From 6af8d3d7c41c2740620c077d7dec530aa440e31c Mon Sep 17 00:00:00 2001
From: Nicola Vigano <nicola.vigano@esrf.fr>
Date: Wed, 8 Apr 2015 17:27:23 +0200
Subject: [PATCH] 6D-C++: fixed performance issues in blobs<->sinos
transformations
Signed-off-by: Nicola Vigano <nicola.vigano@esrf.fr>
---
.../include/gt6DBlobsSinogramsTransforms.h | 101 +++++++++++-------
1 file changed, 61 insertions(+), 40 deletions(-)
diff --git a/zUtil_Cxx/include/gt6DBlobsSinogramsTransforms.h b/zUtil_Cxx/include/gt6DBlobsSinogramsTransforms.h
index 09e67777..76b1ba35 100644
--- a/zUtil_Cxx/include/gt6DBlobsSinogramsTransforms.h
+++ b/zUtil_Cxx/include/gt6DBlobsSinogramsTransforms.h
@@ -11,6 +11,7 @@
#include "internal_cell_defs.h"
#include <algorithm>
+#include <vector>
namespace GT6D {
@@ -364,38 +365,48 @@ namespace GT6D {
const mwSize & line_length = dims_sino[0];
const mwSize & lines_num = dims_sino[2];
const mwSize line_length_unroll = ROUND_DOWN(line_length, 4);
+ Type * const sino_data = ((Type *) mxGetData(sino));
const mwSize sino_skip = dims_sino[0] * dims_sino[1];
- #pragma omp parallel for
- for (mwIndex line_num = 0; line_num < lines_num; line_num++)
+ const mwSize & num_projs = mxGetNumberOfElements(proj_offsets_a);
+
+ std::vector<const Type *> blob_data_bases(num_projs);
+ std::vector<Type *> sino_data_bases(num_projs);
+ std::vector<mwSize> blob_skips(num_projs);
+ std::vector<Type> coeffs_t(num_projs);
+ for (mwIndex m = 0; m < num_projs; m++)
{
- const mwSize line_sino_skip = line_num * sino_skip;
+ const mwIndex p = (const mwIndex) proj_offsets[m]-1;
+ const mwIndex b = (const mwIndex) blob_offsets[m]-1;
+ const mwIndex s = (const mwIndex) sino_offsets[m]-1;
- const mwSize num_projs = mxGetNumberOfElements(proj_offsets_a);
- for (mwIndex m = 0; m < num_projs; m++)
- {
- const mwIndex p = (const mwIndex) proj_offsets[m]-1;
- const mwIndex b = (const mwIndex) blob_offsets[m]-1;
- const mwIndex s = (const mwIndex) sino_offsets[m]-1;
+ coeffs_t[m] = (const Type) proj_coeffs[m];
- const Type & c = (const Type) proj_coeffs[m];
+ const mwSize sino_initial_offset = line_length * s;
- inner_sum_scale<Type> op(c);
+ const mxArray * const cell_blobs = mxGetCell(blobs, b);
+ sino_data_bases[m] = sino_data + sino_initial_offset;
- const mwSize sino_initial_offset = line_length * s;
+ const mwSize blob_initial_offset = line_length * p;
+ blob_data_bases[m] = ((const Type *) mxGetData(cell_blobs)) + blob_initial_offset;
- const mxArray * const cell_blobs = mxGetCell(blobs, b);
+ const mwSize * const dims_blob = mxGetDimensions(cell_blobs);
+ blob_skips[m] = dims_blob[0] * dims_blob[1];
+ }
- const mwSize * const dims_blob = mxGetDimensions(cell_blobs);
- const mwSize blob_initial_offset = line_length * p;
- const mwSize blob_skip = dims_blob[0] * dims_blob[1];
+ #pragma omp parallel for
+ for (mwIndex line_num = 0; line_num < lines_num; line_num++)
+ {
+ const mwSize line_sino_skip = line_num * sino_skip;
- const Type * const blob_data_base = ((const Type *) mxGetData(cell_blobs)) + blob_initial_offset;
- const Type * const blob_data_line = blob_data_base + line_num * blob_skip;
+ for (mwIndex m = 0; m < num_projs; m++)
+ {
+ const Type & c = coeffs_t[m];
+ inner_sum_scale<Type> op(c);
- Type * const sino_data_base = ((Type *) mxGetData(sino)) + sino_initial_offset;
- Type * const sino_data_line = sino_data_base + line_sino_skip;
+ const Type * const blob_data_line = blob_data_bases[m] + line_num * blob_skips[m];
+ Type * const sino_data_line = sino_data_bases[m] + line_sino_skip;
for (mwIndex elem = 0; elem < line_length_unroll; elem += 4)
{
@@ -431,38 +442,48 @@ namespace GT6D {
const mwSize & line_length = dims_sino[0];
const mwSize & lines_num = dims_sino[2];
const mwSize line_length_unroll = ROUND_DOWN(line_length, 4);
+ const Type * const sino_data = ((const Type *) mxGetData(sino));
const mwSize sino_skip = dims_sino[0] * dims_sino[1];
- #pragma omp parallel for
- for (mwIndex line_num = 0; line_num < lines_num; line_num++)
+ const mwSize & num_projs = mxGetNumberOfElements(proj_offsets_a);
+
+ std::vector<Type *> blob_data_bases(num_projs);
+ std::vector<const Type *> sino_data_bases(num_projs);
+ std::vector<mwSize> blob_skips(num_projs);
+ std::vector<Type> coeffs_t(num_projs);
+ for (mwIndex m = 0; m < num_projs; m++)
{
- const mwSize line_sino_skip = line_num * sino_skip;
+ const mwIndex p = (const mwIndex) proj_offsets[m]-1;
+ const mwIndex b = (const mwIndex) blob_offsets[m]-1;
+ const mwIndex s = (const mwIndex) sino_offsets[m]-1;
- const mwSize num_projs = mxGetNumberOfElements(proj_offsets_a);
- for (mwIndex m = 0; m < num_projs; m++)
- {
- const mwIndex p = (const mwIndex) proj_offsets[m]-1;
- const mwIndex b = (const mwIndex) blob_offsets[m]-1;
- const mwIndex s = (const mwIndex) sino_offsets[m]-1;
+ coeffs_t[m] = (const Type) proj_coeffs[m];
- const Type & c = (const Type) proj_coeffs[m];
+ const mwSize sino_initial_offset = line_length * s;
- inner_sum_scale<Type> op(c);
+ const mxArray * const cell_blobs = mxGetCell(blobs, b);
+ sino_data_bases[m] = sino_data + sino_initial_offset;
- const mwSize sino_initial_offset = line_length * s;
+ const mwSize blob_initial_offset = line_length * p;
+ blob_data_bases[m] = ((Type *) mxGetData(cell_blobs)) + blob_initial_offset;
- mxArray * const cell_blobs = mxGetCell(blobs, b);
+ const mwSize * const dims_blob = mxGetDimensions(cell_blobs);
+ blob_skips[m] = dims_blob[0] * dims_blob[1];
+ }
- const mwSize * const dims_blob = mxGetDimensions(cell_blobs);
- const mwSize blob_initial_offset = line_length * p;
- const mwSize blob_skip = dims_blob[0] * dims_blob[1];
+ #pragma omp parallel for
+ for (mwIndex line_num = 0; line_num < lines_num; line_num++)
+ {
+ const mwSize line_sino_skip = line_num * sino_skip;
- Type * const blob_data_base = ((Type *) mxGetData(cell_blobs)) + blob_initial_offset;
- Type * const blob_data_line = blob_data_base + line_num * blob_skip;
+ for (mwIndex m = 0; m < num_projs; m++)
+ {
+ const Type & c = coeffs_t[m];
+ inner_sum_scale<Type> op(c);
- const Type * const sino_data_base = ((const Type *) mxGetData(sino)) + sino_initial_offset;
- const Type * const sino_data_line = sino_data_base + line_sino_skip;
+ Type * const blob_data_line = blob_data_bases[m] + line_num * blob_skips[m];
+ const Type * const sino_data_line = sino_data_bases[m] + line_sino_skip;
for (mwIndex elem = 0; elem < line_length_unroll; elem += 4)
{
--
GitLab