def jz(self, state):
assert self.yp_tilde_pbo is not None
floatsize = 4 #number of bytes, 32bit precision...
bytesize = self.height * self.width
#state.refresh()
#state.render()
self.pycuda_yp_tilde_pbo.unregister()
self.pycuda_yp_fx_tilde_pbo.unregister()
self.pycuda_yp_fy_tilde_pbo.unregister()
self.pycuda_yp_m_tilde_pbo.unregister()
self._pack_texture_into_PBO(self.yp_tilde_pbo, self.texid, bytesize,
GL_UNSIGNED_BYTE)
self._pack_texture_into_PBO(self.yp_fx_tilde_pbo, self.tex_fx_id,
bytesize * floatsize, GL_FLOAT)
self._pack_texture_into_PBO(self.yp_fy_tilde_pbo, self.tex_fy_id,
bytesize * floatsize, GL_FLOAT)
self._pack_texture_into_PBO(self.yp_m_tilde_pbo, self.tex_m_id,
bytesize, GL_UNSIGNED_BYTE)
self.pycuda_yp_tilde_pbo = cuda_gl.BufferObject(long(
self.yp_tilde_pbo))
self.pycuda_yp_fx_tilde_pbo = cuda_gl.BufferObject(
long(self.yp_fx_tilde_pbo))
self.pycuda_yp_fy_tilde_pbo = cuda_gl.BufferObject(
long(self.yp_fy_tilde_pbo))
self.pycuda_yp_m_tilde_pbo = cuda_gl.BufferObject(
long(self.yp_m_tilde_pbo))
#Copied perturbed image data to CUDA accessible memory, run the Cuda kernel
return self._process_jz()
def create_PBOs(w,h):
global source_pbo, dest_pbo, pycuda_source_pbo, pycuda_dest_pbo
num_texels = w*h
data = numpy.zeros((num_texels,4),numpy.uint8)
source_pbo = glGenBuffers(1)
glBindBuffer(GL_ARRAY_BUFFER, source_pbo)
glBufferData(GL_ARRAY_BUFFER, data, GL_DYNAMIC_DRAW)
glBindBuffer(GL_ARRAY_BUFFER, 0)
pycuda_source_pbo = cuda_gl.BufferObject(long(source_pbo))
dest_pbo = glGenBuffers(1)
glBindBuffer(GL_ARRAY_BUFFER, dest_pbo)
glBufferData(GL_ARRAY_BUFFER, data, GL_DYNAMIC_DRAW)
glBindBuffer(GL_ARRAY_BUFFER, 0)
pycuda_dest_pbo = cuda_gl.BufferObject(long(dest_pbo))
def process_image():
""" copy image and process using CUDA """
global pycuda_source_pbo,source_pbo,current_size, dest_pbo
image_width, image_height = current_size
assert source_pbo is not None
# tell cuda we are going to get into these buffers
pycuda_source_pbo.unregister()
# activate destination buffer
glBindBufferARB(GL_PIXEL_PACK_BUFFER_ARB, long(source_pbo))
# read data into pbo. note: use BGRA format for optimal performance
glReadPixels(
0, #start x
0, #start y
image_width, #end x
image_height, #end y
GL_BGRA, #format
GL_UNSIGNED_BYTE, #output type
ctypes.c_void_p(0))
pycuda_source_pbo = cuda_gl.BufferObject(long(source_pbo))
# run the Cuda kernel
process(image_width, image_height)
# blit convolved texture onto the screen
# download texture from PBO
glBindBuffer(GL_PIXEL_UNPACK_BUFFER_ARB, long(dest_pbo))
glBindTexture(GL_TEXTURE_2D, output_texture)
glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0,
image_width, image_height,
GL_BGRA, GL_UNSIGNED_BYTE, ctypes.c_void_p(0))
def create_PBOs(w, h):
global dest_pbo, pycuda_dest_pbo
num_texels = w * h
dest_pbo = glGenBuffers(1)
glBindBuffer(GL_PIXEL_UNPACK_BUFFER, dest_pbo)
glBufferData(GL_PIXEL_UNPACK_BUFFER, num_texels * 4, None, GL_DYNAMIC_COPY)
pycuda_dest_pbo = cuda_gl.BufferObject(long(dest_pbo))
def _initializePBO(self, data):
pbo = glGenBuffers(1)
glBindBuffer(GL_ARRAY_BUFFER, pbo)
glBufferData(GL_ARRAY_BUFFER, data, GL_DYNAMIC_DRAW)
glBindBuffer(GL_ARRAY_BUFFER, 0)
pycuda_pbo = cuda_gl.BufferObject(long(pbo))
return (pbo, pycuda_pbo)
def j_multi(self, state, deltaX, ee, label):
assert self.yp_tilde_pbo is not None
floatsize = 4
rgbsize = 4 #32-bit
bytesize = self.height*self.width
#Perturb first
#logging.debug('---------- Perturb and render first point')
state.X[ee[:,0],0] = np.squeeze(state.X[ee[:,0],0] + deltaX)
state.refresh(label, hess = True)
state.render()
state.X[ee[:,0],0] = np.squeeze(state.X[ee[:,0],0] - deltaX)
#Load pixel buffers
#logging.debug('---------- map into CUDA accessible memory')
self.pycuda_yp_tilde_pbo.unregister()
self.pycuda_yp_fx_tilde_pbo.unregister()
self.pycuda_yp_fy_tilde_pbo.unregister()
self.pycuda_yp_m_tilde_pbo.unregister()
self._pack_texture_into_PBO(self.yp_tilde_pbo, self.texid, bytesize, GL_UNSIGNED_BYTE)
self._pack_texture_into_PBO(self.yp_fx_tilde_pbo, self.tex_fx_id, bytesize*floatsize, GL_FLOAT)
self._pack_texture_into_PBO(self.yp_fy_tilde_pbo, self.tex_fy_id, bytesize*floatsize, GL_FLOAT)
self._pack_texture_into_PBO(self.yp_m_tilde_pbo, self.tex_m_id, bytesize*rgbsize, GL_UNSIGNED_BYTE, imageformat = GL_RGBA)
self.pycuda_yp_tilde_pbo = cuda_gl.BufferObject(long(self.yp_tilde_pbo))
self.pycuda_yp_fx_tilde_pbo = cuda_gl.BufferObject(long(self.yp_fx_tilde_pbo))
self.pycuda_yp_fy_tilde_pbo = cuda_gl.BufferObject(long(self.yp_fy_tilde_pbo))
self.pycuda_yp_m_tilde_pbo = cuda_gl.BufferObject(long(self.yp_m_tilde_pbo))
#Perturb second
#logging.debug('---------- Perturb and render second point')
state.X[ee[:,1],0] = np.squeeze(state.X[ee[:,1],0] + deltaX)
state.refresh(label, hess = True)
state.render()
state.X[ee[:,1],0] = np.squeeze(state.X[ee[:,1],0] - deltaX)
#Load pixel buffers
#logging.debug('---------- 2nd map into CUDA accessible memory. Unregister')
self.pycuda_ypp_tilde_pbo.unregister()
self.pycuda_ypp_fx_tilde_pbo.unregister()
self.pycuda_ypp_fy_tilde_pbo.unregister()
self.pycuda_ypp_m_tilde_pbo.unregister()
logging.debug('---------- Pack texture into PBO')
self._pack_texture_into_PBO(self.ypp_tilde_pbo, self.texid, bytesize, GL_UNSIGNED_BYTE)
self._pack_texture_into_PBO(self.ypp_fx_tilde_pbo, self.tex_fx_id, bytesize*floatsize, GL_FLOAT)
self._pack_texture_into_PBO(self.ypp_fy_tilde_pbo, self.tex_fy_id, bytesize*floatsize, GL_FLOAT)
self._pack_texture_into_PBO(self.ypp_m_tilde_pbo, self.tex_m_id, bytesize*rgbsize, GL_UNSIGNED_BYTE, imageformat = GL_RGBA)
#logging.debug('---------- Create cuda_gl BufferObjects')
self.pycuda_ypp_tilde_pbo = cuda_gl.BufferObject(long(self.ypp_tilde_pbo))
self.pycuda_ypp_fx_tilde_pbo = cuda_gl.BufferObject(long(self.ypp_fx_tilde_pbo))
self.pycuda_ypp_fy_tilde_pbo = cuda_gl.BufferObject(long(self.ypp_fy_tilde_pbo))
self.pycuda_ypp_m_tilde_pbo = cuda_gl.BufferObject(long(self.ypp_m_tilde_pbo))
#Send to CUDA!
logging.debug('---------- Perform reduction with CUDA')
return self._process_j_multi()
def j(self, state, deltaX, i, j): assert self.yp_tilde_pbo is not None floatsize = 4 rgbsize = 4 #32-bit bytesize = self.height*self.width #Perturb first state.X[i,0] += deltaX state.refresh() state.render() state.X[i,0] -= deltaX #Load pixel buffers self.pycuda_yp_tilde_pbo.unregister() self.pycuda_yp_fx_tilde_pbo.unregister() self.pycuda_yp_fy_tilde_pbo.unregister() self.pycuda_yp_m_tilde_pbo.unregister() self._pack_texture_into_PBO(self.yp_tilde_pbo, self.texid, bytesize, GL_UNSIGNED_BYTE) self._pack_texture_into_PBO(self.yp_fx_tilde_pbo, self.tex_fx_id, bytesize*floatsize, GL_FLOAT) self._pack_texture_into_PBO(self.yp_fy_tilde_pbo, self.tex_fy_id, bytesize*floatsize, GL_FLOAT) self._pack_texture_into_PBO(self.yp_m_tilde_pbo, self.tex_m_id, bytesize*rgbsize, GL_UNSIGNED_BYTE, imageformat = GL_RGBA) self.pycuda_yp_tilde_pbo = cuda_gl.BufferObject(long(self.yp_tilde_pbo)) self.pycuda_yp_fx_tilde_pbo = cuda_gl.BufferObject(long(self.yp_fx_tilde_pbo)) self.pycuda_yp_fy_tilde_pbo = cuda_gl.BufferObject(long(self.yp_fy_tilde_pbo)) self.pycuda_yp_m_tilde_pbo = cuda_gl.BufferObject(long(self.yp_m_tilde_pbo)) #Perturb second state.X[j,0] += deltaX state.refresh() state.render() state.X[j,0] -= deltaX #Load pixel buffers self.pycuda_ypp_tilde_pbo.unregister() self.pycuda_ypp_fx_tilde_pbo.unregister() self.pycuda_ypp_fy_tilde_pbo.unregister() self.pycuda_ypp_m_tilde_pbo.unregister() self._pack_texture_into_PBO(self.ypp_tilde_pbo, self.texid, bytesize, GL_UNSIGNED_BYTE) self._pack_texture_into_PBO(self.ypp_fx_tilde_pbo, self.tex_fx_id, bytesize*floatsize, GL_FLOAT) self._pack_texture_into_PBO(self.ypp_fy_tilde_pbo, self.tex_fy_id, bytesize*floatsize, GL_FLOAT) self._pack_texture_into_PBO(self.ypp_m_tilde_pbo, self.tex_m_id, bytesize*rgbsize, GL_UNSIGNED_BYTE, imageformat = GL_RGBA) self.pycuda_ypp_tilde_pbo = cuda_gl.BufferObject(long(self.ypp_tilde_pbo)) self.pycuda_ypp_fx_tilde_pbo = cuda_gl.BufferObject(long(self.ypp_fx_tilde_pbo)) self.pycuda_ypp_fy_tilde_pbo = cuda_gl.BufferObject(long(self.ypp_fy_tilde_pbo)) self.pycuda_ypp_m_tilde_pbo = cuda_gl.BufferObject(long(self.ypp_m_tilde_pbo)) #Send to CUDA! return self._process_j()
def initData(fn=None):
global pixels, array, pbo_buffer, cuda_pbo_resource, imWidth, imHeight, texid
# Cuda array initialization
array = cuda_driver.matrix_to_array(
pixels, "C") # C-style instead of Fortran-style: row-major
pixels.fill(0) # Resetting the array to 0
pbo_buffer = glGenBuffers(1) # generate 1 buffer reference
glBindBuffer(GL_PIXEL_UNPACK_BUFFER, pbo_buffer) # binding to this buffer
glBufferData(GL_PIXEL_UNPACK_BUFFER, imWidth * imHeight, pixels,
GL_STREAM_DRAW) # Allocate the buffer
bsize = glGetBufferParameteriv(
GL_PIXEL_UNPACK_BUFFER, GL_BUFFER_SIZE) # Check allocated buffer size
assert (bsize == imWidth * imHeight)
glBindBuffer(GL_PIXEL_UNPACK_BUFFER, 0) # Unbind
if ver2011:
cuda_pbo_resource = pycuda.gl.RegisteredBuffer(
int(pbo_buffer), cuda_gl.graphics_map_flags.WRITE_DISCARD)
else:
cuda_pbo_resource = cuda_gl.BufferObject(
int(pbo_buffer)) # Mapping GLBuffer to cuda_resource
glGenTextures(1, texid)
# generate 1 texture reference
glBindTexture(GL_TEXTURE_2D, texid)
# binding to this texture
glTexImage2D(GL_TEXTURE_2D, 0, GL_LUMINANCE, imWidth, imHeight, 0,
GL_LUMINANCE, GL_UNSIGNED_BYTE, None)
# Allocate the texture
glBindTexture(GL_TEXTURE_2D, 0) # Unbind
glPixelStorei(GL_UNPACK_ALIGNMENT, 1) # 1-byte row alignment
glPixelStorei(GL_PACK_ALIGNMENT, 1) # 1-byte row alignment
def initjacobian(self, y_im_flip, y_flow, y_m_flip, test=False):
#Copy y_im, y_fx, y_fy to GPU and copy y_tilde, y_fx_tilde, y_fy_tilde to GPU
print 'initjac: y_tilde_pbo', y_tilde_pbo
y_im = np.flipud(y_im_flip)
y_m = np.flipud(y_m_flip)
yfx = np.flipud(y_flow[:, :, 0])
yfy = np.flipud(y_flow[:, :, 0])
y_flow = np.dstack((yfx, yfy))
#Tell cuda we are going to get into these buffers
self.pycuda_y_tilde_pbo.unregister()
self.pycuda_y_fx_tilde_pbo.unregister()
self.pycuda_y_fy_tilde_pbo.unregister()
self.pycuda_y_m_tilde_pbo.unregister()
self.pycuda_y_im_pbo.unregister()
self.pycuda_y_fx_pbo.unregister()
self.pycuda_y_fy_pbo.unregister()
self.pycuda_y_m_pbo.unregister()
########################################################################
#y_im###################################################################
########################################################################
#Load buffer for packing
bytesize = self.height * self.width
self._pack_texture_into_PBO(self.y_tilde_pbo, self.texid, bytesize,
GL_UNSIGNED_BYTE)
#Load y_im (current frame) info from CPU memory
glBindBufferARB(GL_PIXEL_PACK_BUFFER_ARB, long(self.y_im_pbo))
glBufferData(GL_PIXEL_PACK_BUFFER_ARB, self.width * self.height, y_im,
GL_STREAM_DRAW)
glBindBuffer(GL_PIXEL_PACK_BUFFER_ARB, 0)
glDisable(GL_TEXTURE_2D)
########################################################################
#y_fx###################################################################
########################################################################
floatsize = 4 #32-bit
self._pack_texture_into_PBO(self.y_fx_tilde_pbo, self.tex_fx_id,
bytesize * floatsize, GL_FLOAT)
#Load y_fx (current frame) info from CPU memory
glBindBufferARB(GL_PIXEL_PACK_BUFFER_ARB, long(self.y_fx_pbo))
glBufferData(GL_PIXEL_PACK_BUFFER_ARB,
self.width * self.height * floatsize, y_flow[:, :, 0],
GL_STREAM_DRAW)
glBindBuffer(GL_PIXEL_PACK_BUFFER_ARB, 0)
glDisable(GL_TEXTURE_2D)
########################################################################
#y_fy###################################################################
########################################################################
self._pack_texture_into_PBO(self.y_fy_tilde_pbo, self.tex_fy_id,
bytesize * floatsize, GL_FLOAT)
#Load y_fx (current frame) info from CPU memory
glBindBufferARB(GL_PIXEL_PACK_BUFFER_ARB, long(self.y_fy_pbo))
glBufferData(GL_PIXEL_PACK_BUFFER_ARB,
self.width * self.height * floatsize, y_flow[:, :, 1],
GL_STREAM_DRAW)
glBindBuffer(GL_PIXEL_PACK_BUFFER_ARB, 0)
glDisable(GL_TEXTURE_2D)
########################################################################
#y_m####################################################################
########################################################################
#Load buffer for packing
bytesize = self.height * self.width
self._pack_texture_into_PBO(self.y_m_tilde_pbo, self.tex_m_id,
bytesize, GL_UNSIGNED_BYTE)
#Load y_im (current frame) info from CPU memory
glBindBufferARB(GL_PIXEL_PACK_BUFFER_ARB, long(self.y_m_pbo))
glBufferData(GL_PIXEL_PACK_BUFFER_ARB, self.width * self.height, y_m,
GL_STREAM_DRAW)
glBindBuffer(GL_PIXEL_PACK_BUFFER_ARB, 0)
glDisable(GL_TEXTURE_2D)
self.pycuda_y_tilde_pbo = cuda_gl.BufferObject(long(self.y_tilde_pbo))
self.pycuda_y_fx_tilde_pbo = cuda_gl.BufferObject(
long(self.y_fx_tilde_pbo))
self.pycuda_y_fy_tilde_pbo = cuda_gl.BufferObject(
long(self.y_fy_tilde_pbo))
self.pycuda_y_m_tilde_pbo = cuda_gl.BufferObject(
long(self.y_m_tilde_pbo))
self.pycuda_y_im_pbo = cuda_gl.BufferObject(long(self.y_im_pbo))
self.pycuda_y_fx_pbo = cuda_gl.BufferObject(long(self.y_fx_pbo))
self.pycuda_y_fy_pbo = cuda_gl.BufferObject(long(self.y_fy_pbo))
self.pycuda_y_m_pbo = cuda_gl.BufferObject(long(self.y_m_pbo))
#Loaded all into CUDA accessible memory, can test loaded with the following
if test:
return self._process_initjac_test(TEST_IMAGE)
else:
return None