Change stack alignment on linux to 16 bytes to keep gcc 4.4 happy....

Change stack alignment on linux to 16 bytes to keep gcc 4.4 happy.

This fixes the mksnapshot segfault without requiring -fno-tree-vectorize
which just avoided the problem by not generating code with movdqa.

[Craig Schlenter, 2009-06-03 05:08:33]

Please review. Thank you.

[Kasper Lund, 2009-06-03 06:02:47]

LGTM. Do you want me to land the patch?

[Søren Thygesen Gjesse, 2009-06-03 06:08:11]

Erik, did we ever add a test checking floating point parameter passing when
adding the ARM stack aligment?

http://codereview.chromium.org/118153/diff/1/3
File src/platform-linux.cc (right):

http://codereview.chromium.org/118153/diff/1/3#newcode96
Line 96: return 16;
Currently the ARM code generator depends on this having the value 8. See the
ASSERT in MacroAssembler::EnterExitFrame in src/arm/macro-assembler-arm.cc.
Maybe just adding an #ifdef checking for V8_TARGET_ARCH_ARM and returning 8 in
that case would work for now. Then leave the EABI comment in place as well.

[Craig Schlenter, 2009-06-03 06:41:08]

On 2009/06/03 06:08:11, Søren Gjesse wrote:
> Erik, did we ever add a test checking floating point parameter passing when
> adding the ARM stack aligment?
> 
> http://codereview.chromium.org/118153/diff/1/3
> File src/platform-linux.cc (right):
> 
> http://codereview.chromium.org/118153/diff/1/3#newcode96
> Line 96: return 16;
> Currently the ARM code generator depends on this having the value 8. See the
> ASSERT in MacroAssembler::EnterExitFrame in src/arm/macro-assembler-arm.cc.

Ah, interesting! I had assumed if 8 worked, 16 would work too.

> Maybe just adding an #ifdef checking for V8_TARGET_ARCH_ARM and returning 8 in
> that case would work for now. Then leave the EABI comment in place as well.

I have updated the patch to return 8 on ARM and 16 otherwise. Please check.

Thank you!

[Søren Thygesen Gjesse, 2009-06-03 06:50:57]

LGTM

http://codereview.chromium.org/118153/diff/1007/7
File src/platform-linux.cc (right):

http://codereview.chromium.org/118153/diff/1007/7#newcode92
Line 92: // Floating point code runs faster if the stack is 8-byte aligned.
Please remove the first line of the comment, as 'Floating point code runs faster
if the stack is 8-byte aligned.' is related to Intel Linux not ARM.

[Kasper Lund, 2009-06-03 06:51:11]

Looks even better. Thanks for stepping in Søren!

[Craig Schlenter, 2009-06-03 07:38:49]

On 2009/06/03 06:50:57, Søren Gjesse wrote:
> LGTM
> 
> http://codereview.chromium.org/118153/diff/1007/7
> File src/platform-linux.cc (right):
> 
> http://codereview.chromium.org/118153/diff/1007/7#newcode92
> Line 92: // Floating point code runs faster if the stack is 8-byte aligned.
> Please remove the first line of the comment, as 'Floating point code runs
faster
> if the stack is 8-byte aligned.' is related to Intel Linux not ARM.

Done.

[Kasper Lund, 2009-06-05 06:23:33]

Landed as bleeding_edge@2107. Thanks a lot for the patch, Craig.