MD5 x86_64 assembler update.

[openssl.git] / crypto / md5 / asm / md5-x86_64.pl

diff --git a/crypto/md5/asm/md5-x86_64.pl b/crypto/md5/asm/md5-x86_64.pl
index f570f050e765472b73bfcd1e6d5d9675ec7d4e36..75b4085a7ba24025d5f65f909a1247621d6818fc 100755 (executable)
--- a/crypto/md5/asm/md5-x86_64.pl
+++ b/crypto/md5/asm/md5-x86_64.pl
@@ -15,7 +15,7 @@ my $code;
 #   dst = x + ((dst + F(x,y,z) + X[k] + T_i) <<< s)
 #   %r10d = X[k_next]
 #   %r11d = z' (copy of z for the next step)
-# Each round1_step() takes about 5.71 clocks (9 instructions, 1.58 IPC)
+# Each round1_step() takes about 5.3 clocks (9 instructions, 1.7 IPC)
 sub round1_step
 {
     my ($pos, $dst, $x, $y, $z, $k_next, $T_i, $s) = @_;
@@ -37,22 +37,26 @@ EOF
 # round2_step() does:
 #   dst = x + ((dst + G(x,y,z) + X[k] + T_i) <<< s)
 #   %r10d = X[k_next]
-#   %r11d = y' (copy of y for the next step)
-# Each round2_step() takes about 6.22 clocks (9 instructions, 1.45 IPC)
+#   %r11d = z' (copy of z for the next step)
+#   %r12d = z' (copy of z for the next step)
+# Each round2_step() takes about 5.4 clocks (11 instructions, 2.0 IPC)
 sub round2_step
 {
     my ($pos, $dst, $x, $y, $z, $k_next, $T_i, $s) = @_;
     $code .= " mov     1*4(%rsi),      %r10d           /* (NEXT STEP) X[1] */\n" if ($pos == -1);
-    $code .= " mov     %ecx,           %r11d           /* (NEXT STEP) y' = %ecx */\n" if ($pos == -1);
+    $code .= " mov     %edx,           %r11d           /* (NEXT STEP) z' = %edx */\n" if ($pos == -1);
+    $code .= " mov     %edx,           %r12d           /* (NEXT STEP) z' = %edx */\n" if ($pos == -1);
     $code .= <<EOF;
-       xor     $x,             %r11d           /* x ^ ... */
+       not     %r11d                           /* not z */
        lea     $T_i($dst,%r10d),$dst           /* Const + dst + ... */
-       and     $z,             %r11d           /* z & ... */
-       xor     $y,             %r11d           /* y ^ ... */
+       and     $x,             %r12d           /* x & z */
+       and     $y,             %r11d           /* y & (not z) */
        mov     $k_next*4(%rsi),%r10d           /* (NEXT STEP) X[$k_next] */
-       add     %r11d,          $dst            /* dst += ... */
+       or      %r11d,          %r12d           /* (y & (not z)) | (x & z) */
+       mov     $y,             %r11d           /* (NEXT STEP) z' = $y */
+       add     %r12d,          $dst            /* dst += ... */
+       mov     $y,             %r12d           /* (NEXT STEP) z' = $y */
        rol     \$$s,           $dst            /* dst <<< s */
-       mov     $x,             %r11d           /* (NEXT STEP) y' = $x */
        add     $x,             $dst            /* dst += x */
 EOF
 }
@@ -61,7 +65,7 @@ EOF
 #   dst = x + ((dst + H(x,y,z) + X[k] + T_i) <<< s)
 #   %r10d = X[k_next]
 #   %r11d = y' (copy of y for the next step)
-# Each round3_step() takes about 4.26 clocks (8 instructions, 1.88 IPC)
+# Each round3_step() takes about 4.2 clocks (8 instructions, 1.9 IPC)
 sub round3_step
 {
     my ($pos, $dst, $x, $y, $z, $k_next, $T_i, $s) = @_;
@@ -83,7 +87,7 @@ EOF
 #   dst = x + ((dst + I(x,y,z) + X[k] + T_i) <<< s)
 #   %r10d = X[k_next]
 #   %r11d = not z' (copy of not z for the next step)
-# Each round4_step() takes about 5.27 clocks (9 instructions, 1.71 IPC)
+# Each round4_step() takes about 5.2 clocks (9 instructions, 1.7 IPC)
 sub round4_step
 {
     my ($pos, $dst, $x, $y, $z, $k_next, $T_i, $s) = @_;
@@ -105,19 +109,18 @@ EOF
 }
 
 my $output = shift;
-open STDOUT,">$output" or die "can't open $output: $!";
+open STDOUT,"| $^X ../perlasm/x86_64-xlate.pl $output";
 
 $code .= <<EOF;
 .text
 .align 16
 
 .globl md5_block_asm_host_order
-.type md5_block_asm_host_order,\@function
+.type md5_block_asm_host_order,\@function,3
 md5_block_asm_host_order:
        push    %rbp
        push    %rbx
        push    %r12
-       push    %r13
        push    %r14
        push    %r15
 
@@ -131,7 +134,6 @@ md5_block_asm_host_order:
        mov     1*4(%rbp),      %ebx    # ebx = ctx->B
        mov     2*4(%rbp),      %ecx    # ecx = ctx->C
        mov     3*4(%rbp),      %edx    # edx = ctx->D
-       push    %rbp                    # save ctx
        # end is 'rdi'
        # ptr is 'rsi'
        # A is 'eax'
@@ -140,10 +142,10 @@ md5_block_asm_host_order:
        # D is 'edx'
 
        cmp     %rdi,           %rsi            # cmp end with ptr
-       je      1f                              # jmp if ptr == end
+       je      .Lend                           # jmp if ptr == end
 
        # BEGIN of loop over 16-word blocks
-2:     # save old values of A, B, C, D
+.Lloop:        # save old values of A, B, C, D
        mov     %eax,           %r8d
        mov     %ebx,           %r9d
        mov     %ecx,           %r14d
@@ -226,10 +228,10 @@ $code .= <<EOF;
        # loop control
        add     \$64,           %rsi            # ptr += 64
        cmp     %rdi,           %rsi            # cmp end with ptr
-       jb      2b                              # jmp if ptr < end
+       jb      .Lloop                          # jmp if ptr < end
        # END of loop over 16-word blocks
 
-1:     pop     %rbp                            # restore ctx
+.Lend:
        mov     %eax,           0*4(%rbp)       # ctx->A = A
        mov     %ebx,           1*4(%rbp)       # ctx->B = B
        mov     %ecx,           2*4(%rbp)       # ctx->C = C
@@ -237,13 +239,13 @@ $code .= <<EOF;
 
        pop     %r15
        pop     %r14
-       pop     %r13
        pop     %r12
        pop     %rbx
        pop     %rbp
        ret
-.L_md5_block_asm_host_order_end:
-.size md5_block_asm_host_order,.L_md5_block_asm_host_order_end-md5_block_asm_host_order
+.size md5_block_asm_host_order,.-md5_block_asm_host_order
 EOF
 
 print $code;
+
+close STDOUT;