{ if ($label{$i} eq $_[0]) { $pre=''; last; } }
&out1("call",$pre.$_[0]);
}
+sub main'call_ptr { &out1p("call",@_); }
sub main'ret { &out0("ret"); }
sub main'nop { &out0("nop"); }
sub main'test { &out2("testl",@_); }
sub main'rdtsc { &out0(".byte\t0x0f,0x31"); }
sub main'halt { &out0("hlt"); }
sub main'movz { &out2("movzbl",@_); }
+sub main'neg { &out1("negl",@_); }
+sub main'cld { &out0("cld"); }
# SSE2
sub main'emms { &out0("emms"); }
sub main'movd { &out2("movd",@_); }
-sub main'movq { &out2("movq",@_); }
sub main'movdqu { &out2("movdqu",@_); }
sub main'movdqa { &out2("movdqa",@_); }
sub main'movdq2q{ &out2("movdq2q",@_); }
sub main'pxor { &out2("pxor",@_); }
sub main'por { &out2("por",@_); }
sub main'pand { &out2("pand",@_); }
+sub main'movq {
+ local($p1,$p2,$optimize)=@_;
+ if ($optimize && $p1=~/^mm[0-7]$/ && $p2=~/^mm[0-7]$/)
+ # movq between mmx registers can sink Intel CPUs
+ { push(@out,"\tpshufw\t\$0xe4,%$p2,%$p1\n"); }
+ else { &out2("movq",@_); }
+ }
# The bswapl instruction is new for the 486. Emulate if i386.
sub main'bswap
}
}
+sub main'public_label
+ {
+ $label{$_[0]}="${under}${_[0]}" if (!defined($label{$_[0]}));
+ push(@out,".globl\t$label{$_[0]}\n");
+ }
+
sub main'label
{
if (!defined($label{$_[0]}))
if ($main'elf && grep {/%[x]*mm[0-7]/i} @out) {
local($tmp);
- push (@out,"\n.comm\t${under}OPENSSL_ia32cap_P,4,4\n");
+ push (@out,"\n.section\t.bss\n");
+ push (@out,".comm\t${under}OPENSSL_ia32cap_P,4,4\n");
push (@out,".section\t.init\n");
# One can argue that it's wasteful to craft every
movl %edx,0(%edi)
popl %ebx
popl %edi
+ jmp 1f
.align $align
1:
___
}
}
+sub main'data_byte
+ {
+ push(@out,"\t.byte\t".join(',',@_)."\n");
+ }
+
sub main'data_word
{
push(@out,"\t.long\t".join(',',@_)."\n");
$tmp=<<___;
.section .init
call $under$f
- .align $align
+ jmp .Linitalign
+.align $align
+.Linitalign:
___
}
elsif ($main'coff)