[GHC] #9400: poor performance when compiling modules with many Text literals at -O1

[GHC]

#9400: poor performance when compiling modules with many Text literals at -O1
-------------------------------------+-------------------------------------
              Reporter:  rwbarton    |            Owner:
                  Type:  bug         |           Status:  new
              Priority:  normal      |        Milestone:
             Component:  Compiler    |          Version:  7.8.3
            Resolution:              |         Keywords:
      Operating System:              |     Architecture:  Unknown/Multiple
  Unknown/Multiple                   |       Difficulty:  Unknown
       Type of failure:  Compile-    |       Blocked By:
  time performance bug               |  Related Tickets:  #9370
             Test Case:              |
              Blocking:              |
Differential Revisions:              |
-------------------------------------+-------------------------------------
Description changed by rwbarton:

Old description:

> (Spawned from #9370; see there for original discussion)
>
> Unpack the xmlhtml package and edit `src/Text/XmlHtml/HTML/Meta.hs` and
> remove the `OPTIONS_GHC` line at the top of the file that disables
> optimizations and build with `cabal buidl`. Then GHC takes ~1.5GB and
> over a minute to build this single module.
>
> Preliminary investigation indicates that Text's fromString and its
> constituent parts is being inlined repeatedly:
> {{{
> Inlining done: Data.String.fromString
> Inlining done: Data.Text.$fIsStringText
> Inlining done: Data.Text.pack
> Inlining done: Data.Text.Internal.Fusion.unstream
> Inlining done: Data.Text.Internal.Fusion.Common.map
> Inlining done: Data.Text.Internal.Fusion.Common.streamList
> Inlining done: Data.Text.Internal.safe
> Inlining done: Data.Bits.$fBitsInt_$c.&.
> Inlining done: Data.Text.Internal.Fusion.Types.$WYield
> [ repeats ~4000 times ]
> }}}
> resulting in a very large intermediate program:
> {{{
> *** Checking old interface for xmlhtml-0.2.3.2:Text.XmlHtml.HTML.Meta:
> [ 1 of 10] Compiling Text.XmlHtml.HTML.Meta (
> src/Text/XmlHtml/HTML/Meta.hs, dist/build/Text/XmlHtml/HTML/Meta.o )
> *** Parser:
> *** Renamer/typechecker:
> *** Desugar:
> Result size of Desugar (after optimization)
>   = {terms: 26,260, types: 20,021, coercions: 0}
> *** Simplifier:
> Result size of Simplifier iteration=1
>   = {terms: 1,446,658, types: 953,432, coercions: 314,352}
> ...
> }}}

New description:

 (Spawned from #9370; see there for original discussion)

 Unpack the xmlhtml package and edit `src/Text/XmlHtml/HTML/Meta.hs` and
 remove the `OPTIONS_GHC` line at the top of the file that disables
 optimizations and build with `cabal buidl`. Then GHC takes ~1.5GB and over
 a minute to build this single module.

 Preliminary investigation indicates that Text's fromString and its
 constituent parts is being inlined repeatedly:
 {{{
 Inlining done: Data.String.fromString
 Inlining done: Data.Text.$fIsStringText
 Inlining done: Data.Text.pack
 Inlining done: Data.Text.Internal.Fusion.unstream
 Inlining done: Data.Text.Internal.Fusion.Common.map
 Inlining done: Data.Text.Internal.Fusion.Common.streamList
 Inlining done: Data.Text.Internal.safe
 Inlining done: Data.Bits.$fBitsInt_$c.&.
 Inlining done: Data.Text.Internal.Fusion.Types.$WYield
 [ repeats ~4000 times ]
 }}}
 resulting in a very large intermediate program:
 {{{
 [ 1 of 10] Compiling Text.XmlHtml.HTML.Meta (
 src/Text/XmlHtml/HTML/Meta.hs, dist/build/Text/XmlHtml/HTML/Meta.o )
 *** Parser:
 *** Renamer/typechecker:
 *** Desugar:
 Result size of Desugar (after optimization)
   = {terms: 26,806, types: 14,467, coercions: 0}
 *** Simplifier:
 Result size of Simplifier iteration=1
   = {terms: 31,052, types: 20,719, coercions: 0}
 Result size of Simplifier
   = {terms: 31,052, types: 20,719, coercions: 0}
 *** Specialise:
 Result size of Specialise
   = {terms: 32,254, types: 22,696, coercions: 448}
 *** Float out(FOS {Lam = Just 0, Consts = True, PAPs = False}):
 Result size of Float out(FOS {Lam = Just 0,
                               Consts = True,
                               PAPs = False})
   = {terms: 63,022, types: 59,045, coercions: 448}
 *** Float inwards:
 Result size of Float inwards
   = {terms: 63,022, types: 59,045, coercions: 448}
 *** Simplifier:
 Result size of Simplifier iteration=1
   = {terms: 28,537, types: 18,902, coercions: 654}
 Result size of Simplifier iteration=2
   = {terms: 28,157, types: 18,257, coercions: 152}
 Result size of Simplifier iteration=3
   = {terms: 28,074, types: 18,128, coercions: 140}
 Result size of Simplifier iteration=4
   = {terms: 28,068, types: 18,096, coercions: 61}
 Result size of Simplifier
   = {terms: 28,068, types: 18,096, coercions: 61}
 *** Simplifier:
 Result size of Simplifier iteration=1
   = {terms: 43,941, types: 38,325, coercions: 61}
 Result size of Simplifier
   = {terms: 43,941, types: 38,325, coercions: 61}
 *** Simplifier:
 Result size of Simplifier iteration=1
   = {terms: 689,670, types: 461,960, coercions: 146,725}
 ...
 }}}
 [Edited: old output was not actually for `-O1` as described, but rather
 for the sequence described in #9370 of building with `-O0` after building
 another module with `-O1`.]

--

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/9400#comment:2>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler