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

media/libjpeg/simd/jcsamss2-64.asm@6474c204b198 (annotated)

media/libjpeg/simd/jcsamss2-64.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	; jcsamss2-64.asm - downsampling (64-bit SSE2)
michael@0	3	;
michael@0	4	; Copyright 2009 Pierre Ossman <ossman@cendio.se> for Cendio AB
michael@0	5	; Copyright 2009 D. R. Commander
michael@0	6	;
michael@0	7	; Based on
michael@0	8	; x86 SIMD extension for IJG JPEG library
michael@0	9	; Copyright (C) 1999-2006, MIYASAKA Masaru.
michael@0	10	; For conditions of distribution and use, see copyright notice in jsimdext.inc
michael@0	11	;
michael@0	12	; This file should be assembled with NASM (Netwide Assembler),
michael@0	13	; can *not* be assembled with Microsoft's MASM or any compatible
michael@0	14	; assembler (including Borland's Turbo Assembler).
michael@0	15	; NASM is available from http://nasm.sourceforge.net/ or
michael@0	16	; http://sourceforge.net/project/showfiles.php?group_id=6208
michael@0	17	;
michael@0	18	; [TAB8]
michael@0	19
michael@0	20	%include "jsimdext.inc"
michael@0	21
michael@0	22	; --------------------------------------------------------------------------
michael@0	23	SECTION SEG_TEXT
michael@0	24	BITS 64
michael@0	25	;
michael@0	26	; Downsample pixel values of a single component.
michael@0	27	; This version handles the common case of 2:1 horizontal and 1:1 vertical,
michael@0	28	; without smoothing.
michael@0	29	;
michael@0	30	; GLOBAL(void)
michael@0	31	; jsimd_h2v1_downsample_sse2 (JDIMENSION image_width, int max_v_samp_factor,
michael@0	32	; JDIMENSION v_samp_factor, JDIMENSION width_blocks,
michael@0	33	; JSAMPARRAY input_data, JSAMPARRAY output_data);
michael@0	34	;
michael@0	35
michael@0	36	; r10 = JDIMENSION image_width
michael@0	37	; r11 = int max_v_samp_factor
michael@0	38	; r12 = JDIMENSION v_samp_factor
michael@0	39	; r13 = JDIMENSION width_blocks
michael@0	40	; r14 = JSAMPARRAY input_data
michael@0	41	; r15 = JSAMPARRAY output_data
michael@0	42
michael@0	43	align 16
michael@0	44	global EXTN(jsimd_h2v1_downsample_sse2)
michael@0	45
michael@0	46	EXTN(jsimd_h2v1_downsample_sse2):
michael@0	47	push rbp
michael@0	48	mov rax,rsp
michael@0	49	mov rbp,rsp
michael@0	50	collect_args
michael@0	51
michael@0	52	mov rcx, r13
michael@0	53	shl rcx,3 ; imul rcx,DCTSIZE (rcx = output_cols)
michael@0	54	jz near .return
michael@0	55
michael@0	56	mov rdx, r10
michael@0	57
michael@0	58	; -- expand_right_edge
michael@0	59
michael@0	60	push rcx
michael@0	61	shl rcx,1 ; output_cols * 2
michael@0	62	sub rcx,rdx
michael@0	63	jle short .expand_end
michael@0	64
michael@0	65	mov rax, r11
michael@0	66	test rax,rax
michael@0	67	jle short .expand_end
michael@0	68
michael@0	69	cld
michael@0	70	mov rsi, r14 ; input_data
michael@0	71	.expandloop:
michael@0	72	push rax
michael@0	73	push rcx
michael@0	74
michael@0	75	mov rdi, JSAMPROW [rsi]
michael@0	76	add rdi,rdx
michael@0	77	mov al, JSAMPLE [rdi-1]
michael@0	78
michael@0	79	rep stosb
michael@0	80
michael@0	81	pop rcx
michael@0	82	pop rax
michael@0	83
michael@0	84	add rsi, byte SIZEOF_JSAMPROW
michael@0	85	dec rax
michael@0	86	jg short .expandloop
michael@0	87
michael@0	88	.expand_end:
michael@0	89	pop rcx ; output_cols
michael@0	90
michael@0	91	; -- h2v1_downsample
michael@0	92
michael@0	93	mov rax, r12 ; rowctr
michael@0	94	test eax,eax
michael@0	95	jle near .return
michael@0	96
michael@0	97	mov rdx, 0x00010000 ; bias pattern
michael@0	98	movd xmm7,edx
michael@0	99	pcmpeqw xmm6,xmm6
michael@0	100	pshufd xmm7,xmm7,0x00 ; xmm7={0, 1, 0, 1, 0, 1, 0, 1}
michael@0	101	psrlw xmm6,BYTE_BIT ; xmm6={0xFF 0x00 0xFF 0x00 ..}
michael@0	102
michael@0	103	mov rsi, r14 ; input_data
michael@0	104	mov rdi, r15 ; output_data
michael@0	105	.rowloop:
michael@0	106	push rcx
michael@0	107	push rdi
michael@0	108	push rsi
michael@0	109
michael@0	110	mov rsi, JSAMPROW [rsi] ; inptr
michael@0	111	mov rdi, JSAMPROW [rdi] ; outptr
michael@0	112
michael@0	113	cmp rcx, byte SIZEOF_XMMWORD
michael@0	114	jae short .columnloop
michael@0	115
michael@0	116	.columnloop_r8:
michael@0	117	movdqa xmm0, XMMWORD [rsi+0*SIZEOF_XMMWORD]
michael@0	118	pxor xmm1,xmm1
michael@0	119	mov rcx, SIZEOF_XMMWORD
michael@0	120	jmp short .downsample
michael@0	121
michael@0	122	.columnloop:
michael@0	123	movdqa xmm0, XMMWORD [rsi+0*SIZEOF_XMMWORD]
michael@0	124	movdqa xmm1, XMMWORD [rsi+1*SIZEOF_XMMWORD]
michael@0	125
michael@0	126	.downsample:
michael@0	127	movdqa xmm2,xmm0
michael@0	128	movdqa xmm3,xmm1
michael@0	129
michael@0	130	pand xmm0,xmm6
michael@0	131	psrlw xmm2,BYTE_BIT
michael@0	132	pand xmm1,xmm6
michael@0	133	psrlw xmm3,BYTE_BIT
michael@0	134
michael@0	135	paddw xmm0,xmm2
michael@0	136	paddw xmm1,xmm3
michael@0	137	paddw xmm0,xmm7
michael@0	138	paddw xmm1,xmm7
michael@0	139	psrlw xmm0,1
michael@0	140	psrlw xmm1,1
michael@0	141
michael@0	142	packuswb xmm0,xmm1
michael@0	143
michael@0	144	movdqa XMMWORD [rdi+0*SIZEOF_XMMWORD], xmm0
michael@0	145
michael@0	146	sub rcx, byte SIZEOF_XMMWORD ; outcol
michael@0	147	add rsi, byte 2*SIZEOF_XMMWORD ; inptr
michael@0	148	add rdi, byte 1*SIZEOF_XMMWORD ; outptr
michael@0	149	cmp rcx, byte SIZEOF_XMMWORD
michael@0	150	jae short .columnloop
michael@0	151	test rcx,rcx
michael@0	152	jnz short .columnloop_r8
michael@0	153
michael@0	154	pop rsi
michael@0	155	pop rdi
michael@0	156	pop rcx
michael@0	157
michael@0	158	add rsi, byte SIZEOF_JSAMPROW ; input_data
michael@0	159	add rdi, byte SIZEOF_JSAMPROW ; output_data
michael@0	160	dec rax ; rowctr
michael@0	161	jg near .rowloop
michael@0	162
michael@0	163	.return:
michael@0	164	uncollect_args
michael@0	165	pop rbp
michael@0	166	ret
michael@0	167
michael@0	168	; --------------------------------------------------------------------------
michael@0	169	;
michael@0	170	; Downsample pixel values of a single component.
michael@0	171	; This version handles the standard case of 2:1 horizontal and 2:1 vertical,
michael@0	172	; without smoothing.
michael@0	173	;
michael@0	174	; GLOBAL(void)
michael@0	175	; jsimd_h2v2_downsample_sse2 (JDIMENSION image_width, int max_v_samp_factor,
michael@0	176	; JDIMENSION v_samp_factor, JDIMENSION width_blocks,
michael@0	177	; JSAMPARRAY input_data, JSAMPARRAY output_data);
michael@0	178	;
michael@0	179
michael@0	180	; r10 = JDIMENSION image_width
michael@0	181	; r11 = int max_v_samp_factor
michael@0	182	; r12 = JDIMENSION v_samp_factor
michael@0	183	; r13 = JDIMENSION width_blocks
michael@0	184	; r14 = JSAMPARRAY input_data
michael@0	185	; r15 = JSAMPARRAY output_data
michael@0	186
michael@0	187	align 16
michael@0	188	global EXTN(jsimd_h2v2_downsample_sse2)
michael@0	189
michael@0	190	EXTN(jsimd_h2v2_downsample_sse2):
michael@0	191	push rbp
michael@0	192	mov rax,rsp
michael@0	193	mov rbp,rsp
michael@0	194	collect_args
michael@0	195
michael@0	196	mov rcx, r13
michael@0	197	shl rcx,3 ; imul rcx,DCTSIZE (rcx = output_cols)
michael@0	198	jz near .return
michael@0	199
michael@0	200	mov rdx, r10
michael@0	201
michael@0	202	; -- expand_right_edge
michael@0	203
michael@0	204	push rcx
michael@0	205	shl rcx,1 ; output_cols * 2
michael@0	206	sub rcx,rdx
michael@0	207	jle short .expand_end
michael@0	208
michael@0	209	mov rax, r11
michael@0	210	test rax,rax
michael@0	211	jle short .expand_end
michael@0	212
michael@0	213	cld
michael@0	214	mov rsi, r14 ; input_data
michael@0	215	.expandloop:
michael@0	216	push rax
michael@0	217	push rcx
michael@0	218
michael@0	219	mov rdi, JSAMPROW [rsi]
michael@0	220	add rdi,rdx
michael@0	221	mov al, JSAMPLE [rdi-1]
michael@0	222
michael@0	223	rep stosb
michael@0	224
michael@0	225	pop rcx
michael@0	226	pop rax
michael@0	227
michael@0	228	add rsi, byte SIZEOF_JSAMPROW
michael@0	229	dec rax
michael@0	230	jg short .expandloop
michael@0	231
michael@0	232	.expand_end:
michael@0	233	pop rcx ; output_cols
michael@0	234
michael@0	235	; -- h2v2_downsample
michael@0	236
michael@0	237	mov rax, r12 ; rowctr
michael@0	238	test rax,rax
michael@0	239	jle near .return
michael@0	240
michael@0	241	mov rdx, 0x00020001 ; bias pattern
michael@0	242	movd xmm7,edx
michael@0	243	pcmpeqw xmm6,xmm6
michael@0	244	pshufd xmm7,xmm7,0x00 ; xmm7={1, 2, 1, 2, 1, 2, 1, 2}
michael@0	245	psrlw xmm6,BYTE_BIT ; xmm6={0xFF 0x00 0xFF 0x00 ..}
michael@0	246
michael@0	247	mov rsi, r14 ; input_data
michael@0	248	mov rdi, r15 ; output_data
michael@0	249	.rowloop:
michael@0	250	push rcx
michael@0	251	push rdi
michael@0	252	push rsi
michael@0	253
michael@0	254	mov rdx, JSAMPROW [rsi+0*SIZEOF_JSAMPROW] ; inptr0
michael@0	255	mov rsi, JSAMPROW [rsi+1*SIZEOF_JSAMPROW] ; inptr1
michael@0	256	mov rdi, JSAMPROW [rdi] ; outptr
michael@0	257
michael@0	258	cmp rcx, byte SIZEOF_XMMWORD
michael@0	259	jae short .columnloop
michael@0	260
michael@0	261	.columnloop_r8:
michael@0	262	movdqa xmm0, XMMWORD [rdx+0*SIZEOF_XMMWORD]
michael@0	263	movdqa xmm1, XMMWORD [rsi+0*SIZEOF_XMMWORD]
michael@0	264	pxor xmm2,xmm2
michael@0	265	pxor xmm3,xmm3
michael@0	266	mov rcx, SIZEOF_XMMWORD
michael@0	267	jmp short .downsample
michael@0	268
michael@0	269	.columnloop:
michael@0	270	movdqa xmm0, XMMWORD [rdx+0*SIZEOF_XMMWORD]
michael@0	271	movdqa xmm1, XMMWORD [rsi+0*SIZEOF_XMMWORD]
michael@0	272	movdqa xmm2, XMMWORD [rdx+1*SIZEOF_XMMWORD]
michael@0	273	movdqa xmm3, XMMWORD [rsi+1*SIZEOF_XMMWORD]
michael@0	274
michael@0	275	.downsample:
michael@0	276	movdqa xmm4,xmm0
michael@0	277	movdqa xmm5,xmm1
michael@0	278	pand xmm0,xmm6
michael@0	279	psrlw xmm4,BYTE_BIT
michael@0	280	pand xmm1,xmm6
michael@0	281	psrlw xmm5,BYTE_BIT
michael@0	282	paddw xmm0,xmm4
michael@0	283	paddw xmm1,xmm5
michael@0	284
michael@0	285	movdqa xmm4,xmm2
michael@0	286	movdqa xmm5,xmm3
michael@0	287	pand xmm2,xmm6
michael@0	288	psrlw xmm4,BYTE_BIT
michael@0	289	pand xmm3,xmm6
michael@0	290	psrlw xmm5,BYTE_BIT
michael@0	291	paddw xmm2,xmm4
michael@0	292	paddw xmm3,xmm5
michael@0	293
michael@0	294	paddw xmm0,xmm1
michael@0	295	paddw xmm2,xmm3
michael@0	296	paddw xmm0,xmm7
michael@0	297	paddw xmm2,xmm7
michael@0	298	psrlw xmm0,2
michael@0	299	psrlw xmm2,2
michael@0	300
michael@0	301	packuswb xmm0,xmm2
michael@0	302
michael@0	303	movdqa XMMWORD [rdi+0*SIZEOF_XMMWORD], xmm0
michael@0	304
michael@0	305	sub rcx, byte SIZEOF_XMMWORD ; outcol
michael@0	306	add rdx, byte 2*SIZEOF_XMMWORD ; inptr0
michael@0	307	add rsi, byte 2*SIZEOF_XMMWORD ; inptr1
michael@0	308	add rdi, byte 1*SIZEOF_XMMWORD ; outptr
michael@0	309	cmp rcx, byte SIZEOF_XMMWORD
michael@0	310	jae near .columnloop
michael@0	311	test rcx,rcx
michael@0	312	jnz near .columnloop_r8
michael@0	313
michael@0	314	pop rsi
michael@0	315	pop rdi
michael@0	316	pop rcx
michael@0	317
michael@0	318	add rsi, byte 2*SIZEOF_JSAMPROW ; input_data
michael@0	319	add rdi, byte 1*SIZEOF_JSAMPROW ; output_data
michael@0	320	dec rax ; rowctr
michael@0	321	jg near .rowloop
michael@0	322
michael@0	323	.return:
michael@0	324	uncollect_args
michael@0	325	pop rbp
michael@0	326	ret
michael@0	327
michael@0	328	; For some reason, the OS X linker does not honor the request to align the
michael@0	329	; segment unless we do this.
michael@0	330	align 16