November 2015 - ghc-tickets

[GHC] #10069: CPR related performance issue
by GHC 09 Mar '19

09 Mar '19

#10069: CPR related performance issue -------------------------------------+------------------------------------- Reporter: pacak | Owner: Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 7.10.1-rc2 Keywords: | Operating System: Unknown/Multiple Architecture: | Type of failure: Runtime Unknown/Multiple | performance bug Test Case: | Blocked By: Blocking: | Related Tickets: Differential Revisions: | -------------------------------------+------------------------------------- By default CRP analysis can be too aggressive in trying to pass as much as possible in unboxed tuples, in general it's not a problem but when one big datatype is passed to several consumers it might end up pushed to stack several times instead of once - to heap, things are getting worse when there are sufficient fields to cause stack overflow which otherwise is possible to avoid - in our codebase adding one field with ExistentialQuantification (unused, but that prevents ghc from doing CRP transformation) reduces number of stack overflow by a factor of 1000 and increases overall performance by 10%. In provided example performance for both A and B should be identical and yet B is consistently faster by 3-5% It's possible to increase this performance gap by adding more and more fields. I was able to replicate this issue in ghc 7.8.3 and 7.10,1rc2 {{{#!hs {-# LANGUAGE ExistentialQuantification #-} module Blah where import Criterion import Criterion.Main import Data.Typeable data A = A () !Int !Int !Int !Int !Int !Int !Int !Int !Int !Int !Int !Int !Int !Int !Int !Int !Int !Int !Int !Int !Int !Int !Int !Int !Int !Int !Int !Int !Int !Int !Int !Int data B = forall rep. (Typeable rep) => B rep !Int !Int !Int !Int !Int !Int !Int !Int !Int !Int !Int !Int !Int !Int !Int !Int !Int !Int !Int !Int !Int !Int !Int !Int !Int !Int !Int !Int !Int !Int !Int !Int a :: A a = A () 1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8 b :: B b = B () 1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8 1 2 3 4 5 6 7 8 {-# NOINLINE a1 #-} a1 :: A -> Int a1 (A _ f1 f2 f3 f4 f5 f6 f7 f8 g1 g2 g3 g4 g5 g6 g7 g8 _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _) = f1 {-# NOINLINE a2 #-} a2 :: A -> Int a2 (A _ f1 f2 f3 f4 f5 f6 f7 f8 g1 g2 g3 g4 g5 g6 g7 g8 _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _) = f2 {-# NOINLINE a3 #-} a3 :: A -> Int a3 (A _ f1 f2 f3 f4 f5 f6 f7 f8 g1 g2 g3 g4 g5 g6 g7 g8 _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _) = f3 {-# NOINLINE a4 #-} a4 :: A -> Int a4 (A _ f1 f2 f3 f4 f5 f6 f7 f8 g1 g2 g3 g4 g5 g6 g7 g8 _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _) = f4 {-# NOINLINE b1 #-} b1 :: B -> Int b1 (B _ f1 f2 f3 f4 f5 f6 f7 f8 g1 g2 g3 g4 g5 g6 g7 g8 _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _) = f1 {-# NOINLINE b2 #-} b2 :: B -> Int b2 (B _ f1 f2 f3 f4 f5 f6 f7 f8 g1 g2 g3 g4 g5 g6 g7 g8 _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _) = f2 {-# NOINLINE b3 #-} b3 :: B -> Int b3 (B _ f1 f2 f3 f4 f5 f6 f7 f8 g1 g2 g3 g4 g5 g6 g7 g8 _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _) = f3 {-# NOINLINE b4 #-} b4 :: B -> Int b4 (B _ f1 f2 f3 f4 f5 f6 f7 f8 g1 g2 g3 g4 g5 g6 g7 g8 _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _) = f4 {-# NOINLINE fa #-} fa :: A -> Int fa a = a1 a + a2 a + a3 a + a4 a {-# NOINLINE fb #-} fb :: B -> Int fb b = b1 b + b2 b + b3 b + b4 b main :: IO () main = defaultMain [ bgroup "single call" [ bench "A" $ whnf fa a , bench "B" $ whnf fb b ] ] }}} -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/10069> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 39

[GHC] #9832: Get rid of PERL dependency of `ghc-split`
by GHC 08 Mar '19

08 Mar '19

#9832: Get rid of PERL dependency of `ghc-split` -------------------------------------+------------------------------------- Reporter: hvr | Owner: Type: task | Status: new Priority: normal | Milestone: 7.12.1 Component: Compiler | Version: Keywords: | Operating System: Windows Architecture: Unknown/Multiple | Type of failure: Difficulty: Unknown | None/Unknown Blocked By: | Test Case: Related Tickets: | Blocking: | Differential Revisions: -------------------------------------+------------------------------------- (writeme) -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/9832> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 37

[GHC] #8964: split_marker_entry assert breaks -split-objs and -ddump-opt-cmm
by GHC 08 Mar '19

08 Mar '19

#8964: split_marker_entry assert breaks -split-objs and -ddump-opt-cmm -------------------------------------+------------------------------------- Reporter: ezyang | Owner: Type: bug | Status: new Priority: lowest | Milestone: ⊥ Component: Compiler | Version: 7.9 Keywords: | Operating System: Unknown/Multiple Architecture: Unknown/Multiple | Type of failure: Compile-time Difficulty: Easy (less than 1 | crash hour) | Test Case: Blocked By: | Blocking: Related Tickets: | -------------------------------------+------------------------------------- ...since the pretty-printer attempts to force the blocks list of the split_marker_entry proc. I'm not really sure what the right way to fix this is: just provide an empty block list, or hack the pretty-printer to treat split markers specially. But should be easy to fix. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/8964> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 5

[GHC] #8685: -split-objs doesn't work for executables
by GHC 08 Mar '19

08 Mar '19

#8685: -split-objs doesn't work for executables ------------------------------------+------------------------------------- Reporter: refold | Owner: Type: bug | Status: new Priority: normal | Milestone: Component: Driver | Version: 7.6.3 Keywords: | Operating System: Unknown/Multiple Architecture: Unknown/Multiple | Type of failure: None/Unknown Difficulty: Unknown | Test Case: Blocked By: | Blocking: Related Tickets: | ------------------------------------+------------------------------------- While investigating [this bug report](https://github.com/haskell/cabal/issues/1611), I've noticed that the `-split-objs` option only works for libraries, but not for executables. That is, unlike in the case of libraries, the driver feeds only `A.o B.o ...` files to the linker instead of `A_split_0.o A_split_1.o ... B_split_0.o B_split_1.o ...`. It'd be nice if `-split-objs` could be changed to also work on executables. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/8685> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 5

[GHC] #8629: Option 'split-objs' being ignored when trying to reduce object code size in iOS cross-compilation
by GHC 08 Mar '19

08 Mar '19

#8629: Option 'split-objs' being ignored when trying to reduce object code size in iOS cross-compilation ------------------------------------+--------------------------------- Reporter: f1rstmistake | Owner: Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 7.6.3 Keywords: ios | Operating System: MacOS X Architecture: Unknown/Multiple | Type of failure: None/Unknown Difficulty: Unknown | Test Case: Blocked By: | Blocking: Related Tickets: | ------------------------------------+--------------------------------- I have been trying to statically link Pandoc packages to iOS application following the guide in https://ghc.haskell.org/trac/ghc/wiki/Building/CrossCompiling/iOS and it's great so far! I have finally successfully use Pandoc library in iOS app, but the binary size is very big! 1. More than 50MB when using Pandoc library. 2. Around 5MB for simple haskell code. I try to add “split-objs: True” in .cabal/config file, but when running cabal I got this message: {{{ Preprocessing library pandoc-1.12.2.1... unused terminals: 1 on the commandline: Warning: ignoring -fsplit-objs }}} And if I ignore that, at the end of build, I get another error message that stop the build. {{{ cabal: dist-i386/build/Text/Pandoc_o_split: does not exist }}} I try with another package and got similar error message. It seems that split-objs not only being ignored by cross-compilation ghc, but will also fail the build. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/8629> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 4

[GHC] #8095: TypeFamilies painfully slow
by GHC 08 Mar '19

08 Mar '19

#8095: TypeFamilies painfully slow -------------------------------------+------------------------------------- Reporter: MikeIzbicki | Owner: Type: bug | Status: new Priority: normal | Milestone: Component: Compiler (Type | Version: 7.6.3 checker) | Operating System: Unknown/Multiple Keywords: | Type of failure: Compile-time Architecture: Unknown/Multiple | performance bug Difficulty: Unknown | Test Case: Blocked By: | Blocking: Related Tickets: 5321 | -------------------------------------+------------------------------------- I'm using the TypeFamilies extension to generate types that are quite large. GHC can handle these large types fine when they are created manually, but when type families get involved, GHC's performance dies. Unlike in ticket #5321, using tail recursion does not eliminate the problem, and the order of arguments greatly affects compile time. I've attached a file Types.hs that demonstrates the problems. This file generates another Haskell file which has the problems. It takes 3 flags. The first is the size of the type to generate, the second is which type family function to use, and the third is whether to call the type family or just use a manually generated type. Here are my performance results: Using non-tail recursion, I get these results. I have to increase the stack size based on the size of the type I want to generate. {{{ $ ./Types 200 a a > test.hs && time ghc test.hs > /dev/null -fcontext- stack=250 real 0m2.973s $ ./Types 300 a a > test.hs && time ghc test.hs > /dev/null -fcontext- stack=350 real 0m6.018s $ ./Types 400 a a > test.hs && time ghc test.hs > /dev/null -fcontext- stack=450 real 0m9.995s $ ./Types 500 a a > test.hs && time ghc test.hs > /dev/null -fcontext- stack=550 real 0m15.645s }}} Tail recursion generates much slower compile times for some reason, and I still need to adjust the stack size: {{{ $ ./Types 200 b a > test.hs && time ghc test.hs > /dev/null -fcontext- stack=250 real 0m16.120s }}} Changing the order of arguments to the recursive type family greatly changes the run times: {{{ $ ./Types 200 c a > test.hs && time ghc test.hs > /dev/null -fcontext- stack=250 real 0m6.095s }}} Without the type family, I get MUCH better performance: {{{ $ ./Types 10000 a d > test.hs && time ghc test.hs > /dev/null real 0m2.271s }}} -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/8095> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 64

[GHC] #10892: ApplicativeDo should use *> and <*
by GHC 05 Mar '19

05 Mar '19

#10892: ApplicativeDo should use *> and <* -------------------------------------+------------------------------------- Reporter: simonmar | Owner: Type: task | Status: new Priority: normal | Milestone: 8.0.1 Component: Compiler | Version: 7.11 Keywords: | Operating System: Unknown/Multiple Architecture: | Type of failure: None/Unknown Unknown/Multiple | Test Case: | Blocked By: Blocking: | Related Tickets: Differential Revisions: | -------------------------------------+------------------------------------- From @ekmett: Would it be possible to tweak the generation to use `(<*)` or `(*>)` where appropriate when the result isn't being used? For many Applicatives this can be a massive asymptotic win in terms of sharing and/or computational cost. When desugaring using (<*) you'd just omit any handling of the unused result instead. {{{ (\x y -> ...) <$> foo <* bar <*> baz }}} corresponds to {{{ do x <- foo bar y <- baz return ... }}} -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/10892> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 30

[GHC] #8657: -fregs-graph still has a limit on spill slots
by GHC 04 Mar '19

04 Mar '19

#8657: -fregs-graph still has a limit on spill slots -----------------------------------+--------------------------------------- Reporter: simonmar | Owner: Type: bug | Status: new Priority: normal | Milestone: 7.8.1 Component: Compiler (NCG) | Version: 7.7 Keywords: | Operating System: Unknown/Multiple Architecture: | Type of failure: Compile-time crash Unknown/Multiple | Test Case: Difficulty: Unknown | Blocking: Blocked By: | Related Tickets: | -----------------------------------+--------------------------------------- The limit on spill slots was removed for the linear register allocator in 7.8, but not for `-fregs-graph`. SHA-1 fails to compile with `-fregs-graph` in 7.8. Related: #7679 (`-fregs-graph` generates poor code with the new codegen). -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/8657> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 18

[GHC] #9894: Presence of DWARF debug symbols doesn't result in working backtraces in GDB
by GHC 04 Mar '19

04 Mar '19

#9894: Presence of DWARF debug symbols doesn't result in working backtraces in GDB ----------------------------------+--------------------------------------- Reporter: tibbe | Owner: scpmw Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 7.9 Keywords: | Operating System: Linux Architecture: x86_64 (amd64) | Type of failure: None/Unknown Difficulty: Unknown | Test Case: Blocked By: | Blocking: Related Tickets: | Differential Revisions: ----------------------------------+--------------------------------------- Now when GHC can output DWARF debug symbols we need to take the remaining steps needed to get downstream tool like GDB to do something useful with it. As of today GDB cannot produce a valid stack trace for this code: {{{#!hs module Main where import System.IO f name = print $ "Hello, " ++ name {-# NOINLINE f #-} g name = f name {-# NOINLINE g #-} main = do name <- getLine g name }}} when a breakpoint is set on the line that defines `f` before the program is run. The resulting backtrace looks like: {{{ (gdb) bt #0 Main_f_info () at /tmp/Test.hs:5 #1 0x0000000000481a00 in ?? () Backtrace stopped: frame did not save the PC }}} This bug tracks the remaining work that needs to be done before this will work. Peter, what is the remaining work here? -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/9894> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 5

[GHC] #9910: ghc's dwarf (-g) output does not like Unicode (Char.intToDigit: not a digit 136)
by GHC 04 Mar '19

04 Mar '19

#9910: ghc's dwarf (-g) output does not like Unicode (Char.intToDigit: not a digit 136) -------------------------------------+------------------------------------- Reporter: slyfox | Owner: Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 7.9 Keywords: | Operating System: Architecture: Unknown/Multiple | Unknown/Multiple Difficulty: Unknown | Type of failure: Blocked By: | None/Unknown Related Tickets: | Test Case: | Blocking: | Differential Revisions: -------------------------------------+------------------------------------- Easy to reproduce as: {{{ inplace/bin/ghc-stage2 -w -g -c testsuite/tests/parser/unicode/utf8_024.hs ghc-stage2: panic! (the 'impossible' happened) (GHC version 7.9.20141219 for x86_64-unknown-linux): Char.intToDigit: not a digit 136 Please report this as a GHC bug: http://www.haskell.org/ghc/reportabug }}} {{{ petermw> Yes, Unicode doesn't work with DWARF yet }}} -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/9910> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

1 2