#7307: Share top-level code for strings -------------------------------------+------------------------------------- Reporter: simonpj | Owner: parcs Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 7.6.1 Resolution: | Keywords: strings Operating System: Unknown/Multiple | Architecture: Type of failure: Runtime | Unknown/Multiple performance bug | Test Case: Blocked By: | Blocking: Related Tickets: | Differential Rev(s): Wiki Page: | -------------------------------------+------------------------------------- Changes (by bgamari): * keywords: => strings Old description:

A string constant in GHC turns into {{{ foo :: String foo = unpackCString# "the-string' }}} This is a top-level thunk, and it expands into rather a lot of code like this {{{ .text .align 4,0x90 .long 0 .long 22 .globl _Foo_zdfTypeableTzuds1_info _Foo_zdfTypeableTzuds1_info: .LcvI: movl %esi,%eax leal -12(%ebp),%ecx cmpl 84(%ebx),%ecx jb .LcvQ addl $8,%edi cmpl 92(%ebx),%edi ja .LcvS movl $_stg_CAF_BLACKHOLE_info,-4(%edi) movl 100(%ebx),%ecx movl %ecx,0(%edi) leal -4(%edi),%ecx pushl %ecx pushl %eax pushl %ebx movl %eax,76(%esp) call _newCAF addl $12,%esp testl %eax,%eax je .LcvL movl $_stg_bh_upd_frame_info,-8(%ebp) leal -4(%edi),%eax movl %eax,-4(%ebp) movl $_cvJ_str,-12(%ebp) addl $-12,%ebp jmp _ghczmprim_GHCziCString_unpackCStringzh_info .LcvL: movl 64(%esp),%eax jmp *(%eax) .LcvS: movl $8,116(%ebx) .LcvQ: movl %eax,%esi jmp *-12(%ebx) }}} That's rather a lot of goop for one thunk! Of course we can share this, by making a 2-word thunk like this: {{{ ------------------------------ | TopUnpack_info | -------|-----> "the-string"# ------------------------------ }}} where `TopUnpack_info` is a shared RTS info-table and code that embodies the code fragment above.

This would save useless code bloat for every constant string. (This came up when looking at the code generated by `deriving(Typeable)`.)

New description: A string constant in GHC turns into {{{#!hs foo :: String foo = unpackCString# "the-string' }}} This is a top-level thunk, and it expands into rather a lot of code like this {{{ .text .align 4,0x90 .long 0 .long 22 .globl _Foo_zdfTypeableTzuds1_info _Foo_zdfTypeableTzuds1_info: .LcvI: movl %esi,%eax leal -12(%ebp),%ecx cmpl 84(%ebx),%ecx jb .LcvQ addl $8,%edi cmpl 92(%ebx),%edi ja .LcvS movl $_stg_CAF_BLACKHOLE_info,-4(%edi) movl 100(%ebx),%ecx movl %ecx,0(%edi) leal -4(%edi),%ecx pushl %ecx pushl %eax pushl %ebx movl %eax,76(%esp) call _newCAF addl $12,%esp testl %eax,%eax je .LcvL movl $_stg_bh_upd_frame_info,-8(%ebp) leal -4(%edi),%eax movl %eax,-4(%ebp) movl $_cvJ_str,-12(%ebp) addl $-12,%ebp jmp _ghczmprim_GHCziCString_unpackCStringzh_info .LcvL: movl 64(%esp),%eax jmp *(%eax) .LcvS: movl $8,116(%ebx) .LcvQ: movl %eax,%esi jmp *-12(%ebx) }}} That's rather a lot of goop for one thunk! Of course we can share this, by making a 2-word thunk like this: {{{ ------------------------------ | TopUnpack_info | -------|-----> "the-string"# ------------------------------ }}} where `TopUnpack_info` is a shared RTS info-table and code that embodies the code fragment above. This would save useless code bloat for every constant string. (This came up when looking at the code generated by `deriving(Typeable)`.) -- -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/7307#comment:14 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler