Closed adampauls closed 2 years ago
Note: the example shown in the but can be slightly further minimized with
val outputDim = 1
val inputDim = 3
There seems to be more than one place (like https://github.com/luhenry/netlib/blob/master/blas/src/main/java/dev/ludovic/netlib/blas/Java11BLAS.java#L1832-L1833) where
sum11 = Math.fma(a10, b01, sum11);
sum11 = Math.fma(a10, b02, sum12);
appears in the s
version but the d
version has
sum11 = Math.fma(a10, b01, sum11);
sum12 = Math.fma(a10, b02, sum12);
@adampauls yes, seems like a bug indeed. Let me get a fix to v2.2.1
right away
Thanks for the quick fix!
Escalating this issue, originally found when using Breeze. Executing the offending code snippet arrives in Java11Blas.sgemmTN, so I guessing the bug is somewhere in there.