LLR 3.8.4 / gwnum 26.4 Mac 32bit errors

IBethune · 2010-11-08, 13:51

Hi,

I've been doing the MacIntel builds of LLR for a while now, but since LLR 3.8.2/gwnum 26.2, the code won't run correctly on 32 bits Macs (Intel Core 1 / Prescott CPUs), but continues to work on more recent 64 bit CPUs (Core 2, Core i3/5/7) as expected.

The symptom is that gwsetup fails to find a suitable FFT length for the input, reporting:

Code:

Number sent to gwsetup is too large for the FFTs to handle.

This corresponds to the error code GWERROR_TOO_LARGE. I had a dig through the source as far as possible, but it seems the relevant sizes are set by the assembly code (jmptab->max_exp), at which point I gave up trying to figure out what was going on.

An example input that triggers this error would be 1005*2^109568+1(which is prime). This worked correctly with LLR 3.8.1/gwnum 25.14

Is this some known restriction/bug with gwnum 26? Has anyone seen any similar problems with prime95 or PFGW ?

Cheers

- Iain

Prime95 · 2010-11-08, 14:48

I found the problem.

For the Mac build some FFT implementations are not included (like the AMD-optimized FFTs) to reduce the size of the executable. For the 32-bit Mac case, a typo caused too many FFTs to be excluded.

This will be fixed next source code release.

IBethune · 2010-11-08, 15:26

Thanks for the quick response!

Any idea when the next release might be? In the meantime we'll continue to use the older (25.14-based) LLR for 32 bit Macs.

Cheers

- Iain

Prime95 · 2010-11-13, 23:41

Try http://www.mersenne.org/cpus/source264.zip
It'll be there until I move it to its permanent home.

2010-11-08, 13:51	#1
IBethune Nov 2010 19₁₆ Posts	LLR 3.8.4 / gwnum 26.4 Mac 32bit errors Hi, I've been doing the MacIntel builds of LLR for a while now, but since LLR 3.8.2/gwnum 26.2, the code won't run correctly on 32 bits Macs (Intel Core 1 / Prescott CPUs), but continues to work on more recent 64 bit CPUs (Core 2, Core i3/5/7) as expected. The symptom is that gwsetup fails to find a suitable FFT length for the input, reporting: Code: Number sent to gwsetup is too large for the FFTs to handle. This corresponds to the error code GWERROR_TOO_LARGE. I had a dig through the source as far as possible, but it seems the relevant sizes are set by the assembly code (jmptab->max_exp), at which point I gave up trying to figure out what was going on. An example input that triggers this error would be 1005*2^109568+1(which is prime). This worked correctly with LLR 3.8.1/gwnum 25.14 Is this some known restriction/bug with gwnum 26? Has anyone seen any similar problems with prime95 or PFGW ? Cheers - Iain

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
GWNUM issue	ET_	Software	4	2017-06-15 09:52
Windows 32bit vs 64bit mfaktc/cudalucas	bcp19	GPU Computing	20	2012-03-11 01:24
LLR V3.8.2 using gwnum 26.2 is available!	Jean Penné	Software	25	2010-11-01 15:18
GWNUM?	Unregistered	Information & Answers	3	2010-09-12 19:52
GWNUM as DLL?	Cyclamen Persicum	Software	1	2007-01-02 20:53

2010-11-08, 14:48	#2
Prime95 P90 years forever! Aug 2002 Yeehaw, FL 2·53·71 Posts	I found the problem. For the Mac build some FFT implementations are not included (like the AMD-optimized FFTs) to reduce the size of the executable. For the 32-bit Mac case, a typo caused too many FFTs to be excluded. This will be fixed next source code release.

2010-11-08, 15:26	#3
IBethune Nov 2010 5² Posts	Thanks for the quick response! Any idea when the next release might be? In the meantime we'll continue to use the older (25.14-based) LLR for 32 bit Macs. Cheers - Iain

2010-11-13, 23:41	#4
Prime95 P90 years forever! Aug 2002 Yeehaw, FL 2·53·71 Posts	Try http://www.mersenne.org/cpus/source264.zip It'll be there until I move it to its permanent home.