llvm dev - Jul 2008 - [LLVMdev] Slightly improve bugpoint output

Hi all,

after seeing a near inifinite stream of "Checking instruction
''" from
bugpoint, I thought to improve that a bit. Turns out that bugpoint outputs the
name, which is often empty (especially for the big bitcodes you want to pull
through bugpoint :-).

Below patch changes the behaviour to output the assembly version of the
instruction instead of the name, which makes things a bit clearer. Is this ok
to commit?

I've two doubts here. First, was there a particular reason to not do this in
the first place? Performance perhaps? Shouldn't matter so much, I guess?
Second, I'm including <sstream> directly. Is this allowed, or should
there be
some LLVM wrapper just as for <iostream> ? (Code was copied from
instcombine, though).


Gr.

Matthijs


Index: CrashDebugger.cpp
==================================================================---
CrashDebugger.cpp	(revision 54012)
+++ CrashDebugger.cpp	(working copy)
@@ -28,6 +28,7 @@
 #include "llvm/Support/FileUtilities.h"
 #include "llvm/Support/CommandLine.h"
 #include <fstream>
+#include <sstream>
 #include <set>
 using namespace llvm;
 
@@ -469,7 +470,9 @@
             } else {
               if (BugpointIsInterrupted) goto ExitLoops;
 
-              std::cout << "Checking instruction '"
<< I->getName() << "': ";
+              std::ostringstream SS; I->print(SS);
+
+              std::cout << "Checking instruction '"
<< SS.str() << "': ";
               Module *M = BD.deleteInstructionFromProgram(I, Simplification);
 
               // Find out if the pass still crashes on this pass...
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 189 bytes
Desc: Digital signature
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20080725/e58af24a/attachment.sig>

On Jul 25, 2008, at 7:33 AM, Matthijs Kooijman wrote:
> Hi all,
>
> after seeing a near inifinite stream of "Checking instruction
''" from
> bugpoint, I thought to improve that a bit. Turns out that bugpoint  
> outputs the
> name, which is often empty (especially for the big bitcodes you want  
> to pull
> through bugpoint :-).
>
> Below patch changes the behaviour to output the assembly version of  
> the
> instruction instead of the name, which makes things a bit clearer.  
> Is this ok
> to commit?
Cool, sure, looks fine with one change.
> I've two doubts here. First, was there a particular reason to not do  
> this in
> the first place? Performance perhaps? Shouldn't matter so much, I  
> guess?
> Second, I'm including <sstream> directly. Is this allowed, or
should
> there be
> some LLVM wrapper just as for <iostream> ? (Code was copied from
> instcombine, though).
<iostream> is the only "poisoned" header, all other stream
headers are
ok.  Also, for bugpoint, we don't really care... we care in the  
libraries primarily.
> @@ -469,7 +470,9 @@
> -              std::cout << "Checking instruction '"
<< I->getName()
> << "': ";
> +              std::ostringstream SS; I->print(SS);
> +
> +              std::cout << "Checking instruction '"
<< SS.str() <<
> "': ";
Instead of using sstream, why not just use "... << *I <<
"?

-Chris

> <iostream> is the only "poisoned" header, all other stream
headers are
> ok.  Also, for bugpoint, we don't really care... we care in the  
> libraries primarily.
Ah, good to know.
> > +              std::cout << "Checking instruction
'" << SS.str() <<
> Instead of using sstream, why not just use "... << *I <<
"?
Excellent point. If I had remembered that, I would probably have committed it
right away :-p

Gr.

Matthijs
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 189 bytes
Desc: Digital signature
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20080725/129c17ba/attachment.sig>