Convert all tabs to spaces in assembly sources, unify indentation

Previously the assembly sources had mixed indentation consisting of both spaces and tabs, making it quite hard to read unless the right tab size was used in the editor. Tabs have been interpreted as 4 spaces in most cases, matching the surrounding code.

Martin Storsjö committed 12 years ago

Commit 57f6bcc1 parent faaf62a38 files · +21316 −21316Browse files

Showing the changes to codec/processing/src (7 of 38 files). Show all

Unified · Split

7 files

codec/processing/src/arm/adaptive_quantization.S +46 −46
codec/processing/src/arm/down_sample_neon.S +218 −218
codec/processing/src/arm/pixel_sad_neon.S +17 −17
codec/processing/src/arm/vaa_calc_neon.S +584 −584
codec/processing/src/x86/denoisefilter.asm +170 −170
codec/processing/src/x86/downsample_bilinear.asm +978 −978
codec/processing/src/x86/vaa.asm +1518 −1518

codec/processing/src/arm/adaptive_quantization.S+46−46

@@ -36,17 +36,17 @@ · 35 unmodified lines
36	36
37	37	#ifdef __APPLE__
38	38	.macro SQR_ADD_16BYTES
39		− vmull.u8 q3, $0, $0
40		− vmull.u8 q8, $1, $1
41		− vpadal.u16 $2, q3
42		− vpadal.u16 $2, q8
	39	+ vmull.u8 q3, $0, $0
	40	+ vmull.u8 q8, $1, $1
	41	+ vpadal.u16 $2, q3
	42	+ vpadal.u16 $2, q8
43	43	.endm
44	44	#else
45	45	.macro SQR_ADD_16BYTES arg0, arg1, arg2
46		− vmull.u8 q3, \arg0, \arg0
47		− vmull.u8 q8, \arg1, \arg1
48		− vpadal.u16 \arg2, q3
49		− vpadal.u16 \arg2, q8
	46	+ vmull.u8 q3, \arg0, \arg0
	47	+ vmull.u8 q8, \arg1, \arg1
	48	+ vpadal.u16 \arg2, q3
	49	+ vpadal.u16 \arg2, q8
50	50	.endm
51	51	#endif
52	52
@@ -54,66 +54,66 @@ · 1 unmodified line
54	54	WELS_ASM_FUNC_BEGIN SampleVariance16x16_neon
55	55	stmdb sp!, {r4}
56	56
57		− vld1.8 {q15}, [r0], r1 //save the ref data (16bytes)
58		− vld1.8 {q14}, [r2], r3 //save the src data (16bytes)
	57	+ vld1.8 {q15}, [r0], r1 //save the ref data (16bytes)
	58	+ vld1.8 {q14}, [r2], r3 //save the src data (16bytes)
59	59
60	60
61		− vabd.u8 q13, q14, q15
62		− vmull.u8 q12, d27, d27
63		− vmull.u8 q11, d26, d26
64		− vaddl.u16 q12, d24, d25
65		− vpadal.u16 q12, q11 //sqr
	61	+ vabd.u8 q13, q14, q15
	62	+ vmull.u8 q12, d27, d27
	63	+ vmull.u8 q11, d26, d26
	64	+ vaddl.u16 q12, d24, d25
	65	+ vpadal.u16 q12, q11 //sqr
66	66
67	67	vaddl.u8 q13, d26, d27 //sum
68	68
69		− vaddl.u8 q10, d28, d29 //sum_cur
	69	+ vaddl.u8 q10, d28, d29 //sum_cur
70	70
71		− vmull.u8 q9, d29, d29
72		− vmull.u8 q8, d28, d28
73		− vaddl.u16 q9, d18, d19 //sqr_cur
74		− vpadal.u16 q9, q8
	71	+ vmull.u8 q9, d29, d29
	72	+ vmull.u8 q8, d28, d28
	73	+ vaddl.u16 q9, d18, d19 //sqr_cur
	74	+ vpadal.u16 q9, q8
75	75
76		− mov r4, #15
	76	+ mov r4, #15
77	77	pixel_var_16x16_loop0:
78	78
79		− vld1.8 {q0}, [r0], r1 //save the ref data (16bytes)
80		− vld1.8 {q1}, [r2], r3 //save the src data (16bytes)
	79	+ vld1.8 {q0}, [r0], r1 //save the ref data (16bytes)
	80	+ vld1.8 {q1}, [r2], r3 //save the src data (16bytes)
81	81
82		− vabd.u8 q2, q0, q1
	82	+ vabd.u8 q2, q0, q1
83	83
84		− //q10 save sum_cur
85		− vpadal.u8 q10, q1
	84	+ //q10 save sum_cur
	85	+ vpadal.u8 q10, q1
86	86
87		− //q12 save sqr
88		− SQR_ADD_16BYTES d4, d5, q12
	87	+ //q12 save sqr
	88	+ SQR_ADD_16BYTES d4, d5, q12
89	89
90	90	//q13 save sum
91		− vpadal.u8 q13, q2
	91	+ vpadal.u8 q13, q2
92	92
93		− subs r4, #1
	93	+ subs r4, #1
94	94
95		− //q9 save sqr_cur
96		− SQR_ADD_16BYTES d2, d3, q9
	95	+ //q9 save sqr_cur
	96	+ SQR_ADD_16BYTES d2, d3, q9
97	97
98		− bne pixel_var_16x16_loop0
	98	+ bne pixel_var_16x16_loop0
99	99
100		− vadd.u16 d0, d26, d27 //sum
101		− vadd.u16 d1, d20, d21 //sum_cur
102		− vpaddl.u16 q0, q0
103		− vadd.u32 d2, d24, d25 //sqr
104		− vadd.u32 d3, d18, d19 //sqr_cur
105		− vpadd.u32 d0, d0, d1
106		− vpadd.u32 d1, d2, d3
	100	+ vadd.u16 d0, d26, d27 //sum
	101	+ vadd.u16 d1, d20, d21 //sum_cur
	102	+ vpaddl.u16 q0, q0
	103	+ vadd.u32 d2, d24, d25 //sqr
	104	+ vadd.u32 d3, d18, d19 //sqr_cur
	105	+ vpadd.u32 d0, d0, d1
	106	+ vpadd.u32 d1, d2, d3
107	107
108		− ldr r4, [sp, #4]
	108	+ ldr r4, [sp, #4]
109	109
110		− vshr.u32 q0, q0, #8
111		− vmul.u32 d0, d0
112		− vsub.u32 d0, d1, d0
	110	+ vshr.u32 q0, q0, #8
	111	+ vmul.u32 d0, d0
	112	+ vsub.u32 d0, d1, d0
113	113	vmovl.u32 q0, d0
114		− vst2.16 {d0[0], d1[0]}, [r4]
	114	+ vst2.16 {d0[0], d1[0]}, [r4]
115	115
116		− ldmia sp!, {r4}
	116	+ ldmia sp!, {r4}
117	117
118	118	WELS_ASM_FUNC_END
119	119

codec/processing/src/arm/down_sample_neon.S+218−218

@@ -30,313 +30,313 @@ · 29 unmodified lines
30	30	*
31	31	*/
32	32
33		−#ifdef HAVE_NEON
	33	+#ifdef HAVE_NEON
34	34	.text
35	35	#include "arm_arch_common_macro.S"
36	36
37	37
38		−WELS_ASM_FUNC_BEGIN DyadicBilinearDownsampler_neon
39		− stmdb sp!, {r4-r8, lr}
	38	+WELS_ASM_FUNC_BEGIN DyadicBilinearDownsampler_neon
	39	+ stmdb sp!, {r4-r8, lr}
40	40
41		− //Get the width and height
42		− ldr r4, [sp, #24] //src_width
43		− ldr r5, [sp, #28] //src_height
	41	+ //Get the width and height
	42	+ ldr r4, [sp, #24] //src_width
	43	+ ldr r5, [sp, #28] //src_height
44	44
45		− //Initialize the register
46		− mov r6, r2
47		− mov r8, r0
48		− mov lr, #0
49		− lsr r5, #1
	45	+ //Initialize the register
	46	+ mov r6, r2
	47	+ mov r8, r0
	48	+ mov lr, #0
	49	+ lsr r5, #1
50	50
51		− //Save the tailer for the unasigned size
52		− mla r7, r1, r5, r0
53		− vld1.32 {q15}, [r7]
	51	+ //Save the tailer for the unasigned size
	52	+ mla r7, r1, r5, r0
	53	+ vld1.32 {q15}, [r7]
54	54
55		− add r7, r2, r3
56		− //processing a colume data
	55	+ add r7, r2, r3
	56	+ //processing a colume data
57	57	comp_ds_bilinear_loop0:
58	58
59		− vld1.8 {q0,q1}, [r2]!
60		− vld1.8 {q2,q3}, [r7]!
61		− vpaddl.u8 q0, q0
62		− vpaddl.u8 q1, q1
63		− vpaddl.u8 q2, q2
64		− vpaddl.u8 q3, q3
65		− vrshr.u16 q0, #1
66		− vrshr.u16 q1, #1
67		− vrshr.u16 q2, #1
68		− vrshr.u16 q3, #1
69		− vrhadd.u16 q0, q2
70		− vrhadd.u16 q1, q3
71		− vmovn.u16 d0, q0
72		− vmovn.u16 d1, q1
73		− vst1.32 {q0}, [r0]!
74		− add lr, #32
75		−
76		− cmp lr, r4
77		− movcs lr, #0
78		− addcs r6, r6, r3, lsl #1
79		− movcs r2, r6
80		− addcs r7, r2, r3
81		− addcs r8, r1
82		− movcs r0, r8
83		− subscs r5, #1
84		− bne comp_ds_bilinear_loop0
85		−
86		− //restore the tailer for the unasigned size
87		− vst1.32 {q15}, [r0]
88		−
89		− ldmia sp!, {r4-r8,lr}
	59	+ vld1.8 {q0,q1}, [r2]!
	60	+ vld1.8 {q2,q3}, [r7]!
	61	+ vpaddl.u8 q0, q0
	62	+ vpaddl.u8 q1, q1
	63	+ vpaddl.u8 q2, q2
	64	+ vpaddl.u8 q3, q3
	65	+ vrshr.u16 q0, #1
	66	+ vrshr.u16 q1, #1
	67	+ vrshr.u16 q2, #1
	68	+ vrshr.u16 q3, #1
	69	+ vrhadd.u16 q0, q2
	70	+ vrhadd.u16 q1, q3
	71	+ vmovn.u16 d0, q0
	72	+ vmovn.u16 d1, q1
	73	+ vst1.32 {q0}, [r0]!
	74	+ add lr, #32
	75	+
	76	+ cmp lr, r4
	77	+ movcs lr, #0
	78	+ addcs r6, r6, r3, lsl #1
	79	+ movcs r2, r6
	80	+ addcs r7, r2, r3
	81	+ addcs r8, r1
	82	+ movcs r0, r8
	83	+ subscs r5, #1
	84	+ bne comp_ds_bilinear_loop0
	85	+
	86	+ //restore the tailer for the unasigned size
	87	+ vst1.32 {q15}, [r0]
	88	+
	89	+ ldmia sp!, {r4-r8,lr}
90	90	WELS_ASM_FUNC_END
91	91
92	92
93		−WELS_ASM_FUNC_BEGIN comp_ds_bilinear_w_x8_neon
94		− stmdb sp!, {r4-r7, lr}
	93	+WELS_ASM_FUNC_BEGIN comp_ds_bilinear_w_x8_neon
	94	+ stmdb sp!, {r4-r7, lr}
95	95
96		− //Get the width and height
97		− ldr r4, [sp, #20] //src_width
98		− ldr r5, [sp, #24] //src_height
	96	+ //Get the width and height
	97	+ ldr r4, [sp, #20] //src_width
	98	+ ldr r5, [sp, #24] //src_height
99	99
100		− //Get the difference
101		− sub lr, r3, r4
102		− sub r1, r1, r4, lsr #1
	100	+ //Get the difference
	101	+ sub lr, r3, r4
	102	+ sub r1, r1, r4, lsr #1
103	103
104		− lsr r5, #1
	104	+ lsr r5, #1
105	105
106		− //processing a colume data
	106	+ //processing a colume data
107	107	comp_ds_bilinear_w_x8_loop0:
108	108
109		− lsr r6, r4, #3
110		− add r7, r2, r3
111		− //processing a line data
	109	+ lsr r6, r4, #3
	110	+ add r7, r2, r3
	111	+ //processing a line data
112	112	comp_ds_bilinear_w_x8_loop1:
113	113
114		− vld1.8 {d0}, [r2]!
115		− vld1.8 {d1}, [r7]!
116		− vpaddl.u8 q0, q0
117		− vrshr.u16 q0, #1
118		− vrhadd.u16 d0, d1
	114	+ vld1.8 {d0}, [r2]!
	115	+ vld1.8 {d1}, [r7]!
	116	+ vpaddl.u8 q0, q0
	117	+ vrshr.u16 q0, #1
	118	+ vrhadd.u16 d0, d1
119	119
120		− vmovn.u16 d0, q0
121		− vst1.32 {d0[0]}, [r0]!
122		− subs r6, #1
123		− bne comp_ds_bilinear_w_x8_loop1
	120	+ vmovn.u16 d0, q0
	121	+ vst1.32 {d0[0]}, [r0]!
	122	+ subs r6, #1
	123	+ bne comp_ds_bilinear_w_x8_loop1
124	124
125		− add r2, r7, lr
126		− add r0, r1
127		− subs r5, #1
128		− bne comp_ds_bilinear_w_x8_loop0
	125	+ add r2, r7, lr
	126	+ add r0, r1
	127	+ subs r5, #1
	128	+ bne comp_ds_bilinear_w_x8_loop0
129	129
130		− ldmia sp!, {r4-r7,lr}
	130	+ ldmia sp!, {r4-r7,lr}
131	131	WELS_ASM_FUNC_END
132	132
133	133
134		−WELS_ASM_FUNC_BEGIN comp_ds_bilinear_w_x16_neon
135		− stmdb sp!, {r4-r7, lr}
	134	+WELS_ASM_FUNC_BEGIN comp_ds_bilinear_w_x16_neon
	135	+ stmdb sp!, {r4-r7, lr}
136	136
137		− //Get the width and height
138		− ldr r4, [sp, #20] //src_width
139		− ldr r5, [sp, #24] //src_height
	137	+ //Get the width and height
	138	+ ldr r4, [sp, #20] //src_width
	139	+ ldr r5, [sp, #24] //src_height
140	140
141		− //Get the difference
142		− sub lr, r3, r4
143		− sub r1, r1, r4, lsr #1
	141	+ //Get the difference
	142	+ sub lr, r3, r4
	143	+ sub r1, r1, r4, lsr #1
144	144
145		− lsr r5, #1
	145	+ lsr r5, #1
146	146
147		− //processing a colume data
	147	+ //processing a colume data
148	148	comp_ds_bilinear_w_x16_loop0:
149	149
150		− lsr r6, r4, #4
151		− add r7, r2, r3
152		− //processing a line data
	150	+ lsr r6, r4, #4
	151	+ add r7, r2, r3
	152	+ //processing a line data
153	153	comp_ds_bilinear_w_x16_loop1:
154	154
155		− vld1.8 {q0}, [r2]!
156		− vld1.8 {q1}, [r7]!
157		− vpaddl.u8 q0, q0
158		− vpaddl.u8 q1, q1
159		− vrshr.u16 q0, #1
160		− vrshr.u16 q1, #1
161		− vrhadd.u16 q0, q1
162		−
163		− vmovn.u16 d0, q0
164		− vst1.32 {d0}, [r0]!
165		− subs r6, #1
166		− bne comp_ds_bilinear_w_x16_loop1
167		−
168		− add r2, r7, lr
169		− add r0, r1
170		− subs r5, #1
171		− bne comp_ds_bilinear_w_x16_loop0
172		−
173		− ldmia sp!, {r4-r7,lr}
	155	+ vld1.8 {q0}, [r2]!
	156	+ vld1.8 {q1}, [r7]!
	157	+ vpaddl.u8 q0, q0
	158	+ vpaddl.u8 q1, q1
	159	+ vrshr.u16 q0, #1
	160	+ vrshr.u16 q1, #1
	161	+ vrhadd.u16 q0, q1
	162	+
	163	+ vmovn.u16 d0, q0
	164	+ vst1.32 {d0}, [r0]!
	165	+ subs r6, #1
	166	+ bne comp_ds_bilinear_w_x16_loop1
	167	+
	168	+ add r2, r7, lr
	169	+ add r0, r1
	170	+ subs r5, #1
	171	+ bne comp_ds_bilinear_w_x16_loop0
	172	+
	173	+ ldmia sp!, {r4-r7,lr}
174	174	WELS_ASM_FUNC_END
175	175
176	176
177		−WELS_ASM_FUNC_BEGIN DyadicBilinearDownsamplerWidthx32_neon
178		− stmdb sp!, {r4-r7, lr}
	177	+WELS_ASM_FUNC_BEGIN DyadicBilinearDownsamplerWidthx32_neon
	178	+ stmdb sp!, {r4-r7, lr}
179	179
180		− //Get the width and height
181		− ldr r4, [sp, #20] //src_width
182		− ldr r5, [sp, #24] //src_height
	180	+ //Get the width and height
	181	+ ldr r4, [sp, #20] //src_width
	182	+ ldr r5, [sp, #24] //src_height
183	183
184		− //Get the difference
185		− sub lr, r3, r4
186		− sub r1, r1, r4, lsr #1
	184	+ //Get the difference
	185	+ sub lr, r3, r4
	186	+ sub r1, r1, r4, lsr #1
187	187
188		− lsr r5, #1
	188	+ lsr r5, #1
189	189
190		− //processing a colume data
	190	+ //processing a colume data
191	191	comp_ds_bilinear_w_x32_loop0:
192	192
193		− lsr r6, r4, #5
194		− add r7, r2, r3
195		− //processing a line data
	193	+ lsr r6, r4, #5
	194	+ add r7, r2, r3
	195	+ //processing a line data
196	196	comp_ds_bilinear_w_x32_loop1:
197	197
198		− vld1.8 {q0,q1}, [r2]!
199		− vld1.8 {q2,q3}, [r7]!
200		− vpaddl.u8 q0, q0
201		− vpaddl.u8 q1, q1
202		− vpaddl.u8 q2, q2
203		− vpaddl.u8 q3, q3
204		− vrshr.u16 q0, #1
205		− vrshr.u16 q1, #1
206		− vrshr.u16 q2, #1
207		− vrshr.u16 q3, #1
208		− vrhadd.u16 q0, q2
209		− vrhadd.u16 q1, q3
210		−
211		− vmovn.u16 d0, q0
212		− vmovn.u16 d1, q1
213		− vst1.32 {q0}, [r0]!
214		− subs r6, #1
215		− bne comp_ds_bilinear_w_x32_loop1
216		−
217		− add r2, r7, lr
218		− add r0, r1
219		− subs r5, #1
220		− bne comp_ds_bilinear_w_x32_loop0
221		−
222		− ldmia sp!, {r4-r7,lr}
	198	+ vld1.8 {q0,q1}, [r2]!
	199	+ vld1.8 {q2,q3}, [r7]!
	200	+ vpaddl.u8 q0, q0
	201	+ vpaddl.u8 q1, q1
	202	+ vpaddl.u8 q2, q2
	203	+ vpaddl.u8 q3, q3
	204	+ vrshr.u16 q0, #1
	205	+ vrshr.u16 q1, #1
	206	+ vrshr.u16 q2, #1
	207	+ vrshr.u16 q3, #1
	208	+ vrhadd.u16 q0, q2
	209	+ vrhadd.u16 q1, q3
	210	+
	211	+ vmovn.u16 d0, q0
	212	+ vmovn.u16 d1, q1
	213	+ vst1.32 {q0}, [r0]!
	214	+ subs r6, #1
	215	+ bne comp_ds_bilinear_w_x32_loop1
	216	+
	217	+ add r2, r7, lr
	218	+ add r0, r1
	219	+ subs r5, #1
	220	+ bne comp_ds_bilinear_w_x32_loop0
	221	+
	222	+ ldmia sp!, {r4-r7,lr}
223	223	WELS_ASM_FUNC_END
224	224
225	225
226	226	WELS_ASM_FUNC_BEGIN GeneralBilinearAccurateDownsampler_neon
227	227	stmdb sp!, {r4-r12, lr}
228	228
229		− //Get the data from stack
230		− ldr r4, [sp, #40] //the addr of src
231		− ldr r5, [sp, #44] //the value of src_stride
	229	+ //Get the data from stack
	230	+ ldr r4, [sp, #40] //the addr of src
	231	+ ldr r5, [sp, #44] //the value of src_stride
232	232	ldr r6, [sp, #48] //the value of scaleX
233	233	ldr r7, [sp, #52] //the value of scaleY
234	234
235	235	mov r10, #32768
236	236	sub r10, #1
237		− and r8, r6, r10 // r8 uinc(scaleX mod 32767)
	237	+ and r8, r6, r10 // r8 uinc(scaleX mod 32767)
238	238	mov r11, #-1
239		− mul r11, r8 // r11 -uinc
	239	+ mul r11, r8 // r11 -uinc
240	240
241	241	vdup.s16 d2, r8
242	242	vdup.s16 d0, r11
243	243	vzip.s16 d0, d2 // uinc -uinc uinc -uinc
244	244
245		− and r9, r7, r10 // r9 vinc(scaleY mod 32767)
	245	+ and r9, r7, r10 // r9 vinc(scaleY mod 32767)
246	246	mov r11, #-1
247		− mul r11, r9 // r11 -vinc
	247	+ mul r11, r9 // r11 -vinc
248	248
249		− vdup.s16 d2, r9
250		− vdup.s16 d3, r11
251		− vext.8 d5, d3, d2, #4 // vinc vinc -vinc -vinc
	249	+ vdup.s16 d2, r9
	250	+ vdup.s16 d3, r11
	251	+ vext.8 d5, d3, d2, #4 // vinc vinc -vinc -vinc
252	252
253		− mov r11, #0x40000000
	253	+ mov r11, #0x40000000
254	254	mov r12, #0x4000
255	255	sub r12, #1
256	256	add r11, r12
257		− vdup.s32 d1, r11; //init u 16384 16383 16384 16383
	257	+ vdup.s32 d1, r11; //init u 16384 16383 16384 16383
258	258
259		− mov r11, #16384
	259	+ mov r11, #16384
260	260	vdup.s16 d16, r11
261	261	sub r11, #1
262		− vdup.s16 d17, r11
263		− vext.8 d7, d17, d16, #4 //init v 16384 16384 16383 16383
	262	+ vdup.s16 d17, r11
	263	+ vext.8 d7, d17, d16, #4 //init v 16384 16384 16383 16383
264	264
265		− veor q14, q14
266		− sub r1, r2 // stride - width
267		− mov r8, #16384 // yInverse
268		− sub r3, #1
	265	+ veor q14, q14
	266	+ sub r1, r2 // stride - width
	267	+ mov r8, #16384 // yInverse
	268	+ sub r3, #1
269	269
270	270	_HEIGHT:
271	271	ldr r4, [sp, #40] //the addr of src
272		− mov r11, r8
273		− lsr r11, #15
274		− mul r11, r5
275		− add r11, r4 // get current row address
276		− mov r12, r11
277		− add r12, r5
278		−
279		− mov r9, #16384 // xInverse
280		− sub r10, r2, #1
	272	+ mov r11, r8
	273	+ lsr r11, #15
	274	+ mul r11, r5
	275	+ add r11, r4 // get current row address
	276	+ mov r12, r11
	277	+ add r12, r5
	278	+
	279	+ mov r9, #16384 // xInverse
	280	+ sub r10, r2, #1
281	281	vmov.s16 d6, d1
282	282
283	283	_WIDTH:
284		− mov lr, r9
285		− lsr lr, #15
	284	+ mov lr, r9
	285	+ lsr lr, #15
286	286	add r4, r11,lr
287		− vld2.8 {d28[0],d29[0]}, [r4] //q14: 0000000b0000000a;
	287	+ vld2.8 {d28[0],d29[0]}, [r4] //q14: 0000000b0000000a;
288	288	add r4, r12,lr
289		− vld2.8 {d28[4],d29[4]}, [r4] //q14: 000d000b000c000a;
290		− vzip.32 d28, d29 //q14: 000d000c000b000a;
291		−
292		− vmull.u16 q13, d6, d7 //q13: init u * init v
293		− vmull.u32 q12, d26,d28
294		− vmlal.u32 q12, d27,d29
295		− vqadd.u64 d24, d24,d25
296		− vrshr.u64 d24, #30
297		−
298		− vst1.8 {d24[0]}, [r0]!
299		− add r9, r6
300		− vadd.u16 d6, d0 // inc u
301		− vshl.u16 d6, #1
302		− vshr.u16 d6, #1
303		− subs r10, #1
304		− bne _WIDTH
	289	+ vld2.8 {d28[4],d29[4]}, [r4] //q14: 000d000b000c000a;
	290	+ vzip.32 d28, d29 //q14: 000d000c000b000a;
	291	+
	292	+ vmull.u16 q13, d6, d7 //q13: init u * init v
	293	+ vmull.u32 q12, d26,d28
	294	+ vmlal.u32 q12, d27,d29
	295	+ vqadd.u64 d24, d24,d25
	296	+ vrshr.u64 d24, #30
	297	+
	298	+ vst1.8 {d24[0]}, [r0]!
	299	+ add r9, r6
	300	+ vadd.u16 d6, d0 // inc u
	301	+ vshl.u16 d6, #1
	302	+ vshr.u16 d6, #1
	303	+ subs r10, #1
	304	+ bne _WIDTH
305	305
306	306	WIDTH_END:
307		− lsr r9, #15
	307	+ lsr r9, #15
308	308	add r4,r11,r9
309		− vld1.8 {d24[0]}, [r4]
310		− vst1.8 {d24[0]}, [r0]
311		− add r0, #1
312		− add r8, r7
313		− add r0, r1
314		− vadd.s16 d7, d5 // inc v
315		− vshl.u16 d7, #1
316		− vshr.u16 d7, #1
317		− subs r3, #1
318		− bne _HEIGHT
	309	+ vld1.8 {d24[0]}, [r4]
	310	+ vst1.8 {d24[0]}, [r0]
	311	+ add r0, #1
	312	+ add r8, r7
	313	+ add r0, r1
	314	+ vadd.s16 d7, d5 // inc v
	315	+ vshl.u16 d7, #1
	316	+ vshr.u16 d7, #1
	317	+ subs r3, #1
	318	+ bne _HEIGHT
319	319
320	320	LAST_ROW:
321	321	ldr r4, [sp, #40] //the addr of src
322		− lsr r8, #15
323		− mul r8, r5
324		− add r4, r8 // get current row address
325		− mov r9, #16384
	322	+ lsr r8, #15
	323	+ mul r8, r5
	324	+ add r4, r8 // get current row address
	325	+ mov r9, #16384
326	326
327	327	_LAST_ROW_WIDTH:
328		− mov r11, r9
329		− lsr r11, #15
330		−
331		− add r3, r4,r11
332		− vld1.8 {d0[0]}, [r3]
333		− vst1.8 {d0[0]}, [r0]
334		− add r0, #1
335		− add r9, r6
336		− subs r2, #1
337		− bne _LAST_ROW_WIDTH
338		−
339		− ldmia sp!, {r4-r12, lr}
	328	+ mov r11, r9
	329	+ lsr r11, #15
	330	+
	331	+ add r3, r4,r11
	332	+ vld1.8 {d0[0]}, [r3]
	333	+ vst1.8 {d0[0]}, [r0]
	334	+ add r0, #1
	335	+ add r9, r6
	336	+ subs r2, #1
	337	+ bne _LAST_ROW_WIDTH
	338	+
	339	+ ldmia sp!, {r4-r12, lr}
340	340	WELS_ASM_FUNC_END
341	341
342	342	#endif

codec/processing/src/arm/pixel_sad_neon.S+17−17

@@ -37,32 +37,32 @@ · 36 unmodified lines
37	37
38	38	WELS_ASM_FUNC_BEGIN WelsProcessingSampleSad8x8_neon
39	39	stmdb sp!, {lr}
40		− //Loading a horizontal line data (8 bytes)
41		− vld1.8 {d0}, [r0], r1
42		− vld1.8 {d1}, [r2], r3
	40	+ //Loading a horizontal line data (8 bytes)
	41	+ vld1.8 {d0}, [r0], r1
	42	+ vld1.8 {d1}, [r2], r3
43	43
44		− //Do the SAD for 8 bytes
45		− vabdl.u8 q1, d0, d1
	44	+ //Do the SAD for 8 bytes
	45	+ vabdl.u8 q1, d0, d1
46	46
47		− mov lr, #7
	47	+ mov lr, #7
48	48	pixel_sad_8x8_loop0:
49	49
50	50	//Loading a horizontal line data (8 bytes)
51		− vld1.8 {d0}, [r0], r1
52		− vld1.8 {d1}, [r2], r3
	51	+ vld1.8 {d0}, [r0], r1
	52	+ vld1.8 {d1}, [r2], r3
53	53
54		− subs lr, #1
	54	+ subs lr, #1
55	55
56		− //Do the SAD for 8 bytes
57		− vabal.u8 q1, d0, d1
58		− bne pixel_sad_8x8_loop0
	56	+ //Do the SAD for 8 bytes
	57	+ vabal.u8 q1, d0, d1
	58	+ bne pixel_sad_8x8_loop0
59	59
60		− vadd.u16 d2, d3
61		− vpaddl.u16 d2, d2
62		− vpaddl.u32 d2, d2
63		− vmov.u32 r0, d2[0]//TBO...
	60	+ vadd.u16 d2, d3
	61	+ vpaddl.u16 d2, d2
	62	+ vpaddl.u32 d2, d2
	63	+ vmov.u32 r0, d2[0]//TBO...
64	64
65		− ldmia sp!, {lr}
	65	+ ldmia sp!, {lr}
66	66	WELS_ASM_FUNC_END
67	67
68	68	#endif

codec/processing/src/arm/vaa_calc_neon.S+584−584

@@ -37,61 +37,61 @@ · 36 unmodified lines
37	37	#ifdef __APPLE__
38	38
39	39	.macro ABS_SUB_SUM_16BYTES
40		− vld1.32 {q15}, [$0], $2
41		− vld1.32 {q14}, [$1], $2
42		− vabal.u8 $3, d30, d28
43		− vabal.u8 $4, d31, d29
	40	+ vld1.32 {q15}, [$0], $2
	41	+ vld1.32 {q14}, [$1], $2
	42	+ vabal.u8 $3, d30, d28
	43	+ vabal.u8 $4, d31, d29
44	44	.endm
45	45
46	46	.macro ABS_SUB_SUM_8x16BYTES
47		− vld1.32 {q15}, [$0], $2
48		− vld1.32 {q14}, [$1], $2
49		− vabdl.u8 $3, d30, d28
50		− vabdl.u8 $4, d31, d29
51		−
52		− ABS_SUB_SUM_16BYTES $0, $1, $2, $3, $4
53		− ABS_SUB_SUM_16BYTES $0, $1, $2, $3, $4
54		− ABS_SUB_SUM_16BYTES $0, $1, $2, $3, $4
55		− ABS_SUB_SUM_16BYTES $0, $1, $2, $3, $4
56		− ABS_SUB_SUM_16BYTES $0, $1, $2, $3, $4
57		− ABS_SUB_SUM_16BYTES $0, $1, $2, $3, $4
58		− ABS_SUB_SUM_16BYTES $0, $1, $2, $3, $4
	47	+ vld1.32 {q15}, [$0], $2
	48	+ vld1.32 {q14}, [$1], $2
	49	+ vabdl.u8 $3, d30, d28
	50	+ vabdl.u8 $4, d31, d29
	51	+
	52	+ ABS_SUB_SUM_16BYTES $0, $1, $2, $3, $4
	53	+ ABS_SUB_SUM_16BYTES $0, $1, $2, $3, $4
	54	+ ABS_SUB_SUM_16BYTES $0, $1, $2, $3, $4
	55	+ ABS_SUB_SUM_16BYTES $0, $1, $2, $3, $4
	56	+ ABS_SUB_SUM_16BYTES $0, $1, $2, $3, $4
	57	+ ABS_SUB_SUM_16BYTES $0, $1, $2, $3, $4
	58	+ ABS_SUB_SUM_16BYTES $0, $1, $2, $3, $4
59	59	.endm
60	60
61	61	.macro SAD_8X16BITS
62		− vadd.u16 d31, $0, $1
63		− vpaddl.u16 d31, d31
64		− vpaddl.u32 $2, d31
	62	+ vadd.u16 d31, $0, $1
	63	+ vpaddl.u16 d31, d31
	64	+ vpaddl.u32 $2, d31
65	65	.endm
66	66
67	67	#else
68	68
69	69	.macro ABS_SUB_SUM_16BYTES arg0, arg1, arg2, arg3, arg4
70		− vld1.32 {q15}, [\arg0], \arg2
71		− vld1.32 {q14}, [\arg1], \arg2
72		− vabal.u8 \arg3, d30, d28
73		− vabal.u8 \arg4, d31, d29
	70	+ vld1.32 {q15}, [\arg0], \arg2
	71	+ vld1.32 {q14}, [\arg1], \arg2
	72	+ vabal.u8 \arg3, d30, d28
	73	+ vabal.u8 \arg4, d31, d29
74	74	.endm
75	75
76	76	.macro ABS_SUB_SUM_8x16BYTES arg0, arg1, arg2, arg3, arg4
77		− vld1.32 {q15}, [\arg0], \arg2
78		− vld1.32 {q14}, [\arg1], \arg2
79		− vabdl.u8 \arg3, d30, d28
80		− vabdl.u8 \arg4, d31, d29
81		−
82		− ABS_SUB_SUM_16BYTES \arg0, \arg1, \arg2, \arg3, \arg4
83		− ABS_SUB_SUM_16BYTES \arg0, \arg1, \arg2, \arg3, \arg4
84		− ABS_SUB_SUM_16BYTES \arg0, \arg1, \arg2, \arg3, \arg4
85		− ABS_SUB_SUM_16BYTES \arg0, \arg1, \arg2, \arg3, \arg4
86		− ABS_SUB_SUM_16BYTES \arg0, \arg1, \arg2, \arg3, \arg4
87		− ABS_SUB_SUM_16BYTES \arg0, \arg1, \arg2, \arg3, \arg4
88		− ABS_SUB_SUM_16BYTES \arg0, \arg1, \arg2, \arg3, \arg4
	77	+ vld1.32 {q15}, [\arg0], \arg2
	78	+ vld1.32 {q14}, [\arg1], \arg2
	79	+ vabdl.u8 \arg3, d30, d28
	80	+ vabdl.u8 \arg4, d31, d29
	81	+
	82	+ ABS_SUB_SUM_16BYTES \arg0, \arg1, \arg2, \arg3, \arg4
	83	+ ABS_SUB_SUM_16BYTES \arg0, \arg1, \arg2, \arg3, \arg4
	84	+ ABS_SUB_SUM_16BYTES \arg0, \arg1, \arg2, \arg3, \arg4
	85	+ ABS_SUB_SUM_16BYTES \arg0, \arg1, \arg2, \arg3, \arg4
	86	+ ABS_SUB_SUM_16BYTES \arg0, \arg1, \arg2, \arg3, \arg4
	87	+ ABS_SUB_SUM_16BYTES \arg0, \arg1, \arg2, \arg3, \arg4
	88	+ ABS_SUB_SUM_16BYTES \arg0, \arg1, \arg2, \arg3, \arg4
89	89	.endm
90	90
91	91	.macro SAD_8X16BITS arg0, arg1, arg2
92		− vadd.u16 d31, \arg0, \arg1
93		− vpaddl.u16 d31, d31
94		− vpaddl.u32 \arg2, d31
	92	+ vadd.u16 d31, \arg0, \arg1
	93	+ vpaddl.u16 d31, d31
	94	+ vpaddl.u32 \arg2, d31
95	95	.endm
96	96	#endif
97	97
@@ -100,16 +100,16 @@ WELS_ASM_FUNC_BEGIN VAACalcSad_neon · 2 unmodified lines
100	100
101	101	stmdb sp!, {r4-r8}
102	102
103		− ldr r4, [sp, #20] //load pic_stride
104		− ldr r5, [sp, #28] //load psad8x8
	103	+ ldr r4, [sp, #20] //load pic_stride
	104	+ ldr r5, [sp, #28] //load psad8x8
105	105
106		− //Initial the Q8 register for save the "psadframe"
107		− vmov.s64 q8, #0
	106	+ //Initial the Q8 register for save the "psadframe"
	107	+ vmov.s64 q8, #0
108	108
109		− //Get the jump distance to use on loop codes
110		− lsl r8, r4, #4
111		− sub r7, r8, #16 //R7 keep the 16*pic_stride-16
112		− sub r8, r2 //R8 keep the 16*pic_stride-pic_width
	109	+ //Get the jump distance to use on loop codes
	110	+ lsl r8, r4, #4
	111	+ sub r7, r8, #16 //R7 keep the 16*pic_stride-16
	112	+ sub r8, r2 //R8 keep the 16*pic_stride-pic_width
113	113
114	114	vaa_calc_sad_loop0:
115	115
@@ -118,70 +118,70 @@ vaa_calc_sad_loop0: · 2 unmodified lines
118	118
119	119	vaa_calc_sad_loop1:
120	120
121		− //Process the 16x16 bytes
122		− ABS_SUB_SUM_8x16BYTES r0, r1, r4, q0, q1
123		− ABS_SUB_SUM_8x16BYTES r0, r1, r4, q2, q3
	121	+ //Process the 16x16 bytes
	122	+ ABS_SUB_SUM_8x16BYTES r0, r1, r4, q0, q1
	123	+ ABS_SUB_SUM_8x16BYTES r0, r1, r4, q2, q3
124	124
125		− //Do the SAD
126		− SAD_8X16BITS d0, d1, d0
127		− SAD_8X16BITS d2, d3, d1
128		− SAD_8X16BITS d4, d5, d2
129		− SAD_8X16BITS d6, d7, d3
	125	+ //Do the SAD
	126	+ SAD_8X16BITS d0, d1, d0
	127	+ SAD_8X16BITS d2, d3, d1
	128	+ SAD_8X16BITS d4, d5, d2
	129	+ SAD_8X16BITS d6, d7, d3
130	130
131		− //Write to "psad8x8" buffer
132		− vst4.32 {d0[0],d1[0],d2[0],d3[0]}, [r5]!
	131	+ //Write to "psad8x8" buffer
	132	+ vst4.32 {d0[0],d1[0],d2[0],d3[0]}, [r5]!
133	133
134	134
135		− //Adjust the input address
136		− sub r0, r7
137		− sub r1, r7
	135	+ //Adjust the input address
	136	+ sub r0, r7
	137	+ sub r1, r7
138	138
139		− subs r6, #16
	139	+ subs r6, #16
140	140
141		− //Save to calculate "psadframe"
142		− vadd.u32 q0, q1
143		− vadd.u32 q8, q0
	141	+ //Save to calculate "psadframe"
	142	+ vadd.u32 q0, q1
	143	+ vadd.u32 q8, q0
144	144
145		− bne vaa_calc_sad_loop1
	145	+ bne vaa_calc_sad_loop1
146	146
147		− //Adjust the input address
148		− add r0, r8
149		− add r1, r8
	147	+ //Adjust the input address
	148	+ add r0, r8
	149	+ add r1, r8
150	150
151	151	subs r3, #16
152		− bne vaa_calc_sad_loop0
	152	+ bne vaa_calc_sad_loop0
153	153
154		− ldr r6, [sp, #24] //load psadframe
155		− vadd.u32 d16, d17
156		− vst1.32 {d16[0]}, [r6]
	154	+ ldr r6, [sp, #24] //load psadframe
	155	+ vadd.u32 d16, d17
	156	+ vst1.32 {d16[0]}, [r6]
157	157
158		− ldmia sp!, {r4-r8}
	158	+ ldmia sp!, {r4-r8}
159	159
160	160	WELS_ASM_FUNC_END
161	161
162	162
163	163	#ifdef __APPLE__
164	164	.macro SAD_SD_MAD_16BYTES
165		− vld1.32 {q0}, [$0], $2
166		− vld1.32 {q1}, [$1], $2
	165	+ vld1.32 {q0}, [$0], $2
	166	+ vld1.32 {q1}, [$1], $2
167	167
168		− vpadal.u8 $3, q0
169		− vpadal.u8 $4, q1
	168	+ vpadal.u8 $3, q0
	169	+ vpadal.u8 $4, q1
170	170
171		− vabd.u8 q0, q0, q1
172		− vmax.u8 $5, q0
173		− vpadal.u8 $6, q0
	171	+ vabd.u8 q0, q0, q1
	172	+ vmax.u8 $5, q0
	173	+ vpadal.u8 $6, q0
174	174	.endm
175	175
176	176	.macro SAD_SD_MAD_8x16BYTES
177		− vld1.32 {q0}, [$0], $2
178		− vld1.32 {q1}, [$1], $2
	177	+ vld1.32 {q0}, [$0], $2
	178	+ vld1.32 {q1}, [$1], $2
179	179
180		− vpaddl.u8 q2, q0
181		− vpaddl.u8 q3, q1
	180	+ vpaddl.u8 q2, q0
	181	+ vpaddl.u8 q3, q1
182	182
183		− vabd.u8 $3, q0, q1
184		− vpaddl.u8 $4, $3 //abs_diff
	183	+ vabd.u8 $3, q0, q1
	184	+ vpaddl.u8 $4, $3 //abs_diff
185	185
186	186
187	187	SAD_SD_MAD_16BYTES $0,$1,$2,q2,q3,$3,$4
@@ -192,41 +192,41 @@ WELS_ASM_FUNC_END · 4 unmodified lines
192	192	SAD_SD_MAD_16BYTES $0,$1,$2,q2,q3,$3,$4
193	193	SAD_SD_MAD_16BYTES $0,$1,$2,q2,q3,$3,$4
194	194
195		− vsub.u16 $5, q2, q3
	195	+ vsub.u16 $5, q2, q3
196	196	.endm
197	197
198	198	.macro SAD_SD_MAD_CALC
199		− vpmax.u8 d0, $0, $1 //8bytes
200		− vpmax.u8 d0, d0, d0 //4bytes
201		− vpmax.u8 $2, d0, d0 //2bytes
202		−
203		− vpaddl.u16 $3, $3
204		− vpaddl.u32 $3, $3
205		− vpaddl.s16 $4, $4
206		− vpaddl.s32 $4, $4
	199	+ vpmax.u8 d0, $0, $1 //8bytes
	200	+ vpmax.u8 d0, d0, d0 //4bytes
	201	+ vpmax.u8 $2, d0, d0 //2bytes
	202	+
	203	+ vpaddl.u16 $3, $3
	204	+ vpaddl.u32 $3, $3
	205	+ vpaddl.s16 $4, $4
	206	+ vpaddl.s32 $4, $4
207	207	.endm
208	208	#else
209	209	.macro SAD_SD_MAD_16BYTES arg0, arg1, arg2, arg3, arg4, arg5, arg6
210		− vld1.32 {q0}, [\arg0], \arg2
211		− vld1.32 {q1}, [\arg1], \arg2
	210	+ vld1.32 {q0}, [\arg0], \arg2
	211	+ vld1.32 {q1}, [\arg1], \arg2
212	212
213		− vpadal.u8 \arg3, q0
214		− vpadal.u8 \arg4, q1
	213	+ vpadal.u8 \arg3, q0
	214	+ vpadal.u8 \arg4, q1
215	215
216		− vabd.u8 q0, q0, q1
217		− vmax.u8 \arg5, q0
218		− vpadal.u8 \arg6, q0
	216	+ vabd.u8 q0, q0, q1
	217	+ vmax.u8 \arg5, q0
	218	+ vpadal.u8 \arg6, q0
219	219	.endm
220	220
221	221	.macro SAD_SD_MAD_8x16BYTES arg0, arg1, arg2, arg3, arg4, arg5
222		− vld1.32 {q0}, [\arg0], \arg2
223		− vld1.32 {q1}, [\arg1], \arg2
	222	+ vld1.32 {q0}, [\arg0], \arg2
	223	+ vld1.32 {q1}, [\arg1], \arg2
224	224
225		− vpaddl.u8 q2, q0
226		− vpaddl.u8 q3, q1
	225	+ vpaddl.u8 q2, q0
	226	+ vpaddl.u8 q3, q1
227	227
228		− vabd.u8 \arg3, q0, q1
229		− vpaddl.u8 \arg4, \arg3 //abs_diff
	228	+ vabd.u8 \arg3, q0, q1
	229	+ vpaddl.u8 \arg4, \arg3 //abs_diff
230	230
231	231
232	232	SAD_SD_MAD_16BYTES \arg0,\arg1,\arg2,q2,q3,\arg3,\arg4
@@ -237,18 +237,18 @@ WELS_ASM_FUNC_END · 4 unmodified lines
237	237	SAD_SD_MAD_16BYTES \arg0,\arg1,\arg2,q2,q3,\arg3,\arg4
238	238	SAD_SD_MAD_16BYTES \arg0,\arg1,\arg2,q2,q3,\arg3,\arg4
239	239
240		− vsub.u16 \arg5, q2, q3
	240	+ vsub.u16 \arg5, q2, q3
241	241	.endm
242	242
243	243	.macro SAD_SD_MAD_CALC arg0, arg1, arg2, arg3, arg4
244		− vpmax.u8 d0, \arg0, \arg1 //8bytes
245		− vpmax.u8 d0, d0, d0 //4bytes
246		− vpmax.u8 \arg2, d0, d0 //2bytes
247		−
248		− vpaddl.u16 \arg3, \arg3
249		− vpaddl.u32 \arg3, \arg3
250		− vpaddl.s16 \arg4, \arg4
251		− vpaddl.s32 \arg4, \arg4
	244	+ vpmax.u8 d0, \arg0, \arg1 //8bytes
	245	+ vpmax.u8 d0, d0, d0 //4bytes
	246	+ vpmax.u8 \arg2, d0, d0 //2bytes
	247	+
	248	+ vpaddl.u16 \arg3, \arg3
	249	+ vpaddl.u32 \arg3, \arg3
	250	+ vpaddl.s16 \arg4, \arg4
	251	+ vpaddl.s32 \arg4, \arg4
252	252	.endm
253	253	#endif
254	254
@@ -256,18 +256,18 @@ WELS_ASM_FUNC_BEGIN VAACalcSadBgd_neon · 1 unmodified line
256	256
257	257	stmdb sp!, {r4-r10}
258	258
259		− ldr r4, [sp, #28] //load pic_stride
260		− ldr r5, [sp, #36] //load psad8x8
	259	+ ldr r4, [sp, #28] //load pic_stride
	260	+ ldr r5, [sp, #36] //load psad8x8
261	261	ldr r6, [sp, #40] //load psd8x8
262	262	ldr r7, [sp, #44] //load pmad8x8
263	263
264		− //Initial the Q4 register for save the "psadframe"
265		− vmov.s64 q15, #0
	264	+ //Initial the Q4 register for save the "psadframe"
	265	+ vmov.s64 q15, #0
266	266
267		− //Get the jump distance to use on loop codes
268		− lsl r10, r4, #4
269		− sub r9, r10, #16 //R9 keep the 16*pic_stride-16
270		− sub r10, r2 //R10 keep the 16*pic_stride-pic_width
	267	+ //Get the jump distance to use on loop codes
	268	+ lsl r10, r4, #4
	269	+ sub r9, r10, #16 //R9 keep the 16*pic_stride-16
	270	+ sub r10, r2 //R10 keep the 16*pic_stride-pic_width
271	271
272	272	vaa_calc_sad_bgd_loop0:
273	273
@@ -276,391 +276,391 @@ vaa_calc_sad_bgd_loop0: · 2 unmodified lines
276	276
277	277	vaa_calc_sad_bgd_loop1:
278	278
279		− //Process the 16x16 bytes pmad psad psd
280		− SAD_SD_MAD_8x16BYTES r0, r1, r4, q13, q11, q9
281		− SAD_SD_MAD_8x16BYTES r0, r1, r4, q14, q12, q10
	279	+ //Process the 16x16 bytes pmad psad psd
	280	+ SAD_SD_MAD_8x16BYTES r0, r1, r4, q13, q11, q9
	281	+ SAD_SD_MAD_8x16BYTES r0, r1, r4, q14, q12, q10
282	282
283	283	SAD_SD_MAD_CALC d26, d27, d16, q11, q9
284	284	SAD_SD_MAD_CALC d28, d29, d17, q12, q10
285	285
286		− //Write to "psad8x8" buffer
287		− vst4.32 {d22[0],d23[0],d24[0],d25[0]}, [r5]!
288		− //Adjust the input address
289		− sub r0, r9
290		− sub r1, r9
291		− //Write to "psd8x8" buffer
292		− vst4.32 {d18[0],d19[0],d20[0],d21[0]}, [r6]!
293		− subs r8, #16
294		− //Write to "pmad8x8" buffer
295		− vst2.16 {d16[0],d17[0]}, [r7]!
296		− //Save to calculate "psadframe"
297		− vadd.u32 q11, q12
298		− vadd.u32 q15, q11
299		−
300		− bne vaa_calc_sad_bgd_loop1
301		−
302		− //Adjust the input address
303		− add r0, r10
304		− add r1, r10
	286	+ //Write to "psad8x8" buffer
	287	+ vst4.32 {d22[0],d23[0],d24[0],d25[0]}, [r5]!
	288	+ //Adjust the input address
	289	+ sub r0, r9
	290	+ sub r1, r9
	291	+ //Write to "psd8x8" buffer
	292	+ vst4.32 {d18[0],d19[0],d20[0],d21[0]}, [r6]!
	293	+ subs r8, #16
	294	+ //Write to "pmad8x8" buffer
	295	+ vst2.16 {d16[0],d17[0]}, [r7]!
	296	+ //Save to calculate "psadframe"
	297	+ vadd.u32 q11, q12
	298	+ vadd.u32 q15, q11
	299	+
	300	+ bne vaa_calc_sad_bgd_loop1
	301	+
	302	+ //Adjust the input address
	303	+ add r0, r10
	304	+ add r1, r10
305	305
306	306	subs r3, #16
307		− bne vaa_calc_sad_bgd_loop0
	307	+ bne vaa_calc_sad_bgd_loop0
308	308
309		− ldr r8, [sp, #32] //load psadframe
310		− vadd.u32 d30, d31
311		− vst1.32 {d30[0]}, [r8]
312		− ldmia sp!, {r4-r10}
	309	+ ldr r8, [sp, #32] //load psadframe
	310	+ vadd.u32 d30, d31
	311	+ vst1.32 {d30[0]}, [r8]
	312	+ ldmia sp!, {r4-r10}
313	313
314	314	WELS_ASM_FUNC_END
315	315
316	316
317	317	#ifdef __APPLE__
318	318	.macro SSD_MUL_SUM_16BYTES_RESET
319		− vmull.u8 $3, $0, $0
320		− vpaddl.u16 $2, $3
	319	+ vmull.u8 $3, $0, $0
	320	+ vpaddl.u16 $2, $3
321	321
322		− vmull.u8 $3, $1, $1
323		− vpadal.u16 $2, $3
	322	+ vmull.u8 $3, $1, $1
	323	+ vpadal.u16 $2, $3
324	324	.endm
325	325
326	326	.macro SSD_MUL_SUM_16BYTES
327		− vmull.u8 $3, $0, $0
328		− vpadal.u16 $2, $3
	327	+ vmull.u8 $3, $0, $0
	328	+ vpadal.u16 $2, $3
329	329
330		− vmull.u8 $3, $1, $1
331		− vpadal.u16 $2, $3
	330	+ vmull.u8 $3, $1, $1
	331	+ vpadal.u16 $2, $3
332	332	.endm
333	333
334	334	.macro SAD_SSD_BGD_16
335		− vld1.8 {q0}, [$0], $2 //load cur_row
	335	+ vld1.8 {q0}, [$0], $2 //load cur_row
336	336
337		− vpadal.u8 q3, q0 //add cur_row together
338		− vpadal.u8 q4, q1 //add ref_row together
	337	+ vpadal.u8 q3, q0 //add cur_row together
	338	+ vpadal.u8 q4, q1 //add ref_row together
339	339
340		− vabd.u8 q2, q0, q1 //abs_diff
	340	+ vabd.u8 q2, q0, q1 //abs_diff
341	341
342		− vmax.u8 q5, q2 //l_mad for 16 bytes reset for every 8x16
	342	+ vmax.u8 q5, q2 //l_mad for 16 bytes reset for every 8x16
343	343
344		− vpadal.u8 $3, q2 //l_sad for 16 bytes reset for every 8x16
	344	+ vpadal.u8 $3, q2 //l_sad for 16 bytes reset for every 8x16
345	345
346		− SSD_MUL_SUM_16BYTES d4,d5, q8, q11 //q8 for l_sqiff reset for every 16x16
	346	+ SSD_MUL_SUM_16BYTES d4,d5, q8, q11 //q8 for l_sqiff reset for every 16x16
347	347
348		− vld1.8 {q1}, [$1], $2 //load ref_row
349		− vpadal.u8 q9, q0 //q9 for l_sum reset for every 16x16
	348	+ vld1.8 {q1}, [$1], $2 //load ref_row
	349	+ vpadal.u8 q9, q0 //q9 for l_sum reset for every 16x16
350	350
351		− SSD_MUL_SUM_16BYTES d0,d1, q10, q11 //q10 for lsqsum reset for every 16x16
	351	+ SSD_MUL_SUM_16BYTES d0,d1, q10, q11 //q10 for lsqsum reset for every 16x16
352	352	.endm
353	353
354	354	//the last row of a 16x16 block
355	355	.macro SAD_SSD_BGD_16_end
356		− vld1.8 {q0}, [$0], $1 //load cur_row
	356	+ vld1.8 {q0}, [$0], $1 //load cur_row
357	357
358		− vpadal.u8 q3, q0 //add cur_row together
359		− vpadal.u8 q4, q1 //add ref_row together
	358	+ vpadal.u8 q3, q0 //add cur_row together
	359	+ vpadal.u8 q4, q1 //add ref_row together
360	360
361		− vabd.u8 q2, q0, q1 //abs_diff
	361	+ vabd.u8 q2, q0, q1 //abs_diff
362	362
363		− vmax.u8 q5, q2 //l_mad for 16 bytes reset for every 8x16
	363	+ vmax.u8 q5, q2 //l_mad for 16 bytes reset for every 8x16
364	364
365		− vpadal.u8 $2, q2 //l_sad for 16 bytes reset for every 8x16
	365	+ vpadal.u8 $2, q2 //l_sad for 16 bytes reset for every 8x16
366	366
367		− SSD_MUL_SUM_16BYTES d4,d5, q8, q11 //q8 for l_sqiff reset for every 16x16
	367	+ SSD_MUL_SUM_16BYTES d4,d5, q8, q11 //q8 for l_sqiff reset for every 16x16
368	368
369		− vpadal.u8 q9, q0 //q9 for l_sum reset for every 16x16
	369	+ vpadal.u8 q9, q0 //q9 for l_sum reset for every 16x16
370	370
371		− SSD_MUL_SUM_16BYTES d0,d1, q10, q11 //q10 for lsqsum reset for every 16x16
	371	+ SSD_MUL_SUM_16BYTES d0,d1, q10, q11 //q10 for lsqsum reset for every 16x16
372	372	.endm
373	373
374	374	//for the begin of a 8x16 block, use some instructions to reset the register
375	375	.macro SAD_SSD_BGD_16_RESET_8x8
376		− vld1.8 {q0}, [$0], $2 //load cur_row
	376	+ vld1.8 {q0}, [$0], $2 //load cur_row
377	377
378		− vpaddl.u8 q3, q0 //add cur_row together
379		− vpaddl.u8 q4, q1 //add ref_row together
	378	+ vpaddl.u8 q3, q0 //add cur_row together
	379	+ vpaddl.u8 q4, q1 //add ref_row together
380	380
381		− vabd.u8 q2, q0, q1 //abs_diff
	381	+ vabd.u8 q2, q0, q1 //abs_diff
382	382
383		− vmov q5,q2 //calculate max and avoid reset to zero, l_mad for 16 bytes reset for every 8x16
	383	+ vmov q5,q2 //calculate max and avoid reset to zero, l_mad for 16 bytes reset for every 8x16
384	384
385		− vpaddl.u8 $3, q2 //l_sad for 16 bytes reset for every 8x16
	385	+ vpaddl.u8 $3, q2 //l_sad for 16 bytes reset for every 8x16
386	386
387	387
388		− SSD_MUL_SUM_16BYTES d4,d5, q8, q11 //q8 for l_sqiff reset for every 16x16
	388	+ SSD_MUL_SUM_16BYTES d4,d5, q8, q11 //q8 for l_sqiff reset for every 16x16
389	389
390		− vld1.8 {q1}, [$1], $2 //load ref_row
	390	+ vld1.8 {q1}, [$1], $2 //load ref_row
391	391
392		− vpadal.u8 q9, q0 //q9 for l_sum reset for every 16x16
	392	+ vpadal.u8 q9, q0 //q9 for l_sum reset for every 16x16
393	393
394		− SSD_MUL_SUM_16BYTES d0,d1, q10, q11 //q10 for lsqsum reset for every 16x16
	394	+ SSD_MUL_SUM_16BYTES d0,d1, q10, q11 //q10 for lsqsum reset for every 16x16
395	395	.endm
396	396
397	397	//for the begin of a 16x16 block, use some instructions to reset the register
398	398	.macro SAD_SSD_BGD_16_RESET_16x16
399		− vld1.8 {q0}, [$0], $2 //load cur_row
400		− vld1.8 {q1}, [$1], $2 //load ref_row
	399	+ vld1.8 {q0}, [$0], $2 //load cur_row
	400	+ vld1.8 {q1}, [$1], $2 //load ref_row
401	401
402		− vpaddl.u8 q3, q0 //add cur_row together
403		− vpaddl.u8 q4, q1 //add ref_row together
	402	+ vpaddl.u8 q3, q0 //add cur_row together
	403	+ vpaddl.u8 q4, q1 //add ref_row together
404	404
405		− vabd.u8 q2, q0, q1 //abs_diff
	405	+ vabd.u8 q2, q0, q1 //abs_diff
406	406
407		− vmov q5,q2 //calculate max and avoid reset to zero, l_mad for 16 bytes reset for every 8x16
	407	+ vmov q5,q2 //calculate max and avoid reset to zero, l_mad for 16 bytes reset for every 8x16
408	408
409		− vpaddl.u8 $3, q2 //l_sad for 16 bytes reset for every 8x16
	409	+ vpaddl.u8 $3, q2 //l_sad for 16 bytes reset for every 8x16
410	410
411		− SSD_MUL_SUM_16BYTES_RESET d4,d5,q8, q11 //q8 for l_sqiff reset for every 16x16
	411	+ SSD_MUL_SUM_16BYTES_RESET d4,d5,q8, q11 //q8 for l_sqiff reset for every 16x16
412	412
413		− vld1.8 {q1}, [$1], $2 //load ref_row
	413	+ vld1.8 {q1}, [$1], $2 //load ref_row
414	414
415		− vpaddl.u8 q9, q0 //q9 for l_sum reset for every 16x16
	415	+ vpaddl.u8 q9, q0 //q9 for l_sum reset for every 16x16
416	416
417		− SSD_MUL_SUM_16BYTES_RESET d0,d1,q10,q11 //q10 for lsqsum reset for every 16x16
	417	+ SSD_MUL_SUM_16BYTES_RESET d0,d1,q10,q11 //q10 for lsqsum reset for every 16x16
418	418	.endm
419	419
420	420	//for each 8x16 block
421	421	.macro SAD_SSD_BGD_CALC_8x16
422	422
423		− vpmax.u8 d10, d10, d11 //4 numbers
424		− vpmax.u8 d10, d10, d10 //2 numbers
425		− vpmax.u8 d10, d10, d10 //1 number1
	423	+ vpmax.u8 d10, d10, d11 //4 numbers
	424	+ vpmax.u8 d10, d10, d10 //2 numbers
	425	+ vpmax.u8 d10, d10, d10 //1 number1
426	426
427		− vmov $0, d10 //d26 d27 keeps the l_mad
	427	+ vmov $0, d10 //d26 d27 keeps the l_mad
428	428
429		− //p_sd8x8 fix me
430		− vpaddl.u16 q3, q3
431		− vpaddl.u16 q4, q4
	429	+ //p_sd8x8 fix me
	430	+ vpaddl.u16 q3, q3
	431	+ vpaddl.u16 q4, q4
432	432
433		− vsub.i32 $1, q3, q4
434		− vpaddl.u32 $1, $1
	433	+ vsub.i32 $1, q3, q4
	434	+ vpaddl.u32 $1, $1
435	435
436		− //psad8x8
437		− vpaddl.u16 $2, $2
438		− vpaddl.u32 $2, $2
	436	+ //psad8x8
	437	+ vpaddl.u16 $2, $2
	438	+ vpaddl.u32 $2, $2
439	439
440		− //psadframe
441		− vadd.i32 q12, $2
	440	+ //psadframe
	441	+ vadd.i32 q12, $2
442	442	.endm
443	443
444	444	.macro SAD_SSD_BGD_16x16
445		− //for one 8x16
446		− SAD_SSD_BGD_16_RESET_16x16 $0, $1, $2, q6
447		− SAD_SSD_BGD_16 $0, $1, $2, q6
448		− SAD_SSD_BGD_16 $0, $1, $2, q6
449		− SAD_SSD_BGD_16 $0, $1, $2, q6
450		− SAD_SSD_BGD_16 $0, $1, $2, q6
451		− SAD_SSD_BGD_16 $0, $1, $2, q6
452		− SAD_SSD_BGD_16 $0, $1, $2, q6
453		− SAD_SSD_BGD_16 $0, $1, $2, q6
454		−
455		− SAD_SSD_BGD_CALC_8x16 d26, q14, q6
456		−
457		− //for another 8x16
458		− SAD_SSD_BGD_16_RESET_8x8 $0, $1, $2, q7
459		− SAD_SSD_BGD_16 $0, $1, $2, q7
460		− SAD_SSD_BGD_16 $0, $1, $2, q7
461		− SAD_SSD_BGD_16 $0, $1, $2, q7
462		− SAD_SSD_BGD_16 $0, $1, $2, q7
463		− SAD_SSD_BGD_16 $0, $1, $2, q7
464		− SAD_SSD_BGD_16 $0, $1, $2, q7
465		− SAD_SSD_BGD_16_end $0, $2, q7
466		−
467		− SAD_SSD_BGD_CALC_8x16 d27, q15, q7
	445	+ //for one 8x16
	446	+ SAD_SSD_BGD_16_RESET_16x16 $0, $1, $2, q6
	447	+ SAD_SSD_BGD_16 $0, $1, $2, q6
	448	+ SAD_SSD_BGD_16 $0, $1, $2, q6
	449	+ SAD_SSD_BGD_16 $0, $1, $2, q6
	450	+ SAD_SSD_BGD_16 $0, $1, $2, q6
	451	+ SAD_SSD_BGD_16 $0, $1, $2, q6
	452	+ SAD_SSD_BGD_16 $0, $1, $2, q6
	453	+ SAD_SSD_BGD_16 $0, $1, $2, q6
	454	+
	455	+ SAD_SSD_BGD_CALC_8x16 d26, q14, q6
	456	+
	457	+ //for another 8x16
	458	+ SAD_SSD_BGD_16_RESET_8x8 $0, $1, $2, q7
	459	+ SAD_SSD_BGD_16 $0, $1, $2, q7
	460	+ SAD_SSD_BGD_16 $0, $1, $2, q7
	461	+ SAD_SSD_BGD_16 $0, $1, $2, q7
	462	+ SAD_SSD_BGD_16 $0, $1, $2, q7
	463	+ SAD_SSD_BGD_16 $0, $1, $2, q7
	464	+ SAD_SSD_BGD_16 $0, $1, $2, q7
	465	+ SAD_SSD_BGD_16_end $0, $2, q7
	466	+
	467	+ SAD_SSD_BGD_CALC_8x16 d27, q15, q7
468	468	.endm
469	469
470	470	.macro SSD_SAD_SD_MAD_PADDL
471		− vpaddl.s16 $0, $0
472		− vpaddl.s32 $0, $0
473		− vadd.i32 $1, $1, $2
	471	+ vpaddl.s16 $0, $0
	472	+ vpaddl.s32 $0, $0
	473	+ vadd.i32 $1, $1, $2
474	474	.endm
475	475	#else
476	476	.macro SSD_MUL_SUM_16BYTES_RESET arg0, arg1, arg2, arg3
477		− vmull.u8 \arg3, \arg0, \arg0
478		− vpaddl.u16 \arg2, \arg3
	477	+ vmull.u8 \arg3, \arg0, \arg0
	478	+ vpaddl.u16 \arg2, \arg3
479	479
480		− vmull.u8 \arg3, \arg1, \arg1
481		− vpadal.u16 \arg2, \arg3
	480	+ vmull.u8 \arg3, \arg1, \arg1
	481	+ vpadal.u16 \arg2, \arg3
482	482	.endm
483	483
484	484	.macro SSD_MUL_SUM_16BYTES arg0, arg1, arg2, arg3
485		− vmull.u8 \arg3, \arg0, \arg0
486		− vpadal.u16 \arg2, \arg3
	485	+ vmull.u8 \arg3, \arg0, \arg0
	486	+ vpadal.u16 \arg2, \arg3
487	487
488		− vmull.u8 \arg3, \arg1, \arg1
489		− vpadal.u16 \arg2, \arg3
	488	+ vmull.u8 \arg3, \arg1, \arg1
	489	+ vpadal.u16 \arg2, \arg3
490	490	.endm
491	491
492	492	.macro SAD_SSD_BGD_16 arg0, arg1, arg2, arg3
493		− vld1.8 {q0}, [\arg0], \arg2 //load cur_row
	493	+ vld1.8 {q0}, [\arg0], \arg2 //load cur_row
494	494
495		− vpadal.u8 q3, q0 //add cur_row together
496		− vpadal.u8 q4, q1 //add ref_row together
	495	+ vpadal.u8 q3, q0 //add cur_row together
	496	+ vpadal.u8 q4, q1 //add ref_row together
497	497
498		− vabd.u8 q2, q0, q1 //abs_diff
	498	+ vabd.u8 q2, q0, q1 //abs_diff
499	499
500		− vmax.u8 q5, q2 //l_mad for 16 bytes reset for every 8x16
	500	+ vmax.u8 q5, q2 //l_mad for 16 bytes reset for every 8x16
501	501
502		− vpadal.u8 \arg3, q2 //l_sad for 16 bytes reset for every 8x16
	502	+ vpadal.u8 \arg3, q2 //l_sad for 16 bytes reset for every 8x16
503	503
504		− SSD_MUL_SUM_16BYTES d4,d5, q8, q11 //q8 for l_sqiff reset for every 16x16
	504	+ SSD_MUL_SUM_16BYTES d4,d5, q8, q11 //q8 for l_sqiff reset for every 16x16
505	505
506		− vld1.8 {q1}, [\arg1], \arg2 //load ref_row
507		− vpadal.u8 q9, q0 //q9 for l_sum reset for every 16x16
	506	+ vld1.8 {q1}, [\arg1], \arg2 //load ref_row
	507	+ vpadal.u8 q9, q0 //q9 for l_sum reset for every 16x16
508	508
509		− SSD_MUL_SUM_16BYTES d0,d1, q10, q11 //q10 for lsqsum reset for every 16x16
	509	+ SSD_MUL_SUM_16BYTES d0,d1, q10, q11 //q10 for lsqsum reset for every 16x16
510	510	.endm
511	511
512	512	//the last row of a 16x16 block
513	513	.macro SAD_SSD_BGD_16_end arg0, arg1, arg2
514		− vld1.8 {q0}, [\arg0], \arg1 //load cur_row
	514	+ vld1.8 {q0}, [\arg0], \arg1 //load cur_row
515	515
516		− vpadal.u8 q3, q0 //add cur_row together
517		− vpadal.u8 q4, q1 //add ref_row together
	516	+ vpadal.u8 q3, q0 //add cur_row together
	517	+ vpadal.u8 q4, q1 //add ref_row together
518	518
519		− vabd.u8 q2, q0, q1 //abs_diff
	519	+ vabd.u8 q2, q0, q1 //abs_diff
520	520
521		− vmax.u8 q5, q2 //l_mad for 16 bytes reset for every 8x16
	521	+ vmax.u8 q5, q2 //l_mad for 16 bytes reset for every 8x16
522	522
523		− vpadal.u8 \arg2, q2 //l_sad for 16 bytes reset for every 8x16
	523	+ vpadal.u8 \arg2, q2 //l_sad for 16 bytes reset for every 8x16
524	524
525		− SSD_MUL_SUM_16BYTES d4,d5, q8, q11 //q8 for l_sqiff reset for every 16x16
	525	+ SSD_MUL_SUM_16BYTES d4,d5, q8, q11 //q8 for l_sqiff reset for every 16x16
526	526
527		− vpadal.u8 q9, q0 //q9 for l_sum reset for every 16x16
	527	+ vpadal.u8 q9, q0 //q9 for l_sum reset for every 16x16
528	528
529		− SSD_MUL_SUM_16BYTES d0,d1, q10, q11 //q10 for lsqsum reset for every 16x16
	529	+ SSD_MUL_SUM_16BYTES d0,d1, q10, q11 //q10 for lsqsum reset for every 16x16
530	530	.endm
531	531
532	532	//for the begin of a 8x16 block, use some instructions to reset the register
533	533	.macro SAD_SSD_BGD_16_RESET_8x8 arg0, arg1, arg2, arg3
534		− vld1.8 {q0}, [\arg0], \arg2 //load cur_row
	534	+ vld1.8 {q0}, [\arg0], \arg2 //load cur_row
535	535
536		− vpaddl.u8 q3, q0 //add cur_row together
537		− vpaddl.u8 q4, q1 //add ref_row together
	536	+ vpaddl.u8 q3, q0 //add cur_row together
	537	+ vpaddl.u8 q4, q1 //add ref_row together
538	538
539		− vabd.u8 q2, q0, q1 //abs_diff
	539	+ vabd.u8 q2, q0, q1 //abs_diff
540	540
541		− vmov q5,q2 //calculate max and avoid reset to zero, l_mad for 16 bytes reset for every 8x16
	541	+ vmov q5,q2 //calculate max and avoid reset to zero, l_mad for 16 bytes reset for every 8x16
542	542
543		− vpaddl.u8 \arg3, q2 //l_sad for 16 bytes reset for every 8x16
	543	+ vpaddl.u8 \arg3, q2 //l_sad for 16 bytes reset for every 8x16
544	544
545	545
546		− SSD_MUL_SUM_16BYTES d4,d5, q8, q11 //q8 for l_sqiff reset for every 16x16
	546	+ SSD_MUL_SUM_16BYTES d4,d5, q8, q11 //q8 for l_sqiff reset for every 16x16
547	547
548		− vld1.8 {q1}, [\arg1], \arg2 //load ref_row
	548	+ vld1.8 {q1}, [\arg1], \arg2 //load ref_row
549	549
550		− vpadal.u8 q9, q0 //q9 for l_sum reset for every 16x16
	550	+ vpadal.u8 q9, q0 //q9 for l_sum reset for every 16x16
551	551
552		− SSD_MUL_SUM_16BYTES d0,d1, q10, q11 //q10 for lsqsum reset for every 16x16
	552	+ SSD_MUL_SUM_16BYTES d0,d1, q10, q11 //q10 for lsqsum reset for every 16x16
553	553	.endm
554	554
555	555	//for the begin of a 16x16 block, use some instructions to reset the register
556	556	.macro SAD_SSD_BGD_16_RESET_16x16 arg0, arg1, arg2, arg3
557		− vld1.8 {q0}, [\arg0], \arg2 //load cur_row
558		− vld1.8 {q1}, [\arg1], \arg2 //load ref_row
	557	+ vld1.8 {q0}, [\arg0], \arg2 //load cur_row
	558	+ vld1.8 {q1}, [\arg1], \arg2 //load ref_row
559	559
560		− vpaddl.u8 q3, q0 //add cur_row together
561		− vpaddl.u8 q4, q1 //add ref_row together
	560	+ vpaddl.u8 q3, q0 //add cur_row together
	561	+ vpaddl.u8 q4, q1 //add ref_row together
562	562
563		− vabd.u8 q2, q0, q1 //abs_diff
	563	+ vabd.u8 q2, q0, q1 //abs_diff
564	564
565		− vmov q5,q2 //calculate max and avoid reset to zero, l_mad for 16 bytes reset for every 8x16
	565	+ vmov q5,q2 //calculate max and avoid reset to zero, l_mad for 16 bytes reset for every 8x16
566	566
567		− vpaddl.u8 \arg3, q2 //l_sad for 16 bytes reset for every 8x16
	567	+ vpaddl.u8 \arg3, q2 //l_sad for 16 bytes reset for every 8x16
568	568
569		− SSD_MUL_SUM_16BYTES_RESET d4,d5,q8, q11 //q8 for l_sqiff reset for every 16x16
	569	+ SSD_MUL_SUM_16BYTES_RESET d4,d5,q8, q11 //q8 for l_sqiff reset for every 16x16
570	570
571		− vld1.8 {q1}, [\arg1], \arg2 //load ref_row
	571	+ vld1.8 {q1}, [\arg1], \arg2 //load ref_row
572	572
573		− vpaddl.u8 q9, q0 //q9 for l_sum reset for every 16x16
	573	+ vpaddl.u8 q9, q0 //q9 for l_sum reset for every 16x16
574	574
575		− SSD_MUL_SUM_16BYTES_RESET d0,d1,q10,q11 //q10 for lsqsum reset for every 16x16
	575	+ SSD_MUL_SUM_16BYTES_RESET d0,d1,q10,q11 //q10 for lsqsum reset for every 16x16
576	576	.endm
577	577
578	578	//for each 8x16 block
579	579	.macro SAD_SSD_BGD_CALC_8x16 arg0, arg1, arg2
580	580
581		− vpmax.u8 d10, d10, d11 //4 numbers
582		− vpmax.u8 d10, d10, d10 //2 numbers
583		− vpmax.u8 d10, d10, d10 //1 number1
	581	+ vpmax.u8 d10, d10, d11 //4 numbers
	582	+ vpmax.u8 d10, d10, d10 //2 numbers
	583	+ vpmax.u8 d10, d10, d10 //1 number1
584	584
585		− vmov \arg0, d10 //d26 d27 keeps the l_mad
	585	+ vmov \arg0, d10 //d26 d27 keeps the l_mad
586	586
587		− //p_sd8x8
588		− vpaddl.u16 q3, q3
589		− vpaddl.u16 q4, q4
	587	+ //p_sd8x8
	588	+ vpaddl.u16 q3, q3
	589	+ vpaddl.u16 q4, q4
590	590
591		− vsub.i32 \arg1, q3, q4
592		− vpaddl.u32 \arg1, \arg1
	591	+ vsub.i32 \arg1, q3, q4
	592	+ vpaddl.u32 \arg1, \arg1
593	593
594		− //psad8x8
595		− vpaddl.u16 \arg2, \arg2
596		− vpaddl.u32 \arg2, \arg2
	594	+ //psad8x8
	595	+ vpaddl.u16 \arg2, \arg2
	596	+ vpaddl.u32 \arg2, \arg2
597	597
598		− //psadframe
599		− vadd.i32 q12, \arg2
	598	+ //psadframe
	599	+ vadd.i32 q12, \arg2
600	600	.endm
601	601
602	602	.macro SAD_SSD_BGD_16x16 arg0, arg1, arg2
603		− //for one 8x16
604		− SAD_SSD_BGD_16_RESET_16x16 \arg0, \arg1, \arg2, q6
605		− SAD_SSD_BGD_16 \arg0, \arg1, \arg2, q6
606		− SAD_SSD_BGD_16 \arg0, \arg1, \arg2, q6
607		− SAD_SSD_BGD_16 \arg0, \arg1, \arg2, q6
608		− SAD_SSD_BGD_16 \arg0, \arg1, \arg2, q6
609		− SAD_SSD_BGD_16 \arg0, \arg1, \arg2, q6
610		− SAD_SSD_BGD_16 \arg0, \arg1, \arg2, q6
611		− SAD_SSD_BGD_16 \arg0, \arg1, \arg2, q6
612		−
613		− SAD_SSD_BGD_CALC_8x16 d26, q14, q6
614		−
615		− //for another 8x16
616		− SAD_SSD_BGD_16_RESET_8x8 \arg0, \arg1, \arg2, q7
617		− SAD_SSD_BGD_16 \arg0, \arg1, \arg2, q7
618		− SAD_SSD_BGD_16 \arg0, \arg1, \arg2, q7
619		− SAD_SSD_BGD_16 \arg0, \arg1, \arg2, q7
620		− SAD_SSD_BGD_16 \arg0, \arg1, \arg2, q7
621		− SAD_SSD_BGD_16 \arg0, \arg1, \arg2, q7
622		− SAD_SSD_BGD_16 \arg0, \arg1, \arg2, q7
623		− SAD_SSD_BGD_16_end \arg0, \arg2, q7
624		−
625		− SAD_SSD_BGD_CALC_8x16 d27, q15, q7
	603	+ //for one 8x16
	604	+ SAD_SSD_BGD_16_RESET_16x16 \arg0, \arg1, \arg2, q6
	605	+ SAD_SSD_BGD_16 \arg0, \arg1, \arg2, q6
	606	+ SAD_SSD_BGD_16 \arg0, \arg1, \arg2, q6
	607	+ SAD_SSD_BGD_16 \arg0, \arg1, \arg2, q6
	608	+ SAD_SSD_BGD_16 \arg0, \arg1, \arg2, q6
	609	+ SAD_SSD_BGD_16 \arg0, \arg1, \arg2, q6
	610	+ SAD_SSD_BGD_16 \arg0, \arg1, \arg2, q6
	611	+ SAD_SSD_BGD_16 \arg0, \arg1, \arg2, q6
	612	+
	613	+ SAD_SSD_BGD_CALC_8x16 d26, q14, q6
	614	+
	615	+ //for another 8x16
	616	+ SAD_SSD_BGD_16_RESET_8x8 \arg0, \arg1, \arg2, q7
	617	+ SAD_SSD_BGD_16 \arg0, \arg1, \arg2, q7
	618	+ SAD_SSD_BGD_16 \arg0, \arg1, \arg2, q7
	619	+ SAD_SSD_BGD_16 \arg0, \arg1, \arg2, q7
	620	+ SAD_SSD_BGD_16 \arg0, \arg1, \arg2, q7
	621	+ SAD_SSD_BGD_16 \arg0, \arg1, \arg2, q7
	622	+ SAD_SSD_BGD_16 \arg0, \arg1, \arg2, q7
	623	+ SAD_SSD_BGD_16_end \arg0, \arg2, q7
	624	+
	625	+ SAD_SSD_BGD_CALC_8x16 d27, q15, q7
626	626	.endm
627	627
628	628	.macro SSD_SAD_SD_MAD_PADDL arg0, arg1, arg2
629		− vpaddl.s16 \arg0, \arg0
630		− vpaddl.s32 \arg0, \arg0
631		− vadd.i32 \arg1, \arg1, \arg2
	629	+ vpaddl.s16 \arg0, \arg0
	630	+ vpaddl.s32 \arg0, \arg0
	631	+ vadd.i32 \arg1, \arg1, \arg2
632	632	.endm
633	633	#endif
634	634
635	635
636	636	WELS_ASM_FUNC_BEGIN VAACalcSadSsdBgd_neon
637		− stmdb sp!, {r0-r12, r14}
638		− vpush {q4-q7}
	637	+ stmdb sp!, {r0-r12, r14}
	638	+ vpush {q4-q7}
639	639
640		− ldr r4, [sp, #120] //r4 keeps the pic_stride
	640	+ ldr r4, [sp, #120] //r4 keeps the pic_stride
641	641
642		− sub r5, r4, #1
643		− lsl r5, r5, #4 //r5 keeps the little step
	642	+ sub r5, r4, #1
	643	+ lsl r5, r5, #4 //r5 keeps the little step
644	644
645		− lsl r6, r4, #4
646		− sub r6, r2, r6 //r6 keeps the big step
	645	+ lsl r6, r4, #4
	646	+ sub r6, r2, r6 //r6 keeps the big step
647	647
648	648
649		− ldr r8, [sp, #128]//psad8x8
650		− ldr r9, [sp, #132]//psum16x16
651		− ldr r10, [sp, #136]//psqsum16x16
652		− ldr r11, [sp, #140]//psqdiff16x16
653		− ldr r12, [sp, #144]//p_sd8x8
654		− ldr r14, [sp, #148]//p_mad8x8
	649	+ ldr r8, [sp, #128]//psad8x8
	650	+ ldr r9, [sp, #132]//psum16x16
	651	+ ldr r10, [sp, #136]//psqsum16x16
	652	+ ldr r11, [sp, #140]//psqdiff16x16
	653	+ ldr r12, [sp, #144]//p_sd8x8
	654	+ ldr r14, [sp, #148]//p_mad8x8
655	655
656		− vmov.i8 q12, #0
	656	+ vmov.i8 q12, #0
657	657
658	658	vaa_calc_sad_ssd_bgd_height_loop:
659	659
660	660	mov r7, r2
661	661	vaa_calc_sad_ssd_bgd_width_loop:
662	662
663		− //l_sd q14&q15, l_mad q13, l_sad q6 & q7, l_sqdiff q8, l_sum q9, l_sqsum q10
	663	+ //l_sd q14&q15, l_mad q13, l_sad q6 & q7, l_sqdiff q8, l_sum q9, l_sqsum q10
664	664	SAD_SSD_BGD_16x16 r0,r1,r4
665	665
666	666	//psad8x8
@@ -694,243 +694,243 @@ vaa_calc_sad_ssd_bgd_width_loop: · 27 unmodified lines
694	694
695	695	bne vaa_calc_sad_ssd_bgd_width_loop
696	696
697		− sub r0, r0, r6 //jump to next 16 x width
698		− sub r1, r1, r6 //jump to next 16 x width
	697	+ sub r0, r0, r6 //jump to next 16 x width
	698	+ sub r1, r1, r6 //jump to next 16 x width
699	699
700	700	subs r3, #16
701	701	bne vaa_calc_sad_ssd_bgd_height_loop
702	702
703		− //psadframe
704		− ldr r7, [sp, #124]//psadframe
	703	+ //psadframe
	704	+ ldr r7, [sp, #124]//psadframe
705	705
706		− vadd.i32 d24, d24, d25
707		− vst1.32 {d24[0]}, [r7]
	706	+ vadd.i32 d24, d24, d25
	707	+ vst1.32 {d24[0]}, [r7]
708	708
709		− vpop {q4-q7}
710		− ldmia sp!, {r0-r12, r14}
	709	+ vpop {q4-q7}
	710	+ ldmia sp!, {r0-r12, r14}
711	711
712	712	WELS_ASM_FUNC_END
713	713
714	714
715	715	#ifdef __APPLE__
716	716	.macro SAD_VAR_16
717		− vld1.8 {q0}, [$0], $2 //load cur_row
	717	+ vld1.8 {q0}, [$0], $2 //load cur_row
718	718
719		− vpadal.u8 q3, q0 //add cur_row together
720		− vpadal.u8 q4, q1 //add ref_row together
	719	+ vpadal.u8 q3, q0 //add cur_row together
	720	+ vpadal.u8 q4, q1 //add ref_row together
721	721
722		− vabd.u8 q2, q0, q1 //abs_diff
	722	+ vabd.u8 q2, q0, q1 //abs_diff
723	723
724		− vpadal.u8 $3, q2 //l_sad for 16 bytes reset for every 8x16
	724	+ vpadal.u8 $3, q2 //l_sad for 16 bytes reset for every 8x16
725	725
726		− vld1.8 {q1}, [$1], $2
	726	+ vld1.8 {q1}, [$1], $2
727	727
728		− vpadal.u8 q9, q0 //q9 for l_sum reset for every 16x16
	728	+ vpadal.u8 q9, q0 //q9 for l_sum reset for every 16x16
729	729
730		− SSD_MUL_SUM_16BYTES d0,d1, q10, q11 //q10 for lsqsum reset for every 16x16
	730	+ SSD_MUL_SUM_16BYTES d0,d1, q10, q11 //q10 for lsqsum reset for every 16x16
731	731	.endm
732	732
733	733	.macro SAD_VAR_16_END
734		− vld1.8 {q0}, [$0], $1 //load cur_row
	734	+ vld1.8 {q0}, [$0], $1 //load cur_row
735	735
736		− vpadal.u8 q3, q0 //add cur_row together
737		− vpadal.u8 q4, q1 //add ref_row together
	736	+ vpadal.u8 q3, q0 //add cur_row together
	737	+ vpadal.u8 q4, q1 //add ref_row together
738	738
739		− vabd.u8 q2, q0, q1 //abs_diff
	739	+ vabd.u8 q2, q0, q1 //abs_diff
740	740
741		− vpadal.u8 $2, q2 //l_sad for 16 bytes reset for every 8x16
	741	+ vpadal.u8 $2, q2 //l_sad for 16 bytes reset for every 8x16
742	742
743		− vpadal.u8 q9, q0 //q9 for l_sum reset for every 16x16
	743	+ vpadal.u8 q9, q0 //q9 for l_sum reset for every 16x16
744	744
745		− SSD_MUL_SUM_16BYTES d0,d1, q10, q11 //q10 for lsqsum reset for every 16x16
	745	+ SSD_MUL_SUM_16BYTES d0,d1, q10, q11 //q10 for lsqsum reset for every 16x16
746	746	.endm
747	747
748	748	.macro SAD_VAR_16_RESET_16x16
749		− vld1.8 {q0}, [$0], $2 //load cur_row
750		− vld1.8 {q1}, [$1], $2
	749	+ vld1.8 {q0}, [$0], $2 //load cur_row
	750	+ vld1.8 {q1}, [$1], $2
751	751
752		− vpaddl.u8 q3, q0 //add cur_row together
753		− vpaddl.u8 q4, q1 //add ref_row together
	752	+ vpaddl.u8 q3, q0 //add cur_row together
	753	+ vpaddl.u8 q4, q1 //add ref_row together
754	754
755		− vabd.u8 q2, q0, q1 //abs_diff
	755	+ vabd.u8 q2, q0, q1 //abs_diff
756	756
757		− vpaddl.u8 $3, q2 //l_sad for 16 bytes reset for every 8x16
	757	+ vpaddl.u8 $3, q2 //l_sad for 16 bytes reset for every 8x16
758	758
759		− vld1.8 {q1}, [$1], $2
	759	+ vld1.8 {q1}, [$1], $2
760	760
761		− vpaddl.u8 q9, q0 //q9 for l_sum reset for every 16x16
	761	+ vpaddl.u8 q9, q0 //q9 for l_sum reset for every 16x16
762	762
763		− SSD_MUL_SUM_16BYTES_RESET d0,d1, q10, q11
	763	+ SSD_MUL_SUM_16BYTES_RESET d0,d1, q10, q11
764	764	.endm
765	765
766	766	.macro SAD_VAR_16_RESET_8x8
767		− vld1.8 {q0}, [$0], $2 //load cur_row
	767	+ vld1.8 {q0}, [$0], $2 //load cur_row
768	768
769		− vpaddl.u8 q3, q0 //add cur_row together
770		− vpaddl.u8 q4, q1 //add ref_row together
	769	+ vpaddl.u8 q3, q0 //add cur_row together
	770	+ vpaddl.u8 q4, q1 //add ref_row together
771	771
772		− vabd.u8 q2, q0, q1 //abs_diff
	772	+ vabd.u8 q2, q0, q1 //abs_diff
773	773
774		− vpaddl.u8 $3, q2 //l_sad for 16 bytes reset for every 8x16
	774	+ vpaddl.u8 $3, q2 //l_sad for 16 bytes reset for every 8x16
775	775
776		− vld1.8 {q1}, [$1], $2
	776	+ vld1.8 {q1}, [$1], $2
777	777
778		− vpadal.u8 q9, q0 //q9 for l_sum reset for every 16x16
	778	+ vpadal.u8 q9, q0 //q9 for l_sum reset for every 16x16
779	779
780		− SSD_MUL_SUM_16BYTES d0,d1, q10, q11 //q10 for lsqsum reset for every 16x16
	780	+ SSD_MUL_SUM_16BYTES d0,d1, q10, q11 //q10 for lsqsum reset for every 16x16
781	781	.endm
782	782
783	783	.macro SAD_VAR_16x16
784		− //for one 8x16
785		− SAD_VAR_16_RESET_16x16 $0, $1, $2, q6
786		− SAD_VAR_16 $0, $1, $2, q6
787		− SAD_VAR_16 $0, $1, $2, q6
788		− SAD_VAR_16 $0, $1, $2, q6
789		− SAD_VAR_16 $0, $1, $2, q6
790		− SAD_VAR_16 $0, $1, $2, q6
791		− SAD_VAR_16 $0, $1, $2, q6
792		− SAD_VAR_16 $0, $1, $2, q6
793		−
794		− vpaddl.u16 q6, q6
795		− vpaddl.u32 q6, q6
796		− vadd.i32 q12, q6
797		−
798		− //for another 8x16
799		− SAD_VAR_16_RESET_8x8 $0, $1, $2, q7
800		− SAD_VAR_16 $0, $1, $2, q7
801		− SAD_VAR_16 $0, $1, $2, q7
802		− SAD_VAR_16 $0, $1, $2, q7
803		− SAD_VAR_16 $0, $1, $2, q7
804		− SAD_VAR_16 $0, $1, $2, q7
805		− SAD_VAR_16 $0, $1, $2, q7
806		− SAD_VAR_16_END $0, $2, q7
807		−
808		− vpaddl.u16 q7, q7
809		− vpaddl.u32 q7, q7
810		−
811		− vadd.i32 q12, q7
	784	+ //for one 8x16
	785	+ SAD_VAR_16_RESET_16x16 $0, $1, $2, q6
	786	+ SAD_VAR_16 $0, $1, $2, q6
	787	+ SAD_VAR_16 $0, $1, $2, q6
	788	+ SAD_VAR_16 $0, $1, $2, q6
	789	+ SAD_VAR_16 $0, $1, $2, q6
	790	+ SAD_VAR_16 $0, $1, $2, q6
	791	+ SAD_VAR_16 $0, $1, $2, q6
	792	+ SAD_VAR_16 $0, $1, $2, q6
	793	+
	794	+ vpaddl.u16 q6, q6
	795	+ vpaddl.u32 q6, q6
	796	+ vadd.i32 q12, q6
	797	+
	798	+ //for another 8x16
	799	+ SAD_VAR_16_RESET_8x8 $0, $1, $2, q7
	800	+ SAD_VAR_16 $0, $1, $2, q7
	801	+ SAD_VAR_16 $0, $1, $2, q7
	802	+ SAD_VAR_16 $0, $1, $2, q7
	803	+ SAD_VAR_16 $0, $1, $2, q7
	804	+ SAD_VAR_16 $0, $1, $2, q7
	805	+ SAD_VAR_16 $0, $1, $2, q7
	806	+ SAD_VAR_16_END $0, $2, q7
	807	+
	808	+ vpaddl.u16 q7, q7
	809	+ vpaddl.u32 q7, q7
	810	+
	811	+ vadd.i32 q12, q7
812	812	.endm
813	813	#else
814	814	.macro SAD_VAR_16 arg0, arg1, arg2, arg3
815		− vld1.8 {q0}, [\arg0], \arg2 //load cur_row
	815	+ vld1.8 {q0}, [\arg0], \arg2 //load cur_row
816	816
817		− vpadal.u8 q3, q0 //add cur_row together
818		− vpadal.u8 q4, q1 //add ref_row together
	817	+ vpadal.u8 q3, q0 //add cur_row together
	818	+ vpadal.u8 q4, q1 //add ref_row together
819	819
820		− vabd.u8 q2, q0, q1 //abs_diff
	820	+ vabd.u8 q2, q0, q1 //abs_diff
821	821
822		− vpadal.u8 \arg3, q2 //l_sad for 16 bytes reset for every 8x16
	822	+ vpadal.u8 \arg3, q2 //l_sad for 16 bytes reset for every 8x16
823	823
824		− vld1.8 {q1}, [\arg1], \arg2
	824	+ vld1.8 {q1}, [\arg1], \arg2
825	825
826		− vpadal.u8 q9, q0 //q9 for l_sum reset for every 16x16
	826	+ vpadal.u8 q9, q0 //q9 for l_sum reset for every 16x16
827	827
828		− SSD_MUL_SUM_16BYTES d0,d1, q10, q11 //q10 for lsqsum reset for every 16x16
	828	+ SSD_MUL_SUM_16BYTES d0,d1, q10, q11 //q10 for lsqsum reset for every 16x16
829	829	.endm
830	830
831	831	.macro SAD_VAR_16_END arg0, arg1, arg2
832		− vld1.8 {q0}, [\arg0], \arg1 //load cur_row
	832	+ vld1.8 {q0}, [\arg0], \arg1 //load cur_row
833	833
834		− vpadal.u8 q3, q0 //add cur_row together
835		− vpadal.u8 q4, q1 //add ref_row together
	834	+ vpadal.u8 q3, q0 //add cur_row together
	835	+ vpadal.u8 q4, q1 //add ref_row together
836	836
837		− vabd.u8 q2, q0, q1 //abs_diff
	837	+ vabd.u8 q2, q0, q1 //abs_diff
838	838
839		− vpadal.u8 \arg2, q2 //l_sad for 16 bytes reset for every 8x16
	839	+ vpadal.u8 \arg2, q2 //l_sad for 16 bytes reset for every 8x16
840	840
841		− vpadal.u8 q9, q0 //q9 for l_sum reset for every 16x16
	841	+ vpadal.u8 q9, q0 //q9 for l_sum reset for every 16x16
842	842
843		− SSD_MUL_SUM_16BYTES d0,d1, q10, q11 //q10 for lsqsum reset for every 16x16
	843	+ SSD_MUL_SUM_16BYTES d0,d1, q10, q11 //q10 for lsqsum reset for every 16x16
844	844	.endm
845	845
846	846
847	847	.macro SAD_VAR_16_RESET_16x16 arg0, arg1, arg2, arg3
848		− vld1.8 {q0}, [\arg0], \arg2 //load cur_row
849		− vld1.8 {q1}, [\arg1], \arg2
	848	+ vld1.8 {q0}, [\arg0], \arg2 //load cur_row
	849	+ vld1.8 {q1}, [\arg1], \arg2
850	850
851		− vpaddl.u8 q3, q0 //add cur_row together
852		− vpaddl.u8 q4, q1 //add ref_row together
	851	+ vpaddl.u8 q3, q0 //add cur_row together
	852	+ vpaddl.u8 q4, q1 //add ref_row together
853	853
854		− vabd.u8 q2, q0, q1 //abs_diff
	854	+ vabd.u8 q2, q0, q1 //abs_diff
855	855
856		− vpaddl.u8 \arg3, q2 //l_sad for 16 bytes reset for every 8x16
	856	+ vpaddl.u8 \arg3, q2 //l_sad for 16 bytes reset for every 8x16
857	857
858		− vld1.8 {q1}, [\arg1], \arg2
	858	+ vld1.8 {q1}, [\arg1], \arg2
859	859
860		− vpaddl.u8 q9, q0 //q9 for l_sum reset for every 16x16
	860	+ vpaddl.u8 q9, q0 //q9 for l_sum reset for every 16x16
861	861
862		− SSD_MUL_SUM_16BYTES_RESET d0,d1, q10, q11
	862	+ SSD_MUL_SUM_16BYTES_RESET d0,d1, q10, q11
863	863	.endm
864	864
865	865	.macro SAD_VAR_16_RESET_8x8 arg0, arg1, arg2, arg3
866		− vld1.8 {q0}, [\arg0], \arg2 //load cur_row
	866	+ vld1.8 {q0}, [\arg0], \arg2 //load cur_row
867	867
868		− vpaddl.u8 q3, q0 //add cur_row together
869		− vpaddl.u8 q4, q1 //add ref_row together
	868	+ vpaddl.u8 q3, q0 //add cur_row together
	869	+ vpaddl.u8 q4, q1 //add ref_row together
870	870
871		− vabd.u8 q2, q0, q1 //abs_diff
	871	+ vabd.u8 q2, q0, q1 //abs_diff
872	872
873		− vpaddl.u8 \arg3, q2 //l_sad for 16 bytes reset for every 8x16
	873	+ vpaddl.u8 \arg3, q2 //l_sad for 16 bytes reset for every 8x16
874	874
875		− vld1.8 {q1}, [\arg1], \arg2
	875	+ vld1.8 {q1}, [\arg1], \arg2
876	876
877		− vpadal.u8 q9, q0 //q9 for l_sum reset for every 16x16
	877	+ vpadal.u8 q9, q0 //q9 for l_sum reset for every 16x16
878	878
879		− SSD_MUL_SUM_16BYTES d0,d1, q10, q11 //q10 for lsqsum reset for every 16x16
	879	+ SSD_MUL_SUM_16BYTES d0,d1, q10, q11 //q10 for lsqsum reset for every 16x16
880	880	.endm
881	881
882	882	.macro SAD_VAR_16x16 arg0, arg1, arg2
883		− //for one 8x16
884		− SAD_VAR_16_RESET_16x16 \arg0, \arg1, \arg2, q6
885		− SAD_VAR_16 \arg0, \arg1, \arg2, q6
886		− SAD_VAR_16 \arg0, \arg1, \arg2, q6
887		− SAD_VAR_16 \arg0, \arg1, \arg2, q6
888		− SAD_VAR_16 \arg0, \arg1, \arg2, q6
889		− SAD_VAR_16 \arg0, \arg1, \arg2, q6
890		− SAD_VAR_16 \arg0, \arg1, \arg2, q6
891		− SAD_VAR_16 \arg0, \arg1, \arg2, q6
892		−
893		− vpaddl.u16 q6, q6
894		− vpaddl.u32 q6, q6
895		− vadd.i32 q12, q6
896		−
897		− //for another 8x16
898		− SAD_VAR_16_RESET_8x8 \arg0, \arg1, \arg2, q7
899		− SAD_VAR_16 \arg0, \arg1, \arg2, q7
900		− SAD_VAR_16 \arg0, \arg1, \arg2, q7
901		− SAD_VAR_16 \arg0, \arg1, \arg2, q7
902		− SAD_VAR_16 \arg0, \arg1, \arg2, q7
903		− SAD_VAR_16 \arg0, \arg1, \arg2, q7
904		− SAD_VAR_16 \arg0, \arg1, \arg2, q7
905		− SAD_VAR_16_END \arg0, \arg2, q7
906		−
907		− vpaddl.u16 q7, q7
908		− vpaddl.u32 q7, q7
909		−
910		− vadd.i32 q12, q7
	883	+ //for one 8x16
	884	+ SAD_VAR_16_RESET_16x16 \arg0, \arg1, \arg2, q6
	885	+ SAD_VAR_16 \arg0, \arg1, \arg2, q6
	886	+ SAD_VAR_16 \arg0, \arg1, \arg2, q6
	887	+ SAD_VAR_16 \arg0, \arg1, \arg2, q6
	888	+ SAD_VAR_16 \arg0, \arg1, \arg2, q6
	889	+ SAD_VAR_16 \arg0, \arg1, \arg2, q6
	890	+ SAD_VAR_16 \arg0, \arg1, \arg2, q6
	891	+ SAD_VAR_16 \arg0, \arg1, \arg2, q6
	892	+
	893	+ vpaddl.u16 q6, q6
	894	+ vpaddl.u32 q6, q6
	895	+ vadd.i32 q12, q6
	896	+
	897	+ //for another 8x16
	898	+ SAD_VAR_16_RESET_8x8 \arg0, \arg1, \arg2, q7
	899	+ SAD_VAR_16 \arg0, \arg1, \arg2, q7
	900	+ SAD_VAR_16 \arg0, \arg1, \arg2, q7
	901	+ SAD_VAR_16 \arg0, \arg1, \arg2, q7
	902	+ SAD_VAR_16 \arg0, \arg1, \arg2, q7
	903	+ SAD_VAR_16 \arg0, \arg1, \arg2, q7
	904	+ SAD_VAR_16 \arg0, \arg1, \arg2, q7
	905	+ SAD_VAR_16_END \arg0, \arg2, q7
	906	+
	907	+ vpaddl.u16 q7, q7
	908	+ vpaddl.u32 q7, q7
	909	+
	910	+ vadd.i32 q12, q7
911	911	.endm
912	912	#endif
913	913
914	914
915	915	WELS_ASM_FUNC_BEGIN VAACalcSadVar_neon
916		− stmdb sp!, {r4-r11}
917		− vpush {q4}
918		− vpush {q6-q7}
	916	+ stmdb sp!, {r4-r11}
	917	+ vpush {q4}
	918	+ vpush {q6-q7}
919	919
920		− ldr r4, [sp, #80] //r4 keeps the pic_stride
	920	+ ldr r4, [sp, #80] //r4 keeps the pic_stride
921	921
922		− sub r5, r4, #1
923		− lsl r5, r5, #4 //r5 keeps the little step
	922	+ sub r5, r4, #1
	923	+ lsl r5, r5, #4 //r5 keeps the little step
924	924
925		− lsl r6, r4, #4
926		− sub r6, r2, r6 //r6 keeps the big step
	925	+ lsl r6, r4, #4
	926	+ sub r6, r2, r6 //r6 keeps the big step
927	927
928		− ldr r7, [sp, #84] //psadframe
929		− ldr r8, [sp, #88] //psad8x8
930		− ldr r9, [sp, #92] //psum16x16
931		− ldr r10, [sp, #96] //psqsum16x16
	928	+ ldr r7, [sp, #84] //psadframe
	929	+ ldr r8, [sp, #88] //psad8x8
	930	+ ldr r9, [sp, #92] //psum16x16
	931	+ ldr r10, [sp, #96] //psqsum16x16
932	932
933		− vmov.i8 q12, #0
	933	+ vmov.i8 q12, #0
934	934	vaa_calc_sad_var_height_loop:
935	935
936	936	mov r11, r2
@@ -956,154 +956,154 @@ vaa_calc_sad_var_width_loop: · 19 unmodified lines
956	956
957	957	bne vaa_calc_sad_var_width_loop
958	958
959		− sub r0, r0, r6 //jump to next 16 x width
960		− sub r1, r1, r6 //jump to next 16 x width
	959	+ sub r0, r0, r6 //jump to next 16 x width
	960	+ sub r1, r1, r6 //jump to next 16 x width
961	961
962	962	subs r3, #16
963	963	bne vaa_calc_sad_var_height_loop
964	964
965		− vadd.i32 d24, d24, d25
966		− vst1.32 {d24[0]}, [r7]
	965	+ vadd.i32 d24, d24, d25
	966	+ vst1.32 {d24[0]}, [r7]
967	967
968		− vpop {q6-q7}
969		− vpop {q4}
970		− ldmia sp!, {r4-r11}
	968	+ vpop {q6-q7}
	969	+ vpop {q4}
	970	+ ldmia sp!, {r4-r11}
971	971	WELS_ASM_FUNC_END
972	972
973	973
974	974	#ifdef __APPLE__
975	975	.macro SAD_SSD_16
976		− SAD_VAR_16 $0, $1, $2, $3
	976	+ SAD_VAR_16 $0, $1, $2, $3
977	977
978		− SSD_MUL_SUM_16BYTES d4,d5,q8, q11
	978	+ SSD_MUL_SUM_16BYTES d4,d5,q8, q11
979	979	.endm
980	980
981	981	.macro SAD_SSD_16_END
982		− SAD_VAR_16_END $0, $1, $2
	982	+ SAD_VAR_16_END $0, $1, $2
983	983
984		− SSD_MUL_SUM_16BYTES d4,d5,q8, q11 //q8 for l_sqiff reset for every 16x16
	984	+ SSD_MUL_SUM_16BYTES d4,d5,q8, q11 //q8 for l_sqiff reset for every 16x16
985	985	.endm
986	986
987	987	.macro SAD_SSD_16_RESET_16x16
988		− SAD_VAR_16_RESET_16x16 $0, $1, $2, $3
	988	+ SAD_VAR_16_RESET_16x16 $0, $1, $2, $3
989	989
990		− SSD_MUL_SUM_16BYTES_RESET d4,d5,q8, q11 //q8 for l_sqiff reset for every 16x16
	990	+ SSD_MUL_SUM_16BYTES_RESET d4,d5,q8, q11 //q8 for l_sqiff reset for every 16x16
991	991	.endm
992	992
993	993	.macro SAD_SSD_16_RESET_8x8
994		− SAD_VAR_16_RESET_8x8 $0, $1, $2, $3
	994	+ SAD_VAR_16_RESET_8x8 $0, $1, $2, $3
995	995
996		− SSD_MUL_SUM_16BYTES d4,d5,q8, q11 //q8 for l_sqiff reset for every 16x16
	996	+ SSD_MUL_SUM_16BYTES d4,d5,q8, q11 //q8 for l_sqiff reset for every 16x16
997	997	.endm
998	998
999	999	.macro SAD_SSD_16x16
1000		− //for one 8x16
1001		− SAD_SSD_16_RESET_16x16 $0, $1, $2, q6
1002		− SAD_SSD_16 $0, $1, $2, q6
1003		− SAD_SSD_16 $0, $1, $2, q6
1004		− SAD_SSD_16 $0, $1, $2, q6
1005		− SAD_SSD_16 $0, $1, $2, q6
1006		− SAD_SSD_16 $0, $1, $2, q6
1007		− SAD_SSD_16 $0, $1, $2, q6
1008		− SAD_SSD_16 $0, $1, $2, q6
1009		−
1010		− vpaddl.u16 q6, q6
1011		− vpaddl.u32 q6, q6
1012		− vadd.i32 q12, q6
1013		−
1014		− //for another 8x16
1015		− SAD_SSD_16_RESET_8x8 $0, $1, $2, q7
1016		− SAD_SSD_16 $0, $1, $2, q7
1017		− SAD_SSD_16 $0, $1, $2, q7
1018		− SAD_SSD_16 $0, $1, $2, q7
1019		− SAD_SSD_16 $0, $1, $2, q7
1020		− SAD_SSD_16 $0, $1, $2, q7
1021		− SAD_SSD_16 $0, $1, $2, q7
1022		− SAD_SSD_16_END $0, $2, q7
1023		−
1024		− vpaddl.u16 q7, q7
1025		− vpaddl.u32 q7, q7
1026		−
1027		− vadd.i32 q12, q7
	1000	+ //for one 8x16
	1001	+ SAD_SSD_16_RESET_16x16 $0, $1, $2, q6
	1002	+ SAD_SSD_16 $0, $1, $2, q6
	1003	+ SAD_SSD_16 $0, $1, $2, q6
	1004	+ SAD_SSD_16 $0, $1, $2, q6
	1005	+ SAD_SSD_16 $0, $1, $2, q6
	1006	+ SAD_SSD_16 $0, $1, $2, q6
	1007	+ SAD_SSD_16 $0, $1, $2, q6
	1008	+ SAD_SSD_16 $0, $1, $2, q6
	1009	+
	1010	+ vpaddl.u16 q6, q6
	1011	+ vpaddl.u32 q6, q6
	1012	+ vadd.i32 q12, q6
	1013	+
	1014	+ //for another 8x16
	1015	+ SAD_SSD_16_RESET_8x8 $0, $1, $2, q7
	1016	+ SAD_SSD_16 $0, $1, $2, q7
	1017	+ SAD_SSD_16 $0, $1, $2, q7
	1018	+ SAD_SSD_16 $0, $1, $2, q7
	1019	+ SAD_SSD_16 $0, $1, $2, q7
	1020	+ SAD_SSD_16 $0, $1, $2, q7
	1021	+ SAD_SSD_16 $0, $1, $2, q7
	1022	+ SAD_SSD_16_END $0, $2, q7
	1023	+
	1024	+ vpaddl.u16 q7, q7
	1025	+ vpaddl.u32 q7, q7
	1026	+
	1027	+ vadd.i32 q12, q7
1028	1028	.endm
1029	1029	#else
1030	1030	.macro SAD_SSD_16 arg0, arg1, arg2, arg3
1031		− SAD_VAR_16 \arg0, \arg1, \arg2, \arg3
	1031	+ SAD_VAR_16 \arg0, \arg1, \arg2, \arg3
1032	1032
1033		− SSD_MUL_SUM_16BYTES d4,d5,q8, q11
	1033	+ SSD_MUL_SUM_16BYTES d4,d5,q8, q11
1034	1034	.endm
1035	1035
1036	1036	.macro SAD_SSD_16_END arg0, arg1, arg2
1037		− SAD_VAR_16_END \arg0, \arg1, \arg2
	1037	+ SAD_VAR_16_END \arg0, \arg1, \arg2
1038	1038
1039		− SSD_MUL_SUM_16BYTES d4,d5,q8, q11 //q8 for l_sqiff reset for every 16x16
	1039	+ SSD_MUL_SUM_16BYTES d4,d5,q8, q11 //q8 for l_sqiff reset for every 16x16
1040	1040	.endm
1041	1041
1042	1042	.macro SAD_SSD_16_RESET_16x16 arg0, arg1, arg2, arg3
1043		− SAD_VAR_16_RESET_16x16 \arg0, \arg1, \arg2, \arg3
	1043	+ SAD_VAR_16_RESET_16x16 \arg0, \arg1, \arg2, \arg3
1044	1044
1045		− SSD_MUL_SUM_16BYTES_RESET d4,d5,q8, q11 //q8 for l_sqiff reset for every 16x16
	1045	+ SSD_MUL_SUM_16BYTES_RESET d4,d5,q8, q11 //q8 for l_sqiff reset for every 16x16
1046	1046	.endm
1047	1047
1048	1048	.macro SAD_SSD_16_RESET_8x8 arg0, arg1, arg2, arg3
1049		− SAD_VAR_16_RESET_8x8 \arg0, \arg1, \arg2, \arg3
	1049	+ SAD_VAR_16_RESET_8x8 \arg0, \arg1, \arg2, \arg3
1050	1050
1051		− SSD_MUL_SUM_16BYTES d4,d5,q8, q11 //q8 for l_sqiff reset for every 16x16
	1051	+ SSD_MUL_SUM_16BYTES d4,d5,q8, q11 //q8 for l_sqiff reset for every 16x16
1052	1052	.endm
1053	1053
1054	1054	.macro SAD_SSD_16x16 arg0, arg1, arg2
1055		− //for one 8x16
1056		− SAD_SSD_16_RESET_16x16 \arg0, \arg1, \arg2, q6
1057		− SAD_SSD_16 \arg0, \arg1, \arg2, q6
1058		− SAD_SSD_16 \arg0, \arg1, \arg2, q6
1059		− SAD_SSD_16 \arg0, \arg1, \arg2, q6
1060		− SAD_SSD_16 \arg0, \arg1, \arg2, q6
1061		− SAD_SSD_16 \arg0, \arg1, \arg2, q6
1062		− SAD_SSD_16 \arg0, \arg1, \arg2, q6
1063		− SAD_SSD_16 \arg0, \arg1, \arg2, q6
1064		−
1065		− vpaddl.u16 q6, q6
1066		− vpaddl.u32 q6, q6
1067		− vadd.i32 q12, q6
1068		−
1069		− //for another 8x16
1070		− SAD_SSD_16_RESET_8x8 \arg0, \arg1, \arg2, q7
1071		− SAD_SSD_16 \arg0, \arg1, \arg2, q7
1072		− SAD_SSD_16 \arg0, \arg1, \arg2, q7
1073		− SAD_SSD_16 \arg0, \arg1, \arg2, q7
1074		− SAD_SSD_16 \arg0, \arg1, \arg2, q7
1075		− SAD_SSD_16 \arg0, \arg1, \arg2, q7
1076		− SAD_SSD_16 \arg0, \arg1, \arg2, q7
1077		− SAD_SSD_16_END \arg0, \arg2, q7
1078		−
1079		− vpaddl.u16 q7, q7
1080		− vpaddl.u32 q7, q7
1081		−
1082		− vadd.i32 q12, q7
	1055	+ //for one 8x16
	1056	+ SAD_SSD_16_RESET_16x16 \arg0, \arg1, \arg2, q6
	1057	+ SAD_SSD_16 \arg0, \arg1, \arg2, q6
	1058	+ SAD_SSD_16 \arg0, \arg1, \arg2, q6
	1059	+ SAD_SSD_16 \arg0, \arg1, \arg2, q6
	1060	+ SAD_SSD_16 \arg0, \arg1, \arg2, q6
	1061	+ SAD_SSD_16 \arg0, \arg1, \arg2, q6
	1062	+ SAD_SSD_16 \arg0, \arg1, \arg2, q6
	1063	+ SAD_SSD_16 \arg0, \arg1, \arg2, q6
	1064	+
	1065	+ vpaddl.u16 q6, q6
	1066	+ vpaddl.u32 q6, q6
	1067	+ vadd.i32 q12, q6
	1068	+
	1069	+ //for another 8x16
	1070	+ SAD_SSD_16_RESET_8x8 \arg0, \arg1, \arg2, q7
	1071	+ SAD_SSD_16 \arg0, \arg1, \arg2, q7
	1072	+ SAD_SSD_16 \arg0, \arg1, \arg2, q7
	1073	+ SAD_SSD_16 \arg0, \arg1, \arg2, q7
	1074	+ SAD_SSD_16 \arg0, \arg1, \arg2, q7
	1075	+ SAD_SSD_16 \arg0, \arg1, \arg2, q7
	1076	+ SAD_SSD_16 \arg0, \arg1, \arg2, q7
	1077	+ SAD_SSD_16_END \arg0, \arg2, q7
	1078	+
	1079	+ vpaddl.u16 q7, q7
	1080	+ vpaddl.u32 q7, q7
	1081	+
	1082	+ vadd.i32 q12, q7
1083	1083	.endm
1084	1084	#endif
1085	1085
1086	1086
1087	1087	WELS_ASM_FUNC_BEGIN VAACalcSadSsd_neon
1088		− stmdb sp!, {r4-r12}
1089		− vpush {q4}
1090		− vpush {q6-q7}
	1088	+ stmdb sp!, {r4-r12}
	1089	+ vpush {q4}
	1090	+ vpush {q6-q7}
1091	1091
1092		− ldr r4, [sp, #84] //r4 keeps the pic_stride
	1092	+ ldr r4, [sp, #84] //r4 keeps the pic_stride
1093	1093
1094		− sub r5, r4, #1
1095		− lsl r5, r5, #4 //r5 keeps the little step
	1094	+ sub r5, r4, #1
	1095	+ lsl r5, r5, #4 //r5 keeps the little step
1096	1096
1097		− lsl r6, r4, #4
1098		− sub r6, r2, r6 //r6 keeps the big step
	1097	+ lsl r6, r4, #4
	1098	+ sub r6, r2, r6 //r6 keeps the big step
1099	1099
1100		− ldr r7, [sp, #88] //psadframe
1101		− ldr r8, [sp, #92] //psad8x8
1102		− ldr r9, [sp, #96] //psum16x16
1103		− ldr r10, [sp, #100] //psqsum16x16
1104		− ldr r11, [sp, #104] //psqdiff16x16
	1100	+ ldr r7, [sp, #88] //psadframe
	1101	+ ldr r8, [sp, #92] //psad8x8
	1102	+ ldr r9, [sp, #96] //psum16x16
	1103	+ ldr r10, [sp, #100] //psqsum16x16
	1104	+ ldr r11, [sp, #104] //psqdiff16x16
1105	1105
1106		− vmov.i8 q12, #0
	1106	+ vmov.i8 q12, #0
1107	1107	vaa_calc_sad_ssd_height_loop:
1108	1108
1109	1109	mov r12, r2
@@ -1136,18 +1136,18 @@ vaa_calc_sad_ssd_width_loop: · 26 unmodified lines
1136	1136
1137	1137	bne vaa_calc_sad_ssd_width_loop
1138	1138
1139		− sub r0, r0, r6 //jump to next 16 x width
1140		− sub r1, r1, r6 //jump to next 16 x width
	1139	+ sub r0, r0, r6 //jump to next 16 x width
	1140	+ sub r1, r1, r6 //jump to next 16 x width
1141	1141
1142	1142	subs r3, #16
1143		− bne vaa_calc_sad_ssd_height_loop
	1143	+ bne vaa_calc_sad_ssd_height_loop
1144	1144
1145		− vadd.i32 d24, d24, d25
1146		− vst1.32 {d24[0]}, [r7]
	1145	+ vadd.i32 d24, d24, d25
	1146	+ vst1.32 {d24[0]}, [r7]
1147	1147
1148		− vpop {q6-q7}
1149		− vpop {q4}
1150		− ldmia sp!, {r4-r12}
	1148	+ vpop {q6-q7}
	1149	+ vpop {q4}
	1150	+ ldmia sp!, {r4-r12}
1151	1151	WELS_ASM_FUNC_END
1152	1152
1153	1153	#endif

codec/processing/src/x86/denoisefilter.asm+170−170

@@ -56,217 +56,217 @@ sse2_20 times 8 dw 20 · 55 unmodified lines
56	56	;***********************************************************************
57	57	SECTION .text
58	58
59		−%macro WEIGHT_LINE 9
60		− movq %2, %9
61		− punpcklbw %2, %7
62		− movdqa %8, %2
63		−
64		− movdqa %1, %6
65		− psubusb %1, %8
66		− psubusb %8, %6
67		− por %8, %1 ; ABS(curPixel - centerPixel);
68		−
69		− movdqa %1, %3
70		− psubusb %1, %8
71		−
72		− pmullw %1, %1
73		− psrlw %1, 5
74		− pmullw %2, %1
75		− paddusw %4, %1
76		− paddusw %5, %2
	59	+%macro WEIGHT_LINE 9
	60	+ movq %2, %9
	61	+ punpcklbw %2, %7
	62	+ movdqa %8, %2
	63	+
	64	+ movdqa %1, %6
	65	+ psubusb %1, %8
	66	+ psubusb %8, %6
	67	+ por %8, %1 ; ABS(curPixel - centerPixel);
	68	+
	69	+ movdqa %1, %3
	70	+ psubusb %1, %8
	71	+
	72	+ pmullw %1, %1
	73	+ psrlw %1, 5
	74	+ pmullw %2, %1
	75	+ paddusw %4, %1
	76	+ paddusw %5, %2
77	77	%endmacro
78	78
79		−%macro WEIGHT_LINE1_UV 4
80		− movdqa %2, %1
81		− punpcklbw %2, %4
82		− paddw %3, %2
83		−
84		− movdqa %2, %1
85		− psrldq %2, 1
86		− punpcklbw %2, %4
87		− paddw %3, %2
88		−
89		− movdqa %2, %1
90		− psrldq %2, 2
91		− punpcklbw %2, %4
92		− psllw %2, 1
93		− paddw %3, %2
94		−
95		− movdqa %2, %1
96		− psrldq %2, 3
97		− punpcklbw %2, %4
98		− paddw %3, %2
99		−
100		− movdqa %2, %1
101		− psrldq %2, 4
102		− punpcklbw %2, %4
103		− paddw %3, %2
	79	+%macro WEIGHT_LINE1_UV 4
	80	+ movdqa %2, %1
	81	+ punpcklbw %2, %4
	82	+ paddw %3, %2
	83	+
	84	+ movdqa %2, %1
	85	+ psrldq %2, 1
	86	+ punpcklbw %2, %4
	87	+ paddw %3, %2
	88	+
	89	+ movdqa %2, %1
	90	+ psrldq %2, 2
	91	+ punpcklbw %2, %4
	92	+ psllw %2, 1
	93	+ paddw %3, %2
	94	+
	95	+ movdqa %2, %1
	96	+ psrldq %2, 3
	97	+ punpcklbw %2, %4
	98	+ paddw %3, %2
	99	+
	100	+ movdqa %2, %1
	101	+ psrldq %2, 4
	102	+ punpcklbw %2, %4
	103	+ paddw %3, %2
104	104	%endmacro
105	105
106		−%macro WEIGHT_LINE2_UV 4
107		− movdqa %2, %1
108		− punpcklbw %2, %4
109		− paddw %3, %2
110		−
111		− movdqa %2, %1
112		− psrldq %2, 1
113		− punpcklbw %2, %4
114		− psllw %2, 1
115		− paddw %3, %2
116		−
117		− movdqa %2, %1
118		− psrldq %2, 2
119		− punpcklbw %2, %4
120		− psllw %2, 2
121		− paddw %3, %2
122		−
123		− movdqa %2, %1
124		− psrldq %2, 3
125		− punpcklbw %2, %4
126		− psllw %2, 1
127		− paddw %3, %2
128		−
129		− movdqa %2, %1
130		− psrldq %2, 4
131		− punpcklbw %2, %4
132		− paddw %3, %2
	106	+%macro WEIGHT_LINE2_UV 4
	107	+ movdqa %2, %1
	108	+ punpcklbw %2, %4
	109	+ paddw %3, %2
	110	+
	111	+ movdqa %2, %1
	112	+ psrldq %2, 1
	113	+ punpcklbw %2, %4
	114	+ psllw %2, 1
	115	+ paddw %3, %2
	116	+
	117	+ movdqa %2, %1
	118	+ psrldq %2, 2
	119	+ punpcklbw %2, %4
	120	+ psllw %2, 2
	121	+ paddw %3, %2
	122	+
	123	+ movdqa %2, %1
	124	+ psrldq %2, 3
	125	+ punpcklbw %2, %4
	126	+ psllw %2, 1
	127	+ paddw %3, %2
	128	+
	129	+ movdqa %2, %1
	130	+ psrldq %2, 4
	131	+ punpcklbw %2, %4
	132	+ paddw %3, %2
133	133	%endmacro
134	134
135		−%macro WEIGHT_LINE3_UV 4
136		− movdqa %2, %1
137		− punpcklbw %2, %4
138		− psllw %2, 1
139		− paddw %3, %2
140		−
141		− movdqa %2, %1
142		− psrldq %2, 1
143		− punpcklbw %2, %4
144		− psllw %2, 2
145		− paddw %3, %2
146		−
147		− movdqa %2, %1
148		− psrldq %2, 2
149		− punpcklbw %2, %4
150		− pmullw %2, [sse2_20]
151		− paddw %3, %2
152		−
153		− movdqa %2, %1
154		− psrldq %2, 3
155		− punpcklbw %2, %4
156		− psllw %2, 2
157		− paddw %3, %2
158		−
159		− movdqa %2, %1
160		− psrldq %2, 4
161		− punpcklbw %2, %4
162		− psllw %2, 1
163		− paddw %3, %2
	135	+%macro WEIGHT_LINE3_UV 4
	136	+ movdqa %2, %1
	137	+ punpcklbw %2, %4
	138	+ psllw %2, 1
	139	+ paddw %3, %2
	140	+
	141	+ movdqa %2, %1
	142	+ psrldq %2, 1
	143	+ punpcklbw %2, %4
	144	+ psllw %2, 2
	145	+ paddw %3, %2
	146	+
	147	+ movdqa %2, %1
	148	+ psrldq %2, 2
	149	+ punpcklbw %2, %4
	150	+ pmullw %2, [sse2_20]
	151	+ paddw %3, %2
	152	+
	153	+ movdqa %2, %1
	154	+ psrldq %2, 3
	155	+ punpcklbw %2, %4
	156	+ psllw %2, 2
	157	+ paddw %3, %2
	158	+
	159	+ movdqa %2, %1
	160	+ psrldq %2, 4
	161	+ punpcklbw %2, %4
	162	+ psllw %2, 1
	163	+ paddw %3, %2
164	164	%endmacro
165	165
166	166	;***********************************************************************
167	167	; BilateralLumaFilter8_sse2(uint8_t *pixels, int stride);
168	168	;***********************************************************************
169		−; 1 2 3
170		−; 4 0 5
171		−; 6 7 8
172		−; 0: the center point
	169	+; 1 2 3
	170	+; 4 0 5
	171	+; 6 7 8
	172	+; 0: the center point
173	173
174	174	WELS_EXTERN BilateralLumaFilter8_sse2
175	175
176		− push r3
177		− %assign push_num 1
178		− LOAD_2_PARA
179		− PUSH_XMM 8
	176	+ push r3
	177	+ %assign push_num 1
	178	+ LOAD_2_PARA
	179	+ PUSH_XMM 8
180	180
181		− pxor xmm7, xmm7
	181	+ pxor xmm7, xmm7
182	182
183		− mov r3, r0
	183	+ mov r3, r0
184	184
185		− movq xmm6, [r0]
186		− punpcklbw xmm6, xmm7
187		− movdqa xmm3, [sse2_32]
188		− pxor xmm4, xmm4 ; nTotWeight
189		− pxor xmm5, xmm5 ; nSum
	185	+ movq xmm6, [r0]
	186	+ punpcklbw xmm6, xmm7
	187	+ movdqa xmm3, [sse2_32]
	188	+ pxor xmm4, xmm4 ; nTotWeight
	189	+ pxor xmm5, xmm5 ; nSum
190	190
191		− dec r0
192		− WEIGHT_LINE xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7, xmm0, [r0] ; pixel 4
193		− WEIGHT_LINE xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7, xmm0, [r0 + 2] ; pixel 5
	191	+ dec r0
	192	+ WEIGHT_LINE xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7, xmm0, [r0] ; pixel 4
	193	+ WEIGHT_LINE xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7, xmm0, [r0 + 2] ; pixel 5
194	194
195		− sub r0, r1
196		− WEIGHT_LINE xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7, xmm0, [r0] ; pixel 1
197		− WEIGHT_LINE xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7, xmm0, [r0 + 1] ; pixel 2
198		− WEIGHT_LINE xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7, xmm0, [r0 + 2] ; pixel 3
	195	+ sub r0, r1
	196	+ WEIGHT_LINE xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7, xmm0, [r0] ; pixel 1
	197	+ WEIGHT_LINE xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7, xmm0, [r0 + 1] ; pixel 2
	198	+ WEIGHT_LINE xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7, xmm0, [r0 + 2] ; pixel 3
199	199
200		− lea r0, [r0 + r1 * 2]
201		− WEIGHT_LINE xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7, xmm0, [r0] ; pixel 6
202		− WEIGHT_LINE xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7, xmm0, [r0 + 1] ; pixel 7
203		− WEIGHT_LINE xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7, xmm0, [r0 + 2] ; pixel 8
	200	+ lea r0, [r0 + r1 * 2]
	201	+ WEIGHT_LINE xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7, xmm0, [r0] ; pixel 6
	202	+ WEIGHT_LINE xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7, xmm0, [r0 + 1] ; pixel 7
	203	+ WEIGHT_LINE xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7, xmm0, [r0 + 2] ; pixel 8
204	204
205		− pcmpeqw xmm0, xmm0
206		− psrlw xmm0, 15
207		− psllw xmm0, 8
208		− psubusw xmm0, xmm4
209		− pmullw xmm0, xmm6
210		− paddusw xmm5, xmm0
211		− psrlw xmm5, 8
212		− packuswb xmm5, xmm5
213		− movq [r3], xmm5
	205	+ pcmpeqw xmm0, xmm0
	206	+ psrlw xmm0, 15
	207	+ psllw xmm0, 8
	208	+ psubusw xmm0, xmm4
	209	+ pmullw xmm0, xmm6
	210	+ paddusw xmm5, xmm0
	211	+ psrlw xmm5, 8
	212	+ packuswb xmm5, xmm5
	213	+ movq [r3], xmm5
214	214
215	215
216		− POP_XMM
217		− pop r3
218		− %assign push_num 0
	216	+ POP_XMM
	217	+ pop r3
	218	+ %assign push_num 0
219	219
220		− ret
	220	+ ret
221	221
222	222	;***********************************************************************
223		−; void WaverageChromaFilter8_sse2(uint8_t *pixels, int stride);
	223	+; void WaverageChromaFilter8_sse2(uint8_t *pixels, int stride);
224	224	;***********************************************************************
225	225	;5x5 filter:
226		−;1 1 2 1 1
227		−;1 2 4 2 1
228		−;2 4 20 4 2
229		−;1 2 4 2 1
230		−;1 1 2 1 1
	226	+;1 1 2 1 1
	227	+;1 2 4 2 1
	228	+;2 4 20 4 2
	229	+;1 2 4 2 1
	230	+;1 1 2 1 1
231	231
232	232	WELS_EXTERN WaverageChromaFilter8_sse2
233	233
234		− push r3
	234	+ push r3
235	235
236		− %assign push_num 1
	236	+ %assign push_num 1
237	237
238		− LOAD_2_PARA
	238	+ LOAD_2_PARA
239	239
240		− mov r3, r1
241		− add r3, r3
242		− sub r0, r3 ; pixels - 2 * stride
243		− sub r0, 2
	240	+ mov r3, r1
	241	+ add r3, r3
	242	+ sub r0, r3 ; pixels - 2 * stride
	243	+ sub r0, 2
244	244
245		− pxor xmm0, xmm0
246		− pxor xmm3, xmm3
	245	+ pxor xmm0, xmm0
	246	+ pxor xmm3, xmm3
247	247
248		− movdqu xmm1, [r0]
249		− WEIGHT_LINE1_UV xmm1, xmm2, xmm3, xmm0
	248	+ movdqu xmm1, [r0]
	249	+ WEIGHT_LINE1_UV xmm1, xmm2, xmm3, xmm0
250	250
251		− movdqu xmm1, [r0 + r1]
252		− WEIGHT_LINE2_UV xmm1, xmm2, xmm3, xmm0
	251	+ movdqu xmm1, [r0 + r1]
	252	+ WEIGHT_LINE2_UV xmm1, xmm2, xmm3, xmm0
253	253
254		− add r0, r3
255		− movdqu xmm1, [r0]
256		− WEIGHT_LINE3_UV xmm1, xmm2, xmm3, xmm0
	254	+ add r0, r3
	255	+ movdqu xmm1, [r0]
	256	+ WEIGHT_LINE3_UV xmm1, xmm2, xmm3, xmm0
257	257
258		− movdqu xmm1, [r0 + r1]
259		− WEIGHT_LINE2_UV xmm1, xmm2, xmm3, xmm0
	258	+ movdqu xmm1, [r0 + r1]
	259	+ WEIGHT_LINE2_UV xmm1, xmm2, xmm3, xmm0
260	260
261		− movdqu xmm1, [r0 + r1 * 2]
262		− WEIGHT_LINE1_UV xmm1, xmm2, xmm3, xmm0
	261	+ movdqu xmm1, [r0 + r1 * 2]
	262	+ WEIGHT_LINE1_UV xmm1, xmm2, xmm3, xmm0
263	263
264		− psrlw xmm3, 6
265		− packuswb xmm3, xmm3
266		− movq [r0 + 2], xmm3
	264	+ psrlw xmm3, 6
	265	+ packuswb xmm3, xmm3
	266	+ movq [r0 + 2], xmm3
267	267
268	268
269		− pop r3
	269	+ pop r3
270	270
271		− %assign push_num 0
272		− ret
	271	+ %assign push_num 0
	272	+ ret

codec/processing/src/x86/downsample_bilinear.asm+978−978

@@ -29,13 +29,13 @@ · 28 unmodified lines
29	29	;* POSSIBILITY OF SUCH DAMAGE.
30	30	;*
31	31	;*
32		−;* upsampling.asm
	32	+;* upsampling.asm
33	33	;*
34	34	;* Abstract
35		−;* SIMD for pixel domain down sampling
	35	+;* SIMD for pixel domain down sampling
36	36	;*
37	37	;* History
38		−;* 10/22/2009 Created
	38	+;* 10/22/2009 Created
39	39	;*
40	40	;*************************************************************************/
41	41	%include "asm_inc.asm"
@@ -61,9 +61,9 @@ SECTION .rodata align=16 · 19 unmodified lines
61	61
62	62	ALIGN 16
63	63	shufb_mask_low:
64		− db 00h, 80h, 02h, 80h, 04h, 80h, 06h, 80h, 08h, 80h, 0ah, 80h, 0ch, 80h, 0eh, 80h
	64	+ db 00h, 80h, 02h, 80h, 04h, 80h, 06h, 80h, 08h, 80h, 0ah, 80h, 0ch, 80h, 0eh, 80h
65	65	shufb_mask_high:
66		− db 01h, 80h, 03h, 80h, 05h, 80h, 07h, 80h, 09h, 80h, 0bh, 80h, 0dh, 80h, 0fh, 80h
	66	+ db 01h, 80h, 03h, 80h, 05h, 80h, 07h, 80h, 09h, 80h, 0bh, 80h, 0dh, 80h, 0fh, 80h
67	67
68	68
69	69	;***********************************************************************
@@ -73,737 +73,737 @@ shufb_mask_high: · 3 unmodified lines
73	73	SECTION .text
74	74
75	75	;***********************************************************************
76		−; void DyadicBilinearDownsamplerWidthx32_sse( unsigned char* pDst, const int iDstStride,
77		−; unsigned char* pSrc, const int iSrcStride,
78		−; const int iSrcWidth, const int iSrcHeight );
	76	+; void DyadicBilinearDownsamplerWidthx32_sse( unsigned char* pDst, const int iDstStride,
	77	+; unsigned char* pSrc, const int iSrcStride,
	78	+; const int iSrcWidth, const int iSrcHeight );
79	79	;***********************************************************************
80	80	WELS_EXTERN DyadicBilinearDownsamplerWidthx32_sse
81		− push ebx
82		− push edx
83		− push esi
84		− push edi
85		− push ebp
	81	+ push ebx
	82	+ push edx
	83	+ push esi
	84	+ push edi
	85	+ push ebp
86	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
	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	92
93		− sar ebp, $01 ; iSrcHeight >> 1
	93	+ sar ebp, $01 ; iSrcHeight >> 1
94	94
95	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)
101		− ; each loop = source bandwidth: 32 bytes
	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)
	101	+ ; each loop = source bandwidth: 32 bytes
102	102	.xloops:
103		− ; 1st part horizonal loop: x16 bytes
104		− ; mem hi<- ->lo
105		− ;1st Line Src: mm0: d D c C b B a A mm1: h H g G f F e E
106		− ;2nd Line Src: mm2: l L k K j J i I mm3: p P o O n N m M
107		− ;=> target:
108		− ;: H G F E D C B A, P O N M L K J I
109		− ;: h g f e d c b a, p o n m l k j i
110		− ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
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
115		−
116		− ; to handle mm0, mm1, mm2, mm3
117		− pshufw mm4, mm0, 0d8h ; d D b B c C a A ; 11011000 B
118		− pshufw mm5, mm4, 04eh ; c C a A d D b B ; 01001110 B
119		− punpcklbw mm4, mm5 ; d c D C b a B A
120		− pshufw mm4, mm4, 0d8h ; d c b a D C B A ; 11011000 B: mm4
121		−
122		− pshufw mm5, mm1, 0d8h ; h H f F g G e E ; 11011000 B
123		− pshufw mm6, mm5, 04eh ; g G e E h H f F ; 01001110 B
124		− punpcklbw mm5, mm6 ; h g H G f e F E
125		− pshufw mm5, mm5, 0d8h ; h g f e H G F E ; 11011000 B: mm5
126		−
127		− pshufw mm6, mm2, 0d8h ; l L j J k K i I ; 11011000 B
128		− pshufw mm7, mm6, 04eh ; k K i I l L j J ; 01001110 B
129		− punpcklbw mm6, mm7 ; l k L K j i J I
130		− pshufw mm6, mm6, 0d8h ; l k j i L K J I ; 11011000 B: mm6
131		−
132		− pshufw mm7, mm3, 0d8h ; p P n N o O m M ; 11011000 B
133		− pshufw mm0, mm7, 04eh ; o O m M p P n N ; 01001110 B
134		− punpcklbw mm7, mm0 ; p o P O n m N M
135		− pshufw mm7, mm7, 0d8h ; p o n m P O N M ; 11011000 B: mm7
136		−
137		− ; to handle mm4, mm5, mm6, mm7
138		− movq mm0, mm4 ;
139		− punpckldq mm0, mm5 ; H G F E D C B A
140		− punpckhdq mm4, mm5 ; h g f e d c b a
141		−
142		− movq mm1, mm6
143		− punpckldq mm1, mm7 ; P O N M L K J I
144		− punpckhdq mm6, mm7 ; p o n m l k j i
145		−
146		− ; avg within MB horizon width (16 x 2 lines)
147		− pavgb mm0, mm4 ; (A+a+1)>>1, .., (H+h+1)>>1, temp_row1
148		− pavgb mm1, mm6 ; (I+i+1)>>1, .., (P+p+1)>>1, temp_row2
149		− pavgb mm0, mm1 ; (temp_row1+temp_row2+1)>>1, pending here and wait another horizonal part done then write memory once
150		−
151		− ; 2nd part horizonal loop: x16 bytes
152		− ; mem hi<- ->lo
153		− ;1st Line Src: mm0: d D c C b B a A mm1: h H g G f F e E
154		− ;2nd Line Src: mm2: l L k K j J i I mm3: p P o O n N m M
155		− ;=> target:
156		− ;: H G F E D C B A, P O N M L K J I
157		− ;: h g f e d c b a, p o n m l k j i
158		− ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
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
163		−
164		− ; to handle mm1, mm2, mm3, mm4
165		− pshufw mm5, mm1, 0d8h ; d D b B c C a A ; 11011000 B
166		− pshufw mm6, mm5, 04eh ; c C a A d D b B ; 01001110 B
167		− punpcklbw mm5, mm6 ; d c D C b a B A
168		− pshufw mm5, mm5, 0d8h ; d c b a D C B A ; 11011000 B: mm5
169		−
170		− pshufw mm6, mm2, 0d8h ; h H f F g G e E ; 11011000 B
171		− pshufw mm7, mm6, 04eh ; g G e E h H f F ; 01001110 B
172		− punpcklbw mm6, mm7 ; h g H G f e F E
173		− pshufw mm6, mm6, 0d8h ; h g f e H G F E ; 11011000 B: mm6
174		−
175		− pshufw mm7, mm3, 0d8h ; l L j J k K i I ; 11011000 B
176		− pshufw mm1, mm7, 04eh ; k K i I l L j J ; 01001110 B
177		− punpcklbw mm7, mm1 ; l k L K j i J I
178		− pshufw mm7, mm7, 0d8h ; l k j i L K J I ; 11011000 B: mm7
179		−
180		− pshufw mm1, mm4, 0d8h ; p P n N o O m M ; 11011000 B
181		− pshufw mm2, mm1, 04eh ; o O m M p P n N ; 01001110 B
182		− punpcklbw mm1, mm2 ; p o P O n m N M
183		− pshufw mm1, mm1, 0d8h ; p o n m P O N M ; 11011000 B: mm1
184		−
185		− ; to handle mm5, mm6, mm7, mm1
186		− movq mm2, mm5
187		− punpckldq mm2, mm6 ; H G F E D C B A
188		− punpckhdq mm5, mm6 ; h g f e d c b a
189		−
190		− movq mm3, mm7
191		− punpckldq mm3, mm1 ; P O N M L K J I
192		− punpckhdq mm7, mm1 ; p o n m l k j i
193		−
194		− ; avg within MB horizon width (16 x 2 lines)
195		− pavgb mm2, mm5 ; (A+a+1)>>1, .., (H+h+1)>>1, temp_row1
196		− pavgb mm3, mm7 ; (I+i+1)>>1, .., (P+p+1)>>1, temp_row2
197		− pavgb mm2, mm3 ; (temp_row1+temp_row2+1)>>1, done in another 2nd horizonal part
198		−
199		− movq [edi ], mm0
200		− movq [edi+8], mm2
201		−
202		− ; next SMB
203		− lea esi, [esi+32]
204		− lea edi, [edi+16]
205		−
206		− dec eax
207		− jg near .xloops
208		−
209		− ; 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]
214		−
215		− dec ebp
216		− jg near .yloops
217		−
218		− WELSEMMS
219		− pop ebp
220		− pop edi
221		− pop esi
222		− pop edx
223		− pop ebx
224		− ret
	103	+ ; 1st part horizonal loop: x16 bytes
	104	+ ; mem hi<- ->lo
	105	+ ;1st Line Src: mm0: d D c C b B a A mm1: h H g G f F e E
	106	+ ;2nd Line Src: mm2: l L k K j J i I mm3: p P o O n N m M
	107	+ ;=> target:
	108	+ ;: H G F E D C B A, P O N M L K J I
	109	+ ;: h g f e d c b a, p o n m l k j i
	110	+ ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
	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
	115	+
	116	+ ; to handle mm0, mm1, mm2, mm3
	117	+ pshufw mm4, mm0, 0d8h ; d D b B c C a A ; 11011000 B
	118	+ pshufw mm5, mm4, 04eh ; c C a A d D b B ; 01001110 B
	119	+ punpcklbw mm4, mm5 ; d c D C b a B A
	120	+ pshufw mm4, mm4, 0d8h ; d c b a D C B A ; 11011000 B: mm4
	121	+
	122	+ pshufw mm5, mm1, 0d8h ; h H f F g G e E ; 11011000 B
	123	+ pshufw mm6, mm5, 04eh ; g G e E h H f F ; 01001110 B
	124	+ punpcklbw mm5, mm6 ; h g H G f e F E
	125	+ pshufw mm5, mm5, 0d8h ; h g f e H G F E ; 11011000 B: mm5
	126	+
	127	+ pshufw mm6, mm2, 0d8h ; l L j J k K i I ; 11011000 B
	128	+ pshufw mm7, mm6, 04eh ; k K i I l L j J ; 01001110 B
	129	+ punpcklbw mm6, mm7 ; l k L K j i J I
	130	+ pshufw mm6, mm6, 0d8h ; l k j i L K J I ; 11011000 B: mm6
	131	+
	132	+ pshufw mm7, mm3, 0d8h ; p P n N o O m M ; 11011000 B
	133	+ pshufw mm0, mm7, 04eh ; o O m M p P n N ; 01001110 B
	134	+ punpcklbw mm7, mm0 ; p o P O n m N M
	135	+ pshufw mm7, mm7, 0d8h ; p o n m P O N M ; 11011000 B: mm7
	136	+
	137	+ ; to handle mm4, mm5, mm6, mm7
	138	+ movq mm0, mm4 ;
	139	+ punpckldq mm0, mm5 ; H G F E D C B A
	140	+ punpckhdq mm4, mm5 ; h g f e d c b a
	141	+
	142	+ movq mm1, mm6
	143	+ punpckldq mm1, mm7 ; P O N M L K J I
	144	+ punpckhdq mm6, mm7 ; p o n m l k j i
	145	+
	146	+ ; avg within MB horizon width (16 x 2 lines)
	147	+ pavgb mm0, mm4 ; (A+a+1)>>1, .., (H+h+1)>>1, temp_row1
	148	+ pavgb mm1, mm6 ; (I+i+1)>>1, .., (P+p+1)>>1, temp_row2
	149	+ pavgb mm0, mm1 ; (temp_row1+temp_row2+1)>>1, pending here and wait another horizonal part done then write memory once
	150	+
	151	+ ; 2nd part horizonal loop: x16 bytes
	152	+ ; mem hi<- ->lo
	153	+ ;1st Line Src: mm0: d D c C b B a A mm1: h H g G f F e E
	154	+ ;2nd Line Src: mm2: l L k K j J i I mm3: p P o O n N m M
	155	+ ;=> target:
	156	+ ;: H G F E D C B A, P O N M L K J I
	157	+ ;: h g f e d c b a, p o n m l k j i
	158	+ ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
	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
	163	+
	164	+ ; to handle mm1, mm2, mm3, mm4
	165	+ pshufw mm5, mm1, 0d8h ; d D b B c C a A ; 11011000 B
	166	+ pshufw mm6, mm5, 04eh ; c C a A d D b B ; 01001110 B
	167	+ punpcklbw mm5, mm6 ; d c D C b a B A
	168	+ pshufw mm5, mm5, 0d8h ; d c b a D C B A ; 11011000 B: mm5
	169	+
	170	+ pshufw mm6, mm2, 0d8h ; h H f F g G e E ; 11011000 B
	171	+ pshufw mm7, mm6, 04eh ; g G e E h H f F ; 01001110 B
	172	+ punpcklbw mm6, mm7 ; h g H G f e F E
	173	+ pshufw mm6, mm6, 0d8h ; h g f e H G F E ; 11011000 B: mm6
	174	+
	175	+ pshufw mm7, mm3, 0d8h ; l L j J k K i I ; 11011000 B
	176	+ pshufw mm1, mm7, 04eh ; k K i I l L j J ; 01001110 B
	177	+ punpcklbw mm7, mm1 ; l k L K j i J I
	178	+ pshufw mm7, mm7, 0d8h ; l k j i L K J I ; 11011000 B: mm7
	179	+
	180	+ pshufw mm1, mm4, 0d8h ; p P n N o O m M ; 11011000 B
	181	+ pshufw mm2, mm1, 04eh ; o O m M p P n N ; 01001110 B
	182	+ punpcklbw mm1, mm2 ; p o P O n m N M
	183	+ pshufw mm1, mm1, 0d8h ; p o n m P O N M ; 11011000 B: mm1
	184	+
	185	+ ; to handle mm5, mm6, mm7, mm1
	186	+ movq mm2, mm5
	187	+ punpckldq mm2, mm6 ; H G F E D C B A
	188	+ punpckhdq mm5, mm6 ; h g f e d c b a
	189	+
	190	+ movq mm3, mm7
	191	+ punpckldq mm3, mm1 ; P O N M L K J I
	192	+ punpckhdq mm7, mm1 ; p o n m l k j i
	193	+
	194	+ ; avg within MB horizon width (16 x 2 lines)
	195	+ pavgb mm2, mm5 ; (A+a+1)>>1, .., (H+h+1)>>1, temp_row1
	196	+ pavgb mm3, mm7 ; (I+i+1)>>1, .., (P+p+1)>>1, temp_row2
	197	+ pavgb mm2, mm3 ; (temp_row1+temp_row2+1)>>1, done in another 2nd horizonal part
	198	+
	199	+ movq [edi ], mm0
	200	+ movq [edi+8], mm2
	201	+
	202	+ ; next SMB
	203	+ lea esi, [esi+32]
	204	+ lea edi, [edi+16]
	205	+
	206	+ dec eax
	207	+ jg near .xloops
	208	+
	209	+ ; 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]
	214	+
	215	+ dec ebp
	216	+ jg near .yloops
	217	+
	218	+ WELSEMMS
	219	+ pop ebp
	220	+ pop edi
	221	+ pop esi
	222	+ pop edx
	223	+ pop ebx
	224	+ ret
225	225
226	226	;***********************************************************************
227		−; void DyadicBilinearDownsamplerWidthx16_sse( unsigned char* pDst, const int iDstStride,
228		−; unsigned char* pSrc, const int iSrcStride,
229		−; const int iSrcWidth, const int iSrcHeight );
	227	+; void DyadicBilinearDownsamplerWidthx16_sse( unsigned char* pDst, const int iDstStride,
	228	+; unsigned char* pSrc, const int iSrcStride,
	229	+; const int iSrcWidth, const int iSrcHeight );
230	230	;***********************************************************************
231	231	WELS_EXTERN DyadicBilinearDownsamplerWidthx16_sse
232		− push ebx
233		− push edx
234		− push esi
235		− push edi
236		− push ebp
	232	+ push ebx
	233	+ push edx
	234	+ push esi
	235	+ push edi
	236	+ push ebp
237	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
	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	243
244		− sar ebp, $01 ; iSrcHeight >> 1
	244	+ sar ebp, $01 ; iSrcHeight >> 1
245	245
246	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)
252		− ; each loop = source bandwidth: 16 bytes
	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)
	252	+ ; each loop = source bandwidth: 16 bytes
253	253	.xloops:
254		− ; 1st part horizonal loop: x16 bytes
255		− ; mem hi<- ->lo
256		− ;1st Line Src: mm0: d D c C b B a A mm1: h H g G f F e E
257		− ;2nd Line Src: mm2: l L k K j J i I mm3: p P o O n N m M
258		− ;=> target:
259		− ;: H G F E D C B A, P O N M L K J I
260		− ;: h g f e d c b a, p o n m l k j i
261		− ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
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
266		−
267		− ; to handle mm0, mm1, mm2, mm3
268		− pshufw mm4, mm0, 0d8h ; d D b B c C a A ; 11011000 B
269		− pshufw mm5, mm4, 04eh ; c C a A d D b B ; 01001110 B
270		− punpcklbw mm4, mm5 ; d c D C b a B A
271		− pshufw mm4, mm4, 0d8h ; d c b a D C B A ; 11011000 B: mm4
272		−
273		− pshufw mm5, mm1, 0d8h ; h H f F g G e E ; 11011000 B
274		− pshufw mm6, mm5, 04eh ; g G e E h H f F ; 01001110 B
275		− punpcklbw mm5, mm6 ; h g H G f e F E
276		− pshufw mm5, mm5, 0d8h ; h g f e H G F E ; 11011000 B: mm5
277		−
278		− pshufw mm6, mm2, 0d8h ; l L j J k K i I ; 11011000 B
279		− pshufw mm7, mm6, 04eh ; k K i I l L j J ; 01001110 B
280		− punpcklbw mm6, mm7 ; l k L K j i J I
281		− pshufw mm6, mm6, 0d8h ; l k j i L K J I ; 11011000 B: mm6
282		−
283		− pshufw mm7, mm3, 0d8h ; p P n N o O m M ; 11011000 B
284		− pshufw mm0, mm7, 04eh ; o O m M p P n N ; 01001110 B
285		− punpcklbw mm7, mm0 ; p o P O n m N M
286		− pshufw mm7, mm7, 0d8h ; p o n m P O N M ; 11011000 B: mm7
287		−
288		− ; to handle mm4, mm5, mm6, mm7
289		− movq mm0, mm4 ;
290		− punpckldq mm0, mm5 ; H G F E D C B A
291		− punpckhdq mm4, mm5 ; h g f e d c b a
292		−
293		− movq mm1, mm6
294		− punpckldq mm1, mm7 ; P O N M L K J I
295		− punpckhdq mm6, mm7 ; p o n m l k j i
296		−
297		− ; avg within MB horizon width (16 x 2 lines)
298		− pavgb mm0, mm4 ; (A+a+1)>>1, .., (H+h+1)>>1, temp_row1
299		− pavgb mm1, mm6 ; (I+i+1)>>1, .., (P+p+1)>>1, temp_row2
300		− pavgb mm0, mm1 ; (temp_row1+temp_row2+1)>>1, pending here and wait another horizonal part done then write memory once
301		−
302		− movq [edi ], mm0
303		−
304		− ; next SMB
305		− lea esi, [esi+16]
306		− lea edi, [edi+8]
307		−
308		− dec eax
309		− jg near .xloops
310		−
311		− ; 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]
316		−
317		− dec ebp
318		− jg near .yloops
319		−
320		− WELSEMMS
321		− pop ebp
322		− pop edi
323		− pop esi
324		− pop edx
325		− pop ebx
326		− ret
	254	+ ; 1st part horizonal loop: x16 bytes
	255	+ ; mem hi<- ->lo
	256	+ ;1st Line Src: mm0: d D c C b B a A mm1: h H g G f F e E
	257	+ ;2nd Line Src: mm2: l L k K j J i I mm3: p P o O n N m M
	258	+ ;=> target:
	259	+ ;: H G F E D C B A, P O N M L K J I
	260	+ ;: h g f e d c b a, p o n m l k j i
	261	+ ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
	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
	266	+
	267	+ ; to handle mm0, mm1, mm2, mm3
	268	+ pshufw mm4, mm0, 0d8h ; d D b B c C a A ; 11011000 B
	269	+ pshufw mm5, mm4, 04eh ; c C a A d D b B ; 01001110 B
	270	+ punpcklbw mm4, mm5 ; d c D C b a B A
	271	+ pshufw mm4, mm4, 0d8h ; d c b a D C B A ; 11011000 B: mm4
	272	+
	273	+ pshufw mm5, mm1, 0d8h ; h H f F g G e E ; 11011000 B
	274	+ pshufw mm6, mm5, 04eh ; g G e E h H f F ; 01001110 B
	275	+ punpcklbw mm5, mm6 ; h g H G f e F E
	276	+ pshufw mm5, mm5, 0d8h ; h g f e H G F E ; 11011000 B: mm5
	277	+
	278	+ pshufw mm6, mm2, 0d8h ; l L j J k K i I ; 11011000 B
	279	+ pshufw mm7, mm6, 04eh ; k K i I l L j J ; 01001110 B
	280	+ punpcklbw mm6, mm7 ; l k L K j i J I
	281	+ pshufw mm6, mm6, 0d8h ; l k j i L K J I ; 11011000 B: mm6
	282	+
	283	+ pshufw mm7, mm3, 0d8h ; p P n N o O m M ; 11011000 B
	284	+ pshufw mm0, mm7, 04eh ; o O m M p P n N ; 01001110 B
	285	+ punpcklbw mm7, mm0 ; p o P O n m N M
	286	+ pshufw mm7, mm7, 0d8h ; p o n m P O N M ; 11011000 B: mm7
	287	+
	288	+ ; to handle mm4, mm5, mm6, mm7
	289	+ movq mm0, mm4 ;
	290	+ punpckldq mm0, mm5 ; H G F E D C B A
	291	+ punpckhdq mm4, mm5 ; h g f e d c b a
	292	+
	293	+ movq mm1, mm6
	294	+ punpckldq mm1, mm7 ; P O N M L K J I
	295	+ punpckhdq mm6, mm7 ; p o n m l k j i
	296	+
	297	+ ; avg within MB horizon width (16 x 2 lines)
	298	+ pavgb mm0, mm4 ; (A+a+1)>>1, .., (H+h+1)>>1, temp_row1
	299	+ pavgb mm1, mm6 ; (I+i+1)>>1, .., (P+p+1)>>1, temp_row2
	300	+ pavgb mm0, mm1 ; (temp_row1+temp_row2+1)>>1, pending here and wait another horizonal part done then write memory once
	301	+
	302	+ movq [edi ], mm0
	303	+
	304	+ ; next SMB
	305	+ lea esi, [esi+16]
	306	+ lea edi, [edi+8]
	307	+
	308	+ dec eax
	309	+ jg near .xloops
	310	+
	311	+ ; 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]
	316	+
	317	+ dec ebp
	318	+ jg near .yloops
	319	+
	320	+ WELSEMMS
	321	+ pop ebp
	322	+ pop edi
	323	+ pop esi
	324	+ pop edx
	325	+ pop ebx
	326	+ ret
327	327
328	328	;***********************************************************************
329		−; void DyadicBilinearDownsamplerWidthx8_sse( unsigned char* pDst, const int iDstStride,
330		−; unsigned char* pSrc, const int iSrcStride,
331		−; const int iSrcWidth, const int iSrcHeight );
	329	+; void DyadicBilinearDownsamplerWidthx8_sse( unsigned char* pDst, const int iDstStride,
	330	+; unsigned char* pSrc, const int iSrcStride,
	331	+; const int iSrcWidth, const int iSrcHeight );
332	332	;***********************************************************************
333	333	WELS_EXTERN DyadicBilinearDownsamplerWidthx8_sse
334		− push ebx
335		− push edx
336		− push esi
337		− push edi
338		− push ebp
	334	+ push ebx
	335	+ push edx
	336	+ push esi
	337	+ push edi
	338	+ push ebp
339	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
	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	345
346		− sar ebp, $01 ; iSrcHeight >> 1
	346	+ sar ebp, $01 ; iSrcHeight >> 1
347	347
348	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)
354		− ; each loop = source bandwidth: 8 bytes
	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)
	354	+ ; each loop = source bandwidth: 8 bytes
355	355	.xloops:
356		− ; 1st part horizonal loop: x8 bytes
357		− ; mem hi<- ->lo
358		− ;1st Line Src: mm0: d D c C b B a A
359		− ;2nd Line Src: mm1: h H g G f F e E
360		− ;=> target:
361		− ;: H G F E D C B A
362		− ;: h g f e d c b a
363		− ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
364		− movq mm0, [esi] ; 1st pSrc line
365		− movq mm1, [esi+ecx] ; 2nd pSrc line
366		−
367		− ; to handle mm0, mm1, mm2, mm3
368		− pshufw mm2, mm0, 0d8h ; d D b B c C a A ; 11011000 B
369		− pshufw mm3, mm2, 04eh ; c C a A d D b B ; 01001110 B
370		− punpcklbw mm2, mm3 ; d c D C b a B A
371		− pshufw mm2, mm2, 0d8h ; d c b a D C B A ; 11011000 B: mm4
372		−
373		− pshufw mm4, mm1, 0d8h ; h H f F g G e E ; 11011000 B
374		− pshufw mm5, mm4, 04eh ; g G e E h H f F ; 01001110 B
375		− punpcklbw mm4, mm5 ; h g H G f e F E
376		− pshufw mm4, mm4, 0d8h ; h g f e H G F E ; 11011000 B: mm5
377		−
378		− ; to handle mm2, mm4
379		− movq mm0, mm2 ;
380		− punpckldq mm0, mm4 ; H G F E D C B A
381		− punpckhdq mm2, mm4 ; h g f e d c b a
382		−
383		− ; avg within MB horizon width (16 x 2 lines)
384		− pavgb mm0, mm2 ; (H+h+1)>>1, .., (A+a+1)>>1, temp_row1, 2
385		− pshufw mm1, mm0, 04eh ; 01001110 B
386		− pavgb mm0, mm1 ; (temp_row1+temp_row2+1)>>1, pending here and wait another horizonal part done then write memory once
387		−
388		− movd [edi], mm0
389		−
390		− ; next unit
391		− lea esi, [esi+8]
392		− lea edi, [edi+4]
393		−
394		− dec eax
395		− jg near .xloops
396		−
397		− ; 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]
402		−
403		− dec ebp
404		− jg near .yloops
405		−
406		− WELSEMMS
407		− pop ebp
408		− pop edi
409		− pop esi
410		− pop edx
411		− pop ebx
412		− ret
	356	+ ; 1st part horizonal loop: x8 bytes
	357	+ ; mem hi<- ->lo
	358	+ ;1st Line Src: mm0: d D c C b B a A
	359	+ ;2nd Line Src: mm1: h H g G f F e E
	360	+ ;=> target:
	361	+ ;: H G F E D C B A
	362	+ ;: h g f e d c b a
	363	+ ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
	364	+ movq mm0, [esi] ; 1st pSrc line
	365	+ movq mm1, [esi+ecx] ; 2nd pSrc line
	366	+
	367	+ ; to handle mm0, mm1, mm2, mm3
	368	+ pshufw mm2, mm0, 0d8h ; d D b B c C a A ; 11011000 B
	369	+ pshufw mm3, mm2, 04eh ; c C a A d D b B ; 01001110 B
	370	+ punpcklbw mm2, mm3 ; d c D C b a B A
	371	+ pshufw mm2, mm2, 0d8h ; d c b a D C B A ; 11011000 B: mm4
	372	+
	373	+ pshufw mm4, mm1, 0d8h ; h H f F g G e E ; 11011000 B
	374	+ pshufw mm5, mm4, 04eh ; g G e E h H f F ; 01001110 B
	375	+ punpcklbw mm4, mm5 ; h g H G f e F E
	376	+ pshufw mm4, mm4, 0d8h ; h g f e H G F E ; 11011000 B: mm5
	377	+
	378	+ ; to handle mm2, mm4
	379	+ movq mm0, mm2 ;
	380	+ punpckldq mm0, mm4 ; H G F E D C B A
	381	+ punpckhdq mm2, mm4 ; h g f e d c b a
	382	+
	383	+ ; avg within MB horizon width (16 x 2 lines)
	384	+ pavgb mm0, mm2 ; (H+h+1)>>1, .., (A+a+1)>>1, temp_row1, 2
	385	+ pshufw mm1, mm0, 04eh ; 01001110 B
	386	+ pavgb mm0, mm1 ; (temp_row1+temp_row2+1)>>1, pending here and wait another horizonal part done then write memory once
	387	+
	388	+ movd [edi], mm0
	389	+
	390	+ ; next unit
	391	+ lea esi, [esi+8]
	392	+ lea edi, [edi+4]
	393	+
	394	+ dec eax
	395	+ jg near .xloops
	396	+
	397	+ ; 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]
	402	+
	403	+ dec ebp
	404	+ jg near .yloops
	405	+
	406	+ WELSEMMS
	407	+ pop ebp
	408	+ pop edi
	409	+ pop esi
	410	+ pop edx
	411	+ pop ebx
	412	+ ret
413	413
414	414
415	415
416	416	; got about 50% improvement over DyadicBilinearDownsamplerWidthx32_sse
417	417	;***********************************************************************
418		−; void DyadicBilinearDownsamplerWidthx32_ssse3( unsigned char* pDst, const int iDstStride,
419		−; unsigned char* pSrc, const int iSrcStride,
420		−; const int iSrcWidth, const int iSrcHeight );
	418	+; void DyadicBilinearDownsamplerWidthx32_ssse3( unsigned char* pDst, const int iDstStride,
	419	+; unsigned char* pSrc, const int iSrcStride,
	420	+; const int iSrcWidth, const int iSrcHeight );
421	421	;***********************************************************************
422	422	WELS_EXTERN DyadicBilinearDownsamplerWidthx32_ssse3
423		− push ebx
424		− push edx
425		− push esi
426		− push edi
427		− push ebp
	423	+ push ebx
	424	+ push edx
	425	+ push esi
	426	+ push edi
	427	+ push ebp
428	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
	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	434
435		− sar ebp, $01 ; iSrcHeight >> 1
	435	+ sar ebp, $01 ; iSrcHeight >> 1
436	436
437		− movdqa xmm7, [shufb_mask_low] ; mask low
438		− movdqa xmm6, [shufb_mask_high] ; mask high
	437	+ movdqa xmm7, [shufb_mask_low] ; mask low
	438	+ movdqa xmm6, [shufb_mask_high] ; mask high
439	439
440	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)
446		− ; each loop = source bandwidth: 32 bytes
	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)
	446	+ ; each loop = source bandwidth: 32 bytes
447	447	.xloops:
448		− ; 1st part horizonal loop: x16 bytes
449		− ; mem hi<- ->lo
450		− ;1st Line Src: xmm0: h H g G f F e E d D c C b B a A
451		− ; xmm1: p P o O n N m M l L k K j J i I
452		− ;2nd Line Src: xmm2: h H g G f F e E d D c C b B a A
453		− ; xmm3: p P o O n N m M l L k K j J i I
454		− ;=> target:
455		− ;: P O N M L K J I H G F E D C B A
456		− ;: p o n m l k j i h g f e d c b a
457		− ;: P .. A
458		− ;: p .. a
459		−
460		− ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
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
465		−
466		− ; packing & avg
467		− movdqa xmm4, xmm0 ; h H g G f F e E d D c C b B a A
468		− pshufb xmm0, xmm7 ; 0 H 0 G 0 F 0 E 0 D 0 C 0 B 0 A
469		− pshufb xmm4, xmm6 ; 0 h 0 g 0 f 0 e 0 d 0 c 0 b 0 a
470		− ; another implementation for xmm4 high bits
471		−; psubb xmm4, xmm0 ; h 0 g 0 f 0 e 0 d 0 c 0 b 0 a 0
472		−; psrlw xmm4, 8 ; 0 h 0 g 0 f 0 e 0 d 0 c 0 b 0 a
473		− pavgb xmm0, xmm4
474		−
475		− movdqa xmm5, xmm1
476		− pshufb xmm1, xmm7
477		− pshufb xmm5, xmm6
478		−; psubb xmm5, xmm1
479		−; psrlw xmm5, 8
480		− pavgb xmm1, xmm5
481		−
482		− movdqa xmm4, xmm2
483		− pshufb xmm2, xmm7
484		− pshufb xmm4, xmm6
485		−; psubb xmm4, xmm2
486		−; psrlw xmm4, 8
487		− pavgb xmm2, xmm4
488		−
489		− movdqa xmm5, xmm3
490		− pshufb xmm3, xmm7
491		− pshufb xmm5, xmm6
492		−; psubb xmm5, xmm3
493		−; psrlw xmm5, 8
494		− pavgb xmm3, xmm5
495		−
496		− packuswb xmm0, xmm1
497		− packuswb xmm2, xmm3
498		− pavgb xmm0, xmm2
499		−
500		− ; write pDst
501		− movdqa [edi], xmm0
502		−
503		− ; next SMB
504		− lea esi, [esi+32]
505		− lea edi, [edi+16]
506		−
507		− dec eax
508		− jg near .xloops
509		−
510		− ; 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
524		− ret
	448	+ ; 1st part horizonal loop: x16 bytes
	449	+ ; mem hi<- ->lo
	450	+ ;1st Line Src: xmm0: h H g G f F e E d D c C b B a A
	451	+ ; xmm1: p P o O n N m M l L k K j J i I
	452	+ ;2nd Line Src: xmm2: h H g G f F e E d D c C b B a A
	453	+ ; xmm3: p P o O n N m M l L k K j J i I
	454	+ ;=> target:
	455	+ ;: P O N M L K J I H G F E D C B A
	456	+ ;: p o n m l k j i h g f e d c b a
	457	+ ;: P .. A
	458	+ ;: p .. a
	459	+
	460	+ ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
	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
	465	+
	466	+ ; packing & avg
	467	+ movdqa xmm4, xmm0 ; h H g G f F e E d D c C b B a A
	468	+ pshufb xmm0, xmm7 ; 0 H 0 G 0 F 0 E 0 D 0 C 0 B 0 A
	469	+ pshufb xmm4, xmm6 ; 0 h 0 g 0 f 0 e 0 d 0 c 0 b 0 a
	470	+ ; another implementation for xmm4 high bits
	471	+; psubb xmm4, xmm0 ; h 0 g 0 f 0 e 0 d 0 c 0 b 0 a 0
	472	+; psrlw xmm4, 8 ; 0 h 0 g 0 f 0 e 0 d 0 c 0 b 0 a
	473	+ pavgb xmm0, xmm4
	474	+
	475	+ movdqa xmm5, xmm1
	476	+ pshufb xmm1, xmm7
	477	+ pshufb xmm5, xmm6
	478	+; psubb xmm5, xmm1
	479	+; psrlw xmm5, 8
	480	+ pavgb xmm1, xmm5
	481	+
	482	+ movdqa xmm4, xmm2
	483	+ pshufb xmm2, xmm7
	484	+ pshufb xmm4, xmm6
	485	+; psubb xmm4, xmm2
	486	+; psrlw xmm4, 8
	487	+ pavgb xmm2, xmm4
	488	+
	489	+ movdqa xmm5, xmm3
	490	+ pshufb xmm3, xmm7
	491	+ pshufb xmm5, xmm6
	492	+; psubb xmm5, xmm3
	493	+; psrlw xmm5, 8
	494	+ pavgb xmm3, xmm5
	495	+
	496	+ packuswb xmm0, xmm1
	497	+ packuswb xmm2, xmm3
	498	+ pavgb xmm0, xmm2
	499	+
	500	+ ; write pDst
	501	+ movdqa [edi], xmm0
	502	+
	503	+ ; next SMB
	504	+ lea esi, [esi+32]
	505	+ lea edi, [edi+16]
	506	+
	507	+ dec eax
	508	+ jg near .xloops
	509	+
	510	+ ; 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
	524	+ ret
525	525
526	526	;***********************************************************************
527		−; void DyadicBilinearDownsamplerWidthx16_ssse3( unsigned char* pDst, const int iDstStride,
528		−; unsigned char* pSrc, const int iSrcStride,
529		−; const int iSrcWidth, const int iSrcHeight );
	527	+; void DyadicBilinearDownsamplerWidthx16_ssse3( unsigned char* pDst, const int iDstStride,
	528	+; unsigned char* pSrc, const int iSrcStride,
	529	+; const int iSrcWidth, const int iSrcHeight );
530	530	;***********************************************************************
531	531	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
	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
547	547
548	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)
554		− ; each loop = source bandwidth: 16 bytes
	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)
	554	+ ; each loop = source bandwidth: 16 bytes
555	555	.xloops:
556		− ; horizonal loop: x16 bytes by source
557		− ; mem hi<- ->lo
558		− ;1st line pSrc: xmm0: h H g G f F e E d D c C b B a A
559		− ;2nd line pSrc: xmm1: p P o O n N m M l L k K j J i I
560		− ;=> target:
561		− ;: H G F E D C B A, P O N M L K J I
562		− ;: h g f e d c b a, p o n m l k j i
563		−
564		− ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
565		− movdqa xmm0, [esi] ; 1st_src_line
566		− movdqa xmm1, [esi+ecx] ; 2nd_src_line
567		−
568		− ; packing & avg
569		− 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
572		− ; another implementation for xmm2 high bits
573		−; psubb xmm2, xmm0 ; h 0 g 0 f 0 e 0 d 0 c 0 b 0 a 0
574		−; psrlw xmm2, 8 ; 0 h 0 g 0 f 0 e 0 d 0 c 0 b 0 a
575		− pavgb xmm0, xmm2
576		−
577		− movdqa xmm3, xmm1
578		− pshufb xmm1, xmm7
579		− pshufb xmm3, xmm6
580		−; psubb xmm3, xmm1
581		−; psrlw xmm3, 8
582		− pavgb xmm1, xmm3
583		−
584		− pavgb xmm0, xmm1
585		− packuswb xmm0, xmm1
586		−
587		− ; write pDst
588		− movq [edi], xmm0
589		−
590		− ; next SMB
591		− lea esi, [esi+16]
592		− lea edi, [edi+8]
593		−
594		− dec eax
595		− jg near .xloops
596		−
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
611		− ret
	556	+ ; horizonal loop: x16 bytes by source
	557	+ ; mem hi<- ->lo
	558	+ ;1st line pSrc: xmm0: h H g G f F e E d D c C b B a A
	559	+ ;2nd line pSrc: xmm1: p P o O n N m M l L k K j J i I
	560	+ ;=> target:
	561	+ ;: H G F E D C B A, P O N M L K J I
	562	+ ;: h g f e d c b a, p o n m l k j i
	563	+
	564	+ ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
	565	+ movdqa xmm0, [esi] ; 1st_src_line
	566	+ movdqa xmm1, [esi+ecx] ; 2nd_src_line
	567	+
	568	+ ; packing & avg
	569	+ 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
	572	+ ; another implementation for xmm2 high bits
	573	+; psubb xmm2, xmm0 ; h 0 g 0 f 0 e 0 d 0 c 0 b 0 a 0
	574	+; psrlw xmm2, 8 ; 0 h 0 g 0 f 0 e 0 d 0 c 0 b 0 a
	575	+ pavgb xmm0, xmm2
	576	+
	577	+ movdqa xmm3, xmm1
	578	+ pshufb xmm1, xmm7
	579	+ pshufb xmm3, xmm6
	580	+; psubb xmm3, xmm1
	581	+; psrlw xmm3, 8
	582	+ pavgb xmm1, xmm3
	583	+
	584	+ pavgb xmm0, xmm1
	585	+ packuswb xmm0, xmm1
	586	+
	587	+ ; write pDst
	588	+ movq [edi], xmm0
	589	+
	590	+ ; next SMB
	591	+ lea esi, [esi+16]
	592	+ lea edi, [edi+8]
	593	+
	594	+ dec eax
	595	+ jg near .xloops
	596	+
	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
	611	+ ret
612	612
613	613	; got about 65% improvement over DyadicBilinearDownsamplerWidthx32_sse
614	614	;***********************************************************************
615		−; void DyadicBilinearDownsamplerWidthx32_sse4( unsigned char* pDst, const int iDstStride,
616		−; unsigned char* pSrc, const int iSrcStride,
617		−; const int iSrcWidth, const int iSrcHeight );
	615	+; void DyadicBilinearDownsamplerWidthx32_sse4( unsigned char* pDst, const int iDstStride,
	616	+; unsigned char* pSrc, const int iSrcStride,
	617	+; const int iSrcWidth, const int iSrcHeight );
618	618	;***********************************************************************
619	619	WELS_EXTERN DyadicBilinearDownsamplerWidthx32_sse4
620		− push ebx
621		− push edx
622		− push esi
623		− push edi
624		− push ebp
	620	+ push ebx
	621	+ push edx
	622	+ push esi
	623	+ push edi
	624	+ push ebp
625	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
	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	631
632		− sar ebp, $01 ; iSrcHeight >> 1
	632	+ sar ebp, $01 ; iSrcHeight >> 1
633	633
634		− movdqa xmm7, [shufb_mask_low] ; mask low
635		− movdqa xmm6, [shufb_mask_high] ; mask high
	634	+ movdqa xmm7, [shufb_mask_low] ; mask low
	635	+ movdqa xmm6, [shufb_mask_high] ; mask high
636	636
637	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)
643		− ; each loop = source bandwidth: 32 bytes
	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)
	643	+ ; each loop = source bandwidth: 32 bytes
644	644	.xloops:
645		− ; 1st part horizonal loop: x16 bytes
646		− ; mem hi<- ->lo
647		− ;1st Line Src: xmm0: h H g G f F e E d D c C b B a A
648		− ; xmm1: p P o O n N m M l L k K j J i I
649		− ;2nd Line Src: xmm2: h H g G f F e E d D c C b B a A
650		− ; xmm3: p P o O n N m M l L k K j J i I
651		− ;=> target:
652		− ;: P O N M L K J I H G F E D C B A
653		− ;: p o n m l k j i h g f e d c b a
654		− ;: P .. A
655		− ;: p .. a
656		−
657		− ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
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
662		−
663		− ; packing & avg
664		− movdqa xmm4, xmm0 ; h H g G f F e E d D c C b B a A
665		− pshufb xmm0, xmm7 ; 0 H 0 G 0 F 0 E 0 D 0 C 0 B 0 A
666		− pshufb xmm4, xmm6 ; 0 h 0 g 0 f 0 e 0 d 0 c 0 b 0 a
667		−; psubb xmm4, xmm0 ; h 0 g 0 f 0 e 0 d 0 c 0 b 0 a 0
668		−; psrlw xmm4, 8 ; 0 h 0 g 0 f 0 e 0 d 0 c 0 b 0 a
669		− pavgb xmm0, xmm4
670		−
671		− movdqa xmm5, xmm1
672		− pshufb xmm1, xmm7
673		− pshufb xmm5, xmm6
674		−; psubb xmm5, xmm1
675		−; psrlw xmm5, 8
676		− pavgb xmm1, xmm5
677		−
678		− movdqa xmm4, xmm2
679		− pshufb xmm2, xmm7
680		− pshufb xmm4, xmm6
681		−; psubb xmm4, xmm2
682		−; psrlw xmm4, 8
683		− pavgb xmm2, xmm4
684		−
685		− movdqa xmm5, xmm3
686		− pshufb xmm3, xmm7
687		− pshufb xmm5, xmm6
688		−; psubb xmm5, xmm3
689		−; psrlw xmm5, 8
690		− pavgb xmm3, xmm5
691		−
692		− packuswb xmm0, xmm1
693		− packuswb xmm2, xmm3
694		− pavgb xmm0, xmm2
695		−
696		− ; write pDst
697		− movdqa [edi], xmm0
698		−
699		− ; next SMB
700		− lea esi, [esi+32]
701		− lea edi, [edi+16]
702		−
703		− dec eax
704		− jg near .xloops
705		−
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
720		− ret
	645	+ ; 1st part horizonal loop: x16 bytes
	646	+ ; mem hi<- ->lo
	647	+ ;1st Line Src: xmm0: h H g G f F e E d D c C b B a A
	648	+ ; xmm1: p P o O n N m M l L k K j J i I
	649	+ ;2nd Line Src: xmm2: h H g G f F e E d D c C b B a A
	650	+ ; xmm3: p P o O n N m M l L k K j J i I
	651	+ ;=> target:
	652	+ ;: P O N M L K J I H G F E D C B A
	653	+ ;: p o n m l k j i h g f e d c b a
	654	+ ;: P .. A
	655	+ ;: p .. a
	656	+
	657	+ ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
	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
	662	+
	663	+ ; packing & avg
	664	+ movdqa xmm4, xmm0 ; h H g G f F e E d D c C b B a A
	665	+ pshufb xmm0, xmm7 ; 0 H 0 G 0 F 0 E 0 D 0 C 0 B 0 A
	666	+ pshufb xmm4, xmm6 ; 0 h 0 g 0 f 0 e 0 d 0 c 0 b 0 a
	667	+; psubb xmm4, xmm0 ; h 0 g 0 f 0 e 0 d 0 c 0 b 0 a 0
	668	+; psrlw xmm4, 8 ; 0 h 0 g 0 f 0 e 0 d 0 c 0 b 0 a
	669	+ pavgb xmm0, xmm4
	670	+
	671	+ movdqa xmm5, xmm1
	672	+ pshufb xmm1, xmm7
	673	+ pshufb xmm5, xmm6
	674	+; psubb xmm5, xmm1
	675	+; psrlw xmm5, 8
	676	+ pavgb xmm1, xmm5
	677	+
	678	+ movdqa xmm4, xmm2
	679	+ pshufb xmm2, xmm7
	680	+ pshufb xmm4, xmm6
	681	+; psubb xmm4, xmm2
	682	+; psrlw xmm4, 8
	683	+ pavgb xmm2, xmm4
	684	+
	685	+ movdqa xmm5, xmm3
	686	+ pshufb xmm3, xmm7
	687	+ pshufb xmm5, xmm6
	688	+; psubb xmm5, xmm3
	689	+; psrlw xmm5, 8
	690	+ pavgb xmm3, xmm5
	691	+
	692	+ packuswb xmm0, xmm1
	693	+ packuswb xmm2, xmm3
	694	+ pavgb xmm0, xmm2
	695	+
	696	+ ; write pDst
	697	+ movdqa [edi], xmm0
	698	+
	699	+ ; next SMB
	700	+ lea esi, [esi+32]
	701	+ lea edi, [edi+16]
	702	+
	703	+ dec eax
	704	+ jg near .xloops
	705	+
	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
	720	+ ret
721	721
722	722	;***********************************************************************
723		−; void DyadicBilinearDownsamplerWidthx16_sse4( unsigned char* pDst, const int iDstStride,
724		−; unsigned char* pSrc, const int iSrcStride,
725		−; const int iSrcWidth, const int iSrcHeight );
	723	+; void DyadicBilinearDownsamplerWidthx16_sse4( unsigned char* pDst, const int iDstStride,
	724	+; unsigned char* pSrc, const int iSrcStride,
	725	+; const int iSrcWidth, const int iSrcHeight );
726	726	;***********************************************************************
727	727	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
	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
743	743
744	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)
750		− ; each loop = source bandwidth: 16 bytes
	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)
	750	+ ; each loop = source bandwidth: 16 bytes
751	751	.xloops:
752		− ; horizonal loop: x16 bytes by source
753		− ; mem hi<- ->lo
754		− ;1st line pSrc: xmm0: h H g G f F e E d D c C b B a A
755		− ;2nd line pSrc: xmm1: p P o O n N m M l L k K j J i I
756		− ;=> target:
757		− ;: H G F E D C B A, P O N M L K J I
758		− ;: h g f e d c b a, p o n m l k j i
	752	+ ; horizonal loop: x16 bytes by source
	753	+ ; mem hi<- ->lo
	754	+ ;1st line pSrc: xmm0: h H g G f F e E d D c C b B a A
	755	+ ;2nd line pSrc: xmm1: p P o O n N m M l L k K j J i I
	756	+ ;=> target:
	757	+ ;: H G F E D C B A, P O N M L K J I
	758	+ ;: h g f e d c b a, p o n m l k j i
759	759
760		− ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
761		− movntdqa xmm0, [esi] ; 1st_src_line
762		− movntdqa xmm1, [esi+ecx] ; 2nd_src_line
	760	+ ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
	761	+ movntdqa xmm0, [esi] ; 1st_src_line
	762	+ movntdqa xmm1, [esi+ecx] ; 2nd_src_line
763	763
764		− ; packing & avg
765		− 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
768		−; psubb xmm2, xmm0 ; h 0 g 0 f 0 e 0 d 0 c 0 b 0 a 0
769		−; psrlw xmm2, 8 ; 0 h 0 g 0 f 0 e 0 d 0 c 0 b 0 a
770		− pavgb xmm0, xmm2
	764	+ ; packing & avg
	765	+ 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
	768	+; psubb xmm2, xmm0 ; h 0 g 0 f 0 e 0 d 0 c 0 b 0 a 0
	769	+; psrlw xmm2, 8 ; 0 h 0 g 0 f 0 e 0 d 0 c 0 b 0 a
	770	+ pavgb xmm0, xmm2
771	771
772		− movdqa xmm3, xmm1
773		− pshufb xmm1, xmm7
774		− pshufb xmm3, xmm6
775		−; psubb xmm3, xmm1
776		−; psrlw xmm3, 8
777		− pavgb xmm1, xmm3
	772	+ movdqa xmm3, xmm1
	773	+ pshufb xmm1, xmm7
	774	+ pshufb xmm3, xmm6
	775	+; psubb xmm3, xmm1
	776	+; psrlw xmm3, 8
	777	+ pavgb xmm1, xmm3
778	778
779		− pavgb xmm0, xmm1
780		− packuswb xmm0, xmm1
	779	+ pavgb xmm0, xmm1
	780	+ packuswb xmm0, xmm1
781	781
782		− ; write pDst
783		− movq [edi], xmm0
	782	+ ; write pDst
	783	+ movq [edi], xmm0
784	784
785		− ; next SMB
786		− lea esi, [esi+16]
787		− lea edi, [edi+8]
	785	+ ; next SMB
	786	+ lea esi, [esi+16]
	787	+ lea edi, [edi+8]
788	788
789		− dec eax
790		− jg near .xloops
	789	+ dec eax
	790	+ jg near .xloops
791	791
792		− ; 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]
	792	+ ; 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	797
798		− dec ebp
799		− jg near .yloops
	798	+ dec ebp
	799	+ jg near .yloops
800	800
801		− pop ebp
802		− pop edi
803		− pop esi
804		− pop edx
805		− pop ebx
806		− ret
	801	+ pop ebp
	802	+ pop edi
	803	+ pop esi
	804	+ pop edx
	805	+ pop ebx
	806	+ ret
807	807
808	808
809	809
@@ -811,395 +811,395 @@ WELS_EXTERN DyadicBilinearDownsamplerWidthx16_sse4 · 1 unmodified line
811	811
812	812	;**************************************************************************************************************
813	813	;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,
	814	+; unsigned char* pSrc, const int iSrcStride, const int iSrcWidth, const int iSrcHeight,
815	815	; unsigned int uiScaleX, unsigned int uiScaleY );
816	816	;{
817	817	;**************************************************************************************************************
818	818
819		−WELS_EXTERN GeneralBilinearAccurateDownsampler_sse2
820		− push ebp
821		− push esi
822		− push edi
823		− push ebx
824		−%define pushsize 16
825		−%define localsize 28
826		−%define pDstData esp + pushsize + localsize + 4
827		−%define dwDstStride esp + pushsize + localsize + 8
828		−%define dwDstWidth esp + pushsize + localsize + 12
829		−%define dwDstHeight esp + pushsize + localsize + 16
830		−%define pSrcData esp + pushsize + localsize + 20
831		−%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
841		− sub esp, localsize
842		−
843		− pxor xmm0, xmm0
844		− mov edx, 32767
845		− mov eax, [uiScaleX]
846		− and eax, 32767
847		− mov ebx, eax
848		− neg ebx
849		− and ebx, 32767
850		− movd xmm1, eax ; uinc(uiScaleX mod 32767)
851		− movd xmm2, ebx ; -uinc
852		− psllq xmm1, 32
853		− por xmm1, xmm2 ; 0 0 uinc -uinc (dword)
854		− pshufd xmm7, xmm1, 01000100b ; xmm7: uinc -uinc uinc -uinc
855		−
856		− mov eax, [uiScaleY]
857		− and eax, 32767
858		− mov ebx, eax
859		− neg ebx
860		− and ebx, 32767
861		− movd xmm6, eax ; vinc(uiScaleY mod 32767)
862		− movd xmm2, ebx ; -vinc
863		− psllq xmm6, 32
864		− por xmm6, xmm2 ; 0 0 vinc -vinc (dword)
865		− pshufd xmm6, xmm6, 01010000b ; xmm6: vinc vinc -vinc -vinc
866		−
867		− mov edx, 40003fffh
868		− movd xmm5, edx
869		− punpcklwd xmm5, xmm0 ; 16384 16383
870		− pshufd xmm5, xmm5, 01000100b ; xmm5: 16384 16383 16384 16383
	819	+WELS_EXTERN GeneralBilinearAccurateDownsampler_sse2
	820	+ push ebp
	821	+ push esi
	822	+ push edi
	823	+ push ebx
	824	+%define pushsize 16
	825	+%define localsize 28
	826	+%define pDstData esp + pushsize + localsize + 4
	827	+%define dwDstStride esp + pushsize + localsize + 8
	828	+%define dwDstWidth esp + pushsize + localsize + 12
	829	+%define dwDstHeight esp + pushsize + localsize + 16
	830	+%define pSrcData esp + pushsize + localsize + 20
	831	+%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
	841	+ sub esp, localsize
	842	+
	843	+ pxor xmm0, xmm0
	844	+ mov edx, 32767
	845	+ mov eax, [uiScaleX]
	846	+ and eax, 32767
	847	+ mov ebx, eax
	848	+ neg ebx
	849	+ and ebx, 32767
	850	+ movd xmm1, eax ; uinc(uiScaleX mod 32767)
	851	+ movd xmm2, ebx ; -uinc
	852	+ psllq xmm1, 32
	853	+ por xmm1, xmm2 ; 0 0 uinc -uinc (dword)
	854	+ pshufd xmm7, xmm1, 01000100b ; xmm7: uinc -uinc uinc -uinc
	855	+
	856	+ mov eax, [uiScaleY]
	857	+ and eax, 32767
	858	+ mov ebx, eax
	859	+ neg ebx
	860	+ and ebx, 32767
	861	+ movd xmm6, eax ; vinc(uiScaleY mod 32767)
	862	+ movd xmm2, ebx ; -vinc
	863	+ psllq xmm6, 32
	864	+ por xmm6, xmm2 ; 0 0 vinc -vinc (dword)
	865	+ pshufd xmm6, xmm6, 01010000b ; xmm6: vinc vinc -vinc -vinc
	866	+
	867	+ mov edx, 40003fffh
	868	+ movd xmm5, edx
	869	+ punpcklwd xmm5, xmm0 ; 16384 16383
	870	+ pshufd xmm5, xmm5, 01000100b ; xmm5: 16384 16383 16384 16383
871	871
872	872
873	873	DOWNSAMPLE:
874	874
875		− mov eax, [dwDstHeight]
876		− mov edi, [pDstData]
877		− mov edx, [dwDstStride]
878		− mov ecx, [dwDstWidth]
879		− sub edx, ecx
880		− mov [dstStep], edx ; stride - width
881		− dec eax
882		− mov [tmpHeight], eax
883		− mov eax, 16384
884		− mov [yInverse], eax
	875	+ mov eax, [dwDstHeight]
	876	+ mov edi, [pDstData]
	877	+ mov edx, [dwDstStride]
	878	+ mov ecx, [dwDstWidth]
	879	+ sub edx, ecx
	880	+ mov [dstStep], edx ; stride - width
	881	+ dec eax
	882	+ mov [tmpHeight], eax
	883	+ mov eax, 16384
	884	+ mov [yInverse], eax
885	885
886		− pshufd xmm4, xmm5, 01010000b ; initial v to 16384 16384 16383 16383
	886	+ pshufd xmm4, xmm5, 01010000b ; initial v to 16384 16384 16383 16383
887	887
888	888	HEIGHT:
889		− mov eax, [yInverse]
890		− mov esi, [pSrcData]
891		− shr eax, 15
892		− mul dword [dwSrcStride]
893		− add esi, eax ; get current row address
894		− mov ebp, esi
895		− add ebp, [dwSrcStride]
	889	+ mov eax, [yInverse]
	890	+ mov esi, [pSrcData]
	891	+ shr eax, 15
	892	+ mul dword [dwSrcStride]
	893	+ add esi, eax ; get current row address
	894	+ mov ebp, esi
	895	+ add ebp, [dwSrcStride]
896	896
897		− mov eax, 16384
898		− mov [xInverse], eax
899		− mov ecx, [dwDstWidth]
900		− dec ecx
	897	+ mov eax, 16384
	898	+ mov [xInverse], eax
	899	+ mov ecx, [dwDstWidth]
	900	+ dec ecx
901	901
902		− movdqa xmm3, xmm5 ; initial u to 16384 16383 16384 16383
	902	+ movdqa xmm3, xmm5 ; initial u to 16384 16383 16384 16383
903	903
904	904	WIDTH:
905		− mov eax, [xInverse]
906		− shr eax, 15
907		−
908		− movd xmm1, [esi+eax] ; xxxxxxba
909		− movd xmm2, [ebp+eax] ; xxxxxxdc
910		− pxor xmm0, xmm0
911		− punpcklwd xmm1, xmm2 ; xxxxdcba
912		− punpcklbw xmm1, xmm0 ; 0d0c0b0a
913		− punpcklwd xmm1, xmm0 ; 000d000c000b000a
914		−
915		− movdqa xmm2, xmm4 ; xmm2: vv(1-v)(1-v) tmpv
916		− pmaddwd xmm2, xmm3 ; mul u(1-u)u(1-u) on xmm2
917		− movdqa xmm0, xmm2
918		− pmuludq xmm2, xmm1
919		− psrlq xmm0, 32
920		− psrlq xmm1, 32
921		− pmuludq xmm0, xmm1
922		− paddq xmm2, xmm0
923		− pshufd xmm1, xmm2, 00001110b
924		− paddq xmm2, xmm1
925		− psrlq xmm2, 29
926		−
927		− movd eax, xmm2
928		− inc eax
929		− shr eax, 1
930		− mov [edi], al
931		− inc edi
932		−
933		− mov eax, [uiScaleX]
934		− add [xInverse], eax
935		−
936		− paddw xmm3, xmm7 ; inc u
937		− psllw xmm3, 1
938		− psrlw xmm3, 1
939		−
940		− loop WIDTH
	905	+ mov eax, [xInverse]
	906	+ shr eax, 15
	907	+
	908	+ movd xmm1, [esi+eax] ; xxxxxxba
	909	+ movd xmm2, [ebp+eax] ; xxxxxxdc
	910	+ pxor xmm0, xmm0
	911	+ punpcklwd xmm1, xmm2 ; xxxxdcba
	912	+ punpcklbw xmm1, xmm0 ; 0d0c0b0a
	913	+ punpcklwd xmm1, xmm0 ; 000d000c000b000a
	914	+
	915	+ movdqa xmm2, xmm4 ; xmm2: vv(1-v)(1-v) tmpv
	916	+ pmaddwd xmm2, xmm3 ; mul u(1-u)u(1-u) on xmm2
	917	+ movdqa xmm0, xmm2
	918	+ pmuludq xmm2, xmm1
	919	+ psrlq xmm0, 32
	920	+ psrlq xmm1, 32
	921	+ pmuludq xmm0, xmm1
	922	+ paddq xmm2, xmm0
	923	+ pshufd xmm1, xmm2, 00001110b
	924	+ paddq xmm2, xmm1
	925	+ psrlq xmm2, 29
	926	+
	927	+ movd eax, xmm2
	928	+ inc eax
	929	+ shr eax, 1
	930	+ mov [edi], al
	931	+ inc edi
	932	+
	933	+ mov eax, [uiScaleX]
	934	+ add [xInverse], eax
	935	+
	936	+ paddw xmm3, xmm7 ; inc u
	937	+ psllw xmm3, 1
	938	+ psrlw xmm3, 1
	939	+
	940	+ loop WIDTH
941	941
942	942	WIDTH_END:
943		− mov eax, [xInverse]
944		− shr eax, 15
945		− mov cl, [esi+eax]
946		− mov [edi], cl
947		− inc edi
	943	+ mov eax, [xInverse]
	944	+ shr eax, 15
	945	+ mov cl, [esi+eax]
	946	+ mov [edi], cl
	947	+ inc edi
948	948
949		− mov eax, [uiScaleY]
950		− add [yInverse], eax
951		− add edi, [dstStep]
	949	+ mov eax, [uiScaleY]
	950	+ add [yInverse], eax
	951	+ add edi, [dstStep]
952	952
953		− paddw xmm4, xmm6 ; inc v
954		− psllw xmm4, 1
955		− psrlw xmm4, 1
	953	+ paddw xmm4, xmm6 ; inc v
	954	+ psllw xmm4, 1
	955	+ psrlw xmm4, 1
956	956
957		− dec dword [tmpHeight]
958		− jg HEIGHT
	957	+ dec dword [tmpHeight]
	958	+ jg HEIGHT
959	959
960	960
961	961	LAST_ROW:
962		− mov eax, [yInverse]
963		− mov esi, [pSrcData]
964		− shr eax, 15
965		− mul dword [dwSrcStride]
966		− add esi, eax ; get current row address
	962	+ mov eax, [yInverse]
	963	+ mov esi, [pSrcData]
	964	+ shr eax, 15
	965	+ mul dword [dwSrcStride]
	966	+ add esi, eax ; get current row address
967	967
968		− mov eax, 16384
969		− mov [xInverse], eax
970		− mov ecx, [dwDstWidth]
	968	+ mov eax, 16384
	969	+ mov [xInverse], eax
	970	+ mov ecx, [dwDstWidth]
971	971
972	972	LAST_ROW_WIDTH:
973		− mov eax, [xInverse]
974		− shr eax, 15
	973	+ mov eax, [xInverse]
	974	+ shr eax, 15
975	975
976		− mov al, [esi+eax]
977		− mov [edi], al
978		− inc edi
	976	+ mov al, [esi+eax]
	977	+ mov [edi], al
	978	+ inc edi
979	979
980		− mov eax, [uiScaleX]
981		− add [xInverse], eax
	980	+ mov eax, [uiScaleX]
	981	+ add [xInverse], eax
982	982
983		− loop LAST_ROW_WIDTH
	983	+ loop LAST_ROW_WIDTH
984	984
985	985	LAST_ROW_END:
986	986
987		− add esp, localsize
988		− pop ebx
989		− pop edi
990		− pop esi
991		− pop ebp
992		−%undef pushsize
993		−%undef localsize
994		−%undef pSrcData
995		−%undef dwSrcWidth
996		−%undef dwSrcHeight
997		−%undef dwSrcStride
998		−%undef pDstData
999		−%undef dwDstWidth
1000		−%undef dwDstHeight
1001		−%undef dwDstStride
1002		−%undef scale
1003		−%undef uiScaleX
1004		−%undef uiScaleY
1005		−%undef tmpHeight
1006		−%undef yInverse
1007		−%undef xInverse
1008		−%undef dstStep
1009		− ret
	987	+ add esp, localsize
	988	+ pop ebx
	989	+ pop edi
	990	+ pop esi
	991	+ pop ebp
	992	+%undef pushsize
	993	+%undef localsize
	994	+%undef pSrcData
	995	+%undef dwSrcWidth
	996	+%undef dwSrcHeight
	997	+%undef dwSrcStride
	998	+%undef pDstData
	999	+%undef dwDstWidth
	1000	+%undef dwDstHeight
	1001	+%undef dwDstStride
	1002	+%undef scale
	1003	+%undef uiScaleX
	1004	+%undef uiScaleY
	1005	+%undef tmpHeight
	1006	+%undef yInverse
	1007	+%undef xInverse
	1008	+%undef dstStep
	1009	+ ret
1010	1010
1011	1011
1012	1012
1013	1013
1014	1014	;**************************************************************************************************************
1015	1015	;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,
	1016	+; unsigned char* pSrc, const int iSrcStride, const int iSrcWidth, const int iSrcHeight,
1017	1017	; unsigned int uiScaleX, unsigned int uiScaleY );
1018	1018	;{
1019	1019	;**************************************************************************************************************
1020	1020
1021		−WELS_EXTERN GeneralBilinearFastDownsampler_sse2
1022		− push ebp
1023		− push esi
1024		− push edi
1025		− push ebx
1026		−%define pushsize 16
1027		−%define localsize 28
1028		−%define pDstData esp + pushsize + localsize + 4
1029		−%define dwDstStride esp + pushsize + localsize + 8
1030		−%define dwDstWidth esp + pushsize + localsize + 12
1031		−%define dwDstHeight esp + pushsize + localsize + 16
1032		−%define pSrcData esp + pushsize + localsize + 20
1033		−%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
1043		− sub esp, localsize
1044		−
1045		− pxor xmm0, xmm0
1046		− mov edx, 65535
1047		− mov eax, [uiScaleX]
1048		− and eax, edx
1049		− mov ebx, eax
1050		− neg ebx
1051		− and ebx, 65535
1052		− movd xmm1, eax ; uinc(uiScaleX mod 65536)
1053		− movd xmm2, ebx ; -uinc
1054		− psllq xmm1, 32
1055		− por xmm1, xmm2 ; 0 uinc 0 -uinc
1056		− pshuflw xmm7, xmm1, 10001000b ; xmm7: uinc -uinc uinc -uinc
1057		−
1058		− mov eax, [uiScaleY]
1059		− and eax, 32767
1060		− mov ebx, eax
1061		− neg ebx
1062		− and ebx, 32767
1063		− movd xmm6, eax ; vinc(uiScaleY mod 32767)
1064		− movd xmm2, ebx ; -vinc
1065		− psllq xmm6, 32
1066		− por xmm6, xmm2 ; 0 vinc 0 -vinc
1067		− pshuflw xmm6, xmm6, 10100000b ; xmm6: vinc vinc -vinc -vinc
1068		−
1069		− mov edx, 80007fffh ; 32768 32767
1070		− movd xmm5, edx
1071		− pshuflw xmm5, xmm5, 01000100b ; 32768 32767 32768 32767
1072		− mov ebx, 16384
	1021	+WELS_EXTERN GeneralBilinearFastDownsampler_sse2
	1022	+ push ebp
	1023	+ push esi
	1024	+ push edi
	1025	+ push ebx
	1026	+%define pushsize 16
	1027	+%define localsize 28
	1028	+%define pDstData esp + pushsize + localsize + 4
	1029	+%define dwDstStride esp + pushsize + localsize + 8
	1030	+%define dwDstWidth esp + pushsize + localsize + 12
	1031	+%define dwDstHeight esp + pushsize + localsize + 16
	1032	+%define pSrcData esp + pushsize + localsize + 20
	1033	+%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
	1043	+ sub esp, localsize
	1044	+
	1045	+ pxor xmm0, xmm0
	1046	+ mov edx, 65535
	1047	+ mov eax, [uiScaleX]
	1048	+ and eax, edx
	1049	+ mov ebx, eax
	1050	+ neg ebx
	1051	+ and ebx, 65535
	1052	+ movd xmm1, eax ; uinc(uiScaleX mod 65536)
	1053	+ movd xmm2, ebx ; -uinc
	1054	+ psllq xmm1, 32
	1055	+ por xmm1, xmm2 ; 0 uinc 0 -uinc
	1056	+ pshuflw xmm7, xmm1, 10001000b ; xmm7: uinc -uinc uinc -uinc
	1057	+
	1058	+ mov eax, [uiScaleY]
	1059	+ and eax, 32767
	1060	+ mov ebx, eax
	1061	+ neg ebx
	1062	+ and ebx, 32767
	1063	+ movd xmm6, eax ; vinc(uiScaleY mod 32767)
	1064	+ movd xmm2, ebx ; -vinc
	1065	+ psllq xmm6, 32
	1066	+ por xmm6, xmm2 ; 0 vinc 0 -vinc
	1067	+ pshuflw xmm6, xmm6, 10100000b ; xmm6: vinc vinc -vinc -vinc
	1068	+
	1069	+ mov edx, 80007fffh ; 32768 32767
	1070	+ movd xmm5, edx
	1071	+ pshuflw xmm5, xmm5, 01000100b ; 32768 32767 32768 32767
	1072	+ mov ebx, 16384
1073	1073
1074	1074
1075	1075	FAST_DOWNSAMPLE:
1076	1076
1077		− mov eax, [dwDstHeight]
1078		− mov edi, [pDstData]
1079		− mov edx, [dwDstStride]
1080		− mov ecx, [dwDstWidth]
1081		− sub edx, ecx
1082		− mov [dstStep], edx ; stride - width
1083		− dec eax
1084		− mov [tmpHeight], eax
1085		− mov eax, 16384
1086		− mov [yInverse], eax
	1077	+ mov eax, [dwDstHeight]
	1078	+ mov edi, [pDstData]
	1079	+ mov edx, [dwDstStride]
	1080	+ mov ecx, [dwDstWidth]
	1081	+ sub edx, ecx
	1082	+ mov [dstStep], edx ; stride - width
	1083	+ dec eax
	1084	+ mov [tmpHeight], eax
	1085	+ mov eax, 16384
	1086	+ mov [yInverse], eax
1087	1087
1088		− pshuflw xmm4, xmm5, 01010000b
1089		− psrlw xmm4, 1 ; initial v to 16384 16384 16383 16383
	1088	+ pshuflw xmm4, xmm5, 01010000b
	1089	+ psrlw xmm4, 1 ; initial v to 16384 16384 16383 16383
1090	1090
1091	1091	FAST_HEIGHT:
1092		− mov eax, [yInverse]
1093		− mov esi, [pSrcData]
1094		− shr eax, 15
1095		− mul dword [dwSrcStride]
1096		− add esi, eax ; get current row address
1097		− mov ebp, esi
1098		− add ebp, [dwSrcStride]
	1092	+ mov eax, [yInverse]
	1093	+ mov esi, [pSrcData]
	1094	+ shr eax, 15
	1095	+ mul dword [dwSrcStride]
	1096	+ add esi, eax ; get current row address
	1097	+ mov ebp, esi
	1098	+ add ebp, [dwSrcStride]
1099	1099
1100		− mov eax, 32768
1101		− mov [xInverse], eax
1102		− mov ecx, [dwDstWidth]
1103		− dec ecx
	1100	+ mov eax, 32768
	1101	+ mov [xInverse], eax
	1102	+ mov ecx, [dwDstWidth]
	1103	+ dec ecx
1104	1104
1105		− movdqa xmm3, xmm5 ; initial u to 32768 32767 32768 32767
	1105	+ movdqa xmm3, xmm5 ; initial u to 32768 32767 32768 32767
1106	1106
1107	1107	FAST_WIDTH:
1108		− mov eax, [xInverse]
1109		− shr eax, 16
	1108	+ mov eax, [xInverse]
	1109	+ shr eax, 16
1110	1110
1111		− movd xmm1, [esi+eax] ; xxxxxxba
1112		− movd xmm2, [ebp+eax] ; xxxxxxdc
1113		− punpcklwd xmm1, xmm2 ; xxxxdcba
1114		− punpcklbw xmm1, xmm0 ; 0d0c0b0a
	1111	+ movd xmm1, [esi+eax] ; xxxxxxba
	1112	+ movd xmm2, [ebp+eax] ; xxxxxxdc
	1113	+ punpcklwd xmm1, xmm2 ; xxxxdcba
	1114	+ punpcklbw xmm1, xmm0 ; 0d0c0b0a
1115	1115
1116		− movdqa xmm2, xmm4 ; xmm2: vv(1-v)(1-v) tmpv
1117		− pmulhuw xmm2, xmm3 ; mul u(1-u)u(1-u) on xmm2
1118		− pmaddwd xmm2, xmm1
1119		− pshufd xmm1, xmm2, 00000001b
1120		− paddd xmm2, xmm1
1121		− movd xmm1, ebx
1122		− paddd xmm2, xmm1
1123		− psrld xmm2, 15
	1116	+ movdqa xmm2, xmm4 ; xmm2: vv(1-v)(1-v) tmpv
	1117	+ pmulhuw xmm2, xmm3 ; mul u(1-u)u(1-u) on xmm2
	1118	+ pmaddwd xmm2, xmm1
	1119	+ pshufd xmm1, xmm2, 00000001b
	1120	+ paddd xmm2, xmm1
	1121	+ movd xmm1, ebx
	1122	+ paddd xmm2, xmm1
	1123	+ psrld xmm2, 15
1124	1124
1125		− packuswb xmm2, xmm0
1126		− movd eax, xmm2
1127		− mov [edi], al
1128		− inc edi
	1125	+ packuswb xmm2, xmm0
	1126	+ movd eax, xmm2
	1127	+ mov [edi], al
	1128	+ inc edi
1129	1129
1130		− mov eax, [uiScaleX]
1131		− add [xInverse], eax
	1130	+ mov eax, [uiScaleX]
	1131	+ add [xInverse], eax
1132	1132
1133		− paddw xmm3, xmm7 ; inc u
	1133	+ paddw xmm3, xmm7 ; inc u
1134	1134
1135		− loop FAST_WIDTH
	1135	+ loop FAST_WIDTH
1136	1136
1137	1137	FAST_WIDTH_END:
1138		− mov eax, [xInverse]
1139		− shr eax, 16
1140		− mov cl, [esi+eax]
1141		− mov [edi], cl
1142		− inc edi
	1138	+ mov eax, [xInverse]
	1139	+ shr eax, 16
	1140	+ mov cl, [esi+eax]
	1141	+ mov [edi], cl
	1142	+ inc edi
1143	1143
1144		− mov eax, [uiScaleY]
1145		− add [yInverse], eax
1146		− add edi, [dstStep]
	1144	+ mov eax, [uiScaleY]
	1145	+ add [yInverse], eax
	1146	+ add edi, [dstStep]
1147	1147
1148		− paddw xmm4, xmm6 ; inc v
1149		− psllw xmm4, 1
1150		− psrlw xmm4, 1
	1148	+ paddw xmm4, xmm6 ; inc v
	1149	+ psllw xmm4, 1
	1150	+ psrlw xmm4, 1
1151	1151
1152		− dec dword [tmpHeight]
1153		− jg FAST_HEIGHT
	1152	+ dec dword [tmpHeight]
	1153	+ jg FAST_HEIGHT
1154	1154
1155	1155
1156	1156	FAST_LAST_ROW:
1157		− mov eax, [yInverse]
1158		− mov esi, [pSrcData]
1159		− shr eax, 15
1160		− mul dword [dwSrcStride]
1161		− add esi, eax ; get current row address
	1157	+ mov eax, [yInverse]
	1158	+ mov esi, [pSrcData]
	1159	+ shr eax, 15
	1160	+ mul dword [dwSrcStride]
	1161	+ add esi, eax ; get current row address
1162	1162
1163		− mov eax, 32768
1164		− mov [xInverse], eax
1165		− mov ecx, [dwDstWidth]
	1163	+ mov eax, 32768
	1164	+ mov [xInverse], eax
	1165	+ mov ecx, [dwDstWidth]
1166	1166
1167	1167	FAST_LAST_ROW_WIDTH:
1168		− mov eax, [xInverse]
1169		− shr eax, 16
	1168	+ mov eax, [xInverse]
	1169	+ shr eax, 16
1170	1170
1171		− mov al, [esi+eax]
1172		− mov [edi], al
1173		− inc edi
	1171	+ mov al, [esi+eax]
	1172	+ mov [edi], al
	1173	+ inc edi
1174	1174
1175		− mov eax, [uiScaleX]
1176		− add [xInverse], eax
	1175	+ mov eax, [uiScaleX]
	1176	+ add [xInverse], eax
1177	1177
1178		− loop FAST_LAST_ROW_WIDTH
	1178	+ loop FAST_LAST_ROW_WIDTH
1179	1179
1180	1180	FAST_LAST_ROW_END:
1181	1181
1182		− add esp, localsize
1183		− pop ebx
1184		− pop edi
1185		− pop esi
1186		− pop ebp
1187		−%undef pushsize
1188		−%undef localsize
1189		−%undef pSrcData
1190		−%undef dwSrcWidth
1191		−%undef dwSrcHeight
1192		−%undef dwSrcStride
1193		−%undef pDstData
1194		−%undef dwDstWidth
1195		−%undef dwDstHeight
1196		−%undef dwDstStride
1197		−%undef scale
1198		−%undef uiScaleX
1199		−%undef uiScaleY
1200		−%undef tmpHeight
1201		−%undef yInverse
1202		−%undef xInverse
1203		−%undef dstStep
1204		− ret
	1182	+ add esp, localsize
	1183	+ pop ebx
	1184	+ pop edi
	1185	+ pop esi
	1186	+ pop ebp
	1187	+%undef pushsize
	1188	+%undef localsize
	1189	+%undef pSrcData
	1190	+%undef dwSrcWidth
	1191	+%undef dwSrcHeight
	1192	+%undef dwSrcStride
	1193	+%undef pDstData
	1194	+%undef dwDstWidth
	1195	+%undef dwDstHeight
	1196	+%undef dwDstStride
	1197	+%undef scale
	1198	+%undef uiScaleX
	1199	+%undef uiScaleY
	1200	+%undef tmpHeight
	1201	+%undef yInverse
	1202	+%undef xInverse
	1203	+%undef dstStep
	1204	+ ret
1205	1205	%endif

codec/processing/src/x86/vaa.asm+1518−1518

@@ -48,100 +48,100 @@ · 47 unmodified lines
48	48	; Macros and other preprocessor constants
49	49	;***********************************************************************
50	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
	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	61	%endmacro ; END OF SUM_SQR_SSE2
62	62
63	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]
	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]
74	74	%endmacro
75	75
76	76	; by comparing it outperforms than phaddw(SSSE3) sets
77	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
	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	86	%endmacro ; END of SUM_WORD_8x2_SSE2
87	87
88	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
	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
109	109	%endmacro
110	110
111	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
	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
145	145	%endmacro
146	146
147	147	%macro WELS_SAD_SD_MAD_16x1_SSE2 7 ;esi:%5 edi:%6 ebx:%7
@@ -149,40 +149,40 @@ · 1 unmodified line
149	149	%define sum_cur_reg %2
150	150	%define sum_ref_reg %3
151	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
	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
172	172	%endmacro
173	173
174	174
175	175	%macro WELS_MAX_REG_SSE2 1 ; xmm1, xmm2, xmm3 can be used
176	176	%define max_reg %1
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
	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
186	186	%endmacro
187	187
188	188	%macro WELS_SAD_BGD_SQDIFF_16x1_SSE2 7 ;esi:%5 edi:%6 ebx:%7
@@ -190,50 +190,50 @@ · 1 unmodified line
190	190	%define sum_reg %2
191	191	%define mad_reg %3
192	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
	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
237	237	%endmacro
238	238
239	239
@@ -249,99 +249,99 @@ SECTION .text · 9 unmodified lines
249	249	; void SampleVariance16x16_sse2( uint8_t * y_ref, int32_t y_ref_stride, uint8_t * y_src, int32_t y_src_stride,SMotionTextureUnit* pMotionTexture );
250	250	;***********************************************************************
251	251	WELS_EXTERN SampleVariance16x16_sse2
252		− push esi
253		− push edi
254		− push ebx
255		−
256		− sub esp, 16
257		− %define SUM [esp]
258		− %define SUM_CUR [esp+4]
259		− %define SQR [esp+8]
260		− %define SQR_CUR [esp+12]
261		− %define PUSH_SIZE 28 ; 12 + 16
262		−
263		− mov edi, [esp+PUSH_SIZE+4] ; y_ref
264		− mov edx, [esp+PUSH_SIZE+8] ; y_ref_stride
265		− mov esi, [esp+PUSH_SIZE+12] ; y_src
266		− mov eax, [esp+PUSH_SIZE+16] ; y_src_stride
267		− mov ecx, 010h ; height = 16
268		−
269		− pxor xmm7, xmm7
270		− movdqu SUM, xmm7
	252	+ push esi
	253	+ push edi
	254	+ push ebx
	255	+
	256	+ sub esp, 16
	257	+ %define SUM [esp]
	258	+ %define SUM_CUR [esp+4]
	259	+ %define SQR [esp+8]
	260	+ %define SQR_CUR [esp+12]
	261	+ %define PUSH_SIZE 28 ; 12 + 16
	262	+
	263	+ mov edi, [esp+PUSH_SIZE+4] ; y_ref
	264	+ mov edx, [esp+PUSH_SIZE+8] ; y_ref_stride
	265	+ mov esi, [esp+PUSH_SIZE+12] ; y_src
	266	+ mov eax, [esp+PUSH_SIZE+16] ; y_src_stride
	267	+ mov ecx, 010h ; height = 16
	268	+
	269	+ pxor xmm7, xmm7
	270	+ movdqu SUM, xmm7
271	271
272	272	.hloops:
273		− movdqa xmm0, [edi] ; y_ref
274		− movdqa xmm1, [esi] ; y_src
275		− movdqa xmm2, xmm0 ; store first for future process
276		− movdqa xmm3, xmm1
277		− ; sum += diff;
278		− movdqa xmm4, xmm0
279		− psadbw xmm4, xmm1 ; 2 parts, [0,..,15], [64,..,79]
280		− ; to be continued for sum
281		− pshufd xmm5, xmm4, 0C6h ; 11000110 B
282		− paddw xmm4, xmm5
283		− movd ebx, xmm4
284		− add SUM, ebx
285		−
286		− ; sqr += diff * diff;
287		− pmaxub xmm0, xmm1
288		− pminub xmm1, xmm2
289		− psubb xmm0, xmm1 ; diff
290		− SUM_SQR_SSE2 xmm1, xmm0, xmm7 ; dst, pSrc, zero
291		− movd ebx, xmm1
292		− add SQR, ebx
293		−
294		− ; sum_cur += y_src[x];
295		− movdqa xmm0, xmm3 ; cur_orig
296		− movdqa xmm1, xmm0
297		− punpcklbw xmm0, xmm7
298		− punpckhbw xmm1, xmm7
299		− paddw xmm0, xmm1 ; 8x2
300		− SUM_WORD_8x2_SSE2 xmm0, xmm1
301		− movd ebx, xmm0
302		− and ebx, 0ffffh
303		− add SUM_CUR, ebx
304		−
305		− ; sqr_cur += y_src[x] * y_src[x];
306		− SUM_SQR_SSE2 xmm0, xmm3, xmm7 ; dst, pSrc, zero
307		− movd ebx, xmm0
308		− add SQR_CUR, ebx
309		−
310		− lea edi, [edi+edx]
311		− lea esi, [esi+eax]
312		− dec ecx
313		− jnz near .hloops
314		−
315		− mov ebx, 0
316		− mov bx, word SUM
317		− sar ebx, 8
318		− imul ebx, ebx
319		− mov ecx, SQR
320		− sar ecx, 8
321		− sub ecx, ebx
322		− mov edi, [esp+PUSH_SIZE+20] ; pMotionTexture
323		− mov [edi], cx ; to store uiMotionIndex
324		− mov ebx, 0
325		− mov bx, word SUM_CUR
326		− sar ebx, 8
327		− imul ebx, ebx
328		− mov ecx, SQR_CUR
329		− sar ecx, 8
330		− sub ecx, ebx
331		− mov [edi+2], cx ; to store uiTextureIndex
332		−
333		− %undef SUM
334		− %undef SUM_CUR
335		− %undef SQR
336		− %undef SQR_CUR
337		− %undef PUSH_SIZE
338		−
339		− add esp, 16
340		− pop ebx
341		− pop edi
342		− pop esi
343		−
344		− ret
	273	+ movdqa xmm0, [edi] ; y_ref
	274	+ movdqa xmm1, [esi] ; y_src
	275	+ movdqa xmm2, xmm0 ; store first for future process
	276	+ movdqa xmm3, xmm1
	277	+ ; sum += diff;
	278	+ movdqa xmm4, xmm0
	279	+ psadbw xmm4, xmm1 ; 2 parts, [0,..,15], [64,..,79]
	280	+ ; to be continued for sum
	281	+ pshufd xmm5, xmm4, 0C6h ; 11000110 B
	282	+ paddw xmm4, xmm5
	283	+ movd ebx, xmm4
	284	+ add SUM, ebx
	285	+
	286	+ ; sqr += diff * diff;
	287	+ pmaxub xmm0, xmm1
	288	+ pminub xmm1, xmm2
	289	+ psubb xmm0, xmm1 ; diff
	290	+ SUM_SQR_SSE2 xmm1, xmm0, xmm7 ; dst, pSrc, zero
	291	+ movd ebx, xmm1
	292	+ add SQR, ebx
	293	+
	294	+ ; sum_cur += y_src[x];
	295	+ movdqa xmm0, xmm3 ; cur_orig
	296	+ movdqa xmm1, xmm0
	297	+ punpcklbw xmm0, xmm7
	298	+ punpckhbw xmm1, xmm7
	299	+ paddw xmm0, xmm1 ; 8x2
	300	+ SUM_WORD_8x2_SSE2 xmm0, xmm1
	301	+ movd ebx, xmm0
	302	+ and ebx, 0ffffh
	303	+ add SUM_CUR, ebx
	304	+
	305	+ ; sqr_cur += y_src[x] * y_src[x];
	306	+ SUM_SQR_SSE2 xmm0, xmm3, xmm7 ; dst, pSrc, zero
	307	+ movd ebx, xmm0
	308	+ add SQR_CUR, ebx
	309	+
	310	+ lea edi, [edi+edx]
	311	+ lea esi, [esi+eax]
	312	+ dec ecx
	313	+ jnz near .hloops
	314	+
	315	+ mov ebx, 0
	316	+ mov bx, word SUM
	317	+ sar ebx, 8
	318	+ imul ebx, ebx
	319	+ mov ecx, SQR
	320	+ sar ecx, 8
	321	+ sub ecx, ebx
	322	+ mov edi, [esp+PUSH_SIZE+20] ; pMotionTexture
	323	+ mov [edi], cx ; to store uiMotionIndex
	324	+ mov ebx, 0
	325	+ mov bx, word SUM_CUR
	326	+ sar ebx, 8
	327	+ imul ebx, ebx
	328	+ mov ecx, SQR_CUR
	329	+ sar ecx, 8
	330	+ sub ecx, ebx
	331	+ mov [edi+2], cx ; to store uiTextureIndex
	332	+
	333	+ %undef SUM
	334	+ %undef SUM_CUR
	335	+ %undef SQR
	336	+ %undef SQR_CUR
	337	+ %undef PUSH_SIZE
	338	+
	339	+ add esp, 16
	340	+ pop ebx
	341	+ pop edi
	342	+ pop esi
	343	+
	344	+ ret
345	345
346	346
347	347
@@ -360,67 +360,67 @@ WELS_EXTERN VAACalcSad_sse2 · 12 unmodified lines
360	360	%define psadframe esp + pushsize + 24
361	361	%define psad8x8 esp + pushsize + 28
362	362	%define pushsize 12
363		− push esi
364		− push edi
365		− push ebx
366		− mov esi, [cur_data]
367		− mov edi, [ref_data]
368		− mov ebx, [iPicStride]
369		− mov edx, [psad8x8]
370		− mov eax, ebx
371		−
372		− shr dword [iPicWidth], 4 ; iPicWidth/16
373		− shr dword [iPicHeight], 4 ; iPicHeight/16
374		− shl eax, 4 ; iPicStride*16
375		− pxor xmm0, xmm0
376		− pxor xmm7, xmm7 ; iFrameSad
	363	+ push esi
	364	+ push edi
	365	+ push ebx
	366	+ mov esi, [cur_data]
	367	+ mov edi, [ref_data]
	368	+ mov ebx, [iPicStride]
	369	+ mov edx, [psad8x8]
	370	+ mov eax, ebx
	371	+
	372	+ shr dword [iPicWidth], 4 ; iPicWidth/16
	373	+ shr dword [iPicHeight], 4 ; iPicHeight/16
	374	+ shl eax, 4 ; iPicStride*16
	375	+ pxor xmm0, xmm0
	376	+ pxor xmm7, xmm7 ; iFrameSad
377	377	height_loop:
378		− mov ecx, dword [iPicWidth]
379		− push esi
380		− push edi
	378	+ mov ecx, dword [iPicWidth]
	379	+ push esi
	380	+ push edi
381	381	width_loop:
382		− pxor xmm6, xmm6 ;
383		− WELS_SAD_16x2_SSE2 esi,edi,ebx
384		− WELS_SAD_16x2_SSE2 esi,edi,ebx
385		− WELS_SAD_16x2_SSE2 esi,edi,ebx
386		− WELS_SAD_16x2_SSE2 esi,edi,ebx
387		− paddd xmm7, xmm6
388		− movd [edx], xmm6
389		− psrldq xmm6, 8
390		− movd [edx+4], xmm6
391		−
392		− pxor xmm6, xmm6
393		− WELS_SAD_16x2_SSE2 esi,edi,ebx
394		− WELS_SAD_16x2_SSE2 esi,edi,ebx
395		− WELS_SAD_16x2_SSE2 esi,edi,ebx
396		− WELS_SAD_16x2_SSE2 esi,edi,ebx
397		− paddd xmm7, xmm6
398		− movd [edx+8], xmm6
399		− psrldq xmm6, 8
400		− movd [edx+12], xmm6
401		−
402		− add edx, 16
403		− sub esi, eax
404		− sub edi, eax
405		− add esi, 16
406		− add edi, 16
407		−
408		− dec ecx
409		− jnz width_loop
410		−
411		− pop edi
412		− pop esi
413		− add esi, eax
414		− add edi, eax
415		−
416		− dec dword [iPicHeight]
417		− jnz height_loop
418		−
419		− mov edx, [psadframe]
420		− movdqa xmm5, xmm7
421		− psrldq xmm7, 8
422		− paddd xmm7, xmm5
423		− movd [edx], xmm7
	382	+ pxor xmm6, xmm6 ;
	383	+ WELS_SAD_16x2_SSE2 esi,edi,ebx
	384	+ WELS_SAD_16x2_SSE2 esi,edi,ebx
	385	+ WELS_SAD_16x2_SSE2 esi,edi,ebx
	386	+ WELS_SAD_16x2_SSE2 esi,edi,ebx
	387	+ paddd xmm7, xmm6
	388	+ movd [edx], xmm6
	389	+ psrldq xmm6, 8
	390	+ movd [edx+4], xmm6
	391	+
	392	+ pxor xmm6, xmm6
	393	+ WELS_SAD_16x2_SSE2 esi,edi,ebx
	394	+ WELS_SAD_16x2_SSE2 esi,edi,ebx
	395	+ WELS_SAD_16x2_SSE2 esi,edi,ebx
	396	+ WELS_SAD_16x2_SSE2 esi,edi,ebx
	397	+ paddd xmm7, xmm6
	398	+ movd [edx+8], xmm6
	399	+ psrldq xmm6, 8
	400	+ movd [edx+12], xmm6
	401	+
	402	+ add edx, 16
	403	+ sub esi, eax
	404	+ sub edi, eax
	405	+ add esi, 16
	406	+ add edi, 16
	407	+
	408	+ dec ecx
	409	+ jnz width_loop
	410	+
	411	+ pop edi
	412	+ pop esi
	413	+ add esi, eax
	414	+ add edi, eax
	415	+
	416	+ dec dword [iPicHeight]
	417	+ jnz height_loop
	418	+
	419	+ mov edx, [psadframe]
	420	+ movdqa xmm5, xmm7
	421	+ psrldq xmm7, 8
	422	+ paddd xmm7, xmm5
	423	+ movd [edx], xmm7
424	424
425	425	%undef cur_data
426	426	%undef ref_data
@@ -430,10 +430,10 @@ width_loop: · 3 unmodified lines
430	430	%undef psadframe
431	431	%undef psad8x8
432	432	%undef pushsize
433		− pop ebx
434		− pop edi
435		− pop esi
436		− ret
	433	+ pop ebx
	434	+ pop edi
	435	+ pop esi
	436	+ ret
437	437
438	438	%else ;64-bit
439	439
@@ -441,98 +441,98 @@ width_loop: · 1 unmodified line
441	441	; void SampleVariance16x16_sse2( uint8_t * y_ref, int32_t y_ref_stride, uint8_t * y_src, int32_t y_src_stride,SMotionTextureUnit* pMotionTexture );
442	442	;***********************************************************************
443	443	WELS_EXTERN SampleVariance16x16_sse2
444		− %define SUM r10;[esp]
445		− %define SUM_CUR r11;[esp+4]
446		− %define SQR r13;[esp+8]
447		− %define SQR_CUR r15;[esp+12]
448		−
449		− push r12
450		− push r13
451		− push r14
452		− push r15
453		− %assign push_num 4
454		− LOAD_5_PARA
455		− PUSH_XMM 8
456		− SIGN_EXTENSION r1,r1d
457		− SIGN_EXTENSION r3,r3d
458		−
459		− mov r12,010h
460		− pxor xmm7, xmm7
461		− movq SUM, xmm7
462		− movq SUM_CUR,xmm7
463		− movq SQR,xmm7
464		− movq SQR_CUR,xmm7
	444	+ %define SUM r10;[esp]
	445	+ %define SUM_CUR r11;[esp+4]
	446	+ %define SQR r13;[esp+8]
	447	+ %define SQR_CUR r15;[esp+12]
	448	+
	449	+ push r12
	450	+ push r13
	451	+ push r14
	452	+ push r15
	453	+ %assign push_num 4
	454	+ LOAD_5_PARA
	455	+ PUSH_XMM 8
	456	+ SIGN_EXTENSION r1,r1d
	457	+ SIGN_EXTENSION r3,r3d
	458	+
	459	+ mov r12,010h
	460	+ pxor xmm7, xmm7
	461	+ movq SUM, xmm7
	462	+ movq SUM_CUR,xmm7
	463	+ movq SQR,xmm7
	464	+ movq SQR_CUR,xmm7
465	465
466	466	.hloops:
467		− mov r14,0
468		− movdqa xmm0, [r0] ; y_ref
469		− movdqa xmm1, [r2] ; y_src
470		− movdqa xmm2, xmm0 ; store first for future process
471		− movdqa xmm3, xmm1
472		− ; sum += diff;
473		− movdqa xmm4, xmm0
474		− psadbw xmm4, xmm1 ; 2 parts, [0,..,15], [64,..,79]
475		− ; to be continued for sum
476		− pshufd xmm5, xmm4, 0C6h ; 11000110 B
477		− paddw xmm4, xmm5
478		− movd r14d, xmm4
479		− add SUM, r14
480		−
481		− ; sqr += diff * diff;
482		− pmaxub xmm0, xmm1
483		− pminub xmm1, xmm2
484		− psubb xmm0, xmm1 ; diff
485		− SUM_SQR_SSE2 xmm1, xmm0, xmm7 ; dst, pSrc, zero
486		− movd r14d, xmm1
487		− add SQR, r14
488		−
489		− ; sum_cur += y_src[x];
490		− movdqa xmm0, xmm3 ; cur_orig
491		− movdqa xmm1, xmm0
492		− punpcklbw xmm0, xmm7
493		− punpckhbw xmm1, xmm7
494		− paddw xmm0, xmm1 ; 8x2
495		− SUM_WORD_8x2_SSE2 xmm0, xmm1
496		− movd r14d, xmm0
497		− and r14, 0ffffh
498		− add SUM_CUR, r14
499		−
500		− ; sqr_cur += y_src[x] * y_src[x];
501		− SUM_SQR_SSE2 xmm0, xmm3, xmm7 ; dst, pSrc, zero
502		− movd r14d, xmm0
503		− add SQR_CUR, r14
504		−
505		− lea r0, [r0+r1]
506		− lea r2, [r2+r3]
507		− dec r12
508		− jnz near .hloops
509		−
510		− mov r0, SUM
511		− sar r0, 8
512		− imul r0, r0
513		− mov r1, SQR
514		− sar r1, 8
515		− sub r1, r0
516		− mov [r4], r1w ; to store uiMotionIndex
517		− mov r0, SUM_CUR
518		− sar r0, 8
519		− imul r0, r0
520		− mov r1, SQR_CUR
521		− sar r1, 8
522		− sub r1, r0
523		− mov [r4+2], r1w ; to store uiTextureIndex
524		−
525		− POP_XMM
526		− LOAD_5_PARA_POP
527		− pop r15
528		− pop r14
529		− pop r13
530		− pop r12
531		−
532		−
533		− %assign push_num 0
534		−
535		− ret
	467	+ mov r14,0
	468	+ movdqa xmm0, [r0] ; y_ref
	469	+ movdqa xmm1, [r2] ; y_src
	470	+ movdqa xmm2, xmm0 ; store first for future process
	471	+ movdqa xmm3, xmm1
	472	+ ; sum += diff;
	473	+ movdqa xmm4, xmm0
	474	+ psadbw xmm4, xmm1 ; 2 parts, [0,..,15], [64,..,79]
	475	+ ; to be continued for sum
	476	+ pshufd xmm5, xmm4, 0C6h ; 11000110 B
	477	+ paddw xmm4, xmm5
	478	+ movd r14d, xmm4
	479	+ add SUM, r14
	480	+
	481	+ ; sqr += diff * diff;
	482	+ pmaxub xmm0, xmm1
	483	+ pminub xmm1, xmm2
	484	+ psubb xmm0, xmm1 ; diff
	485	+ SUM_SQR_SSE2 xmm1, xmm0, xmm7 ; dst, pSrc, zero
	486	+ movd r14d, xmm1
	487	+ add SQR, r14
	488	+
	489	+ ; sum_cur += y_src[x];
	490	+ movdqa xmm0, xmm3 ; cur_orig
	491	+ movdqa xmm1, xmm0
	492	+ punpcklbw xmm0, xmm7
	493	+ punpckhbw xmm1, xmm7
	494	+ paddw xmm0, xmm1 ; 8x2
	495	+ SUM_WORD_8x2_SSE2 xmm0, xmm1
	496	+ movd r14d, xmm0
	497	+ and r14, 0ffffh
	498	+ add SUM_CUR, r14
	499	+
	500	+ ; sqr_cur += y_src[x] * y_src[x];
	501	+ SUM_SQR_SSE2 xmm0, xmm3, xmm7 ; dst, pSrc, zero
	502	+ movd r14d, xmm0
	503	+ add SQR_CUR, r14
	504	+
	505	+ lea r0, [r0+r1]
	506	+ lea r2, [r2+r3]
	507	+ dec r12
	508	+ jnz near .hloops
	509	+
	510	+ mov r0, SUM
	511	+ sar r0, 8
	512	+ imul r0, r0
	513	+ mov r1, SQR
	514	+ sar r1, 8
	515	+ sub r1, r0
	516	+ mov [r4], r1w ; to store uiMotionIndex
	517	+ mov r0, SUM_CUR
	518	+ sar r0, 8
	519	+ imul r0, r0
	520	+ mov r1, SQR_CUR
	521	+ sar r1, 8
	522	+ sub r1, r0
	523	+ mov [r4+2], r1w ; to store uiTextureIndex
	524	+
	525	+ POP_XMM
	526	+ LOAD_5_PARA_POP
	527	+ pop r15
	528	+ pop r14
	529	+ pop r13
	530	+ pop r12
	531	+
	532	+
	533	+ %assign push_num 0
	534	+
	535	+ ret
536	536
537	537
538	538	;*************************************************************************************************************
@@ -550,69 +550,69 @@ WELS_EXTERN VAACalcSad_sse2 · 11 unmodified lines
550	550	%define psadframe r5
551	551	%define psad8x8 r6
552	552
553		− push r12
554		− push r13
555		− %assign push_num 2
556		− LOAD_7_PARA
557		− PUSH_XMM 8
558		− SIGN_EXTENSION r2,r2d
559		− SIGN_EXTENSION r3,r3d
560		− SIGN_EXTENSION r4,r4d
561		−
562		− mov r12,r4
563		− shr r2, 4 ; iPicWidth/16
564		− shr r3, 4 ; iPicHeight/16
565		−
566		− shl r12, 4 ; iPicStride*16
567		− pxor xmm0, xmm0
568		− pxor xmm7, xmm7 ; iFrameSad
	553	+ push r12
	554	+ push r13
	555	+ %assign push_num 2
	556	+ LOAD_7_PARA
	557	+ PUSH_XMM 8
	558	+ SIGN_EXTENSION r2,r2d
	559	+ SIGN_EXTENSION r3,r3d
	560	+ SIGN_EXTENSION r4,r4d
	561	+
	562	+ mov r12,r4
	563	+ shr r2, 4 ; iPicWidth/16
	564	+ shr r3, 4 ; iPicHeight/16
	565	+
	566	+ shl r12, 4 ; iPicStride*16
	567	+ pxor xmm0, xmm0
	568	+ pxor xmm7, xmm7 ; iFrameSad
569	569	height_loop:
570		− mov r13, r2
571		− push r0
572		− push r1
	570	+ mov r13, r2
	571	+ push r0
	572	+ push r1
573	573	width_loop:
574		− pxor xmm6, xmm6
575		− WELS_SAD_16x2_SSE2 r0,r1,r4
576		− WELS_SAD_16x2_SSE2 r0,r1,r4
577		− WELS_SAD_16x2_SSE2 r0,r1,r4
578		− WELS_SAD_16x2_SSE2 r0,r1,r4
579		− paddd xmm7, xmm6
580		− movd [r6], xmm6
581		− psrldq xmm6, 8
582		− movd [r6+4], xmm6
583		−
584		− pxor xmm6, xmm6
585		− WELS_SAD_16x2_SSE2 r0,r1,r4
586		− WELS_SAD_16x2_SSE2 r0,r1,r4
587		− WELS_SAD_16x2_SSE2 r0,r1,r4
588		− WELS_SAD_16x2_SSE2 r0,r1,r4
589		− paddd xmm7, xmm6
590		− movd [r6+8], xmm6
591		− psrldq xmm6, 8
592		− movd [r6+12], xmm6
593		−
594		− add r6, 16
595		− sub r0, r12
596		− sub r1, r12
597		− add r0, 16
598		− add r1, 16
599		−
600		− dec r13
601		− jnz width_loop
602		−
603		− pop r1
604		− pop r0
605		− add r0, r12
606		− add r1, r12
607		−
608		− dec r3
609		− jnz height_loop
610		−
611		− ;mov r13, [psadframe]
612		− movdqa xmm5, xmm7
613		− psrldq xmm7, 8
614		− paddd xmm7, xmm5
615		− movd [psadframe], xmm7
	574	+ pxor xmm6, xmm6
	575	+ WELS_SAD_16x2_SSE2 r0,r1,r4
	576	+ WELS_SAD_16x2_SSE2 r0,r1,r4
	577	+ WELS_SAD_16x2_SSE2 r0,r1,r4
	578	+ WELS_SAD_16x2_SSE2 r0,r1,r4
	579	+ paddd xmm7, xmm6
	580	+ movd [r6], xmm6
	581	+ psrldq xmm6, 8
	582	+ movd [r6+4], xmm6
	583	+
	584	+ pxor xmm6, xmm6
	585	+ WELS_SAD_16x2_SSE2 r0,r1,r4
	586	+ WELS_SAD_16x2_SSE2 r0,r1,r4
	587	+ WELS_SAD_16x2_SSE2 r0,r1,r4
	588	+ WELS_SAD_16x2_SSE2 r0,r1,r4
	589	+ paddd xmm7, xmm6
	590	+ movd [r6+8], xmm6
	591	+ psrldq xmm6, 8
	592	+ movd [r6+12], xmm6
	593	+
	594	+ add r6, 16
	595	+ sub r0, r12
	596	+ sub r1, r12
	597	+ add r0, 16
	598	+ add r1, 16
	599	+
	600	+ dec r13
	601	+ jnz width_loop
	602	+
	603	+ pop r1
	604	+ pop r0
	605	+ add r0, r12
	606	+ add r1, r12
	607	+
	608	+ dec r3
	609	+ jnz height_loop
	610	+
	611	+ ;mov r13, [psadframe]
	612	+ movdqa xmm5, xmm7
	613	+ psrldq xmm7, 8
	614	+ paddd xmm7, xmm5
	615	+ movd [psadframe], xmm7
616	616
617	617	%undef cur_data
618	618	%undef ref_data
@@ -622,12 +622,12 @@ width_loop: · 3 unmodified lines
622	622	%undef psadframe
623	623	%undef psad8x8
624	624	%undef pushsize
625		− POP_XMM
626		− LOAD_7_PARA_POP
627		− pop r13
628		− pop r12
629		− %assign push_num 0
630		− ret
	625	+ POP_XMM
	626	+ LOAD_7_PARA_POP
	627	+ pop r13
	628	+ pop r12
	629	+ %assign push_num 0
	630	+ ret
631	631
632	632	%endif
633	633
@@ -653,103 +653,103 @@ WELS_EXTERN VAACalcSadVar_sse2 · 19 unmodified lines
653	653	%define tmp_esi esp + 0
654	654	%define tmp_edi esp + 4
655	655	%define pushsize 16
656		− push ebp
657		− push esi
658		− push edi
659		− push ebx
660		− sub esp, localsize
661		− mov esi, [cur_data]
662		− mov edi, [ref_data]
663		− mov ebx, [iPicStride]
664		− mov edx, [psad8x8]
665		− mov eax, ebx
666		−
667		− shr dword [iPicWidth], 4 ; iPicWidth/16
668		− shr dword [iPicHeight], 4 ; iPicHeight/16
669		− shl eax, 4 ; iPicStride*16
670		− pxor xmm0, xmm0
671		− pxor xmm7, xmm7 ; iFrameSad
	656	+ push ebp
	657	+ push esi
	658	+ push edi
	659	+ push ebx
	660	+ sub esp, localsize
	661	+ mov esi, [cur_data]
	662	+ mov edi, [ref_data]
	663	+ mov ebx, [iPicStride]
	664	+ mov edx, [psad8x8]
	665	+ mov eax, ebx
	666	+
	667	+ shr dword [iPicWidth], 4 ; iPicWidth/16
	668	+ shr dword [iPicHeight], 4 ; iPicHeight/16
	669	+ shl eax, 4 ; iPicStride*16
	670	+ pxor xmm0, xmm0
	671	+ pxor xmm7, xmm7 ; iFrameSad
672	672	var_height_loop:
673		− mov ecx, dword [iPicWidth]
674		− mov [tmp_esi], esi
675		− mov [tmp_edi], edi
	673	+ mov ecx, dword [iPicWidth]
	674	+ mov [tmp_esi], esi
	675	+ mov [tmp_edi], edi
676	676	var_width_loop:
677		− pxor xmm6, xmm6 ; hiQuad_loQuad pSad8x8
678		− pxor xmm5, xmm5 ; pSum16x16
679		− pxor xmm4, xmm4 ; sqsum_16x16
680		− WELS_SAD_SUM_SQSUM_16x1_SSE2 esi,edi,ebx
681		− WELS_SAD_SUM_SQSUM_16x1_SSE2 esi,edi,ebx
682		− WELS_SAD_SUM_SQSUM_16x1_SSE2 esi,edi,ebx
683		− WELS_SAD_SUM_SQSUM_16x1_SSE2 esi,edi,ebx
684		− WELS_SAD_SUM_SQSUM_16x1_SSE2 esi,edi,ebx
685		− WELS_SAD_SUM_SQSUM_16x1_SSE2 esi,edi,ebx
686		− WELS_SAD_SUM_SQSUM_16x1_SSE2 esi,edi,ebx
687		− WELS_SAD_SUM_SQSUM_16x1_SSE2 esi,edi,ebx
688		− paddd xmm7, xmm6
689		− movd [edx], xmm6
690		− psrldq xmm6, 8
691		− movd [edx+4], xmm6
692		−
693		− pxor xmm6, xmm6
694		− WELS_SAD_SUM_SQSUM_16x1_SSE2 esi,edi,ebx
695		− WELS_SAD_SUM_SQSUM_16x1_SSE2 esi,edi,ebx
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		− paddd xmm7, xmm6
703		− movd [edx+8], xmm6
704		− psrldq xmm6, 8
705		− movd [edx+12], xmm6
706		−
707		− mov ebp, [psum16x16]
708		− movdqa xmm1, xmm5
709		− psrldq xmm1, 8
710		− paddd xmm5, xmm1
711		− movd [ebp], xmm5
712		− add dword [psum16x16], 4
713		−
714		− movdqa xmm5, xmm4
715		− psrldq xmm5, 8
716		− paddd xmm4, xmm5
717		− movdqa xmm3, xmm4
718		− psrldq xmm3, 4
719		− paddd xmm4, xmm3
720		−
721		− mov ebp, [psqsum16x16]
722		− movd [ebp], xmm4
723		− add dword [psqsum16x16], 4
724		−
725		− add edx, 16
726		− sub esi, eax
727		− sub edi, eax
728		− add esi, 16
729		− add edi, 16
730		−
731		− dec ecx
732		− jnz var_width_loop
733		−
734		− mov esi, [tmp_esi]
735		− mov edi, [tmp_edi]
736		− add esi, eax
737		− add edi, eax
738		−
739		− dec dword [iPicHeight]
740		− jnz var_height_loop
741		−
742		− mov edx, [psadframe]
743		− movdqa xmm5, xmm7
744		− psrldq xmm7, 8
745		− paddd xmm7, xmm5
746		− movd [edx], xmm7
747		−
748		− add esp, localsize
749		− pop ebx
750		− pop edi
751		− pop esi
752		− pop ebp
	677	+ pxor xmm6, xmm6 ; hiQuad_loQuad pSad8x8
	678	+ pxor xmm5, xmm5 ; pSum16x16
	679	+ pxor xmm4, xmm4 ; sqsum_16x16
	680	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 esi,edi,ebx
	681	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 esi,edi,ebx
	682	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 esi,edi,ebx
	683	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 esi,edi,ebx
	684	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 esi,edi,ebx
	685	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 esi,edi,ebx
	686	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 esi,edi,ebx
	687	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 esi,edi,ebx
	688	+ paddd xmm7, xmm6
	689	+ movd [edx], xmm6
	690	+ psrldq xmm6, 8
	691	+ movd [edx+4], xmm6
	692	+
	693	+ pxor xmm6, xmm6
	694	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 esi,edi,ebx
	695	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 esi,edi,ebx
	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	+ paddd xmm7, xmm6
	703	+ movd [edx+8], xmm6
	704	+ psrldq xmm6, 8
	705	+ movd [edx+12], xmm6
	706	+
	707	+ mov ebp, [psum16x16]
	708	+ movdqa xmm1, xmm5
	709	+ psrldq xmm1, 8
	710	+ paddd xmm5, xmm1
	711	+ movd [ebp], xmm5
	712	+ add dword [psum16x16], 4
	713	+
	714	+ movdqa xmm5, xmm4
	715	+ psrldq xmm5, 8
	716	+ paddd xmm4, xmm5
	717	+ movdqa xmm3, xmm4
	718	+ psrldq xmm3, 4
	719	+ paddd xmm4, xmm3
	720	+
	721	+ mov ebp, [psqsum16x16]
	722	+ movd [ebp], xmm4
	723	+ add dword [psqsum16x16], 4
	724	+
	725	+ add edx, 16
	726	+ sub esi, eax
	727	+ sub edi, eax
	728	+ add esi, 16
	729	+ add edi, 16
	730	+
	731	+ dec ecx
	732	+ jnz var_width_loop
	733	+
	734	+ mov esi, [tmp_esi]
	735	+ mov edi, [tmp_edi]
	736	+ add esi, eax
	737	+ add edi, eax
	738	+
	739	+ dec dword [iPicHeight]
	740	+ jnz var_height_loop
	741	+
	742	+ mov edx, [psadframe]
	743	+ movdqa xmm5, xmm7
	744	+ psrldq xmm7, 8
	745	+ paddd xmm7, xmm5
	746	+ movd [edx], xmm7
	747	+
	748	+ add esp, localsize
	749	+ pop ebx
	750	+ pop edi
	751	+ pop esi
	752	+ pop ebp
753	753	%undef cur_data
754	754	%undef ref_data
755	755	%undef iPicWidth
@@ -763,7 +763,7 @@ var_width_loop: · 7 unmodified lines
763	763	%undef tmp_edi
764	764	%undef pushsize
765	765	%undef localsize
766		− ret
	766	+ ret
767	767
768	768	%else ;64-bit
769	769
@@ -784,112 +784,112 @@ WELS_EXTERN VAACalcSadVar_sse2 · 14 unmodified lines
784	784	%define psum16x16 arg8
785	785	%define psqsum16x16 arg9
786	786
787		− push r12
788		− push r13
789		− push r14
790		− push r15
791		− %assign push_num 4
792		− PUSH_XMM 8
	787	+ push r12
	788	+ push r13
	789	+ push r14
	790	+ push r15
	791	+ %assign push_num 4
	792	+ PUSH_XMM 8
793	793
794	794	%ifdef WIN64
795		− mov r4, arg5 ;iPicStride
796		− mov r5, arg6 ;psad8x8
	795	+ mov r4, arg5 ;iPicStride
	796	+ mov r5, arg6 ;psad8x8
797	797	%endif
798		− mov r14,arg7
799		− SIGN_EXTENSION r2,r2d
800		− SIGN_EXTENSION r3,r3d
801		− SIGN_EXTENSION r4,r4d
802		−
803		− mov r13,r4
804		− shr r2,4
805		− shr r3,4
806		−
807		− shl r13,4 ; iPicStride*16
808		− pxor xmm0, xmm0
809		− pxor xmm7, xmm7 ; iFrameSad
	798	+ mov r14,arg7
	799	+ SIGN_EXTENSION r2,r2d
	800	+ SIGN_EXTENSION r3,r3d
	801	+ SIGN_EXTENSION r4,r4d
	802	+
	803	+ mov r13,r4
	804	+ shr r2,4
	805	+ shr r3,4
	806	+
	807	+ shl r13,4 ; iPicStride*16
	808	+ pxor xmm0, xmm0
	809	+ pxor xmm7, xmm7 ; iFrameSad
810	810	var_height_loop:
811		− push r2
812		− %assign push_num push_num+1
813		− mov r11, r0
814		− mov r12, r1
	811	+ push r2
	812	+ %assign push_num push_num+1
	813	+ mov r11, r0
	814	+ mov r12, r1
815	815	var_width_loop:
816		− pxor xmm6, xmm6 ; hiQuad_loQuad pSad8x8
817		− pxor xmm5, xmm5 ; pSum16x16
818		− pxor xmm4, xmm4 ; sqsum_16x16
819		− WELS_SAD_SUM_SQSUM_16x1_SSE2 r0,r1,r4
820		− WELS_SAD_SUM_SQSUM_16x1_SSE2 r0,r1,r4
821		− WELS_SAD_SUM_SQSUM_16x1_SSE2 r0,r1,r4
822		− WELS_SAD_SUM_SQSUM_16x1_SSE2 r0,r1,r4
823		− WELS_SAD_SUM_SQSUM_16x1_SSE2 r0,r1,r4
824		− WELS_SAD_SUM_SQSUM_16x1_SSE2 r0,r1,r4
825		− WELS_SAD_SUM_SQSUM_16x1_SSE2 r0,r1,r4
826		− WELS_SAD_SUM_SQSUM_16x1_SSE2 r0,r1,r4
827		− paddd xmm7, xmm6
828		− movd [r14], xmm6
829		− psrldq xmm6, 8
830		− movd [r14+4], xmm6
831		−
832		− pxor xmm6, xmm6
833		− WELS_SAD_SUM_SQSUM_16x1_SSE2 r0,r1,r4
834		− WELS_SAD_SUM_SQSUM_16x1_SSE2 r0,r1,r4
835		− WELS_SAD_SUM_SQSUM_16x1_SSE2 r0,r1,r4
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		− paddd xmm7, xmm6
842		− movd [r14+8], xmm6
843		− psrldq xmm6, 8
844		− movd [r14+12], xmm6
845		−
846		− mov r15, psum16x16
847		− movdqa xmm1, xmm5
848		− psrldq xmm1, 8
849		− paddd xmm5, xmm1
850		− movd [r15], xmm5
851		− add dword psum16x16, 4
852		−
853		− movdqa xmm5, xmm4
854		− psrldq xmm5, 8
855		− paddd xmm4, xmm5
856		− movdqa xmm3, xmm4
857		− psrldq xmm3, 4
858		− paddd xmm4, xmm3
859		−
860		− mov r15, psqsum16x16
861		− movd [r15], xmm4
862		− add dword psqsum16x16, 4
863		−
864		− add r14,16
865		− sub r0, r13
866		− sub r1, r13
867		− add r0, 16
868		− add r1, 16
869		−
870		− dec r2
871		− jnz var_width_loop
872		−
873		− pop r2
874		− %assign push_num push_num-1
875		− mov r0, r11
876		− mov r1, r12
877		− add r0, r13
878		− add r1, r13
879		− dec r3
880		− jnz var_height_loop
881		−
882		− mov r15, psadframe
883		− movdqa xmm5, xmm7
884		− psrldq xmm7, 8
885		− paddd xmm7, xmm5
886		− movd [r15], xmm7
887		−
888		− POP_XMM
889		− pop r15
890		− pop r14
891		− pop r13
892		− pop r12
	816	+ pxor xmm6, xmm6 ; hiQuad_loQuad pSad8x8
	817	+ pxor xmm5, xmm5 ; pSum16x16
	818	+ pxor xmm4, xmm4 ; sqsum_16x16
	819	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 r0,r1,r4
	820	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 r0,r1,r4
	821	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 r0,r1,r4
	822	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 r0,r1,r4
	823	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 r0,r1,r4
	824	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 r0,r1,r4
	825	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 r0,r1,r4
	826	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 r0,r1,r4
	827	+ paddd xmm7, xmm6
	828	+ movd [r14], xmm6
	829	+ psrldq xmm6, 8
	830	+ movd [r14+4], xmm6
	831	+
	832	+ pxor xmm6, xmm6
	833	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 r0,r1,r4
	834	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 r0,r1,r4
	835	+ WELS_SAD_SUM_SQSUM_16x1_SSE2 r0,r1,r4
	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	+ paddd xmm7, xmm6
	842	+ movd [r14+8], xmm6
	843	+ psrldq xmm6, 8
	844	+ movd [r14+12], xmm6
	845	+
	846	+ mov r15, psum16x16
	847	+ movdqa xmm1, xmm5
	848	+ psrldq xmm1, 8
	849	+ paddd xmm5, xmm1
	850	+ movd [r15], xmm5
	851	+ add dword psum16x16, 4
	852	+
	853	+ movdqa xmm5, xmm4
	854	+ psrldq xmm5, 8
	855	+ paddd xmm4, xmm5
	856	+ movdqa xmm3, xmm4
	857	+ psrldq xmm3, 4
	858	+ paddd xmm4, xmm3
	859	+
	860	+ mov r15, psqsum16x16
	861	+ movd [r15], xmm4
	862	+ add dword psqsum16x16, 4
	863	+
	864	+ add r14,16
	865	+ sub r0, r13
	866	+ sub r1, r13
	867	+ add r0, 16
	868	+ add r1, 16
	869	+
	870	+ dec r2
	871	+ jnz var_width_loop
	872	+
	873	+ pop r2
	874	+ %assign push_num push_num-1
	875	+ mov r0, r11
	876	+ mov r1, r12
	877	+ add r0, r13
	878	+ add r1, r13
	879	+ dec r3
	880	+ jnz var_height_loop
	881	+
	882	+ mov r15, psadframe
	883	+ movdqa xmm5, xmm7
	884	+ psrldq xmm7, 8
	885	+ paddd xmm7, xmm5
	886	+ movd [r15], xmm7
	887	+
	888	+ POP_XMM
	889	+ pop r15
	890	+ pop r14
	891	+ pop r13
	892	+ pop r12
893	893	%assign push_num 0
894	894	%undef cur_data
895	895	%undef ref_data
@@ -904,7 +904,7 @@ var_width_loop: · 8 unmodified lines
904	904	%undef tmp_edi
905	905	%undef pushsize
906	906	%undef localsize
907		− ret
	907	+ ret
908	908
909	909	%endif
910	910
@@ -932,118 +932,118 @@ WELS_EXTERN VAACalcSadSsd_sse2 · 21 unmodified lines
932	932	%define tmp_edi esp + 4
933	933	%define tmp_sadframe esp + 8
934	934	%define pushsize 16
935		− push ebp
936		− push esi
937		− push edi
938		− push ebx
939		− sub esp, localsize
940		−
941		− mov ecx, [iPicWidth]
942		− mov ecx, [iPicHeight]
943		− mov esi, [cur_data]
944		− mov edi, [ref_data]
945		− mov ebx, [iPicStride]
946		− mov edx, [psad8x8]
947		− mov eax, ebx
948		−
949		− shr dword [iPicWidth], 4 ; iPicWidth/16
950		− shr dword [iPicHeight], 4 ; iPicHeight/16
951		− shl eax, 4 ; iPicStride*16
952		− mov ecx, [iPicWidth]
953		− mov ecx, [iPicHeight]
954		− pxor xmm0, xmm0
955		− movd [tmp_sadframe], xmm0
	935	+ push ebp
	936	+ push esi
	937	+ push edi
	938	+ push ebx
	939	+ sub esp, localsize
	940	+
	941	+ mov ecx, [iPicWidth]
	942	+ mov ecx, [iPicHeight]
	943	+ mov esi, [cur_data]
	944	+ mov edi, [ref_data]
	945	+ mov ebx, [iPicStride]
	946	+ mov edx, [psad8x8]
	947	+ mov eax, ebx
	948	+
	949	+ shr dword [iPicWidth], 4 ; iPicWidth/16
	950	+ shr dword [iPicHeight], 4 ; iPicHeight/16
	951	+ shl eax, 4 ; iPicStride*16
	952	+ mov ecx, [iPicWidth]
	953	+ mov ecx, [iPicHeight]
	954	+ pxor xmm0, xmm0
	955	+ movd [tmp_sadframe], xmm0
956	956	sqdiff_height_loop:
957		− mov ecx, dword [iPicWidth]
958		− mov [tmp_esi], esi
959		− mov [tmp_edi], edi
	957	+ mov ecx, dword [iPicWidth]
	958	+ mov [tmp_esi], esi
	959	+ mov [tmp_edi], edi
960	960	sqdiff_width_loop:
961		− pxor xmm7, xmm7 ; hiQuad_loQuad pSad8x8
962		− pxor xmm6, xmm6 ; pSum16x16
963		− pxor xmm5, xmm5 ; sqsum_16x16 four dword
964		− pxor xmm4, xmm4 ; sqdiff_16x16 four Dword
965		− WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 esi,edi,ebx
966		− WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 esi,edi,ebx
967		− WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 esi,edi,ebx
968		− WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 esi,edi,ebx
969		− WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 esi,edi,ebx
970		− WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 esi,edi,ebx
971		− WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 esi,edi,ebx
972		− WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 esi,edi,ebx
973		− movdqa xmm1, xmm7
974		− movd [edx], xmm7
975		− psrldq xmm7, 8
976		− paddd xmm1, xmm7
977		− movd [edx+4], xmm7
978		− movd ebp, xmm1
979		− add [tmp_sadframe], ebp
980		−
981		− pxor xmm7, xmm7
982		− WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 esi,edi,ebx
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		− movdqa xmm1, xmm7
991		− movd [edx+8], xmm7
992		− psrldq xmm7, 8
993		− paddd xmm1, xmm7
994		− movd [edx+12], xmm7
995		− movd ebp, xmm1
996		− add [tmp_sadframe], ebp
997		−
998		− mov ebp, [psum16x16]
999		− movdqa xmm1, xmm6
1000		− psrldq xmm1, 8
1001		− paddd xmm6, xmm1
1002		− movd [ebp], xmm6
1003		− add dword [psum16x16], 4
1004		−
1005		− mov ebp, [psqsum16x16]
1006		− pshufd xmm6, xmm5, 14 ;00001110
1007		− paddd xmm6, xmm5
1008		− pshufd xmm5, xmm6, 1 ;00000001
1009		− paddd xmm5, xmm6
1010		− movd [ebp], xmm5
1011		− add dword [psqsum16x16], 4
1012		−
1013		− mov ebp, [psqdiff16x16]
1014		− pshufd xmm5, xmm4, 14 ; 00001110
1015		− paddd xmm5, xmm4
1016		− pshufd xmm4, xmm5, 1 ; 00000001
1017		− paddd xmm4, xmm5
1018		− movd [ebp], xmm4
1019		− add dword [psqdiff16x16], 4
1020		−
1021		− add edx, 16
1022		− sub esi, eax
1023		− sub edi, eax
1024		− add esi, 16
1025		− add edi, 16
1026		−
1027		− dec ecx
1028		− jnz sqdiff_width_loop
1029		−
1030		− mov esi, [tmp_esi]
1031		− mov edi, [tmp_edi]
1032		− add esi, eax
1033		− add edi, eax
1034		−
1035		− dec dword [iPicHeight]
1036		− jnz sqdiff_height_loop
1037		−
1038		− mov ebx, [tmp_sadframe]
1039		− mov eax, [psadframe]
1040		− mov [eax], ebx
1041		−
1042		− add esp, localsize
1043		− pop ebx
1044		− pop edi
1045		− pop esi
1046		− pop ebp
	961	+ pxor xmm7, xmm7 ; hiQuad_loQuad pSad8x8
	962	+ pxor xmm6, xmm6 ; pSum16x16
	963	+ pxor xmm5, xmm5 ; sqsum_16x16 four dword
	964	+ pxor xmm4, xmm4 ; sqdiff_16x16 four Dword
	965	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 esi,edi,ebx
	966	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 esi,edi,ebx
	967	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 esi,edi,ebx
	968	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 esi,edi,ebx
	969	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 esi,edi,ebx
	970	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 esi,edi,ebx
	971	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 esi,edi,ebx
	972	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 esi,edi,ebx
	973	+ movdqa xmm1, xmm7
	974	+ movd [edx], xmm7
	975	+ psrldq xmm7, 8
	976	+ paddd xmm1, xmm7
	977	+ movd [edx+4], xmm7
	978	+ movd ebp, xmm1
	979	+ add [tmp_sadframe], ebp
	980	+
	981	+ pxor xmm7, xmm7
	982	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 esi,edi,ebx
	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	+ movdqa xmm1, xmm7
	991	+ movd [edx+8], xmm7
	992	+ psrldq xmm7, 8
	993	+ paddd xmm1, xmm7
	994	+ movd [edx+12], xmm7
	995	+ movd ebp, xmm1
	996	+ add [tmp_sadframe], ebp
	997	+
	998	+ mov ebp, [psum16x16]
	999	+ movdqa xmm1, xmm6
	1000	+ psrldq xmm1, 8
	1001	+ paddd xmm6, xmm1
	1002	+ movd [ebp], xmm6
	1003	+ add dword [psum16x16], 4
	1004	+
	1005	+ mov ebp, [psqsum16x16]
	1006	+ pshufd xmm6, xmm5, 14 ;00001110
	1007	+ paddd xmm6, xmm5
	1008	+ pshufd xmm5, xmm6, 1 ;00000001
	1009	+ paddd xmm5, xmm6
	1010	+ movd [ebp], xmm5
	1011	+ add dword [psqsum16x16], 4
	1012	+
	1013	+ mov ebp, [psqdiff16x16]
	1014	+ pshufd xmm5, xmm4, 14 ; 00001110
	1015	+ paddd xmm5, xmm4
	1016	+ pshufd xmm4, xmm5, 1 ; 00000001
	1017	+ paddd xmm4, xmm5
	1018	+ movd [ebp], xmm4
	1019	+ add dword [psqdiff16x16], 4
	1020	+
	1021	+ add edx, 16
	1022	+ sub esi, eax
	1023	+ sub edi, eax
	1024	+ add esi, 16
	1025	+ add edi, 16
	1026	+
	1027	+ dec ecx
	1028	+ jnz sqdiff_width_loop
	1029	+
	1030	+ mov esi, [tmp_esi]
	1031	+ mov edi, [tmp_edi]
	1032	+ add esi, eax
	1033	+ add edi, eax
	1034	+
	1035	+ dec dword [iPicHeight]
	1036	+ jnz sqdiff_height_loop
	1037	+
	1038	+ mov ebx, [tmp_sadframe]
	1039	+ mov eax, [psadframe]
	1040	+ mov [eax], ebx
	1041	+
	1042	+ add esp, localsize
	1043	+ pop ebx
	1044	+ pop edi
	1045	+ pop esi
	1046	+ pop ebp
1047	1047	%undef cur_data
1048	1048	%undef ref_data
1049	1049	%undef iPicWidth
@@ -1059,7 +1059,7 @@ sqdiff_width_loop: · 9 unmodified lines
1059	1059	%undef tmp_sadframe
1060	1060	%undef pushsize
1061	1061	%undef localsize
1062		− ret
	1062	+ ret
1063	1063
1064	1064	%else
1065	1065
@@ -1083,128 +1083,128 @@ WELS_EXTERN VAACalcSadSsd_sse2 · 17 unmodified lines
1083	1083	%define psqsum16x16 arg9;
1084	1084	%define psqdiff16x16 arg10
1085	1085
1086		− push r12
1087		− push r13
1088		− push r14
1089		− push r15
1090		− %assign push_num 4
1091		− PUSH_XMM 10
	1086	+ push r12
	1087	+ push r13
	1088	+ push r14
	1089	+ push r15
	1090	+ %assign push_num 4
	1091	+ PUSH_XMM 10
1092	1092
1093	1093	%ifdef WIN64
1094		− mov r4,arg5
	1094	+ mov r4,arg5
1095	1095	%endif
1096		− mov r14,arg7
1097		− SIGN_EXTENSION r2,r2d
1098		− SIGN_EXTENSION r3,r3d
1099		− SIGN_EXTENSION r4,r4d
1100		−
1101		− mov r13,r4
1102		− shr r2,4 ; iPicWidth/16
1103		− shr r3,4 ; iPicHeight/16
1104		− shl r13,4 ; iPicStride*16
1105		− pxor xmm0, xmm0
1106		− pxor xmm8, xmm8 ;framesad
1107		− pxor xmm9, xmm9
	1096	+ mov r14,arg7
	1097	+ SIGN_EXTENSION r2,r2d
	1098	+ SIGN_EXTENSION r3,r3d
	1099	+ SIGN_EXTENSION r4,r4d
	1100	+
	1101	+ mov r13,r4
	1102	+ shr r2,4 ; iPicWidth/16
	1103	+ shr r3,4 ; iPicHeight/16
	1104	+ shl r13,4 ; iPicStride*16
	1105	+ pxor xmm0, xmm0
	1106	+ pxor xmm8, xmm8 ;framesad
	1107	+ pxor xmm9, xmm9
1108	1108	sqdiff_height_loop:
1109		− ;mov ecx, dword [iPicWidth]
1110		− ;mov r14,r2
1111		− push r2
1112		− %assign push_num push_num +1
1113		− mov r10, r0
1114		− mov r11, r1
	1109	+ ;mov ecx, dword [iPicWidth]
	1110	+ ;mov r14,r2
	1111	+ push r2
	1112	+ %assign push_num push_num +1
	1113	+ mov r10, r0
	1114	+ mov r11, r1
1115	1115	sqdiff_width_loop:
1116		− pxor xmm7, xmm7 ; hiQuad_loQuad pSad8x8
1117		− pxor xmm6, xmm6 ; pSum16x16
1118		− pxor xmm5, xmm5 ; sqsum_16x16 four dword
1119		− pxor xmm4, xmm4 ; sqdiff_16x16 four Dword
1120		− WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 r0,r1,r4
1121		− WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 r0,r1,r4
1122		− WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 r0,r1,r4
1123		− WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 r0,r1,r4
1124		− WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 r0,r1,r4
1125		− WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 r0,r1,r4
1126		− WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 r0,r1,r4
1127		− WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 r0,r1,r4
1128		− movdqa xmm1, xmm7
1129		− movd [r14], xmm7
1130		− psrldq xmm7, 8
1131		− paddd xmm1, xmm7
1132		− movd [r14+4], xmm7
1133		− movd r15d, xmm1
1134		− movd xmm9, r15d
1135		− paddd xmm8,xmm9
1136		−
1137		−
1138		− pxor xmm7, xmm7
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+8], xmm7
1149		− psrldq xmm7, 8
1150		− paddd xmm1, xmm7
1151		− movd [r14+12], xmm7
1152		− movd r15d, xmm1
1153		− movd xmm9, r15d
1154		− paddd xmm8,xmm9
1155		−
1156		− mov r15, psum16x16
1157		− movdqa xmm1, xmm6
1158		− psrldq xmm1, 8
1159		− paddd xmm6, xmm1
1160		− movd [r15], xmm6
1161		− add dword psum16x16, 4
1162		−
1163		− mov r15, psqsum16x16
1164		− pshufd xmm6, xmm5, 14 ;00001110
1165		− paddd xmm6, xmm5
1166		− pshufd xmm5, xmm6, 1 ;00000001
1167		− paddd xmm5, xmm6
1168		− movd [r15], xmm5
1169		− add dword psqsum16x16, 4
1170		−
1171		− mov r15, psqdiff16x16
1172		− pshufd xmm5, xmm4, 14 ; 00001110
1173		− paddd xmm5, xmm4
1174		− pshufd xmm4, xmm5, 1 ; 00000001
1175		− paddd xmm4, xmm5
1176		− movd [r15], xmm4
1177		− add dword psqdiff16x16, 4
1178		−
1179		− add r14,16
1180		− sub r0, r13
1181		− sub r1, r13
1182		− add r0, 16
1183		− add r1, 16
1184		−
1185		− dec r2
1186		− jnz sqdiff_width_loop
1187		−
1188		− pop r2
1189		− %assign push_num push_num -1
1190		−
1191		− mov r0, r10
1192		− mov r1, r11
1193		− add r0, r13
1194		− add r1, r13
1195		−
1196		− dec r3
1197		− jnz sqdiff_height_loop
1198		−
1199		− mov r13, psadframe
1200		− movd [r13], xmm8
1201		−
1202		− POP_XMM
1203		− pop r15
1204		− pop r14
1205		− pop r13
1206		− pop r12
1207		− %assign push_num 0
	1116	+ pxor xmm7, xmm7 ; hiQuad_loQuad pSad8x8
	1117	+ pxor xmm6, xmm6 ; pSum16x16
	1118	+ pxor xmm5, xmm5 ; sqsum_16x16 four dword
	1119	+ pxor xmm4, xmm4 ; sqdiff_16x16 four Dword
	1120	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 r0,r1,r4
	1121	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 r0,r1,r4
	1122	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 r0,r1,r4
	1123	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 r0,r1,r4
	1124	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 r0,r1,r4
	1125	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 r0,r1,r4
	1126	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 r0,r1,r4
	1127	+ WELS_SAD_SUM_SQSUM_SQDIFF_16x1_SSE2 r0,r1,r4
	1128	+ movdqa xmm1, xmm7
	1129	+ movd [r14], xmm7
	1130	+ psrldq xmm7, 8
	1131	+ paddd xmm1, xmm7
	1132	+ movd [r14+4], xmm7
	1133	+ movd r15d, xmm1
	1134	+ movd xmm9, r15d
	1135	+ paddd xmm8,xmm9
	1136	+
	1137	+
	1138	+ pxor xmm7, xmm7
	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+8], xmm7
	1149	+ psrldq xmm7, 8
	1150	+ paddd xmm1, xmm7
	1151	+ movd [r14+12], xmm7
	1152	+ movd r15d, xmm1
	1153	+ movd xmm9, r15d
	1154	+ paddd xmm8,xmm9
	1155	+
	1156	+ mov r15, psum16x16
	1157	+ movdqa xmm1, xmm6
	1158	+ psrldq xmm1, 8
	1159	+ paddd xmm6, xmm1
	1160	+ movd [r15], xmm6
	1161	+ add dword psum16x16, 4
	1162	+
	1163	+ mov r15, psqsum16x16
	1164	+ pshufd xmm6, xmm5, 14 ;00001110
	1165	+ paddd xmm6, xmm5
	1166	+ pshufd xmm5, xmm6, 1 ;00000001
	1167	+ paddd xmm5, xmm6
	1168	+ movd [r15], xmm5
	1169	+ add dword psqsum16x16, 4
	1170	+
	1171	+ mov r15, psqdiff16x16
	1172	+ pshufd xmm5, xmm4, 14 ; 00001110
	1173	+ paddd xmm5, xmm4
	1174	+ pshufd xmm4, xmm5, 1 ; 00000001
	1175	+ paddd xmm4, xmm5
	1176	+ movd [r15], xmm4
	1177	+ add dword psqdiff16x16, 4
	1178	+
	1179	+ add r14,16
	1180	+ sub r0, r13
	1181	+ sub r1, r13
	1182	+ add r0, 16
	1183	+ add r1, 16
	1184	+
	1185	+ dec r2
	1186	+ jnz sqdiff_width_loop
	1187	+
	1188	+ pop r2
	1189	+ %assign push_num push_num -1
	1190	+
	1191	+ mov r0, r10
	1192	+ mov r1, r11
	1193	+ add r0, r13
	1194	+ add r1, r13
	1195	+
	1196	+ dec r3
	1197	+ jnz sqdiff_height_loop
	1198	+
	1199	+ mov r13, psadframe
	1200	+ movd [r13], xmm8
	1201	+
	1202	+ POP_XMM
	1203	+ pop r15
	1204	+ pop r14
	1205	+ pop r13
	1206	+ pop r12
	1207	+ %assign push_num 0
1208	1208
1209	1209	%undef cur_data
1210	1210	%undef ref_data
@@ -1221,7 +1221,7 @@ sqdiff_width_loop: · 10 unmodified lines
1221	1221	%undef tmp_sadframe
1222	1222	%undef pushsize
1223	1223	%undef localsize
1224		− ret
	1224	+ ret
1225	1225
1226	1226
1227	1227
@@ -1249,145 +1249,145 @@ WELS_EXTERN VAACalcSadBgd_sse2 · 21 unmodified lines
1249	1249	%define tmp_edi esp + 4
1250	1250	%define tmp_ecx esp + 8
1251	1251	%define pushsize 16
1252		− push ebp
1253		− push esi
1254		− push edi
1255		− push ebx
1256		− sub esp, localsize
1257		− mov esi, [cur_data]
1258		− mov edi, [ref_data]
1259		− mov ebx, [iPicStride]
1260		− mov eax, ebx
1261		−
1262		− shr dword [iPicWidth], 4 ; iPicWidth/16
1263		− shr dword [iPicHeight], 4 ; iPicHeight/16
1264		− shl eax, 4 ; iPicStride*16
1265		− xor ebp, ebp
1266		− pxor xmm0, xmm0
	1252	+ push ebp
	1253	+ push esi
	1254	+ push edi
	1255	+ push ebx
	1256	+ sub esp, localsize
	1257	+ mov esi, [cur_data]
	1258	+ mov edi, [ref_data]
	1259	+ mov ebx, [iPicStride]
	1260	+ mov eax, ebx
	1261	+
	1262	+ shr dword [iPicWidth], 4 ; iPicWidth/16
	1263	+ shr dword [iPicHeight], 4 ; iPicHeight/16
	1264	+ shl eax, 4 ; iPicStride*16
	1265	+ xor ebp, ebp
	1266	+ pxor xmm0, xmm0
1267	1267	bgd_height_loop:
1268		− mov ecx, dword [iPicWidth]
1269		− mov [tmp_esi], esi
1270		− mov [tmp_edi], edi
	1268	+ mov ecx, dword [iPicWidth]
	1269	+ mov [tmp_esi], esi
	1270	+ mov [tmp_edi], edi
1271	1271	bgd_width_loop:
1272		− pxor xmm7, xmm7 ; pSad8x8
1273		− pxor xmm6, xmm6 ; sum_cur_8x8
1274		− pxor xmm5, xmm5 ; sum_ref_8x8
1275		− pxor xmm4, xmm4 ; pMad8x8
1276		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
1277		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
1278		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
1279		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
1280		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
1281		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
1282		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
1283		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
1284		−
1285		−
1286		− mov edx, [p_mad8x8]
1287		− WELS_MAX_REG_SSE2 xmm4
1288		−
1289		− ;movdqa xmm1, xmm4
1290		− ;punpcklbw xmm1, xmm0
1291		− ;punpcklwd xmm1, xmm0
1292		− ;movd [edx], xmm1
1293		− ;punpckhbw xmm4, xmm0
1294		− ;punpcklwd xmm4, xmm0
1295		− ;movd [edx+4], xmm4
1296		− ;add edx, 8
1297		− ;mov [p_mad8x8], edx
1298		− mov [tmp_ecx], ecx
1299		− movhlps xmm1, xmm4
1300		− movd ecx, xmm4
1301		− mov [edx], cl
1302		− movd ecx, xmm1
1303		− mov [edx+1],cl
1304		− add edx, 2
1305		− mov [p_mad8x8], edx
1306		−
1307		−
1308		− pslldq xmm7, 4
1309		− pslldq xmm6, 4
1310		− pslldq xmm5, 4
1311		−
1312		−
1313		− pxor xmm4, xmm4 ; pMad8x8
1314		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
1315		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
1316		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
1317		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
1318		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
1319		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
1320		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
1321		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
1322		−
1323		− mov edx, [p_mad8x8]
1324		− WELS_MAX_REG_SSE2 xmm4
1325		−
1326		− ;movdqa xmm1, xmm4
1327		− ;punpcklbw xmm1, xmm0
1328		− ;punpcklwd xmm1, xmm0
1329		− ;movd [edx], xmm1
1330		− ;punpckhbw xmm4, xmm0
1331		− ;punpcklwd xmm4, xmm0
1332		− ;movd [edx+4], xmm4
1333		− ;add edx, 8
1334		− ;mov [p_mad8x8], edx
1335		− movhlps xmm1, xmm4
1336		− movd ecx, xmm4
1337		− mov [edx], cl
1338		− movd ecx, xmm1
1339		− mov [edx+1],cl
1340		− add edx, 2
1341		− mov [p_mad8x8], edx
1342		−
1343		− ; data in xmm7, xmm6, xmm5: D1 D3 D0 D2
1344		−
1345		− mov edx, [psad8x8]
1346		− pshufd xmm1, xmm7, 10001101b ; D3 D2 D1 D0
1347		− movdqa [edx], xmm1
1348		− add edx, 16
1349		− mov [psad8x8], edx ; sad8x8
1350		−
1351		− paddd xmm1, xmm7 ; D1+3 D3+2 D0+1 D2+0
1352		− pshufd xmm2, xmm1, 00000011b
1353		− paddd xmm1, xmm2
1354		− movd edx, xmm1
1355		− add ebp, edx ; sad frame
1356		−
1357		− mov edx, [p_sd8x8]
1358		− psubd xmm6, xmm5
1359		− pshufd xmm1, xmm6, 10001101b
1360		− movdqa [edx], xmm1
1361		− add edx, 16
1362		− mov [p_sd8x8], edx
1363		−
1364		−
1365		− add edx, 16
1366		− sub esi, eax
1367		− sub edi, eax
1368		− add esi, 16
1369		− add edi, 16
1370		−
1371		− mov ecx, [tmp_ecx]
1372		− dec ecx
1373		− jnz bgd_width_loop
1374		−
1375		− mov esi, [tmp_esi]
1376		− mov edi, [tmp_edi]
1377		− add esi, eax
1378		− add edi, eax
1379		−
1380		− dec dword [iPicHeight]
1381		− jnz bgd_height_loop
1382		−
1383		− mov edx, [psadframe]
1384		− mov [edx], ebp
1385		−
1386		− add esp, localsize
1387		− pop ebx
1388		− pop edi
1389		− pop esi
1390		− pop ebp
	1272	+ pxor xmm7, xmm7 ; pSad8x8
	1273	+ pxor xmm6, xmm6 ; sum_cur_8x8
	1274	+ pxor xmm5, xmm5 ; sum_ref_8x8
	1275	+ pxor xmm4, xmm4 ; pMad8x8
	1276	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
	1277	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
	1278	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
	1279	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
	1280	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
	1281	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
	1282	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
	1283	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
	1284	+
	1285	+
	1286	+ mov edx, [p_mad8x8]
	1287	+ WELS_MAX_REG_SSE2 xmm4
	1288	+
	1289	+ ;movdqa xmm1, xmm4
	1290	+ ;punpcklbw xmm1, xmm0
	1291	+ ;punpcklwd xmm1, xmm0
	1292	+ ;movd [edx], xmm1
	1293	+ ;punpckhbw xmm4, xmm0
	1294	+ ;punpcklwd xmm4, xmm0
	1295	+ ;movd [edx+4], xmm4
	1296	+ ;add edx, 8
	1297	+ ;mov [p_mad8x8], edx
	1298	+ mov [tmp_ecx], ecx
	1299	+ movhlps xmm1, xmm4
	1300	+ movd ecx, xmm4
	1301	+ mov [edx], cl
	1302	+ movd ecx, xmm1
	1303	+ mov [edx+1],cl
	1304	+ add edx, 2
	1305	+ mov [p_mad8x8], edx
	1306	+
	1307	+
	1308	+ pslldq xmm7, 4
	1309	+ pslldq xmm6, 4
	1310	+ pslldq xmm5, 4
	1311	+
	1312	+
	1313	+ pxor xmm4, xmm4 ; pMad8x8
	1314	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
	1315	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
	1316	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
	1317	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
	1318	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
	1319	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
	1320	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
	1321	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,esi ,edi, ebx
	1322	+
	1323	+ mov edx, [p_mad8x8]
	1324	+ WELS_MAX_REG_SSE2 xmm4
	1325	+
	1326	+ ;movdqa xmm1, xmm4
	1327	+ ;punpcklbw xmm1, xmm0
	1328	+ ;punpcklwd xmm1, xmm0
	1329	+ ;movd [edx], xmm1
	1330	+ ;punpckhbw xmm4, xmm0
	1331	+ ;punpcklwd xmm4, xmm0
	1332	+ ;movd [edx+4], xmm4
	1333	+ ;add edx, 8
	1334	+ ;mov [p_mad8x8], edx
	1335	+ movhlps xmm1, xmm4
	1336	+ movd ecx, xmm4
	1337	+ mov [edx], cl
	1338	+ movd ecx, xmm1
	1339	+ mov [edx+1],cl
	1340	+ add edx, 2
	1341	+ mov [p_mad8x8], edx
	1342	+
	1343	+ ; data in xmm7, xmm6, xmm5: D1 D3 D0 D2
	1344	+
	1345	+ mov edx, [psad8x8]
	1346	+ pshufd xmm1, xmm7, 10001101b ; D3 D2 D1 D0
	1347	+ movdqa [edx], xmm1
	1348	+ add edx, 16
	1349	+ mov [psad8x8], edx ; sad8x8
	1350	+
	1351	+ paddd xmm1, xmm7 ; D1+3 D3+2 D0+1 D2+0
	1352	+ pshufd xmm2, xmm1, 00000011b
	1353	+ paddd xmm1, xmm2
	1354	+ movd edx, xmm1
	1355	+ add ebp, edx ; sad frame
	1356	+
	1357	+ mov edx, [p_sd8x8]
	1358	+ psubd xmm6, xmm5
	1359	+ pshufd xmm1, xmm6, 10001101b
	1360	+ movdqa [edx], xmm1
	1361	+ add edx, 16
	1362	+ mov [p_sd8x8], edx
	1363	+
	1364	+
	1365	+ add edx, 16
	1366	+ sub esi, eax
	1367	+ sub edi, eax
	1368	+ add esi, 16
	1369	+ add edi, 16
	1370	+
	1371	+ mov ecx, [tmp_ecx]
	1372	+ dec ecx
	1373	+ jnz bgd_width_loop
	1374	+
	1375	+ mov esi, [tmp_esi]
	1376	+ mov edi, [tmp_edi]
	1377	+ add esi, eax
	1378	+ add edi, eax
	1379	+
	1380	+ dec dword [iPicHeight]
	1381	+ jnz bgd_height_loop
	1382	+
	1383	+ mov edx, [psadframe]
	1384	+ mov [edx], ebp
	1385	+
	1386	+ add esp, localsize
	1387	+ pop ebx
	1388	+ pop edi
	1389	+ pop esi
	1390	+ pop ebp
1391	1391	%undef cur_data
1392	1392	%undef ref_data
1393	1393	%undef iPicWidth
@@ -1401,7 +1401,7 @@ bgd_width_loop: · 7 unmodified lines
1401	1401	%undef tmp_edi
1402	1402	%undef pushsize
1403	1403	%undef localsize
1404		− ret
	1404	+ ret
1405	1405
1406	1406
1407	1407
@@ -1431,190 +1431,190 @@ WELS_EXTERN VAACalcSadSsdBgd_sse2 · 23 unmodified lines
1431	1431	%define tmp_sadframe esp + 8
1432	1432	%define tmp_ecx esp + 12
1433	1433	%define pushsize 16
1434		− push ebp
1435		− push esi
1436		− push edi
1437		− push ebx
1438		− sub esp, localsize
1439		− mov esi, [cur_data]
1440		− mov edi, [ref_data]
1441		− mov ebx, [iPicStride]
1442		− mov eax, ebx
1443		−
1444		− shr dword [iPicWidth], 4 ; iPicWidth/16
1445		− shr dword [iPicHeight], 4 ; iPicHeight/16
1446		− shl eax, 4 ; iPicStride*16
1447		− pxor xmm0, xmm0
1448		− movd [tmp_sadframe], xmm0
	1434	+ push ebp
	1435	+ push esi
	1436	+ push edi
	1437	+ push ebx
	1438	+ sub esp, localsize
	1439	+ mov esi, [cur_data]
	1440	+ mov edi, [ref_data]
	1441	+ mov ebx, [iPicStride]
	1442	+ mov eax, ebx
	1443	+
	1444	+ shr dword [iPicWidth], 4 ; iPicWidth/16
	1445	+ shr dword [iPicHeight], 4 ; iPicHeight/16
	1446	+ shl eax, 4 ; iPicStride*16
	1447	+ pxor xmm0, xmm0
	1448	+ movd [tmp_sadframe], xmm0
1449	1449	sqdiff_bgd_height_loop:
1450		− mov ecx, dword [iPicWidth]
1451		− mov [tmp_esi], esi
1452		− mov [tmp_edi], edi
	1450	+ mov ecx, dword [iPicWidth]
	1451	+ mov [tmp_esi], esi
	1452	+ mov [tmp_edi], edi
1453	1453	sqdiff_bgd_width_loop:
1454		− pxor xmm7, xmm7 ; pSad8x8 interleaves sqsum16x16: sqsum1 sad1 sqsum0 sad0
1455		− pxor xmm6, xmm6 ; sum_8x8 interleaves cur and pRef in Dword, Sref1 Scur1 Sref0 Scur0
1456		− pxor xmm5, xmm5 ; pMad8x8
1457		− pxor xmm4, xmm4 ; sqdiff_16x16 four Dword
1458		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
1459		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
1460		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
1461		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
1462		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
1463		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
1464		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
1465		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
1466		−
1467		− mov edx, [psad8x8]
1468		− movdqa xmm2, xmm7
1469		− pshufd xmm1, xmm2, 00001110b
1470		− movd [edx], xmm2
1471		− movd [edx+4], xmm1
1472		− add edx, 8
1473		− mov [psad8x8], edx ; sad8x8
1474		−
1475		− paddd xmm1, xmm2
1476		− movd edx, xmm1
1477		− add [tmp_sadframe], edx ; iFrameSad
1478		−
1479		− mov edx, [psum16x16]
1480		− movdqa xmm1, xmm6
1481		− pshufd xmm2, xmm1, 00001110b
1482		− paddd xmm1, xmm2
1483		− movd [edx], xmm1 ; sum
1484		−
1485		− mov edx, [p_sd8x8]
1486		− pshufd xmm1, xmm6, 11110101b ; Sref1 Sref1 Sref0 Sref0
1487		− psubd xmm6, xmm1 ; 00 diff1 00 diff0
1488		− pshufd xmm1, xmm6, 00001000b ; xx xx diff1 diff0
1489		− movq [edx], xmm1
1490		− add edx, 8
1491		− mov [p_sd8x8], edx
1492		−
1493		− mov edx, [p_mad8x8]
1494		− WELS_MAX_REG_SSE2 xmm5
1495		− ;movdqa xmm1, xmm5
1496		− ;punpcklbw xmm1, xmm0
1497		− ;punpcklwd xmm1, xmm0
1498		− ;movd [edx], xmm1
1499		− ;punpckhbw xmm5, xmm0
1500		− ;punpcklwd xmm5, xmm0
1501		− ;movd [edx+4], xmm5
1502		− ;add edx, 8
1503		− ;mov [p_mad8x8], edx
1504		− mov [tmp_ecx], ecx
1505		− movhlps xmm1, xmm5
1506		− movd ecx, xmm5
1507		− mov [edx], cl
1508		− movd ecx, xmm1
1509		− mov [edx+1],cl
1510		− add edx, 2
1511		− mov [p_mad8x8], edx
1512		−
1513		− psrlq xmm7, 32
1514		− psllq xmm7, 32 ; clear sad
1515		− pxor xmm6, xmm6 ; sum_8x8 interleaves cur and pRef in Dword, Sref1 Scur1 Sref0 Scur0
1516		− pxor xmm5, xmm5 ; pMad8x8
1517		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
1518		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
1519		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
1520		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
1521		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
1522		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
1523		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
1524		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
1525		−
1526		− mov edx, [psad8x8]
1527		− movdqa xmm2, xmm7
1528		− pshufd xmm1, xmm2, 00001110b
1529		− movd [edx], xmm2
1530		− movd [edx+4], xmm1
1531		− add edx, 8
1532		− mov [psad8x8], edx ; sad8x8
1533		−
1534		− paddd xmm1, xmm2
1535		− movd edx, xmm1
1536		− add [tmp_sadframe], edx ; iFrameSad
1537		−
1538		− mov edx, [psum16x16]
1539		− movdqa xmm1, xmm6
1540		− pshufd xmm2, xmm1, 00001110b
1541		− paddd xmm1, xmm2
1542		− movd ebp, xmm1 ; sum
1543		− add [edx], ebp
1544		− add edx, 4
1545		− mov [psum16x16], edx
1546		−
1547		− mov edx, [psqsum16x16]
1548		− psrlq xmm7, 32
1549		− pshufd xmm2, xmm7, 00001110b
1550		− paddd xmm2, xmm7
1551		− movd [edx], xmm2 ; sqsum
1552		− add edx, 4
1553		− mov [psqsum16x16], edx
1554		−
1555		− mov edx, [p_sd8x8]
1556		− pshufd xmm1, xmm6, 11110101b ; Sref1 Sref1 Sref0 Sref0
1557		− psubd xmm6, xmm1 ; 00 diff1 00 diff0
1558		− pshufd xmm1, xmm6, 00001000b ; xx xx diff1 diff0
1559		− movq [edx], xmm1
1560		− add edx, 8
1561		− mov [p_sd8x8], edx
1562		−
1563		− mov edx, [p_mad8x8]
1564		− WELS_MAX_REG_SSE2 xmm5
1565		− ;movdqa xmm1, xmm5
1566		− ;punpcklbw xmm1, xmm0
1567		− ;punpcklwd xmm1, xmm0
1568		− ;movd [edx], xmm1
1569		− ;punpckhbw xmm5, xmm0
1570		− ;punpcklwd xmm5, xmm0
1571		− ;movd [edx+4], xmm5
1572		− ;add edx, 8
1573		− ;mov [p_mad8x8], edx
1574		− movhlps xmm1, xmm5
1575		− movd ecx, xmm5
1576		− mov [edx], cl
1577		− movd ecx, xmm1
1578		− mov [edx+1],cl
1579		− add edx, 2
1580		− mov [p_mad8x8], edx
1581		−
1582		− mov edx, [psqdiff16x16]
1583		− pshufd xmm1, xmm4, 00001110b
1584		− paddd xmm4, xmm1
1585		− pshufd xmm1, xmm4, 00000001b
1586		− paddd xmm4, xmm1
1587		− movd [edx], xmm4
1588		− add edx, 4
1589		− mov [psqdiff16x16], edx
1590		−
1591		− add edx, 16
1592		− sub esi, eax
1593		− sub edi, eax
1594		− add esi, 16
1595		− add edi, 16
1596		−
1597		− mov ecx, [tmp_ecx]
1598		− dec ecx
1599		− jnz sqdiff_bgd_width_loop
1600		−
1601		− mov esi, [tmp_esi]
1602		− mov edi, [tmp_edi]
1603		− add esi, eax
1604		− add edi, eax
1605		−
1606		− dec dword [iPicHeight]
1607		− jnz sqdiff_bgd_height_loop
1608		−
1609		− mov edx, [psadframe]
1610		− mov ebp, [tmp_sadframe]
1611		− mov [edx], ebp
1612		−
1613		− add esp, localsize
1614		− pop ebx
1615		− pop edi
1616		− pop esi
1617		− pop ebp
	1454	+ pxor xmm7, xmm7 ; pSad8x8 interleaves sqsum16x16: sqsum1 sad1 sqsum0 sad0
	1455	+ pxor xmm6, xmm6 ; sum_8x8 interleaves cur and pRef in Dword, Sref1 Scur1 Sref0 Scur0
	1456	+ pxor xmm5, xmm5 ; pMad8x8
	1457	+ pxor xmm4, xmm4 ; sqdiff_16x16 four Dword
	1458	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
	1459	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
	1460	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
	1461	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
	1462	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
	1463	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
	1464	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
	1465	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
	1466	+
	1467	+ mov edx, [psad8x8]
	1468	+ movdqa xmm2, xmm7
	1469	+ pshufd xmm1, xmm2, 00001110b
	1470	+ movd [edx], xmm2
	1471	+ movd [edx+4], xmm1
	1472	+ add edx, 8
	1473	+ mov [psad8x8], edx ; sad8x8
	1474	+
	1475	+ paddd xmm1, xmm2
	1476	+ movd edx, xmm1
	1477	+ add [tmp_sadframe], edx ; iFrameSad
	1478	+
	1479	+ mov edx, [psum16x16]
	1480	+ movdqa xmm1, xmm6
	1481	+ pshufd xmm2, xmm1, 00001110b
	1482	+ paddd xmm1, xmm2
	1483	+ movd [edx], xmm1 ; sum
	1484	+
	1485	+ mov edx, [p_sd8x8]
	1486	+ pshufd xmm1, xmm6, 11110101b ; Sref1 Sref1 Sref0 Sref0
	1487	+ psubd xmm6, xmm1 ; 00 diff1 00 diff0
	1488	+ pshufd xmm1, xmm6, 00001000b ; xx xx diff1 diff0
	1489	+ movq [edx], xmm1
	1490	+ add edx, 8
	1491	+ mov [p_sd8x8], edx
	1492	+
	1493	+ mov edx, [p_mad8x8]
	1494	+ WELS_MAX_REG_SSE2 xmm5
	1495	+ ;movdqa xmm1, xmm5
	1496	+ ;punpcklbw xmm1, xmm0
	1497	+ ;punpcklwd xmm1, xmm0
	1498	+ ;movd [edx], xmm1
	1499	+ ;punpckhbw xmm5, xmm0
	1500	+ ;punpcklwd xmm5, xmm0
	1501	+ ;movd [edx+4], xmm5
	1502	+ ;add edx, 8
	1503	+ ;mov [p_mad8x8], edx
	1504	+ mov [tmp_ecx], ecx
	1505	+ movhlps xmm1, xmm5
	1506	+ movd ecx, xmm5
	1507	+ mov [edx], cl
	1508	+ movd ecx, xmm1
	1509	+ mov [edx+1],cl
	1510	+ add edx, 2
	1511	+ mov [p_mad8x8], edx
	1512	+
	1513	+ psrlq xmm7, 32
	1514	+ psllq xmm7, 32 ; clear sad
	1515	+ pxor xmm6, xmm6 ; sum_8x8 interleaves cur and pRef in Dword, Sref1 Scur1 Sref0 Scur0
	1516	+ pxor xmm5, xmm5 ; pMad8x8
	1517	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
	1518	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
	1519	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
	1520	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
	1521	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
	1522	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
	1523	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
	1524	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, esi , edi , ebx
	1525	+
	1526	+ mov edx, [psad8x8]
	1527	+ movdqa xmm2, xmm7
	1528	+ pshufd xmm1, xmm2, 00001110b
	1529	+ movd [edx], xmm2
	1530	+ movd [edx+4], xmm1
	1531	+ add edx, 8
	1532	+ mov [psad8x8], edx ; sad8x8
	1533	+
	1534	+ paddd xmm1, xmm2
	1535	+ movd edx, xmm1
	1536	+ add [tmp_sadframe], edx ; iFrameSad
	1537	+
	1538	+ mov edx, [psum16x16]
	1539	+ movdqa xmm1, xmm6
	1540	+ pshufd xmm2, xmm1, 00001110b
	1541	+ paddd xmm1, xmm2
	1542	+ movd ebp, xmm1 ; sum
	1543	+ add [edx], ebp
	1544	+ add edx, 4
	1545	+ mov [psum16x16], edx
	1546	+
	1547	+ mov edx, [psqsum16x16]
	1548	+ psrlq xmm7, 32
	1549	+ pshufd xmm2, xmm7, 00001110b
	1550	+ paddd xmm2, xmm7
	1551	+ movd [edx], xmm2 ; sqsum
	1552	+ add edx, 4
	1553	+ mov [psqsum16x16], edx
	1554	+
	1555	+ mov edx, [p_sd8x8]
	1556	+ pshufd xmm1, xmm6, 11110101b ; Sref1 Sref1 Sref0 Sref0
	1557	+ psubd xmm6, xmm1 ; 00 diff1 00 diff0
	1558	+ pshufd xmm1, xmm6, 00001000b ; xx xx diff1 diff0
	1559	+ movq [edx], xmm1
	1560	+ add edx, 8
	1561	+ mov [p_sd8x8], edx
	1562	+
	1563	+ mov edx, [p_mad8x8]
	1564	+ WELS_MAX_REG_SSE2 xmm5
	1565	+ ;movdqa xmm1, xmm5
	1566	+ ;punpcklbw xmm1, xmm0
	1567	+ ;punpcklwd xmm1, xmm0
	1568	+ ;movd [edx], xmm1
	1569	+ ;punpckhbw xmm5, xmm0
	1570	+ ;punpcklwd xmm5, xmm0
	1571	+ ;movd [edx+4], xmm5
	1572	+ ;add edx, 8
	1573	+ ;mov [p_mad8x8], edx
	1574	+ movhlps xmm1, xmm5
	1575	+ movd ecx, xmm5
	1576	+ mov [edx], cl
	1577	+ movd ecx, xmm1
	1578	+ mov [edx+1],cl
	1579	+ add edx, 2
	1580	+ mov [p_mad8x8], edx
	1581	+
	1582	+ mov edx, [psqdiff16x16]
	1583	+ pshufd xmm1, xmm4, 00001110b
	1584	+ paddd xmm4, xmm1
	1585	+ pshufd xmm1, xmm4, 00000001b
	1586	+ paddd xmm4, xmm1
	1587	+ movd [edx], xmm4
	1588	+ add edx, 4
	1589	+ mov [psqdiff16x16], edx
	1590	+
	1591	+ add edx, 16
	1592	+ sub esi, eax
	1593	+ sub edi, eax
	1594	+ add esi, 16
	1595	+ add edi, 16
	1596	+
	1597	+ mov ecx, [tmp_ecx]
	1598	+ dec ecx
	1599	+ jnz sqdiff_bgd_width_loop
	1600	+
	1601	+ mov esi, [tmp_esi]
	1602	+ mov edi, [tmp_edi]
	1603	+ add esi, eax
	1604	+ add edi, eax
	1605	+
	1606	+ dec dword [iPicHeight]
	1607	+ jnz sqdiff_bgd_height_loop
	1608	+
	1609	+ mov edx, [psadframe]
	1610	+ mov ebp, [tmp_sadframe]
	1611	+ mov [edx], ebp
	1612	+
	1613	+ add esp, localsize
	1614	+ pop ebx
	1615	+ pop edi
	1616	+ pop esi
	1617	+ pop ebp
1618	1618	%undef cur_data
1619	1619	%undef ref_data
1620	1620	%undef iPicWidth
@@ -1631,7 +1631,7 @@ sqdiff_bgd_width_loop: · 10 unmodified lines
1631	1631	%undef tmp_edi
1632	1632	%undef pushsize
1633	1633	%undef localsize
1634		− ret
	1634	+ ret
1635	1635	%else
1636	1636
1637	1637	;*************************************************************************************************************
@@ -1651,142 +1651,142 @@ WELS_EXTERN VAACalcSadBgd_sse2 · 13 unmodified lines
1651	1651	%define p_sd8x8 arg8;
1652	1652	%define p_mad8x8 arg9;
1653	1653
1654		− push r12
1655		− push r13
1656		− push r14
1657		− push r15
	1654	+ push r12
	1655	+ push r13
	1656	+ push r14
	1657	+ push r15
1658	1658	%assign push_num 4
1659		− PUSH_XMM 10
	1659	+ PUSH_XMM 10
1660	1660	%ifdef WIN64
1661		− mov r4,arg5
1662		− ; mov r5,arg6
	1661	+ mov r4,arg5
	1662	+ ; mov r5,arg6
1663	1663	%endif
1664		− mov r14,arg7
1665		− SIGN_EXTENSION r2,r2d
1666		− SIGN_EXTENSION r3,r3d
1667		− SIGN_EXTENSION r4,r4d
1668		−
1669		−
1670		− mov r13,r4
1671		− mov r15,r0
1672		− shr r2,4
1673		− shr r3,4
1674		− shl r13,4
1675		− pxor xmm0, xmm0
1676		− pxor xmm8, xmm8
1677		− pxor xmm9, xmm9
	1664	+ mov r14,arg7
	1665	+ SIGN_EXTENSION r2,r2d
	1666	+ SIGN_EXTENSION r3,r3d
	1667	+ SIGN_EXTENSION r4,r4d
	1668	+
	1669	+
	1670	+ mov r13,r4
	1671	+ mov r15,r0
	1672	+ shr r2,4
	1673	+ shr r3,4
	1674	+ shl r13,4
	1675	+ pxor xmm0, xmm0
	1676	+ pxor xmm8, xmm8
	1677	+ pxor xmm9, xmm9
1678	1678	bgd_height_loop:
1679		− ;mov ecx, dword [iPicWidth]
1680		− push r2
1681		− %assign push_num push_num+1
1682		− mov r10, r15
1683		− mov r11, r1
	1679	+ ;mov ecx, dword [iPicWidth]
	1680	+ push r2
	1681	+ %assign push_num push_num+1
	1682	+ mov r10, r15
	1683	+ mov r11, r1
1684	1684	bgd_width_loop:
1685		− pxor xmm7, xmm7 ; pSad8x8
1686		− pxor xmm6, xmm6 ; sum_cur_8x8
1687		− pxor xmm5, xmm5 ; sum_ref_8x8
1688		− pxor xmm4, xmm4 ; pMad8x8
1689		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
1690		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
1691		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
1692		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
1693		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
1694		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
1695		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
1696		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
1697		−
1698		−
1699		− mov r14, p_mad8x8
1700		− WELS_MAX_REG_SSE2 xmm4
1701		−
1702		− ;mov [tmp_ecx], ecx
1703		− movhlps xmm1, xmm4
1704		− movd r0d, xmm4
1705		−
1706		−
1707		− mov [r14], r0b
1708		− movd r0d, xmm1
1709		− mov [r14+1],r0b
1710		− add r14, 2
1711		− ;mov p_mad8x8, r14
1712		−
1713		−
1714		− pslldq xmm7, 4
1715		− pslldq xmm6, 4
1716		− pslldq xmm5, 4
1717		−
1718		−
1719		− pxor xmm4, xmm4 ; pMad8x8
1720		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
1721		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
1722		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
1723		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
1724		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
1725		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
1726		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
1727		− WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
1728		−
1729		− ;mov r14, [p_mad8x8]
1730		− WELS_MAX_REG_SSE2 xmm4
1731		−
1732		− movhlps xmm1, xmm4
1733		− movd r0d, xmm4
1734		− mov [r14], r0b
1735		− movd r0d, xmm1
1736		− mov [r14+1],r0b
1737		− add r14, 2
1738		− mov p_mad8x8, r14
1739		−
1740		− ; data in xmm7, xmm6, xmm5: D1 D3 D0 D2
1741		−
1742		− mov r14, psad8x8
1743		− pshufd xmm1, xmm7, 10001101b ; D3 D2 D1 D0
1744		− movdqa [r14], xmm1
1745		− add r14, 16
1746		− mov psad8x8, r14 ; sad8x8
1747		−
1748		− paddd xmm1, xmm7 ; D1+3 D3+2 D0+1 D2+0
1749		− pshufd xmm2, xmm1, 00000011b
1750		− paddd xmm1, xmm2
1751		− movd r14d, xmm1
1752		− movd xmm9, r14d
1753		− paddd xmm8, xmm9 ; sad frame
1754		−
1755		− mov r14, p_sd8x8
1756		− psubd xmm6, xmm5
1757		− pshufd xmm1, xmm6, 10001101b
1758		− movdqa [r14], xmm1
1759		− add r14, 16
1760		− mov p_sd8x8, r14
1761		−
1762		−
1763		− ;add edx, 16
1764		− sub r15, r13
1765		− sub r1, r13
1766		− add r15, 16
1767		− add r1, 16
1768		−
1769		−
1770		− dec r2
1771		− jnz bgd_width_loop
1772		− pop r2
	1685	+ pxor xmm7, xmm7 ; pSad8x8
	1686	+ pxor xmm6, xmm6 ; sum_cur_8x8
	1687	+ pxor xmm5, xmm5 ; sum_ref_8x8
	1688	+ pxor xmm4, xmm4 ; pMad8x8
	1689	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
	1690	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
	1691	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
	1692	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
	1693	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
	1694	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
	1695	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
	1696	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
	1697	+
	1698	+
	1699	+ mov r14, p_mad8x8
	1700	+ WELS_MAX_REG_SSE2 xmm4
	1701	+
	1702	+ ;mov [tmp_ecx], ecx
	1703	+ movhlps xmm1, xmm4
	1704	+ movd r0d, xmm4
	1705	+
	1706	+
	1707	+ mov [r14], r0b
	1708	+ movd r0d, xmm1
	1709	+ mov [r14+1],r0b
	1710	+ add r14, 2
	1711	+ ;mov p_mad8x8, r14
	1712	+
	1713	+
	1714	+ pslldq xmm7, 4
	1715	+ pslldq xmm6, 4
	1716	+ pslldq xmm5, 4
	1717	+
	1718	+
	1719	+ pxor xmm4, xmm4 ; pMad8x8
	1720	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
	1721	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
	1722	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
	1723	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
	1724	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
	1725	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
	1726	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
	1727	+ WELS_SAD_SD_MAD_16x1_SSE2 xmm7, xmm6, xmm5, xmm4 ,r15 ,r1, r4
	1728	+
	1729	+ ;mov r14, [p_mad8x8]
	1730	+ WELS_MAX_REG_SSE2 xmm4
	1731	+
	1732	+ movhlps xmm1, xmm4
	1733	+ movd r0d, xmm4
	1734	+ mov [r14], r0b
	1735	+ movd r0d, xmm1
	1736	+ mov [r14+1],r0b
	1737	+ add r14, 2
	1738	+ mov p_mad8x8, r14
	1739	+
	1740	+ ; data in xmm7, xmm6, xmm5: D1 D3 D0 D2
	1741	+
	1742	+ mov r14, psad8x8
	1743	+ pshufd xmm1, xmm7, 10001101b ; D3 D2 D1 D0
	1744	+ movdqa [r14], xmm1
	1745	+ add r14, 16
	1746	+ mov psad8x8, r14 ; sad8x8
	1747	+
	1748	+ paddd xmm1, xmm7 ; D1+3 D3+2 D0+1 D2+0
	1749	+ pshufd xmm2, xmm1, 00000011b
	1750	+ paddd xmm1, xmm2
	1751	+ movd r14d, xmm1
	1752	+ movd xmm9, r14d
	1753	+ paddd xmm8, xmm9 ; sad frame
	1754	+
	1755	+ mov r14, p_sd8x8
	1756	+ psubd xmm6, xmm5
	1757	+ pshufd xmm1, xmm6, 10001101b
	1758	+ movdqa [r14], xmm1
	1759	+ add r14, 16
	1760	+ mov p_sd8x8, r14
	1761	+
	1762	+
	1763	+ ;add edx, 16
	1764	+ sub r15, r13
	1765	+ sub r1, r13
	1766	+ add r15, 16
	1767	+ add r1, 16
	1768	+
	1769	+
	1770	+ dec r2
	1771	+ jnz bgd_width_loop
	1772	+ pop r2
1773	1773	%assign push_num push_num-1
1774		− mov r15, r10
1775		− mov r1, r11
1776		− add r15, r13
1777		− add r1, r13
1778		−
1779		− dec r3
1780		− jnz bgd_height_loop
1781		−
1782		− mov r13, psadframe
1783		− movd [r13], xmm8
1784		−
1785		− POP_XMM
1786		− pop r15
1787		− pop r14
1788		− pop r13
1789		− pop r12
	1774	+ mov r15, r10
	1775	+ mov r1, r11
	1776	+ add r15, r13
	1777	+ add r1, r13
	1778	+
	1779	+ dec r3
	1780	+ jnz bgd_height_loop
	1781	+
	1782	+ mov r13, psadframe
	1783	+ movd [r13], xmm8
	1784	+
	1785	+ POP_XMM
	1786	+ pop r15
	1787	+ pop r14
	1788	+ pop r13
	1789	+ pop r12
1790	1790	%assign push_num 0
1791	1791	%undef cur_data
1792	1792	%undef ref_data
@@ -1801,7 +1801,7 @@ bgd_width_loop: · 8 unmodified lines
1801	1801	%undef tmp_edi
1802	1802	%undef pushsize
1803	1803	%undef localsize
1804		− ret
	1804	+ ret
1805	1805
1806	1806
1807	1807
@@ -1826,189 +1826,189 @@ WELS_EXTERN VAACalcSadSsdBgd_sse2 · 18 unmodified lines
1826	1826	%define p_sd8x8 arg11
1827	1827	%define p_mad8x8 arg12
1828	1828
1829		− push r12
1830		− push r13
1831		− push r14
1832		− push r15
	1829	+ push r12
	1830	+ push r13
	1831	+ push r14
	1832	+ push r15
1833	1833	%assign push_num 4
1834		− PUSH_XMM 10
	1834	+ PUSH_XMM 10
1835	1835	%ifdef WIN64
1836		− mov r4,arg5
1837		− ;mov r5,arg6
	1836	+ mov r4,arg5
	1837	+ ;mov r5,arg6
1838	1838	%endif
1839		− SIGN_EXTENSION r2,r2d
1840		− SIGN_EXTENSION r3,r3d
1841		− SIGN_EXTENSION r4,r4d
	1839	+ SIGN_EXTENSION r2,r2d
	1840	+ SIGN_EXTENSION r3,r3d
	1841	+ SIGN_EXTENSION r4,r4d
1842	1842
1843		− mov r13,r4
1844		− shr r2, 4 ; iPicWidth/16
1845		− shr r3, 4 ; iPicHeight/16
1846		− shl r13, 4 ; iPicStride*16
1847		− pxor xmm0, xmm0
1848		− pxor xmm8, xmm8
1849		− pxor xmm9, xmm9
	1843	+ mov r13,r4
	1844	+ shr r2, 4 ; iPicWidth/16
	1845	+ shr r3, 4 ; iPicHeight/16
	1846	+ shl r13, 4 ; iPicStride*16
	1847	+ pxor xmm0, xmm0
	1848	+ pxor xmm8, xmm8
	1849	+ pxor xmm9, xmm9
1850	1850
1851	1851
1852	1852	sqdiff_bgd_height_loop:
1853		− mov r10, r0
1854		− mov r11, r1
1855		− push r2
	1853	+ mov r10, r0
	1854	+ mov r11, r1
	1855	+ push r2
1856	1856	%assign push_num push_num+1
1857	1857	sqdiff_bgd_width_loop:
1858	1858
1859		− pxor xmm7, xmm7 ; pSad8x8 interleaves sqsum16x16: sqsum1 sad1 sqsum0 sad0
1860		− pxor xmm6, xmm6 ; sum_8x8 interleaves cur and pRef in Dword, Sref1 Scur1 Sref0 Scur0
1861		− pxor xmm5, xmm5 ; pMad8x8
1862		− pxor xmm4, xmm4 ; sqdiff_16x16 four Dword
1863		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
1864		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
1865		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
1866		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
1867		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
1868		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
1869		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
1870		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
1871		−
1872		− mov r14, psad8x8
1873		− movdqa xmm2, xmm7
1874		− pshufd xmm1, xmm2, 00001110b
1875		− movd [r14], xmm2
1876		− movd [r14+4], xmm1
1877		− add r14, 8
1878		− mov psad8x8, r14 ; sad8x8
1879		−
1880		− paddd xmm1, xmm2
1881		− movd r14d, xmm1
1882		− movd xmm9,r14d
1883		− paddd xmm8, xmm9 ; iFrameSad
1884		−
1885		− mov r14, psum16x16
1886		− movdqa xmm1, xmm6
1887		− pshufd xmm2, xmm1, 00001110b
1888		− paddd xmm1, xmm2
1889		− movd [r14], xmm1 ; sum
1890		−
1891		− mov r14, p_sd8x8
1892		− pshufd xmm1, xmm6, 11110101b ; Sref1 Sref1 Sref0 Sref0
1893		− psubd xmm6, xmm1 ; 00 diff1 00 diff0
1894		− pshufd xmm1, xmm6, 00001000b ; xx xx diff1 diff0
1895		− movq [r14], xmm1
1896		− add r14, 8
1897		− mov p_sd8x8, r14
1898		−
1899		− mov r14, p_mad8x8
1900		− WELS_MAX_REG_SSE2 xmm5
1901		−
1902		− movhlps xmm1, xmm5
1903		− push r0
1904		− movd r0d, xmm5
1905		− mov [r14], r0b
1906		− movd r0d, xmm1
1907		− mov [r14+1],r0b
1908		− pop r0
1909		− add r14, 2
1910		− mov p_mad8x8, r14
1911		−
1912		− psrlq xmm7, 32
1913		− psllq xmm7, 32 ; clear sad
1914		− pxor xmm6, xmm6 ; sum_8x8 interleaves cur and pRef in Dword, Sref1 Scur1 Sref0 Scur0
1915		− pxor xmm5, xmm5 ; pMad8x8
1916		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
1917		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
1918		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
1919		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
1920		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
1921		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
1922		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
1923		− WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
1924		−
1925		− mov r14, psad8x8
1926		− movdqa xmm2, xmm7
1927		− pshufd xmm1, xmm2, 00001110b
1928		− movd [r14], xmm2
1929		− movd [r14+4], xmm1
1930		− add r14, 8
1931		− mov psad8x8, r14 ; sad8x8
1932		−
1933		− paddd xmm1, xmm2
1934		− movd r14d, xmm1
1935		− movd xmm9, r14d
1936		− paddd xmm8, xmm9 ; iFrameSad
1937		−
1938		− mov r14, psum16x16
1939		− movdqa xmm1, xmm6
1940		− pshufd xmm2, xmm1, 00001110b
1941		− paddd xmm1, xmm2
1942		− movd r15d, xmm1 ; sum
1943		− add [r14], r15d
1944		− add r14, 4
1945		− mov psum16x16, r14
1946		−
1947		− mov r14, psqsum16x16
1948		− psrlq xmm7, 32
1949		− pshufd xmm2, xmm7, 00001110b
1950		− paddd xmm2, xmm7
1951		− movd [r14], xmm2 ; sqsum
1952		− add r14, 4
1953		− mov psqsum16x16, r14
1954		−
1955		− mov r14, p_sd8x8
1956		− pshufd xmm1, xmm6, 11110101b ; Sref1 Sref1 Sref0 Sref0
1957		− psubd xmm6, xmm1 ; 00 diff1 00 diff0
1958		− pshufd xmm1, xmm6, 00001000b ; xx xx diff1 diff0
1959		− movq [r14], xmm1
1960		− add r14, 8
1961		− mov p_sd8x8, r14
1962		−
1963		− mov r14, p_mad8x8
1964		− WELS_MAX_REG_SSE2 xmm5
1965		−
1966		−
1967		− movhlps xmm1, xmm5
1968		− push r0
1969		− movd r0d, xmm5
1970		− mov [r14], r0b
1971		− movd r0d, xmm1
1972		− mov [r14+1],r0b
1973		− pop r0
1974		− add r14, 2
1975		− mov p_mad8x8, r14
1976		−
1977		− mov r14, psqdiff16x16
1978		− pshufd xmm1, xmm4, 00001110b
1979		− paddd xmm4, xmm1
1980		− pshufd xmm1, xmm4, 00000001b
1981		− paddd xmm4, xmm1
1982		− movd [r14], xmm4
1983		− add r14, 4
1984		− mov psqdiff16x16, r14
1985		−
1986		− add r14, 16
1987		− sub r0, r13
1988		− sub r1, r13
1989		− add r0, 16
1990		− add r1, 16
1991		−
1992		− dec r2
1993		− jnz sqdiff_bgd_width_loop
1994		− pop r2
1995		− %assign push_num push_num-1
1996		− mov r0, r10
1997		− mov r1, r11
1998		− add r0, r13
1999		− add r1, r13
2000		−
2001		− dec r3
2002		− jnz sqdiff_bgd_height_loop
2003		−
2004		− mov r14, psadframe
2005		− movd [r14], xmm8
2006		−
2007		− POP_XMM
2008		− pop r15
2009		− pop r14
2010		− pop r13
2011		− pop r12
	1859	+ pxor xmm7, xmm7 ; pSad8x8 interleaves sqsum16x16: sqsum1 sad1 sqsum0 sad0
	1860	+ pxor xmm6, xmm6 ; sum_8x8 interleaves cur and pRef in Dword, Sref1 Scur1 Sref0 Scur0
	1861	+ pxor xmm5, xmm5 ; pMad8x8
	1862	+ pxor xmm4, xmm4 ; sqdiff_16x16 four Dword
	1863	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
	1864	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
	1865	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
	1866	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
	1867	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
	1868	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
	1869	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
	1870	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
	1871	+
	1872	+ mov r14, psad8x8
	1873	+ movdqa xmm2, xmm7
	1874	+ pshufd xmm1, xmm2, 00001110b
	1875	+ movd [r14], xmm2
	1876	+ movd [r14+4], xmm1
	1877	+ add r14, 8
	1878	+ mov psad8x8, r14 ; sad8x8
	1879	+
	1880	+ paddd xmm1, xmm2
	1881	+ movd r14d, xmm1
	1882	+ movd xmm9,r14d
	1883	+ paddd xmm8, xmm9 ; iFrameSad
	1884	+
	1885	+ mov r14, psum16x16
	1886	+ movdqa xmm1, xmm6
	1887	+ pshufd xmm2, xmm1, 00001110b
	1888	+ paddd xmm1, xmm2
	1889	+ movd [r14], xmm1 ; sum
	1890	+
	1891	+ mov r14, p_sd8x8
	1892	+ pshufd xmm1, xmm6, 11110101b ; Sref1 Sref1 Sref0 Sref0
	1893	+ psubd xmm6, xmm1 ; 00 diff1 00 diff0
	1894	+ pshufd xmm1, xmm6, 00001000b ; xx xx diff1 diff0
	1895	+ movq [r14], xmm1
	1896	+ add r14, 8
	1897	+ mov p_sd8x8, r14
	1898	+
	1899	+ mov r14, p_mad8x8
	1900	+ WELS_MAX_REG_SSE2 xmm5
	1901	+
	1902	+ movhlps xmm1, xmm5
	1903	+ push r0
	1904	+ movd r0d, xmm5
	1905	+ mov [r14], r0b
	1906	+ movd r0d, xmm1
	1907	+ mov [r14+1],r0b
	1908	+ pop r0
	1909	+ add r14, 2
	1910	+ mov p_mad8x8, r14
	1911	+
	1912	+ psrlq xmm7, 32
	1913	+ psllq xmm7, 32 ; clear sad
	1914	+ pxor xmm6, xmm6 ; sum_8x8 interleaves cur and pRef in Dword, Sref1 Scur1 Sref0 Scur0
	1915	+ pxor xmm5, xmm5 ; pMad8x8
	1916	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
	1917	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
	1918	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
	1919	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
	1920	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
	1921	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
	1922	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
	1923	+ WELS_SAD_BGD_SQDIFF_16x1_SSE2 xmm7, xmm6, xmm5, xmm4, r0 , r1 , r4
	1924	+
	1925	+ mov r14, psad8x8
	1926	+ movdqa xmm2, xmm7
	1927	+ pshufd xmm1, xmm2, 00001110b
	1928	+ movd [r14], xmm2
	1929	+ movd [r14+4], xmm1
	1930	+ add r14, 8
	1931	+ mov psad8x8, r14 ; sad8x8
	1932	+
	1933	+ paddd xmm1, xmm2
	1934	+ movd r14d, xmm1
	1935	+ movd xmm9, r14d
	1936	+ paddd xmm8, xmm9 ; iFrameSad
	1937	+
	1938	+ mov r14, psum16x16
	1939	+ movdqa xmm1, xmm6
	1940	+ pshufd xmm2, xmm1, 00001110b
	1941	+ paddd xmm1, xmm2
	1942	+ movd r15d, xmm1 ; sum
	1943	+ add [r14], r15d
	1944	+ add r14, 4
	1945	+ mov psum16x16, r14
	1946	+
	1947	+ mov r14, psqsum16x16
	1948	+ psrlq xmm7, 32
	1949	+ pshufd xmm2, xmm7, 00001110b
	1950	+ paddd xmm2, xmm7
	1951	+ movd [r14], xmm2 ; sqsum
	1952	+ add r14, 4
	1953	+ mov psqsum16x16, r14
	1954	+
	1955	+ mov r14, p_sd8x8
	1956	+ pshufd xmm1, xmm6, 11110101b ; Sref1 Sref1 Sref0 Sref0
	1957	+ psubd xmm6, xmm1 ; 00 diff1 00 diff0
	1958	+ pshufd xmm1, xmm6, 00001000b ; xx xx diff1 diff0
	1959	+ movq [r14], xmm1
	1960	+ add r14, 8
	1961	+ mov p_sd8x8, r14
	1962	+
	1963	+ mov r14, p_mad8x8
	1964	+ WELS_MAX_REG_SSE2 xmm5
	1965	+
	1966	+
	1967	+ movhlps xmm1, xmm5
	1968	+ push r0
	1969	+ movd r0d, xmm5
	1970	+ mov [r14], r0b
	1971	+ movd r0d, xmm1
	1972	+ mov [r14+1],r0b
	1973	+ pop r0
	1974	+ add r14, 2
	1975	+ mov p_mad8x8, r14
	1976	+
	1977	+ mov r14, psqdiff16x16
	1978	+ pshufd xmm1, xmm4, 00001110b
	1979	+ paddd xmm4, xmm1
	1980	+ pshufd xmm1, xmm4, 00000001b
	1981	+ paddd xmm4, xmm1
	1982	+ movd [r14], xmm4
	1983	+ add r14, 4
	1984	+ mov psqdiff16x16, r14
	1985	+
	1986	+ add r14, 16
	1987	+ sub r0, r13
	1988	+ sub r1, r13
	1989	+ add r0, 16
	1990	+ add r1, 16
	1991	+
	1992	+ dec r2
	1993	+ jnz sqdiff_bgd_width_loop
	1994	+ pop r2
	1995	+ %assign push_num push_num-1
	1996	+ mov r0, r10
	1997	+ mov r1, r11
	1998	+ add r0, r13
	1999	+ add r1, r13
	2000	+
	2001	+ dec r3
	2002	+ jnz sqdiff_bgd_height_loop
	2003	+
	2004	+ mov r14, psadframe
	2005	+ movd [r14], xmm8
	2006	+
	2007	+ POP_XMM
	2008	+ pop r15
	2009	+ pop r14
	2010	+ pop r13
	2011	+ pop r12
2012	2012	%assign push_num 0
2013	2013	%undef cur_data
2014	2014	%undef ref_data
@@ -2026,5 +2026,5 @@ sqdiff_bgd_width_loop: · 11 unmodified lines
2026	2026	%undef tmp_edi
2027	2027	%undef pushsize
2028	2028	%undef localsize
2029		− ret
	2029	+ ret
2030	2030	%endif

Convert all tabs to spaces in assembly sources, unify indentation

Keyboard shortcuts

General

Issues and pull requests

Issues, pull requests and Inbox

Repository code