From 8406772c33b86457a5a72ddbde71de890d404fbe Mon Sep 17 00:00:00 2001
From: jdsteve2 <jdsteve2@illinois.edu>
Date: Tue, 20 Feb 2018 03:40:54 -0600
Subject: [PATCH] more subgroup_size and count_granularity doc
---
loopy/statistics.py | 22 +++++++++++++++++++++-
1 file changed, 21 insertions(+), 1 deletion(-)
diff --git a/loopy/statistics.py b/loopy/statistics.py
index a63ee41ad..0607a769e 100755
--- a/loopy/statistics.py
+++ b/loopy/statistics.py
@@ -572,7 +572,13 @@ class MemAccess(Record):
.. attribute:: count_granularity
A :class:`str` that specifies whether this operation should be counted
- once per *work-item*, *sub-group*, or *group*.
+ once per *work-item*, *sub-group*, or *group*. A work-item is a single
+ instance of computation executing on a single processor (think
+ 'thread'), a collection of which may be grouped together into a
+ work-group. Each work-group executes on a single compute unit with all
+ work-items within the group sharing local memory. A sub-group is an
+ implementation-dependent grouping of work-items within a work-group,
+ analagous to an NVIDIA CUDA warp.
"""
@@ -1240,6 +1246,13 @@ def get_op_map(knl, numpy_types=True, count_redundant_work=False,
(Likely desirable for performance modeling, but undesirable for code
optimization.)
+ :arg subgroup_size: (currently unused) A :class:`int` that specifies the
+ sub-group size. An OpenCL sub-group is an implementation-dependent
+ grouping of work-items within a work-group, analagous to an NVIDIA CUDA
+ warp. subgroup_size is used, e.g., when counting a :class:`MemAccess`
+ whose count_granularity specifies that it should only be counted once
+ per sub-group.
+
:return: A :class:`ToCountMap` of **{** :class:`Op` **:**
:class:`islpy.PwQPolynomial` **}**.
@@ -1554,6 +1567,13 @@ def get_synchronization_map(knl, subgroup_size=None):
:arg knl: A :class:`loopy.LoopKernel` whose barriers are to be counted.
+ :arg subgroup_size: (currently unused) A :class:`int` that specifies the
+ sub-group size. An OpenCL sub-group is an implementation-dependent
+ grouping of work-items within a work-group, analagous to an NVIDIA CUDA
+ warp. subgroup_size is used, e.g., when counting a :class:`MemAccess`
+ whose count_granularity specifies that it should only be counted once
+ per sub-group.
+
:return: A dictionary mapping each type of synchronization event to a
:class:`islpy.PwQPolynomial` holding the number of events per work-item.
--
GitLab