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

media/libvpx/vp8/common/arm/neon/sixtappredict8x8_neon.asm@ac0c01689b40 (annotated)

media/libvpx/vp8/common/arm/neon/sixtappredict8x8_neon.asm

Thu, 15 Jan 2015 15:59:08 +0100

author

Michael Schloh von Bennewitz <michael@schloh.com>

date

Thu, 15 Jan 2015 15:59:08 +0100

branch

TOR_BUG_9701

changeset 10

ac0c01689b40

permissions

-rw-r--r--

Implement a real Private Browsing Mode condition by changing the API/ABI;
This solves Tor bug #9701, complying with disk avoidance documented in
https://www.torproject.org/projects/torbrowser/design/#disk-avoidance.

michael@0	1	;
michael@0	2	; Copyright (c) 2010 The WebM 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
michael@0	12	EXPORT |vp8_sixtap_predict8x8_neon|
michael@0	13	ARM
michael@0	14	REQUIRE8
michael@0	15	PRESERVE8
michael@0	16
michael@0	17	AREA ||.text||, CODE, READONLY, ALIGN=2
michael@0	18
michael@0	19	filter8_coeff
michael@0	20	DCD 0, 0, 128, 0, 0, 0, 0, 0
michael@0	21	DCD 0, -6, 123, 12, -1, 0, 0, 0
michael@0	22	DCD 2, -11, 108, 36, -8, 1, 0, 0
michael@0	23	DCD 0, -9, 93, 50, -6, 0, 0, 0
michael@0	24	DCD 3, -16, 77, 77, -16, 3, 0, 0
michael@0	25	DCD 0, -6, 50, 93, -9, 0, 0, 0
michael@0	26	DCD 1, -8, 36, 108, -11, 2, 0, 0
michael@0	27	DCD 0, -1, 12, 123, -6, 0, 0, 0
michael@0	28
michael@0	29	; r0 unsigned char *src_ptr,
michael@0	30	; r1 int src_pixels_per_line,
michael@0	31	; r2 int xoffset,
michael@0	32	; r3 int yoffset,
michael@0	33	; stack(r4) unsigned char *dst_ptr,
michael@0	34	; stack(r5) int dst_pitch
michael@0	35
michael@0	36	|vp8_sixtap_predict8x8_neon| PROC
michael@0	37	push {r4-r5, lr}
michael@0	38
michael@0	39	adr r12, filter8_coeff
michael@0	40
michael@0	41	ldr r4, [sp, #12] ;load parameters from stack
michael@0	42	ldr r5, [sp, #16] ;load parameters from stack
michael@0	43
michael@0	44	cmp r2, #0 ;skip first_pass filter if xoffset=0
michael@0	45	beq secondpass_filter8x8_only
michael@0	46
michael@0	47	add r2, r12, r2, lsl #5 ;calculate filter location
michael@0	48
michael@0	49	cmp r3, #0 ;skip second_pass filter if yoffset=0
michael@0	50
michael@0	51	vld1.s32 {q14, q15}, [r2] ;load first_pass filter
michael@0	52
michael@0	53	beq firstpass_filter8x8_only
michael@0	54
michael@0	55	sub sp, sp, #64 ;reserve space on stack for temporary storage
michael@0	56	mov lr, sp
michael@0	57
michael@0	58	vabs.s32 q12, q14
michael@0	59	vabs.s32 q13, q15
michael@0	60
michael@0	61	mov r2, #2 ;loop counter
michael@0	62	sub r0, r0, #2 ;move srcptr back to (line-2) and (column-2)
michael@0	63	sub r0, r0, r1, lsl #1
michael@0	64
michael@0	65	vdup.8 d0, d24[0] ;first_pass filter (d0-d5)
michael@0	66	vdup.8 d1, d24[4]
michael@0	67	vdup.8 d2, d25[0]
michael@0	68
michael@0	69	;First pass: output_height lines x output_width columns (13x8)
michael@0	70	vld1.u8 {q3}, [r0], r1 ;load src data
michael@0	71	vdup.8 d3, d25[4]
michael@0	72	vld1.u8 {q4}, [r0], r1
michael@0	73	vdup.8 d4, d26[0]
michael@0	74	vld1.u8 {q5}, [r0], r1
michael@0	75	vdup.8 d5, d26[4]
michael@0	76	vld1.u8 {q6}, [r0], r1
michael@0	77
michael@0	78	filt_blk2d_fp8x8_loop_neon
michael@0	79	pld [r0]
michael@0	80	pld [r0, r1]
michael@0	81	pld [r0, r1, lsl #1]
michael@0	82
michael@0	83	vmull.u8 q7, d6, d0 ;(src_ptr[-2] * vp8_filter[0])
michael@0	84	vmull.u8 q8, d8, d0
michael@0	85	vmull.u8 q9, d10, d0
michael@0	86	vmull.u8 q10, d12, d0
michael@0	87
michael@0	88	vext.8 d28, d6, d7, #1 ;construct src_ptr[-1]
michael@0	89	vext.8 d29, d8, d9, #1
michael@0	90	vext.8 d30, d10, d11, #1
michael@0	91	vext.8 d31, d12, d13, #1
michael@0	92
michael@0	93	vmlsl.u8 q7, d28, d1 ;-(src_ptr[-1] * vp8_filter[1])
michael@0	94	vmlsl.u8 q8, d29, d1
michael@0	95	vmlsl.u8 q9, d30, d1
michael@0	96	vmlsl.u8 q10, d31, d1
michael@0	97
michael@0	98	vext.8 d28, d6, d7, #4 ;construct src_ptr[2]
michael@0	99	vext.8 d29, d8, d9, #4
michael@0	100	vext.8 d30, d10, d11, #4
michael@0	101	vext.8 d31, d12, d13, #4
michael@0	102
michael@0	103	vmlsl.u8 q7, d28, d4 ;-(src_ptr[2] * vp8_filter[4])
michael@0	104	vmlsl.u8 q8, d29, d4
michael@0	105	vmlsl.u8 q9, d30, d4
michael@0	106	vmlsl.u8 q10, d31, d4
michael@0	107
michael@0	108	vext.8 d28, d6, d7, #2 ;construct src_ptr[0]
michael@0	109	vext.8 d29, d8, d9, #2
michael@0	110	vext.8 d30, d10, d11, #2
michael@0	111	vext.8 d31, d12, d13, #2
michael@0	112
michael@0	113	vmlal.u8 q7, d28, d2 ;(src_ptr[0] * vp8_filter[2])
michael@0	114	vmlal.u8 q8, d29, d2
michael@0	115	vmlal.u8 q9, d30, d2
michael@0	116	vmlal.u8 q10, d31, d2
michael@0	117
michael@0	118	vext.8 d28, d6, d7, #5 ;construct src_ptr[3]
michael@0	119	vext.8 d29, d8, d9, #5
michael@0	120	vext.8 d30, d10, d11, #5
michael@0	121	vext.8 d31, d12, d13, #5
michael@0	122
michael@0	123	vmlal.u8 q7, d28, d5 ;(src_ptr[3] * vp8_filter[5])
michael@0	124	vmlal.u8 q8, d29, d5
michael@0	125	vmlal.u8 q9, d30, d5
michael@0	126	vmlal.u8 q10, d31, d5
michael@0	127
michael@0	128	vext.8 d28, d6, d7, #3 ;construct src_ptr[1]
michael@0	129	vext.8 d29, d8, d9, #3
michael@0	130	vext.8 d30, d10, d11, #3
michael@0	131	vext.8 d31, d12, d13, #3
michael@0	132
michael@0	133	vmull.u8 q3, d28, d3 ;(src_ptr[1] * vp8_filter[3])
michael@0	134	vmull.u8 q4, d29, d3
michael@0	135	vmull.u8 q5, d30, d3
michael@0	136	vmull.u8 q6, d31, d3
michael@0	137
michael@0	138	subs r2, r2, #1
michael@0	139
michael@0	140	vqadd.s16 q7, q3 ;sum of all (src_data*filter_parameters)
michael@0	141	vqadd.s16 q8, q4
michael@0	142	vqadd.s16 q9, q5
michael@0	143	vqadd.s16 q10, q6
michael@0	144
michael@0	145	vld1.u8 {q3}, [r0], r1 ;load src data
michael@0	146
michael@0	147	vqrshrun.s16 d22, q7, #7 ;shift/round/saturate to u8
michael@0	148	vqrshrun.s16 d23, q8, #7
michael@0	149	vqrshrun.s16 d24, q9, #7
michael@0	150	vqrshrun.s16 d25, q10, #7
michael@0	151
michael@0	152	vst1.u8 {d22}, [lr]! ;store result
michael@0	153	vld1.u8 {q4}, [r0], r1
michael@0	154	vst1.u8 {d23}, [lr]!
michael@0	155	vld1.u8 {q5}, [r0], r1
michael@0	156	vst1.u8 {d24}, [lr]!
michael@0	157	vld1.u8 {q6}, [r0], r1
michael@0	158	vst1.u8 {d25}, [lr]!
michael@0	159
michael@0	160	bne filt_blk2d_fp8x8_loop_neon
michael@0	161
michael@0	162	;first_pass filtering on the rest 5-line data
michael@0	163	;vld1.u8 {q3}, [r0], r1 ;load src data
michael@0	164	;vld1.u8 {q4}, [r0], r1
michael@0	165	;vld1.u8 {q5}, [r0], r1
michael@0	166	;vld1.u8 {q6}, [r0], r1
michael@0	167	vld1.u8 {q7}, [r0], r1
michael@0	168
michael@0	169	vmull.u8 q8, d6, d0 ;(src_ptr[-2] * vp8_filter[0])
michael@0	170	vmull.u8 q9, d8, d0
michael@0	171	vmull.u8 q10, d10, d0
michael@0	172	vmull.u8 q11, d12, d0
michael@0	173	vmull.u8 q12, d14, d0
michael@0	174
michael@0	175	vext.8 d27, d6, d7, #1 ;construct src_ptr[-1]
michael@0	176	vext.8 d28, d8, d9, #1
michael@0	177	vext.8 d29, d10, d11, #1
michael@0	178	vext.8 d30, d12, d13, #1
michael@0	179	vext.8 d31, d14, d15, #1
michael@0	180
michael@0	181	vmlsl.u8 q8, d27, d1 ;-(src_ptr[-1] * vp8_filter[1])
michael@0	182	vmlsl.u8 q9, d28, d1
michael@0	183	vmlsl.u8 q10, d29, d1
michael@0	184	vmlsl.u8 q11, d30, d1
michael@0	185	vmlsl.u8 q12, d31, d1
michael@0	186
michael@0	187	vext.8 d27, d6, d7, #4 ;construct src_ptr[2]
michael@0	188	vext.8 d28, d8, d9, #4
michael@0	189	vext.8 d29, d10, d11, #4
michael@0	190	vext.8 d30, d12, d13, #4
michael@0	191	vext.8 d31, d14, d15, #4
michael@0	192
michael@0	193	vmlsl.u8 q8, d27, d4 ;-(src_ptr[2] * vp8_filter[4])
michael@0	194	vmlsl.u8 q9, d28, d4
michael@0	195	vmlsl.u8 q10, d29, d4
michael@0	196	vmlsl.u8 q11, d30, d4
michael@0	197	vmlsl.u8 q12, d31, d4
michael@0	198
michael@0	199	vext.8 d27, d6, d7, #2 ;construct src_ptr[0]
michael@0	200	vext.8 d28, d8, d9, #2
michael@0	201	vext.8 d29, d10, d11, #2
michael@0	202	vext.8 d30, d12, d13, #2
michael@0	203	vext.8 d31, d14, d15, #2
michael@0	204
michael@0	205	vmlal.u8 q8, d27, d2 ;(src_ptr[0] * vp8_filter[2])
michael@0	206	vmlal.u8 q9, d28, d2
michael@0	207	vmlal.u8 q10, d29, d2
michael@0	208	vmlal.u8 q11, d30, d2
michael@0	209	vmlal.u8 q12, d31, d2
michael@0	210
michael@0	211	vext.8 d27, d6, d7, #5 ;construct src_ptr[3]
michael@0	212	vext.8 d28, d8, d9, #5
michael@0	213	vext.8 d29, d10, d11, #5
michael@0	214	vext.8 d30, d12, d13, #5
michael@0	215	vext.8 d31, d14, d15, #5
michael@0	216
michael@0	217	vmlal.u8 q8, d27, d5 ;(src_ptr[3] * vp8_filter[5])
michael@0	218	vmlal.u8 q9, d28, d5
michael@0	219	vmlal.u8 q10, d29, d5
michael@0	220	vmlal.u8 q11, d30, d5
michael@0	221	vmlal.u8 q12, d31, d5
michael@0	222
michael@0	223	vext.8 d27, d6, d7, #3 ;construct src_ptr[1]
michael@0	224	vext.8 d28, d8, d9, #3
michael@0	225	vext.8 d29, d10, d11, #3
michael@0	226	vext.8 d30, d12, d13, #3
michael@0	227	vext.8 d31, d14, d15, #3
michael@0	228
michael@0	229	vmull.u8 q3, d27, d3 ;(src_ptr[1] * vp8_filter[3])
michael@0	230	vmull.u8 q4, d28, d3
michael@0	231	vmull.u8 q5, d29, d3
michael@0	232	vmull.u8 q6, d30, d3
michael@0	233	vmull.u8 q7, d31, d3
michael@0	234
michael@0	235	vqadd.s16 q8, q3 ;sum of all (src_data*filter_parameters)
michael@0	236	vqadd.s16 q9, q4
michael@0	237	vqadd.s16 q10, q5
michael@0	238	vqadd.s16 q11, q6
michael@0	239	vqadd.s16 q12, q7
michael@0	240
michael@0	241	add r3, r12, r3, lsl #5
michael@0	242
michael@0	243	vqrshrun.s16 d26, q8, #7 ;shift/round/saturate to u8
michael@0	244	sub lr, lr, #64
michael@0	245	vqrshrun.s16 d27, q9, #7
michael@0	246	vld1.u8 {q9}, [lr]! ;load intermediate data from stack
michael@0	247	vqrshrun.s16 d28, q10, #7
michael@0	248	vld1.u8 {q10}, [lr]!
michael@0	249
michael@0	250	vld1.s32 {q5, q6}, [r3] ;load second_pass filter
michael@0	251
michael@0	252	vqrshrun.s16 d29, q11, #7
michael@0	253	vld1.u8 {q11}, [lr]!
michael@0	254
michael@0	255	vabs.s32 q7, q5
michael@0	256	vabs.s32 q8, q6
michael@0	257
michael@0	258	vqrshrun.s16 d30, q12, #7
michael@0	259	vld1.u8 {q12}, [lr]!
michael@0	260
michael@0	261	;Second pass: 8x8
michael@0	262	mov r3, #2 ;loop counter
michael@0	263
michael@0	264	vdup.8 d0, d14[0] ;second_pass filter parameters (d0-d5)
michael@0	265	vdup.8 d1, d14[4]
michael@0	266	vdup.8 d2, d15[0]
michael@0	267	vdup.8 d3, d15[4]
michael@0	268	vdup.8 d4, d16[0]
michael@0	269	vdup.8 d5, d16[4]
michael@0	270
michael@0	271	filt_blk2d_sp8x8_loop_neon
michael@0	272	vmull.u8 q3, d18, d0 ;(src_ptr[-2] * vp8_filter[0])
michael@0	273	vmull.u8 q4, d19, d0
michael@0	274	vmull.u8 q5, d20, d0
michael@0	275	vmull.u8 q6, d21, d0
michael@0	276
michael@0	277	vmlsl.u8 q3, d19, d1 ;-(src_ptr[-1] * vp8_filter[1])
michael@0	278	vmlsl.u8 q4, d20, d1
michael@0	279	vmlsl.u8 q5, d21, d1
michael@0	280	vmlsl.u8 q6, d22, d1
michael@0	281
michael@0	282	vmlsl.u8 q3, d22, d4 ;-(src_ptr[2] * vp8_filter[4])
michael@0	283	vmlsl.u8 q4, d23, d4
michael@0	284	vmlsl.u8 q5, d24, d4
michael@0	285	vmlsl.u8 q6, d25, d4
michael@0	286
michael@0	287	vmlal.u8 q3, d20, d2 ;(src_ptr[0] * vp8_filter[2])
michael@0	288	vmlal.u8 q4, d21, d2
michael@0	289	vmlal.u8 q5, d22, d2
michael@0	290	vmlal.u8 q6, d23, d2
michael@0	291
michael@0	292	vmlal.u8 q3, d23, d5 ;(src_ptr[3] * vp8_filter[5])
michael@0	293	vmlal.u8 q4, d24, d5
michael@0	294	vmlal.u8 q5, d25, d5
michael@0	295	vmlal.u8 q6, d26, d5
michael@0	296
michael@0	297	vmull.u8 q7, d21, d3 ;(src_ptr[1] * vp8_filter[3])
michael@0	298	vmull.u8 q8, d22, d3
michael@0	299	vmull.u8 q9, d23, d3
michael@0	300	vmull.u8 q10, d24, d3
michael@0	301
michael@0	302	subs r3, r3, #1
michael@0	303
michael@0	304	vqadd.s16 q7, q3 ;sum of all (src_data*filter_parameters)
michael@0	305	vqadd.s16 q8, q4
michael@0	306	vqadd.s16 q9, q5
michael@0	307	vqadd.s16 q10, q6
michael@0	308
michael@0	309	vqrshrun.s16 d6, q7, #7 ;shift/round/saturate to u8
michael@0	310	vqrshrun.s16 d7, q8, #7
michael@0	311	vqrshrun.s16 d8, q9, #7
michael@0	312	vqrshrun.s16 d9, q10, #7
michael@0	313
michael@0	314	vmov q9, q11
michael@0	315	vst1.u8 {d6}, [r4], r5 ;store result
michael@0	316	vmov q10, q12
michael@0	317	vst1.u8 {d7}, [r4], r5
michael@0	318	vmov q11, q13
michael@0	319	vst1.u8 {d8}, [r4], r5
michael@0	320	vmov q12, q14
michael@0	321	vst1.u8 {d9}, [r4], r5
michael@0	322	vmov d26, d30
michael@0	323
michael@0	324	bne filt_blk2d_sp8x8_loop_neon
michael@0	325
michael@0	326	add sp, sp, #64
michael@0	327	pop {r4-r5,pc}
michael@0	328
michael@0	329	;---------------------
michael@0	330	firstpass_filter8x8_only
michael@0	331	;add r2, r12, r2, lsl #5 ;calculate filter location
michael@0	332	;vld1.s32 {q14, q15}, [r2] ;load first_pass filter
michael@0	333	vabs.s32 q12, q14
michael@0	334	vabs.s32 q13, q15
michael@0	335
michael@0	336	mov r2, #2 ;loop counter
michael@0	337	sub r0, r0, #2 ;move srcptr back to (line-2) and (column-2)
michael@0	338
michael@0	339	vdup.8 d0, d24[0] ;first_pass filter (d0-d5)
michael@0	340	vdup.8 d1, d24[4]
michael@0	341	vdup.8 d2, d25[0]
michael@0	342	vdup.8 d3, d25[4]
michael@0	343	vdup.8 d4, d26[0]
michael@0	344	vdup.8 d5, d26[4]
michael@0	345
michael@0	346	;First pass: output_height lines x output_width columns (8x8)
michael@0	347	filt_blk2d_fpo8x8_loop_neon
michael@0	348	vld1.u8 {q3}, [r0], r1 ;load src data
michael@0	349	vld1.u8 {q4}, [r0], r1
michael@0	350	vld1.u8 {q5}, [r0], r1
michael@0	351	vld1.u8 {q6}, [r0], r1
michael@0	352
michael@0	353	pld [r0]
michael@0	354	pld [r0, r1]
michael@0	355	pld [r0, r1, lsl #1]
michael@0	356
michael@0	357	vmull.u8 q7, d6, d0 ;(src_ptr[-2] * vp8_filter[0])
michael@0	358	vmull.u8 q8, d8, d0
michael@0	359	vmull.u8 q9, d10, d0
michael@0	360	vmull.u8 q10, d12, d0
michael@0	361
michael@0	362	vext.8 d28, d6, d7, #1 ;construct src_ptr[-1]
michael@0	363	vext.8 d29, d8, d9, #1
michael@0	364	vext.8 d30, d10, d11, #1
michael@0	365	vext.8 d31, d12, d13, #1
michael@0	366
michael@0	367	vmlsl.u8 q7, d28, d1 ;-(src_ptr[-1] * vp8_filter[1])
michael@0	368	vmlsl.u8 q8, d29, d1
michael@0	369	vmlsl.u8 q9, d30, d1
michael@0	370	vmlsl.u8 q10, d31, d1
michael@0	371
michael@0	372	vext.8 d28, d6, d7, #4 ;construct src_ptr[2]
michael@0	373	vext.8 d29, d8, d9, #4
michael@0	374	vext.8 d30, d10, d11, #4
michael@0	375	vext.8 d31, d12, d13, #4
michael@0	376
michael@0	377	vmlsl.u8 q7, d28, d4 ;-(src_ptr[2] * vp8_filter[4])
michael@0	378	vmlsl.u8 q8, d29, d4
michael@0	379	vmlsl.u8 q9, d30, d4
michael@0	380	vmlsl.u8 q10, d31, d4
michael@0	381
michael@0	382	vext.8 d28, d6, d7, #2 ;construct src_ptr[0]
michael@0	383	vext.8 d29, d8, d9, #2
michael@0	384	vext.8 d30, d10, d11, #2
michael@0	385	vext.8 d31, d12, d13, #2
michael@0	386
michael@0	387	vmlal.u8 q7, d28, d2 ;(src_ptr[0] * vp8_filter[2])
michael@0	388	vmlal.u8 q8, d29, d2
michael@0	389	vmlal.u8 q9, d30, d2
michael@0	390	vmlal.u8 q10, d31, d2
michael@0	391
michael@0	392	vext.8 d28, d6, d7, #5 ;construct src_ptr[3]
michael@0	393	vext.8 d29, d8, d9, #5
michael@0	394	vext.8 d30, d10, d11, #5
michael@0	395	vext.8 d31, d12, d13, #5
michael@0	396
michael@0	397	vmlal.u8 q7, d28, d5 ;(src_ptr[3] * vp8_filter[5])
michael@0	398	vmlal.u8 q8, d29, d5
michael@0	399	vmlal.u8 q9, d30, d5
michael@0	400	vmlal.u8 q10, d31, d5
michael@0	401
michael@0	402	vext.8 d28, d6, d7, #3 ;construct src_ptr[1]
michael@0	403	vext.8 d29, d8, d9, #3
michael@0	404	vext.8 d30, d10, d11, #3
michael@0	405	vext.8 d31, d12, d13, #3
michael@0	406
michael@0	407	vmull.u8 q3, d28, d3 ;(src_ptr[1] * vp8_filter[3])
michael@0	408	vmull.u8 q4, d29, d3
michael@0	409	vmull.u8 q5, d30, d3
michael@0	410	vmull.u8 q6, d31, d3
michael@0	411	;
michael@0	412	vqadd.s16 q7, q3 ;sum of all (src_data*filter_parameters)
michael@0	413	vqadd.s16 q8, q4
michael@0	414	vqadd.s16 q9, q5
michael@0	415	vqadd.s16 q10, q6
michael@0	416
michael@0	417	subs r2, r2, #1
michael@0	418
michael@0	419	vqrshrun.s16 d22, q7, #7 ;shift/round/saturate to u8
michael@0	420	vqrshrun.s16 d23, q8, #7
michael@0	421	vqrshrun.s16 d24, q9, #7
michael@0	422	vqrshrun.s16 d25, q10, #7
michael@0	423
michael@0	424	vst1.u8 {d22}, [r4], r5 ;store result
michael@0	425	vst1.u8 {d23}, [r4], r5
michael@0	426	vst1.u8 {d24}, [r4], r5
michael@0	427	vst1.u8 {d25}, [r4], r5
michael@0	428
michael@0	429	bne filt_blk2d_fpo8x8_loop_neon
michael@0	430
michael@0	431	pop {r4-r5,pc}
michael@0	432
michael@0	433	;---------------------
michael@0	434	secondpass_filter8x8_only
michael@0	435	sub r0, r0, r1, lsl #1
michael@0	436	add r3, r12, r3, lsl #5
michael@0	437
michael@0	438	vld1.u8 {d18}, [r0], r1 ;load src data
michael@0	439	vld1.s32 {q5, q6}, [r3] ;load second_pass filter
michael@0	440	vld1.u8 {d19}, [r0], r1
michael@0	441	vabs.s32 q7, q5
michael@0	442	vld1.u8 {d20}, [r0], r1
michael@0	443	vabs.s32 q8, q6
michael@0	444	vld1.u8 {d21}, [r0], r1
michael@0	445	mov r3, #2 ;loop counter
michael@0	446	vld1.u8 {d22}, [r0], r1
michael@0	447	vdup.8 d0, d14[0] ;second_pass filter parameters (d0-d5)
michael@0	448	vld1.u8 {d23}, [r0], r1
michael@0	449	vdup.8 d1, d14[4]
michael@0	450	vld1.u8 {d24}, [r0], r1
michael@0	451	vdup.8 d2, d15[0]
michael@0	452	vld1.u8 {d25}, [r0], r1
michael@0	453	vdup.8 d3, d15[4]
michael@0	454	vld1.u8 {d26}, [r0], r1
michael@0	455	vdup.8 d4, d16[0]
michael@0	456	vld1.u8 {d27}, [r0], r1
michael@0	457	vdup.8 d5, d16[4]
michael@0	458	vld1.u8 {d28}, [r0], r1
michael@0	459	vld1.u8 {d29}, [r0], r1
michael@0	460	vld1.u8 {d30}, [r0], r1
michael@0	461
michael@0	462	;Second pass: 8x8
michael@0	463	filt_blk2d_spo8x8_loop_neon
michael@0	464	vmull.u8 q3, d18, d0 ;(src_ptr[-2] * vp8_filter[0])
michael@0	465	vmull.u8 q4, d19, d0
michael@0	466	vmull.u8 q5, d20, d0
michael@0	467	vmull.u8 q6, d21, d0
michael@0	468
michael@0	469	vmlsl.u8 q3, d19, d1 ;-(src_ptr[-1] * vp8_filter[1])
michael@0	470	vmlsl.u8 q4, d20, d1
michael@0	471	vmlsl.u8 q5, d21, d1
michael@0	472	vmlsl.u8 q6, d22, d1
michael@0	473
michael@0	474	vmlsl.u8 q3, d22, d4 ;-(src_ptr[2] * vp8_filter[4])
michael@0	475	vmlsl.u8 q4, d23, d4
michael@0	476	vmlsl.u8 q5, d24, d4
michael@0	477	vmlsl.u8 q6, d25, d4
michael@0	478
michael@0	479	vmlal.u8 q3, d20, d2 ;(src_ptr[0] * vp8_filter[2])
michael@0	480	vmlal.u8 q4, d21, d2
michael@0	481	vmlal.u8 q5, d22, d2
michael@0	482	vmlal.u8 q6, d23, d2
michael@0	483
michael@0	484	vmlal.u8 q3, d23, d5 ;(src_ptr[3] * vp8_filter[5])
michael@0	485	vmlal.u8 q4, d24, d5
michael@0	486	vmlal.u8 q5, d25, d5
michael@0	487	vmlal.u8 q6, d26, d5
michael@0	488
michael@0	489	vmull.u8 q7, d21, d3 ;(src_ptr[1] * vp8_filter[3])
michael@0	490	vmull.u8 q8, d22, d3
michael@0	491	vmull.u8 q9, d23, d3
michael@0	492	vmull.u8 q10, d24, d3
michael@0	493
michael@0	494	subs r3, r3, #1
michael@0	495
michael@0	496	vqadd.s16 q7, q3 ;sum of all (src_data*filter_parameters)
michael@0	497	vqadd.s16 q8, q4
michael@0	498	vqadd.s16 q9, q5
michael@0	499	vqadd.s16 q10, q6
michael@0	500
michael@0	501	vqrshrun.s16 d6, q7, #7 ;shift/round/saturate to u8
michael@0	502	vqrshrun.s16 d7, q8, #7
michael@0	503	vqrshrun.s16 d8, q9, #7
michael@0	504	vqrshrun.s16 d9, q10, #7
michael@0	505
michael@0	506	vmov q9, q11
michael@0	507	vst1.u8 {d6}, [r4], r5 ;store result
michael@0	508	vmov q10, q12
michael@0	509	vst1.u8 {d7}, [r4], r5
michael@0	510	vmov q11, q13
michael@0	511	vst1.u8 {d8}, [r4], r5
michael@0	512	vmov q12, q14
michael@0	513	vst1.u8 {d9}, [r4], r5
michael@0	514	vmov d26, d30
michael@0	515
michael@0	516	bne filt_blk2d_spo8x8_loop_neon
michael@0	517
michael@0	518	pop {r4-r5,pc}
michael@0	519
michael@0	520	ENDP
michael@0	521
michael@0	522	;-----------------
michael@0	523
michael@0	524	END