From 4b34ba94bd1d1fcab727e26773dba75def08d637 Mon Sep 17 00:00:00 2001
From: Andreas Kloeckner <inform@tiker.net>
Date: Mon, 16 Jan 2012 02:30:00 -0500
Subject: [PATCH] Parse vectorized reductions. Spelling: 'sum_vec_float4'
---
loopy/kernel.py | 16 +++++++++-------
1 file changed, 9 insertions(+), 7 deletions(-)
diff --git a/loopy/kernel.py b/loopy/kernel.py
index ed83b30c3..cd590211a 100644
--- a/loopy/kernel.py
+++ b/loopy/kernel.py
@@ -387,12 +387,6 @@ class ProductReductionOperation(TypedReductionOperation):
def __call__(self, operand1, operand2):
return operand1 * operand2
-class FloatingPointMaxOperation(TypedReductionOperation):
- neutral_element = -var("INFINITY")
-
- def __call__(self, operand1, operand2):
- return var("max")(operand1, operand2)
-
class FloatingPointMaxOperation(TypedReductionOperation):
# OpenCL 1.1, section 6.11.2
neutral_element = -var("INFINITY")
@@ -433,15 +427,23 @@ def register_reduction_parser(parser):
def parse_reduction_op(name):
import re
- red_op_match = re.match(r"^([a-z]+)_([a-z0-9]+)$", name)
+ red_op_match = re.match(r"^([a-z]+)_([a-z0-9_]+)$", name)
if red_op_match:
op_name = red_op_match.group(1)
op_type = red_op_match.group(2)
+
try:
op_dtype = np.dtype(op_type)
except TypeError:
op_dtype = None
+ if op_dtype is None and op_type.startswith("vec_"):
+ import pyopencl.array as cl_array
+ try:
+ op_dtype = getattr(cl_array.vec, op_type[4:])
+ except AttributeError:
+ op_dtype = None
+
if op_name in _REDUCTION_OPS and op_dtype is not None:
return _REDUCTION_OPS[op_name](op_dtype)
--
GitLab