Rico,
I tried using NearLabels only in cases where the code generated between the
bind
site and the usages does not have variable size. It's possible to make
mistakes
this way, let's wait for Kevin's input.
Thanks,
Vitaly
http://codereview.chromium.org/3386025/diff/1/2
File src/ia32/code-stubs-ia32.cc (right):
http://codereview.chromium.org/3386025/diff/1/2#newcode2804
src/ia32/code-stubs-ia32.cc:2804: NearLabel unordered;
On 2010/09/23 18:17:30, Rico wrote:
I probably would not use a NearLabel here since there is a number of
code
generating method calls in the non sse2 case.
In both cases unordered is only used after the floating point code.
There are even no macro calls.
http://codereview.chromium.org/3386025/diff/1/2#newcode2989
src/ia32/code-stubs-ia32.cc:2989: __ mov(eax, Operand(esp, (argc_ + 1) *
kPointerSize));
On 2010/09/23 18:17:30, Rico wrote:
receiver_is_value should be fine, as before I would use a nearlabel
for
receiver_is_js_object
AFAIK, Enter/Leave macros do not depend of frame size, scope depth,
whatever.
http://codereview.chromium.org/3386025/diff/1/4
File src/x64/code-stubs-x64.cc (right):
http://codereview.chromium.org/3386025/diff/1/4#newcode3100
src/x64/code-stubs-x64.cc:3100: NearLabel got_char_code;
On 2010/09/23 18:17:30, Rico wrote:
again, some of these could be pretty long jumps
These labels are local to GenerateFast, so I only had to check the
macros used below can't generated unbounded code sequences.
http://codereview.chromium.org/3386025/show
--
v8-dev mailing list
[email protected]
http://groups.google.com/group/v8-dev
