Simple type tracking in the fast code generator.

Simple type tracking in the fast code generator.

Initial implementation of ad hoc must-be-smi tracking in the fast code
generator.  Type information is used to avoid the write barrier for
smi property stores and to avoid the smi check for the inputs/output
of bitwise OR.

Committed: http://code.google.com/p/v8/source/detail?r=3833

[fschneider, 2010-02-10 15:14:33]

http://codereview.chromium.org/597021/diff/1/3
File src/fast-codegen.h (right):

http://codereview.chromium.org/597021/diff/1/3#newcode102
src/fast-codegen.h:102: bool is_smi(Register reg) { return (smi_bits_ &
reg.bit()) != 0; }
What happens if reg == no_reg? Maybe we should 
ASSERT(reg != noreg) here, since reg.bit() of no_reg is -1.

http://codereview.chromium.org/597021/diff/1/3#newcode138
src/fast-codegen.h:138: uint32_t smi_bits_;
smi_bits_ should be initialized to 0?

http://codereview.chromium.org/597021/diff/1/4
File src/ia32/fast-codegen-ia32.cc (right):

http://codereview.chromium.org/597021/diff/1/4#newcode115
src/ia32/fast-codegen-ia32.cc:115: if (is_smi(accumulator0())) {
This condition should be always true, since there is no write barrier generated,
right?

[Kevin Millikin (Chromium), 2010-02-11 08:30:09]

http://codereview.chromium.org/597021/diff/1/3
File src/fast-codegen.h (right):

http://codereview.chromium.org/597021/diff/1/3#newcode102
src/fast-codegen.h:102: bool is_smi(Register reg) { return (smi_bits_ &
reg.bit()) != 0; }
On 2010/02/10 15:14:33, fschneider wrote:
> What happens if reg == no_reg? Maybe we should 
> ASSERT(reg != noreg) here, since reg.bit() of no_reg is -1.

Done.

http://codereview.chromium.org/597021/diff/1/3#newcode138
src/fast-codegen.h:138: uint32_t smi_bits_;
On 2010/02/10 15:14:33, fschneider wrote:
> smi_bits_ should be initialized to 0?

Done.

http://codereview.chromium.org/597021/diff/1/4
File src/ia32/fast-codegen-ia32.cc (right):

http://codereview.chromium.org/597021/diff/1/4#newcode115
src/ia32/fast-codegen-ia32.cc:115: if (is_smi(accumulator0())) {
On 2010/02/10 15:14:33, fschneider wrote:
> This condition should be always true, since there is no write barrier
generated,
> right?

Good catch.  I was trying to be too general and not assume that
needs_write_barrier was a synonym for !is_smi.  I'll change it.