sha256-586.pl: full unroll to deliver additional ~16%, add Sandy Bridge-
authorAndy Polyakov <appro@openssl.org>
Mon, 28 May 2012 17:50:57 +0000 (17:50 +0000)
committerAndy Polyakov <appro@openssl.org>
Mon, 28 May 2012 17:50:57 +0000 (17:50 +0000)
specific code path.

crypto/sha/asm/sha256-586.pl

index 4aa87c8..4ca82b0 100644 (file)
@@ -1,7 +1,7 @@
 #!/usr/bin/env perl
 #
 # ====================================================================
-# Written by Andy Polyakov <appro@fy.chalmers.se> for the OpenSSL
+# Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
 # project. The module is, however, dual licensed under OpenSSL and
 # CRYPTOGAMS licenses depending on where you obtain it. For further
 # details see http://www.openssl.org/~appro/cryptogams/.
 #
 # May 2012.
 #
-# Optimization including one of Pavel Semjanov's ideas resulted in
-# ~5% improvement on AMD and Sandy Bridge, and ~15% on Atom and P4.
-# Pavel also suggested full unroll. While his code runs ~20%/13%/6%
-# faster on K8/Core2/Sandy Bridge, it's 9.6x larger and ~14%/23%/24%
-# slower on P4/Atom/Pentium...
+# Optimization including two of Pavel Semjanov's ideas, alternative
+# Maj and full unroll, resulted in ~20-25% improvement on most CPUs,
+# ~10% on Pentium and P4, ~37% on Atom. As fully unrolled loop body is
+# almost 15x larger, 8KB vs. 560B, it's fired only for longer inputs.
+# But not on P4, where it kills performance, nor Sandy Bridge, where
+# folded loop is just as fast...
 #
 # Performance in clock cycles per processed byte (less is better):
 #
 #              Pentium PIII    P4      AMD K8  Core2   SB(**)  Atom
-# gcc          46      36      41      27      26
-# icc          57      33      38      25      23      
-# x86 asm      39      31      29      19      18      19(**)  30
-# x86_64 asm(*)        -       -       21      16      16      18      25
+# gcc          46      36      41      27      26      25      50
+# icc          57      33      38      25      23      -       -
+# x86 asm(*)   39/36   27/24   30      19/15.5 18/16   16(**)  30/26
+# x86_64 asm(***)      -       17.5    15      16      17.5    23
 #
-# (*)  x86_64 assembler performance is presented for reference
+# (*)  numbers after slash are for unrolled loop, where available;
+# (**) for Sandy Bridge executing code path with ror replaced with
+#      equivalent shrd;
+# (***)        x86_64 assembly performance is presented for reference
 #      purposes.
-# (**) Sandy Bridge results can be improved by ~20% by replacing
-#      ror with equivalent shrd.
 
 $0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
 push(@INC,"${dir}","${dir}../../perlasm");
@@ -40,6 +42,8 @@ require "x86asm.pl";
 
 &asm_init($ARGV[0],"sha512-586.pl",$ARGV[$#ARGV] eq "386");
 
+$unroll_after = 1024;
+
 $A="eax";
 $E="edx";
 $T="ebx";
@@ -54,54 +58,76 @@ $Hoff=&DWP(32,"esp");
 $Xoff=&DWP(36,"esp");
 $K256="ebp";
 
-# *ror = sub { &shrd(@_[0],@_) };
+sub BODY_16_63() {
+       &mov    ($T,"ecx");                     # "ecx" is preloaded
+        &mov   ("esi",&DWP(4*(9+15+16-14),"esp"));
+       &ror    ("ecx",18-7);
+        &mov   ("edi","esi");
+       &ror    ("esi",19-17);
+        &xor   ("ecx",$T);
+       &shr    ($T,3);
+        &xor   ("esi","edi");
+       &ror    ("ecx",7);
+        &xor   ($T,"ecx");                     # T = sigma0(X[-15])
+       &ror    ("esi",17);
+        &add   ($T,&DWP(4*(9+15+16),"esp"));   # T += X[-16]
+       &shr    ("edi",10);
+        &add   ($T,&DWP(4*(9+15+16-9),"esp")); # T += X[-7]
+       #&xor   ("edi","esi")                   # sigma1(X[-2])
+       # &add  ($T,"edi");                     # T += sigma1(X[-2])
+       # &mov  (&DWP(4*(9+15),"esp"),$T);      # save X[0]
 
+       &BODY_00_15(1);
+}
 sub BODY_00_15() {
     my $in_16_63=shift;
 
        &mov    ("ecx",$E);
+        &xor   ("edi","esi")                   if ($in_16_63); # sigma1(X[-2])
         &mov   ("esi",$Foff);
        &ror    ("ecx",25-11);
         &add   ($T,"edi")                      if ($in_16_63); # T += sigma1(X[-2])
         &mov   ("edi",$Goff);
        &xor    ("ecx",$E);
         &xor   ("esi","edi");
+        &mov   ($T,&DWP(4*(9+15),"esp"))       if (!$in_16_63);
         &mov   (&DWP(4*(9+15),"esp"),$T)       if ($in_16_63); # save X[0]
        &ror    ("ecx",11-6);
         &and   ("esi",$E);
-        &mov   ($Eoff,$E);     # modulo-scheduled
+        &mov   ($Eoff,$E);             # modulo-scheduled
        &xor    ($E,"ecx");
-        &xor   ("esi","edi");  # Ch(e,f,g)
-        &add   ($T,$Hoff);     # T += h
-       &ror    ($E,6);         # Sigma1(e)
+        &xor   ("esi","edi");          # Ch(e,f,g)
+        &add   ($T,$Hoff);             # T += h
+       &ror    ($E,6);                 # Sigma1(e)
         &mov   ("ecx",$A);
-        &add   ($T,"esi");     # T += Ch(e,f,g)
+        &add   ($T,"esi");             # T += Ch(e,f,g)
 
        &ror    ("ecx",22-13);
-        &add   ($T,$E);        # T += Sigma1(e)
+        &add   ($T,$E);                # T += Sigma1(e)
         &mov   ("edi",$Boff);
        &xor    ("ecx",$A);
-        &mov   ($Aoff,$A);     # modulo-scheduled
+        &mov   ($Aoff,$A);             # modulo-scheduled
         &lea   ("esp",&DWP(-4,"esp"));
        &ror    ("ecx",13-2);
         &mov   ("esi",&DWP(0,$K256));
        &xor    ("ecx",$A);
-        &mov   ($E,$Eoff);     # e becomes d, which is e in next iteration
-        &xor   ($A,"edi");     # a ^= b
-       &ror    ("ecx",2);      # Sigma0(a)
+        &mov   ($E,$Eoff);             # e in next iteration, d in this one
+        &xor   ($A,"edi");             # a ^= b
+       &ror    ("ecx",2);              # Sigma0(a)
 
-        &add   ($T,"esi");     # T+= K[i]
-        &mov   (&DWP(0,"esp"),$A);             # (b^c) in next round
-       &add    ($E,$T);        # d += T
+        &add   ($T,"esi");             # T+= K[i]
+        &mov   (&DWP(0,"esp"),$A);     # (b^c) in next round
+       &add    ($E,$T);                # d += T
         &and   ($A,&DWP(4,"esp"));     # a &= (b^c)
-       &add    ($T,"ecx");     # T += Sigma0(a)
-        &xor   ($A,"edi");     # h = Maj(a,b,c) = Ch(a^b,c,b)
-
+       &add    ($T,"ecx");             # T += Sigma0(a)
+        &xor   ($A,"edi");             # h = Maj(a,b,c) = Ch(a^b,c,b)
+        &mov   ("ecx",&DWP(4*(9+15+16-1),"esp"))       if ($in_16_63); # preload T
        &add    ($K256,4);
-        &add   ($A,$T);        # h += T
-       &mov    ($T,&DWP(4*(9+15+16-1),"esp"))  if ($in_16_63); # preload T
+        &add   ($A,$T);                # h += T
 }
 
+&external_label("OPENSSL_ia32cap_P")           if (!$i386);
+
 &function_begin("sha256_block_data_order");
        &mov    ("esi",wparam(0));      # ctx
        &mov    ("edi",wparam(1));      # inp
@@ -122,8 +148,28 @@ sub BODY_00_15() {
        &mov    (&DWP(4,"esp"),"edi");  # inp
        &mov    (&DWP(8,"esp"),"eax");  # inp+num*128
        &mov    (&DWP(12,"esp"),"ebx"); # saved sp
+                                               if (!$i386) {
+       &picmeup("edx","OPENSSL_ia32cap_P",$K256,&label("K256"));
+       &mov    ("ecx",&DWP(0,"edx"));
+       &mov    ("edx",&DWP(4,"edx"));
+       &test   ("ecx",1<<20);          # check for P4
+       &jnz    (&label("loop"));
+       &and    ("ecx",1<<30);          # mask "Intel CPU" bit
+       &and    ("edx",1<<28);          # mask AVX bit
+       &or     ("ecx","edx");
+       &cmp    ("ecx",1<<28|1<<30);
+       &je     (&label("loop_shrd"));
+                                               if ($unroll_after) {
+       &sub    ("eax","edi");
+       &cmp    ("eax",$unroll_after);
+       &jge    (&label("unrolled"));
+                                               } }
+       &jmp    (&label("loop"));
 
-&set_label("loop",16);
+sub COMPACT_LOOP() {
+my $suffix=shift;
+
+&set_label("loop$suffix",16);
     # copy input block to stack reversing byte and dword order
     for($i=0;$i<4;$i++) {
        &mov    ("eax",&DWP($i*16+0,"edi"));
@@ -163,52 +209,36 @@ sub BODY_00_15() {
        &mov    ($Goff,"ecx");
        &mov    ($Hoff,"edi");
 
-&set_label("00_15",16);
-       &mov    ($T,&DWP(4*(9+15),"esp"));
+&set_label("00_15$suffix",16);
 
        &BODY_00_15();
 
        &cmp    ("esi",0xc19bf174);
-       &jne    (&label("00_15"));
-
-       &mov    ($T,&DWP(4*(9+15+16-1),"esp")); # preloaded in BODY_00_15(1)
-&set_label("16_63",16);
-       &mov    ("esi",$T);
-        &mov   ("ecx",&DWP(4*(9+15+16-14),"esp"));
-       &ror    ("esi",18-7);
-        &mov   ("edi","ecx");
-       &ror    ("ecx",19-17);
-        &xor   ("esi",$T);
-       &shr    ($T,3);
-        &xor   ("ecx","edi");
-       &ror    ("esi",7);
-        &xor   ($T,"esi");                     # T = sigma0(X[-15])
-       &ror    ("ecx",17);
-        &add   ($T,&DWP(4*(9+15+16),"esp"));   # T += X[-16]
-       &shr    ("edi",10);
-        &add   ($T,&DWP(4*(9+15+16-9),"esp")); # T += X[-7]
-       &xor    ("edi","ecx");                  # sigma1(X[-2])
-       # &add  ($T,"edi");                     # T += sigma1(X[-2])
-       # &mov  (&DWP(4*(9+15),"esp"),$T);      # save X[0]
+       &jne    (&label("00_15$suffix"));
 
-       &BODY_00_15(1);
+       &mov    ("ecx",&DWP(4*(9+15+16-1),"esp"));      # preloaded in BODY_00_15(1)
+       &jmp    (&label("16_63$suffix"));
+
+&set_label("16_63$suffix",16);
+
+       &BODY_16_63();
 
        &cmp    ("esi",0xc67178f2);
-       &jne    (&label("16_63"));
+       &jne    (&label("16_63$suffix"));
 
        &mov    ("esi",&DWP(4*(9+16+64)+0,"esp"));#ctx
        # &mov  ($A,$Aoff);
        &mov    ("ebx",$Boff);
-       &mov    ("ecx",$Coff);
-       &mov    ("edi",$Doff);
+       # &mov  ("edi",$Coff);
+       &mov    ("ecx",$Doff);
        &add    ($A,&DWP(0,"esi"));
        &add    ("ebx",&DWP(4,"esi"));
-       &add    ("ecx",&DWP(8,"esi"));
-       &add    ("edi",&DWP(12,"esi"));
+       &add    ("edi",&DWP(8,"esi"));
+       &add    ("ecx",&DWP(12,"esi"));
        &mov    (&DWP(0,"esi"),$A);
        &mov    (&DWP(4,"esi"),"ebx");
-       &mov    (&DWP(8,"esi"),"ecx");
-       &mov    (&DWP(12,"esi"),"edi");
+       &mov    (&DWP(8,"esi"),"edi");
+       &mov    (&DWP(12,"esi"),"ecx");
        # &mov  ($E,$Eoff);
        &mov    ("eax",$Foff);
        &mov    ("ebx",$Goff);
@@ -227,28 +257,189 @@ sub BODY_00_15() {
        &sub    ($K256,4*64);                   # rewind K
 
        &cmp    ("edi",&DWP(8,"esp"));          # are we done yet?
-       &jb     (&label("loop"));
-
+       &jb     (&label("loop$suffix"));
+}
+       &COMPACT_LOOP();
+       &mov    ("esp",&DWP(12,"esp"));         # restore sp
+&function_end_A();
+                                               if (!$i386) {
+       # ~20% improvement on Sandy Bridge
+       local *ror = sub { &shrd(@_[0],@_) };
+       &COMPACT_LOOP("_shrd");
        &mov    ("esp",&DWP(12,"esp"));         # restore sp
 &function_end_A();
+                                               }
 
 &set_label("K256",64); # Yes! I keep it in the code segment!
-       &data_word(0x428a2f98,0x71374491,0xb5c0fbcf,0xe9b5dba5);
-       &data_word(0x3956c25b,0x59f111f1,0x923f82a4,0xab1c5ed5);
-       &data_word(0xd807aa98,0x12835b01,0x243185be,0x550c7dc3);
-       &data_word(0x72be5d74,0x80deb1fe,0x9bdc06a7,0xc19bf174);
-       &data_word(0xe49b69c1,0xefbe4786,0x0fc19dc6,0x240ca1cc);
-       &data_word(0x2de92c6f,0x4a7484aa,0x5cb0a9dc,0x76f988da);
-       &data_word(0x983e5152,0xa831c66d,0xb00327c8,0xbf597fc7);
-       &data_word(0xc6e00bf3,0xd5a79147,0x06ca6351,0x14292967);
-       &data_word(0x27b70a85,0x2e1b2138,0x4d2c6dfc,0x53380d13);
-       &data_word(0x650a7354,0x766a0abb,0x81c2c92e,0x92722c85);
-       &data_word(0xa2bfe8a1,0xa81a664b,0xc24b8b70,0xc76c51a3);
-       &data_word(0xd192e819,0xd6990624,0xf40e3585,0x106aa070);
-       &data_word(0x19a4c116,0x1e376c08,0x2748774c,0x34b0bcb5);
-       &data_word(0x391c0cb3,0x4ed8aa4a,0x5b9cca4f,0x682e6ff3);
-       &data_word(0x748f82ee,0x78a5636f,0x84c87814,0x8cc70208);
-       &data_word(0x90befffa,0xa4506ceb,0xbef9a3f7,0xc67178f2);
+@K256=(        0x428a2f98,0x71374491,0xb5c0fbcf,0xe9b5dba5,
+       0x3956c25b,0x59f111f1,0x923f82a4,0xab1c5ed5,
+       0xd807aa98,0x12835b01,0x243185be,0x550c7dc3,
+       0x72be5d74,0x80deb1fe,0x9bdc06a7,0xc19bf174,
+       0xe49b69c1,0xefbe4786,0x0fc19dc6,0x240ca1cc,
+       0x2de92c6f,0x4a7484aa,0x5cb0a9dc,0x76f988da,
+       0x983e5152,0xa831c66d,0xb00327c8,0xbf597fc7,
+       0xc6e00bf3,0xd5a79147,0x06ca6351,0x14292967,
+       0x27b70a85,0x2e1b2138,0x4d2c6dfc,0x53380d13,
+       0x650a7354,0x766a0abb,0x81c2c92e,0x92722c85,
+       0xa2bfe8a1,0xa81a664b,0xc24b8b70,0xc76c51a3,
+       0xd192e819,0xd6990624,0xf40e3585,0x106aa070,
+       0x19a4c116,0x1e376c08,0x2748774c,0x34b0bcb5,
+       0x391c0cb3,0x4ed8aa4a,0x5b9cca4f,0x682e6ff3,
+       0x748f82ee,0x78a5636f,0x84c87814,0x8cc70208,
+       0x90befffa,0xa4506ceb,0xbef9a3f7,0xc67178f2     );
+&data_word(@K256);
+
+if (!$i386 && $unroll_after) {
+my @AH=($A,$K256);
+
+&set_label("unrolled",16);
+       &lea    ("esp",&DWP(-96,"esp"));
+       # copy ctx->h[0-7] to A,B,C,D,E,F,G,H on stack
+       &mov    ($AH[0],&DWP(0,"esi"));
+       &mov    ($AH[1],&DWP(4,"esi"));
+       &mov    ("ecx",&DWP(8,"esi"));
+       &mov    ("ebx",&DWP(12,"esi"));
+       #&mov   (&DWP(0,"esp"),$AH[0]);
+       &mov    (&DWP(4,"esp"),$AH[1]);
+       &xor    ($AH[1],"ecx");         # magic
+       &mov    (&DWP(8,"esp"),"ecx");
+       &mov    (&DWP(12,"esp"),"ebx");
+       &mov    ($E,&DWP(16,"esi"));    
+       &mov    ("ebx",&DWP(20,"esi"));
+       &mov    ("ecx",&DWP(24,"esi"));
+       &mov    ("esi",&DWP(28,"esi"));
+       #&mov   (&DWP(16,"esp"),$E);
+       &mov    (&DWP(20,"esp"),"ebx");
+       &mov    (&DWP(24,"esp"),"ecx");
+       &mov    (&DWP(28,"esp"),"esi");
+       &jmp    (&label("grand_loop"));
+
+&set_label("grand_loop",16);
+    # copy input block to stack reversing byte order
+    for($i=0;$i<5;$i++) {
+       &mov    ("ebx",&DWP(12*$i+0,"edi"));
+       &mov    ("ecx",&DWP(12*$i+4,"edi"));
+       &bswap  ("ebx");
+       &mov    ("esi",&DWP(12*$i+8,"edi"));
+       &bswap  ("ecx");
+       &mov    (&DWP(32+12*$i+0,"esp"),"ebx");
+       &bswap  ("esi");
+       &mov    (&DWP(32+12*$i+4,"esp"),"ecx");
+       &mov    (&DWP(32+12*$i+8,"esp"),"esi");
+    }
+       &mov    ("ebx",&DWP($i*12,"edi"));
+       &add    ("edi",64);
+       &bswap  ("ebx");
+       &mov    (&DWP(96+4,"esp"),"edi");
+       &mov    (&DWP(32+12*$i,"esp"),"ebx");
+
+    my ($a,$b,$c,$d,$e,$f,$g,$h)=(0..7);       # offsets
+    sub off { &DWP(4*(((shift)-$i)&7),"esp"); }
+
+    for ($i=0;$i<64;$i++) {
+
+      if ($i>=16) {
+       &mov    ($T,"ecx");                     # "ecx" is preloaded
+       # &mov  ("esi",&DWP(32+4*(($i+14)&15),"esp"));
+       &ror    ("ecx",18-7);
+        &mov   ("edi","esi");
+       &ror    ("esi",19-17);
+        &xor   ("ecx",$T);
+       &shr    ($T,3);
+        &xor   ("esi","edi");
+       &ror    ("ecx",7);
+        &xor   ($T,"ecx");                     # T = sigma0(X[-15])
+       &ror    ("esi",17);
+        &add   ($T,&DWP(32+4*($i&15),"esp"));  # T += X[-16]
+       &shr    ("edi",10);
+        &add   ($T,&DWP(32+4*(($i+9)&15),"esp"));      # T += X[-7]
+       #&xor   ("edi","esi")                   # sigma1(X[-2])
+       # &add  ($T,"edi");                     # T += sigma1(X[-2])
+       # &mov  (&DWP(4*(9+15),"esp"),$T);      # save X[0]
+      }
+       &mov    ("ecx",$E);
+        &xor   ("edi","esi")                   if ($i>=16);    # sigma1(X[-2])
+        &mov   ("esi",&off($f));
+       &ror    ("ecx",25-11);
+        &add   ($T,"edi")                      if ($i>=16);    # T += sigma1(X[-2])
+        &mov   ("edi",&off($g));
+       &xor    ("ecx",$E);
+        &xor   ("esi","edi");
+        &mov   ($T,&DWP(32+4*($i&15),"esp"))   if ($i<16);     # X[i]
+        &mov   (&DWP(32+4*($i&15),"esp"),$T)   if ($i>=16);    # save X[0]
+       &ror    ("ecx",11-6);
+        &and   ("esi",$E);
+        &mov   (&off($e),$E);          # modulo-scheduled
+       &xor    ($E,"ecx");
+        &xor   ("esi","edi");          # Ch(e,f,g)
+        &add   ($T,&off($h));          # T += h
+       &ror    ($E,6);                 # Sigma1(e)
+        &mov   ("ecx",$AH[0]);
+        &add   ($T,"esi");             # T += Ch(e,f,g)
+
+       &ror    ("ecx",22-13);
+        &mov   ("edi",&off($b));
+       &xor    ("ecx",$AH[0]);
+        &mov   (&off($a),$AH[0]);      # modulo-scheduled
+       &ror    ("ecx",13-2);
+        &lea   ($T,&DWP(@K256[$i],$T,$E));     # T += Sigma1(1)+K[i]
+        &mov   ($E,&off($d));          # e in next iteration, d in this one
+       &xor    ("ecx",$AH[0]);
+        &xor   ($AH[0],"edi");         # a ^= b, (b^c) in next round
+       &ror    ("ecx",2);              # Sigma0(a)
+
+       &add    ($E,$T);                # d += T
+        &and   ($AH[1],$AH[0]);        # a &= (b^c)
+       &add    ($T,"ecx");             # T += Sigma0(a)
+        &mov   ("ecx",&DWP(32+4*(($i+2)&15),"esp"))    if ($i>=15 && $i<63);
+       &xor    ($AH[1],"edi");         # h = Maj(a,b,c) = Ch(a^b,c,b)
+        &mov   ("esi",&DWP(32+4*(($i+15)&15),"esp"))   if ($i>=15 && $i<63);
+       &add    ($AH[1],$T);            # h += T
+
+       unshift(@AH,pop(@AH));          # rotate(a,h)
+    }
+       &mov    ("esi",&DWP(96,"esp")); #ctx
+                                       #&mov   ($AH[0],&DWP(0,"esp"));
+       &xor    ($AH[1],"edi");         #&mov   ($AH[1],&DWP(4,"esp"));
+                                       #&mov   ("edi", &DWP(8,"esp"));
+       &mov    ("ecx",&DWP(12,"esp"));
+       &add    ($AH[0],&DWP(0,"esi"));
+       &add    ($AH[1],&DWP(4,"esi"));
+       &add    ("edi",&DWP(8,"esi"));
+       &add    ("ecx",&DWP(12,"esi"));
+       &mov    (&DWP(0,"esi"),$AH[0]);
+       &mov    (&DWP(4,"esi"),$AH[1]);
+       &mov    (&DWP(8,"esi"),"edi");
+       &mov    (&DWP(12,"esi"),"ecx");
+        #&mov  (&DWP(0,"esp"),$AH[0]);
+        &mov   (&DWP(4,"esp"),$AH[1]);
+        &xor   ($AH[1],"edi");         # magic
+        &mov   (&DWP(8,"esp"),"edi");
+        &mov   (&DWP(12,"esp"),"ecx");
+       #&mov   ($E,&DWP(16,"esp"));
+       &mov    ("edi",&DWP(20,"esp"));
+       &mov    ("ebx",&DWP(24,"esp"));
+       &mov    ("ecx",&DWP(28,"esp"));
+       &add    ($E,&DWP(16,"esi"));
+       &add    ("edi",&DWP(20,"esi"));
+       &add    ("ebx",&DWP(24,"esi"));
+       &add    ("ecx",&DWP(28,"esi"));
+       &mov    (&DWP(16,"esi"),$E);
+       &mov    (&DWP(20,"esi"),"edi");
+       &mov    (&DWP(24,"esi"),"ebx");
+       &mov    (&DWP(28,"esi"),"ecx");
+        #&mov  (&DWP(16,"esp"),$E);
+        &mov   (&DWP(20,"esp"),"edi");
+       &mov    ("edi",&DWP(96+4,"esp"));       # inp
+        &mov   (&DWP(24,"esp"),"ebx");
+        &mov   (&DWP(28,"esp"),"ecx");
+
+       &cmp    ("edi",&DWP(96+8,"esp"));       # are we done yet?
+       &jb     (&label("grand_loop"));
+
+       &mov    ("esp",&DWP(96+12,"esp"));      # restore sp
+&function_end_A();
+}
 &function_end_B("sha256_block_data_order");
 &asciz("SHA256 block transform for x86, CRYPTOGAMS by <appro\@openssl.org>");