[KJ] RFC: New todo for loops

[KJ] RFC: New todo for loops

[Daniel Marjamäki]

Hello friends!

I have a suggestion for the TODO.
Making loops faster is important. Therefore in the loops we should use pointers instead of arrays.

Example (fs/reiser/file.c:268):
	int i;
	for (i = 0; i < blocks_to_allocate; i++)
		allocated_blocks[i] = cpu_to_le32(allocated_blocks[i]);

This could be changed to something like:
	struct b_blocknr_t *block, *block_last;
	block_last = &allocated_blocks[allocated_blocks-1];
	for (block = allocated_blocks; block <= block_last; block++)
			*block = cpu_to_le32(*block);

Best Regards,
Daniel Marjamäki

_______________________________________________
Kernel-janitors mailing list
Kernel-janitors@lists.osdl.org
https://lists.osdl.org/mailman/listinfo/kernel-janitors

Re: [KJ] RFC: New todo for loops

[Matthew Wilcox]

[-- Attachment #1: Type: text/plain, Size: 429 bytes --]

On Sun, Dec 18, 2005 at 02:29:53PM +0100, Daniel Marjam?ki wrote:
> I have a suggestion for the TODO.
> Making loops faster is important. Therefore in the loops we should use pointers instead of arrays.

Do you have evidence this actually produces better code?  GCC is pretty
smart, you know.  And the exammple you posted is harder to understand.
BTW, examples using gcc 2.95 don't count; the example has to be for 4.0
or later.

[-- Attachment #2: Type: text/plain, Size: 168 bytes --]

_______________________________________________
Kernel-janitors mailing list
Kernel-janitors@lists.osdl.org
https://lists.osdl.org/mailman/listinfo/kernel-janitors

Re: [KJ] RFC: New todo for loops

[Ricardo Nabinger Sanchez]

Quoting  Daniel Marjamäki <daniel.marjamaki@comhem.se>
Sent on  Sun, 18 Dec 2005 14:29:53 +0100

Hello,

> I have a suggestion for the TODO.
> Making loops faster is important. Therefore in the loops we should use
> pointers instead of arrays.

can you point data that support your statement?

AMD recommends[1] the opposite.

Regards.

[1] AMD Athlon Processor x86 Code Optimization Guide
http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/22007.pdf

-- 
Ricardo Nabinger Sanchez
GNU/Linux #140696 [http://counter.li.org]
Slackware Linux + FreeBSD

  Left to themselves, things tend to go from bad to worse.
_______________________________________________
Kernel-janitors mailing list
Kernel-janitors@lists.osdl.org
https://lists.osdl.org/mailman/listinfo/kernel-janitors

Re: [KJ] RFC: New todo for loops

[walter harms]

Hi daniel,
  i do not like the idea. the code is harder to read and debug, smashing 
pointers is easy done. i would not trade that for speed.

SFIK, array[i] is read as *(array+i) with c-compilers. nice example is 
in "how to write unmaintainable code".


re,
  walter


Daniel Marjamäki wrote:
> Hello friends!
> 
> I have a suggestion for the TODO.
> Making loops faster is important. Therefore in the loops we should use pointers instead of arrays.
> 
> Example (fs/reiser/file.c:268):
> 	int i;
> 	for (i = 0; i < blocks_to_allocate; i++)
> 		allocated_blocks[i] = cpu_to_le32(allocated_blocks[i]);
> 
> This could be changed to something like:
> 	struct b_blocknr_t *block, *block_last;
> 	block_last = &allocated_blocks[allocated_blocks-1];
> 	for (block = allocated_blocks; block <= block_last; block++)
> 			*block = cpu_to_le32(*block);
> 
> Best Regards,
> Daniel Marjamäki
> 
> _______________________________________________
> Kernel-janitors mailing list
> Kernel-janitors@lists.osdl.org
> https://lists.osdl.org/mailman/listinfo/kernel-janitors
> 
> 
> 
_______________________________________________
Kernel-janitors mailing list
Kernel-janitors@lists.osdl.org
https://lists.osdl.org/mailman/listinfo/kernel-janitors