smyalygames/FiniteVolumeGPU_HIP
1
0
Fork 0
mirror of https://github.com/smyalygames/FiniteVolumeGPU_HIP.git synced 2025-05-18 06:24:11 +02:00
Code Issues Actions Packages Projects Releases Wiki Activity

update CudaContext

Browse Source
This commit is contained in:
Hicham Agueny 2024-03-20 16:54:19 +01:00
parent 3fd27d5bb8
commit d5601ec808
1 changed files with 23 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

29
GPUSimulators/CudaContext.py
View File

@ -22,7 +22,7 @@ along with this program. If not, see <http://www.gnu.org/licenses/>.
import os import os
import ctypes
import numpy as np import numpy as np
import time import time
import re import re
@ -43,7 +43,6 @@ from GPUSimulators import Autotuner, Common
""" """
Class which keeps track of the CUDA context and some helper functions Class which keeps track of the CUDA context and some helper functions
""" """
class CudaContext(object):
def hip_check(call_result): def hip_check(call_result):
err = call_result[0] err = call_result[0]
@ -59,6 +58,8 @@ class CudaContext(object):
raise RuntimeError(str(err)) raise RuntimeError(str(err))
return result return result
class CudaContext(object):
def __init__(self, device=None, context_flags=None, use_cache=True, autotuning=True): def __init__(self, device=None, context_flags=None, use_cache=True, autotuning=True):
""" """
Create a new CUDA context Create a new CUDA context
@ -79,30 +80,42 @@ class CudaContext(object):
#Print some info about CUDA #Print some info about CUDA
##self.logger.info("CUDA version %s", str(cuda.get_version())) ##self.logger.info("CUDA version %s", str(cuda.get_version()))
#self.logger.info("Driver version %s", str(cuda.get_driver_version())) #self.logger.info("Driver version %s", str(cuda.get_driver_version()))
self.logger.info("HIP runtime version %s", str(hip_check(hip.hipRuntimeGetVersion())))
self.logger.info("Driver version %s", str(hip_check(hip.hipDriverGetVersion()))) self.logger.info("Driver version %s", str(hip_check(hip.hipDriverGetVersion())))
if device is None: if device is None:
device = 0 device = 0
self.cuda_device = hip.Device(device) hip_check(hip.hipSetDevice(device))
props = hip.hipDeviceProp_t()
hip_check(hip.hipGetDeviceProperties(props,device))
arch = props.gcnArchName
#self.cuda_device = cuda.Device(device)
#self.cuda_device = hip_check(hip.hipCtxGetDevice())
#self.logger.info("Using device %d/%d '%s' (%s) GPU", device, cuda.Device.count(), self.cuda_device.name(), self.cuda_device.pci_bus_id()) #self.logger.info("Using device %d/%d '%s' (%s) GPU", device, cuda.Device.count(), self.cuda_device.name(), self.cuda_device.pci_bus_id())
self.logger.info("Using device %d/%d '%s' (%s) GPU", device, hip_check(hip.hipGetDeviceCount()))
# Allocate memory to store the PCI BusID
pciBusId = ctypes.create_string_buffer(64)
# PCI Bus Id
hip_check(hip.hipDeviceGetPCIBusId(pciBusId, 64, device))
self.logger.info("Using device %d/%d with --arch: '%s', --BusID: %s ", device, hip_check(hip.hipGetDeviceCount()),arch,pciBusId.value.decode('utf-8')[5:7])
#self.logger.debug(" => compute capability: %s", str(self.cuda_device.compute_capability())) #self.logger.debug(" => compute capability: %s", str(self.cuda_device.compute_capability()))
self.logger.debug(" => compute capability: %s", str(self.hip.hipDeviceComputeCapability(device))) self.logger.debug(" => compute capability: %s", hip_check(hip.hipDeviceComputeCapability(device)))
# Create the CUDA context # Create the CUDA context
if context_flags is None: if context_flags is None:
# context_flags=cuda.ctx_flags.SCHED_AUTO # context_flags=cuda.ctx_flags.SCHED_AUTO
context_flags=hip_check(hip.hipSetDeviceFlags(hip.hipDeviceScheduleAuto)) context_flags=hip_check(hip.hipSetDeviceFlags(hip.hipDeviceScheduleAuto))
#self.cuda_context = self.cuda_device.make_context(flags=context_flags) #self.cuda_context = self.cuda_device.make_context(flags=context_flags)
self.cuda_context = self.hip_check(hip.hipCtxCreate(0, device)) self.cuda_context = hip_check(hip.hipCtxCreate(0, device))
#free, total = cuda.mem_get_info() #free, total = cuda.mem_get_info()
total = hip_check(hip.hipDeviceTotalMem(device)) total = hip_check(hip.hipDeviceTotalMem(device))
#self.logger.debug(" => memory: %d / %d MB available", int(free/(1024*1024)), int(total/(1024*1024))) #self.logger.debug(" => memory: %d / %d MB available", int(free/(1024*1024)), int(total/(1024*1024)))
self.logger.debug(" => memory: %d / %d MB available", int(total/(1024*1024))) self.logger.debug(" => Total memory: %d MB available", int(total/(1024*1024)))
self.logger.info("Created context handle <%s>", str(self.cuda_context.handle)) ##self.logger.info("Created context handle <%s>", str(self.cuda_context.handle))
#Create cache dir for cubin files #Create cache dir for cubin files
self.cache_path = os.path.join(self.module_path, "cuda_cache") self.cache_path = os.path.join(self.module_path, "cuda_cache")