%% -*- erlang-indent-level: 2 -*-
%%
%% %CopyrightBegin%
%%
%% Copyright Ericsson AB 2001-2016. All Rights Reserved.
%%
%% Licensed under the Apache License, Version 2.0 (the "License");
%% you may not use this file except in compliance with the License.
%% You may obtain a copy of the License at
%%
%% http://www.apache.org/licenses/LICENSE-2.0
%%
%% Unless required by applicable law or agreed to in writing, software
%% distributed under the License is distributed on an "AS IS" BASIS,
%% WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
%% See the License for the specific language governing permissions and
%% limitations under the License.
%%
%% %CopyrightEnd%
%%
-ifdef(HIPE_AMD64).
-define(HIPE_X86_RA_POSTCONDITIONS, hipe_amd64_ra_postconditions).
-define(HIPE_X86_REGISTERS, hipe_amd64_registers).
-define(HIPE_X86_SPECIFIC, hipe_amd64_specific).
-define(ECX, rcx).
-else.
-define(HIPE_X86_RA_POSTCONDITIONS, hipe_x86_ra_postconditions).
-define(HIPE_X86_REGISTERS, hipe_x86_registers).
-define(HIPE_X86_SPECIFIC, hipe_x86_specific).
-define(ECX, ecx).
-endif.
-module(?HIPE_X86_RA_POSTCONDITIONS).
-export([check_and_rewrite/3]).
-include("../x86/hipe_x86.hrl").
-define(HIPE_INSTRUMENT_COMPILER, true).
-include("../main/hipe.hrl").
-define(count_temp(T), ?cons_counter(counter_mfa_mem_temps, T)).
check_and_rewrite(CFG, Coloring, Strategy) ->
%% io:format("Converting\n"),
TempMap = hipe_temp_map:cols2tuple(Coloring, ?HIPE_X86_SPECIFIC, no_context),
%% io:format("Rewriting\n"),
do_bbs(hipe_x86_cfg:labels(CFG), TempMap, Strategy, CFG, false).
do_bbs([], _, _, CFG, DidSpill) -> {CFG, DidSpill};
do_bbs([Lbl|Lbls], TempMap, Strategy, CFG0, DidSpill0) ->
Code0 = hipe_bb:code(BB = hipe_x86_cfg:bb(CFG0, Lbl)),
{Code, DidSpill} = do_insns(Code0, TempMap, Strategy, [], DidSpill0),
CFG = hipe_x86_cfg:bb_add(CFG0, Lbl, hipe_bb:code_update(BB, Code)),
do_bbs(Lbls, TempMap, Strategy, CFG, DidSpill).
do_insns([I|Insns], TempMap, Strategy, Accum, DidSpill0) ->
{NewIs, DidSpill1} = do_insn(I, TempMap, Strategy),
do_insns(Insns, TempMap, Strategy, lists:reverse(NewIs, Accum), DidSpill0 or DidSpill1);
do_insns([], _TempMap, _Strategy, Accum, DidSpill) ->
{lists:reverse(Accum), DidSpill}.
do_insn(I, TempMap, Strategy) -> % Insn -> {Insn list, DidSpill}
case I of
#alu{} ->
do_alu(I, TempMap, Strategy);
#cmp{} ->
do_cmp(I, TempMap, Strategy);
#imul{} ->
do_imul(I, TempMap, Strategy);
#jmp_switch{} ->
do_jmp_switch(I, TempMap, Strategy);
#lea{} ->
do_lea(I, TempMap, Strategy);
#move{} ->
do_move(I, TempMap, Strategy);
#move64{} ->
do_move64(I, TempMap, Strategy);
#movsx{} ->
do_movx(I, TempMap, Strategy);
#movzx{} ->
do_movx(I, TempMap, Strategy);
#fmove{} ->
do_fmove(I, TempMap, Strategy);
#shift{} ->
do_shift(I, TempMap, Strategy);
_ ->
%% comment, jmp*, label, pseudo_call, pseudo_jcc, pseudo_tailcall,
%% pseudo_tailcall_prepare, push, ret
{[I], false}
end.
%%% Fix an alu op.
do_alu(I, TempMap, Strategy) ->
#alu{src=Src0,dst=Dst0} = I,
{FixSrc,Src,FixDst,Dst,DidSpill} =
do_binary(Src0, Dst0, TempMap, Strategy),
{FixSrc ++ FixDst ++ [I#alu{src=Src,dst=Dst}], DidSpill}.
%%% Fix a cmp op.
do_cmp(I, TempMap, Strategy) ->
#cmp{src=Src0,dst=Dst0} = I,
{FixSrc, Src, FixDst, Dst, DidSpill} =
do_binary(Src0, Dst0, TempMap, Strategy),
{FixSrc ++ FixDst ++ [I#cmp{src=Src,dst=Dst}], DidSpill}.
%%% Fix an imul op.
do_imul(I, TempMap, Strategy) ->
#imul{imm_opt=ImmOpt,src=Src0,temp=Temp0} = I,
{FixSrc,Src,DidSpill1} = fix_src_operand(Src0, TempMap, Strategy), % temp1
{FixTempSrc,Temp,FixTempDst,DidSpill2} =
case is_spilled(Temp0, TempMap) of
false ->
{[], Temp0, [], false};
true ->
Reg = spill_temp0('untagged', Strategy),
{case ImmOpt of
[] -> [hipe_x86:mk_move(Temp0, Reg)]; % temp *= src
_ -> [] % temp = src * imm
end,
Reg,
[hipe_x86:mk_move(Reg, Temp0)],
true}
end,
{FixSrc ++ FixTempSrc ++ [I#imul{src=Src,temp=Temp}] ++ FixTempDst,
DidSpill1 or DidSpill2}.
%%% Fix a jmp_switch op.
-ifdef(HIPE_AMD64).
do_jmp_switch(I, TempMap, Strategy) ->
#jmp_switch{temp=Temp, jtab=Tab} = I,
case is_spilled(Temp, TempMap) of
false ->
case is_spilled(Tab, TempMap) of
false ->
{[I], false};
true ->
NewTab = spill_temp('untagged', Strategy),
{[hipe_x86:mk_move(Tab, NewTab), I#jmp_switch{jtab=Tab}],
true}
end;
true ->
case is_spilled(Tab, TempMap) of
false ->
NewTmp = spill_temp('untagged', Strategy),
{[hipe_x86:mk_move(Temp, NewTmp), I#jmp_switch{temp=NewTmp}],
true};
true ->
NewTmp = spill_temp('untagged', Strategy),
NewTab = spill_temp0('untagged', Strategy),
{[hipe_x86:mk_move(Temp, NewTmp),
hipe_x86:mk_move(Tab, NewTab),
I#jmp_switch{temp=NewTmp, jtab=NewTab}],
true}
end
end.
-else. % not AMD64
do_jmp_switch(I, TempMap, Strategy) ->
#jmp_switch{temp=Temp} = I,
case is_spilled(Temp, TempMap) of
false ->
{[I], false};
true ->
NewTmp = spill_temp('untagged', Strategy),
{[hipe_x86:mk_move(Temp, NewTmp), I#jmp_switch{temp=NewTmp}],
true}
end.
-endif. % not AMD64
%%% Fix a lea op.
do_lea(I, TempMap, Strategy) ->
#lea{mem=Mem0,temp=Temp0} = I,
{FixMem, Mem, DidSpill1} = fix_mem_operand(Mem0, TempMap, temp1(Strategy)),
case Mem of
#x86_mem{base=Base, off=#x86_imm{value=0}} ->
%% We've decayed into a move due to both operands being memory (there's an
%% 'add' in FixMem).
{FixMem ++ [hipe_x86:mk_move(Base, Temp0)], DidSpill1};
#x86_mem{} ->
{StoreTemp, Temp, DidSpill2} =
case is_mem_opnd(Temp0, TempMap) of
false -> {[], Temp0, false};
true ->
Temp1 = clone2(Temp0, temp0(Strategy)),
{[hipe_x86:mk_move(Temp1, Temp0)], Temp1, true}
end,
{FixMem ++ [I#lea{mem=Mem,temp=Temp} | StoreTemp], DidSpill1 or DidSpill2}
end.
%%% Fix a move op.
do_move(I, TempMap, Strategy) ->
#move{src=Src0,dst=Dst0} = I,
{FixSrc, Src, FixDst, Dst, DidSpill} =
do_check_byte_move(Src0, Dst0, TempMap, Strategy),
{FixSrc ++ FixDst ++ [I#move{src=Src,dst=Dst}],
DidSpill}.
-ifdef(HIPE_AMD64).
%%% AMD64 has no issues with byte moves.
do_check_byte_move(Src0, Dst0, TempMap, Strategy) ->
do_binary(Src0, Dst0, TempMap, Strategy).
-else. % not AMD64
%%% x86 can only do byte moves to a subset of the integer registers.
do_check_byte_move(Src0, Dst0, TempMap, Strategy) ->
case Dst0 of
#x86_mem{type=byte} ->
do_byte_move(Src0, Dst0, TempMap, Strategy);
_ ->
do_binary(Src0, Dst0, TempMap, Strategy)
end.
do_byte_move(Src0, Dst0, TempMap, Strategy) ->
{FixSrc, Src, DidSpill1} = fix_src_operand(Src0, TempMap, Strategy),
{FixDst, Dst, DidSpill2} = fix_dst_operand(Dst0, TempMap, Strategy),
Reg = hipe_x86_registers:eax(),
{FixSrc3, Src3} = % XXX: this just checks Src, the result is known!
case Src of
#x86_imm{} ->
{FixSrc, Src};
#x86_temp{reg=Reg} -> % small moves must start from reg 1->4
{FixSrc, Src} % so variable sources are always put in eax
end,
{FixSrc3, Src3, FixDst, Dst,
DidSpill2 or DidSpill1}.
-endif. % not AMD64
%%% Fix a move64 op.
do_move64(I, TempMap, Strategy) ->
#move64{dst=Dst} = I,
case is_spilled(Dst, TempMap) of
false ->
{[I], false};
true ->
Reg = clone(Dst, Strategy),
{[I#move64{dst=Reg}, hipe_x86:mk_move(Reg, Dst)], true}
end.
%%% Fix a movx op.
do_movx(I, TempMap, Strategy) ->
{{FixSrc, Src, DidSpill1}, {FixDst, Dst, DidSpill2}} =
case I of
#movsx{src=Src0,dst=Dst0} ->
{fix_src_operand(Src0, TempMap, Strategy),
fix_dst_operand(Dst0, TempMap, Strategy)};
#movzx{src=Src0,dst=Dst0} ->
{fix_src_operand(Src0, TempMap, Strategy),
fix_dst_operand(Dst0, TempMap, Strategy)}
end,
{I3, DidSpill3} =
case is_spilled(Dst, TempMap) of
false ->
I2 = case I of
#movsx{} ->
[hipe_x86:mk_movsx(Src, Dst)];
#movzx{} ->
[hipe_x86:mk_movzx(Src, Dst)]
end,
{I2, false};
true ->
Dst2 = clone(Dst, Strategy),
I2 =
case I of
#movsx{} ->
[hipe_x86:mk_movsx(Src, Dst2), hipe_x86:mk_move(Dst2, Dst)];
#movzx{} ->
[hipe_x86:mk_movzx(Src, Dst2), hipe_x86:mk_move(Dst2, Dst)]
end,
{I2, true}
end,
{FixSrc++FixDst++I3,
DidSpill3 or DidSpill2 or DidSpill1}.
%%% Fix an fmove op.
do_fmove(I, TempMap, Strategy) ->
#fmove{src=Src0,dst=Dst0} = I,
{FixSrc, Src, DidSpill1} = fix_src_operand(Src0, TempMap, Strategy),
{FixDst, Dst, DidSpill2} = fix_dst_operand(Dst0, TempMap, Strategy),
%% fmoves from memory position to memory position is handled
%% by the f.p. register allocator.
{FixSrc ++ FixDst ++ [I#fmove{src=Src,dst=Dst}],
DidSpill1 or DidSpill2}.
%%% Fix a shift operation.
%%% 1. remove pseudos from any explicit memory operands
%%% 2. if the source is a register or memory position
%%% make sure to move it to %ecx
do_shift(I, TempMap, Strategy) ->
#shift{src=Src0,dst=Dst0} = I,
{FixDst, Dst, DidSpill} = fix_dst_operand(Dst0, TempMap, Strategy),
Reg = ?HIPE_X86_REGISTERS:?ECX(),
case Src0 of
#x86_imm{} ->
{FixDst ++ [I#shift{dst=Dst}], DidSpill};
#x86_temp{reg=Reg} ->
{FixDst ++ [I#shift{dst=Dst}], DidSpill}
end.
%%% Fix the operands of a binary op.
%%% 1. remove pseudos from any explicit memory operands
%%% 2. if both operands are (implicit or explicit) memory operands,
%%% move src to a reg and use reg as src in the original insn
do_binary(Src0, Dst0, TempMap, Strategy) ->
{FixSrc, Src, DidSpill1} = fix_src_operand(Src0, TempMap, Strategy),
{FixDst, Dst, DidSpill2} = fix_dst_operand(Dst0, TempMap, Strategy),
{FixSrc3, Src3, DidSpill3} =
case is_mem_opnd(Src, TempMap) of
false ->
{FixSrc, Src, false};
true ->
case is_mem_opnd(Dst, TempMap) of
false ->
{FixSrc, Src, false};
true ->
Src2 = clone(Src, Strategy),
FixSrc2 = FixSrc ++ [hipe_x86:mk_move(Src, Src2)],
{FixSrc2, Src2, true}
end
end,
{FixSrc3, Src3, FixDst, Dst,
DidSpill3 or DidSpill2 or DidSpill1}.
%%% Fix any x86_mem operand to not refer to any spilled temps.
fix_src_operand(Opnd, TmpMap, Strategy) ->
fix_mem_operand(Opnd, TmpMap, temp1(Strategy)).
temp1('normal') -> [];
temp1('linearscan') -> ?HIPE_X86_REGISTERS:temp1().
fix_dst_operand(Opnd, TempMap, Strategy) ->
fix_mem_operand(Opnd, TempMap, temp0(Strategy)).
temp0('normal') -> [];
temp0('linearscan') -> ?HIPE_X86_REGISTERS:temp0().
fix_mem_operand(Opnd, TempMap, RegOpt) -> % -> {[fixupcode], newop, DidSpill}
case Opnd of
#x86_mem{base=Base,off=Off} ->
case is_mem_opnd(Base, TempMap) of
false ->
case is_mem_opnd(Off, TempMap) of
false ->
{[], Opnd, false};
true ->
Temp = clone2(Off, RegOpt),
{[hipe_x86:mk_move(Off, Temp)],
Opnd#x86_mem{off=Temp},
true}
end;
true ->
Temp = clone2(Base, RegOpt),
case is_mem_opnd(Off, TempMap) of
false -> % imm/reg(pseudo)
{[hipe_x86:mk_move(Base, Temp)],
Opnd#x86_mem{base=Temp},
true};
true -> % pseudo(pseudo)
{[hipe_x86:mk_move(Base, Temp),
hipe_x86:mk_alu('add', Off, Temp)],
Opnd#x86_mem{base=Temp, off=hipe_x86:mk_imm(0)},
true}
end
end;
_ ->
{[], Opnd, false}
end.
%%% Check if an operand denotes a memory cell (mem or pseudo).
is_mem_opnd(Opnd, TempMap) ->
R =
case Opnd of
#x86_mem{} -> true;
#x86_temp{} ->
Reg = hipe_x86:temp_reg(Opnd),
case hipe_x86:temp_is_allocatable(Opnd) of
true ->
case
hipe_temp_map:is_spilled(Reg, TempMap) of
true ->
?count_temp(Reg),
true;
false -> false
end;
false -> true
end;
_ -> false
end,
%% io:format("Op ~w mem: ~w\n",[Opnd,R]),
R.
%%% Check if an operand is a spilled Temp.
is_spilled(Temp, TempMap) ->
case hipe_x86:temp_is_allocatable(Temp) of
true ->
Reg = hipe_x86:temp_reg(Temp),
case hipe_temp_map:is_spilled(Reg, TempMap) of
true ->
?count_temp(Reg),
true;
false ->
false
end;
false -> true
end.
%%% Make Reg a clone of Dst (attach Dst's type to Reg).
clone(Dst, Strategy) ->
Type =
case Dst of
#x86_mem{} -> hipe_x86:mem_type(Dst);
#x86_temp{} -> hipe_x86:temp_type(Dst)
end,
spill_temp(Type, Strategy).
spill_temp0(Type, 'normal') when Type =/= double ->
hipe_x86:mk_new_temp(Type);
spill_temp0(Type, 'linearscan') when Type =/= double ->
hipe_x86:mk_temp(?HIPE_X86_REGISTERS:temp0(), Type).
spill_temp(Type, 'normal') when Type =/= double ->
hipe_x86:mk_new_temp(Type);
spill_temp(Type, 'linearscan') when Type =/= double ->
hipe_x86:mk_temp(?HIPE_X86_REGISTERS:temp1(), Type).
%%% Make a certain reg into a clone of Dst
clone2(Dst, RegOpt) ->
Type =
case Dst of
#x86_mem{} -> hipe_x86:mem_type(Dst);
#x86_temp{} -> hipe_x86:temp_type(Dst)
end,
case RegOpt of
[] when Type =/= double -> hipe_x86:mk_new_temp(Type);
Reg -> hipe_x86:mk_temp(Reg, Type)
end.