#!/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.
+# 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/.
# ====================================================================
#
# SHA256/512_Transform for Itanium.
# because on Itanium 1 stall on MM result is accompanied by
# pipeline flush, which takes 6 cycles:-(
#
-# Resulting performance numbers for 900MHz Itanium 2 system:
+# June 2012
#
-# The 'numbers' are in 1000s of bytes per second processed.
-# type 16 bytes 64 bytes 256 bytes 1024 bytes 8192 bytes
-# sha1(*) 6210.14k 20376.30k 52447.83k 85870.05k 105478.12k
-# sha256 7476.45k 20572.05k 41538.34k 56062.29k 62093.18k
-# sha512 4996.56k 20026.28k 47597.20k 85278.79k 111501.31k
+# Improve performance by 15-20%. Note about "rules of engagement"
+# above. Contemporary cores are equipped with additional shifter,
+# so that they should perform even better than below, presumably
+# by ~10%.
#
-# (*) SHA1 numbers are for HP-UX compiler and are presented purely
-# for reference purposes. I bet it can improved too...
+######################################################################
+# Current performance in cycles per processed byte for Itanium 2
+# pre-9000 series [little-endian] system:
+#
+# SHA1(*) 5.7
+# SHA256 12.6
+# SHA512 6.7
+#
+# (*) SHA1 result is presented purely for reference purposes.
#
# To generate code, pass the file name with either 256 or 512 in its
# name and compiler flags.
$ADD="add";
$SHRU="shr.u";
$TABLE="K512";
- $func="sha512_block";
+ $func="sha512_block_data_order";
@Sigma0=(28,34,39);
@Sigma1=(14,18,41);
@sigma0=(1, 8, 7);
$ADD="padd4";
$SHRU="pshr4.u";
$TABLE="K256";
- $func="sha256_block";
+ $func="sha256_block_data_order";
@Sigma0=( 2,13,22);
@Sigma1=( 6,11,25);
@sigma0=( 7,18, 3);
{ $big_endian=(unpack('L',pack('N',1))==1); }
$code=<<___;
-.ident \"$output, version 1.0\"
-.ident \"IA-64 ISA artwork by Andy Polyakov <appro\@fy.chalmers.se>\"
+.ident \"$output, version 2.0\"
+.ident \"IA-64 ISA artwork by Andy Polyakov <appro\@openssl.org>\"
.explicit
.text
+pfssave=r2;
+lcsave=r3;
prsave=r14;
K=r15;
-A=r16; B=r17; C=r18; D=r19;
-E=r20; F=r21; G=r22; H=r23;
+A_=r16; B_=r17; C_=r18; D_=r19;
+E_=r20; F_=r21; G_=r22; H_=r23;
T1=r24; T2=r25;
s0=r26; s1=r27; t0=r28; t1=r29;
Ktbl=r30;
ctx=r31; // 1st arg
-input=r48; // 2nd arg
-num=r49; // 3rd arg
-sgm0=r50; sgm1=r51; // small constants
+input=r56; // 2nd arg
+num=r57; // 3rd arg
+sgm0=r58; sgm1=r59; // small constants
// void $func (SHA_CTX *ctx, const void *in,size_t num[,int host])
.global $func#
.proc $func#
.align 32
+.skip 16
$func:
.prologue
- .save ar.pfs,r2
-{ .mmi; alloc r2=ar.pfs,3,17,0,16
+ .save ar.pfs,pfssave
+{ .mmi; alloc pfssave=ar.pfs,3,25,0,24
$ADDP ctx=0,r32 // 1st arg
- .save ar.lc,r3
- mov r3=ar.lc }
+ .save ar.lc,lcsave
+ mov lcsave=ar.lc }
{ .mmi; $ADDP input=0,r33 // 2nd arg
- addl Ktbl=\@ltoff($TABLE#),gp
+ mov num=r34 // 3rd arg
.save pr,prsave
mov prsave=pr };;
.body
-{ .mii; ld8 Ktbl=[Ktbl]
- mov num=r34 };; // 3rd arg
-
{ .mib; add r8=0*$SZ,ctx
- add r9=1*$SZ,ctx
- brp.loop.imp .L_first16,.L_first16_ctop
- }
+ add r9=1*$SZ,ctx }
{ .mib; add r10=2*$SZ,ctx
- add r11=3*$SZ,ctx
- brp.loop.imp .L_rest,.L_rest_ctop
- };;
-// load A-H
-{ .mmi; $LDW A=[r8],4*$SZ
- $LDW B=[r9],4*$SZ
- mov sgm0=$sigma0[2] }
-{ .mmi; $LDW C=[r10],4*$SZ
- $LDW D=[r11],4*$SZ
- mov sgm1=$sigma1[2] };;
-{ .mmi; $LDW E=[r8]
- $LDW F=[r9] }
-{ .mmi; $LDW G=[r10]
- $LDW H=[r11]
- cmp.ne p15,p14=0,r35 };; // used in sha256_block
+ add r11=3*$SZ,ctx };;
+// load A-H
+.Lpic_point:
+{ .mmi; $LDW A_=[r8],4*$SZ
+ $LDW B_=[r9],4*$SZ
+ mov Ktbl=ip }
+{ .mmi; $LDW C_=[r10],4*$SZ
+ $LDW D_=[r11],4*$SZ
+ mov sgm0=$sigma0[2] };;
+{ .mmi; $LDW E_=[r8]
+ $LDW F_=[r9]
+ add Ktbl=($TABLE#-.Lpic_point),Ktbl }
+{ .mmi; $LDW G_=[r10]
+ $LDW H_=[r11]
+ cmp.ne p0,p16=0,r0 };;
+___
+$code.=<<___ if ($BITS==64);
+{ .mii; and r8=7,input
+ and input=~7,input;;
+ cmp.eq p9,p0=1,r8 }
+{ .mmi; cmp.eq p10,p0=2,r8
+ cmp.eq p11,p0=3,r8
+ cmp.eq p12,p0=4,r8 }
+{ .mmi; cmp.eq p13,p0=5,r8
+ cmp.eq p14,p0=6,r8
+ cmp.eq p15,p0=7,r8 };;
+___
+$code.=<<___;
.L_outer:
-{ .mii; mov ar.lc=15
- mov ar.ec=1 };;
+.rotr R[8],X[16]
+A=R[0]; B=R[1]; C=R[2]; D=R[3]; E=R[4]; F=R[5]; G=R[6]; H=R[7]
+{ .mmi; ld1 X[15]=[input],$SZ // eliminated in sha512
+ mov A=A_
+ mov ar.lc=14 }
+{ .mmi; mov B=B_
+ mov C=C_
+ mov D=D_ }
+{ .mmi; mov E=E_
+ mov F=F_
+ mov ar.ec=2 };;
+{ .mmi; mov G=G_
+ mov H=H_
+ mov sgm1=$sigma1[2] }
+{ .mib; mov r8=0
+ add r9=1-$SZ,input
+ brp.loop.imp .L_first16,.L_first16_end-16 };;
+___
+$t0="A", $t1="E", $code.=<<___ if ($BITS==64);
+// in sha512 case I load whole X[16] at once and take care of alignment...
+{ .mmi; add r8=1*$SZ,input
+ add r9=2*$SZ,input
+ add r10=3*$SZ,input };;
+{ .mmb; $LDW X[15]=[input],4*$SZ
+ $LDW X[14]=[r8],4*$SZ
+(p9) br.cond.dpnt.many .L1byte };;
+{ .mmb; $LDW X[13]=[r9],4*$SZ
+ $LDW X[12]=[r10],4*$SZ
+(p10) br.cond.dpnt.many .L2byte };;
+{ .mmb; $LDW X[11]=[input],4*$SZ
+ $LDW X[10]=[r8],4*$SZ
+(p11) br.cond.dpnt.many .L3byte };;
+{ .mmb; $LDW X[ 9]=[r9],4*$SZ
+ $LDW X[ 8]=[r10],4*$SZ
+(p12) br.cond.dpnt.many .L4byte };;
+{ .mmb; $LDW X[ 7]=[input],4*$SZ
+ $LDW X[ 6]=[r8],4*$SZ
+(p13) br.cond.dpnt.many .L5byte };;
+{ .mmb; $LDW X[ 5]=[r9],4*$SZ
+ $LDW X[ 4]=[r10],4*$SZ
+(p14) br.cond.dpnt.many .L6byte };;
+{ .mmb; $LDW X[ 3]=[input],4*$SZ
+ $LDW X[ 2]=[r8],4*$SZ
+(p15) br.cond.dpnt.many .L7byte };;
+{ .mmb; $LDW X[ 1]=[r9],4*$SZ
+ $LDW X[ 0]=[r10],4*$SZ }
+{ .mib; mov r8=0
+ mux1 X[15]=X[15],\@rev // eliminated on big-endian
+ br.many .L_first16 };;
+.L1byte:
+{ .mmi; $LDW X[13]=[r9],4*$SZ
+ $LDW X[12]=[r10],4*$SZ
+ shrp X[15]=X[15],X[14],56 };;
+{ .mmi; $LDW X[11]=[input],4*$SZ
+ $LDW X[10]=[r8],4*$SZ
+ shrp X[14]=X[14],X[13],56 }
+{ .mmi; $LDW X[ 9]=[r9],4*$SZ
+ $LDW X[ 8]=[r10],4*$SZ
+ shrp X[13]=X[13],X[12],56 };;
+{ .mmi; $LDW X[ 7]=[input],4*$SZ
+ $LDW X[ 6]=[r8],4*$SZ
+ shrp X[12]=X[12],X[11],56 }
+{ .mmi; $LDW X[ 5]=[r9],4*$SZ
+ $LDW X[ 4]=[r10],4*$SZ
+ shrp X[11]=X[11],X[10],56 };;
+{ .mmi; $LDW X[ 3]=[input],4*$SZ
+ $LDW X[ 2]=[r8],4*$SZ
+ shrp X[10]=X[10],X[ 9],56 }
+{ .mmi; $LDW X[ 1]=[r9],4*$SZ
+ $LDW X[ 0]=[r10],4*$SZ
+ shrp X[ 9]=X[ 9],X[ 8],56 };;
+{ .mii; $LDW T1=[input]
+ shrp X[ 8]=X[ 8],X[ 7],56
+ shrp X[ 7]=X[ 7],X[ 6],56 }
+{ .mii; shrp X[ 6]=X[ 6],X[ 5],56
+ shrp X[ 5]=X[ 5],X[ 4],56 };;
+{ .mii; shrp X[ 4]=X[ 4],X[ 3],56
+ shrp X[ 3]=X[ 3],X[ 2],56 }
+{ .mii; shrp X[ 2]=X[ 2],X[ 1],56
+ shrp X[ 1]=X[ 1],X[ 0],56 }
+{ .mib; shrp X[ 0]=X[ 0],T1,56 }
+{ .mib; mov r8=0
+ mux1 X[15]=X[15],\@rev // eliminated on big-endian
+ br.many .L_first16 };;
+.L2byte:
+{ .mmi; $LDW X[11]=[input],4*$SZ
+ $LDW X[10]=[r8],4*$SZ
+ shrp X[15]=X[15],X[14],48 }
+{ .mmi; $LDW X[ 9]=[r9],4*$SZ
+ $LDW X[ 8]=[r10],4*$SZ
+ shrp X[14]=X[14],X[13],48 };;
+{ .mmi; $LDW X[ 7]=[input],4*$SZ
+ $LDW X[ 6]=[r8],4*$SZ
+ shrp X[13]=X[13],X[12],48 }
+{ .mmi; $LDW X[ 5]=[r9],4*$SZ
+ $LDW X[ 4]=[r10],4*$SZ
+ shrp X[12]=X[12],X[11],48 };;
+{ .mmi; $LDW X[ 3]=[input],4*$SZ
+ $LDW X[ 2]=[r8],4*$SZ
+ shrp X[11]=X[11],X[10],48 }
+{ .mmi; $LDW X[ 1]=[r9],4*$SZ
+ $LDW X[ 0]=[r10],4*$SZ
+ shrp X[10]=X[10],X[ 9],48 };;
+{ .mii; $LDW T1=[input]
+ shrp X[ 9]=X[ 9],X[ 8],48
+ shrp X[ 8]=X[ 8],X[ 7],48 }
+{ .mii; shrp X[ 7]=X[ 7],X[ 6],48
+ shrp X[ 6]=X[ 6],X[ 5],48 };;
+{ .mii; shrp X[ 5]=X[ 5],X[ 4],48
+ shrp X[ 4]=X[ 4],X[ 3],48 }
+{ .mii; shrp X[ 3]=X[ 3],X[ 2],48
+ shrp X[ 2]=X[ 2],X[ 1],48 }
+{ .mii; shrp X[ 1]=X[ 1],X[ 0],48
+ shrp X[ 0]=X[ 0],T1,48 }
+{ .mib; mov r8=0
+ mux1 X[15]=X[15],\@rev // eliminated on big-endian
+ br.many .L_first16 };;
+.L3byte:
+{ .mmi; $LDW X[ 9]=[r9],4*$SZ
+ $LDW X[ 8]=[r10],4*$SZ
+ shrp X[15]=X[15],X[14],40 };;
+{ .mmi; $LDW X[ 7]=[input],4*$SZ
+ $LDW X[ 6]=[r8],4*$SZ
+ shrp X[14]=X[14],X[13],40 }
+{ .mmi; $LDW X[ 5]=[r9],4*$SZ
+ $LDW X[ 4]=[r10],4*$SZ
+ shrp X[13]=X[13],X[12],40 };;
+{ .mmi; $LDW X[ 3]=[input],4*$SZ
+ $LDW X[ 2]=[r8],4*$SZ
+ shrp X[12]=X[12],X[11],40 }
+{ .mmi; $LDW X[ 1]=[r9],4*$SZ
+ $LDW X[ 0]=[r10],4*$SZ
+ shrp X[11]=X[11],X[10],40 };;
+{ .mii; $LDW T1=[input]
+ shrp X[10]=X[10],X[ 9],40
+ shrp X[ 9]=X[ 9],X[ 8],40 }
+{ .mii; shrp X[ 8]=X[ 8],X[ 7],40
+ shrp X[ 7]=X[ 7],X[ 6],40 };;
+{ .mii; shrp X[ 6]=X[ 6],X[ 5],40
+ shrp X[ 5]=X[ 5],X[ 4],40 }
+{ .mii; shrp X[ 4]=X[ 4],X[ 3],40
+ shrp X[ 3]=X[ 3],X[ 2],40 }
+{ .mii; shrp X[ 2]=X[ 2],X[ 1],40
+ shrp X[ 1]=X[ 1],X[ 0],40 }
+{ .mib; shrp X[ 0]=X[ 0],T1,40 }
+{ .mib; mov r8=0
+ mux1 X[15]=X[15],\@rev // eliminated on big-endian
+ br.many .L_first16 };;
+.L4byte:
+{ .mmi; $LDW X[ 7]=[input],4*$SZ
+ $LDW X[ 6]=[r8],4*$SZ
+ shrp X[15]=X[15],X[14],32 }
+{ .mmi; $LDW X[ 5]=[r9],4*$SZ
+ $LDW X[ 4]=[r10],4*$SZ
+ shrp X[14]=X[14],X[13],32 };;
+{ .mmi; $LDW X[ 3]=[input],4*$SZ
+ $LDW X[ 2]=[r8],4*$SZ
+ shrp X[13]=X[13],X[12],32 }
+{ .mmi; $LDW X[ 1]=[r9],4*$SZ
+ $LDW X[ 0]=[r10],4*$SZ
+ shrp X[12]=X[12],X[11],32 };;
+{ .mii; $LDW T1=[input]
+ shrp X[11]=X[11],X[10],32
+ shrp X[10]=X[10],X[ 9],32 }
+{ .mii; shrp X[ 9]=X[ 9],X[ 8],32
+ shrp X[ 8]=X[ 8],X[ 7],32 };;
+{ .mii; shrp X[ 7]=X[ 7],X[ 6],32
+ shrp X[ 6]=X[ 6],X[ 5],32 }
+{ .mii; shrp X[ 5]=X[ 5],X[ 4],32
+ shrp X[ 4]=X[ 4],X[ 3],32 }
+{ .mii; shrp X[ 3]=X[ 3],X[ 2],32
+ shrp X[ 2]=X[ 2],X[ 1],32 }
+{ .mii; shrp X[ 1]=X[ 1],X[ 0],32
+ shrp X[ 0]=X[ 0],T1,32 }
+{ .mib; mov r8=0
+ mux1 X[15]=X[15],\@rev // eliminated on big-endian
+ br.many .L_first16 };;
+.L5byte:
+{ .mmi; $LDW X[ 5]=[r9],4*$SZ
+ $LDW X[ 4]=[r10],4*$SZ
+ shrp X[15]=X[15],X[14],24 };;
+{ .mmi; $LDW X[ 3]=[input],4*$SZ
+ $LDW X[ 2]=[r8],4*$SZ
+ shrp X[14]=X[14],X[13],24 }
+{ .mmi; $LDW X[ 1]=[r9],4*$SZ
+ $LDW X[ 0]=[r10],4*$SZ
+ shrp X[13]=X[13],X[12],24 };;
+{ .mii; $LDW T1=[input]
+ shrp X[12]=X[12],X[11],24
+ shrp X[11]=X[11],X[10],24 }
+{ .mii; shrp X[10]=X[10],X[ 9],24
+ shrp X[ 9]=X[ 9],X[ 8],24 };;
+{ .mii; shrp X[ 8]=X[ 8],X[ 7],24
+ shrp X[ 7]=X[ 7],X[ 6],24 }
+{ .mii; shrp X[ 6]=X[ 6],X[ 5],24
+ shrp X[ 5]=X[ 5],X[ 4],24 }
+{ .mii; shrp X[ 4]=X[ 4],X[ 3],24
+ shrp X[ 3]=X[ 3],X[ 2],24 }
+{ .mii; shrp X[ 2]=X[ 2],X[ 1],24
+ shrp X[ 1]=X[ 1],X[ 0],24 }
+{ .mib; shrp X[ 0]=X[ 0],T1,24 }
+{ .mib; mov r8=0
+ mux1 X[15]=X[15],\@rev // eliminated on big-endian
+ br.many .L_first16 };;
+.L6byte:
+{ .mmi; $LDW X[ 3]=[input],4*$SZ
+ $LDW X[ 2]=[r8],4*$SZ
+ shrp X[15]=X[15],X[14],16 }
+{ .mmi; $LDW X[ 1]=[r9],4*$SZ
+ $LDW X[ 0]=[r10],4*$SZ
+ shrp X[14]=X[14],X[13],16 };;
+{ .mii; $LDW T1=[input]
+ shrp X[13]=X[13],X[12],16
+ shrp X[12]=X[12],X[11],16 }
+{ .mii; shrp X[11]=X[11],X[10],16
+ shrp X[10]=X[10],X[ 9],16 };;
+{ .mii; shrp X[ 9]=X[ 9],X[ 8],16
+ shrp X[ 8]=X[ 8],X[ 7],16 }
+{ .mii; shrp X[ 7]=X[ 7],X[ 6],16
+ shrp X[ 6]=X[ 6],X[ 5],16 }
+{ .mii; shrp X[ 5]=X[ 5],X[ 4],16
+ shrp X[ 4]=X[ 4],X[ 3],16 }
+{ .mii; shrp X[ 3]=X[ 3],X[ 2],16
+ shrp X[ 2]=X[ 2],X[ 1],16 }
+{ .mii; shrp X[ 1]=X[ 1],X[ 0],16
+ shrp X[ 0]=X[ 0],T1,16 }
+{ .mib; mov r8=0
+ mux1 X[15]=X[15],\@rev // eliminated on big-endian
+ br.many .L_first16 };;
+.L7byte:
+{ .mmi; $LDW X[ 1]=[r9],4*$SZ
+ $LDW X[ 0]=[r10],4*$SZ
+ shrp X[15]=X[15],X[14],8 };;
+{ .mii; $LDW T1=[input]
+ shrp X[14]=X[14],X[13],8
+ shrp X[13]=X[13],X[12],8 }
+{ .mii; shrp X[12]=X[12],X[11],8
+ shrp X[11]=X[11],X[10],8 };;
+{ .mii; shrp X[10]=X[10],X[ 9],8
+ shrp X[ 9]=X[ 9],X[ 8],8 }
+{ .mii; shrp X[ 8]=X[ 8],X[ 7],8
+ shrp X[ 7]=X[ 7],X[ 6],8 }
+{ .mii; shrp X[ 6]=X[ 6],X[ 5],8
+ shrp X[ 5]=X[ 5],X[ 4],8 }
+{ .mii; shrp X[ 4]=X[ 4],X[ 3],8
+ shrp X[ 3]=X[ 3],X[ 2],8 }
+{ .mii; shrp X[ 2]=X[ 2],X[ 1],8
+ shrp X[ 1]=X[ 1],X[ 0],8 }
+{ .mib; shrp X[ 0]=X[ 0],T1,8 }
+{ .mib; mov r8=0
+ mux1 X[15]=X[15],\@rev };; // eliminated on big-endian
+
.align 32
.L_first16:
-.rotr X[16]
+{ .mmi; $LDW K=[Ktbl],$SZ
+ add A=A,r8 // H+=Sigma(0) from the past
+ _rotr r10=$t1,$Sigma1[0] } // ROTR(e,14)
+{ .mmi; and T1=F,E
+ andcm r8=G,E
+ (p16) mux1 X[14]=X[14],\@rev };; // eliminated on big-endian
+{ .mmi; and T2=A,B
+ and r9=A,C
+ _rotr r11=$t1,$Sigma1[1] } // ROTR(e,41)
+{ .mmi; xor T1=T1,r8 // T1=((e & f) ^ (~e & g))
+ and r8=B,C };;
___
$t0="t0", $t1="t1", $code.=<<___ if ($BITS==32);
-{ .mib; (p14) add r9=1,input
- (p14) add r10=2,input }
-{ .mib; (p14) add r11=3,input
- (p15) br.dptk.few .L_host };;
-{ .mmi; (p14) ld1 r8=[input],$SZ
- (p14) ld1 r9=[r9] }
-{ .mmi; (p14) ld1 r10=[r10]
- (p14) ld1 r11=[r11] };;
-{ .mii; (p14) dep r9=r8,r9,8,8
- (p14) dep r11=r10,r11,8,8 };;
-{ .mib; (p14) dep X[15]=r9,r11,16,16 };;
-.L_host:
-{ .mib; (p15) $LDW X[15]=[input],$SZ // X[i]=*input++
+.align 32
+.L_first16:
+{ .mmi; add A=A,r8 // H+=Sigma(0) from the past
+ add r10=2-$SZ,input
+ add r11=3-$SZ,input };;
+{ .mmi; ld1 r9=[r9]
+ ld1 r10=[r10]
dep.z $t1=E,32,32 }
-{ .mib; $LDW K=[Ktbl],$SZ
+{ .mmi; ld1 r11=[r11]
+ $LDW K=[Ktbl],$SZ
zxt4 E=E };;
-{ .mmi; or $t1=$t1,E
- and T1=F,E
- and T2=A,B }
-{ .mmi; andcm r8=G,E
- and r9=A,C
+{ .mii; or $t1=$t1,E
+ dep X[15]=X[15],r9,8,8
mux2 $t0=A,0x44 };; // copy lower half to upper
-{ .mib; xor T1=T1,r8 // T1=((e & f) ^ (~e & g))
- _rotr r11=$t1,$Sigma1[0] } // ROTR(e,14)
-{ .mib; and r10=B,C
- xor T2=T2,r9 };;
-___
-$t0="A", $t1="E", $code.=<<___ if ($BITS==64);
-{ .mmi; $LDW X[15]=[input],$SZ // X[i]=*input++
- and T1=F,E
- and T2=A,B }
-{ .mmi; $LDW K=[Ktbl],$SZ
+{ .mmi; and T1=F,E
andcm r8=G,E
- and r9=A,C };;
-{ .mmi; xor T1=T1,r8 //T1=((e & f) ^ (~e & g))
- and r10=B,C
- _rotr r11=$t1,$Sigma1[0] } // ROTR(e,14)
-{ .mmi; xor T2=T2,r9
- mux1 X[15]=X[15],\@rev };; // eliminated in big-endian
+ dep r11=r10,r11,8,8 };;
+{ .mmi; and T2=A,B
+ and r9=A,C
+ dep X[15]=X[15],r11,16,16 };;
+{ .mmi; (p16) ld1 X[15-1]=[input],$SZ // prefetch
+ xor T1=T1,r8 // T1=((e & f) ^ (~e & g))
+ _rotr r10=$t1,$Sigma1[0] } // ROTR(e,14)
+{ .mmi; and r8=B,C
+ _rotr r11=$t1,$Sigma1[1] };; // ROTR(e,18)
___
$code.=<<___;
-{ .mib; add T1=T1,H // T1=Ch(e,f,g)+h
- _rotr r8=$t1,$Sigma1[1] } // ROTR(e,18)
-{ .mib; xor T2=T2,r10 // T2=((a & b) ^ (a & c) ^ (b & c))
- mov H=G };;
-{ .mib; xor r11=r8,r11
- _rotr r9=$t1,$Sigma1[2] } // ROTR(e,41)
-{ .mib; mov G=F
- mov F=E };;
-{ .mib; xor r9=r9,r11 // r9=Sigma1(e)
- _rotr r10=$t0,$Sigma0[0] } // ROTR(a,28)
-{ .mib; add T1=T1,K // T1=Ch(e,f,g)+h+K512[i]
- mov E=D };;
-{ .mib; add T1=T1,r9 // T1+=Sigma1(e)
- _rotr r11=$t0,$Sigma0[1] } // ROTR(a,34)
-{ .mib; mov D=C
- mov C=B };;
-{ .mib; add T1=T1,X[15] // T1+=X[i]
- _rotr r8=$t0,$Sigma0[2] } // ROTR(a,39)
-{ .mib; xor r10=r10,r11
- mux2 X[15]=X[15],0x44 };; // eliminated in 64-bit
-{ .mmi; xor r10=r8,r10 // r10=Sigma0(a)
- mov B=A
- add A=T1,T2 };;
-.L_first16_ctop:
-{ .mib; add E=E,T1
- add A=A,r10 // T2=Maj(a,b,c)+Sigma0(a)
- br.ctop.sptk .L_first16 };;
+{ .mmi; add T1=T1,H // T1=Ch(e,f,g)+h
+ xor r10=r10,r11
+ _rotr r11=$t1,$Sigma1[2] } // ROTR(e,41)
+{ .mmi; xor T2=T2,r9
+ add K=K,X[15] };;
+{ .mmi; add T1=T1,K // T1+=K[i]+X[i]
+ xor T2=T2,r8 // T2=((a & b) ^ (a & c) ^ (b & c))
+ _rotr r8=$t0,$Sigma0[0] } // ROTR(a,28)
+{ .mmi; xor r11=r11,r10 // Sigma1(e)
+ _rotr r9=$t0,$Sigma0[1] };; // ROTR(a,34)
+{ .mmi; add T1=T1,r11 // T+=Sigma1(e)
+ xor r8=r8,r9
+ _rotr r9=$t0,$Sigma0[2] };; // ROTR(a,39)
+{ .mmi; xor r8=r8,r9 // Sigma0(a)
+ add D=D,T1
+ mux2 H=X[15],0x44 } // mov H=X[15] in sha512
+{ .mib; (p16) add r9=1-$SZ,input // not used in sha512
+ add X[15]=T1,T2 // H=T1+Maj(a,b,c)
+ br.ctop.sptk .L_first16 };;
+.L_first16_end:
+
+{ .mib; mov ar.lc=$rounds-17
+ brp.loop.imp .L_rest,.L_rest_end-16 }
+{ .mib; mov ar.ec=1
+ br.many .L_rest };;
-{ .mib; mov ar.lc=$rounds-17 }
-{ .mib; mov ar.ec=1 };;
.align 32
.L_rest:
-.rotr X[16]
-{ .mib; $LDW K=[Ktbl],$SZ
+{ .mmi; $LDW K=[Ktbl],$SZ
+ add A=A,r8 // H+=Sigma0(a) from the past
_rotr r8=X[15-1],$sigma0[0] } // ROTR(s0,1)
-{ .mib; $ADD X[15]=X[15],X[15-9] // X[i&0xF]+=X[(i+9)&0xF]
- $SHRU s0=X[15-1],sgm0 };; // s0=X[(i+1)&0xF]>>7
+{ .mmi; add X[15]=X[15],X[15-9] // X[i&0xF]+=X[(i+9)&0xF]
+ $SHRU s0=X[15-1],sgm0 };; // s0=X[(i+1)&0xF]>>7
{ .mib; and T1=F,E
_rotr r9=X[15-1],$sigma0[1] } // ROTR(s0,8)
{ .mib; andcm r10=G,E
- $SHRU s1=X[15-14],sgm1 };; // s1=X[(i+14)&0xF]>>6
+ $SHRU s1=X[15-14],sgm1 };; // s1=X[(i+14)&0xF]>>6
+// Pair of mmi; splits on Itanium 1 and prevents pipeline flush
+// upon $SHRU output usage
{ .mmi; xor T1=T1,r10 // T1=((e & f) ^ (~e & g))
xor r9=r8,r9
- _rotr r10=X[15-14],$sigma1[0] };;// ROTR(s1,19)
-{ .mib; and T2=A,B
- _rotr r11=X[15-14],$sigma1[1] }// ROTR(s1,61)
-{ .mib; and r8=A,C };;
+ _rotr r10=X[15-14],$sigma1[0] }// ROTR(s1,19)
+{ .mmi; and T2=A,B
+ and r8=A,C
+ _rotr r11=X[15-14],$sigma1[1] };;// ROTR(s1,61)
___
$t0="t0", $t1="t1", $code.=<<___ if ($BITS==32);
-// I adhere to mmi; in order to hold Itanium 1 back and avoid 6 cycle
-// pipeline flush in last bundle. Note that even on Itanium2 the
-// latter stalls for one clock cycle...
-{ .mmi; xor s0=s0,r9 // s0=sigma0(X[(i+1)&0xF])
- dep.z $t1=E,32,32 }
-{ .mmi; xor r10=r11,r10
- zxt4 E=E };;
-{ .mmi; or $t1=$t1,E
- xor s1=s1,r10 // s1=sigma1(X[(i+14)&0xF])
- mux2 $t0=A,0x44 };; // copy lower half to upper
+{ .mib; xor s0=s0,r9 // s0=sigma0(X[(i+1)&0xF])
+ dep.z $t1=E,32,32 }
+{ .mib; xor r10=r11,r10
+ zxt4 E=E };;
+{ .mii; xor s1=s1,r10 // s1=sigma1(X[(i+14)&0xF])
+ shrp r9=E,$t1,32+$Sigma1[0] // ROTR(e,14)
+ mux2 $t0=A,0x44 };; // copy lower half to upper
+// Pair of mmi; splits on Itanium 1 and prevents pipeline flush
+// upon mux2 output usage
{ .mmi; xor T2=T2,r8
- _rotr r9=$t1,$Sigma1[0] } // ROTR(e,14)
+ shrp r8=E,$t1,32+$Sigma1[1]} // ROTR(e,18)
{ .mmi; and r10=B,C
add T1=T1,H // T1=Ch(e,f,g)+h
- $ADD X[15]=X[15],s0 };; // X[i&0xF]+=sigma0(X[(i+1)&0xF])
+ or $t1=$t1,E };;
___
$t0="A", $t1="E", $code.=<<___ if ($BITS==64);
{ .mib; xor s0=s0,r9 // s0=sigma0(X[(i+1)&0xF])
- _rotr r9=$t1,$Sigma1[0] } // ROTR(e,14)
+ _rotr r9=$t1,$Sigma1[0] } // ROTR(e,14)
{ .mib; xor r10=r11,r10
- xor T2=T2,r8 };;
+ xor T2=T2,r8 };;
{ .mib; xor s1=s1,r10 // s1=sigma1(X[(i+14)&0xF])
- add T1=T1,H }
+ _rotr r8=$t1,$Sigma1[1] } // ROTR(e,18)
{ .mib; and r10=B,C
- $ADD X[15]=X[15],s0 };; // X[i&0xF]+=sigma0(X[(i+1)&0xF])
+ add T1=T1,H };; // T1+=H
___
$code.=<<___;
-{ .mmi; xor T2=T2,r10 // T2=((a & b) ^ (a & c) ^ (b & c))
- mov H=G
- _rotr r8=$t1,$Sigma1[1] };; // ROTR(e,18)
-{ .mmi; xor r11=r8,r9
- $ADD X[15]=X[15],s1 // X[i&0xF]+=sigma1(X[(i+14)&0xF])
- _rotr r9=$t1,$Sigma1[2] } // ROTR(e,41)
-{ .mmi; mov G=F
- mov F=E };;
-{ .mib; xor r9=r9,r11 // r9=Sigma1(e)
- _rotr r10=$t0,$Sigma0[0] } // ROTR(a,28)
-{ .mib; add T1=T1,K // T1=Ch(e,f,g)+h+K512[i]
- mov E=D };;
-{ .mib; add T1=T1,r9 // T1+=Sigma1(e)
- _rotr r11=$t0,$Sigma0[1] } // ROTR(a,34)
-{ .mib; mov D=C
- mov C=B };;
-{ .mmi; add T1=T1,X[15] // T1+=X[i]
- xor r10=r10,r11
- _rotr r8=$t0,$Sigma0[2] };; // ROTR(a,39)
-{ .mmi; xor r10=r8,r10 // r10=Sigma0(a)
- mov B=A
- add A=T1,T2 };;
-.L_rest_ctop:
-{ .mib; add E=E,T1
- add A=A,r10 // T2=Maj(a,b,c)+Sigma0(a)
- br.ctop.sptk .L_rest };;
+{ .mib; xor r9=r9,r8
+ _rotr r8=$t1,$Sigma1[2] } // ROTR(e,41)
+{ .mib; xor T2=T2,r10 // T2=((a & b) ^ (a & c) ^ (b & c))
+ add X[15]=X[15],s0 };; // X[i]+=sigma0(X[i+1])
+{ .mmi; xor r9=r9,r8 // Sigma1(e)
+ add X[15]=X[15],s1 // X[i]+=sigma0(X[i+14])
+ _rotr r8=$t0,$Sigma0[0] };; // ROTR(a,28)
+{ .mmi; add K=K,X[15]
+ add T1=T1,r9 // T1+=Sigma1(e)
+ _rotr r9=$t0,$Sigma0[1] };; // ROTR(a,34)
+{ .mmi; add T1=T1,K // T1+=K[i]+X[i]
+ xor r8=r8,r9
+ _rotr r9=$t0,$Sigma0[2] };; // ROTR(a,39)
+{ .mib; add D=D,T1
+ mux2 H=X[15],0x44 } // mov H=X[15] in sha512
+{ .mib; xor r8=r8,r9 // Sigma0(a)
+ add X[15]=T1,T2 // H=T1+Maj(a,b,c)
+ br.ctop.sptk .L_rest };;
+.L_rest_end:
+
+{ .mmi; add A=A,r8 };; // H+=Sigma0(a) from the past
+{ .mmi; add A_=A_,A
+ add B_=B_,B
+ add C_=C_,C }
+{ .mmi; add D_=D_,D
+ add E_=E_,E
+ cmp.ltu p16,p0=1,num };;
+{ .mmi; add F_=F_,F
+ add G_=G_,G
+ add H_=H_,H }
+{ .mmb; add Ktbl=-$SZ*$rounds,Ktbl
+(p16) add num=-1,num
+(p16) br.dptk.many .L_outer };;
{ .mib; add r8=0*$SZ,ctx
add r9=1*$SZ,ctx }
{ .mib; add r10=2*$SZ,ctx
add r11=3*$SZ,ctx };;
-{ .mmi; $LDW r32=[r8],4*$SZ
- $LDW r33=[r9],4*$SZ }
-{ .mmi; $LDW r34=[r10],4*$SZ
- $LDW r35=[r11],4*$SZ
- cmp.ltu p6,p7=1,num };;
-{ .mmi; $LDW r36=[r8],-4*$SZ
- $LDW r37=[r9],-4*$SZ
-(p6) add Ktbl=-$SZ*$rounds,Ktbl }
-{ .mmi; $LDW r38=[r10],-4*$SZ
- $LDW r39=[r11],-4*$SZ
-(p7) mov ar.lc=r3 };;
-{ .mmi; add A=A,r32
- add B=B,r33
- add C=C,r34 }
-{ .mmi; add D=D,r35
- add E=E,r36
- add F=F,r37 };;
-{ .mmi; $STW [r8]=A,4*$SZ
- $STW [r9]=B,4*$SZ
- add G=G,r38 }
-{ .mmi; $STW [r10]=C,4*$SZ
- $STW [r11]=D,4*$SZ
- add H=H,r39 };;
-{ .mmi; $STW [r8]=E
- $STW [r9]=F
-(p6) add num=-1,num }
-{ .mmb; $STW [r10]=G
- $STW [r11]=H
-(p6) br.dptk.many .L_outer };;
-
-{ .mib; mov pr=prsave,0x1ffff
+{ .mmi; $STW [r8]=A_,4*$SZ
+ $STW [r9]=B_,4*$SZ
+ mov ar.lc=lcsave }
+{ .mmi; $STW [r10]=C_,4*$SZ
+ $STW [r11]=D_,4*$SZ
+ mov pr=prsave,0x1ffff };;
+{ .mmb; $STW [r8]=E_
+ $STW [r9]=F_ }
+{ .mmb; $STW [r10]=G_
+ $STW [r11]=H_
br.ret.sptk.many b0 };;
.endp $func#
___
-$code =~ s/\`([^\`]*)\`/eval $1/gem;
-$code =~ s/_rotr(\s+)([^=]+)=([^,]+),([0-9]+)/shrp$1$2=$3,$3,$4/gm;
-if ($BITS==64) {
- $code =~ s/mux2(\s+)\S+/nop.i$1 0x0/gm;
- $code =~ s/mux1(\s+)\S+/nop.i$1 0x0/gm if ($big_endian);
-}
+foreach(split($/,$code)) {
+ s/\`([^\`]*)\`/eval $1/gem;
+ s/_rotr(\s+)([^=]+)=([^,]+),([0-9]+)/shrp$1$2=$3,$3,$4/gm;
+ if ($BITS==64) {
+ s/mux2(\s+)([^=]+)=([^,]+),\S+/mov$1 $2=$3/gm;
+ s/mux1(\s+)\S+/nop.i$1 0x0/gm if ($big_endian);
+ s/(shrp\s+X\[[^=]+)=([^,]+),([^,]+),([1-9]+)/$1=$3,$2,64-$4/gm
+ if (!$big_endian);
+ s/ld1(\s+)X\[\S+/nop.m$1 0x0/gm;
+ }
-print $code;
+ print $_,"\n";
+}
print<<___ if ($BITS==32);
.align 64
data4 0x748f82ee,0x78a5636f,0x84c87814,0x8cc70208
data4 0x90befffa,0xa4506ceb,0xbef9a3f7,0xc67178f2
.size K256#,$SZ*$rounds
+stringz "SHA256 block transform for IA64, CRYPTOGAMS by <appro\@openssl.org>"
___
print<<___ if ($BITS==64);
.align 64
data8 0x4cc5d4becb3e42b6,0x597f299cfc657e2a
data8 0x5fcb6fab3ad6faec,0x6c44198c4a475817
.size K512#,$SZ*$rounds
+stringz "SHA512 block transform for IA64, CRYPTOGAMS by <appro\@openssl.org>"
___
projects / openssl.git / blobdiff