Move pointer Isolate::heap_ above bool Isolate::errors_fatal_ so Ubuntu's gcc 4.8.4 and Linaro's gc…

Move pointer Isolate::heap_ above bool Isolate::errors_fatal_ so Ubuntu's gcc 4.8.4 and Linaro's gcc 4.8.3 agree on its offset.

Committed: https://github.com/dart-lang/sdk/commit/525b6f0b088d3f7ea32cd4b47952ee9ca671d7e3

[rmacnak, 2015-09-15 20:18:21]

rmacnak@google.com changed reviewers:
+ iposva@google.com

[Ivan Posva, 2015-09-15 22:25:40]

LGMT -ip

P.S. The other cleanups can be done in a different CL.

https://codereview.chromium.org/1341993003/diff/1/runtime/vm/isolate.h
File runtime/vm/isolate.h (right):

https://codereview.chromium.org/1341993003/diff/1/runtime/vm/isolate.h#newcod...
runtime/vm/isolate.h:812: Heap* heap_;
Please ensure the following ordering:
64-bit fields
pointer/word sized
32-bit fields
16-bit fields
8-bit fields
bool fields

Thanks!

There is still the remaining question about embedded structs.

[koda, 2015-09-15 22:34:44]

koda@google.com changed reviewers:
+ koda@google.com

[koda, 2015-09-15 22:34:45]

DBC

https://codereview.chromium.org/1341993003/diff/1/runtime/vm/isolate.h
File runtime/vm/isolate.h (right):

https://codereview.chromium.org/1341993003/diff/1/runtime/vm/isolate.h#newcod...
runtime/vm/isolate.h:812: Heap* heap_;
On 2015/09/15 22:25:40, Ivan Posva wrote:
> Please ensure the following ordering:
> 64-bit fields
> pointer/word sized
> 32-bit fields
> 16-bit fields
> 8-bit fields
> bool fields
> 
> Thanks!
> 
> There is still the remaining question about embedded structs.

For the fields that are frequently accessed (e.g., from generated code),
shouldn't we try to fit them within the same cache line(s)?

Also, on ARM, the range of offsets that can be encoded in a single load
instruction is somewhat limited, although I'm not sure we have *that* many
fields yet.

[rmacnak, 2015-09-16 00:36:13]

https://codereview.chromium.org/1341993003/diff/1/runtime/vm/isolate.h
File runtime/vm/isolate.h (right):

https://codereview.chromium.org/1341993003/diff/1/runtime/vm/isolate.h#newcod...
runtime/vm/isolate.h:812: Heap* heap_;
On 2015/09/15 22:34:45, koda wrote:
> On 2015/09/15 22:25:40, Ivan Posva wrote:
> > Please ensure the following ordering:
> > 64-bit fields
> > pointer/word sized
> > 32-bit fields
> > 16-bit fields
> > 8-bit fields
> > bool fields
> > 
> > Thanks!
> > 
> > There is still the remaining question about embedded structs.
> 
> For the fields that are frequently accessed (e.g., from generated code),
> shouldn't we try to fit them within the same cache line(s)?
> 
> Also, on ARM, the range of offsets that can be encoded in a single load
> instruction is somewhat limited, although I'm not sure we have *that* many
> fields yet.

Yes, we should sort those first. Currently 32-bit sizeof(Isolate) is 13952 (!)
which exceeds ARM's 4096 single instruction range.

[rmacnak, 2015-09-16 00:42:20]

Committed patchset #1 (id:1) manually as
525b6f0b088d3f7ea32cd4b47952ee9ca671d7e3 (presubmit successful).