#14754: -O1 changes result at runtime -------------------------------------+------------------------------------- Reporter: Bodigrim | Owner: (none) Type: bug | Status: new Priority: high | Milestone: Component: Compiler | Version: 8.4.1-alpha1 Resolution: | Keywords: Operating System: Unknown/Multiple | Architecture: Type of failure: Incorrect result | Unknown/Multiple at runtime | Test Case: Blocked By: | Blocking: Related Tickets: | Differential Rev(s): Wiki Page: | -------------------------------------+------------------------------------- Comment (by mpickering): Which operating system are you using? -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/14754#comment:1> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

#14754: -O1 changes result at runtime -------------------------------------+------------------------------------- Reporter: Bodigrim | Owner: (none) Type: bug | Status: new Priority: highest | Milestone: 8.4.1 Component: Compiler | Version: 8.4.1-alpha1 Resolution: | Keywords: Operating System: Unknown/Multiple | Architecture: Type of failure: Incorrect result | Unknown/Multiple at runtime | Test Case: Blocked By: | Blocking: Related Tickets: | Differential Rev(s): Wiki Page: | -------------------------------------+------------------------------------- Changes (by bgamari): * priority: high => highest * milestone: => 8.4.1 Old description:

I found a program, which works as expected in GHC 8.4.1-alpha1 with `-O0`, but freezes with `-O1`.

{{{ module Main where

import Debug.Trace

main :: IO () main = print (alg 3 1)

alg :: Word -> Word -> Word alg a b | traceShow (a, b) False = undefined | c < b = alg b c | c > b = alg c b | otherwise = c where c = a - b }}}

{{{ $ ghc --version The Glorious Glasgow Haskell Compilation System, version 8.4.0.20180118 $ ghc -O0 alg.hs [1 of 1] Compiling Main ( alg.hs, alg.o ) Linking alg ... $ ./alg (3,1) (2,1) 1 $ ghc -O1 alg.hs [1 of 1] Compiling Main ( alg.hs, alg.o ) [Optimisation flags changed] Linking alg ... $ ./alg 2>&1 | head (3,1) (1,2) (2,18446744073709551615) (18446744073709551615,3) ^C }}}

For some reason an optimised program chooses a wrong case at the very first invocation of `alg`.

New description: I found a program, which works as expected in GHC 8.4.1-alpha1 with `-O0`, but freezes with `-O1`. {{{#!hs module Main where import Debug.Trace main :: IO () main = print (alg 3 1) alg :: Word -> Word -> Word alg a b | traceShow (a, b) False = undefined | c < b = alg b c | c > b = alg c b | otherwise = c where c = a - b }}} {{{ $ ghc --version The Glorious Glasgow Haskell Compilation System, version 8.4.0.20180118 $ ghc -O0 alg.hs [1 of 1] Compiling Main ( alg.hs, alg.o ) Linking alg ... $ ./alg (3,1) (2,1) 1 $ ghc -O1 alg.hs [1 of 1] Compiling Main ( alg.hs, alg.o ) [Optimisation flags changed] Linking alg ... $ ./alg 2>&1 | head (3,1) (1,2) (2,18446744073709551615) (18446744073709551615,3) ^C }}} For some reason an optimised program chooses a wrong case at the very first invocation of `alg`. -- Comment: Yikes! Thank you for catching this. I guess this means my plans for another alpha release today are on hold. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/14754#comment:3> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

#14754: -O1 changes result at runtime -------------------------------------+------------------------------------- Reporter: Bodigrim | Owner: (none) Type: bug | Status: new Priority: highest | Milestone: 8.4.1 Component: Compiler | Version: 8.4.1-alpha1 Resolution: | Keywords: Operating System: Unknown/Multiple | Architecture: Type of failure: Incorrect result | Unknown/Multiple at runtime | Test Case: Blocked By: | Blocking: Related Tickets: | Differential Rev(s): Wiki Page: | -------------------------------------+------------------------------------- Comment (by bgamari): Indeed the problem here is that the CBE logic in the above patch find these two blocks to be equivalent: {{{ c4N2: // global _s4Kg::I64 = _s4Kg::I64; _s4Kf::I64 = _s4Kr::I64; goto c4M3; c4N5: // global _s4Kf::I64 = _s4Kg::I64; _s4Kg::I64 = _s4Kr::I64; goto c4M3; }}} The reason is that the implementation walks the two blocks, zipping together their nodes and building a correspondence between their local registers. In the first node we have {{{ c4N2 c4N5 ---------------- --------------- s4Kg = s4Kg s4Kf = s4Kg }}} As the RHSs of the assignments are identical, this results in the correspondence `s4Kg(c4N2) ~> s4Kf(s4N5)`. We then consider the second pair of nodes: {{{ c4N2 c4N5 ---------------- --------------- s4Kf = s4Kr s4Kg = s4Kr }}} Since the implementation binding occurrences are simply added to the register substitution instead of compared, these two nodes are considered to be equivalent. This is clearly wrong. I'll need to revisit this in the future but in the meantime I'm going to revert. implementation fails to take into account that logic registers may live beyond the -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/14754#comment:5> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

#14754: -O1 changes result at runtime -------------------------------------+------------------------------------- Reporter: Bodigrim | Owner: (none) Type: bug | Status: new Priority: highest | Milestone: 8.4.1 Component: Compiler | Version: 8.4.1-alpha1 Resolution: | Keywords: Operating System: Unknown/Multiple | Architecture: Type of failure: Incorrect result | Unknown/Multiple at runtime | Test Case: Blocked By: | Blocking: Related Tickets: | Differential Rev(s): Wiki Page: | -------------------------------------+------------------------------------- Comment (by Ben Gamari <ben@…>): In [changeset:"50adbd7c5fe5894d3e6e2a58b353ed07e5f8949d/ghc" 50adbd7c/ghc]: {{{ #!CommitTicketReference repository="ghc" revision="50adbd7c5fe5894d3e6e2a58b353ed07e5f8949d" cmm: Revert more aggressive CBE due to #14226 Trac #14226 noted that the C-- CBE pass frequently fails to common up semantically identical blocks due to the differences in local register naming. These patches fixed this by making the pass consider equality up to alpha-renaming. However, the new logic failed to consider the possibility that local register naming *may* matter across multiple blocks. This lead to the regression #14754. I'll need to do a bit of thinking on a proper solution to this but in the meantime I'm reverting all four patches. This reverts commit a27056f9823f8bbe2302f1924b3ab38fd6752e37. This reverts commit 6f990c54f922beae80362fe62426beededc21290. This reverts commit 9aa73892e10e90a1799b9277da593e816a827364. This reverts commit 7920a7d9c53083b234e060a3e72f00b601a46808. }}} -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/14754#comment:6> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

#14754: -O1 changes result at runtime -------------------------------------+------------------------------------- Reporter: Bodigrim | Owner: (none) Type: bug | Status: closed Priority: highest | Milestone: 8.4.1 Component: Compiler | Version: 8.4.1-alpha1 Resolution: fixed | Keywords: Operating System: Unknown/Multiple | Architecture: Type of failure: Incorrect result | Unknown/Multiple at runtime | Test Case: Blocked By: | Blocking: Related Tickets: #14226 | Differential Rev(s): Wiki Page: | -------------------------------------+------------------------------------- Comment (by bgamari): Merged to `ghc-8.4`. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/14754#comment:9> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler

#14754: -O1 changes result at runtime -------------------------------------+------------------------------------- Reporter: Bodigrim | Owner: (none) Type: bug | Status: closed Priority: highest | Milestone: 8.4.1 Component: Compiler | Version: 8.4.1-alpha1 Resolution: fixed | Keywords: Operating System: Unknown/Multiple | Architecture: Type of failure: Incorrect result | Unknown/Multiple at runtime | Test Case: Blocked By: | Blocking: Related Tickets: #14226 | Differential Rev(s): Wiki Page: | -------------------------------------+------------------------------------- Comment (by bgamari): In the `optasm` testsuite way, yes. -- Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/14754#comment:11> GHC <http://www.haskell.org/ghc/> The Glasgow Haskell Compiler