cuda porting

README.md

@@ -1,57 +1,51 @@
 # ASTR code 
 Version 2.0 
 
-ASTR code is a high-order finite-difference flow solver for compressible turbulence research.
+ASTR code is a high-order finite-difference flow solver for compressible turbulence research. This project explores the usage of CUDA-Fortran to parallelise the ASTR code. The tgv solver for a 3D case has been parallelised using CUDA.
 
 # Download, Installation and Compilation
-Required dependencies: Fortran 90, MPI, HDF5
+Required dependencies: Fortran 90, NVIDIA HPC SDK, CMAKE
 
-## Download the astr code:
+The installation guide for NVIDIA HPC SDK can be found at [Installation Guide](https://docs.nvidia.com/hpc-sdk/hpc-sdk-install-guide/index.html)
 
-git clone git@github.com:astr-code/astr.git
+## Download the gpu accelerated astr code:
 
-## Compilation and installation:
-ASTR 2.0 supports both gnu make and cmake.
-For the use of gnu make, do:
-make 
-in the directory containing src folder, and the executable will be found as ./bin/astr 
+Clone the git repository:
+```
+$ git clone https://github.com/terencel411/astr.git
+```
+## Compilation:
+The `Makefile` gives a complete and safe way of compiling the code.
 
-The cmake gives a more complete and safe way of compiling and installing the code.
-By default ASTR solves equations under non-dimensional form, unless the chemstry is included.
+Go to the directory where the miniapps code is present
 
-create a case folder, e.g.
-mkdir test_case
-sh path_to_the_source/script/astr.case.creater #create a new case
+```
+$ cd astr/miniapps/tgv_solver_3d
+```
 
-cmake path_to_the_source
-cmake --build 
-cmake --install 
-ctest -L nondim
+The cpu and the gpu accelerated codes (`tgvsolver_cpu.f90` and `tgvsolver_gpu.cuf`) are present in the same directory, which can be compiled using the following cmake commands
 
-The code will be installed in test_case and excutable can be found at test_case/bin/astr
+Compile and execute the cpu & gpu code
 
-If you want to use the chemstry function, you need first to install cantera. After download and unpack the cantera, you can use the following script to install:
-python scons/scripts/scons.py build python_package=none FORTRAN=<your fortran compiler> f90_interface=y prefix=<the directory of centera to install> boost_inc_dir=<to boost directory>
+```
+$ cmake all
+```
 
-python scons/scripts/scons.py test
+The cpu and gpu code can also be compiled and executed separately
 
-python scons/scripts/scons.py install
+```
+$ cmake cpu
+$ cmake gpu
+```
 
-you may need to make and test ASTR with chemstry with following cmake commands:
+## Acceleration Comparison
+The time acceleration statistics can be obtained by running the following command. 
 
-cmake -DCHEMISTRY=TRUE -DCANTERA_DIR=path_to_cantera path_to_the_source
+```
+$ cmake compare
+```
 
-cmake --build 
-
-cmake --install 
-
-ctest -L combustion
-
-
-## Run the solver:
-
-Once the excutable is built, a typical simulation can be run as,
-mpirun -np 8 ./astr run ./datin/input_file
+A text file `time_report.txt` will be generated with the accelerations statistics.
 
 
 

miniapps/tgv_solver_3d/Makefile

@@ -0,0 +1,44 @@
+# Compiler
+FC = nvfortran
+
+# Executable files
+EXEC1 = tgv_cpu
+EXEC2 = tgv_gpu
+EXEC3 = time_stats
+
+# Flags
+FLAGS = -O1
+
+SRC1 = tgvsolver_cpu.f90
+SRC2 = tgvsolver_gpu.cuf
+SRC3 = compstats.f90
+
+# all: $(EXEC1) $(EXEC2) $(EXEC3)
+# runs cpu & gpu, then compares
+all: cpu gpu
+
+# Compile
+# nvfortran -O3 tgvsolver.cuf -o tgv
+# $^ is the source file
+# $@ is the exec file
+
+# $(EXEC1): $(SRC1)
+# 	$(FC) $^ -o $@
+
+cpu: $(SRC1)
+	$(FC) $(FLAGS) $^ -o $(EXEC1)
+	./$(EXEC1)
+
+gpu: $(SRC2)
+	$(FC) $(FLAGS) $^ -o $(EXEC2)
+	./$(EXEC2)
+
+compare: $(SRC3)
+	$(FC) $^ -o $(EXEC3)
+	./$(EXEC3)
+
+clean:
+	rm -f $(EXEC1) $(EXEC2) $(EXEC2)
+
+# Phony targets
+.PHONY: all clean cpu gpu compare

miniapps/tgv_solver_3d/comparison.f90

@@ -0,0 +1,75 @@
+program comparison
+    implicit none
+    character(len=200) :: line1, line2
+    integer :: i, ios1, ios2, choice
+    logical :: flag1=.true.
+
+    print *, "Enter your choice: "
+    read(*,*) choice
+
+    if(choice == 1) then
+        open(unit=10, file='gradcal_cpu.txt', status='old', action='read')
+        open(unit=20, file='gradcal_gpu.txt', status='old', action='read')
+        open(unit=30, file='gradcal.txt', status='unknown', action='write')
+        open(unit=40, file='gradcal_ne.txt', status='unknown', action='write')
+    else if(choice == 2) then
+        open(unit=10, file='convection_cpu.txt', status='old', action='read')
+        open(unit=20, file='convection_gpu.txt', status='old', action='read')
+        open(unit=30, file='convection.txt', status='unknown', action='write')
+        open(unit=40, file='convection_ne.txt', status='unknown', action='write')
+    else if(choice == 3) then
+        open(unit=10, file='diffusion_cpu.txt', status='old', action='read')
+        open(unit=20, file='diffusion_gpu.txt', status='old', action='read')
+        open(unit=30, file='diffusion.txt', status='unknown', action='write')
+        open(unit=40, file='diffusion_ne.txt', status='unknown', action='write')
+    else if(choice == 4) then
+        open(unit=10, file='filterq_cpu.txt', status='old', action='read')
+        open(unit=20, file='filterq_gpu.txt', status='old', action='read')
+        open(unit=30, file='filterq.txt', status='unknown', action='write')
+        open(unit=40, file='filterq_ne.txt', status='unknown', action='write')
+    else if(choice == 5) then
+        open(unit=10, file='q2fvar_cpu.txt', status='old', action='read')
+        open(unit=20, file='q2fvar_gpu.txt', status='old', action='read')
+        open(unit=30, file='q2fvar.txt', status='unknown', action='write')
+        open(unit=40, file='q2fvar_ne.txt', status='unknown', action='write')
+    else if(choice == 6) then
+        open(unit=10, file='bchomo_cpu.txt', status='old', action='read')
+        open(unit=20, file='bchomo_gpu.txt', status='old', action='read')
+        open(unit=30, file='bchomo.txt', status='unknown', action='write')
+        open(unit=40, file='bchomo_ne.txt', status='unknown', action='write')
+    end if
+
+
+    do
+        read(10, '(A)', iostat=ios1) line1
+        read(20, '(A)', iostat=ios2) line2
+
+        if (ios1 /= 0 .and. ios2 /= 0) exit
+
+        write(30, '(A)') trim(line1)
+        write(30, '(A)') trim(line2)
+        write(30, '(A)') '----------------------------------------------------------------------'
+
+        if(line1/=line2) then
+            flag1 = .false.
+            write(40, '(A)') trim(line1)
+            write(40, '(A)') trim(line2)
+            write(40, '(A)') '----------------------------------------------------------------------'
+        else
+        end if
+
+    end do
+
+    if(flag1) then
+        write(30, '(A)') 'Data is same'
+        write(40, '(A)') '----- NO DATA -----'
+    else
+        write(30, '(A)') 'Data is not same'
+    end if
+    write(30, '(A)') '----------------------------------------------------------------------'
+
+    close(10)
+    close(20)
+    close(30)
+    close(40)
+end program comparison

miniapps/tgv_solver_3d/compstats.f90

@@ -0,0 +1,91 @@
+program merge_files
+    use ieee_arithmetic
+    ! 
+    implicit none
+    ! 
+    character(len=40) :: line1, line2
+    integer :: ios1, ios2, i1, i2
+    real :: num1, num2, num3
+    logical :: flag = .true.
+    character(len=10) :: second_last_word
+    integer :: pos_last_space, pos_second_last_space, ln
+
+    ln = 0
+
+    open(unit=10, file='time_report_cpu.txt', status='old', action='read')
+    open(unit=20, file='time_report_gpu.txt', status='old', action='read')
+    open(unit=30, file='time_report.txt', status='replace', action='write')
+
+    do
+        read(10, '(A)', iostat=ios1) line1
+        read(20, '(A)', iostat=ios2) line2
+
+        if (ios1 /= 0 .or. ios2 /= 0) then
+            write(30, '(A, A, A)') line1, line2, '---------------------'
+            exit
+        end if
+
+        ln = ln + 1
+        flag = .true.
+
+        ! read(line1, '(F13.3)', IOSTAT=i) num1
+        ! read(line1, '(F13.3)', IOSTAT=i) num2
+        ! num3 = num1/num2
+
+        pos_last_space = len_trim(line1)
+        do while (pos_last_space > 1 .and. line1(pos_last_space:pos_last_space) /= ' ')
+          pos_last_space = pos_last_space - 1
+        end do
+
+        pos_second_last_space = pos_last_space - 1
+        do while (pos_second_last_space > 1 .and. line1(pos_second_last_space:pos_second_last_space) /= ' ')
+          pos_second_last_space = pos_second_last_space - 1
+        end do
+
+        second_last_word = adjustl(line1(pos_second_last_space+1:pos_last_space-1))
+        read(second_last_word, '(F13.3)', iostat=i1) num1
+
+        ! print *, line1, second_last_word
+
+        pos_last_space = len_trim(line2)
+        do while (pos_last_space > 1 .and. line2(pos_last_space:pos_last_space) /= ' ')
+          pos_last_space = pos_last_space - 1
+        end do
+
+        pos_second_last_space = pos_last_space - 1
+        do while (pos_second_last_space > 1 .and. line2(pos_second_last_space:pos_second_last_space) /= ' ')
+          pos_second_last_space = pos_second_last_space - 1
+        end do
+
+        second_last_word = adjustl(line2(pos_second_last_space+1:pos_last_space-1))
+        read(second_last_word, '(F13.3)', iostat=i2) num2
+
+        if (ln == 1) then
+            write(30, '(A, A, A)') line1, line2, '-----acceleration----'
+            cycle
+        end if
+
+        if (ln == 2) then
+            write(30, '(A, A, A)') line1, line2, '---------------------'
+            cycle
+        end if
+
+        if(num1 == 0 .or. num2 == 0) flag = .false.
+        ! print *, num1, num2, flag
+
+        if (flag) then
+            num3 = num1/num2
+            write(30, '(A, A, F13.1)') line1, line2, num3
+        else 
+            write(30, '(A, A, A)') line1, line2
+        end if
+
+        if (line1 == '.end' .or. line2 == '.end') print *, "end", line1, line2
+    end do
+
+    close(10)
+    close(20)
+    close(30)
+
+    print*,'-- acceleration statistiscs report generated : time_report.txt --'
+end program merge_files