EPICS Home

Experimental Physics and Industrial Control System


 
2002  2003  2004  2005  2006  2007  2008  2009  2010  2011  <20122013  2014  2015  2016  2017  2018  2019  2020  2021  2022  2023  2024  Index 2002  2003  2004  2005  2006  2007  2008  2009  2010  2011  <20122013  2014  2015  2016  2017  2018  2019  2020  2021  2022  2023  2024 
<== Date ==> <== Thread ==>

Subject: optimization to epicsTime
From: "Hill, Jeff" <[email protected]>
To: "[email protected]" <[email protected]>
Date: Wed, 14 Nov 2012 20:04:47 +0000
Dear All,

I inadvertently committed the attached change to the main trunk on a checkout instead of a branch and therefore bypassed the normal review process. If there is (any) opposition I will back out the change and make a branch. I did run the epicsTime regression test, which didn't detect any issues, before making this commit.

Jeff

Commit message:

After optimizing the epicsTime::addNanoSec function we now have a larger code which uses only unsigned integer arithmetic. This is for the benefit of embedded cpu's lacking a hardware floating point coprocessor at the expense of some additional code to maintain. I measured the cpu load improving from 40% free to 80% free after making this change, on a nios2 soft core processor.

Change:

== modified file src/libCom/osi/epicsTime.cpp
--- src/libCom/osi/epicsTime.cpp	2012-05-03 17:19:34 +0000
+++ src/libCom/osi/epicsTime.cpp	2012-11-14 19:51:30 +0000
@@ -113,8 +113,48 @@
 //
 inline void epicsTime::addNanoSec (long nSecAdj)
 {
-    double secAdj = static_cast <double> (nSecAdj) / nSecPerSec;
-    *this += secAdj;
+    // After optimizing this function we now have a larger
+    // code which uses only unsigned integer arithmetic.
+    // This is for the benefit of embedded cpu's lacking
+    // a hardware floating point coprocessor at the
+    // expense of some additional code to maintain.
+    // joh 14-11-2012
+    if ( nSecAdj >= 0 ) {
+        unsigned long nSecOffsetLong =
+              static_cast < unsigned long > ( nSecAdj );
+        while ( nSecOffsetLong >= nSecPerSec ) {
+            this->secPastEpoch++; // overflow expected
+            nSecOffsetLong -= nSecPerSec;
+        }
+        const epicsUInt32 nSecOffset =
+            static_cast < epicsUInt32 > ( nSecOffsetLong );
+        epicsUInt32 nSecPerSecRemaining = nSecPerSec - nSecOffset;
+        if ( this->nSec >= nSecPerSecRemaining ) {
+            this->secPastEpoch++; // overflow expected
+            this->nSec -= nSecPerSecRemaining;
+        }
+        else {
+            this->nSec += nSecOffset;
+        }
+    }
+    else {
+        unsigned long nSecOffsetLong =
+            static_cast <unsigned long> ( -nSecAdj );
+        while ( nSecOffsetLong >= nSecPerSec ) {
+            this->secPastEpoch--; // underflow expected
+            nSecOffsetLong -= nSecPerSec;
+        }
+        const epicsUInt32 nSecOffset =
+            static_cast < epicsUInt32 > ( nSecOffsetLong );
+        if ( this->nSec >= nSecOffset ) {
+            this->nSec -= nSecOffset;
+        }
+        else {
+            // borrow
+            this->secPastEpoch--; // underflow expected
+            this->nSec += nSecPerSec - nSecOffset;
+        }
+    }
 }
 
 //


Jeff
______________________________________________________
Jeffrey O. Hill           Email        [email protected]
LANL MS H820              Voice        505 665 1831
Los Alamos NM 87545 USA   FAX          505 665 5107

Message content: TSPA





Replies:
Re: optimization to epicsTime Andrew Johnson

Navigate by Date:
Prev: Release timetable for R3.14.12.3 Andrew Johnson
Next: RE: optimization to epicsTime Hill, Jeff
Index: 2002  2003  2004  2005  2006  2007  2008  2009  2010  2011  <20122013  2014  2015  2016  2017  2018  2019  2020  2021  2022  2023  2024 
Navigate by Thread:
Prev: Release timetable for R3.14.12.3 Andrew Johnson
Next: Re: optimization to epicsTime Andrew Johnson
Index: 2002  2003  2004  2005  2006  2007  2008  2009  2010  2011  <20122013  2014  2015  2016  2017  2018  2019  2020  2021  2022  2023  2024