25 |
; * |
; * |
26 |
; *************************************************************************/ |
; *************************************************************************/ |
27 |
|
|
|
; data/text alignment |
|
|
%define ALIGN 8 |
|
|
|
|
28 |
%define SATURATE |
%define SATURATE |
29 |
|
|
30 |
bits 32 |
BITS 32 |
31 |
|
|
32 |
%macro cglobal 1 |
%macro cglobal 1 |
33 |
%ifdef PREFIX |
%ifdef PREFIX |
47 |
%endif |
%endif |
48 |
%endmacro |
%endmacro |
49 |
|
|
50 |
;*************************************************************************** |
;============================================================================= |
51 |
; Local data |
; Local data (Read Only) |
52 |
;*************************************************************************** |
;============================================================================= |
53 |
|
|
54 |
%ifdef FORMAT_COFF |
%ifdef FORMAT_COFF |
55 |
SECTION .data data |
SECTION .rodata data |
56 |
%else |
%else |
57 |
SECTION .data data align=8 |
SECTION .rodata data align=16 |
58 |
%endif |
%endif |
59 |
|
|
60 |
mmx_one: |
mmx_one: |
61 |
times 4 dw 1 |
times 4 dw 1 |
62 |
|
|
63 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
|
; |
|
64 |
; divide by 2Q table |
; divide by 2Q table |
65 |
; |
;----------------------------------------------------------------------------- |
|
;=========================================================================== |
|
66 |
|
|
67 |
align 16 |
ALIGN 16 |
68 |
mmx_div: |
mmx_div: |
69 |
%assign quant 1 |
times 4 dw 65535 ; the div by 2 formula will overflow for the case |
70 |
%rep 31 |
; quant=1 but we don't care much because quant=1 |
71 |
|
; is handled by a different piece of code that |
72 |
|
; doesn't use this table. |
73 |
|
%assign quant 2 |
74 |
|
%rep 30 |
75 |
times 4 dw (1<<17) / (quant*2) + 1 |
times 4 dw (1<<17) / (quant*2) + 1 |
76 |
%assign quant quant+1 |
%assign quant quant+1 |
77 |
%endrep |
%endrep |
78 |
|
|
79 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
|
; |
|
80 |
; intra matrix |
; intra matrix |
81 |
; |
;----------------------------------------------------------------------------- |
|
;=========================================================================== |
|
82 |
|
|
83 |
cextern intra_matrix |
cextern intra_matrix |
84 |
cextern intra_matrix_fix |
cextern intra_matrix_fix |
85 |
|
|
86 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
|
; |
|
87 |
; inter matrix |
; inter matrix |
88 |
; |
;----------------------------------------------------------------------------- |
|
;=========================================================================== |
|
89 |
|
|
90 |
cextern inter_matrix |
cextern inter_matrix |
91 |
cextern inter_matrix_fix |
cextern inter_matrix_fix |
95 |
%define VM18Q 4 |
%define VM18Q 4 |
96 |
|
|
97 |
|
|
98 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
|
; |
|
99 |
; quantd table |
; quantd table |
100 |
; |
;----------------------------------------------------------------------------- |
|
;=========================================================================== |
|
101 |
|
|
102 |
quantd: |
quantd: |
103 |
%assign quant 1 |
%assign quant 1 |
106 |
%assign quant quant+1 |
%assign quant quant+1 |
107 |
%endrep |
%endrep |
108 |
|
|
109 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
|
; |
|
110 |
; multiple by 2Q table |
; multiple by 2Q table |
111 |
; |
;----------------------------------------------------------------------------- |
|
;=========================================================================== |
|
112 |
|
|
113 |
mmx_mul_quant: |
mmx_mul_quant: |
114 |
%assign quant 1 |
%assign quant 1 |
117 |
%assign quant quant+1 |
%assign quant quant+1 |
118 |
%endrep |
%endrep |
119 |
|
|
120 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
|
; |
|
121 |
; saturation limits |
; saturation limits |
122 |
; |
;----------------------------------------------------------------------------- |
|
;=========================================================================== |
|
123 |
|
|
124 |
align 16 |
ALIGN 16 |
125 |
|
|
126 |
mmx_32767_minus_2047: |
mmx_32767_minus_2047: |
127 |
times 4 dw (32767-2047) |
times 4 dw (32767-2047) |
134 |
zero: |
zero: |
135 |
times 4 dw 0 |
times 4 dw 0 |
136 |
|
|
137 |
;*************************************************************************** |
;============================================================================= |
138 |
; Local data |
; Code |
139 |
;*************************************************************************** |
;============================================================================= |
140 |
|
|
141 |
section .text |
SECTION .text |
142 |
|
|
143 |
;=========================================================================== |
cglobal quant_mpeg_intra_mmx |
144 |
|
cglobal quant_mpeg_inter_mmx |
145 |
|
cglobal dequant_mpeg_intra_mmx |
146 |
|
cglobal dequant_mpeg_inter_mmx |
147 |
|
|
148 |
|
;----------------------------------------------------------------------------- |
149 |
; |
; |
150 |
; uint32_t quant_mpeg_intra_mmx(int16_t * coeff, |
; uint32_t quant_mpeg_intra_mmx(int16_t * coeff, |
151 |
; const int16_t const * data, |
; const int16_t const * data, |
152 |
; const uint32_t quant, |
; const uint32_t quant, |
153 |
; const uint32_t dcscalar); |
; const uint32_t dcscalar); |
154 |
; |
; |
155 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
156 |
|
|
157 |
align ALIGN |
ALIGN 16 |
|
cglobal quant_mpeg_intra_mmx |
|
158 |
quant_mpeg_intra_mmx: |
quant_mpeg_intra_mmx: |
159 |
|
|
160 |
push ecx |
push ecx |
176 |
|
|
177 |
movq mm7, [mmx_div + eax * 8 - 8] ; multipliers[quant] -> mm7 |
movq mm7, [mmx_div + eax * 8 - 8] ; multipliers[quant] -> mm7 |
178 |
|
|
179 |
align ALIGN |
ALIGN 16 |
180 |
.loop |
.loop |
181 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
182 |
movq mm3, [esi + 8*ecx + 8] ; |
movq mm3, [esi + 8*ecx + 8] ; |
|
|
|
183 |
pxor mm1, mm1 ; mm1 = 0 |
pxor mm1, mm1 ; mm1 = 0 |
184 |
pxor mm4, mm4 |
pxor mm4, mm4 |
|
|
|
185 |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
186 |
pcmpgtw mm4, mm3 |
pcmpgtw mm4, mm3 |
|
|
|
187 |
pxor mm0, mm1 ; mm0 = |mm0| |
pxor mm0, mm1 ; mm0 = |mm0| |
188 |
pxor mm3, mm4 ; |
pxor mm3, mm4 ; |
189 |
psubw mm0, mm1 ; displace |
psubw mm0, mm1 ; displace |
190 |
psubw mm3, mm4 ; |
psubw mm3, mm4 ; |
|
|
|
191 |
psllw mm0, 4 ; level << 4 |
psllw mm0, 4 ; level << 4 |
192 |
psllw mm3, 4 ; |
psllw mm3, 4 |
|
|
|
193 |
movq mm2, [intra_matrix + 8*ecx] |
movq mm2, [intra_matrix + 8*ecx] |
194 |
psrlw mm2, 1 ; intra_matrix[i]>>1 |
psrlw mm2, 1 ; intra_matrix[i]>>1 |
195 |
paddw mm0, mm2 |
paddw mm0, mm2 |
|
|
|
196 |
movq mm2, [intra_matrix_fix + ecx*8] |
movq mm2, [intra_matrix_fix + ecx*8] |
197 |
pmulhw mm0, mm2 ; (level<<4 + intra_matrix[i]>>1) / intra_matrix[i] |
pmulhw mm0, mm2 ; (level<<4 + intra_matrix[i]>>1) / intra_matrix[i] |
|
|
|
198 |
movq mm2, [intra_matrix + 8*ecx + 8] |
movq mm2, [intra_matrix + 8*ecx + 8] |
199 |
psrlw mm2, 1 |
psrlw mm2, 1 |
200 |
paddw mm3, mm2 |
paddw mm3, mm2 |
|
|
|
201 |
movq mm2, [intra_matrix_fix + ecx*8 + 8] |
movq mm2, [intra_matrix_fix + ecx*8 + 8] |
202 |
pmulhw mm3, mm2 |
pmulhw mm3, mm2 |
|
|
|
203 |
paddw mm0, mm5 ; + quantd |
paddw mm0, mm5 ; + quantd |
204 |
paddw mm3, mm5 |
paddw mm3, mm5 |
205 |
pmulhw mm0, mm7 ; mm0 = (mm0 / 2Q) >> 16 |
pmulhw mm0, mm7 ; mm0 = (mm0 / 2Q) >> 16 |
206 |
pmulhw mm3, mm7 ; |
pmulhw mm3, mm7 ; |
207 |
psrlw mm0, 1 ; additional shift by 1 => 16 + 1 = 17 |
psrlw mm0, 1 ; additional shift by 1 => 16 + 1 = 17 |
208 |
psrlw mm3, 1 |
psrlw mm3, 1 |
|
|
|
209 |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
210 |
pxor mm3, mm4 ; |
pxor mm3, mm4 ; |
211 |
psubw mm0, mm1 ; undisplace |
psubw mm0, mm1 ; undisplace |
220 |
|
|
221 |
.done |
.done |
222 |
; caclulate data[0] // (int32_t)dcscalar) |
; caclulate data[0] // (int32_t)dcscalar) |
|
|
|
223 |
mov ecx, [esp + 12 + 16] ; dcscalar |
mov ecx, [esp + 12 + 16] ; dcscalar |
224 |
mov edx, ecx |
mov edx, ecx |
225 |
movsx eax, word [esi] ; data[0] |
movsx eax, word [esi] ; data[0] |
244 |
xor eax, eax ; return(0); |
xor eax, eax ; return(0); |
245 |
ret |
ret |
246 |
|
|
247 |
align ALIGN |
ALIGN 16 |
248 |
.q1loop |
.q1loop |
249 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
250 |
movq mm3, [esi + 8*ecx + 8] ; |
movq mm3, [esi + 8*ecx + 8] ; |
|
|
|
251 |
pxor mm1, mm1 ; mm1 = 0 |
pxor mm1, mm1 ; mm1 = 0 |
252 |
pxor mm4, mm4 ; |
pxor mm4, mm4 ; |
|
|
|
253 |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
254 |
pcmpgtw mm4, mm3 ; |
pcmpgtw mm4, mm3 ; |
|
|
|
255 |
pxor mm0, mm1 ; mm0 = |mm0| |
pxor mm0, mm1 ; mm0 = |mm0| |
256 |
pxor mm3, mm4 ; |
pxor mm3, mm4 ; |
257 |
psubw mm0, mm1 ; displace |
psubw mm0, mm1 ; displace |
258 |
psubw mm3, mm4 ; |
psubw mm3, mm4 ; |
|
|
|
259 |
psllw mm0, 4 |
psllw mm0, 4 |
260 |
psllw mm3, 4 |
psllw mm3, 4 |
|
|
|
261 |
movq mm2, [intra_matrix + 8*ecx] |
movq mm2, [intra_matrix + 8*ecx] |
262 |
psrlw mm2, 1 |
psrlw mm2, 1 |
263 |
paddw mm0, mm2 |
paddw mm0, mm2 |
|
|
|
264 |
movq mm2, [intra_matrix_fix + ecx*8] |
movq mm2, [intra_matrix_fix + ecx*8] |
265 |
pmulhw mm0, mm2 ; (level<<4 + intra_matrix[i]>>1) / intra_matrix[i] |
pmulhw mm0, mm2 ; (level<<4 + intra_matrix[i]>>1) / intra_matrix[i] |
|
|
|
266 |
movq mm2, [intra_matrix + 8*ecx + 8] |
movq mm2, [intra_matrix + 8*ecx + 8] |
267 |
psrlw mm2, 1 |
psrlw mm2, 1 |
268 |
paddw mm3, mm2 |
paddw mm3, mm2 |
|
|
|
269 |
movq mm2, [intra_matrix_fix + ecx*8 + 8] |
movq mm2, [intra_matrix_fix + ecx*8 + 8] |
270 |
pmulhw mm3, mm2 |
pmulhw mm3, mm2 |
|
|
|
271 |
paddw mm0, mm5 |
paddw mm0, mm5 |
272 |
paddw mm3, mm5 |
paddw mm3, mm5 |
|
|
|
273 |
psrlw mm0, 1 ; mm0 >>= 1 (/2) |
psrlw mm0, 1 ; mm0 >>= 1 (/2) |
274 |
psrlw mm3, 1 ; |
psrlw mm3, 1 ; |
|
|
|
275 |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
276 |
pxor mm3, mm4 ; |
pxor mm3, mm4 ; |
277 |
psubw mm0, mm1 ; undisplace |
psubw mm0, mm1 ; undisplace |
278 |
psubw mm3, mm4 ; |
psubw mm3, mm4 ; |
|
|
|
279 |
movq [edi + 8*ecx], mm0 |
movq [edi + 8*ecx], mm0 |
280 |
movq [edi + 8*ecx + 8], mm3 |
movq [edi + 8*ecx + 8], mm3 |
281 |
|
|
285 |
jmp near .done |
jmp near .done |
286 |
|
|
287 |
|
|
288 |
align ALIGN |
ALIGN 16 |
289 |
.q2loop |
.q2loop |
290 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
291 |
movq mm3, [esi + 8*ecx + 8] ; |
movq mm3, [esi + 8*ecx + 8] ; |
|
|
|
292 |
pxor mm1, mm1 ; mm1 = 0 |
pxor mm1, mm1 ; mm1 = 0 |
293 |
pxor mm4, mm4 ; |
pxor mm4, mm4 ; |
|
|
|
294 |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
295 |
pcmpgtw mm4, mm3 ; |
pcmpgtw mm4, mm3 ; |
|
|
|
296 |
pxor mm0, mm1 ; mm0 = |mm0| |
pxor mm0, mm1 ; mm0 = |mm0| |
297 |
pxor mm3, mm4 ; |
pxor mm3, mm4 ; |
298 |
psubw mm0, mm1 ; displace |
psubw mm0, mm1 ; displace |
299 |
psubw mm3, mm4 ; |
psubw mm3, mm4 ; |
|
|
|
300 |
psllw mm0, 4 |
psllw mm0, 4 |
301 |
psllw mm3, 4 |
psllw mm3, 4 |
|
|
|
302 |
movq mm2, [intra_matrix + 8*ecx] |
movq mm2, [intra_matrix + 8*ecx] |
303 |
psrlw mm2, 1 |
psrlw mm2, 1 |
304 |
paddw mm0, mm2 |
paddw mm0, mm2 |
|
|
|
305 |
movq mm2, [intra_matrix_fix + ecx*8] |
movq mm2, [intra_matrix_fix + ecx*8] |
306 |
pmulhw mm0, mm2 ; (level<<4 + intra_matrix[i]>>1) / intra_matrix[i] |
pmulhw mm0, mm2 ; (level<<4 + intra_matrix[i]>>1) / intra_matrix[i] |
|
|
|
307 |
movq mm2, [intra_matrix + 8*ecx + 8] |
movq mm2, [intra_matrix + 8*ecx + 8] |
308 |
psrlw mm2, 1 |
psrlw mm2, 1 |
309 |
paddw mm3, mm2 |
paddw mm3, mm2 |
|
|
|
310 |
movq mm2, [intra_matrix_fix + ecx*8 + 8] |
movq mm2, [intra_matrix_fix + ecx*8 + 8] |
311 |
pmulhw mm3, mm2 |
pmulhw mm3, mm2 |
|
|
|
312 |
paddw mm0, mm5 |
paddw mm0, mm5 |
313 |
paddw mm3, mm5 |
paddw mm3, mm5 |
|
|
|
314 |
psrlw mm0, 2 ; mm0 >>= 1 (/4) |
psrlw mm0, 2 ; mm0 >>= 1 (/4) |
315 |
psrlw mm3, 2 ; |
psrlw mm3, 2 ; |
|
|
|
316 |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
317 |
pxor mm3, mm4 ; |
pxor mm3, mm4 ; |
318 |
psubw mm0, mm1 ; undisplace |
psubw mm0, mm1 ; undisplace |
319 |
psubw mm3, mm4 ; |
psubw mm3, mm4 ; |
|
|
|
320 |
movq [edi + 8*ecx], mm0 |
movq [edi + 8*ecx], mm0 |
321 |
movq [edi + 8*ecx + 8], mm3 |
movq [edi + 8*ecx + 8], mm3 |
322 |
|
|
326 |
jmp near .done |
jmp near .done |
327 |
|
|
328 |
|
|
329 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
330 |
; |
; |
331 |
; uint32_t quant_mpeg_inter_mmx(int16_t * coeff, |
; uint32_t quant_mpeg_inter_mmx(int16_t * coeff, |
332 |
; const int16_t const * data, |
; const int16_t const * data, |
333 |
; const uint32_t quant); |
; const uint32_t quant); |
334 |
; |
; |
335 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
336 |
|
|
337 |
align ALIGN |
ALIGN 16 |
|
cglobal quant_mpeg_inter_mmx |
|
338 |
quant_mpeg_inter_mmx: |
quant_mpeg_inter_mmx: |
339 |
|
|
340 |
push ecx |
push ecx |
357 |
|
|
358 |
movq mm7, [mmx_div + eax * 8 - 8] ; divider |
movq mm7, [mmx_div + eax * 8 - 8] ; divider |
359 |
|
|
360 |
align ALIGN |
ALIGN 16 |
361 |
.loop |
.loop |
362 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
363 |
movq mm3, [esi + 8*ecx + 8] ; |
movq mm3, [esi + 8*ecx + 8] ; |
369 |
pxor mm3, mm4 ; |
pxor mm3, mm4 ; |
370 |
psubw mm0, mm1 ; displace |
psubw mm0, mm1 ; displace |
371 |
psubw mm3, mm4 ; |
psubw mm3, mm4 ; |
|
|
|
372 |
psllw mm0, 4 |
psllw mm0, 4 |
373 |
psllw mm3, 4 |
psllw mm3, 4 |
|
|
|
374 |
movq mm2, [inter_matrix + 8*ecx] |
movq mm2, [inter_matrix + 8*ecx] |
375 |
psrlw mm2, 1 |
psrlw mm2, 1 |
376 |
paddw mm0, mm2 |
paddw mm0, mm2 |
|
|
|
377 |
movq mm2, [inter_matrix_fix + ecx*8] |
movq mm2, [inter_matrix_fix + ecx*8] |
378 |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
|
|
|
379 |
movq mm2, [inter_matrix + 8*ecx + 8] |
movq mm2, [inter_matrix + 8*ecx + 8] |
380 |
psrlw mm2, 1 |
psrlw mm2, 1 |
381 |
paddw mm3, mm2 |
paddw mm3, mm2 |
|
|
|
382 |
movq mm2, [inter_matrix_fix + ecx*8 + 8] |
movq mm2, [inter_matrix_fix + ecx*8 + 8] |
383 |
pmulhw mm3, mm2 |
pmulhw mm3, mm2 |
|
|
|
384 |
pmulhw mm0, mm7 ; mm0 = (mm0 / 2Q) >> 16 |
pmulhw mm0, mm7 ; mm0 = (mm0 / 2Q) >> 16 |
385 |
pmulhw mm3, mm7 ; |
pmulhw mm3, mm7 ; |
386 |
psrlw mm0, 1 ; additional shift by 1 => 16 + 1 = 17 |
psrlw mm0, 1 ; additional shift by 1 => 16 + 1 = 17 |
387 |
psrlw mm3, 1 |
psrlw mm3, 1 |
|
|
|
388 |
paddw mm5, mm0 ; sum += mm0 |
paddw mm5, mm0 ; sum += mm0 |
389 |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
390 |
paddw mm5, mm3 ; |
paddw mm5, mm3 ; |
411 |
|
|
412 |
ret |
ret |
413 |
|
|
414 |
align ALIGN |
ALIGN 16 |
415 |
.q1loop |
.q1loop |
416 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
417 |
movq mm3, [esi + 8*ecx+ 8] |
movq mm3, [esi + 8*ecx+ 8] |
|
; |
|
418 |
pxor mm1, mm1 ; mm1 = 0 |
pxor mm1, mm1 ; mm1 = 0 |
419 |
pxor mm4, mm4 ; |
pxor mm4, mm4 ; |
|
|
|
420 |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
421 |
pcmpgtw mm4, mm3 ; |
pcmpgtw mm4, mm3 ; |
|
|
|
422 |
pxor mm0, mm1 ; mm0 = |mm0| |
pxor mm0, mm1 ; mm0 = |mm0| |
423 |
pxor mm3, mm4 ; |
pxor mm3, mm4 ; |
424 |
psubw mm0, mm1 ; displace |
psubw mm0, mm1 ; displace |
425 |
psubw mm3, mm4 ; |
psubw mm3, mm4 ; |
|
|
|
426 |
psllw mm0, 4 |
psllw mm0, 4 |
427 |
psllw mm3, 4 |
psllw mm3, 4 |
|
|
|
428 |
movq mm2, [inter_matrix + 8*ecx] |
movq mm2, [inter_matrix + 8*ecx] |
429 |
psrlw mm2, 1 |
psrlw mm2, 1 |
430 |
paddw mm0, mm2 |
paddw mm0, mm2 |
|
|
|
431 |
movq mm2, [inter_matrix_fix + ecx*8] |
movq mm2, [inter_matrix_fix + ecx*8] |
432 |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
|
|
|
433 |
movq mm2, [inter_matrix + 8*ecx + 8] |
movq mm2, [inter_matrix + 8*ecx + 8] |
434 |
psrlw mm2, 1 |
psrlw mm2, 1 |
435 |
paddw mm3, mm2 |
paddw mm3, mm2 |
|
|
|
436 |
movq mm2, [inter_matrix_fix + ecx*8 + 8] |
movq mm2, [inter_matrix_fix + ecx*8 + 8] |
437 |
pmulhw mm3, mm2 |
pmulhw mm3, mm2 |
|
|
|
438 |
psrlw mm0, 1 ; mm0 >>= 1 (/2) |
psrlw mm0, 1 ; mm0 >>= 1 (/2) |
439 |
psrlw mm3, 1 ; |
psrlw mm3, 1 ; |
|
|
|
440 |
paddw mm5, mm0 ; sum += mm0 |
paddw mm5, mm0 ; sum += mm0 |
441 |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
442 |
paddw mm5, mm3 ; |
paddw mm5, mm3 ; |
443 |
pxor mm3, mm4 ; |
pxor mm3, mm4 ; |
444 |
psubw mm0, mm1 ; undisplace |
psubw mm0, mm1 ; undisplace |
445 |
psubw mm3, mm4 |
psubw mm3, mm4 |
|
|
|
446 |
movq [edi + 8*ecx], mm0 |
movq [edi + 8*ecx], mm0 |
447 |
movq [edi + 8*ecx + 8], mm3 |
movq [edi + 8*ecx + 8], mm3 |
448 |
|
|
453 |
jmp .done |
jmp .done |
454 |
|
|
455 |
|
|
456 |
align ALIGN |
ALIGN 16 |
457 |
.q2loop |
.q2loop |
458 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
459 |
movq mm3, [esi + 8*ecx+ 8] |
movq mm3, [esi + 8*ecx+ 8] |
|
; |
|
460 |
pxor mm1, mm1 ; mm1 = 0 |
pxor mm1, mm1 ; mm1 = 0 |
461 |
pxor mm4, mm4 ; |
pxor mm4, mm4 ; |
|
|
|
462 |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
463 |
pcmpgtw mm4, mm3 ; |
pcmpgtw mm4, mm3 ; |
|
|
|
464 |
pxor mm0, mm1 ; mm0 = |mm0| |
pxor mm0, mm1 ; mm0 = |mm0| |
465 |
pxor mm3, mm4 ; |
pxor mm3, mm4 ; |
466 |
psubw mm0, mm1 ; displace |
psubw mm0, mm1 ; displace |
467 |
psubw mm3, mm4 ; |
psubw mm3, mm4 ; |
|
|
|
468 |
psllw mm0, 4 |
psllw mm0, 4 |
469 |
psllw mm3, 4 |
psllw mm3, 4 |
|
|
|
470 |
movq mm2, [inter_matrix + 8*ecx] |
movq mm2, [inter_matrix + 8*ecx] |
471 |
psrlw mm2, 1 |
psrlw mm2, 1 |
472 |
paddw mm0, mm2 |
paddw mm0, mm2 |
|
|
|
473 |
movq mm2, [inter_matrix_fix + ecx*8] |
movq mm2, [inter_matrix_fix + ecx*8] |
474 |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
|
|
|
475 |
movq mm2, [inter_matrix + 8*ecx + 8] |
movq mm2, [inter_matrix + 8*ecx + 8] |
476 |
psrlw mm2, 1 |
psrlw mm2, 1 |
477 |
paddw mm3, mm2 |
paddw mm3, mm2 |
|
|
|
478 |
movq mm2, [inter_matrix_fix + ecx*8 + 8] |
movq mm2, [inter_matrix_fix + ecx*8 + 8] |
479 |
pmulhw mm3, mm2 |
pmulhw mm3, mm2 |
|
|
|
480 |
psrlw mm0, 2 ; mm0 >>= 1 (/2) |
psrlw mm0, 2 ; mm0 >>= 1 (/2) |
481 |
psrlw mm3, 2 ; |
psrlw mm3, 2 ; |
|
|
|
482 |
paddw mm5, mm0 ; sum += mm0 |
paddw mm5, mm0 ; sum += mm0 |
483 |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
484 |
paddw mm5, mm3 ; |
paddw mm5, mm3 ; |
485 |
pxor mm3, mm4 ; |
pxor mm3, mm4 ; |
486 |
psubw mm0, mm1 ; undisplace |
psubw mm0, mm1 ; undisplace |
487 |
psubw mm3, mm4 |
psubw mm3, mm4 |
|
|
|
488 |
movq [edi + 8*ecx], mm0 |
movq [edi + 8*ecx], mm0 |
489 |
movq [edi + 8*ecx + 8], mm3 |
movq [edi + 8*ecx + 8], mm3 |
490 |
|
|
495 |
jmp .done |
jmp .done |
496 |
|
|
497 |
|
|
498 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
499 |
; |
; |
500 |
; uint32_t dequant_mpeg_intra_mmx(int16_t *data, |
; uint32_t dequant_mpeg_intra_mmx(int16_t *data, |
501 |
; const int16_t const *coeff, |
; const int16_t const *coeff, |
502 |
; const uint32_t quant, |
; const uint32_t quant, |
503 |
; const uint32_t dcscalar); |
; const uint32_t dcscalar); |
504 |
; |
; |
505 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
506 |
|
|
507 |
; Note: in order to saturate 'easily', we pre-shift the quantifier |
; Note: in order to saturate 'easily', we pre-shift the quantifier |
508 |
; by 4. Then, the high-word of (coeff[]*matrix[i]*quant) are used to |
; by 4. Then, the high-word of (coeff[]*matrix[i]*quant) are used to |
543 |
|
|
544 |
;******************************************************************** |
;******************************************************************** |
545 |
|
|
546 |
align 16 |
ALIGN 16 |
|
cglobal dequant_mpeg_intra_mmx |
|
547 |
dequant_mpeg_intra_mmx: |
dequant_mpeg_intra_mmx: |
548 |
|
|
549 |
mov edx, [esp+4] ; data |
mov edx, [esp+4] ; data |
551 |
mov eax, [esp+12] ; quant |
mov eax, [esp+12] ; quant |
552 |
|
|
553 |
movq mm7, [mmx_mul_quant + eax*8 - 8] |
movq mm7, [mmx_mul_quant + eax*8 - 8] |
554 |
mov eax, -16 ; to keep aligned, we regularly process coeff[0] |
mov eax, -16 ; to keep ALIGNed, we regularly process coeff[0] |
555 |
psllw mm7, 2 ; << 2. See comment. |
psllw mm7, 2 ; << 2. See comment. |
556 |
pxor mm6, mm6 ; this is a NOP |
pxor mm6, mm6 ; this is a NOP |
557 |
|
|
558 |
align 16 |
ALIGN 16 |
559 |
.loop |
.loop |
560 |
movq mm0, [ecx+8*eax + 8*16] ; mm0 = c = coeff[i] |
movq mm0, [ecx+8*eax + 8*16] ; mm0 = c = coeff[i] |
561 |
movq mm3, [ecx+8*eax + 8*16 +8]; mm3 = c' = coeff[i+1] |
movq mm3, [ecx+8*eax + 8*16 +8]; mm3 = c' = coeff[i+1] |
612 |
jnz near .loop |
jnz near .loop |
613 |
|
|
614 |
; deal with DC |
; deal with DC |
|
|
|
615 |
movd mm0, [ecx] |
movd mm0, [ecx] |
616 |
pmullw mm0, [esp+16] ; dcscalar |
pmullw mm0, [esp+16] ; dcscalar |
617 |
movq mm2, [mmx_32767_minus_2047] |
movq mm2, [mmx_32767_minus_2047] |
626 |
xor eax, eax |
xor eax, eax |
627 |
ret |
ret |
628 |
|
|
629 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
630 |
; |
; |
631 |
; uint32_t dequant_mpeg_inter_mmx(int16_t * data, |
; uint32_t dequant_mpeg_inter_mmx(int16_t * data, |
632 |
; const int16_t * const coeff, |
; const int16_t * const coeff, |
633 |
; const uint32_t quant); |
; const uint32_t quant); |
634 |
; |
; |
635 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
636 |
|
|
637 |
; Note: We use (2*c + sgn(c) - sgn(-c)) as multiplier |
; Note: We use (2*c + sgn(c) - sgn(-c)) as multiplier |
638 |
; so we handle the 3 cases: c<0, c==0, and c>0 in one shot. |
; so we handle the 3 cases: c<0, c==0, and c>0 in one shot. |
639 |
; sgn(x) is the result of 'pcmpgtw 0,x': 0 if x>=0, -1 if x<0. |
; sgn(x) is the result of 'pcmpgtw 0,x': 0 if x>=0, -1 if x<0. |
640 |
; It's mixed with the extraction of the absolute value. |
; It's mixed with the extraction of the absolute value. |
641 |
|
|
642 |
align 16 |
ALIGN 16 |
|
cglobal dequant_mpeg_inter_mmx |
|
643 |
dequant_mpeg_inter_mmx: |
dequant_mpeg_inter_mmx: |
644 |
|
|
645 |
mov edx, [esp+ 4] ; data |
mov edx, [esp+ 4] ; data |
650 |
paddw mm7, mm7 ; << 1 |
paddw mm7, mm7 ; << 1 |
651 |
pxor mm6, mm6 ; mismatch sum |
pxor mm6, mm6 ; mismatch sum |
652 |
|
|
653 |
align 16 |
ALIGN 16 |
654 |
.loop |
.loop |
655 |
movq mm0, [ecx+8*eax + 8*16 ] ; mm0 = coeff[i] |
movq mm0, [ecx+8*eax + 8*16 ] ; mm0 = coeff[i] |
656 |
movq mm2, [ecx+8*eax + 8*16 +8] ; mm2 = coeff[i+1] |
movq mm2, [ecx+8*eax + 8*16 +8] ; mm2 = coeff[i+1] |