gaspersic/triton
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Fixed over-head bug in the auto-tuner (not in the benchmarks)

Browse Source
This commit is contained in:
Philippe Tillet
2014-11-06 16:14:46 -05:00
parent 8dd492de23
commit 6595d20c34
9 changed files with 110 additions and 63 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
python/autotune/pysrc/misc_tools.py
View File

@@ -7,6 +7,7 @@ import sys
import pyopencl as cl
import pyviennacl as vcl
import pyatidlas as atd
import numpy as np
class PhysicalLimitsNV:
@@ -214,13 +215,16 @@ def benchmark(template, statement, device):
if occupancy_record.occupancy < 15 :
raise ValueError("Template has too low occupancy")
else:
template.execute(statement, True)
vcl_statements = statements.vcl_tuple
vcl_context = statement.result.context.vcl_sub_context
model = atd._atidlas.model(template._vcl_template, vcl_context, vcl_context.current_device)
model.execute(vcl_statements, False, True)
statement.result.context.finish_all_queues()
current_time = 0
timings = []
while current_time < 1e-1:
time_before = time.time()
template.execute(statement,False)
model.execute(vcl_statements, False, False)
statement.result.context.finish_all_queues()
timings.append(time.time() - time_before)
current_time = current_time + timings[-1]