modes/asm/ghashv8-armx.pl: handle lengths not divisible by 4x.
authorAndy Polyakov <appro@openssl.org>
Fri, 1 Dec 2017 21:32:48 +0000 (22:32 +0100)
committerAndy Polyakov <appro@openssl.org>
Mon, 4 Dec 2017 16:21:23 +0000 (17:21 +0100)
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/4830)

crypto/modes/asm/ghashv8-armx.pl

index 8cb2e45..7e57238 100644 (file)
@@ -255,9 +255,8 @@ $code.=<<___;
 gcm_ghash_v8:
 ___
 $code.=<<___   if ($flavour =~ /64/);
-       bic             $inc,$len,#63
-       cmp             $len,$inc
-       b.eq            .Lgcm_ghash_v8_4x
+       cmp             $len,#64
+       b.hs            .Lgcm_ghash_v8_4x
 ___
 $code.=<<___           if ($flavour !~ /64/);
        vstmdb          sp!,{d8-d15}            @ 32-bit ABI says so
@@ -421,12 +420,10 @@ gcm_ghash_v8_4x:
        vmov.i8         $xC2,#0xe1
        vld1.64         {$H3-$H4},[$Htbl]       @ load twisted H^3, ..., H^4
        vshl.u64        $xC2,$xC2,#57           @ compose 0xc2.0 constant
-#ifndef __ARMEB__
-       vrev64.8        $Xl,$Xl
-#endif
 
        vld1.64         {$I0-$j3},[$inp],#64
 #ifndef __ARMEB__
+       vrev64.8        $Xl,$Xl
        vrev64.8        $j1,$j1
        vrev64.8        $j2,$j2
        vrev64.8        $j3,$j3
@@ -459,8 +456,8 @@ gcm_ghash_v8_4x:
        veor            $Yh,$Yh,$I1
        veor            $Ym,$Ym,$j1
 
-       subs            $len,$len,#64
-       b.eq            .Ltail4x
+       subs            $len,$len,#128
+       b.lo            .Ltail4x
 
        b               .Loop4x
 
@@ -525,7 +522,7 @@ gcm_ghash_v8_4x:
         veor           $Ym,$Ym,$j1
 
        subs            $len,$len,#64
-       b.ne            .Loop4x
+       b.hs            .Loop4x
 
 .Ltail4x:
        veor            $t0,$I0,$Xl
@@ -540,6 +537,137 @@ gcm_ghash_v8_4x:
        veor            $Xh,$Xh,$Yh
        veor            $Xm,$Xm,$Ym
 
+       adds            $len,$len,#64
+       b.eq            .Ldone4x
+
+       cmp             $len,#32
+       b.lo            .Lone
+       b.eq            .Ltwo
+.Lthree:
+       vext.8          $t1,$Xl,$Xh,#8          @ Karatsuba post-processing
+       veor            $t2,$Xl,$Xh
+       veor            $Xm,$Xm,$t1
+        vld1.64        {$I0-$j2},[$inp]
+       veor            $Xm,$Xm,$t2
+#ifndef        __ARMEB__
+        vrev64.8       $j1,$j1
+        vrev64.8       $j2,$j2
+        vrev64.8       $I0,$I0
+#endif
+
+       vpmull.p64      $t2,$Xl,$xC2            @ 1st phase of reduction
+       vmov            $Xh#lo,$Xm#hi           @ Xh|Xm - 256-bit result
+       vmov            $Xm#hi,$Xl#lo           @ Xm is rotated Xl
+        vext.8         $I2,$j2,$j2,#8
+        vext.8         $I1,$j1,$j1,#8
+       veor            $Xl,$Xm,$t2
+
+        vpmull.p64     $Yl,$H,$I2              @ H·Ii+2
+        veor           $j2,$j2,$I2
+
+       vext.8          $t2,$Xl,$Xl,#8          @ 2nd phase of reduction
+       vpmull.p64      $Xl,$Xl,$xC2
+       veor            $t2,$t2,$Xh
+        vpmull2.p64    $Yh,$H,$I2
+        vpmull.p64     $Ym,$Hhl,$j2
+       veor            $Xl,$Xl,$t2
+        vpmull.p64     $j3,$H2,$I1             @ H^2·Ii+1
+        veor           $j1,$j1,$I1
+       vext.8          $Xl,$Xl,$Xl,#8
+
+        vpmull2.p64    $I1,$H2,$I1
+       veor            $t0,$I0,$Xl
+        vpmull2.p64    $j1,$Hhl,$j1
+       vext.8          $IN,$t0,$t0,#8
+
+        veor           $Yl,$Yl,$j3
+        veor           $Yh,$Yh,$I1
+        veor           $Ym,$Ym,$j1
+
+       vpmull.p64      $Xl,$H3,$IN             @ H^3·(Xi+Ii)
+       veor            $t0,$t0,$IN
+       vpmull2.p64     $Xh,$H3,$IN
+       vpmull.p64      $Xm,$H34,$t0
+
+       veor            $Xl,$Xl,$Yl
+       veor            $Xh,$Xh,$Yh
+       veor            $Xm,$Xm,$Ym
+       b               .Ldone4x
+
+.align 4
+.Ltwo:
+       vext.8          $t1,$Xl,$Xh,#8          @ Karatsuba post-processing
+       veor            $t2,$Xl,$Xh
+       veor            $Xm,$Xm,$t1
+        vld1.64        {$I0-$j1},[$inp]
+       veor            $Xm,$Xm,$t2
+#ifndef        __ARMEB__
+        vrev64.8       $j1,$j1
+        vrev64.8       $I0,$I0
+#endif
+
+       vpmull.p64      $t2,$Xl,$xC2            @ 1st phase of reduction
+       vmov            $Xh#lo,$Xm#hi           @ Xh|Xm - 256-bit result
+       vmov            $Xm#hi,$Xl#lo           @ Xm is rotated Xl
+        vext.8         $I1,$j1,$j1,#8
+       veor            $Xl,$Xm,$t2
+
+       vext.8          $t2,$Xl,$Xl,#8          @ 2nd phase of reduction
+       vpmull.p64      $Xl,$Xl,$xC2
+       veor            $t2,$t2,$Xh
+       veor            $Xl,$Xl,$t2
+       vext.8          $Xl,$Xl,$Xl,#8
+
+        vpmull.p64     $Yl,$H,$I1              @ H·Ii+1
+        veor           $j1,$j1,$I1
+
+       veor            $t0,$I0,$Xl
+       vext.8          $IN,$t0,$t0,#8
+
+        vpmull2.p64    $Yh,$H,$I1
+        vpmull.p64     $Ym,$Hhl,$j1
+
+       vpmull.p64      $Xl,$H2,$IN             @ H^2·(Xi+Ii)
+       veor            $t0,$t0,$IN
+       vpmull2.p64     $Xh,$H2,$IN
+       vpmull2.p64     $Xm,$Hhl,$t0
+
+       veor            $Xl,$Xl,$Yl
+       veor            $Xh,$Xh,$Yh
+       veor            $Xm,$Xm,$Ym
+       b               .Ldone4x
+
+.align 4
+.Lone:
+       vext.8          $t1,$Xl,$Xh,#8          @ Karatsuba post-processing
+       veor            $t2,$Xl,$Xh
+       veor            $Xm,$Xm,$t1
+        vld1.64        {$I0},[$inp]
+       veor            $Xm,$Xm,$t2
+#ifndef        __ARMEB__
+        vrev64.8       $I0,$I0
+#endif
+
+       vpmull.p64      $t2,$Xl,$xC2            @ 1st phase of reduction
+       vmov            $Xh#lo,$Xm#hi           @ Xh|Xm - 256-bit result
+       vmov            $Xm#hi,$Xl#lo           @ Xm is rotated Xl
+       veor            $Xl,$Xm,$t2
+
+       vext.8          $t2,$Xl,$Xl,#8          @ 2nd phase of reduction
+       vpmull.p64      $Xl,$Xl,$xC2
+       veor            $t2,$t2,$Xh
+       veor            $Xl,$Xl,$t2
+       vext.8          $Xl,$Xl,$Xl,#8
+
+       veor            $t0,$I0,$Xl
+       vext.8          $IN,$t0,$t0,#8
+
+       vpmull.p64      $Xl,$H,$IN
+       veor            $t0,$t0,$IN
+       vpmull2.p64     $Xh,$H,$IN
+       vpmull.p64      $Xm,$Hhl,$t0
+
+.Ldone4x:
        vext.8          $t1,$Xl,$Xh,#8          @ Karatsuba post-processing
        veor            $t2,$Xl,$Xh
        veor            $Xm,$Xm,$t1