504219 - TM: Inline malloc counting function

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect)

Description

[Andreas Gal :gal]

About 3% speedup on non-empty object allocation, i.e. while (true) ([1]).

Comment 1

[Andreas Gal :gal]

Attachment: patch

Comment 2

[Brendan Eich [:brendan]]

(In reply to comment #0)
> About 3% speedup on non-empty object allocation, i.e. while (true) ([1]).

BTW, won't we at some point notice that the value is unused and (now that there's no observable effect like calling a hijacked Array implied here, as was the case with ECMA-262 Edition 3) the expression has no side effects, and optimize away the allocation altogether? Better if the loop has a counter and limit ;-), but I am really fishing for a bug on file.

/be

Comment 3

[Jeff Walden [:Waldo]]

Comment on attachment 388600 [details] [diff] [review]
patch

>diff --git a/js/src/jscntxt.h b/js/src/jscntxt.h

>+    /* Call this after succesful malloc of memory for GC-related things. */
>+    inline void
>+    updateMallocCounter(size_t nbytes)

inline here, inside the class declaration with a method definition in-place which already implies inline, is redundant as far as I understand, would prefer it removed.


>+    {
>+        uint32 *pbytes, bytes;
>+
>+#ifdef JS_THREADSAFE
>+        pbytes = &thread->gcMallocBytes;
>+#else
>+        pbytes = &runtime->gcMallocBytes;
>+#endif
>+        bytes = *pbytes;
>+        *pbytes = ((uint32)-1 - bytes <= nbytes) ? (uint32)-1 : bytes + nbytes;

I thought we had a macro like CACHE_LOCUS(cx) that hid the thread/runtime distinction, but I can't find it on a quick search, no matter.  Please convert the C-style (uint32)-1 casts to C++-style uint32(-1) casts while you're here.

Comment 4

[Andreas Gal :gal]

Attachment: patch

Comment 5

[Jeff Walden [:Waldo]]

Comment on attachment 388609 [details] [diff] [review]
patch

Looks cool, although it occurs to me that we should add SIZE_MAX and friends to our stdint.h replacement for more standards-stylistic code, sometime.  I filed bug 504233 to do this.

Comment 6

[Andreas Gal :gal]

http://hg.mozilla.org/tracemonkey/rev/85875bc69317

Comment 7

[Robert Sayre]

http://hg.mozilla.org/mozilla-central/rev/85875bc69317