Implemented FastAsciiStringJoin in X64 full codegen.

Implemented FastAsciiStringJoin in X64 full codegen.

Committed: http://code.google.com/p/v8/source/detail?r=7345

[William Hesse, 2011-03-24 12:02:21]

LGTM.

http://codereview.chromium.org/6729016/diff/1/src/x64/full-codegen-x64.cc
File src/x64/full-codegen-x64.cc (right):

http://codereview.chromium.org/6729016/diff/1/src/x64/full-codegen-x64.cc#new...
src/x64/full-codegen-x64.cc:3348: // Check that the separator is a flat ASCII
string.
sequential ASCII string?

http://codereview.chromium.org/6729016/diff/1/src/x64/full-codegen-x64.cc#new...
src/x64/full-codegen-x64.cc:3355: __ cmpb(scratch,
Immediate(ASCII_STRING_TYPE));
Why don't we use ASCII_STRING_TYPE above?  And why not ASCII_SEQ_STRING_TYPE?

http://codereview.chromium.org/6729016/diff/1/src/x64/full-codegen-x64.cc#new...
src/x64/full-codegen-x64.cc:3465: // a loop limit.
You should try this change on ia32, to see if it makes a difference.  Or else
just use another register for the loop limit.

http://codereview.chromium.org/6729016/diff/1/src/x64/full-codegen-x64.cc#new...
src/x64/full-codegen-x64.cc:3490: //   rsp[0x10] (separator_operand): Address of
first char of separator.
why not separator_operand (rsp[0x10])?

http://codereview.chromium.org/6729016/diff/1/src/x64/full-codegen-x64.cc#new...
src/x64/full-codegen-x64.cc:3506: __ j(less, &loop_3);  // Loop while (index <
0).
I think if you use not_equal for the condition, you can use incq, since inc sets
the zero flag.

http://codereview.chromium.org/6729016/diff/1/src/x64/macro-assembler-x64.cc
File src/x64/macro-assembler-x64.cc (right):

http://codereview.chromium.org/6729016/diff/1/src/x64/macro-assembler-x64.cc#...
src/x64/macro-assembler-x64.cc:1464: addl(dst, Operand(src, kSmiShift /
kBitsPerByte));
ASSERT(kSmiShift % kBitsPerByte == 0);

http://codereview.chromium.org/6729016/diff/1/src/x64/macro-assembler-x64.cc#...
src/x64/macro-assembler-x64.cc:2721: // Experimentation shows that the short
string loop is faster if length < 10.
Comment disagrees with chosen constant.

[Lasse Reichstein, 2011-03-24 12:14:08]

http://codereview.chromium.org/6729016/diff/1/src/x64/full-codegen-x64.cc
File src/x64/full-codegen-x64.cc (right):

http://codereview.chromium.org/6729016/diff/1/src/x64/full-codegen-x64.cc#new...
src/x64/full-codegen-x64.cc:3348: // Check that the separator is a flat ASCII
string.
Done.

http://codereview.chromium.org/6729016/diff/1/src/x64/full-codegen-x64.cc#new...
src/x64/full-codegen-x64.cc:3355: __ cmpb(scratch,
Immediate(ASCII_STRING_TYPE));
No idea. I'll just make it use the or-of-three-things like above.

http://codereview.chromium.org/6729016/diff/1/src/x64/full-codegen-x64.cc#new...
src/x64/full-codegen-x64.cc:3465: // a loop limit.
Good idea for another CL.

http://codereview.chromium.org/6729016/diff/1/src/x64/full-codegen-x64.cc#new...
src/x64/full-codegen-x64.cc:3490: //   rsp[0x10] (separator_operand): Address of
first char of separator.
That's better, thanks.

http://codereview.chromium.org/6729016/diff/1/src/x64/full-codegen-x64.cc#new...
src/x64/full-codegen-x64.cc:3506: __ j(less, &loop_3);  // Loop while (index <
0).
Done.

http://codereview.chromium.org/6729016/diff/1/src/x64/macro-assembler-x64.cc
File src/x64/macro-assembler-x64.cc (right):

http://codereview.chromium.org/6729016/diff/1/src/x64/macro-assembler-x64.cc#...
src/x64/macro-assembler-x64.cc:1464: addl(dst, Operand(src, kSmiShift /
kBitsPerByte));
Done.

http://codereview.chromium.org/6729016/diff/1/src/x64/macro-assembler-x64.cc#...
src/x64/macro-assembler-x64.cc:2721: // Experimentation shows that the short
string loop is faster if length < 10.
Comment removed.
We should do tests to see what the best constant is for X64.