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

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

media/libyuv/source/rotate_neon.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_NEON) && defined(__ARM_NEON__)
michael@0	21	static uvec8 kVTbl4x4Transpose =
michael@0	22	{ 0, 4, 8, 12, 1, 5, 9, 13, 2, 6, 10, 14, 3, 7, 11, 15 };
michael@0	23
michael@0	24	void TransposeWx8_NEON(const uint8* src, int src_stride,
michael@0	25	uint8* dst, int dst_stride,
michael@0	26	int width) {
michael@0	27	asm volatile (
michael@0	28	// loops are on blocks of 8. loop will stop when
michael@0	29	// counter gets to or below 0. starting the counter
michael@0	30	// at w-8 allow for this
michael@0	31	"sub %4, #8 \n"
michael@0	32
michael@0	33	// handle 8x8 blocks. this should be the majority of the plane
michael@0	34	".p2align 2 \n"
michael@0	35	"1: \n"
michael@0	36	"mov r9, %0 \n"
michael@0	37
michael@0	38	"vld1.8 {d0}, [r9], %1 \n"
michael@0	39	"vld1.8 {d1}, [r9], %1 \n"
michael@0	40	"vld1.8 {d2}, [r9], %1 \n"
michael@0	41	"vld1.8 {d3}, [r9], %1 \n"
michael@0	42	"vld1.8 {d4}, [r9], %1 \n"
michael@0	43	"vld1.8 {d5}, [r9], %1 \n"
michael@0	44	"vld1.8 {d6}, [r9], %1 \n"
michael@0	45	"vld1.8 {d7}, [r9] \n"
michael@0	46
michael@0	47	"vtrn.8 d1, d0 \n"
michael@0	48	"vtrn.8 d3, d2 \n"
michael@0	49	"vtrn.8 d5, d4 \n"
michael@0	50	"vtrn.8 d7, d6 \n"
michael@0	51
michael@0	52	"vtrn.16 d1, d3 \n"
michael@0	53	"vtrn.16 d0, d2 \n"
michael@0	54	"vtrn.16 d5, d7 \n"
michael@0	55	"vtrn.16 d4, d6 \n"
michael@0	56
michael@0	57	"vtrn.32 d1, d5 \n"
michael@0	58	"vtrn.32 d0, d4 \n"
michael@0	59	"vtrn.32 d3, d7 \n"
michael@0	60	"vtrn.32 d2, d6 \n"
michael@0	61
michael@0	62	"vrev16.8 q0, q0 \n"
michael@0	63	"vrev16.8 q1, q1 \n"
michael@0	64	"vrev16.8 q2, q2 \n"
michael@0	65	"vrev16.8 q3, q3 \n"
michael@0	66
michael@0	67	"mov r9, %2 \n"
michael@0	68
michael@0	69	"vst1.8 {d1}, [r9], %3 \n"
michael@0	70	"vst1.8 {d0}, [r9], %3 \n"
michael@0	71	"vst1.8 {d3}, [r9], %3 \n"
michael@0	72	"vst1.8 {d2}, [r9], %3 \n"
michael@0	73	"vst1.8 {d5}, [r9], %3 \n"
michael@0	74	"vst1.8 {d4}, [r9], %3 \n"
michael@0	75	"vst1.8 {d7}, [r9], %3 \n"
michael@0	76	"vst1.8 {d6}, [r9] \n"
michael@0	77
michael@0	78	"add %0, #8 \n" // src += 8
michael@0	79	"add %2, %2, %3, lsl #3 \n" // dst += 8 * dst_stride
michael@0	80	"subs %4, #8 \n" // w -= 8
michael@0	81	"bge 1b \n"
michael@0	82
michael@0	83	// add 8 back to counter. if the result is 0 there are
michael@0	84	// no residuals.
michael@0	85	"adds %4, #8 \n"
michael@0	86	"beq 4f \n"
michael@0	87
michael@0	88	// some residual, so between 1 and 7 lines left to transpose
michael@0	89	"cmp %4, #2 \n"
michael@0	90	"blt 3f \n"
michael@0	91
michael@0	92	"cmp %4, #4 \n"
michael@0	93	"blt 2f \n"
michael@0	94
michael@0	95	// 4x8 block
michael@0	96	"mov r9, %0 \n"
michael@0	97	"vld1.32 {d0[0]}, [r9], %1 \n"
michael@0	98	"vld1.32 {d0[1]}, [r9], %1 \n"
michael@0	99	"vld1.32 {d1[0]}, [r9], %1 \n"
michael@0	100	"vld1.32 {d1[1]}, [r9], %1 \n"
michael@0	101	"vld1.32 {d2[0]}, [r9], %1 \n"
michael@0	102	"vld1.32 {d2[1]}, [r9], %1 \n"
michael@0	103	"vld1.32 {d3[0]}, [r9], %1 \n"
michael@0	104	"vld1.32 {d3[1]}, [r9] \n"
michael@0	105
michael@0	106	"mov r9, %2 \n"
michael@0	107
michael@0	108	"vld1.8 {q3}, [%5] \n"
michael@0	109
michael@0	110	"vtbl.8 d4, {d0, d1}, d6 \n"
michael@0	111	"vtbl.8 d5, {d0, d1}, d7 \n"
michael@0	112	"vtbl.8 d0, {d2, d3}, d6 \n"
michael@0	113	"vtbl.8 d1, {d2, d3}, d7 \n"
michael@0	114
michael@0	115	// TODO(frkoenig): Rework shuffle above to
michael@0	116	// write out with 4 instead of 8 writes.
michael@0	117	"vst1.32 {d4[0]}, [r9], %3 \n"
michael@0	118	"vst1.32 {d4[1]}, [r9], %3 \n"
michael@0	119	"vst1.32 {d5[0]}, [r9], %3 \n"
michael@0	120	"vst1.32 {d5[1]}, [r9] \n"
michael@0	121
michael@0	122	"add r9, %2, #4 \n"
michael@0	123	"vst1.32 {d0[0]}, [r9], %3 \n"
michael@0	124	"vst1.32 {d0[1]}, [r9], %3 \n"
michael@0	125	"vst1.32 {d1[0]}, [r9], %3 \n"
michael@0	126	"vst1.32 {d1[1]}, [r9] \n"
michael@0	127
michael@0	128	"add %0, #4 \n" // src += 4
michael@0	129	"add %2, %2, %3, lsl #2 \n" // dst += 4 * dst_stride
michael@0	130	"subs %4, #4 \n" // w -= 4
michael@0	131	"beq 4f \n"
michael@0	132
michael@0	133	// some residual, check to see if it includes a 2x8 block,
michael@0	134	// or less
michael@0	135	"cmp %4, #2 \n"
michael@0	136	"blt 3f \n"
michael@0	137
michael@0	138	// 2x8 block
michael@0	139	"2: \n"
michael@0	140	"mov r9, %0 \n"
michael@0	141	"vld1.16 {d0[0]}, [r9], %1 \n"
michael@0	142	"vld1.16 {d1[0]}, [r9], %1 \n"
michael@0	143	"vld1.16 {d0[1]}, [r9], %1 \n"
michael@0	144	"vld1.16 {d1[1]}, [r9], %1 \n"
michael@0	145	"vld1.16 {d0[2]}, [r9], %1 \n"
michael@0	146	"vld1.16 {d1[2]}, [r9], %1 \n"
michael@0	147	"vld1.16 {d0[3]}, [r9], %1 \n"
michael@0	148	"vld1.16 {d1[3]}, [r9] \n"
michael@0	149
michael@0	150	"vtrn.8 d0, d1 \n"
michael@0	151
michael@0	152	"mov r9, %2 \n"
michael@0	153
michael@0	154	"vst1.64 {d0}, [r9], %3 \n"
michael@0	155	"vst1.64 {d1}, [r9] \n"
michael@0	156
michael@0	157	"add %0, #2 \n" // src += 2
michael@0	158	"add %2, %2, %3, lsl #1 \n" // dst += 2 * dst_stride
michael@0	159	"subs %4, #2 \n" // w -= 2
michael@0	160	"beq 4f \n"
michael@0	161
michael@0	162	// 1x8 block
michael@0	163	"3: \n"
michael@0	164	"vld1.8 {d0[0]}, [%0], %1 \n"
michael@0	165	"vld1.8 {d0[1]}, [%0], %1 \n"
michael@0	166	"vld1.8 {d0[2]}, [%0], %1 \n"
michael@0	167	"vld1.8 {d0[3]}, [%0], %1 \n"
michael@0	168	"vld1.8 {d0[4]}, [%0], %1 \n"
michael@0	169	"vld1.8 {d0[5]}, [%0], %1 \n"
michael@0	170	"vld1.8 {d0[6]}, [%0], %1 \n"
michael@0	171	"vld1.8 {d0[7]}, [%0] \n"
michael@0	172
michael@0	173	"vst1.64 {d0}, [%2] \n"
michael@0	174
michael@0	175	"4: \n"
michael@0	176
michael@0	177	: "+r"(src), // %0
michael@0	178	"+r"(src_stride), // %1
michael@0	179	"+r"(dst), // %2
michael@0	180	"+r"(dst_stride), // %3
michael@0	181	"+r"(width) // %4
michael@0	182	: "r"(&kVTbl4x4Transpose) // %5
michael@0	183	: "memory", "cc", "r9", "q0", "q1", "q2", "q3"
michael@0	184	);
michael@0	185	}
michael@0	186
michael@0	187	static uvec8 kVTbl4x4TransposeDi =
michael@0	188	{ 0, 8, 1, 9, 2, 10, 3, 11, 4, 12, 5, 13, 6, 14, 7, 15 };
michael@0	189
michael@0	190	void TransposeUVWx8_NEON(const uint8* src, int src_stride,
michael@0	191	uint8* dst_a, int dst_stride_a,
michael@0	192	uint8* dst_b, int dst_stride_b,
michael@0	193	int width) {
michael@0	194	asm volatile (
michael@0	195	// loops are on blocks of 8. loop will stop when
michael@0	196	// counter gets to or below 0. starting the counter
michael@0	197	// at w-8 allow for this
michael@0	198	"sub %6, #8 \n"
michael@0	199
michael@0	200	// handle 8x8 blocks. this should be the majority of the plane
michael@0	201	".p2align 2 \n"
michael@0	202	"1: \n"
michael@0	203	"mov r9, %0 \n"
michael@0	204
michael@0	205	"vld2.8 {d0, d1}, [r9], %1 \n"
michael@0	206	"vld2.8 {d2, d3}, [r9], %1 \n"
michael@0	207	"vld2.8 {d4, d5}, [r9], %1 \n"
michael@0	208	"vld2.8 {d6, d7}, [r9], %1 \n"
michael@0	209	"vld2.8 {d16, d17}, [r9], %1 \n"
michael@0	210	"vld2.8 {d18, d19}, [r9], %1 \n"
michael@0	211	"vld2.8 {d20, d21}, [r9], %1 \n"
michael@0	212	"vld2.8 {d22, d23}, [r9] \n"
michael@0	213
michael@0	214	"vtrn.8 q1, q0 \n"
michael@0	215	"vtrn.8 q3, q2 \n"
michael@0	216	"vtrn.8 q9, q8 \n"
michael@0	217	"vtrn.8 q11, q10 \n"
michael@0	218
michael@0	219	"vtrn.16 q1, q3 \n"
michael@0	220	"vtrn.16 q0, q2 \n"
michael@0	221	"vtrn.16 q9, q11 \n"
michael@0	222	"vtrn.16 q8, q10 \n"
michael@0	223
michael@0	224	"vtrn.32 q1, q9 \n"
michael@0	225	"vtrn.32 q0, q8 \n"
michael@0	226	"vtrn.32 q3, q11 \n"
michael@0	227	"vtrn.32 q2, q10 \n"
michael@0	228
michael@0	229	"vrev16.8 q0, q0 \n"
michael@0	230	"vrev16.8 q1, q1 \n"
michael@0	231	"vrev16.8 q2, q2 \n"
michael@0	232	"vrev16.8 q3, q3 \n"
michael@0	233	"vrev16.8 q8, q8 \n"
michael@0	234	"vrev16.8 q9, q9 \n"
michael@0	235	"vrev16.8 q10, q10 \n"
michael@0	236	"vrev16.8 q11, q11 \n"
michael@0	237
michael@0	238	"mov r9, %2 \n"
michael@0	239
michael@0	240	"vst1.8 {d2}, [r9], %3 \n"
michael@0	241	"vst1.8 {d0}, [r9], %3 \n"
michael@0	242	"vst1.8 {d6}, [r9], %3 \n"
michael@0	243	"vst1.8 {d4}, [r9], %3 \n"
michael@0	244	"vst1.8 {d18}, [r9], %3 \n"
michael@0	245	"vst1.8 {d16}, [r9], %3 \n"
michael@0	246	"vst1.8 {d22}, [r9], %3 \n"
michael@0	247	"vst1.8 {d20}, [r9] \n"
michael@0	248
michael@0	249	"mov r9, %4 \n"
michael@0	250
michael@0	251	"vst1.8 {d3}, [r9], %5 \n"
michael@0	252	"vst1.8 {d1}, [r9], %5 \n"
michael@0	253	"vst1.8 {d7}, [r9], %5 \n"
michael@0	254	"vst1.8 {d5}, [r9], %5 \n"
michael@0	255	"vst1.8 {d19}, [r9], %5 \n"
michael@0	256	"vst1.8 {d17}, [r9], %5 \n"
michael@0	257	"vst1.8 {d23}, [r9], %5 \n"
michael@0	258	"vst1.8 {d21}, [r9] \n"
michael@0	259
michael@0	260	"add %0, #8*2 \n" // src += 8*2
michael@0	261	"add %2, %2, %3, lsl #3 \n" // dst_a += 8 * dst_stride_a
michael@0	262	"add %4, %4, %5, lsl #3 \n" // dst_b += 8 * dst_stride_b
michael@0	263	"subs %6, #8 \n" // w -= 8
michael@0	264	"bge 1b \n"
michael@0	265
michael@0	266	// add 8 back to counter. if the result is 0 there are
michael@0	267	// no residuals.
michael@0	268	"adds %6, #8 \n"
michael@0	269	"beq 4f \n"
michael@0	270
michael@0	271	// some residual, so between 1 and 7 lines left to transpose
michael@0	272	"cmp %6, #2 \n"
michael@0	273	"blt 3f \n"
michael@0	274
michael@0	275	"cmp %6, #4 \n"
michael@0	276	"blt 2f \n"
michael@0	277
michael@0	278	//TODO(frkoenig): Clean this up
michael@0	279	// 4x8 block
michael@0	280	"mov r9, %0 \n"
michael@0	281	"vld1.64 {d0}, [r9], %1 \n"
michael@0	282	"vld1.64 {d1}, [r9], %1 \n"
michael@0	283	"vld1.64 {d2}, [r9], %1 \n"
michael@0	284	"vld1.64 {d3}, [r9], %1 \n"
michael@0	285	"vld1.64 {d4}, [r9], %1 \n"
michael@0	286	"vld1.64 {d5}, [r9], %1 \n"
michael@0	287	"vld1.64 {d6}, [r9], %1 \n"
michael@0	288	"vld1.64 {d7}, [r9] \n"
michael@0	289
michael@0	290	"vld1.8 {q15}, [%7] \n"
michael@0	291
michael@0	292	"vtrn.8 q0, q1 \n"
michael@0	293	"vtrn.8 q2, q3 \n"
michael@0	294
michael@0	295	"vtbl.8 d16, {d0, d1}, d30 \n"
michael@0	296	"vtbl.8 d17, {d0, d1}, d31 \n"
michael@0	297	"vtbl.8 d18, {d2, d3}, d30 \n"
michael@0	298	"vtbl.8 d19, {d2, d3}, d31 \n"
michael@0	299	"vtbl.8 d20, {d4, d5}, d30 \n"
michael@0	300	"vtbl.8 d21, {d4, d5}, d31 \n"
michael@0	301	"vtbl.8 d22, {d6, d7}, d30 \n"
michael@0	302	"vtbl.8 d23, {d6, d7}, d31 \n"
michael@0	303
michael@0	304	"mov r9, %2 \n"
michael@0	305
michael@0	306	"vst1.32 {d16[0]}, [r9], %3 \n"
michael@0	307	"vst1.32 {d16[1]}, [r9], %3 \n"
michael@0	308	"vst1.32 {d17[0]}, [r9], %3 \n"
michael@0	309	"vst1.32 {d17[1]}, [r9], %3 \n"
michael@0	310
michael@0	311	"add r9, %2, #4 \n"
michael@0	312	"vst1.32 {d20[0]}, [r9], %3 \n"
michael@0	313	"vst1.32 {d20[1]}, [r9], %3 \n"
michael@0	314	"vst1.32 {d21[0]}, [r9], %3 \n"
michael@0	315	"vst1.32 {d21[1]}, [r9] \n"
michael@0	316
michael@0	317	"mov r9, %4 \n"
michael@0	318
michael@0	319	"vst1.32 {d18[0]}, [r9], %5 \n"
michael@0	320	"vst1.32 {d18[1]}, [r9], %5 \n"
michael@0	321	"vst1.32 {d19[0]}, [r9], %5 \n"
michael@0	322	"vst1.32 {d19[1]}, [r9], %5 \n"
michael@0	323
michael@0	324	"add r9, %4, #4 \n"
michael@0	325	"vst1.32 {d22[0]}, [r9], %5 \n"
michael@0	326	"vst1.32 {d22[1]}, [r9], %5 \n"
michael@0	327	"vst1.32 {d23[0]}, [r9], %5 \n"
michael@0	328	"vst1.32 {d23[1]}, [r9] \n"
michael@0	329
michael@0	330	"add %0, #4*2 \n" // src += 4 * 2
michael@0	331	"add %2, %2, %3, lsl #2 \n" // dst_a += 4 * dst_stride_a
michael@0	332	"add %4, %4, %5, lsl #2 \n" // dst_b += 4 * dst_stride_b
michael@0	333	"subs %6, #4 \n" // w -= 4
michael@0	334	"beq 4f \n"
michael@0	335
michael@0	336	// some residual, check to see if it includes a 2x8 block,
michael@0	337	// or less
michael@0	338	"cmp %6, #2 \n"
michael@0	339	"blt 3f \n"
michael@0	340
michael@0	341	// 2x8 block
michael@0	342	"2: \n"
michael@0	343	"mov r9, %0 \n"
michael@0	344	"vld2.16 {d0[0], d2[0]}, [r9], %1 \n"
michael@0	345	"vld2.16 {d1[0], d3[0]}, [r9], %1 \n"
michael@0	346	"vld2.16 {d0[1], d2[1]}, [r9], %1 \n"
michael@0	347	"vld2.16 {d1[1], d3[1]}, [r9], %1 \n"
michael@0	348	"vld2.16 {d0[2], d2[2]}, [r9], %1 \n"
michael@0	349	"vld2.16 {d1[2], d3[2]}, [r9], %1 \n"
michael@0	350	"vld2.16 {d0[3], d2[3]}, [r9], %1 \n"
michael@0	351	"vld2.16 {d1[3], d3[3]}, [r9] \n"
michael@0	352
michael@0	353	"vtrn.8 d0, d1 \n"
michael@0	354	"vtrn.8 d2, d3 \n"
michael@0	355
michael@0	356	"mov r9, %2 \n"
michael@0	357
michael@0	358	"vst1.64 {d0}, [r9], %3 \n"
michael@0	359	"vst1.64 {d2}, [r9] \n"
michael@0	360
michael@0	361	"mov r9, %4 \n"
michael@0	362
michael@0	363	"vst1.64 {d1}, [r9], %5 \n"
michael@0	364	"vst1.64 {d3}, [r9] \n"
michael@0	365
michael@0	366	"add %0, #2*2 \n" // src += 2 * 2
michael@0	367	"add %2, %2, %3, lsl #1 \n" // dst_a += 2 * dst_stride_a
michael@0	368	"add %4, %4, %5, lsl #1 \n" // dst_b += 2 * dst_stride_b
michael@0	369	"subs %6, #2 \n" // w -= 2
michael@0	370	"beq 4f \n"
michael@0	371
michael@0	372	// 1x8 block
michael@0	373	"3: \n"
michael@0	374	"vld2.8 {d0[0], d1[0]}, [%0], %1 \n"
michael@0	375	"vld2.8 {d0[1], d1[1]}, [%0], %1 \n"
michael@0	376	"vld2.8 {d0[2], d1[2]}, [%0], %1 \n"
michael@0	377	"vld2.8 {d0[3], d1[3]}, [%0], %1 \n"
michael@0	378	"vld2.8 {d0[4], d1[4]}, [%0], %1 \n"
michael@0	379	"vld2.8 {d0[5], d1[5]}, [%0], %1 \n"
michael@0	380	"vld2.8 {d0[6], d1[6]}, [%0], %1 \n"
michael@0	381	"vld2.8 {d0[7], d1[7]}, [%0] \n"
michael@0	382
michael@0	383	"vst1.64 {d0}, [%2] \n"
michael@0	384	"vst1.64 {d1}, [%4] \n"
michael@0	385
michael@0	386	"4: \n"
michael@0	387
michael@0	388	: "+r"(src), // %0
michael@0	389	"+r"(src_stride), // %1
michael@0	390	"+r"(dst_a), // %2
michael@0	391	"+r"(dst_stride_a), // %3
michael@0	392	"+r"(dst_b), // %4
michael@0	393	"+r"(dst_stride_b), // %5
michael@0	394	"+r"(width) // %6
michael@0	395	: "r"(&kVTbl4x4TransposeDi) // %7
michael@0	396	: "memory", "cc", "r9",
michael@0	397	"q0", "q1", "q2", "q3", "q8", "q9", "q10", "q11"
michael@0	398	);
michael@0	399	}
michael@0	400	#endif
michael@0	401
michael@0	402	#ifdef __cplusplus
michael@0	403	} // extern "C"
michael@0	404	} // namespace libyuv
michael@0	405	#endif