sim: Pre-compute the clock period

Information
Submitter:	Andreas Hansson
Repository:	gem5
Branch:	default
Bugs:
Depends On:
Reviewers
Groups:	Default
People:

Description

Changeset 9761:a04648a0d2c6
---------------------------
sim: Pre-compute the clock period

This patch changes the source and derived clock domain such that no
virtual functions or multiplications are needed when calling
clockPeriod. Instead, the clock period is pre-computed when any
changes occur. For this to be possible, a clock domain must be aware
of its children and the corresponding data structures and methods have
been added to facilitate this.

Testing Done

All regressions pass

I guess I spoke too soon in the other patch.

Nilay Vaish June 1, 2013, 9:18 a.m. (June 1, 2013, 9:18 a.m.)

I think you should merge this patch with the clock domain patch when you commit this code.

Andreas Hansson June 2, 2013, 2:23 a.m. (June 2, 2013, 2:23 a.m.)
```
Yeah I'll fold the two together before pushing :-)
```

Description:

~		Changeset 9741:32c0e8ed4c8a
	~	Changeset 9761:a04648a0d2c6

		sim: Pre-compute the clock period

		This patch changes the source and derived clock domain such that no
		virtual functions or multiplications are needed when calling
		clockPeriod. Instead, the clock period is pre-computed when any
		changes occur. For this to be possible, a clock domain must be aware
		of its children and the corresponding data structures and methods have
		been added to facilitate this.

Diff:

Revision 2 (+77 -43)

Show changes

	src/sim/clock_domain.hh
	src/sim/clock_domain.cc

Status: Discarded

Out of curiosity, was this any faster, or did you really just do it because I thought it was crazy.  While I feel better about the code working this way, you indicated that there wasn't much of a performance impact.  I hope you didn't do this because I objected.  I really do feel that a performance measurement is worth more than gut reaction.  (Perhaps you didn't do your performance test using something that stressed the clock and didn't want to?)

Andreas Hansson June 15, 2013, 10:24 p.m. (June 15, 2013, 10:24 p.m.)

I ran the regressions multiple time and it did have an impact in the end. After compensating for fluctuations, it was probably in the order of ~5% of the regressions time, so closer to ~10% for the clock intensive runs. In short, your gut feeling was pretty spot on :-)

You have a pending review.

Review Board 2.0.15

This change has been discarded.

Screenshots

Files

Description:

Changeset 9741:32c0e8ed4c8a

Changeset 9761:a04648a0d2c6

Diff:

Status: Discarded