Sākums Izpētīt Palīdzība
Pierakstīties
RONCC
/
Big-Data-HPC-Platform
Vērot 1
Pievienot zvaigznīti 0
Atdalīts 0
Faili Problēmas 0 Izmaiņu pieprasījumi 0 Vikivietne
Atzars: master
Atzari Tagi
Big-Data-HPC...
/
plugins
/
eclblas
Gordon Smith de2939e222 HPCC-27136 Clean up CMake warnings and errors 3 gadi atpakaļ
..
CMakeLists.txt de2939e222 HPCC-27136 Clean up CMake warnings and errors 3 gadi atpakaļ
README.md adae826495 HPCC-16282 Add BLAS wrappers as standard library attributes and plugins 8 gadi atpakaļ
dasum.cpp 937173f4fe HPCC-16571 Mangle the functions exported from eclblas plugin 8 gadi atpakaļ
daxpy.cpp 937173f4fe HPCC-16571 Mangle the functions exported from eclblas plugin 8 gadi atpakaļ
dgemm.cpp bcf397408e HPCC-17543 Work around misaligned double access in eclblas::dgemm 8 gadi atpakaļ
dgetf2.cpp 937173f4fe HPCC-16571 Mangle the functions exported from eclblas plugin 8 gadi atpakaļ
dpotf2.cpp 937173f4fe HPCC-16571 Mangle the functions exported from eclblas plugin 8 gadi atpakaļ
dscal.cpp 937173f4fe HPCC-16571 Mangle the functions exported from eclblas plugin 8 gadi atpakaļ
dsyrk.cpp 937173f4fe HPCC-16571 Mangle the functions exported from eclblas plugin 8 gadi atpakaļ
dtrsm.cpp 937173f4fe HPCC-16571 Mangle the functions exported from eclblas plugin 8 gadi atpakaļ
eclblas.cpp adae826495 HPCC-16282 Add BLAS wrappers as standard library attributes and plugins 8 gadi atpakaļ
eclblas.hpp 116fab0239 Merge branch 'candidate-6.2.0' into candidate-6.4.0 8 gadi atpakaļ
extract_tri.cpp fa2e97e93e HPCC-16655 Extract_Tri anomaly in eclblas 8 gadi atpakaļ
lib_eclblas.ecllib fa2e97e93e HPCC-16655 Extract_Tri anomaly in eclblas 8 gadi atpakaļ
make_diag.cpp 937173f4fe HPCC-16571 Mangle the functions exported from eclblas plugin 8 gadi atpakaļ
sourcedoc.xml adae826495 HPCC-16282 Add BLAS wrappers as standard library attributes and plugins 8 gadi atpakaļ

README.md

ECL BLAS Plugin

This is the ECL plugin to utilize the Basic Linear Algebra Sysem (BLAS). It utilizes the C wrappers for BLAS

Installation and Dependencies

To build the plugin with the HPCC-Platform, is required. 

sudo apt-get install atlas-dev

Getting started

The libcblas.so library can be found in several packages including ATLAS and LAPACK.

The Actual Plugin

The eclblas plugin features are exposed via the ECL Standard library Std.BLAS module.

###An ECL Example

IMPORT Std.BLAS AS BLAS;
IMPORT BLAS.Types AS Types;
Types.matrix_t init1 := [1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0];
Types.matrix_t init2 := [9.0, 8.0, 7.0, 6.0, 5.0, 4.0, 3.0, 2.0, 1.0];
Types.matrix_t init3 := [1.0, 2.0, 3.0];
Types.matrix_t init4 := [2.0, 2.0, 2.0];

Test1_mat := BLAS.dgemm(FALSE, TRUE, 3, 3, 1, 1.0, init3, init4);
Test2_mat := BLAS.dgemm(TRUE, FALSE, 1, 1, 3, 1.0, init3, init4);
Test3_mat := BLAS.dgemm(FALSE, TRUE, 3, 1, 3, -1.0, init1, init4);
Test4_mat := BLAS.dgemm(FALSE, TRUE, 3, 1, 3, 1.0, init1, init4, 5, init3);
OUTPUT(Test1_mat);
OUTPUT(Test2_mat);
OUTPUT(Test3_mat);
OUTPUT(test4_mat);

Yields the various matrix products.

Behavior and Implementation Details

All of the arrays are column major. The underlying Fortran code expects column major, so we do not need the library to interpret the data as row major (usual C convention).