X-Git-Url: http://git.megacz.com/?a=blobdiff_plain;f=ghc%2Fdriver%2Fmangler%2Fghc-asm.lprl;h=29cbfb10780150550a1a488041dd2905ec22ae0d;hb=5ecfa1f1d48111f5c95ec2fdae895fa24f8e321a;hp=73d05b96d634fb3b0e190d8360e427d84a2ba0c8;hpb=f0e0a108cc121f9b0afdb4311f8b6849efb4c907;p=ghc-hetmet.git diff --git a/ghc/driver/mangler/ghc-asm.lprl b/ghc/driver/mangler/ghc-asm.lprl index 73d05b9..29cbfb1 100644 --- a/ghc/driver/mangler/ghc-asm.lprl +++ b/ghc/driver/mangler/ghc-asm.lprl @@ -96,8 +96,8 @@ sub init_TARGET_STUFF { $T_CONST_LBL = '^\$L?C(\d+):$'; # regexp for what such a lbl looks like $T_POST_LBL = ':'; - $T_MOVE_DIRVS = '^(\s*(\.align\s+\d+|\.(globl|ent)\s+\S+|\#.*|\.(file|loc)\s+\S+\s+\S+|\.text|\.r?data)\n)'; - $T_COPY_DIRVS = '^\s*(\#|\.(file|globl|ent|loc))'; + $T_MOVE_DIRVS = '^(\s*(\$.*\.\.ng:|\.align\s+\d+|\.(globl|ent)\s+\S+|\#.*|\.(file|loc)\s+\S+\s+\S+|\.text|\.r?data)\n)'; + $T_COPY_DIRVS = '^\s*(\$.*\.\.ng:|\#|\.(file|globl|ent|loc))'; $T_hsc_cc_PAT = '\.ascii.*\)(hsc|cc) (.*)\\\\11"\n\t\.ascii\s+"(.*)\\\\0"'; $T_DOT_WORD = '\.(long|quad|byte|word)'; @@ -110,7 +110,6 @@ sub init_TARGET_STUFF { $T_HDR_srt = "\.text\n\t\.align 3\n"; $T_HDR_info = "\.text\n\t\.align 3\n"; $T_HDR_entry = "\.text\n\t\.align 3\n"; - $T_HDR_fast = "\.text\n\t\.align 3\n"; $T_HDR_vector = "\.text\n\t\.align 3\n"; $T_HDR_direct = "\.text\n\t\.align 3\n"; $T_create_word = "\t.quad"; @@ -138,7 +137,6 @@ sub init_TARGET_STUFF { $T_HDR_srt = "\t.SPACE \$PRIVATE\$\n\t.SUBSPA \$DATA\$\n\t\.align 4\n"; $T_HDR_info = "\t.SPACE \$TEXT\$\n\t.SUBSPA \$CODE\$\n\t\.align 4\n"; $T_HDR_entry = "\t.SPACE \$TEXT\$\n\t.SUBSPA \$CODE\$\n\t\.align 4\n"; - $T_HDR_fast = "\t.SPACE \$TEXT\$\n\t.SUBSPA \$CODE\$\n\t\.align 4\n"; $T_HDR_vector = "\t.SPACE \$TEXT\$\n\t.SUBSPA \$CODE\$\n\t\.align 4\n"; $T_HDR_direct = "\t.SPACE \$TEXT\$\n\t.SUBSPA \$CODE\$\n\t\.align 4\n"; $T_create_word = "\t.word"; @@ -166,11 +164,9 @@ sub init_TARGET_STUFF { $T_HDR_data = "\.data\n\t\.align 2\n"; $T_HDR_consist = "\.text\n"; $T_HDR_closure = "\.data\n\t\.align 2\n"; - $T_HDR_closure = "\.data\n\t\.align 2\n\t.long 0\n" if ( $TargetPlatform =~ /.*-mingw32$/ ); $T_HDR_srt = "\.text\n\t\.align 2\n"; $T_HDR_info = "\.text\n\t\.align 2\n"; # NB: requires padding $T_HDR_entry = "\.text\n"; # no .align so we're right next to _info (arguably wrong...?) - $T_HDR_fast = "\.text\n\t\.align 2,0x90\n"; $T_HDR_vector = "\.text\n\t\.align 2\n"; # NB: requires padding $T_HDR_direct = "\.text\n\t\.align 2,0x90\n"; $T_create_word = "\t.word"; @@ -188,8 +184,8 @@ sub init_TARGET_STUFF { $T_X86_PRE_LLBL = '.L'; $T_X86_BADJMP = '^\tjmp\s+[^\.\*]'; - $T_MOVE_DIRVS = '^(\s*(\.(p2)?align\s+\d+(,\s*0x90)?|\.globl\s+\S+|\.text|\.data|\.section\s+.*|\.type\s+.*|\.Lfe.*\n\s*\.size\s+.*|\.size\s+.*|\.ident.*)\n)'; - $T_COPY_DIRVS = '\.(globl)'; + $T_MOVE_DIRVS = '^(\s*(\.(p2)?align\s+\d+(,\s*0x90)?|\.globl\s+\S+|\.text|\.data|\.section\s+.*|\.type\s+.*|\.size\s+\S+\s*,\s*\d+|\.ident.*|\.local.*)\n)'; + $T_COPY_DIRVS = '^\s*\.(globl|type|size|local)'; if ( $TargetPlatform =~ /freebsd|netbsd/ ) { $T_hsc_cc_PAT = '\.ascii.*\)(hsc|cc) (.*)\\\\11"\n\t\.ascii\s+"(.*)\\\\0"'; @@ -197,7 +193,7 @@ sub init_TARGET_STUFF { $T_hsc_cc_PAT = '\.string.*\)(hsc|cc) (.*)\\\\t(.*)"'; } - $T_DOT_WORD = '\.(long|value|byte|zero)'; + $T_DOT_WORD = '\.(long|value|word|byte|zero)'; $T_DOT_GLOBAL = '\.globl'; $T_HDR_literal = "\.section\t\.rodata\n"; # or just use .text??? (WDP 95/11) $T_HDR_misc = "\.text\n\t\.align 4\n"; @@ -207,12 +203,38 @@ sub init_TARGET_STUFF { $T_HDR_srt = "\.text\n\t\.align 4\n"; # ToDo: change align? $T_HDR_info = "\.text\n\t\.align 4\n"; # NB: requires padding $T_HDR_entry = "\.text\n"; # no .align so we're right next to _info (arguably wrong...?) - $T_HDR_fast = "\.text\n\t\.align 4\n"; $T_HDR_vector = "\.text\n\t\.align 4\n"; # NB: requires padding $T_HDR_direct = "\.text\n\t\.align 4\n"; $T_create_word = "\t.word"; #--------------------------------------------------------# + } elsif ( $TargetPlatform =~ /^ia64-.*-linux$/ ) { + + $T_STABBY = 0; # 1 iff .stab things (usually if a.out format) + $T_US = ''; # _ if symbols have an underscore on the front + $T_PRE_APP = '#'; + $T_CONST_LBL = '^\.LC(\d+):$'; # regexp for what such a lbl looks like + $T_POST_LBL = ':'; + + $T_MOVE_DIRVS = '^(\s*\.(global|proc|pred\.safe_across_calls|text|data|section|subsection|align|size|type|ident)\s+.*\n)'; + $T_COPY_DIRVS = '\.(global|proc)'; + + $T_hsc_cc_PAT = '\.string.*\)(hsc|cc) (.*)\\\\t(.*)"'; + $T_DOT_WORD = '\.(long|value|byte|zero)'; + $T_DOT_GLOBAL = '\.global'; + $T_HDR_literal = "\.section\t\.rodata\n"; + $T_HDR_misc = "\.text\n\t\.align 8\n"; + $T_HDR_data = "\.data\n\t\.align 8\n"; + $T_HDR_consist = "\.text\n"; + $T_HDR_closure = "\.data\n\t\.align 8\n"; + $T_HDR_srt = "\.text\n\t\.align 8\n"; + $T_HDR_info = "\.text\n\t\.align 8\n"; + $T_HDR_entry = "\.text\n\t\.align 16\n"; + $T_HDR_vector = "\.text\n\t\.align 8\n"; + $T_HDR_direct = "\.text\n\t\.align 8\n"; + $T_create_word = "\t.word"; + + #--------------------------------------------------------# } elsif ( $TargetPlatform =~ /^m68k-.*-sunos4/ ) { $T_STABBY = 1; # 1 iff .stab things (usually if a.out format) @@ -235,7 +257,6 @@ sub init_TARGET_STUFF { $T_HDR_srt = "\.text\n\t\.even\n"; $T_HDR_info = "\.text\n\t\.even\n"; $T_HDR_entry = "\.text\n\t\.even\n"; - $T_HDR_fast = "\.text\n\t\.even\n"; $T_HDR_vector = "\.text\n\t\.even\n"; $T_HDR_direct = "\.text\n\t\.even\n"; $T_create_word = "\t.long"; @@ -263,7 +284,6 @@ sub init_TARGET_STUFF { $T_HDR_srt = "\t\.text\n\t\.align 2\n"; $T_HDR_info = "\t\.text\n\t\.align 2\n"; $T_HDR_entry = "\t\.text\n\t\.align 2\n"; - $T_HDR_fast = "\t\.text\n\t\.align 2\n"; $T_HDR_vector = "\t\.text\n\t\.align 2\n"; $T_HDR_direct = "\t\.text\n\t\.align 2\n"; $T_create_word = "\t.word"; @@ -277,11 +297,11 @@ sub init_TARGET_STUFF { $T_CONST_LBL = '^\LC\d+:'; # regexp for what such a lbl looks like $T_POST_LBL = ':'; - $T_MOVE_DIRVS = '^(\s*(\.align \d+|\.text|\.data|\.cstring|\.non_lazy_symbol_pointer|\.const|\.static_const|\.literal4|\.literal8|\.static_data|\.globl \S+)\n)'; + $T_MOVE_DIRVS = '^(\s*(\.align \d+|\.text|\.data|\.const_data|\.cstring|\.non_lazy_symbol_pointer|\.const|\.static_const|\.literal4|\.literal8|\.static_data|\.globl \S+)\n)'; $T_COPY_DIRVS = '\.(globl)'; $T_hsc_cc_PAT = '\.byte.*\)(hsc|cc) (.*)"\n\t\.byte \d+\n\t\.byte "(.*)"\n\t\.byte \d+'; - $T_DOT_WORD = '\.(long|short|byte)'; + $T_DOT_WORD = '\.(long|short|byte|fill|space)'; $T_DOT_GLOBAL = '\.globl'; $T_HDR_toc = "\.toc\n"; $T_HDR_literal = "\t\.const_data\n\t\.align 2\n"; @@ -292,7 +312,6 @@ sub init_TARGET_STUFF { $T_HDR_srt = "\t\.text\n\t\.align 2\n"; $T_HDR_info = "\t\.text\n\t\.align 2\n"; $T_HDR_entry = "\t\.text\n\t\.align 2\n"; - $T_HDR_fast = "\t\.text\n\t\.align 2\n"; $T_HDR_vector = "\t\.text\n\t\.align 2\n"; $T_HDR_direct = "\t\.text\n\t\.align 2\n"; $T_create_word = "\t.long"; @@ -310,7 +329,7 @@ sub init_TARGET_STUFF { $T_COPY_DIRVS = '\.(global|proc|stab)'; $T_hsc_cc_PAT = '\.asciz.*\)(hsc|cc) (.*)\\\\t(.*)"'; - $T_DOT_WORD = '\.(word|byte|half|skip|uahalf|uaword)'; + $T_DOT_WORD = '\.(long|word|byte|half|skip|uahalf|uaword)'; $T_DOT_GLOBAL = '^\t\.global'; $T_HDR_literal = "\.text\n\t\.align 8\n"; $T_HDR_misc = "\.text\n\t\.align 4\n"; @@ -320,7 +339,6 @@ sub init_TARGET_STUFF { $T_HDR_srt = "\.data\n\t\.align 4\n"; $T_HDR_info = "\.text\n\t\.align 4\n"; $T_HDR_entry = "\.text\n\t\.align 4\n"; - $T_HDR_fast = "\.text\n\t\.align 4\n"; $T_HDR_vector = "\.text\n\t\.align 4\n"; $T_HDR_direct = "\.text\n\t\.align 4\n"; $T_create_word = "\t.word"; @@ -348,7 +366,6 @@ sub init_TARGET_STUFF { $T_HDR_srt = "\.data\n\t\.align 4\n"; $T_HDR_info = "\.text\n\t\.align 4\n"; $T_HDR_entry = "\.text\n\t\.align 4\n"; - $T_HDR_fast = "\.text\n\t\.align 4\n"; $T_HDR_vector = "\.text\n\t\.align 4\n"; $T_HDR_direct = "\.text\n\t\.align 4\n"; $T_create_word = "\t.word"; @@ -381,7 +398,6 @@ print STDERR "T_HDR_consist: $T_HDR_consist\n"; print STDERR "T_HDR_closure: $T_HDR_closure\n"; print STDERR "T_HDR_info: $T_HDR_info\n"; print STDERR "T_HDR_entry: $T_HDR_entry\n"; -print STDERR "T_HDR_fast: $T_HDR_fast\n"; print STDERR "T_HDR_vector: $T_HDR_vector\n"; print STDERR "T_HDR_direct: $T_HDR_direct\n"; } @@ -419,8 +435,7 @@ sub mangle_asm { $numchks = 0; # number of them @chkcat = (); # what category of thing in each chunk @chksymb = (); # what symbol(base) is defined in this chunk - %slowchk = (); # ditto, its regular "slow" entry code - %fastchk = (); # ditto, fast entry code + %entrychk = (); # ditto, its entry code %closurechk = (); # ditto, the (static) closure %srtchk = (); # ditto, its SRT (for top-level things) %infochk = (); # given a symbol base, say what chunk its info tbl is in @@ -437,17 +452,22 @@ sub mangle_asm { next if /${T_PRE_APP}(NO_)?APP/o; next if /^;/ && $TargetPlatform =~ /^hppa/; - next if /(^$|^\t\.file\t|^ # )/ && $TargetPlatform =~ /^(mips)-/; + next if /(^$|^\t\.file\t|^ # )/ && $TargetPlatform =~ /^(mips|ia64)-/; if ( $TargetPlatform =~ /^mips-/ && /^\t\.(globl\S+\.text|comm\t)/ ) { $EXTERN_DECLS .= $_ unless /(__DISCARD__|\b(PK_|ASSIGN_)(FLT|DBL)\b)/; - - # As a temporary solution for compiling "foreign export" declarations, - # we use global variables to pass arguments from C to STG land. - # These declarations live in the .hc file and not in the generated C - # stub file, so we let them pass through here. - } elsif ( /^\t\.comm\t__fexp_.*$/ ) { + # Treat .comm variables as data. These show up in two (known) places: + # + # - the module_registered variable used in the __stginit fragment. + # even though these are declared static and initialised, gcc 3.3 + # likes to make them .comm, presumably to save space in the + # object file. + # + # - global variables used to pass arguments from C to STG in + # a foreign export. (is this still true? --SDM) + # + } elsif ( /^\t\.comm.*$/ ) { $chk[++$i] = $_; $chkcat[$i] = 'data'; $chksymb[$i] = ''; @@ -489,17 +509,10 @@ sub mangle_asm { } elsif ( /^${T_US}([A-Za-z0-9_]+)_(entry|ret)${T_POST_LBL}$/o ) { $chk[++$i] = $_; - $chkcat[$i] = 'slow'; + $chkcat[$i] = 'entry'; $chksymb[$i] = $1; - $slowchk{$1} = $i; - - } elsif ( /^${T_US}([A-Za-z0-9_]+)_fast\d*${T_POST_LBL}$/o ) { - $chk[++$i] = $_; - $chkcat[$i] = 'fast'; - $chksymb[$i] = $1; - - $fastchk{$1} = $i; + $entrychk{$1} = $i; } elsif ( /^${T_US}([A-Za-z0-9_]+)_closure${T_POST_LBL}$/o ) { $chk[++$i] = $_; @@ -520,10 +533,6 @@ sub mangle_asm { $chkcat[$i] = 'data'; $chksymb[$i] = ''; - } elsif ( /^${T_US}ghc.*c_ID${T_POST_LBL}/o ) { - $chk[++$i] = $_; - $chkcat[$i] = 'consist'; - } elsif ( /^(${T_US}__gnu_compiled_c|gcc2_compiled\.)${T_POST_LBL}/o ) { ; # toss it @@ -558,15 +567,6 @@ sub mangle_asm { $vectorchk{$1} = $i; - # As a temporary solution for compiling "foreign export" declarations, - # we use global variables to pass arguments from C to STG land. - # These declarations live in the .hc file and not in the generated C - # stub file, so we let them pass through here. - } elsif ( /^[\t ]+\.comm[\t ]+__fexp_.*$/ ) { - $chk[++$i] = $_; - $chkcat[$i] = 'data'; - $chksymb[$i] = ''; - } elsif ( $TargetPlatform =~ /^i386-.*-solaris2/ && /^[A-Za-z0-9][A-Za-z0-9_]*:/ ) { # Some Solaris system headers contain function definitions (as @@ -585,19 +585,48 @@ sub mangle_asm { || ! /^L\$\d+$/ ) ) { local($thing); chop($thing = $_); - print STDERR "Funny global thing?: $_" + $thing =~ s/:$//; + print STDERR "Warning: retaining unknown function \`$thing' in output from C compiler\n" unless # $KNOWN_FUNNY_THING{$thing} /^${T_US}stg_.*${T_POST_LBL}$/o # RTS internals || /^${T_US}__stg_.*${T_POST_LBL}$/o # more RTS internals || /^${T_US}__fexp_.*${T_POST_LBL}$/o # foreign export + || /^${T_US}.*_slow${T_POST_LBL}$/o # slow entry || /^${T_US}__stginit.*${T_POST_LBL}$/o # __stginit || /^${T_US}.*_btm${T_POST_LBL}$/o # large bitmaps + || /^${T_US}.*_srtd${T_POST_LBL}$/o # large bitmaps + || /^${T_US}.*_fast${T_POST_LBL}$/o # primops || /^${T_US}.*_closure_tbl${T_POST_LBL}$/o # closure tables || /^_uname:/o; # x86/Solaris2 $chk[++$i] = $_; $chkcat[$i] = 'misc'; $chksymb[$i] = ''; + } elsif ( $TargetPlatform =~ /^powerpc-apple-.*/ && /^\.picsymbol_stub/ ) + { + $chk[++$i] = $_; + $chkcat[$i] = 'dyld'; + $chksymb[$i] = ''; + } elsif ( $TargetPlatform =~ /^powerpc-apple-.*/ && /^\.symbol_stub/ ) + { + $chk[++$i] = $_; + $chkcat[$i] = 'dyld'; + $chksymb[$i] = ''; + } elsif ( $TargetPlatform =~ /^powerpc-apple-.*/ && /^\.lazy_symbol_pointer/ ) + { + $chk[++$i] = $_; + $chkcat[$i] = 'dyld'; + $chksymb[$i] = ''; + } elsif ( $TargetPlatform =~ /^powerpc-apple-.*/ && /^\.non_lazy_symbol_pointer/ ) + { + $chk[++$i] = $_; + $chkcat[$i] = 'dyld'; + $chksymb[$i] = ''; + } elsif ( $TargetPlatform =~ /^powerpc-apple-.*/ && /^\.data/ && $chkcat[$i] eq 'dyld') + { # non_lazy_symbol_ptrs that point to local symbols + $chk[++$i] = $_; + $chkcat[$i] = 'dyld'; + $chksymb[$i] = ''; } else { # simple line (duplicated at the top) $chk[$i] .= $_; @@ -605,11 +634,9 @@ sub mangle_asm { } $numchks = $#chk + 1; - open CHUNKS, ">/tmp/chunks1" or die "Cannot open /tmp/chunks1: $!\n"; - for (my $i = 0; $i < @chk; ++$i) { print CHUNKS "======= $i =======\n", $chk[$i] } - close CHUNKS; - - # ######## + # open CHUNKS, ">/tmp/chunks1" or die "Cannot open /tmp/chunks1: $!\n"; + # for (my $i = 0; $i < @chk; ++$i) { print CHUNKS "======= $i =======\n", $chk[$i] } + # close CHUNKS; # the division into chunks is imperfect; # we throw some things over the fence into the next @@ -639,19 +666,49 @@ sub mangle_asm { # toss all prologue stuff; HPPA is pretty weird # (see elsewhere) - $c = &mash_hppa_prologue($c) if $TargetPlatform =~ /^hppa/; + $c = &hppa_mash_prologue($c) if $TargetPlatform =~ /^hppa-/; # be slightly paranoid to make sure there's # nothing surprising in there if ( $c =~ /--- BEGIN ---/ ) { if (($p, $r) = split(/--- BEGIN ---/, $c)) { + # remove junk whitespace around the split point + $p =~ s/\t+$//; + $r =~ s/^\s*\n//; + if ($TargetPlatform =~ /^i386-/) { $p =~ s/^\tpushl\s+\%edi\n//; $p =~ s/^\tpushl\s+\%esi\n//; $p =~ s/^\tpushl\s+\%ebx\n//; + $p =~ s/^\tmovl\s+\%esi,\s*\d*\(\%esp\)\n//; + $p =~ s/^\tmovl\s+\%edi,\s*\d*\(\%esp\)\n//; $p =~ s/^\tsubl\s+\$\d+,\s*\%esp\n//; $p =~ s/^\tmovl\s+\$\d+,\s*\%eax\n\tcall\s+__alloca\n// if ($TargetPlatform =~ /^.*-cygwin32/); + + # GCC 3.1 is in the habit of adding spurious writes to the + # stack in the prologue. Just to be on the safe side, + # chuck these over the fence into the main code. + while ($p =~ /^\tmovl\s+\$\d+,\s*\d*\(\%esp\)\n/) { + # print "Spurious instruction: $&"; + $p = $` . $'; + $r = $& . $r; + } + + } elsif ($TargetPlatform =~ /^ia64-/) { + $p =~ s/^\t\.prologue .*\n//; + $p =~ s/^\t\.save ar\.pfs, r\d+\n\talloc r\d+ = ar\.pfs, 0, 3[12], \d+, 0\n//; + $p =~ s/^\t\.fframe \d+\n\tadds r12 = -\d+, r12\n//; + $p =~ s/^\t\.save rp, r\d+\n\tmov r\d+ = b0\n//; + $p =~ s/^\t\.(mii|mmi)\n//g; # bundling is no longer sensible + $p =~ s/^\t;;\n//g; # discard stops + $p =~ s/^\t\/\/.*\n//g; # gcc inserts timings in // comments + + # GCC 3.3 saves r1 in the prologue, move this to the body + if ($p =~ /^\tmov r\d+ = r1\n/) { + $p = $` . $'; + $r = $& . $r; + } } elsif ($TargetPlatform =~ /^m68k-/) { $p =~ s/^\tlink a6,#-?\d.*\n//; $p =~ s/^\tpea a6@\n\tmovel sp,a6\n//; @@ -676,33 +733,45 @@ sub mangle_asm { $p =~ s/^\tsw\t\$28,\d+\(\$sp\)\n//; $p =~ s/__FRAME__/$FRAME/; } elsif ($TargetPlatform =~ /^powerpc-apple-.*/) { - $p =~ s/^\tmflr r0\n//; + $pcrel_label = $p; + $pcrel_label =~ s/(.|\n)*^(L\d+\$pb):\n(.|\n)*/$2/ or $pcrel_label = ""; + + $p =~ s/^\tmflr r0\n//; $p =~ s/^\tbl saveFP # f\d+\n//; + $p =~ s/^\tbl saveFP ; save f\d+-f\d+\n//; + $p =~ s/^L\d+\$pb:\n//; $p =~ s/^\tstmw r\d+,-\d+\(r1\)\n//; $p =~ s/^\tstfd f\d+,-\d+\(r1\)\n//g; $p =~ s/^\tstw r0,\d+\(r1\)\n//g; $p =~ s/^\tstwu r1,-\d+\(r1\)\n//; $p =~ s/^\tstw r\d+,-\d+\(r1\)\n//g; + $p =~ s/^\tbcl 20,31,L\d+\$pb\n//; + $p =~ s/^L\d+\$pb:\n//; + $p =~ s/^\tmflr r31\n//; + + # This is bad: GCC 3 seems to zero-fill some local variables in the prologue + # under some circumstances, only when generating position dependent code. + # I have no idea why, and I don't think it is necessary, so let's toss it. + $p =~ s/^\tli r\d+,0\n//g; + $p =~ s/^\tstw r\d+,\d+\(r1\)\n//g; } else { print STDERR "$Pgm: unknown prologue mangling? $TargetPlatform\n"; } - - if ($TargetPlatform =~ /^powerpc-apple-.*/) { - # on PowerPC, we have to keep a part of the prologue - # (which loads the current instruction pointer into register r31) - $u_p = $p; # $u_p is for unexpected prologue - $u_p =~ s/^\tbcl 20,31,L\d+\$pb\n//; - $u_p =~ s/^L\d+\$pb:\n//; - $u_p =~ s/^\tmflr r31\n//; - die "Prologue junk?: $u_p\n" if $u_p =~ /^\t[^\.]/ - } else { - # HWL HACK: dont die, just print a warning - #print stderr "HWL: this should die! Prologue junk?: $p\n" if $p =~ /^\t[^\.]/; - die "Prologue junk?: $p\n" if $p =~ /^\t[^\.]/; + + # HWL HACK: dont die, just print a warning + #print stderr "HWL: this should die! Prologue junk?: $p\n" if $p =~ /^\t[^\.]/; + die "Prologue junk?: $p\n" if $p =~ /^\s+[^\s\.]/; + + if ($TargetPlatform =~ /^powerpc-apple-.*/ && $pcrel_label ne "") { + # on PowerPC, we have to keep a part of the prologue + # (which loads the current instruction pointer into register r31) + $p .= "bcl 20,31,$pcrel_label\n"; + $p .= "$pcrel_label:\n"; + $p .= "\tmflr r31\n"; } + # glue together what's left $c = $p . $r; - $c =~ s/\n\t\n/\n/; # junk blank line } } @@ -722,6 +791,14 @@ sub mangle_asm { $e =~ s/^\tpopl\s+\%ecx\n//; $e =~ s/^\taddl\s+\$\d+,\s*\%esp\n//; $e =~ s/^\tsubl\s+\$-\d+,\s*\%esp\n//; + } elsif ($TargetPlatform =~ /^ia64-/) { + $e =~ s/^\tmov ar\.pfs = r\d+\n//; + $e =~ s/^\tmov b0 = r\d+\n//; + $e =~ s/^\t\.restore sp\n\tadds r12 = \d+, r12\n//; + $e =~ s/^\tbr\.ret\.sptk\.many b0\n//; + $e =~ s/^\t\.(mii|mmi|mib)\n//g; # bundling is no longer sensible + $e =~ s/^\t;;\n//g; # discard stops - stop at end of body is sufficient + $e =~ s/^\t\/\/.*\n//g; # gcc inserts timings in // comments } elsif ($TargetPlatform =~ /^m68k-/) { $e =~ s/^\tunlk a6\n//; $e =~ s/^\trts\n//; @@ -752,7 +829,7 @@ sub mangle_asm { # toss the register-windowing save/restore/ret* instructions # directly: if ( $TargetPlatform =~ /^sparc-/ ) { - $c =~ s/^\t(save.*|restore|ret|retl)\n//g; + $c =~ s/^\t(save.*|restore.*|ret|retl)\n//g; # throw away PROLOGUE comments $c =~ s/^\t!#PROLOGUE# 0\n\t!#PROLOGUE# 1\n//; } @@ -761,7 +838,17 @@ sub mangle_asm { # toss all calls to __DISCARD__ $c =~ s/^\t(call|jbsr|jal)\s+${T_US}__DISCARD__\n//go; -# ###WTT### $c =~ s/^\tbl\s+L___DISCARD__\$stub\n//go if $TargetPlatform =~ /^powerpc-apple-.*/; + $c =~ s/^\tjsr\s+\$26\s*,\s*${T_US}__DISCARD__\n//go if $TargetPlatform =~ /^alpha-/; + $c =~ s/^\tbl\s+L___DISCARD__\$stub\n//go if $TargetPlatform =~ /^powerpc-apple-.*/; + + # IA64: mangle tailcalls into jumps here + if ($TargetPlatform =~ /^ia64-/) { + while ($c =~ s/^\tbr\.call\.sptk\.many b0 = (.*)\n(?:^\.L([0-9]*):\n)?(?:\t;;\n)?(?:\tmov r1 = r\d+\n)?(?:\t;;\n)?\t--- TAILCALL ---\n(?:\t;;\n\tbr \.L\d+\n)?/\tbr\.few $1\n/) { + # Eek, the gcc optimiser is getting smarter... if we see a jump to the --- TAILCALL --- + # marker then we reapply the substitution at the source sites + $c =~ s/^\tbr \.L$2\n/\t--- TAILCALL ---\n/g if ($2); + } + } # MIPS: that may leave some gratuitous asm macros around # (no harm done; but we get rid of them to be tidier) @@ -798,7 +885,7 @@ sub mangle_asm { # blah_closure: # ... # - if ( $TargetPlatform =~ /^(i386|sparc)/ && $to_move =~ /${T_COPY_DIRVS}/ ) { + if ( $TargetPlatform =~ /^(i386|sparc|powerpc)/ && $to_move =~ /${T_COPY_DIRVS}/ ) { $j = $i + 1; while ( $j < $numchks && $chk[$j] =~ /$T_CONST_LBL/) { $j++; @@ -821,16 +908,18 @@ sub mangle_asm { if ( $TargetPlatform =~ /^alpha-/ && $c =~ /^\t\.ent\s+(\S+)/ ) { $ent = $1; # toss all prologue stuff, except for loading gp, and the ..ng address - if (($p, $r) = split(/^\t\.prologue/, $c)) { - if (($keep, $junk) = split(/\.\.ng:/, $p)) { - $keep =~ s/^\t\.frame.*\n/\t.frame \$30,0,\$26,0\n/; - $keep =~ s/^\t\.(mask|fmask).*\n//g; - $c = $keep . "..ng:\n"; - } else { - print STDERR "malformed code block ($ent)?\n" - } - } - $c .= "\t.prologue" . $r; + unless ($c =~ /\.ent.*\n\$.*\.\.ng:/) { + if (($p, $r) = split(/^\t\.prologue/, $c)) { + if (($keep, $junk) = split(/\.\.ng:/, $p)) { + $keep =~ s/^\t\.frame.*\n/\t.frame \$30,0,\$26,0\n/; + $keep =~ s/^\t\.(mask|fmask).*\n//g; + $c = $keep . "..ng:\n"; + } else { + print STDERR "malformed code block ($ent)?\n" + } + } + $c .= "\t.prologue" . $r; + } } $c =~ s/FUNNY#END#THING//; @@ -962,8 +1051,7 @@ sub mangle_asm { } elsif ( $chkcat[$i] eq 'closure' || $chkcat[$i] eq 'srt' || $chkcat[$i] eq 'infotbl' - || $chkcat[$i] eq 'slow' - || $chkcat[$i] eq 'fast' ) { # do them in that order + || $chkcat[$i] eq 'entry') { # do them in that order $symb = $chksymb[$i]; # CLOSURE @@ -992,74 +1080,17 @@ sub mangle_asm { } # STD ENTRY POINT - if ( defined($slowchk{$symb}) ) { - - # teach it to drop through to the fast entry point: - $c = $chk[$slowchk{$symb}]; - - if ( defined($fastchk{$symb}) ) { - if ( $TargetPlatform =~ /^alpha-/ ) { - $c =~ s/^\tjmp \$31,\(\$27\),0\n\t\.align 4\n\t\.end/\t.align 4\n\t.end/; - } elsif ( $TargetPlatform =~ /^hppa/ ) { - $c =~ s/^\s+ldil.*\n\s+ldo.*\n\s+bv.*\n(.*\n)?\s+\.EXIT/$1\t.EXIT/; - } elsif ( $TargetPlatform =~ /^i386-/ ) { - # Reg alloc depending, gcc generated code may jump to the fast entry point via - # a number of registers. - $c =~ s/^\tmovl\s+\$${T_US}${symb}_fast\d*,\s*\%edx\n\tjmp\s+\*\%edx\n//; - $c =~ s/^\tmovl\s+\$${T_US}${symb}_fast\d*,\s*\%ecx\n\tjmp\s+\*\%ecx\n//; - $c =~ s/^\tmovl\s+\$${T_US}${symb}_fast\d*,\s*\%eax\n\tjmp\s+\*\%eax\n//; - # The next two only apply if we're not stealing %esi or %edi. - $c =~ s/^\tmovl\s+\$${T_US}${symb}_fast\d*,\s*\%esi\n\tjmp\s+\*\%esi\n// if ($StolenX86Regs < 3); - $c =~ s/^\tmovl\s+\$${T_US}${symb}_fast\d*,\s*\%edi\n\tjmp\s+\*\%edi\n// if ($StolenX86Regs < 4); - } elsif ( $TargetPlatform =~ /^mips-/ ) { - $c =~ s/^\tjmp \$31,\(\$27\),0\n\t\.align 4\n\t\.end/\t.align 4\n\t.end/; - } elsif ( $TargetPlatform =~ /^m68k-/ ) { - $c =~ s/^\tjmp ${T_US}${symb}_fast\d+.*\n\tnop\n//; - $c =~ s/^\tjmp ${T_US}${symb}_fast\d+.*\n//; - } elsif ( $TargetPlatform =~ /^powerpc-apple-.*/ ) { - $c =~ s/^\taddis r9,r31,ha16\(L_${symb}_fast\d*\$non_lazy_ptr-L\d+\$pb\)\n\tlwz r9,lo16\(L_${symb}_fast\d*\$non_lazy_ptr-L\d+\$pb\)\(r9\)\n\tmtctr r9\n\tbctr\n//; - # addis r9,r31,ha16(L_XXXX_fast3$non_lazy_ptr-L32$pb) - # lwz r9,lo16(L_XXXX_fast3$non_lazy_ptr-L32$pb)(r9) - # mtctr r9 - # bctr - } elsif ( $TargetPlatform =~ /^sparc-/ ) { - $c =~ s/^\tcall\s+${T_US}${symb}_fast\d+,.*\n\t\s*nop\n//; - $c =~ s/^\tcall\s+${T_US}${symb}_fast\d+,.*\n(\t\s*[a-z].*\n)/$1/; - } else { - print STDERR "$Pgm: mystery slow-fast dropthrough: $TargetPlatform\n"; - } - } + if ( defined($entrychk{$symb}) ) { - if ( $TargetPlatform !~ /^(alpha-|hppa|mips-|powerpc-apple-)/) { - # On alphas, hppa: no very good way to look for "dangling" - # references to fast-entry point. - # (questionable re hppa and mips...) - print STDERR "still has jump to fast entry point:\n$c" - if $c =~ /\b${T_US}${symb}_fast/; - } + $c = $chk[$entrychk{$symb}]; print OUTASM $T_HDR_entry; &print_doctored($c, 1); # NB: the 1!!! - $chkcat[$slowchk{$symb}] = 'DONE ALREADY'; + $chkcat[$entrychk{$symb}] = 'DONE ALREADY'; } - # FAST ENTRY POINT - if ( defined($fastchk{$symb}) ) { - $c = $chk[$fastchk{$symb}]; - if ( ! defined($slowchk{$symb}) - # ToDo: the || clause can go once we're no longer - # concerned about producing exactly the same output as before -#OLD: || $TargetPlatform =~ /^(m68k|sparc|i386)-/ - ) { - print OUTASM $T_HDR_fast; - } - - &print_doctored($c, 0); - $chkcat[$fastchk{$symb}] = 'DONE ALREADY'; - } - } elsif ( $chkcat[$i] eq 'vector' || $chkcat[$i] eq 'direct' ) { # do them in that order $symb = $chksymb[$i]; @@ -1067,27 +1098,7 @@ sub mangle_asm { # VECTOR TABLE if ( defined($vectorchk{$symb}) ) { print OUTASM $T_HDR_vector; - print OUTASM &rev_tbl($symb, $chk[$vectorchk{$symb}], 0); - # DO NOT DELETE THE NEXT LINE. It fixes a rather subtle GC bug - # which showed up as a segfault reported by Ryszard Kubiak. - # Problem is with vector tables. They wind up as follows: - # .word some-word - # .word some-other-word - # fooble_vtbl: - # Problem is that we want the label fooble_vtbl to be considered - # in the same section as the vtbl itself, but the label actually - # lives at the next word along. If a data segment should happen - # to immediately follow the vtbl, as it can in GHCi, the label will - # be malclassified as in the data rather than text segment (during - # GC), and so we will regard references to it as static closure - # pointers rather than as code pointers, which is an error which - # usually crashes the garbage collectors. - # To fix this, we place a dummy word after the label, so as to - # ensure that the label is in the same segment as the vtbl proper. - # The native code generator has an analogous fix; see - # ghc/compiler/nativeGen/AbsCStixGen.lhs line 107. - print OUTASM "${T_create_word} 0\n"; # direct return code will be put here! $chkcat[$vectorchk{$symb}] = 'DONE ALREADY'; @@ -1116,6 +1127,12 @@ sub mangle_asm { $j++; } + } elsif ( $TargetPlatform =~ /^powerpc-apple-.*/ && $chkcat[$i] eq 'dyld' ) { + # powerpc-apple: dynamic linker stubs + if($chk[$i] !~ /\.indirect_symbol ___DISCARD__/) + { # print them out unchanged, but remove the stubs for __DISCARD__ + print OUTASM $chk[$i]; + } } else { &tidy_up_and_die(1,"$Pgm: unknown chkcat (ghc-asm: $TargetPlatform)\n$chkcat[$i]\n$chk[$i]\n"); } @@ -1130,7 +1147,7 @@ sub mangle_asm { \end{code} \begin{code} -sub mash_hppa_prologue { # OK, epilogue, too +sub hppa_mash_prologue { # OK, epilogue, too local($_) = @_; # toss all prologue stuff @@ -1203,7 +1220,7 @@ sub print_doctored { # jmp * # - s/^\tmovl\s+\$${T_US}(.*),(\%e[abcd]x)\n\tjmp\s+\*\2/\tjmp $T_US$1/g; + s/^\tmovl\s+\$${T_US}(.*),\s*(\%e[abcd]x)\n\tjmp\s+\*\2/\tjmp $T_US$1/g; if ($StolenX86Regs <= 2 ) { # YURGH! spurious uses of esi? s/^\tmovl\s+(.*),\s*\%esi\n\tjmp\s+\*%esi\n/\tmovl $1,\%eax\n\tjmp \*\%eax\n/g; @@ -1226,7 +1243,7 @@ sub print_doctored { # Offsets into register table - you'd better update these magic # numbers should you change its contents! # local($OFFSET_R1)=0; No offset for R1 in new RTS. - local($OFFSET_Hp)=92; + local($OFFSET_Hp)=88; # Note funky ".=" stuff; we're *adding* to these _patch guys if ( $StolenX86Regs <= 2 @@ -1303,6 +1320,8 @@ right after the table itself. (The code pasting is done elsewhere.) sub rev_tbl { local($symb, $tbl, $discard1) = @_; + return ($tbl) if ($TargetPlatform =~ /^ia64-/); + local($before) = ''; local($label) = ''; local(@imports) = (); # hppa only @@ -1339,13 +1358,18 @@ sub rev_tbl { } } - # now throw away the first word (SRT) iff it is empty. + # Now throw away any initial zero word from the table. This is a hack + # that lets us reduce the size of info tables when the SRT field is not + # needed: see comments StgFunInfoTable in InfoTables.h. + # # The .zero business is for Linux/ELF. # The .skip business is for Sparc/Solaris/ELF. # The .blockz business is for HPPA. - if ($discard1 && $words[0] =~ /^\t?(${T_DOT_WORD}\s+0|\.zero\s+4|\.skip\s+4|\.blockz\s+4)/) { - shift(@words) - } +# if ($discard1) { +# if ($words[0] =~ /^\t?(${T_DOT_WORD}\s+0|\.zero\s+4|\.skip\s+4|\.blockz\s+4)/) { +# shift(@words); +# } +# } for (; $i <= $#lines; $i++) { $after .= $lines[$i] . "\n";