View Single Post
Old 2019-02-18, 03:57   #2
P90 years forever!
Prime95's Avatar
Aug 2002
Yeehaw, FL

1BF716 Posts

Placeholder for bugs reported and bugs fixed.

1) Gerbicz PRP tests of (2^N+1)/factors fails in the last few iterations. Fixed in build 2.
2) Bulldozer users got a "no available FFT lengths" error trying to torture test with FMA3 or AVX FFTs. Fixed in build 2. Note that for the first time ever Bulldozer users can torture test using FMA3 and AVX FFTs. However, SSE2 FFTs may be more stressful.
3) x87 FFTs broken. Fixed in future build 2.
4) In weak torture test, Windows and Mac users are allowed to select options that cause "no FFT lengths" found errors. Fixed in build 6.
5) ScaleOutputFrequency=1 did not work testing ting numbers on AVX-512 machines. Fixed in build 3.
6) Older architectures, such as Pentium-4, may not select the best FFT implementation from gwnum.txt benchmark data. Fixed in build 3.
7) Older architectures, such as Pentium M or 4, do not display L1/L2 cache sizes. Fixed in build 3.
8) Torture test dialog box incorrectly calculated which FFTs will fit in the L3 or L4 cache. Fixed in build 7.
9) AVX-512 FFT fails on 18347731*109^1536-1 (a "rational" FFT -- no FFT weights). Fixed in 29.7.
10) FMA3 FFTs for exponents from 595.7M to 922.6M failed. Fixed in 29.7.
11) Some SkylakeX CPUs would say "no FFT sizes available" for default small torture test. Upper bound on FFT size changed to use L2+L3 cache size since the L3 cache is not inclusive. Fixed in 29.7.
12) The ability to stop individual torture test threads was not working. Running a benchmark without first stopping the torture test could cause a hang or crash. Fixed in 29.7.
13) Windows only!!! Zero-padded AVX-512 FFTs are not working. Fixed in 29.8.
14) If an error occurs writing worktodo.txt, then prime95 will hang at some later point in time. Fixed in 29.8 build 2.
15) If an error occurs deleting a worktodo.txt entry, the worker stopped computing. In 29.8 build 3, the worker will ignore the error and move onto the next entry in worktodo.txt.
16) Incorrect default value for torture test type in Linux menuing system. Fixed in 29.8 build 4.
17) URL of Mersenne Wiki was wrong. Fixed in 29.8 build 4.
18) Mojave dark mode not supported properly. Relinked with latest xcode. Fixed in 29.8 build 4, but untested for pre-Mojave OS.
19) Setting NumCPUs in local.txt to less than the number of L2 caches can cause the Torture Test dialog box to crash. Fixed in 29.8 build 4.
20) Benchmark states results are written to results.txt. Changed message to say results.bench.txt. Fixed in 29.8 build 5.
21) For PRP work, TF depth was not being written to worktodo.txt unless P-1 was required. This caused Test/Status to underestimate the chance that the PRP test would find a new Mersenne prime. Fixed in 29.8 build 5.
22) P-1 is frequently missing factors of numbers of the form 2*3^n+1. Fixed in 29.8 build 6.
23) In a throughput benchmark on machines with multiple L3 caches, some combinations of number-of-cores / number-of-workers would raise errors setting affinity or crash. Fixed in 29.8 build 6.

Last fiddled with by Prime95 on 2019-08-18 at 18:35
Prime95 is offline   Reply With Quote