Re: machine learning for loop unrolling

3 msg	FW: [gnu.org #336636] Needed Java Compilers for...
5 msg	Miscompilations for cris-elf on dataflow-branch
1 msg	LTO Status Report (2007-06-10)
2 msg	[RFC] Aggregate mapping
1 msg	Fwd: gcc 3.1.2 build on Solaris 2.9 - success
1 msg	general register used as saving comparison result
1 msg	gcc 3.1.2 build on Solaris 2.9
1 msg	MIPS results on the pointer_plus branch.
1 msg	gcc-4.3-20070608 is now available
1 msg	[fixed-point] A diff file for all changes
10 msg	Generating DWARF CIE info
2 msg	note_stores vs. PRE_DEC, POST_INC and so on
1 msg	ada, gcc 4.3 and darwin?
1 msg	room for GCC Summit

machine learning for loop unrolling
\ Stefan Ciobaca (8 Jun 2007)
. \ Stefan Ciobaca (8 Jun 2007)
. . \ Kenneth Hoste (8 Jun 2007)
. . \ Ken Raeburn (18 Jun 2007)
. \ Zdenek Dvorak (9 Jun 2007)
. . \ Maxim Kuvyrkov (9 Jun 2007)
. \ Grigori Fursin (10 Jun 2007)
. \ Stefan Ciobaca (15 Jun 2007)
. . \ Zdenek Dvorak (15 Jun 2007)

27 msg	GCC 4.3.0 Status Report (2007-06-07)
4 msg	Writing to Subversion via Git
1 msg	TeX sources for tuples document uploaded to the...
1 msg	[mingw] arm-elf-gcc build error with fixinclude...
4 msg	DWARF-2 unwinder versus MIPS n32

Subject:	Re: machine learning for loop unrolling
Group:	Gcc
From:	Ken Raeburn
Date:	18 Jun 2007

> - compile with the loop unrolled 1x, 2x, 4x, 8x, 16x, 32x and
> measure the time the benchmark takes

The optimal unrolling factor may not be a power of two, depending on
icache size (11 times the loop body size?), iteration count (13*n for
some unknown n?), and whether there are actions performed inside the
loop once or twice every N passes (for N not a power of two).

The powers of two would probably hit a lot of the common cases, but
you might want to throw in some intermediate values too, if it's too
costly to check all practical values.

Ken