Questions about the test

[SilverPoker]

Hi Marius, I have some questions about your code in main.cpp under the test folder
The function test_multiexp() has _size variable for setting different sizes, it works well using your initial code. However, when I changed the _size from 1<<18 to 1<<10. (Since 1<<10 needs shorter time), then the result doesn't match. I didn't find the reason why it mismatches, could you help me check why? The only change I made is changing "1<<18" to "1<<10"
I found that, it's also wrong if _size is less than 1<<14 (like 1<<13, 1<<12, 1<<10, and etc)
The error is below.

TEST MULTI_EXP
Field size: 96, Field count: 1024
Device FFT took 365
Host FFT took 4
Missmatch:
3832513521, 4114774966, 396713337, 473830846, 2659081024, 1847764696, 4867011, 2310256010, 2653309445, 1284315027, 4070234458, 3507217355, 518278877, 1714081604, 1468053814, 3791946234, 981627699, 3593337843, 358896652, 3408152632, 1224301292, 1561265861, 1175238034, 39629,
1218468944, 581656546, 1276554598, 4217098761, 339338039, 2296720683, 3362761856, 4177621453, 3482569788, 3598741332, 1672683522, 2895558447, 3948325594, 104581546, 909519637, 2054422881, 1545362927, 2025652791, 4237826918, 2431456373, 4175224577, 72776181, 119637567, 90433,
fft_test: /home/silverpoker/github/gpusnarks/test/../cuda/device_field.h:212: static void fields::Scalar::testEquality(fields::Scalar, fields::Scalar): Assertion `!"missmatch"' failed.
Aborted (core dumped)

[MariusVanDerWijden]

Hey,
you also need to change the size here: https://github.com/MariusVanDerWijden/gpusnarks/blob/master/cuda/multi_exp.cu#L30
Unfortunately the size has to be set at compile time.

[SilverPoker]

I see, thanks for your reply! I will read that part carefully. But it seems that 1<<14 or bigger than this doesn't need to change the size there? (Also, btw, does it support other sizes which are not powers of 2?)

[MariusVanDerWijden]

Currently it doesn't support sizes not powers of 2. In theory you can just leave the empty elements as zero. However since the input data is in monty form not real form, the "zero element" becomes the neutral element regarding multiplication in the finite field I think.