# fit for Core2 and therefore the code was modified to skip cloop8 on
# this CPU.
-$output=shift;
+# Intel Westmere was observed to perform suboptimally. Adding yet
+# another movzb to cloop1 improved performance by almost 50%! Core2
+# performance is improved too, but nominally...
+
+$flavour = shift;
+$output = shift;
+if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
+
+$win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
die "can't locate x86_64-xlate.pl";
-open STDOUT,"| $^X $xlate $output";
+open STDOUT,"| $^X $xlate $flavour $output";
$dat="%rdi"; # arg1
$len="%rsi"; # arg2
jne .Lentry
ret
.Lentry:
+ push %rbx
push %r12
push %r13
+.Lprologue:
add \$8,$dat
movl -8($dat),$XX[0]#d
jnz .Lloop8
cmp \$0,$len
jne .Lloop1
-___
-$code.=<<___;
-.Lexit:
- sub \$1,$XX[0]#b
- movl $XX[0]#d,-8($dat)
- movl $YY#d,-4($dat)
+ jmp .Lexit
- pop %r13
- pop %r12
- ret
.align 16
.Lloop1:
add $TX[0]#b,$YY#b
jz .Lcloop1
cmpl \$0,260($dat)
jnz .Lcloop1
- push %rbx
jmp .Lcloop8
.align 16
.Lcloop8:
test \$-8,$len
jnz .Lcloop8
- pop %rbx
cmp \$0,$len
jne .Lcloop1
jmp .Lexit
.align 16
.Lcloop1:
add $TX[0]#b,$YY#b
+ movzb $YY#b,$YY#d
movzb ($dat,$YY),$TY#d
movb $TX[0]#b,($dat,$YY)
movb $TY#b,($dat,$XX[0])
sub \$1,$len
jnz .Lcloop1
jmp .Lexit
+
+.align 16
+.Lexit:
+ sub \$1,$XX[0]#b
+ movl $XX[0]#d,-8($dat)
+ movl $YY#d,-4($dat)
+
+ mov (%rsp),%r13
+ mov 8(%rsp),%r12
+ mov 16(%rsp),%rbx
+ add \$24,%rsp
+.Lepilogue:
+ ret
.size RC4,.-RC4
___
.size RC4_set_key,.-RC4_set_key
.globl RC4_options
-.type RC4_options,\@function,0
+.type RC4_options,\@abi-omnipotent
.align 16
RC4_options:
- .picmeup %rax
- lea .Lopts-.(%rax),%rax
+ lea .Lopts(%rip),%rax
mov OPENSSL_ia32cap_P(%rip),%edx
bt \$20,%edx
jnc .Ldone
.size RC4_options,.-RC4_options
___
+# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
+# CONTEXT *context,DISPATCHER_CONTEXT *disp)
+if ($win64) {
+$rec="%rcx";
+$frame="%rdx";
+$context="%r8";
+$disp="%r9";
+
+$code.=<<___;
+.extern __imp_RtlVirtualUnwind
+.type stream_se_handler,\@abi-omnipotent
+.align 16
+stream_se_handler:
+ push %rsi
+ push %rdi
+ push %rbx
+ push %rbp
+ push %r12
+ push %r13
+ push %r14
+ push %r15
+ pushfq
+ sub \$64,%rsp
+
+ mov 120($context),%rax # pull context->Rax
+ mov 248($context),%rbx # pull context->Rip
+
+ lea .Lprologue(%rip),%r10
+ cmp %r10,%rbx # context->Rip<prologue label
+ jb .Lin_prologue
+
+ mov 152($context),%rax # pull context->Rsp
+
+ lea .Lepilogue(%rip),%r10
+ cmp %r10,%rbx # context->Rip>=epilogue label
+ jae .Lin_prologue
+
+ lea 24(%rax),%rax
+
+ mov -8(%rax),%rbx
+ mov -16(%rax),%r12
+ mov -24(%rax),%r13
+ mov %rbx,144($context) # restore context->Rbx
+ mov %r12,216($context) # restore context->R12
+ mov %r13,224($context) # restore context->R13
+
+.Lin_prologue:
+ mov 8(%rax),%rdi
+ mov 16(%rax),%rsi
+ mov %rax,152($context) # restore context->Rsp
+ mov %rsi,168($context) # restore context->Rsi
+ mov %rdi,176($context) # restore context->Rdi
+
+ jmp .Lcommon_seh_exit
+.size stream_se_handler,.-stream_se_handler
+
+.type key_se_handler,\@abi-omnipotent
+.align 16
+key_se_handler:
+ push %rsi
+ push %rdi
+ push %rbx
+ push %rbp
+ push %r12
+ push %r13
+ push %r14
+ push %r15
+ pushfq
+ sub \$64,%rsp
+
+ mov 152($context),%rax # pull context->Rsp
+ mov 8(%rax),%rdi
+ mov 16(%rax),%rsi
+ mov %rsi,168($context) # restore context->Rsi
+ mov %rdi,176($context) # restore context->Rdi
+
+.Lcommon_seh_exit:
+
+ mov 40($disp),%rdi # disp->ContextRecord
+ mov $context,%rsi # context
+ mov \$154,%ecx # sizeof(CONTEXT)
+ .long 0xa548f3fc # cld; rep movsq
+
+ mov $disp,%rsi
+ xor %rcx,%rcx # arg1, UNW_FLAG_NHANDLER
+ mov 8(%rsi),%rdx # arg2, disp->ImageBase
+ mov 0(%rsi),%r8 # arg3, disp->ControlPc
+ mov 16(%rsi),%r9 # arg4, disp->FunctionEntry
+ mov 40(%rsi),%r10 # disp->ContextRecord
+ lea 56(%rsi),%r11 # &disp->HandlerData
+ lea 24(%rsi),%r12 # &disp->EstablisherFrame
+ mov %r10,32(%rsp) # arg5
+ mov %r11,40(%rsp) # arg6
+ mov %r12,48(%rsp) # arg7
+ mov %rcx,56(%rsp) # arg8, (NULL)
+ call *__imp_RtlVirtualUnwind(%rip)
+
+ mov \$1,%eax # ExceptionContinueSearch
+ add \$64,%rsp
+ popfq
+ pop %r15
+ pop %r14
+ pop %r13
+ pop %r12
+ pop %rbp
+ pop %rbx
+ pop %rdi
+ pop %rsi
+ ret
+.size key_se_handler,.-key_se_handler
+
+.section .pdata
+.align 4
+ .rva .LSEH_begin_RC4
+ .rva .LSEH_end_RC4
+ .rva .LSEH_info_RC4
+
+ .rva .LSEH_begin_RC4_set_key
+ .rva .LSEH_end_RC4_set_key
+ .rva .LSEH_info_RC4_set_key
+
+.section .xdata
+.align 8
+.LSEH_info_RC4:
+ .byte 9,0,0,0
+ .rva stream_se_handler
+.LSEH_info_RC4_set_key:
+ .byte 9,0,0,0
+ .rva key_se_handler
+___
+}
+
$code =~ s/#([bwd])/$1/gm;
print $code;
projects / openssl.git / blobdiff