def _generate_cu(self):
cu_gen = CUDAGenerator('{}/code_gen/cudamemop.cu'.format(pkg_dir))
cu_gen.blank_line(2)
cu_gen.include("../c_code/helper_cuda.h")
cu_gen.include("cudamemop.h")
cu_gen.blank_line(2)
for i in self.types:
cu_gen.line_no_end("{}* gpu_malloc_{}(int size)".format(i, i), 0)
cu_gen.open_brace(0)
cu_gen.line('{} * gpu = NULL'.format(i))
cu_gen.malloc_gpu(ret='gpu', type_=str(i), num='size')
cu_gen.line(line='return gpu')
cu_gen.close_brace(0)
cu_gen.blank_line()
cu_gen.line_no_end(
"void cpu2gpu_{}({} *cpu, {} *gpu, int size)".format(i, i, i),
0)
cu_gen.open_brace(0)
cu_gen.cpu_to_gpu(gpu='gpu', cpu='cpu', type_=str(i), num='size')
cu_gen.close_brace(0)
cu_gen.blank_line()
cu_gen.line_no_end(
"void gpu2cpu_{}({} *gpu, {} *cpu, int size)".format(i, i, i),
0)
cu_gen.open_brace(0)
cu_gen.gpu_to_cpu(gpu='gpu', cpu='cpu', type_=str(i), num='size')
cu_gen.close_brace(0)
cu_gen.blank_line()
cu_gen.line_no_end(
"{}* to_gpu_{}({} *cpu, int size)".format(i, i, i), 0)
cu_gen.open_brace(0)
cu_gen.line(line='{} * gpu = NULL'.format(i))
cu_gen.to_gpu(ret='gpu', cpu='cpu', type_=str(i), num='size')
cu_gen.line(line='return gpu', tab=1)
cu_gen.close_brace(0)
cu_gen.blank_line()
cu_gen.line_no_end(
"{}* from_gpu_{}({} *gpu, int size)".format(i, i, i), 0)
cu_gen.open_brace(0)
cu_gen.from_gpu(gpu='gpu', ret='cpu', type_=str(i), num='size')
cu_gen.line(line='return cpu', tab=1)
cu_gen.close_brace(0)
cu_gen.blank_line()
cu_gen.blank_line()
cu_gen.close()
return
def _generate_data_cu(self):
cu_gen = CUDAGenerator(
'{}/code_gen/connection.data.cu'.format(pkg_dir))
cu_gen.blank_line(2)
if self.debug:
cu_gen.include_std('stdio.h')
cu_gen.include_std('stdlib.h')
cu_gen.blank_line()
cu_gen.include('../c_code/helper_cuda.h')
cu_gen.include('connection.h')
cu_gen.blank_line(2)
cu_gen.line_no_end("CConnection * to_gpu_connection(CConnection *cpu)",
0)
cu_gen.open_brace()
cu_gen.line(
'CConnection * gpu = (CConnection*)malloc(sizeof(CConnection))')
cu_gen.line('gpu->n_len = cpu->n_len')
cu_gen.line('gpu->r_n_len = cpu->r_n_len')
cu_gen.line('gpu->s_len = cpu->s_len')
cu_gen.to_gpu(ret='gpu->delay_start',
cpu='cpu->delay_start',
type_='int',
num='cpu->n_len')
cu_gen.to_gpu(ret='gpu->delay_num',
cpu='cpu->delay_num',
type_='int',
num='cpu->n_len')
cu_gen.to_gpu(ret='gpu->rev_delay_start',
cpu='cpu->rev_delay_start',
type_='int',
num='cpu->r_n_len')
cu_gen.to_gpu(ret='gpu->rev_delay_num',
cpu='cpu->rev_delay_num',
type_='int',
num='cpu->r_n_len')
cu_gen.to_gpu(ret='gpu->rev_map2sid',
cpu='cpu->rev_map2sid',
type_='int',
num='cpu->s_len')
cu_gen.line('CConnection * ret = NULL')
cu_gen.to_gpu(ret='ret', cpu='gpu', type_='CConnection')
if self.debug:
cu_gen.line(line=r'printf("GPU CConnection Pointer: %p\n", ret)')
cu_gen.line(
line=
r'printf("GPU n_len: %d r_n_len: %d s_len: %d\n", gpu->n_len, gpu->r_n_len, gpu->s_len)'
)
cu_gen.line('return ret')
cu_gen.close_brace()
cu_gen.blank_line()
cu_gen.line_no_end(
"CConnection * from_gpu_connection(CConnection *gpu)", 0)
cu_gen.open_brace()
cu_gen.from_gpu(gpu='gpu', ret='ret', type_='CConnection')
if self.debug:
cu_gen.line(line=r'printf("CPU CConnection Pointer: %p\n", ret)')
cu_gen.line(
line=
r'printf("CPU n_len: %d r_n_len: %d s_len: %d\n", ret->n_len, ret->r_n_len, ret->s_len)'
)
cu_gen.line('return ret')
cu_gen.close_brace()
cu_gen.blank_line()
cu_gen.close()
return
def generate_data_cu(self, debug=False):
cu_gen = CUDAGenerator('{}/code_gen/{}.data.cu'.format(
pkg_dir, self.name.lower()))
cu_gen.blank_line(2)
cu_gen.include_std('stdlib.h')
cu_gen.blank_line()
cu_gen.include("../c_code/helper_cuda.h")
cu_gen.include("{}.h".format(self.name.lower()))
cu_gen.blank_line(2)
cu_gen.line_no_end(
"{} * to_gpu_{}({} *cpu, int num)".format(self.name.capitalize(),
self.name.lower(),
self.name.capitalize()),
0)
cu_gen.open_brace()
cu_gen.malloc(ret='gpu', type_=self.name.capitalize())
for i in self.parameters['special']:
if i != 'delay':
cu_gen.to_gpu(ret='gpu->p_{}'.format(i),
cpu='cpu->p_{}'.format(i),
num='num',
type_='int')
for i in self.parameters['constant']:
cu_gen.to_gpu(ret='gpu->p_{}'.format(i),
cpu='cpu->p_{}'.format(i),
num='num',
type_=real)
for i in self.parameters['variable']:
cu_gen.to_gpu(ret='gpu->p_{}'.format(i),
cpu='cpu->p_{}'.format(i),
num='num',
type_=real)
cu_gen.line('{} * ret = NULL'.format(self.name.capitalize()))
cu_gen.to_gpu(ret='ret',
cpu='gpu',
num='1',
type_=self.name.capitalize())
cu_gen.line('return ret')
cu_gen.close_brace()
cu_gen.blank_line()
cu_gen.line_no_end(
"{} * from_gpu_{}({} *gpu, int num)".format(
self.name.capitalize(), self.name.lower(),
self.name.capitalize()), 0)
cu_gen.open_brace()
cu_gen.from_gpu(ret='cpu', gpu='gpu', type_=self.name.capitalize())
cu_gen.line('return cpu')
cu_gen.close_brace()
cu_gen.blank_line()
cu_gen.close()
return