forked from mikerabat/mrmath
-
Notifications
You must be signed in to change notification settings - Fork 1
/
AVXMatrixRotations.pas
685 lines (501 loc) · 23 KB
/
AVXMatrixRotations.pas
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
// ###################################################################
// #### This file is part of the mathematics library project, and is
// #### offered under the licence agreement described on
// #### http://www.mrsoft.org/
// ####
// #### Copyright:(c) 2018, Michael R. . All rights reserved.
// ####
// #### Unless required by applicable law or agreed to in writing, software
// #### distributed under the License is distributed on an "AS IS" BASIS,
// #### WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// #### See the License for the specific language governing permissions and
// #### limitations under the License.
// ###################################################################
// Vector/Matrix rotation routines mainly used for the SVD
unit AVXMatrixRotations;
interface
{$IFDEF CPUX64}
{$DEFINE x64}
{$ENDIF}
{$IFDEF cpux86_64}
{$DEFINE x64}
{$ENDIF}
{$IFNDEF x64}
uses MatrixConst;
procedure AVXApplyPlaneRotSeqRVB(width, height : TASMNativeInt; A : PDouble; const LineWidthA : TASMNativeInt; C, S : PConstDoubleArr); {$IFDEF FPC} assembler; {$ELSE} register; {$ENDIF}
procedure AVXApplyPlaneRotSeqRVF(width, height : TASMNativeInt; A : PDouble; const LineWidthA : TASMNativeInt; C, S : PConstDoubleArr); {$IFDEF FPC} assembler; {$ELSE} register; {$ENDIF}
procedure AVXApplyPlaneRotSeqLVF(width, height : TASMNativeInt; A : PDouble; const LineWidthA : TASMNativeInt; C, S : PConstDoubleArr); {$IFDEF FPC} assembler; {$ELSE} register; {$ENDIF}
procedure AVXApplyPlaneRotSeqLVB(width, height : TASMNativeInt; A : PDouble; const LineWidthA : TASMNativeInt; C, S : PConstDoubleArr); {$IFDEF FPC} assembler; {$ELSE} register; {$ENDIF}
procedure AVXMatrixRotate(N : TASMNativeInt; X : PDouble; const LineWidthDX : TASMNativeInt; Y : PDouble; LineWidthDY : TASMNativeInt; const c, s : double);
{$ENDIF}
implementation
{$IFNDEF x64}
{$IFDEF FPC} {$ASMMODE intel} {$S-} {$ENDIF}
const cLocMinusOne : double = -1;
cLocOne : double = 1;
cLocMulM1Bits : Array[0..1] of Int64 = ($8000000000000000, $0);
procedure AVXApplyPlaneRotSeqLVB(width, height : TASMNativeInt; A : PDouble; const LineWidthA : TASMNativeInt; C, S : PConstDoubleArr);
var y2 : TASMNativeInt;
iter : TASMNativeInt;
// eax = width, edx = height, ecx = A
asm
push ebx;
push edi;
push esi;
// store "is odd" flag
mov esi, eax;
and esi, 1;
// if (height < 2) or (width < 1) then exit;
cmp eax, 1;
jl @@endproc;
cmp edx, 2;
jl @@endproc;
// y2 := height - 1;
dec edx;
mov y2, edx;
// iter := -(width and $FFFFFFFE)*sizeof(double);
and eax, $FFFFFFFE;
imul eax, -8;
mov iter, eax;
mov eax, c; // point to y (aka the end)
mov ebx, s;
mov edi, y2;
dec edi;
shl edi, 3; // y2*sizeof(double)
add eax, edi;
add ebx, edi;
//mov ecx, A; // A[y + 1][x]
mov edi, LineWidthA;
imul edi, y2;
add ecx, edi;
sub ecx, iter;
lea edx, cLocOne;
{$IFDEF FPC}vmovsd xmm7, [edx];{$ELSE}db $C5,$FB,$10,$3A;{$ENDIF}
mov edx, ecx; // A[y][x]
sub edx, LineWidthA;
{$IFDEF FPC}vxorpd xmm6, xmm6, xmm6; {$ELSE}db $C5,$C9,$57,$F6;{$ENDIF} // compare reference
@@foryloop:
{$IFDEF FPC}vmovddup xmm0, [eax]; {$ELSE}db $C5,$FB,$12,$00;{$ENDIF} // c[y]
{$IFDEF FPC}vmovddup xmm1, [ebx]; {$ELSE}db $C5,$FB,$12,$0B;{$ENDIF} // s[y]
// ###########################################
// #### if (ctemp <> 1) or (stemp <> 0) then
{$IFDEF FPC}vcomisd xmm0, xmm7; {$ELSE}db $C5,$F9,$2F,$C7;{$ENDIF} // = 1
jne @@beginXLoop;
{$IFDEF FPC}vcomisd xmm1, xmm6; {$ELSE}db $C5,$F9,$2F,$CE;{$ENDIF} // = 0
jne @@beginXLoop;
jmp @@nextLine; // c=1 and stemp=0 next line -> the statement
// ###########################################
// #### for x := 0 to width - 1 do
@@beginXLoop:
// init
mov edi, iter;
test edi, edi;
jz @@LastElem;
@@forxloop:
//temp := pcAy1^[x];
// pcAy1^[x] := cTemp*temp - stemp*pcAy^[x];
// pcAy1^[x + 1] := cTemp*temp1 - stemp*pcAy1[x + 1];
// evaluate 2 values
{$IFDEF FPC}vmovupd xmm2, [edx + edi];{$ELSE}db $C5,$F9,$10,$14,$3A;{$ENDIF}
{$IFDEF FPC}vmovupd xmm3, [ecx + edi];{$ELSE}db $C5,$F9,$10,$1C,$39;{$ENDIF}
// temp store...
{$IFDEF FPC}vmovapd xmm4, xmm2{$ELSE}db $C5,$F9,$28,$E2;{$ENDIF}
{$IFDEF FPC}vmovapd xmm5, xmm3;{$ELSE}db $C5,$F9,$28,$EB;{$ENDIF}
{$IFDEF FPC}vmulpd xmm3, xmm3, xmm0; {$ELSE}db $C5,$E1,$59,$D8;{$ENDIF} // ctemp*pcay1^[x] and ctemp*a[x+1]
{$IFDEF FPC}vmulpd xmm2, xmm2, xmm1; {$ELSE}db $C5,$E9,$59,$D1;{$ENDIF} // stemp*pcAy^[x] and stemp*a[x+1]
{$IFDEF FPC}vsubpd xmm3, xmm3, xmm2;{$ELSE}db $C5,$E1,$5C,$DA;{$ENDIF}
// pcAy^[x] := stemp*temp + ctemp*pcAy^[x];
// pcAy^[x + 1] := stemp*temp1 + ctemp*pcAy^[x + 1]
{$IFDEF FPC}vmulpd xmm4, xmm4, xmm0;{$ELSE}db $C5,$D9,$59,$E0;{$ENDIF}
{$IFDEF FPC}vmulpd xmm5, xmm5, xmm1;{$ELSE}db $C5,$D1,$59,$E9;{$ENDIF}
{$IFDEF FPC}vaddpd xmm5, xmm5, xmm4;{$ELSE}db $C5,$D1,$58,$EC;{$ENDIF}
// write back...
{$IFDEF FPC}vmovupd [edx + edi], xmm5;{$ELSE}db $C5,$F9,$11,$2C,$3A;{$ENDIF}
{$IFDEF FPC}vmovupd [ecx + edi], xmm3;{$ELSE}db $C5,$F9,$11,$1C,$39;{$ENDIF}
add edi, 16;
jnz @@forxloop;
@@LastElem:
// ###########################################
// #### Last element handling
cmp esi, 1;
jne @@nextLine;
// same as above but with single elements
{$IFDEF FPC}vmovsd xmm2, [edx];{$ELSE}db $C5,$FB,$10,$12;{$ENDIF}
{$IFDEF FPC}vmovsd xmm3, [ecx];{$ELSE}db $C5,$FB,$10,$19;{$ENDIF}
{$IFDEF FPC}vmovapd xmm4, xmm2;{$ELSE}db $C5,$F9,$28,$E2;{$ENDIF}
{$IFDEF FPC}vmovapd xmm5, xmm3;{$ELSE}db $C5,$F9,$28,$EB;{$ENDIF}
{$IFDEF FPC}vmulsd xmm3, xmm3, xmm0;{$ELSE}db $C5,$E3,$59,$D8;{$ENDIF}
{$IFDEF FPC}vmulsd xmm2, xmm2, xmm1;{$ELSE}db $C5,$EB,$59,$D1;{$ENDIF}
{$IFDEF FPC}vsubsd xmm3, xmm3, xmm2;{$ELSE}db $C5,$E3,$5C,$DA;{$ENDIF}
{$IFDEF FPC}vmulsd xmm4, xmm4, xmm0;{$ELSE}db $C5,$DB,$59,$E0;{$ENDIF}
{$IFDEF FPC}vmulsd xmm5, xmm5, xmm1;{$ELSE}db $C5,$D3,$59,$E9;{$ENDIF}
{$IFDEF FPC}vaddsd xmm5, xmm5, xmm4;{$ELSE}db $C5,$D3,$58,$EC;{$ENDIF}
{$IFDEF FPC}vmovsd [edx], xmm5;{$ELSE}db $C5,$FB,$11,$2A;{$ENDIF}
{$IFDEF FPC}vmovsd [ecx], xmm3;{$ELSE}db $C5,$FB,$11,$19;{$ENDIF}
// ###########################################
// #### next y
@@nextLine:
sub ebx, 8; // sizeof(double)
sub eax, 8;
sub edx, LineWidthA;
sub ecx, LineWidthA;
dec y2;
jnz @@foryloop;
// epilog
@@endproc:
{$IFDEF FPC}vzeroupper;{$ELSE}db $C5,$F8,$77;{$ENDIF}
pop esi;
pop edi;
pop ebx;
end;
procedure AVXApplyPlaneRotSeqLVF(width, height : TASMNativeInt; A : PDouble; const LineWidthA : TASMNativeInt; C, S : PConstDoubleArr);
var y2 : TASMNativeInt;
iter : TASMNativeInt;
asm
push ebx;
push edi;
push esi;
// if (height < 2) or (width < 1) then
// exit;
mov esi, eax;
and esi, 1;
cmp eax, 1;
jl @@endproc;
cmp edx, 2;
jl @@endproc;
//y2 := height - 1;
//iter := -(width and $FFFFFFFE)*sizeof(double);
dec edx;
mov y2, edx;
and eax, $FFFFFFFE;
imul eax, -8;
mov iter, eax;
mov eax, c;
mov ebx, s;
sub ecx, iter;
lea edx, cLocOne;
{$IFDEF FPC}vmovsd xmm7, [edx];{$ELSE}db $C5,$FB,$10,$3A;{$ENDIF}
mov edx, ecx; // A[y+1][x]
add edx, LineWidthA;
@@foryloop:
{$IFDEF FPC}vmovddup xmm0, [eax]; {$ELSE}db $C5,$FB,$12,$00;{$ENDIF} // c[y]
{$IFDEF FPC}vmovddup xmm1, [ebx]; {$ELSE}db $C5,$FB,$12,$0B;{$ENDIF} // s[y]
// ###########################################
// #### if (ctemp <> 1) or (stemp <> 0) then
{$IFDEF FPC}vcomisd xmm0, xmm7; {$ELSE}db $C5,$F9,$2F,$C7;{$ENDIF} // = 1
jne @@beginXLoop;
{$IFDEF FPC}vcomisd xmm1, xmm6; {$ELSE}db $C5,$F9,$2F,$CE;{$ENDIF} // = 0
jne @@beginXLoop;
jmp @@nextLine; // c=1 and stemp=0 next line -> the statement
// ###########################################
// #### for x := 0 to width - 1 do
@@beginXLoop:
// init
mov edi, iter;
test edi, edi;
jz @@LastElem;
@@forxloop:
//temp := pcAy1^[x];
// pcAy1^[x] := cTemp*temp - stemp*pcAy^[x];
// pcAy1^[x + 1] := cTemp*temp1 - stemp*pcAy1[x + 1];
// evaluate 2 values
{$IFDEF FPC}vmovupd xmm2, [ecx + edi];{$ELSE}db $C5,$F9,$10,$14,$39;{$ENDIF}
{$IFDEF FPC}vmovupd xmm3, [edx + edi];{$ELSE}db $C5,$F9,$10,$1C,$3A;{$ENDIF}
{$IFDEF FPC}vmulpd xmm5, xmm3, xmm0; {$ELSE}db $C5,$E1,$59,$E8;{$ENDIF} // ctemp*pcay1^[x] and ctemp*a[x+1]
{$IFDEF FPC}vmulpd xmm4, xmm2, xmm1; {$ELSE}db $C5,$E9,$59,$E1;{$ENDIF} // stemp*pcAy^[x] and stemp*a[x+1]
//subpd xmm3, xmm2;
{$IFDEF FPC}vsubpd xmm5, xmm5, xmm4;{$ELSE}db $C5,$D1,$5C,$EC;{$ENDIF}
// pcAy^[x] := stemp*temp + ctemp*pcAy^[x];
// pcAy^[x + 1] := stemp*temp1 + ctemp*pcAy^[x + 1]
{$IFDEF FPC}vmulpd xmm2, xmm2, xmm0;{$ELSE}db $C5,$E9,$59,$D0;{$ENDIF}
{$IFDEF FPC}vmulpd xmm3, xmm3, xmm1;{$ELSE}db $C5,$E1,$59,$D9;{$ENDIF}
//addpd xmm5, xmm4;
{$IFDEF FPC}vaddpd xmm3, xmm3, xmm2;{$ELSE}db $C5,$E1,$58,$DA;{$ENDIF}
// write back...
{$IFDEF FPC}vmovupd [ecx + edi], xmm3;{$ELSE}db $C5,$F9,$11,$1C,$39;{$ENDIF}
{$IFDEF FPC}vmovupd [edx + edi], xmm5;{$ELSE}db $C5,$F9,$11,$2C,$3A;{$ENDIF}
add edi, 16;
jnz @@forxloop;
@@LastElem:
// ###########################################
// #### Last element handling
cmp esi, 1;
jne @@nextLine;
// same as above but with single elements
{$IFDEF FPC}vmovsd xmm4, [ecx];{$ELSE}db $C5,$FB,$10,$21;{$ENDIF}
{$IFDEF FPC}vmovsd xmm5, [edx];{$ELSE}db $C5,$FB,$10,$2A;{$ENDIF}
{$IFDEF FPC}vmulsd xmm3, xmm5, xmm0;{$ELSE}db $C5,$D3,$59,$D8;{$ENDIF}
{$IFDEF FPC}vmulsd xmm2, xmm4, xmm1;{$ELSE}db $C5,$DB,$59,$D1;{$ENDIF}
{$IFDEF FPC}vsubsd xmm3, xmm3, xmm2;{$ELSE}db $C5,$E3,$5C,$DA;{$ENDIF}
{$IFDEF FPC}vmulsd xmm4, xmm4, xmm0;{$ELSE}db $C5,$DB,$59,$E0;{$ENDIF}
{$IFDEF FPC}vmulsd xmm5, xmm5, xmm1;{$ELSE}db $C5,$D3,$59,$E9;{$ENDIF}
{$IFDEF FPC}vaddsd xmm5, xmm5, xmm4;{$ELSE}db $C5,$D3,$58,$EC;{$ENDIF}
{$IFDEF FPC}vmovsd [ecx], xmm5;{$ELSE}db $C5,$FB,$11,$29;{$ENDIF}
{$IFDEF FPC}vmovsd [edx], xmm3;{$ELSE}db $C5,$FB,$11,$1A;{$ENDIF}
// ###########################################
// #### next y
@@nextLine:
add ebx, 8; // sizeof(double)
add eax, 8;
add ecx, LineWidthA;
add edx, LineWidthA;
dec y2;
jnz @@foryloop;
@@endproc:
{$IFDEF FPC}vzeroupper;{$ELSE}db $C5,$F8,$77;{$ENDIF}
pop esi;
pop edi;
pop ebx;
end;
procedure AVXApplyPlaneRotSeqRVB(width, height : TASMNativeInt; A : PDouble; const LineWidthA : TASMNativeInt; C, S : PConstDoubleArr);
// eax = width, edx = height, ecx = A
asm
// if (height <= 0) or (width <= 1) then exit;
cmp eax, 1;
jle @@endProc;
cmp edx, 0;
jle @@endProc;
push ebx;
push edi;
push esi;
// w2 := width - 1;
// iter := w2*sizeof(double);
dec eax;
imul eax, 8;
mov esi, c;
mov ebx, s;
lea edi, cLocMulM1Bits;
{$IFDEF FPC}vmovupd xmm7, [edi];{$ELSE}db $C5,$F9,$10,$3F;{$ENDIF}
@@foryloop:
mov edi, eax; //iter;
{$IFDEF FPC}vmovhpd xmm2, xmm2, [ecx + edi];{$ELSE}db $C5,$E9,$16,$14,$39;{$ENDIF}
// for x := width - 2 downto 0
@@forxloop:
{$IFDEF FPC}vmovsd xmm4, [esi + edi - 8]; {$ELSE}db $C5,$FB,$10,$64,$3E,$F8;{$ENDIF} // store c
{$IFDEF FPC}vmovsd xmm3, [ebx + edi - 8]; {$ELSE}db $C5,$FB,$10,$5C,$3B,$F8;{$ENDIF} // store s
{$IFDEF FPC}vmovlpd xmm2, xmm2, [ecx + edi - 8]; {$ELSE}db $C5,$E9,$12,$54,$39,$F8;{$ENDIF} // a[x], a[x+1]
// handle x, x+1
// ####################################
// #### x, x+ 1
{$IFDEF FPC}vmovlhps xmm3, xmm3, xmm4;{$ELSE}db $C5,$E0,$16,$DC;{$ENDIF}
{$IFDEF FPC}vmovlhps xmm4, xmm4, xmm3;{$ELSE}db $C5,$D8,$16,$E3;{$ENDIF}
{$IFDEF FPC}vxorpd xmm3, xmm3, xmm7; {$ELSE}db $C5,$E1,$57,$DF;{$ENDIF} // -s, c
{$IFDEF FPC}vmulpd xmm3, xmm3, xmm2; {$ELSE}db $C5,$E1,$59,$DA;{$ENDIF} // a[x+1)*c[x] - s[x]*a[x]
{$IFDEF FPC}vhaddpd xmm3, xmm3, xmm3;{$ELSE}db $C5,$E1,$7C,$DB;{$ENDIF}
{$IFDEF FPC}vmulpd xmm4, xmm4, xmm2; {$ELSE}db $C5,$D9,$59,$E2;{$ENDIF} // a[x+1]*s[x] + a[x]*c[x]
{$IFDEF FPC}vhaddpd xmm4, xmm4, xmm4;{$ELSE}db $C5,$D9,$7C,$E4;{$ENDIF}
// write back first two values
{$IFDEF FPC}vmovlhps xmm2, xmm2, xmm4;{$ELSE}db $C5,$E8,$16,$D4;{$ENDIF}
{$IFDEF FPC}vmovsd [ecx + edi], xmm3;{$ELSE}db $C5,$FB,$11,$1C,$39;{$ENDIF}
// next one
sub edi, 8;
jnz @@forxloop;
{$IFDEF FPC}vmovsd [ecx + edi], xmm4;{$ELSE}db $C5,$FB,$11,$24,$39;{$ENDIF}
add ecx, LineWidthA;
dec edx;
jnz @@foryloop;
// epilog
{$IFDEF FPC}vzeroupper;{$ELSE}db $C5,$F8,$77;{$ENDIF}
pop esi;
pop edi;
pop ebx;
@@endProc:
end;
procedure AVXApplyPlaneRotSeqRVF(width, height : TASMNativeInt; A : PDouble; const LineWidthA : TASMNativeInt; C, S : PConstDoubleArr);
// eax = width, edx = height, ecx = A
asm
// if (height <= 0) or (width <= 1) then exit;
cmp eax, 1;
jle @@endProc;
cmp edx, 0;
jle @@endProc;
push ebx;
push edi;
push esi;
// iter
dec eax;
imul eax, -8;
mov esi, c;
mov ebx, s;
sub esi, eax;
sub ebx, eax;
sub ecx, eax;
lea edi, cLocMulM1Bits;
{$IFDEF FPC}vmovupd xmm7, [edi];{$ELSE}db $C5,$F9,$10,$3F;{$ENDIF}
@@foryloop:
mov edi, eax;
movsd xmm2, [ecx + edi];
@@forxloop:
{$IFDEF FPC}vmovsd xmm4, [esi + edi]; {$ELSE}db $C5,$FB,$10,$24,$3E;{$ENDIF} // store c
{$IFDEF FPC}vmovhpd xmm4, xmm4, [ebx + edi]; {$ELSE}db $C5,$D9,$16,$24,$3B;{$ENDIF} // store s
{$IFDEF FPC}vshufpd xmm3, xmm4, xmm4, 1;{$ELSE}db $C5,$D9,$C6,$DC,$01;{$ENDIF}
{$IFDEF FPC}vmovhpd xmm2, xmm2, [ecx + edi + 8]; {$ELSE}db $C5,$E9,$16,$54,$39,$08;{$ENDIF} // a[x], a[x+1]
// handle x, x+1
// ####################################
// #### x, x+ 1
{$IFDEF FPC}vxorpd xmm3, xmm3, xmm7; {$ELSE}db $C5,$E1,$57,$DF;{$ENDIF} // -s, c
{$IFDEF FPC}vmulpd xmm3, xmm3, xmm2; {$ELSE}db $C5,$E1,$59,$DA;{$ENDIF} // a[x+1)*c[x] - s[x]*a[x]
{$IFDEF FPC}vhaddpd xmm3, xmm3, xmm3;{$ELSE}db $C5,$E1,$7C,$DB;{$ENDIF}
{$IFDEF FPC}vmulpd xmm4, xmm4, xmm2; {$ELSE}db $C5,$D9,$59,$E2;{$ENDIF} // a[x+1]*s[x] + a[x]*c[x]
{$IFDEF FPC}vhaddpd xmm4, xmm4, xmm4;{$ELSE}db $C5,$D9,$7C,$E4;{$ENDIF}
// write back first two values
{$IFDEF FPC}vmovsd xmm2, xmm2, xmm3;{$ELSE}db $C5,$EB,$10,$D3;{$ENDIF}
{$IFDEF FPC}vmovsd [ecx + edi], xmm4;{$ELSE}db $C5,$FB,$11,$24,$39;{$ENDIF}
// next one
add edi, 8;
jnz @@forxloop;
{$IFDEF FPC}vmovsd [ecx + edi], xmm2;{$ELSE}db $C5,$FB,$11,$14,$39;{$ENDIF}
add ecx, LineWidthA;
dec edx;
jnz @@foryloop;
// epilog
{$IFDEF FPC}vzeroupper;{$ELSE}db $C5,$F8,$77;{$ENDIF}
pop esi;
pop edi;
pop ebx;
@@endProc:
end;
// its assumed that Linewidthdx and linewidthdy = sizeof(double)
procedure AVXMatrixRotateAligned(N : TASMNativeInt; X : PDouble;
Y : PDouble; c, s : double); {$IFDEF FPC} assembler; {$ELSE} register; {$ENDIF}
// eax = N; edx = X; ecx = Y
asm
push ebx;
push esi;
push edi;
lea esi, s;
{$IFDEF FPC}vmovsd xmm2, [esi];{$ELSE}db $C5,$FB,$10,$16;{$ENDIF}
lea edi, cLocMinusOne;
{$IFDEF FPC}vmulsd xmm2, xmm2, [edi];{$ELSE}db $C5,$EB,$59,$17;{$ENDIF}
{$IFDEF FPC}vmovddup xmm0, xmm2;{$ELSE}db $C5,$FB,$12,$C2;{$ENDIF}
lea edi, c;
{$IFDEF FPC}vmovddup xmm1, [edi];{$ELSE}db $C5,$FB,$12,$0F;{$ENDIF}
{$IFDEF FPC}vmovddup xmm2, [esi];{$ELSE}db $C5,$FB,$12,$16;{$ENDIF}
mov edi, eax;
//mov edx, X;
//mov ecx, Y;
xor esi, esi;
shr eax, 1;
test eax, eax;
jz @@exitLoop;
@@forNloop:
// do a full load -> intermediate store in xmm5, and xmm6
{$IFDEF FPC}vmovupd xmm5, [edx + esi]; {$ELSE}db $C5,$F9,$10,$2C,$32;{$ENDIF} // x, x+1
{$IFDEF FPC}vmovupd xmm6, [ecx + esi]; {$ELSE}db $C5,$F9,$10,$34,$31;{$ENDIF} // y, y+1
{$IFDEF FPC}vmulpd xmm3, xmm5, xmm0; {$ELSE}db $C5,$D1,$59,$D8;{$ENDIF} // x, x+1 * -s
{$IFDEF FPC}vmulpd xmm5, xmm5, xmm1; {$ELSE}db $C5,$D1,$59,$E9;{$ENDIF} // x, x+1 * c
{$IFDEF FPC}vmulpd xmm4, xmm6, xmm1; {$ELSE}db $C5,$C9,$59,$E1;{$ENDIF} // y, y+1 * c
{$IFDEF FPC}vmulpd xmm6, xmm6, xmm2; {$ELSE}db $C5,$C9,$59,$F2;{$ENDIF} // y, y+1 * s
{$IFDEF FPC}vaddpd xmm5, xmm5, xmm6; {$ELSE}db $C5,$D1,$58,$EE;{$ENDIF} // c*x + s*y , c*(x+1) + s*(y+1)
{$IFDEF FPC}vaddpd xmm3, xmm3, xmm4; {$ELSE}db $C5,$E1,$58,$DC;{$ENDIF} // -s*x + c*y, -s(x+1) + c*(y+1)
// write back
{$IFDEF FPC}vmovupd [edx + esi], xmm5;{$ELSE}db $C5,$F9,$11,$2C,$32;{$ENDIF}
{$IFDEF FPC}vmovupd [ecx + esi], xmm3;{$ELSE}db $C5,$F9,$11,$1C,$31;{$ENDIF}
add esi, 16;
dec eax;
jnz @@forNloop;
@@exitLoop:
// test for an odd N
test edi, 1;
jz @@endProc;
// handle last element
{$IFDEF FPC}vmovsd xmm5, [edx + esi];{$ELSE}db $C5,$FB,$10,$2C,$32;{$ENDIF}
{$IFDEF FPC}vmovsd xmm6, [ecx + esi];{$ELSE}db $C5,$FB,$10,$34,$31;{$ENDIF}
//dtemp := c*pX^[i] + s*pY^[i];
//pY^[i] := - s*pX^[i] + c*pY^[i];
//px^[i] := dtemp;
{$IFDEF FPC}vmulsd xmm3, xmm5, xmm0; {$ELSE}db $C5,$D3,$59,$D8;{$ENDIF} // x * -s
{$IFDEF FPC}vmulsd xmm4, xmm6, xmm1; {$ELSE}db $C5,$CB,$59,$E1;{$ENDIF} // y * c
{$IFDEF FPC}vmulsd xmm5, xmm5, xmm1; {$ELSE}db $C5,$D3,$59,$E9;{$ENDIF} // x * c
{$IFDEF FPC}vmulsd xmm6, xmm6, xmm2; {$ELSE}db $C5,$CB,$59,$F2;{$ENDIF} // y * s
{$IFDEF FPC}vaddsd xmm5, xmm5, xmm6; {$ELSE}db $C5,$D3,$58,$EE;{$ENDIF} // c*x + s*y
{$IFDEF FPC}vaddsd xmm3, xmm3, xmm4; {$ELSE}db $C5,$E3,$58,$DC;{$ENDIF} // -s*x + c*y
// write back
{$IFDEF FPC}vmovsd [edx + esi], xmm5;{$ELSE}db $C5,$FB,$11,$2C,$32;{$ENDIF}
{$IFDEF FPC}vmovsd [ecx + esi], xmm3;{$ELSE}db $C5,$FB,$11,$1C,$31;{$ENDIF}
@@endProc:
{$IFDEF FPC}vzeroupper;{$ELSE}db $C5,$F8,$77;{$ENDIF}
pop edi;
pop esi;
pop ebx;
end;
procedure AVXMatrixRotateUnaligned(N : TASMNativeInt; X : PDouble; const LineWidthDX : TASMNativeInt;
Y : PDouble; LineWidthDY : TASMNativeInt; c, s : double); {$IFDEF FPC} assembler; {$ELSE} register; {$ENDIF}
// eax = N, edx = X, ecx = LineWidthDX
asm
push ebx;
push edi;
push esi;
lea esi, s;
{$IFDEF FPC}vmovsd xmm2, [esi];{$ELSE}db $C5,$FB,$10,$16;{$ENDIF}
lea edi, cLocMinusOne;
{$IFDEF FPC}vmulsd xmm2, xmm2, [edi];{$ELSE}db $C5,$EB,$59,$17;{$ENDIF}
{$IFDEF FPC}vmovddup xmm0, xmm2;{$ELSE}db $C5,$FB,$12,$C2;{$ENDIF}
lea edi, c;
{$IFDEF FPC}vmovddup xmm1, [edi];{$ELSE}db $C5,$FB,$12,$0F;{$ENDIF}
{$IFDEF FPC}vmovddup xmm2, [esi];{$ELSE}db $C5,$FB,$12,$16;{$ENDIF}
mov edx, X;
mov ebx, Y;
mov edi, LineWidthDY;
sub eax, 2;
jl @@exitLoop;
@@forNloop:
// do a full load -> intermediate store in xmm5, and xmm6
{$IFDEF FPC}vmovlpd xmm5, xmm5, [edx]; {$ELSE}db $C5,$D1,$12,$2A;{$ENDIF} // load x, x+1
{$IFDEF FPC}vmovhpd xmm5, xmm5, [edx + ecx];{$ELSE}db $C5,$D1,$16,$2C,$0A;{$ENDIF}
{$IFDEF FPC}vmovlpd xmm6, xmm6, [ebx]; {$ELSE}db $C5,$C9,$12,$33;{$ENDIF} // load y, y+1
{$IFDEF FPC}vmovhpd xmm6, xmm6, [ebx + edi];{$ELSE}db $C5,$C9,$16,$34,$3B;{$ENDIF}
{$IFDEF FPC}vmulpd xmm3, xmm5, xmm0; {$ELSE}db $C5,$D1,$59,$D8;{$ENDIF} // x, x+1 * -s
{$IFDEF FPC}vmulpd xmm5, xmm5, xmm1; {$ELSE}db $C5,$D1,$59,$E9;{$ENDIF} // x, x+1 * c
{$IFDEF FPC}vmulpd xmm4, xmm6, xmm1; {$ELSE}db $C5,$C9,$59,$E1;{$ENDIF} // y, y+1 * c
{$IFDEF FPC}vmulpd xmm6, xmm6, xmm2; {$ELSE}db $C5,$C9,$59,$F2;{$ENDIF} // y, y+1 * s
{$IFDEF FPC}vaddpd xmm5, xmm5, xmm6; {$ELSE}db $C5,$D1,$58,$EE;{$ENDIF} // c*x + s*y , c*(x+1) + s*(y+1)
{$IFDEF FPC}vaddpd xmm3, xmm3, xmm4; {$ELSE}db $C5,$E1,$58,$DC;{$ENDIF} // -s*x + c*y, -s(x+1) + c*(y+1)
// write back
{$IFDEF FPC}vmovlpd [edx], xmm5;{$ELSE}db $C5,$F9,$13,$2A;{$ENDIF}
{$IFDEF FPC}vmovhpd [edx + ecx], xmm5;{$ELSE}db $C5,$F9,$17,$2C,$0A;{$ENDIF}
{$IFDEF FPC}vmovlpd [ebx], xmm3;{$ELSE}db $C5,$F9,$13,$1B;{$ENDIF}
{$IFDEF FPC}vmovhpd [ebx + edi], xmm3;{$ELSE}db $C5,$F9,$17,$1C,$3B;{$ENDIF}
add edx, ecx;
add edx, ecx;
add ebx, edi;
add ebx, edi;
sub eax, 2;
jge @@forNloop;
@@exitLoop:
// test for an odd N
add eax, 2;
jz @@endProc;
// handle last element
{$IFDEF FPC}vmovsd xmm5, [edx];{$ELSE}db $C5,$FB,$10,$2A;{$ENDIF}
{$IFDEF FPC}vmovsd xmm6, [ebx];{$ELSE}db $C5,$FB,$10,$33;{$ENDIF}
//dtemp := c*pX^[i] + s*pY^[i];
//pY^[i] := - s*pX^[i] + c*pY^[i];
//px^[i] := dtemp;
{$IFDEF FPC}vmulsd xmm3, xmm5, xmm0; {$ELSE}db $C5,$D3,$59,$D8;{$ENDIF} // x * -s
{$IFDEF FPC}vmulsd xmm5, xmm5, xmm1; {$ELSE}db $C5,$D3,$59,$E9;{$ENDIF} // x * c
{$IFDEF FPC}vmulsd xmm4, xmm6, xmm1; {$ELSE}db $C5,$CB,$59,$E1;{$ENDIF} // y * c
{$IFDEF FPC}vmulsd xmm6, xmm6, xmm2; {$ELSE}db $C5,$CB,$59,$F2;{$ENDIF} // y * s
{$IFDEF FPC}vaddsd xmm5, xmm5, xmm6; {$ELSE}db $C5,$D3,$58,$EE;{$ENDIF} // c*x + s*y
{$IFDEF FPC}vaddsd xmm3, xmm3, xmm4; {$ELSE}db $C5,$E3,$58,$DC;{$ENDIF} // -s*x + c*y
// write back
{$IFDEF FPC}vmovsd [edx], xmm5;{$ELSE}db $C5,$FB,$11,$2A;{$ENDIF}
{$IFDEF FPC}vmovsd [ebx], xmm3;{$ELSE}db $C5,$FB,$11,$1B;{$ENDIF}
@@endProc:
{$IFDEF FPC}vzeroupper;{$ELSE}db $C5,$F8,$77;{$ENDIF}
pop esi;
pop edi;
pop ebx;
end;
procedure AVXMatrixRotate(N : TASMNativeInt; X : PDouble; const LineWidthDX : TASMNativeInt; Y : PDouble; LineWidthDY : TASMNativeInt; const c, s : double);
begin
if N <= 0 then
exit;
if (LineWidthDX = sizeof(double)) and (LineWidthDY = sizeof(double))
then
AVXMatrixRotateAligned(N, X, Y, c, s)
else
AVXMatrixRotateUnAligned(N, X, LineWidthDX, Y, LineWidthDY, c, s)
end;
{$ENDIF}
end.