Apologies, I fat fingered the previous email on my iPad.

All our tests have been conducted from a z/OS UNIX shell, which has a maximum 
region size. Using precompiled headers won’t make much difference since most of 
the header files being read are part of the runtime and are not precompiled. 
The XLC compiler used to include precompiled header files, but IBM dropped 
them, stating they intended to improve compiler performance, making them 
unnecessary.

It gets worse. The new compiler does not generate compiler listings. Neither 
does Clang, but at least it provides the llvm-objdump utility, which, when used 
with debug files, can produce something useful for debugging. Unfortunately, 
that tool isn’t included in the z/OS toolchain, so god knows how a customer is 
supposed to support their code in the field.



> On 17 Mar 2025, at 7:11 am, David Crayford <[email protected]> wrote:
> 
> T sting region size is pointless. All our tests have been conducted in the 
> shell which has a maximum regi
> 
>> On 17 Mar 2025, at 06:53, Charles Mills <[email protected]> wrote:
>> 
>> Both tests specified REGION=0M. Here are the stats from the two compiles:
>> 
>> Open:
>> CPU:     0 HR  00 MIN  01.29 SEC    SRB:     0 HR  00 MIN  00.03 SEC      
>> VIRT:    20K  SYS:   244K  EXT:   243432K  SYS:    25660K                 
>> ATB- REAL:                 68856K  SLOTS:                     0K          
>>    VIRT- ALLOC:      88M SHRD:       0M                                      
>>                 
>> 
>> Legacy:
>> CPU:     0 HR  00 MIN  00.36 SEC    SRB:     0 HR  00 MIN  00.00 SEC  
>> VIRT:    80K  SYS:   252K  EXT:   108328K  SYS:    16728K             
>> ATB- REAL:                   232K  SLOTS:                     0K      
>>    VIRT- ALLOC:      13M SHRD:       0M                             
>> 
>> Charles
>> 
>>> On Sun, 16 Mar 2025 15:48:18 -0500, Charles Mills <[email protected]> wrote:
>>> 
>>> Trying to hold down the noise by doing one reply for all of your great 
>>> suggestions.
>>> 
>>> @Peter, no, did not know about .pch. The documentation is sorely lacking. 
>>> Would like to give that a try but no idea how.
>>>> Did you run both of those tests using the same REGION value on the same 
>>>> LPAR?
>>> Same exact virtual machine on both but I am not certain about the REGION 
>>> size. Suspect it was the same. I will look when I get back online and 
>>> possibly re-test.
>>> 
>>> @David Crayford, agree, this product does not seem quite yet ready for 
>>> prime time. Sadly. I had hoped to move forward with the new compiler.
>>> 
>>> @David Cole, well, both compile and run times are important. If you pay for 
>>> compile cycles, as we do, then compile CPU time is important. If you value 
>>> developer time, then compile elapsed time is important. Of course run time 
>>> is important, but harder to measure: what inputs, on what machine?
>>> 
>>> @Allen: noted. I will re-run the tests with larger and consistent region 
>>> sizes. I suspect they were the same, but I am not online at the moment.
>>> 
>>> Also not sure what the OPT value was for the Open compiler test. I did not 
>>> specify, and the default does not seem to be documented :-( OPT value was 0 
>>> for the legacy test. When I retest I will specify OPT 0.
>> 
>> ----------------------------------------------------------------------
>> For IBM-MAIN subscribe / signoff / archive access instructions,
>> send email to [email protected] with the message: INFO IBM-MAIN


----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO IBM-MAIN

Reply via email to