very long ulp execution time for simple stuff

On 10.05.2016 14:57, Richard Hammerl wrote:

But maybe for your purpose it is enough to use the following ULP?

It runs on sheet level; of course bogus connection to pwr pins

are missed (as it is with your schematic on sheet 17, IC1100,

Net: MGTAVCC_L).

 

--------------------------

string tmpstr;

 

 

int i=0;

int t=time();

sprintf(tmpstr,"%s(%s):Start\n",tmpstr,t2string(t,"h:mm:ss:zzz"));

 

sheet(S) S.instances(I) if(ingroup(I)) {

  S.nets(N) N.segments(SEG) SEG.pinrefs(PR) {

    i++;

    if (PR.part.name==I.part.name && PR.instance.name==I.name)

       sprintf(tmpstr,"%s(%s,%d):Found

%s\n",tmpstr,t2string(time(),"h:mm:ss"),i,N.name);

  }

}

 

t=time();

sprintf(tmpstr,"%s(%s,%d):End\n",tmpstr,t2string(t,"h:mm:ss"),i);

dlgMessageBox(tmpstr);

 

--------------------------

Thank you Richard! This was the speedy way that solved the problem

Am 10.05.2016 um 11:02 schrieb Morten Leikvoll:

On 10.05.2016 10:13, Richard Hammerl wrote:

we can reproduce this effect, but do not see this huge time difference.

A schematic with about 3700 pin refs is about three times slower in V7

compared to V6. Could you please send me an example schematic where

you have this long execution time of the ULP?

 

I have sent you my files. Let me know if they don't show up.

 

Okay, now I know a bit more....

What strikes me first when running the ULP in V6 and V7 besides

different time is the different number of found pinrefs.

In V6 we have 7384, since 7.3 we have 7395; 11 more.

In some situations it could happen that the output of pinfefs of

nets in the schematic did not take into consideration unconnected/

implicit power pins. To achieve this the effort for the pinrefs loop

is significantly higher.

But maybe for your purpose it is enough to use the following ULP?

It runs on sheet level; of course bogus connection to pwr pins

are missed (as it is with your schematic on sheet 17, IC1100,

Net: MGTAVCC_L).

-

string tmpstr;

int i=0;

int t=time();

sprintf(tmpstr,"%s(%s):Start\n",tmpstr,t2string(t,"h:mm:ss:zzz"));

sheet(S) S.instances(I) if(ingroup(I)) {

   S.nets(N) N.segments(SEG) SEG.pinrefs(PR) {

     i++;

     if (PR.part.name==I.part.name && PR.instance.name==I.name)

        sprintf(tmpstr,"%s(%s,%d):Found

%s\n",tmpstr,t2string(time(),"h:mm:ss"),i,N.name);

   }

}

t=time();

sprintf(tmpstr,"%s(%s,%d):End\n",tmpstr,t2string(t,"h:mm:ss"),i);

dlgMessageBox(tmpstr);

-

Nevertheless we will try to speed up this pinrefs loop.

--

Mit freundlichen Gruessen / Best regards

Richard Hammerl

  CadSoft Support -- hotline@cadsoft.de

  FAQ: http://www.cadsoft.de/training/faq/

On 10.05.2016 10:13, Richard Hammerl wrote:

we can reproduce this effect, but do not see this huge time difference.

A schematic with about 3700 pin refs is about three times slower in V7

compared to V6. Could you please send me an example schematic where

you have this long execution time of the ULP?

I have sent you my files. Let me know if they don't show up.

Am 09.05.2016 um 11:03 schrieb Morten Leikvoll:

On 06.05.2016 17:33, Morten Leikvoll wrote:

 

I just tested this on old 6.5.4, and it is ALOT faster there.

An almost empty loop like:

 

schematic(SCH) SCH.nets(N) N.pinrefs(PR) i++;

 

..takes 3 seconds vs on Eagle 6.5.4 vs 138(!!!) seconds on Eagle 7.5.4.

 

Same design. I use no modules.

Imo it should take <<3sec too.

 

Hi Morten,

we can reproduce this effect, but do not see this huge time difference.

A schematic with about 3700 pin refs is about three times slower in V7

compared to V6. Could you please send me an example schematic where

you have this long execution time of the ULP?

--

Mit freundlichen Gruessen / Best regards

Richard Hammerl

  CadSoft Support -- hotline@cadsoft.de

  FAQ: http://www.cadsoft.de/training/faq/

On 06.05.2016 17:33, Morten Leikvoll wrote:

I just tested this on old 6.5.4, and it is ALOT faster there.

An almost empty loop like:

schematic(SCH) SCH.nets(N) N.pinrefs(PR) i++;

..takes 3 seconds vs on Eagle 6.5.4 vs 138(!!!) seconds on Eagle 7.5.4.

Same design. I use no modules.

Imo it should take <<3sec too.

On 07.05.2016 07:17, warrenbrayshaw wrote:

The maximum time will occur if your target instance is the last instance

on the last sheet.

So to improve the average time it takes to run the ulp, break out as

soon as you find your instance.

 

Then you look at the nets and their pinrefs. The nets of interest could

have been the last created and so it will take a long time to get to them.

 

In your example you reviewed 6244 pinrefs before you find your first

pinref so that where most of the time was used.

 

Once you found the first one it took 13 seconds over 628 pinrefs.

Hmmm... that,s not quick.

 

I tried the example Arduino Schematic and got 84 pin refs in a second so

I would expect 8 seconds max for 628 (my PC Win7 i7 3.4 Ghz)

 

In the past I have detected slow performance and had to not use sprintf

to concatenate strings.

 

Another thought, try placing foundinst.part.name and foundinst.name into

a string variable once. Fetching them repeatedly that way may give a

speed advantage over digging into the foundinst object.

The problem is more fundamental. Even doing the pinrefs loop without any

if condition will take 20+ seconds. It looks like Eagle is preparing a

lot of data for each time you access the pinref loop.

On 7/05/2016 3:33 a.m., Morten Leikvoll wrote:

To find all the nest belonging to a right clicked symbol (7.5.0 and

7.5.4 beta tested), I run this ULP on a context menu of

the instance object, debug info included:

 

---------------

string tmpstr;

UL_INSTANCE foundinst;

schematic(SCH) SCH.sheets(S) S.instances(I) if(ingroup(I)) foundinst=I;

int i=0;

int t=time();

sprintf(tmpstr,"%s(%s):Start\n",tmpstr,t2string(t,"h:mm:ss:zzz"));

schematic(SCH) SCH.nets(N) N.pinrefs(PR)

{

i++;

if(PR.part.name==foundinst.part.name &&

PR.instance.name==foundinst.name) sprintf(tmpstr,"%s(%s,%d):Found

%s\n",tmpstr,t2string(time(),"h:mm:ss"),i,N.name);

}

t=time();

sprintf(tmpstr,"%s(%s,%d):End\n",tmpstr,t2string(t,"h:mm:ss"),i);

dlgMessageBox(tmpstr);

----------

(sorry for the NNTP line breaks)

 

Now I have a large design, but not large enough to defend 2min execution

time for ~7 contactrefs? Whats going on here?

Here is my messagebox output for a random symbol:

 

(17:31:11:000):Start

(17:32:58,6245):Found GND

(17:32:58,6246):Found GND

(17:32:58,6247):Found GND

(17:33:00,6446):Found LCD_VDDIO

(17:33:00,6448):Found LCD_VDDLCD

(17:33:04,6622):Found N$20

(17:33:04,6624):Found N$22

(17:33:09,6849):Found TFT_DB0

(17:33:09,6851):Found TFT_DB1

(17:33:10,6853):Found TFT_DB2

(17:33:10,6855):Found TFT_DB3

(17:33:10,6857):Found TFT_DB4

(17:33:10,6859):Found TFT_DB5

(17:33:10,6861):Found TFT_DB6

(17:33:10,6863):Found TFT_DB7

(17:33:10,6865):Found TFT_DNC

(17:33:10,6867):Found TFT_NCS

(17:33:10,6869):Found TFT_NRD

(17:33:10,6871):Found TFT_NRST

(17:33:11,6873):Found TFT_WR

(17:33:21,7386):End

Hi Morten

The maximum time will occur if your target instance is the last instance

on the last sheet.

So to improve the average time it takes to run the ulp, break out as

soon as you find your instance.

Then you look at the nets and their pinrefs. The nets of interest could

have been the last created and so it will take a long time to get to them.

In your example you reviewed 6244 pinrefs before you find your first

pinref so that where most of the time was used.

Once you found the first one it took 13 seconds over 628 pinrefs.

Hmmm... that,s not quick.

I tried the example Arduino Schematic and got 84 pin refs in a second so

I would expect 8 seconds max for 628 (my PC Win7 i7 3.4 Ghz)

In the past I have detected slow performance and had to not use sprintf

to concatenate strings.

Another thought, try placing foundinst.part.name and foundinst.name into

a string variable once. Fetching them repeatedly that way may give a

speed advantage over digging into the foundinst object.

HTH

Warren