#8860: Optimized Cmm isn't -------------------------------------+------------------------------------ Reporter: tibbe | Owner: Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 7.6.3 Resolution: | Keywords: Operating System: Unknown/Multiple | Architecture: Unknown/Multiple Type of failure: None/Unknown | Difficulty: Unknown Test Case: | Blocked By: Blocking: | Related Tickets: -------------------------------------+------------------------------------ Changes (by slyfox): * cc: slyfox@… (added) -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/8860#comment:1 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler

#8860: Optimized Cmm isn't -------------------------------------+------------------------------------ Reporter: tibbe | Owner: Type: bug | Status: new Priority: normal | Milestone: Component: Compiler | Version: 7.6.3 Resolution: | Keywords: Operating System: Unknown/Multiple | Architecture: Unknown/Multiple Type of failure: None/Unknown | Difficulty: Unknown Test Case: | Blocked By: Blocking: | Related Tickets: -------------------------------------+------------------------------------ Description changed by tibbe: Old description:

The optimizer seems to miss very basic optimizations. For example, look at this segment (full source below):

{{{ _c1Cd::I64 = 0; _c1Ce::I64 = 0; _c1Cf::I64 = 8; call MO_Memset(_c1Cc::I64, _c1Cd::I64, _c1Ce::I64, _c1Cf::I64); }}}

I'd expect all these temporaries (that are only used once) to be inlined, but they are not and thus we fail to see the static arguments to the `MO_Memset` call, which leads to further missed optimizations.

This used to work in the old codegen.

Furthermore, there are useless basic blocks in the output:

{{{ c1Cm: goto c1C4; c1C4: }}}

I'd expect them to be eliminated.

.dump-cmm file:

{{{ ==================== Cmm produced by new codegen ==================== 2014-03-08 08:25:31.728672 UTC

[section "data" { a_r1za_closure: const a_r1za_info; }, a_r1za_entry() // [] { info_tbl: [(c1Cm, label: a_r1za_info rep:HeapRep static { Fun {arity: 1 fun_type: ArgSpec 3} })] stack_info: arg_space: 8 updfr_space: Just 8 } {offset c1Cm: goto c1C4; c1C4: if ((old + 0) - <highSp> < SpLim) goto c1Cp; else goto c1Cq; c1Cq: goto c1C3; c1C3: Hp = Hp + 152; if (Hp > HpLim) goto c1Cs; else goto c1Cr; c1Cs: HpAlloc = 152; goto c1Cp; c1Cp: R1 = a_r1za_closure; call (stg_gc_fun)(R1) args: 8, res: 0, upd: 8; c1Cr: I64[Hp - 144] = stg_MUT_ARR_PTRS_DIRTY_info; I64[Hp - 136] = 16; I64[Hp - 128] = 16; _c1C6::I64 = Hp - 144; _c1C7::I64 = _c1C6::I64 + 24; goto c1C8; c1C8: if (_c1C7::I64 < (_c1C6::I64 + 128)) goto c1Ca; else goto c1C9; c1Ca: P64[_c1C7::I64] = ()_closure+1; _c1C7::I64 = _c1C7::I64 + 8; goto c1C8; c1C9: _c1Cb::I64 = _c1C6::I64 + 24; _c1Cc::I64 = _c1Cb::I64 + 128; _c1Cd::I64 = 0; _c1Ce::I64 = 0; _c1Cf::I64 = 8; call MO_Memset(_c1Cc::I64, _c1Cd::I64, _c1Ce::I64, _c1Cf::I64); _s1BY::P64 = _c1C6::I64; _s1BY::P64 = _s1BY::P64; goto c1Ch; c1Ch: _s1C1::P64 = _s1BY::P64; goto c1Ck; c1Ck: R1 = ()_closure+1; call (P64[(old + 8)])(R1) args: 8, res: 0, upd: 8; } }]

==================== Post control-flow optimisations ==================== 2014-03-08 08:25:31.729764 UTC

{offset c1Cm: if ((old + 0) - <highSp> < SpLim) goto c1Cp; else goto c1Cq; c1Cq: Hp = Hp + 152; if (Hp > HpLim) goto c1Cs; else goto c1Cr; c1Cs: HpAlloc = 152; goto c1Cp; c1Cp: R1 = a_r1za_closure; call (stg_gc_fun)(R1) args: 8, res: 0, upd: 8; c1Cr: I64[Hp - 144] = stg_MUT_ARR_PTRS_DIRTY_info; I64[Hp - 136] = 16; I64[Hp - 128] = 16; _c1C6::I64 = Hp - 144; _c1C7::I64 = _c1C6::I64 + 24; goto c1C8; c1C8: if (_c1C7::I64 < (_c1C6::I64 + 128)) goto c1Ca; else goto c1C9; c1Ca: P64[_c1C7::I64] = ()_closure+1; _c1C7::I64 = _c1C7::I64 + 8; goto c1C8; c1C9: _c1Cb::I64 = _c1C6::I64 + 24; _c1Cc::I64 = _c1Cb::I64 + 128; _c1Cd::I64 = 0; _c1Ce::I64 = 0; _c1Cf::I64 = 8; call MO_Memset(_c1Cc::I64, _c1Cd::I64, _c1Ce::I64, _c1Cf::I64); _s1BY::P64 = _c1C6::I64; _s1BY::P64 = _s1BY::P64; _s1C1::P64 = _s1BY::P64; R1 = ()_closure+1; call (P64[(old + 8)])(R1) args: 8, res: 0, upd: 8; }

==================== Layout Stack ==================== 2014-03-08 08:25:31.730253 UTC

{offset c1Cm: goto c1Cq; c1Cq: Hp = Hp + 152; if (Hp > HpLim) goto c1Cs; else goto c1Cr; c1Cs: HpAlloc = 152; goto c1Cp; c1Cp: R1 = a_r1za_closure; call (stg_gc_fun)(R1) args: 8, res: 0, upd: 8; c1Cr: I64[Hp - 144] = stg_MUT_ARR_PTRS_DIRTY_info; I64[Hp - 136] = 16; I64[Hp - 128] = 16; _c1C6::I64 = Hp - 144; _c1C7::I64 = _c1C6::I64 + 24; goto c1C8; c1C8: if (_c1C7::I64 < (_c1C6::I64 + 128)) goto c1Ca; else goto c1C9; c1Ca: P64[_c1C7::I64] = ()_closure+1; _c1C7::I64 = _c1C7::I64 + 8; goto c1C8; c1C9: _c1Cb::I64 = _c1C6::I64 + 24; _c1Cc::I64 = _c1Cb::I64 + 128; _c1Cd::I64 = 0; _c1Ce::I64 = 0; _c1Cf::I64 = 8; call MO_Memset(_c1Cc::I64, _c1Cd::I64, _c1Ce::I64, _c1Cf::I64); _s1BY::P64 = _c1C6::I64; _s1BY::P64 = _s1BY::P64; _s1C1::P64 = _s1BY::P64; R1 = ()_closure+1; call (P64[Sp])(R1) args: 8, res: 0, upd: 8; }

==================== CAFEnv ==================== 2014-03-08 08:25:31.73074 UTC

[(c1C8, {}), (c1C9, {}), (c1Ca, {}), (c1Cm, {}), (c1Cp, {}), (c1Cq, {}), (c1Cr, {}), (c1Cs, {})]

==================== after setInfoTableStackMap ==================== 2014-03-08 08:25:31.730895 UTC

a_r1za_entry() // [] { info_tbl: [(c1Cm, label: a_r1za_info rep:HeapRep static { Fun {arity: 1 fun_type: ArgSpec 3} })] stack_info: arg_space: 8 updfr_space: Just 8 } {offset c1Cm: goto c1Cq; c1Cq: Hp = Hp + 152; if (Hp > HpLim) goto c1Cs; else goto c1Cr; c1Cs: HpAlloc = 152; goto c1Cp; c1Cp: R1 = a_r1za_closure; call (stg_gc_fun)(R1) args: 8, res: 0, upd: 8; c1Cr: I64[Hp - 144] = stg_MUT_ARR_PTRS_DIRTY_info; I64[Hp - 136] = 16; I64[Hp - 128] = 16; _c1C6::I64 = Hp - 144; _c1C7::I64 = _c1C6::I64 + 24; goto c1C8; c1C8: if (_c1C7::I64 < (_c1C6::I64 + 128)) goto c1Ca; else goto c1C9; c1Ca: P64[_c1C7::I64] = ()_closure+1; _c1C7::I64 = _c1C7::I64 + 8; goto c1C8; c1C9: _c1Cb::I64 = _c1C6::I64 + 24; _c1Cc::I64 = _c1Cb::I64 + 128; _c1Cd::I64 = 0; _c1Ce::I64 = 0; _c1Cf::I64 = 8; call MO_Memset(_c1Cc::I64, _c1Cd::I64, _c1Ce::I64, _c1Cf::I64); _s1BY::P64 = _c1C6::I64; _s1BY::P64 = _s1BY::P64; _s1C1::P64 = _s1BY::P64; R1 = ()_closure+1; call (P64[Sp])(R1) args: 8, res: 0, upd: 8; } }

==================== Post control-flow optimisations ==================== 2014-03-08 08:25:31.731383 UTC

a_r1za_entry() // [] { info_tbl: [(c1Cm, label: a_r1za_info rep:HeapRep static { Fun {arity: 1 fun_type: ArgSpec 3} })] stack_info: arg_space: 8 updfr_space: Just 8 } {offset c1Cm: goto c1Cq; c1Cq: Hp = Hp + 152; if (Hp > HpLim) goto c1Cs; else goto c1Cr; c1Cs: HpAlloc = 152; goto c1Cp; c1Cp: R1 = a_r1za_closure; call (stg_gc_fun)(R1) args: 8, res: 0, upd: 8; c1Cr: I64[Hp - 144] = stg_MUT_ARR_PTRS_DIRTY_info; I64[Hp - 136] = 16; I64[Hp - 128] = 16; _c1C6::I64 = Hp - 144; _c1C7::I64 = _c1C6::I64 + 24; goto c1C8; c1C8: if (_c1C7::I64 < (_c1C6::I64 + 128)) goto c1Ca; else goto c1C9; c1Ca: P64[_c1C7::I64] = ()_closure+1; _c1C7::I64 = _c1C7::I64 + 8; goto c1C8; c1C9: _c1Cb::I64 = _c1C6::I64 + 24; _c1Cc::I64 = _c1Cb::I64 + 128; _c1Cd::I64 = 0; _c1Ce::I64 = 0; _c1Cf::I64 = 8; call MO_Memset(_c1Cc::I64, _c1Cd::I64, _c1Ce::I64, _c1Cf::I64); _s1BY::P64 = _c1C6::I64; _s1BY::P64 = _s1BY::P64; _s1C1::P64 = _s1BY::P64; R1 = ()_closure+1; call (P64[Sp])(R1) args: 8, res: 0, upd: 8; } }

==================== Post CPS Cmm ==================== 2014-03-08 08:25:31.731772 UTC

[section "data" { a_r1za_closure: const a_r1za_info; }, a_r1za_entry() // [] { info_tbl: [(c1Cm, label: a_r1za_info rep:HeapRep static { Fun {arity: 1 fun_type: ArgSpec 3} })] stack_info: arg_space: 8 updfr_space: Just 8 } {offset c1Cm: goto c1Cq; c1Cq: Hp = Hp + 152; if (Hp > HpLim) goto c1Cs; else goto c1Cr; c1Cs: HpAlloc = 152; goto c1Cp; c1Cp: R1 = a_r1za_closure; call (stg_gc_fun)(R1) args: 8, res: 0, upd: 8; c1Cr: I64[Hp - 144] = stg_MUT_ARR_PTRS_DIRTY_info; I64[Hp - 136] = 16; I64[Hp - 128] = 16; _c1C6::I64 = Hp - 144; _c1C7::I64 = _c1C6::I64 + 24; goto c1C8; c1C8: if (_c1C7::I64 < (_c1C6::I64 + 128)) goto c1Ca; else goto c1C9; c1Ca: P64[_c1C7::I64] = ()_closure+1; _c1C7::I64 = _c1C7::I64 + 8; goto c1C8; c1C9: _c1Cb::I64 = _c1C6::I64 + 24; _c1Cc::I64 = _c1Cb::I64 + 128; _c1Cd::I64 = 0; _c1Ce::I64 = 0; _c1Cf::I64 = 8; call MO_Memset(_c1Cc::I64, _c1Cd::I64, _c1Ce::I64, _c1Cf::I64); _s1BY::P64 = _c1C6::I64; _s1BY::P64 = _s1BY::P64; _s1C1::P64 = _s1BY::P64; R1 = ()_closure+1; call (P64[Sp])(R1) args: 8, res: 0, upd: 8; } }]

==================== Output Cmm ==================== 2014-03-08 08:25:31.73228 UTC

[section "data" { a_r1za_closure: const a_r1za_info; }, a_r1za_entry() // [] { info_tbl: [(c1Cm, label: a_r1za_info rep:HeapRep static { Fun {arity: 1 fun_type: ArgSpec 3} })] stack_info: arg_space: 8 updfr_space: Just 8 } {offset c1Cm: goto c1Cq; c1Cq: Hp = Hp + 152; if (Hp > HpLim) goto c1Cs; else goto c1Cr; c1Cs: HpAlloc = 152; goto c1Cp; c1Cp: R1 = a_r1za_closure; call (stg_gc_fun)(R1) args: 8, res: 0, upd: 8; c1Cr: I64[Hp - 144] = stg_MUT_ARR_PTRS_DIRTY_info; I64[Hp - 136] = 16; I64[Hp - 128] = 16; _c1C6::I64 = Hp - 144; _c1C7::I64 = _c1C6::I64 + 24; goto c1C8; c1C8: if (_c1C7::I64 < (_c1C6::I64 + 128)) goto c1Ca; else goto c1C9; c1Ca: P64[_c1C7::I64] = ()_closure+1; _c1C7::I64 = _c1C7::I64 + 8; goto c1C8; c1C9: _c1Cb::I64 = _c1C6::I64 + 24; _c1Cc::I64 = _c1Cb::I64 + 128; _c1Cd::I64 = 0; _c1Ce::I64 = 0; _c1Cf::I64 = 8; call MO_Memset(_c1Cc::I64, _c1Cd::I64, _c1Ce::I64, _c1Cf::I64); _s1BY::P64 = _c1C6::I64; _s1BY::P64 = _s1BY::P64; _s1C1::P64 = _s1BY::P64; R1 = ()_closure+1; call (P64[Sp])(R1) args: 8, res: 0, upd: 8; } }] }}}

.dump-opt-cmm file: {{{ ==================== Optimised Cmm ==================== 2014-03-08 08:25:31.732971 UTC

a_r1za_entry() // [] { [(c1Cm, a_r1za_info: const 4294967299; const 0; const 15;)] } {offset c1Cm: goto c1Cq; c1Cq: Hp = Hp + 152; if (Hp > I64[BaseReg + 856]) goto c1Cs; else goto c1Cr; c1Cs: I64[BaseReg + 904] = 152; goto c1Cp; c1Cp: R1 = PicBaseReg + a_r1za_closure; call (I64[BaseReg - 8])(R1) args: 8, res: 0, upd: 8; c1Cr: I64[Hp - 144] = I64[PicBaseReg + stg_MUT_ARR_PTRS_DIRTY_info@GOTPCREL]; I64[Hp - 136] = 16; I64[Hp - 128] = 16; _c1C6::I64 = Hp - 144; _c1C7::I64 = _c1C6::I64 + 24; goto c1C8; c1C8: if (_c1C7::I64 < (_c1C6::I64 + 128)) goto c1Ca; else goto c1C9; c1Ca: I64[_c1C7::I64] = PicBaseReg + (()_closure+1); _c1C7::I64 = _c1C7::I64 + 8; goto c1C8; c1C9: _c1Cb::I64 = _c1C6::I64 + 24; _c1Cc::I64 = _c1Cb::I64 + 128; _c1Cd::I64 = 0; _c1Ce::I64 = 0; _c1Cf::I64 = 8; call MO_Memset(_c1Cc::I64, _c1Cd::I64, _c1Ce::I64, _c1Cf::I64); _s1BY::P64 = _c1C6::I64; // nop _s1C1::P64 = _s1BY::P64; R1 = PicBaseReg + (()_closure+1); call (P64[Sp])(R1) args: 8, res: 0, upd: 8; } } }}}

New description: The optimizer seems to miss very basic optimizations. For example, look at this segment (full source below): {{{ _c1Cd::I64 = 0; _c1Ce::I64 = 0; _c1Cf::I64 = 8; call MO_Memset(_c1Cc::I64, _c1Cd::I64, _c1Ce::I64, _c1Cf::I64); }}} I'd expect all these temporaries (that are only used once) to be inlined, but they are not and thus we fail to see the static arguments to the `MO_Memset` call, which leads to further missed optimizations. This used to work in the old codegen. Furthermore, there are useless basic blocks in the output: {{{ c1Cm: goto c1C4; c1C4: }}} I'd expect them to be eliminated. Here's another non-sensical sequence: {{{ _s1BY::P64 = _c1C6::I64; // nop _s1C1::P64 = _s1BY::P64; }}} Neither of these temporaries are ever used. .dump-cmm file: {{{ ==================== Cmm produced by new codegen ==================== 2014-03-08 08:25:31.728672 UTC [section "data" { a_r1za_closure: const a_r1za_info; }, a_r1za_entry() // [] { info_tbl: [(c1Cm, label: a_r1za_info rep:HeapRep static { Fun {arity: 1 fun_type: ArgSpec 3} })] stack_info: arg_space: 8 updfr_space: Just 8 } {offset c1Cm: goto c1C4; c1C4: if ((old + 0) - <highSp> < SpLim) goto c1Cp; else goto c1Cq; c1Cq: goto c1C3; c1C3: Hp = Hp + 152; if (Hp > HpLim) goto c1Cs; else goto c1Cr; c1Cs: HpAlloc = 152; goto c1Cp; c1Cp: R1 = a_r1za_closure; call (stg_gc_fun)(R1) args: 8, res: 0, upd: 8; c1Cr: I64[Hp - 144] = stg_MUT_ARR_PTRS_DIRTY_info; I64[Hp - 136] = 16; I64[Hp - 128] = 16; _c1C6::I64 = Hp - 144; _c1C7::I64 = _c1C6::I64 + 24; goto c1C8; c1C8: if (_c1C7::I64 < (_c1C6::I64 + 128)) goto c1Ca; else goto c1C9; c1Ca: P64[_c1C7::I64] = ()_closure+1; _c1C7::I64 = _c1C7::I64 + 8; goto c1C8; c1C9: _c1Cb::I64 = _c1C6::I64 + 24; _c1Cc::I64 = _c1Cb::I64 + 128; _c1Cd::I64 = 0; _c1Ce::I64 = 0; _c1Cf::I64 = 8; call MO_Memset(_c1Cc::I64, _c1Cd::I64, _c1Ce::I64, _c1Cf::I64); _s1BY::P64 = _c1C6::I64; _s1BY::P64 = _s1BY::P64; goto c1Ch; c1Ch: _s1C1::P64 = _s1BY::P64; goto c1Ck; c1Ck: R1 = ()_closure+1; call (P64[(old + 8)])(R1) args: 8, res: 0, upd: 8; } }] ==================== Post control-flow optimisations ==================== 2014-03-08 08:25:31.729764 UTC {offset c1Cm: if ((old + 0) - <highSp> < SpLim) goto c1Cp; else goto c1Cq; c1Cq: Hp = Hp + 152; if (Hp > HpLim) goto c1Cs; else goto c1Cr; c1Cs: HpAlloc = 152; goto c1Cp; c1Cp: R1 = a_r1za_closure; call (stg_gc_fun)(R1) args: 8, res: 0, upd: 8; c1Cr: I64[Hp - 144] = stg_MUT_ARR_PTRS_DIRTY_info; I64[Hp - 136] = 16; I64[Hp - 128] = 16; _c1C6::I64 = Hp - 144; _c1C7::I64 = _c1C6::I64 + 24; goto c1C8; c1C8: if (_c1C7::I64 < (_c1C6::I64 + 128)) goto c1Ca; else goto c1C9; c1Ca: P64[_c1C7::I64] = ()_closure+1; _c1C7::I64 = _c1C7::I64 + 8; goto c1C8; c1C9: _c1Cb::I64 = _c1C6::I64 + 24; _c1Cc::I64 = _c1Cb::I64 + 128; _c1Cd::I64 = 0; _c1Ce::I64 = 0; _c1Cf::I64 = 8; call MO_Memset(_c1Cc::I64, _c1Cd::I64, _c1Ce::I64, _c1Cf::I64); _s1BY::P64 = _c1C6::I64; _s1BY::P64 = _s1BY::P64; _s1C1::P64 = _s1BY::P64; R1 = ()_closure+1; call (P64[(old + 8)])(R1) args: 8, res: 0, upd: 8; } ==================== Layout Stack ==================== 2014-03-08 08:25:31.730253 UTC {offset c1Cm: goto c1Cq; c1Cq: Hp = Hp + 152; if (Hp > HpLim) goto c1Cs; else goto c1Cr; c1Cs: HpAlloc = 152; goto c1Cp; c1Cp: R1 = a_r1za_closure; call (stg_gc_fun)(R1) args: 8, res: 0, upd: 8; c1Cr: I64[Hp - 144] = stg_MUT_ARR_PTRS_DIRTY_info; I64[Hp - 136] = 16; I64[Hp - 128] = 16; _c1C6::I64 = Hp - 144; _c1C7::I64 = _c1C6::I64 + 24; goto c1C8; c1C8: if (_c1C7::I64 < (_c1C6::I64 + 128)) goto c1Ca; else goto c1C9; c1Ca: P64[_c1C7::I64] = ()_closure+1; _c1C7::I64 = _c1C7::I64 + 8; goto c1C8; c1C9: _c1Cb::I64 = _c1C6::I64 + 24; _c1Cc::I64 = _c1Cb::I64 + 128; _c1Cd::I64 = 0; _c1Ce::I64 = 0; _c1Cf::I64 = 8; call MO_Memset(_c1Cc::I64, _c1Cd::I64, _c1Ce::I64, _c1Cf::I64); _s1BY::P64 = _c1C6::I64; _s1BY::P64 = _s1BY::P64; _s1C1::P64 = _s1BY::P64; R1 = ()_closure+1; call (P64[Sp])(R1) args: 8, res: 0, upd: 8; } ==================== CAFEnv ==================== 2014-03-08 08:25:31.73074 UTC [(c1C8, {}), (c1C9, {}), (c1Ca, {}), (c1Cm, {}), (c1Cp, {}), (c1Cq, {}), (c1Cr, {}), (c1Cs, {})] ==================== after setInfoTableStackMap ==================== 2014-03-08 08:25:31.730895 UTC a_r1za_entry() // [] { info_tbl: [(c1Cm, label: a_r1za_info rep:HeapRep static { Fun {arity: 1 fun_type: ArgSpec 3} })] stack_info: arg_space: 8 updfr_space: Just 8 } {offset c1Cm: goto c1Cq; c1Cq: Hp = Hp + 152; if (Hp > HpLim) goto c1Cs; else goto c1Cr; c1Cs: HpAlloc = 152; goto c1Cp; c1Cp: R1 = a_r1za_closure; call (stg_gc_fun)(R1) args: 8, res: 0, upd: 8; c1Cr: I64[Hp - 144] = stg_MUT_ARR_PTRS_DIRTY_info; I64[Hp - 136] = 16; I64[Hp - 128] = 16; _c1C6::I64 = Hp - 144; _c1C7::I64 = _c1C6::I64 + 24; goto c1C8; c1C8: if (_c1C7::I64 < (_c1C6::I64 + 128)) goto c1Ca; else goto c1C9; c1Ca: P64[_c1C7::I64] = ()_closure+1; _c1C7::I64 = _c1C7::I64 + 8; goto c1C8; c1C9: _c1Cb::I64 = _c1C6::I64 + 24; _c1Cc::I64 = _c1Cb::I64 + 128; _c1Cd::I64 = 0; _c1Ce::I64 = 0; _c1Cf::I64 = 8; call MO_Memset(_c1Cc::I64, _c1Cd::I64, _c1Ce::I64, _c1Cf::I64); _s1BY::P64 = _c1C6::I64; _s1BY::P64 = _s1BY::P64; _s1C1::P64 = _s1BY::P64; R1 = ()_closure+1; call (P64[Sp])(R1) args: 8, res: 0, upd: 8; } } ==================== Post control-flow optimisations ==================== 2014-03-08 08:25:31.731383 UTC a_r1za_entry() // [] { info_tbl: [(c1Cm, label: a_r1za_info rep:HeapRep static { Fun {arity: 1 fun_type: ArgSpec 3} })] stack_info: arg_space: 8 updfr_space: Just 8 } {offset c1Cm: goto c1Cq; c1Cq: Hp = Hp + 152; if (Hp > HpLim) goto c1Cs; else goto c1Cr; c1Cs: HpAlloc = 152; goto c1Cp; c1Cp: R1 = a_r1za_closure; call (stg_gc_fun)(R1) args: 8, res: 0, upd: 8; c1Cr: I64[Hp - 144] = stg_MUT_ARR_PTRS_DIRTY_info; I64[Hp - 136] = 16; I64[Hp - 128] = 16; _c1C6::I64 = Hp - 144; _c1C7::I64 = _c1C6::I64 + 24; goto c1C8; c1C8: if (_c1C7::I64 < (_c1C6::I64 + 128)) goto c1Ca; else goto c1C9; c1Ca: P64[_c1C7::I64] = ()_closure+1; _c1C7::I64 = _c1C7::I64 + 8; goto c1C8; c1C9: _c1Cb::I64 = _c1C6::I64 + 24; _c1Cc::I64 = _c1Cb::I64 + 128; _c1Cd::I64 = 0; _c1Ce::I64 = 0; _c1Cf::I64 = 8; call MO_Memset(_c1Cc::I64, _c1Cd::I64, _c1Ce::I64, _c1Cf::I64); _s1BY::P64 = _c1C6::I64; _s1BY::P64 = _s1BY::P64; _s1C1::P64 = _s1BY::P64; R1 = ()_closure+1; call (P64[Sp])(R1) args: 8, res: 0, upd: 8; } } ==================== Post CPS Cmm ==================== 2014-03-08 08:25:31.731772 UTC [section "data" { a_r1za_closure: const a_r1za_info; }, a_r1za_entry() // [] { info_tbl: [(c1Cm, label: a_r1za_info rep:HeapRep static { Fun {arity: 1 fun_type: ArgSpec 3} })] stack_info: arg_space: 8 updfr_space: Just 8 } {offset c1Cm: goto c1Cq; c1Cq: Hp = Hp + 152; if (Hp > HpLim) goto c1Cs; else goto c1Cr; c1Cs: HpAlloc = 152; goto c1Cp; c1Cp: R1 = a_r1za_closure; call (stg_gc_fun)(R1) args: 8, res: 0, upd: 8; c1Cr: I64[Hp - 144] = stg_MUT_ARR_PTRS_DIRTY_info; I64[Hp - 136] = 16; I64[Hp - 128] = 16; _c1C6::I64 = Hp - 144; _c1C7::I64 = _c1C6::I64 + 24; goto c1C8; c1C8: if (_c1C7::I64 < (_c1C6::I64 + 128)) goto c1Ca; else goto c1C9; c1Ca: P64[_c1C7::I64] = ()_closure+1; _c1C7::I64 = _c1C7::I64 + 8; goto c1C8; c1C9: _c1Cb::I64 = _c1C6::I64 + 24; _c1Cc::I64 = _c1Cb::I64 + 128; _c1Cd::I64 = 0; _c1Ce::I64 = 0; _c1Cf::I64 = 8; call MO_Memset(_c1Cc::I64, _c1Cd::I64, _c1Ce::I64, _c1Cf::I64); _s1BY::P64 = _c1C6::I64; _s1BY::P64 = _s1BY::P64; _s1C1::P64 = _s1BY::P64; R1 = ()_closure+1; call (P64[Sp])(R1) args: 8, res: 0, upd: 8; } }] ==================== Output Cmm ==================== 2014-03-08 08:25:31.73228 UTC [section "data" { a_r1za_closure: const a_r1za_info; }, a_r1za_entry() // [] { info_tbl: [(c1Cm, label: a_r1za_info rep:HeapRep static { Fun {arity: 1 fun_type: ArgSpec 3} })] stack_info: arg_space: 8 updfr_space: Just 8 } {offset c1Cm: goto c1Cq; c1Cq: Hp = Hp + 152; if (Hp > HpLim) goto c1Cs; else goto c1Cr; c1Cs: HpAlloc = 152; goto c1Cp; c1Cp: R1 = a_r1za_closure; call (stg_gc_fun)(R1) args: 8, res: 0, upd: 8; c1Cr: I64[Hp - 144] = stg_MUT_ARR_PTRS_DIRTY_info; I64[Hp - 136] = 16; I64[Hp - 128] = 16; _c1C6::I64 = Hp - 144; _c1C7::I64 = _c1C6::I64 + 24; goto c1C8; c1C8: if (_c1C7::I64 < (_c1C6::I64 + 128)) goto c1Ca; else goto c1C9; c1Ca: P64[_c1C7::I64] = ()_closure+1; _c1C7::I64 = _c1C7::I64 + 8; goto c1C8; c1C9: _c1Cb::I64 = _c1C6::I64 + 24; _c1Cc::I64 = _c1Cb::I64 + 128; _c1Cd::I64 = 0; _c1Ce::I64 = 0; _c1Cf::I64 = 8; call MO_Memset(_c1Cc::I64, _c1Cd::I64, _c1Ce::I64, _c1Cf::I64); _s1BY::P64 = _c1C6::I64; _s1BY::P64 = _s1BY::P64; _s1C1::P64 = _s1BY::P64; R1 = ()_closure+1; call (P64[Sp])(R1) args: 8, res: 0, upd: 8; } }] }}} .dump-opt-cmm file: {{{ ==================== Optimised Cmm ==================== 2014-03-08 08:25:31.732971 UTC a_r1za_entry() // [] { [(c1Cm, a_r1za_info: const 4294967299; const 0; const 15;)] } {offset c1Cm: goto c1Cq; c1Cq: Hp = Hp + 152; if (Hp > I64[BaseReg + 856]) goto c1Cs; else goto c1Cr; c1Cs: I64[BaseReg + 904] = 152; goto c1Cp; c1Cp: R1 = PicBaseReg + a_r1za_closure; call (I64[BaseReg - 8])(R1) args: 8, res: 0, upd: 8; c1Cr: I64[Hp - 144] = I64[PicBaseReg + stg_MUT_ARR_PTRS_DIRTY_info@GOTPCREL]; I64[Hp - 136] = 16; I64[Hp - 128] = 16; _c1C6::I64 = Hp - 144; _c1C7::I64 = _c1C6::I64 + 24; goto c1C8; c1C8: if (_c1C7::I64 < (_c1C6::I64 + 128)) goto c1Ca; else goto c1C9; c1Ca: I64[_c1C7::I64] = PicBaseReg + (()_closure+1); _c1C7::I64 = _c1C7::I64 + 8; goto c1C8; c1C9: _c1Cb::I64 = _c1C6::I64 + 24; _c1Cc::I64 = _c1Cb::I64 + 128; _c1Cd::I64 = 0; _c1Ce::I64 = 0; _c1Cf::I64 = 8; call MO_Memset(_c1Cc::I64, _c1Cd::I64, _c1Ce::I64, _c1Cf::I64); _s1BY::P64 = _c1C6::I64; // nop _s1C1::P64 = _s1BY::P64; R1 = PicBaseReg + (()_closure+1); call (P64[Sp])(R1) args: 8, res: 0, upd: 8; } } }}} -- -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/8860#comment:3 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler

#8860: Optimized Cmm isn't -------------------------------------+------------------------------------ Reporter: tibbe | Owner: Type: bug | Status: closed Priority: normal | Milestone: Component: Compiler | Version: 7.6.3 Resolution: invalid | Keywords: Operating System: Unknown/Multiple | Architecture: Unknown/Multiple Type of failure: None/Unknown | Difficulty: Unknown Test Case: | Blocked By: Blocking: | Related Tickets: -------------------------------------+------------------------------------ Changes (by tibbe): * status: new => closed * resolution: => invalid Comment: Apologies for being stupid. I wasn't. -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/8860#comment:5 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler