sha/asm/sha512-armv8.pl: add hardware-assisted SHA512 subroutine.
authorAndy Polyakov <appro@openssl.org>
Sun, 11 Feb 2018 11:29:47 +0000 (12:29 +0100)
committerAndy Polyakov <appro@openssl.org>
Mon, 12 Feb 2018 13:05:05 +0000 (14:05 +0100)
Reviewed-by: Rich Salz <rsalz@openssl.org>
crypto/sha/asm/sha512-armv8.pl

index a4755e6c364095a9cec2146357f4ae02649e6ca3..2f8644c0a24b969c257a9415caf15a77784eb6c1 100644 (file)
@@ -197,8 +197,6 @@ $code.=<<___;
 .type  $func,%function
 .align 6
 $func:
-___
-$code.=<<___   if ($SZ==4);
 #ifndef        __KERNEL__
 # ifdef        __ILP32__
        ldrsw   x16,.LOPENSSL_armcap_P
@@ -208,13 +206,19 @@ $code.=<<___      if ($SZ==4);
        adr     x17,.LOPENSSL_armcap_P
        add     x16,x16,x17
        ldr     w16,[x16]
+___
+$code.=<<___   if ($SZ==4);
        tst     w16,#ARMV8_SHA256
        b.ne    .Lv8_entry
        tst     w16,#ARMV7_NEON
        b.ne    .Lneon_entry
-#endif
+___
+$code.=<<___   if ($SZ==8);
+       tst     w16,#ARMV8_SHA512
+       b.ne    .Lv8_entry
 ___
 $code.=<<___;
+#endif
        stp     x29,x30,[sp,#-128]!
        add     x29,sp,#0
 
@@ -732,6 +736,108 @@ $code.=<<___;
 ___
 }
 
+if ($SZ==8) {
+my $Ktbl="x3";
+
+my @H = map("v$_.16b",(0..4));
+my ($fg,$de,$m9_10)=map("v$_.16b",(5..7));
+my @MSG=map("v$_.16b",(16..23));
+my ($W0,$W1)=("v24.2d","v25.2d");
+my ($AB,$CD,$EF,$GH)=map("v$_.16b",(26..29));
+
+$code.=<<___;
+#ifndef        __KERNEL__
+.type  sha512_block_armv8,%function
+.align 6
+sha512_block_armv8:
+.Lv8_entry:
+       stp             x29,x30,[sp,#-16]!
+       add             x29,sp,#0
+
+       ld1             {@MSG[0]-@MSG[3]},[$inp],#64    // load input
+       ld1             {@MSG[4]-@MSG[7]},[$inp],#64
+
+       ld1.64          {@H[0]-@H[3]},[$ctx]            // load context
+       adr             $Ktbl,.LK512
+
+       rev64           @MSG[0],@MSG[0]
+       rev64           @MSG[1],@MSG[1]
+       rev64           @MSG[2],@MSG[2]
+       rev64           @MSG[3],@MSG[3]
+       rev64           @MSG[4],@MSG[4]
+       rev64           @MSG[5],@MSG[5]
+       rev64           @MSG[6],@MSG[6]
+       rev64           @MSG[7],@MSG[7]
+       b               .Loop_hw
+
+.align 4
+.Loop_hw:
+       ld1.64          {$W0},[$Ktbl],#16
+       subs            $num,$num,#1
+       sub             x4,$inp,#128
+       orr             $AB,@H[0],@H[0]                 // offload
+       orr             $CD,@H[1],@H[1]
+       orr             $EF,@H[2],@H[2]
+       orr             $GH,@H[3],@H[3]
+       csel            $inp,$inp,x4,ne                 // conditional rewind
+___
+for($i=0;$i<32;$i++) {
+$code.=<<___;
+       add.i64         $W0,$W0,@MSG[0]
+       ld1.64          {$W1},[$Ktbl],#16
+       ext             $W0,$W0,$W0,#8
+       ext             $fg,@H[2],@H[3],#8
+       ext             $de,@H[1],@H[2],#8
+       add.i64         @H[3],@H[3],$W0                 // "T1 + H + K512[i]"
+        sha512su0      @MSG[0],@MSG[1]
+        ext            $m9_10,@MSG[4],@MSG[5],#8
+       sha512h         @H[3],$fg,$de
+        sha512su1      @MSG[0],@MSG[7],$m9_10
+       add.i64         @H[4],@H[1],@H[3]               // "D + T1"
+       sha512h2        @H[3],$H[1],@H[0]
+___
+       ($W0,$W1)=($W1,$W0);    push(@MSG,shift(@MSG));
+       @H = (@H[3],@H[0],@H[4],@H[2],@H[1]);
+}
+for(;$i<40;$i++) {
+$code.=<<___   if ($i<39);
+       ld1.64          {$W1},[$Ktbl],#16
+___
+$code.=<<___   if ($i==39);
+       sub             $Ktbl,$Ktbl,#$rounds*$SZ        // rewind
+___
+$code.=<<___;
+       add.i64         $W0,$W0,@MSG[0]
+        ld1            {@MSG[0]},[$inp],#16            // load next input
+       ext             $W0,$W0,$W0,#8
+       ext             $fg,@H[2],@H[3],#8
+       ext             $de,@H[1],@H[2],#8
+       add.i64         @H[3],@H[3],$W0                 // "T1 + H + K512[i]"
+       sha512h         @H[3],$fg,$de
+        rev64          @MSG[0],@MSG[0]
+       add.i64         @H[4],@H[1],@H[3]               // "D + T1"
+       sha512h2        @H[3],$H[1],@H[0]
+___
+       ($W0,$W1)=($W1,$W0);    push(@MSG,shift(@MSG));
+       @H = (@H[3],@H[0],@H[4],@H[2],@H[1]);
+}
+$code.=<<___;
+       add.i64         @H[0],@H[0],$AB                 // accumulate
+       add.i64         @H[1],@H[1],$CD
+       add.i64         @H[2],@H[2],$EF
+       add.i64         @H[3],@H[3],$GH
+
+       cbnz            $num,.Loop_hw
+
+       st1.64          {@H[0]-@H[3]},[$ctx]            // store context
+
+       ldr             x29,[sp],#16
+       ret
+.size  sha512_block_armv8,.-sha512_block_armv8
+#endif
+___
+}
+
 $code.=<<___;
 #ifndef        __KERNEL__
 .comm  OPENSSL_armcap_P,4,4
@@ -753,6 +859,21 @@ ___
     }
 }
 
+{   my  %opcode = (
+       "sha512h"       => 0xce608000,  "sha512h2"      => 0xce608400,
+       "sha512su0"     => 0xcec08000,  "sha512su1"     => 0xce608800   );
+
+    sub unsha512 {
+       my ($mnemonic,$arg)=@_;
+
+       $arg =~ m/[qv]([0-9]+)[^,]*,\s*[qv]([0-9]+)[^,]*(?:,\s*[qv]([0-9]+))?/o
+       &&
+       sprintf ".inst\t0x%08x\t//%s %s",
+                       $opcode{$mnemonic}|$1|($2<<5)|($3<<16),
+                       $mnemonic,$arg;
+    }
+}
+
 open SELF,$0;
 while(<SELF>) {
         next if (/^#!/);
@@ -765,12 +886,15 @@ foreach(split("\n",$code)) {
 
        s/\`([^\`]*)\`/eval($1)/ge;
 
+       s/\b(sha512\w+)\s+([qv].*)/unsha512($1,$2)/ge   or
        s/\b(sha256\w+)\s+([qv].*)/unsha256($1,$2)/ge;
 
        s/\bq([0-9]+)\b/v$1.16b/g;              # old->new registers
 
        s/\.[ui]?8(\s)/$1/;
+       s/\.\w?64\b//           and s/\.16b/\.2d/g      or
        s/\.\w?32\b//           and s/\.16b/\.4s/g;
+       m/\bext\b/              and s/\.2d/\.16b/g      or
        m/(ld|st)1[^\[]+\[0\]/  and s/\.4s/\.s/g;
 
        print $_,"\n";