smyalygames/FiniteVolumeGPU
1
0
Fork 0
mirror of https://github.com/smyalygames/FiniteVolumeGPU.git synced 2025-05-18 06:24:13 +02:00
Code Issues Actions Packages Projects Releases Wiki Activity

Bugfix common download / upload

Browse Source
This commit is contained in:
André R. Brodtkorb 2018-11-21 12:03:52 +01:00
parent a3f063267a
commit 6b7de7b755
1 changed files with 18 additions and 34 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

52
GPUSimulators/Common.py
View File

@ -308,31 +308,9 @@ class CudaArray2D:
assert not np.isfortran(cpu_data), "Wrong datatype (Fortran, expected C)" assert not np.isfortran(cpu_data), "Wrong datatype (Fortran, expected C)"
#Create copy object from host to device #Create copy object from host to device
copy = cuda.Memcpy2D() x = (nx_halo - cpu_data.shape[1]) // 2
copy.set_src_host(cpu_data) y = (ny_halo - cpu_data.shape[0]) // 2
copy.set_dst_device(self.data.gpudata) self.upload(stream, cpu_data, extent=[x, y, cpu_data.shape[1], cpu_data.shape[0]])
#Set offsets of upload in destination
# This handles the cases where cpu_data contains ghost cell values
# and also when it does not
x_offset = (nx_halo - cpu_data.shape[1]) // 2
y_offset = (ny_halo - cpu_data.shape[0]) // 2
copy.dst_x_in_bytes = x_offset*self.data.strides[1]
copy.dst_y = y_offset
#Set destination pitch
copy.dst_pitch = self.data.strides[0]
#Set width in bytes to copy for each row and
#number of rows to copy
width = max(self.nx, cpu_data.shape[1])
height = max(self.ny, cpu_data.shape[0])
copy.width_in_bytes = width*cpu_data.itemsize
copy.height = height
#Perform the copy
copy(stream)
#self.logger.debug("Buffer <%s> [%dx%d]: Allocated ", int(self.data.gpudata), self.nx, self.ny) #self.logger.debug("Buffer <%s> [%dx%d]: Allocated ", int(self.data.gpudata), self.nx, self.ny)
@ -344,19 +322,25 @@ class CudaArray2D:
""" """
Enables downloading data from GPU to Python Enables downloading data from GPU to Python
""" """
def download(self, stream, async=False, extent=None): def download(self, stream, cpu_data=None, async=False, extent=None):
if (extent == None): if (extent is None):
x = self.x_halo x = self.x_halo
y = self.y_halo y = self.y_halo
nx = self.nx nx = self.nx
ny = self.ny ny = self.ny
else: else:
x, y, nx, ny = extent x, y, nx, ny = extent
#self.logger.debug("Downloading [%dx%d] buffer", self.nx, self.ny) if (cpu_data is None):
#Allocate host memory #self.logger.debug("Downloading [%dx%d] buffer", self.nx, self.ny)
#cpu_data = cuda.pagelocked_empty((self.ny, self.nx), np.float32) #Allocate host memory
cpu_data = np.empty((ny, nx), dtype=np.float32) #cpu_data = cuda.pagelocked_empty((self.ny, self.nx), np.float32)
cpu_data = np.empty((ny, nx), dtype=np.float32)
assert nx == cpu_data.shape[1]
assert ny == cpu_data.shape[0]
assert x+nx <= self.nx + 2*self.x_halo
assert y+ny <= self.ny + 2*self.y_halo
#Create copy object from device to host #Create copy object from device to host
copy = cuda.Memcpy2D() copy = cuda.Memcpy2D()
@ -380,8 +364,8 @@ class CudaArray2D:
return cpu_data return cpu_data
def upload(self, cpu_data, stream, extent=None): def upload(self, stream, cpu_data, extent=None):
if (extent == None): if (extent is None):
x = self.x_halo x = self.x_halo
y = self.y_halo y = self.y_halo
nx = self.nx nx = self.nx