diff options
author | Nikolay Bogoychev <nheart@gmail.com> | 2019-05-21 18:07:26 +0300 |
---|---|---|
committer | Nikolay Bogoychev <nheart@gmail.com> | 2019-05-21 18:07:26 +0300 |
commit | 8efca65a7ef3696f3e66a3014e0acab203d1731d (patch) | |
tree | f51735da31085300d67b0ae920bea1ba664eafbb /README.md | |
parent | f9bbff340eb454ad7995737596028fabfdaa81f7 (diff) |
Update readme and remove outdated file
Diffstat (limited to 'README.md')
-rw-r--r-- | README.md | 4 |
1 files changed, 3 insertions, 1 deletions
@@ -32,12 +32,14 @@ intgemm::Int16::PrepareA(A, A_prepared, quant_mult, A_rows, width); /* Prepare B for multiplication. This is typically done offline. */ intgemm::Int16::PrepareB(B, B_prepared, quant_mult, width, B_cols); /* Multiply and produce results in C */ -intgemm::Int16::Multiply(A_prepared, B_prepared, C, 1.0 / (quant_mult * quant_mult), A_rows, width, B_cols); +intgemm::Int16::Multiply<intgemm::JustUnquantizeC>(A_prepared.get(), B_prepared.get(), intgemm::JustUnquantizeC(C.get(), 1.0 / (quant_mult * quant_mult)), A_rows, width, B_cols); ``` For 8-bit, use `Int8` instead of `Int16`. When repesented as floats, all of A, B, and C are in row-major format. +You can write your own PostProcessing functions on C and use them as a template argument to `Multiply`. For details, see [cops.h](cops.h). + ## Quantization Floating-point values are multiplied by a user-specified constant then rounded to an integer. |