605428 - Implement NJ_EXPANDED_LOADSTORE for SPARC

Status: RESOLVED FIXED

(Core Graveyard :: Nanojit, defect)

Comment 1

[Ginn Chen]

Attachment: patch

Rewrite asm_load64() like i386 backend.
The patch of Bug 605427 is required to make it work.

Verified with 'make check' on both nanojit-central and tracemonkey.

Comment 2

[William Maddox]

A quick look at lirasm/tests shows that LIR_ldf2d, for example, is given a few spot checks, but these changes need a better workout.  Tamarin makes little use of ldf2d, only for the MOPs used by Alchemy-generated code.  I'm not personally very familiar with TM internals, but I only see one useage of ldf2d.  Similar issues likely apply to some of the other expanded load/store instructions.  How extensively have you tested this patch for good coverage on all of the new instructions?  I'd like to know that some non-trivial test cases have exercised these.

Comment 3

[Ginn Chen]

I can see 6099 LIR_ldf2d in make check in tracemonkey.

Thousands of uses are in

trace-test/tests/v8-v5/check-earley-boyer.js
trace-test/tests/v8-v5/check-raytrace.js

For other instructions, 
LIR_sti2s 115143 uses
LIR_std2f 12049 uses

LIR_ldc2i 
LIR_lds2i
are not used by TM, they are tested by Nanojit bigendian/ldc2i.in, bigendian/lds2i.in

Comment 4

[William Maddox]

Comment on attachment 485261 [details] [diff] [review]
patch

R+

Can we start cleaning up the underrunProtect() mess?  See bug 608105.

nits:

In asm_store64, should assert that 'rv' is in FpRegs, as done for 'rr' in asm_load64.

Comment 5

[Ginn Chen]

http://hg.mozilla.org/projects/nanojit-central/rev/e3715e4e883c

Comment 6

[Nicholas Nethercote [inactive]]

http://hg.mozilla.org/tracemonkey/rev/a2d93fed3671

Comment 7

[Robert Sayre]

http://hg.mozilla.org/mozilla-central/rev/a2d93fed3671