Differences

This shows you the differences between two versions of the page.

--- technical:whitepaper:r-runtime-blas-lapack [2018-12-10 11:36] – [Substituting an alternate library] frey
+++ technical:whitepaper:r-runtime-blas-lapack [2018-12-10 12:07] – frey
@@ Line 30: / Line 30: @@
 A simple way to organize multiple underlying BLAS/LAPACK libraries in a single R installation is to create subdirectories for each variant:
-^  Path^Description^
+^Path  ^Description^
-|  ''**R_PREFIX**/lib64/R/lib''|base directory where R looks for shared libraries by default|
+|''**R_PREFIX**/lib64/R/lib''  |base directory where R looks for shared libraries by default|
-|  ''**R_PREFIX**/lib64/R/lib/atlas''|directory to hold ''libatlas.so''|
+|''**R_PREFIX**/lib64/R/lib/libRblas.so''  |symlink to chosen BLAS library (from one of the subdirectories herein)|
-|  ''**R_PREFIX**/lib64/R/lib/rblas''|directory to hold the bundled ''libRblas.so'' and ''libRlapack.so'' produced by R build procedure|
+|''**R_PREFIX**/lib64/R/lib/libRlapack.so''  |symlink to chosen LAPACK library (from one of the subdirectories herein)|
-|  ''**R_PREFIX**/lib64/R/lib/mkl''|directory to hold MKL variants|
+|''**R_PREFIX**/lib64/R/lib/atlas''  |directory to hold ''libatlas.so''|
-|  ''**R_PREFIX**/lib64/R/lib/mkl/seq''|directory to hold sequential MKL variant|
+|''**R_PREFIX**/lib64/R/lib/rblas''  |directory to hold the bundled ''libRblas.so'' and ''libRlapack.so'' produced by R build procedure|
-|  ''**R_PREFIX**/lib64/R/lib/mkl/thr''|directory to hold threaded MKL variant|
+|''**R_PREFIX**/lib64/R/lib/mkl''  |directory to hold MKL variants|
+|''**R_PREFIX**/lib64/R/lib/mkl/seq''  |directory to hold sequential MKL variant|
+|''**R_PREFIX**/lib64/R/lib/mkl/thr''  |directory to hold threaded MKL variant|
+==== R BLAS/LAPACK ====
+When we restructured the R ''lib64/R/lib'' directory, the bundled ''libRblas.so'' and ''libRlapack.so'' shared library files were moved to the ''rblas'' subdirectory.  To configure R to use its bundled libraries:
+<code bash>
+$ cd ${R_PREFIX}/lib64/R/lib
+$ rm -f libR{blas,lapack}.so
+$ ln -s rblas/libRblas.so .
+$ ln -s rblas/libRlapack.so .
+</code>
+==== ATLAS ====
+The ATLAS library contains both BLAS and LAPACK APIs in a single shared library.  With ''libatlas.so'' copied into the ''**R_PREFIX**/lib64/R/lib/atlas'' subdirectory, we configure R to use ATLAS:
+<code bash>
+$ cd ${R_PREFIX}/lib64/R/lib
+$ rm -f libR{blas,lapack}.so
+$ ln -s atlas/libRblas.so .
+$ ln -s atlas/libRlapack.so .
+</code>
+==== Sequential MKL ====
+A C source file containing a dummy function was created in ''**R_PREFIX**/lib64/R/lib/mkl/seq'':
+<file C shim.c>
+int
+mkl_shim_dummy(void)
+{
+	return 0;
+}
+</file>
+The shim library is then created thusly:
+<code bash>
+$ cd ${R_PREFIX}/lib64/R/lib/mkl/seq
+$ icc -shared -o libRblas.so -mkl=sequential shim.c
+$ ln -s libRblas.so libRlapack.so
+</code>
+To configure R to use the sequential MKL:
+<code bash>
+$ cd ${R_PREFIX}/lib64/R/lib
+$ rm -f libR{blas,lapack}.so
+$ ln -s mkl/seq/libRblas.so .
+$ ln -s mkl/seq/libRlapack.so .
+</code>
+==== Threaded MKL ====
+A C source file containing a dummy function was created in ''**R_PREFIX**/lib64/R/lib/mkl/thr'':
+<file C shim.c>
+int
+mkl_shim_dummy(void)
+{
+	return 0;
+}
+</file>
+Since our R build used the GNU C compiler, the threaded MKL variant only works if the shim library is built against the GNU OpenMP runtime.  Using just "-mkl=parallel" links against the Intel OpenMP runtime which in testing yielded numerical issues (not actual crashes).  The shim library is then created thusly:
+<code bash>
+$ cd ${R_PREFIX}/lib64/R/lib/mkl/thr
+$ icc -shared -o libRblas.so shim.c -lmkl_gnu_thread -lmkl_core -lmkl_intel_lp64
+$ ln -s libRblas.so libRlapack.so
+</code>
+To configure R to use the threaded MKL:
+<code bash>
+$ cd ${R_PREFIX}/lib64/R/lib
+$ rm -f libR{blas,lapack}.so
+$ ln -s mkl/thr/libRblas.so .
+$ ln -s mkl/thr/libRlapack.so .
+</code>
 ===== Runtime-configurable substitution =====
+By stashing each BLAS/LAPACK variant in its own subdirectory our copy of R is actually fairly close to being