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

media/libvpx/vp9/common/x86/vp9_subpixel_8t_sse2.asm@6474c204b198 (annotated)

media/libvpx/vp9/common/x86/vp9_subpixel_8t_sse2.asm

Wed, 31 Dec 2014 06:09:35 +0100

author

Michael Schloh von Bennewitz <michael@schloh.com>

date

Wed, 31 Dec 2014 06:09:35 +0100

changeset 0

6474c204b198

permissions

-rw-r--r--

Cloned upstream origin tor-browser at tor-browser-31.3.0esr-4.5-1-build1
revision ID fc1c9ff7c1b2defdbc039f12214767608f46423f for hacking purpose.

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	;Note: tap3 and tap4 have to be applied and added after other taps to avoid
michael@0	15	;overflow.
michael@0	16
michael@0	17	%macro GET_FILTERS_4 0
michael@0	18	mov rdx, arg(5) ;filter ptr
michael@0	19	mov rcx, 0x0400040
michael@0	20
michael@0	21	movdqa xmm7, [rdx] ;load filters
michael@0	22	pshuflw xmm0, xmm7, 0b ;k0
michael@0	23	pshuflw xmm1, xmm7, 01010101b ;k1
michael@0	24	pshuflw xmm2, xmm7, 10101010b ;k2
michael@0	25	pshuflw xmm3, xmm7, 11111111b ;k3
michael@0	26	psrldq xmm7, 8
michael@0	27	pshuflw xmm4, xmm7, 0b ;k4
michael@0	28	pshuflw xmm5, xmm7, 01010101b ;k5
michael@0	29	pshuflw xmm6, xmm7, 10101010b ;k6
michael@0	30	pshuflw xmm7, xmm7, 11111111b ;k7
michael@0	31
michael@0	32	punpcklqdq xmm0, xmm1
michael@0	33	punpcklqdq xmm2, xmm3
michael@0	34	punpcklqdq xmm5, xmm4
michael@0	35	punpcklqdq xmm6, xmm7
michael@0	36
michael@0	37	movdqa k0k1, xmm0
michael@0	38	movdqa k2k3, xmm2
michael@0	39	movdqa k5k4, xmm5
michael@0	40	movdqa k6k7, xmm6
michael@0	41
michael@0	42	movq xmm6, rcx
michael@0	43	pshufd xmm6, xmm6, 0
michael@0	44	movdqa krd, xmm6
michael@0	45
michael@0	46	pxor xmm7, xmm7
michael@0	47	movdqa zero, xmm7
michael@0	48	%endm
michael@0	49
michael@0	50	%macro APPLY_FILTER_4 1
michael@0	51	punpckldq xmm0, xmm1 ;two row in one register
michael@0	52	punpckldq xmm6, xmm7
michael@0	53	punpckldq xmm2, xmm3
michael@0	54	punpckldq xmm5, xmm4
michael@0	55
michael@0	56	punpcklbw xmm0, zero ;unpack to word
michael@0	57	punpcklbw xmm6, zero
michael@0	58	punpcklbw xmm2, zero
michael@0	59	punpcklbw xmm5, zero
michael@0	60
michael@0	61	pmullw xmm0, k0k1 ;multiply the filter factors
michael@0	62	pmullw xmm6, k6k7
michael@0	63	pmullw xmm2, k2k3
michael@0	64	pmullw xmm5, k5k4
michael@0	65
michael@0	66	paddsw xmm0, xmm6 ;sum
michael@0	67	movdqa xmm1, xmm0
michael@0	68	psrldq xmm1, 8
michael@0	69	paddsw xmm0, xmm1
michael@0	70	paddsw xmm0, xmm2
michael@0	71	psrldq xmm2, 8
michael@0	72	paddsw xmm0, xmm5
michael@0	73	psrldq xmm5, 8
michael@0	74	paddsw xmm0, xmm2
michael@0	75	paddsw xmm0, xmm5
michael@0	76
michael@0	77	paddsw xmm0, krd ;rounding
michael@0	78	psraw xmm0, 7 ;shift
michael@0	79	packuswb xmm0, xmm0 ;pack to byte
michael@0	80
michael@0	81	%if %1
michael@0	82	movd xmm1, [rdi]
michael@0	83	pavgb xmm0, xmm1
michael@0	84	%endif
michael@0	85	movd [rdi], xmm0
michael@0	86	%endm
michael@0	87
michael@0	88	%macro GET_FILTERS 0
michael@0	89	mov rdx, arg(5) ;filter ptr
michael@0	90	mov rsi, arg(0) ;src_ptr
michael@0	91	mov rdi, arg(2) ;output_ptr
michael@0	92	mov rcx, 0x0400040
michael@0	93
michael@0	94	movdqa xmm7, [rdx] ;load filters
michael@0	95	pshuflw xmm0, xmm7, 0b ;k0
michael@0	96	pshuflw xmm1, xmm7, 01010101b ;k1
michael@0	97	pshuflw xmm2, xmm7, 10101010b ;k2
michael@0	98	pshuflw xmm3, xmm7, 11111111b ;k3
michael@0	99	pshufhw xmm4, xmm7, 0b ;k4
michael@0	100	pshufhw xmm5, xmm7, 01010101b ;k5
michael@0	101	pshufhw xmm6, xmm7, 10101010b ;k6
michael@0	102	pshufhw xmm7, xmm7, 11111111b ;k7
michael@0	103
michael@0	104	punpcklwd xmm0, xmm0
michael@0	105	punpcklwd xmm1, xmm1
michael@0	106	punpcklwd xmm2, xmm2
michael@0	107	punpcklwd xmm3, xmm3
michael@0	108	punpckhwd xmm4, xmm4
michael@0	109	punpckhwd xmm5, xmm5
michael@0	110	punpckhwd xmm6, xmm6
michael@0	111	punpckhwd xmm7, xmm7
michael@0	112
michael@0	113	movdqa k0, xmm0 ;store filter factors on stack
michael@0	114	movdqa k1, xmm1
michael@0	115	movdqa k2, xmm2
michael@0	116	movdqa k3, xmm3
michael@0	117	movdqa k4, xmm4
michael@0	118	movdqa k5, xmm5
michael@0	119	movdqa k6, xmm6
michael@0	120	movdqa k7, xmm7
michael@0	121
michael@0	122	movq xmm6, rcx
michael@0	123	pshufd xmm6, xmm6, 0
michael@0	124	movdqa krd, xmm6 ;rounding
michael@0	125
michael@0	126	pxor xmm7, xmm7
michael@0	127	movdqa zero, xmm7
michael@0	128	%endm
michael@0	129
michael@0	130	%macro LOAD_VERT_8 1
michael@0	131	movq xmm0, [rsi + %1] ;0
michael@0	132	movq xmm1, [rsi + rax + %1] ;1
michael@0	133	movq xmm6, [rsi + rdx * 2 + %1] ;6
michael@0	134	lea rsi, [rsi + rax]
michael@0	135	movq xmm7, [rsi + rdx * 2 + %1] ;7
michael@0	136	movq xmm2, [rsi + rax + %1] ;2
michael@0	137	movq xmm3, [rsi + rax * 2 + %1] ;3
michael@0	138	movq xmm4, [rsi + rdx + %1] ;4
michael@0	139	movq xmm5, [rsi + rax * 4 + %1] ;5
michael@0	140	%endm
michael@0	141
michael@0	142	%macro APPLY_FILTER_8 2
michael@0	143	punpcklbw xmm0, zero
michael@0	144	punpcklbw xmm1, zero
michael@0	145	punpcklbw xmm6, zero
michael@0	146	punpcklbw xmm7, zero
michael@0	147	punpcklbw xmm2, zero
michael@0	148	punpcklbw xmm5, zero
michael@0	149	punpcklbw xmm3, zero
michael@0	150	punpcklbw xmm4, zero
michael@0	151
michael@0	152	pmullw xmm0, k0
michael@0	153	pmullw xmm1, k1
michael@0	154	pmullw xmm6, k6
michael@0	155	pmullw xmm7, k7
michael@0	156	pmullw xmm2, k2
michael@0	157	pmullw xmm5, k5
michael@0	158	pmullw xmm3, k3
michael@0	159	pmullw xmm4, k4
michael@0	160
michael@0	161	paddsw xmm0, xmm1
michael@0	162	paddsw xmm0, xmm6
michael@0	163	paddsw xmm0, xmm7
michael@0	164	paddsw xmm0, xmm2
michael@0	165	paddsw xmm0, xmm5
michael@0	166	paddsw xmm0, xmm3
michael@0	167	paddsw xmm0, xmm4
michael@0	168
michael@0	169	paddsw xmm0, krd ;rounding
michael@0	170	psraw xmm0, 7 ;shift
michael@0	171	packuswb xmm0, xmm0 ;pack back to byte
michael@0	172	%if %1
michael@0	173	movq xmm1, [rdi + %2]
michael@0	174	pavgb xmm0, xmm1
michael@0	175	%endif
michael@0	176	movq [rdi + %2], xmm0
michael@0	177	%endm
michael@0	178
michael@0	179	;void vp9_filter_block1d4_v8_sse2
michael@0	180	;(
michael@0	181	; unsigned char *src_ptr,
michael@0	182	; unsigned int src_pitch,
michael@0	183	; unsigned char *output_ptr,
michael@0	184	; unsigned int out_pitch,
michael@0	185	; unsigned int output_height,
michael@0	186	; short *filter
michael@0	187	;)
michael@0	188	global sym(vp9_filter_block1d4_v8_sse2) PRIVATE
michael@0	189	sym(vp9_filter_block1d4_v8_sse2):
michael@0	190	push rbp
michael@0	191	mov rbp, rsp
michael@0	192	SHADOW_ARGS_TO_STACK 6
michael@0	193	SAVE_XMM 7
michael@0	194	push rsi
michael@0	195	push rdi
michael@0	196	push rbx
michael@0	197	; end prolog
michael@0	198
michael@0	199	ALIGN_STACK 16, rax
michael@0	200	sub rsp, 16 * 6
michael@0	201	%define k0k1 [rsp + 16 * 0]
michael@0	202	%define k2k3 [rsp + 16 * 1]
michael@0	203	%define k5k4 [rsp + 16 * 2]
michael@0	204	%define k6k7 [rsp + 16 * 3]
michael@0	205	%define krd [rsp + 16 * 4]
michael@0	206	%define zero [rsp + 16 * 5]
michael@0	207
michael@0	208	GET_FILTERS_4
michael@0	209
michael@0	210	mov rsi, arg(0) ;src_ptr
michael@0	211	mov rdi, arg(2) ;output_ptr
michael@0	212
michael@0	213	movsxd rax, DWORD PTR arg(1) ;pixels_per_line
michael@0	214	movsxd rbx, DWORD PTR arg(3) ;out_pitch
michael@0	215	lea rdx, [rax + rax * 2]
michael@0	216	movsxd rcx, DWORD PTR arg(4) ;output_height
michael@0	217
michael@0	218	.loop:
michael@0	219	movd xmm0, [rsi] ;load src: row 0
michael@0	220	movd xmm1, [rsi + rax] ;1
michael@0	221	movd xmm6, [rsi + rdx * 2] ;6
michael@0	222	lea rsi, [rsi + rax]
michael@0	223	movd xmm7, [rsi + rdx * 2] ;7
michael@0	224	movd xmm2, [rsi + rax] ;2
michael@0	225	movd xmm3, [rsi + rax * 2] ;3
michael@0	226	movd xmm4, [rsi + rdx] ;4
michael@0	227	movd xmm5, [rsi + rax * 4] ;5
michael@0	228
michael@0	229	APPLY_FILTER_4 0
michael@0	230
michael@0	231	lea rdi, [rdi + rbx]
michael@0	232	dec rcx
michael@0	233	jnz .loop
michael@0	234
michael@0	235	add rsp, 16 * 6
michael@0	236	pop rsp
michael@0	237	pop rbx
michael@0	238	; begin epilog
michael@0	239	pop rdi
michael@0	240	pop rsi
michael@0	241	RESTORE_XMM
michael@0	242	UNSHADOW_ARGS
michael@0	243	pop rbp
michael@0	244	ret
michael@0	245
michael@0	246	;void vp9_filter_block1d8_v8_sse2
michael@0	247	;(
michael@0	248	; unsigned char *src_ptr,
michael@0	249	; unsigned int src_pitch,
michael@0	250	; unsigned char *output_ptr,
michael@0	251	; unsigned int out_pitch,
michael@0	252	; unsigned int output_height,
michael@0	253	; short *filter
michael@0	254	;)
michael@0	255	global sym(vp9_filter_block1d8_v8_sse2) PRIVATE
michael@0	256	sym(vp9_filter_block1d8_v8_sse2):
michael@0	257	push rbp
michael@0	258	mov rbp, rsp
michael@0	259	SHADOW_ARGS_TO_STACK 6
michael@0	260	SAVE_XMM 7
michael@0	261	push rsi
michael@0	262	push rdi
michael@0	263	push rbx
michael@0	264	; end prolog
michael@0	265
michael@0	266	ALIGN_STACK 16, rax
michael@0	267	sub rsp, 16 * 10
michael@0	268	%define k0 [rsp + 16 * 0]
michael@0	269	%define k1 [rsp + 16 * 1]
michael@0	270	%define k2 [rsp + 16 * 2]
michael@0	271	%define k3 [rsp + 16 * 3]
michael@0	272	%define k4 [rsp + 16 * 4]
michael@0	273	%define k5 [rsp + 16 * 5]
michael@0	274	%define k6 [rsp + 16 * 6]
michael@0	275	%define k7 [rsp + 16 * 7]
michael@0	276	%define krd [rsp + 16 * 8]
michael@0	277	%define zero [rsp + 16 * 9]
michael@0	278
michael@0	279	GET_FILTERS
michael@0	280
michael@0	281	movsxd rax, DWORD PTR arg(1) ;pixels_per_line
michael@0	282	movsxd rbx, DWORD PTR arg(3) ;out_pitch
michael@0	283	lea rdx, [rax + rax * 2]
michael@0	284	movsxd rcx, DWORD PTR arg(4) ;output_height
michael@0	285
michael@0	286	.loop:
michael@0	287	LOAD_VERT_8 0
michael@0	288	APPLY_FILTER_8 0, 0
michael@0	289
michael@0	290	lea rdi, [rdi + rbx]
michael@0	291	dec rcx
michael@0	292	jnz .loop
michael@0	293
michael@0	294	add rsp, 16 * 10
michael@0	295	pop rsp
michael@0	296	pop rbx
michael@0	297	; begin epilog
michael@0	298	pop rdi
michael@0	299	pop rsi
michael@0	300	RESTORE_XMM
michael@0	301	UNSHADOW_ARGS
michael@0	302	pop rbp
michael@0	303	ret
michael@0	304
michael@0	305	;void vp9_filter_block1d16_v8_sse2
michael@0	306	;(
michael@0	307	; unsigned char *src_ptr,
michael@0	308	; unsigned int src_pitch,
michael@0	309	; unsigned char *output_ptr,
michael@0	310	; unsigned int out_pitch,
michael@0	311	; unsigned int output_height,
michael@0	312	; short *filter
michael@0	313	;)
michael@0	314	global sym(vp9_filter_block1d16_v8_sse2) PRIVATE
michael@0	315	sym(vp9_filter_block1d16_v8_sse2):
michael@0	316	push rbp
michael@0	317	mov rbp, rsp
michael@0	318	SHADOW_ARGS_TO_STACK 6
michael@0	319	SAVE_XMM 7
michael@0	320	push rsi
michael@0	321	push rdi
michael@0	322	push rbx
michael@0	323	; end prolog
michael@0	324
michael@0	325	ALIGN_STACK 16, rax
michael@0	326	sub rsp, 16 * 10
michael@0	327	%define k0 [rsp + 16 * 0]
michael@0	328	%define k1 [rsp + 16 * 1]
michael@0	329	%define k2 [rsp + 16 * 2]
michael@0	330	%define k3 [rsp + 16 * 3]
michael@0	331	%define k4 [rsp + 16 * 4]
michael@0	332	%define k5 [rsp + 16 * 5]
michael@0	333	%define k6 [rsp + 16 * 6]
michael@0	334	%define k7 [rsp + 16 * 7]
michael@0	335	%define krd [rsp + 16 * 8]
michael@0	336	%define zero [rsp + 16 * 9]
michael@0	337
michael@0	338	GET_FILTERS
michael@0	339
michael@0	340	movsxd rax, DWORD PTR arg(1) ;pixels_per_line
michael@0	341	movsxd rbx, DWORD PTR arg(3) ;out_pitch
michael@0	342	lea rdx, [rax + rax * 2]
michael@0	343	movsxd rcx, DWORD PTR arg(4) ;output_height
michael@0	344
michael@0	345	.loop:
michael@0	346	LOAD_VERT_8 0
michael@0	347	APPLY_FILTER_8 0, 0
michael@0	348	sub rsi, rax
michael@0	349
michael@0	350	LOAD_VERT_8 8
michael@0	351	APPLY_FILTER_8 0, 8
michael@0	352	add rdi, rbx
michael@0	353
michael@0	354	dec rcx
michael@0	355	jnz .loop
michael@0	356
michael@0	357	add rsp, 16 * 10
michael@0	358	pop rsp
michael@0	359	pop rbx
michael@0	360	; begin epilog
michael@0	361	pop rdi
michael@0	362	pop rsi
michael@0	363	RESTORE_XMM
michael@0	364	UNSHADOW_ARGS
michael@0	365	pop rbp
michael@0	366	ret
michael@0	367
michael@0	368	global sym(vp9_filter_block1d4_v8_avg_sse2) PRIVATE
michael@0	369	sym(vp9_filter_block1d4_v8_avg_sse2):
michael@0	370	push rbp
michael@0	371	mov rbp, rsp
michael@0	372	SHADOW_ARGS_TO_STACK 6
michael@0	373	SAVE_XMM 7
michael@0	374	push rsi
michael@0	375	push rdi
michael@0	376	push rbx
michael@0	377	; end prolog
michael@0	378
michael@0	379	ALIGN_STACK 16, rax
michael@0	380	sub rsp, 16 * 6
michael@0	381	%define k0k1 [rsp + 16 * 0]
michael@0	382	%define k2k3 [rsp + 16 * 1]
michael@0	383	%define k5k4 [rsp + 16 * 2]
michael@0	384	%define k6k7 [rsp + 16 * 3]
michael@0	385	%define krd [rsp + 16 * 4]
michael@0	386	%define zero [rsp + 16 * 5]
michael@0	387
michael@0	388	GET_FILTERS_4
michael@0	389
michael@0	390	mov rsi, arg(0) ;src_ptr
michael@0	391	mov rdi, arg(2) ;output_ptr
michael@0	392
michael@0	393	movsxd rax, DWORD PTR arg(1) ;pixels_per_line
michael@0	394	movsxd rbx, DWORD PTR arg(3) ;out_pitch
michael@0	395	lea rdx, [rax + rax * 2]
michael@0	396	movsxd rcx, DWORD PTR arg(4) ;output_height
michael@0	397
michael@0	398	.loop:
michael@0	399	movd xmm0, [rsi] ;load src: row 0
michael@0	400	movd xmm1, [rsi + rax] ;1
michael@0	401	movd xmm6, [rsi + rdx * 2] ;6
michael@0	402	lea rsi, [rsi + rax]
michael@0	403	movd xmm7, [rsi + rdx * 2] ;7
michael@0	404	movd xmm2, [rsi + rax] ;2
michael@0	405	movd xmm3, [rsi + rax * 2] ;3
michael@0	406	movd xmm4, [rsi + rdx] ;4
michael@0	407	movd xmm5, [rsi + rax * 4] ;5
michael@0	408
michael@0	409	APPLY_FILTER_4 1
michael@0	410
michael@0	411	lea rdi, [rdi + rbx]
michael@0	412	dec rcx
michael@0	413	jnz .loop
michael@0	414
michael@0	415	add rsp, 16 * 6
michael@0	416	pop rsp
michael@0	417	pop rbx
michael@0	418	; begin epilog
michael@0	419	pop rdi
michael@0	420	pop rsi
michael@0	421	RESTORE_XMM
michael@0	422	UNSHADOW_ARGS
michael@0	423	pop rbp
michael@0	424	ret
michael@0	425
michael@0	426	global sym(vp9_filter_block1d8_v8_avg_sse2) PRIVATE
michael@0	427	sym(vp9_filter_block1d8_v8_avg_sse2):
michael@0	428	push rbp
michael@0	429	mov rbp, rsp
michael@0	430	SHADOW_ARGS_TO_STACK 6
michael@0	431	SAVE_XMM 7
michael@0	432	push rsi
michael@0	433	push rdi
michael@0	434	push rbx
michael@0	435	; end prolog
michael@0	436
michael@0	437	ALIGN_STACK 16, rax
michael@0	438	sub rsp, 16 * 10
michael@0	439	%define k0 [rsp + 16 * 0]
michael@0	440	%define k1 [rsp + 16 * 1]
michael@0	441	%define k2 [rsp + 16 * 2]
michael@0	442	%define k3 [rsp + 16 * 3]
michael@0	443	%define k4 [rsp + 16 * 4]
michael@0	444	%define k5 [rsp + 16 * 5]
michael@0	445	%define k6 [rsp + 16 * 6]
michael@0	446	%define k7 [rsp + 16 * 7]
michael@0	447	%define krd [rsp + 16 * 8]
michael@0	448	%define zero [rsp + 16 * 9]
michael@0	449
michael@0	450	GET_FILTERS
michael@0	451
michael@0	452	movsxd rax, DWORD PTR arg(1) ;pixels_per_line
michael@0	453	movsxd rbx, DWORD PTR arg(3) ;out_pitch
michael@0	454	lea rdx, [rax + rax * 2]
michael@0	455	movsxd rcx, DWORD PTR arg(4) ;output_height
michael@0	456	.loop:
michael@0	457	LOAD_VERT_8 0
michael@0	458	APPLY_FILTER_8 1, 0
michael@0	459
michael@0	460	lea rdi, [rdi + rbx]
michael@0	461	dec rcx
michael@0	462	jnz .loop
michael@0	463
michael@0	464	add rsp, 16 * 10
michael@0	465	pop rsp
michael@0	466	pop rbx
michael@0	467	; begin epilog
michael@0	468	pop rdi
michael@0	469	pop rsi
michael@0	470	RESTORE_XMM
michael@0	471	UNSHADOW_ARGS
michael@0	472	pop rbp
michael@0	473	ret
michael@0	474
michael@0	475	global sym(vp9_filter_block1d16_v8_avg_sse2) PRIVATE
michael@0	476	sym(vp9_filter_block1d16_v8_avg_sse2):
michael@0	477	push rbp
michael@0	478	mov rbp, rsp
michael@0	479	SHADOW_ARGS_TO_STACK 6
michael@0	480	SAVE_XMM 7
michael@0	481	push rsi
michael@0	482	push rdi
michael@0	483	push rbx
michael@0	484	; end prolog
michael@0	485
michael@0	486	ALIGN_STACK 16, rax
michael@0	487	sub rsp, 16 * 10
michael@0	488	%define k0 [rsp + 16 * 0]
michael@0	489	%define k1 [rsp + 16 * 1]
michael@0	490	%define k2 [rsp + 16 * 2]
michael@0	491	%define k3 [rsp + 16 * 3]
michael@0	492	%define k4 [rsp + 16 * 4]
michael@0	493	%define k5 [rsp + 16 * 5]
michael@0	494	%define k6 [rsp + 16 * 6]
michael@0	495	%define k7 [rsp + 16 * 7]
michael@0	496	%define krd [rsp + 16 * 8]
michael@0	497	%define zero [rsp + 16 * 9]
michael@0	498
michael@0	499	GET_FILTERS
michael@0	500
michael@0	501	movsxd rax, DWORD PTR arg(1) ;pixels_per_line
michael@0	502	movsxd rbx, DWORD PTR arg(3) ;out_pitch
michael@0	503	lea rdx, [rax + rax * 2]
michael@0	504	movsxd rcx, DWORD PTR arg(4) ;output_height
michael@0	505	.loop:
michael@0	506	LOAD_VERT_8 0
michael@0	507	APPLY_FILTER_8 1, 0
michael@0	508	sub rsi, rax
michael@0	509
michael@0	510	LOAD_VERT_8 8
michael@0	511	APPLY_FILTER_8 1, 8
michael@0	512	add rdi, rbx
michael@0	513
michael@0	514	dec rcx
michael@0	515	jnz .loop
michael@0	516
michael@0	517	add rsp, 16 * 10
michael@0	518	pop rsp
michael@0	519	pop rbx
michael@0	520	; begin epilog
michael@0	521	pop rdi
michael@0	522	pop rsi
michael@0	523	RESTORE_XMM
michael@0	524	UNSHADOW_ARGS
michael@0	525	pop rbp
michael@0	526	ret
michael@0	527
michael@0	528	;void vp9_filter_block1d4_h8_sse2
michael@0	529	;(
michael@0	530	; unsigned char *src_ptr,
michael@0	531	; unsigned int src_pixels_per_line,
michael@0	532	; unsigned char *output_ptr,
michael@0	533	; unsigned int output_pitch,
michael@0	534	; unsigned int output_height,
michael@0	535	; short *filter
michael@0	536	;)
michael@0	537	global sym(vp9_filter_block1d4_h8_sse2) PRIVATE
michael@0	538	sym(vp9_filter_block1d4_h8_sse2):
michael@0	539	push rbp
michael@0	540	mov rbp, rsp
michael@0	541	SHADOW_ARGS_TO_STACK 6
michael@0	542	SAVE_XMM 7
michael@0	543	push rsi
michael@0	544	push rdi
michael@0	545	; end prolog
michael@0	546
michael@0	547	ALIGN_STACK 16, rax
michael@0	548	sub rsp, 16 * 6
michael@0	549	%define k0k1 [rsp + 16 * 0]
michael@0	550	%define k2k3 [rsp + 16 * 1]
michael@0	551	%define k5k4 [rsp + 16 * 2]
michael@0	552	%define k6k7 [rsp + 16 * 3]
michael@0	553	%define krd [rsp + 16 * 4]
michael@0	554	%define zero [rsp + 16 * 5]
michael@0	555
michael@0	556	GET_FILTERS_4
michael@0	557
michael@0	558	mov rsi, arg(0) ;src_ptr
michael@0	559	mov rdi, arg(2) ;output_ptr
michael@0	560
michael@0	561	movsxd rax, DWORD PTR arg(1) ;pixels_per_line
michael@0	562	movsxd rdx, DWORD PTR arg(3) ;out_pitch
michael@0	563	movsxd rcx, DWORD PTR arg(4) ;output_height
michael@0	564
michael@0	565	.loop:
michael@0	566	movdqu xmm0, [rsi - 3] ;load src
michael@0	567
michael@0	568	movdqa xmm1, xmm0
michael@0	569	movdqa xmm6, xmm0
michael@0	570	movdqa xmm7, xmm0
michael@0	571	movdqa xmm2, xmm0
michael@0	572	movdqa xmm3, xmm0
michael@0	573	movdqa xmm5, xmm0
michael@0	574	movdqa xmm4, xmm0
michael@0	575
michael@0	576	psrldq xmm1, 1
michael@0	577	psrldq xmm6, 6
michael@0	578	psrldq xmm7, 7
michael@0	579	psrldq xmm2, 2
michael@0	580	psrldq xmm3, 3
michael@0	581	psrldq xmm5, 5
michael@0	582	psrldq xmm4, 4
michael@0	583
michael@0	584	APPLY_FILTER_4 0
michael@0	585
michael@0	586	lea rsi, [rsi + rax]
michael@0	587	lea rdi, [rdi + rdx]
michael@0	588	dec rcx
michael@0	589	jnz .loop
michael@0	590
michael@0	591	add rsp, 16 * 6
michael@0	592	pop rsp
michael@0	593
michael@0	594	; begin epilog
michael@0	595	pop rdi
michael@0	596	pop rsi
michael@0	597	RESTORE_XMM
michael@0	598	UNSHADOW_ARGS
michael@0	599	pop rbp
michael@0	600	ret
michael@0	601
michael@0	602	;void vp9_filter_block1d8_h8_sse2
michael@0	603	;(
michael@0	604	; unsigned char *src_ptr,
michael@0	605	; unsigned int src_pixels_per_line,
michael@0	606	; unsigned char *output_ptr,
michael@0	607	; unsigned int output_pitch,
michael@0	608	; unsigned int output_height,
michael@0	609	; short *filter
michael@0	610	;)
michael@0	611	global sym(vp9_filter_block1d8_h8_sse2) PRIVATE
michael@0	612	sym(vp9_filter_block1d8_h8_sse2):
michael@0	613	push rbp
michael@0	614	mov rbp, rsp
michael@0	615	SHADOW_ARGS_TO_STACK 6
michael@0	616	SAVE_XMM 7
michael@0	617	push rsi
michael@0	618	push rdi
michael@0	619	; end prolog
michael@0	620
michael@0	621	ALIGN_STACK 16, rax
michael@0	622	sub rsp, 16 * 10
michael@0	623	%define k0 [rsp + 16 * 0]
michael@0	624	%define k1 [rsp + 16 * 1]
michael@0	625	%define k2 [rsp + 16 * 2]
michael@0	626	%define k3 [rsp + 16 * 3]
michael@0	627	%define k4 [rsp + 16 * 4]
michael@0	628	%define k5 [rsp + 16 * 5]
michael@0	629	%define k6 [rsp + 16 * 6]
michael@0	630	%define k7 [rsp + 16 * 7]
michael@0	631	%define krd [rsp + 16 * 8]
michael@0	632	%define zero [rsp + 16 * 9]
michael@0	633
michael@0	634	GET_FILTERS
michael@0	635
michael@0	636	movsxd rax, DWORD PTR arg(1) ;pixels_per_line
michael@0	637	movsxd rdx, DWORD PTR arg(3) ;out_pitch
michael@0	638	movsxd rcx, DWORD PTR arg(4) ;output_height
michael@0	639
michael@0	640	.loop:
michael@0	641	movdqu xmm0, [rsi - 3] ;load src
michael@0	642
michael@0	643	movdqa xmm1, xmm0
michael@0	644	movdqa xmm6, xmm0
michael@0	645	movdqa xmm7, xmm0
michael@0	646	movdqa xmm2, xmm0
michael@0	647	movdqa xmm5, xmm0
michael@0	648	movdqa xmm3, xmm0
michael@0	649	movdqa xmm4, xmm0
michael@0	650
michael@0	651	psrldq xmm1, 1
michael@0	652	psrldq xmm6, 6
michael@0	653	psrldq xmm7, 7
michael@0	654	psrldq xmm2, 2
michael@0	655	psrldq xmm5, 5
michael@0	656	psrldq xmm3, 3
michael@0	657	psrldq xmm4, 4
michael@0	658
michael@0	659	APPLY_FILTER_8 0, 0
michael@0	660
michael@0	661	lea rsi, [rsi + rax]
michael@0	662	lea rdi, [rdi + rdx]
michael@0	663	dec rcx
michael@0	664	jnz .loop
michael@0	665
michael@0	666	add rsp, 16 * 10
michael@0	667	pop rsp
michael@0	668
michael@0	669	; begin epilog
michael@0	670	pop rdi
michael@0	671	pop rsi
michael@0	672	RESTORE_XMM
michael@0	673	UNSHADOW_ARGS
michael@0	674	pop rbp
michael@0	675	ret
michael@0	676
michael@0	677	;void vp9_filter_block1d16_h8_sse2
michael@0	678	;(
michael@0	679	; unsigned char *src_ptr,
michael@0	680	; unsigned int src_pixels_per_line,
michael@0	681	; unsigned char *output_ptr,
michael@0	682	; unsigned int output_pitch,
michael@0	683	; unsigned int output_height,
michael@0	684	; short *filter
michael@0	685	;)
michael@0	686	global sym(vp9_filter_block1d16_h8_sse2) PRIVATE
michael@0	687	sym(vp9_filter_block1d16_h8_sse2):
michael@0	688	push rbp
michael@0	689	mov rbp, rsp
michael@0	690	SHADOW_ARGS_TO_STACK 6
michael@0	691	SAVE_XMM 7
michael@0	692	push rsi
michael@0	693	push rdi
michael@0	694	; end prolog
michael@0	695
michael@0	696	ALIGN_STACK 16, rax
michael@0	697	sub rsp, 16 * 10
michael@0	698	%define k0 [rsp + 16 * 0]
michael@0	699	%define k1 [rsp + 16 * 1]
michael@0	700	%define k2 [rsp + 16 * 2]
michael@0	701	%define k3 [rsp + 16 * 3]
michael@0	702	%define k4 [rsp + 16 * 4]
michael@0	703	%define k5 [rsp + 16 * 5]
michael@0	704	%define k6 [rsp + 16 * 6]
michael@0	705	%define k7 [rsp + 16 * 7]
michael@0	706	%define krd [rsp + 16 * 8]
michael@0	707	%define zero [rsp + 16 * 9]
michael@0	708
michael@0	709	GET_FILTERS
michael@0	710
michael@0	711	movsxd rax, DWORD PTR arg(1) ;pixels_per_line
michael@0	712	movsxd rdx, DWORD PTR arg(3) ;out_pitch
michael@0	713	movsxd rcx, DWORD PTR arg(4) ;output_height
michael@0	714
michael@0	715	.loop:
michael@0	716	movdqu xmm0, [rsi - 3] ;load src
michael@0	717
michael@0	718	movdqa xmm1, xmm0
michael@0	719	movdqa xmm6, xmm0
michael@0	720	movdqa xmm7, xmm0
michael@0	721	movdqa xmm2, xmm0
michael@0	722	movdqa xmm5, xmm0
michael@0	723	movdqa xmm3, xmm0
michael@0	724	movdqa xmm4, xmm0
michael@0	725
michael@0	726	psrldq xmm1, 1
michael@0	727	psrldq xmm6, 6
michael@0	728	psrldq xmm7, 7
michael@0	729	psrldq xmm2, 2
michael@0	730	psrldq xmm5, 5
michael@0	731	psrldq xmm3, 3
michael@0	732	psrldq xmm4, 4
michael@0	733
michael@0	734	APPLY_FILTER_8 0, 0
michael@0	735
michael@0	736	movdqu xmm0, [rsi + 5] ;load src
michael@0	737
michael@0	738	movdqa xmm1, xmm0
michael@0	739	movdqa xmm6, xmm0
michael@0	740	movdqa xmm7, xmm0
michael@0	741	movdqa xmm2, xmm0
michael@0	742	movdqa xmm5, xmm0
michael@0	743	movdqa xmm3, xmm0
michael@0	744	movdqa xmm4, xmm0
michael@0	745
michael@0	746	psrldq xmm1, 1
michael@0	747	psrldq xmm6, 6
michael@0	748	psrldq xmm7, 7
michael@0	749	psrldq xmm2, 2
michael@0	750	psrldq xmm5, 5
michael@0	751	psrldq xmm3, 3
michael@0	752	psrldq xmm4, 4
michael@0	753
michael@0	754	APPLY_FILTER_8 0, 8
michael@0	755
michael@0	756	lea rsi, [rsi + rax]
michael@0	757	lea rdi, [rdi + rdx]
michael@0	758	dec rcx
michael@0	759	jnz .loop
michael@0	760
michael@0	761	add rsp, 16 * 10
michael@0	762	pop rsp
michael@0	763
michael@0	764	; begin epilog
michael@0	765	pop rdi
michael@0	766	pop rsi
michael@0	767	RESTORE_XMM
michael@0	768	UNSHADOW_ARGS
michael@0	769	pop rbp
michael@0	770	ret
michael@0	771
michael@0	772	global sym(vp9_filter_block1d4_h8_avg_sse2) PRIVATE
michael@0	773	sym(vp9_filter_block1d4_h8_avg_sse2):
michael@0	774	push rbp
michael@0	775	mov rbp, rsp
michael@0	776	SHADOW_ARGS_TO_STACK 6
michael@0	777	SAVE_XMM 7
michael@0	778	push rsi
michael@0	779	push rdi
michael@0	780	; end prolog
michael@0	781
michael@0	782	ALIGN_STACK 16, rax
michael@0	783	sub rsp, 16 * 6
michael@0	784	%define k0k1 [rsp + 16 * 0]
michael@0	785	%define k2k3 [rsp + 16 * 1]
michael@0	786	%define k5k4 [rsp + 16 * 2]
michael@0	787	%define k6k7 [rsp + 16 * 3]
michael@0	788	%define krd [rsp + 16 * 4]
michael@0	789	%define zero [rsp + 16 * 5]
michael@0	790
michael@0	791	GET_FILTERS_4
michael@0	792
michael@0	793	mov rsi, arg(0) ;src_ptr
michael@0	794	mov rdi, arg(2) ;output_ptr
michael@0	795
michael@0	796	movsxd rax, DWORD PTR arg(1) ;pixels_per_line
michael@0	797	movsxd rdx, DWORD PTR arg(3) ;out_pitch
michael@0	798	movsxd rcx, DWORD PTR arg(4) ;output_height
michael@0	799
michael@0	800	.loop:
michael@0	801	movdqu xmm0, [rsi - 3] ;load src
michael@0	802
michael@0	803	movdqa xmm1, xmm0
michael@0	804	movdqa xmm6, xmm0
michael@0	805	movdqa xmm7, xmm0
michael@0	806	movdqa xmm2, xmm0
michael@0	807	movdqa xmm3, xmm0
michael@0	808	movdqa xmm5, xmm0
michael@0	809	movdqa xmm4, xmm0
michael@0	810
michael@0	811	psrldq xmm1, 1
michael@0	812	psrldq xmm6, 6
michael@0	813	psrldq xmm7, 7
michael@0	814	psrldq xmm2, 2
michael@0	815	psrldq xmm3, 3
michael@0	816	psrldq xmm5, 5
michael@0	817	psrldq xmm4, 4
michael@0	818
michael@0	819	APPLY_FILTER_4 1
michael@0	820
michael@0	821	lea rsi, [rsi + rax]
michael@0	822	lea rdi, [rdi + rdx]
michael@0	823	dec rcx
michael@0	824	jnz .loop
michael@0	825
michael@0	826	add rsp, 16 * 6
michael@0	827	pop rsp
michael@0	828
michael@0	829	; begin epilog
michael@0	830	pop rdi
michael@0	831	pop rsi
michael@0	832	RESTORE_XMM
michael@0	833	UNSHADOW_ARGS
michael@0	834	pop rbp
michael@0	835	ret
michael@0	836
michael@0	837	global sym(vp9_filter_block1d8_h8_avg_sse2) PRIVATE
michael@0	838	sym(vp9_filter_block1d8_h8_avg_sse2):
michael@0	839	push rbp
michael@0	840	mov rbp, rsp
michael@0	841	SHADOW_ARGS_TO_STACK 6
michael@0	842	SAVE_XMM 7
michael@0	843	push rsi
michael@0	844	push rdi
michael@0	845	; end prolog
michael@0	846
michael@0	847	ALIGN_STACK 16, rax
michael@0	848	sub rsp, 16 * 10
michael@0	849	%define k0 [rsp + 16 * 0]
michael@0	850	%define k1 [rsp + 16 * 1]
michael@0	851	%define k2 [rsp + 16 * 2]
michael@0	852	%define k3 [rsp + 16 * 3]
michael@0	853	%define k4 [rsp + 16 * 4]
michael@0	854	%define k5 [rsp + 16 * 5]
michael@0	855	%define k6 [rsp + 16 * 6]
michael@0	856	%define k7 [rsp + 16 * 7]
michael@0	857	%define krd [rsp + 16 * 8]
michael@0	858	%define zero [rsp + 16 * 9]
michael@0	859
michael@0	860	GET_FILTERS
michael@0	861
michael@0	862	movsxd rax, DWORD PTR arg(1) ;pixels_per_line
michael@0	863	movsxd rdx, DWORD PTR arg(3) ;out_pitch
michael@0	864	movsxd rcx, DWORD PTR arg(4) ;output_height
michael@0	865
michael@0	866	.loop:
michael@0	867	movdqu xmm0, [rsi - 3] ;load src
michael@0	868
michael@0	869	movdqa xmm1, xmm0
michael@0	870	movdqa xmm6, xmm0
michael@0	871	movdqa xmm7, xmm0
michael@0	872	movdqa xmm2, xmm0
michael@0	873	movdqa xmm5, xmm0
michael@0	874	movdqa xmm3, xmm0
michael@0	875	movdqa xmm4, xmm0
michael@0	876
michael@0	877	psrldq xmm1, 1
michael@0	878	psrldq xmm6, 6
michael@0	879	psrldq xmm7, 7
michael@0	880	psrldq xmm2, 2
michael@0	881	psrldq xmm5, 5
michael@0	882	psrldq xmm3, 3
michael@0	883	psrldq xmm4, 4
michael@0	884
michael@0	885	APPLY_FILTER_8 1, 0
michael@0	886
michael@0	887	lea rsi, [rsi + rax]
michael@0	888	lea rdi, [rdi + rdx]
michael@0	889	dec rcx
michael@0	890	jnz .loop
michael@0	891
michael@0	892	add rsp, 16 * 10
michael@0	893	pop rsp
michael@0	894
michael@0	895	; begin epilog
michael@0	896	pop rdi
michael@0	897	pop rsi
michael@0	898	RESTORE_XMM
michael@0	899	UNSHADOW_ARGS
michael@0	900	pop rbp
michael@0	901	ret
michael@0	902
michael@0	903	global sym(vp9_filter_block1d16_h8_avg_sse2) PRIVATE
michael@0	904	sym(vp9_filter_block1d16_h8_avg_sse2):
michael@0	905	push rbp
michael@0	906	mov rbp, rsp
michael@0	907	SHADOW_ARGS_TO_STACK 6
michael@0	908	SAVE_XMM 7
michael@0	909	push rsi
michael@0	910	push rdi
michael@0	911	; end prolog
michael@0	912
michael@0	913	ALIGN_STACK 16, rax
michael@0	914	sub rsp, 16 * 10
michael@0	915	%define k0 [rsp + 16 * 0]
michael@0	916	%define k1 [rsp + 16 * 1]
michael@0	917	%define k2 [rsp + 16 * 2]
michael@0	918	%define k3 [rsp + 16 * 3]
michael@0	919	%define k4 [rsp + 16 * 4]
michael@0	920	%define k5 [rsp + 16 * 5]
michael@0	921	%define k6 [rsp + 16 * 6]
michael@0	922	%define k7 [rsp + 16 * 7]
michael@0	923	%define krd [rsp + 16 * 8]
michael@0	924	%define zero [rsp + 16 * 9]
michael@0	925
michael@0	926	GET_FILTERS
michael@0	927
michael@0	928	movsxd rax, DWORD PTR arg(1) ;pixels_per_line
michael@0	929	movsxd rdx, DWORD PTR arg(3) ;out_pitch
michael@0	930	movsxd rcx, DWORD PTR arg(4) ;output_height
michael@0	931
michael@0	932	.loop:
michael@0	933	movdqu xmm0, [rsi - 3] ;load src
michael@0	934
michael@0	935	movdqa xmm1, xmm0
michael@0	936	movdqa xmm6, xmm0
michael@0	937	movdqa xmm7, xmm0
michael@0	938	movdqa xmm2, xmm0
michael@0	939	movdqa xmm5, xmm0
michael@0	940	movdqa xmm3, xmm0
michael@0	941	movdqa xmm4, xmm0
michael@0	942
michael@0	943	psrldq xmm1, 1
michael@0	944	psrldq xmm6, 6
michael@0	945	psrldq xmm7, 7
michael@0	946	psrldq xmm2, 2
michael@0	947	psrldq xmm5, 5
michael@0	948	psrldq xmm3, 3
michael@0	949	psrldq xmm4, 4
michael@0	950
michael@0	951	APPLY_FILTER_8 1, 0
michael@0	952
michael@0	953	movdqu xmm0, [rsi + 5] ;load src
michael@0	954
michael@0	955	movdqa xmm1, xmm0
michael@0	956	movdqa xmm6, xmm0
michael@0	957	movdqa xmm7, xmm0
michael@0	958	movdqa xmm2, xmm0
michael@0	959	movdqa xmm5, xmm0
michael@0	960	movdqa xmm3, xmm0
michael@0	961	movdqa xmm4, xmm0
michael@0	962
michael@0	963	psrldq xmm1, 1
michael@0	964	psrldq xmm6, 6
michael@0	965	psrldq xmm7, 7
michael@0	966	psrldq xmm2, 2
michael@0	967	psrldq xmm5, 5
michael@0	968	psrldq xmm3, 3
michael@0	969	psrldq xmm4, 4
michael@0	970
michael@0	971	APPLY_FILTER_8 1, 8
michael@0	972
michael@0	973	lea rsi, [rsi + rax]
michael@0	974	lea rdi, [rdi + rdx]
michael@0	975	dec rcx
michael@0	976	jnz .loop
michael@0	977
michael@0	978	add rsp, 16 * 10
michael@0	979	pop rsp
michael@0	980
michael@0	981	; begin epilog
michael@0	982	pop rdi
michael@0	983	pop rsi
michael@0	984	RESTORE_XMM
michael@0	985	UNSHADOW_ARGS
michael@0	986	pop rbp
michael@0	987	ret