Any other required tidying up.
\end{itemize}
+General note [chak]: Many regexps are very fragile because they rely on white
+space being in the right place. This caused trouble with gcc 2.95 (at least
+on Linux), where the use of white space in .s files generated by gcc suddenly
+changed. To guarantee compatibility across different versions of gcc, make
+sure (at least on i386-.*-linux) that regexps tolerate varying amounts of white
+space between an assembler statement and its arguments as well as after a the
+comma separating multiple arguments.
+
+\emph{For the time being, I have corrected the regexps for i386-.*-linux. I
+didn't touch all the regexps for other i386 platforms, as I don't have
+a box to test these changes.}
+
HPPA specific notes:
\begin{itemize}
\item
%************************************************************************
%* *
+\subsection{Top-level code}
+%* *
+%************************************************************************
+
+\begin{code}
+$TargetPlatform = $TARGETPLATFORM;
+
+($Pgm = $0) =~ s|.*/||;
+$ifile = $ARGV[0];
+$ofile = $ARGV[1];
+
+if ( $TargetPlatform =~ /^i386-/ ) {
+ if ($ARGV[2] eq '') {
+ $StolenX86Regs = 4;
+ } else {
+ $StolenX86Regs = $ARGV[2];
+ }
+}
+
+&mangle_asm($ifile,$ofile);
+
+exit(0);
+\end{code}
+
+%************************************************************************
+%* *
\subsection{Constants for various architectures}
%* *
%************************************************************************
$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
+ $T_CONST_LBL = '^\$L?C(\d+):$'; # 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_hsc_cc_PAT = '\.ascii.*\)(hsc|cc) (.*)\\\\11"\n\t\.ascii\s+"(.*)\\\\0"';
- $T_DOT_WORD = '\.quad';
- $T_DOT_GLOBAL = "\t\.globl";
+ $T_DOT_WORD = '\.(long|quad|byte|word)';
+ $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_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_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_hsc_cc_PAT = '\.STRING.*\)(hsc|cc) (.*)\\\\x09(.*)\\\\x00"';
+ $T_DOT_WORD = '\.(blockz|word|half|byte)';
+ $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_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_direct = "\t.SPACE \$TEXT\$\n\t.SUBSPA \$CODE\$\n\t\.align 4\n";
#--------------------------------------------------------#
- } elsif ( $TargetPlatform =~ /^i386-.*-(linuxaout|freebsd)/ ) {
+ } elsif ( $TargetPlatform =~ /^i386-.*-(linuxaout|freebsd2|openbsd|nextstep3|cygwin32|mingw32)$/ ) {
+ # 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 = '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_MOVE_DIRVS = '^(\s*(\.(p2)?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_WORD = '\.(long|word|value|byte|space)';
$T_DOT_GLOBAL = '\.globl';
- $T_HDR_literal = "\.text\n\t\.align 2\n"; # .align 4 is 486-cache friendly
+ $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"; # ToDo: change align??
+ $T_HDR_data = "\.data\n\t\.align 2\n";
$T_HDR_consist = "\.text\n";
- $T_HDR_closure = "\.data\n\t\.align 2\n"; # ToDo: change align?
+ $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_direct = "\.text\n\t\.align 2,0x90\n";
#--------------------------------------------------------#
- } elsif ( $TargetPlatform =~ /^i386-.*-(solaris2|linux)$/ ) {
+ } elsif ( $TargetPlatform =~ /^i386-.*-(solaris2|linux|freebsd|netbsd)$/ ) {
$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
+ $T_PRE_APP = # regexp that says what comes before APP/NO_APP
+ ($TargetPlatform =~ /-(linux|freebsd|netbsd)$/) ? '#' : '/' ;
$T_CONST_LBL = '^\.LC(\d+):$'; # 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_X86_BADJMP = '^\tjmp\s+[^\.\*]';
- $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_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_hsc_cc_PAT = '\.string.*\)(hsc|cc) (.*)\\\\t(.*)"';
- $T_DOT_WORD = '\.long';
+ if ( $TargetPlatform =~ /freebsd|netbsd/ ) {
+ $T_hsc_cc_PAT = '\.ascii.*\)(hsc|cc) (.*)\\\\11"\n\t\.ascii\s+"(.*)\\\\0"';
+ } else {
+ $T_hsc_cc_PAT = '\.string.*\)(hsc|cc) (.*)\\\\t(.*)"';
+ }
+
+ $T_DOT_WORD = '\.(long|value|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 16\n";
+ $T_HDR_misc = "\.text\n\t\.align 4\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_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 16\n";
- $T_HDR_vector = "\.text\n\t\.align 16\n"; # NB: requires padding
- $T_HDR_direct = "\.text\n\t\.align 16\n";
+ $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";
#--------------------------------------------------------#
} 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_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_HDR_data = "\.data\n\t\.even\n";
$T_HDR_consist = "\.text\n";
$T_HDR_closure = "\.data\n\t\.even\n";
+ $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_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_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_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_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_direct = "\t\.text\n\t\.align 2\n";
#--------------------------------------------------------#
- } elsif ( $TargetPlatform =~ /^powerpc-.*/ ) {
+ } 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_US = ''; # _ if symbols have an underscore on the front
$T_PRE_APP = 'NOT APPLICABLE'; # regexp that says what comes before APP/NO_APP
- $T_CONST_LBL = '^LC\.\.(\d+):$'; # regexp for what such a lbl looks like
+ $T_CONST_LBL = 'NOT APPLICABLE'; # regexp for what such a lbl looks like
$T_POST_LBL = ':';
- $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_MOVE_DIRVS = '^(\s*(\.toc|\.align \d+|\.csect \S+|\t\.?l?globl \S+)\n)';
+ $T_COPY_DIRVS = '\.(l?globl)';
- $T_hsc_cc_PAT = '\.string.*\)(hsc|cc) (.*)\\\\t(.*)"';
+ $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_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";
+ $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_srt = "# 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_COPY_DIRVS = '\.(global|proc|stab)';
$T_hsc_cc_PAT = '\.asciz.*\)(hsc|cc) (.*)\\\\t(.*)"';
- $T_DOT_WORD = '\.word';
- $T_DOT_GLOBAL = '\.global';
+ $T_DOT_WORD = '\.(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";
$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_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_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_HDR_data = "\.data\n\t\.align 8\n";
$T_HDR_consist = "\.text\n";
$T_HDR_closure = "\.data\n\t\.align 4\n";
+ $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";
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: $T_POST_LBL\n";
# multi-line regexp matching:
local($*) = 1;
local($i, $c);
+
+
&init_TARGET_STUFF();
&init_FUNNY_THINGS();
%slowchk = (); # ditto, its regular "slow" entry code
%fastchk = (); # ditto, fast 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
%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>) {
+ tr/\r//d if $TargetPlatform =~ /-mingw32$/;
next if $T_STABBY && /^\.stab.*${T_US}__stg_split_marker/o;
next if $T_STABBY && /^\.stab.*ghc.*c_ID/;
- next if /${T_PRE_APP}(NO_)?APP/o;
-
+ next if /^\t\.def.*endef$/;
+ next if /${T_PRE_APP}(NO_)?APP/o;
next if /^;/ && $TargetPlatform =~ /^hppa/;
- next if /(^$|^\t\.file\t|^ # )/ && $TargetPlatform =~ /^mips-/;
+ next if /(^$|^\t\.file\t|^ # )/ && $TargetPlatform =~ /^(mips|powerpc|rs6000)-/;
+
+ last if /^_section_\.text:$/ && $TargetPlatform =~ /^powerpc-|^rs6000-/;
if ( $TargetPlatform =~ /^mips-/
- && /^\t\.(globl \S+ \.text|comm\t)/ ) {
+ && /^\t\.(globl\S+\.text|comm\t)/ ) {
$EXTERN_DECLS .= $_ unless /(__DISCARD__|\b(PK_|ASSIGN_)(FLT|DBL)\b)/;
+ # As a temporary solution for compiling "foreign export" declarations,
+ # we use global variables to pass arguments from C to STG land.
+ # These declarations live in the .hc file and not in the generated C
+ # stub file, so we let them pass through here.
+ } elsif ( /^\t\.comm\t__fexp_.*$/ ) {
+ $chk[++$i] = $_;
+ $chkcat[$i] = 'data';
+ $chksymb[$i] = '';
+
} elsif ( /^\s+/ ) { # most common case first -- a simple line!
# duplicated from the bottom
$chkcat[$i] = 'literal';
$chksymb[$i] = $1;
- } elsif ( /^${T_US}__stg_split_marker(\d+)${T_POST_LBL}$/o ) {
+ } elsif ( /^${T_US}__stg_split_marker(\d*)${T_POST_LBL}$/o ) {
$chk[++$i] = $_;
$chkcat[$i] = 'splitmarker';
$chksymb[$i] = $1;
$infochk{$symb} = $i;
- } elsif ( /^${T_US}([A-Za-z0-9_]+)_entry${T_POST_LBL}$/o ) {
+ } elsif ( /^${T_US}([A-Za-z0-9_]+)_(entry|ret)${T_POST_LBL}$/o ) {
$chk[++$i] = $_;
$chkcat[$i] = 'slow';
$chksymb[$i] = $1;
$slowchk{$1} = $i;
- } elsif ( /^${T_US}([A-Za-z0-9_]+)_fast\d+${T_POST_LBL}$/o ) {
+ } elsif ( /^${T_US}([A-Za-z0-9_]+)_fast\d*${T_POST_LBL}$/o ) {
$chk[++$i] = $_;
$chkcat[$i] = 'fast';
$chksymb[$i] = $1;
$closurechk{$1} = $i;
+ } elsif ( /^${T_US}([A-Za-z0-9_]+)_srt${T_POST_LBL}$/o ) {
+ $chk[++$i] = $_;
+ $chkcat[$i] = 'srt';
+ $chksymb[$i] = $1;
+
+ $srtchk{$1} = $i;
+
+ } elsif ( /^${T_US}([A-Za-z0-9_]+)_ct${T_POST_LBL}$/o ) {
+ $chk[++$i] = $_;
+ $chkcat[$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
- } elsif ( /^${T_US}ErrorIO_call_count${T_POST_LBL}$/o # HACK!!!!
- || /^${T_US}[A-Za-z0-9_]+\.\d+${T_POST_LBL}$/o
+ } elsif ( /^${T_US}[A-Za-z0-9_]+\.\d+${T_POST_LBL}$/o
|| /^${T_US}.*_CAT${T_POST_LBL}$/o # PROF: _entryname_CAT
- || /^${T_US}CC_.*_struct${T_POST_LBL}$/o # PROF: _CC_ccident_struct
|| /^${T_US}.*_done${T_POST_LBL}$/o # PROF: _module_done
|| /^${T_US}_module_registered${T_POST_LBL}$/o # PROF: _module_registered
) {
} elsif ( /^([A-Za-z0-9_]+)\s+\.comm/ && $TargetPlatform =~ /^hppa/ ) {
$chk[++$i] = $_;
$chkcat[$i] = 'bss';
- $chksymb[$i] = $1;
+ $chksymb[$i] = '';
- } elsif ( /^${T_US}(ret_|djn_)/o ) {
+ } elsif ( $TargetPlatform =~ /^powerpc-|^rs6000-/ && /^LC\.\.([0-9]+)/ ) {
+ $chk[++$i] = $_;
+ $chkcat[$i] = 'toc';
+ $chksymb[$i] = $1;
+
+ } elsif ( /^${T_US}([A-Za-z0-9_]+)_cc(s)?${T_POST_LBL}$/o ) {
+ # all CC_ symbols go in the data section...
+ $chk[++$i] = $_;
+ $chkcat[$i] = 'data';
+ $chksymb[$i] = '';
+
+ } elsif ( /^${T_US}([A-Za-z0-9_]+)_(alt|dflt)${T_POST_LBL}$/o ) {
$chk[++$i] = $_;
$chkcat[$i] = 'misc';
$chksymb[$i] = '';
+ #$symbtmp = $1;
+ #$chksymb[$i] = $symbtmp if ($TargetPlatform =~ /^powerpc-|^rs6000-/) ; #rm andre
- } elsif ( /^${T_US}vtbl_([A-Za-z0-9_]+)${T_POST_LBL}$/o ) {
+ } elsif ( /^${T_US}([A-Za-z0-9_]+)_vtbl${T_POST_LBL}$/o ) {
$chk[++$i] = $_;
$chkcat[$i] = 'vector';
$chksymb[$i] = $1;
$vectorchk{$1} = $i;
- } elsif ( /^${T_US}([A-Za-z0-9_]+)DirectReturn${T_POST_LBL}$/o ) {
- $chk[++$i] = $_;
- $chkcat[$i] = 'direct';
- $chksymb[$i] = $1;
-
- $directchk{$1} = $i;
-
- } elsif ( /^${T_US}[A-Za-z0-9_]+_upd${T_POST_LBL}$/o ) {
+ # 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] = 'misc';
+ $chkcat[$i] = 'data';
$chksymb[$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
+ && /^[A-Za-z0-9][A-Za-z0-9_]*:/ ) {
+ # Some Solaris system headers contain function definitions (as
+ # opposed to mere prototypes), which end up in the .hc file when
+ # a Haskell module foreign imports the corresponding system
+ # functions (most notably stat()). We put them into the text
+ # segment. Note that this currently does not extend to function
+ # names starting with an underscore.
+ # - chak 7/2001
$chk[++$i] = $_;
- $chkcat[$i] = 'toss';
+ $chkcat[$i] = 'misc';
$chksymb[$i] = $1;
} elsif ( /^${T_US}[A-Za-z0-9_]/o
&& ( $TargetPlatform !~ /^hppa/ # need to avoid local labels in this case
- || /^L\$\d+$/ ) ) {
+ || ! /^L\$\d+$/ )
+ && ( $TargetPlatform !~ /^powerpc|^rs6000/ # ditto
+ || ! /^(L\.\.\d+|LT\.\..*):$/ ) ) {
local($thing);
chop($thing = $_);
print STDERR "Funny global thing?: $_"
unless $KNOWN_FUNNY_THING{$thing}
- || /^${T_US}_(PRIn|PRStart).*${T_POST_LBL}$/o # pointer reversal GC routines
- || /^${T_US}CC_.*${T_POST_LBL}$/o # PROF: _CC_ccident
- || /^${T_US}_reg.*${T_POST_LBL}$/o; # PROF: __reg<module>
+ || /^${T_US}stg_.*${T_POST_LBL}$/o # RTS internals
+ || /^${T_US}__fexp_.*${T_POST_LBL}$/o # foreign export
+ || /^${T_US}__stginit.*${T_POST_LBL}$/o # __stginit<module>
+ || /^${T_US}.*_btm${T_POST_LBL}$/o # large bitmaps
+ || /^${T_US}.*_closure_tbl${T_POST_LBL}$/o # closure tables
+ || /^_uname:/o; # x86/Solaris2
$chk[++$i] = $_;
$chkcat[$i] = 'misc';
- $chksymb[$i] = '';
+ if ($TargetPlatform =~ /^powerpc-|^rs6000-/)
+ { $chksymb[$i] = $thing; }
+ else { $chksymb[$i] = ''; };
} else { # simple line (duplicated at the top)
}
$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;
+
# the division into chunks is imperfect;
# we throw some things over the fence into the next
# chunk.
# output.
local($FIRST_MANGLABLE) = ($TargetPlatform =~ /^(alpha-|hppa|mips-)/) ? 1 : 0;
+ local($FIRST_TOSSABLE ) = ($TargetPlatform =~ /^(hppa|mips-)/) ? 1 : 0;
# print STDERR "first chunk to mangle: $FIRST_MANGLABLE\n";
# Alphas: NB: we start meddling at chunk 1, not chunk 0
# The first ".rdata" is quite magical; as of GCC 2.7.x, it
- # spits a ".quad 0" in after the v first ".rdata"; we
+ # spits a ".quad 0" in after the very first ".rdata"; we
# detect this special case (tossing the ".quad 0")!
local($magic_rdata_seen) = 0;
# HPPAs, MIPSen: also start medding at chunk 1
- for ($i = $FIRST_MANGLABLE; $i < $numchks; $i++) {
+ if ($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/ )
+#ToDo: instead of all these changes, just fix mangle_powerpc_tailjump and delete/ignore these tocs?
+ { $chk[$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+)\[TC\],(alt_\S+)/\t\.tc \1\[TC\],\.\2/;
+ $chk[$i] =~ s/\t\.tc (\S+)\[TC\],(vtbl_\S+)/\t\.tc \1\[TC\],\.\2/;
+
+ $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_TOSSABLE; $i < $numchks; $i++) {
$c = $chk[$i]; # convenience copy
# print STDERR "\nCHK $i (BEFORE) (",$chkcat[$i],"):\n", $c;
if (($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/^\tpushl\s+\%edi\n//;
+ $p =~ s/^\tpushl\s+\%esi\n//;
+ $p =~ s/^\tpushl\s+\%ebx\n//;
+ $p =~ s/^\tsubl\s+\$\d+,\s*\%esp\n//;
+ $p =~ s/^\tmovl\s+\$\d+,\s*\%eax\n\tcall\s+__alloca\n// if ($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?
$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//g;
+ $p =~ s/^\tstw? 1,\d+\(1\)\n//g; #mc
+ $p =~ s/^\tlw?z 0,0\(1\)\n//g; #mc
+ $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[^\.]/;
+ # HWL HACK: dont die, just print a warning
+ #print stderr "HWL: this should die! Prologue junk?: $p\n" if $p =~ /^\t[^\.]/
+ # && $TargetPlatform !~ /^powerpc-/; #ToDo: remove test
+ 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//;
+ $e =~ s/^\tpopl\s+\%edi\n//;
+ $e =~ s/^\tpopl\s+\%esi\n//;
+ $e =~ s/^\tpopl\s+\%edx\n//;
+ $e =~ s/^\tpopl\s+\%ecx\n//;
+ $e =~ s/^\taddl\s+\$\d+,\s*\%esp\n//;
+ $e =~ s/^\tsubl\s+\$-\d+,\s*\%esp\n//;
} elsif ($TargetPlatform =~ /^m68k-/) {
$e =~ s/^\tunlk a6\n//;
$e =~ s/^\trts\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[^\.]/;
+
+ print STDERR "WARNING: Epilogue junk?: $e\n" if $e =~ /^\t\s*[^\.\s\n]/
+ && $TargetPlatform !~ /^powerpc-/; #ToDo: remove test
# glue together what's left
$c = $r . $e;
# 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//;
}
# On Alphas, the prologue mangling is done a little later (below)
# toss all calls to __DISCARD__
- $c =~ s/^\t(call|jbsr|jal) ${T_US}__DISCARD__\n//go;
+ $c =~ s/^\t(call|jbsr|jal)\s+${T_US}__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 .= 'FUNNY#END#THING';
while ( $c =~ /${T_MOVE_DIRVS}FUNNY#END#THING/o ) {
+
$to_move = $1;
- if ( $i < ($numchks - 1)
+ # on x86 we try not to copy any directives into a literal
+ # chunk, rather we keep looking for the next real chunk. This
+ # is because we get things like
+ #
+ # .globl blah_closure
+ # .LC32
+ # .string "..."
+ # blah_closure:
+ # ...
+ #
+ if ( $TargetPlatform =~ /^(i386|sparc)/ && $to_move =~ /${T_COPY_DIRVS}/ ) {
+ $j = $i + 1;
+ while ( $j < $numchks && $chk[$j] =~ /$T_CONST_LBL/) {
+ $j++;
+ }
+ if ( $j < $numchks ) {
+ $chk[$j] = $to_move . $chk[$j];
+ }
+ }
+
+ elsif ( $i < ($numchks - 1)
&& ( $to_move =~ /${T_COPY_DIRVS}/
|| ($TargetPlatform =~ /^hppa/ && $to_move =~ /align/ && $chkcat[$i+1] eq 'literal') )) {
$chk[$i + 1] = $to_move . $chk[$i + 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.frame \$30,0,\$26,0\n\t.prologue" . $r;
+ $c .= "\t.prologue" . $r;
}
$c =~ s/FUNNY#END#THING//;
$chk[$i] = $c; # update w/ convenience copy
}
+ # open CHUNKS, ">/tmp/chunks2" or die "Cannot open /tmp/chunks2: $!\n";
+ # for (my $i = 0; $i < @chk; ++$i) { print CHUNKS "======= $i =======\n", $chk[$i] }
+ # close CHUNKS;
+
if ( $TargetPlatform =~ /^alpha-/ ) {
# print out the header stuff first
$chk[0] =~ s/^(\t\.file.*)"(ghc\d+\.c)"/$1"$ifile_root.hc"/;
if ( $chkcat[$i] eq 'misc' ) {
if ($chk[$i] ne '') {
print OUTASM $T_HDR_misc;
- &print_doctored($chk[$i], 0);
+ if ($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 '' && $chksymb[$i] !~ /ret_[a-z]/ && $chksymb[$i] !~ /djn_[a-z]/)
+ 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-/ && $printDS) {
+#ok if ($chksymb[$i] !~ /\__stginit_Main/) {
+ print OUTASM "\.csect ${chksymb[$i]}[DS]\n";
+ print OUTASM "${p}TOC[tc0], 0\n";
+#ok }
+ }
}
} elsif ( $chkcat[$i] eq 'toss' ) {
$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"
- if $TargetPlatform !~ /^mips-/; # we just don't try in that case
- } else {
+ #
+ # Using a cygnus-2.7-96q4 gcc build on hppas, the
+ # consistency chunk for ghc_cc_ID often (but not always!)
+ # gets lumped with a bunch of .IMPORT directives containing info on
+ # the code or data space nature of external symbols. We can't
+ # toss these, so once the consistency ID has been turned into
+ # a representable symbol, we substitute it for the symbol
+ # that the string was attached to in the first place (ghc_cc_ID.)
+ # (The original string is also substituted away.)
+ #
+ # This change may affect the code output on other platforms in
+ # adverse ways, hence we restrict this hack hppa targets only.
+ #
+ # -- 2/98 SOF
+ if ( $TargetPlatform =~ /^hppa/ ) {
+ $chk[$i] =~ s/^${T_US}ghc.*c_ID$TPOSTLBL/$consist/o;
+ $chk[$i] =~ s/\t$T_hsc_cc_PAT/$T_HDR_misc/o;
+ $consist = $chk[$i]; #clumsily
+ }
+ 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];
}
print OUTASM "___stg_split_marker",$chksymb[$i],"${T_POST_LBL}\n";
} 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
$chkcat[$closurechk{$symb}] = 'DONE ALREADY';
}
+ # SRT
+ if ( defined($srtchk{$symb}) ) {
+ print OUTASM $T_HDR_srt;
+ print OUTASM $chk[$srtchk{$symb}];
+ $chkcat[$srtchk{$symb}] = 'DONE ALREADY';
+ }
+
# INFO TABLE
if ( defined($infochk{$symb}) ) {
print OUTASM $T_HDR_info;
- print OUTASM &rev_tbl($symb, $chk[$infochk{$symb}], 1);
+ if ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
+ if ( !defined($slowchk{$symb}) && defined($fastchk{$symb}) ) {
+ $fastname = $chk[$fastchk{$symb}];
+ $fastname =~ s/([_A-Za-z]*_fast\d+):.*(.*\n)*/\1/;
+ $chk[$infochk{$symb}] =~ s/\.long StdErrorCode/\.long $fastname/;
+ }
+ $chk[$infochk{$symb}] =~ s/\.long ([_A-Za-z]\S+_entry)/\.long \.\1/;
+ $chk[$infochk{$symb}] =~ s/\.long ([A-Za-z]\S+_upd)/\.long \.\1/;
+ 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{$symb}] =~ /${T_DOT_WORD}\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';
}
# 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//;
+ # 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-|^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/;
+ $c =~ s/^\tcall\s+${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: $TargetPlatform\n";
}
# references to fast-entry point.
# (questionable re hppa and mips...)
print STDERR "still has jump to fast entry point:\n$c"
- if $c =~ /${T_US}${symb}_fast/; # NB: paranoia
+ if $c =~ /\b${T_US}${symb}_fast/;
}
print OUTASM $T_HDR_entry;
# FAST ENTRY POINT
if ( defined($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
- || $TargetPlatform =~ /^(m68k|sparc|i386)-/
+#OLD: || $TargetPlatform =~ /^(m68k|sparc|i386)-/
) {
print OUTASM $T_HDR_fast;
}
- &print_doctored($chk[$fastchk{$symb}], 0);
+ 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{$symb}] = 'DONE ALREADY';
}
# VECTOR TABLE
if ( defined($vectorchk{$symb}) ) {
print OUTASM $T_HDR_vector;
- print OUTASM &rev_tbl($symb, $chk[$vectorchk{$symb}], 0);
+ if ($TargetPlatform =~ /^powerpc-|^rs6000-/) {
+ if ( $symb =~ /^[A-Z]/) {
+ print OUTASM "\t\.globl \.${symb}_vtbl\n";
+ print OUTASM "\t\.globl ${symb}_vtbl\n";
+ };
+ $chk[$vectorchk{$symb}] =~ s/\.long (\S+)/\.long \.\1/g;
+ print OUTASM ".${symb}_vtbl:\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 $T_HDR_direct;
- &print_doctored($chk[$directchk{$symb}], 0);
- $chkcat[$directchk{$symb}] = 'DONE ALREADY';
} elsif ( $TargetPlatform =~ /^alpha-/ ) {
# Alphas: the commented nop is for the splitter, to ensure
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')
+ { if ( $chk[$j] !~ /\.tc UpdatePAP\[TC\]/ # not needed: always turned into a jump.
+ )
+ {
+ 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 $in_asmf\n");
local($_, $need_fallthru_patch) = @_;
if ( $TargetPlatform !~ /^i386-/
- || ! /^\t[a-z]/ ) { # no instructions in here, apparently
+ || ! /^\t[a-z]/ # no instructions in here, apparently
+ || /^${T_US}__stginit_[A-Za-z0-9_]+${T_POST_LBL}/) {
print OUTASM $_;
return;
}
local($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:
# movl $_blah,<bad-reg>
# jmp *<bad-reg>
#
- # which is easily fixed as:
- #
- # sigh! try to hack around it...
- #
+
+ s/^\tmovl\s+\$${T_US}(.*),(\%e[abcd]x)\n\tjmp\s+\*\2/\tjmp $T_US$1/g;
if ($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;
+ s/^\tmovl\s+(.*),\s*\%esi\n\tjmp\s+\*%esi\n/\tmovl $1,\%eax\n\tjmp \*\%eax\n/g;
+ s/^\tjmp\s+\*(-?\d*)\((.*\%esi.*)\)\n/\tmovl $2,\%eax\n\tjmp \*$1\(\%eax\)\n/g;
+ s/^\tjmp\s+\*\%esi\n/\tmovl \%esi,\%eax\n\tjmp \*\%eax\n/g;
die "$Pgm: (mangler) still have jump involving \%esi!\n$_"
- if /(jmp|call) .*\%esi/;
+ if /(jmp|call)\s+.*\%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;
+ s/^\tmovl\s+(.*),\s*\%edi\n\tjmp\s+\*%edi\n/\tmovl $1,\%eax\n\tjmp \*\%eax\n/g;
+ s/^\tjmp\s+\*(-?\d*\(.*\%edi.*\))\n/\tmovl $1,\%eax\n\tjmp \*\%eax\n/g;
+ s/^\tjmp\s+\*\%edi\n/\tmovl \%edi,\%eax\n\tjmp \*\%eax\n/g;
die "$Pgm: (mangler) still have jump involving \%edi!\n$_"
- if /(jmp|call) .*\%edi/;
+ if /(jmp|call)\s+.*\%edi/;
}
-#= if ($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:
+
+ # 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;
+
+ # Note funky ".=" stuff; we're *adding* to these _patch guys
if ( $StolenX86Regs <= 2
- && ( /32\(\%ebx\)/ || /\%esi/ || /^\tcmps/ ) ) { # R1 (esi)
- $entry_patch .= "\tmovl \%esi,32(\%ebx)\n";
- $exit_patch .= "\tmovl 32(\%ebx),\%esi\n";
+ && ( /[^0-9]\(\%ebx\)/ || /\%esi/ || /^\tcmps/ ) ) { # R1 (esi)
+ $entry_patch .= "\tmovl \%esi,(\%ebx)\n";
+ $exit_patch .= "\tmovl (\%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";
- $exit_patch .= "\tmovl 64(\%ebx),\%edi\n";
+ && ( /${OFFSET_Hp}\(\%ebx\)/ || /\%edi/ || /^\t(scas|cmps)/ ) ) { # Hp (edi)
+ $entry_patch .= "\tmovl \%edi,${OFFSET_Hp}(\%ebx)\n";
+ $exit_patch .= "\tmovl ${OFFSET_Hp}(\%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
+
+# Before calling GC we must set up the exit condition before the call
+# and entry condition when we come back
# fix _all_ non-local jumps:
- s/^\tjmp \*${T_X86_PRE_LLBL_PAT}/\tJMP___SL/go;
- s/^\tjmp ${T_X86_PRE_LLBL_PAT}/\tJMP___L/go;
+ s/^\tjmp\s+\*${T_X86_PRE_LLBL_PAT}/\tJMP___SL/go;
+ s/^\tjmp\s+${T_X86_PRE_LLBL_PAT}/\tJMP___L/go;
- s/^(\tjmp .*\n)/$exit_patch$1/g; # here's the fix...
+ s/^(\tjmp\s+.*\n)/$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)/;
+ if /^\t(jmp|call)\s+.*\%e(si|di)/;
} 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/;
+ if /^\t(jmp|call)\s+.*\%edi/;
}
- # final peephole fixes
-
- s/^\tmovl \%eax,36\(\%ebx\)\n\tjmp \*36\(\%ebx\)\n/\tmovl \%eax,36\(\%ebx\)\n\tjmp \*\%eax\n/;
- s/^\tmovl \$_(.*),(\%e[abcd]x)\n\tjmp \*$2/\tjmp _$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$1/g;
-
# --------------------------------------------------------
# that's it -- print it
#
\begin{code}
sub init_FUNNY_THINGS {
%KNOWN_FUNNY_THING = (
- "${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}_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${T_POST_LBL}", 1
+ # example
+ # "${T_US}stg_.*{T_POST_LBL}", 1,
);
}
\end{code}
local(@words) = ();
local($after) = '';
local(@lines) = split(/\n/, $tbl);
- local($i, $extra, $words_to_pad, $j);
+ local($i, $j);
- for ($i = 0; $i <= $#lines && $lines[$i] !~ /^\t${T_DOT_WORD}\s+/o; $i++) {
+ # Deal with the header...
+ for ($i = 0; $i <= $#lines && $lines[$i] !~ /^\t?${T_DOT_WORD}\s+/o; $i++) {
$label .= $lines[$i] . "\n",
next if $lines[$i] =~ /^[A-Za-z0-9_]+_info${T_POST_LBL}$/o
- || $lines[$i] =~ /^${T_DOT_GLOBAL}/o
- || $lines[$i] =~ /^${T_US}vtbl_\S+${T_POST_LBL}$/o;
+ || $lines[$i] =~ /${T_DOT_GLOBAL}/o
+ || $lines[$i] =~ /^${T_US}\S+_vtbl${T_POST_LBL}$/o;
$before .= $lines[$i] . "\n"; # otherwise...
}
+ # Grab the table data...
if ( $TargetPlatform !~ /^hppa/ ) {
- for ( ; $i <= $#lines && $lines[$i] =~ /^\t${T_DOT_WORD}\s+/o; $i++) {
+ for ( ; $i <= $#lines && $lines[$i] =~ /^\t?${T_DOT_WORD}\s+/o; $i++) {
push(@words, $lines[$i]);
}
} else { # hppa weirdness
- for ( ; $i <= $#lines && $lines[$i] =~ /^\s+\.(word|IMPORT)/; $i++) {
+ for ( ; $i <= $#lines && $lines[$i] =~ /^\s+(${T_DOT_WORD}|\.IMPORT)/; $i++) {
if ($lines[$i] =~ /^\s+\.IMPORT/) {
push(@imports, $lines[$i]);
} else {
}
}
- # 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"); }
+ # now throw away the first word (SRT) iff it is empty.
+ # 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)
+ }
for (; $i <= $#lines; $i++) {
$after .= $lines[$i] . "\n";
}
- # Alphas:If we have anonymous text (not part of a procedure), the
+ # Alphas: If we have anonymous text (not part of a procedure), the
# linker may complain about missing exception information. Bleh.
+ # To suppress this, we place a .ent/.end pair around the code.
+ # At the same time, we have to be careful and not enclose any leading
+ # .file/.loc directives.
if ( $TargetPlatform =~ /^alpha-/ && $label =~ /^([A-Za-z0-9_]+):$/) {
- $before = "\t.ent $1\n" . $before;
- $after .= "\t.end $1\n";
+ local ($ident) = $1;
+ $before =~ s/^((\s*\.(file|loc)\s+[^\n]*\n)*)/$1\t.ent $ident\n/;
+ $after .= "\t.end $ident\n";
+ }
+
+ # Alphas: The heroic Simon Marlow found a bug in the Digital UNIX
+ # assembler (!) wherein .quad constants inside .text sections are
+ # first narrowed to 32 bits then sign-extended back to 64 bits.
+ # This obviously screws up our 64-bit bitmaps, so we work around
+ # the bug by replacing .quad with .align 3 + .long + .long [ccshan]
+ if ( $TargetPlatform =~ /^alpha-/ ) {
+ foreach (@words) {
+ if (/^\s*\.quad\s+([-+0-9].*\S)\s*$/ && length $1 >= 10) {
+ local ($number) = $1;
+ if ($number =~ /^([-+])?(0x?)?([0-9]+)$/) {
+ local ($sign, $base, $digits) = ($1, $2, $3);
+ $base = (10, 8, 16)[length $base];
+ local ($hi, $lo) = (0, 0);
+ foreach $i (split(//, $digits)) {
+ $j = $lo * $base + $i;
+ $lo = $j % 4294967296;
+ $hi = $hi * $base + ($j - $lo) / 4294967296;
+ }
+ ($hi, $lo) = (4294967295 - $hi, 4294967296 - $lo)
+ if $sign eq "-";
+ $_ = "\t.align 3\n\t.long $lo\n\t.long $hi\n";
+ # printf STDERR "TURNING %s into 0x %08x %08x\n", $number, $hi, $lo;
+ } else {
+ print STDERR "Cannot handle \".quad $number\" in info table\n";
+ exit 1;
+ }
+ }
+ }
}
$tbl = $before
. (($TargetPlatform !~ /^hppa/) ? '' : join("\n", @imports) . "\n")
- . join("\n", (reverse @words)) . "\n"
+ . join("\n", @words) . "\n"
. $label . $after;
# print STDERR "before=$before\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_i386 {
+sub mini_mangle_asm_hppa {
local($in_asmf, $out_asmf) = @_;
- &init_TARGET_STUFF();
-
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 `$out_asmf' (to write)\n");
while (<INASM>) {
+ s/_info,DATA/_info,CODE/; # Move _info references to code space
+ s/P%_PR/_PR/;
print OUTASM;
-
- next unless
- /^${T_US}(PerformGC|StackOverflow|Yield|PerformReschedule)_wrapper${T_POST_LBL}\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 $in_asmf\n");
}
-\end{code}
-
-The HP is a major nuisance. The threaded code mangler moved info
-tables from data space to code space, but unthreaded code in the RTS
-still has references to info tables in data space. Since the HP
-linker is very precise about where symbols live, we need to patch the
-references in the unthreaded RTS as well.
+\end{code}
+
\begin{code}
-sub mini_mangle_asm_hppa {
+sub mini_mangle_asm_powerpc {
local($in_asmf, $out_asmf) = @_;
open(INASM, "< $in_asmf")
|| &tidy_up_and_die(1,"$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/;
+ 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/;
+ s/\.tc EnterNodeCode\[TC]\,EnterNodeCode\[DS\]/\.tc EnterNodeCode\[TC]\,.EnterNodeCode/; # CONC
+ s/\.tc CheckHeapCode\[TC]\,CheckHeapCode\[DS\]/\.tc CheckHeapCode\[TC]\,.CheckHeapCode/; # CONC
print OUTASM;
}
close(INASM) || &tidy_up_and_die(1,"Failed reading from $in_asmf\n");
}
-# make "require"r happy...
-1;
+sub mangle_powerpc_tailjump {
+ local($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/;
+ }
+ };
+ $c;
+}
+\end{code}
+
+\begin{code}
+sub tidy_up_and_die {
+ local($return_val, $msg) = @_;
+ print STDERR $msg;
+ exit (($return_val == 0) ? 0 : 1);
+}
\end{code}