Merge pull request #1140 from zhilwang/x86_64-downsample

Add X86 64bit asm code for downsample

dongzha committed 12 years ago

Commit da0f65e2 parents 6310017 eaf7d654 files · +1062 −373Browse files

Showing the changes to codec/processing/src (4 of 4 files). Show all

Unified · Split

4 files

codec/processing/src/downsample/downsample.cpp +9 −9
codec/processing/src/downsample/downsample.h +2 −6
codec/processing/src/downsample/downsamplefuncs.cpp +25 −25
codec/processing/src/x86/downsample_bilinear.asm +1026 −333

codec/processing/src/downsample/downsample.cpp+9−9

@@ -57,21 +57,21 @@ void CDownsampling::InitDownsampleFuncs (SDownsampleFuncs& sDownsampleFunc, int · 56 unmodified lines
57	57	sDownsampleFunc.pfGeneralRatioLuma = GeneralBilinearFastDownsampler_c;
58	58	#if defined(X86_ASM)
59	59	if (iCpuFlag & WELS_CPU_SSE) {
60		− /* sDownsampleFunc.pfHalfAverage[0] = DyadicBilinearDownsamplerWidthx32_sse;
61		− sDownsampleFunc.pfHalfAverage[1] = DyadicBilinearDownsamplerWidthx16_sse;
62		− sDownsampleFunc.pfHalfAverage[2] = DyadicBilinearDownsamplerWidthx8_sse;*/
	60	+ sDownsampleFunc.pfHalfAverage[0] = DyadicBilinearDownsamplerWidthx32_sse;
	61	+ sDownsampleFunc.pfHalfAverage[1] = DyadicBilinearDownsamplerWidthx16_sse;
	62	+ sDownsampleFunc.pfHalfAverage[2] = DyadicBilinearDownsamplerWidthx8_sse;
63	63	}
64	64	if (iCpuFlag & WELS_CPU_SSE2) {
65		− // sDownsampleFunc.pfGeneralRatioChroma = GeneralBilinearAccurateDownsamplerWrap_sse2;
66		− // sDownsampleFunc.pfGeneralRatioLuma = GeneralBilinearFastDownsamplerWrap_sse2;
	65	+ sDownsampleFunc.pfGeneralRatioChroma = GeneralBilinearAccurateDownsamplerWrap_sse2;
	66	+ sDownsampleFunc.pfGeneralRatioLuma = GeneralBilinearFastDownsamplerWrap_sse2;
67	67	}
68	68	if (iCpuFlag & WELS_CPU_SSSE3) {
69		− // sDownsampleFunc.pfHalfAverage[0] = DyadicBilinearDownsamplerWidthx32_ssse3;
70		− // sDownsampleFunc.pfHalfAverage[1] = DyadicBilinearDownsamplerWidthx16_ssse3;
	69	+ sDownsampleFunc.pfHalfAverage[0] = DyadicBilinearDownsamplerWidthx32_ssse3;
	70	+ sDownsampleFunc.pfHalfAverage[1] = DyadicBilinearDownsamplerWidthx16_ssse3;
71	71	}
72	72	if (iCpuFlag & WELS_CPU_SSE41) {
73		− // sDownsampleFunc.pfHalfAverage[0] = DyadicBilinearDownsamplerWidthx32_sse4;
74		− // sDownsampleFunc.pfHalfAverage[1] = DyadicBilinearDownsamplerWidthx16_sse4;
	73	+ sDownsampleFunc.pfHalfAverage[0] = DyadicBilinearDownsamplerWidthx32_sse4;
	74	+ sDownsampleFunc.pfHalfAverage[1] = DyadicBilinearDownsamplerWidthx16_sse4;
75	75	}
76	76	#endif//X86_ASM
77	77

codec/processing/src/downsample/downsample.h+2−6

@@ -94,13 +94,9 @@ GeneralDownsampleFunc GeneralBilinearFastDownsamplerWrap_sse2; · 93 unmodified lines
94	94	GeneralDownsampleFunc GeneralBilinearAccurateDownsamplerWrap_sse2;
95	95
96	96	void GeneralBilinearFastDownsampler_sse2 (uint8_t* pDst, const int32_t kiDstStride, const int32_t kiDstWidth,
97		− const int32_t kiDstHeight,
98		− uint8_t* pSrc, const int32_t kiSrcStride, const int32_t kiSrcWidth, const int32_t kiSrcHeight,
99		− const uint32_t kuiScaleX, const uint32_t kuiScaleY);
	97	+ const int32_t kiDstHeight, uint8_t* pSrc, const int32_t kiSrcStride, const uint32_t kuiScaleX, const uint32_t kuiScaleY);
100	98	void GeneralBilinearAccurateDownsampler_sse2 (uint8_t* pDst, const int32_t kiDstStride, const int32_t kiDstWidth,
101		− const int32_t kiDstHeight,
102		− uint8_t* pSrc, const int32_t kiSrcStride, const int32_t kiSrcWidth, const int32_t kiSrcHeight,
103		− const uint32_t kuiScaleX, const uint32_t kuiScaleY);
	99	+ const int32_t kiDstHeight, uint8_t* pSrc, const int32_t kiSrcStride, const uint32_t kuiScaleX, const uint32_t kuiScaleY);
104	100	WELSVP_EXTERN_C_END
105	101	#endif
106	102

codec/processing/src/downsample/downsamplefuncs.cpp+25−25

@@ -202,31 +202,31 @@ void GeneralBilinearAccurateDownsampler_c (uint8_t* pDst, const int32_t kiDstStr · 201 unmodified lines
202	202
203	203
204	204	#ifdef X86_ASM
205		−//void GeneralBilinearFastDownsamplerWrap_sse2 (uint8_t* pDst, const int32_t kiDstStride, const int32_t kiDstWidth,
206		−// const int32_t kiDstHeight,
207		−// uint8_t* pSrc, const int32_t kiSrcStride, const int32_t kiSrcWidth, const int32_t kiSrcHeight) {
208		−// const int32_t kiScaleBitWidth = 16, kiScaleBitHeight = 15;
209		−// const uint32_t kuiScaleWidth = (1 << kiScaleBitWidth), kuiScaleHeight = (1 << kiScaleBitHeight);
210		−//
211		−// uint32_t uiScalex = WELS_ROUND ((float)kiSrcWidth / (float)kiDstWidth * kuiScaleWidth);
212		−// uint32_t uiScaley = WELS_ROUND ((float)kiSrcHeight / (float)kiDstHeight * kuiScaleHeight);
213		−//
214		−// GeneralBilinearFastDownsampler_sse2 (pDst, kiDstStride, kiDstWidth, kiDstHeight,
215		−// pSrc, kiSrcStride, kiSrcWidth, kiSrcHeight, uiScalex, uiScaley);
216		−//}
217		−//
218		−//void GeneralBilinearAccurateDownsamplerWrap_sse2 (uint8_t* pDst, const int32_t kiDstStride, const int32_t kiDstWidth,
219		−// const int32_t kiDstHeight,
220		−// uint8_t* pSrc, const int32_t kiSrcStride, const int32_t kiSrcWidth, const int32_t kiSrcHeight) {
221		−// const int32_t kiScaleBit = 15;
222		−// const uint32_t kuiScale = (1 << kiScaleBit);
223		−//
224		−// uint32_t uiScalex = WELS_ROUND ((float)kiSrcWidth / (float)kiDstWidth * kuiScale);
225		−// uint32_t uiScaley = WELS_ROUND ((float)kiSrcHeight / (float)kiDstHeight * kuiScale);
226		−//
227		−// GeneralBilinearAccurateDownsampler_sse2 (pDst, kiDstStride, kiDstWidth, kiDstHeight,
228		−// pSrc, kiSrcStride, kiSrcWidth, kiSrcHeight, uiScalex, uiScaley);
229		−//}
	205	+void GeneralBilinearFastDownsamplerWrap_sse2 (uint8_t* pDst, const int32_t kiDstStride, const int32_t kiDstWidth,
	206	+ const int32_t kiDstHeight,
	207	+ uint8_t* pSrc, const int32_t kiSrcStride, const int32_t kiSrcWidth, const int32_t kiSrcHeight) {
	208	+ const int32_t kiScaleBitWidth = 16, kiScaleBitHeight = 15;
	209	+ const uint32_t kuiScaleWidth = (1 << kiScaleBitWidth), kuiScaleHeight = (1 << kiScaleBitHeight);
	210	+
	211	+ uint32_t uiScalex = WELS_ROUND ((float)kiSrcWidth / (float)kiDstWidth * kuiScaleWidth);
	212	+ uint32_t uiScaley = WELS_ROUND ((float)kiSrcHeight / (float)kiDstHeight * kuiScaleHeight);
	213	+
	214	+ GeneralBilinearFastDownsampler_sse2 (pDst, kiDstStride, kiDstWidth, kiDstHeight,
	215	+ pSrc, kiSrcStride, uiScalex, uiScaley);
	216	+}
	217	+
	218	+void GeneralBilinearAccurateDownsamplerWrap_sse2 (uint8_t* pDst, const int32_t kiDstStride, const int32_t kiDstWidth,
	219	+ const int32_t kiDstHeight,
	220	+ uint8_t* pSrc, const int32_t kiSrcStride, const int32_t kiSrcWidth, const int32_t kiSrcHeight) {
	221	+ const int32_t kiScaleBit = 15;
	222	+ const uint32_t kuiScale = (1 << kiScaleBit);
	223	+
	224	+ uint32_t uiScalex = WELS_ROUND ((float)kiSrcWidth / (float)kiDstWidth * kuiScale);
	225	+ uint32_t uiScaley = WELS_ROUND ((float)kiSrcHeight / (float)kiDstHeight * kuiScale);
	226	+
	227	+ GeneralBilinearAccurateDownsampler_sse2 (pDst, kiDstStride, kiDstWidth, kiDstHeight,
	228	+ pSrc, kiSrcStride, uiScalex, uiScaley);
	229	+}
230	230	#endif //X86_ASM
231	231
232	232	#ifdef HAVE_NEON

codec/processing/src/x86/downsample_bilinear.asm+1026−333

@@ -39,7 +39,7 @@ · 38 unmodified lines
39	39	;*
40	40	;*************************************************************************/
41	41	%include "asm_inc.asm"
42		−%ifdef X86_32
	42	+
43	43	;***********************************************************************
44	44	; Macros and other preprocessor constants
45	45	;***********************************************************************
@@ -64,6 +64,8 @@ shufb_mask_low: · 18 unmodified lines
64	64	db 00h, 80h, 02h, 80h, 04h, 80h, 06h, 80h, 08h, 80h, 0ah, 80h, 0ch, 80h, 0eh, 80h
65	65	shufb_mask_high:
66	66	db 01h, 80h, 03h, 80h, 05h, 80h, 07h, 80h, 09h, 80h, 0bh, 80h, 0dh, 80h, 0fh, 80h
	67	+add_extra_half:
	68	+ dd 16384,0,0,0
67	69
68	70
69	71	;***********************************************************************
@@ -78,28 +80,36 @@ SECTION .text · 8 unmodified lines
78	80	; const int iSrcWidth, const int iSrcHeight );
79	81	;***********************************************************************
80	82	WELS_EXTERN DyadicBilinearDownsamplerWidthx32_sse
81		− push ebx
82		− push edx
83		− push esi
84		− push edi
85		− push ebp
86		−
87		− mov edi, [esp+24] ; pDst
88		− mov edx, [esp+28] ; iDstStride
89		− mov esi, [esp+32] ; pSrc
90		− mov ecx, [esp+36] ; iSrcStride
91		− mov ebp, [esp+44] ; iSrcHeight
92		−
93		− sar ebp, $01 ; iSrcHeight >> 1
94		−
95		−.yloops:
96		− mov eax, [esp+40] ; iSrcWidth
97		− sar eax, $01 ; iSrcWidth >> 1
98		− mov ebx, eax ; iDstWidth restored at ebx
99		− sar eax, $04 ; (iSrcWidth >> 1) / 16 ; loop count = num_of_mb
100		− neg ebx ; - (iSrcWidth >> 1)
	83	+%ifdef X86_32
	84	+ push r6
	85	+ %assign push_num 1
	86	+%else
	87	+ %assign push_num 0
	88	+%endif
	89	+ LOAD_6_PARA
	90	+ SIGN_EXTENSION r1, r1d
	91	+ SIGN_EXTENSION r3, r3d
	92	+ SIGN_EXTENSION r4, r4d
	93	+ SIGN_EXTENSION r5, r5d
	94	+
	95	+%ifndef X86_32
	96	+ push r12
	97	+ mov r12, r4
	98	+%endif
	99	+ sar r5, $01 ; iSrcHeight >> 1
	100	+
	101	+.yloops1:
	102	+%ifdef X86_32
	103	+ mov r4, arg5
	104	+%else
	105	+ mov r4, r12
	106	+%endif
	107	+ sar r4, $01 ; iSrcWidth >> 1
	108	+ mov r6, r4 ; iDstWidth restored at ebx
	109	+ sar r4, $04 ; (iSrcWidth >> 1) / 16 ; loop count = num_of_mb
	110	+ neg r6 ; - (iSrcWidth >> 1)
101	111	; each loop = source bandwidth: 32 bytes
102		−.xloops:
	112	+.xloops1:
103	113	; 1st part horizonal loop: x16 bytes
104	114	; mem hi<- ->lo
105	115	;1st Line Src: mm0: d D c C b B a A mm1: h H g G f F e E
@@ -108,10 +118,10 @@ WELS_EXTERN DyadicBilinearDownsamplerWidthx32_sse · 2 unmodified lines
108	118	;: H G F E D C B A, P O N M L K J I
109	119	;: h g f e d c b a, p o n m l k j i
110	120	;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
111		− movq mm0, [esi] ; 1st pSrc line
112		− movq mm1, [esi+8] ; 1st pSrc line + 8
113		− movq mm2, [esi+ecx] ; 2nd pSrc line
114		− movq mm3, [esi+ecx+8] ; 2nd pSrc line + 8
	121	+ movq mm0, [r2] ; 1st pSrc line
	122	+ movq mm1, [r2+8] ; 1st pSrc line + 8
	123	+ movq mm2, [r2+r3] ; 2nd pSrc line
	124	+ movq mm3, [r2+r3+8] ; 2nd pSrc line + 8
115	125
116	126	; to handle mm0, mm1, mm2, mm3
117	127	pshufw mm4, mm0, 0d8h ; d D b B c C a A ; 11011000 B
@@ -156,10 +166,10 @@ WELS_EXTERN DyadicBilinearDownsamplerWidthx32_sse · 38 unmodified lines
156	166	;: H G F E D C B A, P O N M L K J I
157	167	;: h g f e d c b a, p o n m l k j i
158	168	;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
159		− movq mm1, [esi+16] ; 1st pSrc line + 16
160		− movq mm2, [esi+24] ; 1st pSrc line + 24
161		− movq mm3, [esi+ecx+16] ; 2nd pSrc line + 16
162		− movq mm4, [esi+ecx+24] ; 2nd pSrc line + 24
	169	+ movq mm1, [r2+16] ; 1st pSrc line + 16
	170	+ movq mm2, [r2+24] ; 1st pSrc line + 24
	171	+ movq mm3, [r2+r3+16] ; 2nd pSrc line + 16
	172	+ movq mm4, [r2+r3+24] ; 2nd pSrc line + 24
163	173
164	174	; to handle mm1, mm2, mm3, mm4
165	175	pshufw mm5, mm1, 0d8h ; d D b B c C a A ; 11011000 B
@@ -196,31 +206,33 @@ WELS_EXTERN DyadicBilinearDownsamplerWidthx32_sse · 30 unmodified lines
196	206	pavgb mm3, mm7 ; (I+i+1)>>1, .., (P+p+1)>>1, temp_row2
197	207	pavgb mm2, mm3 ; (temp_row1+temp_row2+1)>>1, done in another 2nd horizonal part
198	208
199		− movq [edi ], mm0
200		− movq [edi+8], mm2
	209	+ movq [r0 ], mm0
	210	+ movq [r0+8], mm2
201	211
202	212	; next SMB
203		− lea esi, [esi+32]
204		− lea edi, [edi+16]
	213	+ lea r2, [r2+32]
	214	+ lea r0, [r0+16]
205	215
206		− dec eax
207		− jg near .xloops
	216	+ dec r4
	217	+ jg near .xloops1
208	218
209	219	; next line
210		− lea esi, [esi+2*ecx] ; next end of lines
211		− lea esi, [esi+2ebx] ; reset to base 0 [- 2 iDstWidth]
212		− lea edi, [edi+edx]
213		− lea edi, [edi+ebx] ; reset to base 0 [- iDstWidth]
	220	+ lea r2, [r2+2*r3] ; next end of lines
	221	+ lea r2, [r2+2r6] ; reset to base 0 [- 2 iDstWidth]
	222	+ lea r0, [r0+r1]
	223	+ lea r0, [r0+r6] ; reset to base 0 [- iDstWidth]
214	224
215		− dec ebp
216		− jg near .yloops
	225	+ dec r5
	226	+ jg near .yloops1
217	227
218	228	WELSEMMS
219		− pop ebp
220		− pop edi
221		− pop esi
222		− pop edx
223		− pop ebx
	229	+%ifndef X86_32
	230	+ pop r12
	231	+%endif
	232	+ LOAD_6_PARA_POP
	233	+%ifdef X86_32
	234	+ pop r6
	235	+%endif
224	236	ret
225	237
226	238	;***********************************************************************
@@ -229,28 +241,36 @@ WELS_EXTERN DyadicBilinearDownsamplerWidthx32_sse · 2 unmodified lines
229	241	; const int iSrcWidth, const int iSrcHeight );
230	242	;***********************************************************************
231	243	WELS_EXTERN DyadicBilinearDownsamplerWidthx16_sse
232		− push ebx
233		− push edx
234		− push esi
235		− push edi
236		− push ebp
237		−
238		− mov edi, [esp+24] ; pDst
239		− mov edx, [esp+28] ; iDstStride
240		− mov esi, [esp+32] ; pSrc
241		− mov ecx, [esp+36] ; iSrcStride
242		− mov ebp, [esp+44] ; iSrcHeight
243		−
244		− sar ebp, $01 ; iSrcHeight >> 1
245		−
246		−.yloops:
247		− mov eax, [esp+40] ; iSrcWidth
248		− sar eax, $01 ; iSrcWidth >> 1
249		− mov ebx, eax ; iDstWidth restored at ebx
250		− sar eax, $03 ; (iSrcWidth >> 1) / 8 ; loop count = num_of_mb
251		− neg ebx ; - (iSrcWidth >> 1)
	244	+%ifdef X86_32
	245	+ push r6
	246	+ %assign push_num 1
	247	+%else
	248	+ %assign push_num 0
	249	+%endif
	250	+ LOAD_6_PARA
	251	+ SIGN_EXTENSION r1, r1d
	252	+ SIGN_EXTENSION r3, r3d
	253	+ SIGN_EXTENSION r4, r4d
	254	+ SIGN_EXTENSION r5, r5d
	255	+
	256	+%ifndef X86_32
	257	+ push r12
	258	+ mov r12, r4
	259	+%endif
	260	+ sar r5, $01 ; iSrcHeight >> 1
	261	+
	262	+.yloops2:
	263	+%ifdef X86_32
	264	+ mov r4, arg5
	265	+%else
	266	+ mov r4, r12
	267	+%endif
	268	+ sar r4, $01 ; iSrcWidth >> 1
	269	+ mov r6, r4 ; iDstWidth restored at ebx
	270	+ sar r4, $03 ; (iSrcWidth >> 1) / 8 ; loop count = num_of_mb
	271	+ neg r6 ; - (iSrcWidth >> 1)
252	272	; each loop = source bandwidth: 16 bytes
253		−.xloops:
	273	+.xloops2:
254	274	; 1st part horizonal loop: x16 bytes
255	275	; mem hi<- ->lo
256	276	;1st Line Src: mm0: d D c C b B a A mm1: h H g G f F e E
@@ -259,10 +279,10 @@ WELS_EXTERN DyadicBilinearDownsamplerWidthx16_sse · 2 unmodified lines
259	279	;: H G F E D C B A, P O N M L K J I
260	280	;: h g f e d c b a, p o n m l k j i
261	281	;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
262		− movq mm0, [esi] ; 1st pSrc line
263		− movq mm1, [esi+8] ; 1st pSrc line + 8
264		− movq mm2, [esi+ecx] ; 2nd pSrc line
265		− movq mm3, [esi+ecx+8] ; 2nd pSrc line + 8
	282	+ movq mm0, [r2] ; 1st pSrc line
	283	+ movq mm1, [r2+8] ; 1st pSrc line + 8
	284	+ movq mm2, [r2+r3] ; 2nd pSrc line
	285	+ movq mm3, [r2+r3+8] ; 2nd pSrc line + 8
266	286
267	287	; to handle mm0, mm1, mm2, mm3
268	288	pshufw mm4, mm0, 0d8h ; d D b B c C a A ; 11011000 B
@@ -299,30 +319,32 @@ WELS_EXTERN DyadicBilinearDownsamplerWidthx16_sse · 30 unmodified lines
299	319	pavgb mm1, mm6 ; (I+i+1)>>1, .., (P+p+1)>>1, temp_row2
300	320	pavgb mm0, mm1 ; (temp_row1+temp_row2+1)>>1, pending here and wait another horizonal part done then write memory once
301	321
302		− movq [edi ], mm0
	322	+ movq [r0 ], mm0
303	323
304	324	; next SMB
305		− lea esi, [esi+16]
306		− lea edi, [edi+8]
	325	+ lea r2, [r2+16]
	326	+ lea r0, [r0+8]
307	327
308		− dec eax
309		− jg near .xloops
	328	+ dec r4
	329	+ jg near .xloops2
310	330
311	331	; next line
312		− lea esi, [esi+2*ecx] ; next end of lines
313		− lea esi, [esi+2ebx] ; reset to base 0 [- 2 iDstWidth]
314		− lea edi, [edi+edx]
315		− lea edi, [edi+ebx] ; reset to base 0 [- iDstWidth]
	332	+ lea r2, [r2+2*r3] ; next end of lines
	333	+ lea r2, [r2+2r6] ; reset to base 0 [- 2 iDstWidth]
	334	+ lea r0, [r0+r1]
	335	+ lea r0, [r0+r6] ; reset to base 0 [- iDstWidth]
316	336
317		− dec ebp
318		− jg near .yloops
	337	+ dec r5
	338	+ jg near .yloops2
319	339
320	340	WELSEMMS
321		− pop ebp
322		− pop edi
323		− pop esi
324		− pop edx
325		− pop ebx
	341	+%ifndef X86_32
	342	+ pop r12
	343	+%endif
	344	+ LOAD_6_PARA_POP
	345	+%ifdef X86_32
	346	+ pop r6
	347	+%endif
326	348	ret
327	349
328	350	;***********************************************************************
@@ -331,28 +353,36 @@ WELS_EXTERN DyadicBilinearDownsamplerWidthx16_sse · 2 unmodified lines
331	353	; const int iSrcWidth, const int iSrcHeight );
332	354	;***********************************************************************
333	355	WELS_EXTERN DyadicBilinearDownsamplerWidthx8_sse
334		− push ebx
335		− push edx
336		− push esi
337		− push edi
338		− push ebp
339		−
340		− mov edi, [esp+24] ; pDst
341		− mov edx, [esp+28] ; iDstStride
342		− mov esi, [esp+32] ; pSrc
343		− mov ecx, [esp+36] ; iSrcStride
344		− mov ebp, [esp+44] ; iSrcHeight
345		−
346		− sar ebp, $01 ; iSrcHeight >> 1
347		−
348		−.yloops:
349		− mov eax, [esp+40] ; iSrcWidth
350		− sar eax, $01 ; iSrcWidth >> 1
351		− mov ebx, eax ; iDstWidth restored at ebx
352		− sar eax, $02 ; (iSrcWidth >> 1) / 4 ; loop count = num_of_mb
353		− neg ebx ; - (iSrcWidth >> 1)
	356	+%ifdef X86_32
	357	+ push r6
	358	+ %assign push_num 1
	359	+%else
	360	+ %assign push_num 0
	361	+%endif
	362	+ LOAD_6_PARA
	363	+ SIGN_EXTENSION r1, r1d
	364	+ SIGN_EXTENSION r3, r3d
	365	+ SIGN_EXTENSION r4, r4d
	366	+ SIGN_EXTENSION r5, r5d
	367	+
	368	+%ifndef X86_32
	369	+ push r12
	370	+ mov r12, r4
	371	+%endif
	372	+ sar r5, $01 ; iSrcHeight >> 1
	373	+
	374	+.yloops3:
	375	+%ifdef X86_32
	376	+ mov r4, arg5
	377	+%else
	378	+ mov r4, r12
	379	+%endif
	380	+ sar r4, $01 ; iSrcWidth >> 1
	381	+ mov r6, r4 ; iDstWidth restored at ebx
	382	+ sar r4, $02 ; (iSrcWidth >> 1) / 4 ; loop count = num_of_mb
	383	+ neg r6 ; - (iSrcWidth >> 1)
354	384	; each loop = source bandwidth: 8 bytes
355		−.xloops:
	385	+.xloops3:
356	386	; 1st part horizonal loop: x8 bytes
357	387	; mem hi<- ->lo
358	388	;1st Line Src: mm0: d D c C b B a A
@@ -361,8 +391,8 @@ WELS_EXTERN DyadicBilinearDownsamplerWidthx8_sse · 2 unmodified lines
361	391	;: H G F E D C B A
362	392	;: h g f e d c b a
363	393	;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
364		− movq mm0, [esi] ; 1st pSrc line
365		− movq mm1, [esi+ecx] ; 2nd pSrc line
	394	+ movq mm0, [r2] ; 1st pSrc line
	395	+ movq mm1, [r2+r3] ; 2nd pSrc line
366	396
367	397	; to handle mm0, mm1, mm2, mm3
368	398	pshufw mm2, mm0, 0d8h ; d D b B c C a A ; 11011000 B
@@ -385,30 +415,32 @@ WELS_EXTERN DyadicBilinearDownsamplerWidthx8_sse · 16 unmodified lines
385	415	pshufw mm1, mm0, 04eh ; 01001110 B
386	416	pavgb mm0, mm1 ; (temp_row1+temp_row2+1)>>1, pending here and wait another horizonal part done then write memory once
387	417
388		− movd [edi], mm0
	418	+ movd [r0], mm0
389	419
390	420	; next unit
391		− lea esi, [esi+8]
392		− lea edi, [edi+4]
	421	+ lea r2, [r2+8]
	422	+ lea r0, [r0+4]
393	423
394		− dec eax
395		− jg near .xloops
	424	+ dec r4
	425	+ jg near .xloops3
396	426
397	427	; next line
398		− lea esi, [esi+2*ecx] ; next end of lines
399		− lea esi, [esi+2ebx] ; reset to base 0 [- 2 iDstWidth]
400		− lea edi, [edi+edx]
401		− lea edi, [edi+ebx] ; reset to base 0 [- iDstWidth]
	428	+ lea r2, [r2+2*r3] ; next end of lines
	429	+ lea r2, [r2+2r6] ; reset to base 0 [- 2 iDstWidth]
	430	+ lea r0, [r0+r1]
	431	+ lea r0, [r0+r6] ; reset to base 0 [- iDstWidth]
402	432
403		− dec ebp
404		− jg near .yloops
	433	+ dec r5
	434	+ jg near .yloops3
405	435
406	436	WELSEMMS
407		− pop ebp
408		− pop edi
409		− pop esi
410		− pop edx
411		− pop ebx
	437	+%ifndef X86_32
	438	+ pop r12
	439	+%endif
	440	+ LOAD_6_PARA_POP
	441	+%ifdef X86_32
	442	+ pop r6
	443	+%endif
412	444	ret
413	445
414	446
@@ -420,31 +452,56 @@ WELS_EXTERN DyadicBilinearDownsamplerWidthx8_sse · 5 unmodified lines
420	452	; const int iSrcWidth, const int iSrcHeight );
421	453	;***********************************************************************
422	454	WELS_EXTERN DyadicBilinearDownsamplerWidthx32_ssse3
423		− push ebx
424		− push edx
425		− push esi
426		− push edi
427		− push ebp
428		−
429		− mov edi, [esp+24] ; pDst
430		− mov edx, [esp+28] ; iDstStride
431		− mov esi, [esp+32] ; pSrc
432		− mov ecx, [esp+36] ; iSrcStride
433		− mov ebp, [esp+44] ; iSrcHeight
434		−
435		− sar ebp, $01 ; iSrcHeight >> 1
	455	+ ;push ebx
	456	+ ;push edx
	457	+ ;push esi
	458	+ ;push edi
	459	+ ;push ebp
	460	+
	461	+ ;mov edi, [esp+24] ; pDst
	462	+ ;mov edx, [esp+28] ; iDstStride
	463	+ ;mov esi, [esp+32] ; pSrc
	464	+ ;mov ecx, [esp+36] ; iSrcStride
	465	+ ;mov ebp, [esp+44] ; iSrcHeight
	466	+%ifdef X86_32
	467	+ push r6
	468	+ %assign push_num 1
	469	+%else
	470	+ %assign push_num 0
	471	+%endif
	472	+ LOAD_6_PARA
	473	+ PUSH_XMM 8
	474	+ SIGN_EXTENSION r1, r1d
	475	+ SIGN_EXTENSION r3, r3d
	476	+ SIGN_EXTENSION r4, r4d
	477	+ SIGN_EXTENSION r5, r5d
	478	+
	479	+%ifndef X86_32
	480	+ push r12
	481	+ mov r12, r4
	482	+%endif
	483	+ sar r5, $01 ; iSrcHeight >> 1
436	484
437	485	movdqa xmm7, [shufb_mask_low] ; mask low
438	486	movdqa xmm6, [shufb_mask_high] ; mask high
439	487
440		−.yloops:
441		− mov eax, [esp+40] ; iSrcWidth
442		− sar eax, $01 ; iSrcWidth >> 1
443		− mov ebx, eax ; iDstWidth restored at ebx
444		− sar eax, $04 ; (iSrcWidth >> 1) / 16 ; loop count = num_of_mb
445		− neg ebx ; - (iSrcWidth >> 1)
	488	+.yloops4:
	489	+ ;mov eax, [esp+40] ; iSrcWidth
	490	+ ;sar eax, $01 ; iSrcWidth >> 1
	491	+ ;mov ebx, eax ; iDstWidth restored at ebx
	492	+ ;sar eax, $04 ; (iSrcWidth >> 1) / 16 ; loop count = num_of_mb
	493	+ ;neg ebx ; - (iSrcWidth >> 1)
	494	+%ifdef X86_32
	495	+ mov r4, arg5
	496	+%else
	497	+ mov r4, r12
	498	+%endif
	499	+ sar r4, $01 ; iSrcWidth >> 1
	500	+ mov r6, r4 ; iDstWidth restored at ebx
	501	+ sar r4, $04 ; (iSrcWidth >> 1) / 16 ; loop count = num_of_mb
	502	+ neg r6 ; - (iSrcWidth >> 1)
446	503	; each loop = source bandwidth: 32 bytes
447		−.xloops:
	504	+.xloops4:
448	505	; 1st part horizonal loop: x16 bytes
449	506	; mem hi<- ->lo
450	507	;1st Line Src: xmm0: h H g G f F e E d D c C b B a A
@@ -458,10 +515,10 @@ WELS_EXTERN DyadicBilinearDownsamplerWidthx32_ssse3 · 7 unmodified lines
458	515	;: p .. a
459	516
460	517	;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
461		− movdqa xmm0, [esi] ; 1st_src_line
462		− movdqa xmm1, [esi+16] ; 1st_src_line + 16
463		− movdqa xmm2, [esi+ecx] ; 2nd_src_line
464		− movdqa xmm3, [esi+ecx+16] ; 2nd_src_line + 16
	518	+ movdqa xmm0, [r2] ; 1st_src_line
	519	+ movdqa xmm1, [r2+16] ; 1st_src_line + 16
	520	+ movdqa xmm2, [r2+r3] ; 2nd_src_line
	521	+ movdqa xmm3, [r2+r3+16] ; 2nd_src_line + 16
465	522
466	523	; packing & avg
467	524	movdqa xmm4, xmm0 ; h H g G f F e E d D c C b B a A
@@ -498,29 +555,33 @@ WELS_EXTERN DyadicBilinearDownsamplerWidthx32_ssse3 · 30 unmodified lines
498	555	pavgb xmm0, xmm2
499	556
500	557	; write pDst
501		− movdqa [edi], xmm0
	558	+ movdqa [r0], xmm0
502	559
503	560	; next SMB
504		− lea esi, [esi+32]
505		− lea edi, [edi+16]
	561	+ lea r2, [r2+32]
	562	+ lea r0, [r0+16]
506	563
507		− dec eax
508		− jg near .xloops
	564	+ dec r4
	565	+ jg near .xloops4
509	566
510	567	; next line
511		− lea esi, [esi+2*ecx] ; next end of lines
512		− lea esi, [esi+2ebx] ; reset to base 0 [- 2 iDstWidth]
513		− lea edi, [edi+edx]
514		− lea edi, [edi+ebx] ; reset to base 0 [- iDstWidth]
515		−
516		− dec ebp
517		− jg near .yloops
518		−
519		− pop ebp
520		− pop edi
521		− pop esi
522		− pop edx
523		− pop ebx
	568	+ lea r2, [r2+2*r3] ; next end of lines
	569	+ lea r2, [r2+2r6] ; reset to base 0 [- 2 iDstWidth]
	570	+ lea r0, [r0+r1]
	571	+ lea r0, [r0+r6] ; reset to base 0 [- iDstWidth]
	572	+
	573	+ dec r5
	574	+ jg near .yloops4
	575	+
	576	+%ifndef X86_32
	577	+ pop r12
	578	+%endif
	579	+
	580	+ POP_XMM
	581	+ LOAD_6_PARA_POP
	582	+%ifdef X86_32
	583	+ pop r6
	584	+%endif
524	585	ret
525	586
526	587	;***********************************************************************
@@ -529,30 +590,39 @@ WELS_EXTERN DyadicBilinearDownsamplerWidthx32_ssse3 · 2 unmodified lines
529	590	; const int iSrcWidth, const int iSrcHeight );
530	591	;***********************************************************************
531	592	WELS_EXTERN DyadicBilinearDownsamplerWidthx16_ssse3
532		− push ebx
533		− push edx
534		− push esi
535		− push edi
536		− push ebp
537		−
538		− mov edi, [esp+24] ; pDst
539		− mov edx, [esp+28] ; iDstStride
540		− mov esi, [esp+32] ; pSrc
541		− mov ecx, [esp+36] ; iSrcStride
542		− mov ebp, [esp+44] ; iSrcHeight
543		−
544		− sar ebp, $01 ; iSrcHeight >> 1
545		− movdqa xmm7, [shufb_mask_low] ; mask low
546		− movdqa xmm6, [shufb_mask_high] ; mask high
	593	+%ifdef X86_32
	594	+ push r6
	595	+ %assign push_num 1
	596	+%else
	597	+ %assign push_num 0
	598	+%endif
	599	+ LOAD_6_PARA
	600	+ PUSH_XMM 6
	601	+ SIGN_EXTENSION r1, r1d
	602	+ SIGN_EXTENSION r3, r3d
	603	+ SIGN_EXTENSION r4, r4d
	604	+ SIGN_EXTENSION r5, r5d
	605	+
	606	+%ifndef X86_32
	607	+ push r12
	608	+ mov r12, r4
	609	+%endif
	610	+ sar r5, $01 ; iSrcHeight >> 1
	611	+ movdqa xmm5, [shufb_mask_low] ; mask low
	612	+ movdqa xmm4, [shufb_mask_high] ; mask high
547	613
548		−.yloops:
549		− mov eax, [esp+40] ; iSrcWidth
550		− sar eax, $01 ; iSrcWidth >> 1
551		− mov ebx, eax ; iDstWidth restored at ebx
552		− sar eax, $03 ; (iSrcWidth >> 1) / 8 ; loop count = num_of_mb
553		− neg ebx ; - (iSrcWidth >> 1)
	614	+.yloops5:
	615	+%ifdef X86_32
	616	+ mov r4, arg5
	617	+%else
	618	+ mov r4, r12
	619	+%endif
	620	+ sar r4, $01 ; iSrcWidth >> 1
	621	+ mov r6, r4 ; iDstWidth restored at ebx
	622	+ sar r4, $03 ; (iSrcWidth >> 1) / 8 ; loop count = num_of_mb
	623	+ neg r6 ; - (iSrcWidth >> 1)
554	624	; each loop = source bandwidth: 16 bytes
555		−.xloops:
	625	+.xloops5:
556	626	; horizonal loop: x16 bytes by source
557	627	; mem hi<- ->lo
558	628	;1st line pSrc: xmm0: h H g G f F e E d D c C b B a A
@@ -562,21 +632,21 @@ WELS_EXTERN DyadicBilinearDownsamplerWidthx16_ssse3 · 3 unmodified lines
562	632	;: h g f e d c b a, p o n m l k j i
563	633
564	634	;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
565		− movdqa xmm0, [esi] ; 1st_src_line
566		− movdqa xmm1, [esi+ecx] ; 2nd_src_line
	635	+ movdqa xmm0, [r2] ; 1st_src_line
	636	+ movdqa xmm1, [r2+r3] ; 2nd_src_line
567	637
568	638	; packing & avg
569	639	movdqa xmm2, xmm0 ; h H g G f F e E d D c C b B a A
570		− pshufb xmm0, xmm7 ; 0 H 0 G 0 F 0 E 0 D 0 C 0 B 0 A
571		− pshufb xmm2, xmm6 ; 0 h 0 g 0 f 0 e 0 d 0 c 0 b 0 a
	640	+ pshufb xmm0, xmm5 ; 0 H 0 G 0 F 0 E 0 D 0 C 0 B 0 A
	641	+ pshufb xmm2, xmm4 ; 0 h 0 g 0 f 0 e 0 d 0 c 0 b 0 a
572	642	; another implementation for xmm2 high bits
573	643	; psubb xmm2, xmm0 ; h 0 g 0 f 0 e 0 d 0 c 0 b 0 a 0
574	644	; psrlw xmm2, 8 ; 0 h 0 g 0 f 0 e 0 d 0 c 0 b 0 a
575	645	pavgb xmm0, xmm2
576	646
577	647	movdqa xmm3, xmm1
578		− pshufb xmm1, xmm7
579		− pshufb xmm3, xmm6
	648	+ pshufb xmm1, xmm5
	649	+ pshufb xmm3, xmm4
580	650	; psubb xmm3, xmm1
581	651	; psrlw xmm3, 8
582	652	pavgb xmm1, xmm3
@@ -585,29 +655,32 @@ WELS_EXTERN DyadicBilinearDownsamplerWidthx16_ssse3 · 2 unmodified lines
585	655	packuswb xmm0, xmm1
586	656
587	657	; write pDst
588		− movq [edi], xmm0
	658	+ movq [r0], xmm0
589	659
590	660	; next SMB
591		− lea esi, [esi+16]
592		− lea edi, [edi+8]
	661	+ lea r2, [r2+16]
	662	+ lea r0, [r0+8]
593	663
594		− dec eax
595		− jg near .xloops
	664	+ dec r4
	665	+ jg near .xloops5
596	666
597		− ; next line
598		− lea esi, [esi+2*ecx] ; next end of lines
599		− lea esi, [esi+2ebx] ; reset to base 0 [- 2 iDstWidth]
600		− lea edi, [edi+edx]
601		− lea edi, [edi+ebx] ; reset to base 0 [- iDstWidth]
602		−
603		− dec ebp
604		− jg near .yloops
605		−
606		− pop ebp
607		− pop edi
608		− pop esi
609		− pop edx
610		− pop ebx
	667	+ lea r2, [r2+2*r3] ; next end of lines
	668	+ lea r2, [r2+2r6] ; reset to base 0 [- 2 iDstWidth]
	669	+ lea r0, [r0+r1]
	670	+ lea r0, [r0+r6] ; reset to base 0 [- iDstWidth]
	671	+
	672	+ dec r5
	673	+ jg near .yloops5
	674	+
	675	+%ifndef X86_32
	676	+ pop r12
	677	+%endif
	678	+
	679	+ POP_XMM
	680	+ LOAD_6_PARA_POP
	681	+%ifdef X86_32
	682	+ pop r6
	683	+%endif
611	684	ret
612	685
613	686	; got about 65% improvement over DyadicBilinearDownsamplerWidthx32_sse
@@ -617,31 +690,40 @@ WELS_EXTERN DyadicBilinearDownsamplerWidthx16_ssse3 · 3 unmodified lines
617	690	; const int iSrcWidth, const int iSrcHeight );
618	691	;***********************************************************************
619	692	WELS_EXTERN DyadicBilinearDownsamplerWidthx32_sse4
620		− push ebx
621		− push edx
622		− push esi
623		− push edi
624		− push ebp
625		−
626		− mov edi, [esp+24] ; pDst
627		− mov edx, [esp+28] ; iDstStride
628		− mov esi, [esp+32] ; pSrc
629		− mov ecx, [esp+36] ; iSrcStride
630		− mov ebp, [esp+44] ; iSrcHeight
631		−
632		− sar ebp, $01 ; iSrcHeight >> 1
	693	+%ifdef X86_32
	694	+ push r6
	695	+ %assign push_num 1
	696	+%else
	697	+ %assign push_num 0
	698	+%endif
	699	+ LOAD_6_PARA
	700	+ PUSH_XMM 8
	701	+ SIGN_EXTENSION r1, r1d
	702	+ SIGN_EXTENSION r3, r3d
	703	+ SIGN_EXTENSION r4, r4d
	704	+ SIGN_EXTENSION r5, r5d
	705	+
	706	+%ifndef X86_32
	707	+ push r12
	708	+ mov r12, r4
	709	+%endif
	710	+ sar r5, $01 ; iSrcHeight >> 1
633	711
634	712	movdqa xmm7, [shufb_mask_low] ; mask low
635	713	movdqa xmm6, [shufb_mask_high] ; mask high
636	714
637		−.yloops:
638		− mov eax, [esp+40] ; iSrcWidth
639		− sar eax, $01 ; iSrcWidth >> 1
640		− mov ebx, eax ; iDstWidth restored at ebx
641		− sar eax, $04 ; (iSrcWidth >> 1) / 16 ; loop count = num_of_mb
642		− neg ebx ; - (iSrcWidth >> 1)
	715	+.yloops6:
	716	+%ifdef X86_32
	717	+ mov r4, arg5
	718	+%else
	719	+ mov r4, r12
	720	+%endif
	721	+ sar r4, $01 ; iSrcWidth >> 1
	722	+ mov r6, r4 ; iDstWidth restored at ebx
	723	+ sar r4, $04 ; (iSrcWidth >> 1) / 16 ; loop count = num_of_mb
	724	+ neg r6 ; - (iSrcWidth >> 1)
643	725	; each loop = source bandwidth: 32 bytes
644		−.xloops:
	726	+.xloops6:
645	727	; 1st part horizonal loop: x16 bytes
646	728	; mem hi<- ->lo
647	729	;1st Line Src: xmm0: h H g G f F e E d D c C b B a A
@@ -655,10 +737,10 @@ WELS_EXTERN DyadicBilinearDownsamplerWidthx32_sse4 · 7 unmodified lines
655	737	;: p .. a
656	738
657	739	;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
658		− movntdqa xmm0, [esi] ; 1st_src_line
659		− movntdqa xmm1, [esi+16] ; 1st_src_line + 16
660		− movntdqa xmm2, [esi+ecx] ; 2nd_src_line
661		− movntdqa xmm3, [esi+ecx+16] ; 2nd_src_line + 16
	740	+ movntdqa xmm0, [r2] ; 1st_src_line
	741	+ movntdqa xmm1, [r2+16] ; 1st_src_line + 16
	742	+ movntdqa xmm2, [r2+r3] ; 2nd_src_line
	743	+ movntdqa xmm3, [r2+r3+16] ; 2nd_src_line + 16
662	744
663	745	; packing & avg
664	746	movdqa xmm4, xmm0 ; h H g G f F e E d D c C b B a A
@@ -694,29 +776,32 @@ WELS_EXTERN DyadicBilinearDownsamplerWidthx32_sse4 · 29 unmodified lines
694	776	pavgb xmm0, xmm2
695	777
696	778	; write pDst
697		− movdqa [edi], xmm0
	779	+ movdqa [r0], xmm0
698	780
699	781	; next SMB
700		− lea esi, [esi+32]
701		− lea edi, [edi+16]
	782	+ lea r2, [r2+32]
	783	+ lea r0, [r0+16]
702	784
703		− dec eax
704		− jg near .xloops
	785	+ dec r4
	786	+ jg near .xloops6
705	787
706		− ; next line
707		− lea esi, [esi+2*ecx] ; next end of lines
708		− lea esi, [esi+2ebx] ; reset to base 0 [- 2 iDstWidth]
709		− lea edi, [edi+edx]
710		− lea edi, [edi+ebx] ; reset to base 0 [- iDstWidth]
711		−
712		− dec ebp
713		− jg near .yloops
714		−
715		− pop ebp
716		− pop edi
717		− pop esi
718		− pop edx
719		− pop ebx
	788	+ lea r2, [r2+2*r3] ; next end of lines
	789	+ lea r2, [r2+2r6] ; reset to base 0 [- 2 iDstWidth]
	790	+ lea r0, [r0+r1]
	791	+ lea r0, [r0+r6] ; reset to base 0 [- iDstWidth]
	792	+
	793	+ dec r5
	794	+ jg near .yloops6
	795	+
	796	+%ifndef X86_32
	797	+ pop r12
	798	+%endif
	799	+
	800	+ POP_XMM
	801	+ LOAD_6_PARA_POP
	802	+%ifdef X86_32
	803	+ pop r6
	804	+%endif
720	805	ret
721	806
722	807	;***********************************************************************
@@ -725,30 +810,39 @@ WELS_EXTERN DyadicBilinearDownsamplerWidthx32_sse4 · 2 unmodified lines
725	810	; const int iSrcWidth, const int iSrcHeight );
726	811	;***********************************************************************
727	812	WELS_EXTERN DyadicBilinearDownsamplerWidthx16_sse4
728		− push ebx
729		− push edx
730		− push esi
731		− push edi
732		− push ebp
733		−
734		− mov edi, [esp+24] ; pDst
735		− mov edx, [esp+28] ; iDstStride
736		− mov esi, [esp+32] ; pSrc
737		− mov ecx, [esp+36] ; iSrcStride
738		− mov ebp, [esp+44] ; iSrcHeight
739		−
740		− sar ebp, $01 ; iSrcHeight >> 1
741		− movdqa xmm7, [shufb_mask_low] ; mask low
742		− movdqa xmm6, [shufb_mask_high] ; mask high
	813	+%ifdef X86_32
	814	+ push r6
	815	+ %assign push_num 1
	816	+%else
	817	+ %assign push_num 0
	818	+%endif
	819	+ LOAD_6_PARA
	820	+ PUSH_XMM 6
	821	+ SIGN_EXTENSION r1, r1d
	822	+ SIGN_EXTENSION r3, r3d
	823	+ SIGN_EXTENSION r4, r4d
	824	+ SIGN_EXTENSION r5, r5d
	825	+
	826	+%ifndef X86_32
	827	+ push r12
	828	+ mov r12, r4
	829	+%endif
	830	+ sar r5, $01 ; iSrcHeight >> 1
	831	+ movdqa xmm5, [shufb_mask_low] ; mask low
	832	+ movdqa xmm4, [shufb_mask_high] ; mask high
743	833
744		−.yloops:
745		− mov eax, [esp+40] ; iSrcWidth
746		− sar eax, $01 ; iSrcWidth >> 1
747		− mov ebx, eax ; iDstWidth restored at ebx
748		− sar eax, $03 ; (iSrcWidth >> 1) / 8 ; loop count = num_of_mb
749		− neg ebx ; - (iSrcWidth >> 1)
	834	+.yloops7:
	835	+%ifdef X86_32
	836	+ mov r4, arg5
	837	+%else
	838	+ mov r4, r12
	839	+%endif
	840	+ sar r4, $01 ; iSrcWidth >> 1
	841	+ mov r6, r4 ; iDstWidth restored at ebx
	842	+ sar r4, $03 ; (iSrcWidth >> 1) / 8 ; loop count = num_of_mb
	843	+ neg r6 ; - (iSrcWidth >> 1)
750	844	; each loop = source bandwidth: 16 bytes
751		−.xloops:
	845	+.xloops7:
752	846	; horizonal loop: x16 bytes by source
753	847	; mem hi<- ->lo
754	848	;1st line pSrc: xmm0: h H g G f F e E d D c C b B a A
@@ -758,20 +852,20 @@ WELS_EXTERN DyadicBilinearDownsamplerWidthx16_sse4 · 3 unmodified lines
758	852	;: h g f e d c b a, p o n m l k j i
759	853
760	854	;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
761		− movntdqa xmm0, [esi] ; 1st_src_line
762		− movntdqa xmm1, [esi+ecx] ; 2nd_src_line
	855	+ movntdqa xmm0, [r2] ; 1st_src_line
	856	+ movntdqa xmm1, [r2+r3] ; 2nd_src_line
763	857
764	858	; packing & avg
765	859	movdqa xmm2, xmm0 ; h H g G f F e E d D c C b B a A
766		− pshufb xmm0, xmm7 ; 0 H 0 G 0 F 0 E 0 D 0 C 0 B 0 A
767		− pshufb xmm2, xmm6 ; 0 h 0 g 0 f 0 e 0 d 0 c 0 b 0 a
	860	+ pshufb xmm0, xmm5 ; 0 H 0 G 0 F 0 E 0 D 0 C 0 B 0 A
	861	+ pshufb xmm2, xmm4 ; 0 h 0 g 0 f 0 e 0 d 0 c 0 b 0 a
768	862	; psubb xmm2, xmm0 ; h 0 g 0 f 0 e 0 d 0 c 0 b 0 a 0
769	863	; psrlw xmm2, 8 ; 0 h 0 g 0 f 0 e 0 d 0 c 0 b 0 a
770	864	pavgb xmm0, xmm2
771	865
772	866	movdqa xmm3, xmm1
773		− pshufb xmm1, xmm7
774		− pshufb xmm3, xmm6
	867	+ pshufb xmm1, xmm5
	868	+ pshufb xmm3, xmm4
775	869	; psubb xmm3, xmm1
776	870	; psrlw xmm3, 8
777	871	pavgb xmm1, xmm3
@@ -780,38 +874,40 @@ WELS_EXTERN DyadicBilinearDownsamplerWidthx16_sse4 · 2 unmodified lines
780	874	packuswb xmm0, xmm1
781	875
782	876	; write pDst
783		− movq [edi], xmm0
	877	+ movq [r0], xmm0
784	878
785	879	; next SMB
786		− lea esi, [esi+16]
787		− lea edi, [edi+8]
	880	+ lea r2, [r2+16]
	881	+ lea r0, [r0+8]
788	882
789		− dec eax
790		− jg near .xloops
	883	+ dec r4
	884	+ jg near .xloops7
791	885
792	886	; next line
793		− lea esi, [esi+2*ecx] ; next end of lines
794		− lea esi, [esi+2ebx] ; reset to base 0 [- 2 iDstWidth]
795		− lea edi, [edi+edx]
796		− lea edi, [edi+ebx] ; reset to base 0 [- iDstWidth]
797		−
798		− dec ebp
799		− jg near .yloops
800		−
801		− pop ebp
802		− pop edi
803		− pop esi
804		− pop edx
805		− pop ebx
806		− ret
	887	+ lea r2, [r2+2*r3] ; next end of lines
	888	+ lea r2, [r2+2r6] ; reset to base 0 [- 2 iDstWidth]
	889	+ lea r0, [r0+r1]
	890	+ lea r0, [r0+r6] ; reset to base 0 [- iDstWidth]
807	891
	892	+ dec r5
	893	+ jg near .yloops7
808	894
	895	+%ifndef X86_32
	896	+ pop r12
	897	+%endif
809	898
	899	+ POP_XMM
	900	+ LOAD_6_PARA_POP
	901	+%ifdef X86_32
	902	+ pop r6
	903	+%endif
	904	+ ret
810	905
811	906
	907	+%ifdef X86_32
812	908	;**************************************************************************************************************
813	909	;int GeneralBilinearAccurateDownsampler_sse2( unsigned char* pDst, const int iDstStride, const int iDstWidth, const int iDstHeight,
814		−; unsigned char* pSrc, const int iSrcStride, const int iSrcWidth, const int iSrcHeight,
	910	+; unsigned char* pSrc, const int iSrcStride,
815	911	; unsigned int uiScaleX, unsigned int uiScaleY );
816	912	;{
817	913	;**************************************************************************************************************
@@ -822,26 +918,22 @@ WELS_EXTERN GeneralBilinearAccurateDownsampler_sse2 · 4 unmodified lines
822	918	push edi
823	919	push ebx
824	920	%define pushsize 16
825		−%define localsize 28
	921	+%define localsize 16
826	922	%define pDstData esp + pushsize + localsize + 4
827	923	%define dwDstStride esp + pushsize + localsize + 8
828	924	%define dwDstWidth esp + pushsize + localsize + 12
829	925	%define dwDstHeight esp + pushsize + localsize + 16
830	926	%define pSrcData esp + pushsize + localsize + 20
831	927	%define dwSrcStride esp + pushsize + localsize + 24
832		−%define dwSrcWidth esp + pushsize + localsize + 28
833		−%define dwSrcHeight esp + pushsize + localsize + 32
834		−%define scale esp + 0
835		−%define uiScaleX esp + pushsize + localsize + 36
836		−%define uiScaleY esp + pushsize + localsize + 40
837		−%define tmpHeight esp + 12
838		−%define yInverse esp + 16
839		−%define xInverse esp + 20
840		−%define dstStep esp + 24
	928	+%define uiScaleX esp + pushsize + localsize + 28
	929	+%define uiScaleY esp + pushsize + localsize + 32
	930	+%define tmpHeight esp + 0
	931	+%define yInverse esp + 4
	932	+%define xInverse esp + 8
	933	+%define dstStep esp + 12
841	934	sub esp, localsize
842	935
843	936	pxor xmm0, xmm0
844		− mov edx, 32767
845	937	mov eax, [uiScaleX]
846	938	and eax, 32767
847	939	mov ebx, eax
@@ -999,7 +1091,6 @@ LAST_ROW_END: · 151 unmodified lines
999	1091	%undef dwDstWidth
1000	1092	%undef dwDstHeight
1001	1093	%undef dwDstStride
1002		−%undef scale
1003	1094	%undef uiScaleX
1004	1095	%undef uiScaleY
1005	1096	%undef tmpHeight
@@ -1013,7 +1104,7 @@ LAST_ROW_END: · 7 unmodified lines
1013	1104
1014	1105	;**************************************************************************************************************
1015	1106	;int GeneralBilinearFastDownsampler_sse2( unsigned char* pDst, const int iDstStride, const int iDstWidth, const int iDstHeight,
1016		−; unsigned char* pSrc, const int iSrcStride, const int iSrcWidth, const int iSrcHeight,
	1107	+; unsigned char* pSrc, const int iSrcStride,
1017	1108	; unsigned int uiScaleX, unsigned int uiScaleY );
1018	1109	;{
1019	1110	;**************************************************************************************************************
@@ -1024,22 +1115,19 @@ WELS_EXTERN GeneralBilinearFastDownsampler_sse2 · 4 unmodified lines
1024	1115	push edi
1025	1116	push ebx
1026	1117	%define pushsize 16
1027		−%define localsize 28
	1118	+%define localsize 16
1028	1119	%define pDstData esp + pushsize + localsize + 4
1029	1120	%define dwDstStride esp + pushsize + localsize + 8
1030	1121	%define dwDstWidth esp + pushsize + localsize + 12
1031	1122	%define dwDstHeight esp + pushsize + localsize + 16
1032	1123	%define pSrcData esp + pushsize + localsize + 20
1033	1124	%define dwSrcStride esp + pushsize + localsize + 24
1034		−%define dwSrcWidth esp + pushsize + localsize + 28
1035		−%define dwSrcHeight esp + pushsize + localsize + 32
1036		−%define scale esp + 0
1037		−%define uiScaleX esp + pushsize + localsize + 36
1038		−%define uiScaleY esp + pushsize + localsize + 40
1039		−%define tmpHeight esp + 12
1040		−%define yInverse esp + 16
1041		−%define xInverse esp + 20
1042		−%define dstStep esp + 24
	1125	+%define uiScaleX esp + pushsize + localsize + 28
	1126	+%define uiScaleY esp + pushsize + localsize + 32
	1127	+%define tmpHeight esp + 0
	1128	+%define yInverse esp + 4
	1129	+%define xInverse esp + 8
	1130	+%define dstStep esp + 12
1043	1131	sub esp, localsize
1044	1132
1045	1133	pxor xmm0, xmm0
@@ -1191,10 +1279,7 @@ FAST_LAST_ROW_END: · 145 unmodified lines
1191	1279	%undef dwSrcHeight
1192	1280	%undef dwSrcStride
1193	1281	%undef pDstData
1194		−%undef dwDstWidth
1195		−%undef dwDstHeight
1196	1282	%undef dwDstStride
1197		−%undef scale
1198	1283	%undef uiScaleX
1199	1284	%undef uiScaleY
1200	1285	%undef tmpHeight
@@ -1202,4 +1287,612 @@ FAST_LAST_ROW_END: · 1 unmodified line
1202	1287	%undef xInverse
1203	1288	%undef dstStep
1204	1289	ret
	1290	+
	1291	+%elifdef WIN64
	1292	+
	1293	+;**************************************************************************************************************
	1294	+;int GeneralBilinearAccurateDownsampler_sse2( unsigned char* pDst, const int iDstStride, const int iDstWidth, const int iDstHeight,
	1295	+; unsigned char* pSrc, const int iSrcStride,
	1296	+; unsigned int uiScaleX, unsigned int uiScaleY );
	1297	+;{
	1298	+;**************************************************************************************************************
	1299	+
	1300	+WELS_EXTERN GeneralBilinearAccurateDownsampler_sse2
	1301	+ push r12
	1302	+ push r13
	1303	+ push r14
	1304	+ push r15
	1305	+ push rsi
	1306	+ push rdi
	1307	+ push rbx
	1308	+ push rbp
	1309	+ %assign push_num 8
	1310	+ LOAD_7_PARA
	1311	+ PUSH_XMM 8
	1312	+ SIGN_EXTENSION r1, r1d
	1313	+ SIGN_EXTENSION r2, r2d
	1314	+ SIGN_EXTENSION r3, r3d
	1315	+ SIGN_EXTENSION r5, r5d
	1316	+ SIGN_EXTENSION r6, r6d
	1317	+
	1318	+ pxor xmm0, xmm0
	1319	+ mov r12d, r6d
	1320	+ and r12d, 32767
	1321	+ mov r13d, r12d
	1322	+ neg r13d
	1323	+ and r13d, 32767
	1324	+ movd xmm1, r12d ; uinc(uiScaleX mod 32767)
	1325	+ movd xmm2, r13d ; -uinc
	1326	+ psllq xmm1, 32
	1327	+ por xmm1, xmm2 ; 0 0 uinc -uinc (dword)
	1328	+ pshufd xmm7, xmm1, 01000100b ; xmm7: uinc -uinc uinc -uinc
	1329	+
	1330	+ mov r12, arg8
	1331	+ SIGN_EXTENSION r12, r12d
	1332	+ mov rbp, r12
	1333	+ and r12d, 32767
	1334	+ mov r13d, r12d
	1335	+ neg r13d
	1336	+ and r13d, 32767
	1337	+ movd xmm6, r12d ; vinc(uiScaleY mod 32767)
	1338	+ movd xmm2, r13d ; -vinc
	1339	+ psllq xmm6, 32
	1340	+ por xmm6, xmm2 ; 0 0 vinc -vinc (dword)
	1341	+ pshufd xmm6, xmm6, 01010000b ; xmm6: vinc vinc -vinc -vinc
	1342	+
	1343	+ mov r12d, 40003fffh
	1344	+ movd xmm5, r12d
	1345	+ punpcklwd xmm5, xmm0 ; 16384 16383
	1346	+ pshufd xmm5, xmm5, 01000100b ; xmm5: 16384 16383 16384 16383
	1347	+
	1348	+DOWNSAMPLE:
	1349	+ sub r1, r2 ; stride - width
	1350	+ dec r3
	1351	+ mov r14,16384
	1352	+ pshufd xmm4, xmm5, 01010000b ; initial v to 16384 16384 16383 16383
	1353	+
	1354	+HEIGHT:
	1355	+ ;mov r12, r4
	1356	+ mov r12, r14
	1357	+ shr r12, 15
	1358	+ imul r12, r5
	1359	+ add r12, r4 ; get current row address
	1360	+ mov r13, r12
	1361	+ add r13, r5
	1362	+
	1363	+ mov r15, 16384
	1364	+ mov rsi, r2
	1365	+ dec rsi
	1366	+ movdqa xmm3, xmm5 ; initial u to 16384 16383 16384 16383
	1367	+
	1368	+WIDTH:
	1369	+ mov rdi, r15
	1370	+ shr rdi, 15
	1371	+
	1372	+ movd xmm1, [r12+rdi] ; xxxxxxba
	1373	+ movd xmm2, [r13+rdi] ; xxxxxxdc
	1374	+ pxor xmm0, xmm0
	1375	+ punpcklwd xmm1, xmm2 ; xxxxdcba
	1376	+ punpcklbw xmm1, xmm0 ; 0d0c0b0a
	1377	+ punpcklwd xmm1, xmm0 ; 000d000c000b000a
	1378	+
	1379	+ movdqa xmm2, xmm4 ; xmm2: vv(1-v)(1-v) tmpv
	1380	+ pmaddwd xmm2, xmm3 ; mul u(1-u)u(1-u) on xmm2
	1381	+ movdqa xmm0, xmm2
	1382	+ pmuludq xmm2, xmm1
	1383	+ psrlq xmm0, 32
	1384	+ psrlq xmm1, 32
	1385	+ pmuludq xmm0, xmm1
	1386	+ paddq xmm2, xmm0
	1387	+ pshufd xmm1, xmm2, 00001110b
	1388	+ paddq xmm2, xmm1
	1389	+ psrlq xmm2, 29
	1390	+
	1391	+ movd ebx, xmm2
	1392	+ inc ebx
	1393	+ shr ebx, 1
	1394	+ mov [r0], bl
	1395	+ inc r0
	1396	+
	1397	+ add r15, r6
	1398	+ paddw xmm3, xmm7 ; inc u
	1399	+ psllw xmm3, 1
	1400	+ psrlw xmm3, 1
	1401	+
	1402	+ dec rsi
	1403	+ jg WIDTH
	1404	+
	1405	+WIDTH_END:
	1406	+ shr r15, 15
	1407	+ mov bl, [r12+r15]
	1408	+ mov [r0],bl
	1409	+ inc r0
	1410	+ add r14, rbp
	1411	+ add r0, r1
	1412	+
	1413	+ paddw xmm4, xmm6 ; inc v
	1414	+ psllw xmm4, 1
	1415	+ psrlw xmm4, 1
	1416	+
	1417	+ dec r3
	1418	+ jg HEIGHT
	1419	+
	1420	+LAST_ROW:
	1421	+ shr r14, 15
	1422	+ imul r14, r5
	1423	+ add r4, r14
	1424	+ mov r15, 16384
	1425	+
	1426	+LAST_ROW_WIDTH:
	1427	+ mov rdi, r15
	1428	+ shr rdi, 15
	1429	+ mov bl, [r4+rdi]
	1430	+ mov [r0],bl
	1431	+ inc r0
	1432	+
	1433	+ add r15, r6
	1434	+ dec r2
	1435	+ jg LAST_ROW_WIDTH
	1436	+
	1437	+LAST_ROW_END:
	1438	+
	1439	+ POP_XMM
	1440	+ pop rbp
	1441	+ pop rbx
	1442	+ pop rdi
	1443	+ pop rsi
	1444	+ pop r15
	1445	+ pop r14
	1446	+ pop r13
	1447	+ pop r12
	1448	+ ret
	1449	+
	1450	+;**************************************************************************************************************
	1451	+;int GeneralBilinearFastDownsampler_sse2( unsigned char* pDst, const int iDstStride, const int iDstWidth, const int iDstHeight,
	1452	+; unsigned char* pSrc, const int iSrcStride,
	1453	+; unsigned int uiScaleX, unsigned int uiScaleY );
	1454	+;{
	1455	+;**************************************************************************************************************
	1456	+
	1457	+WELS_EXTERN GeneralBilinearFastDownsampler_sse2
	1458	+ push r12
	1459	+ push r13
	1460	+ push r14
	1461	+ push r15
	1462	+ push rsi
	1463	+ push rdi
	1464	+ push rbx
	1465	+ push rbp
	1466	+ %assign push_num 8
	1467	+ LOAD_7_PARA
	1468	+ PUSH_XMM 8
	1469	+ SIGN_EXTENSION r1, r1d
	1470	+ SIGN_EXTENSION r2, r2d
	1471	+ SIGN_EXTENSION r3, r3d
	1472	+ SIGN_EXTENSION r5, r5d
	1473	+ SIGN_EXTENSION r6, r6d
	1474	+
	1475	+ pxor xmm0, xmm0
	1476	+ mov r12d, r6d
	1477	+ and r12d, 65535
	1478	+ mov r13d, r12d
	1479	+ neg r13d
	1480	+ and r13d, 65535
	1481	+ movd xmm1, r12d ; uinc(uiScaleX mod 65536)
	1482	+ movd xmm2, r13d ; -uinc
	1483	+ psllq xmm1, 32
	1484	+ por xmm1, xmm2 ; 0 uinc 0 -uinc
	1485	+ pshuflw xmm7, xmm1, 10001000b ; xmm7: uinc -uinc uinc -uinc
	1486	+
	1487	+ mov r12, arg8
	1488	+ SIGN_EXTENSION r12, r12d
	1489	+ mov rbp, r12
	1490	+ and r12d, 32767
	1491	+ mov r13d, r12d
	1492	+ neg r13d
	1493	+ and r13d, 32767
	1494	+ movd xmm6, r12d ; vinc(uiScaleY mod 32767)
	1495	+ movd xmm2, r13d ; -vinc
	1496	+ psllq xmm6, 32
	1497	+ por xmm6, xmm2 ; 0 vinc 0 -vinc
	1498	+ pshuflw xmm6, xmm6, 10100000b ; xmm6: vinc vinc -vinc -vinc
	1499	+
	1500	+ mov r12d, 80007fffh ; 32768 32767
	1501	+ movd xmm5, r12d
	1502	+ pshuflw xmm5, xmm5, 01000100b ; 32768 32767 32768 32767
	1503	+
	1504	+FAST_DOWNSAMPLE:
	1505	+ sub r1, r2 ; stride - width
	1506	+ dec r3
	1507	+ mov r14,16384
	1508	+
	1509	+ pshuflw xmm4, xmm5, 01010000b
	1510	+ psrlw xmm4, 1 ; initial v to 16384 16384 16383 16383
	1511	+
	1512	+FAST_HEIGHT:
	1513	+ mov r12, r14
	1514	+ shr r12, 15
	1515	+ imul r12, r5
	1516	+ add r12, r4 ; get current row address
	1517	+ mov r13, r12
	1518	+ add r13, r5
	1519	+
	1520	+ mov r15, 32768
	1521	+ mov rsi, r2
	1522	+ dec rsi
	1523	+
	1524	+ movdqa xmm3, xmm5 ; initial u to 32768 32767 32768 32767
	1525	+
	1526	+FAST_WIDTH:
	1527	+ mov rdi, r15
	1528	+ shr rdi, 16
	1529	+
	1530	+ movd xmm1, [r12+rdi] ; xxxxxxba
	1531	+ movd xmm2, [r13+rdi] ; xxxxxxdc
	1532	+ punpcklwd xmm1, xmm2 ; xxxxdcba
	1533	+ punpcklbw xmm1, xmm0 ; 0d0c0b0a
	1534	+
	1535	+ movdqa xmm2, xmm4 ; xmm2: vv(1-v)(1-v) tmpv
	1536	+ pmulhuw xmm2, xmm3 ; mul u(1-u)u(1-u) on xmm2
	1537	+ pmaddwd xmm2, xmm1
	1538	+ pshufd xmm1, xmm2, 00000001b
	1539	+ paddd xmm2, xmm1
	1540	+ movdqa xmm1, [add_extra_half]
	1541	+ paddd xmm2, xmm1
	1542	+ psrld xmm2, 15
	1543	+
	1544	+ packuswb xmm2, xmm0
	1545	+ movd ebx, xmm2
	1546	+ mov [r0], bl
	1547	+ inc r0
	1548	+
	1549	+ add r15, r6
	1550	+
	1551	+ paddw xmm3, xmm7 ; inc u
	1552	+ dec rsi
	1553	+ jg FAST_WIDTH
	1554	+
	1555	+FAST_WIDTH_END:
	1556	+ shr r15, 16
	1557	+ mov bl, [r12+r15]
	1558	+ mov [r0],bl
	1559	+ inc r0
	1560	+ add r14, rbp
	1561	+ add r0, r1
	1562	+
	1563	+ paddw xmm4, xmm6 ; inc v
	1564	+ psllw xmm4, 1
	1565	+ psrlw xmm4, 1
	1566	+
	1567	+ dec r3
	1568	+ jg FAST_HEIGHT
	1569	+
	1570	+
	1571	+FAST_LAST_ROW:
	1572	+ shr r14, 15
	1573	+ imul r14, r5
	1574	+ add r4, r14
	1575	+ mov r15, 32768
	1576	+
	1577	+FAST_LAST_ROW_WIDTH:
	1578	+ mov rdi, r15
	1579	+ shr rdi, 16
	1580	+ mov bl, [r4+rdi]
	1581	+ mov [r0],bl
	1582	+ inc r0
	1583	+
	1584	+ add r15, r6
	1585	+ dec r2
	1586	+ jg FAST_LAST_ROW_WIDTH
	1587	+
	1588	+FAST_LAST_ROW_END:
	1589	+
	1590	+ POP_XMM
	1591	+ pop rbp
	1592	+ pop rbx
	1593	+ pop rdi
	1594	+ pop rsi
	1595	+ pop r15
	1596	+ pop r14
	1597	+ pop r13
	1598	+ pop r12
	1599	+ ret
	1600	+
	1601	+%elifdef UNIX64
	1602	+
	1603	+;**************************************************************************************************************
	1604	+;int GeneralBilinearAccurateDownsampler_sse2( unsigned char* pDst, const int iDstStride, const int iDstWidth, const int iDstHeight,
	1605	+; unsigned char* pSrc, const int iSrcStride,
	1606	+; unsigned int uiScaleX, unsigned int uiScaleY );
	1607	+;{
	1608	+;**************************************************************************************************************
	1609	+
	1610	+WELS_EXTERN GeneralBilinearAccurateDownsampler_sse2
	1611	+ push r12
	1612	+ push r13
	1613	+ push r14
	1614	+ push r15
	1615	+ push rbx
	1616	+ push rbp
	1617	+ %assign push_num 6
	1618	+ LOAD_7_PARA
	1619	+ SIGN_EXTENSION r1, r1d
	1620	+ SIGN_EXTENSION r2, r2d
	1621	+ SIGN_EXTENSION r3, r3d
	1622	+ SIGN_EXTENSION r5, r5d
	1623	+ SIGN_EXTENSION r6, r6d
	1624	+
	1625	+ pxor xmm0, xmm0
	1626	+ mov r12d, r6d
	1627	+ and r12d, 32767
	1628	+ mov r13d, r12d
	1629	+ neg r13d
	1630	+ and r13d, 32767
	1631	+ movd xmm1, r12d ; uinc(uiScaleX mod 32767)
	1632	+ movd xmm2, r13d ; -uinc
	1633	+ psllq xmm1, 32
	1634	+ por xmm1, xmm2 ; 0 0 uinc -uinc (dword)
	1635	+ pshufd xmm7, xmm1, 01000100b ; xmm7: uinc -uinc uinc -uinc
	1636	+
	1637	+ mov r12, arg8
	1638	+ SIGN_EXTENSION r12, r12d
	1639	+ mov rbp, r12
	1640	+ and r12d, 32767
	1641	+ mov r13d, r12d
	1642	+ neg r13d
	1643	+ and r13d, 32767
	1644	+ movd xmm6, r12d ; vinc(uiScaleY mod 32767)
	1645	+ movd xmm2, r13d ; -vinc
	1646	+ psllq xmm6, 32
	1647	+ por xmm6, xmm2 ; 0 0 vinc -vinc (dword)
	1648	+ pshufd xmm6, xmm6, 01010000b ; xmm6: vinc vinc -vinc -vinc
	1649	+
	1650	+ mov r12d, 40003fffh
	1651	+ movd xmm5, r12d
	1652	+ punpcklwd xmm5, xmm0 ; 16384 16383
	1653	+ pshufd xmm5, xmm5, 01000100b ; xmm5: 16384 16383 16384 16383
	1654	+
	1655	+DOWNSAMPLE:
	1656	+ sub r1, r2 ; stride - width
	1657	+ dec r3
	1658	+ mov r14,16384
	1659	+ pshufd xmm4, xmm5, 01010000b ; initial v to 16384 16384 16383 16383
	1660	+
	1661	+HEIGHT:
	1662	+ ;mov r12, r4
	1663	+ mov r12, r14
	1664	+ shr r12, 15
	1665	+ imul r12, r5
	1666	+ add r12, r4 ; get current row address
	1667	+ mov r13, r12
	1668	+ add r13, r5
	1669	+
	1670	+ mov r15, 16384
	1671	+ mov rax, r2
	1672	+ dec rax
	1673	+ movdqa xmm3, xmm5 ; initial u to 16384 16383 16384 16383
	1674	+
	1675	+WIDTH:
	1676	+ mov r11, r15
	1677	+ shr r11, 15
	1678	+
	1679	+ movd xmm1, [r12+r11] ; xxxxxxba
	1680	+ movd xmm2, [r13+r11] ; xxxxxxdc
	1681	+ pxor xmm0, xmm0
	1682	+ punpcklwd xmm1, xmm2 ; xxxxdcba
	1683	+ punpcklbw xmm1, xmm0 ; 0d0c0b0a
	1684	+ punpcklwd xmm1, xmm0 ; 000d000c000b000a
	1685	+
	1686	+ movdqa xmm2, xmm4 ; xmm2: vv(1-v)(1-v) tmpv
	1687	+ pmaddwd xmm2, xmm3 ; mul u(1-u)u(1-u) on xmm2
	1688	+ movdqa xmm0, xmm2
	1689	+ pmuludq xmm2, xmm1
	1690	+ psrlq xmm0, 32
	1691	+ psrlq xmm1, 32
	1692	+ pmuludq xmm0, xmm1
	1693	+ paddq xmm2, xmm0
	1694	+ pshufd xmm1, xmm2, 00001110b
	1695	+ paddq xmm2, xmm1
	1696	+ psrlq xmm2, 29
	1697	+
	1698	+ movd ebx, xmm2
	1699	+ inc ebx
	1700	+ shr ebx, 1
	1701	+ mov [r0], bl
	1702	+ inc r0
	1703	+
	1704	+ add r15, r6
	1705	+ paddw xmm3, xmm7 ; inc u
	1706	+ psllw xmm3, 1
	1707	+ psrlw xmm3, 1
	1708	+
	1709	+ dec rax
	1710	+ jg WIDTH
	1711	+
	1712	+WIDTH_END:
	1713	+ shr r15, 15
	1714	+ mov bl, [r12+r15]
	1715	+ mov [r0],bl
	1716	+ inc r0
	1717	+ add r14, rbp
	1718	+ add r0, r1
	1719	+
	1720	+ paddw xmm4, xmm6 ; inc v
	1721	+ psllw xmm4, 1
	1722	+ psrlw xmm4, 1
	1723	+
	1724	+ dec r3
	1725	+ jg HEIGHT
	1726	+
	1727	+LAST_ROW:
	1728	+ shr r14, 15
	1729	+ imul r14, r5
	1730	+ add r4, r14
	1731	+ mov r15, 16384
	1732	+
	1733	+LAST_ROW_WIDTH:
	1734	+ mov r11, r15
	1735	+ shr r11, 15
	1736	+ mov bl, [r4+r11]
	1737	+ mov [r0],bl
	1738	+ inc r0
	1739	+
	1740	+ add r15, r6
	1741	+ dec r2
	1742	+ jg LAST_ROW_WIDTH
	1743	+
	1744	+LAST_ROW_END:
	1745	+
	1746	+ pop rbp
	1747	+ pop rbx
	1748	+ pop r15
	1749	+ pop r14
	1750	+ pop r13
	1751	+ pop r12
	1752	+ ret
	1753	+
	1754	+;**************************************************************************************************************
	1755	+;int GeneralBilinearFastDownsampler_sse2( unsigned char* pDst, const int iDstStride, const int iDstWidth, const int iDstHeight,
	1756	+; unsigned char* pSrc, const int iSrcStride,
	1757	+; unsigned int uiScaleX, unsigned int uiScaleY );
	1758	+;{
	1759	+;**************************************************************************************************************
	1760	+
	1761	+WELS_EXTERN GeneralBilinearFastDownsampler_sse2
	1762	+ push r12
	1763	+ push r13
	1764	+ push r14
	1765	+ push r15
	1766	+ push rbx
	1767	+ push rbp
	1768	+ %assign push_num 6
	1769	+ LOAD_7_PARA
	1770	+ SIGN_EXTENSION r1, r1d
	1771	+ SIGN_EXTENSION r2, r2d
	1772	+ SIGN_EXTENSION r3, r3d
	1773	+ SIGN_EXTENSION r5, r5d
	1774	+ SIGN_EXTENSION r6, r6d
	1775	+
	1776	+ pxor xmm0, xmm0
	1777	+ mov r12d, r6d
	1778	+ and r12d, 65535
	1779	+ mov r13d, r12d
	1780	+ neg r13d
	1781	+ and r13d, 65535
	1782	+ movd xmm1, r12d ; uinc(uiScaleX mod 65536)
	1783	+ movd xmm2, r13d ; -uinc
	1784	+ psllq xmm1, 32
	1785	+ por xmm1, xmm2 ; 0 uinc 0 -uinc
	1786	+ pshuflw xmm7, xmm1, 10001000b ; xmm7: uinc -uinc uinc -uinc
	1787	+
	1788	+ mov r12, arg8
	1789	+ SIGN_EXTENSION r12, r12d
	1790	+ mov rbp, r12
	1791	+ and r12d, 32767
	1792	+ mov r13d, r12d
	1793	+ neg r13d
	1794	+ and r13d, 32767
	1795	+ movd xmm6, r12d ; vinc(uiScaleY mod 32767)
	1796	+ movd xmm2, r13d ; -vinc
	1797	+ psllq xmm6, 32
	1798	+ por xmm6, xmm2 ; 0 vinc 0 -vinc
	1799	+ pshuflw xmm6, xmm6, 10100000b ; xmm6: vinc vinc -vinc -vinc
	1800	+
	1801	+ mov r12d, 80007fffh ; 32768 32767
	1802	+ movd xmm5, r12d
	1803	+ pshuflw xmm5, xmm5, 01000100b ; 32768 32767 32768 32767
	1804	+
	1805	+FAST_DOWNSAMPLE:
	1806	+ sub r1, r2 ; stride - width
	1807	+ dec r3
	1808	+ mov r14,16384
	1809	+
	1810	+ pshuflw xmm4, xmm5, 01010000b
	1811	+ psrlw xmm4, 1 ; initial v to 16384 16384 16383 16383
	1812	+
	1813	+FAST_HEIGHT:
	1814	+ mov r12, r14
	1815	+ shr r12, 15
	1816	+ imul r12, r5
	1817	+ add r12, r4 ; get current row address
	1818	+ mov r13, r12
	1819	+ add r13, r5
	1820	+
	1821	+ mov r15, 32768
	1822	+ mov rax, r2
	1823	+ dec rax
	1824	+
	1825	+ movdqa xmm3, xmm5 ; initial u to 32768 32767 32768 32767
	1826	+
	1827	+FAST_WIDTH:
	1828	+ mov r11, r15
	1829	+ shr r11, 16
	1830	+
	1831	+ movd xmm1, [r12+r11] ; xxxxxxba
	1832	+ movd xmm2, [r13+r11] ; xxxxxxdc
	1833	+ punpcklwd xmm1, xmm2 ; xxxxdcba
	1834	+ punpcklbw xmm1, xmm0 ; 0d0c0b0a
	1835	+
	1836	+ movdqa xmm2, xmm4 ; xmm2: vv(1-v)(1-v) tmpv
	1837	+ pmulhuw xmm2, xmm3 ; mul u(1-u)u(1-u) on xmm2
	1838	+ pmaddwd xmm2, xmm1
	1839	+ pshufd xmm1, xmm2, 00000001b
	1840	+ paddd xmm2, xmm1
	1841	+ movdqa xmm1, [add_extra_half]
	1842	+ paddd xmm2, xmm1
	1843	+ psrld xmm2, 15
	1844	+
	1845	+ packuswb xmm2, xmm0
	1846	+ movd ebx, xmm2
	1847	+ mov [r0], bl
	1848	+ inc r0
	1849	+
	1850	+ add r15, r6
	1851	+
	1852	+ paddw xmm3, xmm7 ; inc u
	1853	+ dec rax
	1854	+ jg FAST_WIDTH
	1855	+
	1856	+FAST_WIDTH_END:
	1857	+ shr r15, 16
	1858	+ mov bl, [r12+r15]
	1859	+ mov [r0],bl
	1860	+ inc r0
	1861	+ add r14, rbp
	1862	+ add r0, r1
	1863	+
	1864	+ paddw xmm4, xmm6 ; inc v
	1865	+ psllw xmm4, 1
	1866	+ psrlw xmm4, 1
	1867	+
	1868	+ dec r3
	1869	+ jg FAST_HEIGHT
	1870	+
	1871	+
	1872	+FAST_LAST_ROW:
	1873	+ shr r14, 15
	1874	+ imul r14, r5
	1875	+ add r4, r14
	1876	+ mov r15, 32768
	1877	+
	1878	+FAST_LAST_ROW_WIDTH:
	1879	+ mov r11, r15
	1880	+ shr r11, 16
	1881	+ mov bl, [r4+r11]
	1882	+ mov [r0],bl
	1883	+ inc r0
	1884	+
	1885	+ add r15, r6
	1886	+ dec r2
	1887	+ jg FAST_LAST_ROW_WIDTH
	1888	+
	1889	+FAST_LAST_ROW_END:
	1890	+
	1891	+ pop rbp
	1892	+ pop rbx
	1893	+ pop r15
	1894	+ pop r14
	1895	+ pop r13
	1896	+ pop r12
	1897	+ ret
1205	1898	%endif

Merge pull request #1140 from zhilwang/x86_64-downsample

Keyboard shortcuts

General

Issues and pull requests

Issues, pull requests and Inbox

Repository code