`ml64 2>&1` =~ /Version ([0-9]+)\./ &&
$1>=10);
+$shaext=1; ### set to zero if compiling for 1.0.1
+
$stitched_decrypt=0;
open OUT,"| \"$^X\" $xlate $flavour $output";
# caller should check for SSSE3 and AES-NI bits
mov OPENSSL_ia32cap_P+0(%rip),%r10d
mov OPENSSL_ia32cap_P+4(%rip),%r11
+___
+$code.=<<___ if ($shaext);
bt \$61,%r11 # check SHA bit
jc aesni_cbc_sha1_enc_shaext
___
.asciz "AESNI-CBC+SHA1 stitch for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
.align 64
___
- {{{
+ if ($shaext) {{{
($in0,$out,$len,$key,$ivp,$ctx,$inp)=("%rdi","%rsi","%rdx","%rcx","%r8","%r9","%r10");
$rounds="%r11d";
mov `($win64?56:8)`(%rsp),$inp # load 7th argument
___
$code.=<<___ if ($win64);
- lea `-8-4*16`(%rsp),%rsp
+ lea `-8-10*16`(%rsp),%rsp
movaps %xmm6,-8-10*16(%rax)
movaps %xmm7,-8-9*16(%rax)
movaps %xmm8,-8-8*16(%rax)
lea (%rsi,%r10),%r10 # epilogue label
cmp %r10,%rbx # context->Rip>=epilogue label
jae .Lcommon_seh_tail
+___
+$code.=<<___ if ($shaext);
+ lea aesni_cbc_sha1_enc_shaext(%rip),%r10
+ cmp %r10,%rbx
+ jb .Lseh_no_shaext
+ lea (%rax),%rsi
+ lea 512($context),%rdi # &context.Xmm6
+ mov \$20,%ecx
+ .long 0xa548f3fc # cld; rep movsq
+ lea 168(%rax),%rax # adjust stack pointer
+ jmp .Lcommon_seh_tail
+.Lseh_no_shaext:
+___
+$code.=<<___;
lea 96(%rax),%rsi
lea 512($context),%rdi # &context.Xmm6
mov \$20,%ecx
.rva .LSEH_end_aesni_cbc_sha1_enc_avx
.rva .LSEH_info_aesni_cbc_sha1_enc_avx
___
+$code.=<<___ if ($shaext);
+ .rva .LSEH_begin_aesni_cbc_sha1_enc_shaext
+ .rva .LSEH_end_aesni_cbc_sha1_enc_shaext
+ .rva .LSEH_info_aesni_cbc_sha1_enc_shaext
+___
$code.=<<___;
.section .xdata
.align 8
.rva ssse3_handler
.rva .Lprologue_avx,.Lepilogue_avx # HandlerData[]
___
+$code.=<<___ if ($shaext);
+.LSEH_info_aesni_cbc_sha1_enc_shaext:
+ .byte 9,0,0,0
+ .rva ssse3_handler
+ .rva .Lprologue_shaext,.Lepilogue_shaext # HandlerData[]
+___
}
####################################################################
$avx = ($1>=10) + ($1>=11);
}
+$shaext=1; ### set to zero if compiling for 1.0.1
+$avx=1 if (!$shaext && $avx);
+
open OUT,"| \"$^X\" $xlate $flavour $output";
*STDOUT=*OUT;
je .Lprobe
mov 0(%r11),%eax
mov 4(%r11),%r10
-
+___
+$code.=<<___ if ($shaext);
bt \$61,%r10 # check for SHA
jc ${func}_shaext
-
+___
+$code.=<<___;
mov %r10,%r11
shr \$32,%r11
$r++; unshift(@rndkey,pop(@rndkey));
};
+if ($shaext) {
+my $Tbl="%rax";
+
$code.=<<___;
.type ${func}_shaext,\@function,6
.align 32
${func}_shaext:
- mov %rsp,%rax
mov `($win64?56:8)`(%rsp),$inp # load 7th argument
- push %rbx
___
$code.=<<___ if ($win64);
- lea `-4*16`(%rsp),%rsp
+ lea `-8-10*16`(%rsp),%rsp
movaps %xmm6,-8-10*16(%rax)
movaps %xmm7,-8-9*16(%rax)
movaps %xmm8,-8-8*16(%rax)
movdqu $CDGH,16($ctx)
___
$code.=<<___ if ($win64);
- movaps -8-10*16(%rax),%xmm6
- movaps -8-9*16(%rax),%xmm7
- movaps -8-8*16(%rax),%xmm8
- movaps -8-7*16(%rax),%xmm9
- movaps -8-6*16(%rax),%xmm10
- movaps -8-5*16(%rax),%xmm11
- movaps -8-4*16(%rax),%xmm12
- movaps -8-3*16(%rax),%xmm13
- movaps -8-2*16(%rax),%xmm14
- movaps -8-1*16(%rax),%xmm15
+ movaps 0*16(%rsp),%xmm6
+ movaps 1*16(%rsp),%xmm7
+ movaps 2*16(%rsp),%xmm8
+ movaps 3*16(%rsp),%xmm9
+ movaps 4*16(%rsp),%xmm10
+ movaps 5*16(%rsp),%xmm11
+ movaps 6*16(%rsp),%xmm12
+ movaps 7*16(%rsp),%xmm13
+ movaps 8*16(%rsp),%xmm14
+ movaps 9*16(%rsp),%xmm15
+ lea 8+10*16(%rsp),%rsp
.Lepilogue_shaext:
___
$code.=<<___;
- mov -8(%rax),%rbx
- mov %rax,%rsp
ret
.size ${func}_shaext,.-${func}_shaext
___
+}
}}}}}
# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
cmp %r10,%rbx # context->Rip>=epilogue label
jae .Lin_prologue
___
+$code.=<<___ if ($shaext);
+ lea aesni_cbc_sha256_enc_shaext(%rip),%r10
+ cmp %r10,%rbx
+ jb .Lnot_in_shaext
+
+ lea (%rax),%rsi
+ lea 512($context),%rdi # &context.Xmm6
+ mov \$20,%ecx
+ .long 0xa548f3fc # cld; rep movsq
+ lea 168(%rax),%rax # adjust stack pointer
+ jmp .Lin_prologue
+.Lnot_in_shaext:
+___
$code.=<<___ if ($avx>1);
lea .Lavx2_shortcut(%rip),%r10
cmp %r10,%rbx # context->Rip<avx2_shortcut
.rva .LSEH_end_${func}_avx2
.rva .LSEH_info_${func}_avx2
___
+$code.=<<___ if ($shaext);
+ .rva .LSEH_begin_${func}_shaext
+ .rva .LSEH_end_${func}_shaext
+ .rva .LSEH_info_${func}_shaext
+___
$code.=<<___ if ($avx);
.section .xdata
.align 8
.rva se_handler
.rva .Lprologue_avx2,.Lepilogue_avx2 # HandlerData[]
___
+$code.=<<___ if ($shaext);
+.LSEH_info_${func}_shaext:
+ .byte 9,0,0,0
+ .rva se_handler
+ .rva .Lprologue_shaext,.Lepilogue_shaext # HandlerData[]
+___
}
####################################################################
`ml 2>&1` =~ /Version ([0-9]+)\./ &&
$1>=10); # first version supporting AVX
+$shaext=$xmm; ### set to zero if compiling for 1.0.1
+
&external_label("OPENSSL_ia32cap_P") if ($xmm);
&function_begin("sha1_block_data_order");
if ($xmm) {
- &static_label("shaext_shortcut");
+ &static_label("shaext_shortcut") if ($shaext);
&static_label("ssse3_shortcut");
&static_label("avx_shortcut") if ($ymm);
&static_label("K_XX_XX");
&mov ($C,&DWP(8,$T));
&test ($A,1<<24); # check FXSR bit
&jz (&label("x86"));
- &test ($C,1<<29); # check SHA bit
- &jnz (&label("shaext_shortcut"));
+ if ($shaext) {
+ &test ($C,1<<29); # check SHA bit
+ &jnz (&label("shaext_shortcut"));
+ }
if ($ymm) {
&and ($D,1<<28); # mask AVX bit
&and ($A,1<<30); # mask "Intel CPU" bit
&function_end("sha1_block_data_order");
if ($xmm) {
-{
+if ($shaext) {
######################################################################
# Intel SHA Extensions implementation of SHA1 update function.
#
$avx = ($1>=10) + ($1>=11);
}
+$shaext=1; ### set to zero if compiling for 1.0.1
+$avx=1 if (!$shaext && $avx);
+
open OUT,"| \"$^X\" $xlate $flavour $output";
*STDOUT=*OUT;
mov OPENSSL_ia32cap_P+8(%rip),%r10d
test \$`1<<9`,%r8d # check SSSE3 bit
jz .Lialu
-
+___
+$code.=<<___ if ($shaext);
test \$`1<<29`,%r10d # check SHA bit
jnz _shaext_shortcut
___
ret
.size sha1_block_data_order,.-sha1_block_data_order
___
-{{{
+if ($shaext) {{{
######################################################################
# Intel SHA Extensions implementation of SHA1 update function.
#
.rva .LSEH_begin_sha1_block_data_order
.rva .LSEH_end_sha1_block_data_order
.rva .LSEH_info_sha1_block_data_order
+___
+$code.=<<___ if ($shaext);
.rva .LSEH_begin_sha1_block_data_order_shaext
.rva .LSEH_end_sha1_block_data_order_shaext
.rva .LSEH_info_sha1_block_data_order_shaext
+___
+$code.=<<___;
.rva .LSEH_begin_sha1_block_data_order_ssse3
.rva .LSEH_end_sha1_block_data_order_ssse3
.rva .LSEH_info_sha1_block_data_order_ssse3
$avx = ($1>=10) + ($1>=11);
}
+$shaext=$xmm; ### set to zero if compiling for 1.0.1
+
$unroll_after = 64*4; # If pre-evicted from L1P cache first spin of
# fully unrolled loop was measured to run about
# 3-4x slower. If slowdown coefficient is N and
&jz ($unroll_after?&label("no_xmm"):&label("loop"));
&and ("ecx",1<<30); # mask "Intel CPU" bit
&and ("ebx",1<<28|1<<9); # mask AVX and SSSE3 bits
- &test ("edx",1<<29) if ($xmm); # check for SHA
- &jnz (&label("shaext")) if ($xmm);
+ &test ("edx",1<<29) if ($shaext); # check for SHA
+ &jnz (&label("shaext")) if ($shaext);
&or ("ecx","ebx");
&and ("ecx",1<<28|1<<30);
&cmp ("ecx",1<<28|1<<30);
&function_end_A();
}
if (!$i386 && $xmm) {{{
-{
+if ($shaext) {
######################################################################
# Intel SHA Extensions implementation of SHA256 update function.
#
$avx = ($1>=10) + ($1>=11);
}
+$shaext=1; ### set to zero if compiling for 1.0.1
+$avx=1 if (!$shaext && $avx);
+
open OUT,"| \"$^X\" $xlate $flavour $output";
*STDOUT=*OUT;
mov 4(%r11),%r10d
mov 8(%r11),%r11d
___
-$code.=<<___ if ($SZ==4);
+$code.=<<___ if ($SZ==4 && $shaext);
test \$`1<<29`,%r11d # check for SHA
jnz _shaext_shortcut
___
######################################################################
# SIMD code paths
#
-if ($SZ==4) {{{
+if ($SZ==4 && $shaext) {{{
######################################################################
# Intel SHA Extensions implementation of SHA256 update function.
#
.rva .LSEH_end_$func
.rva .LSEH_info_$func
___
-$code.=<<___ if ($SZ==4);
+$code.=<<___ if ($SZ==4 && $shext);
.rva .LSEH_begin_${func}_shaext
.rva .LSEH_end_${func}_shaext
.rva .LSEH_info_${func}_shaext
+___
+$code.=<<___ if ($SZ==4);
.rva .LSEH_begin_${func}_ssse3
.rva .LSEH_end_${func}_ssse3
.rva .LSEH_info_${func}_ssse3
projects / openssl / commitdiff