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

media/libvpx/vp9/encoder/x86/vp9_variance_impl_mmx.asm@ac0c01689b40 (annotated)

media/libvpx/vp9/encoder/x86/vp9_variance_impl_mmx.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_mmx( short *src_ptr )
michael@0	15	global sym(vp9_get_mb_ss_mmx) PRIVATE
michael@0	16	sym(vp9_get_mb_ss_mmx):
michael@0	17	push rbp
michael@0	18	mov rbp, rsp
michael@0	19	SHADOW_ARGS_TO_STACK 7
michael@0	20	GET_GOT rbx
michael@0	21	push rsi
michael@0	22	push rdi
michael@0	23	sub rsp, 8
michael@0	24	; end prolog
michael@0	25
michael@0	26	mov rax, arg(0) ;src_ptr
michael@0	27	mov rcx, 16
michael@0	28	pxor mm4, mm4
michael@0	29
michael@0	30	.NEXTROW:
michael@0	31	movq mm0, [rax]
michael@0	32	movq mm1, [rax+8]
michael@0	33	movq mm2, [rax+16]
michael@0	34	movq mm3, [rax+24]
michael@0	35	pmaddwd mm0, mm0
michael@0	36	pmaddwd mm1, mm1
michael@0	37	pmaddwd mm2, mm2
michael@0	38	pmaddwd mm3, mm3
michael@0	39
michael@0	40	paddd mm4, mm0
michael@0	41	paddd mm4, mm1
michael@0	42	paddd mm4, mm2
michael@0	43	paddd mm4, mm3
michael@0	44
michael@0	45	add rax, 32
michael@0	46	dec rcx
michael@0	47	ja .NEXTROW
michael@0	48	movq QWORD PTR [rsp], mm4
michael@0	49
michael@0	50	;return sum[0]+sum[1];
michael@0	51	movsxd rax, dword ptr [rsp]
michael@0	52	movsxd rcx, dword ptr [rsp+4]
michael@0	53	add rax, rcx
michael@0	54
michael@0	55
michael@0	56	; begin epilog
michael@0	57	add rsp, 8
michael@0	58	pop rdi
michael@0	59	pop rsi
michael@0	60	RESTORE_GOT
michael@0	61	UNSHADOW_ARGS
michael@0	62	pop rbp
michael@0	63	ret
michael@0	64
michael@0	65
michael@0	66	;unsigned int vp9_get8x8var_mmx
michael@0	67	;(
michael@0	68	; unsigned char *src_ptr,
michael@0	69	; int source_stride,
michael@0	70	; unsigned char *ref_ptr,
michael@0	71	; int recon_stride,
michael@0	72	; unsigned int *SSE,
michael@0	73	; int *Sum
michael@0	74	;)
michael@0	75	global sym(vp9_get8x8var_mmx) PRIVATE
michael@0	76	sym(vp9_get8x8var_mmx):
michael@0	77	push rbp
michael@0	78	mov rbp, rsp
michael@0	79	SHADOW_ARGS_TO_STACK 6
michael@0	80	push rsi
michael@0	81	push rdi
michael@0	82	push rbx
michael@0	83	sub rsp, 16
michael@0	84	; end prolog
michael@0	85
michael@0	86
michael@0	87	pxor mm5, mm5 ; Blank mmx6
michael@0	88	pxor mm6, mm6 ; Blank mmx7
michael@0	89	pxor mm7, mm7 ; Blank mmx7
michael@0	90
michael@0	91	mov rax, arg(0) ;[src_ptr] ; Load base addresses
michael@0	92	mov rbx, arg(2) ;[ref_ptr]
michael@0	93	movsxd rcx, dword ptr arg(1) ;[source_stride]
michael@0	94	movsxd rdx, dword ptr arg(3) ;[recon_stride]
michael@0	95
michael@0	96	; Row 1
michael@0	97	movq mm0, [rax] ; Copy eight bytes to mm0
michael@0	98	movq mm1, [rbx] ; Copy eight bytes to mm1
michael@0	99	movq mm2, mm0 ; Take copies
michael@0	100	movq mm3, mm1 ; Take copies
michael@0	101
michael@0	102	punpcklbw mm0, mm6 ; unpack to higher prrcision
michael@0	103	punpcklbw mm1, mm6
michael@0	104	punpckhbw mm2, mm6 ; unpack to higher prrcision
michael@0	105	punpckhbw mm3, mm6
michael@0	106	psubsw mm0, mm1 ; A-B (low order) to MM0
michael@0	107	psubsw mm2, mm3 ; A-B (high order) to MM2
michael@0	108
michael@0	109	paddw mm5, mm0 ; accumulate differences in mm5
michael@0	110	paddw mm5, mm2 ; accumulate differences in mm5
michael@0	111
michael@0	112	pmaddwd mm0, mm0 ; square and accumulate
michael@0	113	pmaddwd mm2, mm2 ; square and accumulate
michael@0	114	add rbx,rdx ; Inc pointer into ref data
michael@0	115	add rax,rcx ; Inc pointer into the new data
michael@0	116	movq mm1, [rbx] ; Copy eight bytes to mm1
michael@0	117	paddd mm7, mm0 ; accumulate in mm7
michael@0	118	paddd mm7, mm2 ; accumulate in mm7
michael@0	119
michael@0	120
michael@0	121	; Row 2
michael@0	122	movq mm0, [rax] ; Copy eight bytes to mm0
michael@0	123	movq mm2, mm0 ; Take copies
michael@0	124	movq mm3, mm1 ; Take copies
michael@0	125
michael@0	126	punpcklbw mm0, mm6 ; unpack to higher prrcision
michael@0	127	punpcklbw mm1, mm6
michael@0	128	punpckhbw mm2, mm6 ; unpack to higher prrcision
michael@0	129	punpckhbw mm3, mm6
michael@0	130	psubsw mm0, mm1 ; A-B (low order) to MM0
michael@0	131	psubsw mm2, mm3 ; A-B (high order) to MM2
michael@0	132
michael@0	133	paddw mm5, mm0 ; accumulate differences in mm5
michael@0	134	paddw mm5, mm2 ; accumulate differences in mm5
michael@0	135
michael@0	136	pmaddwd mm0, mm0 ; square and accumulate
michael@0	137	pmaddwd mm2, mm2 ; square and accumulate
michael@0	138	add rbx,rdx ; Inc pointer into ref data
michael@0	139	add rax,rcx ; Inc pointer into the new data
michael@0	140	movq mm1, [rbx] ; Copy eight bytes to mm1
michael@0	141	paddd mm7, mm0 ; accumulate in mm7
michael@0	142	paddd mm7, mm2 ; accumulate in mm7
michael@0	143
michael@0	144	; Row 3
michael@0	145	movq mm0, [rax] ; Copy eight bytes to mm0
michael@0	146	movq mm2, mm0 ; Take copies
michael@0	147	movq mm3, mm1 ; Take copies
michael@0	148
michael@0	149	punpcklbw mm0, mm6 ; unpack to higher prrcision
michael@0	150	punpcklbw mm1, mm6
michael@0	151	punpckhbw mm2, mm6 ; unpack to higher prrcision
michael@0	152	punpckhbw mm3, mm6
michael@0	153	psubsw mm0, mm1 ; A-B (low order) to MM0
michael@0	154	psubsw mm2, mm3 ; A-B (high order) to MM2
michael@0	155
michael@0	156	paddw mm5, mm0 ; accumulate differences in mm5
michael@0	157	paddw mm5, mm2 ; accumulate differences in mm5
michael@0	158
michael@0	159	pmaddwd mm0, mm0 ; square and accumulate
michael@0	160	pmaddwd mm2, mm2 ; square and accumulate
michael@0	161	add rbx,rdx ; Inc pointer into ref data
michael@0	162	add rax,rcx ; Inc pointer into the new data
michael@0	163	movq mm1, [rbx] ; Copy eight bytes to mm1
michael@0	164	paddd mm7, mm0 ; accumulate in mm7
michael@0	165	paddd mm7, mm2 ; accumulate in mm7
michael@0	166
michael@0	167	; Row 4
michael@0	168	movq mm0, [rax] ; Copy eight bytes to mm0
michael@0	169	movq mm2, mm0 ; Take copies
michael@0	170	movq mm3, mm1 ; Take copies
michael@0	171
michael@0	172	punpcklbw mm0, mm6 ; unpack to higher prrcision
michael@0	173	punpcklbw mm1, mm6
michael@0	174	punpckhbw mm2, mm6 ; unpack to higher prrcision
michael@0	175	punpckhbw mm3, mm6
michael@0	176	psubsw mm0, mm1 ; A-B (low order) to MM0
michael@0	177	psubsw mm2, mm3 ; A-B (high order) to MM2
michael@0	178
michael@0	179	paddw mm5, mm0 ; accumulate differences in mm5
michael@0	180	paddw mm5, mm2 ; accumulate differences in mm5
michael@0	181
michael@0	182	pmaddwd mm0, mm0 ; square and accumulate
michael@0	183	pmaddwd mm2, mm2 ; square and accumulate
michael@0	184	add rbx,rdx ; Inc pointer into ref data
michael@0	185	add rax,rcx ; Inc pointer into the new data
michael@0	186	movq mm1, [rbx] ; Copy eight bytes to mm1
michael@0	187	paddd mm7, mm0 ; accumulate in mm7
michael@0	188	paddd mm7, mm2 ; accumulate in mm7
michael@0	189
michael@0	190	; Row 5
michael@0	191	movq mm0, [rax] ; Copy eight bytes to mm0
michael@0	192	movq mm2, mm0 ; Take copies
michael@0	193	movq mm3, mm1 ; Take copies
michael@0	194
michael@0	195	punpcklbw mm0, mm6 ; unpack to higher prrcision
michael@0	196	punpcklbw mm1, mm6
michael@0	197	punpckhbw mm2, mm6 ; unpack to higher prrcision
michael@0	198	punpckhbw mm3, mm6
michael@0	199	psubsw mm0, mm1 ; A-B (low order) to MM0
michael@0	200	psubsw mm2, mm3 ; A-B (high order) to MM2
michael@0	201
michael@0	202	paddw mm5, mm0 ; accumulate differences in mm5
michael@0	203	paddw mm5, mm2 ; accumulate differences in mm5
michael@0	204
michael@0	205	pmaddwd mm0, mm0 ; square and accumulate
michael@0	206	pmaddwd mm2, mm2 ; square and accumulate
michael@0	207	add rbx,rdx ; Inc pointer into ref data
michael@0	208	add rax,rcx ; Inc pointer into the new data
michael@0	209	movq mm1, [rbx] ; Copy eight bytes to mm1
michael@0	210	; movq mm4, [rbx + rdx]
michael@0	211	paddd mm7, mm0 ; accumulate in mm7
michael@0	212	paddd mm7, mm2 ; accumulate in mm7
michael@0	213
michael@0	214	; Row 6
michael@0	215	movq mm0, [rax] ; Copy eight bytes to mm0
michael@0	216	movq mm2, mm0 ; Take copies
michael@0	217	movq mm3, mm1 ; Take copies
michael@0	218
michael@0	219	punpcklbw mm0, mm6 ; unpack to higher prrcision
michael@0	220	punpcklbw mm1, mm6
michael@0	221	punpckhbw mm2, mm6 ; unpack to higher prrcision
michael@0	222	punpckhbw mm3, mm6
michael@0	223	psubsw mm0, mm1 ; A-B (low order) to MM0
michael@0	224	psubsw mm2, mm3 ; A-B (high order) to MM2
michael@0	225
michael@0	226	paddw mm5, mm0 ; accumulate differences in mm5
michael@0	227	paddw mm5, mm2 ; accumulate differences in mm5
michael@0	228
michael@0	229	pmaddwd mm0, mm0 ; square and accumulate
michael@0	230	pmaddwd mm2, mm2 ; square and accumulate
michael@0	231	add rbx,rdx ; Inc pointer into ref data
michael@0	232	add rax,rcx ; Inc pointer into the new data
michael@0	233	movq mm1, [rbx] ; Copy eight bytes to mm1
michael@0	234	paddd mm7, mm0 ; accumulate in mm7
michael@0	235	paddd mm7, mm2 ; accumulate in mm7
michael@0	236
michael@0	237	; Row 7
michael@0	238	movq mm0, [rax] ; Copy eight bytes to mm0
michael@0	239	movq mm2, mm0 ; Take copies
michael@0	240	movq mm3, mm1 ; Take copies
michael@0	241
michael@0	242	punpcklbw mm0, mm6 ; unpack to higher prrcision
michael@0	243	punpcklbw mm1, mm6
michael@0	244	punpckhbw mm2, mm6 ; unpack to higher prrcision
michael@0	245	punpckhbw mm3, mm6
michael@0	246	psubsw mm0, mm1 ; A-B (low order) to MM0
michael@0	247	psubsw mm2, mm3 ; A-B (high order) to MM2
michael@0	248
michael@0	249	paddw mm5, mm0 ; accumulate differences in mm5
michael@0	250	paddw mm5, mm2 ; accumulate differences in mm5
michael@0	251
michael@0	252	pmaddwd mm0, mm0 ; square and accumulate
michael@0	253	pmaddwd mm2, mm2 ; square and accumulate
michael@0	254	add rbx,rdx ; Inc pointer into ref data
michael@0	255	add rax,rcx ; Inc pointer into the new data
michael@0	256	movq mm1, [rbx] ; Copy eight bytes to mm1
michael@0	257	paddd mm7, mm0 ; accumulate in mm7
michael@0	258	paddd mm7, mm2 ; accumulate in mm7
michael@0	259
michael@0	260	; Row 8
michael@0	261	movq mm0, [rax] ; Copy eight bytes to mm0
michael@0	262	movq mm2, mm0 ; Take copies
michael@0	263	movq mm3, mm1 ; Take copies
michael@0	264
michael@0	265	punpcklbw mm0, mm6 ; unpack to higher prrcision
michael@0	266	punpcklbw mm1, mm6
michael@0	267	punpckhbw mm2, mm6 ; unpack to higher prrcision
michael@0	268	punpckhbw mm3, mm6
michael@0	269	psubsw mm0, mm1 ; A-B (low order) to MM0
michael@0	270	psubsw mm2, mm3 ; A-B (high order) to MM2
michael@0	271
michael@0	272	paddw mm5, mm0 ; accumulate differences in mm5
michael@0	273	paddw mm5, mm2 ; accumulate differences in mm5
michael@0	274
michael@0	275	pmaddwd mm0, mm0 ; square and accumulate
michael@0	276	pmaddwd mm2, mm2 ; square and accumulate
michael@0	277	add rbx,rdx ; Inc pointer into ref data
michael@0	278	add rax,rcx ; Inc pointer into the new data
michael@0	279	paddd mm7, mm0 ; accumulate in mm7
michael@0	280	paddd mm7, mm2 ; accumulate in mm7
michael@0	281
michael@0	282	; Now accumulate the final results.
michael@0	283	movq QWORD PTR [rsp+8], mm5 ; copy back accumulated results into normal memory
michael@0	284	movq QWORD PTR [rsp], mm7 ; copy back accumulated results into normal memory
michael@0	285	movsx rdx, WORD PTR [rsp+8]
michael@0	286	movsx rcx, WORD PTR [rsp+10]
michael@0	287	movsx rbx, WORD PTR [rsp+12]
michael@0	288	movsx rax, WORD PTR [rsp+14]
michael@0	289	add rdx, rcx
michael@0	290	add rbx, rax
michael@0	291	add rdx, rbx ;XSum
michael@0	292	movsxd rax, DWORD PTR [rsp]
michael@0	293	movsxd rcx, DWORD PTR [rsp+4]
michael@0	294	add rax, rcx ;XXSum
michael@0	295	mov rsi, arg(4) ;SSE
michael@0	296	mov rdi, arg(5) ;Sum
michael@0	297	mov dword ptr [rsi], eax
michael@0	298	mov dword ptr [rdi], edx
michael@0	299	xor rax, rax ; return 0
michael@0	300
michael@0	301
michael@0	302	; begin epilog
michael@0	303	add rsp, 16
michael@0	304	pop rbx
michael@0	305	pop rdi
michael@0	306	pop rsi
michael@0	307	UNSHADOW_ARGS
michael@0	308	pop rbp
michael@0	309	ret
michael@0	310
michael@0	311
michael@0	312
michael@0	313	;unsigned int
michael@0	314	;vp9_get4x4var_mmx
michael@0	315	;(
michael@0	316	; unsigned char *src_ptr,
michael@0	317	; int source_stride,
michael@0	318	; unsigned char *ref_ptr,
michael@0	319	; int recon_stride,
michael@0	320	; unsigned int *SSE,
michael@0	321	; int *Sum
michael@0	322	;)
michael@0	323	global sym(vp9_get4x4var_mmx) PRIVATE
michael@0	324	sym(vp9_get4x4var_mmx):
michael@0	325	push rbp
michael@0	326	mov rbp, rsp
michael@0	327	SHADOW_ARGS_TO_STACK 6
michael@0	328	push rsi
michael@0	329	push rdi
michael@0	330	push rbx
michael@0	331	sub rsp, 16
michael@0	332	; end prolog
michael@0	333
michael@0	334
michael@0	335	pxor mm5, mm5 ; Blank mmx6
michael@0	336	pxor mm6, mm6 ; Blank mmx7
michael@0	337	pxor mm7, mm7 ; Blank mmx7
michael@0	338
michael@0	339	mov rax, arg(0) ;[src_ptr] ; Load base addresses
michael@0	340	mov rbx, arg(2) ;[ref_ptr]
michael@0	341	movsxd rcx, dword ptr arg(1) ;[source_stride]
michael@0	342	movsxd rdx, dword ptr arg(3) ;[recon_stride]
michael@0	343
michael@0	344	; Row 1
michael@0	345	movd mm0, [rax] ; Copy 4 bytes to mm0
michael@0	346	movd mm1, [rbx] ; Copy 4 bytes to mm1
michael@0	347	punpcklbw mm0, mm6 ; unpack to higher prrcision
michael@0	348	punpcklbw mm1, mm6
michael@0	349	psubsw mm0, mm1 ; A-B (low order) to MM0
michael@0	350	paddw mm5, mm0 ; accumulate differences in mm5
michael@0	351	pmaddwd mm0, mm0 ; square and accumulate
michael@0	352	add rbx,rdx ; Inc pointer into ref data
michael@0	353	add rax,rcx ; Inc pointer into the new data
michael@0	354	movd mm1, [rbx] ; Copy 4 bytes to mm1
michael@0	355	paddd mm7, mm0 ; accumulate in mm7
michael@0	356
michael@0	357
michael@0	358	; Row 2
michael@0	359	movd mm0, [rax] ; Copy 4 bytes to mm0
michael@0	360	punpcklbw mm0, mm6 ; unpack to higher prrcision
michael@0	361	punpcklbw mm1, mm6
michael@0	362	psubsw mm0, mm1 ; A-B (low order) to MM0
michael@0	363	paddw mm5, mm0 ; accumulate differences in mm5
michael@0	364
michael@0	365	pmaddwd mm0, mm0 ; square and accumulate
michael@0	366	add rbx,rdx ; Inc pointer into ref data
michael@0	367	add rax,rcx ; Inc pointer into the new data
michael@0	368	movd mm1, [rbx] ; Copy 4 bytes to mm1
michael@0	369	paddd mm7, mm0 ; accumulate in mm7
michael@0	370
michael@0	371	; Row 3
michael@0	372	movd mm0, [rax] ; Copy 4 bytes to mm0
michael@0	373	punpcklbw mm0, mm6 ; unpack to higher prrcision
michael@0	374	punpcklbw mm1, mm6
michael@0	375	psubsw mm0, mm1 ; A-B (low order) to MM0
michael@0	376	paddw mm5, mm0 ; accumulate differences in mm5
michael@0	377
michael@0	378	pmaddwd mm0, mm0 ; square and accumulate
michael@0	379	add rbx,rdx ; Inc pointer into ref data
michael@0	380	add rax,rcx ; Inc pointer into the new data
michael@0	381	movd mm1, [rbx] ; Copy 4 bytes to mm1
michael@0	382	paddd mm7, mm0 ; accumulate in mm7
michael@0	383
michael@0	384	; Row 4
michael@0	385	movd mm0, [rax] ; Copy 4 bytes to mm0
michael@0	386
michael@0	387	punpcklbw mm0, mm6 ; unpack to higher prrcision
michael@0	388	punpcklbw mm1, mm6
michael@0	389	psubsw mm0, mm1 ; A-B (low order) to MM0
michael@0	390
michael@0	391	paddw mm5, mm0 ; accumulate differences in mm5
michael@0	392
michael@0	393	pmaddwd mm0, mm0 ; square and accumulate
michael@0	394	paddd mm7, mm0 ; accumulate in mm7
michael@0	395
michael@0	396
michael@0	397	; Now accumulate the final results.
michael@0	398	movq QWORD PTR [rsp+8], mm5 ; copy back accumulated results into normal memory
michael@0	399	movq QWORD PTR [rsp], mm7 ; copy back accumulated results into normal memory
michael@0	400	movsx rdx, WORD PTR [rsp+8]
michael@0	401	movsx rcx, WORD PTR [rsp+10]
michael@0	402	movsx rbx, WORD PTR [rsp+12]
michael@0	403	movsx rax, WORD PTR [rsp+14]
michael@0	404	add rdx, rcx
michael@0	405	add rbx, rax
michael@0	406	add rdx, rbx ;XSum
michael@0	407	movsxd rax, DWORD PTR [rsp]
michael@0	408	movsxd rcx, DWORD PTR [rsp+4]
michael@0	409	add rax, rcx ;XXSum
michael@0	410	mov rsi, arg(4) ;SSE
michael@0	411	mov rdi, arg(5) ;Sum
michael@0	412	mov dword ptr [rsi], eax
michael@0	413	mov dword ptr [rdi], edx
michael@0	414	xor rax, rax ; return 0
michael@0	415
michael@0	416
michael@0	417	; begin epilog
michael@0	418	add rsp, 16
michael@0	419	pop rbx
michael@0	420	pop rdi
michael@0	421	pop rsi
michael@0	422	UNSHADOW_ARGS
michael@0	423	pop rbp
michael@0	424	ret
michael@0	425
michael@0	426
michael@0	427
michael@0	428	;unsigned int
michael@0	429	;vp9_get4x4sse_cs_mmx
michael@0	430	;(
michael@0	431	; unsigned char *src_ptr,
michael@0	432	; int source_stride,
michael@0	433	; unsigned char *ref_ptr,
michael@0	434	; int recon_stride
michael@0	435	;)
michael@0	436	global sym(vp9_get4x4sse_cs_mmx) PRIVATE
michael@0	437	sym(vp9_get4x4sse_cs_mmx):
michael@0	438	push rbp
michael@0	439	mov rbp, rsp
michael@0	440	SHADOW_ARGS_TO_STACK 4
michael@0	441	push rsi
michael@0	442	push rdi
michael@0	443	push rbx
michael@0	444	; end prolog
michael@0	445
michael@0	446
michael@0	447	pxor mm6, mm6 ; Blank mmx7
michael@0	448	pxor mm7, mm7 ; Blank mmx7
michael@0	449
michael@0	450	mov rax, arg(0) ;[src_ptr] ; Load base addresses
michael@0	451	mov rbx, arg(2) ;[ref_ptr]
michael@0	452	movsxd rcx, dword ptr arg(1) ;[source_stride]
michael@0	453	movsxd rdx, dword ptr arg(3) ;[recon_stride]
michael@0	454	; Row 1
michael@0	455	movd mm0, [rax] ; Copy eight bytes to mm0
michael@0	456	movd mm1, [rbx] ; Copy eight bytes to mm1
michael@0	457	punpcklbw mm0, mm6 ; unpack to higher prrcision
michael@0	458	punpcklbw mm1, mm6
michael@0	459	psubsw mm0, mm1 ; A-B (low order) to MM0
michael@0	460	pmaddwd mm0, mm0 ; square and accumulate
michael@0	461	add rbx,rdx ; Inc pointer into ref data
michael@0	462	add rax,rcx ; Inc pointer into the new data
michael@0	463	movd mm1, [rbx] ; Copy eight bytes to mm1
michael@0	464	paddd mm7, mm0 ; accumulate in mm7
michael@0	465
michael@0	466	; Row 2
michael@0	467	movd mm0, [rax] ; Copy eight bytes to mm0
michael@0	468	punpcklbw mm0, mm6 ; unpack to higher prrcision
michael@0	469	punpcklbw mm1, mm6
michael@0	470	psubsw mm0, mm1 ; A-B (low order) to MM0
michael@0	471	pmaddwd mm0, mm0 ; square and accumulate
michael@0	472	add rbx,rdx ; Inc pointer into ref data
michael@0	473	add rax,rcx ; Inc pointer into the new data
michael@0	474	movd mm1, [rbx] ; Copy eight bytes to mm1
michael@0	475	paddd mm7, mm0 ; accumulate in mm7
michael@0	476
michael@0	477	; Row 3
michael@0	478	movd mm0, [rax] ; Copy eight bytes to mm0
michael@0	479	punpcklbw mm1, mm6
michael@0	480	punpcklbw mm0, mm6 ; unpack to higher prrcision
michael@0	481	psubsw mm0, mm1 ; A-B (low order) to MM0
michael@0	482
michael@0	483	pmaddwd mm0, mm0 ; square and accumulate
michael@0	484	add rbx,rdx ; Inc pointer into ref data
michael@0	485	add rax,rcx ; Inc pointer into the new data
michael@0	486	movd mm1, [rbx] ; Copy eight bytes to mm1
michael@0	487	paddd mm7, mm0 ; accumulate in mm7
michael@0	488
michael@0	489	; Row 4
michael@0	490	movd mm0, [rax] ; Copy eight bytes to mm0
michael@0	491	punpcklbw mm0, mm6 ; unpack to higher prrcision
michael@0	492	punpcklbw mm1, mm6
michael@0	493	psubsw mm0, mm1 ; A-B (low order) to MM0
michael@0	494	pmaddwd mm0, mm0 ; square and accumulate
michael@0	495	paddd mm7, mm0 ; accumulate in mm7
michael@0	496
michael@0	497	movq mm0, mm7 ;
michael@0	498	psrlq mm7, 32
michael@0	499
michael@0	500	paddd mm0, mm7
michael@0	501	movq rax, mm0
michael@0	502
michael@0	503
michael@0	504	; begin epilog
michael@0	505	pop rbx
michael@0	506	pop rdi
michael@0	507	pop rsi
michael@0	508	UNSHADOW_ARGS
michael@0	509	pop rbp
michael@0	510	ret