-
Notifications
You must be signed in to change notification settings - Fork 36
/
Copy pathAVXMatrixSqrtOperations.pas
198 lines (140 loc) · 6.97 KB
/
AVXMatrixSqrtOperations.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
// ###################################################################
// #### 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 AVXMatrixSqrtOperations;
// #####################################################
// #### SQRT opertaion applied to every element in a matrix
// #####################################################
interface
{$I 'mrMath_CPU.inc'}
{$IFNDEF x64}
procedure AVXMatrixSQRTAligned(Dest : PDouble; const LineWidth, Width, Height : NativeInt); {$IFDEF FPC} assembler; {$ELSE} register; {$ENDIF}
procedure AVXMatrixSQRTUnAligned(Dest : PDouble; const LineWidth, Width, Height : NativeInt); {$IFDEF FPC} assembler; {$ELSE} register; {$ENDIF}
{$ENDIF}
implementation
{$IFNDEF x64}
procedure AVXMatrixSQRTAligned(Dest : PDouble; const LineWidth, Width, Height : NativeInt);
asm
push edi;
push esi;
//iters := -width*sizeof(double);
imul ecx, -8;
// helper registers dest pointers
sub eax, ecx;
// for y := 0 to height - 1:
mov esi, Height;
@@addforyloop:
// for x := 0 to w - 1;
// prepare for reverse loop
mov edi, ecx;
@addforxloop:
add edi, 128;
jg @loopEnd;
// prefetch data...
// prefetchw [eax + edi];
// elementwise sqrt
{$IFDEF AVXSUP}vsqrtpd ymm0, [eax + edi - 128]; {$ELSE}db $C5,$FD,$51,$44,$38,$80;{$ENDIF}
{$IFDEF AVXSUP}vmovapd [eax + edi - 128], ymm0; {$ELSE}db $C5,$FD,$29,$44,$38,$80;{$ENDIF}
{$IFDEF AVXSUP}vsqrtpd ymm1, [eax + edi - 96]; {$ELSE}db $C5,$FD,$51,$4C,$38,$A0;{$ENDIF}
{$IFDEF AVXSUP}vmovapd [eax + edi - 96], ymm1; {$ELSE}db $C5,$FD,$29,$4C,$38,$A0;{$ENDIF}
{$IFDEF AVXSUP}vsqrtpd ymm0, [eax + edi - 64]; {$ELSE}db $C5,$FD,$51,$44,$38,$C0;{$ENDIF}
{$IFDEF AVXSUP}vmovapd [eax + edi - 64], ymm0; {$ELSE}db $C5,$FD,$29,$44,$38,$C0;{$ENDIF}
{$IFDEF AVXSUP}vsqrtpd ymm1, [eax + edi - 32]; {$ELSE}db $C5,$FD,$51,$4C,$38,$E0;{$ENDIF}
{$IFDEF AVXSUP}vmovapd [eax + edi - 32], ymm1; {$ELSE}db $C5,$FD,$29,$4C,$38,$E0;{$ENDIF}
jmp @addforxloop
@loopEnd:
sub edi, 128;
jz @nextLine;
@addforxloop2:
add edi, 16;
jg @addforxloop2end;
{$IFDEF AVXSUP}vsqrtpd xmm0, [eax + edi - 16]; {$ELSE}db $C5,$F9,$51,$44,$38,$F0;{$ENDIF}
{$IFDEF AVXSUP}vmovapd [eax + edi - 16], xmm0; {$ELSE}db $C5,$F9,$29,$44,$38,$F0;{$ENDIF}
jmp @addforxloop2;
@addforxloop2end:
sub edi, 16;
jz @nextLine;
{$IFDEF AVXSUP}vmovsd xmm0, [eax + edi]; {$ELSE}db $C5,$FB,$10,$04,$38;{$ENDIF}
{$IFDEF AVXSUP}vsqrtsd xmm0, xmm0, xmm0; {$ELSE}db $C5,$FB,$51,$C0;{$ENDIF}
{$IFDEF AVXSUP}vmovsd [eax + edi], xmm0; {$ELSE}db $C5,$FB,$11,$04,$38;{$ENDIF}
@nextLine:
// next line:
add eax, edx;
// loop y end
dec esi;
jnz @@addforyloop;
{$IFDEF AVXSUP}vzeroupper; {$ELSE}db $C5,$F8,$77;{$ENDIF}
pop esi;
pop edi;
end;
procedure AVXMatrixSQRTUnAligned(Dest : PDouble; const LineWidth, Width, Height : NativeInt);
asm
push edi;
push esi
//iters := -width*sizeof(double);
imul ecx, -8;
// helper registers dest pointers
sub eax, ecx;
// for y := 0 to height - 1:
mov esi, Height;
@@addforyloop:
// for x := 0 to w - 1;
// prepare for reverse loop
mov edi, ecx;
@addforxloop:
add edi, 128;
jg @loopEnd;
// prefetch data...
// prefetchw [eax + edi];
// elementwise sqrt
{$IFDEF AVXSUP}vmovupd ymm2, [eax + edi - 128]; {$ELSE}db $C5,$FD,$10,$54,$38,$80;{$ENDIF}
{$IFDEF AVXSUP}vsqrtpd ymm0, ymm2; {$ELSE}db $C5,$FD,$51,$C2;{$ENDIF}
{$IFDEF AVXSUP}vmovupd [eax + edi - 128], ymm0; {$ELSE}db $C5,$FD,$11,$44,$38,$80;{$ENDIF}
{$IFDEF AVXSUP}vmovupd ymm3, [eax + edi - 96]; {$ELSE}db $C5,$FD,$10,$5C,$38,$A0;{$ENDIF}
{$IFDEF AVXSUP}vsqrtpd ymm1, ymm3; {$ELSE}db $C5,$FD,$51,$CB;{$ENDIF}
{$IFDEF AVXSUP}vmovupd [eax + edi - 96], ymm1; {$ELSE}db $C5,$FD,$11,$4C,$38,$A0;{$ENDIF}
{$IFDEF AVXSUP}vmovupd ymm2, [eax + edi - 64]; {$ELSE}db $C5,$FD,$10,$54,$38,$C0;{$ENDIF}
{$IFDEF AVXSUP}vsqrtpd ymm0, ymm2; {$ELSE}db $C5,$FD,$51,$C2;{$ENDIF}
{$IFDEF AVXSUP}vmovupd [eax + edi - 64], ymm0; {$ELSE}db $C5,$FD,$11,$44,$38,$C0;{$ENDIF}
{$IFDEF AVXSUP}vmovupd ymm3, [eax + edi - 32]; {$ELSE}db $C5,$FD,$10,$5C,$38,$E0;{$ENDIF}
{$IFDEF AVXSUP}vsqrtpd ymm1, ymm3; {$ELSE}db $C5,$FD,$51,$CB;{$ENDIF}
{$IFDEF AVXSUP}vmovupd [eax + edi - 32], ymm1; {$ELSE}db $C5,$FD,$11,$4C,$38,$E0;{$ENDIF}
jmp @addforxloop
@loopEnd:
sub edi, 128;
jz @nextLine;
@addforxloop2:
add edi, 16;
jg @addforxloop2end;
{$IFDEF AVXSUP}vmovupd xmm1, [eax + edi - 16]; {$ELSE}db $C5,$F9,$10,$4C,$38,$F0;{$ENDIF}
{$IFDEF AVXSUP}vsqrtpd xmm0, xmm1; {$ELSE}db $C5,$F9,$51,$C1;{$ENDIF}
{$IFDEF AVXSUP}vmovupd [eax + edi - 16], xmm0; {$ELSE}db $C5,$F9,$11,$44,$38,$F0;{$ENDIF}
jmp @addforxloop2;
@addforxloop2end:
sub edi, 16;
jz @nextLine;
{$IFDEF AVXSUP}vmovsd xmm0, [eax + edi]; {$ELSE}db $C5,$FB,$10,$04,$38;{$ENDIF}
{$IFDEF AVXSUP}vsqrtsd xmm0, xmm0, xmm0; {$ELSE}db $C5,$FB,$51,$C0;{$ENDIF}
{$IFDEF AVXSUP}vmovsd [eax + edi], xmm0; {$ELSE}db $C5,$FB,$11,$04,$38;{$ENDIF}
@nextLine:
// next line:
add eax, edx;
// loop y end
dec esi;
jnz @@addforyloop;
{$IFDEF AVXSUP}vzeroupper; {$ELSE}db $C5,$F8,$77;{$ENDIF}
pop esi;
pop edi;
end;
{$ENDIF}
end.