532240 - implement NJ_EXPANDED_LOADSTORE_SUPPORTED for x64 backend

Status: RESOLVED FIXED

(Core Graveyard :: Nanojit, defect)

Comment 1

[Steven Johnson]

Attachment: Patch

Initial implementation.

Comment 2

[Nicholas Nethercote [inactive]]

Comment on attachment 416667 [details] [diff] [review]
Patch

>@@ -438,6 +443,8 @@
>+    void Assembler::CVTSS2SD(R l, R r)  { emitprr(X64_cvtss2sd,l,r); asm_output("cvtss2sd %s, %s",RQ(l),RL(r)); }
>+    void Assembler::CVTSD2SS(R l, R r)  { emitprr(X64_cvtsd2ss,l,r); asm_output("cvtsd2ss %s, %s",RQ(l),RL(r)); }

I think the RQ/RL calls should be reversed in the second one.


>+    void Assembler::MOVBMR(R r1, I d, R r2)     { emitrm8(X64_movbmr,r1,d,r2); asm_output("movb %d(%s), %s",d,RQ(r1),RL(r2)); }
>+    void Assembler::MOVSMR(R r1, I d, R r2)     { emitrm_wide(X64_movsmr,r1,d,r2); asm_output("movs %d(%s), %s",d,RQ(r1),RL(r2)); }

For the first one, RL(r2) should be RB(r2), I think.  And we need to add RS() for the second one.

It's probably worth re-checking all the other RQ/RL calls for similar problems.

With those fixed, r=me.

Comment 3

[Steven Johnson]

Comment on attachment 416667 [details] [diff] [review]
Patch

buggy, new patch coming soon

Comment 4

[Steven Johnson]

Attachment: Patch #2

Improved (functional) patch.

Comment 5

[Steven Johnson]

pushed to nj-c as changeset:   1106:cd0b46495c95

Comment 6

[Steven Johnson]

pushed to tr as changeset:   3309:61d155d042c7

Comment 7

[Nicholas Nethercote [inactive]]

http://hg.mozilla.org/tracemonkey/rev/c1206e65e43f (on December 11, 2009)
http://hg.mozilla.org/tracemonkey/rev/55d3f8fc69fc

Comment 8

[Steven Johnson]

Pretty sure that this has been pushed to all relevant repos, shall we close it?

Comment 9

[Steven Johnson]

Nick, since this is (pretty much) closed out, shall I remove your outstanding r?

Comment 10

[Nicholas Nethercote [inactive]]

Comment on attachment 416813 [details] [diff] [review]
Patch #2

Sorry, missed that.  Fixed now.