/[escript]/trunk/paso/src/Pattern_unrollBlocks.c
ViewVC logotype

Contents of /trunk/paso/src/Pattern_unrollBlocks.c

Parent Directory Parent Directory | Revision Log Revision Log


Revision 3283 - (show annotations)
Mon Oct 18 22:39:28 2010 UTC (10 years, 4 months ago) by gross
File MIME type: text/plain
File size: 3609 byte(s)
AMG reengineered, BUG is Smoother fixed.


1
2 /*******************************************************
3 *
4 * Copyright (c) 2003-2010 by University of Queensland
5 * Earth Systems Science Computational Center (ESSCC)
6 * http://www.uq.edu.au/esscc
7 *
8 * Primary Business: Queensland, Australia
9 * Licensed under the Open Software License version 3.0
10 * http://www.opensource.org/licenses/osl-3.0.php
11 *
12 *******************************************************/
13
14
15 /**************************************************************/
16
17 /* Paso: Pattern_unrollBlocks */
18
19 /**************************************************************/
20
21 /* Author: Lutz Gross, l.gross@uq.edu.au */
22
23 /**************************************************************/
24
25 #include "Paso.h"
26 #include "Pattern.h"
27
28 /**************************************************************/
29
30 /* creates Pattern */
31
32 Paso_Pattern* Paso_Pattern_unrollBlocks(Paso_Pattern* pattern, \
33 int type, dim_t output_block_size,dim_t input_block_size) {
34 Paso_Pattern*out=NULL;
35 index_t *ptr=NULL,*index=NULL,iPtr;
36 dim_t i,j,k, block_size, new_len, new_numOutput, new_numInput;
37 index_t index_offset_in=(pattern->type & PATTERN_FORMAT_OFFSET1 ? 1:0);
38 index_t index_offset_out=(type & PATTERN_FORMAT_OFFSET1 ? 1:0);
39
40 Esys_resetError();
41 if ( ( output_block_size == 1 ) && (input_block_size == 1) && ((pattern->type & PATTERN_FORMAT_OFFSET1) == (type & PATTERN_FORMAT_OFFSET1) ) ) {
42 out = Paso_Pattern_getReference(pattern);
43 } else {
44 block_size=output_block_size*input_block_size;
45 new_len=(pattern->len)*block_size;
46 new_numOutput=(pattern->numOutput)*output_block_size;
47 new_numInput=(pattern->numInput)*input_block_size;
48
49 ptr=MEMALLOC(new_numOutput+1,index_t);
50 index=MEMALLOC(new_len,index_t);
51 if (! ( Esys_checkPtr(ptr) || Esys_checkPtr(index) ) ) {
52 #pragma omp parallel
53 {
54 #pragma omp for private(i) schedule(static)
55 for (i=0;i<new_numOutput+1;++i) ptr[i]=index_offset_out;
56
57 #pragma omp single
58 ptr[new_numOutput]=new_len+index_offset_out;
59
60 #pragma omp for private(i,k) schedule(static)
61 for (i=0;i<pattern->numOutput;++i) {
62 for (k=0;k<output_block_size;++k) {
63 ptr[i*output_block_size+k]=(pattern->ptr[i]-index_offset_in)*block_size+
64 (pattern->ptr[i+1]-pattern->ptr[i])*input_block_size*k+index_offset_out;
65 }
66 }
67
68 #pragma omp for private(i,iPtr) schedule(static)
69 for (i=0;i<new_numOutput;++i) {
70 #pragma ivdep
71 for (iPtr=ptr[i]-index_offset_out;iPtr<ptr[i+1]-index_offset_out;++iPtr) index[iPtr]=index_offset_out;
72 }
73
74 #pragma omp for private(i,j,iPtr,k) schedule(static)
75 for (i=0;i<pattern->numOutput;++i) {
76 for (iPtr=pattern->ptr[i]-index_offset_in;iPtr<pattern->ptr[i+1]-index_offset_in;++iPtr) {
77 for (k=0;k<output_block_size;++k) {
78 #pragma ivdep
79 for (j=0;j<input_block_size;++j) {
80 index[ptr[i*output_block_size+k]-index_offset_out+(iPtr-(pattern->ptr[i]-index_offset_in))*input_block_size+j]=(pattern->index[iPtr]-index_offset_in)*input_block_size+j+index_offset_out;
81 }
82 }
83 }
84 }
85 }
86 out=Paso_Pattern_alloc(type,new_numOutput,new_numInput,ptr,index);
87 }
88 if (! Esys_noError()) {
89 MEMFREE(index);
90 MEMFREE(ptr);
91 }
92 }
93 return out;
94 }

  ViewVC Help
Powered by ViewVC 1.1.26