|
From: Akshay S. <ak...@us...> - 2013-01-27 08:30:04
|
This is an automated email from the git hooks/post-receive script. It was
generated because a ref change was pushed to the repository containing
the project "matlisp".
The branch, tensor has been updated
via 23ae23fc820468a4dc5149850f7f85bd524ebd69 (commit)
via d7359f74d27661c0f8674d139ff196325426e828 (commit)
from dca1995e47f24342e2a8f12fdcbd4b667b76a67e (commit)
Those revisions listed above that are new to this repository have
not appeared on any other notification email; so we list those
revisions in full, below.
- Log -----------------------------------------------------------------
commit 23ae23fc820468a4dc5149850f7f85bd524ebd69
Author: Akshay Srinivasan <aks...@gm...>
Date: Sat Jan 26 19:40:55 2013 -0800
Reduced the default call bounds for L3 stuff.
diff --git a/src/base/standard-tensor.lisp b/src/base/standard-tensor.lisp
index 60b3d8d..5113980 100644
--- a/src/base/standard-tensor.lisp
+++ b/src/base/standard-tensor.lisp
@@ -113,7 +113,7 @@
(assert (= (rank old) 1) nil 'tensor-not-vector :rank (rank old)))
;;
-(defparameter *tensor-class-optimizations* (make-hash-table)
+(defvar *tensor-class-optimizations* (make-hash-table)
"
Contains a either:
o A property list containing:
diff --git a/src/base/tweakable.lisp b/src/base/tweakable.lisp
index a8cc8cf..536af42 100644
--- a/src/base/tweakable.lisp
+++ b/src/base/tweakable.lisp
@@ -28,7 +28,7 @@
Default set with SBCL on x86-64 linux. A reasonable value
is something between 400 and 1000.")
;;Level 3--------------------------------------------------------;;
-(defparameter *real-l3-fcall-lb* 100
+(defparameter *real-l3-fcall-lb* 50
"
If the maximum dimension in the MM is lower than this
parameter, then the lisp code is used by default, instead of
@@ -37,7 +37,7 @@
Default set with SBCL on x86-64 linux. A reasonable value
is something between 20 and 200.")
-(defparameter *complex-l3-fcall-lb* 60
+(defparameter *complex-l3-fcall-lb* 30
"
If the maximum dimension in the MM is lower than this
parameter, then the lisp code is used by default, instead of
commit d7359f74d27661c0f8674d139ff196325426e828
Author: Akshay Srinivasan <aks...@gm...>
Date: Sat Jan 26 19:17:14 2013 -0800
Fix deprecated comments.
diff --git a/src/level-2/gemv.lisp b/src/level-2/gemv.lisp
index 6c05e2c..033d36e 100644
--- a/src/level-2/gemv.lisp
+++ b/src/level-2/gemv.lisp
@@ -3,9 +3,6 @@
(defmacro generate-typed-gemv! (func
(tensor-class blas-gemv-func
fortran-call-lb))
- ;;Be very careful when using functions generated by this macro.
- ;;Indexes can be tricky and this has no safety net.
- ;;Use only after checking the arguments for compatibility.
(let* ((opt (if-ret (get-tensor-class-optimization-hashtable tensor-class)
(error 'tensor-cannot-find-optimization :tensor-class tensor-class)))
(matrix-class (getf opt :matrix))
diff --git a/src/level-3/gemm.lisp b/src/level-3/gemm.lisp
index 9ed038c..d6597e0 100644
--- a/src/level-3/gemm.lisp
+++ b/src/level-3/gemm.lisp
@@ -42,7 +42,6 @@
(declare (type ,(getf opt :element-type) alpha beta)
(type ,matrix-class A B C)
(type symbol job))
- ;;The big done-in-lisp-gemm, loop-ordering was inspired by the BLAS dgemm reference implementation.
,(let
((lisp-routine
`(let-typed ((nr-C (nrows C) :type index-type)
@@ -60,7 +59,7 @@
(rstp-C (row-stride C) :type index-type)
(cstp-C (col-stride C) :type index-type)
(hd-C (head C) :type index-type))
- ;;Replace with separate loops to maximize Row-ordered MM performance
+ ;;
(when (eq job-A :t)
(rotatef rstp-A cstp-A))
(when (eq job-B :t)
-----------------------------------------------------------------------
Summary of changes:
src/base/standard-tensor.lisp | 2 +-
src/base/tweakable.lisp | 4 ++--
src/level-2/gemv.lisp | 3 ---
src/level-3/gemm.lisp | 3 +--
4 files changed, 4 insertions(+), 8 deletions(-)
hooks/post-receive
--
matlisp
|