mirror-github-bspeice/cuburn
1
0
Fork 0
mirror of https://github.com/stevenrobertson/cuburn.git synced 2025-02-05 11:40:04 -05:00
Code Issues Actions Packages Projects Releases Wiki Activity

Add --sync option.

Browse Source
This commit is contained in:
Steven Robertson 2011-10-13 07:53:55 -04:00
parent 9a348f0517
commit e6e2c4a8d7
2 changed files with 33 additions and 14 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

24
cuburn/render.py
View File

@ -234,6 +234,9 @@ class _AnimRenderer(object):
# used, no matter the number of time steps.
PAL_HEIGHT = 16
# Use synchronous launches
sync = False
def __init__(self, anim):
self.anim = anim
self.pending = False
@ -268,6 +271,9 @@ class _AnimRenderer(object):
# It's less than ideal, but we lock some memory ahead of time
self.h_infos_locked = cuda.pagelocked_empty((info_size/4,), np.float32)
if self.sync:
self.stream = self.alt_stream = None
def render(self, cen_time):
assert not self.pending, "Tried to render with results pending!"
self.pending = True
@ -281,6 +287,7 @@ class _AnimRenderer(object):
util.BaseCode.zero_dptr(a.mod, self.d_accum, 4 * self.nbins,
self.stream)
# Ensure all main stream tasks are done before starting alt stream
if not self.sync:
self.alt_stream.wait_for_event(cuda.Event().record(self.stream))
dpal = cuda.make_multichannel_2d_array(palette, 'C')
@ -311,10 +318,13 @@ class _AnimRenderer(object):
if not d_seeds:
seeds = mwc.MWC.make_seeds(iter.IterCode.NTHREADS *
self.cps_per_block)
h_seeds = cuda.pagelocked_empty(seeds.shape, seeds.dtype)
h_seeds[:] = seeds
if self.sync:
d_seeds = cuda.to_device(seeds)
else:
size = seeds.dtype.itemsize * seeds.size
d_seeds = cuda.mem_alloc(size)
h_seeds = cuda.pagelocked_empty(seeds.shape, seeds.dtype)
h_seeds[:] = seeds
cuda.memcpy_htod_async(d_seeds, h_seeds, stream)
if on_main:
self.d_seeds = d_seeds
@ -341,10 +351,12 @@ class _AnimRenderer(object):
infos = np.concatenate(infos)
offset = b * packer.align * self.cps_per_block
d_info_off = int(self.d_infos) + offset
if self.sync:
cuda.memcpy_htod(d_info_off, infos)
else:
h_infos = self.h_infos_locked[offset/4:offset/4+len(infos)]
h_infos[:] = infos
# TODO: portable across 32/64-bit arches?
d_info_off = int(self.d_infos) + offset
cuda.memcpy_htod_async(d_info_off, h_infos, stream)
# TODO: get block config from IterCode
@ -353,6 +365,7 @@ class _AnimRenderer(object):
texrefs=[tref], stream=stream)
# Now ensure all alt stream tasks are done before continuing main
if not self.sync:
self.stream.wait_for_event(cuda.Event().record(self.alt_stream))
util.BaseCode.zero_dptr(a.mod, self.d_out, 4 * self.nbins,
@ -408,9 +421,12 @@ class _AnimRenderer(object):
return pal
def done(self):
if self.sync:
return True
return self.stream.is_done()
def get_result(self):
if not self.sync:
self.stream.synchronize()
self.pending = False
a = self.anim

7
main.py
View File

@ -23,7 +23,7 @@ import pycuda.autoinit
import cuburn._pyflam3_hacks
from fr0stlib import pyflam3
from cuburn.render import *
from cuburn import render
from cuburn.code.mwc import MWCTest
np.set_printoptions(precision=5, edgeitems=20)
@ -110,7 +110,8 @@ def main(args):
if (args.start is None or t >= args.start)
and (args.end is None or t < args.end)]
anim = Animation(genomes)
render._AnimRenderer.sync = args.sync
anim = render.Animation(genomes)
if args.debug:
anim.cmp_options.append('-G')
anim.keep = args.keep or args.debug
@ -216,6 +217,8 @@ if __name__ == "__main__":
help='Keep compilation directory (disables kernel caching)')
debug.add_argument('--debug', action='store_true', dest='debug',
help='Compile kernel with debugging enabled (implies --keep)')
debug.add_argument('--sync', action='store_true', dest='sync',
help='Use synchronous launches whenever possible')
args = parser.parse_args()