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

media/libvpx/vp8/common/x86/idctllm_mmx.asm@6474c204b198 (annotated)

media/libvpx/vp8/common/x86/idctllm_mmx.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	; /****************************************************************************
michael@0	15	; * Notes:
michael@0	16	; *
michael@0	17	; * This implementation makes use of 16 bit fixed point version of two multiply
michael@0	18	; * constants:
michael@0	19	; * 1. sqrt(2) * cos (pi/8)
michael@0	20	; * 2. sqrt(2) * sin (pi/8)
michael@0	21	; * Because the first constant is bigger than 1, to maintain the same 16 bit
michael@0	22	; * fixed point precision as the second one, we use a trick of
michael@0	23	; * x * a = x + x*(a-1)
michael@0	24	; * so
michael@0	25	; * x * sqrt(2) * cos (pi/8) = x + x * (sqrt(2) *cos(pi/8)-1).
michael@0	26	; *
michael@0	27	; * For the second constant, because of the 16bit version is 35468, which
michael@0	28	; * is bigger than 32768, in signed 16 bit multiply, it becomes a negative
michael@0	29	; * number.
michael@0	30	; * (x * (unsigned)35468 >> 16) = x * (signed)35468 >> 16 + x
michael@0	31	; *
michael@0	32	; **************************************************************************/
michael@0	33
michael@0	34
michael@0	35	;void vp8_short_idct4x4llm_mmx(short *input, unsigned char *pred,
michael@0	36	;int pitch, unsigned char *dest,int stride)
michael@0	37	global sym(vp8_short_idct4x4llm_mmx) PRIVATE
michael@0	38	sym(vp8_short_idct4x4llm_mmx):
michael@0	39	push rbp
michael@0	40	mov rbp, rsp
michael@0	41	SHADOW_ARGS_TO_STACK 5
michael@0	42	GET_GOT rbx
michael@0	43	push rsi
michael@0	44	push rdi
michael@0	45	; end prolog
michael@0	46
michael@0	47	mov rax, arg(0) ;input
michael@0	48	mov rsi, arg(1) ;pred
michael@0	49
michael@0	50	movq mm0, [rax ]
michael@0	51	movq mm1, [rax+ 8]
michael@0	52	movq mm2, [rax+16]
michael@0	53	movq mm3, [rax+24]
michael@0	54
michael@0	55	%if 0
michael@0	56	pxor mm7, mm7
michael@0	57	movq [rax], mm7
michael@0	58	movq [rax+8], mm7
michael@0	59	movq [rax+16],mm7
michael@0	60	movq [rax+24],mm7
michael@0	61	%endif
michael@0	62	movsxd rax, dword ptr arg(2) ;pitch
michael@0	63	mov rdx, arg(3) ;dest
michael@0	64	movsxd rdi, dword ptr arg(4) ;stride
michael@0	65
michael@0	66
michael@0	67	psubw mm0, mm2 ; b1= 0-2
michael@0	68	paddw mm2, mm2 ;
michael@0	69
michael@0	70	movq mm5, mm1
michael@0	71	paddw mm2, mm0 ; a1 =0+2
michael@0	72
michael@0	73	pmulhw mm5, [GLOBAL(x_s1sqr2)];
michael@0	74	paddw mm5, mm1 ; ip1 * sin(pi/8) * sqrt(2)
michael@0	75
michael@0	76	movq mm7, mm3 ;
michael@0	77	pmulhw mm7, [GLOBAL(x_c1sqr2less1)];
michael@0	78
michael@0	79	paddw mm7, mm3 ; ip3 * cos(pi/8) * sqrt(2)
michael@0	80	psubw mm7, mm5 ; c1
michael@0	81
michael@0	82	movq mm5, mm1
michael@0	83	movq mm4, mm3
michael@0	84
michael@0	85	pmulhw mm5, [GLOBAL(x_c1sqr2less1)]
michael@0	86	paddw mm5, mm1
michael@0	87
michael@0	88	pmulhw mm3, [GLOBAL(x_s1sqr2)]
michael@0	89	paddw mm3, mm4
michael@0	90
michael@0	91	paddw mm3, mm5 ; d1
michael@0	92	movq mm6, mm2 ; a1
michael@0	93
michael@0	94	movq mm4, mm0 ; b1
michael@0	95	paddw mm2, mm3 ;0
michael@0	96
michael@0	97	paddw mm4, mm7 ;1
michael@0	98	psubw mm0, mm7 ;2
michael@0	99
michael@0	100	psubw mm6, mm3 ;3
michael@0	101
michael@0	102	movq mm1, mm2 ; 03 02 01 00
michael@0	103	movq mm3, mm4 ; 23 22 21 20
michael@0	104
michael@0	105	punpcklwd mm1, mm0 ; 11 01 10 00
michael@0	106	punpckhwd mm2, mm0 ; 13 03 12 02
michael@0	107
michael@0	108	punpcklwd mm3, mm6 ; 31 21 30 20
michael@0	109	punpckhwd mm4, mm6 ; 33 23 32 22
michael@0	110
michael@0	111	movq mm0, mm1 ; 11 01 10 00
michael@0	112	movq mm5, mm2 ; 13 03 12 02
michael@0	113
michael@0	114	punpckldq mm0, mm3 ; 30 20 10 00
michael@0	115	punpckhdq mm1, mm3 ; 31 21 11 01
michael@0	116
michael@0	117	punpckldq mm2, mm4 ; 32 22 12 02
michael@0	118	punpckhdq mm5, mm4 ; 33 23 13 03
michael@0	119
michael@0	120	movq mm3, mm5 ; 33 23 13 03
michael@0	121
michael@0	122	psubw mm0, mm2 ; b1= 0-2
michael@0	123	paddw mm2, mm2 ;
michael@0	124
michael@0	125	movq mm5, mm1
michael@0	126	paddw mm2, mm0 ; a1 =0+2
michael@0	127
michael@0	128	pmulhw mm5, [GLOBAL(x_s1sqr2)];
michael@0	129	paddw mm5, mm1 ; ip1 * sin(pi/8) * sqrt(2)
michael@0	130
michael@0	131	movq mm7, mm3 ;
michael@0	132	pmulhw mm7, [GLOBAL(x_c1sqr2less1)];
michael@0	133
michael@0	134	paddw mm7, mm3 ; ip3 * cos(pi/8) * sqrt(2)
michael@0	135	psubw mm7, mm5 ; c1
michael@0	136
michael@0	137	movq mm5, mm1
michael@0	138	movq mm4, mm3
michael@0	139
michael@0	140	pmulhw mm5, [GLOBAL(x_c1sqr2less1)]
michael@0	141	paddw mm5, mm1
michael@0	142
michael@0	143	pmulhw mm3, [GLOBAL(x_s1sqr2)]
michael@0	144	paddw mm3, mm4
michael@0	145
michael@0	146	paddw mm3, mm5 ; d1
michael@0	147	paddw mm0, [GLOBAL(fours)]
michael@0	148
michael@0	149	paddw mm2, [GLOBAL(fours)]
michael@0	150	movq mm6, mm2 ; a1
michael@0	151
michael@0	152	movq mm4, mm0 ; b1
michael@0	153	paddw mm2, mm3 ;0
michael@0	154
michael@0	155	paddw mm4, mm7 ;1
michael@0	156	psubw mm0, mm7 ;2
michael@0	157
michael@0	158	psubw mm6, mm3 ;3
michael@0	159	psraw mm2, 3
michael@0	160
michael@0	161	psraw mm0, 3
michael@0	162	psraw mm4, 3
michael@0	163
michael@0	164	psraw mm6, 3
michael@0	165
michael@0	166	movq mm1, mm2 ; 03 02 01 00
michael@0	167	movq mm3, mm4 ; 23 22 21 20
michael@0	168
michael@0	169	punpcklwd mm1, mm0 ; 11 01 10 00
michael@0	170	punpckhwd mm2, mm0 ; 13 03 12 02
michael@0	171
michael@0	172	punpcklwd mm3, mm6 ; 31 21 30 20
michael@0	173	punpckhwd mm4, mm6 ; 33 23 32 22
michael@0	174
michael@0	175	movq mm0, mm1 ; 11 01 10 00
michael@0	176	movq mm5, mm2 ; 13 03 12 02
michael@0	177
michael@0	178	punpckldq mm0, mm3 ; 30 20 10 00
michael@0	179	punpckhdq mm1, mm3 ; 31 21 11 01
michael@0	180
michael@0	181	punpckldq mm2, mm4 ; 32 22 12 02
michael@0	182	punpckhdq mm5, mm4 ; 33 23 13 03
michael@0	183
michael@0	184	pxor mm7, mm7
michael@0	185
michael@0	186	movd mm4, [rsi]
michael@0	187	punpcklbw mm4, mm7
michael@0	188	paddsw mm0, mm4
michael@0	189	packuswb mm0, mm7
michael@0	190	movd [rdx], mm0
michael@0	191
michael@0	192	movd mm4, [rsi+rax]
michael@0	193	punpcklbw mm4, mm7
michael@0	194	paddsw mm1, mm4
michael@0	195	packuswb mm1, mm7
michael@0	196	movd [rdx+rdi], mm1
michael@0	197
michael@0	198	movd mm4, [rsi+2*rax]
michael@0	199	punpcklbw mm4, mm7
michael@0	200	paddsw mm2, mm4
michael@0	201	packuswb mm2, mm7
michael@0	202	movd [rdx+rdi*2], mm2
michael@0	203
michael@0	204	add rdx, rdi
michael@0	205	add rsi, rax
michael@0	206
michael@0	207	movd mm4, [rsi+2*rax]
michael@0	208	punpcklbw mm4, mm7
michael@0	209	paddsw mm5, mm4
michael@0	210	packuswb mm5, mm7
michael@0	211	movd [rdx+rdi*2], mm5
michael@0	212
michael@0	213	; begin epilog
michael@0	214	pop rdi
michael@0	215	pop rsi
michael@0	216	RESTORE_GOT
michael@0	217	UNSHADOW_ARGS
michael@0	218	pop rbp
michael@0	219	ret
michael@0	220
michael@0	221	;void vp8_dc_only_idct_add_mmx(
michael@0	222	;short input_dc,
michael@0	223	;unsigned char *pred_ptr,
michael@0	224	;int pred_stride,
michael@0	225	;unsigned char *dst_ptr,
michael@0	226	;int stride)
michael@0	227	global sym(vp8_dc_only_idct_add_mmx) PRIVATE
michael@0	228	sym(vp8_dc_only_idct_add_mmx):
michael@0	229	push rbp
michael@0	230	mov rbp, rsp
michael@0	231	SHADOW_ARGS_TO_STACK 5
michael@0	232	GET_GOT rbx
michael@0	233	; end prolog
michael@0	234
michael@0	235	movd mm5, arg(0) ;input_dc
michael@0	236	mov rax, arg(1) ;pred_ptr
michael@0	237	movsxd rdx, dword ptr arg(2) ;pred_stride
michael@0	238
michael@0	239	pxor mm0, mm0
michael@0	240
michael@0	241	paddw mm5, [GLOBAL(fours)]
michael@0	242	lea rcx, [rdx + rdx*2]
michael@0	243
michael@0	244	psraw mm5, 3
michael@0	245
michael@0	246	punpcklwd mm5, mm5
michael@0	247
michael@0	248	punpckldq mm5, mm5
michael@0	249
michael@0	250	movd mm1, [rax]
michael@0	251	movd mm2, [rax+rdx]
michael@0	252	movd mm3, [rax+2*rdx]
michael@0	253	movd mm4, [rax+rcx]
michael@0	254
michael@0	255	mov rax, arg(3) ;d -- destination
michael@0	256	movsxd rdx, dword ptr arg(4) ;dst_stride
michael@0	257
michael@0	258	punpcklbw mm1, mm0
michael@0	259	paddsw mm1, mm5
michael@0	260	packuswb mm1, mm0 ; pack and unpack to saturate
michael@0	261	lea rcx, [rdx + rdx*2]
michael@0	262
michael@0	263	punpcklbw mm2, mm0
michael@0	264	paddsw mm2, mm5
michael@0	265	packuswb mm2, mm0 ; pack and unpack to saturate
michael@0	266
michael@0	267	punpcklbw mm3, mm0
michael@0	268	paddsw mm3, mm5
michael@0	269	packuswb mm3, mm0 ; pack and unpack to saturate
michael@0	270
michael@0	271	punpcklbw mm4, mm0
michael@0	272	paddsw mm4, mm5
michael@0	273	packuswb mm4, mm0 ; pack and unpack to saturate
michael@0	274
michael@0	275	movd [rax], mm1
michael@0	276	movd [rax+rdx], mm2
michael@0	277	movd [rax+2*rdx], mm3
michael@0	278	movd [rax+rcx], mm4
michael@0	279
michael@0	280	; begin epilog
michael@0	281	RESTORE_GOT
michael@0	282	UNSHADOW_ARGS
michael@0	283	pop rbp
michael@0	284	ret
michael@0	285
michael@0	286	SECTION_RODATA
michael@0	287	align 16
michael@0	288	x_s1sqr2:
michael@0	289	times 4 dw 0x8A8C
michael@0	290	align 16
michael@0	291	x_c1sqr2less1:
michael@0	292	times 4 dw 0x4E7B
michael@0	293	align 16
michael@0	294	fours:
michael@0	295	times 4 dw 0x0004