RFR: JDK-8205055: Fix compilation optimization errors with new Solaris toolchain

classic Classic list List threaded Threaded
3 messages Options
Reply | Threaded
Open this post in threaded view
|

RFR: JDK-8205055: Fix compilation optimization errors with new Solaris toolchain

Gary Adams-3
Locally tested the fix below and via mach5 jobs.
The ~50 failed tests all completed successfully.
It'd be good for Erik to repeat his full testing with
this patch included in case it blocked other issues
from being observed.

This fix can be integrated after the sync from jdk/client.

  Issue: https://bugs.openjdk.java.net/browse/JDK-8205055

Proposed fix:

diff --git a/make/hotspot/lib/JvmOverrideFiles.gmk
b/make/hotspot/lib/JvmOverrideFiles.gmk
--- a/make/hotspot/lib/JvmOverrideFiles.gmk
+++ b/make/hotspot/lib/JvmOverrideFiles.gmk
@@ -74,6 +74,8 @@

  else ifeq ($(OPENJDK_TARGET_OS), solaris)
    ifneq ($(DEBUG_LEVEL), slowdebug)
+ # dev studio 12.6 workaround
+ BUILD_LIBJVM_arguments.cpp_OPTIMIZATION := LOW

Reply | Threaded
Open this post in threaded view
|

Re: RFR: JDK-8205055: Fix compilation optimization errors with new Solaris toolchain

Magnus Ihse Bursie
On 2018-06-15 01:26, [hidden email] wrote:

> Locally tested the fix below and via mach5 jobs.
> The ~50 failed tests all completed successfully.
> It'd be good for Erik to repeat his full testing with
> this patch included in case it blocked other issues
> from being observed.
>
> This fix can be integrated after the sync from jdk/client.
>
>  Issue: https://bugs.openjdk.java.net/browse/JDK-8205055
>
> Proposed fix:
>
> diff --git a/make/hotspot/lib/JvmOverrideFiles.gmk
> b/make/hotspot/lib/JvmOverrideFiles.gmk
> --- a/make/hotspot/lib/JvmOverrideFiles.gmk
> +++ b/make/hotspot/lib/JvmOverrideFiles.gmk
> @@ -74,6 +74,8 @@
>
>  else ifeq ($(OPENJDK_TARGET_OS), solaris)
>    ifneq ($(DEBUG_LEVEL), slowdebug)
> + # dev studio 12.6 workaround
> + BUILD_LIBJVM_arguments.cpp_OPTIMIZATION := LOW

The if should test
ifeq ($(TOOLCHAIN_TYPE), solstudio)
since this is a compiler issue, not an OS issue.

Ideally, it should also test if using version 12.6, but unfortunately
there's currently no elegant way of expressing that in the makefiles, so
I'm OK with the workaround of just having a comment. I assume, from the
name and a quick glance at the file, that the contents makes it unlikely
that a lowevered optimization level wouldprovoke a performance
regression in pre-12.6 solstudio compilations.

Still, it would be nice if you are able to drill down the issue to
figure out exactly why a higher optimization level causes this file to
be incorrectly compiled. It might be code in hotspot with undefined
behavior, or it might be bugs in solstudio. In any case, it should be
addressed. As an initial fix to allow compilation to proceed, this is
fine, though.

/Magnus
Reply | Threaded
Open this post in threaded view
|

Re: RFR: JDK-8205055: Fix compilation optimization errors with new Solaris toolchain

Erik Joelsson
Hello,

On 2018-06-15 05:44, Magnus Ihse Bursie wrote:

> On 2018-06-15 01:26, [hidden email] wrote:
>> Locally tested the fix below and via mach5 jobs.
>> The ~50 failed tests all completed successfully.
>> It'd be good for Erik to repeat his full testing with
>> this patch included in case it blocked other issues
>> from being observed.
>>
>> This fix can be integrated after the sync from jdk/client.
>>
>>  Issue: https://bugs.openjdk.java.net/browse/JDK-8205055
>>
>> Proposed fix:
>>
>> diff --git a/make/hotspot/lib/JvmOverrideFiles.gmk
>> b/make/hotspot/lib/JvmOverrideFiles.gmk
>> --- a/make/hotspot/lib/JvmOverrideFiles.gmk
>> +++ b/make/hotspot/lib/JvmOverrideFiles.gmk
>> @@ -74,6 +74,8 @@
>>
>>  else ifeq ($(OPENJDK_TARGET_OS), solaris)
>>    ifneq ($(DEBUG_LEVEL), slowdebug)
>> + # dev studio 12.6 workaround
>> + BUILD_LIBJVM_arguments.cpp_OPTIMIZATION := LOW
>
> The if should test
> ifeq ($(TOOLCHAIN_TYPE), solstudio)
> since this is a compiler issue, not an OS issue.
>
> Ideally, it should also test if using version 12.6, but unfortunately
> there's currently no elegant way of expressing that in the makefiles,
> so I'm OK with the workaround of just having a comment. I assume, from
> the name and a quick glance at the file, that the contents makes it
> unlikely that a lowevered optimization level wouldprovoke a
> performance regression in pre-12.6 solstudio compilations.
>
> Still, it would be nice if you are able to drill down the issue to
> figure out exactly why a higher optimization level causes this file to
> be incorrectly compiled. It might be code in hotspot with undefined
> behavior, or it might be bugs in solstudio. In any case, it should be
> addressed. As an initial fix to allow compilation to proceed, this is
> fine, though.
>
I have run more testing using this patch and it does look like it works
pretty well with just this change. Nothing sticks out.

I agree with Magnus that it needs a TOOLCHAIN_TYPE check.

Will you file a followup for investigating what actually goes wrong?

/Erik