ARM: Update the full compiler to handle all code...

ARM: Update the full compiler to handle all code

The full compiler is now complete on ARM as well. The syntax checker is still used to determine whether to use it for top level code or not during normal execution. When debugging is enabled all code will be compiled with the full compiler.

This change removes the temporary flag --force-full-compiler and now the flag --always-full-compiler enables the full compiler for all code on all platforms.

This also fixes building on Intel platform without debugger support (ENABLE_DEBUGGER_SUPPORT not defined) and adds full check for the full compiler for lazily compiled code.

Committed: http://code.google.com/p/v8/source/detail?r=4716

[Søren Thygesen Gjesse, 2010-05-25 12:21:34]

Runs on the ARM simulator with flag  --special-command="@ --full-compiler"
passed to the test script. Will run all the tests on ARM hardware before
committing.

Erik please look at the ARM code
Mads the rest (and please double check the selection criteria in compiler.cc)

[Erik Corry, 2010-05-25 12:55:24]

ARM part LGTM.

http://codereview.chromium.org/2163006/diff/10001/11002
File src/arm/full-codegen-arm.cc (right):

http://codereview.chromium.org/2163006/diff/10001/11002#newcode919
src/arm/full-codegen-arm.cc:919: __ push(r2);  // Enumeration cache.
Move this push down two instructions and you can do a multi-register push and
get more code density?

http://codereview.chromium.org/2163006/diff/10001/11002#newcode920
src/arm/full-codegen-arm.cc:920: __ ldr(r0, FieldMemOperand(r2,
FixedArray::kLengthOffset));
And make this r1 so you can do 3 pushes in one?

http://codereview.chromium.org/2163006/diff/10001/11002#newcode931
src/arm/full-codegen-arm.cc:931: __ ldr(r0, FieldMemOperand(r0,
FixedArray::kLengthOffset));
And make this r1 so you can push 2 in one?

http://codereview.chromium.org/2163006/diff/10001/11002#newcode1071
src/arm/full-codegen-arm.cc:1071: __ LoadRoot(ip,
Heap::kUndefinedValueRootIndex);
Why not just load this directly into r0?

http://codereview.chromium.org/2163006/diff/10001/11002#newcode1489
src/arm/full-codegen-arm.cc:1489: __ ldr(r2, target);
Well spotted!

http://codereview.chromium.org/2163006/diff/10001/11002#newcode1713
src/arm/full-codegen-arm.cc:1713: __ str(r1, MemOperand(sp, arg_count *
kPointerSize));
Is this strd or am I getting it back to front?

http://codereview.chromium.org/2163006/diff/10001/11002#newcode2174
src/arm/full-codegen-arm.cc:2174: __ mov(r0,
Operand(Factory::function_class_symbol()));
There's a root for that (two places).

[Mads Ager (chromium), 2010-05-25 13:00:51]

The non-code-generation part LGTM.

[Søren Thygesen Gjesse, 2010-05-25 14:08:04]

http://codereview.chromium.org/2163006/diff/10001/11002
File src/arm/full-codegen-arm.cc (right):

http://codereview.chromium.org/2163006/diff/10001/11002#newcode919
src/arm/full-codegen-arm.cc:919: __ push(r2);  // Enumeration cache.
On 2010/05/25 12:55:25, Erik Corry wrote:
> Move this push down two instructions and you can do a multi-register push and
> get more code density?

Done.

http://codereview.chromium.org/2163006/diff/10001/11002#newcode920
src/arm/full-codegen-arm.cc:920: __ ldr(r0, FieldMemOperand(r2,
FixedArray::kLengthOffset));
On 2010/05/25 12:55:25, Erik Corry wrote:
> And make this r1 so you can do 3 pushes in one?

Done.

http://codereview.chromium.org/2163006/diff/10001/11002#newcode931
src/arm/full-codegen-arm.cc:931: __ ldr(r0, FieldMemOperand(r0,
FixedArray::kLengthOffset));
On 2010/05/25 12:55:25, Erik Corry wrote:
> And make this r1 so you can push 2 in one?

Done.

http://codereview.chromium.org/2163006/diff/10001/11002#newcode1071
src/arm/full-codegen-arm.cc:1071: __ LoadRoot(ip,
Heap::kUndefinedValueRootIndex);
On 2010/05/25 12:55:25, Erik Corry wrote:
> Why not just load this directly into r0?

Done.

http://codereview.chromium.org/2163006/diff/10001/11002#newcode1713
src/arm/full-codegen-arm.cc:1713: __ str(r1, MemOperand(sp, arg_count *
kPointerSize));
On 2010/05/25 12:55:25, Erik Corry wrote:
> Is this strd or am I getting it back to front?

strd puts r0 in lowest memory, so it is not applicable here.

http://codereview.chromium.org/2163006/diff/10001/11002#newcode2174
src/arm/full-codegen-arm.cc:2174: __ mov(r0,
Operand(Factory::function_class_symbol()));
On 2010/05/25 12:55:25, Erik Corry wrote:
> There's a root for that (two places).

Done.