mirror-github-bspeice/cuburn
1
0
Fork 0
mirror of https://github.com/stevenrobertson/cuburn.git synced 2025-07-12 03:05:14 -04:00
Code Issues Actions Packages Projects Releases Wiki Activity

Speed enhancement: alpha packing.

Browse Source 
When the alpha channel is used in a color palette, the code now replaces
the blue channel in the accumulation buffer with a pair of two U16s,
which encode the values of the blue and alpha channels as a fraction of
the value of the density. When the alpha channel is always 1.0, the blue
channel works as normal. Density is now always the last element in the
accumulation buffer.

Eliminating the separate IO operations improved total runtime by more
than 30% on my card, while the extra calculations reduced that to 20%
when alpha was present (though that can be optimized further).
This commit is contained in:
Steven Robertson
2011-10-11 09:57:37 -04:00
parent a052f7f4c5
commit 618b51b1b1
4 changed files with 75 additions and 19 deletions
Download Patch File Download Diff File Expand all files Collapse all files

17
cuburn/render.py
View File

@ -229,7 +229,6 @@ class _AnimRenderer(object):
memset(byref(self._cen_cp), 0, sizeof(self._cen_cp))
self.nbins = anim.features.acc_height * anim.features.acc_stride
self.d_den = cuda.mem_alloc(4 * self.nbins)
self.d_accum = cuda.mem_alloc(16 * self.nbins)
self.d_out = cuda.mem_alloc(16 * self.nbins)
self.d_infos = cuda.mem_alloc(anim._iter.packer.align * self.ncps)
@ -245,8 +244,6 @@ class _AnimRenderer(object):
a._interp(cen_time, cen_cp)
palette = self._interp_colors(cen_time, cen_cp)
util.BaseCode.zero_dptr(a.mod, self.d_den, self.nbins,
self.stream)
util.BaseCode.zero_dptr(a.mod, self.d_accum, 4 * self.nbins,
self.stream)
@ -311,7 +308,7 @@ class _AnimRenderer(object):
# TODO: get block config from IterCode
# TODO: print timing information
iter_fun(self.d_seeds[b], np.uint64(d_info_off),
self.d_accum, self.d_den, texrefs=[tref],
self.d_accum, texrefs=[tref],
block=(32, 16, 1), grid=(len(block_times), 1),
stream=self.stream)
@ -327,8 +324,7 @@ class _AnimRenderer(object):
util.BaseCode.zero_dptr(a.mod, self.d_out, 4 * self.nbins,
self.stream)
a._de.invoke(a.mod, Genome(cen_cp),
self.d_accum, self.d_out, self.d_den,
a._de.invoke(a.mod, Genome(cen_cp), self.d_accum, self.d_out,
self.stream)
f = np.float32
@ -381,11 +377,6 @@ class _AnimRenderer(object):
g = a.features.gutter
obuf_dim = (a.features.acc_height, a.features.acc_stride, 4)
out = cuda.from_device(self.d_out, obuf_dim, np.float32)
#dacc = cuda.from_device(self.d_accum, obuf_dim, np.float32)
#daccw = dacc[:,:,3]
#print daccw.sum()
# TODO: performance?
g = a.features.gutter
out = np.delete(out, np.s_[:g], axis=0)
out = np.delete(out, np.s_[:g], axis=1)
out = np.delete(out, np.s_[-g:], axis=0)
@ -447,6 +438,10 @@ class Features(object):
else:
self.final_xform_index = None
alphas = np.array([c.color[3] for g in genomes
for c in g.palette.entries])
self.pal_has_alpha = np.any(alphas != 1.0)
self.max_cps = max([cp.ntemporal_samples for cp in genomes])
self.width = genomes[0].width