Changes between Version 7 and Version 8 of Frisby2013Q1

Feb 11, 2013 4:28:12 PM (3 years ago)



  • Frisby2013Q1

    v7 v8  
     117Using -flate-float-in-thunk-limit=10, -fprotect-last-arg, and -O1, I tested the libraries+NoFib for four variants.
     119  * nn - do not float a binding that applies one of its free variables.
     120  * yn - do not float a binding that applies one of its free variables saturated or oversaturated.
     121  * ny - do not float a binding that applies one of its free variables undersaturated.
     122  * yy - do not restrict application of bindings free variables
     124Roughly, we expect that more floating means (barely) less allocation but worse runtime (by how much?) because some known calls become unknown calls.
     126==== Anchoring let-no-escapes ====
     128I discovered this while experimenting with the fast preservation variants above.
     130TODO: Mitigate it.
     132In fish (1.6%), hpg (~4.5%), and sphere (10.4%), allocation gets worse for ny and yy compared to nn and yn. The nn and ny do not change the allocation compared to the baseline library (ie no LLF).
     134The nn -> ny comparison is counter to our rough idea: floating more bindings (those that saturate/oversaturate some free variables) worsens allocation. Thus, I investigate.
     136The sphere program hammers `hPutStr`. Its extra allocation is mostly due to a regression in `GHC.IO.Encoding.UTF8`. Here's the situation.
     138With the nn variant:
     141outer a b c ... =
     142  let-no-escape f x = CTX[let-no-escape $j y = ... (f ...) ... in CTX2[$j]]
     143  in ...
     146In this case, `$j` is not floated because it applies `f`. With the ny variant, `$j` gets floated.
     149poly_$j a b c ... f y = ...
     151outer a b c ... =
     152  let f x = CTX[CTX2[poly_$j a b c ... f]]
     153  in ...
     156Thus `f` cannot be let-no-escape because it now occurs as an argument to `poly_$j`.
     158This contributes to sphere's 1 megabyte of extra allocation for two reasons:
     160  * `outer` is entered about 60,000 times.
     161  * The RHS of `f` has 13 free variables, so it's closure is rather large.
     16313*60,000 ~ 750,000. I suspect the rest of sphere's increase is due to a similar issue in `GHC.IO.Handle`.
     165In hpg, it's principally due to GHC`.IO.Encoding.UTF8` again, with a second place contributor of `GHC.IO.FD`, where the function `$wa17` is again like the `outer` example above, but with fewer free variables and thus less effect.
    117167=== Discovered Benefits of LLF ===