Re: --enable-haifa
[EMAIL PROTECTED] (W Gerald Hicks) writes: > I don't have a shiny new K7 yet, where I might expect the haifa > build to make more of a difference than my crusty old Pentium... Processors with out-of-order execution benefit *less* from scheduling than non-OOO superscalar processors. To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-hackers" in the body of the message
Re: --enable-haifa
> Out of curiosity, were the benchmarks done with any of the Haifa > command-line-options, notably -fsched-interblock, -fsched-spec, > -fsched-spec-load, and -fbranch-count-reg ? > (toplev.c, grep for '#ifdef HAIFA', and read the comments > scattered elsewhere (haifa-sched.c)) Not yet, but thanks for the pointers! :-) I don't have a shiny new K7 yet, where I might expect the haifa build to make more of a difference than my crusty old Pentium... Cheers, Jerry Hicks [EMAIL PROTECTED] To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-hackers" in the body of the message
Re: --enable-haifa
Out of curiosity, were the benchmarks done with any of the Haifa command-line-options, notably -fsched-interblock, -fsched-spec, -fsched-spec-load, and -fbranch-count-reg ? (toplev.c, grep for '#ifdef HAIFA', and read the comments scattered elsewhere (haifa-sched.c)) Sean. To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-hackers" in the body of the message
Re: --enable-haifa
On Thu, 14 Oct 1999, Narvi wrote: [snip] > According to my vague recollectiosn from the early times of egcs when > faifa was integrated, etc. part of the problem is that to have haifa be > really effective, all old kluges need be removed... > > Something that back then happened/had happened only to a few cpus, like > HP-PA. Not sure what may have happened since, but I doubt somebody has > redone x86 code generation and fully integrated haifa ins scheduler. Richard Henderson rewrote the whole ia32 backend and the merger was completed on September 2, 1999 . I don't know if haifa was fully integrated, but I bet it was (IIRC, R. Henderson did this for alpha). Martin > > > -- > > Ollivier ROBERT -=- FreeBSD: The Power to Serve! -=- [EMAIL PROTECTED] > > FreeBSD keltia.freenix.fr 4.0-CURRENT #74: Thu Sep 9 00:20:51 CEST 1999 > > > > > > To Unsubscribe: send mail to [EMAIL PROTECTED] > with "unsubscribe freebsd-hackers" in the body of the message > To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-hackers" in the body of the message
Re: --enable-haifa
On Thu, 14 Oct 1999, Ollivier Robert wrote: > According to W Gerald Hicks: > > Just curious what effect using the --enable-haifa flag for building > > gcc-2.95.1/x86 would have so I did a comparison using the Dhrystone > > benchmark from /usr/ports/benchmarks/bytebench. > > I think the Haifa scheduler is only really effective on pure RISC processors > like the Alpha or PA-8000. I remember doing some tests with older versions of > egcs and not seeing any significant changes... According to my vague recollectiosn from the early times of egcs when faifa was integrated, etc. part of the problem is that to have haifa be really effective, all old kluges need be removed... Something that back then happened/had happened only to a few cpus, like HP-PA. Not sure what may have happened since, but I doubt somebody has redone x86 code generation and fully integrated haifa ins scheduler. > -- > Ollivier ROBERT -=- FreeBSD: The Power to Serve! -=- [EMAIL PROTECTED] > FreeBSD keltia.freenix.fr 4.0-CURRENT #74: Thu Sep 9 00:20:51 CEST 1999 > To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-hackers" in the body of the message
Re: --enable-haifa
According to W Gerald Hicks: > Just curious what effect using the --enable-haifa flag for building > gcc-2.95.1/x86 would have so I did a comparison using the Dhrystone > benchmark from /usr/ports/benchmarks/bytebench. I think the Haifa scheduler is only really effective on pure RISC processors like the Alpha or PA-8000. I remember doing some tests with older versions of egcs and not seeing any significant changes... -- Ollivier ROBERT -=- FreeBSD: The Power to Serve! -=- [EMAIL PROTECTED] FreeBSD keltia.freenix.fr 4.0-CURRENT #74: Thu Sep 9 00:20:51 CEST 1999 To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-hackers" in the body of the message
Re: --enable-haifa
On Wed, 13 Oct 1999, W Gerald Hicks wrote: > Just curious what effect using the --enable-haifa flag for building > gcc-2.95.1/x86 would have so I did a comparison using the Dhrystone > benchmark from /usr/ports/benchmarks/bytebench. This seems marginal, in other words. How did the results vary when you ran the tests multiple times? Kris XOR for AES -- join the campaign! To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-hackers" in the body of the message
--enable-haifa
Just curious what effect using the --enable-haifa flag for building gcc-2.95.1/x86 would have so I did a comparison using the Dhrystone benchmark from /usr/ports/benchmarks/bytebench. This was run on a Ziatech 200Mhz Pentium cPCI system. I don't know if this is really worth further testing but it seems interesting. Cheers, Jerry Hicks [EMAIL PROTECTED] stock -current (1999/10/05) gcc = Dhrystone 2 without register variables 321955.8 lps (10 secs, 6 samples) BASELINE RESULT INDEX Dhrystone 2 without register variables 22366.3 321955.8 14.4 i386-portbld-FreeBSD4.0 --enable-haifa = Dhrystone 2 without register variables 323767.4 lps (10 secs, 6 samples) BASELINE RESULT INDEX Dhrystone 2 without register variables 22366.3 323767.4 14.5 To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-hackers" in the body of the message