diff options
author | Yann Herklotz <git@yannherklotz.com> | 2021-04-17 12:12:47 +0100 |
---|---|---|
committer | Yann Herklotz <git@yannherklotz.com> | 2021-04-17 12:12:47 +0100 |
commit | 2db9260b7ed82e4e097185a3334ae1f4c14a365d (patch) | |
tree | 51a64a4111cfbd0a9d2f931a5f6b3117db310077 /conclusion.tex | |
parent | c1796951a900b58dfe339809df8204221e8f1a54 (diff) | |
download | oopsla21_fvhls-2db9260b7ed82e4e097185a3334ae1f4c14a365d.tar.gz oopsla21_fvhls-2db9260b7ed82e4e097185a3334ae1f4c14a365d.zip |
Add
Diffstat (limited to 'conclusion.tex')
-rw-r--r-- | conclusion.tex | 2 |
1 files changed, 1 insertions, 1 deletions
diff --git a/conclusion.tex b/conclusion.tex index 5029bae..23476da 100644 --- a/conclusion.tex +++ b/conclusion.tex @@ -9,7 +9,7 @@ We evaluated \vericert{} against the existing \legup{} HLS tool on the \polybenc %Our Polybench hardware is guaranteed to preserve the C behaviour, whilst \legup{} cannot provide any such guarantee about its generated hardware. Currently, our hardware is \slowdownOrig$\times$ slower and \areaIncr$\times$ larger compared to \legup{}, though it is only \slowdownDiv$\times$ slower if inefficient divisions are removed. -There are abundant opportunities for improving \vericert{}'s performance. For instance, as discussed in Section~\ref{sec:evaluation}, simply replacing the na\"ive single-cycle division and modulo operations with C implementations increases clock frequency by $5.6\times$. +There are abundant opportunities for improving \vericert{}'s performance. For instance, as discussed in Section~\ref{sec:evaluation}, simply replacing the na\"ive single-cycle division and modulo operations with C implementations increases clock frequency by $8.2\times$. %Going forward, we envision introducing HLS-specific optimisations that are intended to improve the hardware quality of \vericert{}, whilst maintaining correctness. % However, to make the tool more usable there are many more optimisations that could be implemented to get the performance closer to \legup{}. Beyond this, we plan to implement scheduling and loop pipelining, since this allows more operations to be packed into fewer clock cycles; recent work by \citet{six+20} indicates how these scheduling algorithms can be implemented in \compcert. |