llrCUDA

[Jean Penné]

Hi All,

This GPU version of the LLR program has the same features as the previous 3.8.3, but I think that the updates made from it deserve to change the version number.
Here are some of these updates :

- K*2^n+c numbers can now be tested using IBDWT, even if abs(c) != 1
- All 64 bits residues match with those obtained with LLR 3.8.24
- several issues found on llrCUDA 3.8.3b3 are now fixed.

Please, let me know if you have any problem to run the binary on Linux and/or to build it on your system.

I wish you many successes in prime hunting!
Best Regards,
Jean

[henryzz]

I have a more recent version of CUDA than you linked in(11.4 rather than 8). As such I tried to recompile. I got the following error:

/usr/local/cuda/bin/nvcc -g -O3 -I/usr/local/cuda/include -I. -Wno-deprecated-gpu-targets lprime.cu -c
Llr.c(7786): error: more than one instance of overloaded function "abs" matches the argument list:
            function "abs(int)"
/usr/local/cuda/include/crt/math_functions.h(213): here
            function "std::abs(__int128)"
/usr/include/c++/9/bits/std_abs.h(85): here
            function "std::abs(long double)"
/usr/include/c++/9/bits/std_abs.h(79): here
            function "std::abs(float)"
/usr/include/c++/9/bits/std_abs.h(75): here
            function "std::abs(double)"
/usr/include/c++/9/bits/std_abs.h(71): here
            function "std::abs(long long)"
/usr/include/c++/9/bits/std_abs.h(61): here
            function "std::abs(long)"
/usr/include/c++/9/bits/std_abs.h(56): here
            argument types are: (unsigned long)

Llr.c(7786): error: more than one instance of overloaded function "abs" matches the argument list:
            function "abs(int)"
/usr/local/cuda/include/crt/math_functions.h(213): here
            function "std::abs(__int128)"
/usr/include/c++/9/bits/std_abs.h(85): here
            function "std::abs(long double)"
/usr/include/c++/9/bits/std_abs.h(79): here
            function "std::abs(float)"
/usr/include/c++/9/bits/std_abs.h(75): here
            function "std::abs(double)"
/usr/include/c++/9/bits/std_abs.h(71): here
            function "std::abs(long long)"
/usr/include/c++/9/bits/std_abs.h(61): here
            function "std::abs(long)"
/usr/include/c++/9/bits/std_abs.h(56): here
            argument types are: (unsigned long)

Llr.c(7863): error: more than one instance of overloaded function "abs" matches the argument list:
            function "abs(int)"
/usr/local/cuda/include/crt/math_functions.h(213): here
            function "std::abs(__int128)"
/usr/include/c++/9/bits/std_abs.h(85): here
            function "std::abs(long double)"
/usr/include/c++/9/bits/std_abs.h(79): here
            function "std::abs(float)"
/usr/include/c++/9/bits/std_abs.h(75): here
            function "std::abs(double)"
/usr/include/c++/9/bits/std_abs.h(71): here
            function "std::abs(long long)"
/usr/include/c++/9/bits/std_abs.h(61): here
            function "std::abs(long)"
/usr/include/c++/9/bits/std_abs.h(56): here
            argument types are: (unsigned long)

Llr.c(7863): error: more than one instance of overloaded function "abs" matches the argument list:
            function "abs(int)"
/usr/local/cuda/include/crt/math_functions.h(213): here
            function "std::abs(__int128)"
/usr/include/c++/9/bits/std_abs.h(85): here
            function "std::abs(long double)"
/usr/include/c++/9/bits/std_abs.h(79): here
            function "std::abs(float)"
/usr/include/c++/9/bits/std_abs.h(75): here
            function "std::abs(double)"
/usr/include/c++/9/bits/std_abs.h(71): here
            function "std::abs(long long)"
/usr/include/c++/9/bits/std_abs.h(61): here
            function "std::abs(long)"
/usr/include/c++/9/bits/std_abs.h(56): here
            argument types are: (unsigned long)

4 errors detected in the compilation of "lprime.cu".
make: *** [Makefile:11: lprime.o] Error 1

[Jean Penné]

Hi All,

This GPU version of the LLR program has the same features as the previous 3.8.3, but I think that the updates made from it deserve to change the version number.
Here are some of these updates :

- In previous Version 3.8.4, one call to free() function was missing in
Gerbicz error checking code ; this caused an important memory leak...
This issue is now fixed here!

- K*2^n+c numbers can now be tested using IBDWT, even if abs(c) != 1
- All 64 bits residues match with those obtained with LLR 3.8.24
- several issues found on llrCUDA 3.8.3b3 are now fixed.

Please, let me know if you have any problem to run the binary on Linux and/or to build it on your system.

I wish you many successes in prime hunting!
Best Regards,
Jean

[henryzz]

The error in post #376 still exists. Once I avoided that(not sure if I correctly did) it compiles and runs.


I think Q is probably wrongly typed. Taking the absolute value of an unsigned integer doesn't make much sense. The relevant line numbers in llr.c are now 7843 and 7920 in the new version.

[Jean Penné]

Originally Posted by henryzz

The error in post #376 still exists. Once I avoided that(not sure if I correctly did) it compiles and runs.


I think Q is probably wrongly typed. Taking the absolute value of an unsigned integer doesn't make much sense. The relevant line numbers in llr.c are now 7843 and 7920 in the new version.

I think you are right!
So, I updated the source accordingly, suppressing the calls to abs() in the these two lines.
Also, I uploaded the compressed source to-day.
Please, would you tell me if the corrected source compiles and runs now OK in your environment.
Thank you by advance, and Best Regards,

Jean

[henryzz]

Originally Posted by Jean Penné

I think you are right!
So, I updated the source accordingly, suppressing the calls to abs() in the these two lines.
Also, I uploaded the compressed source to-day.
Please, would you tell me if the corrected source compiles and runs now OK in your environment.
Thank you by advance, and Best Regards,

Jean

That seems to compile and run. I am not sure I triggered those codepaths though.

[Jean Penné]

Originally Posted by henryzz

That seems to compile and run. I am not sure I triggered those codepaths though.

Many thanks!
Jean

[Jean Penné]

Hi All,

This GPU version of the LLR program has the same features as the previous 3.8.3, but I think that the updates made from it deserve to change the version number.
Here are some of these updates :

- May 2022 : The range of available FFT lengthes has been extended using
SSE2 Woltman tables. This allowed this version to test successfully
M82589933 in less than 8 days!

- The maximum value of bits per input double word that was defaulted to 35.0 is now defaulted to 37.0.
moreover, it may be changed by user, using -oMAXBPD=xx.x option.
It can be useful to decrease this value if too much round off errors occur...

- In previous Version 3.8.4, one call to free() function was missing in Gerbicz
error checking code ; this caused an important memory leak...
This issue is now fixed here!
No much new feature, but some improvements related to reliability and speed.

Please, let me know if you have any problem to run the binary on Linux and/or to build it on your system.

I wish you many successes in prime hunting!
Best Regards,
Jean