-
Notifications
You must be signed in to change notification settings - Fork 35
/
AVXMatrixTransposeOperations.pas
385 lines (294 loc) · 12.9 KB
/
AVXMatrixTransposeOperations.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
// ###################################################################
// #### 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.
// ###################################################################
unit AVXMatrixTransposeOperations;
interface
{$I 'mrMath_CPU.inc'}
{$IFNDEF x64}
procedure AVXMatrixTransposeAligned(dest : PDouble; const destLineWidth : NativeInt; mt : PDouble; const LineWidth : NativeInt; width : NativeInt; height : NativeInt); {$IFDEF FPC} assembler; {$ELSE} register; {$ENDIF}
procedure AVXMatrixTransposeUnaligned(dest : PDouble; const destLineWidth : NativeInt; mt : PDouble; const LineWidth : NativeInt; width : NativeInt; height : NativeInt); {$IFDEF FPC} assembler; {$ELSE} register; {$ENDIF}
// Inplace Trasnposition of an N x N matrix
procedure AVXMatrixTransposeInplace(mt : PDouble; const LineWidth : NativeInt; N : NativeInt); {$IFDEF FPC} assembler; {$ELSE} register; {$ENDIF}
{$ENDIF}
implementation
{$IFNDEF x64}
procedure AVXMatrixTransposeAligned(dest : PDouble; const destLineWidth : NativeInt; mt : PDouble; const LineWidth : NativeInt; width : NativeInt; height : NativeInt);
var iter : NativeInt;
asm
// prolog:
push ebx;
push edi;
push esi;
// iters := -width*sizeof(double);
mov edi, width;
imul edi, -8;
mov iter, edi;
sub ecx, edi;
mov ebx, LineWidth;
mov edi, Height;
sub edi, 4;
jl @@foryloop4End;
// we have at least a height of 4 -> try to transpose 4x4 blocks
@foryloop4:
// prepare pointers to mt
push eax;
// 4x4 blockwise transposition
mov esi, iter;
@forxloop4:
add esi, 32;
jg @loopend4;
{$IFDEF AVXSUP}vmovapd ymm0, [ecx + esi - 32]; {$ELSE}db $C5,$FD,$28,$44,$31,$E0;{$ENDIF}
add ecx, ebx;
{$IFDEF AVXSUP}vmovapd ymm1, [ecx + esi - 32]; {$ELSE}db $C5,$FD,$28,$4C,$31,$E0;{$ENDIF}
add ecx, ebx;
{$IFDEF AVXSUP}vmovapd ymm2, [ecx + esi - 32]; {$ELSE}db $C5,$FD,$28,$54,$31,$E0;{$ENDIF}
add ecx, ebx;
{$IFDEF AVXSUP}vmovapd ymm3, [ecx + esi - 32]; {$ELSE}db $C5,$FD,$28,$5C,$31,$E0;{$ENDIF}
add ecx, ebx;
{$IFDEF AVXSUP}vunpckhpd ymm4, ymm0, ymm1; {$ELSE}db $C5,$FD,$15,$E1;{$ENDIF}
{$IFDEF AVXSUP}vunpckhpd ymm5, ymm2, ymm3; {$ELSE}db $C5,$ED,$15,$EB;{$ENDIF}
{$IFDEF AVXSUP}vunpcklpd ymm7, ymm2, ymm3; {$ELSE}db $C5,$ED,$14,$FB;{$ENDIF}
shl ebx, 2;
{$IFDEF AVXSUP}vperm2f128 ymm3, ymm4, ymm5, $31; {$ELSE}db $C4,$E3,$5D,$06,$DD,$31;{$ENDIF}
{$IFDEF AVXSUP}vunpcklpd ymm6, ymm0, ymm1; {$ELSE}db $C5,$FD,$14,$F1;{$ENDIF}
{$IFDEF AVXSUP}vinsertf128 ymm1, ymm4, xmm5, 1; {$ELSE}db $C4,$E3,$5D,$18,$CD,$01;{$ENDIF}
{$IFDEF AVXSUP}vperm2f128 ymm2, ymm6, ymm7, $31; {$ELSE}db $C4,$E3,$4D,$06,$D7,$31;{$ENDIF}
{$IFDEF AVXSUP}vinsertf128 ymm0, ymm6, xmm7, 1; {$ELSE}db $C4,$E3,$4D,$18,$C7,$01;{$ENDIF}
// recover ecx
sub ecx, ebx;
{$IFDEF AVXSUP}vmovapd [eax], ymm0; {$ELSE}db $C5,$FD,$29,$00;{$ENDIF}
add eax, edx;
{$IFDEF AVXSUP}vmovapd [eax], ymm1; {$ELSE}db $C5,$FD,$29,$08;{$ENDIF}
add eax, edx;
shr ebx, 2;
{$IFDEF AVXSUP}vmovapd [eax], ymm2; {$ELSE}db $C5,$FD,$29,$10;{$ENDIF}
add eax, edx;
{$IFDEF AVXSUP}vmovapd [eax], ymm3; {$ELSE}db $C5,$FD,$29,$18;{$ENDIF}
add eax, edx;
jmp @forxloop4;
@loopend4:
sub esi, 32;
jz @nextline4;
// handle the missing columns
@forxloop:
{$IFDEF AVXSUP}vmovsd xmm0, [ecx + esi]; {$ELSE}db $C5,$FB,$10,$04,$31;{$ENDIF}
add ecx, ebx;
{$IFDEF AVXSUP}vmovsd [eax], xmm0; {$ELSE}db $C5,$FB,$11,$00;{$ENDIF}
{$IFDEF AVXSUP}vmovsd xmm0, [ecx + esi]; {$ELSE}db $C5,$FB,$10,$04,$31;{$ENDIF}
{$IFDEF AVXSUP}vmovsd [eax + 8], xmm0; {$ELSE}db $C5,$FB,$11,$40,$08;{$ENDIF}
add ecx, ebx;
{$IFDEF AVXSUP}vmovsd xmm0, [ecx + esi]; {$ELSE}db $C5,$FB,$10,$04,$31;{$ENDIF}
{$IFDEF AVXSUP}vmovsd [eax + 16], xmm0; {$ELSE}db $C5,$FB,$11,$40,$10;{$ENDIF}
add ecx, ebx;
{$IFDEF AVXSUP}vmovsd xmm0, [ecx + esi]; {$ELSE}db $C5,$FB,$10,$04,$31;{$ENDIF}
add ecx, ebx;
{$IFDEF AVXSUP}vmovsd [eax + 24], xmm0; {$ELSE}db $C5,$FB,$11,$40,$18;{$ENDIF}
add eax, edx;
shl ebx, 2;
sub ecx, ebx;
shr ebx, 2;
add esi, 8;
jnz @forxloop;
@nextline4:
pop eax;
lea ecx, [ecx + 4*ebx];
// dest := dest + 4*sizeof(double)
add eax, 32;
sub edi, 4;
jge @foryloop4;
@@foryloop4End:
add edi, 4;
jz @endproc;
// #############################
// #### handle the last max 4 rows
@@foryloop:
mov esi, iter;
push eax;
@@forxloop:
{$IFDEF AVXSUP}vmovsd xmm0, [ecx + esi]; {$ELSE}db $C5,$FB,$10,$04,$31;{$ENDIF}
{$IFDEF AVXSUP}vmovsd [eax], xmm0; {$ELSE}db $C5,$FB,$11,$00;{$ENDIF}
add eax, edx;
add esi, 8;
jnz @@forxloop;
pop eax;
add ecx, LineWidth;
add eax, 8;
dec edi;
jnz @@foryloop;
@endproc:
// epilog
{$IFDEF AVXSUP}vzeroupper; {$ELSE}db $C5,$F8,$77;{$ENDIF}
pop esi;
pop edi;
pop ebx;
end;
procedure AVXMatrixTransposeUnaligned(dest : PDouble; const destLineWidth : NativeInt; mt : PDouble; const LineWidth : NativeInt; width : NativeInt; height : NativeInt);
var iter : NativeInt;
asm
// prolog:
push ebx;
push edi;
push esi;
// iters := -width*sizeof(double);
mov edi, width;
imul edi, -8;
mov iter, edi;
sub ecx, edi;
mov ebx, LineWidth;
mov edi, Height;
sub edi, 4;
jl @@foryloop4End;
// we have at least a height of 4 -> try to transpose 4x4 blocks
@foryloop4:
// prepare pointers to mt
push eax;
// 4x4 blockwise transposition
mov esi, iter;
@forxloop4:
add esi, 32;
jg @loopend4;
{$IFDEF AVXSUP}vmovupd ymm0, [ecx + esi - 32]; {$ELSE}db $C5,$FD,$10,$44,$31,$E0;{$ENDIF}
add ecx, ebx;
{$IFDEF AVXSUP}vmovupd ymm1, [ecx + esi - 32]; {$ELSE}db $C5,$FD,$10,$4C,$31,$E0;{$ENDIF}
add ecx, ebx;
{$IFDEF AVXSUP}vmovupd ymm2, [ecx + esi - 32]; {$ELSE}db $C5,$FD,$10,$54,$31,$E0;{$ENDIF}
add ecx, ebx;
{$IFDEF AVXSUP}vmovupd ymm3, [ecx + esi - 32]; {$ELSE}db $C5,$FD,$10,$5C,$31,$E0;{$ENDIF}
add ecx, ebx;
{$IFDEF AVXSUP}vunpckhpd ymm4, ymm0, ymm1; {$ELSE}db $C5,$FD,$15,$E1;{$ENDIF}
{$IFDEF AVXSUP}vunpckhpd ymm5, ymm2, ymm3; {$ELSE}db $C5,$ED,$15,$EB;{$ENDIF}
{$IFDEF AVXSUP}vunpcklpd ymm7, ymm2, ymm3; {$ELSE}db $C5,$ED,$14,$FB;{$ENDIF}
shl ebx, 2;
{$IFDEF AVXSUP}vperm2f128 ymm3, ymm4, ymm5, $31; {$ELSE}db $C4,$E3,$5D,$06,$DD,$31;{$ENDIF}
{$IFDEF AVXSUP}vunpcklpd ymm6, ymm0, ymm1; {$ELSE}db $C5,$FD,$14,$F1;{$ENDIF}
{$IFDEF AVXSUP}vinsertf128 ymm1, ymm4, xmm5, 1; {$ELSE}db $C4,$E3,$5D,$18,$CD,$01;{$ENDIF}
{$IFDEF AVXSUP}vperm2f128 ymm2, ymm6, ymm7, $31; {$ELSE}db $C4,$E3,$4D,$06,$D7,$31;{$ENDIF}
{$IFDEF AVXSUP}vinsertf128 ymm0, ymm6, xmm7, 1; {$ELSE}db $C4,$E3,$4D,$18,$C7,$01;{$ENDIF}
sub ecx, ebx;
// invert so lea works
{$IFDEF AVXSUP}vmovupd [eax], ymm0; {$ELSE}db $C5,$FD,$11,$00;{$ENDIF}
add eax, edx;
{$IFDEF AVXSUP}vmovupd [eax], ymm1; {$ELSE}db $C5,$FD,$11,$08;{$ENDIF}
add eax, edx;
{$IFDEF AVXSUP}vmovupd [eax], ymm2; {$ELSE}db $C5,$FD,$11,$10;{$ENDIF}
add eax, edx;
{$IFDEF AVXSUP}vmovupd [eax], ymm3; {$ELSE}db $C5,$FD,$11,$18;{$ENDIF}
add eax, edx;
shr ebx, 2;
jmp @forxloop4;
@loopend4:
sub esi, 32;
jz @nextline4;
// handle the missing columns
@forxloop:
{$IFDEF AVXSUP}vmovsd xmm0, [ecx + esi]; {$ELSE}db $C5,$FB,$10,$04,$31;{$ENDIF}
add ecx, ebx;
{$IFDEF AVXSUP}vmovsd [eax], xmm0; {$ELSE}db $C5,$FB,$11,$00;{$ENDIF}
{$IFDEF AVXSUP}vmovsd xmm0, [ecx + esi]; {$ELSE}db $C5,$FB,$10,$04,$31;{$ENDIF}
add ecx, ebx;
{$IFDEF AVXSUP}vmovsd [eax + 8], xmm0; {$ELSE}db $C5,$FB,$11,$40,$08;{$ENDIF}
{$IFDEF AVXSUP}vmovsd xmm0, [ecx + esi]; {$ELSE}db $C5,$FB,$10,$04,$31;{$ENDIF}
add ecx, ebx;
{$IFDEF AVXSUP}vmovsd [eax + 16], xmm0; {$ELSE}db $C5,$FB,$11,$40,$10;{$ENDIF}
{$IFDEF AVXSUP}vmovsd xmm0, [ecx + esi]; {$ELSE}db $C5,$FB,$10,$04,$31;{$ENDIF}
add ecx, ebx;
{$IFDEF AVXSUP}vmovsd [eax + 24], xmm0; {$ELSE}db $C5,$FB,$11,$40,$18;{$ENDIF}
add eax, edx;
shl ebx, 2;
sub ecx, ebx;
shr ebx, 2;
add esi, 8;
jnz @forxloop;
@nextline4:
pop eax;
lea ecx, [ecx + 4*ebx];
// dest := dest + 4*sizeof(double)
add eax, 32;
sub edi, 4;
jge @foryloop4;
@@foryloop4End:
add edi, 4;
jz @endproc;
// #############################
// #### handle the last max 4 rows
@@foryloop:
mov esi, iter;
push eax;
@@forxloop:
{$IFDEF AVXSUP}vmovsd xmm0, [ecx + esi]; {$ELSE}db $C5,$FB,$10,$04,$31;{$ENDIF}
{$IFDEF AVXSUP}vmovsd [eax], xmm0; {$ELSE}db $C5,$FB,$11,$00;{$ENDIF}
add eax, edx;
add esi, 8;
jnz @@forxloop;
pop eax;
add ecx, LineWidth;
add eax, 8;
dec edi;
jnz @@foryloop;
@endproc:
// epilog
{$IFDEF AVXSUP}vzeroupper; {$ELSE}db $C5,$F8,$77;{$ENDIF}
pop esi;
pop edi;
pop ebx;
end;
// simple Inplace Trasnposition of an N x N matrix
procedure AVXMatrixTransposeInplace(mt : PDouble; const LineWidth : NativeInt; N : NativeInt);
// eax = mt, edx = LineWidth, ecx = N
var aN : NativeInt;
asm
push ebx;
push edi;
push esi;
mov ecx, N;
cmp ecx, 2;
jl @@exitProc;
// iter: -N*sizeof(Double)
mov edi, ecx;
imul edi, -8;
dec ecx;
mov aN, ecx;
mov ebx, eax; // pDest1: genptr(mt, 0, 1, linewidth)
add ebx, edx;
sub eax, edi; // mt + iter
// for y := 0 to n - 2
@@foryloop:
mov esi, edi; // iter aka x
add esi, 8;
mov ecx, ebx;
// for x := y + 1 to n-1 do
@@forxloop:
{$IFDEF AVXSUP}vmovsd xmm0, [eax + esi]; {$ELSE}db $C5,$FB,$10,$04,$30;{$ENDIF}
{$IFDEF AVXSUP}vmovsd xmm1, [ecx]; {$ELSE}db $C5,$FB,$10,$09;{$ENDIF}
{$IFDEF AVXSUP}vmovsd [eax + esi], xmm1; {$ELSE}db $C5,$FB,$11,$0C,$30;{$ENDIF}
{$IFDEF AVXSUP}vmovsd [ecx], xmm0; {$ELSE}db $C5,$FB,$11,$01;{$ENDIF}
add ecx, edx;
add esi, 8;
jnz @@forxloop;
add edi, 8; // iter + sizeof(double);
//pDest := PConstDoubleArr( GenPtr(dest, 0, y, destLineWidth) );
add eax, edx;
// GenPtr(dest, y, y + 1, destLineWidth);
add ebx, edx;
add ebx, 8;
dec aN;
jnz @@foryloop;
@@exitProc:
// epilog
{$IFDEF AVXSUP}vzeroupper; {$ELSE}db $C5,$F8,$77;{$ENDIF}
pop esi;
pop edi;
pop ebx;
end;
{$ENDIF}
end.