476716 - Unmacroize ip_adj helpers.

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect)

Description

[Jason Orendorff [:jorendorff]]

Attachment: v1

I did this in the process of debugging bug 462027 comment 54, and I'd like to commit the change.

Comment 1

[Brendan Eich [:brendan]]

Need to get this into 1.9.1 to avoid merge hell.

/be

Comment 2

[Brendan Eich [:brendan]]

Comment on attachment 360365 [details] [diff] [review]
v1

The NAMING_CONVENTION shouts MACRO! but that's ok -- can clean up later, or leave as a warning and a mystery ;-).

/be

Comment 3

[Brendan Eich [:brendan]]

Comment on attachment 360365 [details] [diff] [review]
v1

>+FI_IMACRO_PC(const FrameInfo &fi, JSStackFrame *fp)
>+{
>+    if (IMACRO_PC_ADJ(fi.ip_adj))
>+        return imacro_code[*FI_SCRIPT_PC(fi, fp)] + IMACRO_PC_ADJ(fi.ip_adj);
>+    return NULL;
>+}
[...]
>+DECODE_IP_ADJ(uintptr_t ip, JSStackFrame *fp)
>+{
>+    if (IMACRO_PC_ADJ(ip)) {
>+        fp->imacpc = fp->script->code + SCRIPT_PC_ADJ(ip);
>+        fp->regs->pc = imacro_code[*fp->imacpc] + IMACRO_PC_ADJ(ip);

Now that these are inlines, would it be clearer to avoid repeating calls and counting on CSE, and use a local var to capture the result of one call to IMACRO_PC_ADJ(ip)?

/be

Comment 4

[Jason Orendorff [:jorendorff]]

Attachment: v2

Completely different patch.

A minor drawback with this approach is that the numbers in the debug spew for LIR guard instructions are now pointers, not bytecode offsets:

    xf1: xf eq1 -> 0x30db81:0x0 sp+0 rp+0

I can live with it.

Comment 5

[Brendan Eich [:brendan]]

Comment on attachment 361607 [details] [diff] [review]
v2

Great! Bye-byte imacro_code. Thanks,

/be

Comment 6

[Jason Orendorff [:jorendorff]]

http://hg.mozilla.org/tracemonkey/rev/ab8216b1e596

Comment 7

[Jason Orendorff [:jorendorff]]

Reopening because this is nominated for 1.9.1 (nothing wrong with the patch, which is still in tracemonkey).

Comment 8

[Robert Sayre]

http://hg.mozilla.org/releases/mozilla-1.9.1/rev/8b558153923a