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
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
|
import time
from stl import *
import numpy as np
from pycuda import autoinit
from pycuda.compiler import SourceModule
import pycuda.driver as cuda
from pycuda import gpuarray
from string import Template
def array2vector(arr, dtype=gpuarray.vec.float3):
if len(arr.shape) != 2 or arr.shape[-1] != 3:
raise Exception('shape mismatch')
x = np.empty(arr.shape[0], dtype=dtype)
x['x'] = arr[:,0]
x['y'] = arr[:,1]
x['z'] = arr[:,2]
return x
print 'device %s' % autoinit.device.name()
source = open('src/intersect.cu').read()
mod = SourceModule(source, options=['-I /home/tlatorre/projects/chroma/src'], no_extern_c=True, arch='sm_13')
rotate = mod.get_function('rotate')
translate = mod.get_function('translate')
intersect_mesh = mod.get_function('intersect_mesh')
import pygame
size = width, height = 800, 600
screen = pygame.display.set_mode(size, (pygame.NOFRAME | pygame.DOUBLEBUF))
film_size = (0.035, 0.024)
focal_length = 0.05
grid = []
for x in np.linspace(-film_size[0]/2, film_size[0]/2, width):
for z in np.linspace(-film_size[1]/2, film_size[1]/2, height):
grid.append((x,0,z))
grid = np.array(grid)
grid += (0,focal_length,0)
grid += (0,300,0)
x = array2vector(grid)
x_gpu = cuda.to_device(x)
p = (0,300,0)-grid
for i in range(p.shape[0]):
p[i] /= np.linalg.norm(p[i])
p = array2vector(p)
p_gpu = cuda.to_device(p)
from zcurve import *
mesh = read_stl('models/tie_interceptor6.stl')
mesh = mesh.reshape(mesh.shape[0]//3,3,3)
mesh = morton_order(mesh)
mesh = mesh.reshape(mesh.shape[0]*3, 3)
mesh3 = array2vector(mesh)
from build import Graph
rotate(np.int32(mesh3.size), cuda.InOut(mesh3), np.float32(-np.pi/2), gpuarray.vec.make_float3(1,0,0), block=(256,1,1), grid=(mesh3.size//256+1,1))
translate(np.int32(mesh3.size), cuda.InOut(mesh3), gpuarray.vec.make_float3(0,30,0), block=(256,1,1), grid=(mesh3.size//256+1,1))
graph = Graph(mesh3)
lower = array2vector(graph.lower, dtype=gpuarray.vec.float4)
upper = array2vector(graph.upper, dtype=gpuarray.vec.float4)
start = graph.start.astype(np.uint32)
count = graph.count.astype(np.uint32)
stack = np.zeros(lower.size, dtype=np.int32)
lower_gpu = cuda.to_device(lower)
upper_gpu = cuda.to_device(upper)
lower_tex = mod.get_texref('lower_bound_arr')
upper_tex = mod.get_texref('upper_bound_arr')
lower_tex.set_address(lower_gpu, lower.nbytes)
upper_tex.set_address(upper_gpu, upper.nbytes)
lower_tex.set_format(cuda.array_format.FLOAT, 4)
upper_tex.set_format(cuda.array_format.FLOAT, 4)
start_gpu = cuda.to_device(start)
count_gpu = cuda.to_device(count)
stack_gpu = cuda.mem_alloc(stack.nbytes)
cuda.memcpy_htod(stack_gpu, stack)
child_map_tex = mod.get_texref('child_map_arr')
child_len_tex = mod.get_texref('child_len_arr')
child_map_tex.set_address(start_gpu, start.nbytes)
child_len_tex.set_address(count_gpu, count.nbytes)
child_map_tex.set_format(cuda.array_format.UNSIGNED_INT32, 1)
child_len_tex.set_format(cuda.array_format.UNSIGNED_INT32, 1)
mesh = np.empty(mesh3.size, dtype=gpuarray.vec.float4)
mesh['x'] = mesh3['x']
mesh['y'] = mesh3['y']
mesh['z'] = mesh3['z']
mesh_gpu = cuda.to_device(mesh)
mesh_tex = mod.get_texref('mesh')
mesh_tex.set_address(mesh_gpu, mesh.nbytes)
mesh_tex.set_format(cuda.array_format.FLOAT, 4)
pixel = np.empty(size, dtype=np.int32).flatten()
pixel_gpu = cuda.mem_alloc(pixel.nbytes)
cuda.memcpy_htod(pixel_gpu, pixel)
speed = []
elapsed = []
t0total = time.time()
block_size = 64
for i in range(100):
rotate(np.int32(x.size), x_gpu, np.float32(np.pi/100), gpuarray.vec.make_float3(0,0,1), block=(block_size,1,1), grid=(width*height//block_size+1,1))
rotate(np.int32(p.size), p_gpu, np.float32(np.pi/100), gpuarray.vec.make_float3(0,0,1), block=(block_size,1,1), grid=(width*height//block_size+1,1))
t0 = time.time()
intersect_mesh(np.int32(x.size), x_gpu, p_gpu, pixel_gpu, np.int32(graph.first_leaf), block=(block_size,1,1), grid=(width*height//block_size+1,1), texrefs=[mesh_tex, upper_tex, lower_tex, child_map_tex, child_len_tex])
cuda.Context.synchronize()
elapsed.append(time.time() - t0)
print '%i triangles, %i photons, %f sec; (%f photons/sec)' % \
(mesh.size//3, pixel.size, elapsed[-1], pixel.size/elapsed[-1])
speed.append(pixel.size/elapsed[-1])
cuda.memcpy_dtoh(pixel, pixel_gpu)
pygame.surfarray.blit_array(screen, pixel.reshape(size))
pygame.display.flip()
print 'average time = %f sec' % np.mean(elapsed)
print 'average speed = %f photons/sec' % np.mean(speed)
print 'total time = %f sec' % (time.time() - t0total)
raw_input('press enter to exit')
|
