scnc
/
pyFAI
mirror of https://github.com/silx-kit/pyFAI.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

input image in a texture: helps.

This commit is contained in:
Jerome Kieffer 2012-10-16 20:46:28 +02:00
parent 46e7d0cd54
commit 4e0e44df40
2 changed files with 87 additions and 11 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
openCL/ocl_azim_LUT.cl
View File

@ -263,7 +263,7 @@ lut_integrate_image( __read_only image2d_t weights,
coef = lut[k].coef;
if((idx == 0) && (coef <= 0.0))
break;
data = read_imagef(weights, sampler, (int2)(idx/dimX , idx%dimX)).s0;
data = read_imagef(weights, sampler, (int2)(idx%dimY , idx/dimY)).s0;
//data = weights[idx];
if( (!do_dummy) || (delta_dummy && (fabs(data-dummy) > delta_dummy))|| (data!=dummy) )
{

96
src/test_splitBBox.py
View File

@ -87,7 +87,7 @@ q = pyopencl.CommandQueue(ctx)
program = pyopencl.Program(ctx, open("../openCL/ocl_azim_LUT.cl").read()).build()
t3 = time.time()
weights_buf = pyopencl.Buffer(ctx, mf.READ_ONLY | mf.COPY_HOST_PTR, hostbuf=data)
#weights_img = pyopencl.image_from_array(ctx, ary=img.data.astype(numpy.float32), mode="r", norm_int=False, num_channels=1)
weights_img = pyopencl.image_from_array(ctx, ary=img.data.astype(numpy.float32), mode="r", norm_int=False, num_channels=1)
#print co.INTENSITY, ct.FLOAT,
#imf = pyopencl.ImageFormat(numpy.uint32(co.INTENSITY), numpy.uint32(ct.FLOAT))
#weights_img = pyopencl.Image(ctx, flags=mf.READ_ONLY | mf.COPY_HOST_PTR,
@ -96,19 +96,57 @@ weights_buf = pyopencl.Buffer(ctx, mf.READ_ONLY | mf.COPY_HOST_PTR, hostbuf=data
# pitches=(img.data.shape[-1],))
#image_from_array(ctx, ary=img.data.astype(numpy.float32), mode="r", norm_int=False, num_channels=1)
#lut_idx_buf = pyopencl.Buffer(ctx, mf.READ_ONLY | mf.COPY_HOST_PTR, hostbuf=integ.lut_idx.astype(numpy.uint32))
#lut_coef_buf = pyopencl.Buffer(ctx, mf.READ_ONLY | mf.COPY_HOST_PTR, hostbuf=integ.lut_coef)
lut_idx_buf = pyopencl.Buffer(ctx, mf.READ_ONLY | mf.COPY_HOST_PTR, hostbuf=integ.lut_idx.astype(numpy.uint32))
lut_coef_buf = pyopencl.Buffer(ctx, mf.READ_ONLY | mf.COPY_HOST_PTR, hostbuf=integ.lut_coef)
lut_buf = pyopencl.Buffer(ctx, mf.READ_ONLY | mf.COPY_HOST_PTR, hostbuf=integ.lut)
None_buf = pyopencl.Buffer(ctx, mf.READ_ONLY | mf.COPY_HOST_PTR, hostbuf=numpy.zeros(1, dtype=numpy.float32))
outData_buf = pyopencl.Buffer(ctx, mf.WRITE_ONLY, numpy.dtype(numpy.float32).itemsize * bins)
outCount_buf = pyopencl.Buffer(ctx, mf.WRITE_ONLY, numpy.dtype(numpy.float32).itemsize * bins)
outMerge_buf = pyopencl.Buffer(ctx, mf.WRITE_ONLY, numpy.dtype(numpy.float32).itemsize * bins)
args = (#weights_img, numpy.uint32(img.dim1), numpy.uint32(img.dim0),
args_orig = (#weights_img, numpy.uint32(img.dim1), numpy.uint32(img.dim0),
weights_buf,
numpy.uint32(2048),
numpy.uint32(integ.lut_size),
# lut_idx_buf,
# lut_coef_buf,
lut_idx_buf,
lut_coef_buf,
# lut_buf,
numpy.int32(0),
numpy.float32(0),
numpy.float32(0),
numpy.int32(0),
None_buf,
numpy.int32(0),
None_buf,
outData_buf,
outCount_buf,
outMerge_buf)
t4 = time.time()
program.lut_integrate(q, (bins,), (16,), *args_orig)
b = numpy.empty(bins, dtype=numpy.float32)
c = numpy.empty(bins, dtype=numpy.float32)
d = numpy.empty(bins, dtype=numpy.float32)
pyopencl.enqueue_copy(q, c, outData_buf)
pyopencl.enqueue_copy(q, d, outCount_buf)
pyopencl.enqueue_copy(q, b, outMerge_buf).wait()
t5 = time.time()
pylab.plot(a, b, label="OpenCL_orig")
print "OpenCL speed-up: %s setup: %.2fms \texec: %.2fms" % (0.001 * ref_time / (t5 - t3), 1000 * (t4 - t3), 1000 * (t5 - t4))
print abs(ra - a).max(), abs(rb - b).max(), abs(rc - c).max(), abs(rd - d).max()
for i in range(10):
j = 2 ** i
st = time.time()
program.lut_integrate(q, (bins,), (j,), * args_orig)
pyopencl.enqueue_copy(q, b, outMerge_buf).wait()
print("Size: %s \ttime: %.2fms" % (j, 1000 * (time.time() - st)))
args_single = (#weights_img, numpy.uint32(img.dim1), numpy.uint32(img.dim0),
weights_buf,
numpy.uint32(2048),
numpy.uint32(integ.lut_size),
#lut_idx_buf,
#lut_coef_buf,
lut_buf,
numpy.int32(0),
numpy.float32(0),
@ -121,8 +159,7 @@ args = (#weights_img, numpy.uint32(img.dim1), numpy.uint32(img.dim0),
outCount_buf,
outMerge_buf)
t4 = time.time()
print len(args)
program.lut_integrate_single(q, (bins,), (16,), *args)
program.lut_integrate_single(q, (bins,), (16,), *args_single)
b = numpy.empty(bins, dtype=numpy.float32)
c = numpy.empty(bins, dtype=numpy.float32)
d = numpy.empty(bins, dtype=numpy.float32)
@ -130,17 +167,56 @@ pyopencl.enqueue_copy(q, c, outData_buf)
pyopencl.enqueue_copy(q, d, outCount_buf)
pyopencl.enqueue_copy(q, b, outMerge_buf).wait()
t5 = time.time()
pylab.plot(a, b, label="OpenCL")
pylab.plot(a, b, label="OpenCL_single")
print "OpenCL speed-up: %s setup: %.2fms \texec: %.2fms" % (0.001 * ref_time / (t5 - t3), 1000 * (t4 - t3), 1000 * (t5 - t4))
print abs(ra - a).max(), abs(rb - b).max(), abs(rc - c).max(), abs(rd - d).max()
for i in range(10):
j = 2 ** i
st = time.time()
program.lut_integrate_single(q, (bins,), (j,), * args)
program.lut_integrate_single(q, (bins,), (j,), * args_single)
pyopencl.enqueue_copy(q, b, outMerge_buf).wait()
print("Size: %s \ttime: %.2fms" % (j, 1000 * (time.time() - st)))
args_image = (weights_img, numpy.uint32(img.dim2), numpy.uint32(img.dim1),
# weights_buf,
numpy.uint32(2048),
numpy.uint32(integ.lut_size),
#lut_idx_buf,
#lut_coef_buf,
lut_buf,
numpy.int32(0),
numpy.float32(0),
numpy.float32(0),
numpy.int32(0),
None_buf,
numpy.int32(0),
None_buf,
outData_buf,
outCount_buf,
outMerge_buf)
t4 = time.time()
program.lut_integrate_image(q, (bins,), (16,), *args_image)
b = numpy.empty(bins, dtype=numpy.float32)
c = numpy.empty(bins, dtype=numpy.float32)
d = numpy.empty(bins, dtype=numpy.float32)
pyopencl.enqueue_copy(q, c, outData_buf)
pyopencl.enqueue_copy(q, d, outCount_buf)
pyopencl.enqueue_copy(q, b, outMerge_buf).wait()
t5 = time.time()
pylab.plot(a, b, label="OpenCL_image")
print "OpenCL speed-up: %s setup: %.2fms \texec: %.2fms" % (0.001 * ref_time / (t5 - t3), 1000 * (t4 - t3), 1000 * (t5 - t4))
print abs(ra - a).max(), abs(rb - b).max(), abs(rc - c).max(), abs(rd - d).max()
for i in range(10):
j = 2 ** i
st = time.time()
program.lut_integrate_image(q, (bins,), (j,), * args_image)
pyopencl.enqueue_copy(q, b, outMerge_buf).wait()
print("Size: %s \ttime: %.2fms" % (j, 1000 * (time.time() - st)))
#plot(ee)
#pylab.plot(a, b, label="OpenCL")
pylab.legend()