Most efficient way to do tensor-dot-product

[ClmnsRck]

C = sum(A(a,b,c) * B(c,d,e),c);

Is this the fastest way right now, to do a Tensor-Dot-Product(Tensor-Inner-product) in Blitz++?

[citibeth]

I don’t know. Blitz us more arrays not tensors. Have you considered Eigen’s tensor capabilities?

On Fri, Mar 29, 2019 at 07:26 ClmnsRck notifications@github.com wrote:

C = sum(A(a,b,c) * B(c,d,e),c);

Is this the fastest way right now, to do a Tensor-Dot-Product(Tensor-Inner-product) in Blitz++?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/blitzpp/blitz/issues/107, or mute the thread https://github.com/notifications/unsubscribe-auth/AB1cd9iJnJrLZ5LUm4vIQGUoiurKsUIQks5vbhSCgaJpZM4cSYpJ .

[ClmnsRck]

I am currently comparing several libraries with tensor-capabilities, such as Fastor, eigen, FTensor, blitz++ and xtensor, which is why i would like to be so fair as to use the most efficient way to do certain tasks.

EDIT: If you know of other libs, i would gladly test those to.

[citibeth]

Thanks I’m curious about the results!

I use blitz over Eigen for multi dim arrays for reasons other than tensor algorithms. (I’m also heavily reliant on Eigen for sparse on sparse matrix multiplication). The people at Eigen are really smart and, my guess is, hard to beat in speed.

On Fri, Mar 29, 2019 at 07:47 ClmnsRck notifications@github.com wrote:

I am currently comparing several libraries with tensor-capabilities, such as Fastor, eigen, FTensor, blitz++ and xtensor, which is why i would like to be so fair as to use the most efficient way to do certain tasks.

— You are receiving this because you commented.

Reply to this email directly, view it on GitHub https://github.com/blitzpp/blitz/issues/107#issuecomment-478003530, or mute the thread https://github.com/notifications/unsubscribe-auth/AB1cdzTPJPvBiG9DZgnynRBWG9hKNrVOks5vbhlygaJpZM4cSYpJ .

[ClmnsRck]

eigens 3D-Tensor multiplication is extremely well optimised, even to the point, that 1GB tensors times other 1GB tensors only take something aroung 5-10 seconds to complete. I'll make sure to come back to show the results once my tests are reliable enough.

[citibeth]

I would also try examples where youbiterate thrrough the loop explicitly

On Fri, Mar 29, 2019 at 07:54 ClmnsRck notifications@github.com wrote:

eigens 3D-Tensor multiplication is extremely well optimised, even to the point, that 1GB tensors times other 1GB tensors only take something aroung 5-10 seconds to complete. I'll make sure to come back to show the results once my tests are reliable enough.

— You are receiving this because you commented.

Reply to this email directly, view it on GitHub https://github.com/blitzpp/blitz/issues/107#issuecomment-478005761, or mute the thread https://github.com/notifications/unsubscribe-auth/AB1cd8h5diKdPfY298OPINZtBMz_BBLMks5vbhr8gaJpZM4cSYpJ .

[ClmnsRck]

So like a base case? sequentially going through every index, to see a baseline performance?

[citibeth]

I mean, just write it yourself as a nested loop like one would do in fortran

On Fri, Mar 29, 2019 at 08:01 ClmnsRck notifications@github.com wrote:

So like a base case? sequentially going through every index, to see a baseline performance?

— You are receiving this because you commented.

Reply to this email directly, view it on GitHub https://github.com/blitzpp/blitz/issues/107#issuecomment-478008139, or mute the thread https://github.com/notifications/unsubscribe-auth/AB1cdztBe6ly11DmtgMjh5qrqMFjhtW0ks5vbhyigaJpZM4cSYpJ .