SHA1 assembler for IA-64.
authorAndy Polyakov <appro@openssl.org>
Thu, 9 Dec 2004 11:57:38 +0000 (11:57 +0000)
committerAndy Polyakov <appro@openssl.org>
Thu, 9 Dec 2004 11:57:38 +0000 (11:57 +0000)
crypto/sha/asm/sha1-ia64.pl [new file with mode: 0644]

diff --git a/crypto/sha/asm/sha1-ia64.pl b/crypto/sha/asm/sha1-ia64.pl
new file mode 100644 (file)
index 0000000..cb9dfad
--- /dev/null
@@ -0,0 +1,549 @@
+#!/usr/bin/env perl
+#
+# ====================================================================
+# Written by Andy Polyakov <appro@fy.chalmers.se> for the OpenSSL
+# project. Rights for redistribution and usage in source and binary
+# forms are granted according to the OpenSSL license.
+# ====================================================================
+#
+# Eternal question is what's wrong with compiler generated code? The
+# trick is that it's possible to reduce the number of shifts required
+# to perform rotations by maintaining copy of 32-bit value in upper
+# bits of 64-bit register. Just follow mux2 and shrp instructions...
+# Performance under big-endian OS such as HP-UX is 179MBps*1GHz, which
+# is >50% better than HP C and >2x better than gcc. As of this moment
+# performance under little-endian OS such as Linux and Windows will be
+# a bit lower, because data has to be picked in reverse byte-order.
+# It's possible to resolve this issue by implementing third function,
+# sha1_block_asm_data_order_aligned, which would temporarily flip
+# BE field in User Mask register...
+
+$code=<<___;
+.ident  \"sha1-ia64.s, version 1.0\"
+.ident  \"IA-64 ISA artwork by Andy Polyakov <appro\@fy.chalmers.se>\"
+.explicit
+
+___
+
+
+if ($^O eq "hpux") {
+    $ADDP="addp4";
+    for (@ARGV) { $ADDP="add" if (/[\+DD|\-mlp]64/); }
+} else { $ADDP="add"; }
+for (@ARGV) {  $big_endian=1 if (/\-DB_ENDIAN/);
+               $big_endian=0 if (/\-DL_ENDIAN/);   }
+if (!defined($big_endian))
+           {   $big_endian=(unpack('L',pack('N',1))==1);   }
+
+#$human=1;
+if ($human) {  # useful for visual code auditing...
+       ($A,$B,$C,$D,$E,$T)   = ("A","B","C","D","E","T");
+       ($h0,$h1,$h2,$h3,$h4) = ("h0","h1","h2","h3","h4");
+       ($K_00_19, $K_20_39, $K_40_59, $K_60_79) =
+           (   "K_00_19","K_20_39","K_40_59","K_60_79" );
+       @X= (   "X0", "X1", "X2", "X3", "X4", "X5", "X6", "X7",
+               "X8", "X9","X10","X11","X12","X13","X14","X15"  );
+}
+else {
+       ($A,$B,$C,$D,$E,$T)   = ("loc0","loc1","loc2","loc3","loc4","loc5");
+       ($h0,$h1,$h2,$h3,$h4) = ("loc6","loc7","loc8","loc9","loc10");
+       ($K_00_19, $K_20_39, $K_40_59, $K_60_79) =
+           (   "r14", "r15", "loc11", "loc12"  );
+       @X= (   "r16", "r17", "r18", "r19", "r20", "r21", "r22", "r23",
+               "r24", "r25", "r26", "r27", "r28", "r29", "r30", "r31"  );
+}
+
+sub BODY_00_15 {
+local  *code=shift;
+local  ($i,$a,$b,$c,$d,$e,$f,$unaligned)=@_;
+
+if ($unaligned) {
+       $code.=<<___;
+{ .mmi;        ld1     tmp0=[inp],2                // MSB
+       ld1     tmp1=[tmp3],2           };;
+{ .mmi;        ld1     tmp2=[inp],2
+       ld1     $X[$i&0xf]=[tmp3],2         // LSB
+       dep     tmp1=tmp0,tmp1,8,8      };;
+{ .mii;        cmp.ne  p16,p0=r0,r0                // no misaligned prefetch
+       dep     $X[$i&0xf]=tmp2,$X[$i&0xf],8,8;;
+       dep     $X[$i&0xf]=tmp1,$X[$i&0xf],16,16        };;
+{ .mmi;        nop.m   0
+___
+       }
+elsif ($i<15) {
+       $code.=<<___;
+{ .mmi;        ld4     $X[($i+1)&0xf]=[inp],4  // prefetch
+___
+       }
+else   {
+       $code.=<<___;
+{ .mmi;        nop.m   0
+___
+       }
+if ($i<15) {
+       $code.=<<___;
+       and     tmp0=$c,$b
+       dep.z   tmp5=$a,5,27            }   // a<<5
+{ .mmi;        andcm   tmp1=$d,$b
+       add     tmp4=$e,$K_00_19        };;
+{ .mmi;        or      tmp0=tmp0,tmp1              // F_00_19(b,c,d)=(b&c)|(~b&d)
+       add     $f=tmp4,$X[$i&0xf]          // f=xi+e+K_00_19
+       extr.u  tmp1=$a,27,5            };; // a>>27
+{ .mib;        add     $f=$f,tmp0                  // f+=F_00_19(b,c,d)
+       shrp    $b=tmp6,tmp6,2          }   // b=ROTATE(b,30)
+{ .mib;        or      tmp1=tmp1,tmp5              // ROTATE(a,5)
+       mux2    tmp6=$a,0x44            };; // see b in next iteration
+{ .mii;        add     $f=$f,tmp1                  // f+=ROTATE(a,5)
+       mux2    $X[$i&0xf]=$X[$i&0xf],0x44
+       nop.i   0                       };;
+
+___
+       }
+else   {
+       $code.=<<___;
+       and     tmp0=$c,$b
+       dep.z   tmp5=$a,5,27            }   // a<<5 ;;?
+{ .mmi;        andcm   tmp1=$d,$b
+       add     tmp4=$e,$K_00_19        };;
+{ .mmi;        or      tmp0=tmp0,tmp1              // F_00_19(b,c,d)=(b&c)|(~b&d)
+       add     $f=tmp4,$X[$i&0xf]          // f=xi+e+K_00_19
+       extr.u  tmp1=$a,27,5            }   // a>>27
+{ .mmi;        xor     tmp2=$X[($i+0+1)&0xf],$X[($i+2+1)&0xf]  // +1
+       xor     tmp3=$X[($i+8+1)&0xf],$X[($i+13+1)&0xf] // +1
+       nop.i   0                       };;
+{ .mmi;        add     $f=$f,tmp0                  // f+=F_00_19(b,c,d)
+       xor     tmp2=tmp2,tmp3              // +1
+       shrp    $b=tmp6,tmp6,2          }   // b=ROTATE(b,30)
+{ .mmi; or     tmp1=tmp1,tmp5              // ROTATE(a,5)
+       mux2    tmp6=$a,0x44            };; // see b in next iteration
+{ .mii;        add     $f=$f,tmp1                  // f+=ROTATE(a,5)
+       shrp    $e=tmp2,tmp2,31             // f+1=ROTATE(x[0]^x[2]^x[8]^x[13],1)
+       mux2    $X[$i&0xf]=$X[$i&0xf],0x44  };;
+
+___
+       }
+}
+
+sub BODY_16_19 {
+local  *code=shift;
+local  ($i,$a,$b,$c,$d,$e,$f)=@_;
+
+$code.=<<___;
+{ .mmi;        mov     $X[$i&0xf]=$f               // Xupdate
+       and     tmp0=$c,$b
+       dep.z   tmp5=$a,5,27            }   // a<<5
+{ .mmi;        andcm   tmp1=$d,$b
+       add     tmp4=$e,$K_00_19        };;
+{ .mmi;        or      tmp0=tmp0,tmp1              // F_00_19(b,c,d)=(b&c)|(~b&d)
+       add     $f=$f,tmp4                  // f+=e+K_00_19
+       extr.u  tmp1=$a,27,5            }   // a>>27
+{ .mmi;        xor     tmp2=$X[($i+0+1)&0xf],$X[($i+2+1)&0xf]  // +1
+       xor     tmp3=$X[($i+8+1)&0xf],$X[($i+13+1)&0xf] // +1
+       nop.i   0                       };;
+{ .mmi;        add     $f=$f,tmp0                  // f+=F_00_19(b,c,d)
+       xor     tmp2=tmp2,tmp3              // +1
+       shrp    $b=tmp6,tmp6,2          }   // b=ROTATE(b,30)
+{ .mmi;        or      tmp1=tmp1,tmp5              // ROTATE(a,5)
+       mux2    tmp6=$a,0x44            };; // see b in next iteration
+{ .mii;        add     $f=$f,tmp1                  // f+=ROTATE(a,5)
+       shrp    $e=tmp2,tmp2,31             // f+1=ROTATE(x[0]^x[2]^x[8]^x[13],1)
+       nop.i   0                       };;
+
+___
+}
+
+sub BODY_20_39 {
+local  *code=shift;
+local  ($i,$a,$b,$c,$d,$e,$f,$Konst)=@_;
+       $Konst = $K_20_39 if (!defined($Konst));
+
+if ($i<79) {
+$code.=<<___;
+{ .mib;        mov     $X[$i&0xf]=$f               // Xupdate
+       dep.z   tmp5=$a,5,27            }   // a<<5
+{ .mib;        xor     tmp0=$c,$b
+       add     tmp4=$e,$Konst          };;
+{ .mmi;        xor     tmp0=tmp0,$d                // F_20_39(b,c,d)=b^c^d
+       add     $f=$f,tmp4                  // f+=e+K_20_39
+       extr.u  tmp1=$a,27,5            }   // a>>27
+{ .mmi;        xor     tmp2=$X[($i+0+1)&0xf],$X[($i+2+1)&0xf]  // +1
+       xor     tmp3=$X[($i+8+1)&0xf],$X[($i+13+1)&0xf] // +1
+       nop.i   0                       };;
+{ .mmi;        add     $f=$f,tmp0                  // f+=F_20_39(b,c,d)
+       xor     tmp2=tmp2,tmp3              // +1
+       shrp    $b=tmp6,tmp6,2          }   // b=ROTATE(b,30)
+{ .mmi;        or      tmp1=tmp1,tmp5              // ROTATE(a,5)
+       mux2    tmp6=$a,0x44            };; // see b in next iteration
+{ .mii;        add     $f=$f,tmp1                  // f+=ROTATE(a,5)
+       shrp    $e=tmp2,tmp2,31             // f+1=ROTATE(x[0]^x[2]^x[8]^x[13],1)
+       nop.i   0                       };;
+
+___
+}
+else {
+$code.=<<___;
+{ .mib;        mov     $X[$i&0xf]=$f               // Xupdate
+       dep.z   tmp5=$a,5,27            }   // a<<5
+{ .mib;        xor     tmp0=$c,$b
+       add     tmp4=$e,$Konst          };;
+{ .mib;        xor     tmp0=tmp0,$d                // F_20_39(b,c,d)=b^c^d
+       extr.u  tmp1=$a,27,5            }   // a>>27
+{ .mib;        add     $f=$f,tmp4                  // f+=e+K_20_39
+       add     $h1=$h1,$a              };; // wrap up
+{ .mmi;
+(p16)  ld4.s   $X[0]=[inp],4               // non-faulting prefetch
+       add     $f=$f,tmp0                  // f+=F_20_39(b,c,d)
+       shrp    $b=tmp6,tmp6,2          }   // b=ROTATE(b,30) ;;?
+{ .mmi;        or      tmp1=tmp1,tmp5              // ROTATE(a,5)
+       add     $h3=$h3,$c              };; // wrap up
+{ .mib;        add     tmp3=1,inp                  // used in unaligned codepath
+       add     $f=$f,tmp1              }   // f+=ROTATE(a,5)
+{ .mib;        add     $h2=$h2,$b                  // wrap up
+       add     $h4=$h4,$d              };; // wrap up
+
+___
+}
+}
+
+sub BODY_40_59 {
+local  *code=shift;
+local  ($i,$a,$b,$c,$d,$e,$f)=@_;
+
+$code.=<<___;
+{ .mmi;        mov     $X[$i&0xf]=$f               // Xupdate
+       and     tmp0=$c,$b
+       dep.z   tmp5=$a,5,27            }   // a<<5
+{ .mmi;        and     tmp1=$d,$b
+       add     tmp4=$e,$K_40_59        };;
+{ .mmi;        or      tmp0=tmp0,tmp1              // (b&c)|(b&d)
+       add     $f=$f,tmp4                  // f+=e+K_40_59
+       extr.u  tmp1=$a,27,5            }   // a>>27
+{ .mmi;        and     tmp4=$c,$d
+       xor     tmp2=$X[($i+0+1)&0xf],$X[($i+2+1)&0xf]  // +1
+       xor     tmp3=$X[($i+8+1)&0xf],$X[($i+13+1)&0xf] // +1
+       };;
+{ .mmi;        or      tmp1=tmp1,tmp5              // ROTATE(a,5)
+       xor     tmp2=tmp2,tmp3              // +1
+       shrp    $b=tmp6,tmp6,2          }   // b=ROTATE(b,30)
+{ .mmi;        or      tmp0=tmp0,tmp4              // F_40_59(b,c,d)=(b&c)|(b&d)|(c&d)
+       mux2    tmp6=$a,0x44            };; // see b in next iteration
+{ .mii;        add     $f=$f,tmp0                  // f+=F_40_59(b,c,d)
+       shrp    $e=tmp2,tmp2,31;;           // f+1=ROTATE(x[0]^x[2]^x[8]^x[13],1)
+       add     $f=$f,tmp1              };; // f+=ROTATE(a,5)
+
+___
+}
+sub BODY_60_79 { &BODY_20_39(@_,$K_60_79); }
+
+$code.=<<___;
+.text
+
+tmp0=r8;
+tmp1=r9;
+tmp2=r10;
+tmp3=r11;
+ctx=r32;       // in0
+inp=r33;       // in1
+
+// void sha1_block_asm_host_order(SHA_CTX *c,const void *p,size_t num);
+.global        sha1_block_asm_host_order#
+.proc  sha1_block_asm_host_order#
+.align 32
+sha1_block_asm_host_order:
+       .prologue
+       .fframe 0
+       .save   ar.pfs,r0
+       .save   ar.lc,r3
+{ .mmi;        alloc   tmp1=ar.pfs,3,15,0,0
+       $ADDP   tmp0=4,ctx
+       mov     r3=ar.lc                }
+{ .mmi;        $ADDP   ctx=0,ctx
+       $ADDP   inp=0,inp
+       mov     r2=pr                   };;
+tmp4=in2;
+tmp5=loc13;
+tmp6=loc14;
+       .body
+{ .mlx;        ld4     $h0=[ctx],8
+       movl    $K_00_19=0x5a827999     }
+{ .mlx;        ld4     $h1=[tmp0],8
+       movl    $K_20_39=0x6ed9eba1     };;
+{ .mlx;        ld4     $h2=[ctx],8
+       movl    $K_40_59=0x8f1bbcdc     }
+{ .mlx;        ld4     $h3=[tmp0]
+       movl    $K_60_79=0xca62c1d6     };;
+{ .mmi;        ld4     $h4=[ctx],-16
+       add     in2=-1,in2                  // adjust num for ar.lc
+       mov     ar.ec=1                 };;
+{ .mmi;        ld4     $X[0]=[inp],4               // prefetch
+       cmp.ne  p16,p0=r0,in2               // prefecth at loop end
+       mov     ar.lc=in2               };; // brp.loop.imp: too far
+
+.Lhtop:
+{ .mmi;        mov     $A=$h0
+       mov     $B=$h1
+       mux2    tmp6=$h1,0x44           }
+{ .mmi;        mov     $C=$h2
+       mov     $D=$h3
+       mov     $E=$h4                  };;
+
+___
+
+       &BODY_00_15(\$code, 0,$A,$B,$C,$D,$E,$T);
+       &BODY_00_15(\$code, 1,$T,$A,$B,$C,$D,$E);
+       &BODY_00_15(\$code, 2,$E,$T,$A,$B,$C,$D);
+       &BODY_00_15(\$code, 3,$D,$E,$T,$A,$B,$C);
+       &BODY_00_15(\$code, 4,$C,$D,$E,$T,$A,$B);
+       &BODY_00_15(\$code, 5,$B,$C,$D,$E,$T,$A);
+       &BODY_00_15(\$code, 6,$A,$B,$C,$D,$E,$T);
+       &BODY_00_15(\$code, 7,$T,$A,$B,$C,$D,$E);
+       &BODY_00_15(\$code, 8,$E,$T,$A,$B,$C,$D);
+       &BODY_00_15(\$code, 9,$D,$E,$T,$A,$B,$C);
+       &BODY_00_15(\$code,10,$C,$D,$E,$T,$A,$B);
+       &BODY_00_15(\$code,11,$B,$C,$D,$E,$T,$A);
+       &BODY_00_15(\$code,12,$A,$B,$C,$D,$E,$T);
+       &BODY_00_15(\$code,13,$T,$A,$B,$C,$D,$E);
+       &BODY_00_15(\$code,14,$E,$T,$A,$B,$C,$D);
+       &BODY_00_15(\$code,15,$D,$E,$T,$A,$B,$C);
+
+       &BODY_16_19(\$code,16,$C,$D,$E,$T,$A,$B);
+       &BODY_16_19(\$code,17,$B,$C,$D,$E,$T,$A);
+       &BODY_16_19(\$code,18,$A,$B,$C,$D,$E,$T);
+       &BODY_16_19(\$code,19,$T,$A,$B,$C,$D,$E);
+
+       &BODY_20_39(\$code,20,$E,$T,$A,$B,$C,$D);
+       &BODY_20_39(\$code,21,$D,$E,$T,$A,$B,$C);
+       &BODY_20_39(\$code,22,$C,$D,$E,$T,$A,$B);
+       &BODY_20_39(\$code,23,$B,$C,$D,$E,$T,$A);
+       &BODY_20_39(\$code,24,$A,$B,$C,$D,$E,$T);
+       &BODY_20_39(\$code,25,$T,$A,$B,$C,$D,$E);
+       &BODY_20_39(\$code,26,$E,$T,$A,$B,$C,$D);
+       &BODY_20_39(\$code,27,$D,$E,$T,$A,$B,$C);
+       &BODY_20_39(\$code,28,$C,$D,$E,$T,$A,$B);
+       &BODY_20_39(\$code,29,$B,$C,$D,$E,$T,$A);
+       &BODY_20_39(\$code,30,$A,$B,$C,$D,$E,$T);
+       &BODY_20_39(\$code,31,$T,$A,$B,$C,$D,$E);
+       &BODY_20_39(\$code,32,$E,$T,$A,$B,$C,$D);
+       &BODY_20_39(\$code,33,$D,$E,$T,$A,$B,$C);
+       &BODY_20_39(\$code,34,$C,$D,$E,$T,$A,$B);
+       &BODY_20_39(\$code,35,$B,$C,$D,$E,$T,$A);
+       &BODY_20_39(\$code,36,$A,$B,$C,$D,$E,$T);
+       &BODY_20_39(\$code,37,$T,$A,$B,$C,$D,$E);
+       &BODY_20_39(\$code,38,$E,$T,$A,$B,$C,$D);
+       &BODY_20_39(\$code,39,$D,$E,$T,$A,$B,$C);
+
+       &BODY_40_59(\$code,40,$C,$D,$E,$T,$A,$B);
+       &BODY_40_59(\$code,41,$B,$C,$D,$E,$T,$A);
+       &BODY_40_59(\$code,42,$A,$B,$C,$D,$E,$T);
+       &BODY_40_59(\$code,43,$T,$A,$B,$C,$D,$E);
+       &BODY_40_59(\$code,44,$E,$T,$A,$B,$C,$D);
+       &BODY_40_59(\$code,45,$D,$E,$T,$A,$B,$C);
+       &BODY_40_59(\$code,46,$C,$D,$E,$T,$A,$B);
+       &BODY_40_59(\$code,47,$B,$C,$D,$E,$T,$A);
+       &BODY_40_59(\$code,48,$A,$B,$C,$D,$E,$T);
+       &BODY_40_59(\$code,49,$T,$A,$B,$C,$D,$E);
+       &BODY_40_59(\$code,50,$E,$T,$A,$B,$C,$D);
+       &BODY_40_59(\$code,51,$D,$E,$T,$A,$B,$C);
+       &BODY_40_59(\$code,52,$C,$D,$E,$T,$A,$B);
+       &BODY_40_59(\$code,53,$B,$C,$D,$E,$T,$A);
+       &BODY_40_59(\$code,54,$A,$B,$C,$D,$E,$T);
+       &BODY_40_59(\$code,55,$T,$A,$B,$C,$D,$E);
+       &BODY_40_59(\$code,56,$E,$T,$A,$B,$C,$D);
+       &BODY_40_59(\$code,57,$D,$E,$T,$A,$B,$C);
+       &BODY_40_59(\$code,58,$C,$D,$E,$T,$A,$B);
+       &BODY_40_59(\$code,59,$B,$C,$D,$E,$T,$A);
+
+       &BODY_60_79(\$code,60,$A,$B,$C,$D,$E,$T);
+       &BODY_60_79(\$code,61,$T,$A,$B,$C,$D,$E);
+       &BODY_60_79(\$code,62,$E,$T,$A,$B,$C,$D);
+       &BODY_60_79(\$code,63,$D,$E,$T,$A,$B,$C);
+       &BODY_60_79(\$code,64,$C,$D,$E,$T,$A,$B);
+       &BODY_60_79(\$code,65,$B,$C,$D,$E,$T,$A);
+       &BODY_60_79(\$code,66,$A,$B,$C,$D,$E,$T);
+       &BODY_60_79(\$code,67,$T,$A,$B,$C,$D,$E);
+       &BODY_60_79(\$code,68,$E,$T,$A,$B,$C,$D);
+       &BODY_60_79(\$code,69,$D,$E,$T,$A,$B,$C);
+       &BODY_60_79(\$code,70,$C,$D,$E,$T,$A,$B);
+       &BODY_60_79(\$code,71,$B,$C,$D,$E,$T,$A);
+       &BODY_60_79(\$code,72,$A,$B,$C,$D,$E,$T);
+       &BODY_60_79(\$code,73,$T,$A,$B,$C,$D,$E);
+       &BODY_60_79(\$code,74,$E,$T,$A,$B,$C,$D);
+       &BODY_60_79(\$code,75,$D,$E,$T,$A,$B,$C);
+       &BODY_60_79(\$code,76,$C,$D,$E,$T,$A,$B);
+       &BODY_60_79(\$code,77,$B,$C,$D,$E,$T,$A);
+       &BODY_60_79(\$code,78,$A,$B,$C,$D,$E,$T);
+       &BODY_60_79(\$code,79,$T,$A,$B,$C,$D,$E);
+
+$code.=<<___;
+{ .mmb;        add     $h0=$h0,$E
+       nop.m   0
+       br.ctop.dptk.many       .Lhtop  };;
+.Lhend:
+{ .mmi;        add     tmp0=4,ctx
+       mov     ar.lc=r3                };;
+{ .mmi;        st4     [ctx]=$h0,8
+       st4     [tmp0]=$h1,8            };;
+{ .mmi;        st4     [ctx]=$h2,8
+       st4     [tmp0]=$h3              };;
+{ .mib;        st4     [ctx]=$h4,-16
+       mov     pr=r2,0x1ffff
+       br.ret.sptk.many        b0      };;
+.endp  sha1_block_asm_host_order#
+___
+
+
+$code.=<<___;
+// void sha1_block_asm_data_order(SHA_CTX *c,const void *p,size_t num);
+.global        sha1_block_asm_data_order#
+.proc  sha1_block_asm_data_order#
+.align 32
+sha1_block_asm_data_order:
+___
+$code.=<<___ if ($big_endian);
+{ .mmi;        and     r2=3,inp                                };;
+{ .mib;        cmp.eq  p6,p0=r0,r2
+(p6)   br.dptk.many    sha1_block_asm_host_order       };;
+___
+$code.=<<___;
+       .prologue
+       .fframe 0
+       .save   ar.pfs,r0
+       .save   ar.lc,r3
+{ .mmi;        alloc   tmp1=ar.pfs,3,15,0,0
+       $ADDP   tmp0=4,ctx
+       mov     r3=ar.lc                }
+{ .mmi;        $ADDP   ctx=0,ctx
+       $ADDP   inp=0,inp
+       mov     r2=pr                   };;
+tmp4=in2;
+tmp5=loc13;
+tmp6=loc14;
+       .body
+{ .mlx;        ld4     $h0=[ctx],8
+       movl    $K_00_19=0x5a827999     }
+{ .mlx;        ld4     $h1=[tmp0],8
+       movl    $K_20_39=0x6ed9eba1     };;
+{ .mlx;        ld4     $h2=[ctx],8
+       movl    $K_40_59=0x8f1bbcdc     }
+{ .mlx;        ld4     $h3=[tmp0]
+       movl    $K_60_79=0xca62c1d6     };;
+{ .mmi;        ld4     $h4=[ctx],-16
+       add     in2=-1,in2                  // adjust num for ar.lc
+       mov     ar.ec=1                 };;
+{ .mmi;        nop.m   0
+       add     tmp3=1,inp
+       mov     ar.lc=in2               };; // brp.loop.imp: too far
+
+.Ldtop:
+{ .mmi;        mov     $A=$h0
+       mov     $B=$h1
+       mux2    tmp6=$h1,0x44           }
+{ .mmi;        mov     $C=$h2
+       mov     $D=$h3
+       mov     $E=$h4                  };;
+
+___
+
+       &BODY_00_15(\$code, 0,$A,$B,$C,$D,$E,$T,1);
+       &BODY_00_15(\$code, 1,$T,$A,$B,$C,$D,$E,1);
+       &BODY_00_15(\$code, 2,$E,$T,$A,$B,$C,$D,1);
+       &BODY_00_15(\$code, 3,$D,$E,$T,$A,$B,$C,1);
+       &BODY_00_15(\$code, 4,$C,$D,$E,$T,$A,$B,1);
+       &BODY_00_15(\$code, 5,$B,$C,$D,$E,$T,$A,1);
+       &BODY_00_15(\$code, 6,$A,$B,$C,$D,$E,$T,1);
+       &BODY_00_15(\$code, 7,$T,$A,$B,$C,$D,$E,1);
+       &BODY_00_15(\$code, 8,$E,$T,$A,$B,$C,$D,1);
+       &BODY_00_15(\$code, 9,$D,$E,$T,$A,$B,$C,1);
+       &BODY_00_15(\$code,10,$C,$D,$E,$T,$A,$B,1);
+       &BODY_00_15(\$code,11,$B,$C,$D,$E,$T,$A,1);
+       &BODY_00_15(\$code,12,$A,$B,$C,$D,$E,$T,1);
+       &BODY_00_15(\$code,13,$T,$A,$B,$C,$D,$E,1);
+       &BODY_00_15(\$code,14,$E,$T,$A,$B,$C,$D,1);
+       &BODY_00_15(\$code,15,$D,$E,$T,$A,$B,$C,1);
+
+       &BODY_16_19(\$code,16,$C,$D,$E,$T,$A,$B);
+       &BODY_16_19(\$code,17,$B,$C,$D,$E,$T,$A);
+       &BODY_16_19(\$code,18,$A,$B,$C,$D,$E,$T);
+       &BODY_16_19(\$code,19,$T,$A,$B,$C,$D,$E);
+
+       &BODY_20_39(\$code,20,$E,$T,$A,$B,$C,$D);
+       &BODY_20_39(\$code,21,$D,$E,$T,$A,$B,$C);
+       &BODY_20_39(\$code,22,$C,$D,$E,$T,$A,$B);
+       &BODY_20_39(\$code,23,$B,$C,$D,$E,$T,$A);
+       &BODY_20_39(\$code,24,$A,$B,$C,$D,$E,$T);
+       &BODY_20_39(\$code,25,$T,$A,$B,$C,$D,$E);
+       &BODY_20_39(\$code,26,$E,$T,$A,$B,$C,$D);
+       &BODY_20_39(\$code,27,$D,$E,$T,$A,$B,$C);
+       &BODY_20_39(\$code,28,$C,$D,$E,$T,$A,$B);
+       &BODY_20_39(\$code,29,$B,$C,$D,$E,$T,$A);
+       &BODY_20_39(\$code,30,$A,$B,$C,$D,$E,$T);
+       &BODY_20_39(\$code,31,$T,$A,$B,$C,$D,$E);
+       &BODY_20_39(\$code,32,$E,$T,$A,$B,$C,$D);
+       &BODY_20_39(\$code,33,$D,$E,$T,$A,$B,$C);
+       &BODY_20_39(\$code,34,$C,$D,$E,$T,$A,$B);
+       &BODY_20_39(\$code,35,$B,$C,$D,$E,$T,$A);
+       &BODY_20_39(\$code,36,$A,$B,$C,$D,$E,$T);
+       &BODY_20_39(\$code,37,$T,$A,$B,$C,$D,$E);
+       &BODY_20_39(\$code,38,$E,$T,$A,$B,$C,$D);
+       &BODY_20_39(\$code,39,$D,$E,$T,$A,$B,$C);
+
+       &BODY_40_59(\$code,40,$C,$D,$E,$T,$A,$B);
+       &BODY_40_59(\$code,41,$B,$C,$D,$E,$T,$A);
+       &BODY_40_59(\$code,42,$A,$B,$C,$D,$E,$T);
+       &BODY_40_59(\$code,43,$T,$A,$B,$C,$D,$E);
+       &BODY_40_59(\$code,44,$E,$T,$A,$B,$C,$D);
+       &BODY_40_59(\$code,45,$D,$E,$T,$A,$B,$C);
+       &BODY_40_59(\$code,46,$C,$D,$E,$T,$A,$B);
+       &BODY_40_59(\$code,47,$B,$C,$D,$E,$T,$A);
+       &BODY_40_59(\$code,48,$A,$B,$C,$D,$E,$T);
+       &BODY_40_59(\$code,49,$T,$A,$B,$C,$D,$E);
+       &BODY_40_59(\$code,50,$E,$T,$A,$B,$C,$D);
+       &BODY_40_59(\$code,51,$D,$E,$T,$A,$B,$C);
+       &BODY_40_59(\$code,52,$C,$D,$E,$T,$A,$B);
+       &BODY_40_59(\$code,53,$B,$C,$D,$E,$T,$A);
+       &BODY_40_59(\$code,54,$A,$B,$C,$D,$E,$T);
+       &BODY_40_59(\$code,55,$T,$A,$B,$C,$D,$E);
+       &BODY_40_59(\$code,56,$E,$T,$A,$B,$C,$D);
+       &BODY_40_59(\$code,57,$D,$E,$T,$A,$B,$C);
+       &BODY_40_59(\$code,58,$C,$D,$E,$T,$A,$B);
+       &BODY_40_59(\$code,59,$B,$C,$D,$E,$T,$A);
+
+       &BODY_60_79(\$code,60,$A,$B,$C,$D,$E,$T);
+       &BODY_60_79(\$code,61,$T,$A,$B,$C,$D,$E);
+       &BODY_60_79(\$code,62,$E,$T,$A,$B,$C,$D);
+       &BODY_60_79(\$code,63,$D,$E,$T,$A,$B,$C);
+       &BODY_60_79(\$code,64,$C,$D,$E,$T,$A,$B);
+       &BODY_60_79(\$code,65,$B,$C,$D,$E,$T,$A);
+       &BODY_60_79(\$code,66,$A,$B,$C,$D,$E,$T);
+       &BODY_60_79(\$code,67,$T,$A,$B,$C,$D,$E);
+       &BODY_60_79(\$code,68,$E,$T,$A,$B,$C,$D);
+       &BODY_60_79(\$code,69,$D,$E,$T,$A,$B,$C);
+       &BODY_60_79(\$code,70,$C,$D,$E,$T,$A,$B);
+       &BODY_60_79(\$code,71,$B,$C,$D,$E,$T,$A);
+       &BODY_60_79(\$code,72,$A,$B,$C,$D,$E,$T);
+       &BODY_60_79(\$code,73,$T,$A,$B,$C,$D,$E);
+       &BODY_60_79(\$code,74,$E,$T,$A,$B,$C,$D);
+       &BODY_60_79(\$code,75,$D,$E,$T,$A,$B,$C);
+       &BODY_60_79(\$code,76,$C,$D,$E,$T,$A,$B);
+       &BODY_60_79(\$code,77,$B,$C,$D,$E,$T,$A);
+       &BODY_60_79(\$code,78,$A,$B,$C,$D,$E,$T);
+       &BODY_60_79(\$code,79,$T,$A,$B,$C,$D,$E);
+
+$code.=<<___;
+{ .mmb;        add     $h0=$h0,$E
+       nop.m   0
+       br.ctop.dptk.many       .Ldtop  };;
+.Ldend:
+{ .mmi;        add     tmp0=4,ctx
+       mov     ar.lc=r3                };;
+{ .mmi;        st4     [ctx]=$h0,8
+       st4     [tmp0]=$h1,8            };;
+{ .mmi;        st4     [ctx]=$h2,8
+       st4     [tmp0]=$h3              };;
+{ .mib;        st4     [ctx]=$h4,-16
+       mov     pr=r2,0x1ffff
+       br.ret.sptk.many        b0      };;
+.endp  sha1_block_asm_data_order#
+___
+
+print $code;