Repository navigation
Expand file tree
/
Copy pathmain.rb
More file actions
108 lines (83 loc) · 3.72 KB
/
Copy pathmain.rb
File metadata and controls
108 lines (83 loc) · 3.72 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
require 'ffi'
module Vector
extend FFI::Library
dll_path = File.join(__dir__, 'Vector.dll')
ffi_lib dll_path
# Attach function
attach_function :CudaMalloc, [:int], :pointer
attach_function :CudaMemcpy, [:pointer, :pointer, :int, :int], :void
attach_function :Cuda_Add, [:int, :int, :pointer, :pointer, :pointer], :void
attach_function :Cuda___powf,[:int, :int, :pointer, :double, :pointer], :void
attach_function :CudaDeviceSynchronize, [], :void
attach_function :CudaFree, [:pointer], :void
attach_function :Cuda_Conv2d, [:int, :int, :int, :int, :int, :int, :pointer, :pointer, :pointer], :void
end
cudaMemcpyKind = {
cudaMemcpyHostToHost: 0,
cudaMemcpyHostToDevice: 1,
cudaMemcpyDeviceToHost: 2,
cudaMemcpyDeviceToDevice: 3,
cudaMemcpyDefault: 4
}
num_elements, threadsPerBlock = 1000000, 256
a_arr = Array.new num_elements, 0
b_arr = Array.new num_elements, 0
for i in 0..5-1 do
a_arr[i] = 10
b_arr[i] = 10
end
a_arr_ptr = FFI::MemoryPointer.new :double, a_arr.size
b_arr_ptr = FFI::MemoryPointer.new :double, b_arr.size
c_arr_ptr = FFI::MemoryPointer.new :double, a_arr.size
a_arr_ptr.write_array_of_double a_arr
b_arr_ptr.write_array_of_double b_arr
byte_size_arr = num_elements * 8
cu_a_arr = Vector.CudaMalloc byte_size_arr
cu_b_arr = Vector.CudaMalloc byte_size_arr
cu_c_arr = Vector.CudaMalloc byte_size_arr
Vector.CudaMemcpy cu_a_arr, a_arr_ptr, byte_size_arr, cudaMemcpyKind[:cudaMemcpyHostToDevice]
Vector.CudaMemcpy cu_b_arr, b_arr_ptr, byte_size_arr, cudaMemcpyKind[:cudaMemcpyHostToDevice]
Vector.Cuda_Add threadsPerBlock, num_elements, cu_a_arr, cu_b_arr, cu_c_arr
Vector.Cuda___powf threadsPerBlock, num_elements, cu_c_arr, 2.0, cu_c_arr
Vector.CudaDeviceSynchronize
Vector.CudaMemcpy c_arr_ptr, cu_c_arr, byte_size_arr, cudaMemcpyKind[:cudaMemcpyDeviceToHost]
c_arr = c_arr_ptr.read_array_of_double a_arr.size
for i in 0..5-1 do
puts c_arr[i]
end
Vector.CudaFree cu_a_arr
Vector.CudaFree cu_b_arr
Vector.CudaFree cu_c_arr
data_arr = [
1, 2, 3, 4, 5, 6, 7, 8, 9,
10, 11, 12, 13, 14, 15, 16, 17, 18,
19, 20, 21, 22, 23, 24, 25, 26, 27,
28, 29, 30, 31, 32, 33, 34, 35, 36 ]
data_arr = data_arr.map { |i| i.to_f }
mask_arr = [
3, 4, 5,
6, 7, 8,
9, 10, 11 ]
mask_arr = mask_arr.map { |i| i.to_f }
data_row, data_col = 4, 9
mask_row, mask_col = 3, 3
data_arr_ptr = FFI::MemoryPointer.new :double, data_arr.size
mask_arr_ptr = FFI::MemoryPointer.new :double, mask_arr.size
result_arr_ptr = FFI::MemoryPointer.new :double, ((data_row-mask_row+1)*(data_col-mask_col+1))
data_arr_ptr.write_array_of_double data_arr
mask_arr_ptr.write_array_of_double mask_arr
cu_d_a_arr = Vector.CudaMalloc data_arr.size*FFI.type_size(:double)
cu_m_b_arr = Vector.CudaMalloc mask_arr.size*FFI.type_size(:double)
cu_res_c_arr = Vector.CudaMalloc ((data_row-mask_row+1)*(data_col-mask_col+1)*FFI.type_size(:double))
Vector.CudaMemcpy cu_d_a_arr, data_arr_ptr, data_arr.size*FFI.type_size(:double), cudaMemcpyKind[:cudaMemcpyHostToDevice]
Vector.CudaMemcpy cu_m_b_arr, mask_arr_ptr, mask_arr.size*FFI.type_size(:double), cudaMemcpyKind[:cudaMemcpyHostToDevice]
Vector.Cuda_Conv2d threadsPerBlock, num_elements, data_row, data_col, mask_row, mask_col, cu_d_a_arr, cu_m_b_arr, cu_res_c_arr
Vector.CudaDeviceSynchronize
Vector.CudaMemcpy result_arr_ptr, cu_res_c_arr, ((data_row-mask_row+1)*(data_col-mask_col+1)*FFI.type_size(:double)), cudaMemcpyKind[:cudaMemcpyDeviceToHost]
result_arr = result_arr_ptr.read_array_of_double ((data_row-mask_row+1)*(data_col-mask_col+1))
result_arr.each do |item|
puts item
end
Vector.CudaFree cu_d_a_arr
Vector.CudaFree cu_m_b_arr
Vector.CudaFree cu_res_c_arr