Closed Bug 530828 Opened 15 years ago Closed 4 years ago

Force javascript engine to activate intel speedstep during dynamic recompile

Categories

(Core :: JavaScript Engine, enhancement)

x86
Windows Vista
enhancement
Not set
normal

Tracking

()

RESOLVED WONTFIX

People

(Reporter: fxtech, Unassigned)

Details

(Whiteboard: [platform-rel-Intel])

Attachments

(1 file)

User-Agent: Mozilla/5.0 (Windows; U; Windows NT 6.0; en-US; rv:1.9.3a1pre) Gecko/20091124 Minefield/3.7a1pre Build Identifier: Mozilla/5.0 (Windows; U; Windows NT 6.0; en-US; rv:1.9.3a1pre) Gecko/20091124 Minefield/3.7a1pre some week ago i noticed that firefox sunspider benchmark was a lot faster on windows 7 compared to vista , i understand why i have core 2 duo 2,6 ghz , under seven even if the cpu workload was not eavy OS trigger the cpu to the nominal frequency (avoiding speedstep downscale), under vista the cpu stay in low power mode 1,8 ghz, can speedstep be triggered by software ? could we ask to firefox to done it before start dynamic recompilation on Tracemonkey ? [code] TEST COMPARISON FROM TO DETAILS ============================================================================= ** TOTAL **: 1.43x as fast 1071.4ms +/- 1.3% 747.2ms +/- 3.1% significant ============================================================================= 3d: 1.40x as fast 194.4ms +/- 2.0% 139.2ms +/- 16.4% significant cube: 1.42x as fast 55.0ms +/- 0.0% 38.8ms +/- 11.6% significant morph: 1.45x as fast 63.4ms +/- 1.1% 43.6ms +/- 1.6% significant raytrace: 1.34x as fast 76.0ms +/- 5.7% 56.8ms +/- 32.1% significant access: 1.50x as fast 164.6ms +/- 1.4% 109.8ms +/- 0.9% significant binary-trees: 1.53x as fast 27.2ms +/- 2.0% 17.8ms +/- 5.8% significant fannkuch: 1.51x as fast 96.4ms +/- 2.2% 64.0ms +/- 0.0% significant nbody: 1.43x as fast 26.4ms +/- 4.2% 18.4ms +/- 3.7% significant nsieve: 1.52x as fast 14.6ms +/- 4.7% 9.6ms +/- 7.1% significant bitops: 1.50x as fast 49.6ms +/- 2.2% 33.0ms +/- 2.7% significant 3bit-bits-in-byte: - 1.8ms +/- 30.9% 1.4ms +/- 48.6% bits-in-byte: 1.55x as fast 15.2ms +/- 3.7% 9.8ms +/- 5.7% significant bitwise-and: 1.40x as fast 2.8ms +/- 19.9% 2.0ms +/- 0.0% significant nsieve-bits: 1.51x as fast 29.8ms +/- 1.9% 19.8ms +/- 2.8% significant controlflow: 1.58x as fast 11.4ms +/- 22.6% 7.2ms +/- 7.7% significant recursive: 1.58x as fast 11.4ms +/- 22.6% 7.2ms +/- 7.7% significant crypto: 1.55x as fast 65.0ms +/- 3.0% 42.0ms +/- 3.6% significant aes: 1.52x as fast 38.2ms +/- 5.3% 25.2ms +/- 5.4% significant md5: 1.56x as fast 16.8ms +/- 3.3% 10.8ms +/- 5.1% significant sha1: 1.67x as fast 10.0ms +/- 0.0% 6.0ms +/- 0.0% significant date: 1.48x as fast 167.6ms +/- 3.5% 113.0ms +/- 0.8% significant format-tofte: 1.49x as fast 83.2ms +/- 1.3% 56.0ms +/- 1.6% significant format-xparb: 1.48x as fast 84.4ms +/- 6.6% 57.0ms +/- 0.0% significant math: 1.52x as fast 69.8ms +/- 2.0% 46.0ms +/- 3.8% significant cordic: 1.51x as fast 38.6ms +/- 1.8% 25.6ms +/- 2.7% significant partial-sums: 1.55x as fast 22.6ms +/- 3.0% 14.6ms +/- 4.7% significant spectral-norm: 1.48x as fast 8.6ms +/- 7.9% 5.8ms +/- 9.6% significant regexp: 1.28x as fast 63.2ms +/- 12.5% 49.4ms +/- 5.5% significant dna: 1.28x as fast 63.2ms +/- 12.5% 49.4ms +/- 5.5% significant string: 1.38x as fast 285.8ms +/- 3.3% 207.6ms +/- 7.8% significant base64: 1.57x as fast 12.6ms +/- 5.4% 8.0ms +/- 0.0% significant fasta: 1.54x as fast 69.2ms +/- 1.5% 45.0ms +/- 2.0% significant tagcloud: 1.30x as fast 86.4ms +/- 8.4% 66.4ms +/- 2.8% significant unpack-code: 1.31x as fast 83.6ms +/- 5.0% 63.8ms +/- 25.3% significant validate-input: 1.39x as fast 34.0ms +/- 2.6% 24.4ms +/- 10.6% significant [/code] Reproducible: Always Steps to Reproduce: 1. start sunspider benchmark under Vista 2. step up of intel speedstep is not triggered (reduced frequency) 3. start sunspider benchmark under Win 7 4. step up of intel speedstep is triggered (nominal frequency) Thread opened on forum http://forums.mozillazine.org/viewtopic.php?f=23&t=1609925
Component: General → JavaScript Engine
Product: Firefox → Core
Version: unspecified → Trunk
Assignee: nobody → general
QA Contact: general → general
Isn't it better to leave such a decision that has a broad system-wide impact to the system Power Manager instead of the browser? Attached is a screenshot of the Lenovo ThinkPad Power Manager. You can define and configure multiple "power schemes" which control how the system should operate (e.g., depending on whether the system is connected to a power outlet or running on a battery). Of course, one can think of having such options in the browser as well, but this seems more a system-wide decision. -moh
Assignee: general → nobody
Whiteboard: [platform-rel-Intel]
platform-rel: --- → ?
platform-rel: ? → ---

Improving performance on Windows Vista seems hardly relevant anymore...

Status: UNCONFIRMED → RESOLVED
Closed: 4 years ago
Resolution: --- → WONTFIX
You need to log in before you can comment on or make changes to this bug.

Attachment

General

Creator:
Created:
Updated:
Size: