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

media/libyuv/source/row_x86.asm@b8a032363ba2 (annotated)

media/libyuv/source/row_x86.asm

Thu, 22 Jan 2015 13:21:57 +0100

author

Michael Schloh von Bennewitz <michael@schloh.com>

date

Thu, 22 Jan 2015 13:21:57 +0100

branch

TOR_BUG_9701

changeset 15

b8a032363ba2

permissions

-rw-r--r--

Incorporate requested changes from Mozilla in review:
https://bugzilla.mozilla.org/show_bug.cgi?id=1123480#c6

michael@0	1	;
michael@0	2	; Copyright 2012 The LibYuv 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	%ifdef __YASM_VERSION_ID__
michael@0	12	%if __YASM_VERSION_ID__ < 01020000h
michael@0	13	%error AVX2 is supported only by yasm 1.2.0 or later.
michael@0	14	%endif
michael@0	15	%endif
michael@0	16	%include "x86inc.asm"
michael@0	17
michael@0	18	SECTION .text
michael@0	19
michael@0	20	; cglobal numeric constants are parameters, gpr regs, mm regs
michael@0	21
michael@0	22	; void YUY2ToYRow_SSE2(const uint8* src_yuy2, uint8* dst_y, int pix)
michael@0	23
michael@0	24	%macro YUY2TOYROW 2-3
michael@0	25	cglobal %1ToYRow%3, 3, 3, 3, src_yuy2, dst_y, pix
michael@0	26	%ifidn %1,YUY2
michael@0	27	pcmpeqb m2, m2, m2 ; generate mask 0x00ff00ff
michael@0	28	psrlw m2, m2, 8
michael@0	29	%endif
michael@0	30
michael@0	31	ALIGN 4
michael@0	32	.convertloop:
michael@0	33	mov%2 m0, [src_yuy2q]
michael@0	34	mov%2 m1, [src_yuy2q + mmsize]
michael@0	35	lea src_yuy2q, [src_yuy2q + mmsize * 2]
michael@0	36	%ifidn %1,YUY2
michael@0	37	pand m0, m0, m2 ; YUY2 even bytes are Y
michael@0	38	pand m1, m1, m2
michael@0	39	%else
michael@0	40	psrlw m0, m0, 8 ; UYVY odd bytes are Y
michael@0	41	psrlw m1, m1, 8
michael@0	42	%endif
michael@0	43	packuswb m0, m0, m1
michael@0	44	%if cpuflag(AVX2)
michael@0	45	vpermq m0, m0, 0xd8
michael@0	46	%endif
michael@0	47	sub pixd, mmsize
michael@0	48	mov%2 [dst_yq], m0
michael@0	49	lea dst_yq, [dst_yq + mmsize]
michael@0	50	jg .convertloop
michael@0	51	REP_RET
michael@0	52	%endmacro
michael@0	53
michael@0	54	; TODO(fbarchard): Remove MMX. Add SSSE3 pshufb version.
michael@0	55	INIT_MMX MMX
michael@0	56	YUY2TOYROW YUY2,a,
michael@0	57	YUY2TOYROW YUY2,u,_Unaligned
michael@0	58	YUY2TOYROW UYVY,a,
michael@0	59	YUY2TOYROW UYVY,u,_Unaligned
michael@0	60	INIT_XMM SSE2
michael@0	61	YUY2TOYROW YUY2,a,
michael@0	62	YUY2TOYROW YUY2,u,_Unaligned
michael@0	63	YUY2TOYROW UYVY,a,
michael@0	64	YUY2TOYROW UYVY,u,_Unaligned
michael@0	65	INIT_YMM AVX2
michael@0	66	YUY2TOYROW YUY2,a,
michael@0	67	YUY2TOYROW UYVY,a,
michael@0	68
michael@0	69	; void SplitUVRow_SSE2(const uint8* src_uv, uint8* dst_u, uint8* dst_v, int pix)
michael@0	70
michael@0	71	%macro SplitUVRow 1-2
michael@0	72	cglobal SplitUVRow%2, 4, 4, 5, src_uv, dst_u, dst_v, pix
michael@0	73	pcmpeqb m4, m4, m4 ; generate mask 0x00ff00ff
michael@0	74	psrlw m4, m4, 8
michael@0	75	sub dst_vq, dst_uq
michael@0	76
michael@0	77	ALIGN 4
michael@0	78	.convertloop:
michael@0	79	mov%1 m0, [src_uvq]
michael@0	80	mov%1 m1, [src_uvq + mmsize]
michael@0	81	lea src_uvq, [src_uvq + mmsize * 2]
michael@0	82	psrlw m2, m0, 8 ; odd bytes
michael@0	83	psrlw m3, m1, 8
michael@0	84	pand m0, m0, m4 ; even bytes
michael@0	85	pand m1, m1, m4
michael@0	86	packuswb m0, m0, m1
michael@0	87	packuswb m2, m2, m3
michael@0	88	%if cpuflag(AVX2)
michael@0	89	vpermq m0, m0, 0xd8
michael@0	90	vpermq m2, m2, 0xd8
michael@0	91	%endif
michael@0	92	mov%1 [dst_uq], m0
michael@0	93	mov%1 [dst_uq + dst_vq], m2
michael@0	94	lea dst_uq, [dst_uq + mmsize]
michael@0	95	sub pixd, mmsize
michael@0	96	jg .convertloop
michael@0	97	REP_RET
michael@0	98	%endmacro
michael@0	99
michael@0	100	INIT_MMX MMX
michael@0	101	SplitUVRow a,
michael@0	102	SplitUVRow u,_Unaligned
michael@0	103	INIT_XMM SSE2
michael@0	104	SplitUVRow a,
michael@0	105	SplitUVRow u,_Unaligned
michael@0	106	INIT_YMM AVX2
michael@0	107	SplitUVRow a,
michael@0	108
michael@0	109	; void MergeUVRow_SSE2(const uint8* src_u, const uint8* src_v, uint8* dst_uv,
michael@0	110	; int width);
michael@0	111
michael@0	112	%macro MergeUVRow_ 1-2
michael@0	113	cglobal MergeUVRow_%2, 4, 4, 3, src_u, src_v, dst_uv, pix
michael@0	114	sub src_vq, src_uq
michael@0	115
michael@0	116	ALIGN 4
michael@0	117	.convertloop:
michael@0	118	mov%1 m0, [src_uq]
michael@0	119	mov%1 m1, [src_vq]
michael@0	120	lea src_uq, [src_uq + mmsize]
michael@0	121	punpcklbw m2, m0, m1 // first 8 UV pairs
michael@0	122	punpckhbw m0, m0, m1 // next 8 UV pairs
michael@0	123	%if cpuflag(AVX2)
michael@0	124	vperm2i128 m1, m2, m0, 0x20 // low 128 of ymm2 and low 128 of ymm0
michael@0	125	vperm2i128 m2, m2, m0, 0x31 // high 128 of ymm2 and high 128 of ymm0
michael@0	126	mov%1 [dst_uvq], m1
michael@0	127	mov%1 [dst_uvq + mmsize], m2
michael@0	128	%else
michael@0	129	mov%1 [dst_uvq], m2
michael@0	130	mov%1 [dst_uvq + mmsize], m0
michael@0	131	%endif
michael@0	132	lea dst_uvq, [dst_uvq + mmsize * 2]
michael@0	133	sub pixd, mmsize
michael@0	134	jg .convertloop
michael@0	135	REP_RET
michael@0	136	%endmacro
michael@0	137
michael@0	138	INIT_MMX MMX
michael@0	139	MergeUVRow_ a,
michael@0	140	MergeUVRow_ u,_Unaligned
michael@0	141	INIT_XMM SSE2
michael@0	142	MergeUVRow_ a,
michael@0	143	MergeUVRow_ u,_Unaligned
michael@0	144	INIT_YMM AVX2
michael@0	145	MergeUVRow_ a,
michael@0	146