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

media/libvpx/vp9/common/x86/vp9_postproc_mmx.asm@ac0c01689b40 (annotated)

media/libvpx/vp9/common/x86/vp9_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 VP9_FILTER_WEIGHT 128
michael@0	15	%define VP9_FILTER_SHIFT 7
michael@0	16
michael@0	17	;void vp9_post_proc_down_and_across_mmx
michael@0	18	;(
michael@0	19	; unsigned char *src_ptr,
michael@0	20	; unsigned char *dst_ptr,
michael@0	21	; int src_pixels_per_line,
michael@0	22	; int dst_pixels_per_line,
michael@0	23	; int rows,
michael@0	24	; int cols,
michael@0	25	; int flimit
michael@0	26	;)
michael@0	27	global sym(vp9_post_proc_down_and_across_mmx) PRIVATE
michael@0	28	sym(vp9_post_proc_down_and_across_mmx):
michael@0	29	push rbp
michael@0	30	mov rbp, rsp
michael@0	31	SHADOW_ARGS_TO_STACK 7
michael@0	32	GET_GOT rbx
michael@0	33	push rsi
michael@0	34	push rdi
michael@0	35	; end prolog
michael@0	36
michael@0	37	%if ABI_IS_32BIT=1 && CONFIG_PIC=1
michael@0	38	; move the global rd onto the stack, since we don't have enough registers
michael@0	39	; to do PIC addressing
michael@0	40	movq mm0, [GLOBAL(rd)]
michael@0	41	sub rsp, 8
michael@0	42	movq [rsp], mm0
michael@0	43	%define RD [rsp]
michael@0	44	%else
michael@0	45	%define RD [GLOBAL(rd)]
michael@0	46	%endif
michael@0	47
michael@0	48	push rbx
michael@0	49	lea rbx, [GLOBAL(Blur)]
michael@0	50	movd mm2, dword ptr arg(6) ;flimit
michael@0	51	punpcklwd mm2, mm2
michael@0	52	punpckldq mm2, mm2
michael@0	53
michael@0	54	mov rsi, arg(0) ;src_ptr
michael@0	55	mov rdi, arg(1) ;dst_ptr
michael@0	56
michael@0	57	movsxd rcx, DWORD PTR arg(4) ;rows
michael@0	58	movsxd rax, DWORD PTR arg(2) ;src_pixels_per_line ; destination pitch?
michael@0	59	pxor mm0, mm0 ; mm0 = 00000000
michael@0	60
michael@0	61	.nextrow:
michael@0	62
michael@0	63	xor rdx, rdx ; clear out rdx for use as loop counter
michael@0	64	.nextcol:
michael@0	65
michael@0	66	pxor mm7, mm7 ; mm7 = 00000000
michael@0	67	movq mm6, [rbx + 32 ] ; mm6 = kernel 2 taps
michael@0	68	movq mm3, [rsi] ; mm4 = r0 p0..p7
michael@0	69	punpcklbw mm3, mm0 ; mm3 = p0..p3
michael@0	70	movq mm1, mm3 ; mm1 = p0..p3
michael@0	71	pmullw mm3, mm6 ; mm3 *= kernel 2 modifiers
michael@0	72
michael@0	73	movq mm6, [rbx + 48] ; mm6 = kernel 3 taps
michael@0	74	movq mm5, [rsi + rax] ; mm4 = r1 p0..p7
michael@0	75	punpcklbw mm5, mm0 ; mm5 = r1 p0..p3
michael@0	76	pmullw mm6, mm5 ; mm6 *= p0..p3 * kernel 3 modifiers
michael@0	77	paddusw mm3, mm6 ; mm3 += mm6
michael@0	78
michael@0	79	; thresholding
michael@0	80	movq mm7, mm1 ; mm7 = r0 p0..p3
michael@0	81	psubusw mm7, mm5 ; mm7 = r0 p0..p3 - r1 p0..p3
michael@0	82	psubusw mm5, mm1 ; mm5 = r1 p0..p3 - r0 p0..p3
michael@0	83	paddusw mm7, mm5 ; mm7 = abs(r0 p0..p3 - r1 p0..p3)
michael@0	84	pcmpgtw mm7, mm2
michael@0	85
michael@0	86	movq mm6, [rbx + 64 ] ; mm6 = kernel 4 modifiers
michael@0	87	movq mm5, [rsi + 2*rax] ; mm4 = r2 p0..p7
michael@0	88	punpcklbw mm5, mm0 ; mm5 = r2 p0..p3
michael@0	89	pmullw mm6, mm5 ; mm5 *= kernel 4 modifiers
michael@0	90	paddusw mm3, mm6 ; mm3 += mm5
michael@0	91
michael@0	92	; thresholding
michael@0	93	movq mm6, mm1 ; mm6 = r0 p0..p3
michael@0	94	psubusw mm6, mm5 ; mm6 = r0 p0..p3 - r2 p0..p3
michael@0	95	psubusw mm5, mm1 ; mm5 = r2 p0..p3 - r2 p0..p3
michael@0	96	paddusw mm6, mm5 ; mm6 = abs(r0 p0..p3 - r2 p0..p3)
michael@0	97	pcmpgtw mm6, mm2
michael@0	98	por mm7, mm6 ; accumulate thresholds
michael@0	99
michael@0	100
michael@0	101	neg rax
michael@0	102	movq mm6, [rbx ] ; kernel 0 taps
michael@0	103	movq mm5, [rsi+2*rax] ; mm4 = r-2 p0..p7
michael@0	104	punpcklbw mm5, mm0 ; mm5 = r-2 p0..p3
michael@0	105	pmullw mm6, mm5 ; mm5 *= kernel 0 modifiers
michael@0	106	paddusw mm3, mm6 ; mm3 += mm5
michael@0	107
michael@0	108	; thresholding
michael@0	109	movq mm6, mm1 ; mm6 = r0 p0..p3
michael@0	110	psubusw mm6, mm5 ; mm6 = p0..p3 - r-2 p0..p3
michael@0	111	psubusw mm5, mm1 ; mm5 = r-2 p0..p3 - p0..p3
michael@0	112	paddusw mm6, mm5 ; mm6 = abs(r0 p0..p3 - r-2 p0..p3)
michael@0	113	pcmpgtw mm6, mm2
michael@0	114	por mm7, mm6 ; accumulate thresholds
michael@0	115
michael@0	116	movq mm6, [rbx + 16] ; kernel 1 taps
michael@0	117	movq mm4, [rsi+rax] ; mm4 = r-1 p0..p7
michael@0	118	punpcklbw mm4, mm0 ; mm4 = r-1 p0..p3
michael@0	119	pmullw mm6, mm4 ; mm4 *= kernel 1 modifiers.
michael@0	120	paddusw mm3, mm6 ; mm3 += mm5
michael@0	121
michael@0	122	; thresholding
michael@0	123	movq mm6, mm1 ; mm6 = r0 p0..p3
michael@0	124	psubusw mm6, mm4 ; mm6 = p0..p3 - r-2 p0..p3
michael@0	125	psubusw mm4, mm1 ; mm5 = r-1 p0..p3 - p0..p3
michael@0	126	paddusw mm6, mm4 ; mm6 = abs(r0 p0..p3 - r-1 p0..p3)
michael@0	127	pcmpgtw mm6, mm2
michael@0	128	por mm7, mm6 ; accumulate thresholds
michael@0	129
michael@0	130
michael@0	131	paddusw mm3, RD ; mm3 += round value
michael@0	132	psraw mm3, VP9_FILTER_SHIFT ; mm3 /= 128
michael@0	133
michael@0	134	pand mm1, mm7 ; mm1 select vals > thresh from source
michael@0	135	pandn mm7, mm3 ; mm7 select vals < thresh from blurred result
michael@0	136	paddusw mm1, mm7 ; combination
michael@0	137
michael@0	138	packuswb mm1, mm0 ; pack to bytes
michael@0	139
michael@0	140	movd [rdi], mm1 ;
michael@0	141	neg rax ; pitch is positive
michael@0	142
michael@0	143
michael@0	144	add rsi, 4
michael@0	145	add rdi, 4
michael@0	146	add rdx, 4
michael@0	147
michael@0	148	cmp edx, dword ptr arg(5) ;cols
michael@0	149	jl .nextcol
michael@0	150	; done with the all cols, start the across filtering in place
michael@0	151	sub rsi, rdx
michael@0	152	sub rdi, rdx
michael@0	153
michael@0	154
michael@0	155	push rax
michael@0	156	xor rdx, rdx
michael@0	157	mov rax, [rdi-4];
michael@0	158
michael@0	159	.acrossnextcol:
michael@0	160	pxor mm7, mm7 ; mm7 = 00000000
michael@0	161	movq mm6, [rbx + 32 ] ;
michael@0	162	movq mm4, [rdi+rdx] ; mm4 = p0..p7
michael@0	163	movq mm3, mm4 ; mm3 = p0..p7
michael@0	164	punpcklbw mm3, mm0 ; mm3 = p0..p3
michael@0	165	movq mm1, mm3 ; mm1 = p0..p3
michael@0	166	pmullw mm3, mm6 ; mm3 *= kernel 2 modifiers
michael@0	167
michael@0	168	movq mm6, [rbx + 48]
michael@0	169	psrlq mm4, 8 ; mm4 = p1..p7
michael@0	170	movq mm5, mm4 ; mm5 = p1..p7
michael@0	171	punpcklbw mm5, mm0 ; mm5 = p1..p4
michael@0	172	pmullw mm6, mm5 ; mm6 *= p1..p4 * kernel 3 modifiers
michael@0	173	paddusw mm3, mm6 ; mm3 += mm6
michael@0	174
michael@0	175	; thresholding
michael@0	176	movq mm7, mm1 ; mm7 = p0..p3
michael@0	177	psubusw mm7, mm5 ; mm7 = p0..p3 - p1..p4
michael@0	178	psubusw mm5, mm1 ; mm5 = p1..p4 - p0..p3
michael@0	179	paddusw mm7, mm5 ; mm7 = abs(p0..p3 - p1..p4)
michael@0	180	pcmpgtw mm7, mm2
michael@0	181
michael@0	182	movq mm6, [rbx + 64 ]
michael@0	183	psrlq mm4, 8 ; mm4 = p2..p7
michael@0	184	movq mm5, mm4 ; mm5 = p2..p7
michael@0	185	punpcklbw mm5, mm0 ; mm5 = p2..p5
michael@0	186	pmullw mm6, mm5 ; mm5 *= kernel 4 modifiers
michael@0	187	paddusw mm3, mm6 ; mm3 += mm5
michael@0	188
michael@0	189	; thresholding
michael@0	190	movq mm6, mm1 ; mm6 = p0..p3
michael@0	191	psubusw mm6, mm5 ; mm6 = p0..p3 - p1..p4
michael@0	192	psubusw mm5, mm1 ; mm5 = p1..p4 - p0..p3
michael@0	193	paddusw mm6, mm5 ; mm6 = abs(p0..p3 - p1..p4)
michael@0	194	pcmpgtw mm6, mm2
michael@0	195	por mm7, mm6 ; accumulate thresholds
michael@0	196
michael@0	197
michael@0	198	movq mm6, [rbx ]
michael@0	199	movq mm4, [rdi+rdx-2] ; mm4 = p-2..p5
michael@0	200	movq mm5, mm4 ; mm5 = p-2..p5
michael@0	201	punpcklbw mm5, mm0 ; mm5 = p-2..p1
michael@0	202	pmullw mm6, mm5 ; mm5 *= kernel 0 modifiers
michael@0	203	paddusw mm3, mm6 ; mm3 += mm5
michael@0	204
michael@0	205	; thresholding
michael@0	206	movq mm6, mm1 ; mm6 = p0..p3
michael@0	207	psubusw mm6, mm5 ; mm6 = p0..p3 - p1..p4
michael@0	208	psubusw mm5, mm1 ; mm5 = p1..p4 - p0..p3
michael@0	209	paddusw mm6, mm5 ; mm6 = abs(p0..p3 - p1..p4)
michael@0	210	pcmpgtw mm6, mm2
michael@0	211	por mm7, mm6 ; accumulate thresholds
michael@0	212
michael@0	213	movq mm6, [rbx + 16]
michael@0	214	psrlq mm4, 8 ; mm4 = p-1..p5
michael@0	215	punpcklbw mm4, mm0 ; mm4 = p-1..p2
michael@0	216	pmullw mm6, mm4 ; mm4 *= kernel 1 modifiers.
michael@0	217	paddusw mm3, mm6 ; mm3 += mm5
michael@0	218
michael@0	219	; thresholding
michael@0	220	movq mm6, mm1 ; mm6 = p0..p3
michael@0	221	psubusw mm6, mm4 ; mm6 = p0..p3 - p1..p4
michael@0	222	psubusw mm4, mm1 ; mm5 = p1..p4 - p0..p3
michael@0	223	paddusw mm6, mm4 ; mm6 = abs(p0..p3 - p1..p4)
michael@0	224	pcmpgtw mm6, mm2
michael@0	225	por mm7, mm6 ; accumulate thresholds
michael@0	226
michael@0	227	paddusw mm3, RD ; mm3 += round value
michael@0	228	psraw mm3, VP9_FILTER_SHIFT ; mm3 /= 128
michael@0	229
michael@0	230	pand mm1, mm7 ; mm1 select vals > thresh from source
michael@0	231	pandn mm7, mm3 ; mm7 select vals < thresh from blurred result
michael@0	232	paddusw mm1, mm7 ; combination
michael@0	233
michael@0	234	packuswb mm1, mm0 ; pack to bytes
michael@0	235	mov DWORD PTR [rdi+rdx-4], eax ; store previous four bytes
michael@0	236	movd eax, mm1
michael@0	237
michael@0	238	add rdx, 4
michael@0	239	cmp edx, dword ptr arg(5) ;cols
michael@0	240	jl .acrossnextcol;
michael@0	241
michael@0	242	mov DWORD PTR [rdi+rdx-4], eax
michael@0	243	pop rax
michael@0	244
michael@0	245	; done with this rwo
michael@0	246	add rsi,rax ; next line
michael@0	247	movsxd rax, dword ptr arg(3) ;dst_pixels_per_line ; destination pitch?
michael@0	248	add rdi,rax ; next destination
michael@0	249	movsxd rax, dword ptr arg(2) ;src_pixels_per_line ; destination pitch?
michael@0	250
michael@0	251	dec rcx ; decrement count
michael@0	252	jnz .nextrow ; next row
michael@0	253	pop rbx
michael@0	254
michael@0	255	; begin epilog
michael@0	256	pop rdi
michael@0	257	pop rsi
michael@0	258	RESTORE_GOT
michael@0	259	UNSHADOW_ARGS
michael@0	260	pop rbp
michael@0	261	ret
michael@0	262	%undef RD
michael@0	263
michael@0	264
michael@0	265	;void vp9_mbpost_proc_down_mmx(unsigned char *dst,
michael@0	266	; int pitch, int rows, int cols,int flimit)
michael@0	267	extern sym(vp9_rv)
michael@0	268	global sym(vp9_mbpost_proc_down_mmx) PRIVATE
michael@0	269	sym(vp9_mbpost_proc_down_mmx):
michael@0	270	push rbp
michael@0	271	mov rbp, rsp
michael@0	272	SHADOW_ARGS_TO_STACK 5
michael@0	273	GET_GOT rbx
michael@0	274	push rsi
michael@0	275	push rdi
michael@0	276	; end prolog
michael@0	277
michael@0	278	ALIGN_STACK 16, rax
michael@0	279	sub rsp, 136
michael@0	280
michael@0	281	; unsigned char d[16][8] at [rsp]
michael@0	282	; create flimit2 at [rsp+128]
michael@0	283	mov eax, dword ptr arg(4) ;flimit
michael@0	284	mov [rsp+128], eax
michael@0	285	mov [rsp+128+4], eax
michael@0	286	%define flimit2 [rsp+128]
michael@0	287
michael@0	288	%if ABI_IS_32BIT=0
michael@0	289	lea r8, [GLOBAL(sym(vp9_rv))]
michael@0	290	%endif
michael@0	291
michael@0	292	;rows +=8;
michael@0	293	add dword ptr arg(2), 8
michael@0	294
michael@0	295	;for(c=0; c<cols; c+=4)
michael@0	296	.loop_col:
michael@0	297	mov rsi, arg(0) ;s
michael@0	298	pxor mm0, mm0 ;
michael@0	299
michael@0	300	movsxd rax, dword ptr arg(1) ;pitch ;
michael@0	301	neg rax ; rax = -pitch
michael@0	302
michael@0	303	lea rsi, [rsi + rax*8]; ; rdi = s[-pitch*8]
michael@0	304	neg rax
michael@0	305
michael@0	306
michael@0	307	pxor mm5, mm5
michael@0	308	pxor mm6, mm6 ;
michael@0	309
michael@0	310	pxor mm7, mm7 ;
michael@0	311	mov rdi, rsi
michael@0	312
michael@0	313	mov rcx, 15 ;
michael@0	314
michael@0	315	.loop_initvar:
michael@0	316	movd mm1, DWORD PTR [rdi];
michael@0	317	punpcklbw mm1, mm0 ;
michael@0	318
michael@0	319	paddw mm5, mm1 ;
michael@0	320	pmullw mm1, mm1 ;
michael@0	321
michael@0	322	movq mm2, mm1 ;
michael@0	323	punpcklwd mm1, mm0 ;
michael@0	324
michael@0	325	punpckhwd mm2, mm0 ;
michael@0	326	paddd mm6, mm1 ;
michael@0	327
michael@0	328	paddd mm7, mm2 ;
michael@0	329	lea rdi, [rdi+rax] ;
michael@0	330
michael@0	331	dec rcx
michael@0	332	jne .loop_initvar
michael@0	333	;save the var and sum
michael@0	334	xor rdx, rdx
michael@0	335	.loop_row:
michael@0	336	movd mm1, DWORD PTR [rsi] ; [s-pitch*8]
michael@0	337	movd mm2, DWORD PTR [rdi] ; [s+pitch*7]
michael@0	338
michael@0	339	punpcklbw mm1, mm0
michael@0	340	punpcklbw mm2, mm0
michael@0	341
michael@0	342	paddw mm5, mm2
michael@0	343	psubw mm5, mm1
michael@0	344
michael@0	345	pmullw mm2, mm2
michael@0	346	movq mm4, mm2
michael@0	347
michael@0	348	punpcklwd mm2, mm0
michael@0	349	punpckhwd mm4, mm0
michael@0	350
michael@0	351	paddd mm6, mm2
michael@0	352	paddd mm7, mm4
michael@0	353
michael@0	354	pmullw mm1, mm1
michael@0	355	movq mm2, mm1
michael@0	356
michael@0	357	punpcklwd mm1, mm0
michael@0	358	psubd mm6, mm1
michael@0	359
michael@0	360	punpckhwd mm2, mm0
michael@0	361	psubd mm7, mm2
michael@0	362
michael@0	363
michael@0	364	movq mm3, mm6
michael@0	365	pslld mm3, 4
michael@0	366
michael@0	367	psubd mm3, mm6
michael@0	368	movq mm1, mm5
michael@0	369
michael@0	370	movq mm4, mm5
michael@0	371	pmullw mm1, mm1
michael@0	372
michael@0	373	pmulhw mm4, mm4
michael@0	374	movq mm2, mm1
michael@0	375
michael@0	376	punpcklwd mm1, mm4
michael@0	377	punpckhwd mm2, mm4
michael@0	378
michael@0	379	movq mm4, mm7
michael@0	380	pslld mm4, 4
michael@0	381
michael@0	382	psubd mm4, mm7
michael@0	383
michael@0	384	psubd mm3, mm1
michael@0	385	psubd mm4, mm2
michael@0	386
michael@0	387	psubd mm3, flimit2
michael@0	388	psubd mm4, flimit2
michael@0	389
michael@0	390	psrad mm3, 31
michael@0	391	psrad mm4, 31
michael@0	392
michael@0	393	packssdw mm3, mm4
michael@0	394	packsswb mm3, mm0
michael@0	395
michael@0	396	movd mm1, DWORD PTR [rsi+rax*8]
michael@0	397
michael@0	398	movq mm2, mm1
michael@0	399	punpcklbw mm1, mm0
michael@0	400
michael@0	401	paddw mm1, mm5
michael@0	402	mov rcx, rdx
michael@0	403
michael@0	404	and rcx, 127
michael@0	405	%if ABI_IS_32BIT=1 && CONFIG_PIC=1
michael@0	406	push rax
michael@0	407	lea rax, [GLOBAL(sym(vp9_rv))]
michael@0	408	movq mm4, [rax + rcx*2] ;vp9_rv[rcx*2]
michael@0	409	pop rax
michael@0	410	%elif ABI_IS_32BIT=0
michael@0	411	movq mm4, [r8 + rcx*2] ;vp9_rv[rcx*2]
michael@0	412	%else
michael@0	413	movq mm4, [sym(vp9_rv) + rcx*2]
michael@0	414	%endif
michael@0	415	paddw mm1, mm4
michael@0	416	;paddw xmm1, eight8s
michael@0	417	psraw mm1, 4
michael@0	418
michael@0	419	packuswb mm1, mm0
michael@0	420	pand mm1, mm3
michael@0	421
michael@0	422	pandn mm3, mm2
michael@0	423	por mm1, mm3
michael@0	424
michael@0	425	and rcx, 15
michael@0	426	movd DWORD PTR [rsp+rcx*4], mm1 ;d[rcx*4]
michael@0	427
michael@0	428	mov rcx, rdx
michael@0	429	sub rcx, 8
michael@0	430
michael@0	431	and rcx, 15
michael@0	432	movd mm1, DWORD PTR [rsp+rcx*4] ;d[rcx*4]
michael@0	433
michael@0	434	movd [rsi], mm1
michael@0	435	lea rsi, [rsi+rax]
michael@0	436
michael@0	437	lea rdi, [rdi+rax]
michael@0	438	add rdx, 1
michael@0	439
michael@0	440	cmp edx, dword arg(2) ;rows
michael@0	441	jl .loop_row
michael@0	442
michael@0	443
michael@0	444	add dword arg(0), 4 ; s += 4
michael@0	445	sub dword arg(3), 4 ; cols -= 4
michael@0	446	cmp dword arg(3), 0
michael@0	447	jg .loop_col
michael@0	448
michael@0	449	add rsp, 136
michael@0	450	pop rsp
michael@0	451
michael@0	452	; begin epilog
michael@0	453	pop rdi
michael@0	454	pop rsi
michael@0	455	RESTORE_GOT
michael@0	456	UNSHADOW_ARGS
michael@0	457	pop rbp
michael@0	458	ret
michael@0	459	%undef flimit2
michael@0	460
michael@0	461
michael@0	462	;void vp9_plane_add_noise_mmx (unsigned char *start, unsigned char *noise,
michael@0	463	; unsigned char blackclamp[16],
michael@0	464	; unsigned char whiteclamp[16],
michael@0	465	; unsigned char bothclamp[16],
michael@0	466	; unsigned int width, unsigned int height, int pitch)
michael@0	467	extern sym(rand)
michael@0	468	global sym(vp9_plane_add_noise_mmx) PRIVATE
michael@0	469	sym(vp9_plane_add_noise_mmx):
michael@0	470	push rbp
michael@0	471	mov rbp, rsp
michael@0	472	SHADOW_ARGS_TO_STACK 8
michael@0	473	GET_GOT rbx
michael@0	474	push rsi
michael@0	475	push rdi
michael@0	476	; end prolog
michael@0	477
michael@0	478	.addnoise_loop:
michael@0	479	call sym(rand) WRT_PLT
michael@0	480	mov rcx, arg(1) ;noise
michael@0	481	and rax, 0xff
michael@0	482	add rcx, rax
michael@0	483
michael@0	484	; we rely on the fact that the clamping vectors are stored contiguously
michael@0	485	; in black/white/both order. Note that we have to reload this here because
michael@0	486	; rdx could be trashed by rand()
michael@0	487	mov rdx, arg(2) ; blackclamp
michael@0	488
michael@0	489
michael@0	490	mov rdi, rcx
michael@0	491	movsxd rcx, dword arg(5) ;[Width]
michael@0	492	mov rsi, arg(0) ;Pos
michael@0	493	xor rax,rax
michael@0	494
michael@0	495	.addnoise_nextset:
michael@0	496	movq mm1,[rsi+rax] ; get the source
michael@0	497
michael@0	498	psubusb mm1, [rdx] ;blackclamp ; clamp both sides so we don't outrange adding noise
michael@0	499	paddusb mm1, [rdx+32] ;bothclamp
michael@0	500	psubusb mm1, [rdx+16] ;whiteclamp
michael@0	501
michael@0	502	movq mm2,[rdi+rax] ; get the noise for this line
michael@0	503	paddb mm1,mm2 ; add it in
michael@0	504	movq [rsi+rax],mm1 ; store the result
michael@0	505
michael@0	506	add rax,8 ; move to the next line
michael@0	507
michael@0	508	cmp rax, rcx
michael@0	509	jl .addnoise_nextset
michael@0	510
michael@0	511	movsxd rax, dword arg(7) ; Pitch
michael@0	512	add arg(0), rax ; Start += Pitch
michael@0	513	sub dword arg(6), 1 ; Height -= 1
michael@0	514	jg .addnoise_loop
michael@0	515
michael@0	516	; begin epilog
michael@0	517	pop rdi
michael@0	518	pop rsi
michael@0	519	RESTORE_GOT
michael@0	520	UNSHADOW_ARGS
michael@0	521	pop rbp
michael@0	522	ret
michael@0	523
michael@0	524
michael@0	525	SECTION_RODATA
michael@0	526	align 16
michael@0	527	Blur:
michael@0	528	times 16 dw 16
michael@0	529	times 8 dw 64
michael@0	530	times 16 dw 16
michael@0	531	times 8 dw 0
michael@0	532
michael@0	533	rd:
michael@0	534	times 4 dw 0x40