75
76
STORE_DIFFx2 m2, m3, m4, m5, m7, 6, %1, %3
79
80
; ff_h264_idct_add_mmx(uint8_t *dst, int16_t *block, int stride)
80
cglobal h264_idct_add_8_mmx, 3, 3, 0
81
cglobal h264_idct_add_8, 3, 3, 0
81
82
IDCT4_ADD r0, r1, r2
179
180
STORE_DIFFx2 m1, m2, m5, m6, m7, 6, %1, %3
183
184
; ff_h264_idct8_add_mmx(uint8_t *dst, int16_t *block, int stride)
184
cglobal h264_idct8_add_8_mmx, 3, 4, 0
185
cglobal h264_idct8_add_8, 3, 4, 0
185
186
%assign pad 128+4-(stack_offset&7)
198
199
; %1=uint8_t *dst, %2=int16_t *block, %3=int stride
199
200
%macro IDCT8_ADD_SSE 4
202
203
TRANSPOSE8x8W 0, 1, 2, 3, 4, 5, 6, 7, 8
204
205
TRANSPOSE8x8W 0, 1, 2, 3, 4, 5, 6, 7, [%2], [%2+16]
206
207
paddw m0, [pw_32]
211
212
IDCT8_1D [%2], [%2+ 16]
225
226
STORE_DIFF m1, m6, m7, [%1+%3 ]
226
227
STORE_DIFF m2, m6, m7, [%1+%3*2]
227
228
STORE_DIFF m3, m6, m7, [%1+%4 ]
239
240
STORE_DIFF m1, m6, m7, [%1+%4 ]
243
244
; ff_h264_idct8_add_sse2(uint8_t *dst, int16_t *block, int stride)
244
cglobal h264_idct8_add_8_sse2, 3, 4, 10
245
cglobal h264_idct8_add_8, 3, 4, 10
245
246
IDCT8_ADD_SSE r0, r1, r2, r3
248
%macro DC_ADD_MMX2_INIT 2-3
249
%macro DC_ADD_MMXEXT_INIT 2-3
250
251
movsx %1, word [%1]
288
; ff_h264_idct_dc_add_mmx2(uint8_t *dst, int16_t *block, int stride)
289
cglobal h264_idct_dc_add_8_mmx2, 3, 3, 0
290
DC_ADD_MMX2_INIT r1, r2
291
DC_ADD_MMX2_OP movh, r0, r2, r1
289
; ff_h264_idct_dc_add_mmxext(uint8_t *dst, int16_t *block, int stride)
290
cglobal h264_idct_dc_add_8, 3, 3, 0
291
DC_ADD_MMXEXT_INIT r1, r2
292
DC_ADD_MMXEXT_OP movh, r0, r2, r1
294
; ff_h264_idct8_dc_add_mmx2(uint8_t *dst, int16_t *block, int stride)
295
cglobal h264_idct8_dc_add_8_mmx2, 3, 3, 0
296
DC_ADD_MMX2_INIT r1, r2
297
DC_ADD_MMX2_OP mova, r0, r2, r1
295
; ff_h264_idct8_dc_add_mmxext(uint8_t *dst, int16_t *block, int stride)
296
cglobal h264_idct8_dc_add_8, 3, 3, 0
297
DC_ADD_MMXEXT_INIT r1, r2
298
DC_ADD_MMXEXT_OP mova, r0, r2, r1
298
299
lea r0, [r0+r2*4]
299
DC_ADD_MMX2_OP mova, r0, r2, r1
300
DC_ADD_MMXEXT_OP mova, r0, r2, r1
302
304
; ff_h264_idct_add16_mmx(uint8_t *dst, const int *block_offset,
303
305
; DCTELEM *block, int stride, const uint8_t nnzc[6*8])
304
cglobal h264_idct_add16_8_mmx, 5, 7, 0
306
cglobal h264_idct_add16_8, 5, 7 + npicregs, 0, dst, block_offset, block, stride, nnzc, cntr, coeff, picreg
309
lea picregq, [scan8_mem]
310
312
movzx r6, byte [scan8+r5]
311
313
movzx r6, byte [r4+r6]
324
326
; ff_h264_idct8_add4_mmx(uint8_t *dst, const int *block_offset,
325
327
; DCTELEM *block, int stride, const uint8_t nnzc[6*8])
326
cglobal h264_idct8_add4_8_mmx, 5, 7, 0
328
cglobal h264_idct8_add4_8, 5, 7 + npicregs, 0, dst, block_offset, block, stride, nnzc, cntr, coeff, picreg
327
329
%assign pad 128+4-(stack_offset&7)
334
lea picregq, [scan8_mem]
335
337
movzx r6, byte [scan8+r5]
336
338
movzx r6, byte [r4+r6]
356
; ff_h264_idct_add16_mmx2(uint8_t *dst, const int *block_offset,
357
; DCTELEM *block, int stride, const uint8_t nnzc[6*8])
358
cglobal h264_idct_add16_8_mmx2, 5, 7, 0
359
; ff_h264_idct_add16_mmxext(uint8_t *dst, const int *block_offset,
360
; DCTELEM *block, int stride, const uint8_t nnzc[6*8])
361
cglobal h264_idct_add16_8, 5, 8 + npicregs, 0, dst1, block_offset, block, stride, nnzc, cntr, coeff, dst2, picreg
364
lea picregq, [scan8_mem]
364
367
movzx r6, byte [scan8+r5]
365
368
movzx r6, byte [r4+r6]
370
373
movsx r6, word [r2]
373
DC_ADD_MMX2_INIT r2, r3, r6
376
%define dst_regd r10d
376
DC_ADD_MMXEXT_INIT r2, r3, r6
381
mov dst_regd, dword [r1+r5*4]
382
lea dst_reg, [r0+dst_reg]
383
DC_ADD_MMX2_OP movh, dst_reg, r3, r6
381
mov dst2d, dword [r1+r5*4]
382
lea dst2q, [r0+dst2q]
383
DC_ADD_MMXEXT_OP movh, dst2q, r3, r6
393
393
mov r6d, dword [r1+r5*4]
395
395
IDCT4_ADD r6, r2, r3
403
404
; ff_h264_idct_add16intra_mmx(uint8_t *dst, const int *block_offset,
404
405
; DCTELEM *block, int stride, const uint8_t nnzc[6*8])
405
cglobal h264_idct_add16intra_8_mmx, 5, 7, 0
406
cglobal h264_idct_add16intra_8, 5, 7 + npicregs, 0, dst, block_offset, block, stride, nnzc, cntr, coeff, picreg
409
lea picregq, [scan8_mem]
411
412
movzx r6, byte [scan8+r5]
412
413
movzx r6, byte [r4+r6]
413
414
or r6w, word [r2]
416
417
mov r6d, dword [r1+r5*4]
418
419
IDCT4_ADD r6, r2, r3
426
; ff_h264_idct_add16intra_mmx2(uint8_t *dst, const int *block_offset,
427
; DCTELEM *block, int stride, const uint8_t nnzc[6*8])
428
cglobal h264_idct_add16intra_8_mmx2, 5, 7, 0
428
; ff_h264_idct_add16intra_mmxext(uint8_t *dst, const int *block_offset,
429
; DCTELEM *block, int stride,
430
; const uint8_t nnzc[6*8])
431
cglobal h264_idct_add16intra_8, 5, 8 + npicregs, 0, dst1, block_offset, block, stride, nnzc, cntr, coeff, dst2, picreg
434
lea picregq, [scan8_mem]
434
437
movzx r6, byte [scan8+r5]
435
438
movzx r6, byte [r4+r6]
447
450
movsx r6, word [r2]
450
DC_ADD_MMX2_INIT r2, r3, r6
453
%define dst_regd r10d
453
DC_ADD_MMXEXT_INIT r2, r3, r6
458
mov dst_regd, dword [r1+r5*4]
460
DC_ADD_MMX2_OP movh, dst_reg, r3, r6
458
mov dst2d, dword [r1+r5*4]
460
DC_ADD_MMXEXT_OP movh, dst2q, r3, r6
471
; ff_h264_idct8_add4_mmx2(uint8_t *dst, const int *block_offset,
472
; DCTELEM *block, int stride, const uint8_t nnzc[6*8])
473
cglobal h264_idct8_add4_8_mmx2, 5, 7, 0
471
; ff_h264_idct8_add4_mmxext(uint8_t *dst, const int *block_offset,
472
; DCTELEM *block, int stride,
473
; const uint8_t nnzc[6*8])
474
cglobal h264_idct8_add4_8, 5, 8 + npicregs, 0, dst1, block_offset, block, stride, nnzc, cntr, coeff, dst2, picreg
474
475
%assign pad 128+4-(stack_offset&7)
480
lea picregq, [scan8_mem]
482
483
movzx r6, byte [scan8+r5]
483
484
movzx r6, byte [r4+r6]
488
489
movsx r6, word [r2]
491
DC_ADD_MMX2_INIT r2, r3, r6
494
%define dst_regd r10d
492
DC_ADD_MMXEXT_INIT r2, r3, r6
499
mov dst_regd, dword [r1+r5*4]
500
lea dst_reg, [r0+dst_reg]
501
DC_ADD_MMX2_OP mova, dst_reg, r3, r6
502
lea dst_reg, [dst_reg+r3*4]
503
DC_ADD_MMX2_OP mova, dst_reg, r3, r6
497
mov dst2d, dword [r1+r5*4]
498
lea dst2q, [r0+dst2q]
499
DC_ADD_MMXEXT_OP mova, dst2q, r3, r6
500
lea dst2q, [dst2q+r3*4]
501
DC_ADD_MMXEXT_OP mova, dst2q, r3, r6
534
532
; ff_h264_idct8_add4_sse2(uint8_t *dst, const int *block_offset,
535
533
; DCTELEM *block, int stride, const uint8_t nnzc[6*8])
536
cglobal h264_idct8_add4_8_sse2, 5, 7, 10
534
cglobal h264_idct8_add4_8, 5, 8 + npicregs, 10, dst1, block_offset, block, stride, nnzc, cntr, coeff, dst2, picreg
537
lea picregq, [scan8_mem]
542
540
movzx r6, byte [scan8+r5]
543
541
movzx r6, byte [r4+r6]
548
546
movsx r6, word [r2]
552
DC_ADD_MMX2_INIT r2, r3, r6
555
%define dst_regd r10d
550
DC_ADD_MMXEXT_INIT r2, r3, r6
560
mov dst_regd, dword [r1+r5*4]
562
DC_ADD_MMX2_OP mova, dst_reg, r3, r6
563
lea dst_reg, [dst_reg+r3*4]
564
DC_ADD_MMX2_OP mova, dst_reg, r3, r6
555
mov dst2d, dword [r1+r5*4]
557
DC_ADD_MMXEXT_OP mova, dst2q, r3, r6
558
lea dst2q, [dst2q+r3*4]
559
DC_ADD_MMXEXT_OP mova, dst2q, r3, r6
575
mov dst_regd, dword [r1+r5*4]
577
IDCT8_ADD_SSE dst_reg, r2, r3, r6
570
mov dst2d, dword [r1+r5*4]
572
IDCT8_ADD_SSE dst2q, r2, r3, r6
589
584
h264_idct_add8_mmx_plane:
591
586
movzx r6, byte [scan8+r5]
592
587
movzx r6, byte [r4+r6]
593
588
or r6w, word [r2]
597
592
mov r0d, dword [r1+r5*4]
600
595
mov r0, r1m ; XXX r1m here is actually r0m of the calling func
602
597
add r0, dword [r1+r5*4]
604
599
IDCT4_ADD r0, r2, r3
612
607
; ff_h264_idct_add8_mmx(uint8_t **dest, const int *block_offset,
613
608
; DCTELEM *block, int stride, const uint8_t nnzc[6*8])
614
cglobal h264_idct_add8_8_mmx, 5, 7, 0
609
cglobal h264_idct_add8_8, 5, 8 + npicregs, 0, dst1, block_offset, block, stride, nnzc, cntr, coeff, dst2, picreg
613
lea picregq, [scan8_mem]
623
618
call h264_idct_add8_mmx_plane
629
624
add r0mp, gprsize
631
626
call h264_idct_add8_mmx_plane
634
h264_idct_add8_mmx2_plane
629
h264_idct_add8_mmxext_plane:
636
631
movzx r6, byte [scan8+r5]
637
632
movzx r6, byte [r4+r6]
641
636
mov r0d, dword [r1+r5*4]
644
639
mov r0, r1m ; XXX r1m here is actually r0m of the calling func
655
650
movsx r6, word [r2]
658
DC_ADD_MMX2_INIT r2, r3, r6
653
DC_ADD_MMXEXT_INIT r2, r3, r6
660
655
mov r0d, dword [r1+r5*4]
663
658
mov r0, r1m ; XXX r1m here is actually r0m of the calling func
665
660
add r0, dword [r1+r5*4]
667
DC_ADD_MMX2_OP movh, r0, r3, r6
662
DC_ADD_MMXEXT_OP movh, r0, r3, r6
675
; ff_h264_idct_add8_mmx2(uint8_t **dest, const int *block_offset,
676
; DCTELEM *block, int stride, const uint8_t nnzc[6*8])
677
cglobal h264_idct_add8_8_mmx2, 5, 7, 0
671
; ff_h264_idct_add8_mmxext(uint8_t **dest, const int *block_offset,
672
; DCTELEM *block, int stride, const uint8_t nnzc[6*8])
673
cglobal h264_idct_add8_8, 5, 8 + npicregs, 0, dst1, block_offset, block, stride, nnzc, cntr, coeff, dst2, picreg
680
lea picregq, [scan8_mem]
686
call h264_idct_add8_mmx2_plane
682
call h264_idct_add8_mmxext_plane
692
688
add r0mp, gprsize
694
call h264_idct_add8_mmx2_plane
690
call h264_idct_add8_mmxext_plane
698
693
; r0 = uint8_t *dst, r2 = int16_t *block, r3 = int stride, r6=clobbered
699
h264_idct_dc_add8_mmx2:
694
h264_idct_dc_add8_mmxext:
700
695
movd m0, [r2 ] ; 0 0 X D
701
696
punpcklwd m0, [r2+32] ; x X d D
702
697
paddsw m0, [pw_32]
708
703
pshufw m1, m0, 0xFA ; -d-d-d-d-D-D-D-D
709
704
punpcklwd m0, m0 ; d d d d D D D D
711
DC_ADD_MMX2_OP movq, r0, r3, r6
706
DC_ADD_MMXEXT_OP movq, r0, r3, r6
716
711
; r0 = uint8_t *dst (clobbered), r2 = int16_t *block, r3 = int stride
717
x264_add8x4_idct_sse2:
712
h264_add8x4_idct_sse2:
753
748
; ff_h264_idct_add16_sse2(uint8_t *dst, const int *block_offset,
754
749
; DCTELEM *block, int stride, const uint8_t nnzc[6*8])
755
cglobal h264_idct_add16_8_sse2, 5, 5, 8
750
cglobal h264_idct_add16_8, 5, 5 + ARCH_X86_64, 8
759
754
; unrolling of the loop leads to an average performance gain of
800
795
; ff_h264_idct_add16intra_sse2(uint8_t *dst, const int *block_offset,
801
796
; DCTELEM *block, int stride, const uint8_t nnzc[6*8])
802
cglobal h264_idct_add16intra_8_sse2, 5, 7, 8
797
cglobal h264_idct_add16intra_8, 5, 7 + ARCH_X86_64, 8
806
801
add16intra_sse2_cycle 0, 0xc
807
802
add16intra_sse2_cycle 1, 0x14
817
812
movzx r0, word [r4+%2]
821
816
mov r0d, dword [r1+(%1&1)*8+64*(1+(%1>>1))]
826
821
add r0, dword [r1+(%1&1)*8+64*(1+(%1>>1))]
828
call x264_add8x4_idct_sse2
823
call h264_add8x4_idct_sse2
831
826
movsx r0, word [r2 ]
832
827
or r0w, word [r2+32]
835
830
mov r0d, dword [r1+(%1&1)*8+64*(1+(%1>>1))]
840
835
add r0, dword [r1+(%1&1)*8+64*(1+(%1>>1))]
842
call h264_idct_dc_add8_mmx2
837
call h264_idct_dc_add8_mmxext
851
846
; ff_h264_idct_add8_sse2(uint8_t **dest, const int *block_offset,
852
847
; DCTELEM *block, int stride, const uint8_t nnzc[6*8])
853
cglobal h264_idct_add8_8_sse2, 5, 7, 8
848
cglobal h264_idct_add8_8, 5, 7 + ARCH_X86_64, 8
858
853
add8_sse2_cycle 0, 0x34
859
854
add8_sse2_cycle 1, 0x3c
863
858
add r0mp, gprsize
899
%macro STORE_WORDS_MMX 5
911
%macro DEQUANT_STORE_MMX 1
912
DEQUANT_MMX m0, m1, %1
913
STORE_WORDS_MMX m0, 0, 1, 4, 5
914
STORE_WORDS_MMX m1, 2, 3, 6, 7
916
DEQUANT_MMX m2, m3, %1
917
STORE_WORDS_MMX m2, 8, 9, 12, 13
918
STORE_WORDS_MMX m3, 10, 11, 14, 15
921
%macro STORE_WORDS_SSE 9
894
%macro STORE_WORDS 5-9
963
949
packssdw xmm0, xmm1
964
950
packssdw xmm2, xmm3
965
STORE_WORDS_SSE xmm0, 0, 1, 4, 5, 2, 3, 6, 7
966
STORE_WORDS_SSE xmm2, 8, 9, 12, 13, 10, 11, 14, 15
951
STORE_WORDS xmm0, 0, 1, 4, 5, 2, 3, 6, 7
952
STORE_WORDS xmm2, 8, 9, 12, 13, 10, 11, 14, 15
954
DEQUANT_MMX m0, m1, %1
955
STORE_WORDS m0, 0, 1, 4, 5
956
STORE_WORDS m1, 2, 3, 6, 7
958
DEQUANT_MMX m2, m3, %1
959
STORE_WORDS m2, 8, 9, 12, 13
960
STORE_WORDS m3, 10, 11, 14, 15
969
%macro IDCT_DC_DEQUANT 2
970
cglobal h264_luma_dc_dequant_idct_%1, 3,4,%2
964
%macro IDCT_DC_DEQUANT 1
965
cglobal h264_luma_dc_dequant_idct, 3, 4, %1
966
; manually spill XMM registers for Win64 because
967
; the code here is initialized with INIT_MMX