".set push \n"
michael@0	27	".set noreorder \n"
michael@0	28
michael@0	29	"srl $t9, %[dst_width], 4 \n" // iterations -> by 16
michael@0	30	"beqz $t9, 2f \n"
michael@0	31	" nop \n"
michael@0	32
michael@0	33	".p2align 2 \n"
michael@0	34	"1: \n"
michael@0	35	"lw $t0, 0(%[src_ptr]) \n" // |3|2|1|0|
michael@0	36	"lw $t1, 4(%[src_ptr]) \n" // |7|6|5|4|
michael@0	37	"lw $t2, 8(%[src_ptr]) \n" // |11|10|9|8|
michael@0	38	"lw $t3, 12(%[src_ptr]) \n" // |15|14|13|12|
michael@0	39	"lw $t4, 16(%[src_ptr]) \n" // |19|18|17|16|
michael@0	40	"lw $t5, 20(%[src_ptr]) \n" // |23|22|21|20|
michael@0	41	"lw $t6, 24(%[src_ptr]) \n" // |27|26|25|24|
michael@0	42	"lw $t7, 28(%[src_ptr]) \n" // |31|30|29|28|
michael@0	43	// TODO(fbarchard): Use odd pixels instead of even.
michael@0	44	"precr.qb.ph $t8, $t1, $t0 \n" // |6|4|2|0|
michael@0	45	"precr.qb.ph $t0, $t3, $t2 \n" // |14|12|10|8|
michael@0	46	"precr.qb.ph $t1, $t5, $t4 \n" // |22|20|18|16|
michael@0	47	"precr.qb.ph $t2, $t7, $t6 \n" // |30|28|26|24|
michael@0	48	"addiu %[src_ptr], %[src_ptr], 32 \n"
michael@0	49	"addiu $t9, $t9, -1 \n"
michael@0	50	"sw $t8, 0(%[dst]) \n"
michael@0	51	"sw $t0, 4(%[dst]) \n"
michael@0	52	"sw $t1, 8(%[dst]) \n"
michael@0	53	"sw $t2, 12(%[dst]) \n"
michael@0	54	"bgtz $t9, 1b \n"
michael@0	55	" addiu %[dst], %[dst], 16 \n"
michael@0	56
michael@0	57	"2: \n"
michael@0	58	"andi $t9, %[dst_width], 0xf \n" // residue
michael@0	59	"beqz $t9, 3f \n"
michael@0	60	" nop \n"
michael@0	61
michael@0	62	"21: \n"
michael@0	63	"lbu $t0, 0(%[src_ptr]) \n"
michael@0	64	"addiu %[src_ptr], %[src_ptr], 2 \n"
michael@0	65	"addiu $t9, $t9, -1 \n"
michael@0	66	"sb $t0, 0(%[dst]) \n"
michael@0	67	"bgtz $t9, 21b \n"
michael@0	68	" addiu %[dst], %[dst], 1 \n"
michael@0	69
michael@0	70	"3: \n"
michael@0	71	".set pop \n"
michael@0	72	: [src_ptr] "+r" (src_ptr),
michael@0	73	[dst] "+r" (dst)
michael@0	74	: [dst_width] "r" (dst_width)
michael@0	75	: "t0", "t1", "t2", "t3", "t4", "t5",
michael@0	76	"t6", "t7", "t8", "t9"
michael@0	77	);
michael@0	78	}
michael@0	79
michael@0	80	void ScaleRowDown2Box_MIPS_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
michael@0	81	uint8* dst, int dst_width) {
michael@0	82	const uint8* t = src_ptr + src_stride;
michael@0	83
michael@0	84	__asm__ __volatile__ (
michael@0	85	".set push \n"
michael@0	86	".set noreorder \n"
michael@0	87
michael@0	88	"srl $t9, %[dst_width], 3 \n" // iterations -> step 8
michael@0	89	"bltz $t9, 2f \n"
michael@0	90	" nop \n"
michael@0	91
michael@0	92	".p2align 2 \n"
michael@0	93	"1: \n"
michael@0	94	"lw $t0, 0(%[src_ptr]) \n" // |3|2|1|0|
michael@0	95	"lw $t1, 4(%[src_ptr]) \n" // |7|6|5|4|
michael@0	96	"lw $t2, 8(%[src_ptr]) \n" // |11|10|9|8|
michael@0	97	"lw $t3, 12(%[src_ptr]) \n" // |15|14|13|12|
michael@0	98	"lw $t4, 0(%[t]) \n" // |19|18|17|16|
michael@0	99	"lw $t5, 4(%[t]) \n" // |23|22|21|20|
michael@0	100	"lw $t6, 8(%[t]) \n" // |27|26|25|24|
michael@0	101	"lw $t7, 12(%[t]) \n" // |31|30|29|28|
michael@0	102	"addiu $t9, $t9, -1 \n"
michael@0	103	"srl $t8, $t0, 16 \n" // |X|X|3|2|
michael@0	104	"ins $t0, $t4, 16, 16 \n" // |17|16|1|0|
michael@0	105	"ins $t4, $t8, 0, 16 \n" // |19|18|3|2|
michael@0	106	"raddu.w.qb $t0, $t0 \n" // |17+16+1+0|
michael@0	107	"raddu.w.qb $t4, $t4 \n" // |19+18+3+2|
michael@0	108	"shra_r.w $t0, $t0, 2 \n" // |t0+2|>>2
michael@0	109	"shra_r.w $t4, $t4, 2 \n" // |t4+2|>>2
michael@0	110	"srl $t8, $t1, 16 \n" // |X|X|7|6|
michael@0	111	"ins $t1, $t5, 16, 16 \n" // |21|20|5|4|
michael@0	112	"ins $t5, $t8, 0, 16 \n" // |22|23|7|6|
michael@0	113	"raddu.w.qb $t1, $t1 \n" // |21+20+5+4|
michael@0	114	"raddu.w.qb $t5, $t5 \n" // |23+22+7+6|
michael@0	115	"shra_r.w $t1, $t1, 2 \n" // |t1+2|>>2
michael@0	116	"shra_r.w $t5, $t5, 2 \n" // |t5+2|>>2
michael@0	117	"srl $t8, $t2, 16 \n" // |X|X|11|10|
michael@0	118	"ins $t2, $t6, 16, 16 \n" // |25|24|9|8|
michael@0	119	"ins $t6, $t8, 0, 16 \n" // |27|26|11|10|
michael@0	120	"raddu.w.qb $t2, $t2 \n" // |25+24+9+8|
michael@0	121	"raddu.w.qb $t6, $t6 \n" // |27+26+11+10|
michael@0	122	"shra_r.w $t2, $t2, 2 \n" // |t2+2|>>2
michael@0	123	"shra_r.w $t6, $t6, 2 \n" // |t5+2|>>2
michael@0	124	"srl $t8, $t3, 16 \n" // |X|X|15|14|
michael@0	125	"ins $t3, $t7, 16, 16 \n" // |29|28|13|12|
michael@0	126	"ins $t7, $t8, 0, 16 \n" // |31|30|15|14|
michael@0	127	"raddu.w.qb $t3, $t3 \n" // |29+28+13+12|
michael@0	128	"raddu.w.qb $t7, $t7 \n" // |31+30+15+14|
michael@0	129	"shra_r.w $t3, $t3, 2 \n" // |t3+2|>>2
michael@0	130	"shra_r.w $t7, $t7, 2 \n" // |t7+2|>>2
michael@0	131	"addiu %[src_ptr], %[src_ptr], 16 \n"
michael@0	132	"addiu %[t], %[t], 16 \n"
michael@0	133	"sb $t0, 0(%[dst]) \n"
michael@0	134	"sb $t4, 1(%[dst]) \n"
michael@0	135	"sb $t1, 2(%[dst]) \n"
michael@0	136	"sb $t5, 3(%[dst]) \n"
michael@0	137	"sb $t2, 4(%[dst]) \n"
michael@0	138	"sb $t6, 5(%[dst]) \n"
michael@0	139	"sb $t3, 6(%[dst]) \n"
michael@0	140	"sb $t7, 7(%[dst]) \n"
michael@0	141	"bgtz $t9, 1b \n"
michael@0	142	" addiu %[dst], %[dst], 8 \n"
michael@0	143
michael@0	144	"2: \n"
michael@0	145	"andi $t9, %[dst_width], 0x7 \n" // x = residue
michael@0	146	"beqz $t9, 3f \n"
michael@0	147	" nop \n"
michael@0	148
michael@0	149	"21: \n"
michael@0	150	"lwr $t1, 0(%[src_ptr]) \n"
michael@0	151	"lwl $t1, 3(%[src_ptr]) \n"
michael@0	152	"lwr $t2, 0(%[t]) \n"
michael@0	153	"lwl $t2, 3(%[t]) \n"
michael@0	154	"srl $t8, $t1, 16 \n"
michael@0	155	"ins $t1, $t2, 16, 16 \n"
michael@0	156	"ins $t2, $t8, 0, 16 \n"
michael@0	157	"raddu.w.qb $t1, $t1 \n"
michael@0	158	"raddu.w.qb $t2, $t2 \n"
michael@0	159	"shra_r.w $t1, $t1, 2 \n"
michael@0	160	"shra_r.w $t2, $t2, 2 \n"
michael@0	161	"sb $t1, 0(%[dst]) \n"
michael@0	162	"sb $t2, 1(%[dst]) \n"
michael@0	163	"addiu %[src_ptr], %[src_ptr], 4 \n"
michael@0	164	"addiu $t9, $t9, -2 \n"
michael@0	165	"addiu %[t], %[t], 4 \n"
michael@0	166	"bgtz $t9, 21b \n"
michael@0	167	" addiu %[dst], %[dst], 2 \n"
michael@0	168
michael@0	169	"3: \n"
michael@0	170	".set pop \n"
michael@0	171
michael@0	172	: [src_ptr] "+r" (src_ptr),
michael@0	173	[dst] "+r" (dst), [t] "+r" (t)
michael@0	174	: [dst_width] "r" (dst_width)
michael@0	175	: "t0", "t1", "t2", "t3", "t4", "t5",
michael@0	176	"t6", "t7", "t8", "t9"
michael@0	177	);
michael@0	178	}
michael@0	179
michael@0	180	void ScaleRowDown4_MIPS_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
michael@0	181	uint8* dst, int dst_width) {
michael@0	182	__asm__ __volatile__ (
michael@0	183	".set push \n"
michael@0	184	".set noreorder \n"
michael@0	185
michael@0	186	"srl $t9, %[dst_width], 3 \n"
michael@0	187	"beqz $t9, 2f \n"
michael@0	188	" nop \n"
michael@0	189
michael@0	190	".p2align 2 \n"
michael@0	191	"1: \n"
michael@0	192	"lw $t1, 0(%[src_ptr]) \n" // |3|2|1|0|
michael@0	193	"lw $t2, 4(%[src_ptr]) \n" // |7|6|5|4|
michael@0	194	"lw $t3, 8(%[src_ptr]) \n" // |11|10|9|8|
michael@0	195	"lw $t4, 12(%[src_ptr]) \n" // |15|14|13|12|
michael@0	196	"lw $t5, 16(%[src_ptr]) \n" // |19|18|17|16|
michael@0	197	"lw $t6, 20(%[src_ptr]) \n" // |23|22|21|20|
michael@0	198	"lw $t7, 24(%[src_ptr]) \n" // |27|26|25|24|
michael@0	199	"lw $t8, 28(%[src_ptr]) \n" // |31|30|29|28|
michael@0	200	"precr.qb.ph $t1, $t2, $t1 \n" // |6|4|2|0|
michael@0	201	"precr.qb.ph $t2, $t4, $t3 \n" // |14|12|10|8|
michael@0	202	"precr.qb.ph $t5, $t6, $t5 \n" // |22|20|18|16|
michael@0	203	"precr.qb.ph $t6, $t8, $t7 \n" // |30|28|26|24|
michael@0	204	"precr.qb.ph $t1, $t2, $t1 \n" // |12|8|4|0|
michael@0	205	"precr.qb.ph $t5, $t6, $t5 \n" // |28|24|20|16|
michael@0	206	"addiu %[src_ptr], %[src_ptr], 32 \n"
michael@0	207	"addiu $t9, $t9, -1 \n"
michael@0	208	"sw $t1, 0(%[dst]) \n"
michael@0	209	"sw $t5, 4(%[dst]) \n"
michael@0	210	"bgtz $t9, 1b \n"
michael@0	211	" addiu %[dst], %[dst], 8 \n"
michael@0	212
michael@0	213	"2: \n"
michael@0	214	"andi $t9, %[dst_width], 7 \n" // residue
michael@0	215	"beqz $t9, 3f \n"
michael@0	216	" nop \n"
michael@0	217
michael@0	218	"21: \n"
michael@0	219	"lbu $t1, 0(%[src_ptr]) \n"
michael@0	220	"addiu %[src_ptr], %[src_ptr], 4 \n"
michael@0	221	"addiu $t9, $t9, -1 \n"
michael@0	222	"sb $t1, 0(%[dst]) \n"
michael@0	223	"bgtz $t9, 21b \n"
michael@0	224	" addiu %[dst], %[dst], 1 \n"
michael@0	225
michael@0	226	"3: \n"
michael@0	227	".set pop \n"
michael@0	228	: [src_ptr] "+r" (src_ptr),
michael@0	229	[dst] "+r" (dst)
michael@0	230	: [dst_width] "r" (dst_width)
michael@0	231	: "t1", "t2", "t3", "t4", "t5",
michael@0	232	"t6", "t7", "t8", "t9"
michael@0	233	);
michael@0	234	}
michael@0	235
michael@0	236	void ScaleRowDown4Box_MIPS_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
michael@0	237	uint8* dst, int dst_width) {
michael@0	238	intptr_t stride = src_stride;
michael@0	239	const uint8* s1 = src_ptr + stride;
michael@0	240	const uint8* s2 = s1 + stride;
michael@0	241	const uint8* s3 = s2 + stride;
michael@0	242
michael@0	243	__asm__ __volatile__ (
michael@0	244	".set push \n"
michael@0	245	".set noreorder \n"
michael@0	246
michael@0	247	"srl $t9, %[dst_width], 1 \n"
michael@0	248	"andi $t8, %[dst_width], 1 \n"
michael@0	249
michael@0	250	".p2align 2 \n"
michael@0	251	"1: \n"
michael@0	252	"lw $t0, 0(%[src_ptr]) \n" // |3|2|1|0|
michael@0	253	"lw $t1, 0(%[s1]) \n" // |7|6|5|4|
michael@0	254	"lw $t2, 0(%[s2]) \n" // |11|10|9|8|
michael@0	255	"lw $t3, 0(%[s3]) \n" // |15|14|13|12|
michael@0	256	"lw $t4, 4(%[src_ptr]) \n" // |19|18|17|16|
michael@0	257	"lw $t5, 4(%[s1]) \n" // |23|22|21|20|
michael@0	258	"lw $t6, 4(%[s2]) \n" // |27|26|25|24|
michael@0	259	"lw $t7, 4(%[s3]) \n" // |31|30|29|28|
michael@0	260	"raddu.w.qb $t0, $t0 \n" // |3 + 2 + 1 + 0|
michael@0	261	"raddu.w.qb $t1, $t1 \n" // |7 + 6 + 5 + 4|
michael@0	262	"raddu.w.qb $t2, $t2 \n" // |11 + 10 + 9 + 8|
michael@0	263	"raddu.w.qb $t3, $t3 \n" // |15 + 14 + 13 + 12|
michael@0	264	"raddu.w.qb $t4, $t4 \n" // |19 + 18 + 17 + 16|
michael@0	265	"raddu.w.qb $t5, $t5 \n" // |23 + 22 + 21 + 20|
michael@0	266	"raddu.w.qb $t6, $t6 \n" // |27 + 26 + 25 + 24|
michael@0	267	"raddu.w.qb $t7, $t7 \n" // |31 + 30 + 29 + 28|
michael@0	268	"add $t0, $t0, $t1 \n"
michael@0	269	"add $t1, $t2, $t3 \n"
michael@0	270	"add $t0, $t0, $t1 \n"
michael@0	271	"add $t4, $t4, $t5 \n"
michael@0	272	"add $t6, $t6, $t7 \n"
michael@0	273	"add $t4, $t4, $t6 \n"
michael@0	274	"shra_r.w $t0, $t0, 4 \n"
michael@0	275	"shra_r.w $t4, $t4, 4 \n"
michael@0	276	"sb $t0, 0(%[dst]) \n"
michael@0	277	"sb $t4, 1(%[dst]) \n"
michael@0	278	"addiu %[src_ptr], %[src_ptr], 8 \n"
michael@0	279	"addiu %[s1], %[s1], 8 \n"
michael@0	280	"addiu %[s2], %[s2], 8 \n"
michael@0	281	"addiu %[s3], %[s3], 8 \n"
michael@0	282	"addiu $t9, $t9, -1 \n"
michael@0	283	"bgtz $t9, 1b \n"
michael@0	284	" addiu %[dst], %[dst], 2 \n"
michael@0	285	"beqz $t8, 2f \n"
michael@0	286	" nop \n"
michael@0	287
michael@0	288	"lw $t0, 0(%[src_ptr]) \n" // |3|2|1|0|
michael@0	289	"lw $t1, 0(%[s1]) \n" // |7|6|5|4|
michael@0	290	"lw $t2, 0(%[s2]) \n" // |11|10|9|8|
michael@0	291	"lw $t3, 0(%[s3]) \n" // |15|14|13|12|
michael@0	292	"raddu.w.qb $t0, $t0 \n" // |3 + 2 + 1 + 0|
michael@0	293	"raddu.w.qb $t1, $t1 \n" // |7 + 6 + 5 + 4|
michael@0	294	"raddu.w.qb $t2, $t2 \n" // |11 + 10 + 9 + 8|
michael@0	295	"raddu.w.qb $t3, $t3 \n" // |15 + 14 + 13 + 12|
michael@0	296	"add $t0, $t0, $t1 \n"
michael@0	297	"add $t1, $t2, $t3 \n"
michael@0	298	"add $t0, $t0, $t1 \n"
michael@0	299	"shra_r.w $t0, $t0, 4 \n"
michael@0	300	"sb $t0, 0(%[dst]) \n"
michael@0	301
michael@0	302	"2: \n"
michael@0	303	".set pop \n"
michael@0	304
michael@0	305	: [src_ptr] "+r" (src_ptr),
michael@0	306	[dst] "+r" (dst),
michael@0	307	[s1] "+r" (s1),
michael@0	308	[s2] "+r" (s2),
michael@0	309	[s3] "+r" (s3)
michael@0	310	: [dst_width] "r" (dst_width)
michael@0	311	: "t0", "t1", "t2", "t3", "t4", "t5",
michael@0	312	"t6","t7", "t8", "t9"
michael@0	313	);
michael@0	314	}
michael@0	315
michael@0	316	void ScaleRowDown34_MIPS_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
michael@0	317	uint8* dst, int dst_width) {
michael@0	318	__asm__ __volatile__ (
michael@0	319	".set push \n"
michael@0	320	".set noreorder \n"
michael@0	321	".p2align 2 \n"
michael@0	322	"1: \n"
michael@0	323	"lw $t1, 0(%[src_ptr]) \n" // |3|2|1|0|
michael@0	324	"lw $t2, 4(%[src_ptr]) \n" // |7|6|5|4|
michael@0	325	"lw $t3, 8(%[src_ptr]) \n" // |11|10|9|8|
michael@0	326	"lw $t4, 12(%[src_ptr]) \n" // |15|14|13|12|
michael@0	327	"lw $t5, 16(%[src_ptr]) \n" // |19|18|17|16|
michael@0	328	"lw $t6, 20(%[src_ptr]) \n" // |23|22|21|20|
michael@0	329	"lw $t7, 24(%[src_ptr]) \n" // |27|26|25|24|
michael@0	330	"lw $t8, 28(%[src_ptr]) \n" // |31|30|29|28|
michael@0	331	"precrq.qb.ph $t0, $t2, $t4 \n" // |7|5|15|13|
michael@0	332	"precrq.qb.ph $t9, $t6, $t8 \n" // |23|21|31|30|
michael@0	333	"addiu %[dst_width], %[dst_width], -24 \n"
michael@0	334	"ins $t1, $t1, 8, 16 \n" // |3|1|0|X|
michael@0	335	"ins $t4, $t0, 8, 16 \n" // |X|15|13|12|
michael@0	336	"ins $t5, $t5, 8, 16 \n" // |19|17|16|X|
michael@0	337	"ins $t8, $t9, 8, 16 \n" // |X|31|29|28|
michael@0	338	"addiu %[src_ptr], %[src_ptr], 32 \n"
michael@0	339	"packrl.ph $t0, $t3, $t0 \n" // |9|8|7|5|
michael@0	340	"packrl.ph $t9, $t7, $t9 \n" // |25|24|23|21|
michael@0	341	"prepend $t1, $t2, 8 \n" // |4|3|1|0|
michael@0	342	"prepend $t3, $t4, 24 \n" // |15|13|12|11|
michael@0	343	"prepend $t5, $t6, 8 \n" // |20|19|17|16|
michael@0	344	"prepend $t7, $t8, 24 \n" // |31|29|28|27|
michael@0	345	"sw $t1, 0(%[dst]) \n"
michael@0	346	"sw $t0, 4(%[dst]) \n"
michael@0	347	"sw $t3, 8(%[dst]) \n"
michael@0	348	"sw $t5, 12(%[dst]) \n"
michael@0	349	"sw $t9, 16(%[dst]) \n"
michael@0	350	"sw $t7, 20(%[dst]) \n"
michael@0	351	"bnez %[dst_width], 1b \n"
michael@0	352	" addiu %[dst], %[dst], 24 \n"
michael@0	353	".set pop \n"
michael@0	354	: [src_ptr] "+r" (src_ptr),
michael@0	355	[dst] "+r" (dst),
michael@0	356	[dst_width] "+r" (dst_width)
michael@0	357	:
michael@0	358	: "t0", "t1", "t2", "t3", "t4", "t5",
michael@0	359	"t6","t7", "t8", "t9"
michael@0	360	);
michael@0	361	}
michael@0	362
michael@0	363	void ScaleRowDown34_0_Box_MIPS_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
michael@0	364	uint8* d, int dst_width) {
michael@0	365	__asm__ __volatile__ (
michael@0	366	".set push \n"
michael@0	367	".set noreorder \n"
michael@0	368	"repl.ph $t3, 3 \n" // 0x00030003
michael@0	369
michael@0	370	".p2align 2 \n"
michael@0	371	"1: \n"
michael@0	372	"lw $t0, 0(%[src_ptr]) \n" // |S3|S2|S1|S0|
michael@0	373	"lwx $t1, %[src_stride](%[src_ptr]) \n" // |T3|T2|T1|T0|
michael@0	374	"rotr $t2, $t0, 8 \n" // |S0|S3|S2|S1|
michael@0	375	"rotr $t6, $t1, 8 \n" // |T0|T3|T2|T1|
michael@0	376	"muleu_s.ph.qbl $t4, $t2, $t3 \n" // |S0*3|S3*3|
michael@0	377	"muleu_s.ph.qbl $t5, $t6, $t3 \n" // |T0*3|T3*3|
michael@0	378	"andi $t0, $t2, 0xFFFF \n" // |0|0|S2|S1|
michael@0	379	"andi $t1, $t6, 0xFFFF \n" // |0|0|T2|T1|
michael@0	380	"raddu.w.qb $t0, $t0 \n"
michael@0	381	"raddu.w.qb $t1, $t1 \n"
michael@0	382	"shra_r.w $t0, $t0, 1 \n"
michael@0	383	"shra_r.w $t1, $t1, 1 \n"
michael@0	384	"preceu.ph.qbr $t2, $t2 \n" // |0|S2|0|S1|
michael@0	385	"preceu.ph.qbr $t6, $t6 \n" // |0|T2|0|T1|
michael@0	386	"rotr $t2, $t2, 16 \n" // |0|S1|0|S2|
michael@0	387	"rotr $t6, $t6, 16 \n" // |0|T1|0|T2|
michael@0	388	"addu.ph $t2, $t2, $t4 \n"
michael@0	389	"addu.ph $t6, $t6, $t5 \n"
michael@0	390	"sll $t5, $t0, 1 \n"
michael@0	391	"add $t0, $t5, $t0 \n"
michael@0	392	"shra_r.ph $t2, $t2, 2 \n"
michael@0	393	"shra_r.ph $t6, $t6, 2 \n"
michael@0	394	"shll.ph $t4, $t2, 1 \n"
michael@0	395	"addq.ph $t4, $t4, $t2 \n"
michael@0	396	"addu $t0, $t0, $t1 \n"
michael@0	397	"addiu %[src_ptr], %[src_ptr], 4 \n"
michael@0	398	"shra_r.w $t0, $t0, 2 \n"
michael@0	399	"addu.ph $t6, $t6, $t4 \n"
michael@0	400	"shra_r.ph $t6, $t6, 2 \n"
michael@0	401	"srl $t1, $t6, 16 \n"
michael@0	402	"addiu %[dst_width], %[dst_width], -3 \n"
michael@0	403	"sb $t1, 0(%[d]) \n"
michael@0	404	"sb $t0, 1(%[d]) \n"
michael@0	405	"sb $t6, 2(%[d]) \n"
michael@0	406	"bgtz %[dst_width], 1b \n"
michael@0	407	" addiu %[d], %[d], 3 \n"
michael@0	408	"3: \n"
michael@0	409	".set pop \n"
michael@0	410	: [src_ptr] "+r" (src_ptr),
michael@0	411	[src_stride] "+r" (src_stride),
michael@0	412	[d] "+r" (d),
michael@0	413	[dst_width] "+r" (dst_width)
michael@0	414	:
michael@0	415	: "t0", "t1", "t2", "t3",
michael@0	416	"t4", "t5", "t6"
michael@0	417	);
michael@0	418	}
michael@0	419
michael@0	420	void ScaleRowDown34_1_Box_MIPS_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
michael@0	421	uint8* d, int dst_width) {
michael@0	422	__asm__ __volatile__ (
michael@0	423	".set push \n"
michael@0	424	".set noreorder \n"
michael@0	425	"repl.ph $t2, 3 \n" // 0x00030003
michael@0	426
michael@0	427	".p2align 2 \n"
michael@0	428	"1: \n"
michael@0	429	"lw $t0, 0(%[src_ptr]) \n" // |S3|S2|S1|S0|
michael@0	430	"lwx $t1, %[src_stride](%[src_ptr]) \n" // |T3|T2|T1|T0|
michael@0	431	"rotr $t4, $t0, 8 \n" // |S0|S3|S2|S1|
michael@0	432	"rotr $t6, $t1, 8 \n" // |T0|T3|T2|T1|
michael@0	433	"muleu_s.ph.qbl $t3, $t4, $t2 \n" // |S0*3|S3*3|
michael@0	434	"muleu_s.ph.qbl $t5, $t6, $t2 \n" // |T0*3|T3*3|
michael@0	435	"andi $t0, $t4, 0xFFFF \n" // |0|0|S2|S1|
michael@0	436	"andi $t1, $t6, 0xFFFF \n" // |0|0|T2|T1|
michael@0	437	"raddu.w.qb $t0, $t0 \n"
michael@0	438	"raddu.w.qb $t1, $t1 \n"
michael@0	439	"shra_r.w $t0, $t0, 1 \n"
michael@0	440	"shra_r.w $t1, $t1, 1 \n"
michael@0	441	"preceu.ph.qbr $t4, $t4 \n" // |0|S2|0|S1|
michael@0	442	"preceu.ph.qbr $t6, $t6 \n" // |0|T2|0|T1|
michael@0	443	"rotr $t4, $t4, 16 \n" // |0|S1|0|S2|
michael@0	444	"rotr $t6, $t6, 16 \n" // |0|T1|0|T2|
michael@0	445	"addu.ph $t4, $t4, $t3 \n"
michael@0	446	"addu.ph $t6, $t6, $t5 \n"
michael@0	447	"shra_r.ph $t6, $t6, 2 \n"
michael@0	448	"shra_r.ph $t4, $t4, 2 \n"
michael@0	449	"addu.ph $t6, $t6, $t4 \n"
michael@0	450	"addiu %[src_ptr], %[src_ptr], 4 \n"
michael@0	451	"shra_r.ph $t6, $t6, 1 \n"
michael@0	452	"addu $t0, $t0, $t1 \n"
michael@0	453	"addiu %[dst_width], %[dst_width], -3 \n"
michael@0	454	"shra_r.w $t0, $t0, 1 \n"
michael@0	455	"srl $t1, $t6, 16 \n"
michael@0	456	"sb $t1, 0(%[d]) \n"
michael@0	457	"sb $t0, 1(%[d]) \n"
michael@0	458	"sb $t6, 2(%[d]) \n"
michael@0	459	"bgtz %[dst_width], 1b \n"
michael@0	460	" addiu %[d], %[d], 3 \n"
michael@0	461	"3: \n"
michael@0	462	".set pop \n"
michael@0	463	: [src_ptr] "+r" (src_ptr),
michael@0	464	[src_stride] "+r" (src_stride),
michael@0	465	[d] "+r" (d),
michael@0	466	[dst_width] "+r" (dst_width)
michael@0	467	:
michael@0	468	: "t0", "t1", "t2", "t3",
michael@0	469	"t4", "t5", "t6"
michael@0	470	);
michael@0	471	}
michael@0	472
michael@0	473	void ScaleRowDown38_MIPS_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
michael@0	474	uint8* dst, int dst_width) {
michael@0	475	__asm__ __volatile__ (
michael@0	476	".set push \n"
michael@0	477	".set noreorder \n"
michael@0	478
michael@0	479	".p2align 2 \n"
michael@0	480	"1: \n"
michael@0	481	"lw $t0, 0(%[src_ptr]) \n" // |3|2|1|0|
michael@0	482	"lw $t1, 4(%[src_ptr]) \n" // |7|6|5|4|
michael@0	483	"lw $t2, 8(%[src_ptr]) \n" // |11|10|9|8|
michael@0	484	"lw $t3, 12(%[src_ptr]) \n" // |15|14|13|12|
michael@0	485	"lw $t4, 16(%[src_ptr]) \n" // |19|18|17|16|
michael@0	486	"lw $t5, 20(%[src_ptr]) \n" // |23|22|21|20|
michael@0	487	"lw $t6, 24(%[src_ptr]) \n" // |27|26|25|24|
michael@0	488	"lw $t7, 28(%[src_ptr]) \n" // |31|30|29|28|
michael@0	489	"wsbh $t0, $t0 \n" // |2|3|0|1|
michael@0	490	"wsbh $t6, $t6 \n" // |26|27|24|25|
michael@0	491	"srl $t0, $t0, 8 \n" // |X|2|3|0|
michael@0	492	"srl $t3, $t3, 16 \n" // |X|X|15|14|
michael@0	493	"srl $t5, $t5, 16 \n" // |X|X|23|22|
michael@0	494	"srl $t7, $t7, 16 \n" // |X|X|31|30|
michael@0	495	"ins $t1, $t2, 24, 8 \n" // |8|6|5|4|
michael@0	496	"ins $t6, $t5, 0, 8 \n" // |26|27|24|22|
michael@0	497	"ins $t1, $t0, 0, 16 \n" // |8|6|3|0|
michael@0	498	"ins $t6, $t7, 24, 8 \n" // |30|27|24|22|
michael@0	499	"prepend $t2, $t3, 24 \n" // |X|15|14|11|
michael@0	500	"ins $t4, $t4, 16, 8 \n" // |19|16|17|X|
michael@0	501	"ins $t4, $t2, 0, 16 \n" // |19|16|14|11|
michael@0	502	"addiu %[src_ptr], %[src_ptr], 32 \n"
michael@0	503	"addiu %[dst_width], %[dst_width], -12 \n"
michael@0	504	"addiu $t8,%[dst_width], -12 \n"
michael@0	505	"sw $t1, 0(%[dst]) \n"
michael@0	506	"sw $t4, 4(%[dst]) \n"
michael@0	507	"sw $t6, 8(%[dst]) \n"
michael@0	508	"bgez $t8, 1b \n"
michael@0	509	" addiu %[dst], %[dst], 12 \n"
michael@0	510	".set pop \n"
michael@0	511	: [src_ptr] "+r" (src_ptr),
michael@0	512	[dst] "+r" (dst),
michael@0	513	[dst_width] "+r" (dst_width)
michael@0	514	:
michael@0	515	: "t0", "t1", "t2", "t3", "t4",
michael@0	516	"t5", "t6", "t7", "t8"
michael@0	517	);
michael@0	518	}
michael@0	519
michael@0	520	void ScaleRowDown38_2_Box_MIPS_DSPR2(const uint8* src_ptr, ptrdiff_t src_stride,
michael@0	521	uint8* dst_ptr, int dst_width) {
michael@0	522	intptr_t stride = src_stride;
michael@0	523	const uint8* t = src_ptr + stride;
michael@0	524	const int c = 0x2AAA;
michael@0	525
michael@0	526	__asm__ __volatile__ (
michael@0	527	".set push \n"
michael@0	528	".set noreorder \n"
michael@0	529
michael@0	530	".p2align 2 \n"
michael@0	531	"1: \n"
michael@0	532	"lw $t0, 0(%[src_ptr]) \n" // |S3|S2|S1|S0|
michael@0	533	"lw $t1, 4(%[src_ptr]) \n" // |S7|S6|S5|S4|
michael@0	534	"lw $t2, 0(%[t]) \n" // |T3|T2|T1|T0|
michael@0	535	"lw $t3, 4(%[t]) \n" // |T7|T6|T5|T4|
michael@0	536	"rotr $t1, $t1, 16 \n" // |S5|S4|S7|S6|
michael@0	537	"packrl.ph $t4, $t1, $t3 \n" // |S7|S6|T7|T6|
michael@0	538	"packrl.ph $t5, $t3, $t1 \n" // |T5|T4|S5|S4|
michael@0	539	"raddu.w.qb $t4, $t4 \n" // S7+S6+T7+T6
michael@0	540	"raddu.w.qb $t5, $t5 \n" // T5+T4+S5+S4
michael@0	541	"precrq.qb.ph $t6, $t0, $t2 \n" // |S3|S1|T3|T1|
michael@0	542	"precrq.qb.ph $t6, $t6, $t6 \n" // |S3|T3|S3|T3|
michael@0	543	"srl $t4, $t4, 2 \n" // t4 / 4
michael@0	544	"srl $t6, $t6, 16 \n" // |0|0|S3|T3|
michael@0	545	"raddu.w.qb $t6, $t6 \n" // 0+0+S3+T3
michael@0	546	"addu $t6, $t5, $t6 \n"
michael@0	547	"mul $t6, $t6, %[c] \n" // t6 * 0x2AAA
michael@0	548	"sll $t0, $t0, 8 \n" // |S2|S1|S0|0|
michael@0	549	"sll $t2, $t2, 8 \n" // |T2|T1|T0|0|
michael@0	550	"raddu.w.qb $t0, $t0 \n" // S2+S1+S0+0
michael@0	551	"raddu.w.qb $t2, $t2 \n" // T2+T1+T0+0
michael@0	552	"addu $t0, $t0, $t2 \n"
michael@0	553	"mul $t0, $t0, %[c] \n" // t0 * 0x2AAA
michael@0	554	"addiu %[src_ptr], %[src_ptr], 8 \n"
michael@0	555	"addiu %[t], %[t], 8 \n"
michael@0	556	"addiu %[dst_width], %[dst_width], -3 \n"
michael@0	557	"addiu %[dst_ptr], %[dst_ptr], 3 \n"
michael@0	558	"srl $t6, $t6, 16 \n"
michael@0	559	"srl $t0, $t0, 16 \n"
michael@0	560	"sb $t4, -1(%[dst_ptr]) \n"
michael@0	561	"sb $t6, -2(%[dst_ptr]) \n"
michael@0	562	"bgtz %[dst_width], 1b \n"
michael@0	563	" sb $t0, -3(%[dst_ptr]) \n"
michael@0	564	".set pop \n"
michael@0	565	: [src_ptr] "+r" (src_ptr),
michael@0	566	[dst_ptr] "+r" (dst_ptr),
michael@0	567	[t] "+r" (t),
michael@0	568	[dst_width] "+r" (dst_width)
michael@0	569	: [c] "r" (c)
michael@0	570	: "t0", "t1", "t2", "t3", "t4", "t5", "t6"
michael@0	571	);
michael@0	572	}
michael@0	573
michael@0	574	void ScaleRowDown38_3_Box_MIPS_DSPR2(const uint8* src_ptr,
michael@0	575	ptrdiff_t src_stride,
michael@0	576	uint8* dst_ptr, int dst_width) {
michael@0	577	intptr_t stride = src_stride;
michael@0	578	const uint8* s1 = src_ptr + stride;
michael@0	579	stride += stride;
michael@0	580	const uint8* s2 = src_ptr + stride;
michael@0	581	const int c1 = 0x1C71;
michael@0	582	const int c2 = 0x2AAA;
michael@0	583
michael@0	584	__asm__ __volatile__ (
michael@0	585	".set push \n"
michael@0	586	".set noreorder \n"
michael@0	587
michael@0	588	".p2align 2 \n"
michael@0	589	"1: \n"
michael@0	590	"lw $t0, 0(%[src_ptr]) \n" // |S3|S2|S1|S0|
michael@0	591	"lw $t1, 4(%[src_ptr]) \n" // |S7|S6|S5|S4|
michael@0	592	"lw $t2, 0(%[s1]) \n" // |T3|T2|T1|T0|
michael@0	593	"lw $t3, 4(%[s1]) \n" // |T7|T6|T5|T4|
michael@0	594	"lw $t4, 0(%[s2]) \n" // |R3|R2|R1|R0|
michael@0	595	"lw $t5, 4(%[s2]) \n" // |R7|R6|R5|R4|
michael@0	596	"rotr $t1, $t1, 16 \n" // |S5|S4|S7|S6|
michael@0	597	"packrl.ph $t6, $t1, $t3 \n" // |S7|S6|T7|T6|
michael@0	598	"raddu.w.qb $t6, $t6 \n" // S7+S6+T7+T6
michael@0	599	"packrl.ph $t7, $t3, $t1 \n" // |T5|T4|S5|S4|
michael@0	600	"raddu.w.qb $t7, $t7 \n" // T5+T4+S5+S4
michael@0	601	"sll $t8, $t5, 16 \n" // |R5|R4|0|0|
michael@0	602	"raddu.w.qb $t8, $t8 \n" // R5+R4
michael@0	603	"addu $t7, $t7, $t8 \n"
michael@0	604	"srl $t8, $t5, 16 \n" // |0|0|R7|R6|
michael@0	605	"raddu.w.qb $t8, $t8 \n" // R7 + R6
michael@0	606	"addu $t6, $t6, $t8 \n"
michael@0	607	"mul $t6, $t6, %[c2] \n" // t6 * 0x2AAA
michael@0	608	"precrq.qb.ph $t8, $t0, $t2 \n" // |S3|S1|T3|T1|
michael@0	609	"precrq.qb.ph $t8, $t8, $t4 \n" // |S3|T3|R3|R1|
michael@0	610	"srl $t8, $t8, 8 \n" // |0|S3|T3|R3|
michael@0	611	"raddu.w.qb $t8, $t8 \n" // S3 + T3 + R3
michael@0	612	"addu $t7, $t7, $t8 \n"
michael@0	613	"mul $t7, $t7, %[c1] \n" // t7 * 0x1C71
michael@0	614	"sll $t0, $t0, 8 \n" // |S2|S1|S0|0|
michael@0	615	"sll $t2, $t2, 8 \n" // |T2|T1|T0|0|
michael@0	616	"sll $t4, $t4, 8 \n" // |R2|R1|R0|0|
michael@0	617	"raddu.w.qb $t0, $t0 \n"
michael@0	618	"raddu.w.qb $t2, $t2 \n"
michael@0	619	"raddu.w.qb $t4, $t4 \n"
michael@0	620	"addu $t0, $t0, $t2 \n"
michael@0	621	"addu $t0, $t0, $t4 \n"
michael@0	622	"mul $t0, $t0, %[c1] \n" // t0 * 0x1C71
michael@0	623	"addiu %[src_ptr], %[src_ptr], 8 \n"
michael@0	624	"addiu %[s1], %[s1], 8 \n"
michael@0	625	"addiu %[s2], %[s2], 8 \n"
michael@0	626	"addiu %[dst_width], %[dst_width], -3 \n"
michael@0	627	"addiu %[dst_ptr], %[dst_ptr], 3 \n"
michael@0	628	"srl $t6, $t6, 16 \n"
michael@0	629	"srl $t7, $t7, 16 \n"
michael@0	630	"srl $t0, $t0, 16 \n"
michael@0	631	"sb $t6, -1(%[dst_ptr]) \n"
michael@0	632	"sb $t7, -2(%[dst_ptr]) \n"
michael@0	633	"bgtz %[dst_width], 1b \n"
michael@0	634	" sb $t0, -3(%[dst_ptr]) \n"
michael@0	635	".set pop \n"
michael@0	636	: [src_ptr] "+r" (src_ptr),
michael@0	637	[dst_ptr] "+r" (dst_ptr),
michael@0	638	[s1] "+r" (s1),
michael@0	639	[s2] "+r" (s2),
michael@0	640	[dst_width] "+r" (dst_width)
michael@0	641	: [c1] "r" (c1), [c2] "r" (c2)
michael@0	642	: "t0", "t1", "t2", "t3", "t4",
michael@0	643	"t5", "t6", "t7", "t8"
michael@0	644	);
michael@0	645	}
michael@0	646
michael@0	647	#endif // defined(__mips_dsp) && (__mips_dsp_rev >= 2)
michael@0	648
michael@0	649	#ifdef __cplusplus
michael@0	650	} // extern "C"
michael@0	651	} // namespace libyuv
michael@0	652	#endif
michael@0	653