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

media/libvpx/vp8/common/x86/postproc_mmx.asm@ac0c01689b40 (annotated)

media/libvpx/vp8/common/x86/postproc_mmx.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	%include "vpx_ports/x86_abi_support.asm"
michael@0	13
michael@0	14	%define VP8_FILTER_WEIGHT 128
michael@0	15	%define VP8_FILTER_SHIFT 7
michael@0	16
michael@0	17	;void vp8_mbpost_proc_down_mmx(unsigned char *dst,
michael@0	18	; int pitch, int rows, int cols,int flimit)
michael@0	19	extern sym(vp8_rv)
michael@0	20	global sym(vp8_mbpost_proc_down_mmx) PRIVATE
michael@0	21	sym(vp8_mbpost_proc_down_mmx):
michael@0	22	push rbp
michael@0	23	mov rbp, rsp
michael@0	24	SHADOW_ARGS_TO_STACK 5
michael@0	25	GET_GOT rbx
michael@0	26	push rsi
michael@0	27	push rdi
michael@0	28	; end prolog
michael@0	29
michael@0	30	ALIGN_STACK 16, rax
michael@0	31	sub rsp, 136
michael@0	32
michael@0	33	; unsigned char d[16][8] at [rsp]
michael@0	34	; create flimit2 at [rsp+128]
michael@0	35	mov eax, dword ptr arg(4) ;flimit
michael@0	36	mov [rsp+128], eax
michael@0	37	mov [rsp+128+4], eax
michael@0	38	%define flimit2 [rsp+128]
michael@0	39
michael@0	40	%if ABI_IS_32BIT=0
michael@0	41	lea r8, [GLOBAL(sym(vp8_rv))]
michael@0	42	%endif
michael@0	43
michael@0	44	;rows +=8;
michael@0	45	add dword ptr arg(2), 8
michael@0	46
michael@0	47	;for(c=0; c<cols; c+=4)
michael@0	48	.loop_col:
michael@0	49	mov rsi, arg(0) ;s
michael@0	50	pxor mm0, mm0 ;
michael@0	51
michael@0	52	movsxd rax, dword ptr arg(1) ;pitch ;
michael@0	53
michael@0	54	; this copies the last row down into the border 8 rows
michael@0	55	mov rdi, rsi
michael@0	56	mov rdx, arg(2)
michael@0	57	sub rdx, 9
michael@0	58	imul rdx, rax
michael@0	59	lea rdi, [rdi+rdx]
michael@0	60	movq mm1, QWORD ptr[rdi] ; first row
michael@0	61	mov rcx, 8
michael@0	62	.init_borderd ; initialize borders
michael@0	63	lea rdi, [rdi + rax]
michael@0	64	movq [rdi], mm1
michael@0	65
michael@0	66	dec rcx
michael@0	67	jne .init_borderd
michael@0	68
michael@0	69	neg rax ; rax = -pitch
michael@0	70
michael@0	71	; this copies the first row up into the border 8 rows
michael@0	72	mov rdi, rsi
michael@0	73	movq mm1, QWORD ptr[rdi] ; first row
michael@0	74	mov rcx, 8
michael@0	75	.init_border ; initialize borders
michael@0	76	lea rdi, [rdi + rax]
michael@0	77	movq [rdi], mm1
michael@0	78
michael@0	79	dec rcx
michael@0	80	jne .init_border
michael@0	81
michael@0	82
michael@0	83	lea rsi, [rsi + rax*8]; ; rdi = s[-pitch*8]
michael@0	84	neg rax
michael@0	85
michael@0	86
michael@0	87	pxor mm5, mm5
michael@0	88	pxor mm6, mm6 ;
michael@0	89
michael@0	90	pxor mm7, mm7 ;
michael@0	91	mov rdi, rsi
michael@0	92
michael@0	93	mov rcx, 15 ;
michael@0	94
michael@0	95	.loop_initvar:
michael@0	96	movd mm1, DWORD PTR [rdi];
michael@0	97	punpcklbw mm1, mm0 ;
michael@0	98
michael@0	99	paddw mm5, mm1 ;
michael@0	100	pmullw mm1, mm1 ;
michael@0	101
michael@0	102	movq mm2, mm1 ;
michael@0	103	punpcklwd mm1, mm0 ;
michael@0	104
michael@0	105	punpckhwd mm2, mm0 ;
michael@0	106	paddd mm6, mm1 ;
michael@0	107
michael@0	108	paddd mm7, mm2 ;
michael@0	109	lea rdi, [rdi+rax] ;
michael@0	110
michael@0	111	dec rcx
michael@0	112	jne .loop_initvar
michael@0	113	;save the var and sum
michael@0	114	xor rdx, rdx
michael@0	115	.loop_row:
michael@0	116	movd mm1, DWORD PTR [rsi] ; [s-pitch*8]
michael@0	117	movd mm2, DWORD PTR [rdi] ; [s+pitch*7]
michael@0	118
michael@0	119	punpcklbw mm1, mm0
michael@0	120	punpcklbw mm2, mm0
michael@0	121
michael@0	122	paddw mm5, mm2
michael@0	123	psubw mm5, mm1
michael@0	124
michael@0	125	pmullw mm2, mm2
michael@0	126	movq mm4, mm2
michael@0	127
michael@0	128	punpcklwd mm2, mm0
michael@0	129	punpckhwd mm4, mm0
michael@0	130
michael@0	131	paddd mm6, mm2
michael@0	132	paddd mm7, mm4
michael@0	133
michael@0	134	pmullw mm1, mm1
michael@0	135	movq mm2, mm1
michael@0	136
michael@0	137	punpcklwd mm1, mm0
michael@0	138	psubd mm6, mm1
michael@0	139
michael@0	140	punpckhwd mm2, mm0
michael@0	141	psubd mm7, mm2
michael@0	142
michael@0	143
michael@0	144	movq mm3, mm6
michael@0	145	pslld mm3, 4
michael@0	146
michael@0	147	psubd mm3, mm6
michael@0	148	movq mm1, mm5
michael@0	149
michael@0	150	movq mm4, mm5
michael@0	151	pmullw mm1, mm1
michael@0	152
michael@0	153	pmulhw mm4, mm4
michael@0	154	movq mm2, mm1
michael@0	155
michael@0	156	punpcklwd mm1, mm4
michael@0	157	punpckhwd mm2, mm4
michael@0	158
michael@0	159	movq mm4, mm7
michael@0	160	pslld mm4, 4
michael@0	161
michael@0	162	psubd mm4, mm7
michael@0	163
michael@0	164	psubd mm3, mm1
michael@0	165	psubd mm4, mm2
michael@0	166
michael@0	167	psubd mm3, flimit2
michael@0	168	psubd mm4, flimit2
michael@0	169
michael@0	170	psrad mm3, 31
michael@0	171	psrad mm4, 31
michael@0	172
michael@0	173	packssdw mm3, mm4
michael@0	174	packsswb mm3, mm0
michael@0	175
michael@0	176	movd mm1, DWORD PTR [rsi+rax*8]
michael@0	177
michael@0	178	movq mm2, mm1
michael@0	179	punpcklbw mm1, mm0
michael@0	180
michael@0	181	paddw mm1, mm5
michael@0	182	mov rcx, rdx
michael@0	183
michael@0	184	and rcx, 127
michael@0	185	%if ABI_IS_32BIT=1 && CONFIG_PIC=1
michael@0	186	push rax
michael@0	187	lea rax, [GLOBAL(sym(vp8_rv))]
michael@0	188	movq mm4, [rax + rcx*2] ;vp8_rv[rcx*2]
michael@0	189	pop rax
michael@0	190	%elif ABI_IS_32BIT=0
michael@0	191	movq mm4, [r8 + rcx*2] ;vp8_rv[rcx*2]
michael@0	192	%else
michael@0	193	movq mm4, [sym(vp8_rv) + rcx*2]
michael@0	194	%endif
michael@0	195	paddw mm1, mm4
michael@0	196	psraw mm1, 4
michael@0	197
michael@0	198	packuswb mm1, mm0
michael@0	199	pand mm1, mm3
michael@0	200
michael@0	201	pandn mm3, mm2
michael@0	202	por mm1, mm3
michael@0	203
michael@0	204	and rcx, 15
michael@0	205	movd DWORD PTR [rsp+rcx*4], mm1 ;d[rcx*4]
michael@0	206
michael@0	207	mov rcx, rdx
michael@0	208	sub rcx, 8
michael@0	209
michael@0	210	and rcx, 15
michael@0	211	movd mm1, DWORD PTR [rsp+rcx*4] ;d[rcx*4]
michael@0	212
michael@0	213	movd [rsi], mm1
michael@0	214	lea rsi, [rsi+rax]
michael@0	215
michael@0	216	lea rdi, [rdi+rax]
michael@0	217	add rdx, 1
michael@0	218
michael@0	219	cmp edx, dword arg(2) ;rows
michael@0	220	jl .loop_row
michael@0	221
michael@0	222
michael@0	223	add dword arg(0), 4 ; s += 4
michael@0	224	sub dword arg(3), 4 ; cols -= 4
michael@0	225	cmp dword arg(3), 0
michael@0	226	jg .loop_col
michael@0	227
michael@0	228	add rsp, 136
michael@0	229	pop rsp
michael@0	230
michael@0	231	; begin epilog
michael@0	232	pop rdi
michael@0	233	pop rsi
michael@0	234	RESTORE_GOT
michael@0	235	UNSHADOW_ARGS
michael@0	236	pop rbp
michael@0	237	ret
michael@0	238	%undef flimit2
michael@0	239
michael@0	240
michael@0	241	;void vp8_plane_add_noise_mmx (unsigned char *Start, unsigned char *noise,
michael@0	242	; unsigned char blackclamp[16],
michael@0	243	; unsigned char whiteclamp[16],
michael@0	244	; unsigned char bothclamp[16],
michael@0	245	; unsigned int Width, unsigned int Height, int Pitch)
michael@0	246	extern sym(rand)
michael@0	247	global sym(vp8_plane_add_noise_mmx) PRIVATE
michael@0	248	sym(vp8_plane_add_noise_mmx):
michael@0	249	push rbp
michael@0	250	mov rbp, rsp
michael@0	251	SHADOW_ARGS_TO_STACK 8
michael@0	252	GET_GOT rbx
michael@0	253	push rsi
michael@0	254	push rdi
michael@0	255	; end prolog
michael@0	256
michael@0	257	.addnoise_loop:
michael@0	258	call sym(rand) WRT_PLT
michael@0	259	mov rcx, arg(1) ;noise
michael@0	260	and rax, 0xff
michael@0	261	add rcx, rax
michael@0	262
michael@0	263	; we rely on the fact that the clamping vectors are stored contiguously
michael@0	264	; in black/white/both order. Note that we have to reload this here because
michael@0	265	; rdx could be trashed by rand()
michael@0	266	mov rdx, arg(2) ; blackclamp
michael@0	267
michael@0	268
michael@0	269	mov rdi, rcx
michael@0	270	movsxd rcx, dword arg(5) ;[Width]
michael@0	271	mov rsi, arg(0) ;Pos
michael@0	272	xor rax,rax
michael@0	273
michael@0	274	.addnoise_nextset:
michael@0	275	movq mm1,[rsi+rax] ; get the source
michael@0	276
michael@0	277	psubusb mm1, [rdx] ;blackclamp ; clamp both sides so we don't outrange adding noise
michael@0	278	paddusb mm1, [rdx+32] ;bothclamp
michael@0	279	psubusb mm1, [rdx+16] ;whiteclamp
michael@0	280
michael@0	281	movq mm2,[rdi+rax] ; get the noise for this line
michael@0	282	paddb mm1,mm2 ; add it in
michael@0	283	movq [rsi+rax],mm1 ; store the result
michael@0	284
michael@0	285	add rax,8 ; move to the next line
michael@0	286
michael@0	287	cmp rax, rcx
michael@0	288	jl .addnoise_nextset
michael@0	289
michael@0	290	movsxd rax, dword arg(7) ; Pitch
michael@0	291	add arg(0), rax ; Start += Pitch
michael@0	292	sub dword arg(6), 1 ; Height -= 1
michael@0	293	jg .addnoise_loop
michael@0	294
michael@0	295	; begin epilog
michael@0	296	pop rdi
michael@0	297	pop rsi
michael@0	298	RESTORE_GOT
michael@0	299	UNSHADOW_ARGS
michael@0	300	pop rbp
michael@0	301	ret
michael@0	302
michael@0	303
michael@0	304	SECTION_RODATA
michael@0	305	align 16
michael@0	306	Blur:
michael@0	307	times 16 dw 16
michael@0	308	times 8 dw 64
michael@0	309	times 16 dw 16
michael@0	310	times 8 dw 0
michael@0	311
michael@0	312	rd:
michael@0	313	times 4 dw 0x40