TREX/qmckl
1
0
Fork 0
mirror of https://github.com/TREX-CoE/qmckl.git synced 2024-11-19 20:42:50 +01:00
Code Issues Projects Releases Wiki Activity

Adding documentation to ORG file.

Browse Source
This commit is contained in:
Francois Coppens 2023-02-13 17:44:11 +01:00
parent c0d4f766b1
commit 707fa17e09
1 changed files with 18 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

22
org/qmckl_sherman_morrison_woodbury.org
View File

@ -81,6 +81,9 @@ from applying the updates to the original matrix.
| ~determinant~ | ~double~ | inout | Determinant of the Slater-matrix |
** Pedagogical kernel source (in Fortran)
The following source code written in Fortran is inteded to illustrate how the kernel works. Even though the kernel is
able to do numerically correct computations, it does not do it in the most efficient way possible. It should therefore
not be used in real workloads.
#+begin_src f90 :tangle (eval f)
integer function qmckl_sherman_morrison_naive_doc_f(context, &
@ -118,6 +121,8 @@ end function qmckl_sherman_morrison_naive_doc_f
#+end_src
*** C interface to the pedagogical kernel
The following interface block in Fortran makes sure that the pedagogical kernel,
written in Fortran, can be called from C using the ~ISO_C_BINDING~.
#+CALL: generate_c_interface(table=qmckl_sherman_morrison_naive_args,rettyp=get_value("CRetType"),fname="qmckl_sherman_morrison_naive_doc")
@ -210,7 +215,7 @@ qmckl_exit_code qmckl_sherman_morrison_naive_doc (
#+end_src
** C sources
Common includes and macros used by all the Sherman-Morrison-Woodbury kernels.
#+begin_src c :tangle (eval c) :comments org
#include <stdbool.h>
#include <math.h>
@ -235,6 +240,15 @@ qmckl_exit_code qmckl_sherman_morrison_naive_doc (
#endif
#+end_src
~qmckl_sherman_morrison_naive_hpc~ is a high performance variation of
~qmckl_sherman_morrison_naive~ written in C. It is used in cases when ~Dim~ is
smaller than the leading dimension ~LDS~, irrespective of whetether ~LDS~
includes zero padding to benefit from SIMD instructions or not. Cases like this
include situations where one wants to apply updates to a square submatrix of the
full matrix.
It takes advantage of memory aligned data and assumes no data dependencies
inside the loops. The loops are fully vectorised whenever ~Dim~ is an integer
multiple of ~SIMD_LEGTH~.
#+begin_src c :tangle (eval c) :comments org
qmckl_exit_code qmckl_sherman_morrison_naive_hpc(
const qmckl_context context,
@ -265,7 +279,7 @@ qmckl_exit_code qmckl_sherman_morrison_naive_hpc(
C[i] = 0.0f;
IVDEP
ALIGNED
for (uint64_t j = 0; j < LDS; j++) {
for (uint64_t j = 0; j < Dim; j++) {
C[i] += Slater_inv[i * LDS + j] * Updates[l * LDS + j];
}
}
@ -286,7 +300,7 @@ qmckl_exit_code qmckl_sherman_morrison_naive_hpc(
// selecting column: v_l^T * S_inv
IVDEP
ALIGNED
for (uint64_t j = 0; j < LDS; j++) {
for (uint64_t j = 0; j < Dim; j++) {
D[j] = Slater_inv[cui * LDS + j];
}
@ -294,7 +308,7 @@ qmckl_exit_code qmckl_sherman_morrison_naive_hpc(
for (uint64_t i = 0; i < Dim; i++) {
IVDEP
ALIGNED
for (uint64_t j = 0; j < LDS; j++) {
for (uint64_t j = 0; j < Dim; j++) {
const double update = C[i] * D[j] * iden;
Slater_inv[i * LDS + j] -= update;
}