632037 - Optimize more for speed and less for size

Status: RESOLVED WONTFIX

(Firefox for Android Graveyard :: General, defect)

Description

[Mark Finkle (:mfinkle) (use needinfo?)]

Attachment: patch for Android

We currently optimize for space (Os). Doug suggested trying to optimize more for speed (O2) and test the results. We score consistently 15% faster on local Tp results. The down size is a space increase: 12330427 bytes -> 13211583 bytes (a 860KB increase)

Personally, I'd rather have the speed.

I think we should consider the same changes for Maemo too.

Comment 1

[Mark Finkle (:mfinkle) (use needinfo?)]

The size measurements were of the final APKs

Comment 2

[Brian Crowder]

What's the difference in size of the final installed libs, after a first-run?

Comment 3

[Mark Finkle (:mfinkle) (use needinfo?)]

(In reply to comment #2)
> What's the difference in size of the final installed libs, after a first-run?

I did not check> Here is the APK http://people.mozilla.com/~mfinkle/fennec/fennec-4.0b5pre.en-US.optimizedforspeed.apk

Comment 4

[Brad Lassey [:blassey] (use needinfo?)]

for anyone following long at home, mfinkle is going to get numbers for other available optimization flags.

Comment 5

[Mark Finkle (:mfinkle) (use needinfo?)]

I tired -O3 and found no measurable improvement over -O2. I did find the APK size jumped to 15186695 bytes, a 2789KB increase over the current -Os flag.

Comment 6

[Mark Finkle (:mfinkle) (use needinfo?)]

s/tired/tried

Comment 7

[Brad Lassey [:blassey] (use needinfo?)]

Comment on attachment 510279 [details] [diff] [review]
patch for Android

Please make the same change in js's configure

Comment 8

[Mark Finkle (:mfinkle) (use needinfo?)]

Attachment: patch 2

Adds JS configure and removes a redundant flag.

Comment 9

[Chris Leary [:cdleary] (not checking bugmail)]

Comment on attachment 510933 [details] [diff] [review]
patch 2

r+, but did we try with just-JS at -O3? Curious as to whether that would change anything in the engine -- we may have some important inlines that aren't always-inline-annotated that we'd lose out on with -O2.

Comment 10

[Mark Finkle (:mfinkle) (use needinfo?)]

Stuart suggested looking into more fine-grained use of opt flags so we don't grow the lib size too much.

Comment 11

[(dormant account)]

(In reply to comment #10)
> Stuart suggested looking into more fine-grained use of opt flags so we don't
> grow the lib size too much.

which  is what PGO does, ie bug 632954

Comment 12

[Mark Finkle (:mfinkle) (use needinfo?)]

pushed to test affects. we can backout and wait for PGO if needed:
http://hg.mozilla.org/mozilla-central/rev/c49735ad9fa0

Comment 13

[Stuart Parmenter]

The JS changes in this patch don't do anything to how JS is compiled, because JS overwrites with MODULE_OPTIMIZE_FLAGS

Comment 14

[Mark Finkle (:mfinkle) (use needinfo?)]

backed out. we have no post-landing perf data to verify this improved anything:
http://hg.mozilla.org/mozilla-central/rev/63817ae85b30

not worth the size increase