fix BNNM docs
bug: 22507076
Change-Id: I5fdffdde597d97b1dde77574d096b81bff4aa382
(cherry picked from commit d7d413ad0b)
This commit is contained in:
@@ -1554,7 +1554,7 @@ public final class ScriptIntrinsicBLAS extends ScriptIntrinsic {
|
||||
|
||||
|
||||
/**
|
||||
* 8-bit GEMM-like operation for neural networks: C = B.transposed() * A
|
||||
* 8-bit GEMM-like operation for neural networks: C = A * Transpose(B)
|
||||
* Calculations are done in 1.10.21 fixed-point format for the final output,
|
||||
* just before there's a shift down to drop the fractional parts. The output
|
||||
* values are gated to 0 to 255 to fit in a byte, but the 10-bit format
|
||||
|
||||
Reference in New Issue
Block a user