21 |
; * along with this program ; if not, write to the Free Software |
; * along with this program ; if not, write to the Free Software |
22 |
; * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
; * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
23 |
; * |
; * |
24 |
; * $Id: quantize_mpeg_mmx.asm,v 1.1.2.2 2003-10-09 18:50:22 edgomez Exp $ |
; * $Id: quantize_mpeg_mmx.asm,v 1.1.2.3 2003-10-28 22:23:03 edgomez Exp $ |
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 |
SECTION .rodata |
|
SECTION .data data |
|
|
%else |
|
|
SECTION .data data align=8 |
|
|
%endif |
|
55 |
|
|
56 |
mmx_one: |
mmx_one: |
57 |
times 4 dw 1 |
times 4 dw 1 |
58 |
|
|
59 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
|
; |
|
60 |
; divide by 2Q table |
; divide by 2Q table |
61 |
; |
;----------------------------------------------------------------------------- |
|
;=========================================================================== |
|
62 |
|
|
63 |
align 16 |
ALIGN 16 |
64 |
mmx_div: |
mmx_div: |
65 |
%assign quant 1 |
times 4 dw 65535 ; the div by 2 formula will overflow for the case |
66 |
%rep 31 |
; quant=1 but we don't care much because quant=1 |
67 |
|
; is handled by a different piece of code that |
68 |
|
; doesn't use this table. |
69 |
|
%assign quant 2 |
70 |
|
%rep 30 |
71 |
times 4 dw (1<<17) / (quant*2) + 1 |
times 4 dw (1<<17) / (quant*2) + 1 |
72 |
%assign quant quant+1 |
%assign quant quant+1 |
73 |
%endrep |
%endrep |
74 |
|
|
75 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
|
; |
|
76 |
; intra matrix |
; intra matrix |
77 |
; |
;----------------------------------------------------------------------------- |
|
;=========================================================================== |
|
78 |
|
|
79 |
cextern intra_matrix |
cextern intra_matrix |
80 |
cextern intra_matrix_fix |
cextern intra_matrix_fix |
81 |
|
|
82 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
|
; |
|
83 |
; inter matrix |
; inter matrix |
84 |
; |
;----------------------------------------------------------------------------- |
|
;=========================================================================== |
|
85 |
|
|
86 |
cextern inter_matrix |
cextern inter_matrix |
87 |
cextern inter_matrix_fix |
cextern inter_matrix_fix |
91 |
%define VM18Q 4 |
%define VM18Q 4 |
92 |
|
|
93 |
|
|
94 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
|
; |
|
95 |
; quantd table |
; quantd table |
96 |
; |
;----------------------------------------------------------------------------- |
|
;=========================================================================== |
|
97 |
|
|
98 |
quantd: |
quantd: |
99 |
%assign quant 1 |
%assign quant 1 |
102 |
%assign quant quant+1 |
%assign quant quant+1 |
103 |
%endrep |
%endrep |
104 |
|
|
105 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
|
; |
|
106 |
; multiple by 2Q table |
; multiple by 2Q table |
107 |
; |
;----------------------------------------------------------------------------- |
|
;=========================================================================== |
|
108 |
|
|
109 |
mmx_mul_quant: |
mmx_mul_quant: |
110 |
%assign quant 1 |
%assign quant 1 |
113 |
%assign quant quant+1 |
%assign quant quant+1 |
114 |
%endrep |
%endrep |
115 |
|
|
116 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
|
; |
|
117 |
; saturation limits |
; saturation limits |
118 |
; |
;----------------------------------------------------------------------------- |
|
;=========================================================================== |
|
119 |
|
|
120 |
align 16 |
ALIGN 16 |
121 |
|
|
122 |
mmx_32767_minus_2047: |
mmx_32767_minus_2047: |
123 |
times 4 dw (32767-2047) |
times 4 dw (32767-2047) |
130 |
zero: |
zero: |
131 |
times 4 dw 0 |
times 4 dw 0 |
132 |
|
|
133 |
;*************************************************************************** |
;============================================================================= |
134 |
; Local data |
; Code |
135 |
;*************************************************************************** |
;============================================================================= |
136 |
|
|
137 |
section .text |
SECTION .text |
138 |
|
|
139 |
;=========================================================================== |
cglobal quant_mpeg_intra_mmx |
140 |
|
cglobal quant_mpeg_inter_mmx |
141 |
|
cglobal dequant_mpeg_intra_mmx |
142 |
|
cglobal dequant_mpeg_inter_mmx |
143 |
|
|
144 |
|
;----------------------------------------------------------------------------- |
145 |
; |
; |
146 |
; uint32_t quant_mpeg_intra_mmx(int16_t * coeff, |
; uint32_t quant_mpeg_intra_mmx(int16_t * coeff, |
147 |
; const int16_t const * data, |
; const int16_t const * data, |
148 |
; const uint32_t quant, |
; const uint32_t quant, |
149 |
; const uint32_t dcscalar); |
; const uint32_t dcscalar); |
150 |
; |
; |
151 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
152 |
|
|
153 |
align ALIGN |
ALIGN 16 |
|
cglobal quant_mpeg_intra_mmx |
|
154 |
quant_mpeg_intra_mmx: |
quant_mpeg_intra_mmx: |
155 |
|
|
156 |
push ecx |
push ecx |
172 |
|
|
173 |
movq mm7, [mmx_div + eax * 8 - 8] ; multipliers[quant] -> mm7 |
movq mm7, [mmx_div + eax * 8 - 8] ; multipliers[quant] -> mm7 |
174 |
|
|
175 |
align ALIGN |
ALIGN 16 |
176 |
.loop |
.loop |
177 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
178 |
movq mm3, [esi + 8*ecx + 8] ; |
movq mm3, [esi + 8*ecx + 8] ; |
|
|
|
179 |
pxor mm1, mm1 ; mm1 = 0 |
pxor mm1, mm1 ; mm1 = 0 |
180 |
pxor mm4, mm4 |
pxor mm4, mm4 |
|
|
|
181 |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
182 |
pcmpgtw mm4, mm3 |
pcmpgtw mm4, mm3 |
|
|
|
183 |
pxor mm0, mm1 ; mm0 = |mm0| |
pxor mm0, mm1 ; mm0 = |mm0| |
184 |
pxor mm3, mm4 ; |
pxor mm3, mm4 ; |
185 |
psubw mm0, mm1 ; displace |
psubw mm0, mm1 ; displace |
186 |
psubw mm3, mm4 ; |
psubw mm3, mm4 ; |
|
|
|
187 |
psllw mm0, 4 ; level << 4 |
psllw mm0, 4 ; level << 4 |
188 |
psllw mm3, 4 ; |
psllw mm3, 4 |
|
|
|
189 |
movq mm2, [intra_matrix + 8*ecx] |
movq mm2, [intra_matrix + 8*ecx] |
190 |
psrlw mm2, 1 ; intra_matrix[i]>>1 |
psrlw mm2, 1 ; intra_matrix[i]>>1 |
191 |
paddw mm0, mm2 |
paddw mm0, mm2 |
|
|
|
192 |
movq mm2, [intra_matrix_fix + ecx*8] |
movq mm2, [intra_matrix_fix + ecx*8] |
193 |
pmulhw mm0, mm2 ; (level<<4 + intra_matrix[i]>>1) / intra_matrix[i] |
pmulhw mm0, mm2 ; (level<<4 + intra_matrix[i]>>1) / intra_matrix[i] |
|
|
|
194 |
movq mm2, [intra_matrix + 8*ecx + 8] |
movq mm2, [intra_matrix + 8*ecx + 8] |
195 |
psrlw mm2, 1 |
psrlw mm2, 1 |
196 |
paddw mm3, mm2 |
paddw mm3, mm2 |
|
|
|
197 |
movq mm2, [intra_matrix_fix + ecx*8 + 8] |
movq mm2, [intra_matrix_fix + ecx*8 + 8] |
198 |
pmulhw mm3, mm2 |
pmulhw mm3, mm2 |
|
|
|
199 |
paddw mm0, mm5 ; + quantd |
paddw mm0, mm5 ; + quantd |
200 |
paddw mm3, mm5 |
paddw mm3, mm5 |
201 |
pmulhw mm0, mm7 ; mm0 = (mm0 / 2Q) >> 16 |
pmulhw mm0, mm7 ; mm0 = (mm0 / 2Q) >> 16 |
202 |
pmulhw mm3, mm7 ; |
pmulhw mm3, mm7 ; |
203 |
psrlw mm0, 1 ; additional shift by 1 => 16 + 1 = 17 |
psrlw mm0, 1 ; additional shift by 1 => 16 + 1 = 17 |
204 |
psrlw mm3, 1 |
psrlw mm3, 1 |
|
|
|
205 |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
206 |
pxor mm3, mm4 ; |
pxor mm3, mm4 ; |
207 |
psubw mm0, mm1 ; undisplace |
psubw mm0, mm1 ; undisplace |
216 |
|
|
217 |
.done |
.done |
218 |
; caclulate data[0] // (int32_t)dcscalar) |
; caclulate data[0] // (int32_t)dcscalar) |
|
|
|
219 |
mov ecx, [esp + 12 + 16] ; dcscalar |
mov ecx, [esp + 12 + 16] ; dcscalar |
220 |
mov edx, ecx |
mov edx, ecx |
221 |
movsx eax, word [esi] ; data[0] |
movsx eax, word [esi] ; data[0] |
240 |
xor eax, eax ; return(0); |
xor eax, eax ; return(0); |
241 |
ret |
ret |
242 |
|
|
243 |
align ALIGN |
ALIGN 16 |
244 |
.q1loop |
.q1loop |
245 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
246 |
movq mm3, [esi + 8*ecx + 8] ; |
movq mm3, [esi + 8*ecx + 8] ; |
|
|
|
247 |
pxor mm1, mm1 ; mm1 = 0 |
pxor mm1, mm1 ; mm1 = 0 |
248 |
pxor mm4, mm4 ; |
pxor mm4, mm4 ; |
|
|
|
249 |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
250 |
pcmpgtw mm4, mm3 ; |
pcmpgtw mm4, mm3 ; |
|
|
|
251 |
pxor mm0, mm1 ; mm0 = |mm0| |
pxor mm0, mm1 ; mm0 = |mm0| |
252 |
pxor mm3, mm4 ; |
pxor mm3, mm4 ; |
253 |
psubw mm0, mm1 ; displace |
psubw mm0, mm1 ; displace |
254 |
psubw mm3, mm4 ; |
psubw mm3, mm4 ; |
|
|
|
255 |
psllw mm0, 4 |
psllw mm0, 4 |
256 |
psllw mm3, 4 |
psllw mm3, 4 |
|
|
|
257 |
movq mm2, [intra_matrix + 8*ecx] |
movq mm2, [intra_matrix + 8*ecx] |
258 |
psrlw mm2, 1 |
psrlw mm2, 1 |
259 |
paddw mm0, mm2 |
paddw mm0, mm2 |
|
|
|
260 |
movq mm2, [intra_matrix_fix + ecx*8] |
movq mm2, [intra_matrix_fix + ecx*8] |
261 |
pmulhw mm0, mm2 ; (level<<4 + intra_matrix[i]>>1) / intra_matrix[i] |
pmulhw mm0, mm2 ; (level<<4 + intra_matrix[i]>>1) / intra_matrix[i] |
|
|
|
262 |
movq mm2, [intra_matrix + 8*ecx + 8] |
movq mm2, [intra_matrix + 8*ecx + 8] |
263 |
psrlw mm2, 1 |
psrlw mm2, 1 |
264 |
paddw mm3, mm2 |
paddw mm3, mm2 |
|
|
|
265 |
movq mm2, [intra_matrix_fix + ecx*8 + 8] |
movq mm2, [intra_matrix_fix + ecx*8 + 8] |
266 |
pmulhw mm3, mm2 |
pmulhw mm3, mm2 |
|
|
|
267 |
paddw mm0, mm5 |
paddw mm0, mm5 |
268 |
paddw mm3, mm5 |
paddw mm3, mm5 |
|
|
|
269 |
psrlw mm0, 1 ; mm0 >>= 1 (/2) |
psrlw mm0, 1 ; mm0 >>= 1 (/2) |
270 |
psrlw mm3, 1 ; |
psrlw mm3, 1 ; |
|
|
|
271 |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
272 |
pxor mm3, mm4 ; |
pxor mm3, mm4 ; |
273 |
psubw mm0, mm1 ; undisplace |
psubw mm0, mm1 ; undisplace |
274 |
psubw mm3, mm4 ; |
psubw mm3, mm4 ; |
|
|
|
275 |
movq [edi + 8*ecx], mm0 |
movq [edi + 8*ecx], mm0 |
276 |
movq [edi + 8*ecx + 8], mm3 |
movq [edi + 8*ecx + 8], mm3 |
277 |
|
|
281 |
jmp near .done |
jmp near .done |
282 |
|
|
283 |
|
|
284 |
align ALIGN |
ALIGN 16 |
285 |
.q2loop |
.q2loop |
286 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
287 |
movq mm3, [esi + 8*ecx + 8] ; |
movq mm3, [esi + 8*ecx + 8] ; |
|
|
|
288 |
pxor mm1, mm1 ; mm1 = 0 |
pxor mm1, mm1 ; mm1 = 0 |
289 |
pxor mm4, mm4 ; |
pxor mm4, mm4 ; |
|
|
|
290 |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
291 |
pcmpgtw mm4, mm3 ; |
pcmpgtw mm4, mm3 ; |
|
|
|
292 |
pxor mm0, mm1 ; mm0 = |mm0| |
pxor mm0, mm1 ; mm0 = |mm0| |
293 |
pxor mm3, mm4 ; |
pxor mm3, mm4 ; |
294 |
psubw mm0, mm1 ; displace |
psubw mm0, mm1 ; displace |
295 |
psubw mm3, mm4 ; |
psubw mm3, mm4 ; |
|
|
|
296 |
psllw mm0, 4 |
psllw mm0, 4 |
297 |
psllw mm3, 4 |
psllw mm3, 4 |
|
|
|
298 |
movq mm2, [intra_matrix + 8*ecx] |
movq mm2, [intra_matrix + 8*ecx] |
299 |
psrlw mm2, 1 |
psrlw mm2, 1 |
300 |
paddw mm0, mm2 |
paddw mm0, mm2 |
|
|
|
301 |
movq mm2, [intra_matrix_fix + ecx*8] |
movq mm2, [intra_matrix_fix + ecx*8] |
302 |
pmulhw mm0, mm2 ; (level<<4 + intra_matrix[i]>>1) / intra_matrix[i] |
pmulhw mm0, mm2 ; (level<<4 + intra_matrix[i]>>1) / intra_matrix[i] |
|
|
|
303 |
movq mm2, [intra_matrix + 8*ecx + 8] |
movq mm2, [intra_matrix + 8*ecx + 8] |
304 |
psrlw mm2, 1 |
psrlw mm2, 1 |
305 |
paddw mm3, mm2 |
paddw mm3, mm2 |
|
|
|
306 |
movq mm2, [intra_matrix_fix + ecx*8 + 8] |
movq mm2, [intra_matrix_fix + ecx*8 + 8] |
307 |
pmulhw mm3, mm2 |
pmulhw mm3, mm2 |
|
|
|
308 |
paddw mm0, mm5 |
paddw mm0, mm5 |
309 |
paddw mm3, mm5 |
paddw mm3, mm5 |
|
|
|
310 |
psrlw mm0, 2 ; mm0 >>= 1 (/4) |
psrlw mm0, 2 ; mm0 >>= 1 (/4) |
311 |
psrlw mm3, 2 ; |
psrlw mm3, 2 ; |
|
|
|
312 |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
313 |
pxor mm3, mm4 ; |
pxor mm3, mm4 ; |
314 |
psubw mm0, mm1 ; undisplace |
psubw mm0, mm1 ; undisplace |
315 |
psubw mm3, mm4 ; |
psubw mm3, mm4 ; |
|
|
|
316 |
movq [edi + 8*ecx], mm0 |
movq [edi + 8*ecx], mm0 |
317 |
movq [edi + 8*ecx + 8], mm3 |
movq [edi + 8*ecx + 8], mm3 |
318 |
|
|
322 |
jmp near .done |
jmp near .done |
323 |
|
|
324 |
|
|
325 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
326 |
; |
; |
327 |
; uint32_t quant_mpeg_inter_mmx(int16_t * coeff, |
; uint32_t quant_mpeg_inter_mmx(int16_t * coeff, |
328 |
; const int16_t const * data, |
; const int16_t const * data, |
329 |
; const uint32_t quant); |
; const uint32_t quant); |
330 |
; |
; |
331 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
332 |
|
|
333 |
align ALIGN |
ALIGN 16 |
|
cglobal quant_mpeg_inter_mmx |
|
334 |
quant_mpeg_inter_mmx: |
quant_mpeg_inter_mmx: |
335 |
|
|
336 |
push ecx |
push ecx |
353 |
|
|
354 |
movq mm7, [mmx_div + eax * 8 - 8] ; divider |
movq mm7, [mmx_div + eax * 8 - 8] ; divider |
355 |
|
|
356 |
align ALIGN |
ALIGN 16 |
357 |
.loop |
.loop |
358 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
359 |
movq mm3, [esi + 8*ecx + 8] ; |
movq mm3, [esi + 8*ecx + 8] ; |
365 |
pxor mm3, mm4 ; |
pxor mm3, mm4 ; |
366 |
psubw mm0, mm1 ; displace |
psubw mm0, mm1 ; displace |
367 |
psubw mm3, mm4 ; |
psubw mm3, mm4 ; |
|
|
|
368 |
psllw mm0, 4 |
psllw mm0, 4 |
369 |
psllw mm3, 4 |
psllw mm3, 4 |
|
|
|
370 |
movq mm2, [inter_matrix + 8*ecx] |
movq mm2, [inter_matrix + 8*ecx] |
371 |
psrlw mm2, 1 |
psrlw mm2, 1 |
372 |
paddw mm0, mm2 |
paddw mm0, mm2 |
|
|
|
373 |
movq mm2, [inter_matrix_fix + ecx*8] |
movq mm2, [inter_matrix_fix + ecx*8] |
374 |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
|
|
|
375 |
movq mm2, [inter_matrix + 8*ecx + 8] |
movq mm2, [inter_matrix + 8*ecx + 8] |
376 |
psrlw mm2, 1 |
psrlw mm2, 1 |
377 |
paddw mm3, mm2 |
paddw mm3, mm2 |
|
|
|
378 |
movq mm2, [inter_matrix_fix + ecx*8 + 8] |
movq mm2, [inter_matrix_fix + ecx*8 + 8] |
379 |
pmulhw mm3, mm2 |
pmulhw mm3, mm2 |
|
|
|
380 |
pmulhw mm0, mm7 ; mm0 = (mm0 / 2Q) >> 16 |
pmulhw mm0, mm7 ; mm0 = (mm0 / 2Q) >> 16 |
381 |
pmulhw mm3, mm7 ; |
pmulhw mm3, mm7 ; |
382 |
psrlw mm0, 1 ; additional shift by 1 => 16 + 1 = 17 |
psrlw mm0, 1 ; additional shift by 1 => 16 + 1 = 17 |
383 |
psrlw mm3, 1 |
psrlw mm3, 1 |
|
|
|
384 |
paddw mm5, mm0 ; sum += mm0 |
paddw mm5, mm0 ; sum += mm0 |
385 |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
386 |
paddw mm5, mm3 ; |
paddw mm5, mm3 ; |
407 |
|
|
408 |
ret |
ret |
409 |
|
|
410 |
align ALIGN |
ALIGN 16 |
411 |
.q1loop |
.q1loop |
412 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
413 |
movq mm3, [esi + 8*ecx+ 8] |
movq mm3, [esi + 8*ecx+ 8] |
|
; |
|
414 |
pxor mm1, mm1 ; mm1 = 0 |
pxor mm1, mm1 ; mm1 = 0 |
415 |
pxor mm4, mm4 ; |
pxor mm4, mm4 ; |
|
|
|
416 |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
417 |
pcmpgtw mm4, mm3 ; |
pcmpgtw mm4, mm3 ; |
|
|
|
418 |
pxor mm0, mm1 ; mm0 = |mm0| |
pxor mm0, mm1 ; mm0 = |mm0| |
419 |
pxor mm3, mm4 ; |
pxor mm3, mm4 ; |
420 |
psubw mm0, mm1 ; displace |
psubw mm0, mm1 ; displace |
421 |
psubw mm3, mm4 ; |
psubw mm3, mm4 ; |
|
|
|
422 |
psllw mm0, 4 |
psllw mm0, 4 |
423 |
psllw mm3, 4 |
psllw mm3, 4 |
|
|
|
424 |
movq mm2, [inter_matrix + 8*ecx] |
movq mm2, [inter_matrix + 8*ecx] |
425 |
psrlw mm2, 1 |
psrlw mm2, 1 |
426 |
paddw mm0, mm2 |
paddw mm0, mm2 |
|
|
|
427 |
movq mm2, [inter_matrix_fix + ecx*8] |
movq mm2, [inter_matrix_fix + ecx*8] |
428 |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
|
|
|
429 |
movq mm2, [inter_matrix + 8*ecx + 8] |
movq mm2, [inter_matrix + 8*ecx + 8] |
430 |
psrlw mm2, 1 |
psrlw mm2, 1 |
431 |
paddw mm3, mm2 |
paddw mm3, mm2 |
|
|
|
432 |
movq mm2, [inter_matrix_fix + ecx*8 + 8] |
movq mm2, [inter_matrix_fix + ecx*8 + 8] |
433 |
pmulhw mm3, mm2 |
pmulhw mm3, mm2 |
|
|
|
434 |
psrlw mm0, 1 ; mm0 >>= 1 (/2) |
psrlw mm0, 1 ; mm0 >>= 1 (/2) |
435 |
psrlw mm3, 1 ; |
psrlw mm3, 1 ; |
|
|
|
436 |
paddw mm5, mm0 ; sum += mm0 |
paddw mm5, mm0 ; sum += mm0 |
437 |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
438 |
paddw mm5, mm3 ; |
paddw mm5, mm3 ; |
439 |
pxor mm3, mm4 ; |
pxor mm3, mm4 ; |
440 |
psubw mm0, mm1 ; undisplace |
psubw mm0, mm1 ; undisplace |
441 |
psubw mm3, mm4 |
psubw mm3, mm4 |
|
|
|
442 |
movq [edi + 8*ecx], mm0 |
movq [edi + 8*ecx], mm0 |
443 |
movq [edi + 8*ecx + 8], mm3 |
movq [edi + 8*ecx + 8], mm3 |
444 |
|
|
449 |
jmp .done |
jmp .done |
450 |
|
|
451 |
|
|
452 |
align ALIGN |
ALIGN 16 |
453 |
.q2loop |
.q2loop |
454 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
455 |
movq mm3, [esi + 8*ecx+ 8] |
movq mm3, [esi + 8*ecx+ 8] |
|
; |
|
456 |
pxor mm1, mm1 ; mm1 = 0 |
pxor mm1, mm1 ; mm1 = 0 |
457 |
pxor mm4, mm4 ; |
pxor mm4, mm4 ; |
|
|
|
458 |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
459 |
pcmpgtw mm4, mm3 ; |
pcmpgtw mm4, mm3 ; |
|
|
|
460 |
pxor mm0, mm1 ; mm0 = |mm0| |
pxor mm0, mm1 ; mm0 = |mm0| |
461 |
pxor mm3, mm4 ; |
pxor mm3, mm4 ; |
462 |
psubw mm0, mm1 ; displace |
psubw mm0, mm1 ; displace |
463 |
psubw mm3, mm4 ; |
psubw mm3, mm4 ; |
|
|
|
464 |
psllw mm0, 4 |
psllw mm0, 4 |
465 |
psllw mm3, 4 |
psllw mm3, 4 |
|
|
|
466 |
movq mm2, [inter_matrix + 8*ecx] |
movq mm2, [inter_matrix + 8*ecx] |
467 |
psrlw mm2, 1 |
psrlw mm2, 1 |
468 |
paddw mm0, mm2 |
paddw mm0, mm2 |
|
|
|
469 |
movq mm2, [inter_matrix_fix + ecx*8] |
movq mm2, [inter_matrix_fix + ecx*8] |
470 |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
pmulhw mm0, mm2 ; (level<<4 + inter_matrix[i]>>1) / inter_matrix[i] |
|
|
|
471 |
movq mm2, [inter_matrix + 8*ecx + 8] |
movq mm2, [inter_matrix + 8*ecx + 8] |
472 |
psrlw mm2, 1 |
psrlw mm2, 1 |
473 |
paddw mm3, mm2 |
paddw mm3, mm2 |
|
|
|
474 |
movq mm2, [inter_matrix_fix + ecx*8 + 8] |
movq mm2, [inter_matrix_fix + ecx*8 + 8] |
475 |
pmulhw mm3, mm2 |
pmulhw mm3, mm2 |
|
|
|
476 |
psrlw mm0, 2 ; mm0 >>= 1 (/2) |
psrlw mm0, 2 ; mm0 >>= 1 (/2) |
477 |
psrlw mm3, 2 ; |
psrlw mm3, 2 ; |
|
|
|
478 |
paddw mm5, mm0 ; sum += mm0 |
paddw mm5, mm0 ; sum += mm0 |
479 |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
pxor mm0, mm1 ; mm0 *= sign(mm0) |
480 |
paddw mm5, mm3 ; |
paddw mm5, mm3 ; |
481 |
pxor mm3, mm4 ; |
pxor mm3, mm4 ; |
482 |
psubw mm0, mm1 ; undisplace |
psubw mm0, mm1 ; undisplace |
483 |
psubw mm3, mm4 |
psubw mm3, mm4 |
|
|
|
484 |
movq [edi + 8*ecx], mm0 |
movq [edi + 8*ecx], mm0 |
485 |
movq [edi + 8*ecx + 8], mm3 |
movq [edi + 8*ecx + 8], mm3 |
486 |
|
|
491 |
jmp .done |
jmp .done |
492 |
|
|
493 |
|
|
494 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
495 |
; |
; |
496 |
; uint32_t dequant_mpeg_intra_mmx(int16_t *data, |
; uint32_t dequant_mpeg_intra_mmx(int16_t *data, |
497 |
; const int16_t const *coeff, |
; const int16_t const *coeff, |
498 |
; const uint32_t quant, |
; const uint32_t quant, |
499 |
; const uint32_t dcscalar); |
; const uint32_t dcscalar); |
500 |
; |
; |
501 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
502 |
|
|
503 |
; Note: in order to saturate 'easily', we pre-shift the quantifier |
; Note: in order to saturate 'easily', we pre-shift the quantifier |
504 |
; 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 |
539 |
|
|
540 |
;******************************************************************** |
;******************************************************************** |
541 |
|
|
542 |
align 16 |
ALIGN 16 |
|
cglobal dequant_mpeg_intra_mmx |
|
543 |
dequant_mpeg_intra_mmx: |
dequant_mpeg_intra_mmx: |
544 |
|
|
545 |
mov edx, [esp+4] ; data |
mov edx, [esp+4] ; data |
547 |
mov eax, [esp+12] ; quant |
mov eax, [esp+12] ; quant |
548 |
|
|
549 |
movq mm7, [mmx_mul_quant + eax*8 - 8] |
movq mm7, [mmx_mul_quant + eax*8 - 8] |
550 |
mov eax, -16 ; to keep aligned, we regularly process coeff[0] |
mov eax, -16 ; to keep ALIGNed, we regularly process coeff[0] |
551 |
psllw mm7, 2 ; << 2. See comment. |
psllw mm7, 2 ; << 2. See comment. |
552 |
pxor mm6, mm6 ; this is a NOP |
pxor mm6, mm6 ; this is a NOP |
553 |
|
|
554 |
align 16 |
ALIGN 16 |
555 |
.loop |
.loop |
556 |
movq mm0, [ecx+8*eax + 8*16] ; mm0 = c = coeff[i] |
movq mm0, [ecx+8*eax + 8*16] ; mm0 = c = coeff[i] |
557 |
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] |
608 |
jnz near .loop |
jnz near .loop |
609 |
|
|
610 |
; deal with DC |
; deal with DC |
|
|
|
611 |
movd mm0, [ecx] |
movd mm0, [ecx] |
612 |
pmullw mm0, [esp+16] ; dcscalar |
pmullw mm0, [esp+16] ; dcscalar |
613 |
movq mm2, [mmx_32767_minus_2047] |
movq mm2, [mmx_32767_minus_2047] |
622 |
xor eax, eax |
xor eax, eax |
623 |
ret |
ret |
624 |
|
|
625 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
626 |
; |
; |
627 |
; uint32_t dequant_mpeg_inter_mmx(int16_t * data, |
; uint32_t dequant_mpeg_inter_mmx(int16_t * data, |
628 |
; const int16_t * const coeff, |
; const int16_t * const coeff, |
629 |
; const uint32_t quant); |
; const uint32_t quant); |
630 |
; |
; |
631 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
632 |
|
|
633 |
; Note: We use (2*c + sgn(c) - sgn(-c)) as multiplier |
; Note: We use (2*c + sgn(c) - sgn(-c)) as multiplier |
634 |
; 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. |
635 |
; 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. |
636 |
; It's mixed with the extraction of the absolute value. |
; It's mixed with the extraction of the absolute value. |
637 |
|
|
638 |
align 16 |
ALIGN 16 |
|
cglobal dequant_mpeg_inter_mmx |
|
639 |
dequant_mpeg_inter_mmx: |
dequant_mpeg_inter_mmx: |
640 |
|
|
641 |
mov edx, [esp+ 4] ; data |
mov edx, [esp+ 4] ; data |
646 |
paddw mm7, mm7 ; << 1 |
paddw mm7, mm7 ; << 1 |
647 |
pxor mm6, mm6 ; mismatch sum |
pxor mm6, mm6 ; mismatch sum |
648 |
|
|
649 |
align 16 |
ALIGN 16 |
650 |
.loop |
.loop |
651 |
movq mm0, [ecx+8*eax + 8*16 ] ; mm0 = coeff[i] |
movq mm0, [ecx+8*eax + 8*16 ] ; mm0 = coeff[i] |
652 |
movq mm2, [ecx+8*eax + 8*16 +8] ; mm2 = coeff[i+1] |
movq mm2, [ecx+8*eax + 8*16 +8] ; mm2 = coeff[i+1] |