TREX/qmckl
1
0
Fork 0
mirror of https://github.com/TREX-CoE/qmckl.git synced 2024-07-23 11:17:44 +02:00
Code Issues Projects Releases Wiki Activity

Started adding the pedagogical kernels for the HAVE_DOC builds.

Browse Source
This commit is contained in:
Francois Coppens 2023-02-02 17:04:34 +01:00
parent 2e45927e04
commit d3aebe52ff
1 changed files with 366 additions and 290 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

302
org/qmckl_sherman_morrison_woodbury.org
View File

@ -9,18 +9,18 @@
* Headers * Headers
#+begin_src elisp :noexport :results none :exports none #+begin_src elisp :noexport :results none :exports none
(org-babel-lob-ingest "../tools/lib.org") (org-babel-lob-ingest "../tools/lib.org")
#+end_src #+end_src
#+begin_src c :comments link :tangle (eval c_test) :noweb yes #+begin_src c :comments link :tangle (eval c_test) :noweb yes
#include "qmckl.h" #include "qmckl.h"
#include "assert.h" #include "assert.h"
#ifdef HAVE_CONFIG_H #ifdef HAVE_CONFIG_H
#include "config.h" #include "config.h"
#endif #endif
#include <math.h> #include <math.h>
int main() { int main() {
qmckl_context context; qmckl_context context;
context = qmckl_context_create(); context = qmckl_context_create();
qmckl_exit_code rc; qmckl_exit_code rc;
@ -28,11 +28,10 @@ int main() {
#+NAME:kernel_generator_range #+NAME:kernel_generator_range
#+begin_src python :noweb yes :exports none #+begin_src python :noweb yes :exports none
range(2, 22) range(2, 22)
#+end_src #+end_src
* Naïve Sherman-Morrison * Naïve Sherman-Morrison
** ~qmckl_sherman_morrison_naive~ ** ~qmckl_sherman_morrison_naive~
:PROPERTIES: :PROPERTIES:
:Name: qmckl_sherman_morrison_naive :Name: qmckl_sherman_morrison_naive
@ -90,6 +89,31 @@ range(2, 22)
* ~breakdown~ is a small number such that $0 < breakdown << 1$ * ~breakdown~ is a small number such that $0 < breakdown << 1$
* ~Slater_inv~ is allocated with $Dim \times Dim$ elements * ~Slater_inv~ is allocated with $Dim \times Dim$ elements
*** Fortran source
#+begin_src f90 :tangle (eval f)
integer function qmckl_sherman_morrison_naive_doc_f &
(context, Dim, N_updates, Updates, Updates_index, breakdown, Slater_inv, determinant) &
bind(C)
use, intrinsic :: iso_c_binding
implicit none
integer (c_int64_t) , intent(in) , value :: context
integer (c_int64_t) , intent(in) , value :: Dim
integer (c_int64_t) , intent(in) , value :: N_updates
real (c_double ) , intent(in) :: Updates(N_updates*Dim)
integer (c_int64_t) , intent(in) :: Updates_index(N_updates)
real (c_double ) , intent(in) , value :: breakdown
real (c_double ) , intent(inout) :: Slater_inv(Dim*Dim)
real (c_double ) , intent(inout) :: determinant
write(*,*) "Function 'qmckl_sherman_morrison_naive_doc_f' does nothing for now..."
end function qmckl_sherman_morrison_naive_doc_f
#+end_src
*** C header *** C header
#+CALL: generate_c_header(table=qmckl_sherman_morrison_naive_args,rettyp=get_value("CRetType"),fname=get_value("Name")) #+CALL: generate_c_header(table=qmckl_sherman_morrison_naive_args,rettyp=get_value("CRetType"),fname=get_value("Name"))
@ -108,34 +132,44 @@ range(2, 22)
double* determinant); double* determinant);
#+end_src #+end_src
#+begin_src c :tangle (eval h_func) :comments org
qmckl_exit_code qmckl_sherman_morrison_naive_doc(
const qmckl_context context,
const uint64_t Dim,
const uint64_t N_updates,
const double* Updates,
const uint64_t* Updates_index,
const double breakdown,
double* Slater_inv,
double* determinant);
#+end_src
*** C source
#+begin_src c :tangle (eval c) :comments org #+begin_src c :tangle (eval c) :comments org
#include <stdbool.h> #include <stdbool.h>
#include <math.h> #include <math.h>
#include "qmckl.h" #include "qmckl.h"
#include "config.h" #include "config.h"
// Order important because // Order important because
// __GNUC__ also set in ICC, ICX and CLANG // __GNUC__ also set in ICC, ICX and CLANG
// __clang__ also set in ICX // __clang__ also set in ICX
#if defined(__INTEL_COMPILER) #if defined(__INTEL_COMPILER)
#define IVDEP _Pragma("ivdep") #define IVDEP _Pragma("ivdep")
#define ALIGNED _Pragma("vector aligned") #define ALIGNED _Pragma("vector aligned")
#elif defined(__INTEL_LLVM_COMPILER) #elif defined(__INTEL_LLVM_COMPILER)
#define IVDEP _Pragma("ivdep") #define IVDEP _Pragma("ivdep")
#define ALIGNED _Pragma("vector aligned") #define ALIGNED _Pragma("vector aligned")
#elif defined(__clang__) #elif defined(__clang__)
#define IVDEP _Pragma("clang loop vectorize(enable)") #define IVDEP _Pragma("clang loop vectorize(enable)")
#define ALIGNED #define ALIGNED
#elif defined(__GNUC__) #elif defined(__GNUC__)
#define IVDEP _Pragma("GCC ivdep") #define IVDEP _Pragma("GCC ivdep")
#define ALIGNED #define ALIGNED
#endif #endif
qmckl_exit_code qmckl_sherman_morrison_naive_hpc( qmckl_exit_code qmckl_sherman_morrison_naive_hpc(
const qmckl_context context, const qmckl_context context,
const uint64_t LDS, const uint64_t LDS,
const uint64_t Dim, const uint64_t Dim,
@ -173,9 +207,9 @@ qmckl_exit_code qmckl_sherman_morrison_naive_hpc(
const int cui = Updates_index[l] - 1; const int cui = Updates_index[l] - 1;
double den = 1.0f + C[cui]; double den = 1.0f + C[cui];
if (fabs(den) < breakdown) { if (fabs(den) < breakdown)
return QMCKL_FAILURE; return QMCKL_FAILURE;
}
double iden = 1.0f / den; double iden = 1.0f / den;
// Update det(A) // Update det(A)
@ -201,16 +235,12 @@ qmckl_exit_code qmckl_sherman_morrison_naive_hpc(
l += 1; l += 1;
} }
return QMCKL_SUCCESS; return QMCKL_SUCCESS;
} }
#+end_src #+end_src
#+NAME:naive_template_code #+NAME:naive_template_code
#+begin_src c #+begin_src c
static inline qmckl_exit_code qmckl_sherman_morrison_naive_{Dim}( static inline qmckl_exit_code qmckl_sherman_morrison_naive_{Dim}(
const qmckl_context context, const qmckl_context context,
const uint64_t N_updates, const uint64_t N_updates,
const double* __restrict Updates, const double* __restrict Updates,
@ -220,7 +250,7 @@ static inline qmckl_exit_code qmckl_sherman_morrison_naive_{Dim}(
double* __restrict determinant) { double* __restrict determinant) {
if (qmckl_context_check(context) == QMCKL_NULL_CONTEXT) { if (qmckl_context_check(context) == QMCKL_NULL_CONTEXT) {
return qmckl_failwith( context, return qmckl_failwith(context,
QMCKL_NULL_CONTEXT, QMCKL_NULL_CONTEXT,
"qmckl_sherman_morrison_naive_{Dim}", "qmckl_sherman_morrison_naive_{Dim}",
NULL); NULL);
@ -278,13 +308,9 @@ static inline qmckl_exit_code qmckl_sherman_morrison_naive_{Dim}(
} }
return QMCKL_SUCCESS; return QMCKL_SUCCESS;
} }
#+end_src #+end_src
#+NAME:naive_kernel_generator #+NAME:naive_kernel_generator
#+begin_src python :noweb yes :exports none #+begin_src python :noweb yes :exports none
text=""" text="""
@ -298,10 +324,6 @@ for Dim in <<kernel_generator_range>>:
return '\n'.join(result) return '\n'.join(result)
#+end_src #+end_src
#+NAME:naive_switch-case_generator #+NAME:naive_switch-case_generator
#+begin_src python :noweb yes :exports none #+begin_src python :noweb yes :exports none
text=""" text="""
@ -322,10 +344,6 @@ for Dim in <<kernel_generator_range>>:
return '\n'.join(result) return '\n'.join(result)
#+end_src #+end_src
#+begin_src c :tangle (eval c) :comments org :noweb yes #+begin_src c :tangle (eval c) :comments org :noweb yes
<<naive_kernel_generator()>> <<naive_kernel_generator()>>
@ -346,6 +364,7 @@ qmckl_exit_code qmckl_sherman_morrison_naive(const qmckl_context context,
NULL); NULL);
} }
#ifdef HAVE_HPC
if (LDS == (1+(Dim-1)/SIMD_LENGTH)*SIMD_LENGTH) { // Most cases if (LDS == (1+(Dim-1)/SIMD_LENGTH)*SIMD_LENGTH) { // Most cases
switch (Dim) { switch (Dim) {
<<naive_switch-case_generator()>> <<naive_switch-case_generator()>>
@ -363,20 +382,115 @@ qmckl_exit_code qmckl_sherman_morrison_naive(const qmckl_context context,
determinant); determinant);
} }
#else
return qmckl_sherman_morrison_naive_doc(context,
Dim,
N_updates,
Updates,
Updates_index,
breakdown,
Slater_inv,
determinant);
#endif
return QMCKL_FAILURE; return QMCKL_FAILURE;
} }
#+end_src #+end_src
*** Test :noexport:
The tests for the kernels are executed on datasets that are extracted from a run of
QMC=Chem on Benzene (21 spin-up/21 spin down electrons) using 329 unique alpha determinants.
The tests are run such that the kernels reject the computed inverse whenever the computed
intermediate determinants or denominators are smaller than 1e-3. This is the default value in
QMC=Chem. The tests will return QMCKL_SUCCESS whenever all the elements of the final matrix
$R=S.S^-1 - 1$ are smaller than the given tolerance value of 1e-3, and will return
QMCKL_FAILURE if the values are larger than this tolerance value.
#+begin_src c :tangle (eval c_test)
const uint64_t Dim = 21;
const uint64_t LDS = (1+(Dim-1)/SIMD_LENGTH)*SIMD_LENGTH;
const double breakdown = 1e-3;
const double tolerance = 1e-3;
double res[441];
#include "sm_test.h"
*** Performance assert(Updates1 != NULL);
assert(Updates_index1 != NULL);
assert(Slater_inv1 != NULL);
This function performs best when there is only 1 rank-1 update in the update cycle. It is not useful to // original determinant of Slater1 (before applying updates)
use Sherman-Morrison with update splitting for these cycles since splitting can never resolve a situation double det = 3.407025646103221e-10;
where applying the update causes singular behaviour. rc = qmckl_sherman_morrison_naive(context,
LDS,
Dim,
N_updates1,
Updates1,
Updates_index1,
breakdown,
Slater_inv1,
&det);
// Check that the determinant is updated properly
assert(fabs(det + 4.120398385068217e-10) < 1e-15);
for (unsigned int i = 0; i < Dim; i++) {
for (unsigned int j = 0; j < Dim; j++) {
res[i * Dim + j] = 0;
for (unsigned int k = 0; k < Dim; k++) {
res[i * Dim + j] += Slater1[i * Dim + k] * Slater_inv1[k * LDS + j];
}
}
}
rc = QMCKL_SUCCESS;
for (unsigned int i = 0; i < Dim; i++) {
for (unsigned int j = 0; j < Dim; j++) {
if (i == j && fabs(res[i * Dim + j] - 1) > tolerance) {
rc = QMCKL_FAILURE;
}
if (i != j && fabs(res[i * Dim + j]) > tolerance) {
rc = QMCKL_FAILURE;
}
}
}
assert(rc == QMCKL_SUCCESS);
#+end_src
** Performance
This function performs best when there is only 1 rank-1 update in the update cycle. It is
not useful to use Sherman-Morrison with update splitting for these cycles since splitting
can never resolve a situation where applying the update causes singular behaviour.
** C interface
#+begin_src f90 :tangle (eval f) :comments org :exports none
interface
integer(c_int32_t) function qmckl_sherman_morrison_naive_doc &
(context, Dim, N_updates, Updates, Updates_index, breakdown, Slater_inv, determinant) &
bind(C)
use, intrinsic :: iso_c_binding
import
implicit none
integer (c_int64_t) , intent(in) , value :: context
integer (c_int64_t) , intent(in) , value :: Dim
integer (c_int64_t) , intent(in) , value :: N_updates
real (c_double ) , intent(in) :: Updates(N_updates*Dim)
integer (c_int64_t) , intent(in) :: Updates_index(N_updates)
real (c_double ) , intent(in) , value :: breakdown
real (c_double ) , intent(inout) :: Slater_inv(Dim*Dim)
real (c_double ) , intent(inout) :: determinant
integer(c_int32_t), external :: qmckl_sherman_morrison_naive_doc_f
info = qmckl_sherman_morrison_naive_doc_f &
(context, Dim, N_updates, Updates, &
Updates_index, breakdown, Slater_inv, determinant)
end function qmckl_sherman_morrison_naive_doc
end interface
#+end_src
** Fortran interface :noexport: ** Fortran interface :noexport:
:PROPERTIES: :PROPERTIES:
@ -412,63 +526,31 @@ qmckl_exit_code qmckl_sherman_morrison_naive(const qmckl_context context,
end interface end interface
#+end_src #+end_src
*** Test :noexport: #+begin_src f90 :tangle (eval fh_func) :comments org :exports none
interface
integer(c_int32_t) function qmckl_sherman_morrison_naive_doc &
(context, Dim, N_updates, Updates, Updates_index, breakdown, Slater_inv, determinant) &
bind(C)
The tests for the kernels are executed on datasets that are extracted from a run of QMC=Chem on Benzene (21 spin-up/21 spin down electrons) using 329 unique alpha determinants. The tests are run such that the kernels reject the computed inverse whenever the computed intermediate determinants or denominators are smaller than 1e-3. This is the default value in QMC=Chem. The tests will return QMCKL_SUCCESS whenever all the elements of the final matrix $R=S.S^-1 - 1$ are smaller than the given tolerance value of 1e-3, and will return QMCKL_FAILURE if the values are larger than this tolerance value. use, intrinsic :: iso_c_binding
import
implicit none
#+begin_src c :tangle (eval c_test) integer (c_int64_t) , intent(in) , value :: context
const uint64_t Dim = 21; integer (c_int64_t) , intent(in) , value :: Dim
const uint64_t LDS = (1+(Dim-1)/SIMD_LENGTH)*SIMD_LENGTH; integer (c_int64_t) , intent(in) , value :: N_updates
const double breakdown = 1e-3; real (c_double ) , intent(in) :: Updates(N_updates*Dim)
const double tolerance = 1e-3; integer (c_int64_t) , intent(in) :: Updates_index(N_updates)
double res[441]; real (c_double ) , intent(in) , value :: breakdown
real (c_double ) , intent(inout) :: Slater_inv(Dim*Dim)
real (c_double ) , intent(inout) :: determinant
#include "sm_test.h" end function qmckl_sherman_morrison_naive_doc
end interface
assert(Updates1 != NULL);
assert(Updates_index1 != NULL);
assert(Slater_inv1 != NULL);
// original determinant of Slater1 (before applying updates)
double det = 3.407025646103221e-10;
rc = qmckl_sherman_morrison_naive(context,
LDS,
Dim,
N_updates1,
Updates1,
Updates_index1,
breakdown,
Slater_inv1,
&det);
// Check that the determinant is updated properly
assert(fabs(det + 4.120398385068217e-10) < 1e-15);
for (unsigned int i = 0; i < Dim; i++) {
for (unsigned int j = 0; j < Dim; j++) {
res[i * Dim + j] = 0;
for (unsigned int k = 0; k < Dim; k++) {
res[i * Dim + j] += Slater1[i * Dim + k] * Slater_inv1[k * LDS + j];
}
}
}
rc = QMCKL_SUCCESS;
for (unsigned int i = 0; i < Dim; i++) {
for (unsigned int j = 0; j < Dim; j++) {
if (i == j && fabs(res[i * Dim + j] - 1) > tolerance) {
rc = QMCKL_FAILURE;
}
if (i != j && fabs(res[i * Dim + j]) > tolerance) {
rc = QMCKL_FAILURE;
}
}
}
assert(rc == QMCKL_SUCCESS);
#+end_src #+end_src
* Woodbury 2x2 * Woodbury 2x2
** ~qmckl_woodbury_2x2~
** ~qmckl_woodbury_2x2~
:PROPERTIES: :PROPERTIES:
:Name: qmckl_woodbury_2x2 :Name: qmckl_woodbury_2x2
:CRetType: qmckl_exit_code :CRetType: qmckl_exit_code
@ -790,13 +872,12 @@ qmckl_exit_code qmckl_woodbury_2x2(const qmckl_context context,
*** Performance *** Performance
This function is most efficient when used in cases where there are only 2 rank-1 updates and This function is most efficient when used in cases where there are only 2 rank-1 updates and
it is sure they will not result in a singular matrix. it is sure they will not result in a singular matrix.
** Fortran interface :noexport: ** Fortran interface :noexport:
:PROPERTIES: :PROPERTIES:
:Name: qmckl_woodbury_2x2 :Name: qmckl_woodbury_2x2
:CRetType: qmckl_exit_code :CRetType: qmckl_exit_code
@ -860,7 +941,6 @@ assert(rc == QMCKL_SUCCESS);
#+end_src #+end_src
* Woodbury 3x3 * Woodbury 3x3
** ~qmckl_woodbury_3x3~ ** ~qmckl_woodbury_3x3~
:PROPERTIES: :PROPERTIES:
:Name: qmckl_woodbury_3x3 :Name: qmckl_woodbury_3x3
@ -1209,13 +1289,12 @@ qmckl_exit_code qmckl_woodbury_3x3(const qmckl_context context,
#+end_src #+end_src
*** Performance... *** Performance...
This function is most efficient when used in cases where there are only 3 rank-1 updates and This function is most efficient when used in cases where there are only 3 rank-1 updates and
it is sure they will not result in a singular matrix. it is sure they will not result in a singular matrix.
** Fortran interface :noexport: ** Fortran interface :noexport:
:PROPERTIES: :PROPERTIES:
:Name: qmckl_woodbury_3x3 :Name: qmckl_woodbury_3x3
:CRetType: qmckl_exit_code :CRetType: qmckl_exit_code
@ -1279,7 +1358,6 @@ assert(rc == QMCKL_SUCCESS);
#+end_src #+end_src
* Sherman-Morrison with update splitting * Sherman-Morrison with update splitting
** ~qmckl_sherman_morrison_splitting~ ** ~qmckl_sherman_morrison_splitting~
:PROPERTIES: :PROPERTIES:
:Name: qmckl_sherman_morrison_splitting :Name: qmckl_sherman_morrison_splitting
@ -1314,7 +1392,6 @@ assert(rc == QMCKL_SUCCESS);
If the determinant of the Slater-matrix is passed, it will be updated to the determinant resulting If the determinant of the Slater-matrix is passed, it will be updated to the determinant resulting
from applying the updates to the original matrix. from applying the updates to the original matrix.
*** Requirements *** Requirements
* ~context~ is not ~QMCKL_NULL_CONTEXT~ * ~context~ is not ~QMCKL_NULL_CONTEXT~
@ -1454,8 +1531,7 @@ assert(rc == QMCKL_SUCCESS);
#+end_src #+end_src
* Woodbury 3x3 and 2x2 with Sherman-Morrison and update splitting * Woodbury 3x3 and 2x2 with Sherman-Morrison and update splitting
** ~qmckl_sherman_morrison_smw32s~
** ~qmckl_sherman_morrison_smw32s~
:PROPERTIES: :PROPERTIES:
:Name: qmckl_sherman_morrison_smw32s :Name: qmckl_sherman_morrison_smw32s
:CRetType: qmckl_exit_code :CRetType: qmckl_exit_code
@ -1639,7 +1715,7 @@ qmckl_exit_code qmckl_sherman_morrison_smw32s(const qmckl_context context,
This kernel performs best for update cycles with 2 or more rank-1 updates and the fail-rate is low. This kernel performs best for update cycles with 2 or more rank-1 updates and the fail-rate is low.
** Fortran interface :noexport: ** Fortran interface :noexport:
:PROPERTIES: :PROPERTIES:
:Name: qmckl_sherman_morrison_smw32s :Name: qmckl_sherman_morrison_smw32s
:CRetType: qmckl_exit_code :CRetType: qmckl_exit_code