-
Notifications
You must be signed in to change notification settings - Fork 35
/
AVXMatrixTransposeOperationsx64.pas
430 lines (344 loc) · 16.4 KB
/
AVXMatrixTransposeOperationsx64.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
// ###################################################################
// #### This file is part of the mathematics library project, and is
// #### offered under the licence agreement described on
// #### http://www.mrsoft.org/
// ####
// #### Copyright:(c) 2017, 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 AVXMatrixTransposeOperationsx64;
interface
{$I 'mrMath_CPU.inc'}
{$IFDEF x64}
procedure AVXMatrixTransposeAligned(dest : PDouble; const destLineWidth : NativeInt; mt : PDouble; const LineWidth : NativeInt; {$ifdef UNIX}unixWidth{$ELSE}width{$endif}, {$ifdef UNIX}unixHeight{$ELSE}height{$endif} : NativeInt); {$IFDEF FPC}assembler;{$ENDIF}
procedure AVXMatrixTransposeUnaligned(dest : PDouble; const destLineWidth : NativeInt; mt : PDouble; const LineWidth : NativeInt; {$ifdef UNIX}unixWidth{$ELSE}width{$endif}, {$ifdef UNIX}unixHeight{$ELSE}height{$endif} : NativeInt); {$IFDEF FPC}assembler;{$ENDIF}
// Inplace Trasnposition of an N x N matrix
procedure AVXMatrixTransposeInplace(mt : PDouble; const LineWidth : NativeInt; N : NativeInt); {$IFDEF FPC}assembler;{$ENDIF}
{$ENDIF}
implementation
{$IFDEF x64}
procedure AVXMatrixTransposeAligned(dest : PDouble; const destLineWidth : NativeInt;
mt : PDouble; const LineWidth : NativeInt; {$ifdef UNIX}unixWidth{$ELSE}width{$endif}, {$ifdef UNIX}unixHeight{$ELSE}height{$endif} : NativeInt); {$IFDEF FPC}assembler;{$ENDIF}
var iR12, iR13, iR14, iR15, iRSI, iRDI : NativeInt;
{$ifdef UNIX}
width : NativeInt;
height : NativeInt;
{$ENDIF}
asm
{$IFDEF UNIX}
// Linux uses a diffrent ABI -> copy over the registers so they meet with winABI
// (note that the 5th and 6th parameter are are on the stack)
// The parameters are passed in the following order:
// RDI, RSI, RDX, RCX -> mov to RCX, RDX, R8, R9
mov width, r8;
mov height, r9;
mov r8, rdx;
mov r9, rcx;
mov rcx, rdi;
mov rdx, rsi;
{$ENDIF}
// rcx = dest, rdx = destLineWidth, r8 = mt, r9 = LineWidth
mov iR12, r12;
mov iR13, r13;
mov iR14, r14;
mov iR15, r15;
mov iRSI, rsi;
mov iRDI, rdi;
// stack handling
sub rsp, $40;
{$IFDEF AVXSUP}vmovupd [rsp + $00], xmm4; {$ELSE}db $C5,$F9,$11,$24,$24;{$ENDIF}
{$IFDEF AVXSUP}vmovupd [rsp + $10], xmm5; {$ELSE}db $C5,$F9,$11,$6C,$24,$10;{$ENDIF}
{$IFDEF AVXSUP}vmovupd [rsp + $20], xmm6; {$ELSE}db $C5,$F9,$11,$74,$24,$20;{$ENDIF}
{$IFDEF AVXSUP}vmovupd [rsp + $30], xmm7; {$ELSE}db $C5,$F9,$11,$7C,$24,$30;{$ENDIF}
// iters := -width*sizeof(double);
mov r10, width;
shl r10, 3;
imul r10, -1;
sub r8, r10;
// mtLineWidth2 := 4**LineWidth
mov r11, r9;
shl r11, 2;
mov rdi, height;
sub rdi, 4;
jl @@foryloop4End;
// we have at least a height of 4 -> try to transpose 4x4 blocks
@foryloop4:
// prepare pointers to mt
mov r12, r8;
mov r13, r12;
add r13, r9; // + linewidth
mov r14, r13;
add r14, r9;
mov r15, r14;
add r15, r9;
mov rsi, rcx;
// 4x4 blockwise transposition
mov rax, r10;
@forxloop4:
add rax, 32;
jg @loopend4;
{$IFDEF AVXSUP}vmovapd ymm0, [r12 + rax - 32]; {$ELSE}db $C4,$C1,$7D,$28,$44,$04,$E0;{$ENDIF}
{$IFDEF AVXSUP}vmovapd ymm1, [r13 + rax - 32]; {$ELSE}db $C4,$C1,$7D,$28,$4C,$05,$E0;{$ENDIF}
{$IFDEF AVXSUP}vmovapd ymm2, [r14 + rax - 32]; {$ELSE}db $C4,$C1,$7D,$28,$54,$06,$E0;{$ENDIF}
{$IFDEF AVXSUP}vmovapd ymm3, [r15 + rax - 32]; {$ELSE}db $C4,$C1,$7D,$28,$5C,$07,$E0;{$ENDIF}
{$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;{$ENDIF}
{$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}
{$IFDEF AVXSUP}vmovapd [rsi], ymm0; {$ELSE}db $C5,$FD,$29,$06;{$ENDIF}
add rsi, rdx;
{$IFDEF AVXSUP}vmovapd [rsi], ymm1; {$ELSE}db $C5,$FD,$29,$0E;{$ENDIF}
add rsi, rdx;
{$IFDEF AVXSUP}vmovapd [rsi], ymm2; {$ELSE}db $C5,$FD,$29,$16;{$ENDIF}
add rsi, rdx;
{$IFDEF AVXSUP}vmovapd [rsi], ymm3; {$ELSE}db $C5,$FD,$29,$1E;{$ENDIF}
add rsi, rdx;
jmp @forxloop4;
@loopend4:
sub rax, 32;
jz @nextline4;
// handle the missing columns
@forxloop:
{$IFDEF AVXSUP}vmovsd xmm0, [r12 + rax]; {$ELSE}db $C4,$C1,$7B,$10,$04,$04;{$ENDIF}
{$IFDEF AVXSUP}vmovsd [rsi], xmm0; {$ELSE}db $C5,$FB,$11,$06;{$ENDIF}
{$IFDEF AVXSUP}vmovsd xmm0, [r13 + rax]; {$ELSE}db $C4,$A1,$7B,$10,$04,$28;{$ENDIF}
{$IFDEF AVXSUP}vmovsd [rsi + 8], xmm0; {$ELSE}db $C5,$FB,$11,$46,$08;{$ENDIF}
{$IFDEF AVXSUP}vmovsd xmm0, [r14 + rax]; {$ELSE}db $C4,$C1,$7B,$10,$04,$06;{$ENDIF}
{$IFDEF AVXSUP}vmovsd [rsi + 16], xmm0; {$ELSE}db $C5,$FB,$11,$46,$10;{$ENDIF}
{$IFDEF AVXSUP}vmovsd xmm0, [r15 + rax]; {$ELSE}db $C4,$C1,$7B,$10,$04,$07;{$ENDIF}
{$IFDEF AVXSUP}vmovsd [rsi + 24], xmm0; {$ELSE}db $C5,$FB,$11,$46,$18;{$ENDIF}
add rsi, rdx;
add rax, 8;
jnz @forxloop;
@nextline4:
// dest := dest + 4*sizeof(double)
add rcx, 32;
add r8, r11; // src := src + 4*LineWidth
sub rdi, 4;
jge @foryloop4;
@@foryloop4End:
add rdi, 4;
jz @endproc;
// #############################
// #### handle the last max 4 rows
@@foryloop:
mov rsi, rcx;
mov rax, r10;
@@forxloop:
{$IFDEF AVXSUP}vmovsd xmm0, [r8 + rax]; {$ELSE}db $C4,$C1,$7B,$10,$04,$00;{$ENDIF}
{$IFDEF AVXSUP}vmovsd [rsi], xmm0; {$ELSE}db $C5,$FB,$11,$06;{$ENDIF}
add rsi, rdx;
add rax, 8;
jnz @@forxloop;
add rcx, 8;
add r8, r9;
dec rdi;
jnz @@foryloop;
@endproc:
// epilog
mov r12, iR12;
mov r13, iR13;
mov r14, iR14;
mov r15, iR15;
mov rsi, iRSI;
mov rdi, iRDI;
{$IFDEF AVXSUP}vmovupd xmm4, [rsp + $00]; {$ELSE}db $C5,$F9,$10,$24,$24;{$ENDIF}
{$IFDEF AVXSUP}vmovupd xmm5, [rsp + $10]; {$ELSE}db $C5,$F9,$10,$6C,$24,$10;{$ENDIF}
{$IFDEF AVXSUP}vmovupd xmm6, [rsp + $20]; {$ELSE}db $C5,$F9,$10,$74,$24,$20;{$ENDIF}
{$IFDEF AVXSUP}vmovupd xmm7, [rsp + $30]; {$ELSE}db $C5,$F9,$10,$7C,$24,$30;{$ENDIF}
add rsp, $40;
{$IFDEF AVXSUP}vzeroupper; {$ELSE}db $C5,$F8,$77;{$ENDIF}
end;
procedure AVXMatrixTransposeUnaligned(dest : PDouble; const destLineWidth : NativeInt; mt : PDouble; const LineWidth : NativeInt; {$ifdef UNIX}unixWidth{$ELSE}width{$endif}, {$ifdef UNIX}unixHeight{$ELSE}height{$endif} : NativeInt); {$IFDEF FPC}assembler;{$ENDIF}
var iR12, iR13, iR14, iR15, iRSI, iRDI : NativeInt;
{$ifdef UNIX}
width : NativeInt;
height : NativeInt;
{$ENDIF}
asm
{$IFDEF UNIX}
// Linux uses a diffrent ABI -> copy over the registers so they meet with winABI
// (note that the 5th and 6th parameter are are on the stack)
// The parameters are passed in the following order:
// RDI, RSI, RDX, RCX -> mov to RCX, RDX, R8, R9
mov width, r8;
mov height, r9;
mov r8, rdx;
mov r9, rcx;
mov rcx, rdi;
mov rdx, rsi;
{$ENDIF}
// rcx = dest, rdx = destLineWidth, r8 = mt, r9 = LineWidth
mov iR12, r12;
mov iR13, r13;
mov iR14, r14;
mov iR15, r15;
mov iRSI, rsi;
mov iRDI, rdi;
sub rsp, $40;
{$IFDEF AVXSUP}vmovupd [rsp + $00], xmm4; {$ELSE}db $C5,$F9,$11,$24,$24;{$ENDIF}
{$IFDEF AVXSUP}vmovupd [rsp + $10], xmm5; {$ELSE}db $C5,$F9,$11,$6C,$24,$10;{$ENDIF}
{$IFDEF AVXSUP}vmovupd [rsp + $20], xmm6; {$ELSE}db $C5,$F9,$11,$74,$24,$20;{$ENDIF}
{$IFDEF AVXSUP}vmovupd [rsp + $30], xmm7; {$ELSE}db $C5,$F9,$11,$7C,$24,$30;{$ENDIF}
// iters := -width*sizeof(double);
mov r10, width;
imul r10, -8;
sub r8, r10;
// mtLineWidth2 := 4**LineWidth
mov r11, r9;
shl r11, 2;
mov rdi, height;
sub rdi, 4;
jl @@foryloop4End;
// we have at least a height of 4 -> try to transpose 4x4 blocks
@foryloop4:
// prepare pointers to mt
mov r12, r8;
mov r13, r12;
add r13, r9; // + linewidth
mov r14, r13;
add r14, r9;
mov r15, r14;
add r15, r9;
mov rsi, rcx;
// 4x4 blockwise transposition
mov rax, r10;
@forxloop4:
add rax, 32;
jg @loopend4;
{$IFDEF AVXSUP}vmovupd ymm0, [r12 + rax - 32]; {$ELSE}db $C4,$C1,$7D,$10,$44,$04,$E0;{$ENDIF}
{$IFDEF AVXSUP}vmovupd ymm1, [r13 + rax - 32]; {$ELSE}db $C4,$C1,$7D,$10,$4C,$05,$E0;{$ENDIF}
{$IFDEF AVXSUP}vmovupd ymm2, [r14 + rax - 32]; {$ELSE}db $C4,$C1,$7D,$10,$54,$06,$E0;{$ENDIF}
{$IFDEF AVXSUP}vmovupd ymm3, [r15 + rax - 32]; {$ELSE}db $C4,$C1,$7D,$10,$5C,$07,$E0;{$ENDIF}
{$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;{$ENDIF}
{$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}
{$IFDEF AVXSUP}vmovupd [rsi], ymm0; {$ELSE}db $C5,$FD,$11,$06;{$ENDIF}
add rsi, rdx;
{$IFDEF AVXSUP}vmovupd [rsi], ymm1; {$ELSE}db $C5,$FD,$11,$0E;{$ENDIF}
add rsi, rdx;
{$IFDEF AVXSUP}vmovupd [rsi], ymm2; {$ELSE}db $C5,$FD,$11,$16;{$ENDIF}
add rsi, rdx;
{$IFDEF AVXSUP}vmovupd [rsi], ymm3; {$ELSE}db $C5,$FD,$11,$1E;{$ENDIF}
add rsi, rdx;
jmp @forxloop4;
@loopend4:
sub rax, 32;
jz @nextline4;
// handle the missing columns
@forxloop:
{$IFDEF AVXSUP}vmovsd xmm0, [r12 + rax]; {$ELSE}db $C4,$C1,$7B,$10,$04,$04;{$ENDIF}
{$IFDEF AVXSUP}vmovsd [rsi], xmm0; {$ELSE}db $C5,$FB,$11,$06;{$ENDIF}
{$IFDEF AVXSUP}vmovsd xmm0, [r13 + rax]; {$ELSE}db $C4,$A1,$7B,$10,$04,$28;{$ENDIF}
{$IFDEF AVXSUP}vmovsd [rsi + 8], xmm0; {$ELSE}db $C5,$FB,$11,$46,$08;{$ENDIF}
{$IFDEF AVXSUP}vmovsd xmm0, [r14 + rax]; {$ELSE}db $C4,$C1,$7B,$10,$04,$06;{$ENDIF}
{$IFDEF AVXSUP}vmovsd [rsi + 16], xmm0; {$ELSE}db $C5,$FB,$11,$46,$10;{$ENDIF}
{$IFDEF AVXSUP}vmovsd xmm0, [r15 + rax]; {$ELSE}db $C4,$C1,$7B,$10,$04,$07;{$ENDIF}
{$IFDEF AVXSUP}vmovsd [rsi + 24], xmm0; {$ELSE}db $C5,$FB,$11,$46,$18;{$ENDIF}
add rsi, rdx;
add rax, 8;
jnz @forxloop;
@nextline4:
// dest := dest + 4*sizeof(double)
add rcx, 32;
add r8, r11; // src := src + 4*LineWidth
sub rdi, 4;
jge @foryloop4;
@@foryloop4End:
add rdi, 4;
jz @endproc;
// #############################
// #### handle the last max 4 rows
@@foryloop:
mov rsi, rcx;
mov rax, r10;
@@forxloop:
{$IFDEF AVXSUP}vmovsd xmm0, [r8 + rax]; {$ELSE}db $C4,$C1,$7B,$10,$04,$00;{$ENDIF}
{$IFDEF AVXSUP}vmovsd [rsi], xmm0; {$ELSE}db $C5,$FB,$11,$06;{$ENDIF}
add rsi, rdx;
add rax, 8;
jnz @@forxloop;
add rcx, 8;
add r8, r9;
dec rdi;
jnz @@foryloop;
@endproc:
// epilog
mov r12, iR12;
mov r13, iR13;
mov r14, iR14;
mov r15, iR15;
mov rsi, iRSI;
mov rdi, iRDI;
{$IFDEF AVXSUP}vmovupd xmm4, [rsp + $00]; {$ELSE}db $C5,$F9,$10,$24,$24;{$ENDIF}
{$IFDEF AVXSUP}vmovupd xmm5, [rsp + $10]; {$ELSE}db $C5,$F9,$10,$6C,$24,$10;{$ENDIF}
{$IFDEF AVXSUP}vmovupd xmm6, [rsp + $20]; {$ELSE}db $C5,$F9,$10,$74,$24,$20;{$ENDIF}
{$IFDEF AVXSUP}vmovupd xmm7, [rsp + $30]; {$ELSE}db $C5,$F9,$10,$7C,$24,$30;{$ENDIF}
add rsp, $40;
{$IFDEF AVXSUP}vzeroupper; {$ELSE}db $C5,$F8,$77;{$ENDIF}
end;
// simple Inplace Trasnposition of an N x N matrix
procedure AVXMatrixTransposeInplace(mt : PDouble; const LineWidth : NativeInt; N : NativeInt); {$IFDEF FPC}assembler;{$ENDIF}
// rcx: mt, rdx, LineWidth, r8: N
asm
{$IFDEF UNIX}
// Linux uses a diffrent ABI -> copy over the registers so they meet with winABI
// (note that the 5th and 6th parameter are are on the stack)
// The parameters are passed in the following order:
// RDI, RSI, RDX, RCX -> mov to RCX, RDX, R8, R9
mov r8, rdx;
mov r9, rcx;
mov rcx, rdi;
mov rdx, rsi;
{$ENDIF}
// check if N < 2 -> do nothing an a 1x1 matrix
cmp r8, 2;
jl @@exitProc;
// iter: -N*sizeof(Double)
mov rax, r8;
imul rax, -8;
mov r9, rcx; // pDest1: genptr(mt, 0, 1, linewidth)
add r9, rdx;
sub rcx, rax; // mt + iter
// for y := 0 to n - 2
dec r8;
@@foryloop:
mov r10, rax; // iter aka x
add r10, 8;
mov r11, r9;
// for x := y + 1 to n-1 do
@@forxloop:
{$IFDEF AVXSUP}vmovsd xmm0, [rcx + r10]; {$ELSE}db $C4,$A1,$7B,$10,$04,$11;{$ENDIF}
{$IFDEF AVXSUP}vmovsd xmm1, [r11]; {$ELSE}db $C4,$C1,$7B,$10,$0B;{$ENDIF}
{$IFDEF AVXSUP}vmovsd [rcx + r10], xmm1; {$ELSE}db $C4,$A1,$7B,$11,$0C,$11;{$ENDIF}
{$IFDEF AVXSUP}vmovsd [r11], xmm0; {$ELSE}db $C4,$C1,$7B,$11,$03;{$ENDIF}
add r11, rdx;
add r10, 8;
jnz @@forxloop;
add rax, 8; // iter + sizeof(double);
//pDest := PConstDoubleArr( GenPtr(dest, 0, y, destLineWidth) );
add rcx, rdx;
// GenPtr(dest, y, y + 1, destLineWidth);
add r9, rdx;
add r9, 8;
dec r8;
jnz @@foryloop;
@@exitProc:
{$IFDEF AVXSUP}vzeroupper; {$ELSE}db $C5,$F8,$77;{$ENDIF}
end;
{$ENDIF}
end.