[turbofan] Add checked load/store operators.

[turbofan] Add checked load/store operators.

TEST=mjsunit,cctest,unittests
R=jarin@chromium.org

Committed: https://chromium.googlesource.com/v8/v8/+/c516d4f09408879602520748333754d3866224fb

[Benedikt Meurer, 2014-12-01 14:42:54]

bmeurer@chromium.org changed reviewers:
+ jarin@chromium.org

[Benedikt Meurer, 2014-12-01 14:42:55]

PTAL

[Rodolph Perfetta, 2014-12-01 15:16:33]

rodolph.perfetta@gmail.com changed reviewers:
+ rodolph.perfetta@gmail.com

[Rodolph Perfetta, 2014-12-01 15:16:34]

DBC

https://codereview.chromium.org/763963002/diff/100001/src/compiler/arm64/code...
File src/compiler/arm64/code-generator-arm64.cc (right):

https://codereview.chromium.org/763963002/diff/100001/src/compiler/arm64/code...
src/compiler/arm64/code-generator-arm64.cc:185: __ Fmov(result_,
std::numeric_limits<float>::quiet_NaN());
This won't do what you expect. Since result is a double register the float NaN
will be promoted to double NaN.

If you use FPRegister (which can be 32 or 64) instead then you can squash both
32 and 64 handler in one, just query the register size to work out which NaN
value to pass.

https://codereview.chromium.org/763963002/diff/100001/src/compiler/arm64/code...
src/compiler/arm64/code-generator-arm64.cc:230: __ bind(ool->exit());           
                                \
nit, use the upper case variant: Bind. Same below.

[Rodolph Perfetta, 2014-12-01 15:37:30]

Argh, scratch that, DoubleRegister is an alias for FPRegister so it doesn't have
to be a 64-bit float. It was just looking strange to pass a 32-bit float as
DoubleRegister but it will work.

Sorry for the noise.

On 2014/12/01 15:16:34, Rodolph Perfetta wrote:
> DBC
> 
>
https://codereview.chromium.org/763963002/diff/100001/src/compiler/arm64/code...
> File src/compiler/arm64/code-generator-arm64.cc (right):
> 
>
https://codereview.chromium.org/763963002/diff/100001/src/compiler/arm64/code...
> src/compiler/arm64/code-generator-arm64.cc:185: __ Fmov(result_,
> std::numeric_limits<float>::quiet_NaN());
> This won't do what you expect. Since result is a double register the float NaN
> will be promoted to double NaN.
> 
> If you use FPRegister (which can be 32 or 64) instead then you can squash both
> 32 and 64 handler in one, just query the register size to work out which NaN
> value to pass.
> 
>
https://codereview.chromium.org/763963002/diff/100001/src/compiler/arm64/code...
> src/compiler/arm64/code-generator-arm64.cc:230: __ bind(ool->exit());         
 
>                                 \
> nit, use the upper case variant: Bind. Same below.

[Benedikt Meurer, 2014-12-01 17:58:40]

https://codereview.chromium.org/763963002/diff/100001/src/compiler/arm64/code...
File src/compiler/arm64/code-generator-arm64.cc (right):

https://codereview.chromium.org/763963002/diff/100001/src/compiler/arm64/code...
src/compiler/arm64/code-generator-arm64.cc:185: __ Fmov(result_,
std::numeric_limits<float>::quiet_NaN());
Yeah the arm64 macro assembler can be quite confusing, but this should do the
right thing (hopefully) :-)

https://codereview.chromium.org/763963002/diff/100001/src/compiler/arm64/code...
src/compiler/arm64/code-generator-arm64.cc:230: __ bind(ool->exit());           
                                \
On 2014/12/01 15:16:34, Rodolph Perfetta wrote:
> nit, use the upper case variant: Bind. Same below.

Done.

[Jarin, 2014-12-01 21:37:11]

lgtm with a question.

https://codereview.chromium.org/763963002/diff/120001/src/compiler/typer.cc
File src/compiler/typer.cc (right):

https://codereview.chromium.org/763963002/diff/120001/src/compiler/typer.cc#n...
src/compiler/typer.cc:72: signed8_ = Type::Range(f->NewNumber(-0x7F),
f->NewNumber(0x7F - 1), zone);
As Ben pointed out, why is not this -0x80 to 0x7F? (Similarly below.)

[Jarin, 2014-12-01 21:37:13]

lgtm with a question.

[Benedikt Meurer, 2014-12-02 04:49:22]

Committed patchset #9 (id:160001) manually as
c516d4f09408879602520748333754d3866224fb (presubmit successful).

[Benedikt Meurer, 2014-12-02 04:49:55]

https://codereview.chromium.org/763963002/diff/120001/src/compiler/typer.cc
File src/compiler/typer.cc (right):

https://codereview.chromium.org/763963002/diff/120001/src/compiler/typer.cc#n...
src/compiler/typer.cc:72: signed8_ = Type::Range(f->NewNumber(-0x7F),
f->NewNumber(0x7F - 1), zone);
I don't know, just moved around the code...but I guess I can fix that with this
CL.