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

media/libvpx/vp8/common/arm/neon/vp8_subpixelvariance16x16_neon.asm@ac0c01689b40 (annotated)

media/libvpx/vp8/common/arm/neon/vp8_subpixelvariance16x16_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	;-----------------
michael@0	13
michael@0	14	EXPORT |vp8_sub_pixel_variance16x16_neon_func|
michael@0	15	ARM
michael@0	16	REQUIRE8
michael@0	17	PRESERVE8
michael@0	18
michael@0	19	AREA ||.text||, CODE, READONLY, ALIGN=2
michael@0	20	; r0 unsigned char *src_ptr,
michael@0	21	; r1 int src_pixels_per_line,
michael@0	22	; r2 int xoffset,
michael@0	23	; r3 int yoffset,
michael@0	24	; stack(r4) unsigned char *dst_ptr,
michael@0	25	; stack(r5) int dst_pixels_per_line,
michael@0	26	; stack(r6) unsigned int *sse
michael@0	27	;note: most of the code is copied from bilinear_predict16x16_neon and vp8_variance16x16_neon.
michael@0	28
michael@0	29	bilinear_taps_coeff
michael@0	30	DCD 128, 0, 112, 16, 96, 32, 80, 48, 64, 64, 48, 80, 32, 96, 16, 112
michael@0	31
michael@0	32	|vp8_sub_pixel_variance16x16_neon_func| PROC
michael@0	33	push {r4-r6, lr}
michael@0	34
michael@0	35	adr r12, bilinear_taps_coeff
michael@0	36	ldr r4, [sp, #16] ;load *dst_ptr from stack
michael@0	37	ldr r5, [sp, #20] ;load dst_pixels_per_line from stack
michael@0	38	ldr r6, [sp, #24] ;load *sse from stack
michael@0	39
michael@0	40	cmp r2, #0 ;skip first_pass filter if xoffset=0
michael@0	41	beq secondpass_bfilter16x16_only
michael@0	42
michael@0	43	add r2, r12, r2, lsl #3 ;calculate filter location
michael@0	44
michael@0	45	cmp r3, #0 ;skip second_pass filter if yoffset=0
michael@0	46
michael@0	47	vld1.s32 {d31}, [r2] ;load first_pass filter
michael@0	48
michael@0	49	beq firstpass_bfilter16x16_only
michael@0	50
michael@0	51	sub sp, sp, #272 ;reserve space on stack for temporary storage
michael@0	52	vld1.u8 {d2, d3, d4}, [r0], r1 ;load src data
michael@0	53	mov lr, sp
michael@0	54	vld1.u8 {d5, d6, d7}, [r0], r1
michael@0	55
michael@0	56	mov r2, #3 ;loop counter
michael@0	57	vld1.u8 {d8, d9, d10}, [r0], r1
michael@0	58
michael@0	59	vdup.8 d0, d31[0] ;first_pass filter (d0 d1)
michael@0	60	vld1.u8 {d11, d12, d13}, [r0], r1
michael@0	61
michael@0	62	vdup.8 d1, d31[4]
michael@0	63
michael@0	64	;First Pass: output_height lines x output_width columns (17x16)
michael@0	65	vp8e_filt_blk2d_fp16x16_loop_neon
michael@0	66	pld [r0]
michael@0	67	pld [r0, r1]
michael@0	68	pld [r0, r1, lsl #1]
michael@0	69
michael@0	70	vmull.u8 q7, d2, d0 ;(src_ptr[0] * Filter[0])
michael@0	71	vmull.u8 q8, d3, d0
michael@0	72	vmull.u8 q9, d5, d0
michael@0	73	vmull.u8 q10, d6, d0
michael@0	74	vmull.u8 q11, d8, d0
michael@0	75	vmull.u8 q12, d9, d0
michael@0	76	vmull.u8 q13, d11, d0
michael@0	77	vmull.u8 q14, d12, d0
michael@0	78
michael@0	79	vext.8 d2, d2, d3, #1 ;construct src_ptr[1]
michael@0	80	vext.8 d5, d5, d6, #1
michael@0	81	vext.8 d8, d8, d9, #1
michael@0	82	vext.8 d11, d11, d12, #1
michael@0	83
michael@0	84	vmlal.u8 q7, d2, d1 ;(src_ptr[0] * Filter[1])
michael@0	85	vmlal.u8 q9, d5, d1
michael@0	86	vmlal.u8 q11, d8, d1
michael@0	87	vmlal.u8 q13, d11, d1
michael@0	88
michael@0	89	vext.8 d3, d3, d4, #1
michael@0	90	vext.8 d6, d6, d7, #1
michael@0	91	vext.8 d9, d9, d10, #1
michael@0	92	vext.8 d12, d12, d13, #1
michael@0	93
michael@0	94	vmlal.u8 q8, d3, d1 ;(src_ptr[0] * Filter[1])
michael@0	95	vmlal.u8 q10, d6, d1
michael@0	96	vmlal.u8 q12, d9, d1
michael@0	97	vmlal.u8 q14, d12, d1
michael@0	98
michael@0	99	subs r2, r2, #1
michael@0	100
michael@0	101	vqrshrn.u16 d14, q7, #7 ;shift/round/saturate to u8
michael@0	102	vqrshrn.u16 d15, q8, #7
michael@0	103	vqrshrn.u16 d16, q9, #7
michael@0	104	vqrshrn.u16 d17, q10, #7
michael@0	105	vqrshrn.u16 d18, q11, #7
michael@0	106	vqrshrn.u16 d19, q12, #7
michael@0	107	vqrshrn.u16 d20, q13, #7
michael@0	108
michael@0	109	vld1.u8 {d2, d3, d4}, [r0], r1 ;load src data
michael@0	110	vqrshrn.u16 d21, q14, #7
michael@0	111	vld1.u8 {d5, d6, d7}, [r0], r1
michael@0	112
michael@0	113	vst1.u8 {d14, d15, d16, d17}, [lr]! ;store result
michael@0	114	vld1.u8 {d8, d9, d10}, [r0], r1
michael@0	115	vst1.u8 {d18, d19, d20, d21}, [lr]!
michael@0	116	vld1.u8 {d11, d12, d13}, [r0], r1
michael@0	117
michael@0	118	bne vp8e_filt_blk2d_fp16x16_loop_neon
michael@0	119
michael@0	120	;First-pass filtering for rest 5 lines
michael@0	121	vld1.u8 {d14, d15, d16}, [r0], r1
michael@0	122
michael@0	123	vmull.u8 q9, d2, d0 ;(src_ptr[0] * Filter[0])
michael@0	124	vmull.u8 q10, d3, d0
michael@0	125	vmull.u8 q11, d5, d0
michael@0	126	vmull.u8 q12, d6, d0
michael@0	127	vmull.u8 q13, d8, d0
michael@0	128	vmull.u8 q14, d9, d0
michael@0	129
michael@0	130	vext.8 d2, d2, d3, #1 ;construct src_ptr[1]
michael@0	131	vext.8 d5, d5, d6, #1
michael@0	132	vext.8 d8, d8, d9, #1
michael@0	133
michael@0	134	vmlal.u8 q9, d2, d1 ;(src_ptr[0] * Filter[1])
michael@0	135	vmlal.u8 q11, d5, d1
michael@0	136	vmlal.u8 q13, d8, d1
michael@0	137
michael@0	138	vext.8 d3, d3, d4, #1
michael@0	139	vext.8 d6, d6, d7, #1
michael@0	140	vext.8 d9, d9, d10, #1
michael@0	141
michael@0	142	vmlal.u8 q10, d3, d1 ;(src_ptr[0] * Filter[1])
michael@0	143	vmlal.u8 q12, d6, d1
michael@0	144	vmlal.u8 q14, d9, d1
michael@0	145
michael@0	146	vmull.u8 q1, d11, d0
michael@0	147	vmull.u8 q2, d12, d0
michael@0	148	vmull.u8 q3, d14, d0
michael@0	149	vmull.u8 q4, d15, d0
michael@0	150
michael@0	151	vext.8 d11, d11, d12, #1 ;construct src_ptr[1]
michael@0	152	vext.8 d14, d14, d15, #1
michael@0	153
michael@0	154	vmlal.u8 q1, d11, d1 ;(src_ptr[0] * Filter[1])
michael@0	155	vmlal.u8 q3, d14, d1
michael@0	156
michael@0	157	vext.8 d12, d12, d13, #1
michael@0	158	vext.8 d15, d15, d16, #1
michael@0	159
michael@0	160	vmlal.u8 q2, d12, d1 ;(src_ptr[0] * Filter[1])
michael@0	161	vmlal.u8 q4, d15, d1
michael@0	162
michael@0	163	vqrshrn.u16 d10, q9, #7 ;shift/round/saturate to u8
michael@0	164	vqrshrn.u16 d11, q10, #7
michael@0	165	vqrshrn.u16 d12, q11, #7
michael@0	166	vqrshrn.u16 d13, q12, #7
michael@0	167	vqrshrn.u16 d14, q13, #7
michael@0	168	vqrshrn.u16 d15, q14, #7
michael@0	169	vqrshrn.u16 d16, q1, #7
michael@0	170	vqrshrn.u16 d17, q2, #7
michael@0	171	vqrshrn.u16 d18, q3, #7
michael@0	172	vqrshrn.u16 d19, q4, #7
michael@0	173
michael@0	174	vst1.u8 {d10, d11, d12, d13}, [lr]! ;store result
michael@0	175	vst1.u8 {d14, d15, d16, d17}, [lr]!
michael@0	176	vst1.u8 {d18, d19}, [lr]!
michael@0	177
michael@0	178	;Second pass: 16x16
michael@0	179	;secondpass_filter
michael@0	180	add r3, r12, r3, lsl #3
michael@0	181	sub lr, lr, #272
michael@0	182
michael@0	183	vld1.u32 {d31}, [r3] ;load second_pass filter
michael@0	184
michael@0	185	sub sp, sp, #256
michael@0	186	mov r3, sp
michael@0	187
michael@0	188	vld1.u8 {d22, d23}, [lr]! ;load src data
michael@0	189
michael@0	190	vdup.8 d0, d31[0] ;second_pass filter parameters (d0 d1)
michael@0	191	vdup.8 d1, d31[4]
michael@0	192	mov r12, #4 ;loop counter
michael@0	193
michael@0	194	vp8e_filt_blk2d_sp16x16_loop_neon
michael@0	195	vld1.u8 {d24, d25}, [lr]!
michael@0	196	vmull.u8 q1, d22, d0 ;(src_ptr[0] * Filter[0])
michael@0	197	vld1.u8 {d26, d27}, [lr]!
michael@0	198	vmull.u8 q2, d23, d0
michael@0	199	vld1.u8 {d28, d29}, [lr]!
michael@0	200	vmull.u8 q3, d24, d0
michael@0	201	vld1.u8 {d30, d31}, [lr]!
michael@0	202
michael@0	203	vmull.u8 q4, d25, d0
michael@0	204	vmull.u8 q5, d26, d0
michael@0	205	vmull.u8 q6, d27, d0
michael@0	206	vmull.u8 q7, d28, d0
michael@0	207	vmull.u8 q8, d29, d0
michael@0	208
michael@0	209	vmlal.u8 q1, d24, d1 ;(src_ptr[pixel_step] * Filter[1])
michael@0	210	vmlal.u8 q2, d25, d1
michael@0	211	vmlal.u8 q3, d26, d1
michael@0	212	vmlal.u8 q4, d27, d1
michael@0	213	vmlal.u8 q5, d28, d1
michael@0	214	vmlal.u8 q6, d29, d1
michael@0	215	vmlal.u8 q7, d30, d1
michael@0	216	vmlal.u8 q8, d31, d1
michael@0	217
michael@0	218	subs r12, r12, #1
michael@0	219
michael@0	220	vqrshrn.u16 d2, q1, #7 ;shift/round/saturate to u8
michael@0	221	vqrshrn.u16 d3, q2, #7
michael@0	222	vqrshrn.u16 d4, q3, #7
michael@0	223	vqrshrn.u16 d5, q4, #7
michael@0	224	vqrshrn.u16 d6, q5, #7
michael@0	225	vqrshrn.u16 d7, q6, #7
michael@0	226	vqrshrn.u16 d8, q7, #7
michael@0	227	vqrshrn.u16 d9, q8, #7
michael@0	228
michael@0	229	vst1.u8 {d2, d3}, [r3]! ;store result
michael@0	230	vst1.u8 {d4, d5}, [r3]!
michael@0	231	vst1.u8 {d6, d7}, [r3]!
michael@0	232	vmov q11, q15
michael@0	233	vst1.u8 {d8, d9}, [r3]!
michael@0	234
michael@0	235	bne vp8e_filt_blk2d_sp16x16_loop_neon
michael@0	236
michael@0	237	b sub_pixel_variance16x16_neon
michael@0	238
michael@0	239	;--------------------
michael@0	240	firstpass_bfilter16x16_only
michael@0	241	mov r2, #4 ;loop counter
michael@0	242	sub sp, sp, #528 ;reserve space on stack for temporary storage
michael@0	243	vdup.8 d0, d31[0] ;first_pass filter (d0 d1)
michael@0	244	vdup.8 d1, d31[4]
michael@0	245	mov r3, sp
michael@0	246
michael@0	247	;First Pass: output_height lines x output_width columns (16x16)
michael@0	248	vp8e_filt_blk2d_fpo16x16_loop_neon
michael@0	249	vld1.u8 {d2, d3, d4}, [r0], r1 ;load src data
michael@0	250	vld1.u8 {d5, d6, d7}, [r0], r1
michael@0	251	vld1.u8 {d8, d9, d10}, [r0], r1
michael@0	252	vld1.u8 {d11, d12, d13}, [r0], r1
michael@0	253
michael@0	254	pld [r0]
michael@0	255	pld [r0, r1]
michael@0	256	pld [r0, r1, lsl #1]
michael@0	257
michael@0	258	vmull.u8 q7, d2, d0 ;(src_ptr[0] * Filter[0])
michael@0	259	vmull.u8 q8, d3, d0
michael@0	260	vmull.u8 q9, d5, d0
michael@0	261	vmull.u8 q10, d6, d0
michael@0	262	vmull.u8 q11, d8, d0
michael@0	263	vmull.u8 q12, d9, d0
michael@0	264	vmull.u8 q13, d11, d0
michael@0	265	vmull.u8 q14, d12, d0
michael@0	266
michael@0	267	vext.8 d2, d2, d3, #1 ;construct src_ptr[1]
michael@0	268	vext.8 d5, d5, d6, #1
michael@0	269	vext.8 d8, d8, d9, #1
michael@0	270	vext.8 d11, d11, d12, #1
michael@0	271
michael@0	272	vmlal.u8 q7, d2, d1 ;(src_ptr[0] * Filter[1])
michael@0	273	vmlal.u8 q9, d5, d1
michael@0	274	vmlal.u8 q11, d8, d1
michael@0	275	vmlal.u8 q13, d11, d1
michael@0	276
michael@0	277	vext.8 d3, d3, d4, #1
michael@0	278	vext.8 d6, d6, d7, #1
michael@0	279	vext.8 d9, d9, d10, #1
michael@0	280	vext.8 d12, d12, d13, #1
michael@0	281
michael@0	282	vmlal.u8 q8, d3, d1 ;(src_ptr[0] * Filter[1])
michael@0	283	vmlal.u8 q10, d6, d1
michael@0	284	vmlal.u8 q12, d9, d1
michael@0	285	vmlal.u8 q14, d12, d1
michael@0	286
michael@0	287	subs r2, r2, #1
michael@0	288
michael@0	289	vqrshrn.u16 d14, q7, #7 ;shift/round/saturate to u8
michael@0	290	vqrshrn.u16 d15, q8, #7
michael@0	291	vqrshrn.u16 d16, q9, #7
michael@0	292	vqrshrn.u16 d17, q10, #7
michael@0	293	vqrshrn.u16 d18, q11, #7
michael@0	294	vqrshrn.u16 d19, q12, #7
michael@0	295	vqrshrn.u16 d20, q13, #7
michael@0	296	vst1.u8 {d14, d15}, [r3]! ;store result
michael@0	297	vqrshrn.u16 d21, q14, #7
michael@0	298
michael@0	299	vst1.u8 {d16, d17}, [r3]!
michael@0	300	vst1.u8 {d18, d19}, [r3]!
michael@0	301	vst1.u8 {d20, d21}, [r3]!
michael@0	302
michael@0	303	bne vp8e_filt_blk2d_fpo16x16_loop_neon
michael@0	304
michael@0	305	b sub_pixel_variance16x16_neon
michael@0	306
michael@0	307	;---------------------
michael@0	308	secondpass_bfilter16x16_only
michael@0	309	;Second pass: 16x16
michael@0	310	;secondpass_filter
michael@0	311	sub sp, sp, #528 ;reserve space on stack for temporary storage
michael@0	312	add r3, r12, r3, lsl #3
michael@0	313	mov r12, #4 ;loop counter
michael@0	314	vld1.u32 {d31}, [r3] ;load second_pass filter
michael@0	315	vld1.u8 {d22, d23}, [r0], r1 ;load src data
michael@0	316	mov r3, sp
michael@0	317
michael@0	318	vdup.8 d0, d31[0] ;second_pass filter parameters (d0 d1)
michael@0	319	vdup.8 d1, d31[4]
michael@0	320
michael@0	321	vp8e_filt_blk2d_spo16x16_loop_neon
michael@0	322	vld1.u8 {d24, d25}, [r0], r1
michael@0	323	vmull.u8 q1, d22, d0 ;(src_ptr[0] * Filter[0])
michael@0	324	vld1.u8 {d26, d27}, [r0], r1
michael@0	325	vmull.u8 q2, d23, d0
michael@0	326	vld1.u8 {d28, d29}, [r0], r1
michael@0	327	vmull.u8 q3, d24, d0
michael@0	328	vld1.u8 {d30, d31}, [r0], r1
michael@0	329
michael@0	330	vmull.u8 q4, d25, d0
michael@0	331	vmull.u8 q5, d26, d0
michael@0	332	vmull.u8 q6, d27, d0
michael@0	333	vmull.u8 q7, d28, d0
michael@0	334	vmull.u8 q8, d29, d0
michael@0	335
michael@0	336	vmlal.u8 q1, d24, d1 ;(src_ptr[pixel_step] * Filter[1])
michael@0	337	vmlal.u8 q2, d25, d1
michael@0	338	vmlal.u8 q3, d26, d1
michael@0	339	vmlal.u8 q4, d27, d1
michael@0	340	vmlal.u8 q5, d28, d1
michael@0	341	vmlal.u8 q6, d29, d1
michael@0	342	vmlal.u8 q7, d30, d1
michael@0	343	vmlal.u8 q8, d31, d1
michael@0	344
michael@0	345	vqrshrn.u16 d2, q1, #7 ;shift/round/saturate to u8
michael@0	346	vqrshrn.u16 d3, q2, #7
michael@0	347	vqrshrn.u16 d4, q3, #7
michael@0	348	vqrshrn.u16 d5, q4, #7
michael@0	349	vqrshrn.u16 d6, q5, #7
michael@0	350	vqrshrn.u16 d7, q6, #7
michael@0	351	vqrshrn.u16 d8, q7, #7
michael@0	352	vqrshrn.u16 d9, q8, #7
michael@0	353
michael@0	354	vst1.u8 {d2, d3}, [r3]! ;store result
michael@0	355	subs r12, r12, #1
michael@0	356	vst1.u8 {d4, d5}, [r3]!
michael@0	357	vmov q11, q15
michael@0	358	vst1.u8 {d6, d7}, [r3]!
michael@0	359	vst1.u8 {d8, d9}, [r3]!
michael@0	360
michael@0	361	bne vp8e_filt_blk2d_spo16x16_loop_neon
michael@0	362
michael@0	363	b sub_pixel_variance16x16_neon
michael@0	364
michael@0	365	;----------------------------
michael@0	366	;variance16x16
michael@0	367	sub_pixel_variance16x16_neon
michael@0	368	vmov.i8 q8, #0 ;q8 - sum
michael@0	369	vmov.i8 q9, #0 ;q9, q10 - sse
michael@0	370	vmov.i8 q10, #0
michael@0	371
michael@0	372	sub r3, r3, #256
michael@0	373	mov r12, #8
michael@0	374
michael@0	375	sub_pixel_variance16x16_neon_loop
michael@0	376	vld1.8 {q0}, [r3]! ;Load up source and reference
michael@0	377	vld1.8 {q2}, [r4], r5
michael@0	378	vld1.8 {q1}, [r3]!
michael@0	379	vld1.8 {q3}, [r4], r5
michael@0	380
michael@0	381	vsubl.u8 q11, d0, d4 ;diff
michael@0	382	vsubl.u8 q12, d1, d5
michael@0	383	vsubl.u8 q13, d2, d6
michael@0	384	vsubl.u8 q14, d3, d7
michael@0	385
michael@0	386	vpadal.s16 q8, q11 ;sum
michael@0	387	vmlal.s16 q9, d22, d22 ;sse
michael@0	388	vmlal.s16 q10, d23, d23
michael@0	389
michael@0	390	subs r12, r12, #1
michael@0	391
michael@0	392	vpadal.s16 q8, q12
michael@0	393	vmlal.s16 q9, d24, d24
michael@0	394	vmlal.s16 q10, d25, d25
michael@0	395	vpadal.s16 q8, q13
michael@0	396	vmlal.s16 q9, d26, d26
michael@0	397	vmlal.s16 q10, d27, d27
michael@0	398	vpadal.s16 q8, q14
michael@0	399	vmlal.s16 q9, d28, d28
michael@0	400	vmlal.s16 q10, d29, d29
michael@0	401
michael@0	402	bne sub_pixel_variance16x16_neon_loop
michael@0	403
michael@0	404	vadd.u32 q10, q9, q10 ;accumulate sse
michael@0	405	vpaddl.s32 q0, q8 ;accumulate sum
michael@0	406
michael@0	407	vpaddl.u32 q1, q10
michael@0	408	vadd.s64 d0, d0, d1
michael@0	409	vadd.u64 d1, d2, d3
michael@0	410
michael@0	411	vmull.s32 q5, d0, d0
michael@0	412	vst1.32 {d1[0]}, [r6] ;store sse
michael@0	413	vshr.u32 d10, d10, #8
michael@0	414	vsub.u32 d0, d1, d10
michael@0	415
michael@0	416	add sp, sp, #528
michael@0	417	vmov.32 r0, d0[0] ;return
michael@0	418
michael@0	419	pop {r4-r6,pc}
michael@0	420
michael@0	421	ENDP
michael@0	422
michael@0	423	END