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 1.4 2004-08-22 11:46:10 edgomez Exp $ |
; * $Id: quantize_mpeg_mmx.asm,v 1.10 2008-11-11 20:46:24 Isibaar Exp $ |
25 |
; * |
; * |
26 |
; *************************************************************************/ |
; *************************************************************************/ |
27 |
|
|
32 |
%macro cglobal 1 |
%macro cglobal 1 |
33 |
%ifdef PREFIX |
%ifdef PREFIX |
34 |
%ifdef MARK_FUNCS |
%ifdef MARK_FUNCS |
35 |
global _%1:function |
global _%1:function %1.endfunc-%1 |
36 |
%define %1 _%1:function |
%define %1 _%1:function %1.endfunc-%1 |
37 |
|
%define ENDFUNC .endfunc |
38 |
%else |
%else |
39 |
global _%1 |
global _%1 |
40 |
%define %1 _%1 |
%define %1 _%1 |
41 |
|
%define ENDFUNC |
42 |
%endif |
%endif |
43 |
%else |
%else |
44 |
%ifdef MARK_FUNCS |
%ifdef MARK_FUNCS |
45 |
global %1:function |
global %1:function %1.endfunc-%1 |
46 |
|
%define ENDFUNC .endfunc |
47 |
%else |
%else |
48 |
global %1 |
global %1 |
49 |
|
%define ENDFUNC |
50 |
%endif |
%endif |
51 |
%endif |
%endif |
52 |
%endmacro |
%endmacro |
133 |
times 4 dw 0 |
times 4 dw 0 |
134 |
|
|
135 |
;============================================================================= |
;============================================================================= |
136 |
|
; rounding |
137 |
|
;============================================================================= |
138 |
|
|
139 |
|
mmx_rounding: |
140 |
|
dw (1<<13) |
141 |
|
dw 0 |
142 |
|
dw (1<<13) |
143 |
|
dw 0 |
144 |
|
|
145 |
|
;============================================================================= |
146 |
; Code |
; Code |
147 |
;============================================================================= |
;============================================================================= |
148 |
|
|
153 |
cglobal dequant_mpeg_intra_mmx |
cglobal dequant_mpeg_intra_mmx |
154 |
cglobal dequant_mpeg_inter_mmx |
cglobal dequant_mpeg_inter_mmx |
155 |
|
|
156 |
|
|
157 |
|
%macro QUANT_MMX 1 |
158 |
|
movq mm0, [eax + 16*(%1)] ; data |
159 |
|
movq mm2, [ecx + 16*(%1) + 128] ; intra_matrix_rec |
160 |
|
movq mm4, [eax + 16*(%1) + 8] ; data |
161 |
|
movq mm6, [ecx + 16*(%1) + 128 + 8] ; intra_matrix_rec |
162 |
|
|
163 |
|
movq mm1, mm0 |
164 |
|
movq mm5, mm4 |
165 |
|
|
166 |
|
pmullw mm0, mm2 ; low results |
167 |
|
pmulhw mm1, mm2 ; high results |
168 |
|
pmullw mm4, mm6 ; low results |
169 |
|
pmulhw mm5, mm6 ; high results |
170 |
|
|
171 |
|
movq mm2, mm0 |
172 |
|
movq mm6, mm4 |
173 |
|
|
174 |
|
punpckhwd mm0, mm1 |
175 |
|
punpcklwd mm2, mm1 |
176 |
|
punpckhwd mm4, mm5 |
177 |
|
punpcklwd mm6, mm5 |
178 |
|
|
179 |
|
paddd mm2, mm7 |
180 |
|
paddd mm0, mm7 |
181 |
|
paddd mm6, mm7 |
182 |
|
paddd mm4, mm7 |
183 |
|
|
184 |
|
psrad mm2, 14 |
185 |
|
psrad mm0, 14 |
186 |
|
psrad mm6, 14 |
187 |
|
psrad mm4, 14 |
188 |
|
|
189 |
|
packssdw mm2, mm0 |
190 |
|
packssdw mm6, mm4 |
191 |
|
|
192 |
|
movq [edi + 16*(%1)], mm2 |
193 |
|
movq [edi + 16*(%1)+8], mm6 |
194 |
|
%endmacro |
195 |
|
|
196 |
;----------------------------------------------------------------------------- |
;----------------------------------------------------------------------------- |
197 |
; |
; |
198 |
; uint32_t quant_mpeg_intra_mmx(int16_t * coeff, |
; uint32_t quant_mpeg_intra_mmx(int16_t * coeff, |
206 |
ALIGN 16 |
ALIGN 16 |
207 |
quant_mpeg_intra_mmx: |
quant_mpeg_intra_mmx: |
208 |
|
|
|
push ecx |
|
|
push esi |
|
209 |
push edi |
push edi |
210 |
push ebx |
movq mm7, [mmx_rounding] |
211 |
|
|
212 |
mov edi, [esp + 16 + 4] ; coeff |
mov eax, [esp + 4 + 8] ; data |
213 |
mov esi, [esp + 16 + 8] ; data |
mov ecx, [esp + 4 + 20] ; mpeg_quant_matrices |
214 |
mov eax, [esp + 16 + 12] ; quant |
mov edi, [esp + 4 + 4] ; coeff |
215 |
mov ebx, [esp + 16 + 20] ; mpeg_quant_matrices |
|
216 |
|
QUANT_MMX(0) |
217 |
movq mm5, [quantd + eax * 8 - 8] ; quantd -> mm5 |
QUANT_MMX(1) |
218 |
|
QUANT_MMX(2) |
219 |
xor ecx, ecx |
QUANT_MMX(3) |
220 |
cmp al, 1 |
QUANT_MMX(4) |
221 |
jz near .q1loop |
QUANT_MMX(5) |
222 |
|
QUANT_MMX(6) |
223 |
cmp al, 2 |
QUANT_MMX(7) |
224 |
jz near .q2loop |
|
225 |
|
; calculate DC |
226 |
movq mm7, [mmx_div + eax * 8 - 8] ; multipliers[quant] -> mm7 |
movsx eax, word [eax] ; data[0] |
227 |
|
mov ecx, [esp + 4 + 16] ; dcscalar |
228 |
|
mov edx, eax |
229 |
|
mov edi, ecx |
230 |
|
shr ecx, 1 ; ecx = dcscalar/2 |
231 |
|
sar edx, 31 ; edx = sign extend of eax (ready for division too) |
232 |
|
xor ecx, edx ; adjust ecx according to the sign of data[0] |
233 |
|
sub ecx, edx |
234 |
|
add eax, ecx |
235 |
|
|
236 |
|
mov ecx, [esp + 4 + 4] ; coeff again |
237 |
|
idiv edi ; eax = edx:eax / dcscalar |
238 |
|
mov [ecx], ax ; coeff[0] = ax |
239 |
|
|
|
ALIGN 16 |
|
|
.loop |
|
|
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
|
|
movq mm3, [esi + 8*ecx + 8] ; |
|
|
pxor mm1, mm1 ; mm1 = 0 |
|
|
pxor mm4, mm4 |
|
|
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
|
|
pcmpgtw mm4, mm3 |
|
|
pxor mm0, mm1 ; mm0 = |mm0| |
|
|
pxor mm3, mm4 ; |
|
|
psubw mm0, mm1 ; displace |
|
|
psubw mm3, mm4 ; |
|
|
psllw mm0, 4 ; level << 4 |
|
|
psllw mm3, 4 |
|
|
movq mm2, [ebx + 8*ecx] |
|
|
psrlw mm2, 1 ; intra_matrix[i]>>1 |
|
|
paddw mm0, mm2 |
|
|
movq mm2, [ebx + 256 + ecx*8] |
|
|
pmulhw mm0, mm2 ; (level<<4 + intra_matrix[i]>>1) / intra_matrix[i] |
|
|
movq mm2, [ebx + 8*ecx + 8] |
|
|
psrlw mm2, 1 |
|
|
paddw mm3, mm2 |
|
|
movq mm2, [ebx + 256 + ecx*8 + 8] |
|
|
pmulhw mm3, mm2 |
|
|
paddw mm0, mm5 ; + quantd |
|
|
paddw mm3, mm5 |
|
|
pmulhw mm0, mm7 ; mm0 = (mm0 / 2Q) >> 16 |
|
|
pmulhw mm3, mm7 ; |
|
|
psrlw mm0, 1 ; additional shift by 1 => 16 + 1 = 17 |
|
|
psrlw mm3, 1 |
|
|
pxor mm0, mm1 ; mm0 *= sign(mm0) |
|
|
pxor mm3, mm4 ; |
|
|
psubw mm0, mm1 ; undisplace |
|
|
psubw mm3, mm4 ; |
|
|
|
|
|
movq [edi + 8*ecx], mm0 |
|
|
movq [edi + 8*ecx + 8], mm3 |
|
|
|
|
|
add ecx,2 |
|
|
cmp ecx,16 |
|
|
jnz near .loop |
|
|
|
|
|
.done |
|
|
; caclulate data[0] // (int32_t)dcscalar) |
|
|
mov ecx, [esp + 16 + 16] ; dcscalar |
|
|
mov edx, ecx |
|
|
movsx eax, word [esi] ; data[0] |
|
|
shr edx, 1 ; edx = dcscalar /2 |
|
|
cmp eax, 0 |
|
|
jg .gtzero |
|
|
|
|
|
sub eax, edx |
|
|
jmp short .mul |
|
|
.gtzero |
|
|
add eax, edx |
|
|
.mul |
|
|
cdq ; expand eax -> edx:eax |
|
|
idiv ecx ; eax = edx:eax / dcscalar |
|
|
|
|
|
mov [edi], ax ; coeff[0] = ax |
|
|
|
|
|
pop ebx |
|
240 |
pop edi |
pop edi |
|
pop esi |
|
|
pop ecx |
|
241 |
|
|
242 |
xor eax, eax ; return(0); |
xor eax, eax ; return(0); |
243 |
ret |
ret |
244 |
|
ENDFUNC |
|
ALIGN 16 |
|
|
.q1loop |
|
|
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
|
|
movq mm3, [esi + 8*ecx + 8] ; |
|
|
pxor mm1, mm1 ; mm1 = 0 |
|
|
pxor mm4, mm4 ; |
|
|
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
|
|
pcmpgtw mm4, mm3 ; |
|
|
pxor mm0, mm1 ; mm0 = |mm0| |
|
|
pxor mm3, mm4 ; |
|
|
psubw mm0, mm1 ; displace |
|
|
psubw mm3, mm4 ; |
|
|
psllw mm0, 4 |
|
|
psllw mm3, 4 |
|
|
movq mm2, [ebx + 8*ecx] |
|
|
psrlw mm2, 1 |
|
|
paddw mm0, mm2 |
|
|
movq mm2, [ebx + 256 + ecx*8] |
|
|
pmulhw mm0, mm2 ; (level<<4 + intra_matrix[i]>>1) / intra_matrix[i] |
|
|
movq mm2, [ebx + 8*ecx + 8] |
|
|
psrlw mm2, 1 |
|
|
paddw mm3, mm2 |
|
|
movq mm2, [ebx + 256 + ecx*8 + 8] |
|
|
pmulhw mm3, mm2 |
|
|
paddw mm0, mm5 |
|
|
paddw mm3, mm5 |
|
|
psrlw mm0, 1 ; mm0 >>= 1 (/2) |
|
|
psrlw mm3, 1 ; |
|
|
pxor mm0, mm1 ; mm0 *= sign(mm0) |
|
|
pxor mm3, mm4 ; |
|
|
psubw mm0, mm1 ; undisplace |
|
|
psubw mm3, mm4 ; |
|
|
movq [edi + 8*ecx], mm0 |
|
|
movq [edi + 8*ecx + 8], mm3 |
|
|
|
|
|
add ecx, 2 |
|
|
cmp ecx, 16 |
|
|
jnz near .q1loop |
|
|
jmp near .done |
|
|
|
|
|
|
|
|
ALIGN 16 |
|
|
.q2loop |
|
|
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
|
|
movq mm3, [esi + 8*ecx + 8] ; |
|
|
pxor mm1, mm1 ; mm1 = 0 |
|
|
pxor mm4, mm4 ; |
|
|
pcmpgtw mm1, mm0 ; mm1 = (0 > mm0) |
|
|
pcmpgtw mm4, mm3 ; |
|
|
pxor mm0, mm1 ; mm0 = |mm0| |
|
|
pxor mm3, mm4 ; |
|
|
psubw mm0, mm1 ; displace |
|
|
psubw mm3, mm4 ; |
|
|
psllw mm0, 4 |
|
|
psllw mm3, 4 |
|
|
movq mm2, [ebx + 8*ecx] |
|
|
psrlw mm2, 1 |
|
|
paddw mm0, mm2 |
|
|
movq mm2, [ebx + 256 + ecx*8] |
|
|
pmulhw mm0, mm2 ; (level<<4 + intra_matrix[i]>>1) / intra_matrix[i] |
|
|
movq mm2, [ebx + 8*ecx + 8] |
|
|
psrlw mm2, 1 |
|
|
paddw mm3, mm2 |
|
|
movq mm2, [ebx + 256 + ecx*8 + 8] |
|
|
pmulhw mm3, mm2 |
|
|
paddw mm0, mm5 |
|
|
paddw mm3, mm5 |
|
|
psrlw mm0, 2 ; mm0 >>= 1 (/4) |
|
|
psrlw mm3, 2 ; |
|
|
pxor mm0, mm1 ; mm0 *= sign(mm0) |
|
|
pxor mm3, mm4 ; |
|
|
psubw mm0, mm1 ; undisplace |
|
|
psubw mm3, mm4 ; |
|
|
movq [edi + 8*ecx], mm0 |
|
|
movq [edi + 8*ecx + 8], mm3 |
|
|
|
|
|
add ecx,2 |
|
|
cmp ecx,16 |
|
|
jnz near .q2loop |
|
|
jmp near .done |
|
245 |
|
|
246 |
|
|
247 |
;----------------------------------------------------------------------------- |
;----------------------------------------------------------------------------- |
279 |
movq mm7, [mmx_div + eax * 8 - 8] ; divider |
movq mm7, [mmx_div + eax * 8 - 8] ; divider |
280 |
|
|
281 |
ALIGN 16 |
ALIGN 16 |
282 |
.loop |
.loop: |
283 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
284 |
movq mm3, [esi + 8*ecx + 8] ; |
movq mm3, [esi + 8*ecx + 8] ; |
285 |
pxor mm1, mm1 ; mm1 = 0 |
pxor mm1, mm1 ; mm1 = 0 |
319 |
cmp ecx, 16 |
cmp ecx, 16 |
320 |
jnz near .loop |
jnz near .loop |
321 |
|
|
322 |
.done |
.done: |
323 |
pmaddwd mm5, [mmx_one] |
pmaddwd mm5, [mmx_one] |
324 |
movq mm0, mm5 |
movq mm0, mm5 |
325 |
psrlq mm5, 32 |
psrlq mm5, 32 |
334 |
ret |
ret |
335 |
|
|
336 |
ALIGN 16 |
ALIGN 16 |
337 |
.q1loop |
.q1loop: |
338 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
339 |
movq mm3, [esi + 8*ecx+ 8] |
movq mm3, [esi + 8*ecx+ 8] |
340 |
pxor mm1, mm1 ; mm1 = 0 |
pxor mm1, mm1 ; mm1 = 0 |
374 |
|
|
375 |
jmp .done |
jmp .done |
376 |
|
|
|
|
|
377 |
ALIGN 16 |
ALIGN 16 |
378 |
.q2loop |
.q2loop: |
379 |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
movq mm0, [esi + 8*ecx] ; mm0 = [1st] |
380 |
movq mm3, [esi + 8*ecx+ 8] |
movq mm3, [esi + 8*ecx+ 8] |
381 |
pxor mm1, mm1 ; mm1 = 0 |
pxor mm1, mm1 ; mm1 = 0 |
414 |
jnz near .q2loop |
jnz near .q2loop |
415 |
|
|
416 |
jmp .done |
jmp .done |
417 |
|
ENDFUNC |
418 |
|
|
419 |
|
|
420 |
;----------------------------------------------------------------------------- |
;----------------------------------------------------------------------------- |
482 |
pxor mm6, mm6 ; this is a NOP |
pxor mm6, mm6 ; this is a NOP |
483 |
|
|
484 |
ALIGN 16 |
ALIGN 16 |
485 |
.loop |
.loop: |
486 |
movq mm0, [ecx+8*eax + 8*16] ; mm0 = c = coeff[i] |
movq mm0, [ecx+8*eax + 8*16] ; mm0 = c = coeff[i] |
487 |
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] |
488 |
pxor mm1, mm1 |
pxor mm1, mm1 |
554 |
pop ebx |
pop ebx |
555 |
|
|
556 |
ret |
ret |
557 |
|
ENDFUNC |
558 |
|
|
559 |
;----------------------------------------------------------------------------- |
;----------------------------------------------------------------------------- |
560 |
; |
; |
586 |
pxor mm6, mm6 ; mismatch sum |
pxor mm6, mm6 ; mismatch sum |
587 |
|
|
588 |
ALIGN 16 |
ALIGN 16 |
589 |
.loop |
.loop: |
590 |
movq mm0, [ecx+8*eax + 8*16 ] ; mm0 = coeff[i] |
movq mm0, [ecx+8*eax + 8*16 ] ; mm0 = coeff[i] |
591 |
movq mm2, [ecx+8*eax + 8*16 +8] ; mm2 = coeff[i+1] |
movq mm2, [ecx+8*eax + 8*16 +8] ; mm2 = coeff[i+1] |
592 |
add eax, 2 |
add eax, 2 |
668 |
pop ebx |
pop ebx |
669 |
|
|
670 |
ret |
ret |
671 |
|
ENDFUNC |
672 |
|
|
673 |
|
|
674 |
|
%ifidn __OUTPUT_FORMAT__,elf |
675 |
|
section ".note.GNU-stack" noalloc noexec nowrite progbits |
676 |
|
%endif |
677 |
|
|