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

media/libvpx/vp9/encoder/x86/vp9_variance_impl_sse2.asm@ac0c01689b40 (annotated)

media/libvpx/vp9/encoder/x86/vp9_variance_impl_sse2.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	;unsigned int vp9_get_mb_ss_sse2
michael@0	15	;(
michael@0	16	; short *src_ptr
michael@0	17	;)
michael@0	18	global sym(vp9_get_mb_ss_sse2) PRIVATE
michael@0	19	sym(vp9_get_mb_ss_sse2):
michael@0	20	push rbp
michael@0	21	mov rbp, rsp
michael@0	22	SHADOW_ARGS_TO_STACK 1
michael@0	23	GET_GOT rbx
michael@0	24	push rsi
michael@0	25	push rdi
michael@0	26	sub rsp, 16
michael@0	27	; end prolog
michael@0	28
michael@0	29
michael@0	30	mov rax, arg(0) ;[src_ptr]
michael@0	31	mov rcx, 8
michael@0	32	pxor xmm4, xmm4
michael@0	33
michael@0	34	.NEXTROW:
michael@0	35	movdqa xmm0, [rax]
michael@0	36	movdqa xmm1, [rax+16]
michael@0	37	movdqa xmm2, [rax+32]
michael@0	38	movdqa xmm3, [rax+48]
michael@0	39	pmaddwd xmm0, xmm0
michael@0	40	pmaddwd xmm1, xmm1
michael@0	41	pmaddwd xmm2, xmm2
michael@0	42	pmaddwd xmm3, xmm3
michael@0	43
michael@0	44	paddd xmm0, xmm1
michael@0	45	paddd xmm2, xmm3
michael@0	46	paddd xmm4, xmm0
michael@0	47	paddd xmm4, xmm2
michael@0	48
michael@0	49	add rax, 0x40
michael@0	50	dec rcx
michael@0	51	ja .NEXTROW
michael@0	52
michael@0	53	movdqa xmm3,xmm4
michael@0	54	psrldq xmm4,8
michael@0	55	paddd xmm4,xmm3
michael@0	56	movdqa xmm3,xmm4
michael@0	57	psrldq xmm4,4
michael@0	58	paddd xmm4,xmm3
michael@0	59	movq rax,xmm4
michael@0	60
michael@0	61
michael@0	62	; begin epilog
michael@0	63	add rsp, 16
michael@0	64	pop rdi
michael@0	65	pop rsi
michael@0	66	RESTORE_GOT
michael@0	67	UNSHADOW_ARGS
michael@0	68	pop rbp
michael@0	69	ret
michael@0	70
michael@0	71
michael@0	72	;unsigned int vp9_get16x16var_sse2
michael@0	73	;(
michael@0	74	; unsigned char * src_ptr,
michael@0	75	; int source_stride,
michael@0	76	; unsigned char * ref_ptr,
michael@0	77	; int recon_stride,
michael@0	78	; unsigned int * SSE,
michael@0	79	; int * Sum
michael@0	80	;)
michael@0	81	global sym(vp9_get16x16var_sse2) PRIVATE
michael@0	82	sym(vp9_get16x16var_sse2):
michael@0	83	push rbp
michael@0	84	mov rbp, rsp
michael@0	85	SHADOW_ARGS_TO_STACK 6
michael@0	86	SAVE_XMM 7
michael@0	87	push rbx
michael@0	88	push rsi
michael@0	89	push rdi
michael@0	90	; end prolog
michael@0	91
michael@0	92	mov rsi, arg(0) ;[src_ptr]
michael@0	93	mov rdi, arg(2) ;[ref_ptr]
michael@0	94
michael@0	95	movsxd rax, DWORD PTR arg(1) ;[source_stride]
michael@0	96	movsxd rdx, DWORD PTR arg(3) ;[recon_stride]
michael@0	97
michael@0	98	; Prefetch data
michael@0	99	lea rcx, [rax+rax*2]
michael@0	100	prefetcht0 [rsi]
michael@0	101	prefetcht0 [rsi+rax]
michael@0	102	prefetcht0 [rsi+rax*2]
michael@0	103	prefetcht0 [rsi+rcx]
michael@0	104	lea rbx, [rsi+rax*4]
michael@0	105	prefetcht0 [rbx]
michael@0	106	prefetcht0 [rbx+rax]
michael@0	107	prefetcht0 [rbx+rax*2]
michael@0	108	prefetcht0 [rbx+rcx]
michael@0	109
michael@0	110	lea rcx, [rdx+rdx*2]
michael@0	111	prefetcht0 [rdi]
michael@0	112	prefetcht0 [rdi+rdx]
michael@0	113	prefetcht0 [rdi+rdx*2]
michael@0	114	prefetcht0 [rdi+rcx]
michael@0	115	lea rbx, [rdi+rdx*4]
michael@0	116	prefetcht0 [rbx]
michael@0	117	prefetcht0 [rbx+rdx]
michael@0	118	prefetcht0 [rbx+rdx*2]
michael@0	119	prefetcht0 [rbx+rcx]
michael@0	120
michael@0	121	pxor xmm0, xmm0 ; clear xmm0 for unpack
michael@0	122	pxor xmm7, xmm7 ; clear xmm7 for accumulating diffs
michael@0	123
michael@0	124	pxor xmm6, xmm6 ; clear xmm6 for accumulating sse
michael@0	125	mov rcx, 16
michael@0	126
michael@0	127	.var16loop:
michael@0	128	movdqu xmm1, XMMWORD PTR [rsi]
michael@0	129	movdqu xmm2, XMMWORD PTR [rdi]
michael@0	130
michael@0	131	prefetcht0 [rsi+rax*8]
michael@0	132	prefetcht0 [rdi+rdx*8]
michael@0	133
michael@0	134	movdqa xmm3, xmm1
michael@0	135	movdqa xmm4, xmm2
michael@0	136
michael@0	137
michael@0	138	punpcklbw xmm1, xmm0
michael@0	139	punpckhbw xmm3, xmm0
michael@0	140
michael@0	141	punpcklbw xmm2, xmm0
michael@0	142	punpckhbw xmm4, xmm0
michael@0	143
michael@0	144
michael@0	145	psubw xmm1, xmm2
michael@0	146	psubw xmm3, xmm4
michael@0	147
michael@0	148	paddw xmm7, xmm1
michael@0	149	pmaddwd xmm1, xmm1
michael@0	150
michael@0	151	paddw xmm7, xmm3
michael@0	152	pmaddwd xmm3, xmm3
michael@0	153
michael@0	154	paddd xmm6, xmm1
michael@0	155	paddd xmm6, xmm3
michael@0	156
michael@0	157	add rsi, rax
michael@0	158	add rdi, rdx
michael@0	159
michael@0	160	sub rcx, 1
michael@0	161	jnz .var16loop
michael@0	162
michael@0	163
michael@0	164	movdqa xmm1, xmm6
michael@0	165	pxor xmm6, xmm6
michael@0	166
michael@0	167	pxor xmm5, xmm5
michael@0	168	punpcklwd xmm6, xmm7
michael@0	169
michael@0	170	punpckhwd xmm5, xmm7
michael@0	171	psrad xmm5, 16
michael@0	172
michael@0	173	psrad xmm6, 16
michael@0	174	paddd xmm6, xmm5
michael@0	175
michael@0	176	movdqa xmm2, xmm1
michael@0	177	punpckldq xmm1, xmm0
michael@0	178
michael@0	179	punpckhdq xmm2, xmm0
michael@0	180	movdqa xmm7, xmm6
michael@0	181
michael@0	182	paddd xmm1, xmm2
michael@0	183	punpckldq xmm6, xmm0
michael@0	184
michael@0	185	punpckhdq xmm7, xmm0
michael@0	186	paddd xmm6, xmm7
michael@0	187
michael@0	188	movdqa xmm2, xmm1
michael@0	189	movdqa xmm7, xmm6
michael@0	190
michael@0	191	psrldq xmm1, 8
michael@0	192	psrldq xmm6, 8
michael@0	193
michael@0	194	paddd xmm7, xmm6
michael@0	195	paddd xmm1, xmm2
michael@0	196
michael@0	197	mov rax, arg(5) ;[Sum]
michael@0	198	mov rdi, arg(4) ;[SSE]
michael@0	199
michael@0	200	movd DWORD PTR [rax], xmm7
michael@0	201	movd DWORD PTR [rdi], xmm1
michael@0	202
michael@0	203
michael@0	204	; begin epilog
michael@0	205	pop rdi
michael@0	206	pop rsi
michael@0	207	pop rbx
michael@0	208	RESTORE_XMM
michael@0	209	UNSHADOW_ARGS
michael@0	210	pop rbp
michael@0	211	ret
michael@0	212
michael@0	213
michael@0	214
michael@0	215
michael@0	216	;unsigned int vp9_get8x8var_sse2
michael@0	217	;(
michael@0	218	; unsigned char * src_ptr,
michael@0	219	; int source_stride,
michael@0	220	; unsigned char * ref_ptr,
michael@0	221	; int recon_stride,
michael@0	222	; unsigned int * SSE,
michael@0	223	; int * Sum
michael@0	224	;)
michael@0	225	global sym(vp9_get8x8var_sse2) PRIVATE
michael@0	226	sym(vp9_get8x8var_sse2):
michael@0	227	push rbp
michael@0	228	mov rbp, rsp
michael@0	229	SHADOW_ARGS_TO_STACK 6
michael@0	230	SAVE_XMM 7
michael@0	231	GET_GOT rbx
michael@0	232	push rsi
michael@0	233	push rdi
michael@0	234	sub rsp, 16
michael@0	235	; end prolog
michael@0	236
michael@0	237	mov rsi, arg(0) ;[src_ptr]
michael@0	238	mov rdi, arg(2) ;[ref_ptr]
michael@0	239
michael@0	240	movsxd rax, DWORD PTR arg(1) ;[source_stride]
michael@0	241	movsxd rdx, DWORD PTR arg(3) ;[recon_stride]
michael@0	242
michael@0	243	pxor xmm0, xmm0 ; clear xmm0 for unpack
michael@0	244	pxor xmm7, xmm7 ; clear xmm7 for accumulating diffs
michael@0	245
michael@0	246	movq xmm1, QWORD PTR [rsi]
michael@0	247	movq xmm2, QWORD PTR [rdi]
michael@0	248
michael@0	249	punpcklbw xmm1, xmm0
michael@0	250	punpcklbw xmm2, xmm0
michael@0	251
michael@0	252	psubsw xmm1, xmm2
michael@0	253	paddw xmm7, xmm1
michael@0	254
michael@0	255	pmaddwd xmm1, xmm1
michael@0	256
michael@0	257	movq xmm2, QWORD PTR[rsi + rax]
michael@0	258	movq xmm3, QWORD PTR[rdi + rdx]
michael@0	259
michael@0	260	punpcklbw xmm2, xmm0
michael@0	261	punpcklbw xmm3, xmm0
michael@0	262
michael@0	263	psubsw xmm2, xmm3
michael@0	264	paddw xmm7, xmm2
michael@0	265
michael@0	266	pmaddwd xmm2, xmm2
michael@0	267	paddd xmm1, xmm2
michael@0	268
michael@0	269
michael@0	270	movq xmm2, QWORD PTR[rsi + rax * 2]
michael@0	271	movq xmm3, QWORD PTR[rdi + rdx * 2]
michael@0	272
michael@0	273	punpcklbw xmm2, xmm0
michael@0	274	punpcklbw xmm3, xmm0
michael@0	275
michael@0	276	psubsw xmm2, xmm3
michael@0	277	paddw xmm7, xmm2
michael@0	278
michael@0	279	pmaddwd xmm2, xmm2
michael@0	280	paddd xmm1, xmm2
michael@0	281
michael@0	282
michael@0	283	lea rsi, [rsi + rax * 2]
michael@0	284	lea rdi, [rdi + rdx * 2]
michael@0	285	movq xmm2, QWORD PTR[rsi + rax]
michael@0	286	movq xmm3, QWORD PTR[rdi + rdx]
michael@0	287
michael@0	288	punpcklbw xmm2, xmm0
michael@0	289	punpcklbw xmm3, xmm0
michael@0	290
michael@0	291	psubsw xmm2, xmm3
michael@0	292	paddw xmm7, xmm2
michael@0	293
michael@0	294	pmaddwd xmm2, xmm2
michael@0	295	paddd xmm1, xmm2
michael@0	296
michael@0	297	movq xmm2, QWORD PTR[rsi + rax *2]
michael@0	298	movq xmm3, QWORD PTR[rdi + rdx *2]
michael@0	299
michael@0	300	punpcklbw xmm2, xmm0
michael@0	301	punpcklbw xmm3, xmm0
michael@0	302
michael@0	303	psubsw xmm2, xmm3
michael@0	304	paddw xmm7, xmm2
michael@0	305
michael@0	306	pmaddwd xmm2, xmm2
michael@0	307	paddd xmm1, xmm2
michael@0	308
michael@0	309
michael@0	310	lea rsi, [rsi + rax * 2]
michael@0	311	lea rdi, [rdi + rdx * 2]
michael@0	312
michael@0	313
michael@0	314	movq xmm2, QWORD PTR[rsi + rax]
michael@0	315	movq xmm3, QWORD PTR[rdi + rdx]
michael@0	316
michael@0	317	punpcklbw xmm2, xmm0
michael@0	318	punpcklbw xmm3, xmm0
michael@0	319
michael@0	320	psubsw xmm2, xmm3
michael@0	321	paddw xmm7, xmm2
michael@0	322
michael@0	323	pmaddwd xmm2, xmm2
michael@0	324	paddd xmm1, xmm2
michael@0	325
michael@0	326	movq xmm2, QWORD PTR[rsi + rax *2]
michael@0	327	movq xmm3, QWORD PTR[rdi + rdx *2]
michael@0	328
michael@0	329	punpcklbw xmm2, xmm0
michael@0	330	punpcklbw xmm3, xmm0
michael@0	331
michael@0	332	psubsw xmm2, xmm3
michael@0	333	paddw xmm7, xmm2
michael@0	334
michael@0	335	pmaddwd xmm2, xmm2
michael@0	336	paddd xmm1, xmm2
michael@0	337
michael@0	338
michael@0	339	lea rsi, [rsi + rax * 2]
michael@0	340	lea rdi, [rdi + rdx * 2]
michael@0	341
michael@0	342	movq xmm2, QWORD PTR[rsi + rax]
michael@0	343	movq xmm3, QWORD PTR[rdi + rdx]
michael@0	344
michael@0	345	punpcklbw xmm2, xmm0
michael@0	346	punpcklbw xmm3, xmm0
michael@0	347
michael@0	348	psubsw xmm2, xmm3
michael@0	349	paddw xmm7, xmm2
michael@0	350
michael@0	351	pmaddwd xmm2, xmm2
michael@0	352	paddd xmm1, xmm2
michael@0	353
michael@0	354
michael@0	355	movdqa xmm6, xmm7
michael@0	356	punpcklwd xmm6, xmm0
michael@0	357
michael@0	358	punpckhwd xmm7, xmm0
michael@0	359	movdqa xmm2, xmm1
michael@0	360
michael@0	361	paddw xmm6, xmm7
michael@0	362	punpckldq xmm1, xmm0
michael@0	363
michael@0	364	punpckhdq xmm2, xmm0
michael@0	365	movdqa xmm7, xmm6
michael@0	366
michael@0	367	paddd xmm1, xmm2
michael@0	368	punpckldq xmm6, xmm0
michael@0	369
michael@0	370	punpckhdq xmm7, xmm0
michael@0	371	paddw xmm6, xmm7
michael@0	372
michael@0	373	movdqa xmm2, xmm1
michael@0	374	movdqa xmm7, xmm6
michael@0	375
michael@0	376	psrldq xmm1, 8
michael@0	377	psrldq xmm6, 8
michael@0	378
michael@0	379	paddw xmm7, xmm6
michael@0	380	paddd xmm1, xmm2
michael@0	381
michael@0	382	mov rax, arg(5) ;[Sum]
michael@0	383	mov rdi, arg(4) ;[SSE]
michael@0	384
michael@0	385	movq rdx, xmm7
michael@0	386	movsx rcx, dx
michael@0	387
michael@0	388	mov dword ptr [rax], ecx
michael@0	389	movd DWORD PTR [rdi], xmm1
michael@0	390
michael@0	391	; begin epilog
michael@0	392	add rsp, 16
michael@0	393	pop rdi
michael@0	394	pop rsi
michael@0	395	RESTORE_GOT
michael@0	396	RESTORE_XMM
michael@0	397	UNSHADOW_ARGS
michael@0	398	pop rbp
michael@0	399	ret
michael@0	400
michael@0	401	;void vp9_half_horiz_vert_variance8x_h_sse2
michael@0	402	;(
michael@0	403	; unsigned char *ref_ptr,
michael@0	404	; int ref_pixels_per_line,
michael@0	405	; unsigned char *src_ptr,
michael@0	406	; int src_pixels_per_line,
michael@0	407	; unsigned int Height,
michael@0	408	; int *sum,
michael@0	409	; unsigned int *sumsquared
michael@0	410	;)
michael@0	411	global sym(vp9_half_horiz_vert_variance8x_h_sse2) PRIVATE
michael@0	412	sym(vp9_half_horiz_vert_variance8x_h_sse2):
michael@0	413	push rbp
michael@0	414	mov rbp, rsp
michael@0	415	SHADOW_ARGS_TO_STACK 7
michael@0	416	SAVE_XMM 7
michael@0	417	GET_GOT rbx
michael@0	418	push rsi
michael@0	419	push rdi
michael@0	420	; end prolog
michael@0	421
michael@0	422	%if ABI_IS_32BIT=0
michael@0	423	movsxd r8, dword ptr arg(1) ;ref_pixels_per_line
michael@0	424	movsxd r9, dword ptr arg(3) ;src_pixels_per_line
michael@0	425	%endif
michael@0	426
michael@0	427	pxor xmm6, xmm6 ; error accumulator
michael@0	428	pxor xmm7, xmm7 ; sse eaccumulator
michael@0	429	mov rsi, arg(0) ;ref_ptr ;
michael@0	430
michael@0	431	mov rdi, arg(2) ;src_ptr ;
michael@0	432	movsxd rcx, dword ptr arg(4) ;Height ;
michael@0	433	movsxd rax, dword ptr arg(1) ;ref_pixels_per_line
michael@0	434
michael@0	435	pxor xmm0, xmm0 ;
michael@0	436
michael@0	437	movq xmm5, QWORD PTR [rsi] ; xmm5 = s0,s1,s2..s8
michael@0	438	movq xmm3, QWORD PTR [rsi+1] ; xmm3 = s1,s2,s3..s9
michael@0	439	pavgb xmm5, xmm3 ; xmm5 = avg(xmm1,xmm3) horizontal line 1
michael@0	440
michael@0	441	%if ABI_IS_32BIT
michael@0	442	add rsi, dword ptr arg(1) ;ref_pixels_per_line ; next source
michael@0	443	%else
michael@0	444	add rsi, r8
michael@0	445	%endif
michael@0	446
michael@0	447	.half_horiz_vert_variance8x_h_1:
michael@0	448
michael@0	449	movq xmm1, QWORD PTR [rsi] ;
michael@0	450	movq xmm2, QWORD PTR [rsi+1] ;
michael@0	451	pavgb xmm1, xmm2 ; xmm1 = avg(xmm1,xmm3) horizontal line i+1
michael@0	452
michael@0	453	pavgb xmm5, xmm1 ; xmm = vertical average of the above
michael@0	454	punpcklbw xmm5, xmm0 ; xmm5 = words of above
michael@0	455
michael@0	456	movq xmm3, QWORD PTR [rdi] ; xmm3 = d0,d1,d2..d8
michael@0	457	punpcklbw xmm3, xmm0 ; xmm3 = words of above
michael@0	458
michael@0	459	psubw xmm5, xmm3 ; xmm5 -= xmm3
michael@0	460	paddw xmm6, xmm5 ; xmm6 += accumulated column differences
michael@0	461	pmaddwd xmm5, xmm5 ; xmm5 *= xmm5
michael@0	462	paddd xmm7, xmm5 ; xmm7 += accumulated square column differences
michael@0	463
michael@0	464	movdqa xmm5, xmm1 ; save xmm1 for use on the next row
michael@0	465
michael@0	466	%if ABI_IS_32BIT
michael@0	467	add esi, dword ptr arg(1) ;ref_pixels_per_line ; next source
michael@0	468	add edi, dword ptr arg(3) ;src_pixels_per_line ; next destination
michael@0	469	%else
michael@0	470	add rsi, r8
michael@0	471	add rdi, r9
michael@0	472	%endif
michael@0	473
michael@0	474	sub rcx, 1 ;
michael@0	475	jnz .half_horiz_vert_variance8x_h_1 ;
michael@0	476
michael@0	477	movdq2q mm6, xmm6 ;
michael@0	478	movdq2q mm7, xmm7 ;
michael@0	479
michael@0	480	psrldq xmm6, 8
michael@0	481	psrldq xmm7, 8
michael@0	482
michael@0	483	movdq2q mm2, xmm6
michael@0	484	movdq2q mm3, xmm7
michael@0	485
michael@0	486	paddw mm6, mm2
michael@0	487	paddd mm7, mm3
michael@0	488
michael@0	489	pxor mm3, mm3 ;
michael@0	490	pxor mm2, mm2 ;
michael@0	491
michael@0	492	punpcklwd mm2, mm6 ;
michael@0	493	punpckhwd mm3, mm6 ;
michael@0	494
michael@0	495	paddd mm2, mm3 ;
michael@0	496	movq mm6, mm2 ;
michael@0	497
michael@0	498	psrlq mm6, 32 ;
michael@0	499	paddd mm2, mm6 ;
michael@0	500
michael@0	501	psrad mm2, 16 ;
michael@0	502	movq mm4, mm7 ;
michael@0	503
michael@0	504	psrlq mm4, 32 ;
michael@0	505	paddd mm4, mm7 ;
michael@0	506
michael@0	507	mov rsi, arg(5) ; sum
michael@0	508	mov rdi, arg(6) ; sumsquared
michael@0	509
michael@0	510	movd [rsi], mm2 ;
michael@0	511	movd [rdi], mm4 ;
michael@0	512
michael@0	513
michael@0	514	; begin epilog
michael@0	515	pop rdi
michael@0	516	pop rsi
michael@0	517	RESTORE_GOT
michael@0	518	RESTORE_XMM
michael@0	519	UNSHADOW_ARGS
michael@0	520	pop rbp
michael@0	521	ret
michael@0	522
michael@0	523	;void vp9_half_vert_variance8x_h_sse2
michael@0	524	;(
michael@0	525	; unsigned char *ref_ptr,
michael@0	526	; int ref_pixels_per_line,
michael@0	527	; unsigned char *src_ptr,
michael@0	528	; int src_pixels_per_line,
michael@0	529	; unsigned int Height,
michael@0	530	; int *sum,
michael@0	531	; unsigned int *sumsquared
michael@0	532	;)
michael@0	533	global sym(vp9_half_vert_variance8x_h_sse2) PRIVATE
michael@0	534	sym(vp9_half_vert_variance8x_h_sse2):
michael@0	535	push rbp
michael@0	536	mov rbp, rsp
michael@0	537	SHADOW_ARGS_TO_STACK 7
michael@0	538	SAVE_XMM 7
michael@0	539	GET_GOT rbx
michael@0	540	push rsi
michael@0	541	push rdi
michael@0	542	; end prolog
michael@0	543
michael@0	544	%if ABI_IS_32BIT=0
michael@0	545	movsxd r8, dword ptr arg(1) ;ref_pixels_per_line
michael@0	546	movsxd r9, dword ptr arg(3) ;src_pixels_per_line
michael@0	547	%endif
michael@0	548
michael@0	549	pxor xmm6, xmm6 ; error accumulator
michael@0	550	pxor xmm7, xmm7 ; sse eaccumulator
michael@0	551	mov rsi, arg(0) ;ref_ptr ;
michael@0	552
michael@0	553	mov rdi, arg(2) ;src_ptr ;
michael@0	554	movsxd rcx, dword ptr arg(4) ;Height ;
michael@0	555	movsxd rax, dword ptr arg(1) ;ref_pixels_per_line
michael@0	556
michael@0	557	pxor xmm0, xmm0 ;
michael@0	558	.half_vert_variance8x_h_1:
michael@0	559	movq xmm5, QWORD PTR [rsi] ; xmm5 = s0,s1,s2..s8
michael@0	560	movq xmm3, QWORD PTR [rsi+rax] ; xmm3 = s1,s2,s3..s9
michael@0	561
michael@0	562	pavgb xmm5, xmm3 ; xmm5 = avg(xmm1,xmm3)
michael@0	563	punpcklbw xmm5, xmm0 ; xmm5 = words of above
michael@0	564
michael@0	565	movq xmm3, QWORD PTR [rdi] ; xmm3 = d0,d1,d2..d8
michael@0	566	punpcklbw xmm3, xmm0 ; xmm3 = words of above
michael@0	567
michael@0	568	psubw xmm5, xmm3 ; xmm5 -= xmm3
michael@0	569	paddw xmm6, xmm5 ; xmm6 += accumulated column differences
michael@0	570	pmaddwd xmm5, xmm5 ; xmm5 *= xmm5
michael@0	571	paddd xmm7, xmm5 ; xmm7 += accumulated square column differences
michael@0	572
michael@0	573	%if ABI_IS_32BIT
michael@0	574	add esi, dword ptr arg(1) ;ref_pixels_per_line ; next source
michael@0	575	add edi, dword ptr arg(3) ;src_pixels_per_line ; next destination
michael@0	576	%else
michael@0	577	add rsi, r8
michael@0	578	add rdi, r9
michael@0	579	%endif
michael@0	580
michael@0	581	sub rcx, 1 ;
michael@0	582	jnz .half_vert_variance8x_h_1 ;
michael@0	583
michael@0	584	movdq2q mm6, xmm6 ;
michael@0	585	movdq2q mm7, xmm7 ;
michael@0	586
michael@0	587	psrldq xmm6, 8
michael@0	588	psrldq xmm7, 8
michael@0	589
michael@0	590	movdq2q mm2, xmm6
michael@0	591	movdq2q mm3, xmm7
michael@0	592
michael@0	593	paddw mm6, mm2
michael@0	594	paddd mm7, mm3
michael@0	595
michael@0	596	pxor mm3, mm3 ;
michael@0	597	pxor mm2, mm2 ;
michael@0	598
michael@0	599	punpcklwd mm2, mm6 ;
michael@0	600	punpckhwd mm3, mm6 ;
michael@0	601
michael@0	602	paddd mm2, mm3 ;
michael@0	603	movq mm6, mm2 ;
michael@0	604
michael@0	605	psrlq mm6, 32 ;
michael@0	606	paddd mm2, mm6 ;
michael@0	607
michael@0	608	psrad mm2, 16 ;
michael@0	609	movq mm4, mm7 ;
michael@0	610
michael@0	611	psrlq mm4, 32 ;
michael@0	612	paddd mm4, mm7 ;
michael@0	613
michael@0	614	mov rsi, arg(5) ; sum
michael@0	615	mov rdi, arg(6) ; sumsquared
michael@0	616
michael@0	617	movd [rsi], mm2 ;
michael@0	618	movd [rdi], mm4 ;
michael@0	619
michael@0	620
michael@0	621	; begin epilog
michael@0	622	pop rdi
michael@0	623	pop rsi
michael@0	624	RESTORE_GOT
michael@0	625	RESTORE_XMM
michael@0	626	UNSHADOW_ARGS
michael@0	627	pop rbp
michael@0	628	ret
michael@0	629
michael@0	630
michael@0	631	;void vp9_half_horiz_variance8x_h_sse2
michael@0	632	;(
michael@0	633	; unsigned char *ref_ptr,
michael@0	634	; int ref_pixels_per_line,
michael@0	635	; unsigned char *src_ptr,
michael@0	636	; int src_pixels_per_line,
michael@0	637	; unsigned int Height,
michael@0	638	; int *sum,
michael@0	639	; unsigned int *sumsquared
michael@0	640	;)
michael@0	641	global sym(vp9_half_horiz_variance8x_h_sse2) PRIVATE
michael@0	642	sym(vp9_half_horiz_variance8x_h_sse2):
michael@0	643	push rbp
michael@0	644	mov rbp, rsp
michael@0	645	SHADOW_ARGS_TO_STACK 7
michael@0	646	SAVE_XMM 7
michael@0	647	GET_GOT rbx
michael@0	648	push rsi
michael@0	649	push rdi
michael@0	650	; end prolog
michael@0	651
michael@0	652	%if ABI_IS_32BIT=0
michael@0	653	movsxd r8, dword ptr arg(1) ;ref_pixels_per_line
michael@0	654	movsxd r9, dword ptr arg(3) ;src_pixels_per_line
michael@0	655	%endif
michael@0	656
michael@0	657	pxor xmm6, xmm6 ; error accumulator
michael@0	658	pxor xmm7, xmm7 ; sse eaccumulator
michael@0	659	mov rsi, arg(0) ;ref_ptr ;
michael@0	660
michael@0	661	mov rdi, arg(2) ;src_ptr ;
michael@0	662	movsxd rcx, dword ptr arg(4) ;Height ;
michael@0	663
michael@0	664	pxor xmm0, xmm0 ;
michael@0	665	.half_horiz_variance8x_h_1:
michael@0	666	movq xmm5, QWORD PTR [rsi] ; xmm5 = s0,s1,s2..s8
michael@0	667	movq xmm3, QWORD PTR [rsi+1] ; xmm3 = s1,s2,s3..s9
michael@0	668
michael@0	669	pavgb xmm5, xmm3 ; xmm5 = avg(xmm1,xmm3)
michael@0	670	punpcklbw xmm5, xmm0 ; xmm5 = words of above
michael@0	671
michael@0	672	movq xmm3, QWORD PTR [rdi] ; xmm3 = d0,d1,d2..d8
michael@0	673	punpcklbw xmm3, xmm0 ; xmm3 = words of above
michael@0	674
michael@0	675	psubw xmm5, xmm3 ; xmm5 -= xmm3
michael@0	676	paddw xmm6, xmm5 ; xmm6 += accumulated column differences
michael@0	677	pmaddwd xmm5, xmm5 ; xmm5 *= xmm5
michael@0	678	paddd xmm7, xmm5 ; xmm7 += accumulated square column differences
michael@0	679
michael@0	680	%if ABI_IS_32BIT
michael@0	681	add esi, dword ptr arg(1) ;ref_pixels_per_line ; next source
michael@0	682	add edi, dword ptr arg(3) ;src_pixels_per_line ; next destination
michael@0	683	%else
michael@0	684	add rsi, r8
michael@0	685	add rdi, r9
michael@0	686	%endif
michael@0	687	sub rcx, 1 ;
michael@0	688	jnz .half_horiz_variance8x_h_1 ;
michael@0	689
michael@0	690	movdq2q mm6, xmm6 ;
michael@0	691	movdq2q mm7, xmm7 ;
michael@0	692
michael@0	693	psrldq xmm6, 8
michael@0	694	psrldq xmm7, 8
michael@0	695
michael@0	696	movdq2q mm2, xmm6
michael@0	697	movdq2q mm3, xmm7
michael@0	698
michael@0	699	paddw mm6, mm2
michael@0	700	paddd mm7, mm3
michael@0	701
michael@0	702	pxor mm3, mm3 ;
michael@0	703	pxor mm2, mm2 ;
michael@0	704
michael@0	705	punpcklwd mm2, mm6 ;
michael@0	706	punpckhwd mm3, mm6 ;
michael@0	707
michael@0	708	paddd mm2, mm3 ;
michael@0	709	movq mm6, mm2 ;
michael@0	710
michael@0	711	psrlq mm6, 32 ;
michael@0	712	paddd mm2, mm6 ;
michael@0	713
michael@0	714	psrad mm2, 16 ;
michael@0	715	movq mm4, mm7 ;
michael@0	716
michael@0	717	psrlq mm4, 32 ;
michael@0	718	paddd mm4, mm7 ;
michael@0	719
michael@0	720	mov rsi, arg(5) ; sum
michael@0	721	mov rdi, arg(6) ; sumsquared
michael@0	722
michael@0	723	movd [rsi], mm2 ;
michael@0	724	movd [rdi], mm4 ;
michael@0	725
michael@0	726
michael@0	727	; begin epilog
michael@0	728	pop rdi
michael@0	729	pop rsi
michael@0	730	RESTORE_GOT
michael@0	731	RESTORE_XMM
michael@0	732	UNSHADOW_ARGS
michael@0	733	pop rbp
michael@0	734	ret