示例#1
0
import veo
import os
from cffi import FFI

print("\nVEO test:")
print("Pass a cffi built structure to the VE as argument on the stack.")
print("Sum the elements and multiply with a factor. Correct result is 30.")
print("\n")

p = veo.VeoProc(0)
lib = p.load_library(os.getcwd() + "/libvetest6.so")
c = p.open_context()

ffi = FFI()
ffi.cdef("""
    struct abc {
        int a, b, c;
    };
    """)

abc = ffi.new("struct abc *")
abc.a = 1
abc.b = 2
abc.c = 3

# we'll pass the struct * as a void *
lib.multeach.args_type("void *", "int")
lib.multeach.ret_type("int")

req = lib.multeach(c, veo.OnStack(ffi.buffer(abc)), 5)
r = req.wait_result()
示例#2
0
from __future__ import print_function
import numpy as np
import veo as v
import vecblas as vb
import time

#Select a problem size
M = 10000
K = 1000
N = 5000
total_flops = M*K*(2*N+3)

# start VE process
p = v.VeoProc(0)
lib = p.lib["__static__"]

# open a context (worker thread)
ctx = p.open_context()

# Generate input data as numpy arrays
a_np = np.random.rand(M*K).astype(np.float32).reshape(M,K)
b_np = np.random.rand(K*N).astype(np.float32).reshape(K,N)
res_np = np.zeros((M,N)).astype(np.float32)
exp_np = np.copy(res_np)

# Copy memory to VE
a_v = p.alloc_mem(a_np.nbytes)
b_v = p.alloc_mem(b_np.nbytes)
res_v = p.alloc_mem(res_np.nbytes)
p.write_mem(a_v, a_np, a_np.nbytes)
p.write_mem(b_v, b_np, b_np.nbytes)