[QUOTE=S485122;123622]No : you need to have a 64 bits capable CPU to run a 64 bits OS. And you need a 64 bits OS to run 64 bits programs.

Jacob[/QUOTE]

That isn't true on OS X. I could build and run 64-bit apps on 32-bit versions of the OS. I couldn't run 64-bit GUI apps, just 64-bit command line apps via the Terminal app. I'm unfamiliar with other OS's, but I don't see why you couldn't use -m64 to build a 64-bit app to run from a unix shell prompt.

I think OS X 10.4 and up has shipped with a 64-bit version of libsystem and ABI.

It's not going to work in a 32-bit version of Ubuntu.

[QUOTE=AES;123687]I think OS X 10.4 and up has shipped with a 64-bit version of libsystem and ABI.

It's not going to work in a 32-bit version of Ubuntu.[/QUOTE]

That's possible, but as I no longer have a 10.4 install, I can't verify it directly. I'm sure that Apple's website probably has is somewhere, but that doesn't help you.

Pentium-M 1.6GHz - Win32
[code] Bit scalar scalar vector vector
level i386 sse2 i386 sse2
----- ------ ------ ------ ------
58-59 74.0 85.9 39.9 44.0
59-60 76.0 93.4 41.0 44.5
60-61 75.2 88.8 41.5 44.1
61-62 75.8 86.8 42.5 46.0[/code]
Athlon 64 1.8GHz - Win32
[code] Bit scalar scalar vector vector
level i386 sse2 i386 sse2
----- ------ ------ ------ ------
58-59 68.3 66.0 34.6 31.4
59-60 67.9 66.0 35.2 31.3
60-61 67.4 66.0 34.9 31.1
61-62 68.2 66.0 35.7 30.9[/code]

Thanks everyone for the benchmarks. The vector powmod beats the scalar by a wide margin even if SSE2 is not much use on some machines (probably I should disable SSE2 if a pentium M is detected). So I can't explain the A64 slowdown at all :-(.

[QUOTE=Anonymous;123617]Maybe I phrased my question wrong. Actually, what I was wondering is, can I run 64-bit sr(x)sieve on a 32-bit OS as long as I have a 64-bit-compatible CPU? :smile:[/QUOTE]

No. It is possible to build 64-bit executables on a 32-bit Linux system with -m64 (I use a 32-bit P4 for development), but to actually execute them you need to be running a 64-bit kernel.

I sieve with a amd Semperon 1790 mhz 512MB and since I started with 1.7.6 I have noticed a big hit with all my programs web pages take forever to load so I went back to 1.6.18 and everything works fine. On my current range with 1.7.6 eta was 1 feb and with 1.6.18 went to 3 feb I lost about 34k p/sec going back to 1.6.18 but without lag with my programs so I'll stay with 1.6.18 for now

What's strange is that regardless of A64 speed the benchmark shows almost the same results :question:

[QUOTE=Cruelty;123879]What's strange is that regardless of A64 speed the benchmark shows almost the same results :question:[/QUOTE]

The benchmark is measured in clock cycles, so it should be independent of clock speed.

[QUOTE=Cruelty;123770]Pentium-M 1.6GHz - Win32
[code] Bit scalar scalar vector vector
level i386 sse2 i386 sse2
----- ------ ------ ------ ------
58-59 74.0 85.9 39.9 44.0
59-60 76.0 93.4 41.0 44.5
60-61 75.2 88.8 41.5 44.1
61-62 75.8 86.8 42.5 46.0[/code]
[/QUOTE]

It seems that there are two quite different CPUs both called Pentium-M: One identifies as CPU family 6, the other as CPU family 15. Do you know which one this is? I think the info will be in the CPU information panel, but if you can't find it then I can write a small program to test which it is. Thanks.

Here you go.

I found a build in the 64-bit PPC code which affects srsieve, sr1sieve, sr2sieve, sr5sieve, and gcwsieve. From what I have seen it hasn't led to any missed factors, but it can cause a segfault.

The specific fix is to replace the "adde" assembler instruction with "addc" in asm-ppc64.h and asm-ppc64.c. It might be in some other sources, but I haven't looked further yet.

I've notified Geoff of the problem, but he hasn't replied yet, which is why I am posting here.