llvm dev - Sep 2006 - [LLVMdev] compiling the full SPEC CPU2000 suite to LLVM bytecode

On 31 Aug 2006, at 23:46, Chris Lattner wrote:
> On Thu, 31 Aug 2006, Kenneth Hoste wrote:
>> Bummer. I think I'll contact the NAG support for more info on  
>> this. Can you
>> show me the content of your Makefile.nagfortran?
>
> It is identical to yours.
>
>> Also, it is possible to tell make only to compile benchmark X? How  
>> can I
>> enforce this?
>
> Go into the directory for that benchmark, then run 'make' or
whatever.
OK, I seem to be getting closer to the problem... When compiling a  
single benchmark, I noticed the following...

/work/NAG_f95/bin/f95 -w -S -O2 /work/SPEC_CPU2000_1.3_src/benchspec/ 
CFP2000/168.wupwise/src/dcabs1.f -o dcabs1.c -dusty -dcfuns
Evaluation trial version of NAGWare Fortran 95 Release 5.1(216)
/home/kehoste/work/LLVM/1.8/llvm-gcc4-1.8-x86-linux/bin/llvm-gcc - 
DSPEC_CPU2000 -I/home/kehoste/work/LLVM/1.8/llvm/projects/llvm-test/ 
External/SPEC/CFP2000/168.wupwise -I/home/kehoste/work/LLVM/1.8/llvm/ 
projects/llvm-test/External/SPEC/CFP2000/168.wupwise -I/home/kehoste/ 
work/LLVM/1.8/llvm/include -I/home/kehoste/work/LLVM/1.8/llvm/ 
projects/llvm-test/include -I../../../../include -I/home/kehoste/work/ 
LLVM/1.8/llvm/include -D_GNU_SOURCE -D__STDC_LIMIT_MACROS -I /work/ 
SPEC_CPU2000_1.3_src/benchspec/CFP2000/168.wupwise/src/ -I/work/ 
NAG_f95/lib  -O3  -S dcabs1.c -o Output/dcabs1.ll -emit-llvm
In file included from /work/SPEC_CPU2000_1.3_src/benchspec/ 
CFP2000/168.wupwise/src/dcabs1.f:1:
/work/NAG_f95/lib/f95.h:109: error: syntax error before ‘Logical4’
/work/NAG_f95/lib/f95.h:109: warning: data definition has no type or  
storage class
/work/NAG_f95/lib/f95.h:118: error: syntax error before ‘Integer4’
/work/NAG_f95/lib/f95.h:118: warning: data definition has no type or  
storage class

This seems to suggest something is wrong with the f95.h file provided  
by NAG. This would be very strange, since I can't imagine them
releasing something without thorough testing on a common platform  
(Linux/x86). I've mailed the support, but their reply seems to be  
quite slow...

greetings,

Kenneth

-- 

Statistics are like a bikini. What they reveal is suggestive, but  
what they conceal is vital (Aaron Levenstein)

Kenneth Hoste
ELIS - Ghent University
kenneth.hoste at elis.ugent.be
http://www.elis.ugent.be/~kehoste

On 01 Sep 2006, at 10:05, Kenneth Hoste wrote:
>
>>
>>> Also, it is possible to tell make only to compile benchmark X? How
>>> can I
>>> enforce this?
>>
>> Go into the directory for that benchmark, then run 'make' or  
>> whatever.
>
I tried tom compile each of the SPEC CPU2000 benchmarks using the  
make command is each respective subdirectory of
llvm-test/External/SPEC.

My setup:

LLVM-1.8 with the gcc4 frontend (gcc-4.0.1). The NAG f95 compiler is  
v5.1(216). My platform is Fedora Core 4 on a Intel Pentium 4.

I was confronted with 6 different errors for the various benchmarks,  
4 of which are relate to the NAG f95 compiler. See below for details.
I seems most of the error are directly related to f95, which seems to  
have undergone some changes in the 5.1 release. Two errors are related
to LLVM I think, (crtend and the syntax error for perlbmk).

Any hints to solve these problems are welcome. I've contacted the f95  
people on the NAG-specific problems.

*** cannot find library crtend

This seems to be LLVM-specific. I've check the FAQ, and this is  
actually mentioned there. Only, it says to
define the LLVM_LIB_SEARCH_PATH environment variable. Only, I can't  
find libcrtend.a anywhere in the
cfrontend directories. Is there a known workaround for this? And if  
there is, why isn't it in the documentation?

Affected benchmarks: equake, lucas, vpr, gcc, crafty, eon, gcc

*** syntax error

Compiling the perlbmk benchmark produces a syntax error. This may be  
a GCC4 problem.

<path>/SPEC_CPU2000_1.3_src/benchspec/CINT2000/253.perlbmk/src/ 
nt_perlmain.c:80: error: syntax error before ‘int’

(among others)

This specific line of code is:

DllExport int RunPerl(int argc, char **argv, char **env, void *ios);

which does seem very strange syntax to me...

Affected benchmarks: perlbmk.

+++ following errors seem to be NAG-specific +++

*** f95.h

There seem to be some changes in the f95.h file for version 5.1 which  
cause llvm-gcc to throw an error. I suspect there's
a small problem, perhaps with the detection of my system. I guess  
this has more to do with the NAG compiler than with LLVM itself.
Nevertheless, I wanted to mention it.

llvm-gcc <with-a-bunch-of-options>  -O3  -S dcabs1.c -o Output/ 
dcabs1.ll -emit-llvm
In file included from <path>/SPEC_CPU2000_1.3_src/benchspec/ 
CFP2000/168.wupwise/src/dcabs1.f:1:
<path>/lib/f95.h:109: error: syntax error before ‘Logical4’
<path>/lib/f95.h:109: warning: data definition has no type or  
sto        rage class
<path>/lib/f95.h:118: error: syntax error before ‘Integer4’
...

Affected SPEC CPU2000 benchmarks: wupwise, swim, mgrid, applu

*** f95 PANIC

Another f95 specific issue is a PANIC when compiling the code of apsi  
to C code. More specific:

Panic: apsi.f:         Unexpected expr node 47050 type 414 in tree  
flattener
Internal Error -- please report this bug

Again, this is no job for the LLVM people.

Affected benchmarks: apsi

*** libf97.dylib not found

...simply because it's not on my system. I guess this changed in the  
5.1 release of f95.
Also, there's no NAGWare subdirectory of the lib directory, which  
should be the case according to the NAG documentation.
It seems the libraries are directly located in <path>/lib.
This should probably be fixed in the Makefile.nagfortran, but I don't  
know how (since there's no other *dylib file available it seems).
This only generates a warning from gccld, but I think it's the cause  
of the problems which are reported subsequently.

Affected benchmarks: galgel, facerec, sixtrack

*** BEAM (fma3d)

There seem to be some issues with the fma3d benchmark, when  
translating from Fortran to C:

f95 -w -S -O2 <path>/SPEC_CPU2000_1.3_src/benchspec/CFP2000/191.fma3d/ 
src/beam.f90 -o beam.c -dusty -maxcontin=69
Error: <path>/SPEC_CPU2000_1.3_src/benchspec/CFP2000/191.fma3d/src/ 
beam.f
90: Binding label 'beam_' of COMMON/BEAM/ matches module BEAM_
[f95 error termination]

Also a f95-related issue I'd guess.

Affected benchmarks: fma3d.

***

greetings,

Kenneth

-- 

Statistics are like a bikini. What they reveal is suggestive, but  
what they conceal is vital (Aaron Levenstein)

Kenneth Hoste
ELIS - Ghent University
kenneth.hoste at elis.ugent.be
http://www.elis.ugent.be/~kehoste

Hello,

Some problems were solved, new ones arised... Getting closer though...
The fixes for the previous problems are at the bottom of this email,  
bug reports will be submitted when all problems are solved.

+++ New/remaining problems +++

Currently, 9/26 benchmarks compile and run succesfully. One (fma3d)  
still has a f95 related problem (see below).
The other 16 are divided into two groups:

*** 1) *** bytecode builds ok, but exection fails

For 10 benchmarks the execution fails as follows:

JIT problems:
<start>
Running: /home/kehoste/work/LLVM/1.8/llvm/projects/llvm-test/ 
RunSafely.sh 500 galgel.in galgel.out /home/kehoste/work/LLVM/1.8/ 
llvm/Release/bin/lli -force-interpreter=false ../178.galgel.llvm.bc
(no debugging symbols found)
Using host libthread_db library "/lib/libthread_db.so.1".
(no debugging symbols found)
Core was generated by `/home/kehoste/work/LLVM/1.8/llvm/Release/bin/ 
lli-force-interpreter=false ../17'.
Program terminated with signal 6, Aborted.

warning: svr4_current_sos: Can't read pathname for load map: Input/ 
outpu        t error
</end>

Other tests:
<start>
  /home/kehoste/work/LLVM/1.8/llvm/Release/bin/fpcmp: Comparison  
failed, not a numeric difference.
******************** TEST (llc) '178.galgel' FAILED!  
********************
Execution Context Diff:
</end>
Strangely enough, no additional output (error or warnings) is  
generated. Also, the execution diff seems to be empty.
Additionally, I don't think the programs get executed at all, because  
make runs too fast.

For gap/gcc, only the cbe test fails, the other (including JIT) are ok.

Affected benchmarks: galgel, equake, lucas, gcc, gap, facerec,  
sixtrack, wupwise, mgrid, applu, apsi

Since these include both SPECfp and SPECint benchmarks, this has  
nothing todo with f95.

*** 2) *** Undefined references

With 3 benchmarks (vpr, crafty and eon), I'm getting similar  
undefined references:

/tmp/ccGCQxYs.o(.text+0x76b2): In function `init_chan':  
175.vpr.cbe.c: undefined reference to `ltmp_6791_156'
/tmp/ccGCQxYs.o(.text+0x76cb):175.vpr.cbe.c: undefined reference to  
`ltmp_6792_156'
/tmp/ccGCQxYs.o(.text+0x76f1):175.vpr.cbe.c: undefined reference to  
`ltmp_6793_157'
/tmp/ccGCQxYs.o(.text+0x76fb):175.vpr.cbe.c: undefined reference to  
`ltmp_6794_156'

These also include SPECint benchmarks, so again, this has nothing  
todo with f95.
> *** perlbmk syntax error
>
> Compiling the perlbmk benchmark produces a syntax error. This may be
> a GCC4 problem.
>
> <path>/SPEC_CPU2000_1.3_src/benchspec/CINT2000/253.perlbmk/src/
> nt_perlmain.c:80: error: syntax error before ‘int’
>
> (among others)
>
> This specific line of code is:
>
> DllExport int RunPerl(int argc, char **argv, char **env, void *ios);
>
> which does seem very strange syntax to me...
>
> Affected benchmarks: perlbmk.
>
> This one remains. Solutions/ideas are very welcome.
This one remains. Solutions/ideas are very welcome.


+++ Problems solved +++
> *** cannot find library crtend
>
This is a know problem. Fix is shown at http://lists.cs.uiuc.edu/ 
pipermail/llvm-commits/Week-of-Mon-20060828/037184.html
Thanks to Reid for mentioning this (on IRC).

Reason: crtend is a library used in the old frontend, and the  
Makefile weren't aware of this.
>
> +++ following errors seem to be NAG-specific +++
>
The NAG support was really friendly, here's how the NAG problems got  
fixed.
> *** BEAM (fma3d)
>
> There seem to be some issues with the fma3d benchmark, when
> translating from Fortran to C:
>
> f95 -w -S -O2
<path>/SPEC_CPU2000_1.3_src/benchspec/CFP2000/191.fma3d/
> src/beam.f90 -o beam.c -dusty -maxcontin=69
> Error: <path>/SPEC_CPU2000_1.3_src/benchspec/CFP2000/191.fma3d/src/
> beam.f
> 90: Binding label 'beam_' of COMMON/BEAM/ matches module BEAM_
> [f95 error termination]
>
> Also a f95-related issue I'd guess.
>
> Affected benchmarks: fma3d.
This is yet to be solved.
> *** problems with f95.h
Adding -DINT64='long long' to the CPPFLAGS solved this. I will submit  
a bug report for this some time soon.
>
> *** f95 PANIC
This was a known problem, which is fixed in the latest edit  
(available at ftp://ftp.nag.co.uk/pub/nagware_support/ 
nplux51na_rh90_282.tgz).
>
> *** libf97.dylib not found
>
It appears the *.dylib files are Mac specific. Apparently the usage  
of f95 was only tested on Mac (I think Chris did this?).

Solution (for Linux/x86):

modify Makefile.nagfortran:

CPPFLAGS += -I$(F95_DIR)/lib
LDFLAGS += $(F95_DIR)/lib/quickfit.o $(F95_DIR)/lib/libf98.so $ 
(F95_DIR)        /lib/libf98.a

I'll also submit a bug report for this one.

greetings,

Kenneth

-- 

Statistics are like a bikini. What they reveal is suggestive, but  
what they conceal is vital (Aaron Levenstein)

Kenneth Hoste
ELIS - Ghent University
kenneth.hoste at elis.ugent.be
http://www.elis.ugent.be/~kehoste