fix win64 float issue, enable AQ assembly

volvet committed 12 years ago

Commit 5c9f4471 parent ae027b815 files · +2088 −1642Browse files

Showing the changes to codec/processing/src (6 of 15 files). Show all

Unified · Split

6 files

codec/processing/src/adaptivequantization/AdaptiveQuantization.cpp +7 −1
codec/processing/src/adaptivequantization/AdaptiveQuantization.h +2 −0
codec/processing/src/asm/vaa.asm +1907 −1264
codec/processing/src/common/cpu.cpp +0 −196
codec/processing/src/common/cpu.h +0 −102
codec/processing/src/vaacalc/vaacalculation.cpp +5 −5

codec/processing/src/adaptivequantization/AdaptiveQuantization.cpp+7−1

@@ -50,9 +50,11 @@ CAdaptiveQuantization::CAdaptiveQuantization (int32_t iCpuFlag) { · 49 unmodified lines
50	50	m_pfVar = NULL;
51	51	WelsMemset (&m_sAdaptiveQuantParam, 0, sizeof (m_sAdaptiveQuantParam));
52	52	WelsInitVarFunc (m_pfVar, m_CPUFlag);
	53	+ XMMREG_PROTECT_INIT(AdaptiveQuantization);
53	54	}
54	55
55	56	CAdaptiveQuantization::~CAdaptiveQuantization() {
	57	+ XMMREG_PROTECT_UNINIT(AdaptiveQuantization);
56	58	}
57	59
58	60	EResult CAdaptiveQuantization::Process (int32_t iType, SPixMap* pSrcPixMap, SPixMap* pRefPixMap) {
@@ -101,6 +103,7 @@ EResult CAdaptiveQuantization::Process (int32_t iType, SPixMap* pSrcPixMap, SPix · 42 unmodified lines
101	103	pRefFrameTmp = pRefFrameY;
102	104	pCurFrameTmp = pCurFrameY;
103	105	for (i = 0; i < iMbWidth; i++) {
	106	+ XMMREG_PROTECT_STORE(AdaptiveQuantization);
104	107	iSumDiff = pVaaCalcResults->pSad8x8[iMbIndex][0];
105	108	iSumDiff += pVaaCalcResults->pSad8x8[iMbIndex][1];
106	109	iSumDiff += pVaaCalcResults->pSad8x8[iMbIndex][2];
@@ -109,6 +112,7 @@ EResult CAdaptiveQuantization::Process (int32_t iType, SPixMap* pSrcPixMap, SPix · 2 unmodified lines
109	112	iSQDiff = pVaaCalcResults->pSsd16x16[iMbIndex];
110	113	uiSum = pVaaCalcResults->pSum16x16[iMbIndex];
111	114	iSQSum = pVaaCalcResults->pSumOfSquare16x16[iMbIndex];
	115	+ XMMREG_PROTECT_LOAD(AdaptiveQuantization);
112	116
113	117	iSumDiff = iSumDiff >> 8;
114	118	pMotionTexture->uiMotionIndex = (iSQDiff >> 8) - (iSumDiff * iSumDiff);
@@ -131,7 +135,9 @@ EResult CAdaptiveQuantization::Process (int32_t iType, SPixMap* pSrcPixMap, SPix · 16 unmodified lines
131	135	pRefFrameTmp = pRefFrameY;
132	136	pCurFrameTmp = pCurFrameY;
133	137	for (i = 0; i < iMbWidth; i++) {
	138	+ XMMREG_PROTECT_STORE(AdaptiveQuantization);
134	139	m_pfVar (pRefFrameTmp, iRefStride, pCurFrameTmp, iCurStride, pMotionTexture);
	140	+ XMMREG_PROTECT_LOAD(AdaptiveQuantization);
135	141	dAverageMotionIndex += pMotionTexture->uiMotionIndex;
136	142	dAverageTextureIndex += pMotionTexture->uiTextureIndex;
137	143	pMotionTexture++;
@@ -223,7 +229,7 @@ void CAdaptiveQuantization::WelsInitVarFunc (PVarFunc& pfVar, int32_t iCpuFlag) · 85 unmodified lines
223	229
224	230	#ifdef X86_ASM
225	231	if (iCpuFlag & WELS_CPU_SSE2) {
226		− // pfVar = SampleVariance16x16_sse2;
	232	+ pfVar = SampleVariance16x16_sse2;
227	233	}
228	234	#endif
229	235	}

codec/processing/src/adaptivequantization/AdaptiveQuantization.h+2−0

@@ -45,6 +45,7 @@ · 44 unmodified lines
45	45	#include "../common/memory.h"
46	46	#include "../common/WelsFrameWork.h"
47	47	#include "../../interface/IWelsVP.h"
	48	+#include "cpu.h"
48	49
49	50	WELSVP_NAMESPACE_BEGIN
50	51
@@ -78,6 +79,7 @@ class CAdaptiveQuantization : public IStrategy { · 27 unmodified lines
78	79	PVarFunc m_pfVar;
79	80	int32_t m_CPUFlag;
80	81	SAdaptiveQuantizationParam m_sAdaptiveQuantParam;
	82	+ XMMREG_PROTECT_DECLARE(AdaptiveQuantization);
81	83	};
82	84
83	85	WELSVP_NAMESPACE_END

codec/processing/src/asm/vaa.asm+1907−1264

@@ -29,291 +29,211 @@ · 28 unmodified lines
29	29	;* POSSIBILITY OF SUCH DAMAGE.
30	30	;*
31	31	;*
32		−;* vaa.asm
	32	+;* vaa.asm
33	33	;*
34		−;* Abstract
	34	+;* Abstract
35	35	;* sse2 for pVaa routines
36	36	;*
37	37	;* History
38		−;* 04/14/2010 Created
	38	+;* 04/14/2010 Created
	39	+;* 06/07/2010 Added AnalysisVaaInfoIntra_sse2(ssse3)
	40	+;* 06/10/2010 Tune rc_sad_frame_sse2 and got about 40% improvement
	41	+;* 08/11/2010 Added abs_difference_mbrow_sse2 & sum_sqrsum_mbrow_sse2
39	42	;*
40	43	;*************************************************************************/
41	44	%include "asm_inc.asm"
42		−%ifdef X86_32
	45	+
	46	+
43	47	;***********************************************************************
44	48	; Macros and other preprocessor constants
45	49	;***********************************************************************
46		−
47		−;%macro SUM_SSE2 4 ; dst, pSrc, zero, pack1_8x2
48		−; movdqa %1, %2
49		−; punpcklbw %1, %3
50		−; punpckhbw %2, %3
51		−; paddw %1, %2
52		−; pmaddwd %1, %4
53		−; pshufd %2, %1, 04Eh ; 01001110 B
54		−; paddd %1, %2
55		−; pshufd %2, %1, 0B1h ; 10110001 B
56		−; paddd %1, %2
57		−;%endmacro ; END OF SUM_SSE2
58		−
59		−; by comparing it outperforms than phaddw(SSSE3) sets
60		−%macro SUM_WORD_8x2_SSE2 2 ; dst(pSrc), tmp
61		− ; @sum_8x2 begin
62		− pshufd %2, %1, 04Eh ; 01001110 B
63		− paddw %1, %2
64		− pshuflw %2, %1, 04Eh ; 01001110 B
65		− paddw %1, %2
66		− pshuflw %2, %1, 0B1h ; 10110001 B
67		− paddw %1, %2
68		− ; end of @sum_8x2
69		−%endmacro ; END of SUM_WORD_8x2_SSE2
70		−
71		−%macro SUM_SQR_SSE2 3 ; dst, pSrc, zero
72		− movdqa %1, %2
73		− punpcklbw %1, %3
74		− punpckhbw %2, %3
75		− pmaddwd %1, %1
76		− pmaddwd %2, %2
77		− paddd %1, %2
78		− pshufd %2, %1, 04Eh ; 01001110 B
79		− paddd %1, %2
80		− pshufd %2, %1, 0B1h ; 10110001 B
81		− paddd %1, %2
82		−%endmacro ; END OF SUM_SQR_SSE2
83		−
84		−%macro VAA_AVG_BLOCK_SSE2 6 ; dst, t0, t1, t2, t3, t4
85		− movdqa %1, [esi ] ; line 0
86		− movdqa %2, [esi+ecx] ; line 1
87		− movdqa %3, %1
88		− punpcklbw %1, xmm7
89		− punpckhbw %3, xmm7
90		− movdqa %4, %2
91		− punpcklbw %4, xmm7
92		− punpckhbw %2, xmm7
93		− paddw %1, %4
94		− paddw %2, %3
95		− movdqa %3, [esi+ebx] ; line 2
96		− movdqa %4, [esi+edx] ; line 3
97		− movdqa %5, %3
98		− punpcklbw %3, xmm7
99		− punpckhbw %5, xmm7
100		− movdqa %6, %4
101		− punpcklbw %6, xmm7
102		− punpckhbw %4, xmm7
103		− paddw %3, %6
104		− paddw %4, %5
105		− paddw %1, %3 ; block 0, 1
106		− paddw %2, %4 ; block 2, 3
107		− pshufd %3, %1, 0B1h
108		− pshufd %4, %2, 0B1h
109		− paddw %1, %3
110		− paddw %2, %4
111		− movdqa %3, %1
112		− movdqa %4, %2
113		− pshuflw %5, %1, 0B1h
114		− pshufhw %6, %3, 0B1h
115		− paddw %1, %5
116		− paddw %3, %6
117		− pshuflw %5, %2, 0B1h
118		− pshufhw %6, %4, 0B1h
119		− paddw %2, %5
120		− paddw %4, %6
121		− punpcklwd %1, %2
122		− punpckhwd %3, %4
123		− punpcklwd %1, %3
124		− psraw %1, $04
	50	+%macro SUM_SQR_SSE2 3 ; dst, pSrc, zero
	51	+ movdqa %1, %2
	52	+ punpcklbw %1, %3
	53	+ punpckhbw %2, %3
	54	+ pmaddwd %1, %1
	55	+ pmaddwd %2, %2
	56	+ paddd %1, %2
	57	+ pshufd %2, %1, 04Eh ; 01001110 B
	58	+ paddd %1, %2
	59	+ pshufd %2, %1, 0B1h ; 10110001 B
	60	+ paddd %1, %2
	61	+%endmacro ; END OF SUM_SQR_SSE2
	62	+
	63	+%macro WELS_SAD_16x2_SSE2 3 ;esi :%1 edi:%2 ebx:%3
	64	+ movdqa xmm1, [%1]
	65	+ movdqa xmm2, [%2]
	66	+ movdqa xmm3, [%1+%3]
	67	+ movdqa xmm4, [%2+%3]
	68	+ psadbw xmm1, xmm2
	69	+ psadbw xmm3, xmm4
	70	+ paddd xmm6, xmm1
	71	+ paddd xmm6, xmm3
	72	+ lea %1, [%1+%3*2]
	73	+ lea %2, [%2+%3*2]
125	74	%endmacro
126	75
127		−%macro VAA_AVG_BLOCK_SSSE3 6 ; dst, t0, t1, t2, t3, t4
128		− movdqa %1, [esi ] ; line 0
129		− movdqa %2, [esi+ecx] ; line 1
130		− movdqa %3, %1
131		− punpcklbw %1, xmm7
132		− punpckhbw %3, xmm7
133		− movdqa %4, %2
134		− punpcklbw %4, xmm7
135		− punpckhbw %2, xmm7
136		− paddw %1, %4
137		− paddw %2, %3
138		− movdqa %3, [esi+ebx] ; line 2
139		− movdqa %4, [esi+edx] ; line 3
140		− movdqa %5, %3
141		− punpcklbw %3, xmm7
142		− punpckhbw %5, xmm7
143		− movdqa %6, %4
144		− punpcklbw %6, xmm7
145		− punpckhbw %4, xmm7
146		− paddw %3, %6
147		− paddw %4, %5
148		− paddw %1, %3 ; block 0, 1
149		− paddw %2, %4 ; block 2, 3
150		− phaddw %1, %2 ; block[0]: 0-15, 16-31; block[1]: 32-47, 48-63; ..
151		− phaddw %1, xmm7 ; block[0]: 0-15; block[1]: 16-31; block[2]: 32-47; block[3]: 48-63; ....
152		− psraw %1, $04
153		−%endmacro
154		−
155		−%macro WELS_SAD_16x2_SSE2 0
156		− movdqa xmm1, [esi]
157		− movdqa xmm2, [edi]
158		− movdqa xmm3, [esi+ebx]
159		− movdqa xmm4, [edi+ebx]
160		− psadbw xmm1, xmm2
161		− psadbw xmm3, xmm4
162		− paddd xmm6, xmm1
163		− paddd xmm6, xmm3
164		− lea esi, [esi+ebx*2]
165		− lea edi, [edi+ebx*2]
166		−%endmacro
167		−
168		−%macro WELS_SAD_SUM_SQSUM_16x1_SSE2 0
169		− movdqa xmm1, [esi]
170		− movdqa xmm2, [edi]
171		− movdqa xmm3, xmm1
172		− psadbw xmm3, xmm2
173		− paddd xmm6, xmm3
174		−
175		− movdqa xmm3, xmm1
176		− psadbw xmm3, xmm0
177		− paddd xmm5, xmm3
178		−
179		− movdqa xmm2, xmm1
180		− punpcklbw xmm1, xmm0
181		− punpckhbw xmm2, xmm0
182		− pmaddwd xmm1, xmm1
183		− pmaddwd xmm2, xmm2
184		− paddd xmm4, xmm1
185		− paddd xmm4, xmm2
186		−
187		− add esi, ebx
188		− add edi, ebx
	76	+; by comparing it outperforms than phaddw(SSSE3) sets
	77	+%macro SUM_WORD_8x2_SSE2 2 ; dst(pSrc), tmp
	78	+ ; @sum_8x2 begin
	79	+ pshufd %2, %1, 04Eh ; 01001110 B
	80	+ paddw %1, %2
	81	+ pshuflw %2, %1, 04Eh ; 01001110 B
	82	+ paddw %1, %2
	83	+ pshuflw %2, %1, 0B1h ; 10110001 B
	84	+ paddw %1, %2
	85	+ ; end of @sum_8x2
	86	+%endmacro ; END of SUM_WORD_8x2_SSE2
	87	+
	88	+%macro WELS_SAD_SUM_SQSUM_16x1_SSE2 3 ;esi:%1,edi:%2,ebx:%3
	89	+ movdqa xmm1, [%1]
	90	+ movdqa xmm2, [%2]
	91	+ movdqa xmm3, xmm1
	92	+ psadbw xmm3, xmm2
	93	+ paddd xmm6, xmm3
	94	+
	95	+ movdqa xmm3, xmm1
	96	+ psadbw xmm3, xmm0
	97	+ paddd xmm5, xmm3
	98	+
	99	+ movdqa xmm2, xmm1
	100	+ punpcklbw xmm1, xmm0
	101	+ punpckhbw xmm2, xmm0
	102	+ pmaddwd xmm1, xmm1
	103	+ pmaddwd xmm2, xmm2
	104	+ paddd xmm4, xmm1
	105	+ paddd xmm4, xmm2
	106	+
	107	+ add %1, %3
	108	+ add %2, %3
189	109	%endmacro
190	110
191		−%macro WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 0
192		− movdqa xmm1, [esi]
193		− movdqa xmm2, [edi]
194		− movdqa xmm3, xmm1
195		− psadbw xmm3, xmm2
196		− paddd xmm7, xmm3 ; sad
197		−
198		− movdqa xmm3, xmm1
199		− pmaxub xmm3, xmm2
200		− pminub xmm2, xmm1
201		− psubb xmm3, xmm2 ; diff
202		−
203		− movdqa xmm2, xmm1
204		− psadbw xmm2, xmm0
205		− paddd xmm6, xmm2 ; sum
206		−
207		− movdqa xmm2, xmm1
208		− punpcklbw xmm1, xmm0
209		− punpckhbw xmm2, xmm0
210		− pmaddwd xmm1, xmm1
211		− pmaddwd xmm2, xmm2
212		− paddd xmm5, xmm1
213		− paddd xmm5, xmm2 ; sqsum
214		−
215		− movdqa xmm1, xmm3
216		− punpcklbw xmm1, xmm0
217		− punpckhbw xmm3, xmm0
218		− pmaddwd xmm1, xmm1
219		− pmaddwd xmm3, xmm3
220		− paddd xmm4, xmm1
221		− paddd xmm4, xmm3 ; sqdiff
222		−
223		− add esi, ebx
224		− add edi, ebx
	111	+%macro WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 3 ;esi:%1 edi:%2 ebx:%3
	112	+ movdqa xmm1, [%1]
	113	+ movdqa xmm2, [%2]
	114	+ movdqa xmm3, xmm1
	115	+ psadbw xmm3, xmm2
	116	+ paddd xmm7, xmm3 ; sad
	117	+
	118	+ movdqa xmm3, xmm1
	119	+ pmaxub xmm3, xmm2
	120	+ pminub xmm2, xmm1
	121	+ psubb xmm3, xmm2 ; diff
	122	+
	123	+ movdqa xmm2, xmm1
	124	+ psadbw xmm2, xmm0
	125	+ paddd xmm6, xmm2 ; sum
	126	+
	127	+ movdqa xmm2, xmm1
	128	+ punpcklbw xmm1, xmm0
	129	+ punpckhbw xmm2, xmm0
	130	+ pmaddwd xmm1, xmm1
	131	+ pmaddwd xmm2, xmm2
	132	+ paddd xmm5, xmm1
	133	+ paddd xmm5, xmm2 ; sqsum
	134	+
	135	+ movdqa xmm1, xmm3
	136	+ punpcklbw xmm1, xmm0
	137	+ punpckhbw xmm3, xmm0
	138	+ pmaddwd xmm1, xmm1
	139	+ pmaddwd xmm3, xmm3
	140	+ paddd xmm4, xmm1
	141	+ paddd xmm4, xmm3 ; sqdiff
	142	+
	143	+ add %1, %3
	144	+ add %2, %3
225	145	%endmacro
226	146
227		−%macro WELS_SAD_SD_MAD_16x1_SSE2 4
228		−%define sad_reg %1
229		−%define sum_cur_reg %2
230		−%define sum_ref_reg %3
231		−%define mad_reg %4
232		− movdqa xmm1, [esi]
233		− movdqa xmm2, [edi]
234		− movdqa xmm3, xmm1
235		− psadbw xmm3, xmm0
236		− paddd sum_cur_reg, xmm3 ; sum_cur
237		− movdqa xmm3, xmm2
238		− psadbw xmm3, xmm0
239		− paddd sum_ref_reg, xmm3 ; sum_ref
240		−
241		− movdqa xmm3, xmm1
242		− pmaxub xmm3, xmm2
243		− pminub xmm2, xmm1
244		− psubb xmm3, xmm2 ; abs diff
245		− pmaxub mad_reg, xmm3 ; max abs diff
246		−
247		− psadbw xmm3, xmm0
248		− paddd sad_reg, xmm3 ; sad
249		−
250		− add esi, ebx
251		− add edi, ebx
	147	+%macro WELS_SAD_SD_MAD_16x1_SSE2 7 ;esi:%5 edi:%6 ebx:%7
	148	+%define sad_reg %1
	149	+%define sum_cur_reg %2
	150	+%define sum_ref_reg %3
	151	+%define mad_reg %4
	152	+ movdqa xmm1, [%5]
	153	+ movdqa xmm2, [%6]
	154	+ movdqa xmm3, xmm1
	155	+ psadbw xmm3, xmm0
	156	+ paddd sum_cur_reg, xmm3 ; sum_cur
	157	+ movdqa xmm3, xmm2
	158	+ psadbw xmm3, xmm0
	159	+ paddd sum_ref_reg, xmm3 ; sum_ref
	160	+
	161	+ movdqa xmm3, xmm1
	162	+ pmaxub xmm3, xmm2
	163	+ pminub xmm2, xmm1
	164	+ psubb xmm3, xmm2 ; abs diff
	165	+ pmaxub mad_reg, xmm3 ; max abs diff
	166	+
	167	+ psadbw xmm3, xmm0
	168	+ paddd sad_reg, xmm3 ; sad
	169	+
	170	+ add %5, %7
	171	+ add %6, %7
252	172	%endmacro
253	173
254	174
255		−%macro WELS_MAX_REG_SSE2 1 ; xmm1, xmm2, xmm3 can be used
	175	+%macro WELS_MAX_REG_SSE2 1 ; xmm1, xmm2, xmm3 can be used
256	176	%define max_reg %1
257		− movdqa xmm1, max_reg
258		− psrldq xmm1, 4
259		− pmaxub max_reg, xmm1
260		− movdqa xmm1, max_reg
261		− psrldq xmm1, 2
262		− pmaxub max_reg, xmm1
263		− movdqa xmm1, max_reg
264		− psrldq xmm1, 1
265		− pmaxub max_reg, xmm1
	177	+ movdqa xmm1, max_reg
	178	+ psrldq xmm1, 4
	179	+ pmaxub max_reg, xmm1
	180	+ movdqa xmm1, max_reg
	181	+ psrldq xmm1, 2
	182	+ pmaxub max_reg, xmm1
	183	+ movdqa xmm1, max_reg
	184	+ psrldq xmm1, 1
	185	+ pmaxub max_reg, xmm1
266	186	%endmacro
267	187
268		−%macro WELS_SAD_BGD_SQDIFF_16x1_SSE2 4
269		−%define sad_reg %1
270		−%define sum_reg %2
271		−%define mad_reg %3
272		−%define sqdiff_reg %4
273		− movdqa xmm1, [esi]
274		− movdqa xmm2, xmm1
275		− movdqa xmm3, xmm1
276		− punpcklbw xmm2, xmm0
277		− punpckhbw xmm3, xmm0
278		− pmaddwd xmm2, xmm2
279		− pmaddwd xmm3, xmm3
280		− paddd xmm2, xmm3
281		− movdqa xmm3, xmm2
282		− psllq xmm2, 32
283		− psrlq xmm3, 32
284		− psllq xmm3, 32
285		− paddd xmm2, xmm3
286		− paddd sad_reg, xmm2 ; sqsum
287		−
288		− movdqa xmm2, [edi]
289		− movdqa xmm3, xmm1
290		− psadbw xmm3, xmm0
291		− paddd sum_reg, xmm3 ; sum_cur
292		− movdqa xmm3, xmm2
293		− psadbw xmm3, xmm0
294		− pslldq xmm3, 4
295		− paddd sum_reg, xmm3 ; sum_ref
296		−
297		− movdqa xmm3, xmm1
298		− pmaxub xmm3, xmm2
299		− pminub xmm2, xmm1
300		− psubb xmm3, xmm2 ; abs diff
301		− pmaxub mad_reg, xmm3 ; max abs diff
302		−
303		− movdqa xmm1, xmm3
304		− psadbw xmm3, xmm0
305		− paddd sad_reg, xmm3 ; sad
306		−
307		− movdqa xmm3, xmm1
308		− punpcklbw xmm1, xmm0
309		− punpckhbw xmm3, xmm0
310		− pmaddwd xmm1, xmm1
311		− pmaddwd xmm3, xmm3
312		− paddd sqdiff_reg, xmm1
313		− paddd sqdiff_reg, xmm3 ; sqdiff
314		−
315		− add esi, ebx
316		− add edi, ebx
	188	+%macro WELS_SAD_BGD_SQDIFF_16x1_SSE2 7 ;esi:%5 edi:%6 ebx:%7
	189	+%define sad_reg %1
	190	+%define sum_reg %2
	191	+%define mad_reg %3
	192	+%define sqdiff_reg %4
	193	+ movdqa xmm1, [%5]
	194	+ movdqa xmm2, xmm1
	195	+ movdqa xmm3, xmm1
	196	+ punpcklbw xmm2, xmm0
	197	+ punpckhbw xmm3, xmm0
	198	+ pmaddwd xmm2, xmm2
	199	+ pmaddwd xmm3, xmm3
	200	+ paddd xmm2, xmm3
	201	+ movdqa xmm3, xmm2
	202	+ psllq xmm2, 32
	203	+ psrlq xmm3, 32
	204	+ psllq xmm3, 32
	205	+ paddd xmm2, xmm3
	206	+ paddd sad_reg, xmm2 ; sqsum
	207	+
	208	+ movdqa xmm2, [%6]
	209	+ movdqa xmm3, xmm1
	210	+ psadbw xmm3, xmm0
	211	+ paddd sum_reg, xmm3 ; sum_cur
	212	+ movdqa xmm3, xmm2
	213	+ psadbw xmm3, xmm0
	214	+ pslldq xmm3, 4
	215	+ paddd sum_reg, xmm3 ; sum_ref
	216	+
	217	+ movdqa xmm3, xmm1
	218	+ pmaxub xmm3, xmm2
	219	+ pminub xmm2, xmm1
	220	+ psubb xmm3, xmm2 ; abs diff
	221	+ pmaxub mad_reg, xmm3 ; max abs diff
	222	+
	223	+ movdqa xmm1, xmm3
	224	+ psadbw xmm3, xmm0
	225	+ paddd sad_reg, xmm3 ; sad
	226	+
	227	+ movdqa xmm3, xmm1
	228	+ punpcklbw xmm1, xmm0
	229	+ punpckhbw xmm3, xmm0
	230	+ pmaddwd xmm1, xmm1
	231	+ pmaddwd xmm3, xmm3
	232	+ paddd sqdiff_reg, xmm1
	233	+ paddd sqdiff_reg, xmm3 ; sqdiff
	234	+
	235	+ add %5, %7
	236	+ add %6, %7
317	237	%endmacro
318	238
319	239
@@ -325,7 +245,7 @@ · 5 unmodified lines
325	245
326	246	;ALIGN 16
327	247	;pack1_8x2:
328		−; dw 1, 1, 1, 1, 1, 1, 1, 1
	248	+; dw 1, 1, 1, 1, 1, 1, 1, 1
329	249
330	250	;***********************************************************************
331	251	; Code
@@ -333,1082 +253,1805 @@ · 1 unmodified line
333	253
334	254	SECTION .text
335	255
336		−WELS_EXTERN rc_sad_frame_sse2
337		−;***********************************************************************
338		−; uint32_t rc_sad_frame_sse2( uint8_t ref_orig, uint8_t cur_orig, const int mb_width, const int iPicHeight, const int iPicStride );
339		−;***********************************************************************
340		−ALIGN 16
341		−rc_sad_frame_sse2:
342		− push esi
343		− push edi
344		− push ebp
345		− push ebx
346		− push edx
347		−
348		− mov esi, [esp+24]
349		− mov edi, [esp+28]
350		− mov ebx, [esp+32]
351		− mov ecx, [esp+36]
352		− mov edx, [esp+40]
353		− pxor xmm0, xmm0
354		−.hloop:
355		− mov eax, ebx
356		− mov ebp, $00
357		−.wloop:
358		− movdqa xmm1, [esi+ebp]
359		− movdqa xmm2, [edi+ebp]
360		− psadbw xmm1, xmm2
361		− pshufd xmm2, xmm1, 0f6h ; 11110110 B ; movhlps for float
362		− paddd xmm1, xmm2
363		− paddd xmm0, xmm1
364		− add ebp, 010h
365		− dec eax
366		− jnz near .wloop
367		− lea esi, [esi+edx]
368		− lea edi, [edi+edx]
369		− dec ecx
370		− jnz near .hloop
371		−
372		− movd eax, xmm0
373		− pop edx
374		− pop ebx
375		− pop ebp
376		− pop edi
377		− pop esi
378		− ret
379		−
	256	+%ifdef X86_32
380	257
381	258	WELS_EXTERN SampleVariance16x16_sse2
382	259	;***********************************************************************
383		−; void SampleVariance16x16_sse2( uint8_t * y_ref, int32_t y_ref_stride, uint8_t * y_src, int32_t y_src_stride,SMotionTextureUnit* pMotionTexture );
	260	+; void SampleVariance16x16_sse2( uint8_t * y_ref, int32_t y_ref_stride, uint8_t * y_src, int32_t y_src_stride,SMotionTextureUnit* pMotionTexture );
384	261	;***********************************************************************
385	262	ALIGN 16
386	263	SampleVariance16x16_sse2:
387		− push esi
388		− push edi
389		− push ebx
390		−
391		− sub esp, 16
392		− %define SUM [esp]
393		− %define SUM_CUR [esp+4]
394		− %define SQR [esp+8]
395		− %define SQR_CUR [esp+12]
396		− %define PUSH_SIZE 28 ; 12 + 16
397		−
398		− mov edi, [esp+PUSH_SIZE+4] ; y_ref
399		− mov edx, [esp+PUSH_SIZE+8] ; y_ref_stride
400		− mov esi, [esp+PUSH_SIZE+12] ; y_src
401		− mov eax, [esp+PUSH_SIZE+16] ; y_src_stride
402		− mov ecx, 010h ; height = 16
403		−
404		− pxor xmm7, xmm7
405		− movdqu SUM, xmm7
	264	+ push esi
	265	+ push edi
	266	+ push ebx
	267	+
	268	+ sub esp, 16
	269	+ %define SUM [esp]
	270	+ %define SUM_CUR [esp+4]
	271	+ %define SQR [esp+8]
	272	+ %define SQR_CUR [esp+12]
	273	+ %define PUSH_SIZE 28 ; 12 + 16
	274	+
	275	+ mov edi, [esp+PUSH_SIZE+4] ; y_ref
	276	+ mov edx, [esp+PUSH_SIZE+8] ; y_ref_stride
	277	+ mov esi, [esp+PUSH_SIZE+12] ; y_src
	278	+ mov eax, [esp+PUSH_SIZE+16] ; y_src_stride
	279	+ mov ecx, 010h ; height = 16
	280	+
	281	+ pxor xmm7, xmm7
	282	+ movdqu SUM, xmm7
406	283
407	284	.hloops:
408		− movdqa xmm0, [edi] ; y_ref
409		− movdqa xmm1, [esi] ; y_src
410		− movdqa xmm2, xmm0 ; store first for future process
411		− movdqa xmm3, xmm1
412		− ; sum += diff;
413		− movdqa xmm4, xmm0
414		− psadbw xmm4, xmm1 ; 2 parts, [0,..,15], [64,..,79]
415		− ; to be continued for sum
416		− pshufd xmm5, xmm4, 0C6h ; 11000110 B
417		− paddw xmm4, xmm5
418		− movd ebx, xmm4
419		− add SUM, ebx
420		−
421		− ; sqr += diff * diff;
422		− pmaxub xmm0, xmm1
423		− pminub xmm1, xmm2
424		− psubb xmm0, xmm1 ; diff
425		− SUM_SQR_SSE2 xmm1, xmm0, xmm7 ; dst, pSrc, zero
426		− movd ebx, xmm1
427		− add SQR, ebx
428		−
429		− ; sum_cur += y_src[x];
430		− movdqa xmm0, xmm3 ; cur_orig
431		− movdqa xmm1, xmm0
432		− punpcklbw xmm0, xmm7
433		− punpckhbw xmm1, xmm7
434		− paddw xmm0, xmm1 ; 8x2
435		− SUM_WORD_8x2_SSE2 xmm0, xmm1
436		− movd ebx, xmm0
437		− and ebx, 0ffffh
438		− add SUM_CUR, ebx
439		−
440		− ; sqr_cur += y_src[x] * y_src[x];
441		− SUM_SQR_SSE2 xmm0, xmm3, xmm7 ; dst, pSrc, zero
442		− movd ebx, xmm0
443		− add SQR_CUR, ebx
444		−
445		− lea edi, [edi+edx]
446		− lea esi, [esi+eax]
447		− dec ecx
448		− jnz near .hloops
449		−
450		− mov ebx, 0
451		− mov bx, word SUM
452		− sar ebx, 8
453		− imul ebx, ebx
454		− mov ecx, SQR
455		− sar ecx, 8
456		− sub ecx, ebx
457		− mov edi, [esp+PUSH_SIZE+20] ; pMotionTexture
458		− mov [edi], cx ; to store uiMotionIndex
459		− mov ebx, 0
460		− mov bx, word SUM_CUR
461		− sar ebx, 8
462		− imul ebx, ebx
463		− mov ecx, SQR_CUR
464		− sar ecx, 8
465		− sub ecx, ebx
466		− mov [edi+2], cx ; to store uiTextureIndex
467		−
468		− %undef SUM
469		− %undef SUM_CUR
470		− %undef SQR
471		− %undef SQR_CUR
472		− %undef PUSH_SIZE
473		−
474		− add esp, 16
475		− pop ebx
476		− pop edi
477		− pop esi
478		−
479		− ret
480		−
481		−; , 6/7/2010
482		−
483		−
484		−WELS_EXTERN abs_difference_mbrow_sse2
	285	+ movdqa xmm0, [edi] ; y_ref
	286	+ movdqa xmm1, [esi] ; y_src
	287	+ movdqa xmm2, xmm0 ; store first for future process
	288	+ movdqa xmm3, xmm1
	289	+ ; sum += diff;
	290	+ movdqa xmm4, xmm0
	291	+ psadbw xmm4, xmm1 ; 2 parts, [0,..,15], [64,..,79]
	292	+ ; to be continued for sum
	293	+ pshufd xmm5, xmm4, 0C6h ; 11000110 B
	294	+ paddw xmm4, xmm5
	295	+ movd ebx, xmm4
	296	+ add SUM, ebx
	297	+
	298	+ ; sqr += diff * diff;
	299	+ pmaxub xmm0, xmm1
	300	+ pminub xmm1, xmm2
	301	+ psubb xmm0, xmm1 ; diff
	302	+ SUM_SQR_SSE2 xmm1, xmm0, xmm7 ; dst, pSrc, zero
	303	+ movd ebx, xmm1
	304	+ add SQR, ebx
	305	+
	306	+ ; sum_cur += y_src[x];
	307	+ movdqa xmm0, xmm3 ; cur_orig
	308	+ movdqa xmm1, xmm0
	309	+ punpcklbw xmm0, xmm7
	310	+ punpckhbw xmm1, xmm7
	311	+ paddw xmm0, xmm1 ; 8x2
	312	+ SUM_WORD_8x2_SSE2 xmm0, xmm1
	313	+ movd ebx, xmm0
	314	+ and ebx, 0ffffh
	315	+ add SUM_CUR, ebx
	316	+
	317	+ ; sqr_cur += y_src[x] * y_src[x];
	318	+ SUM_SQR_SSE2 xmm0, xmm3, xmm7 ; dst, pSrc, zero
	319	+ movd ebx, xmm0
	320	+ add SQR_CUR, ebx
	321	+
	322	+ lea edi, [edi+edx]
	323	+ lea esi, [esi+eax]
	324	+ dec ecx
	325	+ jnz near .hloops
	326	+
	327	+ mov ebx, 0
	328	+ mov bx, word SUM
	329	+ sar ebx, 8
	330	+ imul ebx, ebx
	331	+ mov ecx, SQR
	332	+ sar ecx, 8
	333	+ sub ecx, ebx
	334	+ mov edi, [esp+PUSH_SIZE+20] ; pMotionTexture
	335	+ mov [edi], cx ; to store uiMotionIndex
	336	+ mov ebx, 0
	337	+ mov bx, word SUM_CUR
	338	+ sar ebx, 8
	339	+ imul ebx, ebx
	340	+ mov ecx, SQR_CUR
	341	+ sar ecx, 8
	342	+ sub ecx, ebx
	343	+ mov [edi+2], cx ; to store uiTextureIndex
	344	+
	345	+ %undef SUM
	346	+ %undef SUM_CUR
	347	+ %undef SQR
	348	+ %undef SQR_CUR
	349	+ %undef PUSH_SIZE
	350	+
	351	+ add esp, 16
	352	+ pop ebx
	353	+ pop edi
	354	+ pop esi
	355	+
	356	+ ret
	357	+
	358	+
	359	+
	360	+WELS_EXTERN VAACalcSad_sse2
485	361	;*************************************************************************************************************
486		−;void abs_difference_mbrow_sse2( uint8_t ref_orig, uint8_t cur_orig, int32_t iPicStride,
487		−; int32_t gom_pixel_num, int32_t *pSum)
	362	+;void VAACalcSad_sse2( uint8_t cur_data, uint8_t ref_data, int32_t iPicWidth, int32_t iPicHeight
	363	+; int32_t iPicStride, int32_t psadframe, int32_t psad8x8)
488	364	;*************************************************************************************************************
	365	+
	366	+
489	367	ALIGN 16
490		−abs_difference_mbrow_sse2:
491		−%define ref_orig esp + pushsize + 4
492		−%define cur_orig esp + pushsize + 8
493		−%define iPicStride esp + pushsize + 12
494		−%define gom_pixel_num esp + pushsize + 16
495		−%define pSum esp + pushsize + 20
496		−%define pushsize 12
497		− push esi
498		− push edi
499		− push ebx
500		− mov esi, [ref_orig]
501		− mov edi, [cur_orig]
502		− mov ebx, [iPicStride]
503		− mov eax, [gom_pixel_num]
504		− mov ecx, 16 ;MB_WIDTH_LUMA
505		− pxor xmm0, xmm0
506		−mb_width_loop_p:
507		− mov edx, esi
508		− add edx, eax ; end address
509		−gom_row_loop_p:
510		− movdqa xmm1, [esi]
511		− movdqa xmm2, [edi]
512		− psadbw xmm1, xmm2
513		− paddd xmm0, xmm1
514		− add esi, 16
515		− add edi, 16
516		− cmp esi, edx
517		− jl gom_row_loop_p
518		−
519		− sub esi, eax
520		− sub edi, eax
521		− add esi, ebx
522		− add edi, ebx
523		− loop mb_width_loop_p
524		−
525		− movdqa xmm1, xmm0
526		− psrldq xmm1, 8
527		− paddd xmm1, xmm0
528		− movd eax, xmm1
529		− mov edx, [pSum] ; pSum
530		− add [edx], eax
531		−
532		−%undef ref_orig
533		−%undef cur_orig
534		−%undef iPicStride
535		−%undef gom_pixel_num
536		−%undef pSum
537		−%undef pushsize
538		− pop ebx
539		− pop edi
540		− pop esi
541		− ret
542		−
543		−
544		−
545		−
546		−WELS_EXTERN sum_sqrsum_mbrow_sse2
547		−;*************************************************************************************************************
548		−;void sum_sqrsum_mbrow_sse2( uint8_t *cur_orig, int32_t iPicStride,
549		−; int32_t gom_pixel_num, int32_t pSum, int32_t pSqrSum)
550		−;*************************************************************************************************************
	368	+VAACalcSad_sse2:
	369	+%define cur_data esp + pushsize + 4
	370	+%define ref_data esp + pushsize + 8
	371	+%define iPicWidth esp + pushsize + 12
	372	+%define iPicHeight esp + pushsize + 16
	373	+%define iPicStride esp + pushsize + 20
	374	+%define psadframe esp + pushsize + 24
	375	+%define psad8x8 esp + pushsize + 28
	376	+%define pushsize 12
	377	+ push esi
	378	+ push edi
	379	+ push ebx
	380	+ mov esi, [cur_data]
	381	+ mov edi, [ref_data]
	382	+ mov ebx, [iPicStride]
	383	+ mov edx, [psad8x8]
	384	+ mov eax, ebx
	385	+
	386	+ shr dword [iPicWidth], 4 ; iPicWidth/16
	387	+ shr dword [iPicHeight], 4 ; iPicHeight/16
	388	+ shl eax, 4 ; iPicStride*16
	389	+ pxor xmm0, xmm0
	390	+ pxor xmm7, xmm7 ; iFrameSad
	391	+height_loop:
	392	+ mov ecx, dword [iPicWidth]
	393	+ push esi
	394	+ push edi
	395	+width_loop:
	396	+ pxor xmm6, xmm6 ;
	397	+ WELS_SAD_16x2_SSE2 esi,edi,ebx
	398	+ WELS_SAD_16x2_SSE2 esi,edi,ebx
	399	+ WELS_SAD_16x2_SSE2 esi,edi,ebx
	400	+ WELS_SAD_16x2_SSE2 esi,edi,ebx
	401	+ paddd xmm7, xmm6
	402	+ movd [edx], xmm6
	403	+ psrldq xmm6, 8
	404	+ movd [edx+4], xmm6
	405	+
	406	+ pxor xmm6, xmm6
	407	+ WELS_SAD_16x2_SSE2 esi,edi,ebx
	408	+ WELS_SAD_16x2_SSE2 esi,edi,ebx
	409	+ WELS_SAD_16x2_SSE2 esi,edi,ebx
	410	+ WELS_SAD_16x2_SSE2 esi,edi,ebx
	411	+ paddd xmm7, xmm6
	412	+ movd [edx+8], xmm6
	413	+ psrldq xmm6, 8
	414	+ movd [edx+12], xmm6
	415	+
	416	+ add edx, 16
	417	+ sub esi, eax
	418	+ sub edi, eax
	419	+ add esi, 16
	420	+ add edi, 16
	421	+
	422	+ dec ecx
	423	+ jnz width_loop
	424	+
	425	+ pop edi
	426	+ pop esi
	427	+ add esi, eax
	428	+ add edi, eax
	429	+
	430	+ dec dword [iPicHeight]
	431	+ jnz height_loop
	432	+
	433	+ mov edx, [psadframe]
	434	+ movdqa xmm5, xmm7
	435	+ psrldq xmm7, 8
	436	+ paddd xmm7, xmm5
	437	+ movd [edx], xmm7
	438	+
	439	+%undef cur_data
	440	+%undef ref_data
	441	+%undef iPicWidth
	442	+%undef iPicHeight
	443	+%undef iPicStride
	444	+%undef psadframe
	445	+%undef psad8x8
	446	+%undef pushsize
	447	+ pop ebx
	448	+ pop edi
	449	+ pop esi
	450	+ ret
	451	+
	452	+%else ;64-bit
	453	+
	454	+WELS_EXTERN SampleVariance16x16_sse2
	455	+;***********************************************************************
	456	+; void SampleVariance16x16_sse2( uint8_t * y_ref, int32_t y_ref_stride, uint8_t * y_src, int32_t y_src_stride,SMotionTextureUnit* pMotionTexture );
	457	+;***********************************************************************
551	458	ALIGN 16
552		−sum_sqrsum_mbrow_sse2:
553		−%define cur_orig esp + pushsize + 4
554		−%define iPicStride esp + pushsize + 8
555		−%define gom_pixel_num esp + pushsize + 12
556		−%define pSum esp + pushsize + 16
557		−%define pSqrSum esp + pushsize + 20
558		−%define pushsize 8
559		− push esi
560		− push ebx
561		− mov esi, [cur_orig]
562		− mov eax, [gom_pixel_num]
563		− mov ebx, [iPicStride]
564		− mov ecx, 16 ;MB_WIDTH_LUMA
565		− pxor xmm0, xmm0 ; zero
566		− pxor xmm1, xmm1 ; sum
567		− pxor xmm2, xmm2 ; sqr sum
568		−mb_width_loop_i:
569		− mov edx, esi
570		− add edx, eax ; end address
571		−gom_row_loop_i:
572		− movdqa xmm3, [esi]
573		− movdqa xmm4, xmm3
574		− psadbw xmm4, xmm0
575		− paddd xmm1, xmm4
576		− movdqa xmm4, xmm3
577		− punpcklbw xmm4, xmm0
578		− punpckhbw xmm3, xmm0
579		− pmaddwd xmm4, xmm4
580		− pmaddwd xmm3, xmm3
581		− paddd xmm2, xmm3
582		− paddd xmm2, xmm4
583		− add esi, 16
584		− cmp esi, edx
585		− jl gom_row_loop_i
586		−
587		− sub esi, eax
588		− add esi, ebx
589		− loop mb_width_loop_i
590		−
591		− movdqa xmm3, xmm1
592		− psrldq xmm3, 8
593		− paddd xmm1, xmm3
594		− movd eax, xmm1
595		− mov edx, [pSum]
596		− add [edx], eax
597		−
598		− movdqa xmm3, xmm2
599		− psrldq xmm3, 8
600		− paddd xmm2, xmm3
601		− movdqa xmm3, xmm2
602		− psrldq xmm3, 4
603		− paddd xmm2, xmm3
604		− movd eax, xmm2
605		− mov edx, [pSqrSum]
606		− add [edx], eax
607		−
608		−
609		−%undef cur_orig
610		−%undef iPicStride
611		−%undef gom_pixel_num
612		−%undef pSum
613		−%undef pSqrSum
614		−%undef pushsize
615		− pop ebx
616		− pop esi
617		− ret
	459	+SampleVariance16x16_sse2:
	460	+ %define SUM r10;[esp]
	461	+ %define SUM_CUR r11;[esp+4]
	462	+ %define SQR r13;[esp+8]
	463	+ %define SQR_CUR r15;[esp+12]
	464	+
	465	+ push r12
	466	+ push r13
	467	+ push r14
	468	+ push r15
	469	+ %assign push_num 4
	470	+ LOAD_5_PARA
	471	+ SIGN_EXTENTION r1,r1d
	472	+ SIGN_EXTENTION r3,r3d
	473	+
	474	+ mov r12,010h
	475	+ pxor xmm7, xmm7
	476	+ movq SUM, xmm7
	477	+ movq SUM_CUR,xmm7
	478	+ movq SQR,xmm7
	479	+ movq SQR_CUR,xmm7
618	480
	481	+.hloops:
	482	+ mov r14,0
	483	+ movdqa xmm0, [r0] ; y_ref
	484	+ movdqa xmm1, [r2] ; y_src
	485	+ movdqa xmm2, xmm0 ; store first for future process
	486	+ movdqa xmm3, xmm1
	487	+ ; sum += diff;
	488	+ movdqa xmm4, xmm0
	489	+ psadbw xmm4, xmm1 ; 2 parts, [0,..,15], [64,..,79]
	490	+ ; to be continued for sum
	491	+ pshufd xmm5, xmm4, 0C6h ; 11000110 B
	492	+ paddw xmm4, xmm5
	493	+ movd r14d, xmm4
	494	+ add SUM, r14
	495	+
	496	+ ; sqr += diff * diff;
	497	+ pmaxub xmm0, xmm1
	498	+ pminub xmm1, xmm2
	499	+ psubb xmm0, xmm1 ; diff
	500	+ SUM_SQR_SSE2 xmm1, xmm0, xmm7 ; dst, pSrc, zero
	501	+ movd r14d, xmm1
	502	+ add SQR, r14
	503	+
	504	+ ; sum_cur += y_src[x];
	505	+ movdqa xmm0, xmm3 ; cur_orig
	506	+ movdqa xmm1, xmm0
	507	+ punpcklbw xmm0, xmm7
	508	+ punpckhbw xmm1, xmm7
	509	+ paddw xmm0, xmm1 ; 8x2
	510	+ SUM_WORD_8x2_SSE2 xmm0, xmm1
	511	+ movd r14d, xmm0
	512	+ and r14, 0ffffh
	513	+ add SUM_CUR, r14
	514	+
	515	+ ; sqr_cur += y_src[x] * y_src[x];
	516	+ SUM_SQR_SSE2 xmm0, xmm3, xmm7 ; dst, pSrc, zero
	517	+ movd r14d, xmm0
	518	+ add SQR_CUR, r14
	519	+
	520	+ lea r0, [r0+r1]
	521	+ lea r2, [r2+r3]
	522	+ dec r12
	523	+ jnz near .hloops
	524	+
	525	+ mov r0, SUM
	526	+ sar r0, 8
	527	+ imul r0, r0
	528	+ mov r1, SQR
	529	+ sar r1, 8
	530	+ sub r1, r0
	531	+ mov [r4], r1w ; to store uiMotionIndex
	532	+ mov r0, SUM_CUR
	533	+ sar r0, 8
	534	+ imul r0, r0
	535	+ mov r1, SQR_CUR
	536	+ sar r1, 8
	537	+ sub r1, r0
	538	+ mov [r4+2], r1w ; to store uiTextureIndex
	539	+
	540	+ LOAD_5_PARA_POP
	541	+ pop r15
	542	+ pop r14
	543	+ pop r13
	544	+ pop r12
	545	+
	546	+
	547	+ %assign push_num 0
	548	+
	549	+ ret
619	550
620	551
621	552	WELS_EXTERN VAACalcSad_sse2
622	553	;*************************************************************************************************************
623	554	;void VAACalcSad_sse2( uint8_t cur_data, uint8_t ref_data, int32_t iPicWidth, int32_t iPicHeight
624		−; int32_t iPicStride, int32_t psadframe, int32_t psad8x8)
	555	+; int32_t iPicStride, int32_t psadframe, int32_t psad8x8)
625	556	;*************************************************************************************************************
626	557
627	558
628	559	ALIGN 16
629	560	VAACalcSad_sse2:
630		−%define cur_data esp + pushsize + 4
631		−%define ref_data esp + pushsize + 8
632		−%define iPicWidth esp + pushsize + 12
633		−%define iPicHeight esp + pushsize + 16
634		−%define iPicStride esp + pushsize + 20
635		−%define psadframe esp + pushsize + 24
636		−%define psad8x8 esp + pushsize + 28
637		−%define pushsize 12
638		− push esi
639		− push edi
640		− push ebx
641		− mov esi, [cur_data]
642		− mov edi, [ref_data]
643		− mov ebx, [iPicStride]
644		− mov edx, [psad8x8]
645		− mov eax, ebx
646		−
647		− shr dword [iPicWidth], 4 ; iPicWidth/16
648		− shr dword [iPicHeight], 4 ; iPicHeight/16
649		− shl eax, 4 ; iPicStride*16
650		− pxor xmm0, xmm0
651		− pxor xmm7, xmm7 ; iFrameSad
	561	+%define cur_data r0
	562	+%define ref_data r1
	563	+%define iPicWidth r2
	564	+%define iPicHeight r3
	565	+%define iPicStride r4
	566	+%define psadframe r5
	567	+%define psad8x8 r6
	568	+
	569	+ push r12
	570	+ push r13
	571	+ %assign push_num 2
	572	+ LOAD_7_PARA
	573	+ SIGN_EXTENTION r2,r2d
	574	+ SIGN_EXTENTION r3,r3d
	575	+ SIGN_EXTENTION r4,r4d
	576	+
	577	+ mov r12,r4
	578	+ shr r2, 4 ; iPicWidth/16
	579	+ shr r3, 4 ; iPicHeight/16
	580	+
	581	+ shl r12, 4 ; iPicStride*16
	582	+ pxor xmm0, xmm0
	583	+ pxor xmm7, xmm7 ; iFrameSad
652	584	height_loop:
653		− mov ecx, dword [iPicWidth]
654		− push esi
655		− push edi
	585	+ mov r13, r2
	586	+ push r0
	587	+ push r1
656	588	width_loop:
657		− pxor xmm6, xmm6 ;
658		− WELS_SAD_16x2_SSE2
659		− WELS_SAD_16x2_SSE2
660		− WELS_SAD_16x2_SSE2
661		− WELS_SAD_16x2_SSE2
662		− paddd xmm7, xmm6
663		− movd [edx], xmm6
664		− psrldq xmm6, 8
665		− movd [edx+4], xmm6
666		−
667		− pxor xmm6, xmm6
668		− WELS_SAD_16x2_SSE2
669		− WELS_SAD_16x2_SSE2
670		− WELS_SAD_16x2_SSE2
671		− WELS_SAD_16x2_SSE2
672		− paddd xmm7, xmm6
673		− movd [edx+8], xmm6
674		− psrldq xmm6, 8
675		− movd [edx+12], xmm6
676		−
677		− add edx, 16
678		− sub esi, eax
679		− sub edi, eax
680		− add esi, 16
681		− add edi, 16
682		−
683		− dec ecx
684		− jnz width_loop
685		−
686		− pop edi
687		− pop esi
688		− add esi, eax
689		− add edi, eax
690		−
691		− dec dword [iPicHeight]
692		− jnz height_loop
693		−
694		− mov edx, [psadframe]
695		− movdqa xmm5, xmm7
696		− psrldq xmm7, 8
697		− paddd xmm7, xmm5
698		− movd [edx], xmm7
699		−
700		−%undef cur_data
701		−%undef ref_data
702		−%undef iPicWidth
703		−%undef iPicHeight
704		−%undef iPicStride
705		−%undef psadframe
706		−%undef psad8x8
707		−%undef pushsize
708		− pop ebx
709		− pop edi
710		− pop esi
711		− ret
	589	+ pxor xmm6, xmm6
	590	+ WELS_SAD_16x2_SSE2 r0,r1,r4
	591	+ WELS_SAD_16x2_SSE2 r0,r1,r4
	592	+ WELS_SAD_16x2_SSE2 r0,r1,r4
	593	+ WELS_SAD_16x2_SSE2 r0,r1,r4
	594	+ paddd xmm7, xmm6
	595	+ movd [r6], xmm6
	596	+ psrldq xmm6, 8
	597	+ movd [r6+4], xmm6
	598	+
	599	+ pxor xmm6, xmm6
	600	+ WELS_SAD_16x2_SSE2 r0,r1,r4
	601	+ WELS_SAD_16x2_SSE2 r0,r1,r4
	602	+ WELS_SAD_16x2_SSE2 r0,r1,r4
	603	+ WELS_SAD_16x2_SSE2 r0,r1,r4
	604	+ paddd xmm7, xmm6
	605	+ movd [r6+8], xmm6
	606	+ psrldq xmm6, 8
	607	+ movd [r6+12], xmm6
	608	+
	609	+ add r6, 16
	610	+ sub r0, r12
	611	+ sub r1, r12
	612	+ add r0, 16
	613	+ add r1, 16
	614	+
	615	+ dec r13
	616	+ jnz width_loop
	617	+
	618	+ pop r1
	619	+ pop r0
	620	+ add r0, r12
	621	+ add r1, r12
	622	+
	623	+ dec r3
	624	+ jnz height_loop
	625	+
	626	+ ;mov r13, [psadframe]
	627	+ movdqa xmm5, xmm7
	628	+ psrldq xmm7, 8
	629	+ paddd xmm7, xmm5
	630	+ movd [psadframe], xmm7
	631	+
	632	+%undef cur_data
	633	+%undef ref_data
	634	+%undef iPicWidth
	635	+%undef iPicHeight
	636	+%undef iPicStride
	637	+%undef psadframe
	638	+%undef psad8x8
	639	+%undef pushsize
	640	+ LOAD_7_PARA_POP
	641	+ pop r13
	642	+ pop r12
	643	+ %assign push_num 0
	644	+ ret
	645	+
	646	+%endif
712	647
713	648
	649	+%ifdef X86_32
714	650	WELS_EXTERN VAACalcSadVar_sse2
715	651	;*************************************************************************************************************
716	652	;void VAACalcSadVar_sse2( uint8_t cur_data, uint8_t ref_data, int32_t iPicWidth, int32_t iPicHeight
717		−; int32_t iPicStride, int32_t psadframe, int32_t psad8x8, int32_t psum16x16, int32_t psqsum16x16)
	653	+; int32_t iPicStride, int32_t psadframe, int32_t psad8x8, int32_t psum16x16, int32_t psqsum16x16)
718	654	;*************************************************************************************************************
719	655
720	656
721	657	ALIGN 16
722	658	VAACalcSadVar_sse2:
723		−%define localsize 8
724		−%define cur_data esp + pushsize + localsize + 4
725		−%define ref_data esp + pushsize + localsize + 8
726		−%define iPicWidth esp + pushsize + localsize + 12
727		−%define iPicHeight esp + pushsize + localsize + 16
728		−%define iPicStride esp + pushsize + localsize + 20
729		−%define psadframe esp + pushsize + localsize + 24
730		−%define psad8x8 esp + pushsize + localsize + 28
731		−%define psum16x16 esp + pushsize + localsize + 32
732		−%define psqsum16x16 esp + pushsize + localsize + 36
733		−%define tmp_esi esp + 0
734		−%define tmp_edi esp + 4
735		−%define pushsize 16
736		− push ebp
737		− push esi
738		− push edi
739		− push ebx
740		− sub esp, localsize
741		− mov esi, [cur_data]
742		− mov edi, [ref_data]
743		− mov ebx, [iPicStride]
744		− mov edx, [psad8x8]
745		− mov eax, ebx
746		−
747		− shr dword [iPicWidth], 4 ; iPicWidth/16
748		− shr dword [iPicHeight], 4 ; iPicHeight/16
749		− shl eax, 4 ; iPicStride*16
750		− pxor xmm0, xmm0
751		− pxor xmm7, xmm7 ; iFrameSad
	659	+%define localsize 8
	660	+%define cur_data esp + pushsize + localsize + 4
	661	+%define ref_data esp + pushsize + localsize + 8
	662	+%define iPicWidth esp + pushsize + localsize + 12
	663	+%define iPicHeight esp + pushsize + localsize + 16
	664	+%define iPicStride esp + pushsize + localsize + 20
	665	+%define psadframe esp + pushsize + localsize + 24
	666	+%define psad8x8 esp + pushsize + localsize + 28
	667	+%define psum16x16 esp + pushsize + localsize + 32
	668	+%define psqsum16x16 esp + pushsize + localsize + 36
	669	+%define tmp_esi esp + 0
	670	+%define tmp_edi esp + 4
	671	+%define pushsize 16
	672	+ push ebp
	673	+ push esi
	674	+ push edi
	675	+ push ebx
	676	+ sub esp, localsize
	677	+ mov esi, [cur_data]
	678	+ mov edi, [ref_data]
	679	+ mov ebx, [iPicStride]
	680	+ mov edx, [psad8x8]
	681	+ mov eax, ebx
	682	+
	683	+ shr dword [iPicWidth], 4 ; iPicWidth/16
	684	+ shr dword [iPicHeight], 4 ; iPicHeight/16
	685	+ shl eax, 4 ; iPicStride*16
	686	+ pxor xmm0, xmm0
	687	+ pxor xmm7, xmm7 ; iFrameSad
752	688	var_height_loop:
753		− mov ecx, dword [iPicWidth]
754		− mov [tmp_esi], esi
755		− mov [tmp_edi], edi
	689	+ mov ecx, dword [iPicWidth]
	690	+ mov [tmp_esi], esi
	691	+ mov [tmp_edi], edi
756	692	var_width_loop:
757		− pxor xmm6, xmm6 ; hiQuad_loQuad pSad8x8
758		− pxor xmm5, xmm5 ; pSum16x16
759		− pxor xmm4, xmm4 ; sqsum_16x16
760		− WELS_SAD_SUM_SQSUM_16x1_SSE2
761		− WELS_SAD_SUM_SQSUM_16x1_SSE2
762		− WELS_SAD_SUM_SQSUM_16x1_SSE2
763		− WELS_SAD_SUM_SQSUM_16x1_SSE2
764		− WELS_SAD_SUM_SQSUM_16x1_SSE2
765		− WELS_SAD_SUM_SQSUM_16x1_SSE2
766		− WELS_SAD_SUM_SQSUM_16x1_SSE2
767		− WELS_SAD_SUM_SQSUM_16x1_SSE2
768		− paddd xmm7, xmm6
769		− movd [edx], xmm6
770		− psrldq xmm6, 8
771		− movd [edx+4], xmm6
772		−
773		− pxor xmm6, xmm6
774		− WELS_SAD_SUM_SQSUM_16x1_SSE2
775		− WELS_SAD_SUM_SQSUM_16x1_SSE2
776		− WELS_SAD_SUM_SQSUM_16x1_SSE2
777		− WELS_SAD_SUM_SQSUM_16x1_SSE2
778		− WELS_SAD_SUM_SQSUM_16x1_SSE2
779		− WELS_SAD_SUM_SQSUM_16x1_SSE2
780		− WELS_SAD_SUM_SQSUM_16x1_SSE2
781		− WELS_SAD_SUM_SQSUM_16x1_SSE2
782		− paddd xmm7, xmm6
783		− movd [edx+8], xmm6
784		− psrldq xmm6, 8
785		− movd [edx+12], xmm6
786		−
787		− mov ebp, [psum16x16]
788		− movdqa xmm1, xmm5
789		− psrldq xmm1, 8
790		− paddd xmm5, xmm1
791		− movd [ebp], xmm5
792		− add dword [psum16x16], 4
793		−
794		− movdqa xmm5, xmm4
795		− psrldq xmm5, 8
796		− paddd xmm4, xmm5
797		− movdqa xmm3, xmm4
798		− psrldq xmm3, 4
799		− paddd xmm4, xmm3
800		−
801		− mov ebp, [psqsum16x16]
802		− movd [ebp], xmm4
803		− add dword [psqsum16x16], 4
804		−
805		− add edx, 16
806		− sub esi, eax
807		− sub edi, eax
808		− add esi, 16
809		− add edi, 16
810		−
811		− dec ecx
812		− jnz var_width_loop
813		−
814		− mov esi, [tmp_esi]
815		− mov edi, [tmp_edi]
816		− add esi, eax
817		− add edi, eax
818		−
819		− dec dword [iPicHeight]
820		− jnz var_height_loop
821		−
822		− mov edx, [psadframe]
823		− movdqa xmm5, xmm7
824		− psrldq xmm7, 8
825		− paddd xmm7, xmm5
826		− movd [edx], xmm7
827		−
828		− add esp, localsize
829		− pop ebx
830		− pop edi
831		− pop esi
832		− pop ebp
833		−%undef cur_data
834		−%undef ref_data
835		−%undef iPicWidth
836		−%undef iPicHeight
837		−%undef iPicStride
838		−%undef psadframe
839		−%undef psad8x8
840		−%undef psum16x16
841		−%undef psqsum16x16
842		−%undef tmp_esi
843		−%undef tmp_edi
844		−%undef pushsize
845		−%undef localsize
846		− ret
	693	+ pxor xmm6, xmm6 ; hiQuad_loQuad pSad8x8
	694	+ pxor xmm5, xmm5 ; pSum16x16
	695	+ pxor xmm4, xmm4 ; sqsum_16x16
	696	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 esi,edi,ebx
	697	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 esi,edi,ebx
	698	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 esi,edi,ebx
	699	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 esi,edi,ebx
	700	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 esi,edi,ebx
	701	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 esi,edi,ebx
	702	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 esi,edi,ebx
	703	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 esi,edi,ebx
	704	+ paddd xmm7, xmm6
	705	+ movd [edx], xmm6
	706	+ psrldq xmm6, 8
	707	+ movd [edx+4], xmm6
	708	+
	709	+ pxor xmm6, xmm6
	710	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 esi,edi,ebx
	711	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 esi,edi,ebx
	712	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 esi,edi,ebx
	713	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 esi,edi,ebx
	714	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 esi,edi,ebx
	715	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 esi,edi,ebx
	716	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 esi,edi,ebx
	717	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 esi,edi,ebx
	718	+ paddd xmm7, xmm6
	719	+ movd [edx+8], xmm6
	720	+ psrldq xmm6, 8
	721	+ movd [edx+12], xmm6
	722	+
	723	+ mov ebp, [psum16x16]
	724	+ movdqa xmm1, xmm5
	725	+ psrldq xmm1, 8
	726	+ paddd xmm5, xmm1
	727	+ movd [ebp], xmm5
	728	+ add dword [psum16x16], 4
	729	+
	730	+ movdqa xmm5, xmm4
	731	+ psrldq xmm5, 8
	732	+ paddd xmm4, xmm5
	733	+ movdqa xmm3, xmm4
	734	+ psrldq xmm3, 4
	735	+ paddd xmm4, xmm3
	736	+
	737	+ mov ebp, [psqsum16x16]
	738	+ movd [ebp], xmm4
	739	+ add dword [psqsum16x16], 4
	740	+
	741	+ add edx, 16
	742	+ sub esi, eax
	743	+ sub edi, eax
	744	+ add esi, 16
	745	+ add edi, 16
	746	+
	747	+ dec ecx
	748	+ jnz var_width_loop
	749	+
	750	+ mov esi, [tmp_esi]
	751	+ mov edi, [tmp_edi]
	752	+ add esi, eax
	753	+ add edi, eax
	754	+
	755	+ dec dword [iPicHeight]
	756	+ jnz var_height_loop
	757	+
	758	+ mov edx, [psadframe]
	759	+ movdqa xmm5, xmm7
	760	+ psrldq xmm7, 8
	761	+ paddd xmm7, xmm5
	762	+ movd [edx], xmm7
	763	+
	764	+ add esp, localsize
	765	+ pop ebx
	766	+ pop edi
	767	+ pop esi
	768	+ pop ebp
	769	+%undef cur_data
	770	+%undef ref_data
	771	+%undef iPicWidth
	772	+%undef iPicHeight
	773	+%undef iPicStride
	774	+%undef psadframe
	775	+%undef psad8x8
	776	+%undef psum16x16
	777	+%undef psqsum16x16
	778	+%undef tmp_esi
	779	+%undef tmp_edi
	780	+%undef pushsize
	781	+%undef localsize
	782	+ ret
	783	+
	784	+%else ;64-bit
847	785
	786	+WELS_EXTERN VAACalcSadVar_sse2
	787	+;*************************************************************************************************************
	788	+;void VAACalcSadVar_sse2( uint8_t cur_data, uint8_t ref_data, int32_t iPicWidth, int32_t iPicHeight
	789	+; int32_t iPicStride, int32_t psadframe, int32_t psad8x8, int32_t psum16x16, int32_t psqsum16x16)
	790	+;*************************************************************************************************************
	791	+
	792	+
	793	+ALIGN 16
	794	+VAACalcSadVar_sse2:
	795	+%define cur_data arg1 ;r0
	796	+%define ref_data arg2 ;r1
	797	+%define iPicWidth arg3 ;r2
	798	+%define iPicHeight arg4 ;r3
	799	+%define iPicStride arg5
	800	+%define psadframe arg6
	801	+%define psad8x8 arg7
	802	+%define psum16x16 arg8
	803	+%define psqsum16x16 arg9
	804	+
	805	+ push r12
	806	+ push r13
	807	+ push r14
	808	+ push r15
	809	+ %assign push_num 4
	810	+
	811	+%ifdef WIN64
	812	+ mov r4, arg5 ;iPicStride
	813	+ mov r5, arg6 ;psad8x8
	814	+%endif
	815	+ mov r14,arg7
	816	+ SIGN_EXTENTION r2,r2d
	817	+ SIGN_EXTENTION r3,r3d
	818	+ SIGN_EXTENTION r4,r4d
	819	+
	820	+ mov r13,r4
	821	+ shr r2,4
	822	+ shr r3,4
	823	+
	824	+ shl r13,4 ; iPicStride*16
	825	+ pxor xmm0, xmm0
	826	+ pxor xmm7, xmm7 ; iFrameSad
	827	+var_height_loop:
	828	+ push r2
	829	+ %assign push_num push_num+1
	830	+ mov r11, r0
	831	+ mov r12, r1
	832	+var_width_loop:
	833	+ pxor xmm6, xmm6 ; hiQuad_loQuad pSad8x8
	834	+ pxor xmm5, xmm5 ; pSum16x16
	835	+ pxor xmm4, xmm4 ; sqsum_16x16
	836	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 r0,r1,r4
	837	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 r0,r1,r4
	838	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 r0,r1,r4
	839	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 r0,r1,r4
	840	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 r0,r1,r4
	841	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 r0,r1,r4
	842	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 r0,r1,r4
	843	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 r0,r1,r4
	844	+ paddd xmm7, xmm6
	845	+ movd [r14], xmm6
	846	+ psrldq xmm6, 8
	847	+ movd [r14+4], xmm6
	848	+
	849	+ pxor xmm6, xmm6
	850	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 r0,r1,r4
	851	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 r0,r1,r4
	852	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 r0,r1,r4
	853	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 r0,r1,r4
	854	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 r0,r1,r4
	855	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 r0,r1,r4
	856	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 r0,r1,r4
	857	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 r0,r1,r4
	858	+ paddd xmm7, xmm6
	859	+ movd [r14+8], xmm6
	860	+ psrldq xmm6, 8
	861	+ movd [r14+12], xmm6
	862	+
	863	+ mov r15, psum16x16
	864	+ movdqa xmm1, xmm5
	865	+ psrldq xmm1, 8
	866	+ paddd xmm5, xmm1
	867	+ movd [r15], xmm5
	868	+ add dword psum16x16, 4
	869	+
	870	+ movdqa xmm5, xmm4
	871	+ psrldq xmm5, 8
	872	+ paddd xmm4, xmm5
	873	+ movdqa xmm3, xmm4
	874	+ psrldq xmm3, 4
	875	+ paddd xmm4, xmm3
	876	+
	877	+ mov r15, psqsum16x16
	878	+ movd [r15], xmm4
	879	+ add dword psqsum16x16, 4
	880	+
	881	+ add r14,16
	882	+ sub r0, r13
	883	+ sub r1, r13
	884	+ add r0, 16
	885	+ add r1, 16
	886	+
	887	+ dec r2
	888	+ jnz var_width_loop
	889	+
	890	+ pop r2
	891	+ %assign push_num push_num-1
	892	+ mov r0, r11
	893	+ mov r1, r12
	894	+ add r0, r13
	895	+ add r1, r13
	896	+ dec r3
	897	+ jnz var_height_loop
	898	+
	899	+ mov r15, psadframe
	900	+ movdqa xmm5, xmm7
	901	+ psrldq xmm7, 8
	902	+ paddd xmm7, xmm5
	903	+ movd [r15], xmm7
	904	+
	905	+ pop r15
	906	+ pop r14
	907	+ pop r13
	908	+ pop r12
	909	+%assign push_num 0
	910	+%undef cur_data
	911	+%undef ref_data
	912	+%undef iPicWidth
	913	+%undef iPicHeight
	914	+%undef iPicStride
	915	+%undef psadframe
	916	+%undef psad8x8
	917	+%undef psum16x16
	918	+%undef psqsum16x16
	919	+%undef tmp_esi
	920	+%undef tmp_edi
	921	+%undef pushsize
	922	+%undef localsize
	923	+ ret
	924	+
	925	+%endif
	926	+
	927	+%ifdef X86_32
	928	+
	929	+WELS_EXTERN VAACalcSadSsd_sse2
	930	+;*************************************************************************************************************
	931	+;void VAACalcSadSsd_sse2(uint8_t cur_data, uint8_t ref_data, int32_t iPicWidth, int32_t iPicHeight,
	932	+; int32_t iPicStride,int32_t psadframe, int32_t psad8x8, int32_t psum16x16, int32_t psqsum16x16, int32_t *psqdiff16x16)
	933	+;*************************************************************************************************************
	934	+
	935	+
	936	+ALIGN 16
	937	+VAACalcSadSsd_sse2:
	938	+%define localsize 12
	939	+%define cur_data esp + pushsize + localsize + 4
	940	+%define ref_data esp + pushsize + localsize + 8
	941	+%define iPicWidth esp + pushsize + localsize + 12
	942	+%define iPicHeight esp + pushsize + localsize + 16
	943	+%define iPicStride esp + pushsize + localsize + 20
	944	+%define psadframe esp + pushsize + localsize + 24
	945	+%define psad8x8 esp + pushsize + localsize + 28
	946	+%define psum16x16 esp + pushsize + localsize + 32
	947	+%define psqsum16x16 esp + pushsize + localsize + 36
	948	+%define psqdiff16x16 esp + pushsize + localsize + 40
	949	+%define tmp_esi esp + 0
	950	+%define tmp_edi esp + 4
	951	+%define tmp_sadframe esp + 8
	952	+%define pushsize 16
	953	+ push ebp
	954	+ push esi
	955	+ push edi
	956	+ push ebx
	957	+ sub esp, localsize
	958	+
	959	+ mov ecx, [iPicWidth]
	960	+ mov ecx, [iPicHeight]
	961	+ mov esi, [cur_data]
	962	+ mov edi, [ref_data]
	963	+ mov ebx, [iPicStride]
	964	+ mov edx, [psad8x8]
	965	+ mov eax, ebx
	966	+
	967	+ shr dword [iPicWidth], 4 ; iPicWidth/16
	968	+ shr dword [iPicHeight], 4 ; iPicHeight/16
	969	+ shl eax, 4 ; iPicStride*16
	970	+ mov ecx, [iPicWidth]
	971	+ mov ecx, [iPicHeight]
	972	+ pxor xmm0, xmm0
	973	+ movd [tmp_sadframe], xmm0
	974	+sqdiff_height_loop:
	975	+ mov ecx, dword [iPicWidth]
	976	+ mov [tmp_esi], esi
	977	+ mov [tmp_edi], edi
	978	+sqdiff_width_loop:
	979	+ pxor xmm7, xmm7 ; hiQuad_loQuad pSad8x8
	980	+ pxor xmm6, xmm6 ; pSum16x16
	981	+ pxor xmm5, xmm5 ; sqsum_16x16 four dword
	982	+ pxor xmm4, xmm4 ; sqdiff_16x16 four Dword
	983	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 esi,edi,ebx
	984	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 esi,edi,ebx
	985	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 esi,edi,ebx
	986	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 esi,edi,ebx
	987	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 esi,edi,ebx
	988	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 esi,edi,ebx
	989	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 esi,edi,ebx
	990	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 esi,edi,ebx
	991	+ movdqa xmm1, xmm7
	992	+ movd [edx], xmm7
	993	+ psrldq xmm7, 8
	994	+ paddd xmm1, xmm7
	995	+ movd [edx+4], xmm7
	996	+ movd ebp, xmm1
	997	+ add [tmp_sadframe], ebp
	998	+
	999	+ pxor xmm7, xmm7
	1000	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 esi,edi,ebx
	1001	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 esi,edi,ebx
	1002	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 esi,edi,ebx
	1003	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 esi,edi,ebx
	1004	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 esi,edi,ebx
	1005	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 esi,edi,ebx
	1006	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 esi,edi,ebx
	1007	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 esi,edi,ebx
	1008	+ movdqa xmm1, xmm7
	1009	+ movd [edx+8], xmm7
	1010	+ psrldq xmm7, 8
	1011	+ paddd xmm1, xmm7
	1012	+ movd [edx+12], xmm7
	1013	+ movd ebp, xmm1
	1014	+ add [tmp_sadframe], ebp
	1015	+
	1016	+ mov ebp, [psum16x16]
	1017	+ movdqa xmm1, xmm6
	1018	+ psrldq xmm1, 8
	1019	+ paddd xmm6, xmm1
	1020	+ movd [ebp], xmm6
	1021	+ add dword [psum16x16], 4
	1022	+
	1023	+ mov ebp, [psqsum16x16]
	1024	+ pshufd xmm6, xmm5, 14 ;00001110
	1025	+ paddd xmm6, xmm5
	1026	+ pshufd xmm5, xmm6, 1 ;00000001
	1027	+ paddd xmm5, xmm6
	1028	+ movd [ebp], xmm5
	1029	+ add dword [psqsum16x16], 4
	1030	+
	1031	+ mov ebp, [psqdiff16x16]
	1032	+ pshufd xmm5, xmm4, 14 ; 00001110
	1033	+ paddd xmm5, xmm4
	1034	+ pshufd xmm4, xmm5, 1 ; 00000001
	1035	+ paddd xmm4, xmm5
	1036	+ movd [ebp], xmm4
	1037	+ add dword [psqdiff16x16], 4
	1038	+
	1039	+ add edx, 16
	1040	+ sub esi, eax
	1041	+ sub edi, eax
	1042	+ add esi, 16
	1043	+ add edi, 16
	1044	+
	1045	+ dec ecx
	1046	+ jnz sqdiff_width_loop
	1047	+
	1048	+ mov esi, [tmp_esi]
	1049	+ mov edi, [tmp_edi]
	1050	+ add esi, eax
	1051	+ add edi, eax
	1052	+
	1053	+ dec dword [iPicHeight]
	1054	+ jnz sqdiff_height_loop
	1055	+
	1056	+ mov ebx, [tmp_sadframe]
	1057	+ mov eax, [psadframe]
	1058	+ mov [eax], ebx
	1059	+
	1060	+ add esp, localsize
	1061	+ pop ebx
	1062	+ pop edi
	1063	+ pop esi
	1064	+ pop ebp
	1065	+%undef cur_data
	1066	+%undef ref_data
	1067	+%undef iPicWidth
	1068	+%undef iPicHeight
	1069	+%undef iPicStride
	1070	+%undef psadframe
	1071	+%undef psad8x8
	1072	+%undef psum16x16
	1073	+%undef psqsum16x16
	1074	+%undef psqdiff16x16
	1075	+%undef tmp_esi
	1076	+%undef tmp_edi
	1077	+%undef tmp_sadframe
	1078	+%undef pushsize
	1079	+%undef localsize
	1080	+ ret
	1081	+
	1082	+%else
848	1083
849	1084
850	1085	WELS_EXTERN VAACalcSadSsd_sse2
851	1086	;*************************************************************************************************************
852	1087	;void VAACalcSadSsd_sse2(uint8_t cur_data, uint8_t ref_data, int32_t iPicWidth, int32_t iPicHeight,
853		−; int32_t iPicStride,int32_t psadframe, int32_t psad8x8, int32_t psum16x16, int32_t psqsum16x16, int32_t *psqdiff16x16)
	1088	+; int32_t iPicStride,int32_t psadframe, int32_t psad8x8, int32_t psum16x16, int32_t psqsum16x16, int32_t *psqdiff16x16)
854	1089	;*************************************************************************************************************
855	1090
856	1091
857	1092	ALIGN 16
858	1093	VAACalcSadSsd_sse2:
859		−%define localsize 12
860		−%define cur_data esp + pushsize + localsize + 4
861		−%define ref_data esp + pushsize + localsize + 8
862		−%define iPicWidth esp + pushsize + localsize + 12
863		−%define iPicHeight esp + pushsize + localsize + 16
864		−%define iPicStride esp + pushsize + localsize + 20
865		−%define psadframe esp + pushsize + localsize + 24
866		−%define psad8x8 esp + pushsize + localsize + 28
867		−%define psum16x16 esp + pushsize + localsize + 32
868		−%define psqsum16x16 esp + pushsize + localsize + 36
869		−%define psqdiff16x16 esp + pushsize + localsize + 40
870		−%define tmp_esi esp + 0
871		−%define tmp_edi esp + 4
872		−%define tmp_sadframe esp + 8
873		−%define pushsize 16
874		− push ebp
875		− push esi
876		− push edi
877		− push ebx
878		− sub esp, localsize
879		− mov ecx, [iPicWidth]
880		− mov ecx, [iPicHeight]
881		− mov esi, [cur_data]
882		− mov edi, [ref_data]
883		− mov ebx, [iPicStride]
884		− mov edx, [psad8x8]
885		− mov eax, ebx
886		−
887		− shr dword [iPicWidth], 4 ; iPicWidth/16
888		− shr dword [iPicHeight], 4 ; iPicHeight/16
889		− shl eax, 4 ; iPicStride*16
890		− mov ecx, [iPicWidth]
891		− mov ecx, [iPicHeight]
892		− pxor xmm0, xmm0
893		− movd [tmp_sadframe], xmm0
	1094	+%define localsize 12
	1095	+%define cur_data arg1;r0
	1096	+%define ref_data arg2;r1
	1097	+%define iPicWidth arg3;r2
	1098	+%define iPicHeight arg4;r3
	1099	+%define iPicStride arg5;
	1100	+%define psadframe arg6;
	1101	+%define psad8x8 arg7;
	1102	+%define psum16x16 arg8;
	1103	+%define psqsum16x16 arg9;
	1104	+%define psqdiff16x16 arg10
	1105	+
	1106	+ push r12
	1107	+ push r13
	1108	+ push r14
	1109	+ push r15
	1110	+ %assign push_num 4
	1111	+
	1112	+%ifdef WIN64
	1113	+ mov r4,arg5
	1114	+%endif
	1115	+ mov r14,arg7
	1116	+ SIGN_EXTENTION r2,r2d
	1117	+ SIGN_EXTENTION r3,r3d
	1118	+ SIGN_EXTENTION r4,r4d
	1119	+
	1120	+ mov r13,r4
	1121	+ shr r2,4 ; iPicWidth/16
	1122	+ shr r3,4 ; iPicHeight/16
	1123	+ shl r13,4 ; iPicStride*16
	1124	+ pxor xmm0, xmm0
	1125	+ pxor xmm8, xmm8 ;framesad
	1126	+ pxor xmm9, xmm9
894	1127	sqdiff_height_loop:
895		− mov ecx, dword [iPicWidth]
896		− mov [tmp_esi], esi
897		− mov [tmp_edi], edi
	1128	+ ;mov ecx, dword [iPicWidth]
	1129	+ ;mov r14,r2
	1130	+ push r2
	1131	+ %assign push_num push_num +1
	1132	+ mov r10, r0
	1133	+ mov r11, r1
898	1134	sqdiff_width_loop:
899		− pxor xmm7, xmm7 ; hiQuad_loQuad pSad8x8
900		− pxor xmm6, xmm6 ; pSum16x16
901		− pxor xmm5, xmm5 ; sqsum_16x16 four dword
902		− pxor xmm4, xmm4 ; sqdiff_16x16 four Dword
903		− WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2
904		− WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2
905		− WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2
906		− WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2
907		− WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2
908		− WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2
909		− WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2
910		− WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2
911		− movdqa xmm1, xmm7
912		− movd [edx], xmm7
913		− psrldq xmm7, 8
914		− paddd xmm1, xmm7
915		− movd [edx+4], xmm7
916		− movd ebp, xmm1
917		− add [tmp_sadframe], ebp
918		−
919		− pxor xmm7, xmm7
920		− WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2
921		− WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2
922		− WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2
923		− WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2
924		− WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2
925		− WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2
926		− WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2
927		− WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2
928		− movdqa xmm1, xmm7
929		− movd [edx+8], xmm7
930		− psrldq xmm7, 8
931		− paddd xmm1, xmm7
932		− movd [edx+12], xmm7
933		− movd ebp, xmm1
934		− add [tmp_sadframe], ebp
935		−
936		− mov ebp, [psum16x16]
937		− movdqa xmm1, xmm6
938		− psrldq xmm1, 8
939		− paddd xmm6, xmm1
940		− movd [ebp], xmm6
941		− add dword [psum16x16], 4
942		−
943		− mov ebp, [psqsum16x16]
944		− pshufd xmm6, xmm5, 14 ;00001110
945		− paddd xmm6, xmm5
946		− pshufd xmm5, xmm6, 1 ;00000001
947		− paddd xmm5, xmm6
948		− movd [ebp], xmm5
949		− add dword [psqsum16x16], 4
950		−
951		− mov ebp, [psqdiff16x16]
952		− pshufd xmm5, xmm4, 14 ; 00001110
953		− paddd xmm5, xmm4
954		− pshufd xmm4, xmm5, 1 ; 00000001
955		− paddd xmm4, xmm5
956		− movd [ebp], xmm4
957		− add dword [psqdiff16x16], 4
958		−
959		− add edx, 16
960		− sub esi, eax
961		− sub edi, eax
962		− add esi, 16
963		− add edi, 16
964		−
965		− dec ecx
966		− jnz sqdiff_width_loop
967		−
968		− mov esi, [tmp_esi]
969		− mov edi, [tmp_edi]
970		− add esi, eax
971		− add edi, eax
972		−
973		− dec dword [iPicHeight]
974		− jnz sqdiff_height_loop
975		−
976		− mov ebx, [tmp_sadframe]
977		− mov eax, [psadframe]
978		− mov [eax], ebx
979		−
980		− add esp, localsize
981		− pop ebx
982		− pop edi
983		− pop esi
984		− pop ebp
985		−%undef cur_data
986		−%undef ref_data
987		−%undef iPicWidth
988		−%undef iPicHeight
989		−%undef iPicStride
990		−%undef psadframe
991		−%undef psad8x8
992		−%undef psum16x16
993		−%undef psqsum16x16
994		−%undef psqdiff16x16
995		−%undef tmp_esi
996		−%undef tmp_edi
997		−%undef tmp_sadframe
998		−%undef pushsize
999		−%undef localsize
1000		− ret
	1135	+ pxor xmm7, xmm7 ; hiQuad_loQuad pSad8x8
	1136	+ pxor xmm6, xmm6 ; pSum16x16
	1137	+ pxor xmm5, xmm5 ; sqsum_16x16 four dword
	1138	+ pxor xmm4, xmm4 ; sqdiff_16x16 four Dword
	1139	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 r0,r1,r4
	1140	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 r0,r1,r4
	1141	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 r0,r1,r4
	1142	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 r0,r1,r4
	1143	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 r0,r1,r4
	1144	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 r0,r1,r4
	1145	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 r0,r1,r4
	1146	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 r0,r1,r4
	1147	+ movdqa xmm1, xmm7
	1148	+ movd [r14], xmm7
	1149	+ psrldq xmm7, 8
	1150	+ paddd xmm1, xmm7
	1151	+ movd [r14+4], xmm7
	1152	+ movd r15d, xmm1
	1153	+ movd xmm9, r15d
	1154	+ paddd xmm8,xmm9
	1155	+
	1156	+
	1157	+ pxor xmm7, xmm7
	1158	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 r0,r1,r4
	1159	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 r0,r1,r4
	1160	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 r0,r1,r4
	1161	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 r0,r1,r4
	1162	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 r0,r1,r4
	1163	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 r0,r1,r4
	1164	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 r0,r1,r4
	1165	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 r0,r1,r4
	1166	+ movdqa xmm1, xmm7
	1167	+ movd [r14+8], xmm7
	1168	+ psrldq xmm7, 8
	1169	+ paddd xmm1, xmm7
	1170	+ movd [r14+12], xmm7
	1171	+ movd r15d, xmm1
	1172	+ movd xmm9, r15d
	1173	+ paddd xmm8,xmm9
	1174	+
	1175	+ mov r15, psum16x16
	1176	+ movdqa xmm1, xmm6
	1177	+ psrldq xmm1, 8
	1178	+ paddd xmm6, xmm1
	1179	+ movd [r15], xmm6
	1180	+ add dword psum16x16, 4
	1181	+
	1182	+ mov r15, psqsum16x16
	1183	+ pshufd xmm6, xmm5, 14 ;00001110
	1184	+ paddd xmm6, xmm5
	1185	+ pshufd xmm5, xmm6, 1 ;00000001
	1186	+ paddd xmm5, xmm6
	1187	+ movd [r15], xmm5
	1188	+ add dword psqsum16x16, 4
	1189	+
	1190	+ mov r15, psqdiff16x16
	1191	+ pshufd xmm5, xmm4, 14 ; 00001110
	1192	+ paddd xmm5, xmm4
	1193	+ pshufd xmm4, xmm5, 1 ; 00000001
	1194	+ paddd xmm4, xmm5
	1195	+ movd [r15], xmm4
	1196	+ add dword psqdiff16x16, 4
	1197	+
	1198	+ add r14,16
	1199	+ sub r0, r13
	1200	+ sub r1, r13
	1201	+ add r0, 16
	1202	+ add r1, 16
	1203	+
	1204	+ dec r2
	1205	+ jnz sqdiff_width_loop
	1206	+
	1207	+ pop r2
	1208	+ %assign push_num push_num -1
	1209	+
	1210	+ mov r0, r10
	1211	+ mov r1, r11
	1212	+ add r0, r13
	1213	+ add r1, r13
	1214	+
	1215	+ dec r3
	1216	+ jnz sqdiff_height_loop
	1217	+
	1218	+ mov r13, psadframe
	1219	+ movd [r13], xmm8
	1220	+
	1221	+ pop r15
	1222	+ pop r14
	1223	+ pop r13
	1224	+ pop r12
	1225	+ %assign push_num 0
	1226	+
	1227	+%undef cur_data
	1228	+%undef ref_data
	1229	+%undef iPicWidth
	1230	+%undef iPicHeight
	1231	+%undef iPicStride
	1232	+%undef psadframe
	1233	+%undef psad8x8
	1234	+%undef psum16x16
	1235	+%undef psqsum16x16
	1236	+%undef psqdiff16x16
	1237	+%undef tmp_esi
	1238	+%undef tmp_edi
	1239	+%undef tmp_sadframe
	1240	+%undef pushsize
	1241	+%undef localsize
	1242	+ ret
	1243	+
	1244	+
	1245	+
	1246	+%endif
	1247	+
	1248	+%ifdef X86_32
	1249	+WELS_EXTERN VAACalcSadBgd_sse2
	1250	+;*************************************************************************************************************
	1251	+;void VAACalcSadBgd_sse2(uint8_t cur_data, uint8_t ref_data, int32_t iPicWidth, int32_t iPicHeight,
	1252	+; int32_t iPicStride, int32_t psadframe, int32_t psad8x8, int32_t p_sd8x8, uint8_t p_mad8x8)
	1253	+;*************************************************************************************************************
	1254	+
	1255	+
	1256	+ALIGN 16
	1257	+VAACalcSadBgd_sse2:
	1258	+%define localsize 12
	1259	+%define cur_data esp + pushsize + localsize + 4
	1260	+%define ref_data esp + pushsize + localsize + 8
	1261	+%define iPicWidth esp + pushsize + localsize + 12
	1262	+%define iPicHeight esp + pushsize + localsize + 16
	1263	+%define iPicStride esp + pushsize + localsize + 20
	1264	+%define psadframe esp + pushsize + localsize + 24
	1265	+%define psad8x8 esp + pushsize + localsize + 28
	1266	+%define p_sd8x8 esp + pushsize + localsize + 32
	1267	+%define p_mad8x8 esp + pushsize + localsize + 36
	1268	+%define tmp_esi esp + 0
	1269	+%define tmp_edi esp + 4
	1270	+%define tmp_ecx esp + 8
	1271	+%define pushsize 16
	1272	+ push ebp
	1273	+ push esi
	1274	+ push edi
	1275	+ push ebx
	1276	+ sub esp, localsize
	1277	+ mov esi, [cur_data]
	1278	+ mov edi, [ref_data]
	1279	+ mov ebx, [iPicStride]
	1280	+ mov eax, ebx
	1281	+
	1282	+ shr dword [iPicWidth], 4 ; iPicWidth/16
	1283	+ shr dword [iPicHeight], 4 ; iPicHeight/16
	1284	+ shl eax, 4 ; iPicStride*16
	1285	+ xor ebp, ebp
	1286	+ pxor xmm0, xmm0
	1287	+bgd_height_loop:
	1288	+ mov ecx, dword [iPicWidth]
	1289	+ mov [tmp_esi], esi
	1290	+ mov [tmp_edi], edi
	1291	+bgd_width_loop:
	1292	+ pxor xmm7, xmm7 ; pSad8x8
	1293	+ pxor xmm6, xmm6 ; sum_cur_8x8
	1294	+ pxor xmm5, xmm5 ; sum_ref_8x8
	1295	+ pxor xmm4, xmm4 ; pMad8x8
	1296	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
	1297	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
	1298	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
	1299	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
	1300	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
	1301	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
	1302	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
	1303	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
	1304	+
	1305	+
	1306	+ mov edx, [p_mad8x8]
	1307	+ WELS_MAX_REG_SSE2 xmm4
	1308	+
	1309	+ ;movdqa xmm1, xmm4
	1310	+ ;punpcklbw xmm1, xmm0
	1311	+ ;punpcklwd xmm1, xmm0
	1312	+ ;movd [edx], xmm1
	1313	+ ;punpckhbw xmm4, xmm0
	1314	+ ;punpcklwd xmm4, xmm0
	1315	+ ;movd [edx+4], xmm4
	1316	+ ;add edx, 8
	1317	+ ;mov [p_mad8x8], edx
	1318	+ mov [tmp_ecx], ecx
	1319	+ movhlps xmm1, xmm4
	1320	+ movd ecx, xmm4
	1321	+ mov [edx], cl
	1322	+ movd ecx, xmm1
	1323	+ mov [edx+1],cl
	1324	+ add edx, 2
	1325	+ mov [p_mad8x8], edx
	1326	+
	1327	+
	1328	+ pslldq xmm7, 4
	1329	+ pslldq xmm6, 4
	1330	+ pslldq xmm5, 4
	1331	+
	1332	+
	1333	+ pxor xmm4, xmm4 ; pMad8x8
	1334	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
	1335	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
	1336	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
	1337	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
	1338	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
	1339	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
	1340	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
	1341	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
	1342	+
	1343	+ mov edx, [p_mad8x8]
	1344	+ WELS_MAX_REG_SSE2 xmm4
	1345	+
	1346	+ ;movdqa xmm1, xmm4
	1347	+ ;punpcklbw xmm1, xmm0
	1348	+ ;punpcklwd xmm1, xmm0
	1349	+ ;movd [edx], xmm1
	1350	+ ;punpckhbw xmm4, xmm0
	1351	+ ;punpcklwd xmm4, xmm0
	1352	+ ;movd [edx+4], xmm4
	1353	+ ;add edx, 8
	1354	+ ;mov [p_mad8x8], edx
	1355	+ movhlps xmm1, xmm4
	1356	+ movd ecx, xmm4
	1357	+ mov [edx], cl
	1358	+ movd ecx, xmm1
	1359	+ mov [edx+1],cl
	1360	+ add edx, 2
	1361	+ mov [p_mad8x8], edx
	1362	+
	1363	+ ; data in xmm7, xmm6, xmm5: D1 D3 D0 D2
	1364	+
	1365	+ mov edx, [psad8x8]
	1366	+ pshufd xmm1, xmm7, 10001101b ; D3 D2 D1 D0
	1367	+ movdqa [edx], xmm1
	1368	+ add edx, 16
	1369	+ mov [psad8x8], edx ; sad8x8
	1370	+
	1371	+ paddd xmm1, xmm7 ; D1+3 D3+2 D0+1 D2+0
	1372	+ pshufd xmm2, xmm1, 00000011b
	1373	+ paddd xmm1, xmm2
	1374	+ movd edx, xmm1
	1375	+ add ebp, edx ; sad frame
	1376	+
	1377	+ mov edx, [p_sd8x8]
	1378	+ psubd xmm6, xmm5
	1379	+ pshufd xmm1, xmm6, 10001101b
	1380	+ movdqa [edx], xmm1
	1381	+ add edx, 16
	1382	+ mov [p_sd8x8], edx
	1383	+
	1384	+
	1385	+ add edx, 16
	1386	+ sub esi, eax
	1387	+ sub edi, eax
	1388	+ add esi, 16
	1389	+ add edi, 16
	1390	+
	1391	+ mov ecx, [tmp_ecx]
	1392	+ dec ecx
	1393	+ jnz bgd_width_loop
	1394	+
	1395	+ mov esi, [tmp_esi]
	1396	+ mov edi, [tmp_edi]
	1397	+ add esi, eax
	1398	+ add edi, eax
	1399	+
	1400	+ dec dword [iPicHeight]
	1401	+ jnz bgd_height_loop
	1402	+
	1403	+ mov edx, [psadframe]
	1404	+ mov [edx], ebp
	1405	+
	1406	+ add esp, localsize
	1407	+ pop ebx
	1408	+ pop edi
	1409	+ pop esi
	1410	+ pop ebp
	1411	+%undef cur_data
	1412	+%undef ref_data
	1413	+%undef iPicWidth
	1414	+%undef iPicHeight
	1415	+%undef iPicStride
	1416	+%undef psadframe
	1417	+%undef psad8x8
	1418	+%undef p_sd8x8
	1419	+%undef p_mad8x8
	1420	+%undef tmp_esi
	1421	+%undef tmp_edi
	1422	+%undef pushsize
	1423	+%undef localsize
	1424	+ ret
	1425	+
1001	1426
1002	1427
	1428	+WELS_EXTERN VAACalcSadSsdBgd_sse2
	1429	+;*************************************************************************************************************
	1430	+;void VAACalcSadSsdBgd_sse2(uint8_t cur_data, uint8_t ref_data, int32_t iPicWidth, int32_t iPicHeight,
	1431	+; int32_t iPicStride, int32_t psadframe, int32_t psad8x8, int32_t psum16x16, int32_t psqsum16x16,
	1432	+; int32_t psqdiff16x16, int32_t p_sd8x8, uint8_t *p_mad8x8)
	1433	+;*************************************************************************************************************
1003	1434
1004	1435
	1436	+ALIGN 16
	1437	+VAACalcSadSsdBgd_sse2:
	1438	+%define localsize 16
	1439	+%define cur_data esp + pushsize + localsize + 4
	1440	+%define ref_data esp + pushsize + localsize + 8
	1441	+%define iPicWidth esp + pushsize + localsize + 12
	1442	+%define iPicHeight esp + pushsize + localsize + 16
	1443	+%define iPicStride esp + pushsize + localsize + 20
	1444	+%define psadframe esp + pushsize + localsize + 24
	1445	+%define psad8x8 esp + pushsize + localsize + 28
	1446	+%define psum16x16 esp + pushsize + localsize + 32
	1447	+%define psqsum16x16 esp + pushsize + localsize + 36
	1448	+%define psqdiff16x16 esp + pushsize + localsize + 40
	1449	+%define p_sd8x8 esp + pushsize + localsize + 44
	1450	+%define p_mad8x8 esp + pushsize + localsize + 48
	1451	+%define tmp_esi esp + 0
	1452	+%define tmp_edi esp + 4
	1453	+%define tmp_sadframe esp + 8
	1454	+%define tmp_ecx esp + 12
	1455	+%define pushsize 16
	1456	+ push ebp
	1457	+ push esi
	1458	+ push edi
	1459	+ push ebx
	1460	+ sub esp, localsize
	1461	+ mov esi, [cur_data]
	1462	+ mov edi, [ref_data]
	1463	+ mov ebx, [iPicStride]
	1464	+ mov eax, ebx
	1465	+
	1466	+ shr dword [iPicWidth], 4 ; iPicWidth/16
	1467	+ shr dword [iPicHeight], 4 ; iPicHeight/16
	1468	+ shl eax, 4 ; iPicStride*16
	1469	+ pxor xmm0, xmm0
	1470	+ movd [tmp_sadframe], xmm0
	1471	+sqdiff_bgd_height_loop:
	1472	+ mov ecx, dword [iPicWidth]
	1473	+ mov [tmp_esi], esi
	1474	+ mov [tmp_edi], edi
	1475	+sqdiff_bgd_width_loop:
	1476	+ pxor xmm7, xmm7 ; pSad8x8 interleaves sqsum16x16: sqsum1 sad1 sqsum0 sad0
	1477	+ pxor xmm6, xmm6 ; sum_8x8 interleaves cur and pRef in Dword, Sref1 Scur1 Sref0 Scur0
	1478	+ pxor xmm5, xmm5 ; pMad8x8
	1479	+ pxor xmm4, xmm4 ; sqdiff_16x16 four Dword
	1480	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
	1481	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
	1482	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
	1483	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
	1484	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
	1485	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
	1486	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
	1487	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
	1488	+
	1489	+ mov edx, [psad8x8]
	1490	+ movdqa xmm2, xmm7
	1491	+ pshufd xmm1, xmm2, 00001110b
	1492	+ movd [edx], xmm2
	1493	+ movd [edx+4], xmm1
	1494	+ add edx, 8
	1495	+ mov [psad8x8], edx ; sad8x8
	1496	+
	1497	+ paddd xmm1, xmm2
	1498	+ movd edx, xmm1
	1499	+ add [tmp_sadframe], edx ; iFrameSad
	1500	+
	1501	+ mov edx, [psum16x16]
	1502	+ movdqa xmm1, xmm6
	1503	+ pshufd xmm2, xmm1, 00001110b
	1504	+ paddd xmm1, xmm2
	1505	+ movd [edx], xmm1 ; sum
	1506	+
	1507	+ mov edx, [p_sd8x8]
	1508	+ pshufd xmm1, xmm6, 11110101b ; Sref1 Sref1 Sref0 Sref0
	1509	+ psubd xmm6, xmm1 ; 00 diff1 00 diff0
	1510	+ pshufd xmm1, xmm6, 00001000b ; xx xx diff1 diff0
	1511	+ movq [edx], xmm1
	1512	+ add edx, 8
	1513	+ mov [p_sd8x8], edx
	1514	+
	1515	+ mov edx, [p_mad8x8]
	1516	+ WELS_MAX_REG_SSE2 xmm5
	1517	+ ;movdqa xmm1, xmm5
	1518	+ ;punpcklbw xmm1, xmm0
	1519	+ ;punpcklwd xmm1, xmm0
	1520	+ ;movd [edx], xmm1
	1521	+ ;punpckhbw xmm5, xmm0
	1522	+ ;punpcklwd xmm5, xmm0
	1523	+ ;movd [edx+4], xmm5
	1524	+ ;add edx, 8
	1525	+ ;mov [p_mad8x8], edx
	1526	+ mov [tmp_ecx], ecx
	1527	+ movhlps xmm1, xmm5
	1528	+ movd ecx, xmm5
	1529	+ mov [edx], cl
	1530	+ movd ecx, xmm1
	1531	+ mov [edx+1],cl
	1532	+ add edx, 2
	1533	+ mov [p_mad8x8], edx
	1534	+
	1535	+ psrlq xmm7, 32
	1536	+ psllq xmm7, 32 ; clear sad
	1537	+ pxor xmm6, xmm6 ; sum_8x8 interleaves cur and pRef in Dword, Sref1 Scur1 Sref0 Scur0
	1538	+ pxor xmm5, xmm5 ; pMad8x8
	1539	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
	1540	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
	1541	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
	1542	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
	1543	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
	1544	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
	1545	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
	1546	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
	1547	+
	1548	+ mov edx, [psad8x8]
	1549	+ movdqa xmm2, xmm7
	1550	+ pshufd xmm1, xmm2, 00001110b
	1551	+ movd [edx], xmm2
	1552	+ movd [edx+4], xmm1
	1553	+ add edx, 8
	1554	+ mov [psad8x8], edx ; sad8x8
	1555	+
	1556	+ paddd xmm1, xmm2
	1557	+ movd edx, xmm1
	1558	+ add [tmp_sadframe], edx ; iFrameSad
	1559	+
	1560	+ mov edx, [psum16x16]
	1561	+ movdqa xmm1, xmm6
	1562	+ pshufd xmm2, xmm1, 00001110b
	1563	+ paddd xmm1, xmm2
	1564	+ movd ebp, xmm1 ; sum
	1565	+ add [edx], ebp
	1566	+ add edx, 4
	1567	+ mov [psum16x16], edx
	1568	+
	1569	+ mov edx, [psqsum16x16]
	1570	+ psrlq xmm7, 32
	1571	+ pshufd xmm2, xmm7, 00001110b
	1572	+ paddd xmm2, xmm7
	1573	+ movd [edx], xmm2 ; sqsum
	1574	+ add edx, 4
	1575	+ mov [psqsum16x16], edx
	1576	+
	1577	+ mov edx, [p_sd8x8]
	1578	+ pshufd xmm1, xmm6, 11110101b ; Sref1 Sref1 Sref0 Sref0
	1579	+ psubd xmm6, xmm1 ; 00 diff1 00 diff0
	1580	+ pshufd xmm1, xmm6, 00001000b ; xx xx diff1 diff0
	1581	+ movq [edx], xmm1
	1582	+ add edx, 8
	1583	+ mov [p_sd8x8], edx
	1584	+
	1585	+ mov edx, [p_mad8x8]
	1586	+ WELS_MAX_REG_SSE2 xmm5
	1587	+ ;movdqa xmm1, xmm5
	1588	+ ;punpcklbw xmm1, xmm0
	1589	+ ;punpcklwd xmm1, xmm0
	1590	+ ;movd [edx], xmm1
	1591	+ ;punpckhbw xmm5, xmm0
	1592	+ ;punpcklwd xmm5, xmm0
	1593	+ ;movd [edx+4], xmm5
	1594	+ ;add edx, 8
	1595	+ ;mov [p_mad8x8], edx
	1596	+ movhlps xmm1, xmm5
	1597	+ movd ecx, xmm5
	1598	+ mov [edx], cl
	1599	+ movd ecx, xmm1
	1600	+ mov [edx+1],cl
	1601	+ add edx, 2
	1602	+ mov [p_mad8x8], edx
	1603	+
	1604	+ mov edx, [psqdiff16x16]
	1605	+ pshufd xmm1, xmm4, 00001110b
	1606	+ paddd xmm4, xmm1
	1607	+ pshufd xmm1, xmm4, 00000001b
	1608	+ paddd xmm4, xmm1
	1609	+ movd [edx], xmm4
	1610	+ add edx, 4
	1611	+ mov [psqdiff16x16], edx
	1612	+
	1613	+ add edx, 16
	1614	+ sub esi, eax
	1615	+ sub edi, eax
	1616	+ add esi, 16
	1617	+ add edi, 16
	1618	+
	1619	+ mov ecx, [tmp_ecx]
	1620	+ dec ecx
	1621	+ jnz sqdiff_bgd_width_loop
	1622	+
	1623	+ mov esi, [tmp_esi]
	1624	+ mov edi, [tmp_edi]
	1625	+ add esi, eax
	1626	+ add edi, eax
	1627	+
	1628	+ dec dword [iPicHeight]
	1629	+ jnz sqdiff_bgd_height_loop
	1630	+
	1631	+ mov edx, [psadframe]
	1632	+ mov ebp, [tmp_sadframe]
	1633	+ mov [edx], ebp
	1634	+
	1635	+ add esp, localsize
	1636	+ pop ebx
	1637	+ pop edi
	1638	+ pop esi
	1639	+ pop ebp
	1640	+%undef cur_data
	1641	+%undef ref_data
	1642	+%undef iPicWidth
	1643	+%undef iPicHeight
	1644	+%undef iPicStride
	1645	+%undef psadframe
	1646	+%undef psad8x8
	1647	+%undef psum16x16
	1648	+%undef psqsum16x16
	1649	+%undef psqdiff16x16
	1650	+%undef p_sd8x8
	1651	+%undef p_mad8x8
	1652	+%undef tmp_esi
	1653	+%undef tmp_edi
	1654	+%undef pushsize
	1655	+%undef localsize
	1656	+ ret
	1657	+%else
1005	1658
1006	1659	WELS_EXTERN VAACalcSadBgd_sse2
1007	1660	;*************************************************************************************************************
1008	1661	;void VAACalcSadBgd_sse2(uint8_t cur_data, uint8_t ref_data, int32_t iPicWidth, int32_t iPicHeight,
1009		−; int32_t iPicStride, int32_t psadframe, int32_t psad8x8, int32_t p_sd8x8, uint8_t p_mad8x8)
	1662	+; int32_t iPicStride, int32_t psadframe, int32_t psad8x8, int32_t p_sd8x8, uint8_t p_mad8x8)
1010	1663	;*************************************************************************************************************
1011	1664
1012	1665
1013	1666	ALIGN 16
1014	1667	VAACalcSadBgd_sse2:
1015		−%define localsize 12
1016		−%define cur_data esp + pushsize + localsize + 4
1017		−%define ref_data esp + pushsize + localsize + 8
1018		−%define iPicWidth esp + pushsize + localsize + 12
1019		−%define iPicHeight esp + pushsize + localsize + 16
1020		−%define iPicStride esp + pushsize + localsize + 20
1021		−%define psadframe esp + pushsize + localsize + 24
1022		−%define psad8x8 esp + pushsize + localsize + 28
1023		−%define p_sd8x8 esp + pushsize + localsize + 32
1024		−%define p_mad8x8 esp + pushsize + localsize + 36
1025		−%define tmp_esi esp + 0
1026		−%define tmp_edi esp + 4
1027		−%define tmp_ecx esp + 8
1028		−%define pushsize 16
1029		− push ebp
1030		− push esi
1031		− push edi
1032		− push ebx
1033		− sub esp, localsize
1034		− mov esi, [cur_data]
1035		− mov edi, [ref_data]
1036		− mov ebx, [iPicStride]
1037		− mov eax, ebx
1038		−
1039		− shr dword [iPicWidth], 4 ; iPicWidth/16
1040		− shr dword [iPicHeight], 4 ; iPicHeight/16
1041		− shl eax, 4 ; iPicStride*16
1042		− xor ebp, ebp
1043		− pxor xmm0, xmm0
	1668	+%define cur_data arg1;
	1669	+%define ref_data arg2;
	1670	+%define iPicWidth arg3;
	1671	+%define iPicHeight arg4;
	1672	+%define iPicStride arg5;
	1673	+%define psadframe arg6;
	1674	+%define psad8x8 arg7;
	1675	+%define p_sd8x8 arg8;
	1676	+%define p_mad8x8 arg9;
	1677	+
	1678	+ push r12
	1679	+ push r13
	1680	+ push r14
	1681	+ push r15
	1682	+%assign push_num 4
	1683	+%ifdef WIN64
	1684	+ mov r4,arg5
	1685	+ ; mov r5,arg6
	1686	+%endif
	1687	+ mov r14,arg7
	1688	+ SIGN_EXTENTION r2,r2d
	1689	+ SIGN_EXTENTION r3,r3d
	1690	+ SIGN_EXTENTION r4,r4d
	1691	+
	1692	+
	1693	+ mov r13,r4
	1694	+ mov r15,r0
	1695	+ shr r2,4
	1696	+ shr r3,4
	1697	+ shl r13,4
	1698	+ pxor xmm0, xmm0
	1699	+ pxor xmm8, xmm8
	1700	+ pxor xmm9, xmm9
1044	1701	bgd_height_loop:
1045		− mov ecx, dword [iPicWidth]
1046		− mov [tmp_esi], esi
1047		− mov [tmp_edi], edi
	1702	+ ;mov ecx, dword [iPicWidth]
	1703	+ push r2
	1704	+ %assign push_num push_num+1
	1705	+ mov r10, r15
	1706	+ mov r11, r1
1048	1707	bgd_width_loop:
1049		− pxor xmm7, xmm7 ; pSad8x8
1050		− pxor xmm6, xmm6 ; sum_cur_8x8
1051		− pxor xmm5, xmm5 ; sum_ref_8x8
1052		− pxor xmm4, xmm4 ; pMad8x8
1053		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4
1054		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4
1055		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4
1056		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4
1057		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4
1058		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4
1059		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4
1060		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4
1061		−
1062		−
1063		− mov edx, [p_mad8x8]
1064		− WELS_MAX_REG_SSE2 xmm4
1065		−
1066		− ;movdqa xmm1, xmm4
1067		− ;punpcklbw xmm1, xmm0
1068		− ;punpcklwd xmm1, xmm0
1069		− ;movd [edx], xmm1
1070		− ;punpckhbw xmm4, xmm0
1071		− ;punpcklwd xmm4, xmm0
1072		− ;movd [edx+4], xmm4
1073		− ;add edx, 8
1074		− ;mov [p_mad8x8], edx
1075		− mov [tmp_ecx], ecx
1076		− movhlps xmm1, xmm4
1077		− movd ecx, xmm4
1078		− mov [edx], cl
1079		− movd ecx, xmm1
1080		− mov [edx+1],cl
1081		− add edx, 2
1082		− mov [p_mad8x8], edx
1083		−
1084		−
1085		− pslldq xmm7, 4
1086		− pslldq xmm6, 4
1087		− pslldq xmm5, 4
1088		−
1089		−
1090		− pxor xmm4, xmm4 ; pMad8x8
1091		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4
1092		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4
1093		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4
1094		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4
1095		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4
1096		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4
1097		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4
1098		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4
1099		−
1100		− mov edx, [p_mad8x8]
1101		− WELS_MAX_REG_SSE2 xmm4
1102		−
1103		− ;movdqa xmm1, xmm4
1104		− ;punpcklbw xmm1, xmm0
1105		− ;punpcklwd xmm1, xmm0
1106		− ;movd [edx], xmm1
1107		− ;punpckhbw xmm4, xmm0
1108		− ;punpcklwd xmm4, xmm0
1109		− ;movd [edx+4], xmm4
1110		− ;add edx, 8
1111		− ;mov [p_mad8x8], edx
1112		− movhlps xmm1, xmm4
1113		− movd ecx, xmm4
1114		− mov [edx], cl
1115		− movd ecx, xmm1
1116		− mov [edx+1],cl
1117		− add edx, 2
1118		− mov [p_mad8x8], edx
1119		−
1120		− ; data in xmm7, xmm6, xmm5: D1 D3 D0 D2
1121		−
1122		− mov edx, [psad8x8]
1123		− pshufd xmm1, xmm7, 10001101b ; D3 D2 D1 D0
1124		− movdqa [edx], xmm1
1125		− add edx, 16
1126		− mov [psad8x8], edx ; sad8x8
1127		−
1128		− paddd xmm1, xmm7 ; D1+3 D3+2 D0+1 D2+0
1129		− pshufd xmm2, xmm1, 00000011b
1130		− paddd xmm1, xmm2
1131		− movd edx, xmm1
1132		− add ebp, edx ; sad frame
1133		−
1134		− mov edx, [p_sd8x8]
1135		− psubd xmm6, xmm5
1136		− pshufd xmm1, xmm6, 10001101b
1137		− movdqa [edx], xmm1
1138		− add edx, 16
1139		− mov [p_sd8x8], edx
1140		−
1141		−
1142		− add edx, 16
1143		− sub esi, eax
1144		− sub edi, eax
1145		− add esi, 16
1146		− add edi, 16
1147		−
1148		− mov ecx, [tmp_ecx]
1149		− dec ecx
1150		− jnz bgd_width_loop
1151		−
1152		− mov esi, [tmp_esi]
1153		− mov edi, [tmp_edi]
1154		− add esi, eax
1155		− add edi, eax
1156		−
1157		− dec dword [iPicHeight]
1158		− jnz bgd_height_loop
1159		−
1160		− mov edx, [psadframe]
1161		− mov [edx], ebp
1162		−
1163		− add esp, localsize
1164		− pop ebx
1165		− pop edi
1166		− pop esi
1167		− pop ebp
1168		−%undef cur_data
1169		−%undef ref_data
1170		−%undef iPicWidth
1171		−%undef iPicHeight
1172		−%undef iPicStride
1173		−%undef psadframe
1174		−%undef psad8x8
1175		−%undef p_sd8x8
1176		−%undef p_mad8x8
1177		−%undef tmp_esi
1178		−%undef tmp_edi
1179		−%undef pushsize
1180		−%undef localsize
1181		− ret
	1708	+ pxor xmm7, xmm7 ; pSad8x8
	1709	+ pxor xmm6, xmm6 ; sum_cur_8x8
	1710	+ pxor xmm5, xmm5 ; sum_ref_8x8
	1711	+ pxor xmm4, xmm4 ; pMad8x8
	1712	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
	1713	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
	1714	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
	1715	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
	1716	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
	1717	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
	1718	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
	1719	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
	1720	+
	1721	+
	1722	+ mov r14, p_mad8x8
	1723	+ WELS_MAX_REG_SSE2 xmm4
	1724	+
	1725	+ ;mov [tmp_ecx], ecx
	1726	+ movhlps xmm1, xmm4
	1727	+ movd r0d, xmm4
	1728	+
	1729	+
	1730	+ mov [r14], r0b
	1731	+ movd r0d, xmm1
	1732	+ mov [r14+1],r0b
	1733	+ add r14, 2
	1734	+ ;mov p_mad8x8, r14
	1735	+
	1736	+
	1737	+ pslldq xmm7, 4
	1738	+ pslldq xmm6, 4
	1739	+ pslldq xmm5, 4
	1740	+
	1741	+
	1742	+ pxor xmm4, xmm4 ; pMad8x8
	1743	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
	1744	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
	1745	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
	1746	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
	1747	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
	1748	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
	1749	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
	1750	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
	1751	+
	1752	+ ;mov r14, [p_mad8x8]
	1753	+ WELS_MAX_REG_SSE2 xmm4
	1754	+
	1755	+ movhlps xmm1, xmm4
	1756	+ movd r0d, xmm4
	1757	+ mov [r14], r0b
	1758	+ movd r0d, xmm1
	1759	+ mov [r14+1],r0b
	1760	+ add r14, 2
	1761	+ mov p_mad8x8, r14
	1762	+
	1763	+ ; data in xmm7, xmm6, xmm5: D1 D3 D0 D2
	1764	+
	1765	+ mov r14, psad8x8
	1766	+ pshufd xmm1, xmm7, 10001101b ; D3 D2 D1 D0
	1767	+ movdqa [r14], xmm1
	1768	+ add r14, 16
	1769	+ mov psad8x8, r14 ; sad8x8
	1770	+
	1771	+ paddd xmm1, xmm7 ; D1+3 D3+2 D0+1 D2+0
	1772	+ pshufd xmm2, xmm1, 00000011b
	1773	+ paddd xmm1, xmm2
	1774	+ movd r14d, xmm1
	1775	+ movd xmm9, r14d
	1776	+ paddd xmm8, xmm9 ; sad frame
	1777	+
	1778	+ mov r14, p_sd8x8
	1779	+ psubd xmm6, xmm5
	1780	+ pshufd xmm1, xmm6, 10001101b
	1781	+ movdqa [r14], xmm1
	1782	+ add r14, 16
	1783	+ mov p_sd8x8, r14
	1784	+
	1785	+
	1786	+ ;add edx, 16
	1787	+ sub r15, r13
	1788	+ sub r1, r13
	1789	+ add r15, 16
	1790	+ add r1, 16
	1791	+
	1792	+
	1793	+ dec r2
	1794	+ jnz bgd_width_loop
	1795	+ pop r2
	1796	+%assign push_num push_num-1
	1797	+ mov r15, r10
	1798	+ mov r1, r11
	1799	+ add r15, r13
	1800	+ add r1, r13
	1801	+
	1802	+ dec r3
	1803	+ jnz bgd_height_loop
	1804	+
	1805	+ mov r13, psadframe
	1806	+ movd [r13], xmm8
	1807	+
	1808	+ pop r15
	1809	+ pop r14
	1810	+ pop r13
	1811	+ pop r12
	1812	+%assign push_num 0
	1813	+%undef cur_data
	1814	+%undef ref_data
	1815	+%undef iPicWidth
	1816	+%undef iPicHeight
	1817	+%undef iPicStride
	1818	+%undef psadframe
	1819	+%undef psad8x8
	1820	+%undef p_sd8x8
	1821	+%undef p_mad8x8
	1822	+%undef tmp_esi
	1823	+%undef tmp_edi
	1824	+%undef pushsize
	1825	+%undef localsize
	1826	+ ret
1182	1827
1183	1828
1184	1829
1185	1830	WELS_EXTERN VAACalcSadSsdBgd_sse2
1186	1831	;*************************************************************************************************************
1187	1832	;void VAACalcSadSsdBgd_sse2(uint8_t cur_data, uint8_t ref_data, int32_t iPicWidth, int32_t iPicHeight,
1188		−; int32_t iPicStride, int32_t psadframe, int32_t psad8x8, int32_t psum16x16, int32_t psqsum16x16,
1189		−; int32_t psqdiff16x16, int32_t p_sd8x8, uint8_t *p_mad8x8)
	1833	+; int32_t iPicStride, int32_t psadframe, int32_t psad8x8, int32_t psum16x16, int32_t psqsum16x16,
	1834	+; int32_t psqdiff16x16, int32_t p_sd8x8, uint8_t *p_mad8x8)
1190	1835	;*************************************************************************************************************
1191	1836
1192	1837
1193	1838	ALIGN 16
1194	1839	VAACalcSadSsdBgd_sse2:
1195		−%define localsize 16
1196		−%define cur_data esp + pushsize + localsize + 4
1197		−%define ref_data esp + pushsize + localsize + 8
1198		−%define iPicWidth esp + pushsize + localsize + 12
1199		−%define iPicHeight esp + pushsize + localsize + 16
1200		−%define iPicStride esp + pushsize + localsize + 20
1201		−%define psadframe esp + pushsize + localsize + 24
1202		−%define psad8x8 esp + pushsize + localsize + 28
1203		−%define psum16x16 esp + pushsize + localsize + 32
1204		−%define psqsum16x16 esp + pushsize + localsize + 36
1205		−%define psqdiff16x16 esp + pushsize + localsize + 40
1206		−%define p_sd8x8 esp + pushsize + localsize + 44
1207		−%define p_mad8x8 esp + pushsize + localsize + 48
1208		−%define tmp_esi esp + 0
1209		−%define tmp_edi esp + 4
1210		−%define tmp_sadframe esp + 8
1211		−%define tmp_ecx esp + 12
1212		−%define pushsize 16
1213		− push ebp
1214		− push esi
1215		− push edi
1216		− push ebx
1217		− sub esp, localsize
1218		− mov esi, [cur_data]
1219		− mov edi, [ref_data]
1220		− mov ebx, [iPicStride]
1221		− mov eax, ebx
1222		−
1223		− shr dword [iPicWidth], 4 ; iPicWidth/16
1224		− shr dword [iPicHeight], 4 ; iPicHeight/16
1225		− shl eax, 4 ; iPicStride*16
1226		− pxor xmm0, xmm0
1227		− movd [tmp_sadframe], xmm0
	1840	+%define cur_data arg1;
	1841	+%define ref_data arg2;
	1842	+%define iPicWidth arg3;
	1843	+%define iPicHeight arg4;
	1844	+%define iPicStride arg5;
	1845	+%define psadframe arg6;
	1846	+%define psad8x8 arg7;
	1847	+%define psum16x16 arg8;
	1848	+%define psqsum16x16 arg9;
	1849	+%define psqdiff16x16 arg10;
	1850	+%ifdef WIN64
	1851	+%define p_sd8x8 [rsp + push_num*8 + 88];
	1852	+%define p_mad8x8 [rsp + push_num*8 + 96];
	1853	+%else ;linux
	1854	+%define p_sd8x8 [rsp + push_num*8 + 40];
	1855	+%define p_mad8x8 [rsp + push_num*8 + 48];
	1856	+%endif
	1857	+
	1858	+ push r12
	1859	+ push r13
	1860	+ push r14
	1861	+ push r15
	1862	+%assign push_num 4
	1863	+%ifdef WIN64
	1864	+ mov r4,arg5
	1865	+ ;mov r5,arg6
	1866	+%endif
	1867	+ SIGN_EXTENTION r2,r2d
	1868	+ SIGN_EXTENTION r3,r3d
	1869	+ SIGN_EXTENTION r4,r4d
	1870	+
	1871	+ mov r13,r4
	1872	+ shr r2, 4 ; iPicWidth/16
	1873	+ shr r3, 4 ; iPicHeight/16
	1874	+ shl r13, 4 ; iPicStride*16
	1875	+ pxor xmm0, xmm0
	1876	+ pxor xmm8, xmm8
	1877	+ pxor xmm9, xmm9
	1878	+
	1879	+
1228	1880	sqdiff_bgd_height_loop:
1229		− mov ecx, dword [iPicWidth]
1230		− mov [tmp_esi], esi
1231		− mov [tmp_edi], edi
	1881	+ mov r10, r0
	1882	+ mov r11, r1
	1883	+ push r2
	1884	+%assign push_num push_num+1
1232	1885	sqdiff_bgd_width_loop:
1233		− pxor xmm7, xmm7 ; pSad8x8 interleaves sqsum16x16: sqsum1 sad1 sqsum0 sad0
1234		− pxor xmm6, xmm6 ; sum_8x8 interleaves cur and pRef in Dword, Sref1 Scur1 Sref0 Scur0
1235		− pxor xmm5, xmm5 ; pMad8x8
1236		− pxor xmm4, xmm4 ; sqdiff_16x16 four Dword
1237		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4
1238		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4
1239		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4
1240		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4
1241		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4
1242		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4
1243		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4
1244		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4
1245		−
1246		− mov edx, [psad8x8]
1247		− movdqa xmm2, xmm7
1248		− pshufd xmm1, xmm2, 00001110b
1249		− movd [edx], xmm2
1250		− movd [edx+4], xmm1
1251		− add edx, 8
1252		− mov [psad8x8], edx ; sad8x8
1253		−
1254		− paddd xmm1, xmm2
1255		− movd edx, xmm1
1256		− add [tmp_sadframe], edx ; iFrameSad
1257		−
1258		− mov edx, [psum16x16]
1259		− movdqa xmm1, xmm6
1260		− pshufd xmm2, xmm1, 00001110b
1261		− paddd xmm1, xmm2
1262		− movd [edx], xmm1 ; sum
1263		−
1264		− mov edx, [p_sd8x8]
1265		− pshufd xmm1, xmm6, 11110101b ; Sref1 Sref1 Sref0 Sref0
1266		− psubd xmm6, xmm1 ; 00 diff1 00 diff0
1267		− pshufd xmm1, xmm6, 00001000b ; xx xx diff1 diff0
1268		− movq [edx], xmm1
1269		− add edx, 8
1270		− mov [p_sd8x8], edx
1271		−
1272		− mov edx, [p_mad8x8]
1273		− WELS_MAX_REG_SSE2 xmm5
1274		− ;movdqa xmm1, xmm5
1275		− ;punpcklbw xmm1, xmm0
1276		− ;punpcklwd xmm1, xmm0
1277		− ;movd [edx], xmm1
1278		− ;punpckhbw xmm5, xmm0
1279		− ;punpcklwd xmm5, xmm0
1280		− ;movd [edx+4], xmm5
1281		− ;add edx, 8
1282		− ;mov [p_mad8x8], edx
1283		− mov [tmp_ecx], ecx
1284		− movhlps xmm1, xmm5
1285		− movd ecx, xmm5
1286		− mov [edx], cl
1287		− movd ecx, xmm1
1288		− mov [edx+1],cl
1289		− add edx, 2
1290		− mov [p_mad8x8], edx
1291		−
1292		− psrlq xmm7, 32
1293		− psllq xmm7, 32 ; clear sad
1294		− pxor xmm6, xmm6 ; sum_8x8 interleaves cur and pRef in Dword, Sref1 Scur1 Sref0 Scur0
1295		− pxor xmm5, xmm5 ; pMad8x8
1296		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4
1297		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4
1298		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4
1299		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4
1300		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4
1301		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4
1302		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4
1303		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4
1304		−
1305		− mov edx, [psad8x8]
1306		− movdqa xmm2, xmm7
1307		− pshufd xmm1, xmm2, 00001110b
1308		− movd [edx], xmm2
1309		− movd [edx+4], xmm1
1310		− add edx, 8
1311		− mov [psad8x8], edx ; sad8x8
1312		−
1313		− paddd xmm1, xmm2
1314		− movd edx, xmm1
1315		− add [tmp_sadframe], edx ; iFrameSad
1316		−
1317		− mov edx, [psum16x16]
1318		− movdqa xmm1, xmm6
1319		− pshufd xmm2, xmm1, 00001110b
1320		− paddd xmm1, xmm2
1321		− movd ebp, xmm1 ; sum
1322		− add [edx], ebp
1323		− add edx, 4
1324		− mov [psum16x16], edx
1325		−
1326		− mov edx, [psqsum16x16]
1327		− psrlq xmm7, 32
1328		− pshufd xmm2, xmm7, 00001110b
1329		− paddd xmm2, xmm7
1330		− movd [edx], xmm2 ; sqsum
1331		− add edx, 4
1332		− mov [psqsum16x16], edx
1333		−
1334		− mov edx, [p_sd8x8]
1335		− pshufd xmm1, xmm6, 11110101b ; Sref1 Sref1 Sref0 Sref0
1336		− psubd xmm6, xmm1 ; 00 diff1 00 diff0
1337		− pshufd xmm1, xmm6, 00001000b ; xx xx diff1 diff0
1338		− movq [edx], xmm1
1339		− add edx, 8
1340		− mov [p_sd8x8], edx
1341		−
1342		− mov edx, [p_mad8x8]
1343		− WELS_MAX_REG_SSE2 xmm5
1344		− ;movdqa xmm1, xmm5
1345		− ;punpcklbw xmm1, xmm0
1346		− ;punpcklwd xmm1, xmm0
1347		− ;movd [edx], xmm1
1348		− ;punpckhbw xmm5, xmm0
1349		− ;punpcklwd xmm5, xmm0
1350		− ;movd [edx+4], xmm5
1351		− ;add edx, 8
1352		− ;mov [p_mad8x8], edx
1353		− movhlps xmm1, xmm5
1354		− movd ecx, xmm5
1355		− mov [edx], cl
1356		− movd ecx, xmm1
1357		− mov [edx+1],cl
1358		− add edx, 2
1359		− mov [p_mad8x8], edx
1360		−
1361		− mov edx, [psqdiff16x16]
1362		− pshufd xmm1, xmm4, 00001110b
1363		− paddd xmm4, xmm1
1364		− pshufd xmm1, xmm4, 00000001b
1365		− paddd xmm4, xmm1
1366		− movd [edx], xmm4
1367		− add edx, 4
1368		− mov [psqdiff16x16], edx
1369		−
1370		− add edx, 16
1371		− sub esi, eax
1372		− sub edi, eax
1373		− add esi, 16
1374		− add edi, 16
1375		−
1376		− mov ecx, [tmp_ecx]
1377		− dec ecx
1378		− jnz sqdiff_bgd_width_loop
1379		−
1380		− mov esi, [tmp_esi]
1381		− mov edi, [tmp_edi]
1382		− add esi, eax
1383		− add edi, eax
1384		−
1385		− dec dword [iPicHeight]
1386		− jnz sqdiff_bgd_height_loop
1387		−
1388		− mov edx, [psadframe]
1389		− mov ebp, [tmp_sadframe]
1390		− mov [edx], ebp
1391		−
1392		− add esp, localsize
1393		− pop ebx
1394		− pop edi
1395		− pop esi
1396		− pop ebp
1397		−%undef cur_data
1398		−%undef ref_data
1399		−%undef iPicWidth
1400		−%undef iPicHeight
1401		−%undef iPicStride
1402		−%undef psadframe
1403		−%undef psad8x8
1404		−%undef psum16x16
1405		−%undef psqsum16x16
1406		−%undef psqdiff16x16
1407		−%undef p_sd8x8
1408		−%undef p_mad8x8
1409		−%undef tmp_esi
1410		−%undef tmp_edi
1411		−%undef pushsize
1412		−%undef localsize
1413		− ret
	1886	+
	1887	+ pxor xmm7, xmm7 ; pSad8x8 interleaves sqsum16x16: sqsum1 sad1 sqsum0 sad0
	1888	+ pxor xmm6, xmm6 ; sum_8x8 interleaves cur and pRef in Dword, Sref1 Scur1 Sref0 Scur0
	1889	+ pxor xmm5, xmm5 ; pMad8x8
	1890	+ pxor xmm4, xmm4 ; sqdiff_16x16 four Dword
	1891	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
	1892	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
	1893	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
	1894	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
	1895	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
	1896	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
	1897	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
	1898	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
	1899	+
	1900	+ mov r14, psad8x8
	1901	+ movdqa xmm2, xmm7
	1902	+ pshufd xmm1, xmm2, 00001110b
	1903	+ movd [r14], xmm2
	1904	+ movd [r14+4], xmm1
	1905	+ add r14, 8
	1906	+ mov psad8x8, r14 ; sad8x8
	1907	+
	1908	+ paddd xmm1, xmm2
	1909	+ movd r14d, xmm1
	1910	+ movd xmm9,r14d
	1911	+ paddd xmm8, xmm9 ; iFrameSad
	1912	+
	1913	+ mov r14, psum16x16
	1914	+ movdqa xmm1, xmm6
	1915	+ pshufd xmm2, xmm1, 00001110b
	1916	+ paddd xmm1, xmm2
	1917	+ movd [r14], xmm1 ; sum
	1918	+
	1919	+ mov r14, p_sd8x8
	1920	+ pshufd xmm1, xmm6, 11110101b ; Sref1 Sref1 Sref0 Sref0
	1921	+ psubd xmm6, xmm1 ; 00 diff1 00 diff0
	1922	+ pshufd xmm1, xmm6, 00001000b ; xx xx diff1 diff0
	1923	+ movq [r14], xmm1
	1924	+ add r14, 8
	1925	+ mov p_sd8x8, r14
	1926	+
	1927	+ mov r14, p_mad8x8
	1928	+ WELS_MAX_REG_SSE2 xmm5
	1929	+
	1930	+ movhlps xmm1, xmm5
	1931	+ push r0
	1932	+ movd r0d, xmm5
	1933	+ mov [r14], r0b
	1934	+ movd r0d, xmm1
	1935	+ mov [r14+1],r0b
	1936	+ pop r0
	1937	+ add r14, 2
	1938	+ mov p_mad8x8, r14
	1939	+
	1940	+ psrlq xmm7, 32
	1941	+ psllq xmm7, 32 ; clear sad
	1942	+ pxor xmm6, xmm6 ; sum_8x8 interleaves cur and pRef in Dword, Sref1 Scur1 Sref0 Scur0
	1943	+ pxor xmm5, xmm5 ; pMad8x8
	1944	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
	1945	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
	1946	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
	1947	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
	1948	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
	1949	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
	1950	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
	1951	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
	1952	+
	1953	+ mov r14, psad8x8
	1954	+ movdqa xmm2, xmm7
	1955	+ pshufd xmm1, xmm2, 00001110b
	1956	+ movd [r14], xmm2
	1957	+ movd [r14+4], xmm1
	1958	+ add r14, 8
	1959	+ mov psad8x8, r14 ; sad8x8
	1960	+
	1961	+ paddd xmm1, xmm2
	1962	+ movd r14d, xmm1
	1963	+ movd xmm9, r14d
	1964	+ paddd xmm8, xmm9 ; iFrameSad
	1965	+
	1966	+ mov r14, psum16x16
	1967	+ movdqa xmm1, xmm6
	1968	+ pshufd xmm2, xmm1, 00001110b
	1969	+ paddd xmm1, xmm2
	1970	+ movd r15d, xmm1 ; sum
	1971	+ add [r14], r15d
	1972	+ add r14, 4
	1973	+ mov psum16x16, r14
	1974	+
	1975	+ mov r14, psqsum16x16
	1976	+ psrlq xmm7, 32
	1977	+ pshufd xmm2, xmm7, 00001110b
	1978	+ paddd xmm2, xmm7
	1979	+ movd [r14], xmm2 ; sqsum
	1980	+ add r14, 4
	1981	+ mov psqsum16x16, r14
	1982	+
	1983	+ mov r14, p_sd8x8
	1984	+ pshufd xmm1, xmm6, 11110101b ; Sref1 Sref1 Sref0 Sref0
	1985	+ psubd xmm6, xmm1 ; 00 diff1 00 diff0
	1986	+ pshufd xmm1, xmm6, 00001000b ; xx xx diff1 diff0
	1987	+ movq [r14], xmm1
	1988	+ add r14, 8
	1989	+ mov p_sd8x8, r14
	1990	+
	1991	+ mov r14, p_mad8x8
	1992	+ WELS_MAX_REG_SSE2 xmm5
	1993	+
	1994	+
	1995	+ movhlps xmm1, xmm5
	1996	+ push r0
	1997	+ movd r0d, xmm5
	1998	+ mov [r14], r0b
	1999	+ movd r0d, xmm1
	2000	+ mov [r14+1],r0b
	2001	+ pop r0
	2002	+ add r14, 2
	2003	+ mov p_mad8x8, r14
	2004	+
	2005	+ mov r14, psqdiff16x16
	2006	+ pshufd xmm1, xmm4, 00001110b
	2007	+ paddd xmm4, xmm1
	2008	+ pshufd xmm1, xmm4, 00000001b
	2009	+ paddd xmm4, xmm1
	2010	+ movd [r14], xmm4
	2011	+ add r14, 4
	2012	+ mov psqdiff16x16, r14
	2013	+
	2014	+ add r14, 16
	2015	+ sub r0, r13
	2016	+ sub r1, r13
	2017	+ add r0, 16
	2018	+ add r1, 16
	2019	+
	2020	+ dec r2
	2021	+ jnz sqdiff_bgd_width_loop
	2022	+ pop r2
	2023	+ %assign push_num push_num-1
	2024	+ mov r0, r10
	2025	+ mov r1, r11
	2026	+ add r0, r13
	2027	+ add r1, r13
	2028	+
	2029	+ dec r3
	2030	+ jnz sqdiff_bgd_height_loop
	2031	+
	2032	+ mov r14, psadframe
	2033	+ movd [r14], xmm8
	2034	+
	2035	+ pop r15
	2036	+ pop r14
	2037	+ pop r13
	2038	+ pop r12
	2039	+%assign push_num 0
	2040	+%undef cur_data
	2041	+%undef ref_data
	2042	+%undef iPicWidth
	2043	+%undef iPicHeight
	2044	+%undef iPicStride
	2045	+%undef psadframe
	2046	+%undef psad8x8
	2047	+%undef psum16x16
	2048	+%undef psqsum16x16
	2049	+%undef psqdiff16x16
	2050	+%undef p_sd8x8
	2051	+%undef p_mad8x8
	2052	+%undef tmp_esi
	2053	+%undef tmp_edi
	2054	+%undef pushsize
	2055	+%undef localsize
	2056	+ ret
1414	2057	%endif

codec/processing/src/common/cpu.cpp+0−196

@@ -1,196 +0,0 @@
1		−/*!
2		− * \copy
3		− * Copyright (c) 2009-2013, Cisco Systems
4		− * All rights reserved.
5		− *
6		− * Redistribution and use in source and binary forms, with or without
7		− * modification, are permitted provided that the following conditions
8		− * are met:
9		− *
10		− * * Redistributions of source code must retain the above copyright
11		− * notice, this list of conditions and the following disclaimer.
12		− *
13		− * * Redistributions in binary form must reproduce the above copyright
14		− * notice, this list of conditions and the following disclaimer in
15		− * the documentation and/or other materials provided with the
16		− * distribution.
17		− *
18		− * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
19		− * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
20		− * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
21		− * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
22		− * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
23		− * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
24		− * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
25		− * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
26		− * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27		− * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
28		− * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29		− * POSSIBILITY OF SUCH DAMAGE.
30		− *
31		− * \file cpu.c
32		− *
33		− * \brief CPU compatibility detection
34		− *
35		− * \date 04/29/2009 Created
36		− *
37		− *************************************************************************************
38		− */
39		−
40		−#include "util.h"
41		−#include "cpu.h"
42		−
43		−WELSVP_NAMESPACE_BEGIN
44		−
45		−#define CPU_Vender_AMD "AuthenticAMD"
46		−#define CPU_Vender_INTEL "GenuineIntel"
47		−#define CPU_Vender_CYRIX "CyrixInstead"
48		−
49		−
50		−#if defined(X86_ASM)
51		−
52		−uint32_t WelsCPUFeatureDetect (int32_t* pNumberOfLogicProcessors) {
53		− uint32_t uiCPU = 0;
54		− uint32_t uiFeatureA = 0, uiFeatureB = 0, uiFeatureC = 0, uiFeatureD = 0;
55		− int32_t CacheLineSize = 0;
56		− int8_t chVenderName[16] = { 0 };
57		−
58		− if (!WelsCPUIdVerify()) {
59		− /* cpuid is not supported in cpu */
60		− return 0;
61		− }
62		−
63		− WelsCPUId (0, &uiFeatureA, (uint32_t)&chVenderName[0], (uint32_t)&chVenderName[8], (uint32_t*)&chVenderName[4]);
64		− if (uiFeatureA == 0) {
65		− /* maximum input value for basic cpuid information */
66		− return 0;
67		− }
68		−
69		− WelsCPUId (1, &uiFeatureA, &uiFeatureB, &uiFeatureC, &uiFeatureD);
70		− if ((uiFeatureD & 0x00800000) == 0) {
71		− /* Basic MMX technology is not support in cpu, mean nothing for us so return here */
72		− return 0;
73		− }
74		−
75		− uiCPU = WELS_CPU_MMX;
76		− if (uiFeatureD & 0x02000000) {
77		− /* SSE technology is identical to AMD MMX extensions */
78		− uiCPU \|= WELS_CPU_MMXEXT \| WELS_CPU_SSE;
79		− }
80		− if (uiFeatureD & 0x04000000) {
81		− /* SSE2 support here */
82		− uiCPU \|= WELS_CPU_SSE2;
83		− }
84		− if (uiFeatureD & 0x00000001) {
85		− /* x87 FPU on-chip checking */
86		− uiCPU \|= WELS_CPU_FPU;
87		− }
88		− if (uiFeatureD & 0x00008000) {
89		− /* CMOV instruction checking */
90		− uiCPU \|= WELS_CPU_CMOV;
91		− }
92		− if (!strcmp ((const str_t*)chVenderName, CPU_Vender_INTEL)) { // confirmed_safe_unsafe_usage
93		− if (uiFeatureD & 0x10000000) {
94		− /* Multi-Threading checking: contains of multiple logic processors */
95		− uiCPU \|= WELS_CPU_HTT;
96		− }
97		− }
98		−
99		− if (uiFeatureC & 0x00000001) {
100		− /* SSE3 support here */
101		− uiCPU \|= WELS_CPU_SSE3;
102		− }
103		− if (uiFeatureC & 0x00000200) {
104		− /* SSSE3 support here */
105		− uiCPU \|= WELS_CPU_SSSE3;
106		− }
107		− if (uiFeatureC & 0x00080000) {
108		− /* SSE4.1 support here, 45nm Penryn processor */
109		− uiCPU \|= WELS_CPU_SSE41;
110		− }
111		− if (uiFeatureC & 0x00100000) {
112		− /* SSE4.2 support here, next generation Nehalem processor */
113		− uiCPU \|= WELS_CPU_SSE42;
114		− }
115		− if (WelsCPUSupportAVX (uiFeatureA, uiFeatureC)) { //
116		− /* AVX supported */
117		− uiCPU \|= WELS_CPU_AVX;
118		− }
119		− if (WelsCPUSupportFMA (uiFeatureA, uiFeatureC)) { //
120		− /* AVX FMA supported */
121		− uiCPU \|= WELS_CPU_FMA;
122		− }
123		− if (uiFeatureC & 0x02000000) {
124		− /* AES checking */
125		− uiCPU \|= WELS_CPU_AES;
126		− }
127		− if (uiFeatureC & 0x00400000) {
128		− /* MOVBE checking */
129		− uiCPU \|= WELS_CPU_MOVBE;
130		− }
131		−
132		− if (pNumberOfLogicProcessors != NULL) {
133		− // HTT enabled on chip
134		− *pNumberOfLogicProcessors = (uiFeatureB & 0x00ff0000) >> 16; // feature bits: 23-16 on returned EBX
135		− }
136		−
137		− WelsCPUId (0x80000000, &uiFeatureA, &uiFeatureB, &uiFeatureC, &uiFeatureD);
138		−
139		− if ((!strcmp ((const str_t*)chVenderName, CPU_Vender_AMD))
140		− && (uiFeatureA >= 0x80000001)) { // confirmed_safe_unsafe_usage
141		− WelsCPUId (0x80000001, &uiFeatureA, &uiFeatureB, &uiFeatureC, &uiFeatureD);
142		− if (uiFeatureD & 0x00400000) {
143		− uiCPU \|= WELS_CPU_MMXEXT;
144		− }
145		− if (uiFeatureD & 0x80000000) {
146		− uiCPU \|= WELS_CPU_3DNOW;
147		− }
148		− }
149		−
150		− if (!strcmp ((const str_t*)chVenderName, CPU_Vender_INTEL)) { // confirmed_safe_unsafe_usage
151		− int32_t family, model;
152		−
153		− WelsCPUId (1, &uiFeatureA, &uiFeatureB, &uiFeatureC, &uiFeatureD);
154		− family = ((uiFeatureA >> 8) & 0xf) + ((uiFeatureA >> 20) & 0xff);
155		− model = ((uiFeatureA >> 4) & 0xf) + ((uiFeatureA >> 12) & 0xf0);
156		−
157		− if ((family == 6) && (model == 9 \|\| model == 13 \|\| model == 14)) {
158		− uiCPU &= ~ (WELS_CPU_SSE2 \| WELS_CPU_SSE3);
159		− }
160		− }
161		−
162		− // get cache line size
163		− if ((!strcmp ((const str_t*)chVenderName, CPU_Vender_INTEL))
164		− \|\| ! (strcmp ((const str_t*)chVenderName, CPU_Vender_CYRIX))) { // confirmed_safe_unsafe_usage
165		− WelsCPUId (1, &uiFeatureA, &uiFeatureB, &uiFeatureC, &uiFeatureD);
166		−
167		− CacheLineSize = (uiFeatureB & 0xff00) >>
168		− 5; // ((clflush_line_size >> 8) << 3), CLFLUSH_line_size * 8 = CacheLineSize_in_byte
169		−
170		− if (CacheLineSize == 128) {
171		− uiCPU \|= WELS_CPU_CACHELINE_128;
172		− } else if (CacheLineSize == 64) {
173		− uiCPU \|= WELS_CPU_CACHELINE_64;
174		− } else if (CacheLineSize == 32) {
175		− uiCPU \|= WELS_CPU_CACHELINE_32;
176		− } else if (CacheLineSize == 16) {
177		− uiCPU \|= WELS_CPU_CACHELINE_16;
178		− }
179		− }
180		−
181		− return uiCPU;
182		−}
183		−
184		−
185		−void WelsCPURestore (const uint32_t kuiCPU) {
186		− if (kuiCPU & (WELS_CPU_MMX \| WELS_CPU_MMXEXT \| WELS_CPU_3DNOW \| WELS_CPU_3DNOWEXT)) {
187		− WelsEmms();
188		− }
189		−}
190		−
191		−#endif
192		−
193		−
194		−WELSVP_NAMESPACE_END
195		−
196		−

codec/processing/src/common/cpu.h+0−102

@@ -1,102 +0,0 @@
1		−/*!
2		− * \copy
3		− * Copyright (c) 2009-2013, Cisco Systems
4		− * All rights reserved.
5		− *
6		− * Redistribution and use in source and binary forms, with or without
7		− * modification, are permitted provided that the following conditions
8		− * are met:
9		− *
10		− * * Redistributions of source code must retain the above copyright
11		− * notice, this list of conditions and the following disclaimer.
12		− *
13		− * * Redistributions in binary form must reproduce the above copyright
14		− * notice, this list of conditions and the following disclaimer in
15		− * the documentation and/or other materials provided with the
16		− * distribution.
17		− *
18		− * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
19		− * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
20		− * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
21		− * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
22		− * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
23		− * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
24		− * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
25		− * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
26		− * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27		− * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
28		− * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29		− * POSSIBILITY OF SUCH DAMAGE.
30		− *
31		− * \file cpu.h
32		− *
33		− * \brief CPU feature compatibility detection
34		− *
35		− * \date 04/29/2009 Created
36		− *
37		− *************************************************************************************
38		− */
39		−
40		−#ifndef WELSVP_CPU_H
41		−#define WELSVP_CPU_H
42		−
43		−#include "typedef.h"
44		−
45		−WELSVP_NAMESPACE_BEGIN
46		−
47		−/*
48		− * WELS CPU feature flags
49		− */
50		−#define WELS_CPU_MMX 0x00000001 /* mmx */
51		−#define WELS_CPU_MMXEXT 0x00000002 /* mmx-ext*/
52		−#define WELS_CPU_SSE 0x00000004 /* sse */
53		−#define WELS_CPU_SSE2 0x00000008 /* sse 2 */
54		−#define WELS_CPU_SSE3 0x00000010 /* sse 3 */
55		−#define WELS_CPU_SSE41 0x00000020 /* sse 4.1 */
56		−#define WELS_CPU_3DNOW 0x00000040 /* 3dnow! */
57		−#define WELS_CPU_3DNOWEXT 0x00000080 /* 3dnow! ext */
58		−#define WELS_CPU_ALTIVEC 0x00000100 /* altivec */
59		−#define WELS_CPU_SSSE3 0x00000200 /* ssse3 */
60		−#define WELS_CPU_SSE42 0x00000400 /* sse 4.2 */
61		−
62		−/* CPU features application extensive */
63		−#define WELS_CPU_AVX 0x00000800 /* Advanced Vector eXtentions */
64		−#define WELS_CPU_FPU 0x00001000 /* x87-FPU on chip */
65		−#define WELS_CPU_HTT 0x00002000 /* Hyper-Threading Technology (HTT), Multi-threading enabled feature:
66		− physical processor package is capable of supporting more than one logic processor
67		− */
68		−#define WELS_CPU_CMOV 0x00004000 /* Conditional Move Instructions,
69		− also if x87-FPU is present at indicated by the CPUID.FPU feature bit, then FCOMI and FCMOV are supported
70		− */
71		−#define WELS_CPU_MOVBE 0x00008000 /* MOVBE instruction */
72		−#define WELS_CPU_AES 0x00010000 /* AES instruction extensions */
73		−#define WELS_CPU_FMA 0x00020000 /* AVX VEX FMA instruction sets */
74		−
75		−#define WELS_CPU_CACHELINE_16 0x10000000 /* CacheLine Size 16 */
76		−#define WELS_CPU_CACHELINE_32 0x20000000 /* CacheLine Size 32 */
77		−#define WELS_CPU_CACHELINE_64 0x40000000 /* CacheLine Size 64 */
78		−#define WELS_CPU_CACHELINE_128 0x80000000 /* CacheLine Size 128 */
79		−
80		−/*
81		− * Interfaces for CPU core feature detection as below
82		− */
83		−
84		−#ifdef X86_ASM
85		−WELSVP_EXTERN_C_BEGIN
86		−
87		−int32_t WelsCPUIdVerify();
88		−
89		−void WelsCPUId (uint32_t uiIndex, uint32_t* pFeatureA, uint32_t* pFeatureB, uint32_t* pFeatureC, uint32_t* pFeatureD);
90		−int32_t WelsCPUSupportAVX (uint32_t eax, uint32_t ecx);
91		−int32_t WelsCPUSupportFMA (uint32_t eax, uint32_t ecx);
92		−
93		−void WelsEmms();
94		−
95		−WELSVP_EXTERN_C_END
96		−#endif
97		−
98		−uint32_t WelsCPUFeatureDetect (int32_t* pNumberOfLogicProcessors);
99		−
100		−WELSVP_NAMESPACE_END
101		−
102		−#endif

codec/processing/src/vaacalc/vaacalculation.cpp+5−5

@@ -58,11 +58,11 @@ void CVAACalculation::InitVaaFuncs (SVaaFuncs& sVaaFuncs, int32_t iCpuFlag) { · 57 unmodified lines
58	58	sVaaFuncs.pfVAACalcSadVar = VAACalcSadVar_c;
59	59	#ifdef X86_ASM
60	60	if ((iCpuFlag & WELS_CPU_SSE2) == WELS_CPU_SSE2) {
61		− /* sVaaFuncs.pfVAACalcSad = VAACalcSad_sse2;
62		− sVaaFuncs.pfVAACalcSadBgd = VAACalcSadBgd_sse2;
63		− sVaaFuncs.pfVAACalcSadSsd = VAACalcSadSsd_sse2;
64		− sVaaFuncs.pfVAACalcSadSsdBgd = VAACalcSadSsdBgd_sse2;
65		− sVaaFuncs.pfVAACalcSadVar = VAACalcSadVar_sse2;*/
	61	+ sVaaFuncs.pfVAACalcSad = VAACalcSad_sse2;
	62	+ sVaaFuncs.pfVAACalcSadBgd = VAACalcSadBgd_sse2;
	63	+ sVaaFuncs.pfVAACalcSadSsd = VAACalcSadSsd_sse2;
	64	+ sVaaFuncs.pfVAACalcSadSsdBgd = VAACalcSadSsdBgd_sse2;
	65	+ sVaaFuncs.pfVAACalcSadVar = VAACalcSadVar_sse2;
66	66	}
67	67	#endif//X86_ASM
68	68	}

fix win64 float issue, enable AQ assembly

Keyboard shortcuts

General

Issues and pull requests

Issues, pull requests and Inbox

Repository code