Review request (s): 7116081: USE_PRECOMPILED_HEADER=0 triggers a single threaded build of the JVM
stefan.karlsson at oracle.com
Tue Nov 29 00:47:01 PST 2011
Thanks for the review.
Updated webrev: http://cr.openjdk.java.net/~stefank/7116081/webrev.2/
The rationale for this change is inlined:
On 11/29/2011 12:48 AM, David Holmes wrote:
> Hi Stefan,
> On 29/11/2011 12:34 AM, Stefan Karlsson wrote:
>> Turning off the precompiled headers is somewhat broken. It triggers a
>> single threaded build even when HOTSPOT_BUILD_JOBS has been set. With
>> this fix the compile times went from around 14 minutes to 2.5 minutes,
>> on an 8 core machine.
> Took me a while to figure out why this was the case :)
>> This affects both Linux and BSD builds, but has only been tested on
>> Linux. It would be great if someone with access to a BSD machine could
>> verify this fix.
> The problem here was using
> ifdef USE_PRECOMPILED_HEADER
> when defining it to 0 is used to turn it off - so it seems to me the
> better fix here was to simply change to:
> ifeq ($(USE_PRECOMPILED_HEADER),1)
I've changed the conditional to check USE_PRECOMPILED_HEADER instead.
While doing that I found another issue. In gcc.make we try to redefine
88 ifneq ($(USE_PRECOMPILED_HEADER),0)
That doesn't work, so setting USE_PRECOMPILED_HEADER=<arbitrary string
(not 0)> on the command line will leave the precompiled headers on, but
219 ifneq ($(USE_PRECOMPILED_HEADER),1)
220 CFLAGS += -DDONT_USE_PRECOMPILED_HEADER
will empty the precompiled.hpp file. From precompiled.hpp:
// Precompiled headers are turned off for Sun Studion,
// or if the user passes USE_PRECOMPILED_HEADER=0 to the makefiles.
In the new webrev I've changed all places where we read
USE_PRECOMPILED_HEADER to always compare against 0.
> The removal of:
> PrecompiledOption = -DUSE_PRECOMPILED_HEADER
> seems okay.
More information about the hotspot-dev