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

media/libvpx/vp8/common/arm/neon/bilinearpredict16x16_neon.asm@ac0c01689b40 (annotated)

media/libvpx/vp8/common/arm/neon/bilinearpredict16x16_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_bilinear_predict16x16_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	; r0 unsigned char *src_ptr,
michael@0	19	; r1 int src_pixels_per_line,
michael@0	20	; r2 int xoffset,
michael@0	21	; r3 int yoffset,
michael@0	22	; r4 unsigned char *dst_ptr,
michael@0	23	; stack(r5) int dst_pitch
michael@0	24
michael@0	25	|vp8_bilinear_predict16x16_neon| PROC
michael@0	26	push {r4-r5, lr}
michael@0	27
michael@0	28	adr r12, bifilter16_coeff
michael@0	29	ldr r4, [sp, #12] ;load parameters from stack
michael@0	30	ldr r5, [sp, #16] ;load parameters from stack
michael@0	31
michael@0	32	cmp r2, #0 ;skip first_pass filter if xoffset=0
michael@0	33	beq secondpass_bfilter16x16_only
michael@0	34
michael@0	35	add r2, r12, r2, lsl #3 ;calculate filter location
michael@0	36
michael@0	37	cmp r3, #0 ;skip second_pass filter if yoffset=0
michael@0	38
michael@0	39	vld1.s32 {d31}, [r2] ;load first_pass filter
michael@0	40
michael@0	41	beq firstpass_bfilter16x16_only
michael@0	42
michael@0	43	sub sp, sp, #272 ;reserve space on stack for temporary storage
michael@0	44	vld1.u8 {d2, d3, d4}, [r0], r1 ;load src data
michael@0	45	mov lr, sp
michael@0	46	vld1.u8 {d5, d6, d7}, [r0], r1
michael@0	47
michael@0	48	mov r2, #3 ;loop counter
michael@0	49	vld1.u8 {d8, d9, d10}, [r0], r1
michael@0	50
michael@0	51	vdup.8 d0, d31[0] ;first_pass filter (d0 d1)
michael@0	52	vld1.u8 {d11, d12, d13}, [r0], r1
michael@0	53
michael@0	54	vdup.8 d1, d31[4]
michael@0	55
michael@0	56	;First Pass: output_height lines x output_width columns (17x16)
michael@0	57	filt_blk2d_fp16x16_loop_neon
michael@0	58	pld [r0]
michael@0	59	pld [r0, r1]
michael@0	60	pld [r0, r1, lsl #1]
michael@0	61
michael@0	62	vmull.u8 q7, d2, d0 ;(src_ptr[0] * vp8_filter[0])
michael@0	63	vmull.u8 q8, d3, d0
michael@0	64	vmull.u8 q9, d5, d0
michael@0	65	vmull.u8 q10, d6, d0
michael@0	66	vmull.u8 q11, d8, d0
michael@0	67	vmull.u8 q12, d9, d0
michael@0	68	vmull.u8 q13, d11, d0
michael@0	69	vmull.u8 q14, d12, d0
michael@0	70
michael@0	71	vext.8 d2, d2, d3, #1 ;construct src_ptr[1]
michael@0	72	vext.8 d5, d5, d6, #1
michael@0	73	vext.8 d8, d8, d9, #1
michael@0	74	vext.8 d11, d11, d12, #1
michael@0	75
michael@0	76	vmlal.u8 q7, d2, d1 ;(src_ptr[0] * vp8_filter[1])
michael@0	77	vmlal.u8 q9, d5, d1
michael@0	78	vmlal.u8 q11, d8, d1
michael@0	79	vmlal.u8 q13, d11, d1
michael@0	80
michael@0	81	vext.8 d3, d3, d4, #1
michael@0	82	vext.8 d6, d6, d7, #1
michael@0	83	vext.8 d9, d9, d10, #1
michael@0	84	vext.8 d12, d12, d13, #1
michael@0	85
michael@0	86	vmlal.u8 q8, d3, d1 ;(src_ptr[0] * vp8_filter[1])
michael@0	87	vmlal.u8 q10, d6, d1
michael@0	88	vmlal.u8 q12, d9, d1
michael@0	89	vmlal.u8 q14, d12, d1
michael@0	90
michael@0	91	subs r2, r2, #1
michael@0	92
michael@0	93	vqrshrn.u16 d14, q7, #7 ;shift/round/saturate to u8
michael@0	94	vqrshrn.u16 d15, q8, #7
michael@0	95	vqrshrn.u16 d16, q9, #7
michael@0	96	vqrshrn.u16 d17, q10, #7
michael@0	97	vqrshrn.u16 d18, q11, #7
michael@0	98	vqrshrn.u16 d19, q12, #7
michael@0	99	vqrshrn.u16 d20, q13, #7
michael@0	100
michael@0	101	vld1.u8 {d2, d3, d4}, [r0], r1 ;load src data
michael@0	102	vqrshrn.u16 d21, q14, #7
michael@0	103	vld1.u8 {d5, d6, d7}, [r0], r1
michael@0	104
michael@0	105	vst1.u8 {d14, d15, d16, d17}, [lr]! ;store result
michael@0	106	vld1.u8 {d8, d9, d10}, [r0], r1
michael@0	107	vst1.u8 {d18, d19, d20, d21}, [lr]!
michael@0	108	vld1.u8 {d11, d12, d13}, [r0], r1
michael@0	109
michael@0	110	bne filt_blk2d_fp16x16_loop_neon
michael@0	111
michael@0	112	;First-pass filtering for rest 5 lines
michael@0	113	vld1.u8 {d14, d15, d16}, [r0], r1
michael@0	114
michael@0	115	vmull.u8 q9, d2, d0 ;(src_ptr[0] * vp8_filter[0])
michael@0	116	vmull.u8 q10, d3, d0
michael@0	117	vmull.u8 q11, d5, d0
michael@0	118	vmull.u8 q12, d6, d0
michael@0	119	vmull.u8 q13, d8, d0
michael@0	120	vmull.u8 q14, d9, d0
michael@0	121
michael@0	122	vext.8 d2, d2, d3, #1 ;construct src_ptr[1]
michael@0	123	vext.8 d5, d5, d6, #1
michael@0	124	vext.8 d8, d8, d9, #1
michael@0	125
michael@0	126	vmlal.u8 q9, d2, d1 ;(src_ptr[0] * vp8_filter[1])
michael@0	127	vmlal.u8 q11, d5, d1
michael@0	128	vmlal.u8 q13, d8, d1
michael@0	129
michael@0	130	vext.8 d3, d3, d4, #1
michael@0	131	vext.8 d6, d6, d7, #1
michael@0	132	vext.8 d9, d9, d10, #1
michael@0	133
michael@0	134	vmlal.u8 q10, d3, d1 ;(src_ptr[0] * vp8_filter[1])
michael@0	135	vmlal.u8 q12, d6, d1
michael@0	136	vmlal.u8 q14, d9, d1
michael@0	137
michael@0	138	vmull.u8 q1, d11, d0
michael@0	139	vmull.u8 q2, d12, d0
michael@0	140	vmull.u8 q3, d14, d0
michael@0	141	vmull.u8 q4, d15, d0
michael@0	142
michael@0	143	vext.8 d11, d11, d12, #1 ;construct src_ptr[1]
michael@0	144	vext.8 d14, d14, d15, #1
michael@0	145
michael@0	146	vmlal.u8 q1, d11, d1 ;(src_ptr[0] * vp8_filter[1])
michael@0	147	vmlal.u8 q3, d14, d1
michael@0	148
michael@0	149	vext.8 d12, d12, d13, #1
michael@0	150	vext.8 d15, d15, d16, #1
michael@0	151
michael@0	152	vmlal.u8 q2, d12, d1 ;(src_ptr[0] * vp8_filter[1])
michael@0	153	vmlal.u8 q4, d15, d1
michael@0	154
michael@0	155	vqrshrn.u16 d10, q9, #7 ;shift/round/saturate to u8
michael@0	156	vqrshrn.u16 d11, q10, #7
michael@0	157	vqrshrn.u16 d12, q11, #7
michael@0	158	vqrshrn.u16 d13, q12, #7
michael@0	159	vqrshrn.u16 d14, q13, #7
michael@0	160	vqrshrn.u16 d15, q14, #7
michael@0	161	vqrshrn.u16 d16, q1, #7
michael@0	162	vqrshrn.u16 d17, q2, #7
michael@0	163	vqrshrn.u16 d18, q3, #7
michael@0	164	vqrshrn.u16 d19, q4, #7
michael@0	165
michael@0	166	vst1.u8 {d10, d11, d12, d13}, [lr]! ;store result
michael@0	167	vst1.u8 {d14, d15, d16, d17}, [lr]!
michael@0	168	vst1.u8 {d18, d19}, [lr]!
michael@0	169
michael@0	170	;Second pass: 16x16
michael@0	171	;secondpass_filter
michael@0	172	add r3, r12, r3, lsl #3
michael@0	173	sub lr, lr, #272
michael@0	174
michael@0	175	vld1.u32 {d31}, [r3] ;load second_pass filter
michael@0	176
michael@0	177	vld1.u8 {d22, d23}, [lr]! ;load src data
michael@0	178
michael@0	179	vdup.8 d0, d31[0] ;second_pass filter parameters (d0 d1)
michael@0	180	vdup.8 d1, d31[4]
michael@0	181	mov r12, #4 ;loop counter
michael@0	182
michael@0	183	filt_blk2d_sp16x16_loop_neon
michael@0	184	vld1.u8 {d24, d25}, [lr]!
michael@0	185	vmull.u8 q1, d22, d0 ;(src_ptr[0] * vp8_filter[0])
michael@0	186	vld1.u8 {d26, d27}, [lr]!
michael@0	187	vmull.u8 q2, d23, d0
michael@0	188	vld1.u8 {d28, d29}, [lr]!
michael@0	189	vmull.u8 q3, d24, d0
michael@0	190	vld1.u8 {d30, d31}, [lr]!
michael@0	191
michael@0	192	vmull.u8 q4, d25, d0
michael@0	193	vmull.u8 q5, d26, d0
michael@0	194	vmull.u8 q6, d27, d0
michael@0	195	vmull.u8 q7, d28, d0
michael@0	196	vmull.u8 q8, d29, d0
michael@0	197
michael@0	198	vmlal.u8 q1, d24, d1 ;(src_ptr[pixel_step] * vp8_filter[1])
michael@0	199	vmlal.u8 q2, d25, d1
michael@0	200	vmlal.u8 q3, d26, d1
michael@0	201	vmlal.u8 q4, d27, d1
michael@0	202	vmlal.u8 q5, d28, d1
michael@0	203	vmlal.u8 q6, d29, d1
michael@0	204	vmlal.u8 q7, d30, d1
michael@0	205	vmlal.u8 q8, d31, d1
michael@0	206
michael@0	207	subs r12, r12, #1
michael@0	208
michael@0	209	vqrshrn.u16 d2, q1, #7 ;shift/round/saturate to u8
michael@0	210	vqrshrn.u16 d3, q2, #7
michael@0	211	vqrshrn.u16 d4, q3, #7
michael@0	212	vqrshrn.u16 d5, q4, #7
michael@0	213	vqrshrn.u16 d6, q5, #7
michael@0	214	vqrshrn.u16 d7, q6, #7
michael@0	215	vqrshrn.u16 d8, q7, #7
michael@0	216	vqrshrn.u16 d9, q8, #7
michael@0	217
michael@0	218	vst1.u8 {d2, d3}, [r4], r5 ;store result
michael@0	219	vst1.u8 {d4, d5}, [r4], r5
michael@0	220	vst1.u8 {d6, d7}, [r4], r5
michael@0	221	vmov q11, q15
michael@0	222	vst1.u8 {d8, d9}, [r4], r5
michael@0	223
michael@0	224	bne filt_blk2d_sp16x16_loop_neon
michael@0	225
michael@0	226	add sp, sp, #272
michael@0	227
michael@0	228	pop {r4-r5,pc}
michael@0	229
michael@0	230	;--------------------
michael@0	231	firstpass_bfilter16x16_only
michael@0	232	mov r2, #4 ;loop counter
michael@0	233	vdup.8 d0, d31[0] ;first_pass filter (d0 d1)
michael@0	234	vdup.8 d1, d31[4]
michael@0	235
michael@0	236	;First Pass: output_height lines x output_width columns (16x16)
michael@0	237	filt_blk2d_fpo16x16_loop_neon
michael@0	238	vld1.u8 {d2, d3, d4}, [r0], r1 ;load src data
michael@0	239	vld1.u8 {d5, d6, d7}, [r0], r1
michael@0	240	vld1.u8 {d8, d9, d10}, [r0], r1
michael@0	241	vld1.u8 {d11, d12, d13}, [r0], r1
michael@0	242
michael@0	243	pld [r0]
michael@0	244	pld [r0, r1]
michael@0	245	pld [r0, r1, lsl #1]
michael@0	246
michael@0	247	vmull.u8 q7, d2, d0 ;(src_ptr[0] * vp8_filter[0])
michael@0	248	vmull.u8 q8, d3, d0
michael@0	249	vmull.u8 q9, d5, d0
michael@0	250	vmull.u8 q10, d6, d0
michael@0	251	vmull.u8 q11, d8, d0
michael@0	252	vmull.u8 q12, d9, d0
michael@0	253	vmull.u8 q13, d11, d0
michael@0	254	vmull.u8 q14, d12, d0
michael@0	255
michael@0	256	vext.8 d2, d2, d3, #1 ;construct src_ptr[1]
michael@0	257	vext.8 d5, d5, d6, #1
michael@0	258	vext.8 d8, d8, d9, #1
michael@0	259	vext.8 d11, d11, d12, #1
michael@0	260
michael@0	261	vmlal.u8 q7, d2, d1 ;(src_ptr[0] * vp8_filter[1])
michael@0	262	vmlal.u8 q9, d5, d1
michael@0	263	vmlal.u8 q11, d8, d1
michael@0	264	vmlal.u8 q13, d11, d1
michael@0	265
michael@0	266	vext.8 d3, d3, d4, #1
michael@0	267	vext.8 d6, d6, d7, #1
michael@0	268	vext.8 d9, d9, d10, #1
michael@0	269	vext.8 d12, d12, d13, #1
michael@0	270
michael@0	271	vmlal.u8 q8, d3, d1 ;(src_ptr[0] * vp8_filter[1])
michael@0	272	vmlal.u8 q10, d6, d1
michael@0	273	vmlal.u8 q12, d9, d1
michael@0	274	vmlal.u8 q14, d12, d1
michael@0	275
michael@0	276	subs r2, r2, #1
michael@0	277
michael@0	278	vqrshrn.u16 d14, q7, #7 ;shift/round/saturate to u8
michael@0	279	vqrshrn.u16 d15, q8, #7
michael@0	280	vqrshrn.u16 d16, q9, #7
michael@0	281	vqrshrn.u16 d17, q10, #7
michael@0	282	vqrshrn.u16 d18, q11, #7
michael@0	283	vqrshrn.u16 d19, q12, #7
michael@0	284	vqrshrn.u16 d20, q13, #7
michael@0	285	vst1.u8 {d14, d15}, [r4], r5 ;store result
michael@0	286	vqrshrn.u16 d21, q14, #7
michael@0	287
michael@0	288	vst1.u8 {d16, d17}, [r4], r5
michael@0	289	vst1.u8 {d18, d19}, [r4], r5
michael@0	290	vst1.u8 {d20, d21}, [r4], r5
michael@0	291
michael@0	292	bne filt_blk2d_fpo16x16_loop_neon
michael@0	293	pop {r4-r5,pc}
michael@0	294
michael@0	295	;---------------------
michael@0	296	secondpass_bfilter16x16_only
michael@0	297	;Second pass: 16x16
michael@0	298	;secondpass_filter
michael@0	299	add r3, r12, r3, lsl #3
michael@0	300	mov r12, #4 ;loop counter
michael@0	301	vld1.u32 {d31}, [r3] ;load second_pass filter
michael@0	302	vld1.u8 {d22, d23}, [r0], r1 ;load src data
michael@0	303
michael@0	304	vdup.8 d0, d31[0] ;second_pass filter parameters (d0 d1)
michael@0	305	vdup.8 d1, d31[4]
michael@0	306
michael@0	307	filt_blk2d_spo16x16_loop_neon
michael@0	308	vld1.u8 {d24, d25}, [r0], r1
michael@0	309	vmull.u8 q1, d22, d0 ;(src_ptr[0] * vp8_filter[0])
michael@0	310	vld1.u8 {d26, d27}, [r0], r1
michael@0	311	vmull.u8 q2, d23, d0
michael@0	312	vld1.u8 {d28, d29}, [r0], r1
michael@0	313	vmull.u8 q3, d24, d0
michael@0	314	vld1.u8 {d30, d31}, [r0], r1
michael@0	315
michael@0	316	vmull.u8 q4, d25, d0
michael@0	317	vmull.u8 q5, d26, d0
michael@0	318	vmull.u8 q6, d27, d0
michael@0	319	vmull.u8 q7, d28, d0
michael@0	320	vmull.u8 q8, d29, d0
michael@0	321
michael@0	322	vmlal.u8 q1, d24, d1 ;(src_ptr[pixel_step] * vp8_filter[1])
michael@0	323	vmlal.u8 q2, d25, d1
michael@0	324	vmlal.u8 q3, d26, d1
michael@0	325	vmlal.u8 q4, d27, d1
michael@0	326	vmlal.u8 q5, d28, d1
michael@0	327	vmlal.u8 q6, d29, d1
michael@0	328	vmlal.u8 q7, d30, d1
michael@0	329	vmlal.u8 q8, d31, d1
michael@0	330
michael@0	331	vqrshrn.u16 d2, q1, #7 ;shift/round/saturate to u8
michael@0	332	vqrshrn.u16 d3, q2, #7
michael@0	333	vqrshrn.u16 d4, q3, #7
michael@0	334	vqrshrn.u16 d5, q4, #7
michael@0	335	vqrshrn.u16 d6, q5, #7
michael@0	336	vqrshrn.u16 d7, q6, #7
michael@0	337	vqrshrn.u16 d8, q7, #7
michael@0	338	vqrshrn.u16 d9, q8, #7
michael@0	339
michael@0	340	vst1.u8 {d2, d3}, [r4], r5 ;store result
michael@0	341	subs r12, r12, #1
michael@0	342	vst1.u8 {d4, d5}, [r4], r5
michael@0	343	vmov q11, q15
michael@0	344	vst1.u8 {d6, d7}, [r4], r5
michael@0	345	vst1.u8 {d8, d9}, [r4], r5
michael@0	346
michael@0	347	bne filt_blk2d_spo16x16_loop_neon
michael@0	348	pop {r4-r5,pc}
michael@0	349
michael@0	350	ENDP
michael@0	351
michael@0	352	;-----------------
michael@0	353
michael@0	354	bifilter16_coeff
michael@0	355	DCD 128, 0, 112, 16, 96, 32, 80, 48, 64, 64, 48, 80, 32, 96, 16, 112
michael@0	356
michael@0	357	END