From fc0076d269b87d211bda01348a70f9a7fa4cfa71 Mon Sep 17 00:00:00 2001
From: Andreas Kloeckner <inform@tiker.net>
Date: Tue, 9 Aug 2011 14:39:26 +0200
Subject: [PATCH] Allow use of constant memory.
---
loopy/__init__.py | 1 -
loopy/codegen/__init__.py | 7 +++++--
loopy/kernel.py | 11 ++++++++++-
3 files changed, 15 insertions(+), 4 deletions(-)
diff --git a/loopy/__init__.py b/loopy/__init__.py
index 049327d03..417957584 100644
--- a/loopy/__init__.py
+++ b/loopy/__init__.py
@@ -11,7 +11,6 @@ register_mpz_with_pymbolic()
-# TODO: Constant memory (plus check for count)
# TODO: Reuse of previously split dimensions for prefetch
# (Or general merging)
diff --git a/loopy/codegen/__init__.py b/loopy/codegen/__init__.py
index e8415bdbb..4048ca971 100644
--- a/loopy/codegen/__init__.py
+++ b/loopy/codegen/__init__.py
@@ -145,7 +145,7 @@ def generate_code(kernel):
Define, Line, Const, LiteralLines, Initializer)
from cgen.opencl import (CLKernel, CLGlobal, CLRequiredWorkGroupSize,
- CLLocal, CLImage)
+ CLLocal, CLImage, CLConstant)
from loopy.symbolic import LoopyCCodeMapper
my_ccm = LoopyCCodeMapper(kernel)
@@ -180,7 +180,10 @@ def generate_code(kernel):
arg_decl = restrict_ptr_if_not_nvidia(
POD(arg.dtype, arg.name))
if arg_decl.name in kernel.input_vectors():
- arg_decl = Const(arg_decl)
+ if arg.constant_mem:
+ arg_decl = CLConstant(Const(arg_decl))
+ else:
+ arg_decl = Const(arg_decl)
arg_decl = CLGlobal(arg_decl)
elif isinstance(arg, ImageArg):
if arg.name in kernel.input_vectors():
diff --git a/loopy/kernel.py b/loopy/kernel.py
index f3dff96bd..b8f9d9753 100644
--- a/loopy/kernel.py
+++ b/loopy/kernel.py
@@ -14,7 +14,7 @@ import pyopencl as cl
class ArrayArg:
def __init__(self, name, dtype, strides=None, shape=None, order="C",
- offset=0):
+ offset=0, constant_mem=False):
"""
All of the following are optional. Specify either strides or shape.
@@ -49,6 +49,8 @@ class ArrayArg:
self.strides = strides
self.offset = offset
+ self.constant_mem = constant_mem
+
def __repr__(self):
return "<ArrayArg '%s' of type %s>" % (self.name, self.dtype)
@@ -493,6 +495,13 @@ class LoopKernel(Record):
msg(4, "using more local memory than available--"
"possibly OK due to cache nature")
+ const_arg_count = sum(
+ 1 for arg in self.args
+ if isinstance(arg, ArrayArg) and arg.constant_mem)
+
+ if const_arg_count > self.device.max_constant_args:
+ msg(5, "too many constant arguments")
+
max_severity = 0
for sev, msg in msgs:
max_severity = max(sev, max_severity)
--
GitLab