From 13a8026acd60e9d4f9024edfe69cb9a2371b233d Mon Sep 17 00:00:00 2001
From: Paul Baumeister <p.baumeister@fz-juelich.de>
Date: Fri, 17 May 2019 13:52:59 +0200
Subject: [PATCH] prepare for a first GPU-accelerated version, make
tfQMRgpu=yes
---
.../source/IterativeSolver/KKRmat_mod.F90 | 24 +++++++++++++++++++
source/KKRnano/source/Makefile | 16 +++++++++++--
2 files changed, 38 insertions(+), 2 deletions(-)
diff --git a/source/KKRnano/source/IterativeSolver/KKRmat_mod.F90 b/source/KKRnano/source/IterativeSolver/KKRmat_mod.F90
index 785b690e2..73a40cee9 100644
--- a/source/KKRnano/source/IterativeSolver/KKRmat_mod.F90
+++ b/source/KKRnano/source/IterativeSolver/KKRmat_mod.F90
@@ -219,6 +219,9 @@ module KKRmat_mod
use fillKKRMatrix_mod, only: dump
use IterativeSolver_mod, only: IterativeSolver, solve
use DirectSolver_mod, only: DirectSolver, solve
+#ifdef has_tfQMRgpu
+ ! use tfqmrgpu, only: tfqmrgpu_bsrsv_complete ! all-in-one GPU solver interface for rapid integration
+#endif
use SparseMatrixDescription_mod, only: dump
use InitialGuess_mod, only: InitialGuess, load, store
use KKROperator_mod, only: KKROperator
@@ -257,6 +260,13 @@ module KKRmat_mod
double complex, allocatable :: dPdE_local(:,:,:), gllke_x(:,:), dgde(:,:), MinvdMdE(:,:,:), TinvMinvdMdE(:,:,:) ! LLY
double complex :: tracek ! LLY
+#ifdef has_tfQMRgpu
+ external :: tfqmrgpu_bsrsv_complete ! subroutine
+ integer :: o = 6
+ integer :: ierr = 1
+ integer :: iterations, lda
+ double precision :: residuum
+#endif
integer :: num_trunc_atoms, lmsd, lm1, idx_lly, i1
double complex :: cfctorinv
@@ -383,6 +393,20 @@ module KKRmat_mod
! store the initial guess
call store(iguess_data, op%mat_X, ik=ikpoint, is=ispin, ie=ienergy)
+#ifdef has_tfQMRgpu
+ case (5) ! GPU solver
+
+ iterations = 1000
+ residuum = 1e-7
+ lda = size(op%mat_A, 1)
+
+ call tfqmrgpu_bsrsv_complete(op%bsr_A%nRows, lda, &
+ op%bsr_A%RowStart, op%bsr_A%ColIndex, op%mat_A(:,:,:,0), 'n', & !! A (in)
+ op%bsr_X%RowStart, op%bsr_X%ColIndex, op%mat_X, 'n', & !! X (out)
+ op%bsr_B%RowStart, op%bsr_B%ColIndex, op%mat_B, 'n', & !! B (in)
+ iterations, residuum, o, ierr)
+#endif
+
case default
warn(6, "No solver selected! Problem is not solved, solver_type ="+solver_type)
endselect ! solver_type
diff --git a/source/KKRnano/source/Makefile b/source/KKRnano/source/Makefile
index 05f24378b..25e3c764e 100644
--- a/source/KKRnano/source/Makefile
+++ b/source/KKRnano/source/Makefile
@@ -7,12 +7,19 @@ TYPE ?= nodebug
### the user may specify SMP=openmp as a command line argument to make
SMP ?= none
+### can we make use of the tfQMRgpu library
+tfQMRgpu ?= no
+
PROGRAM = kkr.exe
# Path to put object files and module files
BUILDDIR = $(HOME)/build
-EXTRA_FLAGS =
+EXTRA_FLAGS ?=
+
+ifeq ($(tfQMRgpu),yes)
+ EXTRA_FLAGS += -D has_tfQMRgpu
+endif
FC90FLAGS =
FCFLAGS =
PPFLAGS =
@@ -375,6 +382,11 @@ OBJS += $(notdir ${SRCS90:.f90=.o})
OBJS += $(notdir ${SRCSFPP:.F=.o})
OBJS += $(notdir ${SRCS90FPP:.F90=.o})
+ifeq ($(tfQMRgpu),yes)
+ OBJS += tfqmrgpu_Fortran.o tfqmrgpu_Fortran_wrappers.o tfqmrgpu.o
+ LDFLAGS += -L/usr/local/cuda/lib64 -lcudart -lnvToolsExt -lm -lcurand
+endif
+
.PHONY: all
all: $(PROGRAM)
@@ -521,7 +533,7 @@ bsrmm_mod.o: CacheOverlap_mod.o
CacheOverlap_mod.o:
ChebMeshData_mod.o: InputParams_mod.o RadialMeshData_mod.o
Truncation_mod.o: Logging_mod.o Exceptions_mod.o TruncationZone_mod.o
-NonCollinearMagnetism_mod.o: RadialMeshData_mod.o ChebMeshData_mod.o
+NonCollinearMagnetism_mod.o: RadialMeshData_mod.o ChebMeshData_mod.o read_formatted_shapefun_mod.o
NonCollinearMagnetismData_mod.o: Exceptions_mod.o
# DO NOT DELETE
--
GitLab