ARM: Track Smis on top 4 stack positions and Smi loop variables....

ARM: Track Smis on top 4 stack positions and Smi loop variables.
Improve code generation for known smis and suspected Smis.

Committed: http://code.google.com/p/v8/source/detail?r=4783

[Søren Thygesen Gjesse, 2010-06-02 08:23:03]

LGTM

http://codereview.chromium.org/2452002/diff/1/2
File src/arm/codegen-arm.cc (right):

http://codereview.chromium.org/2452002/diff/1/2#newcode1228
src/arm/codegen-arm.cc:1228: new DeferredInlineSmiOperation(op, int_value,
reversed, mode, tos);
4 space indent.

http://codereview.chromium.org/2452002/diff/1/2#newcode1255
src/arm/codegen-arm.cc:1255: new DeferredInlineSmiOperation(op, int_value, true,
mode, tos);
4 space indent.

http://codereview.chromium.org/2452002/diff/1/2#newcode1271
src/arm/codegen-arm.cc:1271: }
Maybe add an empty else block around this comment.

http://codereview.chromium.org/2452002/diff/1/2#newcode1296
src/arm/codegen-arm.cc:1296: __ stop("smishift");
Test code (or never hit)?

http://codereview.chromium.org/2452002/diff/1/6
File src/arm/virtual-frame-arm.cc (right):

http://codereview.chromium.org/2452002/diff/1/6#newcode670
src/arm/virtual-frame-arm.cc:670: void VirtualFrame::SetElementAt(Register reg,
int this_far_down) {
No use for TypeInfo argument for this method?

[Erik Corry, 2010-06-02 09:20:44]

http://codereview.chromium.org/2452002/diff/1/2
File src/arm/codegen-arm.cc (right):

http://codereview.chromium.org/2452002/diff/1/2#newcode1228
src/arm/codegen-arm.cc:1228: new DeferredInlineSmiOperation(op, int_value,
reversed, mode, tos);
On 2010/06/02 08:23:03, Søren Gjesse wrote:
> 4 space indent.

Done.

http://codereview.chromium.org/2452002/diff/1/2#newcode1255
src/arm/codegen-arm.cc:1255: new DeferredInlineSmiOperation(op, int_value, true,
mode, tos);
On 2010/06/02 08:23:03, Søren Gjesse wrote:
> 4 space indent.

Done.

http://codereview.chromium.org/2452002/diff/1/2#newcode1271
src/arm/codegen-arm.cc:1271: }
On 2010/06/02 08:23:03, Søren Gjesse wrote:
> Maybe add an empty else block around this comment.

Done.

http://codereview.chromium.org/2452002/diff/1/2#newcode1296
src/arm/codegen-arm.cc:1296: __ stop("smishift");
On 2010/06/02 08:23:03, Søren Gjesse wrote:
> Test code (or never hit)?

Paranoia code.  Removed.

http://codereview.chromium.org/2452002/diff/1/6
File src/arm/virtual-frame-arm.cc (right):

http://codereview.chromium.org/2452002/diff/1/6#newcode670
src/arm/virtual-frame-arm.cc:670: void VirtualFrame::SetElementAt(Register reg,
int this_far_down) {
On 2010/06/02 08:23:03, Søren Gjesse wrote:
> No use for TypeInfo argument for this method?

This method is almost never used for the expression stack, so it didn't seem
worth it.

[Søren Thygesen Gjesse, 2010-06-02 09:24:01]

LGTM

http://codereview.chromium.org/2452002/diff/12001/13001
File src/arm/codegen-arm.cc (right):

http://codereview.chromium.org/2452002/diff/12001/13001#newcode824
src/arm/codegen-arm.cc:824: void
CodeGenerator::VirtualFrameBinaryOperation(Token::Value op,
VirtualFrameBinaryOperation -> GenericBinaryOperation now it is the only one
left?

http://codereview.chromium.org/2452002/diff/12001/13001#newcode868
src/arm/codegen-arm.cc:868: Register lhs = frame_->PopToRegister(rhs);  // Don't
pop to rhs register.
Long line.

http://codereview.chromium.org/2452002/diff/12001/13001#newcode1148
src/arm/codegen-arm.cc:1148: VirtualFrameBinaryOperation(op, mode,
GENERATE_INLINE_SMI, kUnknownIntValue);
Long line.

http://codereview.chromium.org/2452002/diff/12001/13001#newcode3497
src/arm/codegen-arm.cc:3497: if (literal != NULL) inline_smi =
DONT_GENERATE_INLINE_SMI;
Maybe assert !literal.IsSmi() when it is != NULL (more like this below).