From 392813b7d9ed86b80127b58bc6e108cc28530eca Mon Sep 17 00:00:00 2001 From: jsing <> Date: Sun, 4 Sep 2016 14:06:46 +0000 Subject: Less IA64. ok deraadt@ --- src/lib/libcrypto/aes/asm/aes-ia64.S | 1123 --------------------- src/lib/libcrypto/bn/asm/ia64-mont.pl | 851 ---------------- src/lib/libcrypto/bn/asm/ia64.S | 1555 ----------------------------- src/lib/libcrypto/ia64cpuid.S | 121 --- src/lib/libcrypto/md5/asm/md5-ia64.S | 992 ------------------ src/lib/libcrypto/modes/asm/ghash-ia64.pl | 463 --------- src/lib/libcrypto/rc4/asm/rc4-ia64.pl | 755 -------------- src/lib/libcrypto/sha/asm/sha1-ia64.pl | 305 ------ src/lib/libcrypto/sha/asm/sha512-ia64.pl | 672 ------------- src/lib/libcrypto/whrlpool/wp_block.c | 10 +- 10 files changed, 1 insertion(+), 6846 deletions(-) delete mode 100644 src/lib/libcrypto/aes/asm/aes-ia64.S delete mode 100644 src/lib/libcrypto/bn/asm/ia64-mont.pl delete mode 100644 src/lib/libcrypto/bn/asm/ia64.S delete mode 100644 src/lib/libcrypto/ia64cpuid.S delete mode 100644 src/lib/libcrypto/md5/asm/md5-ia64.S delete mode 100755 src/lib/libcrypto/modes/asm/ghash-ia64.pl delete mode 100644 src/lib/libcrypto/rc4/asm/rc4-ia64.pl delete mode 100644 src/lib/libcrypto/sha/asm/sha1-ia64.pl delete mode 100755 src/lib/libcrypto/sha/asm/sha512-ia64.pl (limited to 'src') diff --git a/src/lib/libcrypto/aes/asm/aes-ia64.S b/src/lib/libcrypto/aes/asm/aes-ia64.S deleted file mode 100644 index 7f6c4c3662..0000000000 --- a/src/lib/libcrypto/aes/asm/aes-ia64.S +++ /dev/null @@ -1,1123 +0,0 @@ -// ==================================================================== -// Written by Andy Polyakov for the OpenSSL -// project. Rights for redistribution and usage in source and binary -// forms are granted according to the OpenSSL license. -// ==================================================================== -// -// What's wrong with compiler generated code? Compiler never uses -// variable 'shr' which is pairable with 'extr'/'dep' instructions. -// Then it uses 'zxt' which is an I-type, but can be replaced with -// 'and' which in turn can be assigned to M-port [there're double as -// much M-ports as there're I-ports on Itanium 2]. By sacrificing few -// registers for small constants (255, 24 and 16) to be used with -// 'shr' and 'and' instructions I can achieve better ILP, Intruction -// Level Parallelism, and performance. This code outperforms GCC 3.3 -// generated code by over factor of 2 (two), GCC 3.4 - by 70% and -// HP C - by 40%. Measured best-case scenario, i.e. aligned -// big-endian input, ECB timing on Itanium 2 is (18 + 13*rounds) -// ticks per block, or 9.25 CPU cycles per byte for 128 bit key. - -// Version 1.2 mitigates the hazard of cache-timing attacks by -// a) compressing S-boxes from 8KB to 2KB+256B, b) scheduling -// references to S-boxes for L2 cache latency, c) prefetching T[ed]4 -// prior last round. As result performance dropped to (26 + 15*rounds) -// ticks per block or 11 cycles per byte processed with 128-bit key. -// This is ~16% deterioration. For reference Itanium 2 L1 cache has -// 64 bytes line size and L2 - 128 bytes... - -.ident "aes-ia64.S, version 1.2" -.ident "IA-64 ISA artwork by Andy Polyakov " -.explicit -.text - -rk0=r8; rk1=r9; - -pfssave=r2; -lcsave=r10; -prsave=r3; -maskff=r11; -twenty4=r14; -sixteen=r15; - -te00=r16; te11=r17; te22=r18; te33=r19; -te01=r20; te12=r21; te23=r22; te30=r23; -te02=r24; te13=r25; te20=r26; te31=r27; -te03=r28; te10=r29; te21=r30; te32=r31; - -// these are rotating... -t0=r32; s0=r33; -t1=r34; s1=r35; -t2=r36; s2=r37; -t3=r38; s3=r39; - -te0=r40; te1=r41; te2=r42; te3=r43; - -#if defined(_HPUX_SOURCE) && !defined(_LP64) -# define ADDP addp4 -#else -# define ADDP add -#endif - -// Offsets from Te0 -#define TE0 0 -#define TE2 2 -#if defined(_HPUX_SOURCE) || defined(B_ENDIAN) -#define TE1 3 -#define TE3 1 -#else -#define TE1 1 -#define TE3 3 -#endif - -// This implies that AES_KEY comprises 32-bit key schedule elements -// even on LP64 platforms. -#ifndef KSZ -# define KSZ 4 -# define LDKEY ld4 -#endif - -.proc _ia64_AES_encrypt# -// Input: rk0-rk1 -// te0 -// te3 as AES_KEY->rounds!!! -// s0-s3 -// maskff,twenty4,sixteen -// Output: r16,r20,r24,r28 as s0-s3 -// Clobber: r16-r31,rk0-rk1,r32-r43 -.align 32 -_ia64_AES_encrypt: - .prologue - .altrp b6 - .body -{ .mmi; alloc r16=ar.pfs,12,0,0,8 - LDKEY t0=[rk0],2*KSZ - mov pr.rot=1<<16 } -{ .mmi; LDKEY t1=[rk1],2*KSZ - add te1=TE1,te0 - add te3=-3,te3 };; -{ .mib; LDKEY t2=[rk0],2*KSZ - mov ar.ec=2 } -{ .mib; LDKEY t3=[rk1],2*KSZ - add te2=TE2,te0 - brp.loop.imp .Le_top,.Le_end-16 };; - -{ .mmi; xor s0=s0,t0 - xor s1=s1,t1 - mov ar.lc=te3 } -{ .mmi; xor s2=s2,t2 - xor s3=s3,t3 - add te3=TE3,te0 };; - -.align 32 -.Le_top: -{ .mmi; (p0) LDKEY t0=[rk0],2*KSZ // 0/0:rk[0] - (p0) and te33=s3,maskff // 0/0:s3&0xff - (p0) extr.u te22=s2,8,8 } // 0/0:s2>>8&0xff -{ .mmi; (p0) LDKEY t1=[rk1],2*KSZ // 0/1:rk[1] - (p0) and te30=s0,maskff // 0/1:s0&0xff - (p0) shr.u te00=s0,twenty4 };; // 0/0:s0>>24 -{ .mmi; (p0) LDKEY t2=[rk0],2*KSZ // 1/2:rk[2] - (p0) shladd te33=te33,3,te3 // 1/0:te0+s0>>24 - (p0) extr.u te23=s3,8,8 } // 1/1:s3>>8&0xff -{ .mmi; (p0) LDKEY t3=[rk1],2*KSZ // 1/3:rk[3] - (p0) shladd te30=te30,3,te3 // 1/1:te3+s0 - (p0) shr.u te01=s1,twenty4 };; // 1/1:s1>>24 -{ .mmi; (p0) ld4 te33=[te33] // 2/0:te3[s3&0xff] - (p0) shladd te22=te22,3,te2 // 2/0:te2+s2>>8&0xff - (p0) extr.u te20=s0,8,8 } // 2/2:s0>>8&0xff -{ .mmi; (p0) ld4 te30=[te30] // 2/1:te3[s0] - (p0) shladd te23=te23,3,te2 // 2/1:te2+s3>>8 - (p0) shr.u te02=s2,twenty4 };; // 2/2:s2>>24 -{ .mmi; (p0) ld4 te22=[te22] // 3/0:te2[s2>>8] - (p0) shladd te20=te20,3,te2 // 3/2:te2+s0>>8 - (p0) extr.u te21=s1,8,8 } // 3/3:s1>>8&0xff -{ .mmi; (p0) ld4 te23=[te23] // 3/1:te2[s3>>8] - (p0) shladd te00=te00,3,te0 // 3/0:te0+s0>>24 - (p0) shr.u te03=s3,twenty4 };; // 3/3:s3>>24 -{ .mmi; (p0) ld4 te20=[te20] // 4/2:te2[s0>>8] - (p0) shladd te21=te21,3,te2 // 4/3:te3+s2 - (p0) extr.u te11=s1,16,8 } // 4/0:s1>>16&0xff -{ .mmi; (p0) ld4 te00=[te00] // 4/0:te0[s0>>24] - (p0) shladd te01=te01,3,te0 // 4/1:te0+s1>>24 - (p0) shr.u te13=s3,sixteen };; // 4/2:s3>>16 -{ .mmi; (p0) ld4 te21=[te21] // 5/3:te2[s1>>8] - (p0) shladd te11=te11,3,te1 // 5/0:te1+s1>>16 - (p0) extr.u te12=s2,16,8 } // 5/1:s2>>16&0xff -{ .mmi; (p0) ld4 te01=[te01] // 5/1:te0[s1>>24] - (p0) shladd te02=te02,3,te0 // 5/2:te0+s2>>24 - (p0) and te31=s1,maskff };; // 5/2:s1&0xff -{ .mmi; (p0) ld4 te11=[te11] // 6/0:te1[s1>>16] - (p0) shladd te12=te12,3,te1 // 6/1:te1+s2>>16 - (p0) extr.u te10=s0,16,8 } // 6/3:s0>>16&0xff -{ .mmi; (p0) ld4 te02=[te02] // 6/2:te0[s2>>24] - (p0) shladd te03=te03,3,te0 // 6/3:te1+s0>>16 - (p0) and te32=s2,maskff };; // 6/3:s2&0xff - -{ .mmi; (p0) ld4 te12=[te12] // 7/1:te1[s2>>16] - (p0) shladd te31=te31,3,te3 // 7/2:te3+s1&0xff - (p0) and te13=te13,maskff} // 7/2:s3>>16&0xff -{ .mmi; (p0) ld4 te03=[te03] // 7/3:te0[s3>>24] - (p0) shladd te32=te32,3,te3 // 7/3:te3+s2 - (p0) xor t0=t0,te33 };; // 7/0: -{ .mmi; (p0) ld4 te31=[te31] // 8/2:te3[s1] - (p0) shladd te13=te13,3,te1 // 8/2:te1+s3>>16 - (p0) xor t0=t0,te22 } // 8/0: -{ .mmi; (p0) ld4 te32=[te32] // 8/3:te3[s2] - (p0) shladd te10=te10,3,te1 // 8/3:te1+s0>>16 - (p0) xor t1=t1,te30 };; // 8/1: -{ .mmi; (p0) ld4 te13=[te13] // 9/2:te1[s3>>16] - (p0) ld4 te10=[te10] // 9/3:te1[s0>>16] - (p0) xor t0=t0,te00 };; // 9/0: !L2 scheduling -{ .mmi; (p0) xor t1=t1,te23 // 10[9]/1: - (p0) xor t2=t2,te20 // 10[9]/2: - (p0) xor t3=t3,te21 };; // 10[9]/3: -{ .mmi; (p0) xor t0=t0,te11 // 11[10]/0:done! - (p0) xor t1=t1,te01 // 11[10]/1: - (p0) xor t2=t2,te02 };; // 11[10]/2: !L2 scheduling -{ .mmi; (p0) xor t3=t3,te03 // 12[10]/3: - (p16) cmp.eq p0,p17=r0,r0 };; // 12[10]/clear (p17) -{ .mmi; (p0) xor t1=t1,te12 // 13[11]/1:done! - (p0) xor t2=t2,te31 // 13[11]/2: - (p0) xor t3=t3,te32 } // 13[11]/3: -{ .mmi; (p17) add te0=2048,te0 // 13[11]/ - (p17) add te1=2048+64-TE1,te1};; // 13[11]/ -{ .mib; (p0) xor t2=t2,te13 // 14[12]/2:done! - (p17) add te2=2048+128-TE2,te2} // 14[12]/ -{ .mib; (p0) xor t3=t3,te10 // 14[12]/3:done! - (p17) add te3=2048+192-TE3,te3 // 14[12]/ - br.ctop.sptk .Le_top };; -.Le_end: - - -{ .mmi; ld8 te12=[te0] // prefetch Te4 - ld8 te31=[te1] } -{ .mmi; ld8 te10=[te2] - ld8 te32=[te3] } - -{ .mmi; LDKEY t0=[rk0],2*KSZ // 0/0:rk[0] - and te33=s3,maskff // 0/0:s3&0xff - extr.u te22=s2,8,8 } // 0/0:s2>>8&0xff -{ .mmi; LDKEY t1=[rk1],2*KSZ // 0/1:rk[1] - and te30=s0,maskff // 0/1:s0&0xff - shr.u te00=s0,twenty4 };; // 0/0:s0>>24 -{ .mmi; LDKEY t2=[rk0],2*KSZ // 1/2:rk[2] - add te33=te33,te0 // 1/0:te0+s0>>24 - extr.u te23=s3,8,8 } // 1/1:s3>>8&0xff -{ .mmi; LDKEY t3=[rk1],2*KSZ // 1/3:rk[3] - add te30=te30,te0 // 1/1:te0+s0 - shr.u te01=s1,twenty4 };; // 1/1:s1>>24 -{ .mmi; ld1 te33=[te33] // 2/0:te0[s3&0xff] - add te22=te22,te0 // 2/0:te0+s2>>8&0xff - extr.u te20=s0,8,8 } // 2/2:s0>>8&0xff -{ .mmi; ld1 te30=[te30] // 2/1:te0[s0] - add te23=te23,te0 // 2/1:te0+s3>>8 - shr.u te02=s2,twenty4 };; // 2/2:s2>>24 -{ .mmi; ld1 te22=[te22] // 3/0:te0[s2>>8] - add te20=te20,te0 // 3/2:te0+s0>>8 - extr.u te21=s1,8,8 } // 3/3:s1>>8&0xff -{ .mmi; ld1 te23=[te23] // 3/1:te0[s3>>8] - add te00=te00,te0 // 3/0:te0+s0>>24 - shr.u te03=s3,twenty4 };; // 3/3:s3>>24 -{ .mmi; ld1 te20=[te20] // 4/2:te0[s0>>8] - add te21=te21,te0 // 4/3:te0+s2 - extr.u te11=s1,16,8 } // 4/0:s1>>16&0xff -{ .mmi; ld1 te00=[te00] // 4/0:te0[s0>>24] - add te01=te01,te0 // 4/1:te0+s1>>24 - shr.u te13=s3,sixteen };; // 4/2:s3>>16 -{ .mmi; ld1 te21=[te21] // 5/3:te0[s1>>8] - add te11=te11,te0 // 5/0:te0+s1>>16 - extr.u te12=s2,16,8 } // 5/1:s2>>16&0xff -{ .mmi; ld1 te01=[te01] // 5/1:te0[s1>>24] - add te02=te02,te0 // 5/2:te0+s2>>24 - and te31=s1,maskff };; // 5/2:s1&0xff -{ .mmi; ld1 te11=[te11] // 6/0:te0[s1>>16] - add te12=te12,te0 // 6/1:te0+s2>>16 - extr.u te10=s0,16,8 } // 6/3:s0>>16&0xff -{ .mmi; ld1 te02=[te02] // 6/2:te0[s2>>24] - add te03=te03,te0 // 6/3:te0+s0>>16 - and te32=s2,maskff };; // 6/3:s2&0xff - -{ .mmi; ld1 te12=[te12] // 7/1:te0[s2>>16] - add te31=te31,te0 // 7/2:te0+s1&0xff - dep te33=te22,te33,8,8} // 7/0: -{ .mmi; ld1 te03=[te03] // 7/3:te0[s3>>24] - add te32=te32,te0 // 7/3:te0+s2 - and te13=te13,maskff};; // 7/2:s3>>16&0xff -{ .mmi; ld1 te31=[te31] // 8/2:te0[s1] - add te13=te13,te0 // 8/2:te0+s3>>16 - dep te30=te23,te30,8,8} // 8/1: -{ .mmi; ld1 te32=[te32] // 8/3:te0[s2] - add te10=te10,te0 // 8/3:te0+s0>>16 - shl te00=te00,twenty4};; // 8/0: -{ .mii; ld1 te13=[te13] // 9/2:te0[s3>>16] - dep te33=te11,te33,16,8 // 9/0: - shl te01=te01,twenty4};; // 9/1: -{ .mii; ld1 te10=[te10] // 10/3:te0[s0>>16] - dep te31=te20,te31,8,8 // 10/2: - shl te02=te02,twenty4};; // 10/2: -{ .mii; xor t0=t0,te33 // 11/0: - dep te32=te21,te32,8,8 // 11/3: - shl te12=te12,sixteen};; // 11/1: -{ .mii; xor r16=t0,te00 // 12/0:done! - dep te31=te13,te31,16,8 // 12/2: - shl te03=te03,twenty4};; // 12/3: -{ .mmi; xor t1=t1,te01 // 13/1: - xor t2=t2,te02 // 13/2: - dep te32=te10,te32,16,8};; // 13/3: -{ .mmi; xor t1=t1,te30 // 14/1: - xor r24=t2,te31 // 14/2:done! - xor t3=t3,te32 };; // 14/3: -{ .mib; xor r20=t1,te12 // 15/1:done! - xor r28=t3,te03 // 15/3:done! - br.ret.sptk b6 };; -.endp _ia64_AES_encrypt# - -// void AES_encrypt (const void *in,void *out,const AES_KEY *key); -.global AES_encrypt# -.proc AES_encrypt# -.align 32 -AES_encrypt: - .prologue - .save ar.pfs,pfssave -{ .mmi; alloc pfssave=ar.pfs,3,1,12,0 - and out0=3,in0 - mov r3=ip } -{ .mmi; ADDP in0=0,in0 - mov loc0=psr.um - ADDP out11=KSZ*60,in2 };; // &AES_KEY->rounds - -{ .mmi; ld4 out11=[out11] // AES_KEY->rounds - add out8=(AES_Te#-AES_encrypt#),r3 // Te0 - .save pr,prsave - mov prsave=pr } -{ .mmi; rum 1<<3 // clear um.ac - .save ar.lc,lcsave - mov lcsave=ar.lc };; - - .body -#if defined(_HPUX_SOURCE) // HPUX is big-endian, cut 15+15 cycles... -{ .mib; cmp.ne p6,p0=out0,r0 - add out0=4,in0 -(p6) br.dpnt.many .Le_i_unaligned };; - -{ .mmi; ld4 out1=[in0],8 // s0 - and out9=3,in1 - mov twenty4=24 } -{ .mmi; ld4 out3=[out0],8 // s1 - ADDP rk0=0,in2 - mov sixteen=16 };; -{ .mmi; ld4 out5=[in0] // s2 - cmp.ne p6,p0=out9,r0 - mov maskff=0xff } -{ .mmb; ld4 out7=[out0] // s3 - ADDP rk1=KSZ,in2 - br.call.sptk.many b6=_ia64_AES_encrypt };; - -{ .mib; ADDP in0=4,in1 - ADDP in1=0,in1 -(p6) br.spnt .Le_o_unaligned };; - -{ .mii; mov psr.um=loc0 - mov ar.pfs=pfssave - mov ar.lc=lcsave };; -{ .mmi; st4 [in1]=r16,8 // s0 - st4 [in0]=r20,8 // s1 - mov pr=prsave,0x1ffff };; -{ .mmb; st4 [in1]=r24 // s2 - st4 [in0]=r28 // s3 - br.ret.sptk.many b0 };; -#endif - -.align 32 -.Le_i_unaligned: -{ .mmi; add out0=1,in0 - add out2=2,in0 - add out4=3,in0 };; -{ .mmi; ld1 r16=[in0],4 - ld1 r17=[out0],4 }//;; -{ .mmi; ld1 r18=[out2],4 - ld1 out1=[out4],4 };; // s0 -{ .mmi; ld1 r20=[in0],4 - ld1 r21=[out0],4 }//;; -{ .mmi; ld1 r22=[out2],4 - ld1 out3=[out4],4 };; // s1 -{ .mmi; ld1 r24=[in0],4 - ld1 r25=[out0],4 }//;; -{ .mmi; ld1 r26=[out2],4 - ld1 out5=[out4],4 };; // s2 -{ .mmi; ld1 r28=[in0] - ld1 r29=[out0] }//;; -{ .mmi; ld1 r30=[out2] - ld1 out7=[out4] };; // s3 - -{ .mii; - dep out1=r16,out1,24,8 //;; - dep out3=r20,out3,24,8 }//;; -{ .mii; ADDP rk0=0,in2 - dep out5=r24,out5,24,8 //;; - dep out7=r28,out7,24,8 };; -{ .mii; ADDP rk1=KSZ,in2 - dep out1=r17,out1,16,8 //;; - dep out3=r21,out3,16,8 }//;; -{ .mii; mov twenty4=24 - dep out5=r25,out5,16,8 //;; - dep out7=r29,out7,16,8 };; -{ .mii; mov sixteen=16 - dep out1=r18,out1,8,8 //;; - dep out3=r22,out3,8,8 }//;; -{ .mii; mov maskff=0xff - dep out5=r26,out5,8,8 //;; - dep out7=r30,out7,8,8 };; - -{ .mib; br.call.sptk.many b6=_ia64_AES_encrypt };; - -.Le_o_unaligned: -{ .mii; ADDP out0=0,in1 - extr.u r17=r16,8,8 // s0 - shr.u r19=r16,twenty4 }//;; -{ .mii; ADDP out1=1,in1 - extr.u r18=r16,16,8 - shr.u r23=r20,twenty4 }//;; // s1 -{ .mii; ADDP out2=2,in1 - extr.u r21=r20,8,8 - shr.u r22=r20,sixteen }//;; -{ .mii; ADDP out3=3,in1 - extr.u r25=r24,8,8 // s2 - shr.u r27=r24,twenty4 };; -{ .mii; st1 [out3]=r16,4 - extr.u r26=r24,16,8 - shr.u r31=r28,twenty4 }//;; // s3 -{ .mii; st1 [out2]=r17,4 - extr.u r29=r28,8,8 - shr.u r30=r28,sixteen }//;; - -{ .mmi; st1 [out1]=r18,4 - st1 [out0]=r19,4 };; -{ .mmi; st1 [out3]=r20,4 - st1 [out2]=r21,4 }//;; -{ .mmi; st1 [out1]=r22,4 - st1 [out0]=r23,4 };; -{ .mmi; st1 [out3]=r24,4 - st1 [out2]=r25,4 - mov pr=prsave,0x1ffff }//;; -{ .mmi; st1 [out1]=r26,4 - st1 [out0]=r27,4 - mov ar.pfs=pfssave };; -{ .mmi; st1 [out3]=r28 - st1 [out2]=r29 - mov ar.lc=lcsave }//;; -{ .mmi; st1 [out1]=r30 - st1 [out0]=r31 } -{ .mfb; mov psr.um=loc0 // restore user mask - br.ret.sptk.many b0 };; -.endp AES_encrypt# - -// *AES_decrypt are autogenerated by the following script: -#if 0 -#!/usr/bin/env perl -print "// *AES_decrypt are autogenerated by the following script:\n#if 0\n"; -open(PROG,'<'.$0); while() { print; } close(PROG); -print "#endif\n"; -while(<>) { - $process=1 if (/\.proc\s+_ia64_AES_encrypt/); - next if (!$process); - - #s/te00=s0/td00=s0/; s/te00/td00/g; - s/te11=s1/td13=s3/; s/te11/td13/g; - #s/te22=s2/td22=s2/; s/te22/td22/g; - s/te33=s3/td31=s1/; s/te33/td31/g; - - #s/te01=s1/td01=s1/; s/te01/td01/g; - s/te12=s2/td10=s0/; s/te12/td10/g; - #s/te23=s3/td23=s3/; s/te23/td23/g; - s/te30=s0/td32=s2/; s/te30/td32/g; - - #s/te02=s2/td02=s2/; s/te02/td02/g; - s/te13=s3/td11=s1/; s/te13/td11/g; - #s/te20=s0/td20=s0/; s/te20/td20/g; - s/te31=s1/td33=s3/; s/te31/td33/g; - - #s/te03=s3/td03=s3/; s/te03/td03/g; - s/te10=s0/td12=s2/; s/te10/td12/g; - #s/te21=s1/td21=s1/; s/te21/td21/g; - s/te32=s2/td30=s0/; s/te32/td30/g; - - s/td/te/g; - - s/AES_encrypt/AES_decrypt/g; - s/\.Le_/.Ld_/g; - s/AES_Te#/AES_Td#/g; - - print; - - exit if (/\.endp\s+AES_decrypt/); -} -#endif -.proc _ia64_AES_decrypt# -// Input: rk0-rk1 -// te0 -// te3 as AES_KEY->rounds!!! -// s0-s3 -// maskff,twenty4,sixteen -// Output: r16,r20,r24,r28 as s0-s3 -// Clobber: r16-r31,rk0-rk1,r32-r43 -.align 32 -_ia64_AES_decrypt: - .prologue - .altrp b6 - .body -{ .mmi; alloc r16=ar.pfs,12,0,0,8 - LDKEY t0=[rk0],2*KSZ - mov pr.rot=1<<16 } -{ .mmi; LDKEY t1=[rk1],2*KSZ - add te1=TE1,te0 - add te3=-3,te3 };; -{ .mib; LDKEY t2=[rk0],2*KSZ - mov ar.ec=2 } -{ .mib; LDKEY t3=[rk1],2*KSZ - add te2=TE2,te0 - brp.loop.imp .Ld_top,.Ld_end-16 };; - -{ .mmi; xor s0=s0,t0 - xor s1=s1,t1 - mov ar.lc=te3 } -{ .mmi; xor s2=s2,t2 - xor s3=s3,t3 - add te3=TE3,te0 };; - -.align 32 -.Ld_top: -{ .mmi; (p0) LDKEY t0=[rk0],2*KSZ // 0/0:rk[0] - (p0) and te31=s1,maskff // 0/0:s3&0xff - (p0) extr.u te22=s2,8,8 } // 0/0:s2>>8&0xff -{ .mmi; (p0) LDKEY t1=[rk1],2*KSZ // 0/1:rk[1] - (p0) and te32=s2,maskff // 0/1:s0&0xff - (p0) shr.u te00=s0,twenty4 };; // 0/0:s0>>24 -{ .mmi; (p0) LDKEY t2=[rk0],2*KSZ // 1/2:rk[2] - (p0) shladd te31=te31,3,te3 // 1/0:te0+s0>>24 - (p0) extr.u te23=s3,8,8 } // 1/1:s3>>8&0xff -{ .mmi; (p0) LDKEY t3=[rk1],2*KSZ // 1/3:rk[3] - (p0) shladd te32=te32,3,te3 // 1/1:te3+s0 - (p0) shr.u te01=s1,twenty4 };; // 1/1:s1>>24 -{ .mmi; (p0) ld4 te31=[te31] // 2/0:te3[s3&0xff] - (p0) shladd te22=te22,3,te2 // 2/0:te2+s2>>8&0xff - (p0) extr.u te20=s0,8,8 } // 2/2:s0>>8&0xff -{ .mmi; (p0) ld4 te32=[te32] // 2/1:te3[s0] - (p0) shladd te23=te23,3,te2 // 2/1:te2+s3>>8 - (p0) shr.u te02=s2,twenty4 };; // 2/2:s2>>24 -{ .mmi; (p0) ld4 te22=[te22] // 3/0:te2[s2>>8] - (p0) shladd te20=te20,3,te2 // 3/2:te2+s0>>8 - (p0) extr.u te21=s1,8,8 } // 3/3:s1>>8&0xff -{ .mmi; (p0) ld4 te23=[te23] // 3/1:te2[s3>>8] - (p0) shladd te00=te00,3,te0 // 3/0:te0+s0>>24 - (p0) shr.u te03=s3,twenty4 };; // 3/3:s3>>24 -{ .mmi; (p0) ld4 te20=[te20] // 4/2:te2[s0>>8] - (p0) shladd te21=te21,3,te2 // 4/3:te3+s2 - (p0) extr.u te13=s3,16,8 } // 4/0:s1>>16&0xff -{ .mmi; (p0) ld4 te00=[te00] // 4/0:te0[s0>>24] - (p0) shladd te01=te01,3,te0 // 4/1:te0+s1>>24 - (p0) shr.u te11=s1,sixteen };; // 4/2:s3>>16 -{ .mmi; (p0) ld4 te21=[te21] // 5/3:te2[s1>>8] - (p0) shladd te13=te13,3,te1 // 5/0:te1+s1>>16 - (p0) extr.u te10=s0,16,8 } // 5/1:s2>>16&0xff -{ .mmi; (p0) ld4 te01=[te01] // 5/1:te0[s1>>24] - (p0) shladd te02=te02,3,te0 // 5/2:te0+s2>>24 - (p0) and te33=s3,maskff };; // 5/2:s1&0xff -{ .mmi; (p0) ld4 te13=[te13] // 6/0:te1[s1>>16] - (p0) shladd te10=te10,3,te1 // 6/1:te1+s2>>16 - (p0) extr.u te12=s2,16,8 } // 6/3:s0>>16&0xff -{ .mmi; (p0) ld4 te02=[te02] // 6/2:te0[s2>>24] - (p0) shladd te03=te03,3,te0 // 6/3:te1+s0>>16 - (p0) and te30=s0,maskff };; // 6/3:s2&0xff - -{ .mmi; (p0) ld4 te10=[te10] // 7/1:te1[s2>>16] - (p0) shladd te33=te33,3,te3 // 7/2:te3+s1&0xff - (p0) and te11=te11,maskff} // 7/2:s3>>16&0xff -{ .mmi; (p0) ld4 te03=[te03] // 7/3:te0[s3>>24] - (p0) shladd te30=te30,3,te3 // 7/3:te3+s2 - (p0) xor t0=t0,te31 };; // 7/0: -{ .mmi; (p0) ld4 te33=[te33] // 8/2:te3[s1] - (p0) shladd te11=te11,3,te1 // 8/2:te1+s3>>16 - (p0) xor t0=t0,te22 } // 8/0: -{ .mmi; (p0) ld4 te30=[te30] // 8/3:te3[s2] - (p0) shladd te12=te12,3,te1 // 8/3:te1+s0>>16 - (p0) xor t1=t1,te32 };; // 8/1: -{ .mmi; (p0) ld4 te11=[te11] // 9/2:te1[s3>>16] - (p0) ld4 te12=[te12] // 9/3:te1[s0>>16] - (p0) xor t0=t0,te00 };; // 9/0: !L2 scheduling -{ .mmi; (p0) xor t1=t1,te23 // 10[9]/1: - (p0) xor t2=t2,te20 // 10[9]/2: - (p0) xor t3=t3,te21 };; // 10[9]/3: -{ .mmi; (p0) xor t0=t0,te13 // 11[10]/0:done! - (p0) xor t1=t1,te01 // 11[10]/1: - (p0) xor t2=t2,te02 };; // 11[10]/2: !L2 scheduling -{ .mmi; (p0) xor t3=t3,te03 // 12[10]/3: - (p16) cmp.eq p0,p17=r0,r0 };; // 12[10]/clear (p17) -{ .mmi; (p0) xor t1=t1,te10 // 13[11]/1:done! - (p0) xor t2=t2,te33 // 13[11]/2: - (p0) xor t3=t3,te30 } // 13[11]/3: -{ .mmi; (p17) add te0=2048,te0 // 13[11]/ - (p17) add te1=2048+64-TE1,te1};; // 13[11]/ -{ .mib; (p0) xor t2=t2,te11 // 14[12]/2:done! - (p17) add te2=2048+128-TE2,te2} // 14[12]/ -{ .mib; (p0) xor t3=t3,te12 // 14[12]/3:done! - (p17) add te3=2048+192-TE3,te3 // 14[12]/ - br.ctop.sptk .Ld_top };; -.Ld_end: - - -{ .mmi; ld8 te10=[te0] // prefetch Td4 - ld8 te33=[te1] } -{ .mmi; ld8 te12=[te2] - ld8 te30=[te3] } - -{ .mmi; LDKEY t0=[rk0],2*KSZ // 0/0:rk[0] - and te31=s1,maskff // 0/0:s3&0xff - extr.u te22=s2,8,8 } // 0/0:s2>>8&0xff -{ .mmi; LDKEY t1=[rk1],2*KSZ // 0/1:rk[1] - and te32=s2,maskff // 0/1:s0&0xff - shr.u te00=s0,twenty4 };; // 0/0:s0>>24 -{ .mmi; LDKEY t2=[rk0],2*KSZ // 1/2:rk[2] - add te31=te31,te0 // 1/0:te0+s0>>24 - extr.u te23=s3,8,8 } // 1/1:s3>>8&0xff -{ .mmi; LDKEY t3=[rk1],2*KSZ // 1/3:rk[3] - add te32=te32,te0 // 1/1:te0+s0 - shr.u te01=s1,twenty4 };; // 1/1:s1>>24 -{ .mmi; ld1 te31=[te31] // 2/0:te0[s3&0xff] - add te22=te22,te0 // 2/0:te0+s2>>8&0xff - extr.u te20=s0,8,8 } // 2/2:s0>>8&0xff -{ .mmi; ld1 te32=[te32] // 2/1:te0[s0] - add te23=te23,te0 // 2/1:te0+s3>>8 - shr.u te02=s2,twenty4 };; // 2/2:s2>>24 -{ .mmi; ld1 te22=[te22] // 3/0:te0[s2>>8] - add te20=te20,te0 // 3/2:te0+s0>>8 - extr.u te21=s1,8,8 } // 3/3:s1>>8&0xff -{ .mmi; ld1 te23=[te23] // 3/1:te0[s3>>8] - add te00=te00,te0 // 3/0:te0+s0>>24 - shr.u te03=s3,twenty4 };; // 3/3:s3>>24 -{ .mmi; ld1 te20=[te20] // 4/2:te0[s0>>8] - add te21=te21,te0 // 4/3:te0+s2 - extr.u te13=s3,16,8 } // 4/0:s1>>16&0xff -{ .mmi; ld1 te00=[te00] // 4/0:te0[s0>>24] - add te01=te01,te0 // 4/1:te0+s1>>24 - shr.u te11=s1,sixteen };; // 4/2:s3>>16 -{ .mmi; ld1 te21=[te21] // 5/3:te0[s1>>8] - add te13=te13,te0 // 5/0:te0+s1>>16 - extr.u te10=s0,16,8 } // 5/1:s2>>16&0xff -{ .mmi; ld1 te01=[te01] // 5/1:te0[s1>>24] - add te02=te02,te0 // 5/2:te0+s2>>24 - and te33=s3,maskff };; // 5/2:s1&0xff -{ .mmi; ld1 te13=[te13] // 6/0:te0[s1>>16] - add te10=te10,te0 // 6/1:te0+s2>>16 - extr.u te12=s2,16,8 } // 6/3:s0>>16&0xff -{ .mmi; ld1 te02=[te02] // 6/2:te0[s2>>24] - add te03=te03,te0 // 6/3:te0+s0>>16 - and te30=s0,maskff };; // 6/3:s2&0xff - -{ .mmi; ld1 te10=[te10] // 7/1:te0[s2>>16] - add te33=te33,te0 // 7/2:te0+s1&0xff - dep te31=te22,te31,8,8} // 7/0: -{ .mmi; ld1 te03=[te03] // 7/3:te0[s3>>24] - add te30=te30,te0 // 7/3:te0+s2 - and te11=te11,maskff};; // 7/2:s3>>16&0xff -{ .mmi; ld1 te33=[te33] // 8/2:te0[s1] - add te11=te11,te0 // 8/2:te0+s3>>16 - dep te32=te23,te32,8,8} // 8/1: -{ .mmi; ld1 te30=[te30] // 8/3:te0[s2] - add te12=te12,te0 // 8/3:te0+s0>>16 - shl te00=te00,twenty4};; // 8/0: -{ .mii; ld1 te11=[te11] // 9/2:te0[s3>>16] - dep te31=te13,te31,16,8 // 9/0: - shl te01=te01,twenty4};; // 9/1: -{ .mii; ld1 te12=[te12] // 10/3:te0[s0>>16] - dep te33=te20,te33,8,8 // 10/2: - shl te02=te02,twenty4};; // 10/2: -{ .mii; xor t0=t0,te31 // 11/0: - dep te30=te21,te30,8,8 // 11/3: - shl te10=te10,sixteen};; // 11/1: -{ .mii; xor r16=t0,te00 // 12/0:done! - dep te33=te11,te33,16,8 // 12/2: - shl te03=te03,twenty4};; // 12/3: -{ .mmi; xor t1=t1,te01 // 13/1: - xor t2=t2,te02 // 13/2: - dep te30=te12,te30,16,8};; // 13/3: -{ .mmi; xor t1=t1,te32 // 14/1: - xor r24=t2,te33 // 14/2:done! - xor t3=t3,te30 };; // 14/3: -{ .mib; xor r20=t1,te10 // 15/1:done! - xor r28=t3,te03 // 15/3:done! - br.ret.sptk b6 };; -.endp _ia64_AES_decrypt# - -// void AES_decrypt (const void *in,void *out,const AES_KEY *key); -.global AES_decrypt# -.proc AES_decrypt# -.align 32 -AES_decrypt: - .prologue - .save ar.pfs,pfssave -{ .mmi; alloc pfssave=ar.pfs,3,1,12,0 - and out0=3,in0 - mov r3=ip } -{ .mmi; ADDP in0=0,in0 - mov loc0=psr.um - ADDP out11=KSZ*60,in2 };; // &AES_KEY->rounds - -{ .mmi; ld4 out11=[out11] // AES_KEY->rounds - add out8=(AES_Td#-AES_decrypt#),r3 // Te0 - .save pr,prsave - mov prsave=pr } -{ .mmi; rum 1<<3 // clear um.ac - .save ar.lc,lcsave - mov lcsave=ar.lc };; - - .body -#if defined(_HPUX_SOURCE) // HPUX is big-endian, cut 15+15 cycles... -{ .mib; cmp.ne p6,p0=out0,r0 - add out0=4,in0 -(p6) br.dpnt.many .Ld_i_unaligned };; - -{ .mmi; ld4 out1=[in0],8 // s0 - and out9=3,in1 - mov twenty4=24 } -{ .mmi; ld4 out3=[out0],8 // s1 - ADDP rk0=0,in2 - mov sixteen=16 };; -{ .mmi; ld4 out5=[in0] // s2 - cmp.ne p6,p0=out9,r0 - mov maskff=0xff } -{ .mmb; ld4 out7=[out0] // s3 - ADDP rk1=KSZ,in2 - br.call.sptk.many b6=_ia64_AES_decrypt };; - -{ .mib; ADDP in0=4,in1 - ADDP in1=0,in1 -(p6) br.spnt .Ld_o_unaligned };; - -{ .mii; mov psr.um=loc0 - mov ar.pfs=pfssave - mov ar.lc=lcsave };; -{ .mmi; st4 [in1]=r16,8 // s0 - st4 [in0]=r20,8 // s1 - mov pr=prsave,0x1ffff };; -{ .mmb; st4 [in1]=r24 // s2 - st4 [in0]=r28 // s3 - br.ret.sptk.many b0 };; -#endif - -.align 32 -.Ld_i_unaligned: -{ .mmi; add out0=1,in0 - add out2=2,in0 - add out4=3,in0 };; -{ .mmi; ld1 r16=[in0],4 - ld1 r17=[out0],4 }//;; -{ .mmi; ld1 r18=[out2],4 - ld1 out1=[out4],4 };; // s0 -{ .mmi; ld1 r20=[in0],4 - ld1 r21=[out0],4 }//;; -{ .mmi; ld1 r22=[out2],4 - ld1 out3=[out4],4 };; // s1 -{ .mmi; ld1 r24=[in0],4 - ld1 r25=[out0],4 }//;; -{ .mmi; ld1 r26=[out2],4 - ld1 out5=[out4],4 };; // s2 -{ .mmi; ld1 r28=[in0] - ld1 r29=[out0] }//;; -{ .mmi; ld1 r30=[out2] - ld1 out7=[out4] };; // s3 - -{ .mii; - dep out1=r16,out1,24,8 //;; - dep out3=r20,out3,24,8 }//;; -{ .mii; ADDP rk0=0,in2 - dep out5=r24,out5,24,8 //;; - dep out7=r28,out7,24,8 };; -{ .mii; ADDP rk1=KSZ,in2 - dep out1=r17,out1,16,8 //;; - dep out3=r21,out3,16,8 }//;; -{ .mii; mov twenty4=24 - dep out5=r25,out5,16,8 //;; - dep out7=r29,out7,16,8 };; -{ .mii; mov sixteen=16 - dep out1=r18,out1,8,8 //;; - dep out3=r22,out3,8,8 }//;; -{ .mii; mov maskff=0xff - dep out5=r26,out5,8,8 //;; - dep out7=r30,out7,8,8 };; - -{ .mib; br.call.sptk.many b6=_ia64_AES_decrypt };; - -.Ld_o_unaligned: -{ .mii; ADDP out0=0,in1 - extr.u r17=r16,8,8 // s0 - shr.u r19=r16,twenty4 }//;; -{ .mii; ADDP out1=1,in1 - extr.u r18=r16,16,8 - shr.u r23=r20,twenty4 }//;; // s1 -{ .mii; ADDP out2=2,in1 - extr.u r21=r20,8,8 - shr.u r22=r20,sixteen }//;; -{ .mii; ADDP out3=3,in1 - extr.u r25=r24,8,8 // s2 - shr.u r27=r24,twenty4 };; -{ .mii; st1 [out3]=r16,4 - extr.u r26=r24,16,8 - shr.u r31=r28,twenty4 }//;; // s3 -{ .mii; st1 [out2]=r17,4 - extr.u r29=r28,8,8 - shr.u r30=r28,sixteen }//;; - -{ .mmi; st1 [out1]=r18,4 - st1 [out0]=r19,4 };; -{ .mmi; st1 [out3]=r20,4 - st1 [out2]=r21,4 }//;; -{ .mmi; st1 [out1]=r22,4 - st1 [out0]=r23,4 };; -{ .mmi; st1 [out3]=r24,4 - st1 [out2]=r25,4 - mov pr=prsave,0x1ffff }//;; -{ .mmi; st1 [out1]=r26,4 - st1 [out0]=r27,4 - mov ar.pfs=pfssave };; -{ .mmi; st1 [out3]=r28 - st1 [out2]=r29 - mov ar.lc=lcsave }//;; -{ .mmi; st1 [out1]=r30 - st1 [out0]=r31 } -{ .mfb; mov psr.um=loc0 // restore user mask - br.ret.sptk.many b0 };; -.endp AES_decrypt# - -// leave it in .text segment... -.align 64 -.global AES_Te# -.type AES_Te#,@object -AES_Te: data4 0xc66363a5,0xc66363a5, 0xf87c7c84,0xf87c7c84 - data4 0xee777799,0xee777799, 0xf67b7b8d,0xf67b7b8d - data4 0xfff2f20d,0xfff2f20d, 0xd66b6bbd,0xd66b6bbd - data4 0xde6f6fb1,0xde6f6fb1, 0x91c5c554,0x91c5c554 - data4 0x60303050,0x60303050, 0x02010103,0x02010103 - data4 0xce6767a9,0xce6767a9, 0x562b2b7d,0x562b2b7d - data4 0xe7fefe19,0xe7fefe19, 0xb5d7d762,0xb5d7d762 - data4 0x4dababe6,0x4dababe6, 0xec76769a,0xec76769a - data4 0x8fcaca45,0x8fcaca45, 0x1f82829d,0x1f82829d - data4 0x89c9c940,0x89c9c940, 0xfa7d7d87,0xfa7d7d87 - data4 0xeffafa15,0xeffafa15, 0xb25959eb,0xb25959eb - data4 0x8e4747c9,0x8e4747c9, 0xfbf0f00b,0xfbf0f00b - data4 0x41adadec,0x41adadec, 0xb3d4d467,0xb3d4d467 - data4 0x5fa2a2fd,0x5fa2a2fd, 0x45afafea,0x45afafea - data4 0x239c9cbf,0x239c9cbf, 0x53a4a4f7,0x53a4a4f7 - data4 0xe4727296,0xe4727296, 0x9bc0c05b,0x9bc0c05b - data4 0x75b7b7c2,0x75b7b7c2, 0xe1fdfd1c,0xe1fdfd1c - data4 0x3d9393ae,0x3d9393ae, 0x4c26266a,0x4c26266a - data4 0x6c36365a,0x6c36365a, 0x7e3f3f41,0x7e3f3f41 - data4 0xf5f7f702,0xf5f7f702, 0x83cccc4f,0x83cccc4f - data4 0x6834345c,0x6834345c, 0x51a5a5f4,0x51a5a5f4 - data4 0xd1e5e534,0xd1e5e534, 0xf9f1f108,0xf9f1f108 - data4 0xe2717193,0xe2717193, 0xabd8d873,0xabd8d873 - data4 0x62313153,0x62313153, 0x2a15153f,0x2a15153f - data4 0x0804040c,0x0804040c, 0x95c7c752,0x95c7c752 - data4 0x46232365,0x46232365, 0x9dc3c35e,0x9dc3c35e - data4 0x30181828,0x30181828, 0x379696a1,0x379696a1 - data4 0x0a05050f,0x0a05050f, 0x2f9a9ab5,0x2f9a9ab5 - data4 0x0e070709,0x0e070709, 0x24121236,0x24121236 - data4 0x1b80809b,0x1b80809b, 0xdfe2e23d,0xdfe2e23d - data4 0xcdebeb26,0xcdebeb26, 0x4e272769,0x4e272769 - data4 0x7fb2b2cd,0x7fb2b2cd, 0xea75759f,0xea75759f - data4 0x1209091b,0x1209091b, 0x1d83839e,0x1d83839e - data4 0x582c2c74,0x582c2c74, 0x341a1a2e,0x341a1a2e - data4 0x361b1b2d,0x361b1b2d, 0xdc6e6eb2,0xdc6e6eb2 - data4 0xb45a5aee,0xb45a5aee, 0x5ba0a0fb,0x5ba0a0fb - data4 0xa45252f6,0xa45252f6, 0x763b3b4d,0x763b3b4d - data4 0xb7d6d661,0xb7d6d661, 0x7db3b3ce,0x7db3b3ce - data4 0x5229297b,0x5229297b, 0xdde3e33e,0xdde3e33e - data4 0x5e2f2f71,0x5e2f2f71, 0x13848497,0x13848497 - data4 0xa65353f5,0xa65353f5, 0xb9d1d168,0xb9d1d168 - data4 0x00000000,0x00000000, 0xc1eded2c,0xc1eded2c - data4 0x40202060,0x40202060, 0xe3fcfc1f,0xe3fcfc1f - data4 0x79b1b1c8,0x79b1b1c8, 0xb65b5bed,0xb65b5bed - data4 0xd46a6abe,0xd46a6abe, 0x8dcbcb46,0x8dcbcb46 - data4 0x67bebed9,0x67bebed9, 0x7239394b,0x7239394b - data4 0x944a4ade,0x944a4ade, 0x984c4cd4,0x984c4cd4 - data4 0xb05858e8,0xb05858e8, 0x85cfcf4a,0x85cfcf4a - data4 0xbbd0d06b,0xbbd0d06b, 0xc5efef2a,0xc5efef2a - data4 0x4faaaae5,0x4faaaae5, 0xedfbfb16,0xedfbfb16 - data4 0x864343c5,0x864343c5, 0x9a4d4dd7,0x9a4d4dd7 - data4 0x66333355,0x66333355, 0x11858594,0x11858594 - data4 0x8a4545cf,0x8a4545cf, 0xe9f9f910,0xe9f9f910 - data4 0x04020206,0x04020206, 0xfe7f7f81,0xfe7f7f81 - data4 0xa05050f0,0xa05050f0, 0x783c3c44,0x783c3c44 - data4 0x259f9fba,0x259f9fba, 0x4ba8a8e3,0x4ba8a8e3 - data4 0xa25151f3,0xa25151f3, 0x5da3a3fe,0x5da3a3fe - data4 0x804040c0,0x804040c0, 0x058f8f8a,0x058f8f8a - data4 0x3f9292ad,0x3f9292ad, 0x219d9dbc,0x219d9dbc - data4 0x70383848,0x70383848, 0xf1f5f504,0xf1f5f504 - data4 0x63bcbcdf,0x63bcbcdf, 0x77b6b6c1,0x77b6b6c1 - data4 0xafdada75,0xafdada75, 0x42212163,0x42212163 - data4 0x20101030,0x20101030, 0xe5ffff1a,0xe5ffff1a - data4 0xfdf3f30e,0xfdf3f30e, 0xbfd2d26d,0xbfd2d26d - data4 0x81cdcd4c,0x81cdcd4c, 0x180c0c14,0x180c0c14 - data4 0x26131335,0x26131335, 0xc3ecec2f,0xc3ecec2f - data4 0xbe5f5fe1,0xbe5f5fe1, 0x359797a2,0x359797a2 - data4 0x884444cc,0x884444cc, 0x2e171739,0x2e171739 - data4 0x93c4c457,0x93c4c457, 0x55a7a7f2,0x55a7a7f2 - data4 0xfc7e7e82,0xfc7e7e82, 0x7a3d3d47,0x7a3d3d47 - data4 0xc86464ac,0xc86464ac, 0xba5d5de7,0xba5d5de7 - data4 0x3219192b,0x3219192b, 0xe6737395,0xe6737395 - data4 0xc06060a0,0xc06060a0, 0x19818198,0x19818198 - data4 0x9e4f4fd1,0x9e4f4fd1, 0xa3dcdc7f,0xa3dcdc7f - data4 0x44222266,0x44222266, 0x542a2a7e,0x542a2a7e - data4 0x3b9090ab,0x3b9090ab, 0x0b888883,0x0b888883 - data4 0x8c4646ca,0x8c4646ca, 0xc7eeee29,0xc7eeee29 - data4 0x6bb8b8d3,0x6bb8b8d3, 0x2814143c,0x2814143c - data4 0xa7dede79,0xa7dede79, 0xbc5e5ee2,0xbc5e5ee2 - data4 0x160b0b1d,0x160b0b1d, 0xaddbdb76,0xaddbdb76 - data4 0xdbe0e03b,0xdbe0e03b, 0x64323256,0x64323256 - data4 0x743a3a4e,0x743a3a4e, 0x140a0a1e,0x140a0a1e - data4 0x924949db,0x924949db, 0x0c06060a,0x0c06060a - data4 0x4824246c,0x4824246c, 0xb85c5ce4,0xb85c5ce4 - data4 0x9fc2c25d,0x9fc2c25d, 0xbdd3d36e,0xbdd3d36e - data4 0x43acacef,0x43acacef, 0xc46262a6,0xc46262a6 - data4 0x399191a8,0x399191a8, 0x319595a4,0x319595a4 - data4 0xd3e4e437,0xd3e4e437, 0xf279798b,0xf279798b - data4 0xd5e7e732,0xd5e7e732, 0x8bc8c843,0x8bc8c843 - data4 0x6e373759,0x6e373759, 0xda6d6db7,0xda6d6db7 - data4 0x018d8d8c,0x018d8d8c, 0xb1d5d564,0xb1d5d564 - data4 0x9c4e4ed2,0x9c4e4ed2, 0x49a9a9e0,0x49a9a9e0 - data4 0xd86c6cb4,0xd86c6cb4, 0xac5656fa,0xac5656fa - data4 0xf3f4f407,0xf3f4f407, 0xcfeaea25,0xcfeaea25 - data4 0xca6565af,0xca6565af, 0xf47a7a8e,0xf47a7a8e - data4 0x47aeaee9,0x47aeaee9, 0x10080818,0x10080818 - data4 0x6fbabad5,0x6fbabad5, 0xf0787888,0xf0787888 - data4 0x4a25256f,0x4a25256f, 0x5c2e2e72,0x5c2e2e72 - data4 0x381c1c24,0x381c1c24, 0x57a6a6f1,0x57a6a6f1 - data4 0x73b4b4c7,0x73b4b4c7, 0x97c6c651,0x97c6c651 - data4 0xcbe8e823,0xcbe8e823, 0xa1dddd7c,0xa1dddd7c - data4 0xe874749c,0xe874749c, 0x3e1f1f21,0x3e1f1f21 - data4 0x964b4bdd,0x964b4bdd, 0x61bdbddc,0x61bdbddc - data4 0x0d8b8b86,0x0d8b8b86, 0x0f8a8a85,0x0f8a8a85 - data4 0xe0707090,0xe0707090, 0x7c3e3e42,0x7c3e3e42 - data4 0x71b5b5c4,0x71b5b5c4, 0xcc6666aa,0xcc6666aa - data4 0x904848d8,0x904848d8, 0x06030305,0x06030305 - data4 0xf7f6f601,0xf7f6f601, 0x1c0e0e12,0x1c0e0e12 - data4 0xc26161a3,0xc26161a3, 0x6a35355f,0x6a35355f - data4 0xae5757f9,0xae5757f9, 0x69b9b9d0,0x69b9b9d0 - data4 0x17868691,0x17868691, 0x99c1c158,0x99c1c158 - data4 0x3a1d1d27,0x3a1d1d27, 0x279e9eb9,0x279e9eb9 - data4 0xd9e1e138,0xd9e1e138, 0xebf8f813,0xebf8f813 - data4 0x2b9898b3,0x2b9898b3, 0x22111133,0x22111133 - data4 0xd26969bb,0xd26969bb, 0xa9d9d970,0xa9d9d970 - data4 0x078e8e89,0x078e8e89, 0x339494a7,0x339494a7 - data4 0x2d9b9bb6,0x2d9b9bb6, 0x3c1e1e22,0x3c1e1e22 - data4 0x15878792,0x15878792, 0xc9e9e920,0xc9e9e920 - data4 0x87cece49,0x87cece49, 0xaa5555ff,0xaa5555ff - data4 0x50282878,0x50282878, 0xa5dfdf7a,0xa5dfdf7a - data4 0x038c8c8f,0x038c8c8f, 0x59a1a1f8,0x59a1a1f8 - data4 0x09898980,0x09898980, 0x1a0d0d17,0x1a0d0d17 - data4 0x65bfbfda,0x65bfbfda, 0xd7e6e631,0xd7e6e631 - data4 0x844242c6,0x844242c6, 0xd06868b8,0xd06868b8 - data4 0x824141c3,0x824141c3, 0x299999b0,0x299999b0 - data4 0x5a2d2d77,0x5a2d2d77, 0x1e0f0f11,0x1e0f0f11 - data4 0x7bb0b0cb,0x7bb0b0cb, 0xa85454fc,0xa85454fc - data4 0x6dbbbbd6,0x6dbbbbd6, 0x2c16163a,0x2c16163a -// Te4: - data1 0x63, 0x7c, 0x77, 0x7b, 0xf2, 0x6b, 0x6f, 0xc5 - data1 0x30, 0x01, 0x67, 0x2b, 0xfe, 0xd7, 0xab, 0x76 - data1 0xca, 0x82, 0xc9, 0x7d, 0xfa, 0x59, 0x47, 0xf0 - data1 0xad, 0xd4, 0xa2, 0xaf, 0x9c, 0xa4, 0x72, 0xc0 - data1 0xb7, 0xfd, 0x93, 0x26, 0x36, 0x3f, 0xf7, 0xcc - data1 0x34, 0xa5, 0xe5, 0xf1, 0x71, 0xd8, 0x31, 0x15 - data1 0x04, 0xc7, 0x23, 0xc3, 0x18, 0x96, 0x05, 0x9a - data1 0x07, 0x12, 0x80, 0xe2, 0xeb, 0x27, 0xb2, 0x75 - data1 0x09, 0x83, 0x2c, 0x1a, 0x1b, 0x6e, 0x5a, 0xa0 - data1 0x52, 0x3b, 0xd6, 0xb3, 0x29, 0xe3, 0x2f, 0x84 - data1 0x53, 0xd1, 0x00, 0xed, 0x20, 0xfc, 0xb1, 0x5b - data1 0x6a, 0xcb, 0xbe, 0x39, 0x4a, 0x4c, 0x58, 0xcf - data1 0xd0, 0xef, 0xaa, 0xfb, 0x43, 0x4d, 0x33, 0x85 - data1 0x45, 0xf9, 0x02, 0x7f, 0x50, 0x3c, 0x9f, 0xa8 - data1 0x51, 0xa3, 0x40, 0x8f, 0x92, 0x9d, 0x38, 0xf5 - data1 0xbc, 0xb6, 0xda, 0x21, 0x10, 0xff, 0xf3, 0xd2 - data1 0xcd, 0x0c, 0x13, 0xec, 0x5f, 0x97, 0x44, 0x17 - data1 0xc4, 0xa7, 0x7e, 0x3d, 0x64, 0x5d, 0x19, 0x73 - data1 0x60, 0x81, 0x4f, 0xdc, 0x22, 0x2a, 0x90, 0x88 - data1 0x46, 0xee, 0xb8, 0x14, 0xde, 0x5e, 0x0b, 0xdb - data1 0xe0, 0x32, 0x3a, 0x0a, 0x49, 0x06, 0x24, 0x5c - data1 0xc2, 0xd3, 0xac, 0x62, 0x91, 0x95, 0xe4, 0x79 - data1 0xe7, 0xc8, 0x37, 0x6d, 0x8d, 0xd5, 0x4e, 0xa9 - data1 0x6c, 0x56, 0xf4, 0xea, 0x65, 0x7a, 0xae, 0x08 - data1 0xba, 0x78, 0x25, 0x2e, 0x1c, 0xa6, 0xb4, 0xc6 - data1 0xe8, 0xdd, 0x74, 0x1f, 0x4b, 0xbd, 0x8b, 0x8a - data1 0x70, 0x3e, 0xb5, 0x66, 0x48, 0x03, 0xf6, 0x0e - data1 0x61, 0x35, 0x57, 0xb9, 0x86, 0xc1, 0x1d, 0x9e - data1 0xe1, 0xf8, 0x98, 0x11, 0x69, 0xd9, 0x8e, 0x94 - data1 0x9b, 0x1e, 0x87, 0xe9, 0xce, 0x55, 0x28, 0xdf - data1 0x8c, 0xa1, 0x89, 0x0d, 0xbf, 0xe6, 0x42, 0x68 - data1 0x41, 0x99, 0x2d, 0x0f, 0xb0, 0x54, 0xbb, 0x16 -.size AES_Te#,2048+256 // HP-UX assembler fails to ".-AES_Te#" - -.align 64 -.global AES_Td# -.type AES_Td#,@object -AES_Td: data4 0x51f4a750,0x51f4a750, 0x7e416553,0x7e416553 - data4 0x1a17a4c3,0x1a17a4c3, 0x3a275e96,0x3a275e96 - data4 0x3bab6bcb,0x3bab6bcb, 0x1f9d45f1,0x1f9d45f1 - data4 0xacfa58ab,0xacfa58ab, 0x4be30393,0x4be30393 - data4 0x2030fa55,0x2030fa55, 0xad766df6,0xad766df6 - data4 0x88cc7691,0x88cc7691, 0xf5024c25,0xf5024c25 - data4 0x4fe5d7fc,0x4fe5d7fc, 0xc52acbd7,0xc52acbd7 - data4 0x26354480,0x26354480, 0xb562a38f,0xb562a38f - data4 0xdeb15a49,0xdeb15a49, 0x25ba1b67,0x25ba1b67 - data4 0x45ea0e98,0x45ea0e98, 0x5dfec0e1,0x5dfec0e1 - data4 0xc32f7502,0xc32f7502, 0x814cf012,0x814cf012 - data4 0x8d4697a3,0x8d4697a3, 0x6bd3f9c6,0x6bd3f9c6 - data4 0x038f5fe7,0x038f5fe7, 0x15929c95,0x15929c95 - data4 0xbf6d7aeb,0xbf6d7aeb, 0x955259da,0x955259da - data4 0xd4be832d,0xd4be832d, 0x587421d3,0x587421d3 - data4 0x49e06929,0x49e06929, 0x8ec9c844,0x8ec9c844 - data4 0x75c2896a,0x75c2896a, 0xf48e7978,0xf48e7978 - data4 0x99583e6b,0x99583e6b, 0x27b971dd,0x27b971dd - data4 0xbee14fb6,0xbee14fb6, 0xf088ad17,0xf088ad17 - data4 0xc920ac66,0xc920ac66, 0x7dce3ab4,0x7dce3ab4 - data4 0x63df4a18,0x63df4a18, 0xe51a3182,0xe51a3182 - data4 0x97513360,0x97513360, 0x62537f45,0x62537f45 - data4 0xb16477e0,0xb16477e0, 0xbb6bae84,0xbb6bae84 - data4 0xfe81a01c,0xfe81a01c, 0xf9082b94,0xf9082b94 - data4 0x70486858,0x70486858, 0x8f45fd19,0x8f45fd19 - data4 0x94de6c87,0x94de6c87, 0x527bf8b7,0x527bf8b7 - data4 0xab73d323,0xab73d323, 0x724b02e2,0x724b02e2 - data4 0xe31f8f57,0xe31f8f57, 0x6655ab2a,0x6655ab2a - data4 0xb2eb2807,0xb2eb2807, 0x2fb5c203,0x2fb5c203 - data4 0x86c57b9a,0x86c57b9a, 0xd33708a5,0xd33708a5 - data4 0x302887f2,0x302887f2, 0x23bfa5b2,0x23bfa5b2 - data4 0x02036aba,0x02036aba, 0xed16825c,0xed16825c - data4 0x8acf1c2b,0x8acf1c2b, 0xa779b492,0xa779b492 - data4 0xf307f2f0,0xf307f2f0, 0x4e69e2a1,0x4e69e2a1 - data4 0x65daf4cd,0x65daf4cd, 0x0605bed5,0x0605bed5 - data4 0xd134621f,0xd134621f, 0xc4a6fe8a,0xc4a6fe8a - data4 0x342e539d,0x342e539d, 0xa2f355a0,0xa2f355a0 - data4 0x058ae132,0x058ae132, 0xa4f6eb75,0xa4f6eb75 - data4 0x0b83ec39,0x0b83ec39, 0x4060efaa,0x4060efaa - data4 0x5e719f06,0x5e719f06, 0xbd6e1051,0xbd6e1051 - data4 0x3e218af9,0x3e218af9, 0x96dd063d,0x96dd063d - data4 0xdd3e05ae,0xdd3e05ae, 0x4de6bd46,0x4de6bd46 - data4 0x91548db5,0x91548db5, 0x71c45d05,0x71c45d05 - data4 0x0406d46f,0x0406d46f, 0x605015ff,0x605015ff - data4 0x1998fb24,0x1998fb24, 0xd6bde997,0xd6bde997 - data4 0x894043cc,0x894043cc, 0x67d99e77,0x67d99e77 - data4 0xb0e842bd,0xb0e842bd, 0x07898b88,0x07898b88 - data4 0xe7195b38,0xe7195b38, 0x79c8eedb,0x79c8eedb - data4 0xa17c0a47,0xa17c0a47, 0x7c420fe9,0x7c420fe9 - data4 0xf8841ec9,0xf8841ec9, 0x00000000,0x00000000 - data4 0x09808683,0x09808683, 0x322bed48,0x322bed48 - data4 0x1e1170ac,0x1e1170ac, 0x6c5a724e,0x6c5a724e - data4 0xfd0efffb,0xfd0efffb, 0x0f853856,0x0f853856 - data4 0x3daed51e,0x3daed51e, 0x362d3927,0x362d3927 - data4 0x0a0fd964,0x0a0fd964, 0x685ca621,0x685ca621 - data4 0x9b5b54d1,0x9b5b54d1, 0x24362e3a,0x24362e3a - data4 0x0c0a67b1,0x0c0a67b1, 0x9357e70f,0x9357e70f - data4 0xb4ee96d2,0xb4ee96d2, 0x1b9b919e,0x1b9b919e - data4 0x80c0c54f,0x80c0c54f, 0x61dc20a2,0x61dc20a2 - data4 0x5a774b69,0x5a774b69, 0x1c121a16,0x1c121a16 - data4 0xe293ba0a,0xe293ba0a, 0xc0a02ae5,0xc0a02ae5 - data4 0x3c22e043,0x3c22e043, 0x121b171d,0x121b171d - data4 0x0e090d0b,0x0e090d0b, 0xf28bc7ad,0xf28bc7ad - data4 0x2db6a8b9,0x2db6a8b9, 0x141ea9c8,0x141ea9c8 - data4 0x57f11985,0x57f11985, 0xaf75074c,0xaf75074c - data4 0xee99ddbb,0xee99ddbb, 0xa37f60fd,0xa37f60fd - data4 0xf701269f,0xf701269f, 0x5c72f5bc,0x5c72f5bc - data4 0x44663bc5,0x44663bc5, 0x5bfb7e34,0x5bfb7e34 - data4 0x8b432976,0x8b432976, 0xcb23c6dc,0xcb23c6dc - data4 0xb6edfc68,0xb6edfc68, 0xb8e4f163,0xb8e4f163 - data4 0xd731dcca,0xd731dcca, 0x42638510,0x42638510 - data4 0x13972240,0x13972240, 0x84c61120,0x84c61120 - data4 0x854a247d,0x854a247d, 0xd2bb3df8,0xd2bb3df8 - data4 0xaef93211,0xaef93211, 0xc729a16d,0xc729a16d - data4 0x1d9e2f4b,0x1d9e2f4b, 0xdcb230f3,0xdcb230f3 - data4 0x0d8652ec,0x0d8652ec, 0x77c1e3d0,0x77c1e3d0 - data4 0x2bb3166c,0x2bb3166c, 0xa970b999,0xa970b999 - data4 0x119448fa,0x119448fa, 0x47e96422,0x47e96422 - data4 0xa8fc8cc4,0xa8fc8cc4, 0xa0f03f1a,0xa0f03f1a - data4 0x567d2cd8,0x567d2cd8, 0x223390ef,0x223390ef - data4 0x87494ec7,0x87494ec7, 0xd938d1c1,0xd938d1c1 - data4 0x8ccaa2fe,0x8ccaa2fe, 0x98d40b36,0x98d40b36 - data4 0xa6f581cf,0xa6f581cf, 0xa57ade28,0xa57ade28 - data4 0xdab78e26,0xdab78e26, 0x3fadbfa4,0x3fadbfa4 - data4 0x2c3a9de4,0x2c3a9de4, 0x5078920d,0x5078920d - data4 0x6a5fcc9b,0x6a5fcc9b, 0x547e4662,0x547e4662 - data4 0xf68d13c2,0xf68d13c2, 0x90d8b8e8,0x90d8b8e8 - data4 0x2e39f75e,0x2e39f75e, 0x82c3aff5,0x82c3aff5 - data4 0x9f5d80be,0x9f5d80be, 0x69d0937c,0x69d0937c - data4 0x6fd52da9,0x6fd52da9, 0xcf2512b3,0xcf2512b3 - data4 0xc8ac993b,0xc8ac993b, 0x10187da7,0x10187da7 - data4 0xe89c636e,0xe89c636e, 0xdb3bbb7b,0xdb3bbb7b - data4 0xcd267809,0xcd267809, 0x6e5918f4,0x6e5918f4 - data4 0xec9ab701,0xec9ab701, 0x834f9aa8,0x834f9aa8 - data4 0xe6956e65,0xe6956e65, 0xaaffe67e,0xaaffe67e - data4 0x21bccf08,0x21bccf08, 0xef15e8e6,0xef15e8e6 - data4 0xbae79bd9,0xbae79bd9, 0x4a6f36ce,0x4a6f36ce - data4 0xea9f09d4,0xea9f09d4, 0x29b07cd6,0x29b07cd6 - data4 0x31a4b2af,0x31a4b2af, 0x2a3f2331,0x2a3f2331 - data4 0xc6a59430,0xc6a59430, 0x35a266c0,0x35a266c0 - data4 0x744ebc37,0x744ebc37, 0xfc82caa6,0xfc82caa6 - data4 0xe090d0b0,0xe090d0b0, 0x33a7d815,0x33a7d815 - data4 0xf104984a,0xf104984a, 0x41ecdaf7,0x41ecdaf7 - data4 0x7fcd500e,0x7fcd500e, 0x1791f62f,0x1791f62f - data4 0x764dd68d,0x764dd68d, 0x43efb04d,0x43efb04d - data4 0xccaa4d54,0xccaa4d54, 0xe49604df,0xe49604df - data4 0x9ed1b5e3,0x9ed1b5e3, 0x4c6a881b,0x4c6a881b - data4 0xc12c1fb8,0xc12c1fb8, 0x4665517f,0x4665517f - data4 0x9d5eea04,0x9d5eea04, 0x018c355d,0x018c355d - data4 0xfa877473,0xfa877473, 0xfb0b412e,0xfb0b412e - data4 0xb3671d5a,0xb3671d5a, 0x92dbd252,0x92dbd252 - data4 0xe9105633,0xe9105633, 0x6dd64713,0x6dd64713 - data4 0x9ad7618c,0x9ad7618c, 0x37a10c7a,0x37a10c7a - data4 0x59f8148e,0x59f8148e, 0xeb133c89,0xeb133c89 - data4 0xcea927ee,0xcea927ee, 0xb761c935,0xb761c935 - data4 0xe11ce5ed,0xe11ce5ed, 0x7a47b13c,0x7a47b13c - data4 0x9cd2df59,0x9cd2df59, 0x55f2733f,0x55f2733f - data4 0x1814ce79,0x1814ce79, 0x73c737bf,0x73c737bf - data4 0x53f7cdea,0x53f7cdea, 0x5ffdaa5b,0x5ffdaa5b - data4 0xdf3d6f14,0xdf3d6f14, 0x7844db86,0x7844db86 - data4 0xcaaff381,0xcaaff381, 0xb968c43e,0xb968c43e - data4 0x3824342c,0x3824342c, 0xc2a3405f,0xc2a3405f - data4 0x161dc372,0x161dc372, 0xbce2250c,0xbce2250c - data4 0x283c498b,0x283c498b, 0xff0d9541,0xff0d9541 - data4 0x39a80171,0x39a80171, 0x080cb3de,0x080cb3de - data4 0xd8b4e49c,0xd8b4e49c, 0x6456c190,0x6456c190 - data4 0x7bcb8461,0x7bcb8461, 0xd532b670,0xd532b670 - data4 0x486c5c74,0x486c5c74, 0xd0b85742,0xd0b85742 -// Td4: - data1 0x52, 0x09, 0x6a, 0xd5, 0x30, 0x36, 0xa5, 0x38 - data1 0xbf, 0x40, 0xa3, 0x9e, 0x81, 0xf3, 0xd7, 0xfb - data1 0x7c, 0xe3, 0x39, 0x82, 0x9b, 0x2f, 0xff, 0x87 - data1 0x34, 0x8e, 0x43, 0x44, 0xc4, 0xde, 0xe9, 0xcb - data1 0x54, 0x7b, 0x94, 0x32, 0xa6, 0xc2, 0x23, 0x3d - data1 0xee, 0x4c, 0x95, 0x0b, 0x42, 0xfa, 0xc3, 0x4e - data1 0x08, 0x2e, 0xa1, 0x66, 0x28, 0xd9, 0x24, 0xb2 - data1 0x76, 0x5b, 0xa2, 0x49, 0x6d, 0x8b, 0xd1, 0x25 - data1 0x72, 0xf8, 0xf6, 0x64, 0x86, 0x68, 0x98, 0x16 - data1 0xd4, 0xa4, 0x5c, 0xcc, 0x5d, 0x65, 0xb6, 0x92 - data1 0x6c, 0x70, 0x48, 0x50, 0xfd, 0xed, 0xb9, 0xda - data1 0x5e, 0x15, 0x46, 0x57, 0xa7, 0x8d, 0x9d, 0x84 - data1 0x90, 0xd8, 0xab, 0x00, 0x8c, 0xbc, 0xd3, 0x0a - data1 0xf7, 0xe4, 0x58, 0x05, 0xb8, 0xb3, 0x45, 0x06 - data1 0xd0, 0x2c, 0x1e, 0x8f, 0xca, 0x3f, 0x0f, 0x02 - data1 0xc1, 0xaf, 0xbd, 0x03, 0x01, 0x13, 0x8a, 0x6b - data1 0x3a, 0x91, 0x11, 0x41, 0x4f, 0x67, 0xdc, 0xea - data1 0x97, 0xf2, 0xcf, 0xce, 0xf0, 0xb4, 0xe6, 0x73 - data1 0x96, 0xac, 0x74, 0x22, 0xe7, 0xad, 0x35, 0x85 - data1 0xe2, 0xf9, 0x37, 0xe8, 0x1c, 0x75, 0xdf, 0x6e - data1 0x47, 0xf1, 0x1a, 0x71, 0x1d, 0x29, 0xc5, 0x89 - data1 0x6f, 0xb7, 0x62, 0x0e, 0xaa, 0x18, 0xbe, 0x1b - data1 0xfc, 0x56, 0x3e, 0x4b, 0xc6, 0xd2, 0x79, 0x20 - data1 0x9a, 0xdb, 0xc0, 0xfe, 0x78, 0xcd, 0x5a, 0xf4 - data1 0x1f, 0xdd, 0xa8, 0x33, 0x88, 0x07, 0xc7, 0x31 - data1 0xb1, 0x12, 0x10, 0x59, 0x27, 0x80, 0xec, 0x5f - data1 0x60, 0x51, 0x7f, 0xa9, 0x19, 0xb5, 0x4a, 0x0d - data1 0x2d, 0xe5, 0x7a, 0x9f, 0x93, 0xc9, 0x9c, 0xef - data1 0xa0, 0xe0, 0x3b, 0x4d, 0xae, 0x2a, 0xf5, 0xb0 - data1 0xc8, 0xeb, 0xbb, 0x3c, 0x83, 0x53, 0x99, 0x61 - data1 0x17, 0x2b, 0x04, 0x7e, 0xba, 0x77, 0xd6, 0x26 - data1 0xe1, 0x69, 0x14, 0x63, 0x55, 0x21, 0x0c, 0x7d -.size AES_Td#,2048+256 // HP-UX assembler fails to ".-AES_Td#" diff --git a/src/lib/libcrypto/bn/asm/ia64-mont.pl b/src/lib/libcrypto/bn/asm/ia64-mont.pl deleted file mode 100644 index e258658428..0000000000 --- a/src/lib/libcrypto/bn/asm/ia64-mont.pl +++ /dev/null @@ -1,851 +0,0 @@ -#!/usr/bin/env perl -# -# ==================================================================== -# Written by Andy Polyakov 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/. -# ==================================================================== - -# January 2010 -# -# "Teaser" Montgomery multiplication module for IA-64. There are -# several possibilities for improvement: -# -# - modulo-scheduling outer loop would eliminate quite a number of -# stalls after ldf8, xma and getf.sig outside inner loop and -# improve shorter key performance; -# - shorter vector support [with input vectors being fetched only -# once] should be added; -# - 2x unroll with help of n0[1] would make the code scalable on -# "wider" IA-64, "wider" than Itanium 2 that is, which is not of -# acute interest, because upcoming Tukwila's individual cores are -# reportedly based on Itanium 2 design; -# - dedicated squaring procedure(?); -# -# January 2010 -# -# Shorter vector support is implemented by zero-padding ap and np -# vectors up to 8 elements, or 512 bits. This means that 256-bit -# inputs will be processed only 2 times faster than 512-bit inputs, -# not 4 [as one would expect, because algorithm complexity is n^2]. -# The reason for padding is that inputs shorter than 512 bits won't -# be processed faster anyway, because minimal critical path of the -# core loop happens to match 512-bit timing. Either way, it resulted -# in >100% improvement of 512-bit RSA sign benchmark and 50% - of -# 1024-bit one [in comparison to original version of *this* module]. -# -# So far 'openssl speed rsa dsa' output on 900MHz Itanium 2 *with* -# this module is: -# sign verify sign/s verify/s -# rsa 512 bits 0.000290s 0.000024s 3452.8 42031.4 -# rsa 1024 bits 0.000793s 0.000058s 1261.7 17172.0 -# rsa 2048 bits 0.005908s 0.000148s 169.3 6754.0 -# rsa 4096 bits 0.033456s 0.000469s 29.9 2133.6 -# dsa 512 bits 0.000253s 0.000198s 3949.9 5057.0 -# dsa 1024 bits 0.000585s 0.000607s 1708.4 1647.4 -# dsa 2048 bits 0.001453s 0.001703s 688.1 587.4 -# -# ... and *without* (but still with ia64.S): -# -# rsa 512 bits 0.000670s 0.000041s 1491.8 24145.5 -# rsa 1024 bits 0.001988s 0.000080s 502.9 12499.3 -# rsa 2048 bits 0.008702s 0.000189s 114.9 5293.9 -# rsa 4096 bits 0.043860s 0.000533s 22.8 1875.9 -# dsa 512 bits 0.000441s 0.000427s 2265.3 2340.6 -# dsa 1024 bits 0.000823s 0.000867s 1215.6 1153.2 -# dsa 2048 bits 0.001894s 0.002179s 528.1 458.9 -# -# As it can be seen, RSA sign performance improves by 130-30%, -# hereafter less for longer keys, while verify - by 74-13%. -# DSA performance improves by 115-30%. - -if ($^O eq "hpux") { - $ADDP="addp4"; - for (@ARGV) { $ADDP="add" if (/[\+DD|\-mlp]64/); } -} else { $ADDP="add"; } - -$code=<<___; -.explicit -.text - -// int bn_mul_mont (BN_ULONG *rp,const BN_ULONG *ap, -// const BN_ULONG *bp,const BN_ULONG *np, -// const BN_ULONG *n0p,int num); -.align 64 -.global bn_mul_mont# -.proc bn_mul_mont# -bn_mul_mont: - .prologue - .body -{ .mmi; cmp4.le p6,p7=2,r37;; -(p6) cmp4.lt.unc p8,p9=8,r37 - mov ret0=r0 };; -{ .bbb; -(p9) br.cond.dptk.many bn_mul_mont_8 -(p8) br.cond.dpnt.many bn_mul_mont_general -(p7) br.ret.spnt.many b0 };; -.endp bn_mul_mont# - -prevfs=r2; prevpr=r3; prevlc=r10; prevsp=r11; - -rptr=r8; aptr=r9; bptr=r14; nptr=r15; -tptr=r16; // &tp[0] -tp_1=r17; // &tp[-1] -num=r18; len=r19; lc=r20; -topbit=r21; // carry bit from tmp[num] - -n0=f6; -m0=f7; -bi=f8; - -.align 64 -.local bn_mul_mont_general# -.proc bn_mul_mont_general# -bn_mul_mont_general: - .prologue -{ .mmi; .save ar.pfs,prevfs - alloc prevfs=ar.pfs,6,2,0,8 - $ADDP aptr=0,in1 - .save ar.lc,prevlc - mov prevlc=ar.lc } -{ .mmi; .vframe prevsp - mov prevsp=sp - $ADDP bptr=0,in2 - .save pr,prevpr - mov prevpr=pr };; - - .body - .rotf alo[6],nlo[4],ahi[8],nhi[6] - .rotr a[3],n[3],t[2] - -{ .mmi; ldf8 bi=[bptr],8 // (*bp++) - ldf8 alo[4]=[aptr],16 // ap[0] - $ADDP r30=8,in1 };; -{ .mmi; ldf8 alo[3]=[r30],16 // ap[1] - ldf8 alo[2]=[aptr],16 // ap[2] - $ADDP in4=0,in4 };; -{ .mmi; ldf8 alo[1]=[r30] // ap[3] - ldf8 n0=[in4] // n0 - $ADDP rptr=0,in0 } -{ .mmi; $ADDP nptr=0,in3 - mov r31=16 - zxt4 num=in5 };; -{ .mmi; ldf8 nlo[2]=[nptr],8 // np[0] - shladd len=num,3,r0 - shladd r31=num,3,r31 };; -{ .mmi; ldf8 nlo[1]=[nptr],8 // np[1] - add lc=-5,num - sub r31=sp,r31 };; -{ .mfb; and sp=-16,r31 // alloca - xmpy.hu ahi[2]=alo[4],bi // ap[0]*bp[0] - nop.b 0 } -{ .mfb; nop.m 0 - xmpy.lu alo[4]=alo[4],bi - brp.loop.imp .L1st_ctop,.L1st_cend-16 - };; -{ .mfi; nop.m 0 - xma.hu ahi[1]=alo[3],bi,ahi[2] // ap[1]*bp[0] - add tp_1=8,sp } -{ .mfi; nop.m 0 - xma.lu alo[3]=alo[3],bi,ahi[2] - mov pr.rot=0x20001f<<16 - // ------^----- (p40) at first (p23) - // ----------^^ p[16:20]=1 - };; -{ .mfi; nop.m 0 - xmpy.lu m0=alo[4],n0 // (ap[0]*bp[0])*n0 - mov ar.lc=lc } -{ .mfi; nop.m 0 - fcvt.fxu.s1 nhi[1]=f0 - mov ar.ec=8 };; - -.align 32 -.L1st_ctop: -.pred.rel "mutex",p40,p42 -{ .mfi; (p16) ldf8 alo[0]=[aptr],8 // *(aptr++) - (p18) xma.hu ahi[0]=alo[2],bi,ahi[1] - (p40) add n[2]=n[2],a[2] } // (p23) } -{ .mfi; (p18) ldf8 nlo[0]=[nptr],8 // *(nptr++)(p16) - (p18) xma.lu alo[2]=alo[2],bi,ahi[1] - (p42) add n[2]=n[2],a[2],1 };; // (p23) -{ .mfi; (p21) getf.sig a[0]=alo[5] - (p20) xma.hu nhi[0]=nlo[2],m0,nhi[1] - (p42) cmp.leu p41,p39=n[2],a[2] } // (p23) -{ .mfi; (p23) st8 [tp_1]=n[2],8 - (p20) xma.lu nlo[2]=nlo[2],m0,nhi[1] - (p40) cmp.ltu p41,p39=n[2],a[2] } // (p23) -{ .mmb; (p21) getf.sig n[0]=nlo[3] - (p16) nop.m 0 - br.ctop.sptk .L1st_ctop };; -.L1st_cend: - -{ .mmi; getf.sig a[0]=ahi[6] // (p24) - getf.sig n[0]=nhi[4] - add num=-1,num };; // num-- -{ .mmi; .pred.rel "mutex",p40,p42 -(p40) add n[0]=n[0],a[0] -(p42) add n[0]=n[0],a[0],1 - sub aptr=aptr,len };; // rewind -{ .mmi; .pred.rel "mutex",p40,p42 -(p40) cmp.ltu p41,p39=n[0],a[0] -(p42) cmp.leu p41,p39=n[0],a[0] - sub nptr=nptr,len };; -{ .mmi; .pred.rel "mutex",p39,p41 -(p39) add topbit=r0,r0 -(p41) add topbit=r0,r0,1 - nop.i 0 } -{ .mmi; st8 [tp_1]=n[0] - add tptr=16,sp - add tp_1=8,sp };; - -.Louter: -{ .mmi; ldf8 bi=[bptr],8 // (*bp++) - ldf8 ahi[3]=[tptr] // tp[0] - add r30=8,aptr };; -{ .mmi; ldf8 alo[4]=[aptr],16 // ap[0] - ldf8 alo[3]=[r30],16 // ap[1] - add r31=8,nptr };; -{ .mfb; ldf8 alo[2]=[aptr],16 // ap[2] - xma.hu ahi[2]=alo[4],bi,ahi[3] // ap[0]*bp[i]+tp[0] - brp.loop.imp .Linner_ctop,.Linner_cend-16 - } -{ .mfb; ldf8 alo[1]=[r30] // ap[3] - xma.lu alo[4]=alo[4],bi,ahi[3] - clrrrb.pr };; -{ .mfi; ldf8 nlo[2]=[nptr],16 // np[0] - xma.hu ahi[1]=alo[3],bi,ahi[2] // ap[1]*bp[i] - nop.i 0 } -{ .mfi; ldf8 nlo[1]=[r31] // np[1] - xma.lu alo[3]=alo[3],bi,ahi[2] - mov pr.rot=0x20101f<<16 - // ------^----- (p40) at first (p23) - // --------^--- (p30) at first (p22) - // ----------^^ p[16:20]=1 - };; -{ .mfi; st8 [tptr]=r0 // tp[0] is already accounted - xmpy.lu m0=alo[4],n0 // (ap[0]*bp[i]+tp[0])*n0 - mov ar.lc=lc } -{ .mfi; - fcvt.fxu.s1 nhi[1]=f0 - mov ar.ec=8 };; - -// This loop spins in 4*(n+7) ticks on Itanium 2 and should spin in -// 7*(n+7) ticks on Itanium (the one codenamed Merced). Factor of 7 -// in latter case accounts for two-tick pipeline stall, which means -// that its performance would be ~20% lower than optimal one. No -// attempt was made to address this, because original Itanium is -// hardly represented out in the wild... -.align 32 -.Linner_ctop: -.pred.rel "mutex",p40,p42 -.pred.rel "mutex",p30,p32 -{ .mfi; (p16) ldf8 alo[0]=[aptr],8 // *(aptr++) - (p18) xma.hu ahi[0]=alo[2],bi,ahi[1] - (p40) add n[2]=n[2],a[2] } // (p23) -{ .mfi; (p16) nop.m 0 - (p18) xma.lu alo[2]=alo[2],bi,ahi[1] - (p42) add n[2]=n[2],a[2],1 };; // (p23) -{ .mfi; (p21) getf.sig a[0]=alo[5] - (p16) nop.f 0 - (p40) cmp.ltu p41,p39=n[2],a[2] } // (p23) -{ .mfi; (p21) ld8 t[0]=[tptr],8 - (p16) nop.f 0 - (p42) cmp.leu p41,p39=n[2],a[2] };; // (p23) -{ .mfi; (p18) ldf8 nlo[0]=[nptr],8 // *(nptr++) - (p20) xma.hu nhi[0]=nlo[2],m0,nhi[1] - (p30) add a[1]=a[1],t[1] } // (p22) -{ .mfi; (p16) nop.m 0 - (p20) xma.lu nlo[2]=nlo[2],m0,nhi[1] - (p32) add a[1]=a[1],t[1],1 };; // (p22) -{ .mmi; (p21) getf.sig n[0]=nlo[3] - (p16) nop.m 0 - (p30) cmp.ltu p31,p29=a[1],t[1] } // (p22) -{ .mmb; (p23) st8 [tp_1]=n[2],8 - (p32) cmp.leu p31,p29=a[1],t[1] // (p22) - br.ctop.sptk .Linner_ctop };; -.Linner_cend: - -{ .mmi; getf.sig a[0]=ahi[6] // (p24) - getf.sig n[0]=nhi[4] - nop.i 0 };; - -{ .mmi; .pred.rel "mutex",p31,p33 -(p31) add a[0]=a[0],topbit -(p33) add a[0]=a[0],topbit,1 - mov topbit=r0 };; -{ .mfi; .pred.rel "mutex",p31,p33 -(p31) cmp.ltu p32,p30=a[0],topbit -(p33) cmp.leu p32,p30=a[0],topbit - } -{ .mfi; .pred.rel "mutex",p40,p42 -(p40) add n[0]=n[0],a[0] -(p42) add n[0]=n[0],a[0],1 - };; -{ .mmi; .pred.rel "mutex",p44,p46 -(p40) cmp.ltu p41,p39=n[0],a[0] -(p42) cmp.leu p41,p39=n[0],a[0] -(p32) add topbit=r0,r0,1 } - -{ .mmi; st8 [tp_1]=n[0],8 - cmp4.ne p6,p0=1,num - sub aptr=aptr,len };; // rewind -{ .mmi; sub nptr=nptr,len -(p41) add topbit=r0,r0,1 - add tptr=16,sp } -{ .mmb; add tp_1=8,sp - add num=-1,num // num-- -(p6) br.cond.sptk.many .Louter };; - -{ .mbb; add lc=4,lc - brp.loop.imp .Lsub_ctop,.Lsub_cend-16 - clrrrb.pr };; -{ .mii; nop.m 0 - mov pr.rot=0x10001<<16 - // ------^---- (p33) at first (p17) - mov ar.lc=lc } -{ .mii; nop.m 0 - mov ar.ec=3 - nop.i 0 };; - -.Lsub_ctop: -.pred.rel "mutex",p33,p35 -{ .mfi; (p16) ld8 t[0]=[tptr],8 // t=*(tp++) - (p16) nop.f 0 - (p33) sub n[1]=t[1],n[1] } // (p17) -{ .mfi; (p16) ld8 n[0]=[nptr],8 // n=*(np++) - (p16) nop.f 0 - (p35) sub n[1]=t[1],n[1],1 };; // (p17) -{ .mib; (p18) st8 [rptr]=n[2],8 // *(rp++)=r - (p33) cmp.gtu p34,p32=n[1],t[1] // (p17) - (p18) nop.b 0 } -{ .mib; (p18) nop.m 0 - (p35) cmp.geu p34,p32=n[1],t[1] // (p17) - br.ctop.sptk .Lsub_ctop };; -.Lsub_cend: - -{ .mmb; .pred.rel "mutex",p34,p36 -(p34) sub topbit=topbit,r0 // (p19) -(p36) sub topbit=topbit,r0,1 - brp.loop.imp .Lcopy_ctop,.Lcopy_cend-16 - } -{ .mmb; sub rptr=rptr,len // rewind - sub tptr=tptr,len - clrrrb.pr };; -{ .mmi; and aptr=tptr,topbit - andcm bptr=rptr,topbit - mov pr.rot=1<<16 };; -{ .mii; or nptr=aptr,bptr - mov ar.lc=lc - mov ar.ec=3 };; - -.Lcopy_ctop: -{ .mmb; (p16) ld8 n[0]=[nptr],8 - (p18) st8 [tptr]=r0,8 - (p16) nop.b 0 } -{ .mmb; (p16) nop.m 0 - (p18) st8 [rptr]=n[2],8 - br.ctop.sptk .Lcopy_ctop };; -.Lcopy_cend: - -{ .mmi; mov ret0=1 // signal "handled" - rum 1<<5 // clear um.mfh - mov ar.lc=prevlc } -{ .mib; .restore sp - mov sp=prevsp - mov pr=prevpr,0x1ffff - br.ret.sptk.many b0 };; -.endp bn_mul_mont_general# - -a1=r16; a2=r17; a3=r18; a4=r19; a5=r20; a6=r21; a7=r22; a8=r23; -n1=r24; n2=r25; n3=r26; n4=r27; n5=r28; n6=r29; n7=r30; n8=r31; -t0=r15; - -ai0=f8; ai1=f9; ai2=f10; ai3=f11; ai4=f12; ai5=f13; ai6=f14; ai7=f15; -ni0=f16; ni1=f17; ni2=f18; ni3=f19; ni4=f20; ni5=f21; ni6=f22; ni7=f23; - -.align 64 -.skip 48 // aligns loop body -.local bn_mul_mont_8# -.proc bn_mul_mont_8# -bn_mul_mont_8: - .prologue -{ .mmi; .save ar.pfs,prevfs - alloc prevfs=ar.pfs,6,2,0,8 - .vframe prevsp - mov prevsp=sp - .save ar.lc,prevlc - mov prevlc=ar.lc } -{ .mmi; add r17=-6*16,sp - add sp=-7*16,sp - .save pr,prevpr - mov prevpr=pr };; - -{ .mmi; .save.gf 0,0x10 - stf.spill [sp]=f16,-16 - .save.gf 0,0x20 - stf.spill [r17]=f17,32 - add r16=-5*16,prevsp};; -{ .mmi; .save.gf 0,0x40 - stf.spill [r16]=f18,32 - .save.gf 0,0x80 - stf.spill [r17]=f19,32 - $ADDP aptr=0,in1 };; -{ .mmi; .save.gf 0,0x100 - stf.spill [r16]=f20,32 - .save.gf 0,0x200 - stf.spill [r17]=f21,32 - $ADDP r29=8,in1 };; -{ .mmi; .save.gf 0,0x400 - stf.spill [r16]=f22 - .save.gf 0,0x800 - stf.spill [r17]=f23 - $ADDP rptr=0,in0 };; - - .body - .rotf bj[8],mj[2],tf[2],alo[10],ahi[10],nlo[10],nhi[10] - .rotr t[8] - -// load input vectors padding them to 8 elements -{ .mmi; ldf8 ai0=[aptr],16 // ap[0] - ldf8 ai1=[r29],16 // ap[1] - $ADDP bptr=0,in2 } -{ .mmi; $ADDP r30=8,in2 - $ADDP nptr=0,in3 - $ADDP r31=8,in3 };; -{ .mmi; ldf8 bj[7]=[bptr],16 // bp[0] - ldf8 bj[6]=[r30],16 // bp[1] - cmp4.le p4,p5=3,in5 } -{ .mmi; ldf8 ni0=[nptr],16 // np[0] - ldf8 ni1=[r31],16 // np[1] - cmp4.le p6,p7=4,in5 };; - -{ .mfi; (p4)ldf8 ai2=[aptr],16 // ap[2] - (p5)fcvt.fxu ai2=f0 - cmp4.le p8,p9=5,in5 } -{ .mfi; (p6)ldf8 ai3=[r29],16 // ap[3] - (p7)fcvt.fxu ai3=f0 - cmp4.le p10,p11=6,in5 } -{ .mfi; (p4)ldf8 bj[5]=[bptr],16 // bp[2] - (p5)fcvt.fxu bj[5]=f0 - cmp4.le p12,p13=7,in5 } -{ .mfi; (p6)ldf8 bj[4]=[r30],16 // bp[3] - (p7)fcvt.fxu bj[4]=f0 - cmp4.le p14,p15=8,in5 } -{ .mfi; (p4)ldf8 ni2=[nptr],16 // np[2] - (p5)fcvt.fxu ni2=f0 - addp4 r28=-1,in5 } -{ .mfi; (p6)ldf8 ni3=[r31],16 // np[3] - (p7)fcvt.fxu ni3=f0 - $ADDP in4=0,in4 };; - -{ .mfi; ldf8 n0=[in4] - fcvt.fxu tf[1]=f0 - nop.i 0 } - -{ .mfi; (p8)ldf8 ai4=[aptr],16 // ap[4] - (p9)fcvt.fxu ai4=f0 - mov t[0]=r0 } -{ .mfi; (p10)ldf8 ai5=[r29],16 // ap[5] - (p11)fcvt.fxu ai5=f0 - mov t[1]=r0 } -{ .mfi; (p8)ldf8 bj[3]=[bptr],16 // bp[4] - (p9)fcvt.fxu bj[3]=f0 - mov t[2]=r0 } -{ .mfi; (p10)ldf8 bj[2]=[r30],16 // bp[5] - (p11)fcvt.fxu bj[2]=f0 - mov t[3]=r0 } -{ .mfi; (p8)ldf8 ni4=[nptr],16 // np[4] - (p9)fcvt.fxu ni4=f0 - mov t[4]=r0 } -{ .mfi; (p10)ldf8 ni5=[r31],16 // np[5] - (p11)fcvt.fxu ni5=f0 - mov t[5]=r0 };; - -{ .mfi; (p12)ldf8 ai6=[aptr],16 // ap[6] - (p13)fcvt.fxu ai6=f0 - mov t[6]=r0 } -{ .mfi; (p14)ldf8 ai7=[r29],16 // ap[7] - (p15)fcvt.fxu ai7=f0 - mov t[7]=r0 } -{ .mfi; (p12)ldf8 bj[1]=[bptr],16 // bp[6] - (p13)fcvt.fxu bj[1]=f0 - mov ar.lc=r28 } -{ .mfi; (p14)ldf8 bj[0]=[r30],16 // bp[7] - (p15)fcvt.fxu bj[0]=f0 - mov ar.ec=1 } -{ .mfi; (p12)ldf8 ni6=[nptr],16 // np[6] - (p13)fcvt.fxu ni6=f0 - mov pr.rot=1<<16 } -{ .mfb; (p14)ldf8 ni7=[r31],16 // np[7] - (p15)fcvt.fxu ni7=f0 - brp.loop.imp .Louter_8_ctop,.Louter_8_cend-16 - };; - -// The loop is scheduled for 32*n ticks on Itanium 2. Actual attempt -// to measure with help of Interval Time Counter indicated that the -// factor is a tad higher: 33 or 34, if not 35. Exact measurement and -// addressing the issue is problematic, because I don't have access -// to platform-specific instruction-level profiler. On Itanium it -// should run in 56*n ticks, because of higher xma latency... -.Louter_8_ctop: - .pred.rel "mutex",p40,p42 - .pred.rel "mutex",p48,p50 -{ .mfi; (p16) nop.m 0 // 0: - (p16) xma.hu ahi[0]=ai0,bj[7],tf[1] // ap[0]*b[i]+t[0] - (p40) add a3=a3,n3 } // (p17) a3+=n3 -{ .mfi; (p42) add a3=a3,n3,1 - (p16) xma.lu alo[0]=ai0,bj[7],tf[1] - (p16) nop.i 0 };; -{ .mii; (p17) getf.sig a7=alo[8] // 1: - (p48) add t[6]=t[6],a3 // (p17) t[6]+=a3 - (p50) add t[6]=t[6],a3,1 };; -{ .mfi; (p17) getf.sig a8=ahi[8] // 2: - (p17) xma.hu nhi[7]=ni6,mj[1],nhi[6] // np[6]*m0 - (p40) cmp.ltu p43,p41=a3,n3 } -{ .mfi; (p42) cmp.leu p43,p41=a3,n3 - (p17) xma.lu nlo[7]=ni6,mj[1],nhi[6] - (p16) nop.i 0 };; -{ .mii; (p17) getf.sig n5=nlo[6] // 3: - (p48) cmp.ltu p51,p49=t[6],a3 - (p50) cmp.leu p51,p49=t[6],a3 };; - .pred.rel "mutex",p41,p43 - .pred.rel "mutex",p49,p51 -{ .mfi; (p16) nop.m 0 // 4: - (p16) xma.hu ahi[1]=ai1,bj[7],ahi[0] // ap[1]*b[i] - (p41) add a4=a4,n4 } // (p17) a4+=n4 -{ .mfi; (p43) add a4=a4,n4,1 - (p16) xma.lu alo[1]=ai1,bj[7],ahi[0] - (p16) nop.i 0 };; -{ .mfi; (p49) add t[5]=t[5],a4 // 5: (p17) t[5]+=a4 - (p16) xmpy.lu mj[0]=alo[0],n0 // (ap[0]*b[i]+t[0])*n0 - (p51) add t[5]=t[5],a4,1 };; -{ .mfi; (p16) nop.m 0 // 6: - (p17) xma.hu nhi[8]=ni7,mj[1],nhi[7] // np[7]*m0 - (p41) cmp.ltu p42,p40=a4,n4 } -{ .mfi; (p43) cmp.leu p42,p40=a4,n4 - (p17) xma.lu nlo[8]=ni7,mj[1],nhi[7] - (p16) nop.i 0 };; -{ .mii; (p17) getf.sig n6=nlo[7] // 7: - (p49) cmp.ltu p50,p48=t[5],a4 - (p51) cmp.leu p50,p48=t[5],a4 };; - .pred.rel "mutex",p40,p42 - .pred.rel "mutex",p48,p50 -{ .mfi; (p16) nop.m 0 // 8: - (p16) xma.hu ahi[2]=ai2,bj[7],ahi[1] // ap[2]*b[i] - (p40) add a5=a5,n5 } // (p17) a5+=n5 -{ .mfi; (p42) add a5=a5,n5,1 - (p16) xma.lu alo[2]=ai2,bj[7],ahi[1] - (p16) nop.i 0 };; -{ .mii; (p16) getf.sig a1=alo[1] // 9: - (p48) add t[4]=t[4],a5 // p(17) t[4]+=a5 - (p50) add t[4]=t[4],a5,1 };; -{ .mfi; (p16) nop.m 0 // 10: - (p16) xma.hu nhi[0]=ni0,mj[0],alo[0] // np[0]*m0 - (p40) cmp.ltu p43,p41=a5,n5 } -{ .mfi; (p42) cmp.leu p43,p41=a5,n5 - (p16) xma.lu nlo[0]=ni0,mj[0],alo[0] - (p16) nop.i 0 };; -{ .mii; (p17) getf.sig n7=nlo[8] // 11: - (p48) cmp.ltu p51,p49=t[4],a5 - (p50) cmp.leu p51,p49=t[4],a5 };; - .pred.rel "mutex",p41,p43 - .pred.rel "mutex",p49,p51 -{ .mfi; (p17) getf.sig n8=nhi[8] // 12: - (p16) xma.hu ahi[3]=ai3,bj[7],ahi[2] // ap[3]*b[i] - (p41) add a6=a6,n6 } // (p17) a6+=n6 -{ .mfi; (p43) add a6=a6,n6,1 - (p16) xma.lu alo[3]=ai3,bj[7],ahi[2] - (p16) nop.i 0 };; -{ .mii; (p16) getf.sig a2=alo[2] // 13: - (p49) add t[3]=t[3],a6 // (p17) t[3]+=a6 - (p51) add t[3]=t[3],a6,1 };; -{ .mfi; (p16) nop.m 0 // 14: - (p16) xma.hu nhi[1]=ni1,mj[0],nhi[0] // np[1]*m0 - (p41) cmp.ltu p42,p40=a6,n6 } -{ .mfi; (p43) cmp.leu p42,p40=a6,n6 - (p16) xma.lu nlo[1]=ni1,mj[0],nhi[0] - (p16) nop.i 0 };; -{ .mii; (p16) nop.m 0 // 15: - (p49) cmp.ltu p50,p48=t[3],a6 - (p51) cmp.leu p50,p48=t[3],a6 };; - .pred.rel "mutex",p40,p42 - .pred.rel "mutex",p48,p50 -{ .mfi; (p16) nop.m 0 // 16: - (p16) xma.hu ahi[4]=ai4,bj[7],ahi[3] // ap[4]*b[i] - (p40) add a7=a7,n7 } // (p17) a7+=n7 -{ .mfi; (p42) add a7=a7,n7,1 - (p16) xma.lu alo[4]=ai4,bj[7],ahi[3] - (p16) nop.i 0 };; -{ .mii; (p16) getf.sig a3=alo[3] // 17: - (p48) add t[2]=t[2],a7 // (p17) t[2]+=a7 - (p50) add t[2]=t[2],a7,1 };; -{ .mfi; (p16) nop.m 0 // 18: - (p16) xma.hu nhi[2]=ni2,mj[0],nhi[1] // np[2]*m0 - (p40) cmp.ltu p43,p41=a7,n7 } -{ .mfi; (p42) cmp.leu p43,p41=a7,n7 - (p16) xma.lu nlo[2]=ni2,mj[0],nhi[1] - (p16) nop.i 0 };; -{ .mii; (p16) getf.sig n1=nlo[1] // 19: - (p48) cmp.ltu p51,p49=t[2],a7 - (p50) cmp.leu p51,p49=t[2],a7 };; - .pred.rel "mutex",p41,p43 - .pred.rel "mutex",p49,p51 -{ .mfi; (p16) nop.m 0 // 20: - (p16) xma.hu ahi[5]=ai5,bj[7],ahi[4] // ap[5]*b[i] - (p41) add a8=a8,n8 } // (p17) a8+=n8 -{ .mfi; (p43) add a8=a8,n8,1 - (p16) xma.lu alo[5]=ai5,bj[7],ahi[4] - (p16) nop.i 0 };; -{ .mii; (p16) getf.sig a4=alo[4] // 21: - (p49) add t[1]=t[1],a8 // (p17) t[1]+=a8 - (p51) add t[1]=t[1],a8,1 };; -{ .mfi; (p16) nop.m 0 // 22: - (p16) xma.hu nhi[3]=ni3,mj[0],nhi[2] // np[3]*m0 - (p41) cmp.ltu p42,p40=a8,n8 } -{ .mfi; (p43) cmp.leu p42,p40=a8,n8 - (p16) xma.lu nlo[3]=ni3,mj[0],nhi[2] - (p16) nop.i 0 };; -{ .mii; (p16) getf.sig n2=nlo[2] // 23: - (p49) cmp.ltu p50,p48=t[1],a8 - (p51) cmp.leu p50,p48=t[1],a8 };; -{ .mfi; (p16) nop.m 0 // 24: - (p16) xma.hu ahi[6]=ai6,bj[7],ahi[5] // ap[6]*b[i] - (p16) add a1=a1,n1 } // (p16) a1+=n1 -{ .mfi; (p16) nop.m 0 - (p16) xma.lu alo[6]=ai6,bj[7],ahi[5] - (p17) mov t[0]=r0 };; -{ .mii; (p16) getf.sig a5=alo[5] // 25: - (p16) add t0=t[7],a1 // (p16) t[7]+=a1 - (p42) add t[0]=t[0],r0,1 };; -{ .mfi; (p16) setf.sig tf[0]=t0 // 26: - (p16) xma.hu nhi[4]=ni4,mj[0],nhi[3] // np[4]*m0 - (p50) add t[0]=t[0],r0,1 } -{ .mfi; (p16) cmp.ltu.unc p42,p40=a1,n1 - (p16) xma.lu nlo[4]=ni4,mj[0],nhi[3] - (p16) nop.i 0 };; -{ .mii; (p16) getf.sig n3=nlo[3] // 27: - (p16) cmp.ltu.unc p50,p48=t0,a1 - (p16) nop.i 0 };; - .pred.rel "mutex",p40,p42 - .pred.rel "mutex",p48,p50 -{ .mfi; (p16) nop.m 0 // 28: - (p16) xma.hu ahi[7]=ai7,bj[7],ahi[6] // ap[7]*b[i] - (p40) add a2=a2,n2 } // (p16) a2+=n2 -{ .mfi; (p42) add a2=a2,n2,1 - (p16) xma.lu alo[7]=ai7,bj[7],ahi[6] - (p16) nop.i 0 };; -{ .mii; (p16) getf.sig a6=alo[6] // 29: - (p48) add t[6]=t[6],a2 // (p16) t[6]+=a2 - (p50) add t[6]=t[6],a2,1 };; -{ .mfi; (p16) nop.m 0 // 30: - (p16) xma.hu nhi[5]=ni5,mj[0],nhi[4] // np[5]*m0 - (p40) cmp.ltu p41,p39=a2,n2 } -{ .mfi; (p42) cmp.leu p41,p39=a2,n2 - (p16) xma.lu nlo[5]=ni5,mj[0],nhi[4] - (p16) nop.i 0 };; -{ .mfi; (p16) getf.sig n4=nlo[4] // 31: - (p16) nop.f 0 - (p48) cmp.ltu p49,p47=t[6],a2 } -{ .mfb; (p50) cmp.leu p49,p47=t[6],a2 - (p16) nop.f 0 - br.ctop.sptk.many .Louter_8_ctop };; -.Louter_8_cend: - -// above loop has to execute one more time, without (p16), which is -// replaced with merged move of np[8] to GPR bank - .pred.rel "mutex",p40,p42 - .pred.rel "mutex",p48,p50 -{ .mmi; (p0) getf.sig n1=ni0 // 0: - (p40) add a3=a3,n3 // (p17) a3+=n3 - (p42) add a3=a3,n3,1 };; -{ .mii; (p17) getf.sig a7=alo[8] // 1: - (p48) add t[6]=t[6],a3 // (p17) t[6]+=a3 - (p50) add t[6]=t[6],a3,1 };; -{ .mfi; (p17) getf.sig a8=ahi[8] // 2: - (p17) xma.hu nhi[7]=ni6,mj[1],nhi[6] // np[6]*m0 - (p40) cmp.ltu p43,p41=a3,n3 } -{ .mfi; (p42) cmp.leu p43,p41=a3,n3 - (p17) xma.lu nlo[7]=ni6,mj[1],nhi[6] - (p0) nop.i 0 };; -{ .mii; (p17) getf.sig n5=nlo[6] // 3: - (p48) cmp.ltu p51,p49=t[6],a3 - (p50) cmp.leu p51,p49=t[6],a3 };; - .pred.rel "mutex",p41,p43 - .pred.rel "mutex",p49,p51 -{ .mmi; (p0) getf.sig n2=ni1 // 4: - (p41) add a4=a4,n4 // (p17) a4+=n4 - (p43) add a4=a4,n4,1 };; -{ .mfi; (p49) add t[5]=t[5],a4 // 5: (p17) t[5]+=a4 - (p0) nop.f 0 - (p51) add t[5]=t[5],a4,1 };; -{ .mfi; (p0) getf.sig n3=ni2 // 6: - (p17) xma.hu nhi[8]=ni7,mj[1],nhi[7] // np[7]*m0 - (p41) cmp.ltu p42,p40=a4,n4 } -{ .mfi; (p43) cmp.leu p42,p40=a4,n4 - (p17) xma.lu nlo[8]=ni7,mj[1],nhi[7] - (p0) nop.i 0 };; -{ .mii; (p17) getf.sig n6=nlo[7] // 7: - (p49) cmp.ltu p50,p48=t[5],a4 - (p51) cmp.leu p50,p48=t[5],a4 };; - .pred.rel "mutex",p40,p42 - .pred.rel "mutex",p48,p50 -{ .mii; (p0) getf.sig n4=ni3 // 8: - (p40) add a5=a5,n5 // (p17) a5+=n5 - (p42) add a5=a5,n5,1 };; -{ .mii; (p0) nop.m 0 // 9: - (p48) add t[4]=t[4],a5 // p(17) t[4]+=a5 - (p50) add t[4]=t[4],a5,1 };; -{ .mii; (p0) nop.m 0 // 10: - (p40) cmp.ltu p43,p41=a5,n5 - (p42) cmp.leu p43,p41=a5,n5 };; -{ .mii; (p17) getf.sig n7=nlo[8] // 11: - (p48) cmp.ltu p51,p49=t[4],a5 - (p50) cmp.leu p51,p49=t[4],a5 };; - .pred.rel "mutex",p41,p43 - .pred.rel "mutex",p49,p51 -{ .mii; (p17) getf.sig n8=nhi[8] // 12: - (p41) add a6=a6,n6 // (p17) a6+=n6 - (p43) add a6=a6,n6,1 };; -{ .mii; (p0) getf.sig n5=ni4 // 13: - (p49) add t[3]=t[3],a6 // (p17) t[3]+=a6 - (p51) add t[3]=t[3],a6,1 };; -{ .mii; (p0) nop.m 0 // 14: - (p41) cmp.ltu p42,p40=a6,n6 - (p43) cmp.leu p42,p40=a6,n6 };; -{ .mii; (p0) getf.sig n6=ni5 // 15: - (p49) cmp.ltu p50,p48=t[3],a6 - (p51) cmp.leu p50,p48=t[3],a6 };; - .pred.rel "mutex",p40,p42 - .pred.rel "mutex",p48,p50 -{ .mii; (p0) nop.m 0 // 16: - (p40) add a7=a7,n7 // (p17) a7+=n7 - (p42) add a7=a7,n7,1 };; -{ .mii; (p0) nop.m 0 // 17: - (p48) add t[2]=t[2],a7 // (p17) t[2]+=a7 - (p50) add t[2]=t[2],a7,1 };; -{ .mii; (p0) nop.m 0 // 18: - (p40) cmp.ltu p43,p41=a7,n7 - (p42) cmp.leu p43,p41=a7,n7 };; -{ .mii; (p0) getf.sig n7=ni6 // 19: - (p48) cmp.ltu p51,p49=t[2],a7 - (p50) cmp.leu p51,p49=t[2],a7 };; - .pred.rel "mutex",p41,p43 - .pred.rel "mutex",p49,p51 -{ .mii; (p0) nop.m 0 // 20: - (p41) add a8=a8,n8 // (p17) a8+=n8 - (p43) add a8=a8,n8,1 };; -{ .mmi; (p0) nop.m 0 // 21: - (p49) add t[1]=t[1],a8 // (p17) t[1]+=a8 - (p51) add t[1]=t[1],a8,1 } -{ .mmi; (p17) mov t[0]=r0 - (p41) cmp.ltu p42,p40=a8,n8 - (p43) cmp.leu p42,p40=a8,n8 };; -{ .mmi; (p0) getf.sig n8=ni7 // 22: - (p49) cmp.ltu p50,p48=t[1],a8 - (p51) cmp.leu p50,p48=t[1],a8 } -{ .mmi; (p42) add t[0]=t[0],r0,1 - (p0) add r16=-7*16,prevsp - (p0) add r17=-6*16,prevsp };; - -// subtract np[8] from carrybit|tmp[8] -// carrybit|tmp[8] layout upon exit from above loop is: -// t[0]|t[1]|t[2]|t[3]|t[4]|t[5]|t[6]|t[7]|t0 (least significant) -{ .mmi; (p50)add t[0]=t[0],r0,1 - add r18=-5*16,prevsp - sub n1=t0,n1 };; -{ .mmi; cmp.gtu p34,p32=n1,t0;; - .pred.rel "mutex",p32,p34 - (p32)sub n2=t[7],n2 - (p34)sub n2=t[7],n2,1 };; -{ .mii; (p32)cmp.gtu p35,p33=n2,t[7] - (p34)cmp.geu p35,p33=n2,t[7];; - .pred.rel "mutex",p33,p35 - (p33)sub n3=t[6],n3 } -{ .mmi; (p35)sub n3=t[6],n3,1;; - (p33)cmp.gtu p34,p32=n3,t[6] - (p35)cmp.geu p34,p32=n3,t[6] };; - .pred.rel "mutex",p32,p34 -{ .mii; (p32)sub n4=t[5],n4 - (p34)sub n4=t[5],n4,1;; - (p32)cmp.gtu p35,p33=n4,t[5] } -{ .mmi; (p34)cmp.geu p35,p33=n4,t[5];; - .pred.rel "mutex",p33,p35 - (p33)sub n5=t[4],n5 - (p35)sub n5=t[4],n5,1 };; -{ .mii; (p33)cmp.gtu p34,p32=n5,t[4] - (p35)cmp.geu p34,p32=n5,t[4];; - .pred.rel "mutex",p32,p34 - (p32)sub n6=t[3],n6 } -{ .mmi; (p34)sub n6=t[3],n6,1;; - (p32)cmp.gtu p35,p33=n6,t[3] - (p34)cmp.geu p35,p33=n6,t[3] };; - .pred.rel "mutex",p33,p35 -{ .mii; (p33)sub n7=t[2],n7 - (p35)sub n7=t[2],n7,1;; - (p33)cmp.gtu p34,p32=n7,t[2] } -{ .mmi; (p35)cmp.geu p34,p32=n7,t[2];; - .pred.rel "mutex",p32,p34 - (p32)sub n8=t[1],n8 - (p34)sub n8=t[1],n8,1 };; -{ .mii; (p32)cmp.gtu p35,p33=n8,t[1] - (p34)cmp.geu p35,p33=n8,t[1];; - .pred.rel "mutex",p33,p35 - (p33)sub a8=t[0],r0 } -{ .mmi; (p35)sub a8=t[0],r0,1;; - (p33)cmp.gtu p34,p32=a8,t[0] - (p35)cmp.geu p34,p32=a8,t[0] };; - -// save the result, either tmp[num] or tmp[num]-np[num] - .pred.rel "mutex",p32,p34 -{ .mmi; (p32)st8 [rptr]=n1,8 - (p34)st8 [rptr]=t0,8 - add r19=-4*16,prevsp};; -{ .mmb; (p32)st8 [rptr]=n2,8 - (p34)st8 [rptr]=t[7],8 - (p5)br.cond.dpnt.few .Ldone };; -{ .mmb; (p32)st8 [rptr]=n3,8 - (p34)st8 [rptr]=t[6],8 - (p7)br.cond.dpnt.few .Ldone };; -{ .mmb; (p32)st8 [rptr]=n4,8 - (p34)st8 [rptr]=t[5],8 - (p9)br.cond.dpnt.few .Ldone };; -{ .mmb; (p32)st8 [rptr]=n5,8 - (p34)st8 [rptr]=t[4],8 - (p11)br.cond.dpnt.few .Ldone };; -{ .mmb; (p32)st8 [rptr]=n6,8 - (p34)st8 [rptr]=t[3],8 - (p13)br.cond.dpnt.few .Ldone };; -{ .mmb; (p32)st8 [rptr]=n7,8 - (p34)st8 [rptr]=t[2],8 - (p15)br.cond.dpnt.few .Ldone };; -{ .mmb; (p32)st8 [rptr]=n8,8 - (p34)st8 [rptr]=t[1],8 - nop.b 0 };; -.Ldone: // epilogue -{ .mmi; ldf.fill f16=[r16],64 - ldf.fill f17=[r17],64 - nop.i 0 } -{ .mmi; ldf.fill f18=[r18],64 - ldf.fill f19=[r19],64 - mov pr=prevpr,0x1ffff };; -{ .mmi; ldf.fill f20=[r16] - ldf.fill f21=[r17] - mov ar.lc=prevlc } -{ .mmi; ldf.fill f22=[r18] - ldf.fill f23=[r19] - mov ret0=1 } // signal "handled" -{ .mib; rum 1<<5 - .restore sp - mov sp=prevsp - br.ret.sptk.many b0 };; -.endp bn_mul_mont_8# - -.type copyright#,\@object -copyright: -stringz "Montgomery multiplication for IA-64, CRYPTOGAMS by " -___ - -$output=shift and open STDOUT,">$output"; -print $code; -close STDOUT; diff --git a/src/lib/libcrypto/bn/asm/ia64.S b/src/lib/libcrypto/bn/asm/ia64.S deleted file mode 100644 index 0cf805ddc4..0000000000 --- a/src/lib/libcrypto/bn/asm/ia64.S +++ /dev/null @@ -1,1555 +0,0 @@ -.explicit -.text -.ident "ia64.S, Version 2.1" -.ident "IA-64 ISA artwork by Andy Polyakov " - -// -// ==================================================================== -// Written by Andy Polyakov for the OpenSSL -// project. -// -// Rights for redistribution and usage in source and binary forms are -// granted according to the OpenSSL license. Warranty of any kind is -// disclaimed. -// ==================================================================== -// -// Version 2.x is Itanium2 re-tune. Few words about how Itanum2 is -// different from Itanium to this module viewpoint. Most notably, is it -// "wider" than Itanium? Can you experience loop scalability as -// discussed in commentary sections? Not really:-( Itanium2 has 6 -// integer ALU ports, i.e. it's 2 ports wider, but it's not enough to -// spin twice as fast, as I need 8 IALU ports. Amount of floating point -// ports is the same, i.e. 2, while I need 4. In other words, to this -// module Itanium2 remains effectively as "wide" as Itanium. Yet it's -// essentially different in respect to this module, and a re-tune was -// required. Well, because some intruction latencies has changed. Most -// noticeably those intensively used: -// -// Itanium Itanium2 -// ldf8 9 6 L2 hit -// ld8 2 1 L1 hit -// getf 2 5 -// xma[->getf] 7[+1] 4[+0] -// add[->st8] 1[+1] 1[+0] -// -// What does it mean? You might ratiocinate that the original code -// should run just faster... Because sum of latencies is smaller... -// Wrong! Note that getf latency increased. This means that if a loop is -// scheduled for lower latency (as they were), then it will suffer from -// stall condition and the code will therefore turn anti-scalable, e.g. -// original bn_mul_words spun at 5*n or 2.5 times slower than expected -// on Itanium2! What to do? Reschedule loops for Itanium2? But then -// Itanium would exhibit anti-scalability. So I've chosen to reschedule -// for worst latency for every instruction aiming for best *all-round* -// performance. - -// Q. How much faster does it get? -// A. Here is the output from 'openssl speed rsa dsa' for vanilla -// 0.9.6a compiled with gcc version 2.96 20000731 (Red Hat -// Linux 7.1 2.96-81): -// -// sign verify sign/s verify/s -// rsa 512 bits 0.0036s 0.0003s 275.3 2999.2 -// rsa 1024 bits 0.0203s 0.0011s 49.3 894.1 -// rsa 2048 bits 0.1331s 0.0040s 7.5 250.9 -// rsa 4096 bits 0.9270s 0.0147s 1.1 68.1 -// sign verify sign/s verify/s -// dsa 512 bits 0.0035s 0.0043s 288.3 234.8 -// dsa 1024 bits 0.0111s 0.0135s 90.0 74.2 -// -// And here is similar output but for this assembler -// implementation:-) -// -// sign verify sign/s verify/s -// rsa 512 bits 0.0021s 0.0001s 549.4 9638.5 -// rsa 1024 bits 0.0055s 0.0002s 183.8 4481.1 -// rsa 2048 bits 0.0244s 0.0006s 41.4 1726.3 -// rsa 4096 bits 0.1295s 0.0018s 7.7 561.5 -// sign verify sign/s verify/s -// dsa 512 bits 0.0012s 0.0013s 891.9 756.6 -// dsa 1024 bits 0.0023s 0.0028s 440.4 376.2 -// -// Yes, you may argue that it's not fair comparison as it's -// possible to craft the C implementation with BN_UMULT_HIGH -// inline assembler macro. But of course! Here is the output -// with the macro: -// -// sign verify sign/s verify/s -// rsa 512 bits 0.0020s 0.0002s 495.0 6561.0 -// rsa 1024 bits 0.0086s 0.0004s 116.2 2235.7 -// rsa 2048 bits 0.0519s 0.0015s 19.3 667.3 -// rsa 4096 bits 0.3464s 0.0053s 2.9 187.7 -// sign verify sign/s verify/s -// dsa 512 bits 0.0016s 0.0020s 613.1 510.5 -// dsa 1024 bits 0.0045s 0.0054s 221.0 183.9 -// -// My code is still way faster, huh:-) And I believe that even -// higher performance can be achieved. Note that as keys get -// longer, performance gain is larger. Why? According to the -// profiler there is another player in the field, namely -// BN_from_montgomery consuming larger and larger portion of CPU -// time as keysize decreases. I therefore consider putting effort -// to assembler implementation of the following routine: -// -// void bn_mul_add_mont (BN_ULONG *rp,BN_ULONG *np,int nl,BN_ULONG n0) -// { -// int i,j; -// BN_ULONG v; -// -// for (i=0; i M1, etc. - -.md5_block_loop0: -{ .mmi - ld4 M12_ = [DPtr_], 4 - mov TPtr = CTable - mov TRound = CTable0 -} ;; -{ .mmi - ld4 M13_ = [DPtr_], 4 - mov A_ = AccumA - mov B_ = AccumB -} ;; -{ .mmi - ld4 M14_ = [DPtr_], 4 - mov C_ = AccumC - mov D_ = AccumD -} ;; -{ .mmb - ld4 M15_ = [DPtr_], 4 - add BlockCount = -1, BlockCount - br.call.sptk.many QUICK_RTN = md5_digest_block0 -} ;; - -// Now, we add the new digest values and do some clean-up -// before checking if there's another full block to process - -{ .mmi - add AccumA = AccumA, A_ - add AccumB = AccumB, B_ - cmp.ne pAgain, p0 = 0, BlockCount -} -{ .mib - add AccumC = AccumC, C_ - add AccumD = AccumD, D_ -(pAgain) br.cond.dptk.many .md5_block_loop0 -} ;; - -.md5_exit: -#ifdef HOST_IS_BIG_ENDIAN - sum psr.be;; // switch back to big-endian mode -#endif -{ .mmi - st4 [CtxPtr0] = AccumB, -4 - st4 [CtxPtr1] = AccumD, -4 - mov pr = PRSave, 0x1ffff ;; -} -{ .mmi - st4 [CtxPtr0] = AccumA - st4 [CtxPtr1] = AccumC - mov ar.lc = LCSave -} ;; -{ .mib - mov ar.pfs = PFSSave - br.ret.sptk.few rp -} ;; - -#define MD5UNALIGNED(offset) \ -.md5_process##offset: \ -{ .mib ; \ - nop 0x0 ; \ - GETRW(DTmp, DTmp, offset) ; \ -} ;; \ -.md5_block_loop##offset: \ -{ .mmi ; \ - ld4 Y_ = [DPtr_], 4 ; \ - mov TPtr = CTable ; \ - mov TRound = CTable0 ; \ -} ;; \ -{ .mmi ; \ - ld4 M13_ = [DPtr_], 4 ; \ - mov A_ = AccumA ; \ - mov B_ = AccumB ; \ -} ;; \ -{ .mii ; \ - ld4 M14_ = [DPtr_], 4 ; \ - GETLW(W_, Y_, offset) ; \ - mov C_ = AccumC ; \ -} \ -{ .mmi ; \ - mov D_ = AccumD ;; \ - or M12_ = W_, DTmp ; \ - GETRW(DTmp, Y_, offset) ; \ -} \ -{ .mib ; \ - ld4 M15_ = [DPtr_], 4 ; \ - add BlockCount = -1, BlockCount ; \ - br.call.sptk.many QUICK_RTN = md5_digest_block##offset; \ -} ;; \ -{ .mmi ; \ - add AccumA = AccumA, A_ ; \ - add AccumB = AccumB, B_ ; \ - cmp.ne pAgain, p0 = 0, BlockCount ; \ -} \ -{ .mib ; \ - add AccumC = AccumC, C_ ; \ - add AccumD = AccumD, D_ ; \ -(pAgain) br.cond.dptk.many .md5_block_loop##offset ; \ -} ;; \ -{ .mib ; \ - nop 0x0 ; \ - nop 0x0 ; \ - br.cond.sptk.many .md5_exit ; \ -} ;; - - .align 32 -.md5_unaligned: -// -// Because variable shifts are expensive, we special case each of -// the four alignements. In practice, this won't hurt too much -// since only one working set of code will be loaded. -// -{ .mib - ld4 DTmp = [DPtr_], 4 - cmp.eq pOff, p0 = 1, InAlign -(pOff) br.cond.dpnt.many .md5_process1 -} ;; -{ .mib - cmp.eq pOff, p0 = 2, InAlign - nop 0x0 -(pOff) br.cond.dpnt.many .md5_process2 -} ;; - MD5UNALIGNED(3) - MD5UNALIGNED(1) - MD5UNALIGNED(2) - - .endp md5_block_asm_data_order - - -// MD5 Perform the F function and load -// -// Passed the first 4 words (M0 - M3) and initial (A, B, C, D) values, -// computes the FF() round of functions, then branches to the common -// digest code to finish up with GG(), HH, and II(). -// -// INPUT -// -// rp Return Address - -// -// CODE -// -// v0 PFS bit bucket PFS -// v1 Loop Trip Count LTrip -// pt0 Load next word pMore - -/* For F round: */ -#define LTrip r9 -#define PFS r8 -#define pMore p6 - -/* For GHI rounds: */ -#define T r9 -#define U r10 -#define V r11 - -#define COMPUTE(a, b, s, M, R) \ -{ \ - .mii ; \ - ld4 TRound = [TPtr], 4 ; \ - dep.z Y = Z, 32, 32 ;; \ - shrp Z = Z, Y, 64 - s ; \ -} ;; \ -{ \ - .mmi ; \ - add a = Z, b ; \ - mov R = M ; \ - nop 0x0 ; \ -} ;; - -#define LOOP(a, b, s, M, R, label) \ -{ .mii ; \ - ld4 TRound = [TPtr], 4 ; \ - dep.z Y = Z, 32, 32 ;; \ - shrp Z = Z, Y, 64 - s ; \ -} ;; \ -{ .mib ; \ - add a = Z, b ; \ - mov R = M ; \ - br.ctop.sptk.many label ; \ -} ;; - -// G(B, C, D) = (B & D) | (C & ~D) - -#define G(a, b, c, d, M) \ -{ .mmi ; \ - add Z = M, TRound ; \ - and Y = b, d ; \ - andcm X = c, d ; \ -} ;; \ -{ .mii ; \ - add Z = Z, a ; \ - or Y = Y, X ;; \ - add Z = Z, Y ; \ -} ;; - -// H(B, C, D) = B ^ C ^ D - -#define H(a, b, c, d, M) \ -{ .mmi ; \ - add Z = M, TRound ; \ - xor Y = b, c ; \ - nop 0x0 ; \ -} ;; \ -{ .mii ; \ - add Z = Z, a ; \ - xor Y = Y, d ;; \ - add Z = Z, Y ; \ -} ;; - -// I(B, C, D) = C ^ (B | ~D) -// -// However, since we have an andcm operator, we use the fact that -// -// Y ^ Z == ~Y ^ ~Z -// -// to rewrite the expression as -// -// I(B, C, D) = ~C ^ (~B & D) - -#define I(a, b, c, d, M) \ -{ .mmi ; \ - add Z = M, TRound ; \ - andcm Y = d, b ; \ - andcm X = -1, c ; \ -} ;; \ -{ .mii ; \ - add Z = Z, a ; \ - xor Y = Y, X ;; \ - add Z = Z, Y ; \ -} ;; - -#define GG4(label) \ - G(A, B, C, D, M0) \ - COMPUTE(A, B, 5, M0, RotateM0) \ - G(D, A, B, C, M1) \ - COMPUTE(D, A, 9, M1, RotateM1) \ - G(C, D, A, B, M2) \ - COMPUTE(C, D, 14, M2, RotateM2) \ - G(B, C, D, A, M3) \ - LOOP(B, C, 20, M3, RotateM3, label) - -#define HH4(label) \ - H(A, B, C, D, M0) \ - COMPUTE(A, B, 4, M0, RotateM0) \ - H(D, A, B, C, M1) \ - COMPUTE(D, A, 11, M1, RotateM1) \ - H(C, D, A, B, M2) \ - COMPUTE(C, D, 16, M2, RotateM2) \ - H(B, C, D, A, M3) \ - LOOP(B, C, 23, M3, RotateM3, label) - -#define II4(label) \ - I(A, B, C, D, M0) \ - COMPUTE(A, B, 6, M0, RotateM0) \ - I(D, A, B, C, M1) \ - COMPUTE(D, A, 10, M1, RotateM1) \ - I(C, D, A, B, M2) \ - COMPUTE(C, D, 15, M2, RotateM2) \ - I(B, C, D, A, M3) \ - LOOP(B, C, 21, M3, RotateM3, label) - -#define FFLOAD(a, b, c, d, M, N, s) \ -{ .mii ; \ -(pMore) ld4 N = [DPtr], 4 ; \ - add Z = M, TRound ; \ - and Y = c, b ; \ -} \ -{ .mmi ; \ - andcm X = d, b ;; \ - add Z = Z, a ; \ - or Y = Y, X ; \ -} ;; \ -{ .mii ; \ - ld4 TRound = [TPtr], 4 ; \ - add Z = Z, Y ;; \ - dep.z Y = Z, 32, 32 ; \ -} ;; \ -{ .mii ; \ - nop 0x0 ; \ - shrp Z = Z, Y, 64 - s ;; \ - add a = Z, b ; \ -} ;; - -#define FFLOOP(a, b, c, d, M, N, s, dest) \ -{ .mii ; \ -(pMore) ld4 N = [DPtr], 4 ; \ - add Z = M, TRound ; \ - and Y = c, b ; \ -} \ -{ .mmi ; \ - andcm X = d, b ;; \ - add Z = Z, a ; \ - or Y = Y, X ; \ -} ;; \ -{ .mii ; \ - ld4 TRound = [TPtr], 4 ; \ - add Z = Z, Y ;; \ - dep.z Y = Z, 32, 32 ; \ -} ;; \ -{ .mii ; \ - nop 0x0 ; \ - shrp Z = Z, Y, 64 - s ;; \ - add a = Z, b ; \ -} \ -{ .mib ; \ - cmp.ne pMore, p0 = 0, LTrip ; \ - add LTrip = -1, LTrip ; \ - br.ctop.dptk.many dest ; \ -} ;; - - .type md5_digest_block0, @function - .align 32 - - .proc md5_digest_block0 - .prologue -md5_digest_block0: - .altrp QUICK_RTN - .body -{ .mmi - alloc PFS = ar.pfs, _NINPUTS, _NLOCALS, _NOUTPUT, _NROTATE - mov LTrip = 2 - mov ar.lc = 3 -} ;; -{ .mii - cmp.eq pMore, p0 = r0, r0 - mov ar.ec = 0 - nop 0x0 -} ;; - -.md5_FF_round0: - FFLOAD(A, B, C, D, M12, RotateM0, 7) - FFLOAD(D, A, B, C, M13, RotateM1, 12) - FFLOAD(C, D, A, B, M14, RotateM2, 17) - FFLOOP(B, C, D, A, M15, RotateM3, 22, .md5_FF_round0) - // - // !!! Fall through to md5_digest_GHI - // - .endp md5_digest_block0 - - .type md5_digest_GHI, @function - .align 32 - - .proc md5_digest_GHI - .prologue - .regstk _NINPUTS, _NLOCALS, _NOUTPUT, _NROTATE -md5_digest_GHI: - .altrp QUICK_RTN - .body -// -// The following sequence shuffles the block counstants round for the -// next round: -// -// 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 -// 1 6 11 0 5 10 14 4 9 14 3 8 13 2 7 12 -// -{ .mmi - mov Z = M0 - mov Y = M15 - mov ar.lc = 3 -} -{ .mmi - mov X = M2 - mov W = M9 - mov V = M4 -} ;; - -{ .mmi - mov M0 = M1 - mov M15 = M12 - mov ar.ec = 1 -} -{ .mmi - mov M2 = M11 - mov M9 = M14 - mov M4 = M5 -} ;; - -{ .mmi - mov M1 = M6 - mov M12 = M13 - mov U = M3 -} -{ .mmi - mov M11 = M8 - mov M14 = M7 - mov M5 = M10 -} ;; - -{ .mmi - mov M6 = Y - mov M13 = X - mov M3 = Z -} -{ .mmi - mov M8 = W - mov M7 = V - mov M10 = U -} ;; - -.md5_GG_round: - GG4(.md5_GG_round) - -// The following sequence shuffles the block constants round for the -// next round: -// -// 1 6 11 0 5 10 14 4 9 14 3 8 13 2 7 12 -// 5 8 11 14 1 4 7 10 13 0 3 6 9 12 15 2 - -{ .mmi - mov Z = M0 - mov Y = M1 - mov ar.lc = 3 -} -{ .mmi - mov X = M3 - mov W = M5 - mov V = M6 -} ;; - -{ .mmi - mov M0 = M4 - mov M1 = M11 - mov ar.ec = 1 -} -{ .mmi - mov M3 = M9 - mov U = M8 - mov T = M13 -} ;; - -{ .mmi - mov M4 = Z - mov M11 = Y - mov M5 = M7 -} -{ .mmi - mov M6 = M14 - mov M8 = M12 - mov M13 = M15 -} ;; - -{ .mmi - mov M7 = W - mov M14 = V - nop 0x0 -} -{ .mmi - mov M9 = X - mov M12 = U - mov M15 = T -} ;; - -.md5_HH_round: - HH4(.md5_HH_round) - -// The following sequence shuffles the block constants round for the -// next round: -// -// 5 8 11 14 1 4 7 10 13 0 3 6 9 12 15 2 -// 0 7 14 5 12 3 10 1 8 15 6 13 4 11 2 9 - -{ .mmi - mov Z = M0 - mov Y = M15 - mov ar.lc = 3 -} -{ .mmi - mov X = M10 - mov W = M1 - mov V = M4 -} ;; - -{ .mmi - mov M0 = M9 - mov M15 = M12 - mov ar.ec = 1 -} -{ .mmi - mov M10 = M11 - mov M1 = M6 - mov M4 = M13 -} ;; - -{ .mmi - mov M9 = M14 - mov M12 = M5 - mov U = M3 -} -{ .mmi - mov M11 = M8 - mov M6 = M7 - mov M13 = M2 -} ;; - -{ .mmi - mov M14 = Y - mov M5 = X - mov M3 = Z -} -{ .mmi - mov M8 = W - mov M7 = V - mov M2 = U -} ;; - -.md5_II_round: - II4(.md5_II_round) - -{ .mib - nop 0x0 - nop 0x0 - br.ret.sptk.many QUICK_RTN -} ;; - - .endp md5_digest_GHI - -#define FFLOADU(a, b, c, d, M, P, N, s, offset) \ -{ .mii ; \ -(pMore) ld4 N = [DPtr], 4 ; \ - add Z = M, TRound ; \ - and Y = c, b ; \ -} \ -{ .mmi ; \ - andcm X = d, b ;; \ - add Z = Z, a ; \ - or Y = Y, X ; \ -} ;; \ -{ .mii ; \ - ld4 TRound = [TPtr], 4 ; \ - GETLW(W, P, offset) ; \ - add Z = Z, Y ; \ -} ;; \ -{ .mii ; \ - or W = W, DTmp ; \ - dep.z Y = Z, 32, 32 ;; \ - shrp Z = Z, Y, 64 - s ; \ -} ;; \ -{ .mii ; \ - add a = Z, b ; \ - GETRW(DTmp, P, offset) ; \ - mov P = W ; \ -} ;; - -#define FFLOOPU(a, b, c, d, M, P, N, s, offset) \ -{ .mii ; \ -(pMore) ld4 N = [DPtr], 4 ; \ - add Z = M, TRound ; \ - and Y = c, b ; \ -} \ -{ .mmi ; \ - andcm X = d, b ;; \ - add Z = Z, a ; \ - or Y = Y, X ; \ -} ;; \ -{ .mii ; \ - ld4 TRound = [TPtr], 4 ; \ -(pMore) GETLW(W, P, offset) ; \ - add Z = Z, Y ; \ -} ;; \ -{ .mii ; \ -(pMore) or W = W, DTmp ; \ - dep.z Y = Z, 32, 32 ;; \ - shrp Z = Z, Y, 64 - s ; \ -} ;; \ -{ .mii ; \ - add a = Z, b ; \ -(pMore) GETRW(DTmp, P, offset) ; \ -(pMore) mov P = W ; \ -} \ -{ .mib ; \ - cmp.ne pMore, p0 = 0, LTrip ; \ - add LTrip = -1, LTrip ; \ - br.ctop.sptk.many .md5_FF_round##offset ; \ -} ;; - -#define MD5FBLOCK(offset) \ - .type md5_digest_block##offset, @function ; \ - \ - .align 32 ; \ - .proc md5_digest_block##offset ; \ - .prologue ; \ - .altrp QUICK_RTN ; \ - .body ; \ -md5_digest_block##offset: \ -{ .mmi ; \ - alloc PFS = ar.pfs, _NINPUTS, _NLOCALS, _NOUTPUT, _NROTATE ; \ - mov LTrip = 2 ; \ - mov ar.lc = 3 ; \ -} ;; \ -{ .mii ; \ - cmp.eq pMore, p0 = r0, r0 ; \ - mov ar.ec = 0 ; \ - nop 0x0 ; \ -} ;; \ - \ - .pred.rel "mutex", pLoad, pSkip ; \ -.md5_FF_round##offset: \ - FFLOADU(A, B, C, D, M12, M13, RotateM0, 7, offset) \ - FFLOADU(D, A, B, C, M13, M14, RotateM1, 12, offset) \ - FFLOADU(C, D, A, B, M14, M15, RotateM2, 17, offset) \ - FFLOOPU(B, C, D, A, M15, RotateM0, RotateM3, 22, offset) \ - \ -{ .mib ; \ - nop 0x0 ; \ - nop 0x0 ; \ - br.cond.sptk.many md5_digest_GHI ; \ -} ;; \ - .endp md5_digest_block##offset - -MD5FBLOCK(1) -MD5FBLOCK(2) -MD5FBLOCK(3) - - .align 64 - .type md5_constants, @object -md5_constants: -.md5_tbl_data_order: // To ensure little-endian data - // order, code as bytes. - data1 0x78, 0xa4, 0x6a, 0xd7 // 0 - data1 0x56, 0xb7, 0xc7, 0xe8 // 1 - data1 0xdb, 0x70, 0x20, 0x24 // 2 - data1 0xee, 0xce, 0xbd, 0xc1 // 3 - data1 0xaf, 0x0f, 0x7c, 0xf5 // 4 - data1 0x2a, 0xc6, 0x87, 0x47 // 5 - data1 0x13, 0x46, 0x30, 0xa8 // 6 - data1 0x01, 0x95, 0x46, 0xfd // 7 - data1 0xd8, 0x98, 0x80, 0x69 // 8 - data1 0xaf, 0xf7, 0x44, 0x8b // 9 - data1 0xb1, 0x5b, 0xff, 0xff // 10 - data1 0xbe, 0xd7, 0x5c, 0x89 // 11 - data1 0x22, 0x11, 0x90, 0x6b // 12 - data1 0x93, 0x71, 0x98, 0xfd // 13 - data1 0x8e, 0x43, 0x79, 0xa6 // 14 - data1 0x21, 0x08, 0xb4, 0x49 // 15 - data1 0x62, 0x25, 0x1e, 0xf6 // 16 - data1 0x40, 0xb3, 0x40, 0xc0 // 17 - data1 0x51, 0x5a, 0x5e, 0x26 // 18 - data1 0xaa, 0xc7, 0xb6, 0xe9 // 19 - data1 0x5d, 0x10, 0x2f, 0xd6 // 20 - data1 0x53, 0x14, 0x44, 0x02 // 21 - data1 0x81, 0xe6, 0xa1, 0xd8 // 22 - data1 0xc8, 0xfb, 0xd3, 0xe7 // 23 - data1 0xe6, 0xcd, 0xe1, 0x21 // 24 - data1 0xd6, 0x07, 0x37, 0xc3 // 25 - data1 0x87, 0x0d, 0xd5, 0xf4 // 26 - data1 0xed, 0x14, 0x5a, 0x45 // 27 - data1 0x05, 0xe9, 0xe3, 0xa9 // 28 - data1 0xf8, 0xa3, 0xef, 0xfc // 29 - data1 0xd9, 0x02, 0x6f, 0x67 // 30 - data1 0x8a, 0x4c, 0x2a, 0x8d // 31 - data1 0x42, 0x39, 0xfa, 0xff // 32 - data1 0x81, 0xf6, 0x71, 0x87 // 33 - data1 0x22, 0x61, 0x9d, 0x6d // 34 - data1 0x0c, 0x38, 0xe5, 0xfd // 35 - data1 0x44, 0xea, 0xbe, 0xa4 // 36 - data1 0xa9, 0xcf, 0xde, 0x4b // 37 - data1 0x60, 0x4b, 0xbb, 0xf6 // 38 - data1 0x70, 0xbc, 0xbf, 0xbe // 39 - data1 0xc6, 0x7e, 0x9b, 0x28 // 40 - data1 0xfa, 0x27, 0xa1, 0xea // 41 - data1 0x85, 0x30, 0xef, 0xd4 // 42 - data1 0x05, 0x1d, 0x88, 0x04 // 43 - data1 0x39, 0xd0, 0xd4, 0xd9 // 44 - data1 0xe5, 0x99, 0xdb, 0xe6 // 45 - data1 0xf8, 0x7c, 0xa2, 0x1f // 46 - data1 0x65, 0x56, 0xac, 0xc4 // 47 - data1 0x44, 0x22, 0x29, 0xf4 // 48 - data1 0x97, 0xff, 0x2a, 0x43 // 49 - data1 0xa7, 0x23, 0x94, 0xab // 50 - data1 0x39, 0xa0, 0x93, 0xfc // 51 - data1 0xc3, 0x59, 0x5b, 0x65 // 52 - data1 0x92, 0xcc, 0x0c, 0x8f // 53 - data1 0x7d, 0xf4, 0xef, 0xff // 54 - data1 0xd1, 0x5d, 0x84, 0x85 // 55 - data1 0x4f, 0x7e, 0xa8, 0x6f // 56 - data1 0xe0, 0xe6, 0x2c, 0xfe // 57 - data1 0x14, 0x43, 0x01, 0xa3 // 58 - data1 0xa1, 0x11, 0x08, 0x4e // 59 - data1 0x82, 0x7e, 0x53, 0xf7 // 60 - data1 0x35, 0xf2, 0x3a, 0xbd // 61 - data1 0xbb, 0xd2, 0xd7, 0x2a // 62 - data1 0x91, 0xd3, 0x86, 0xeb // 63 -.size md5_constants#,64*4 diff --git a/src/lib/libcrypto/modes/asm/ghash-ia64.pl b/src/lib/libcrypto/modes/asm/ghash-ia64.pl deleted file mode 100755 index 0354c95444..0000000000 --- a/src/lib/libcrypto/modes/asm/ghash-ia64.pl +++ /dev/null @@ -1,463 +0,0 @@ -#!/usr/bin/env perl - -# ==================================================================== -# Written by Andy Polyakov 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/. -# ==================================================================== -# -# March 2010 -# -# The module implements "4-bit" GCM GHASH function and underlying -# single multiplication operation in GF(2^128). "4-bit" means that it -# uses 256 bytes per-key table [+128 bytes shared table]. Streamed -# GHASH performance was measured to be 6.67 cycles per processed byte -# on Itanium 2, which is >90% better than Microsoft compiler generated -# code. To anchor to something else sha1-ia64.pl module processes one -# byte in 5.7 cycles. On Itanium GHASH should run at ~8.5 cycles per -# byte. - -# September 2010 -# -# It was originally thought that it makes lesser sense to implement -# "528B" variant on Itanium 2 for following reason. Because number of -# functional units is naturally limited, it appeared impossible to -# implement "528B" loop in 4 cycles, only in 5. This would mean that -# theoretically performance improvement couldn't be more than 20%. -# But occasionally you prove yourself wrong:-) I figured out a way to -# fold couple of instructions and having freed yet another instruction -# slot by unrolling the loop... Resulting performance is 4.45 cycles -# per processed byte and 50% better than "256B" version. On original -# Itanium performance should remain the same as the "256B" version, -# i.e. ~8.5 cycles. - -$output=shift and (open STDOUT,">$output" or die "can't open $output: $!"); - -if ($^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); } - -sub loop() { -my $label=shift; -my ($p16,$p17)=(shift)?("p63","p63"):("p16","p17"); # mask references to inp - -# Loop is scheduled for 6 ticks on Itanium 2 and 8 on Itanium, i.e. -# in scalable manner;-) Naturally assuming data in L1 cache... -# Special note about 'dep' instruction, which is used to construct -# &rem_4bit[Zlo&0xf]. It works, because rem_4bit is aligned at 128 -# bytes boundary and lower 7 bits of its address are guaranteed to -# be zero. -$code.=<<___; -$label: -{ .mfi; (p18) ld8 Hlo=[Hi[1]],-8 - (p19) dep rem=Zlo,rem_4bitp,3,4 } -{ .mfi; (p19) xor Zhi=Zhi,Hhi - ($p17) xor xi[1]=xi[1],in[1] };; -{ .mfi; (p18) ld8 Hhi=[Hi[1]] - (p19) shrp Zlo=Zhi,Zlo,4 } -{ .mfi; (p19) ld8 rem=[rem] - (p18) and Hi[1]=mask0xf0,xi[2] };; -{ .mmi; ($p16) ld1 in[0]=[inp],-1 - (p18) xor Zlo=Zlo,Hlo - (p19) shr.u Zhi=Zhi,4 } -{ .mib; (p19) xor Hhi=Hhi,rem - (p18) add Hi[1]=Htbl,Hi[1] };; - -{ .mfi; (p18) ld8 Hlo=[Hi[1]],-8 - (p18) dep rem=Zlo,rem_4bitp,3,4 } -{ .mfi; (p17) shladd Hi[0]=xi[1],4,r0 - (p18) xor Zhi=Zhi,Hhi };; -{ .mfi; (p18) ld8 Hhi=[Hi[1]] - (p18) shrp Zlo=Zhi,Zlo,4 } -{ .mfi; (p18) ld8 rem=[rem] - (p17) and Hi[0]=mask0xf0,Hi[0] };; -{ .mmi; (p16) ld1 xi[0]=[Xi],-1 - (p18) xor Zlo=Zlo,Hlo - (p18) shr.u Zhi=Zhi,4 } -{ .mib; (p18) xor Hhi=Hhi,rem - (p17) add Hi[0]=Htbl,Hi[0] - br.ctop.sptk $label };; -___ -} - -$code=<<___; -.explicit -.text - -prevfs=r2; prevlc=r3; prevpr=r8; -mask0xf0=r21; -rem=r22; rem_4bitp=r23; -Xi=r24; Htbl=r25; -inp=r26; end=r27; -Hhi=r28; Hlo=r29; -Zhi=r30; Zlo=r31; - -.align 128 -.skip 16 // aligns loop body -.global gcm_gmult_4bit# -.proc gcm_gmult_4bit# -gcm_gmult_4bit: - .prologue -{ .mmi; .save ar.pfs,prevfs - alloc prevfs=ar.pfs,2,6,0,8 - $ADDP Xi=15,in0 // &Xi[15] - mov rem_4bitp=ip } -{ .mii; $ADDP Htbl=8,in1 // &Htbl[0].lo - .save ar.lc,prevlc - mov prevlc=ar.lc - .save pr,prevpr - mov prevpr=pr };; - - .body - .rotr in[3],xi[3],Hi[2] - -{ .mib; ld1 xi[2]=[Xi],-1 // Xi[15] - mov mask0xf0=0xf0 - brp.loop.imp .Loop1,.Lend1-16};; -{ .mmi; ld1 xi[1]=[Xi],-1 // Xi[14] - };; -{ .mii; shladd Hi[1]=xi[2],4,r0 - mov pr.rot=0x7<<16 - mov ar.lc=13 };; -{ .mii; and Hi[1]=mask0xf0,Hi[1] - mov ar.ec=3 - xor Zlo=Zlo,Zlo };; -{ .mii; add Hi[1]=Htbl,Hi[1] // &Htbl[nlo].lo - add rem_4bitp=rem_4bit#-gcm_gmult_4bit#,rem_4bitp - xor Zhi=Zhi,Zhi };; -___ - &loop (".Loop1",1); -$code.=<<___; -.Lend1: -{ .mib; xor Zhi=Zhi,Hhi };; // modulo-scheduling artefact -{ .mib; mux1 Zlo=Zlo,\@rev };; -{ .mib; mux1 Zhi=Zhi,\@rev };; -{ .mmi; add Hlo=9,Xi;; // ;; is here to prevent - add Hhi=1,Xi };; // pipeline flush on Itanium -{ .mib; st8 [Hlo]=Zlo - mov pr=prevpr,0x1ffff };; -{ .mib; st8 [Hhi]=Zhi - mov ar.lc=prevlc - br.ret.sptk.many b0 };; -.endp gcm_gmult_4bit# -___ - -###################################################################### -# "528B" (well, "512B" actualy) streamed GHASH -# -$Xip="in0"; -$Htbl="in1"; -$inp="in2"; -$len="in3"; -$rem_8bit="loc0"; -$mask0xff="loc1"; -($sum,$rum) = $big_endian ? ("nop.m","nop.m") : ("sum","rum"); - -sub load_htable() { - for (my $i=0;$i<8;$i++) { - $code.=<<___; -{ .mmi; ld8 r`16+2*$i+1`=[r8],16 // Htable[$i].hi - ld8 r`16+2*$i`=[r9],16 } // Htable[$i].lo -{ .mmi; ldf8 f`32+2*$i+1`=[r10],16 // Htable[`8+$i`].hi - ldf8 f`32+2*$i`=[r11],16 // Htable[`8+$i`].lo -___ - $code.=shift if (($i+$#_)==7); - $code.="\t};;\n" - } -} - -$code.=<<___; -prevsp=r3; - -.align 32 -.skip 16 // aligns loop body -.global gcm_ghash_4bit# -.proc gcm_ghash_4bit# -gcm_ghash_4bit: - .prologue -{ .mmi; .save ar.pfs,prevfs - alloc prevfs=ar.pfs,4,2,0,0 - .vframe prevsp - mov prevsp=sp - mov $rem_8bit=ip };; - .body -{ .mfi; $ADDP r8=0+0,$Htbl - $ADDP r9=0+8,$Htbl } -{ .mfi; $ADDP r10=128+0,$Htbl - $ADDP r11=128+8,$Htbl };; -___ - &load_htable( - " $ADDP $Xip=15,$Xip", # &Xi[15] - " $ADDP $len=$len,$inp", # &inp[len] - " $ADDP $inp=15,$inp", # &inp[15] - " mov $mask0xff=0xff", - " add sp=-512,sp", - " andcm sp=sp,$mask0xff", # align stack frame - " add r14=0,sp", - " add r15=8,sp"); -$code.=<<___; -{ .mmi; $sum 1<<1 // go big-endian - add r8=256+0,sp - add r9=256+8,sp } -{ .mmi; add r10=256+128+0,sp - add r11=256+128+8,sp - add $len=-17,$len };; -___ -for($i=0;$i<8;$i++) { # generate first half of Hshr4[] -my ($rlo,$rhi)=("r".eval(16+2*$i),"r".eval(16+2*$i+1)); -$code.=<<___; -{ .mmi; st8 [r8]=$rlo,16 // Htable[$i].lo - st8 [r9]=$rhi,16 // Htable[$i].hi - shrp $rlo=$rhi,$rlo,4 }//;; -{ .mmi; stf8 [r10]=f`32+2*$i`,16 // Htable[`8+$i`].lo - stf8 [r11]=f`32+2*$i+1`,16 // Htable[`8+$i`].hi - shr.u $rhi=$rhi,4 };; -{ .mmi; st8 [r14]=$rlo,16 // Htable[$i].lo>>4 - st8 [r15]=$rhi,16 }//;; // Htable[$i].hi>>4 -___ -} -$code.=<<___; -{ .mmi; ld8 r16=[r8],16 // Htable[8].lo - ld8 r17=[r9],16 };; // Htable[8].hi -{ .mmi; ld8 r18=[r8],16 // Htable[9].lo - ld8 r19=[r9],16 } // Htable[9].hi -{ .mmi; rum 1<<5 // clear um.mfh - shrp r16=r17,r16,4 };; -___ -for($i=0;$i<6;$i++) { # generate second half of Hshr4[] -$code.=<<___; -{ .mmi; ld8 r`20+2*$i`=[r8],16 // Htable[`10+$i`].lo - ld8 r`20+2*$i+1`=[r9],16 // Htable[`10+$i`].hi - shr.u r`16+2*$i+1`=r`16+2*$i+1`,4 };; -{ .mmi; st8 [r14]=r`16+2*$i`,16 // Htable[`8+$i`].lo>>4 - st8 [r15]=r`16+2*$i+1`,16 // Htable[`8+$i`].hi>>4 - shrp r`18+2*$i`=r`18+2*$i+1`,r`18+2*$i`,4 } -___ -} -$code.=<<___; -{ .mmi; shr.u r`16+2*$i+1`=r`16+2*$i+1`,4 };; -{ .mmi; st8 [r14]=r`16+2*$i`,16 // Htable[`8+$i`].lo>>4 - st8 [r15]=r`16+2*$i+1`,16 // Htable[`8+$i`].hi>>4 - shrp r`18+2*$i`=r`18+2*$i+1`,r`18+2*$i`,4 } -{ .mmi; add $Htbl=256,sp // &Htable[0] - add $rem_8bit=rem_8bit#-gcm_ghash_4bit#,$rem_8bit - shr.u r`18+2*$i+1`=r`18+2*$i+1`,4 };; -{ .mmi; st8 [r14]=r`18+2*$i` // Htable[`8+$i`].lo>>4 - st8 [r15]=r`18+2*$i+1` } // Htable[`8+$i`].hi>>4 -___ - -$in="r15"; -@xi=("r16","r17"); -@rem=("r18","r19"); -($Alo,$Ahi,$Blo,$Bhi,$Zlo,$Zhi)=("r20","r21","r22","r23","r24","r25"); -($Atbl,$Btbl)=("r26","r27"); - -$code.=<<___; # (p16) -{ .mmi; ld1 $in=[$inp],-1 //(p16) *inp-- - ld1 $xi[0]=[$Xip],-1 //(p16) *Xi-- - cmp.eq p0,p6=r0,r0 };; // clear p6 -___ -push (@xi,shift(@xi)); push (@rem,shift(@rem)); # "rotate" registers - -$code.=<<___; # (p16),(p17) -{ .mmi; ld1 $xi[0]=[$Xip],-1 //(p16) *Xi-- - xor $xi[1]=$xi[1],$in };; //(p17) xi=$xi[i]^inp[i] -{ .mii; ld1 $in=[$inp],-1 //(p16) *inp-- - dep $Atbl=$xi[1],$Htbl,4,4 //(p17) &Htable[nlo].lo - and $xi[1]=-16,$xi[1] };; //(p17) nhi=xi&0xf0 -.align 32 -.LOOP: -{ .mmi; -(p6) st8 [$Xip]=$Zhi,13 - xor $Zlo=$Zlo,$Zlo - add $Btbl=$xi[1],$Htbl };; //(p17) &Htable[nhi].lo -___ -push (@xi,shift(@xi)); push (@rem,shift(@rem)); # "rotate" registers - -$code.=<<___; # (p16),(p17),(p18) -{ .mmi; ld8 $Alo=[$Atbl],8 //(p18) Htable[nlo].lo,&Htable[nlo].hi - ld8 $rem[0]=[$Btbl],-256 //(p18) Htable[nhi].lo,&Hshr4[nhi].lo - xor $xi[1]=$xi[1],$in };; //(p17) xi=$xi[i]^inp[i] -{ .mfi; ld8 $Ahi=[$Atbl] //(p18) Htable[nlo].hi - dep $Atbl=$xi[1],$Htbl,4,4 } //(p17) &Htable[nlo].lo -{ .mfi; shladd $rem[0]=$rem[0],4,r0 //(p18) Htable[nhi].lo<<4 - xor $Zlo=$Zlo,$Alo };; //(p18) Z.lo^=Htable[nlo].lo -{ .mmi; ld8 $Blo=[$Btbl],8 //(p18) Hshr4[nhi].lo,&Hshr4[nhi].hi - ld1 $in=[$inp],-1 } //(p16) *inp-- -{ .mmi; xor $rem[0]=$rem[0],$Zlo //(p18) Z.lo^(Htable[nhi].lo<<4) - mov $Zhi=$Ahi //(p18) Z.hi^=Htable[nlo].hi - and $xi[1]=-16,$xi[1] };; //(p17) nhi=xi&0xf0 -{ .mmi; ld8 $Bhi=[$Btbl] //(p18) Hshr4[nhi].hi - ld1 $xi[0]=[$Xip],-1 //(p16) *Xi-- - shrp $Zlo=$Zhi,$Zlo,8 } //(p18) Z.lo=(Z.hi<<56)|(Z.lo>>8) -{ .mmi; and $rem[0]=$rem[0],$mask0xff //(p18) rem=($Zlo^(Htable[nhi].lo<<4))&0xff - add $Btbl=$xi[1],$Htbl };; //(p17) &Htable[nhi] -___ -push (@xi,shift(@xi)); push (@rem,shift(@rem)); # "rotate" registers - -for ($i=1;$i<14;$i++) { -# Above and below fragments are derived from this one by removing -# unsuitable (p??) instructions. -$code.=<<___; # (p16),(p17),(p18),(p19) -{ .mmi; ld8 $Alo=[$Atbl],8 //(p18) Htable[nlo].lo,&Htable[nlo].hi - ld8 $rem[0]=[$Btbl],-256 //(p18) Htable[nhi].lo,&Hshr4[nhi].lo - shr.u $Zhi=$Zhi,8 } //(p19) Z.hi>>=8 -{ .mmi; shladd $rem[1]=$rem[1],1,$rem_8bit //(p19) &rem_8bit[rem] - xor $Zlo=$Zlo,$Blo //(p19) Z.lo^=Hshr4[nhi].lo - xor $xi[1]=$xi[1],$in };; //(p17) xi=$xi[i]^inp[i] -{ .mmi; ld8 $Ahi=[$Atbl] //(p18) Htable[nlo].hi - ld2 $rem[1]=[$rem[1]] //(p19) rem_8bit[rem] - dep $Atbl=$xi[1],$Htbl,4,4 } //(p17) &Htable[nlo].lo -{ .mmi; shladd $rem[0]=$rem[0],4,r0 //(p18) Htable[nhi].lo<<4 - xor $Zlo=$Zlo,$Alo //(p18) Z.lo^=Htable[nlo].lo - xor $Zhi=$Zhi,$Bhi };; //(p19) Z.hi^=Hshr4[nhi].hi -{ .mmi; ld8 $Blo=[$Btbl],8 //(p18) Hshr4[nhi].lo,&Hshr4[nhi].hi - ld1 $in=[$inp],-1 //(p16) *inp-- - shl $rem[1]=$rem[1],48 } //(p19) rem_8bit[rem]<<48 -{ .mmi; xor $rem[0]=$rem[0],$Zlo //(p18) Z.lo^(Htable[nhi].lo<<4) - xor $Zhi=$Zhi,$Ahi //(p18) Z.hi^=Htable[nlo].hi - and $xi[1]=-16,$xi[1] };; //(p17) nhi=xi&0xf0 -{ .mmi; ld8 $Bhi=[$Btbl] //(p18) Hshr4[nhi].hi - ld1 $xi[0]=[$Xip],-1 //(p16) *Xi-- - shrp $Zlo=$Zhi,$Zlo,8 } //(p18) Z.lo=(Z.hi<<56)|(Z.lo>>8) -{ .mmi; and $rem[0]=$rem[0],$mask0xff //(p18) rem=($Zlo^(Htable[nhi].lo<<4))&0xff - xor $Zhi=$Zhi,$rem[1] //(p19) Z.hi^=rem_8bit[rem]<<48 - add $Btbl=$xi[1],$Htbl };; //(p17) &Htable[nhi] -___ -push (@xi,shift(@xi)); push (@rem,shift(@rem)); # "rotate" registers -} - -$code.=<<___; # (p17),(p18),(p19) -{ .mmi; ld8 $Alo=[$Atbl],8 //(p18) Htable[nlo].lo,&Htable[nlo].hi - ld8 $rem[0]=[$Btbl],-256 //(p18) Htable[nhi].lo,&Hshr4[nhi].lo - shr.u $Zhi=$Zhi,8 } //(p19) Z.hi>>=8 -{ .mmi; shladd $rem[1]=$rem[1],1,$rem_8bit //(p19) &rem_8bit[rem] - xor $Zlo=$Zlo,$Blo //(p19) Z.lo^=Hshr4[nhi].lo - xor $xi[1]=$xi[1],$in };; //(p17) xi=$xi[i]^inp[i] -{ .mmi; ld8 $Ahi=[$Atbl] //(p18) Htable[nlo].hi - ld2 $rem[1]=[$rem[1]] //(p19) rem_8bit[rem] - dep $Atbl=$xi[1],$Htbl,4,4 };; //(p17) &Htable[nlo].lo -{ .mmi; shladd $rem[0]=$rem[0],4,r0 //(p18) Htable[nhi].lo<<4 - xor $Zlo=$Zlo,$Alo //(p18) Z.lo^=Htable[nlo].lo - xor $Zhi=$Zhi,$Bhi };; //(p19) Z.hi^=Hshr4[nhi].hi -{ .mmi; ld8 $Blo=[$Btbl],8 //(p18) Hshr4[nhi].lo,&Hshr4[nhi].hi - shl $rem[1]=$rem[1],48 } //(p19) rem_8bit[rem]<<48 -{ .mmi; xor $rem[0]=$rem[0],$Zlo //(p18) Z.lo^(Htable[nhi].lo<<4) - xor $Zhi=$Zhi,$Ahi //(p18) Z.hi^=Htable[nlo].hi - and $xi[1]=-16,$xi[1] };; //(p17) nhi=xi&0xf0 -{ .mmi; ld8 $Bhi=[$Btbl] //(p18) Hshr4[nhi].hi - shrp $Zlo=$Zhi,$Zlo,8 } //(p18) Z.lo=(Z.hi<<56)|(Z.lo>>8) -{ .mmi; and $rem[0]=$rem[0],$mask0xff //(p18) rem=($Zlo^(Htable[nhi].lo<<4))&0xff - xor $Zhi=$Zhi,$rem[1] //(p19) Z.hi^=rem_8bit[rem]<<48 - add $Btbl=$xi[1],$Htbl };; //(p17) &Htable[nhi] -___ -push (@xi,shift(@xi)); push (@rem,shift(@rem)); # "rotate" registers - -$code.=<<___; # (p18),(p19) -{ .mfi; ld8 $Alo=[$Atbl],8 //(p18) Htable[nlo].lo,&Htable[nlo].hi - shr.u $Zhi=$Zhi,8 } //(p19) Z.hi>>=8 -{ .mfi; shladd $rem[1]=$rem[1],1,$rem_8bit //(p19) &rem_8bit[rem] - xor $Zlo=$Zlo,$Blo };; //(p19) Z.lo^=Hshr4[nhi].lo -{ .mfi; ld8 $Ahi=[$Atbl] //(p18) Htable[nlo].hi - xor $Zlo=$Zlo,$Alo } //(p18) Z.lo^=Htable[nlo].lo -{ .mfi; ld2 $rem[1]=[$rem[1]] //(p19) rem_8bit[rem] - xor $Zhi=$Zhi,$Bhi };; //(p19) Z.hi^=Hshr4[nhi].hi -{ .mfi; ld8 $Blo=[$Btbl],8 //(p18) Htable[nhi].lo,&Htable[nhi].hi - shl $rem[1]=$rem[1],48 } //(p19) rem_8bit[rem]<<48 -{ .mfi; shladd $rem[0]=$Zlo,4,r0 //(p18) Z.lo<<4 - xor $Zhi=$Zhi,$Ahi };; //(p18) Z.hi^=Htable[nlo].hi -{ .mfi; ld8 $Bhi=[$Btbl] //(p18) Htable[nhi].hi - shrp $Zlo=$Zhi,$Zlo,4 } //(p18) Z.lo=(Z.hi<<60)|(Z.lo>>4) -{ .mfi; and $rem[0]=$rem[0],$mask0xff //(p18) rem=($Zlo^(Htable[nhi].lo<<4))&0xff - xor $Zhi=$Zhi,$rem[1] };; //(p19) Z.hi^=rem_8bit[rem]<<48 -___ -push (@xi,shift(@xi)); push (@rem,shift(@rem)); # "rotate" registers - -$code.=<<___; # (p19) -{ .mmi; cmp.ltu p6,p0=$inp,$len - add $inp=32,$inp - shr.u $Zhi=$Zhi,4 } //(p19) Z.hi>>=4 -{ .mmi; shladd $rem[1]=$rem[1],1,$rem_8bit //(p19) &rem_8bit[rem] - xor $Zlo=$Zlo,$Blo //(p19) Z.lo^=Hshr4[nhi].lo - add $Xip=9,$Xip };; // &Xi.lo -{ .mmi; ld2 $rem[1]=[$rem[1]] //(p19) rem_8bit[rem] -(p6) ld1 $in=[$inp],-1 //[p16] *inp-- -(p6) extr.u $xi[1]=$Zlo,8,8 } //[p17] Xi[14] -{ .mmi; xor $Zhi=$Zhi,$Bhi //(p19) Z.hi^=Hshr4[nhi].hi -(p6) and $xi[0]=$Zlo,$mask0xff };; //[p16] Xi[15] -{ .mmi; st8 [$Xip]=$Zlo,-8 -(p6) xor $xi[0]=$xi[0],$in //[p17] xi=$xi[i]^inp[i] - shl $rem[1]=$rem[1],48 };; //(p19) rem_8bit[rem]<<48 -{ .mmi; -(p6) ld1 $in=[$inp],-1 //[p16] *inp-- - xor $Zhi=$Zhi,$rem[1] //(p19) Z.hi^=rem_8bit[rem]<<48 -(p6) dep $Atbl=$xi[0],$Htbl,4,4 } //[p17] &Htable[nlo].lo -{ .mib; -(p6) and $xi[0]=-16,$xi[0] //[p17] nhi=xi&0xf0 -(p6) br.cond.dptk.many .LOOP };; - -{ .mib; st8 [$Xip]=$Zhi };; -{ .mib; $rum 1<<1 // return to little-endian - .restore sp - mov sp=prevsp - br.ret.sptk.many b0 };; -.endp gcm_ghash_4bit# -___ -$code.=<<___; -.align 128 -.type rem_4bit#,\@object -rem_4bit: - data8 0x0000<<48, 0x1C20<<48, 0x3840<<48, 0x2460<<48 - data8 0x7080<<48, 0x6CA0<<48, 0x48C0<<48, 0x54E0<<48 - data8 0xE100<<48, 0xFD20<<48, 0xD940<<48, 0xC560<<48 - data8 0x9180<<48, 0x8DA0<<48, 0xA9C0<<48, 0xB5E0<<48 -.size rem_4bit#,128 -.type rem_8bit#,\@object -rem_8bit: - data1 0x00,0x00, 0x01,0xC2, 0x03,0x84, 0x02,0x46, 0x07,0x08, 0x06,0xCA, 0x04,0x8C, 0x05,0x4E - data1 0x0E,0x10, 0x0F,0xD2, 0x0D,0x94, 0x0C,0x56, 0x09,0x18, 0x08,0xDA, 0x0A,0x9C, 0x0B,0x5E - data1 0x1C,0x20, 0x1D,0xE2, 0x1F,0xA4, 0x1E,0x66, 0x1B,0x28, 0x1A,0xEA, 0x18,0xAC, 0x19,0x6E - data1 0x12,0x30, 0x13,0xF2, 0x11,0xB4, 0x10,0x76, 0x15,0x38, 0x14,0xFA, 0x16,0xBC, 0x17,0x7E - data1 0x38,0x40, 0x39,0x82, 0x3B,0xC4, 0x3A,0x06, 0x3F,0x48, 0x3E,0x8A, 0x3C,0xCC, 0x3D,0x0E - data1 0x36,0x50, 0x37,0x92, 0x35,0xD4, 0x34,0x16, 0x31,0x58, 0x30,0x9A, 0x32,0xDC, 0x33,0x1E - data1 0x24,0x60, 0x25,0xA2, 0x27,0xE4, 0x26,0x26, 0x23,0x68, 0x22,0xAA, 0x20,0xEC, 0x21,0x2E - data1 0x2A,0x70, 0x2B,0xB2, 0x29,0xF4, 0x28,0x36, 0x2D,0x78, 0x2C,0xBA, 0x2E,0xFC, 0x2F,0x3E - data1 0x70,0x80, 0x71,0x42, 0x73,0x04, 0x72,0xC6, 0x77,0x88, 0x76,0x4A, 0x74,0x0C, 0x75,0xCE - data1 0x7E,0x90, 0x7F,0x52, 0x7D,0x14, 0x7C,0xD6, 0x79,0x98, 0x78,0x5A, 0x7A,0x1C, 0x7B,0xDE - data1 0x6C,0xA0, 0x6D,0x62, 0x6F,0x24, 0x6E,0xE6, 0x6B,0xA8, 0x6A,0x6A, 0x68,0x2C, 0x69,0xEE - data1 0x62,0xB0, 0x63,0x72, 0x61,0x34, 0x60,0xF6, 0x65,0xB8, 0x64,0x7A, 0x66,0x3C, 0x67,0xFE - data1 0x48,0xC0, 0x49,0x02, 0x4B,0x44, 0x4A,0x86, 0x4F,0xC8, 0x4E,0x0A, 0x4C,0x4C, 0x4D,0x8E - data1 0x46,0xD0, 0x47,0x12, 0x45,0x54, 0x44,0x96, 0x41,0xD8, 0x40,0x1A, 0x42,0x5C, 0x43,0x9E - data1 0x54,0xE0, 0x55,0x22, 0x57,0x64, 0x56,0xA6, 0x53,0xE8, 0x52,0x2A, 0x50,0x6C, 0x51,0xAE - data1 0x5A,0xF0, 0x5B,0x32, 0x59,0x74, 0x58,0xB6, 0x5D,0xF8, 0x5C,0x3A, 0x5E,0x7C, 0x5F,0xBE - data1 0xE1,0x00, 0xE0,0xC2, 0xE2,0x84, 0xE3,0x46, 0xE6,0x08, 0xE7,0xCA, 0xE5,0x8C, 0xE4,0x4E - data1 0xEF,0x10, 0xEE,0xD2, 0xEC,0x94, 0xED,0x56, 0xE8,0x18, 0xE9,0xDA, 0xEB,0x9C, 0xEA,0x5E - data1 0xFD,0x20, 0xFC,0xE2, 0xFE,0xA4, 0xFF,0x66, 0xFA,0x28, 0xFB,0xEA, 0xF9,0xAC, 0xF8,0x6E - data1 0xF3,0x30, 0xF2,0xF2, 0xF0,0xB4, 0xF1,0x76, 0xF4,0x38, 0xF5,0xFA, 0xF7,0xBC, 0xF6,0x7E - data1 0xD9,0x40, 0xD8,0x82, 0xDA,0xC4, 0xDB,0x06, 0xDE,0x48, 0xDF,0x8A, 0xDD,0xCC, 0xDC,0x0E - data1 0xD7,0x50, 0xD6,0x92, 0xD4,0xD4, 0xD5,0x16, 0xD0,0x58, 0xD1,0x9A, 0xD3,0xDC, 0xD2,0x1E - data1 0xC5,0x60, 0xC4,0xA2, 0xC6,0xE4, 0xC7,0x26, 0xC2,0x68, 0xC3,0xAA, 0xC1,0xEC, 0xC0,0x2E - data1 0xCB,0x70, 0xCA,0xB2, 0xC8,0xF4, 0xC9,0x36, 0xCC,0x78, 0xCD,0xBA, 0xCF,0xFC, 0xCE,0x3E - data1 0x91,0x80, 0x90,0x42, 0x92,0x04, 0x93,0xC6, 0x96,0x88, 0x97,0x4A, 0x95,0x0C, 0x94,0xCE - data1 0x9F,0x90, 0x9E,0x52, 0x9C,0x14, 0x9D,0xD6, 0x98,0x98, 0x99,0x5A, 0x9B,0x1C, 0x9A,0xDE - data1 0x8D,0xA0, 0x8C,0x62, 0x8E,0x24, 0x8F,0xE6, 0x8A,0xA8, 0x8B,0x6A, 0x89,0x2C, 0x88,0xEE - data1 0x83,0xB0, 0x82,0x72, 0x80,0x34, 0x81,0xF6, 0x84,0xB8, 0x85,0x7A, 0x87,0x3C, 0x86,0xFE - data1 0xA9,0xC0, 0xA8,0x02, 0xAA,0x44, 0xAB,0x86, 0xAE,0xC8, 0xAF,0x0A, 0xAD,0x4C, 0xAC,0x8E - data1 0xA7,0xD0, 0xA6,0x12, 0xA4,0x54, 0xA5,0x96, 0xA0,0xD8, 0xA1,0x1A, 0xA3,0x5C, 0xA2,0x9E - data1 0xB5,0xE0, 0xB4,0x22, 0xB6,0x64, 0xB7,0xA6, 0xB2,0xE8, 0xB3,0x2A, 0xB1,0x6C, 0xB0,0xAE - data1 0xBB,0xF0, 0xBA,0x32, 0xB8,0x74, 0xB9,0xB6, 0xBC,0xF8, 0xBD,0x3A, 0xBF,0x7C, 0xBE,0xBE -.size rem_8bit#,512 -stringz "GHASH for IA64, CRYPTOGAMS by " -___ - -$code =~ s/mux1(\s+)\S+\@rev/nop.i$1 0x0/gm if ($big_endian); -$code =~ s/\`([^\`]*)\`/eval $1/gem; - -print $code; -close STDOUT; diff --git a/src/lib/libcrypto/rc4/asm/rc4-ia64.pl b/src/lib/libcrypto/rc4/asm/rc4-ia64.pl deleted file mode 100644 index 49cd5b5e69..0000000000 --- a/src/lib/libcrypto/rc4/asm/rc4-ia64.pl +++ /dev/null @@ -1,755 +0,0 @@ -#!/usr/bin/env perl -# -# ==================================================================== -# Written by David Mosberger based on the -# Itanium optimized Crypto code which was released by HP Labs at -# http://www.hpl.hp.com/research/linux/crypto/. -# -# Copyright (c) 2005 Hewlett-Packard Development Company, L.P. -# -# Permission is hereby granted, free of charge, to any person obtaining -# a copy of this software and associated documentation files (the -# "Software"), to deal in the Software without restriction, including -# without limitation the rights to use, copy, modify, merge, publish, -# distribute, sublicense, and/or sell copies of the Software, and to -# permit persons to whom the Software is furnished to do so, subject to -# the following conditions: -# -# The above copyright notice and this permission notice shall be -# included in all copies or substantial portions of the Software. - -# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, -# EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF -# MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND -# NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE -# LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION -# OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION -# WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. */ - - - -# This is a little helper program which generates a software-pipelined -# for RC4 encryption. The basic algorithm looks like this: -# -# for (counter = 0; counter < len; ++counter) -# { -# in = inp[counter]; -# SI = S[I]; -# J = (SI + J) & 0xff; -# SJ = S[J]; -# T = (SI + SJ) & 0xff; -# S[I] = SJ, S[J] = SI; -# ST = S[T]; -# outp[counter] = in ^ ST; -# I = (I + 1) & 0xff; -# } -# -# Pipelining this loop isn't easy, because the stores to the S[] array -# need to be observed in the right order. The loop generated by the -# code below has the following pipeline diagram: -# -# cycle -# | 0 | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 |10 |11 |12 |13 |14 |15 |16 |17 | -# iter -# 1: xxx LDI xxx xxx xxx LDJ xxx SWP xxx LDT xxx xxx -# 2: xxx LDI xxx xxx xxx LDJ xxx SWP xxx LDT xxx xxx -# 3: xxx LDI xxx xxx xxx LDJ xxx SWP xxx LDT xxx xxx -# -# where: -# LDI = load of S[I] -# LDJ = load of S[J] -# SWP = swap of S[I] and S[J] -# LDT = load of S[T] -# -# Note that in the above diagram, the major trouble-spot is that LDI -# of the 2nd iteration is performed BEFORE the SWP of the first -# iteration. Fortunately, this is easy to detect (I of the 1st -# iteration will be equal to J of the 2nd iteration) and when this -# happens, we simply forward the proper value from the 1st iteration -# to the 2nd one. The proper value in this case is simply the value -# of S[I] from the first iteration (thanks to the fact that SWP -# simply swaps the contents of S[I] and S[J]). -# -# Another potential trouble-spot is in cycle 7, where SWP of the 1st -# iteration issues at the same time as the LDI of the 3rd iteration. -# However, thanks to IA-64 execution semantics, this can be taken -# care of simply by placing LDI later in the instruction-group than -# SWP. IA-64 CPUs will automatically forward the value if they -# detect that the SWP and LDI are accessing the same memory-location. - -# The core-loop that can be pipelined then looks like this (annotated -# with McKinley/Madison issue port & latency numbers, assuming L1 -# cache hits for the most part): - -# operation: instruction: issue-ports: latency -# ------------------ ----------------------------- ------------- ------- - -# Data = *inp++ ld1 data = [inp], 1 M0-M1 1 cyc c0 -# shladd Iptr = I, KeyTable, 3 M0-M3, I0, I1 1 cyc -# I = (I + 1) & 0xff padd1 nextI = I, one M0-M3, I0, I1 3 cyc -# ;; -# SI = S[I] ld8 SI = [Iptr] M0-M1 1 cyc c1 * after SWAP! -# ;; -# cmp.eq.unc pBypass = I, J * after J is valid! -# J = SI + J add J = J, SI M0-M3, I0, I1 1 cyc c2 -# (pBypass) br.cond.spnt Bypass -# ;; -# --------------------------------------------------------------------------------------- -# J = J & 0xff zxt1 J = J I0, I1, 1 cyc c3 -# ;; -# shladd Jptr = J, KeyTable, 3 M0-M3, I0, I1 1 cyc c4 -# ;; -# SJ = S[J] ld8 SJ = [Jptr] M0-M1 1 cyc c5 -# ;; -# --------------------------------------------------------------------------------------- -# T = (SI + SJ) add T = SI, SJ M0-M3, I0, I1 1 cyc c6 -# ;; -# T = T & 0xff zxt1 T = T I0, I1 1 cyc -# S[I] = SJ st8 [Iptr] = SJ M2-M3 c7 -# S[J] = SI st8 [Jptr] = SI M2-M3 -# ;; -# shladd Tptr = T, KeyTable, 3 M0-M3, I0, I1 1 cyc c8 -# ;; -# --------------------------------------------------------------------------------------- -# T = S[T] ld8 T = [Tptr] M0-M1 1 cyc c9 -# ;; -# data ^= T xor data = data, T M0-M3, I0, I1 1 cyc c10 -# ;; -# *out++ = Data ^ T dep word = word, data, 8, POS I0, I1 1 cyc c11 -# ;; -# --------------------------------------------------------------------------------------- - -# There are several points worth making here: - -# - Note that due to the bypass/forwarding-path, the first two -# phases of the loop are strangly mingled together. In -# particular, note that the first stage of the pipeline is -# using the value of "J", as calculated by the second stage. -# - Each bundle-pair will have exactly 6 instructions. -# - Pipelined, the loop can execute in 3 cycles/iteration and -# 4 stages. However, McKinley/Madison can issue "st1" to -# the same bank at a rate of at most one per 4 cycles. Thus, -# instead of storing each byte, we accumulate them in a word -# and then write them back at once with a single "st8" (this -# implies that the setup code needs to ensure that the output -# buffer is properly aligned, if need be, by encoding the -# first few bytes separately). -# - There is no space for a "br.ctop" instruction. For this -# reason we can't use module-loop support in IA-64 and have -# to do a traditional, purely software-pipelined loop. -# - We can't replace any of the remaining "add/zxt1" pairs with -# "padd1" because the latency for that instruction is too high -# and would push the loop to the point where more bypasses -# would be needed, which we don't have space for. -# - The above loop runs at around 3.26 cycles/byte, or roughly -# 440 MByte/sec on a 1.5GHz Madison. This is well below the -# system bus bandwidth and hence with judicious use of -# "lfetch" this loop can run at (almost) peak speed even when -# the input and output data reside in memory. The -# max. latency that can be tolerated is (PREFETCH_DISTANCE * -# L2_LINE_SIZE * 3 cyc), or about 384 cycles assuming (at -# least) 1-ahead prefetching of 128 byte cache-lines. Note -# that we do NOT prefetch into L1, since that would only -# interfere with the S[] table values stored there. This is -# acceptable because there is a 10 cycle latency between -# load and first use of the input data. -# - We use a branch to out-of-line bypass-code of cycle-pressure: -# we calculate the next J, check for the need to activate the -# bypass path, and activate the bypass path ALL IN THE SAME -# CYCLE. If we didn't have these constraints, we could do -# the bypass with a simple conditional move instruction. -# Fortunately, the bypass paths get activated relatively -# infrequently, so the extra branches don't cost all that much -# (about 0.04 cycles/byte, measured on a 16396 byte file with -# random input data). -# - -$phases = 4; # number of stages/phases in the pipelined-loop -$unroll_count = 6; # number of times we unrolled it -$pComI = (1 << 0); -$pComJ = (1 << 1); -$pComT = (1 << 2); -$pOut = (1 << 3); - -$NData = 4; -$NIP = 3; -$NJP = 2; -$NI = 2; -$NSI = 3; -$NSJ = 2; -$NT = 2; -$NOutWord = 2; - -# -# $threshold is the minimum length before we attempt to use the -# big software-pipelined loop. It MUST be greater-or-equal -# to: -# PHASES * (UNROLL_COUNT + 1) + 7 -# -# The "+ 7" comes from the fact we may have to encode up to -# 7 bytes separately before the output pointer is aligned. -# -$threshold = (3 * ($phases * ($unroll_count + 1)) + 7); - -sub I { - local *code = shift; - local $format = shift; - $code .= sprintf ("\t\t".$format."\n", @_); -} - -sub P { - local *code = shift; - local $format = shift; - $code .= sprintf ($format."\n", @_); -} - -sub STOP { - local *code = shift; - $code .=<<___; - ;; -___ -} - -sub emit_body { - local *c = shift; - local *bypass = shift; - local ($iteration, $p) = @_; - - local $i0 = $iteration; - local $i1 = $iteration - 1; - local $i2 = $iteration - 2; - local $i3 = $iteration - 3; - local $iw0 = ($iteration - 3) / 8; - local $iw1 = ($iteration > 3) ? ($iteration - 4) / 8 : 1; - local $byte_num = ($iteration - 3) % 8; - local $label = $iteration + 1; - local $pAny = ($p & 0xf) == 0xf; - local $pByp = (($p & $pComI) && ($iteration > 0)); - - $c.=<<___; -////////////////////////////////////////////////// -___ - - if (($p & 0xf) == 0) { - $c.="#ifdef HOST_IS_BIG_ENDIAN\n"; - &I(\$c,"shr.u OutWord[%u] = OutWord[%u], 32;;", - $iw1 % $NOutWord, $iw1 % $NOutWord); - $c.="#endif\n"; - &I(\$c, "st4 [OutPtr] = OutWord[%u], 4", $iw1 % $NOutWord); - return; - } - - # Cycle 0 - &I(\$c, "{ .mmi") if ($pAny); - &I(\$c, "ld1 Data[%u] = [InPtr], 1", $i0 % $NData) if ($p & $pComI); - &I(\$c, "padd1 I[%u] = One, I[%u]", $i0 % $NI, $i1 % $NI)if ($p & $pComI); - &I(\$c, "zxt1 J = J") if ($p & $pComJ); - &I(\$c, "}") if ($pAny); - &I(\$c, "{ .mmi") if ($pAny); - &I(\$c, "LKEY T[%u] = [T[%u]]", $i1 % $NT, $i1 % $NT) if ($p & $pOut); - &I(\$c, "add T[%u] = SI[%u], SJ[%u]", - $i0 % $NT, $i2 % $NSI, $i1 % $NSJ) if ($p & $pComT); - &I(\$c, "KEYADDR(IPr[%u], I[%u])", $i0 % $NIP, $i1 % $NI) if ($p & $pComI); - &I(\$c, "}") if ($pAny); - &STOP(\$c); - - # Cycle 1 - &I(\$c, "{ .mmi") if ($pAny); - &I(\$c, "SKEY [IPr[%u]] = SJ[%u]", $i2 % $NIP, $i1%$NSJ)if ($p & $pComT); - &I(\$c, "SKEY [JP[%u]] = SI[%u]", $i1 % $NJP, $i2%$NSI) if ($p & $pComT); - &I(\$c, "zxt1 T[%u] = T[%u]", $i0 % $NT, $i0 % $NT) if ($p & $pComT); - &I(\$c, "}") if ($pAny); - &I(\$c, "{ .mmi") if ($pAny); - &I(\$c, "LKEY SI[%u] = [IPr[%u]]", $i0 % $NSI, $i0%$NIP)if ($p & $pComI); - &I(\$c, "KEYADDR(JP[%u], J)", $i0 % $NJP) if ($p & $pComJ); - &I(\$c, "xor Data[%u] = Data[%u], T[%u]", - $i3 % $NData, $i3 % $NData, $i1 % $NT) if ($p & $pOut); - &I(\$c, "}") if ($pAny); - &STOP(\$c); - - # Cycle 2 - &I(\$c, "{ .mmi") if ($pAny); - &I(\$c, "LKEY SJ[%u] = [JP[%u]]", $i0 % $NSJ, $i0%$NJP) if ($p & $pComJ); - &I(\$c, "cmp.eq pBypass, p0 = I[%u], J", $i1 % $NI) if ($pByp); - &I(\$c, "dep OutWord[%u] = Data[%u], OutWord[%u], BYTE_POS(%u), 8", - $iw0%$NOutWord, $i3%$NData, $iw1%$NOutWord, $byte_num) if ($p & $pOut); - &I(\$c, "}") if ($pAny); - &I(\$c, "{ .mmb") if ($pAny); - &I(\$c, "add J = J, SI[%u]", $i0 % $NSI) if ($p & $pComI); - &I(\$c, "KEYADDR(T[%u], T[%u])", $i0 % $NT, $i0 % $NT) if ($p & $pComT); - &P(\$c, "(pBypass)\tbr.cond.spnt.many .rc4Bypass%u",$label)if ($pByp); - &I(\$c, "}") if ($pAny); - &STOP(\$c); - - &P(\$c, ".rc4Resume%u:", $label) if ($pByp); - if ($byte_num == 0 && $iteration >= $phases) { - &I(\$c, "st8 [OutPtr] = OutWord[%u], 8", - $iw1 % $NOutWord) if ($p & $pOut); - if ($iteration == (1 + $unroll_count) * $phases - 1) { - if ($unroll_count == 6) { - &I(\$c, "mov OutWord[%u] = OutWord[%u]", - $iw1 % $NOutWord, $iw0 % $NOutWord); - } - &I(\$c, "lfetch.nt1 [InPrefetch], %u", - $unroll_count * $phases); - &I(\$c, "lfetch.excl.nt1 [OutPrefetch], %u", - $unroll_count * $phases); - &I(\$c, "br.cloop.sptk.few .rc4Loop"); - } - } - - if ($pByp) { - &P(\$bypass, ".rc4Bypass%u:", $label); - &I(\$bypass, "sub J = J, SI[%u]", $i0 % $NSI); - &I(\$bypass, "nop 0"); - &I(\$bypass, "nop 0"); - &I(\$bypass, ";;"); - &I(\$bypass, "add J = J, SI[%u]", $i1 % $NSI); - &I(\$bypass, "mov SI[%u] = SI[%u]", $i0 % $NSI, $i1 % $NSI); - &I(\$bypass, "br.sptk.many .rc4Resume%u\n", $label); - &I(\$bypass, ";;"); - } -} - -$code=<<___; -.ident \"rc4-ia64.s, version 3.0\" -.ident \"Copyright (c) 2005 Hewlett-Packard Development Company, L.P.\" - -#define LCSave r8 -#define PRSave r9 - -/* Inputs become invalid once rotation begins! */ - -#define StateTable in0 -#define DataLen in1 -#define InputBuffer in2 -#define OutputBuffer in3 - -#define KTable r14 -#define J r15 -#define InPtr r16 -#define OutPtr r17 -#define InPrefetch r18 -#define OutPrefetch r19 -#define One r20 -#define LoopCount r21 -#define Remainder r22 -#define IFinal r23 -#define EndPtr r24 - -#define tmp0 r25 -#define tmp1 r26 - -#define pBypass p6 -#define pDone p7 -#define pSmall p8 -#define pAligned p9 -#define pUnaligned p10 - -#define pComputeI pPhase[0] -#define pComputeJ pPhase[1] -#define pComputeT pPhase[2] -#define pOutput pPhase[3] - -#define RetVal r8 -#define L_OK p7 -#define L_NOK p8 - -#define _NINPUTS 4 -#define _NOUTPUT 0 - -#define _NROTATE 24 -#define _NLOCALS (_NROTATE - _NINPUTS - _NOUTPUT) - -#ifndef SZ -# define SZ 4 // this must be set to sizeof(RC4_INT) -#endif - -#if SZ == 1 -# define LKEY ld1 -# define SKEY st1 -# define KEYADDR(dst, i) add dst = i, KTable -#elif SZ == 2 -# define LKEY ld2 -# define SKEY st2 -# define KEYADDR(dst, i) shladd dst = i, 1, KTable -#elif SZ == 4 -# define LKEY ld4 -# define SKEY st4 -# define KEYADDR(dst, i) shladd dst = i, 2, KTable -#else -# define LKEY ld8 -# define SKEY st8 -# define KEYADDR(dst, i) shladd dst = i, 3, KTable -#endif - -#if defined(_HPUX_SOURCE) && !defined(_LP64) -# define ADDP addp4 -#else -# define ADDP add -#endif - -/* Define a macro for the bit number of the n-th byte: */ - -#if defined(_HPUX_SOURCE) || defined(B_ENDIAN) -# define HOST_IS_BIG_ENDIAN -# define BYTE_POS(n) (56 - (8 * (n))) -#else -# define BYTE_POS(n) (8 * (n)) -#endif - -/* - We must perform the first phase of the pipeline explicitly since - we will always load from the stable the first time. The br.cexit - will never be taken since regardless of the number of bytes because - the epilogue count is 4. -*/ -/* MODSCHED_RC4 macro was split to _PROLOGUE and _LOOP, because HP-UX - assembler failed on original macro with syntax error. */ -#define MODSCHED_RC4_PROLOGUE \\ - { \\ - ld1 Data[0] = [InPtr], 1; \\ - add IFinal = 1, I[1]; \\ - KEYADDR(IPr[0], I[1]); \\ - } ;; \\ - { \\ - LKEY SI[0] = [IPr[0]]; \\ - mov pr.rot = 0x10000; \\ - mov ar.ec = 4; \\ - } ;; \\ - { \\ - add J = J, SI[0]; \\ - zxt1 I[0] = IFinal; \\ - br.cexit.spnt.few .+16; /* never taken */ \\ - } ;; -#define MODSCHED_RC4_LOOP(label) \\ -label: \\ - { .mmi; \\ - (pComputeI) ld1 Data[0] = [InPtr], 1; \\ - (pComputeI) add IFinal = 1, I[1]; \\ - (pComputeJ) zxt1 J = J; \\ - }{ .mmi; \\ - (pOutput) LKEY T[1] = [T[1]]; \\ - (pComputeT) add T[0] = SI[2], SJ[1]; \\ - (pComputeI) KEYADDR(IPr[0], I[1]); \\ - } ;; \\ - { .mmi; \\ - (pComputeT) SKEY [IPr[2]] = SJ[1]; \\ - (pComputeT) SKEY [JP[1]] = SI[2]; \\ - (pComputeT) zxt1 T[0] = T[0]; \\ - }{ .mmi; \\ - (pComputeI) LKEY SI[0] = [IPr[0]]; \\ - (pComputeJ) KEYADDR(JP[0], J); \\ - (pComputeI) cmp.eq.unc pBypass, p0 = I[1], J; \\ - } ;; \\ - { .mmi; \\ - (pComputeJ) LKEY SJ[0] = [JP[0]]; \\ - (pOutput) xor Data[3] = Data[3], T[1]; \\ - nop 0x0; \\ - }{ .mmi; \\ - (pComputeT) KEYADDR(T[0], T[0]); \\ - (pBypass) mov SI[0] = SI[1]; \\ - (pComputeI) zxt1 I[0] = IFinal; \\ - } ;; \\ - { .mmb; \\ - (pOutput) st1 [OutPtr] = Data[3], 1; \\ - (pComputeI) add J = J, SI[0]; \\ - br.ctop.sptk.few label; \\ - } ;; - - .text - - .align 32 - - .type RC4, \@function - .global RC4 - - .proc RC4 - .prologue - -RC4: - { - .mmi - alloc r2 = ar.pfs, _NINPUTS, _NLOCALS, _NOUTPUT, _NROTATE - - .rotr Data[4], I[2], IPr[3], SI[3], JP[2], SJ[2], T[2], \\ - OutWord[2] - .rotp pPhase[4] - - ADDP InPrefetch = 0, InputBuffer - ADDP KTable = 0, StateTable - } - { - .mmi - ADDP InPtr = 0, InputBuffer - ADDP OutPtr = 0, OutputBuffer - mov RetVal = r0 - } - ;; - { - .mmi - lfetch.nt1 [InPrefetch], 0x80 - ADDP OutPrefetch = 0, OutputBuffer - } - { // Return 0 if the input length is nonsensical - .mib - ADDP StateTable = 0, StateTable - cmp.ge.unc L_NOK, L_OK = r0, DataLen - (L_NOK) br.ret.sptk.few rp - } - ;; - { - .mib - cmp.eq.or L_NOK, L_OK = r0, InPtr - cmp.eq.or L_NOK, L_OK = r0, OutPtr - nop 0x0 - } - { - .mib - cmp.eq.or L_NOK, L_OK = r0, StateTable - nop 0x0 - (L_NOK) br.ret.sptk.few rp - } - ;; - LKEY I[1] = [KTable], SZ -/* Prefetch the state-table. It contains 256 elements of size SZ */ - -#if SZ == 1 - ADDP tmp0 = 1*128, StateTable -#elif SZ == 2 - ADDP tmp0 = 3*128, StateTable - ADDP tmp1 = 2*128, StateTable -#elif SZ == 4 - ADDP tmp0 = 7*128, StateTable - ADDP tmp1 = 6*128, StateTable -#elif SZ == 8 - ADDP tmp0 = 15*128, StateTable - ADDP tmp1 = 14*128, StateTable -#endif - ;; -#if SZ >= 8 - lfetch.fault.nt1 [tmp0], -256 // 15 - lfetch.fault.nt1 [tmp1], -256;; - lfetch.fault.nt1 [tmp0], -256 // 13 - lfetch.fault.nt1 [tmp1], -256;; - lfetch.fault.nt1 [tmp0], -256 // 11 - lfetch.fault.nt1 [tmp1], -256;; - lfetch.fault.nt1 [tmp0], -256 // 9 - lfetch.fault.nt1 [tmp1], -256;; -#endif -#if SZ >= 4 - lfetch.fault.nt1 [tmp0], -256 // 7 - lfetch.fault.nt1 [tmp1], -256;; - lfetch.fault.nt1 [tmp0], -256 // 5 - lfetch.fault.nt1 [tmp1], -256;; -#endif -#if SZ >= 2 - lfetch.fault.nt1 [tmp0], -256 // 3 - lfetch.fault.nt1 [tmp1], -256;; -#endif - { - .mii - lfetch.fault.nt1 [tmp0] // 1 - add I[1]=1,I[1];; - zxt1 I[1]=I[1] - } - { - .mmi - lfetch.nt1 [InPrefetch], 0x80 - lfetch.excl.nt1 [OutPrefetch], 0x80 - .save pr, PRSave - mov PRSave = pr - } ;; - { - .mmi - lfetch.excl.nt1 [OutPrefetch], 0x80 - LKEY J = [KTable], SZ - ADDP EndPtr = DataLen, InPtr - } ;; - { - .mmi - ADDP EndPtr = -1, EndPtr // Make it point to - // last data byte. - mov One = 1 - .save ar.lc, LCSave - mov LCSave = ar.lc - .body - } ;; - { - .mmb - sub Remainder = 0, OutPtr - cmp.gtu pSmall, p0 = $threshold, DataLen -(pSmall) br.cond.dpnt .rc4Remainder // Data too small for - // big loop. - } ;; - { - .mmi - and Remainder = 0x7, Remainder - ;; - cmp.eq pAligned, pUnaligned = Remainder, r0 - nop 0x0 - } ;; - { - .mmb -.pred.rel "mutex",pUnaligned,pAligned -(pUnaligned) add Remainder = -1, Remainder -(pAligned) sub Remainder = EndPtr, InPtr -(pAligned) br.cond.dptk.many .rc4Aligned - } ;; - { - .mmi - nop 0x0 - nop 0x0 - mov.i ar.lc = Remainder - } - -/* Do the initial few bytes via the compact, modulo-scheduled loop - until the output pointer is 8-byte-aligned. */ - - MODSCHED_RC4_PROLOGUE - MODSCHED_RC4_LOOP(.RC4AlignLoop) - - { - .mib - sub Remainder = EndPtr, InPtr - zxt1 IFinal = IFinal - clrrrb // Clear CFM.rrb.pr so - ;; // next "mov pr.rot = N" - // does the right thing. - } - { - .mmi - mov I[1] = IFinal - nop 0x0 - nop 0x0 - } ;; - - -.rc4Aligned: - -/* - Unrolled loop count = (Remainder - ($unroll_count+1)*$phases)/($unroll_count*$phases) - */ - - { - .mlx - add LoopCount = 1 - ($unroll_count + 1)*$phases, Remainder - movl Remainder = 0xaaaaaaaaaaaaaaab - } ;; - { - .mmi - setf.sig f6 = LoopCount // M2, M3 6 cyc - setf.sig f7 = Remainder // M2, M3 6 cyc - nop 0x0 - } ;; - { - .mfb - nop 0x0 - xmpy.hu f6 = f6, f7 - nop 0x0 - } ;; - { - .mmi - getf.sig LoopCount = f6;; // M2 5 cyc - nop 0x0 - shr.u LoopCount = LoopCount, 4 - } ;; - { - .mmi - nop 0x0 - nop 0x0 - mov.i ar.lc = LoopCount - } ;; - -/* Now comes the unrolled loop: */ - -.rc4Prologue: -___ - -$iteration = 0; - -# Generate the prologue: -$predicates = 1; -for ($i = 0; $i < $phases; ++$i) { - &emit_body (\$code, \$bypass, $iteration++, $predicates); - $predicates = ($predicates << 1) | 1; -} - -$code.=<<___; -.rc4Loop: -___ - -# Generate the body: -for ($i = 0; $i < $unroll_count*$phases; ++$i) { - &emit_body (\$code, \$bypass, $iteration++, $predicates); -} - -$code.=<<___; -.rc4Epilogue: -___ - -# Generate the epilogue: -for ($i = 0; $i < $phases; ++$i) { - $predicates <<= 1; - &emit_body (\$code, \$bypass, $iteration++, $predicates); -} - -$code.=<<___; - { - .mmi - lfetch.nt1 [EndPtr] // fetch line with last byte - mov IFinal = I[1] - nop 0x0 - } - -.rc4Remainder: - { - .mmi - sub Remainder = EndPtr, InPtr // Calculate - // # of bytes - // left - 1 - nop 0x0 - nop 0x0 - } ;; - { - .mib - cmp.eq pDone, p0 = -1, Remainder // done already? - mov.i ar.lc = Remainder -(pDone) br.cond.dptk.few .rc4Complete - } - -/* Do the remaining bytes via the compact, modulo-scheduled loop */ - - MODSCHED_RC4_PROLOGUE - MODSCHED_RC4_LOOP(.RC4RestLoop) - -.rc4Complete: - { - .mmi - add KTable = -SZ, KTable - add IFinal = -1, IFinal - mov ar.lc = LCSave - } ;; - { - .mii - SKEY [KTable] = J,-SZ - zxt1 IFinal = IFinal - mov pr = PRSave, 0x1FFFF - } ;; - { - .mib - SKEY [KTable] = IFinal - add RetVal = 1, r0 - br.ret.sptk.few rp - } ;; -___ - -# Last but not least, emit the code for the bypass-code of the unrolled loop: - -$code.=$bypass; - -$code.=<<___; - .endp RC4 -___ - -print $code; diff --git a/src/lib/libcrypto/sha/asm/sha1-ia64.pl b/src/lib/libcrypto/sha/asm/sha1-ia64.pl deleted file mode 100644 index 02d35d1614..0000000000 --- a/src/lib/libcrypto/sha/asm/sha1-ia64.pl +++ /dev/null @@ -1,305 +0,0 @@ -#!/usr/bin/env perl -# -# ==================================================================== -# Written by Andy Polyakov for the OpenSSL -# project. The module is, however, dual licensed under OpenSSL and -# CRYPTOGAMS licenses depending on where you obtain it. For further -# details see http://www.openssl.org/~appro/cryptogams/. -# ==================================================================== -# -# Eternal question is what's wrong with compiler generated code? The -# trick is that it's possible to reduce the number of shifts required -# to perform rotations by maintaining copy of 32-bit value in upper -# bits of 64-bit register. Just follow mux2 and shrp instructions... -# Performance under big-endian OS such as HP-UX is 179MBps*1GHz, which -# is >50% better than HP C and >2x better than gcc. - -$code=<<___; -.ident \"sha1-ia64.s, version 1.3\" -.ident \"IA-64 ISA artwork by Andy Polyakov \" -.explicit - -___ - - -if ($^O eq "hpux") { - $ADDP="addp4"; - for (@ARGV) { $ADDP="add" if (/[\+DD|\-mlp]64/); } -} else { $ADDP="add"; } - -#$human=1; -if ($human) { # useful for visual code auditing... - ($A,$B,$C,$D,$E) = ("A","B","C","D","E"); - ($h0,$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) = ("loc0","loc1","loc2","loc3","loc4"); - ($h0,$h1,$h2,$h3,$h4) = ("loc5","loc6","loc7","loc8","loc9"); - ($K_00_19, $K_20_39, $K_40_59, $K_60_79) = - ( "r14", "r15", "loc10", "loc11" ); - @X= ( "r16", "r17", "r18", "r19", "r20", "r21", "r22", "r23", - "r24", "r25", "r26", "r27", "r28", "r29", "r30", "r31" ); -} - -sub BODY_00_15 { -local *code=shift; -my ($i,$a,$b,$c,$d,$e)=@_; -my $j=$i+1; -my $Xn=@X[$j%16]; - -$code.=<<___ if ($i==0); -{ .mmi; ld1 $X[$i]=[inp],2 // MSB - ld1 tmp2=[tmp3],2 };; -{ .mmi; ld1 tmp0=[inp],2 - ld1 tmp4=[tmp3],2 // LSB - dep $X[$i]=$X[$i],tmp2,8,8 };; -___ -if ($i<15) { - $code.=<<___; -{ .mmi; ld1 $Xn=[inp],2 // forward Xload - nop.m 0x0 - dep tmp1=tmp0,tmp4,8,8 };; -{ .mmi; ld1 tmp2=[tmp3],2 // forward Xload - and tmp4=$c,$b - dep $X[$i]=$X[$i],tmp1,16,16} //;; -{ .mmi; add $e=$e,$K_00_19 // e+=K_00_19 - andcm tmp1=$d,$b - dep.z tmp5=$a,5,27 };; // a<<5 -{ .mmi; add $e=$e,$X[$i] // e+=Xload - or tmp4=tmp4,tmp1 // F_00_19(b,c,d)=(b&c)|(~b&d) - extr.u tmp1=$a,27,5 };; // a>>27 -{ .mmi; ld1 tmp0=[inp],2 // forward Xload - add $e=$e,tmp4 // e+=F_00_19(b,c,d) - shrp $b=tmp6,tmp6,2 } // b=ROTATE(b,30) -{ .mmi; ld1 tmp4=[tmp3],2 // forward Xload - or tmp5=tmp1,tmp5 // ROTATE(a,5) - mux2 tmp6=$a,0x44 };; // see b in next iteration -{ .mii; add $e=$e,tmp5 // e+=ROTATE(a,5) - dep $Xn=$Xn,tmp2,8,8 // forward Xload - mux2 $X[$i]=$X[$i],0x44 } //;; - -___ - } -else { - $code.=<<___; -{ .mii; and tmp3=$c,$b - dep tmp1=tmp0,tmp4,8,8;; - dep $X[$i]=$X[$i],tmp1,16,16} //;; -{ .mmi; add $e=$e,$K_00_19 // e+=K_00_19 - andcm tmp1=$d,$b - dep.z tmp5=$a,5,27 };; // a<<5 -{ .mmi; add $e=$e,$X[$i] // e+=Xupdate - or tmp4=tmp3,tmp1 // F_00_19(b,c,d)=(b&c)|(~b&d) - extr.u tmp1=$a,27,5 } // a>>27 -{ .mmi; xor $Xn=$Xn,$X[($j+2)%16] // forward Xupdate - xor tmp3=$X[($j+8)%16],$X[($j+13)%16] // forward Xupdate - nop.i 0 };; -{ .mmi; add $e=$e,tmp4 // e+=F_00_19(b,c,d) - xor $Xn=$Xn,tmp3 // forward Xupdate - 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 $e=$e,tmp1 // e+=ROTATE(a,5) - shrp $Xn=$Xn,$Xn,31 // ROTATE(x[0]^x[2]^x[8]^x[13],1) - mux2 $X[$i]=$X[$i],0x44 };; - -___ - } -} - -sub BODY_16_19 { -local *code=shift; -my ($i,$a,$b,$c,$d,$e)=@_; -my $j=$i+1; -my $Xn=@X[$j%16]; - -$code.=<<___; -{ .mib; add $e=$e,$K_00_19 // e+=K_00_19 - dep.z tmp5=$a,5,27 } // a<<5 -{ .mib; andcm tmp1=$d,$b - and tmp0=$c,$b };; -{ .mmi; add $e=$e,$X[$i%16] // e+=Xupdate - or tmp0=tmp0,tmp1 // F_00_19(b,c,d)=(b&c)|(~b&d) - extr.u tmp1=$a,27,5 } // a>>27 -{ .mmi; xor $Xn=$Xn,$X[($j+2)%16] // forward Xupdate - xor tmp3=$X[($j+8)%16],$X[($j+13)%16] // forward Xupdate - nop.i 0 };; -{ .mmi; add $e=$e,tmp0 // f+=F_00_19(b,c,d) - xor $Xn=$Xn,tmp3 // forward Xupdate - 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 $e=$e,tmp1 // e+=ROTATE(a,5) - shrp $Xn=$Xn,$Xn,31 // ROTATE(x[0]^x[2]^x[8]^x[13],1) - nop.i 0 };; - -___ -} - -sub BODY_20_39 { -local *code=shift; -my ($i,$a,$b,$c,$d,$e,$Konst)=@_; - $Konst = $K_20_39 if (!defined($Konst)); -my $j=$i+1; -my $Xn=@X[$j%16]; - -if ($i<79) { -$code.=<<___; -{ .mib; add $e=$e,$Konst // e+=K_XX_XX - dep.z tmp5=$a,5,27 } // a<<5 -{ .mib; xor tmp0=$c,$b - xor $Xn=$Xn,$X[($j+2)%16] };; // forward Xupdate -{ .mib; add $e=$e,$X[$i%16] // e+=Xupdate - extr.u tmp1=$a,27,5 } // a>>27 -{ .mib; xor tmp0=tmp0,$d // F_20_39(b,c,d)=b^c^d - xor $Xn=$Xn,$X[($j+8)%16] };; // forward Xupdate -{ .mmi; add $e=$e,tmp0 // e+=F_20_39(b,c,d) - xor $Xn=$Xn,$X[($j+13)%16] // forward Xupdate - 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 $e=$e,tmp1 // e+=ROTATE(a,5) - shrp $Xn=$Xn,$Xn,31 // ROTATE(x[0]^x[2]^x[8]^x[13],1) - nop.i 0 };; - -___ -} -else { -$code.=<<___; -{ .mib; add $e=$e,$Konst // e+=K_60_79 - dep.z tmp5=$a,5,27 } // a<<5 -{ .mib; xor tmp0=$c,$b - add $h1=$h1,$a };; // wrap up -{ .mib; add $e=$e,$X[$i%16] // e+=Xupdate - extr.u tmp1=$a,27,5 } // a>>27 -{ .mib; xor tmp0=tmp0,$d // F_20_39(b,c,d)=b^c^d - add $h3=$h3,$c };; // wrap up -{ .mmi; add $e=$e,tmp0 // e+=F_20_39(b,c,d) - or tmp1=tmp1,tmp5 // ROTATE(a,5) - shrp $b=tmp6,tmp6,2 };; // b=ROTATE(b,30) ;;? -{ .mmi; add $e=$e,tmp1 // e+=ROTATE(a,5) - add tmp3=1,inp // used in unaligned codepath - add $h4=$h4,$d };; // wrap up - -___ -} -} - -sub BODY_40_59 { -local *code=shift; -my ($i,$a,$b,$c,$d,$e)=@_; -my $j=$i+1; -my $Xn=@X[$j%16]; - -$code.=<<___; -{ .mib; add $e=$e,$K_40_59 // e+=K_40_59 - dep.z tmp5=$a,5,27 } // a<<5 -{ .mib; and tmp1=$c,$d - xor tmp0=$c,$d };; -{ .mmi; add $e=$e,$X[$i%16] // e+=Xupdate - add tmp5=tmp5,tmp1 // a<<5+(c&d) - extr.u tmp1=$a,27,5 } // a>>27 -{ .mmi; and tmp0=tmp0,$b - xor $Xn=$Xn,$X[($j+2)%16] // forward Xupdate - xor tmp3=$X[($j+8)%16],$X[($j+13)%16] };; // forward Xupdate -{ .mmi; add $e=$e,tmp0 // e+=b&(c^d) - add tmp5=tmp5,tmp1 // ROTATE(a,5)+(c&d) - shrp $b=tmp6,tmp6,2 } // b=ROTATE(b,30) -{ .mmi; xor $Xn=$Xn,tmp3 - mux2 tmp6=$a,0x44 };; // see b in next iteration -{ .mii; add $e=$e,tmp5 // e+=ROTATE(a,5)+(c&d) - shrp $Xn=$Xn,$Xn,31 // ROTATE(x[0]^x[2]^x[8]^x[13],1) - nop.i 0x0 };; - -___ -} -sub BODY_60_79 { &BODY_20_39(@_,$K_60_79); } - -$code.=<<___; -.text - -tmp0=r8; -tmp1=r9; -tmp2=r10; -tmp3=r11; -ctx=r32; // in0 -inp=r33; // in1 - -// void sha1_block_data_order(SHA_CTX *c,const void *p,size_t num); -.global sha1_block_data_order# -.proc sha1_block_data_order# -.align 32 -sha1_block_data_order: - .prologue -{ .mmi; alloc tmp1=ar.pfs,3,14,0,0 - $ADDP tmp0=4,ctx - .save ar.lc,r3 - mov r3=ar.lc } -{ .mmi; $ADDP ctx=0,ctx - $ADDP inp=0,inp - mov r2=pr };; -tmp4=in2; -tmp5=loc12; -tmp6=loc13; - .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 };; - -___ - -{ my $i; - my @V=($A,$B,$C,$D,$E); - - for($i=0;$i<16;$i++) { &BODY_00_15(\$code,$i,@V); unshift(@V,pop(@V)); } - for(;$i<20;$i++) { &BODY_16_19(\$code,$i,@V); unshift(@V,pop(@V)); } - for(;$i<40;$i++) { &BODY_20_39(\$code,$i,@V); unshift(@V,pop(@V)); } - for(;$i<60;$i++) { &BODY_40_59(\$code,$i,@V); unshift(@V,pop(@V)); } - for(;$i<80;$i++) { &BODY_60_79(\$code,$i,@V); unshift(@V,pop(@V)); } - - (($V[0] eq $A) and ($V[4] eq $E)) or die; # double-check -} - -$code.=<<___; -{ .mmb; add $h0=$h0,$A - add $h2=$h2,$C - br.ctop.dptk.many .Ldtop };; -.Ldend: -{ .mmi; add tmp0=4,ctx - mov ar.lc=r3 };; -{ .mmi; st4 [ctx]=$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_data_order# -stringz "SHA1 block transform for IA64, CRYPTOGAMS by " -___ - -$output=shift and open STDOUT,">$output"; -print $code; diff --git a/src/lib/libcrypto/sha/asm/sha512-ia64.pl b/src/lib/libcrypto/sha/asm/sha512-ia64.pl deleted file mode 100755 index 1c6ce56522..0000000000 --- a/src/lib/libcrypto/sha/asm/sha512-ia64.pl +++ /dev/null @@ -1,672 +0,0 @@ -#!/usr/bin/env perl -# -# ==================================================================== -# Written by Andy Polyakov 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. -# -# sha512_block runs in 1003 cycles on Itanium 2, which is almost 50% -# faster than gcc and >60%(!) faster than code generated by HP-UX -# compiler (yes, HP-UX is generating slower code, because unlike gcc, -# it failed to deploy "shift right pair," 'shrp' instruction, which -# substitutes for 64-bit rotate). -# -# 924 cycles long sha256_block outperforms gcc by over factor of 2(!) -# and HP-UX compiler - by >40% (yes, gcc won sha512_block, but lost -# this one big time). Note that "formally" 924 is about 100 cycles -# too much. I mean it's 64 32-bit rounds vs. 80 virtually identical -# 64-bit ones and 1003*64/80 gives 802. Extra cycles, 2 per round, -# are spent on extra work to provide for 32-bit rotations. 32-bit -# rotations are still handled by 'shrp' instruction and for this -# reason lower 32 bits are deposited to upper half of 64-bit register -# prior 'shrp' issue. And in order to minimize the amount of such -# operations, X[16] values are *maintained* with copies of lower -# halves in upper halves, which is why you'll spot such instructions -# as custom 'mux2', "parallel 32-bit add," 'padd4' and "parallel -# 32-bit unsigned right shift," 'pshr4.u' instructions here. -# -# Rules of engagement. -# -# There is only one integer shifter meaning that if I have two rotate, -# deposit or extract instructions in adjacent bundles, they shall -# split [at run-time if they have to]. But note that variable and -# parallel shifts are performed by multi-media ALU and *are* pairable -# with rotates [and alike]. On the backside MMALU is rather slow: it -# takes 2 extra cycles before the result of integer operation is -# available *to* MMALU and 2(*) extra cycles before the result of MM -# operation is available "back" *to* integer ALU, not to mention that -# MMALU itself has 2 cycles latency. However! I explicitly scheduled -# these MM instructions to avoid MM stalls, so that all these extra -# latencies get "hidden" in instruction-level parallelism. -# -# (*) 2 cycles on Itanium 1 and 1 cycle on Itanium 2. But I schedule -# for 2 in order to provide for best *overall* performance, -# 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: -# -# 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 -# -# (*) SHA1 numbers are for HP-UX compiler and are presented purely -# for reference purposes. I bet it can improved too... -# -# To generate code, pass the file name with either 256 or 512 in its -# name and compiler flags. - -$output=shift; - -if ($output =~ /512.*\.[s|asm]/) { - $SZ=8; - $BITS=8*$SZ; - $LDW="ld8"; - $STW="st8"; - $ADD="add"; - $SHRU="shr.u"; - $TABLE="K512"; - $func="sha512_block_data_order"; - @Sigma0=(28,34,39); - @Sigma1=(14,18,41); - @sigma0=(1, 8, 7); - @sigma1=(19,61, 6); - $rounds=80; -} elsif ($output =~ /256.*\.[s|asm]/) { - $SZ=4; - $BITS=8*$SZ; - $LDW="ld4"; - $STW="st4"; - $ADD="padd4"; - $SHRU="pshr4.u"; - $TABLE="K256"; - $func="sha256_block_data_order"; - @Sigma0=( 2,13,22); - @Sigma1=( 6,11,25); - @sigma0=( 7,18, 3); - @sigma1=(17,19,10); - $rounds=64; -} else { die "nonsense $output"; } - -open STDOUT,">$output" || die "can't open $output: $!"; - -if ($^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)) - { $big_endian=(unpack('L',pack('N',1))==1); } - -$code=<<___; -.ident \"$output, version 1.1\" -.ident \"IA-64 ISA artwork by Andy Polyakov \" -.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; -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 -A_=r54; B_=r55; C_=r56; D_=r57; -E_=r58; F_=r59; G_=r60; H_=r61; - -// void $func (SHA_CTX *ctx, const void *in,size_t num[,int host]) -.global $func# -.proc $func# -.align 32 -$func: - .prologue - .save ar.pfs,pfssave -{ .mmi; alloc pfssave=ar.pfs,3,27,0,16 - $ADDP ctx=0,r32 // 1st arg - .save ar.lc,lcsave - mov lcsave=ar.lc } -{ .mmi; $ADDP input=0,r33 // 2nd arg - mov num=r34 // 3rd arg - .save pr,prsave - mov prsave=pr };; - - .body -{ .mib; add r8=0*$SZ,ctx - add r9=1*$SZ,ctx - brp.loop.imp .L_first16,.L_first16_end-16 } -{ .mib; add r10=2*$SZ,ctx - add r11=3*$SZ,ctx - brp.loop.imp .L_rest,.L_rest_end-16 };; - -// 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] - $LDW H_=[r11] - cmp.ne p0,p16=0,r0 };; // used in sha256_block -___ -$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: -.rotr X[16] -{ .mmi; mov A=A_ - mov B=B_ - mov ar.lc=14 } -{ .mmi; mov C=C_ - mov D=D_ - mov E=E_ } -{ .mmi; mov F=F_ - mov G=G_ - mov ar.ec=2 } -{ .mmi; ld1 X[15]=[input],$SZ // eliminated in 64-bit - mov H=H_ - mov sgm1=$sigma1[2] };; - -___ -$t0="t0", $t1="t1", $code.=<<___ if ($BITS==32); -.align 32 -.L_first16: -{ .mmi; add r9=1-$SZ,input - add r10=2-$SZ,input - add r11=3-$SZ,input };; -{ .mmi; ld1 r9=[r9] - ld1 r10=[r10] - dep.z $t1=E,32,32 } -{ .mmi; $LDW K=[Ktbl],$SZ - ld1 r11=[r11] - zxt4 E=E };; -{ .mii; or $t1=$t1,E - dep X[15]=X[15],r9,8,8 - dep r11=r10,r11,8,8 };; -{ .mmi; and T1=F,E - and T2=A,B - dep X[15]=X[15],r11,16,16 } -{ .mmi; andcm r8=G,E - and r9=A,C - mux2 $t0=A,0x44 };; // copy lower half to upper -{ .mmi; (p16) ld1 X[15-1]=[input],$SZ // prefetch - 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); -// in 64-bit mode 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 - 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 - 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 } -{ .mfb; 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 - 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 } -{ .mfb; 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 - 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 } -{ .mfb; 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 - br.many .L_first16 };; - -.align 32 -.L_first16: -{ .mmi; $LDW K=[Ktbl],$SZ - and T1=F,E - and T2=A,B } -{ .mmi; //$LDW X[15]=[input],$SZ // X[i]=*input++ - 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 -___ -$code.=<<___; -{ .mib; add T1=T1,H // T1=Ch(e,f,g)+h - _rotr r8=$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,$Sigma0[2] } // ROTR(a,39) -{ .mib; xor r10=r10,r11 - mux2 X[15]=X[15],0x44 };; // eliminated in 64-bit -{ .mmi; xor r10=r8,r10 // r10=Sigma0(a) - mov B=A - add A=T1,T2 };; -{ .mib; add E=E,T1 - add A=A,r10 // T2=Maj(a,b,c)+Sigma0(a) - br.ctop.sptk .L_first16 };; -.L_first16_end: - -{ .mii; mov ar.lc=$rounds-17 - mov ar.ec=1 };; - -.align 32 -.L_rest: -.rotr X[16] -{ .mib; $LDW K=[Ktbl],$SZ - _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] - $SHRU s0=X[15-1],sgm0 };; // s0=X[(i+1)&0xF]>>7 -{ .mib; and T1=F,E - _rotr r9=X[15-1],$sigma0[1] } // ROTR(s0,8) -{ .mib; andcm r10=G,E - $SHRU s1=X[15-14],sgm1 };; // s1=X[(i+14)&0xF]>>6 -{ .mmi; xor T1=T1,r10 // T1=((e & f) ^ (~e & g)) - xor r9=r8,r9 - _rotr r10=X[15-14],$sigma1[0] };;// ROTR(s1,19) -{ .mib; and T2=A,B - _rotr r11=X[15-14],$sigma1[1] }// ROTR(s1,61) -{ .mib; and r8=A,C };; -___ -$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 $t0=A,0x44 };; // copy lower half to upper -{ .mmi; xor T2=T2,r8 - _rotr r9=$t1,$Sigma1[0] } // ROTR(e,14) -{ .mmi; and r10=B,C - add T1=T1,H // T1=Ch(e,f,g)+h - $ADD X[15]=X[15],s0 };; // X[i&0xF]+=sigma0(X[(i+1)&0xF]) -___ -$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) -{ .mib; xor r10=r11,r10 - xor T2=T2,r8 };; -{ .mib; xor s1=s1,r10 // s1=sigma1(X[(i+14)&0xF]) - add T1=T1,H } -{ .mib; and r10=B,C - $ADD X[15]=X[15],s0 };; // X[i&0xF]+=sigma0(X[(i+1)&0xF]) -___ -$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,$Sigma0[2] };; // ROTR(a,39) -{ .mmi; xor r10=r8,r10 // r10=Sigma0(a) - mov B=A - add A=T1,T2 };; -{ .mib; add E=E,T1 - add A=A,r10 // T2=Maj(a,b,c)+Sigma0(a) - br.ctop.sptk .L_rest };; -.L_rest_end: - -{ .mmi; add A_=A_,A - add B_=B_,B - add C_=C_,C } -{ .mmi; add D_=D_,D - add E_=E_,E - cmp.ltu p16,p0=1,num };; -{ .mmi; add F_=F_,F - add G_=G_,G - add H_=H_,H } -{ .mmb; add Ktbl=-$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; $STW [r8]=A_,4*$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); - $code =~ s/(shrp\s+X\[[^=]+)=([^,]+),([^,]+),([1-9]+)/$1=$3,$2,64-$4/gm - if (!$big_endian); - $code =~ s/ld1(\s+)X\[\S+/nop.m$1 0x0/gm; -} - -print $code; - -print<<___ if ($BITS==32); -.align 64 -.type K256#,\@object -K256: data4 0x428a2f98,0x71374491,0xb5c0fbcf,0xe9b5dba5 - data4 0x3956c25b,0x59f111f1,0x923f82a4,0xab1c5ed5 - data4 0xd807aa98,0x12835b01,0x243185be,0x550c7dc3 - data4 0x72be5d74,0x80deb1fe,0x9bdc06a7,0xc19bf174 - data4 0xe49b69c1,0xefbe4786,0x0fc19dc6,0x240ca1cc - data4 0x2de92c6f,0x4a7484aa,0x5cb0a9dc,0x76f988da - data4 0x983e5152,0xa831c66d,0xb00327c8,0xbf597fc7 - data4 0xc6e00bf3,0xd5a79147,0x06ca6351,0x14292967 - data4 0x27b70a85,0x2e1b2138,0x4d2c6dfc,0x53380d13 - data4 0x650a7354,0x766a0abb,0x81c2c92e,0x92722c85 - data4 0xa2bfe8a1,0xa81a664b,0xc24b8b70,0xc76c51a3 - data4 0xd192e819,0xd6990624,0xf40e3585,0x106aa070 - data4 0x19a4c116,0x1e376c08,0x2748774c,0x34b0bcb5 - data4 0x391c0cb3,0x4ed8aa4a,0x5b9cca4f,0x682e6ff3 - data4 0x748f82ee,0x78a5636f,0x84c87814,0x8cc70208 - data4 0x90befffa,0xa4506ceb,0xbef9a3f7,0xc67178f2 -.size K256#,$SZ*$rounds -stringz "SHA256 block transform for IA64, CRYPTOGAMS by " -___ -print<<___ if ($BITS==64); -.align 64 -.type K512#,\@object -K512: data8 0x428a2f98d728ae22,0x7137449123ef65cd - data8 0xb5c0fbcfec4d3b2f,0xe9b5dba58189dbbc - data8 0x3956c25bf348b538,0x59f111f1b605d019 - data8 0x923f82a4af194f9b,0xab1c5ed5da6d8118 - data8 0xd807aa98a3030242,0x12835b0145706fbe - data8 0x243185be4ee4b28c,0x550c7dc3d5ffb4e2 - data8 0x72be5d74f27b896f,0x80deb1fe3b1696b1 - data8 0x9bdc06a725c71235,0xc19bf174cf692694 - data8 0xe49b69c19ef14ad2,0xefbe4786384f25e3 - data8 0x0fc19dc68b8cd5b5,0x240ca1cc77ac9c65 - data8 0x2de92c6f592b0275,0x4a7484aa6ea6e483 - data8 0x5cb0a9dcbd41fbd4,0x76f988da831153b5 - data8 0x983e5152ee66dfab,0xa831c66d2db43210 - data8 0xb00327c898fb213f,0xbf597fc7beef0ee4 - data8 0xc6e00bf33da88fc2,0xd5a79147930aa725 - data8 0x06ca6351e003826f,0x142929670a0e6e70 - data8 0x27b70a8546d22ffc,0x2e1b21385c26c926 - data8 0x4d2c6dfc5ac42aed,0x53380d139d95b3df - data8 0x650a73548baf63de,0x766a0abb3c77b2a8 - data8 0x81c2c92e47edaee6,0x92722c851482353b - data8 0xa2bfe8a14cf10364,0xa81a664bbc423001 - data8 0xc24b8b70d0f89791,0xc76c51a30654be30 - data8 0xd192e819d6ef5218,0xd69906245565a910 - data8 0xf40e35855771202a,0x106aa07032bbd1b8 - data8 0x19a4c116b8d2d0c8,0x1e376c085141ab53 - data8 0x2748774cdf8eeb99,0x34b0bcb5e19b48a8 - data8 0x391c0cb3c5c95a63,0x4ed8aa4ae3418acb - data8 0x5b9cca4f7763e373,0x682e6ff3d6b2b8a3 - data8 0x748f82ee5defb2fc,0x78a5636f43172f60 - data8 0x84c87814a1f0ab72,0x8cc702081a6439ec - data8 0x90befffa23631e28,0xa4506cebde82bde9 - data8 0xbef9a3f7b2c67915,0xc67178f2e372532b - data8 0xca273eceea26619c,0xd186b8c721c0c207 - data8 0xeada7dd6cde0eb1e,0xf57d4f7fee6ed178 - data8 0x06f067aa72176fba,0x0a637dc5a2c898a6 - data8 0x113f9804bef90dae,0x1b710b35131c471b - data8 0x28db77f523047d84,0x32caab7b40c72493 - data8 0x3c9ebe0a15c9bebc,0x431d67c49c100d4c - data8 0x4cc5d4becb3e42b6,0x597f299cfc657e2a - data8 0x5fcb6fab3ad6faec,0x6c44198c4a475817 -.size K512#,$SZ*$rounds -stringz "SHA512 block transform for IA64, CRYPTOGAMS by " -___ diff --git a/src/lib/libcrypto/whrlpool/wp_block.c b/src/lib/libcrypto/whrlpool/wp_block.c index 57f5b5df7a..d8c1b89ba3 100644 --- a/src/lib/libcrypto/whrlpool/wp_block.c +++ b/src/lib/libcrypto/whrlpool/wp_block.c @@ -1,4 +1,4 @@ -/* $OpenBSD: wp_block.c,v 1.11 2016/09/04 13:39:48 jsing Exp $ */ +/* $OpenBSD: wp_block.c,v 1.12 2016/09/04 14:06:46 jsing Exp $ */ /** * The Whirlpool hashing function. * @@ -73,14 +73,6 @@ typedef unsigned long long u64; # if defined(__x86_64) || defined(__x86_64__) # define ROTATE(a,n) ({ u64 ret; asm ("rolq %1,%0" \ : "=r"(ret) : "J"(n),"0"(a) : "cc"); ret; }) -# elif defined(__ia64) || defined(__ia64__) -# if BYTE_ORDER == LITTLE_ENDIAN -# define ROTATE(a,n) ({ u64 ret; asm ("shrp %0=%1,%1,%2" \ - : "=r"(ret) : "r"(a),"M"(64-(n))); ret; }) -# else -# define ROTATE(a,n) ({ u64 ret; asm ("shrp %0=%1,%1,%2" \ - : "=r"(ret) : "r"(a),"M"(n)); ret; }) -# endif # endif #endif -- cgit v1.2.3-55-g6feb