diff options
| author | Tom Smeding <tom@tomsmeding.com> | 2025-03-14 21:57:56 +0100 |
|---|---|---|
| committer | Tom Smeding <tom@tomsmeding.com> | 2025-03-14 21:58:51 +0100 |
| commit | 6276ed3c7bcd20c8b860e1275386ecd068671bcc (patch) | |
| tree | b2710f261d12a7a1b73962691c187752663543f6 /src/Data/Array | |
| parent | 308ca9fac150cd28d62afef852f26ae4c40fa5a0 (diff) | |
Optimise reductions and dotprod with more vectorisation
Turns out that if you don't supply -ffast-math, the C compiler will
faithfully reproduce your linear reduction order, which is rather
disastrous for parallelisation with vector units.
This changes the summation order, so numerical results might differ
slightly. To wit: the test suite needed adjustment.
Diffstat (limited to 'src/Data/Array')
| -rw-r--r-- | src/Data/Array/Mixed/Internal/Arith/Foreign.hs | 1 |
1 files changed, 0 insertions, 1 deletions
diff --git a/src/Data/Array/Mixed/Internal/Arith/Foreign.hs b/src/Data/Array/Mixed/Internal/Arith/Foreign.hs index 15fbc79..969a25a 100644 --- a/src/Data/Array/Mixed/Internal/Arith/Foreign.hs +++ b/src/Data/Array/Mixed/Internal/Arith/Foreign.hs @@ -20,7 +20,6 @@ $(do ,("reducefull_" ++ tyn, [t| CInt -> Int64 -> Ptr Int64 -> Ptr Int64 -> Ptr $ttyp -> IO $ttyp |]) ,("extremum_min_" ++ tyn, [t| Ptr Int64 -> Int64 -> Ptr Int64 -> Ptr Int64 -> Ptr $ttyp -> IO () |]) ,("extremum_max_" ++ tyn, [t| Ptr Int64 -> Int64 -> Ptr Int64 -> Ptr Int64 -> Ptr $ttyp -> IO () |]) - ,("dotprod_" ++ tyn, [t| Int64 -> Ptr $ttyp -> Ptr $ttyp -> IO $ttyp |]) ,("dotprod_" ++ tyn ++ "_strided", [t| Int64 -> Int64 -> Int64 -> Ptr $ttyp -> Int64 -> Int64 -> Ptr $ttyp -> IO $ttyp |]) ,("dotprodinner_" ++ tyn, [t| Int64 -> Ptr Int64 -> Ptr $ttyp -> Ptr Int64 -> Ptr $ttyp -> Ptr Int64 -> Ptr $ttyp -> IO () |]) ] |