SHA1 asm Pentium tune-up. Performance loss is not as bad anymore.
authorAndy Polyakov <appro@openssl.org>
Mon, 27 Sep 2004 09:37:03 +0000 (09:37 +0000)
committerAndy Polyakov <appro@openssl.org>
Mon, 27 Sep 2004 09:37:03 +0000 (09:37 +0000)
crypto/sha/asm/sha1-586.pl

index e00f709..9fe1a75 100644 (file)
@@ -9,7 +9,7 @@
 #
 #              compared with original  compared with Intel cc
 #              assembler impl.         generated code
-# Pentium      -25%                    +37%
+# Pentium      -16%                    +48%
 # PIII/AMD     +8%                     +16%
 # P4           +85%(!)                 +45%
 #
@@ -104,19 +104,21 @@ sub BODY_00_15
 
        &comment("00_15 $n");
 
-       &mov($tmp1,$a);
-        &mov($f,$c);                   # f to hold F_00_19(b,c,d)
+       &mov($f,$c);                    # f to hold F_00_19(b,c,d)
+        if ($n==0)  { &mov($tmp1,$a); }
+        else        { &mov($a,$tmp1); }
        &rotl($tmp1,5);                 # tmp1=ROTATE(a,5)
         &xor($f,$d);
        &and($f,$b);
-        &rotr($b,2);                   # b=ROTATE(b,30)
-       &add($tmp1,$e);                 # tmp1+=e;
-        &mov($e,&swtmp($n));           # e becomes volatile and
+        &add($tmp1,$e);                # tmp1+=e;
+       &mov($e,&swtmp($n));            # e becomes volatile and
                                        # is loaded with xi
-       &xor($f,$d);                    # f holds F_00_19(b,c,d)
+        &xor($f,$d);                   # f holds F_00_19(b,c,d)
+       &rotr($b,2);                    # b=ROTATE(b,30)
         &lea($tmp1,&DWP($K,$tmp1,$e,1));# tmp1+=K_00_19+xi
-       
-       &add($f,$tmp1);                 # f+=tmp1
+
+       if ($n==15) { &add($f,$tmp1); } # f+=tmp1
+       else        { &add($tmp1,$f); }
        }
 
 sub BODY_16_19
@@ -132,15 +134,15 @@ sub BODY_16_19
         &xor($tmp1,$d);
        &xor($f,&swtmp($n2));
         &and($tmp1,$b);                # tmp1 holds F_00_19(b,c,d)
-       &xor($f,&swtmp($n3));           # f holds xa^xb^xc^xd
-        &rotr($b,2);                   # b=ROTATE(b,30)
-       &xor($tmp1,$d);                 # tmp1=F_00_19(b,c,d)
-        &rotl($f,1);                   # f=ROATE(f,1)
+       &rotr($b,2);                    # b=ROTATE(b,30)
+        &xor($f,&swtmp($n3));          # f holds xa^xb^xc^xd
+       &rotl($f,1);                    # f=ROATE(f,1)
+        &xor($tmp1,$d);                # tmp1=F_00_19(b,c,d)
        &mov(&swtmp($n0),$f);           # xi=f
        &lea($f,&DWP($K,$f,$e,1));      # f+=K_00_19+e
         &mov($e,$a);                   # e becomes volatile
-       &add($f,$tmp1);                 # f+=F_00_19(b,c,d)
-        &rotl($e,5);                   # e=ROTATE(a,5)
+       &rotl($e,5);                    # e=ROTATE(a,5)
+        &add($f,$tmp1);                # f+=F_00_19(b,c,d)
        &add($f,$e);                    # f+=ROTATE(a,5)
        }
 
@@ -151,20 +153,20 @@ sub BODY_20_39
        &comment("20_39 $n");
        local($n0,$n1,$n2,$n3,$np)=&Na($n);
 
-       &mov($f,&swtmp($n0));           # f to hold Xupdate(xi,xa,xb,xc,xd)
-        &mov($tmp1,$b);                # tmp1 to hold F_20_39(b,c,d)
-       &xor($f,&swtmp($n1));
-        &rotr($b,2);                   # b=ROTATE(b,30)
-       &xor($f,&swtmp($n2));
-        &xor($tmp1,$c);
-       &xor($f,&swtmp($n3));           # f holds xa^xb^xc^xd
-        &xor($tmp1,$d);                # tmp1 holds F_20_39(b,c,d)
+       &mov($tmp1,$b);                 # tmp1 to hold F_20_39(b,c,d)
+        &mov($f,&swtmp($n0));          # f to hold Xupdate(xi,xa,xb,xc,xd)
+       &rotr($b,2);                    # b=ROTATE(b,30)
+        &xor($f,&swtmp($n1));
+       &xor($tmp1,$c);
+        &xor($f,&swtmp($n2));
+       &xor($tmp1,$d);                 # tmp1 holds F_20_39(b,c,d)
+        &xor($f,&swtmp($n3));          # f holds xa^xb^xc^xd
        &rotl($f,1);                    # f=ROTATE(f,1)
+        &add($tmp1,$e);
        &mov(&swtmp($n0),$f);           # xi=f
-       &lea($f,&DWP($K,$f,$e,1));      # f+=K_20_39+e
         &mov($e,$a);                   # e becomes volatile
        &rotl($e,5);                    # e=ROTATE(a,5)
-        &add($f,$tmp1);                # f+=F_20_39(b,c,d)
+        &lea($f,&DWP($K,$f,$tmp1,1));  # f+=K_20_39+e
        &add($f,$e);                    # f+=ROTATE(a,5)
        }
 
@@ -176,14 +178,17 @@ sub BODY_40_59
        local($n0,$n1,$n2,$n3,$np)=&Na($n);
 
        &mov($f,&swtmp($n0));           # f to hold Xupdate(xi,xa,xb,xc,xd)
+        &mov($tmp1,&swtmp($n1));
+       &xor($f,$tmp1);
+        &mov($tmp1,&swtmp($n2));
+       &xor($f,$tmp1);
+        &mov($tmp1,&swtmp($n3));
+       &xor($f,$tmp1);                 # f holds xa^xb^xc^xd
         &mov($tmp1,$b);                # tmp1 to hold F_40_59(b,c,d)
-       &xor($f,&swtmp($n1));
-        &or($tmp1,$c);
-       &xor($f,&swtmp($n2));
-        &and($tmp1,$d);
-       &xor($f,&swtmp($n3));           # f holds xa^xb^xc^xd
        &rotl($f,1);                    # f=ROTATE(f,1)
+        &or($tmp1,$c);
        &mov(&swtmp($n0),$f);           # xi=f
+        &and($tmp1,$d);
        &lea($f,&DWP($K,$f,$e,1));      # f+=K_40_59+e
         &mov($e,$b);                   # e becomes volatile and is used
                                        # to calculate F_40_59(b,c,d)
@@ -192,8 +197,8 @@ sub BODY_40_59
        &or($tmp1,$e);                  # tmp1 holds F_40_59(b,c,d)             
         &mov($e,$a);
        &rotl($e,5);                    # e=ROTATE(a,5)
-       &add($tmp1,$e);                 # tmp1+=ROTATE(a,5)
-       &add($f,$tmp1);                 # f+=tmp1;
+        &add($f,$tmp1);                # f+=tmp1;
+       &add($f,$e);                    # f+=ROTATE(a,5)
        }
 
 sub BODY_60_79