'CUdeviceptr ptr = extract<CUdeviceptr>(gpu_array.attr("gpudata"))',
#Call Thrust routine, compiled into the CudaModule
'my_sort(ptr, length)',
#Return result
'return gpu_array',
]
host_mod.add_function(
FunctionBody(
FunctionDeclaration(Value('object', 'host_entry'),
[Value('object', 'gpu_array')]),
Block([Statement(x) for x in host_statements])))
#Print out generated code, to see what we're actually compiling
print("---------------------- Host code ----------------------")
print(host_mod.generate())
print("--------------------- Device code ---------------------")
print(nvcc_mod.generate())
print("-------------------------------------------------------")
#Compile modules
import codepy.jit, codepy.toolchain
gcc_toolchain = codepy.toolchain.guess_toolchain()
nvcc_toolchain = codepy.toolchain.guess_nvcc_toolchain()
module = nvcc_mod.compile(gcc_toolchain, nvcc_toolchain, debug=True)
'CUdeviceptr ptr = extract<CUdeviceptr>(gpu_array.attr("ptr"))',
#Call Thrust routine, compiled into the CudaModule
'my_sort(ptr, length)',
#Return result
'return gpu_array',
]
host_mod.add_function(
FunctionBody(
FunctionDeclaration(Value('object', 'host_entry'),
[Value('object', 'gpu_array')]),
Block([Statement(x) for x in host_statements])))
#Print out generated code, to see what we're actually compiling
print("---------------------- Host code ----------------------")
print((host_mod.generate()))
print("--------------------- Device code ---------------------")
print((nvcc_mod.generate()))
print("-------------------------------------------------------")
#Compile modules
import codepy.jit, codepy.toolchain
gcc_toolchain = codepy.toolchain.guess_toolchain()
nvcc_toolchain = codepy.toolchain.guess_nvcc_toolchain()
module = nvcc_mod.compile(gcc_toolchain, nvcc_toolchain, debug=True)
length = 100
a = np.array(np.random.rand(length), dtype=np.float32)
print("---------------------- Unsorted -----------------------")
print(a)
