--- /dev/null
+%************************************************************************
+%* *
+\section[Driver-asm-fiddling]{Fiddling with assembler files}
+%* *
+%************************************************************************
+
+Tasks:
+\begin{itemize}
+\item
+Utterly stomp out C functions' prologues and epilogues; i.e., the
+stuff to do with the C stack.
+\item
+Any other required tidying up.
+\end{itemize}
+
+HPPA specific notes:
+\begin{itemize}
+\item
+The HP linker is very picky about symbols being in the appropriate
+space (code vs. data). When we mangle the threaded code to put the
+info tables just prior to the code, they wind up in code space
+rather than data space. This means that references to *_info from
+un-mangled parts of the RTS (e.g. unthreaded GC code) get
+unresolved symbols. Solution: mini-mangler for .c files on HP. I
+think this should really be triggered in the driver by a new -rts
+option, so that user code doesn't get mangled inappropriately.
+\item
+With reversed tables, jumps are to the _info label rather than to
+the _entry label. The _info label is just an address in code
+space, rather than an entry point with the descriptive blob we
+talked about yesterday. As a result, you can't use the call-style
+JMP_ macro. However, some JMP_ macros take _info labels as targets
+and some take code entry points within the RTS. The latter won't
+work with the goto-style JMP_ macro. Sigh. Solution: Use the goto
+style JMP_ macro, and mangle some more assembly, changing all
+"RP'literal" and "LP'literal" references to "R'literal" and
+"L'literal," so that you get the real address of the code, rather
+than the descriptive blob. Also change all ".word P%literal"
+entries in info tables and vector tables to just ".word literal,"
+for the same reason. Advantage: No more ridiculous call sequences.
+\end{itemize}
+
+%************************************************************************
+%* *
+\subsection{Constants for various architectures}
+%* *
+%************************************************************************
+
+\begin{code}
+sub init_TARGET_STUFF {
+
+ #--------------------------------------------------------#
+ if ( $TargetPlatform =~ /^alpha-.*-.*/ ) {
+
+ $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
+ $T_US = ''; # _ if symbols have an underscore on the front
+ $T_DO_GC = 'PerformGC_wrapper';
+ $T_PRE_APP = 'DONT THINK THIS APPLIES'; # regexp that says what comes before APP/NO_APP
+ $T_CONST_LBL = '^\$C(\d+):$'; # 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_hsc_cc_PAT = '\.ascii.*\)(hsc|cc) (.*)\\\\11"\n\t\.ascii\s+"(.*)\\\\0"';
+ $T_DOT_WORD = '\.quad';
+ $T_DOT_GLOBAL = '^\t\.globl';
+ $T_HDR_literal = "\.rdata\n\t\.align 3\n";
+ $T_HDR_misc = "\.text\n\t\.align 3\n";
+ $T_HDR_data = "\.data\n\t\.align 3\n";
+ $T_HDR_consist = "\.text\n";
+ $T_HDR_closure = "\.data\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";
+
+ #--------------------------------------------------------#
+ } elsif ( $TargetPlatform =~ /^hppa/ ) {
+
+ $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
+ $T_US = ''; # _ if symbols have an underscore on the front
+ $T_DO_GC = 'PerformGC_wrapper';
+ $T_PRE_APP = 'DONT THINK THIS APPLIES'; # regexp that says what comes before APP/NO_APP
+ $T_CONST_LBL = '^L\$C(\d+)$'; # regexp for what such a lbl looks like
+ $T_POST_LBL = '';
+
+ $T_MOVE_DIRVS = '^((\s+\.(IMPORT|EXPORT|PARAM).*|\s+\.align\s+\d+|\s+\.(SPACE|SUBSPA)\s+\S+|\s*)\n)';
+ $T_COPY_DIRVS = '^\s+\.(IMPORT|EXPORT)';
+
+ $T_hsc_cc_PAT = '\.STRING.*\)(hsc|cc) (.*)\\\\x09(.*)\\\\x00';
+ $T_DOT_WORD = '\.word';
+ $T_DOT_GLOBAL = '^\s+\.EXPORT';
+ $T_HDR_literal = "\t.SPACE \$TEXT\$\n\t.SUBSPA \$LIT\$\n";
+ $T_HDR_misc = "\t.SPACE \$TEXT\$\n\t.SUBSPA \$CODE\$\n\t\.align 4\n";
+ $T_HDR_data = "\t.SPACE \$PRIVATE\$\n\t.SUBSPA \$DATA\$\n\t\.align 4\n";
+ $T_HDR_consist = "\t.SPACE \$TEXT\$\n\t.SUBSPA \$LIT\$\n";
+ $T_HDR_closure = "\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";
+
+ #--------------------------------------------------------#
+ } elsif ( $TargetPlatform =~ /^i386-.*-(linuxaout|freebsd|nextstep3|cygwin32)/ ) {
+ # NeXT added but not tested. CaS
+
+ $T_STABBY = 1; # 1 iff .stab things (usually if a.out format)
+ $T_US = '_'; # _ if symbols have an underscore on the front
+ $T_DO_GC = '_PerformGC_wrapper';
+ $T_PRE_APP = '^#'; # regexp that says what comes before APP/NO_APP
+ $T_CONST_LBL = '^LC(\d+):$';
+ $T_POST_LBL = ':';
+ $T_X86_PRE_LLBL_PAT = 'L';
+ $T_X86_PRE_LLBL = 'L';
+ $T_X86_BADJMP = '^\tjmp [^L\*]';
+
+ $T_MOVE_DIRVS = '^(\s*(\.align\s+\d+(,0x90)?|\.globl\s+\S+|\.text|\.data|\.stab[^n].*|\.type\s+.*|\.size\s+.*)\n)';
+ $T_COPY_DIRVS = '\.(globl|stab)';
+ $T_hsc_cc_PAT = '\.ascii.*\)(hsc|cc) (.*)\\\\11"\n\t\.ascii\s+"(.*)\\\\0"';
+ $T_DOT_WORD = '\.long';
+ $T_DOT_GLOBAL = '\.globl';
+ $T_HDR_literal = "\.text\n\t\.align 2\n";
+ $T_HDR_misc = "\.text\n\t\.align 2,0x90\n";
+ $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_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";
+
+ #--------------------------------------------------------#
+ } elsif ( $TargetPlatform =~ /^i386-.*-(solaris2|linux)$/ ) {
+
+ $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
+ $T_US = ''; # _ if symbols have an underscore on the front
+ $T_DO_GC = 'PerformGC_wrapper';
+ $T_PRE_APP = # regexp that says what comes before APP/NO_APP
+ ($TargetPlatform =~ /-linux$/) ? '#' : '/' ;
+ $T_CONST_LBL = '^\.LC(\d+):$'; # regexp for what such a lbl looks like
+ $T_POST_LBL = ':';
+ $T_X86_PRE_LLBL_PAT = '\.L';
+ $T_X86_PRE_LLBL = '.L';
+ $T_X86_BADJMP = '^\tjmp [^\.\*]';
+
+ $T_MOVE_DIRVS = '^(\s*(\.align\s+\d+(,0x90)?|\.globl\s+\S+|\.text|\.data|\.section\s+.*|\.type\s+.*|\.Lfe.*\n\t\.size\s+.*|\.size\s+.*|\.ident.*)\n)';
+ $T_COPY_DIRVS = '\.(globl)';
+
+ $T_hsc_cc_PAT = '\.string.*\)(hsc|cc) (.*)\\\\t(.*)"';
+ $T_DOT_WORD = '\.long';
+ $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 16\n";
+ $T_HDR_data = "\.data\n\t\.align 4\n"; # ToDo: change align??
+ $T_HDR_consist = "\.text\n";
+ $T_HDR_closure = "\.data\n\t\.align 4\n"; # ToDo: change align?
+ $T_HDR_info = "\.text\n\t\.align 16\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 16\n";
+ $T_HDR_vector = "\.text\n\t\.align 16\n"; # NB: requires padding
+ $T_HDR_direct = "\.text\n\t\.align 16\n";
+
+ #--------------------------------------------------------#
+ } elsif ( $TargetPlatform =~ /^m68k-.*-sunos4/ ) {
+
+ $T_STABBY = 1; # 1 iff .stab things (usually if a.out format)
+ $T_US = '_'; # _ if symbols have an underscore on the front
+ $T_DO_GC = '_PerformGC_wrapper';
+ $T_PRE_APP = '^# MAY NOT APPLY'; # regexp that says what comes before APP/NO_APP
+ $T_CONST_LBL = '^LC(\d+):$';
+ $T_POST_LBL = ':';
+
+ $T_MOVE_DIRVS = '^(\s*(\.align\s+\d+|\.proc\s+\d+|\.const|\.cstring|\.globl\s+\S+|\.text|\.data|\.even|\.stab[^n].*)\n)';
+ $T_COPY_DIRVS = '\.(globl|proc|stab)';
+ $T_hsc_cc_PAT = '\.ascii.*\)(hsc|cc) (.*)\\\\11"\n\t\.ascii\s+"(.*)\\\\0"';
+
+ $T_DOT_WORD = '\.long';
+ $T_DOT_GLOBAL = '\.globl';
+ $T_HDR_literal = "\.text\n\t\.even\n";
+ $T_HDR_misc = "\.text\n\t\.even\n";
+ $T_HDR_data = "\.data\n\t\.even\n";
+ $T_HDR_consist = "\.text\n";
+ $T_HDR_closure = "\.data\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";
+
+ #--------------------------------------------------------#
+ } elsif ( $TargetPlatform =~ /^mips-.*/ ) {
+
+ $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
+ $T_US = ''; # _ if symbols have an underscore on the front
+ $T_DO_GC = 'PerformGC_wrapper';
+ $T_PRE_APP = '^\s*#'; # regexp that says what comes before APP/NO_APP
+ $T_CONST_LBL = '^\$LC(\d+):$'; # regexp for what such a lbl looks like
+ $T_POST_LBL = ':';
+
+ $T_MOVE_DIRVS = '^(\s*(\.align\s+\d+|\.(globl|ent)\s+\S+|\.text|\.r?data)\n)';
+ $T_COPY_DIRVS = '\.(globl|ent)';
+
+ $T_hsc_cc_PAT = 'I WAS TOO LAZY TO DO THIS BIT (WDP 95/05)';
+ $T_DOT_WORD = '\.word';
+ $T_DOT_GLOBAL = '^\t\.globl';
+ $T_HDR_literal = "\t\.rdata\n\t\.align 2\n";
+ $T_HDR_misc = "\t\.text\n\t\.align 2\n";
+ $T_HDR_data = "\t\.data\n\t\.align 2\n";
+ $T_HDR_consist = 'TOO LAZY TO DO THIS TOO';
+ $T_HDR_closure = "\t\.data\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";
+
+ #--------------------------------------------------------#
+ } elsif ( $TargetPlatform =~ /^powerpc-.*|^rs6000-.*/ ) {
+
+ $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
+ $T_US = ''; # _ if symbols have an underscore on the front
+ $T_DO_GC = '\.PerformGC_wrapper';
+ $T_PRE_APP = 'NOT APPLICABLE'; # regexp that says what comes before APP/NO_APP
+ $T_CONST_LBL = 'NOT APPLICABLE'; # regexp for what such a lbl looks like
+ $T_POST_LBL = ':';
+
+ $T_MOVE_DIRVS = '^(\s*(\.toc|\.align \d+|\.csect \S+|\t\.?l?globl \S+)\n)';
+ $T_COPY_DIRVS = '\.(l?globl)';
+
+ $T_hsc_cc_PAT = '\.byte.*\)(hsc|cc) (.*)"\n\t\.byte \d+\n\t\.byte "(.*)"\n\t\.byte \d+';
+ $T_DOT_WORD = '\.long';
+ $T_DOT_GLOBAL = '\.globl';
+ $T_HDR_toc = "\.toc\n";
+ $T_HDR_literal = "\.csect .data[RW]\n\t\.align 2\n"; #not RO!?
+ $T_HDR_misc = "# misc\n\.csect \.text[PR]\n\t\.align 2\n";
+ $T_HDR_data = "# data\n\.csect \.data[RW]\n\t\.align 2\n";
+ $T_HDR_consist = "# consist\n\.csect \.data[RW]\n\t\.align 2\n";
+ $T_HDR_closure = "# closure\n\.csect \.data[RW]\n\t\.align 2\n";
+ $T_HDR_info = "# info\n\.csect \.data[RW]\n\t\.align 2\n"; #not RO!?
+ $T_HDR_entry = "# entry\n\.csect \.text[PR]\n\t\.align 2\n";
+ $T_HDR_fast = "# fast\n\.csect \.text[PR]\n\t\.align 2\n";
+ $T_HDR_vector = "# vector\n\.csect \.data[RW]\n\t\.align 2\n"; #not RO!?
+ $T_HDR_direct = "# direct\n";
+
+ #--------------------------------------------------------#
+ } elsif ( $TargetPlatform =~ /^sparc-.*-solaris2/ ) {
+
+ $T_STABBY = 0; # 1 iff .stab things (usually if a.out format)
+ $T_US = ''; # _ if symbols have an underscore on the front
+ $T_DO_GC = 'PerformGC_wrapper';
+ $T_PRE_APP = 'DOES NOT SEEM TO APPLY'; # regexp that says what comes before APP/NO_APP
+ $T_CONST_LBL = '^\.LLC(\d+):$'; # regexp for what such a lbl looks like
+ $T_POST_LBL = ':';
+
+ $T_MOVE_DIRVS = '^((\s+\.align\s+\d+|\s+\.proc\s+\d+|\s+\.global\s+\S+|\.text|\.data|\.stab.*|\.section.*|\s+\.type.*|\s+\.size.*)\n)';
+ $T_COPY_DIRVS = '\.(global|proc|stab)';
+
+ $T_hsc_cc_PAT = '\.asciz.*\)(hsc|cc) (.*)\\\\t(.*)"';
+ $T_DOT_WORD = '\.word';
+ $T_DOT_GLOBAL = '^\t\.global';
+ $T_HDR_literal = "\.text\n\t\.align 8\n";
+ $T_HDR_misc = "\.text\n\t\.align 4\n";
+ $T_HDR_data = "\.data\n\t\.align 8\n";
+ $T_HDR_consist = "\.text\n";
+ $T_HDR_closure = "\.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";
+
+ #--------------------------------------------------------#
+ } elsif ( $TargetPlatform =~ /^sparc-.*-sunos4/ ) {
+
+ $T_STABBY = 1; # 1 iff .stab things (usually if a.out format)
+ $T_US = '_'; # _ if symbols have an underscore on the front
+ $T_DO_GC = '_PerformGC_wrapper';
+ $T_PRE_APP = '^# DOES NOT SEEM TO APPLY'; # regexp that says what comes before APP/NO_APP
+ $T_CONST_LBL = '^LC(\d+):$';
+ $T_POST_LBL = ':';
+
+ $T_MOVE_DIRVS = '^((\s+\.align\s+\d+|\s+\.proc\s+\d+|\s+\.global\s+\S+|\.text|\.data|\.stab.*)\n)';
+ $T_COPY_DIRVS = '\.(global|proc|stab)';
+ $T_hsc_cc_PAT = '\.ascii.*\)(hsc|cc) (.*)\\\\11"\n\t\.ascii\s+"(.*)\\\\0"';
+
+ $T_DOT_WORD = '\.word';
+ $T_DOT_GLOBAL = '^\t\.global';
+ $T_HDR_literal = "\.text\n\t\.align 8\n";
+ $T_HDR_misc = "\.text\n\t\.align 4\n";
+ $T_HDR_data = "\.data\n\t\.align 8\n";
+ $T_HDR_consist = "\.text\n";
+ $T_HDR_closure = "\.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";
+ $T_HDR_direct = "\.text\n\t\.align 4\n";
+
+ #--------------------------------------------------------#
+ } else {
+ print STDERR "$Pgm: don't know how to mangle assembly language for: $TargetPlatform\n";
+ exit 1;
+ }
+
+if ( 0 ) {
+print STDERR "T_STABBY: $T_STABBY\n";
+print STDERR "T_US: $T_US\n";
+print STDERR "T_DO_GC: $T_DO_GC\n";
+print STDERR "T_PRE_APP: $T_PRE_APP\n";
+print STDERR "T_CONST_LBL: $T_CONST_LBL\n";
+print STDERR "T_POST_LBL: $T_POST_LBL\n";
+if ( $TargetPlatform =~ /^i386-/ ) {
+ print STDERR "T_X86_PRE_LLBL_PAT: $T_X86_PRE_LLBL_PAT\n";
+ print STDERR "T_X86_PRE_LLBL: $T_X86_PRE_LLBL\n";
+ print STDERR "T_X86_BADJMP: $T_X86_BADJMP\n";
+}
+print STDERR "T_MOVE_DIRVS: $T_MOVE_DIRVS\n";
+print STDERR "T_COPY_DIRVS: $T_COPY_DIRVS\n";
+print STDERR "T_hsc_cc_PAT: $T_hsc_cc_PAT\n";
+print STDERR "T_DOT_WORD: $T_DOT_WORD\n";
+print STDERR "T_HDR_literal: $T_HDR_literal\n";
+print STDERR "T_HDR_misc: $T_HDR_misc\n";
+print STDERR "T_HDR_data: $T_HDR_data\n";
+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: $T_HDR_direct\n";
+}
+
+}
+\end{code}
+
+%************************************************************************
+%* *
+\subsection{Mangle away}
+%* *
+%************************************************************************
+
+\begin{code}
+sub mangle_asm {
+ local($in_asmf, $out_asmf) = @_;
+
+ # multi-line regexp matching:
+ local($*) = 1;
+ local($i, $c);
+
+
+ &init_TARGET_STUFF();
+ &init_FUNNY_THINGS();
+
+ # perl4 on alphas SEGVs when give ${foo} substitutions in patterns.
+ # To avoid them we declare some locals that allows to avoid using curlies.
+ local($TUS) = ${T_US};
+ local($TPOSTLBL) = ${T_POST_LBL};
+ local($TMOVEDIRVS) = ${T_MOVE_DIRVS};
+ local($TPREAPP) = ${T_PRE_APP};
+ local($TCOPYDIRVS) = ${T_COPY_DIRVS};
+ local($TDOTWORD) = ${T_DOT_WORD};
+
+ open(INASM, "< $in_asmf")
+ || &tidy_up_and_die(1,"$Pgm: failed to open `$in_asmf' (to read)\n");
+ open(OUTASM,"> $out_asmf")
+ || &tidy_up_and_die(1,"$Pgm: failed to open `$out_asmf' (to write)\n");
+
+ # read whole file, divide into "chunks":
+ # record some info about what we've found...
+
+ @chk = (); # contents of the chunk
+ $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
+ %closurechk = (); # ditto, the (static) closure
+ %infochk = (); # given a symbol base, say what chunk its info tbl is in
+ %vectorchk = (); # ditto, return vector table
+ %directchk = (); # ditto, direct return code
+ $EXTERN_DECLS = ''; # .globl <foo> .text (MIPS only)
+
+ $i = 0; $chkcat[0] = 'misc'; $chk[0] = '';
+
+ while (<INASM>) {
+ next if $T_STABBY && /^\.stab.*$TUS[@]?__stg_split_marker/o;
+ next if $T_STABBY && /^\.stab.*ghc.*c_ID/;
+ next if /$TPREAPP(NO_)?APP/o;
+ next if /^;/ && $TargetPlatform =~ /^hppa/;
+
+ next if /(^$|^\t\.file\t|^ # )/ && $TargetPlatform =~ /^(mips|powerpc|rs6000)-/;
+
+ last if /^_section_\.text:$/ && $TargetPlatform =~ /^powerpc-|^rs6000-/;
+
+ if ( $TargetPlatform =~ /^mips-/
+ && /^\t\.(globl \S+ \.text|comm\t)/ ) {
+ $EXTERN_DECLS .= $_ unless /(__DISCARD__|\b(PK_|ASSIGN_)(FLT|DBL)\b)/;
+
+ } elsif ( /^\s+/ ) { # most common case first -- a simple line!
+ # duplicated from the bottom
+
+ $chk[$i] .= $_;
+
+ } elsif ( /\.\.ng:$/ && $TargetPlatform =~ /^alpha-/ ) {
+ # Alphas: Local labels not to be confused with new chunks
+ $chk[$i] .= $_;
+
+ # NB: all the rest start with a non-space
+
+ } elsif ( $TargetPlatform =~ /^mips-/
+ && /^\d+:/ ) { # a funny-looking very-local label
+ $chk[$i] .= $_;
+
+ } elsif ( /$T_CONST_LBL/o ) {
+ $chk[++$i] = $_;
+ $chkcat[$i] = 'literal';
+ $chksymb[$i] = $1;
+
+ } elsif ( /^$TUS[@]?__stg_split_marker(\d+)$TPOSTLBL[@]?$/o ) {
+ $chk[++$i] = $_;
+ $chkcat[$i] = 'splitmarker';
+ $chksymb[$i] = $1;
+
+ } elsif ( /^$TUS[@]?([A-Za-z0-9_]+)_info$TPOSTLBL[@]?$/o ) {
+ $symb = $1;
+ $chk[++$i] = $_;
+ $chkcat[$i] = 'infotbl';
+ $chksymb[$i] = $symb;
+
+ die "Info table already? $symb; $i\n" if defined($infochk{$symb});
+
+ $infochk{$symb} = $i;
+
+ } elsif ( /^$TUS[@]?([A-Za-z0-9_]+)_entry$TPOSTLBL[@]?$/o ) {
+ $chk[++$i] = $_;
+ $chkcat[$i] = 'slow';
+ $chksymb[$i] = $1;
+
+ $slowchk{$1} = $i;
+
+ } elsif ( /^$TUS[@]?([A-Za-z0-9_]+)_fast\d+$TPOSTLBL[@]?$/o ) {
+ $chk[++$i] = $_;
+ $chkcat[$i] = 'fast';
+ $chksymb[$i] = $1;
+
+ $fastchk{$1} = $i;
+
+ } elsif ( /^$TUS[@]?([A-Za-z0-9_]+)_closure$TPOSTLBL[@]?$/o ) {
+ $chk[++$i] = $_;
+ $chkcat[$i] = 'closure';
+ $chksymb[$i] = $1;
+
+ $closurechk{$1} = $i;
+
+ } elsif ( /^$TUS[@]?ghc.*c_ID$TPOSTLBL/o ) {
+ $chk[++$i] = $_;
+ $chkcat[$i] = 'consist';
+
+ } elsif ( /^($TUS[@]?__gnu_compiled_c|gcc2_compiled\.)$TPOSTLBL/o ) {
+ ; # toss it
+
+ } elsif ( /^$TUS[@]?ErrorIO_call_count$TPOSTLBL[@]?$/o # HACK!!!!
+ || /^$TUS[A-Za-z0-9_]+\.\d+$TPOSTLBL[@]?$/o
+ || /^$TUS[@]?.*_CAT$TPOSTLBL[@]?$/o # PROF: _entryname_CAT
+ || /^$TUS[@]?CC_.*_struct$TPOSTLBL[@]?$/o # PROF: _CC_ccident_struct
+ || /^$TUS[@]?.*_done$TPOSTLBL[@]?$/o # PROF: _module_done
+ || /^$TUS[@]?_module_registered$TPOSTLBL[@]?$/o # PROF: _module_registered
+ ) {
+ $chk[++$i] = $_;
+ $chkcat[$i] = 'data';
+ $chksymb[$i] = '';
+
+ } elsif ( /^([A-Za-z0-9_]+)\s+\.comm/ && $TargetPlatform =~ /^hppa/ ) {
+ $chk[++$i] = $_;
+ $chkcat[$i] = 'bss';
+ $chksymb[$i] = '';
+
+ } elsif ( $TargetPlatform =~ /^powerpc-|^rs6000-/ && /^LC\.\.([0-9]+)/ ) {
+ $chk[++$i] = $_;
+ $chkcat[$i] = 'toc';
+ $chksymb[$i] = $1;
+
+ } elsif ( /^($TUS[@]?(ret_|djn_)[A-Za-z0-9_]+)/o ) {
+ $chk[++$i] = $_;
+ $chkcat[$i] = 'misc';
+ $chksymb[$i] = '';
+ $symbtmp = $1;
+ $chksymb[$i] = $symbtmp if ($TargetPlatform =~ /^powerpc-|^rs6000-/) ; #rm andre
+
+ } elsif ( /^$TUS[@]?vtbl_([A-Za-z0-9_]+)$TPOSTLBL[@]?$/o ) {
+ $chk[++$i] = $_;
+ $chkcat[$i] = 'vector';
+ $chksymb[$i] = $1;
+
+ $vectorchk{$1} = $i;
+
+ } elsif ( /^$TUS[@]?([A-Za-z0-9_]+)DirectReturn$TPOSTLBL[@]?$/o ) {
+ $chk[++$i] = $_;
+ $chkcat[$i] = 'direct';
+ $chksymb[$i] = $1;
+
+ $directchk{$1} = $i;
+
+ } elsif ( /^$TUS[@]?([A-Za-z0-9_]+)_upd$TPOSTLBL[@]?$/o ) {
+ $chk[++$i] = $_;
+ $chkcat[$i] = 'misc';
+ print STDERR "_upd!!!!! I guess this code is dead!!!!\n";
+ # I guess this is never entered, since all _upds are
+ # either vtbl_'s or ret_'s, caught above. - andre
+ $chksymb[$i] = '';
+
+ } elsif ( $TargetPlatform =~ /^i386-.*-solaris2/
+ && /^(_uname|uname|stat|fstat):/ ) {
+ # for some utterly bizarre reason, this platform
+ # likes to drop little local C routines with these names
+ # into each and every .o file that #includes the
+ # relevant system .h file. Yuck. We just don't
+ # tolerate them in .hc files (which we are processing
+ # here). If you need to call one of these things from
+ # Haskell, make a call to your own C wrapper, then
+ # put that C wrapper (which calls one of these) in a
+ # plain .c file. WDP 95/12
+ $chk[++$i] = $_;
+ $chkcat[$i] = 'toss';
+ $chksymb[$i] = $1;
+
+ } elsif ( /^$TUS[@]?[A-Za-z0-9_]/o
+ && ( $TargetPlatform !~ /^hppa/ # need to avoid local labels in this case
+ || ! /^L\$\d+$/ )
+ && ( $TargetPlatform !~ /^powerpc|^rs6000/ # ditto
+ || ! /^(L\.\.\d+|LT\.\..*):$/ ) ) {
+ local($thing);
+ chop($thing = $_);
+ print STDERR "Funny global thing?: $_"
+ unless $KNOWN_FUNNY_THING{$thing}
+ || /^$TUS[@]?_(PRIn|PRStart).*$TPOSTLBL[@]?$/o # pointer reversal GC routines
+ || /^$TUS[@]?CC_.*$TPOSTLBL$/o # PROF: _CC_ccident ([@]? is a silly hack (see above))
+ || /^$TUS[@]?_reg.*$TPOSTLBL$/o; # PROF: __reg<module>
+ $chk[++$i] = $_;
+ $chkcat[$i] = 'misc';
+ if ($TargetPlatform =~ /^powerpc-|^rs6000-/)
+ { $chksymb[$i] = $thing; }
+ else { $chksymb[$i] = ''; };
+
+ } else { # simple line (duplicated at the top)
+
+ $chk[$i] .= $_;
+ }
+ }
+ $numchks = $#chk + 1;
+
+ # the division into chunks is imperfect;
+ # we throw some things over the fence into the next
+ # chunk.
+ #
+ # also, there are things we would like to know
+ # about the whole module before we start spitting
+ # output.
+
+ local($FIRST_MANGLABLE) = ($TargetPlatform =~ /^(alpha-|hppa|mips-)/) ? 1 : 0;
+
+# print STDERR "first chunk to mangle: $FIRST_MANGLABLE\n";
+
+ # Alphas: NB: we start meddling at chunk 1, not chunk 0
+ # The first ".rdata" is quite magical; as of GCC 2.7.x, it
+ # spits a ".quad 0" in after the v first ".rdata"; we
+ # detect this special case (tossing the ".quad 0")!
+ local($magic_rdata_seen) = 0;
+
+ # HPPAs, MIPSen: also start medding at chunk 1
+
+ if ($TargetPlatform =~ /^powerpc|^rs6000/) {
+ print OUTASM $T_HDR_toc; # yes, we have to put a .toc
+ # in the beginning of every file!
+ %tocequiv = (); # maps toc symbol number to toc symbol
+ %revtocequiv = (); # maps toc symbol to toc symbol number
+ for ($i = 1; $i < $numchks; $i++) {
+ $chk[$i] =~ s/\[RW\]//g;
+ $chk[$i] =~ s/\[DS\]//g;
+ $chk[$i] =~ s/^\.csect .*\[DS\]$//g;
+
+ if ( $chkcat[$i] eq 'toc' && $chk[$i] !~ /\.byte/ )
+ { $chk[$i] =~ s/$T_MOVE_DIRVS//g;
+ $chk[$i] =~ s/\t\.tc (\S+)\[TC\],(\S+_fast\d+)/\t\.tc \1\[TC\],\.\2/;
+ $chk[$i] =~ s/\t\.tc (\S+)\[TC\],(\S+_entry)\n/\t\.tc \1\[TC\],\.\2\n/;
+ $chk[$i] =~ s/\t\.tc (\S+)\[TC\],(ret_\S+)/\t\.tc \1\[TC\],\.\2/;
+ $chk[$i] =~ s/\t\.tc (\S+)\[TC\],(djn_\S+)/\t\.tc \1\[TC\],\.\2/;
+ $chk[$i] =~ s/\t\.tc (\S+)\[TC\],(vtbl_\S+)/\t\.tc \1\[TC\],\.\2/;
+ $chk[$i] =~ s/\t\.tc (\S+)\[TC\],ErrorIO_innards/\t\.tc \1\[TC\],\.ErrorIO_innards/;
+ $chk[$i] =~ s/\t\.tc (\S+)\[TC\],startStgWorld/\t\.tc \1\[TC\],\.startStgWorld/;
+ $chk[$i] =~ s/\.tc UpdatePAP\[TC\],UpdatePAP/\.tc UpdatePAP\[TC\],\.UpdatePAP/;
+ $chk[$i] =~ s/\.tc IndUpdRetDir\[TC\],IndUpdRetDir/\.tc IndUpdRetDir\[TC\],\.IndUpdRetDir/;
+ $chk[$i] =~ s/\t\.tc (_PRStart_\S+)\[TC\],_PRStart_\S+/\t\.tc \1\[TC\],\.\1/;
+
+ $tocnumber = $chksymb[$i];
+ $tocsymb = $chk[$i];
+ $tocsymb =~ s/^LC\.\.\d+:\n//;
+ $tocsymb =~ s/^\t\.tc \S+,(\S+)\n/\1/;
+ $tocequiv{$tocnumber} = $tocsymb;
+
+ } elsif ( $chkcat[$i] eq 'toc' && $chk[$i] =~ /\.byte/ ) {
+ $chkcat[$i] = 'literal';
+ }
+ }
+ };
+
+ for ($i = $FIRST_MANGLABLE; $i < $numchks; $i++) {
+ $c = $chk[$i]; # convenience copy
+
+# print STDERR "\nCHK $i (BEFORE) (",$chkcat[$i],"):\n", $c;
+
+ # toss all prologue stuff; HPPA is pretty weird
+ # (see elsewhere)
+ $c = &mash_hppa_prologue($c) if $TargetPlatform =~ /^hppa/;
+
+ # be slightly paranoid to make sure there's
+ # nothing surprising in there
+ if ( $c =~ /--- BEGIN ---/ ) {
+ if (($p, $r) = split(/--- BEGIN ---/, $c)) {
+
+ if ($TargetPlatform =~ /^i386-/) {
+ $p =~ s/^\tpushl \%edi\n//;
+ $p =~ s/^\tpushl \%esi\n//;
+ $p =~ s/^\tsubl \$\d+,\%esp\n//;
+ $p =~ s/^\tmovl \$\d+,\%eax\n\tcall __alloca\n// if ($TargetPlatform =~ /^.*-cygwin32/);
+ } elsif ($TargetPlatform =~ /^m68k-/) {
+ $p =~ s/^\tlink a6,#-?\d.*\n//;
+ $p =~ s/^\tpea a6@\n\tmovel sp,a6\n//;
+ # The above showed up in the asm code,
+ # so I added it here.
+ # I hope it's correct.
+ # CaS
+ $p =~ s/^\tmovel d2,sp\@-\n//;
+ $p =~ s/^\tmovel d5,sp\@-\n//; # SMmark.* only?
+ $p =~ s/^\tmoveml \#0x[0-9a-f]+,sp\@-\n//; # SMmark.* only?
+ } elsif ($TargetPlatform =~ /^mips-/) {
+ # the .frame/.mask/.fmask that we use is the same
+ # as that produced by GCC for miniInterpret; this
+ # gives GDB some chance of figuring out what happened
+ $FRAME = "\t.frame\t\$sp,2168,\$31\n\t.mask\t0x90000000,-4\n\t.fmask\t0x00000000,0\n";
+ $p =~ s/^\t\.(frame).*\n/__FRAME__/g;
+ $p =~ s/^\t\.(mask|fmask).*\n//g;
+ $p =~ s/^\t\.cprestore.*\n/\t\.cprestore 416\n/; # 16 + 100 4-byte args
+ $p =~ s/^\tsubu\t\$sp,\$sp,\d+\n//;
+ $p =~ s/^\tsw\t\$31,\d+\(\$sp\)\n//;
+ $p =~ s/^\tsw\t\$fp,\d+\(\$sp\)\n//;
+ $p =~ s/^\tsw\t\$28,\d+\(\$sp\)\n//;
+ $p =~ s/__FRAME__/$FRAME/;
+ } elsif ($TargetPlatform =~ /^powerpc-|^rs6000/) {
+ $p =~ s/^\tmflr 0\n//;
+ $p =~ s/^\tstm \d+,-\d+\(1\)\n//;
+ $p =~ s/^\tstw? 0,\d+\(1\)\n//;
+ $p =~ s/^\tstw?u 1,-\d+\(1\)\n//;
+ $p =~ s/^\tstw? \d+,-\d+\(1\)\n//g;
+ $p =~ s/^\tstfd \d+,-\d+\(1\)\n//g;
+ } else {
+ print STDERR "$Pgm: unknown prologue mangling? $TargetPlatform\n";
+ }
+
+ die "Prologue junk?: $p\n" if $p =~ /^\t[^\.]/
+ && $TargetPlatform !~ /^powerpc-/; #ToDo: remove test
+
+ # glue together what's left
+ $c = $p . $r;
+ $c =~ s/\n\t\n/\n/; # junk blank line
+ }
+ }
+
+ if ( $TargetPlatform =~ /^mips-/ ) {
+ # MIPS: first, this basic sequence may occur "--- END ---" or not
+ $c =~ s/^\tlw\t\$31,\d+\(\$sp\)\n\taddu\t\$sp,\$sp,\d+\n\tj\t\$31\n\t\.end/\t\.end/;
+ }
+
+ # toss all epilogue stuff; again, paranoidly
+ if ( $c =~ /--- END ---/ ) {
+ if (($r, $e) = split(/--- END ---/, $c)) {
+ if ($TargetPlatform =~ /^i386-/) {
+ $e =~ s/^\tret\n//;
+ $e =~ s/^\tpopl \%edi\n//;
+ $e =~ s/^\tpopl \%esi\n//;
+ $e =~ s/^\taddl \$\d+,\%esp\n//;
+ } elsif ($TargetPlatform =~ /^m68k-/) {
+ $e =~ s/^\tunlk a6\n//;
+ $e =~ s/^\trts\n//;
+ } elsif ($TargetPlatform =~ /^mips-/) {
+ $e =~ s/^\tlw\t\$31,\d+\(\$sp\)\n//;
+ $e =~ s/^\tlw\t\$fp,\d+\(\$sp\)\n//;
+ $e =~ s/^\taddu\t\$sp,\$sp,\d+\n//;
+ $e =~ s/^\tj\t\$31\n//;
+ } elsif ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
+ $e =~ s/^\taddi 1,1,\d+\n//;
+ $e =~ s/^\tcal 1,\d+\(1\)\n//;
+ $e =~ s/^\tlw?z? \d+,\d+\(1\)\n//;
+ $e =~ s/^\tmtlr 0\n//;
+ $e =~ s/^\tbl?r\n//;
+ } else {
+ print STDERR "$Pgm: unknown epilogue mangling? $TargetPlatform\n";
+ }
+ die "Epilogue junk?: $e\n" if $e =~ /^\t[^\.]/
+ && $TargetPlatform !~ /^powerpc-/; #ToDo: remove test
+
+ # glue together what's left
+ $c = $r . $e;
+ $c =~ s/\n\t\n/\n/; # junk blank line
+ }
+ }
+
+ # On SPARCs, we don't do --- BEGIN/END ---, we just
+ # toss the register-windowing save/restore/ret* instructions
+ # directly:
+ if ( $TargetPlatform =~ /^sparc-/ ) {
+ $c =~ s/^\t(save .*|restore|ret|retl)\n//g;
+ # throw away PROLOGUE comments
+ $c =~ s/^\t!#PROLOGUE# 0\n\t!#PROLOGUE# 1\n//;
+ }
+
+ # On Alphas, the prologue mangling is done a little later (below)
+
+ # toss all calls to __DISCARD__
+ $c =~ s/^\t(call|jbsr|jal)\s+$TUS[@]?__DISCARD__\n//go;
+
+ # MIPS: that may leave some gratuitous asm macros around
+ # (no harm done; but we get rid of them to be tidier)
+ $c =~ s/^\t\.set\tnoreorder\n\t\.set\tnomacro\n\taddu\t(\S+)\n\t\.set\tmacro\n\t\.set\treorder\n/\taddu\t$1\n/
+ if $TargetPlatform =~ /^mips-/;
+
+ # toss stack adjustment after DoSparks
+ $c =~ s/^(\tjbsr _DoSparks\n)\taddqw #8,sp/$1/g
+ if $TargetPlatform =~ /^m68k-/; # this looks old...
+
+ if ( $TargetPlatform =~ /^alpha-/ &&
+ ! $magic_rdata_seen &&
+ $c =~ /^\s*\.rdata\n\t\.quad 0\n\t\.align \d\n/ ) {
+ $c =~ s/^\s*\.rdata\n\t\.quad 0\n\t\.align (\d)\n/\.rdata\n\t\.align $1\n/;
+ $magic_rdata_seen = 1;
+ }
+
+ # pick some end-things and move them to the next chunk
+
+ # pin a funny end-thing on (for easier matching):
+ $c .= 'FUNNY#END#THING';
+
+ while ( $c =~ /$TMOVEDIRVS[@]?FUNNY#END#THING/o ) { # [@]? is a silly hack to avoid having to use curlies for T_PRE_APP
+ # (this SEGVs perl4 on alphas, you see)
+
+ $to_move = $1;
+ if ( $i < ($numchks - 1)
+ && ( $to_move =~ /$TCOPYDIRVS/
+ || ($TargetPlatform =~ /^hppa/ && $to_move =~ /align/ && $chkcat[$i+1] eq 'literal') )) {
+ $chk[$i + 1] = $to_move . $chk[$i + 1];
+ # otherwise they're tossed
+ }
+
+ $c =~ s/$TMOVEDIRVS[@]?FUNNY#END#THING/FUNNY#END#THING/o; # [@]? is a hack (see above)
+ }
+
+ 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)) {
+ $c = $keep . "..ng:\n";
+ } else {
+ print STDERR "malformed code block ($ent)?\n"
+ }
+ }
+ $c .= "\t.frame \$30,0,\$26,0\n\t.prologue" . $r;
+ }
+
+ $c =~ s/FUNNY#END#THING//;
+
+# print STDERR "\nCHK $i (AFTER) (",$chkcat[$i],"):\n", $c;
+
+ $chk[$i] = $c; # update w/ convenience copy
+ }
+
+ if ( $TargetPlatform =~ /^alpha-/ ) {
+ # print out the header stuff first
+ $chk[0] =~ s/^(\t\.file.*)"(ghc\d+\.c)"/$1"$ifile_root.hc"/;
+ print OUTASM $chk[0];
+
+ } elsif ( $TargetPlatform =~ /^hppa/ ) {
+ print OUTASM $chk[0];
+
+ } elsif ( $TargetPlatform =~ /^mips-/ ) {
+ $chk[0] = "\t\.file\t1 \"$ifile_root.hc\"\n" . $chk[0];
+
+ # get rid of horrible "<dollar>Revision: .*$" strings
+ local(@lines0) = split(/\n/, $chk[0]);
+ local($z) = 0;
+ while ( $z <= $#lines0 ) {
+ if ( $lines0[$z] =~ /^\t\.byte\t0x24,0x52,0x65,0x76,0x69,0x73,0x69,0x6f$/ ) {
+ undef($lines0[$z]);
+ $z++;
+ while ( $z <= $#lines0 ) {
+ undef($lines0[$z]);
+ last if $lines0[$z] =~ /[,\t]0x0$/;
+ $z++;
+ }
+ }
+ $z++;
+ }
+ $chk[0] = join("\n", @lines0);
+ $chk[0] =~ s/\n\n+/\n/;
+ print OUTASM $chk[0];
+ }
+
+ # print out all the literal strings next
+ for ($i = 0; $i < $numchks; $i++) {
+ if ( $chkcat[$i] eq 'literal' ) {
+ print OUTASM $T_HDR_literal, $chk[$i];
+ print OUTASM "; end literal\n" if $TargetPlatform =~ /^hppa/; # for the splitter
+
+ $chkcat[$i] = 'DONE ALREADY';
+ }
+ }
+
+ # on the HPPA, print out all the bss next
+ if ( $TargetPlatform =~ /^hppa/ ) {
+ for ($i = 1; $i < $numchks; $i++) {
+ if ( $chkcat[$i] eq 'bss' ) {
+ print OUTASM "\t.SPACE \$PRIVATE\$\n\t.SUBSPA \$BSS\$\n\t.align 4\n";
+ print OUTASM $chk[$i];
+
+ $chkcat[$i] = 'DONE ALREADY';
+ }
+ }
+ }
+
+ for ($i = $FIRST_MANGLABLE; $i < $numchks; $i++) {
+# print STDERR "$i: cat $chkcat[$i], symb $chksymb[$i]\n";
+
+ next if $chkcat[$i] eq 'DONE ALREADY';
+
+ if ( $chkcat[$i] eq 'misc' ) {
+ if ($chk[$i] ne '') {
+ print OUTASM $T_HDR_misc;
+ if ($TargetPlatform =~ /^powerpc-|^rs6000/) {
+ $chksymb[$i] =~ s/://;
+#new if ($chksymb[$i] =~ /ret.*upd/ || $KNOWN_FUNNY_THING{$chksymb[$i]}
+#new || $chksymb[$i] =~ /^$.{T_US}_(PRIn|PRStart).*${T_POST_LBL}$/o )
+#new { print OUTASM "\t\.globl $chksymb[$i]\n"; }
+ if ($chksymb[$i] ne '')
+ { print OUTASM "\t\.globl \.$chksymb[$i]\n"; };
+ if ($chk[$i] =~ /TOC\[tc0\], 0\n/)
+ { ($p, $r) = split(/TOC\[tc0\], 0\n/, $chk[$i]); $printDS = 1;}
+ else { $r = $chk[$i]; $printDS = 0; };
+ $chk[$i] = &mangle_powerpc_tailjump($r);
+ };
+ &print_doctored($chk[$i], 0);
+ if ($TargetPlatform =~ /^powerpc-|^rs6000-/ && $printDS) {
+ print OUTASM "\.csect ${chksymb[$i]}[DS]\n";
+ print OUTASM "${p}TOC[tc0], 0\n";
+ }
+ }
+
+ } elsif ( $chkcat[$i] eq 'toss' ) {
+ print STDERR "*** NB: TOSSING code for $chksymb[$i] !!! ***\n";
+
+ } elsif ( $chkcat[$i] eq 'data' ) {
+ if ($chk[$i] ne '') {
+ print OUTASM $T_HDR_data;
+ print OUTASM $chk[$i];
+ }
+
+ } elsif ( $chkcat[$i] eq 'consist' ) {
+ if ( $chk[$i] =~ /$T_hsc_cc_PAT/o ) {
+ local($consist) = "$1.$2.$3";
+ $consist =~ s/,/./g;
+ $consist =~ s/\//./g;
+ $consist =~ s/-/_/g;
+ $consist =~ s/[^A-Za-z0-9_.]/ZZ/g; # ToDo: properly?
+ print OUTASM $T_HDR_consist, "${consist}${T_POST_LBL}\n";
+
+ } elsif ( $TargetPlatform !~ /^(mips)-/ ) { # we just don't try in those case (ToDo)
+ # on mips: consistency string is just a v
+ # horrible bunch of .bytes,
+ # which I am too lazy to sort out (WDP 95/05)
+
+ print STDERR "Couldn't grok consistency: ", $chk[$i];
+ }
+
+ } elsif ( $chkcat[$i] eq 'splitmarker' ) {
+ # we can just re-constitute this one...
+ # NB: we emit _three_ underscores no matter what,
+ # so ghc-split doesn't have to care.
+ print OUTASM "___stg_split_marker",$chksymb[$i],"${T_POST_LBL}\n";
+
+ } elsif ( $chkcat[$i] eq 'closure'
+ || $chkcat[$i] eq 'infotbl'
+ || $chkcat[$i] eq 'slow'
+ || $chkcat[$i] eq 'fast' ) { # do them in that order
+ $symb = $chksymb[$i];
+
+ # CLOSURE
+ if ( defined($closurechk{$symb}) ) {
+ print OUTASM $T_HDR_closure;
+ print OUTASM $chk[$closurechk{$symb}];
+ $chkcat[$closurechk{$symb}] = 'DONE ALREADY';
+ }
+
+ # INFO TABLE
+ if ( defined($infochk{$symb}) ) {
+
+ print OUTASM $T_HDR_info;
+ if ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
+ $chk[$infochk{$symb}] =~ s/\.long ([_A-Za-z]\S+_entry)/\.long \.\1/;
+ $chk[$infochk{$symb}] =~ s/\.long ([A-Za-z]\S+_upd)/\.long \.\1/;
+ $chk[$infochk{$symb}] =~ s/\.long (IndUpdRet\S+)/\.long \.\1/;
+ $chk[$infochk{$symb}] =~ s/\.long StdErrorCode/\.long \.StdErrorCode/;
+ $chk[$infochk{$symb}] =~ s/\.long UpdErr/\.long \.UpdErr/;
+ print OUTASM $chk[$infochk{$symb}];
+ } else {
+ print OUTASM &rev_tbl($symb, $chk[$infochk{$symb}], 1);
+ }
+ # entry code will be put here!
+
+ # paranoia
+ if ( $chk[$infochk{$symb}] =~ /$TDOTWORD[@]?\s+([A-Za-z0-9_]+_entry)$/o
+ && $1 ne "${T_US}${symb}_entry" ) {
+ print STDERR "!!! entry point???\n",$chk[$infochk{$symb}];
+ }
+
+ $chkcat[$infochk{$symb}] = 'DONE ALREADY';
+ }
+
+ # STD ENTRY POINT
+ if ( defined($slowchk{$symb}) ) {
+
+ # teach it to drop through to the fast entry point:
+ $c = $chk[$slowchk{$symb}];
+
+ if ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
+ ($p, $r) = split(/TOC\[tc0\], 0\n/, $c);
+ if ($symb =~ /^[_A-Z]/)
+ {
+ print OUTASM "\t\.globl \.${chksymb[$i]}_entry\n";
+ print OUTASM "\.csect ${symb}_entry[DS]\n";
+ print OUTASM "${p}TOC[tc0], 0\n";
+ };
+ $r =~ s/\.csect \.text\[PR\]\n//; # todo: properly - andre
+ $c = &mangle_powerpc_tailjump($r);
+ };
+
+ 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-/ ) {
+ $c =~ s/^\tmovl \$${T_US}${symb}_fast\d+,\%edx\n\tjmp \*\%edx\n//;
+ $c =~ s/^\tmovl \$${T_US}${symb}_fast\d+,\%eax\n\tjmp \*\%eax\n//;
+ } 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-|^rs6000-/ ) {
+ $c =~ s/^\tb \.${T_US}${symb}_fast\d+\n//;
+ } elsif ( $TargetPlatform =~ /^sparc-/ ) {
+ $c =~ s/^\tcall ${T_US}${symb}_fast\d+,.*\n\tnop\n//;
+ $c =~ s/^\tcall ${T_US}${symb}_fast\d+,.*\n(\t[a-z].*\n)/$1/;
+ } else {
+ print STDERR "$Pgm: mystery slow-fast dropthrough: $TargetPlatform\n";
+ }
+ }
+
+ if ( $TargetPlatform !~ /^(alpha-|hppa|mips-)/ ) {
+ # On alphas, hppa: no very good way to look for "dangling"
+ # references to fast-entry point.
+ # (questionable re hppa and mips...)
+ print STDERR "still has jump to fast entry point:\n$c"
+ if $c =~ /$TUS[@]?$symb[@]?_fast/; # NB: paranoia
+ }
+
+ print OUTASM $T_HDR_entry;
+
+ &print_doctored($c, 1); # NB: the 1!!!
+
+ $chkcat[$slowchk{$symb}] = 'DONE ALREADY';
+ }
+
+ # FAST ENTRY POINT
+ if ( defined($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 $T_HDR_fast;
+ }
+ if ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
+ local(@lbls) = split(/:/, $c);
+ $fullname = $lbls[0];
+ $fullname =~ s/$T_MOVE_DIRVS//g;
+ if ( $fullname =~ /^[A-Z]/)
+ { print OUTASM "\t\.globl \.${fullname}\n";
+ } else {
+ print OUTASM "\t\.lglobl \.${fullname}\n"; #todo: rm - andre
+ };
+ $c =~ s/((.*\n)*)\t.long \S+, TOC\[tc0\], 0\n\.csect \.text\[PR\]\n((.*\n)*)/\1\3/;
+ $c = &mangle_powerpc_tailjump($c);
+ };
+ &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];
+
+ # VECTOR TABLE
+ if ( defined($vectorchk{$symb}) ) {
+ print OUTASM $T_HDR_vector;
+ if ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
+ if ( $symb =~ /^[A-Z]/) {
+ print OUTASM "\t\.globl \.vtbl_${symb}\n";
+ print OUTASM "\t\.globl vtbl_${symb}\n";
+ };
+ $chk[$vectorchk{$symb}] =~ s/\.long (\S+)/\.long \.\1/g;
+ print OUTASM ".vtbl_${symb}:\n";
+ print OUTASM $chk[$vectorchk{$symb}];
+ } else {
+ print OUTASM &rev_tbl($symb, $chk[$vectorchk{$symb}], 0);
+ }
+ # direct return code will be put here!
+ $chkcat[$vectorchk{$symb}] = 'DONE ALREADY';
+ }
+
+ # DIRECT RETURN
+ if ( defined($directchk{$symb}) ) {
+ print OUTASM $T_HDR_direct;
+ if ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
+ ($p, $r) = split(/TOC\[tc0\], 0\n/, $chk[$directchk{$symb}]);
+ &print_doctored($r, 0);
+ print OUTASM "\.csect ${symb}DirectReturn[DS]\n";
+ print OUTASM "${p}TOC[tc0], 0\n";
+ } else {
+ &print_doctored($chk[$directchk{$symb}], 0);
+ }
+ $chkcat[$directchk{$symb}] = 'DONE ALREADY';
+
+ } elsif ( $TargetPlatform =~ /^alpha-/ ) {
+ # Alphas: the commented nop is for the splitter, to ensure
+ # that no module ends with a label as the very last
+ # thing. (The linker will adjust the label to point
+ # to the first code word of the next module linked in,
+ # even if alignment constraints cause the label to move!)
+
+ print OUTASM "\t# nop\n";
+ }
+
+ } elsif ( $chkcat[$i] eq 'toc' ) {
+ # silly optimisation to print tocs, since they come in groups...
+ print OUTASM $T_HDR_toc;
+ local($j) = $i;
+ while ($chkcat[$j] eq 'toc')
+ { print OUTASM $chk[$j];
+ $chkcat[$j] = 'DONE ALREADY';
+ $j++;
+ }
+
+ } else {
+ &tidy_up_and_die(1,"$Pgm: unknown chkcat (ghc-asm: $TargetPlatform)\n$chkcat[$i]\n$chk[$i]\n");
+ }
+ }
+
+ print OUTASM $EXTERN_DECLS if $TargetPlatform =~ /^mips-/;
+
+ if ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
+ print OUTASM ".csect .text[PR]\n_section_.text:\n.csect .data[RW]\n\t.long _section_.text\n"
+ };
+
+ # finished
+ close(OUTASM) || &tidy_up_and_die(1,"Failed writing to $out_asmf\n");
+ close(INASM) || &tidy_up_and_die(1,"Failed reading from $in_asmf\n");
+}
+\end{code}
+
+\begin{code}
+sub mash_hppa_prologue { # OK, epilogue, too
+ local($_) = @_;
+
+ # toss all prologue stuff
+ s/^\s+\.ENTRY[^\0]*--- BEGIN ---/\t.ENTRY/;
+
+ # Lie about our .CALLINFO
+ s/^\s+\.CALLINFO.*$/\t.CALLINFO NO_CALLS,NO_UNWIND/;
+
+ # Get rid of P'
+
+ s/LP'/L'/g;
+ s/RP'/R'/g;
+
+ # toss all epilogue stuff
+ s/^\s+--- END ---[^\0]*\.EXIT/\t.EXIT/;
+
+ # Sorry; we moved the _info stuff to the code segment.
+ s/_info,DATA/_info,CODE/g;
+
+ return($_);
+}
+\end{code}
+
+\begin{code}
+sub print_doctored {
+ local($_, $need_fallthru_patch) = @_;
+
+ if ( $TargetPlatform !~ /^i386-/
+ || ! /^\t[a-z]/ ) { # no instructions in here, apparently
+ print OUTASM $_;
+ return;
+ }
+ # OK, must do some x86 **HACKING**
+
+ local($entry_patch) = '';
+ local($exit_patch) = '';
+ local($call_entry_patch)= '';
+ local($call_exit_patch) = '';
+
+#OLD: # first, convert calls to *very magic form*: (ToDo: document
+ # for real!) from
+ #
+ # pushl $768
+ # call _?PerformGC_wrapper
+ # addl $4,%esp
+ # to
+ # movl $768, %eax
+ # call _?PerformGC_wrapper
+ #
+ # The reason we do this now is to remove the apparent use of
+ # %esp, which would throw off the "what patch code do we need"
+ # decision.
+ #
+ # Special macros in ghc/includes/COptWraps.lh, used in
+ # ghc/runtime/CallWrap_C.lc, are required for this to work!
+ #
+
+ s/^\tpushl \$(\d+)\n\tcall ${T_DO_GC}\n\taddl \$4,\%esp\n/\tmovl \$$1,\%eax\n\tcall ${T_DO_GC}\n/go;
+ s/^\tpushl \%eax\n\tcall ${T_DO_GC}\n\taddl \$4,\%esp\n/\tcall ${T_DO_GC}\n/go;
+ s/^\tpushl \%edx\n\tcall ${T_DO_GC}\n\taddl \$4,\%esp\n/\tmovl \%edx,\%eax\n\tcall ${T_DO_GC}\n/go;
+
+#= if ( $StolenX86Regs <= 4 ) { # %ecx is ordinary reg
+#= s/^\tpushl \%ecx\n\tcall ${T_DO_GC}\n\taddl \$4,\%esp\n/\tmovl \%ecx,\%eax\n\tcall ${T_DO_GC}\n/go;
+#= }
+
+ # gotta watch out for weird instructions that
+ # invisibly smash various regs:
+ # rep* %ecx used for counting
+ # scas* %edi used for destination index
+ # cmps* %e[sd]i used for indices
+ # loop* %ecx used for counting
+ #
+ # SIGH.
+
+ # We cater for:
+ # * use of STG reg [ nn(%ebx) ] where no machine reg avail
+ #
+ # * GCC used an "STG reg" for its own purposes
+ #
+ # * some secret uses of machine reg, requiring STG reg
+ # to be saved/restored
+
+ # The most dangerous "GCC uses" of an "STG reg" are when
+ # the reg holds the target of a jmp -- it's tricky to
+ # insert the patch-up code before we get to the target!
+ # So here we change the jmps:
+
+ # --------------------------------------------------------
+ # it can happen that we have jumps of the form...
+ # jmp *<something involving %esp>
+ # or
+ # jmp <something involving another naughty register...>
+ #
+ # a reasonably-common case is:
+ #
+ # movl $_blah,<bad-reg>
+ # jmp *<bad-reg>
+ #
+ # which is easily fixed as:
+ #
+ # sigh! try to hack around it...
+ #
+
+ if ($StolenX86Regs <= 2 ) { # YURGH! spurious uses of esi?
+ s/^\tmovl (.*),\%esi\n\tjmp \*%esi\n/\tmovl $1,\%eax\n\tjmp \*\%eax\n/g;
+ s/^\tjmp \*(-?\d*)\((.*\%esi.*)\)\n/\tmovl $2,\%eax\n\tjmp \*$1\(\%eax\)\n/g;
+ s/^\tjmp \*\%esi\n/\tmovl \%esi,\%eax\n\tjmp \*\%eax\n/g;
+ die "$Pgm: (mangler) still have jump involving \%esi!\n$_"
+ if /(jmp|call) .*\%esi/;
+ }
+ if ($StolenX86Regs <= 3 ) { # spurious uses of edi?
+ s/^\tmovl (.*),\%edi\n\tjmp \*%edi\n/\tmovl $1,\%eax\n\tjmp \*\%eax\n/g;
+ s/^\tjmp \*(-?\d*)\((.*\%edi.*)\)\n/\tmovl $2,\%eax\n\tjmp \*$1\(\%eax\)\n/g;
+ s/^\tjmp \*\%edi\n/\tmovl \%edi,\%eax\n\tjmp \*\%eax\n/g;
+ die "$Pgm: (mangler) still have jump involving \%edi!\n$_"
+ if /(jmp|call) .*\%edi/;
+ }
+#= if ($StolenX86Regs <= 4 ) { # spurious uses of ecx?
+#= s/^\tmovl (.*),\%ecx\n\tjmp \*%ecx\n/\tmovl $1,\%eax\n\tjmp \*\%eax\n/g;
+#= s/^\tjmp \*(-?\d*)\((.*\%ecx.*)\)\n/\tmovl $2,\%eax\n\tjmp \*$1\(\%eax\)\n/g;
+#= s/^\tjmp \*\%ecx\n/\tmovl \%ecx,\%eax\n\tjmp \*\%eax\n/g;
+#= die "$Pgm: (mangler) still have jump involving \%ecx!\n$_"
+#= if /(jmp|call) .*\%ecx/;
+#= }
+
+ # OK, now we can decide what our patch-up code is going to
+ # be:
+ if ( $StolenX86Regs <= 2
+ && ( /32\(\%ebx\)/ || /\%esi/ || /^\tcmps/ ) ) { # R1 (esi)
+ $entry_patch .= "\tmovl \%esi,32(\%ebx)\n";
+ $exit_patch .= "\tmovl 32(\%ebx),\%esi\n";
+ # nothing for call_{entry,exit} because %esi is callee-save
+ }
+ if ( $StolenX86Regs <= 3
+ && ( /64\(\%ebx\)/ || /\%edi/ || /^\t(scas|cmps)/ ) ) { # SpA (edi)
+ $entry_patch .= "\tmovl \%edi,64(\%ebx)\n";
+ $exit_patch .= "\tmovl 64(\%ebx),\%edi\n";
+ # nothing for call_{entry,exit} because %edi is callee-save
+ }
+#= if ( $StolenX86Regs <= 4
+#= && ( /80\(\%ebx\)/ || /\%ecx/ || /^\t(rep|loop)/ ) ) { # Hp (ecx)
+#= $entry_patch .= "\tmovl \%ecx,80(\%ebx)\n";
+#= $exit_patch .= "\tmovl 80(\%ebx),\%ecx\n";
+#=
+#= $call_exit_patch .= "\tmovl \%ecx,108(\%ebx)\n";
+#= $call_entry_patch .= "\tmovl 108(\%ebx),\%ecx\n";
+#= # I have a really bad feeling about this if we ever
+#= # have a nested call...
+#= # NB: should just hide it somewhere in the C stack.
+#= }
+ # --------------------------------------------------------
+ # next, here we go with non-%esp patching!
+ #
+ s/^(\t[a-z])/$entry_patch$1/; # before first instruction
+ s/^(\tcall .*\n(\taddl \$\d+,\%esp\n)?)/$call_exit_patch$1$call_entry_patch/g; # _all_ calls
+
+ # fix _all_ non-local jumps:
+
+ s/^\tjmp \*${T_X86_PRE_LLBL_PAT}/\tJMP___SL/go;
+ s/^\tjmp ${T_X86_PRE_LLBL_PAT}/\tJMP___L/go;
+
+ s/^(\tjmp .*\n)/$exit_patch$1/g; # here's the fix...
+
+ s/^\tJMP___SL/\tjmp \*${T_X86_PRE_LLBL}/go;
+ s/^\tJMP___L/\tjmp ${T_X86_PRE_LLBL}/go;
+
+ # fix post-PerformGC wrapper (re-)entries ???
+
+ if ($StolenX86Regs == 2 ) {
+ die "ARGH! Jump uses \%esi or \%edi with -monly-2-regs:\n$_"
+ if /^\t(jmp|call) .*\%e(si|di)/;
+#= die "ARGH! Jump uses \%esi, \%edi, or \%ecx with -monly-2-regs:\n$_"
+#= if /^\t(jmp|call) .*\%e(si|di|cx)/;
+ } elsif ($StolenX86Regs == 3 ) {
+ die "ARGH! Jump uses \%edi with -monly-3-regs:\n$_"
+ if /^\t(jmp|call) .*\%edi/;
+#= die "ARGH! Jump uses \%edi or \%ecx with -monly-3-regs:\n$_"
+#= if /^\t(jmp|call) .*\%e(di|cx)/;
+#= } elsif ($StolenX86Regs == 4 ) {
+#= die "ARGH! Jump uses \%ecx with -monly-4-regs:\n$_"
+#= if /^\t(jmp|call) .*\%ecx/;
+ }
+
+ # final peephole fixes
+
+ s/^\tmovl \%eax,36\(\%ebx\)\n\tjmp \*36\(\%ebx\)\n/\tmovl \%eax,36\(\%ebx\)\n\tjmp \*\%eax\n/;
+# the short form may tickle perl bug:
+# s/^\tmovl \$${T_US}(.*),(\%e[abcd]x)\n\tjmp \*$2/\tjmp $T_US$1/g;
+ s/^\tmovl \$${T_US}(.*),\%eax\n\tjmp \*\%eax/\tjmp $T_US$1/g;
+ s/^\tmovl \$${T_US}(.*),\%ebx\n\tjmp \*\%ebx/\tjmp $T_US$1/g;
+ s/^\tmovl \$${T_US}(.*),\%ecx\n\tjmp \*\%ecx/\tjmp $T_US$1/g;
+ s/^\tmovl \$${T_US}(.*),\%edx\n\tjmp \*\%edx/\tjmp $T_US$1/g;
+
+ # Hacks to eliminate some reloads of Hp. Worth about 5% code size.
+ # We could do much better than this, but at least it catches about
+ # half of the unnecessary reloads.
+ # Note that these will stop working if either:
+ # (i) the offset of Hp from BaseReg changes from 80, or
+ # (ii) the register assignment of BaseReg changes from %ebx
+
+ s/^\tmovl 80\(\%ebx\),\%e.x\n\tmovl \$(.*),(-?[0-9]*)\(\%e.x\)\n\tmovl 80\(\%ebx\),\%e(.)x/\tmovl 80\(\%ebx\),\%e$3x\n\tmovl \$$1,$2\(\%e$3x\)/g;
+
+ s/^\tmovl 80\(\%ebx\),\%e(.)x\n\tmovl (.*),\%e(.)x\n\tmovl \%e$3x,(-?[0-9]*\(\%e$1x\))\n\tmovl 80\(\%ebx\),\%e$1x/\tmovl 80\(\%ebx\),\%e$1x\n\tmovl $2,\%e$3x\n\tmovl \%e$3x,$4/g;
+
+ s/^\tmovl 80\(\%ebx\),\%edx((\n\t(movl|addl) .*,((-?[0-9]*\(.*)|(\%e[abc]x)))+)\n\tmovl 80\(\%ebx\),\%edx/\tmovl 80\(\%ebx\),\%edx$1/g;
+ s/^\tmovl 80\(\%ebx\),\%eax((\n\t(movl|addl) .*,((-?[0-9]*\(.*)|(\%e[bcd]x)))+)\n\tmovl 80\(\%ebx\),\%eax/\tmovl 80\(\%ebx\),\%eax$1/g;
+
+ # --------------------------------------------------------
+ # that's it -- print it
+ #
+ #die "Funny jumps?\n$_" if /${T_X86_BADJMP}/o; # paranoia
+
+ print OUTASM $_;
+
+ if ( $need_fallthru_patch ) { # exit patch for end of slow entry code
+ print OUTASM $exit_patch;
+ # ToDo: make it not print if there is a "jmp" at the end
+ }
+}
+\end{code}
+
+\begin{code}
+sub init_FUNNY_THINGS {
+ %KNOWN_FUNNY_THING = (
+ "${T_US}CheckHeapCode${T_POST_LBL}", 1,
+ "${T_US}CommonUnderflow${T_POST_LBL}", 1,
+ "${T_US}Continue${T_POST_LBL}", 1,
+ "${T_US}EnterNodeCode${T_POST_LBL}", 1,
+ "${T_US}ErrorIO_call_count${T_POST_LBL}", 1,
+ "${T_US}ErrorIO_innards${T_POST_LBL}", 1,
+ "${T_US}IndUpdRetDir${T_POST_LBL}", 1,
+ "${T_US}IndUpdRetV0${T_POST_LBL}", 1,
+ "${T_US}IndUpdRetV1${T_POST_LBL}", 1,
+ "${T_US}IndUpdRetV2${T_POST_LBL}", 1,
+ "${T_US}IndUpdRetV3${T_POST_LBL}", 1,
+ "${T_US}IndUpdRetV4${T_POST_LBL}", 1,
+ "${T_US}IndUpdRetV5${T_POST_LBL}", 1,
+ "${T_US}IndUpdRetV6${T_POST_LBL}", 1,
+ "${T_US}IndUpdRetV7${T_POST_LBL}", 1,
+ "${T_US}PrimUnderflow${T_POST_LBL}", 1,
+ "${T_US}StackUnderflowEnterNode${T_POST_LBL}", 1,
+ "${T_US}StdErrorCode${T_POST_LBL}", 1,
+ "${T_US}UnderflowVect0${T_POST_LBL}", 1,
+ "${T_US}UnderflowVect1${T_POST_LBL}", 1,
+ "${T_US}UnderflowVect2${T_POST_LBL}", 1,
+ "${T_US}UnderflowVect3${T_POST_LBL}", 1,
+ "${T_US}UnderflowVect4${T_POST_LBL}", 1,
+ "${T_US}UnderflowVect5${T_POST_LBL}", 1,
+ "${T_US}UnderflowVect6${T_POST_LBL}", 1,
+ "${T_US}UnderflowVect7${T_POST_LBL}", 1,
+ "${T_US}UpdErr${T_POST_LBL}", 1,
+ "${T_US}UpdatePAP${T_POST_LBL}", 1,
+ "${T_US}WorldStateToken${T_POST_LBL}", 1,
+ "${T_US}_Enter_Internal${T_POST_LBL}", 1,
+ "${T_US}_PRMarking_MarkNextAStack${T_POST_LBL}", 1,
+ "${T_US}_PRMarking_MarkNextBStack${T_POST_LBL}", 1,
+ "${T_US}_PRMarking_MarkNextCAF${T_POST_LBL}", 1,
+ "${T_US}_PRMarking_MarkNextGA${T_POST_LBL}", 1,
+ "${T_US}_PRMarking_MarkNextRoot${T_POST_LBL}", 1,
+ "${T_US}_PRMarking_MarkNextSpark${T_POST_LBL}", 1,
+ "${T_US}_PRMarking_MarkNextEvent${T_POST_LBL}", 1,
+ "${T_US}_PRMarking_MarkNextClosureInFetchBuffer${T_POST_LBL}", 1,
+ "${T_US}_Scavenge_Forward_Ref${T_POST_LBL}", 1,
+ "${T_US}__std_entry_error__${T_POST_LBL}", 1,
+ "${T_US}_startMarkWorld${T_POST_LBL}", 1,
+ "${T_US}resumeThread${T_POST_LBL}", 1,
+ "${T_US}startCcRegisteringWorld${T_POST_LBL}", 1,
+ "${T_US}startEnterFloat${T_POST_LBL}", 1,
+ "${T_US}startEnterInt${T_POST_LBL}", 1,
+ "${T_US}startPerformIO${T_POST_LBL}", 1,
+ "${T_US}startStgWorld${T_POST_LBL}", 1,
+ "${T_US}stopPerformIO${T_POST_LBL}", 1
+ );
+}
+\end{code}
+
+The following table reversal is used for both info tables and return
+vectors. In both cases, we remove the first entry from the table,
+reverse the table, put the label at the end, and paste some code
+(that which is normally referred to by the first entry in the table)
+right after the table itself. (The code pasting is done elsewhere.)
+
+\begin{code}
+sub rev_tbl {
+ local($symb, $tbl, $discard1) = @_;
+
+ local($before) = '';
+ local($label) = '';
+ local(@imports) = (); # hppa only
+ local(@words) = ();
+ local($after) = '';
+ local(@lines) = split(/\n/, $tbl);
+ local($i, $j); #local ($i, $extra, $words_to_pad, $j);
+
+ # see comment in mangleAsm as to why this silliness is needed.
+ local($TDOTWORD) = ${T_DOT_WORD};
+ local($TDOTGLOBAL) = ${T_DOT_GLOBAL};
+ local($TUS) = ${T_US};
+ local($TPOSTLBL) = ${T_POST_LBL};
+
+ for ($i = 0; $i <= $#lines && $lines[$i] !~ /^\t$TDOTWORD\s+/o; $i++) {
+ $label .= $lines[$i] . "\n",
+ next if $lines[$i] =~ /^[A-Za-z0-9_]+_info$TPOSTLBL[@]?$/o
+ || $lines[$i] =~ /$TDOTGLOBAL/o
+ || $lines[$i] =~ /^$TUS[@]?vtbl_\S+$TPOSTLBL[@]?$/o;
+
+ $before .= $lines[$i] . "\n"; # otherwise...
+ }
+
+ if ( $TargetPlatform !~ /^hppa/ ) {
+ for ( ; $i <= $#lines && $lines[$i] =~ /^\t$TDOTWORD\s+/o; $i++) {
+ push(@words, $lines[$i]);
+ }
+ } else { # hppa weirdness
+ for ( ; $i <= $#lines && $lines[$i] =~ /^\s+\.(word|IMPORT)/; $i++) {
+ if ($lines[$i] =~ /^\s+\.IMPORT/) {
+ push(@imports, $lines[$i]);
+ } else {
+ # We don't use HP's ``function pointers''
+ # We just use labels in code space, like normal people
+ $lines[$i] =~ s/P%//;
+ push(@words, $lines[$i]);
+ }
+ }
+ }
+
+ # now throw away the first word (entry code):
+ shift(@words) if $discard1;
+
+# Padding removed to reduce code size and improve performance on Pentiums.
+# Simon M. 13/4/96
+ # for 486-cache-friendliness, we want our tables aligned
+ # on 16-byte boundaries (.align 4). Let's pad:
+# $extra = ($#words + 1) % 4;
+# $words_to_pad = ($extra == 0) ? 0 : 4 - $extra;
+# for ($j = 0; $j < $words_to_pad; $j++) { push(@words, "\t${T_DOT_WORD} 0"); }
+
+ for (; $i <= $#lines; $i++) {
+ $after .= $lines[$i] . "\n";
+ }
+
+ # Alphas:If we have anonymous text (not part of a procedure), the
+ # linker may complain about missing exception information. Bleh.
+ if ( $TargetPlatform =~ /^alpha-/ && $label =~ /^([A-Za-z0-9_]+):$/) {
+ $before = "\t.ent $1\n" . $before;
+ $after .= "\t.end $1\n";
+ }
+
+ $tbl = $before
+ . (($TargetPlatform !~ /^hppa/) ? '' : join("\n", @imports) . "\n")
+ . join("\n", (reverse @words)) . "\n"
+ . $label . $after;
+
+# print STDERR "before=$before\n";
+# print STDERR "label=$label\n";
+# print STDERR "words=",(reverse @words),"\n";
+# print STDERR "after=$after\n";
+
+ $tbl;
+}
+\end{code}
+
+\begin{code}
+sub mini_mangle_asm_i386 {
+ local($in_asmf, $out_asmf) = @_;
+
+ &init_TARGET_STUFF();
+
+ # see mangleAsm comment
+ local($TUS) = ${T_US};
+ local($TPOSTLBL)=${T_POST_LBL};
+
+ open(INASM, "< $in_asmf")
+ || &tidy_up_and_die(1,"$Pgm: failed to open `$in_asmf' (to read)\n");
+ open(OUTASM,"> $out_asmf")
+ || &tidy_up_and_die(1,"$Pgm: failed to open `$out_asmf' (to write)\n");
+
+ while (<INASM>) {
+ print OUTASM;
+
+ next unless
+ /^$TUS[@]?(PerformGC|StackOverflow|Yield|PerformReschedule)_wrapper$TPOSTLBL\n/o;
+ print OUTASM "\tmovl \%esp, ${T_US}__temp_esp\n";
+ print OUTASM "\tmovl \%eax, ${T_US}__temp_eax\n";
+ }
+
+ # finished:
+ close(OUTASM) || &tidy_up_and_die(1,"Failed writing to $out_asmf\n");
+ close(INASM) || &tidy_up_and_die(1,"Failed reading from $in_asmf\n");
+}
+\end{code}
+
+The HP is a major nuisance. The threaded code mangler moved info
+tables from data space to code space, but unthreaded code in the RTS
+still has references to info tables in data space. Since the HP
+linker is very precise about where symbols live, we need to patch the
+references in the unthreaded RTS as well.
+
+\begin{code}
+sub mini_mangle_asm_hppa {
+ local($in_asmf, $out_asmf) = @_;
+
+ open(INASM, "< $in_asmf")
+ || &tidy_up_and_die(1,"$Pgm: failed to open `$in_asmf' (to read)\n");
+ open(OUTASM,"> $out_asmf")
+ || &tidy_up_and_die(1,"$Pgm: failed to open `$out_asmf' (to write)\n");
+
+ while (<INASM>) {
+ s/_info,DATA/_info,CODE/; # Move _info references to code space
+ s/P%_PR/_PR/;
+ print OUTASM;
+ }
+
+ # finished:
+ close(OUTASM) || &tidy_up_and_die(1,"Failed writing to $out_asmf\n");
+ close(INASM) || &tidy_up_and_die(1,"Failed reading from $in_asmf\n");
+}
+
+\end{code}
+
+\begin{code}
+sub mini_mangle_asm_powerpc {
+ local($in_asmf, $out_asmf) = @_;
+
+ open(INASM, "< $in_asmf")
+ || &tidy_up_and_die(1,"$Pgm: failed to open `$in_asmf' (to read)\n");
+ open(OUTASM,"> $out_asmf")
+ || &tidy_up_and_die(1,"$Pgm: failed to open `$out_asmf' (to write)\n");
+
+ while (<INASM>) {
+ s/long _PRStart/long ._PRStart/;
+ s/long _PRIn_/long ._PRIn_/;
+ s/long _Dummy_(\S+)_entry/long ._Dummy_\1_entry/;
+ s/long _PRMarking_MarkNextRoot\[DS\]/long ._PRMarking_MarkNextRoot/;
+ s/long _PRMarking_MarkNextCAF\[DS\]/long ._PRMarking_MarkNextCAF/;
+ s/long _PRMarking_MarkNextAStack\[DS\]/long ._PRMarking_MarkNextAStack/;
+ s/long _PRMarking_MarkNextBStack\[DS\]/long ._PRMarking_MarkNextBStack/;
+ print OUTASM;
+ }
+
+ # finished:
+ close(OUTASM) || &tidy_up_and_die(1,"Failed writing to $out_asmf\n");
+ close(INASM) || &tidy_up_and_die(1,"Failed reading from $in_asmf\n");
+}
+
+sub mangle_powerpc_tailjump {
+ local($c) = @_;
+ local($maybe_more) = 1;
+ while (($c =~ /\tlw?z? \d+,LC\.\.\d+\(2\)\n\tmtctr \d+\n\tbctr\n/) && $maybe_more)
+ { $maybe_more = 0;
+ $lcsymb = $c;
+ $lcsymb =~ s/(.*\n)*\tlw?z? \d+,LC\.\.(\d+)\(2\)\n\tmtctr \d+\n\tbctr\n(.*\n)*/\2/;
+# the checks for r1 and r2 are mostly paranoia...
+ $r1 = $c;
+ $r1 =~ s/(.*\n)*\tlw?z? (\d+),LC\.\.\d+\(2\)\n\tmtctr \d+\n\tbctr\n(.*\n)*/\2/;
+ $r2 = $c;
+ $r2 =~ s/(.*\n)*\tlw?z? \d+,LC\.\.(\d+)\(2\)\n\tmtctr (\d+)\n\tbctr\n(.*\n)*/\3/;
+ if (r1 == r2)
+ { $maybe_more = 1;
+ $c =~ s/((.*\n)*)\tlw?z? \d+,LC\.\.\d+\(2\)\n\tmtctr \d+\n\tbctr\n((.*\n)*)/\1\tb $tocequiv{$lcsymb}\n\3/;
+ }
+ };
+ $c;
+}
+
+# make "require"r happy...
+1;
+\end{code}