/[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 1859 - (show annotations)
Wed Oct 8 03:03:37 2008 UTC (10 years, 11 months ago) by gross
File MIME type: text/plain
File size: 4400 byte(s)
first version of testing for transport solver.
1
2 /*******************************************************
3 *
4 * Copyright (c) 2003-2008 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: gross@access.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 Paso_resetError();
41 if ((pattern->type & PATTERN_FORMAT_SYM) != (type & PATTERN_FORMAT_SYM)) {
42 Paso_setError(TYPE_ERROR,"Paso_Pattern_unrollBlocks: conversion between symmetric and non-symmetric is not implemented yet");
43 return NULL;
44 }
45 if (( (pattern->type & PATTERN_FORMAT_OFFSET1) == (type & PATTERN_FORMAT_OFFSET1)) &&
46 (pattern->input_block_size == input_block_size) &&
47 (pattern->output_block_size == output_block_size)) {
48
49 out = Paso_Pattern_getReference(pattern);
50 } else {
51 if ( ( (pattern->input_block_size >1) && (input_block_size != pattern->input_block_size) ) ||
52 ( (pattern->output_block_size >1) && (output_block_size != pattern->output_block_size) ) ) {
53
54 Paso_setError(TYPE_ERROR,"Paso_Pattern_unrollBlocks: unrolling requires matching block sizes or block size one for input pattern.");
55 return NULL;
56 }
57 /* printf("Information: matrix pattern is unrolled to block size %d x %d with offset %d.\n",output_block_size,input_block_size,index_offset_out); */
58 block_size=output_block_size*input_block_size;
59 new_len=(pattern->len)*block_size;
60 new_numOutput=(pattern->numOutput)*output_block_size;
61 new_numInput=(pattern->numInput)*input_block_size;
62
63 ptr=MEMALLOC(new_numOutput+1,index_t);
64 index=MEMALLOC(new_len,index_t);
65 if (! ( Paso_checkPtr(ptr) || Paso_checkPtr(index) ) ) {
66 #pragma omp parallel
67 {
68 #pragma omp for private(i) schedule(static)
69 for (i=0;i<new_numOutput+1;++i) ptr[i]=index_offset_out;
70
71 #pragma omp single
72 ptr[new_numOutput]=new_len+index_offset_out;
73
74 #pragma omp for private(i,k) schedule(static)
75 for (i=0;i<pattern->numOutput;++i)
76 for (k=0;k<output_block_size;++k) ptr[i*output_block_size+k]=(pattern->ptr[i]-index_offset_in)*block_size+(pattern->ptr[i+1]-pattern->ptr[i])*input_block_size*k+index_offset_out;
77
78 #pragma omp for private(i,iPtr) schedule(static)
79 for (i=0;i<new_numOutput;++i) {
80 #pragma ivdep
81 for (iPtr=ptr[i]-index_offset_out;iPtr<ptr[i+1]-index_offset_out;++iPtr) index[iPtr]=index_offset_out;
82 }
83
84 #pragma omp for private(i,j,iPtr,k) schedule(static)
85 for (i=0;i<pattern->numOutput;++i) {
86 for (iPtr=pattern->ptr[i]-index_offset_in;iPtr<pattern->ptr[i+1]-index_offset_in;++iPtr) {
87 for (k=0;k<output_block_size;++k) {
88 #pragma ivdep
89 for (j=0;j<input_block_size;++j) {
90 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;
91 }
92 }
93 }
94 }
95 }
96 out=Paso_Pattern_alloc(type,pattern->input_block_size * input_block_size,pattern->output_block_size * output_block_size,new_numOutput,new_numInput,ptr,index);
97 }
98 if (! Paso_noError()) {
99 MEMFREE(index);
100 MEMFREE(ptr);
101 }
102 }
103 return out;
104 }

  ViewVC Help
Powered by ViewVC 1.1.26