From eee767f9e825a0097f07cbfda9b6e09578c770a7 Mon Sep 17 00:00:00 2001
From: Martin Bauer <martin.bauer@fau.de>
Date: Tue, 11 Apr 2017 09:27:52 +0200
Subject: [PATCH] Fixed CUDA resource problems in GPU test (too many registers
used)
-> smaller block
---
gpucuda/cudajit.py | 2 +-
gpucuda/indexing.py | 39 +++++++++++++++++++++++++++++++++++----
2 files changed, 36 insertions(+), 5 deletions(-)
diff --git a/gpucuda/cudajit.py b/gpucuda/cudajit.py
index f57ded792..0b18c1359 100644
--- a/gpucuda/cudajit.py
+++ b/gpucuda/cudajit.py
@@ -41,7 +41,7 @@ def makePythonFunction(kernelFunctionNode, argumentDict={}):
shape = _checkArguments(parameters, fullArguments)
indexing = kernelFunctionNode.indexing
- dictWithBlockAndThreadNumbers = indexing.getCallParameters(shape)
+ dictWithBlockAndThreadNumbers = indexing.getCallParameters(shape, func)
args = _buildNumpyArgumentList(parameters, fullArguments)
cache[key] = (args, dictWithBlockAndThreadNumbers)
diff --git a/gpucuda/indexing.py b/gpucuda/indexing.py
index 651340d6d..34de23a2e 100644
--- a/gpucuda/indexing.py
+++ b/gpucuda/indexing.py
@@ -1,4 +1,5 @@
import abc
+
import sympy as sp
import math
import pycuda.driver as cuda
@@ -30,10 +31,12 @@ class AbstractIndexing(abc.ABCMeta('ABC', (object,), {})):
return BLOCK_IDX + THREAD_IDX
@abc.abstractmethod
- def getCallParameters(self, arrShape):
+ def getCallParameters(self, arrShape, functionToCall):
"""
Determine grid and block size for kernel call
:param arrShape: the numeric (not symbolic) shape of the array
+ :param functionToCall: compile kernel function that should be called. Use this object to get information
+ about required resources like number of registers
:return: dict with keys 'blocks' and 'threads' with tuple values for number of (x,y,z) threads and blocks
the kernel should be started with
"""
@@ -84,7 +87,7 @@ class BlockIndexing(AbstractIndexing):
return coordinates[:self._dim]
- def getCallParameters(self, arrShape):
+ def getCallParameters(self, arrShape, functionToCall):
substitutionDict = {sym: value for sym, value in zip(self._symbolicShape, arrShape) if sym is not None}
widths = [end - start for start, end in zip(_getStartFromSlice(self._iterationSlice),
@@ -94,6 +97,7 @@ class BlockIndexing(AbstractIndexing):
grid = tuple(math.ceil(length / blockSize) for length, blockSize in zip(widths, self._blockSize))
extendBs = (1,) * (3 - len(self._blockSize))
extendGr = (1,) * (3 - len(grid))
+
return {'block': self._blockSize + extendBs,
'grid': grid + extendGr}
@@ -160,6 +164,32 @@ class BlockIndexing(AbstractIndexing):
return tuple(blockSize)
+ @staticmethod
+ def limitBlockSizeByRegisterRestriction(blockSize, requiredRegistersPerThread, device=None):
+ """Shrinks the blockSize if there are too many registers used per multiprocessor.
+ This is not done automatically, since the requiredRegistersPerThread are not known before compilation.
+ They can be obtained by ``func.num_regs`` from a pycuda function.
+ :returns smaller blockSize if too many registers are used.
+ """
+ da = cuda.device_attribute
+ if device is None:
+ device = cuda.Context.get_device()
+ availableRegistersPerMP = device.get_attribute(da.MAX_REGISTERS_PER_MULTIPROCESSOR)
+
+ block = blockSize
+
+ while True:
+ numThreads = 1
+ for t in block:
+ numThreads *= t
+ requiredRegistersPerMT = numThreads * requiredRegistersPerThread
+ if requiredRegistersPerMT <= availableRegistersPerMP:
+ return block
+ else:
+ largestGridEntryIdx = max(range(len(block)), key=lambda e: block[e])
+ assert block[largestGridEntryIdx] >= 2
+ block[largestGridEntryIdx] //= 2
+
@staticmethod
def permuteBlockSizeAccordingToLayout(blockSize, layout):
"""Returns modified blockSize such that the fastest coordinate gets the biggest block dimension"""
@@ -200,7 +230,7 @@ class LineIndexing(AbstractIndexing):
def coordinates(self):
return [i + offset for i, offset in zip(self._coordinates, _getStartFromSlice(self._iterationSlice))]
- def getCallParameters(self, arrShape):
+ def getCallParameters(self, arrShape, functionToCall):
substitutionDict = {sym: value for sym, value in zip(self._symbolicShape, arrShape) if sym is not None}
widths = [end - start for start, end in zip(_getStartFromSlice(self._iterationSlice),
@@ -243,4 +273,5 @@ def _getEndFromSlice(iterationSlice, arrShape):
else:
assert isinstance(sliceComponent, int)
res.append(sliceComponent + 1)
- return res
\ No newline at end of file
+ return res
+
--
GitLab