1#! /usr/bin/env perl
2# Copyright 2004-2016 The OpenSSL Project Authors. All Rights Reserved.
3#
4# Licensed under the Apache License 2.0 (the "License").  You may not use
5# this file except in compliance with the License.  You can obtain a copy
6# in the file LICENSE in the source distribution or at
7# https://www.openssl.org/source/license.html
8
9#
10# ====================================================================
11# Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
12# project. The module is, however, dual licensed under OpenSSL and
13# CRYPTOGAMS licenses depending on where you obtain it. For further
14# details see http://www.openssl.org/~appro/cryptogams/.
15# ====================================================================
16#
17# SHA256/512_Transform for Itanium.
18#
19# sha512_block runs in 1003 cycles on Itanium 2, which is almost 50%
20# faster than gcc and >60%(!) faster than code generated by HP-UX
21# compiler (yes, HP-UX is generating slower code, because unlike gcc,
22# it failed to deploy "shift right pair," 'shrp' instruction, which
23# substitutes for 64-bit rotate).
24#
25# 924 cycles long sha256_block outperforms gcc by over factor of 2(!)
26# and HP-UX compiler - by >40% (yes, gcc won sha512_block, but lost
27# this one big time). Note that "formally" 924 is about 100 cycles
28# too much. I mean it's 64 32-bit rounds vs. 80 virtually identical
29# 64-bit ones and 1003*64/80 gives 802. Extra cycles, 2 per round,
30# are spent on extra work to provide for 32-bit rotations. 32-bit
31# rotations are still handled by 'shrp' instruction and for this
32# reason lower 32 bits are deposited to upper half of 64-bit register
33# prior 'shrp' issue. And in order to minimize the amount of such
34# operations, X[16] values are *maintained* with copies of lower
35# halves in upper halves, which is why you'll spot such instructions
36# as custom 'mux2', "parallel 32-bit add," 'padd4' and "parallel
37# 32-bit unsigned right shift," 'pshr4.u' instructions here.
38#
39# Rules of engagement.
40#
41# There is only one integer shifter meaning that if I have two rotate,
42# deposit or extract instructions in adjacent bundles, they shall
43# split [at run-time if they have to]. But note that variable and
44# parallel shifts are performed by multi-media ALU and *are* pairable
45# with rotates [and alike]. On the backside MMALU is rather slow: it
46# takes 2 extra cycles before the result of integer operation is
47# available *to* MMALU and 2(*) extra cycles before the result of MM
48# operation is available "back" *to* integer ALU, not to mention that
49# MMALU itself has 2 cycles latency. However! I explicitly scheduled
50# these MM instructions to avoid MM stalls, so that all these extra
51# latencies get "hidden" in instruction-level parallelism.
52#
53# (*) 2 cycles on Itanium 1 and 1 cycle on Itanium 2. But I schedule
54#     for 2 in order to provide for best *overall* performance,
55#     because on Itanium 1 stall on MM result is accompanied by
56#     pipeline flush, which takes 6 cycles:-(
57#
58# June 2012
59#
60# Improve performance by 15-20%. Note about "rules of engagement"
61# above. Contemporary cores are equipped with additional shifter,
62# so that they should perform even better than below, presumably
63# by ~10%.
64#
65######################################################################
66# Current performance in cycles per processed byte for Itanium 2
67# pre-9000 series [little-endian] system:
68#
69# SHA1(*)	5.7
70# SHA256	12.6
71# SHA512	6.7
72#
73# (*) SHA1 result is presented purely for reference purposes.
74#
75# To generate code, pass the file name with either 256 or 512 in its
76# name and compiler flags.
77
78# $output is the last argument if it looks like a file (it has an extension)
79$output = $#ARGV >= 0 && $ARGV[$#ARGV] =~ m|\.\w+$| ? pop : undef;
80
81if ($output =~ /512.*\.[s|asm]/i) {
82	$SZ=8;
83	$BITS=8*$SZ;
84	$LDW="ld8";
85	$STW="st8";
86	$ADD="add";
87	$SHRU="shr.u";
88	$TABLE="K512";
89	$func="sha512_block_data_order";
90	@Sigma0=(28,34,39);
91	@Sigma1=(14,18,41);
92	@sigma0=(1,  8, 7);
93	@sigma1=(19,61, 6);
94	$rounds=80;
95} elsif ($output =~ /256.*\.[s|asm]/i) {
96	$SZ=4;
97	$BITS=8*$SZ;
98	$LDW="ld4";
99	$STW="st4";
100	$ADD="padd4";
101	$SHRU="pshr4.u";
102	$TABLE="K256";
103	$func="sha256_block_data_order";
104	@Sigma0=( 2,13,22);
105	@Sigma1=( 6,11,25);
106	@sigma0=( 7,18, 3);
107	@sigma1=(17,19,10);
108	$rounds=64;
109} else { die "nonsense $output"; }
110
111$output and (open STDOUT,">$output" or die "can't open $output: $!");
112
113if ($^O eq "hpux") {
114    $ADDP="addp4";
115    for (@ARGV) { $ADDP="add" if (/[\+DD|\-mlp]64/); }
116} else { $ADDP="add"; }
117for (@ARGV)  {	$big_endian=1 if (/\-DB_ENDIAN/);
118		$big_endian=0 if (/\-DL_ENDIAN/);  }
119if (!defined($big_endian))
120             {	$big_endian=(unpack('L',pack('N',1))==1);  }
121
122$code=<<___;
123.ident  \"$output, version 2.0\"
124.ident  \"IA-64 ISA artwork by Andy Polyakov <appro\@openssl.org>\"
125.explicit
126.text
127
128pfssave=r2;
129lcsave=r3;
130prsave=r14;
131K=r15;
132A_=r16; B_=r17; C_=r18; D_=r19;
133E_=r20; F_=r21; G_=r22; H_=r23;
134T1=r24;	T2=r25;
135s0=r26;	s1=r27;	t0=r28;	t1=r29;
136Ktbl=r30;
137ctx=r31;	// 1st arg
138input=r56;	// 2nd arg
139num=r57;	// 3rd arg
140sgm0=r58;	sgm1=r59;	// small constants
141
142// void $func (SHA_CTX *ctx, const void *in,size_t num[,int host])
143.global	$func#
144.proc	$func#
145.align	32
146.skip	16
147$func:
148	.prologue
149	.save	ar.pfs,pfssave
150{ .mmi;	alloc	pfssave=ar.pfs,3,25,0,24
151	$ADDP	ctx=0,r32		// 1st arg
152	.save	ar.lc,lcsave
153	mov	lcsave=ar.lc	}
154{ .mmi;	$ADDP	input=0,r33		// 2nd arg
155	mov	num=r34			// 3rd arg
156	.save	pr,prsave
157	mov	prsave=pr	};;
158
159	.body
160{ .mib;	add	r8=0*$SZ,ctx
161	add	r9=1*$SZ,ctx	}
162{ .mib;	add	r10=2*$SZ,ctx
163	add	r11=3*$SZ,ctx	};;
164
165// load A-H
166.Lpic_point:
167{ .mmi;	$LDW	A_=[r8],4*$SZ
168	$LDW	B_=[r9],4*$SZ
169	mov	Ktbl=ip		}
170{ .mmi;	$LDW	C_=[r10],4*$SZ
171	$LDW	D_=[r11],4*$SZ
172	mov	sgm0=$sigma0[2]	};;
173{ .mmi;	$LDW	E_=[r8]
174	$LDW	F_=[r9]
175	add	Ktbl=($TABLE#-.Lpic_point),Ktbl		}
176{ .mmi;	$LDW	G_=[r10]
177	$LDW	H_=[r11]
178	cmp.ne	p0,p16=0,r0	};;
179___
180$code.=<<___ if ($BITS==64);
181{ .mii;	and	r8=7,input
182	and	input=~7,input;;
183	cmp.eq	p9,p0=1,r8	}
184{ .mmi;	cmp.eq	p10,p0=2,r8
185	cmp.eq	p11,p0=3,r8
186	cmp.eq	p12,p0=4,r8	}
187{ .mmi;	cmp.eq	p13,p0=5,r8
188	cmp.eq	p14,p0=6,r8
189	cmp.eq	p15,p0=7,r8	};;
190___
191$code.=<<___;
192.L_outer:
193.rotr	R[8],X[16]
194A=R[0]; B=R[1]; C=R[2]; D=R[3]; E=R[4]; F=R[5]; G=R[6]; H=R[7]
195{ .mmi;	ld1	X[15]=[input],$SZ		// eliminated in sha512
196	mov	A=A_
197	mov	ar.lc=14	}
198{ .mmi;	mov	B=B_
199	mov	C=C_
200	mov	D=D_		}
201{ .mmi;	mov	E=E_
202	mov	F=F_
203	mov	ar.ec=2		};;
204{ .mmi;	mov	G=G_
205	mov	H=H_
206	mov	sgm1=$sigma1[2]	}
207{ .mib;	mov	r8=0
208	add	r9=1-$SZ,input
209	brp.loop.imp	.L_first16,.L_first16_end-16	};;
210___
211$t0="A", $t1="E", $code.=<<___ if ($BITS==64);
212// in sha512 case I load whole X[16] at once and take care of alignment...
213{ .mmi;	add	r8=1*$SZ,input
214	add	r9=2*$SZ,input
215	add	r10=3*$SZ,input		};;
216{ .mmb;	$LDW	X[15]=[input],4*$SZ
217	$LDW	X[14]=[r8],4*$SZ
218(p9)	br.cond.dpnt.many	.L1byte	};;
219{ .mmb;	$LDW	X[13]=[r9],4*$SZ
220	$LDW	X[12]=[r10],4*$SZ
221(p10)	br.cond.dpnt.many	.L2byte	};;
222{ .mmb;	$LDW	X[11]=[input],4*$SZ
223	$LDW	X[10]=[r8],4*$SZ
224(p11)	br.cond.dpnt.many	.L3byte	};;
225{ .mmb;	$LDW	X[ 9]=[r9],4*$SZ
226	$LDW	X[ 8]=[r10],4*$SZ
227(p12)	br.cond.dpnt.many	.L4byte	};;
228{ .mmb;	$LDW	X[ 7]=[input],4*$SZ
229	$LDW	X[ 6]=[r8],4*$SZ
230(p13)	br.cond.dpnt.many	.L5byte	};;
231{ .mmb;	$LDW	X[ 5]=[r9],4*$SZ
232	$LDW	X[ 4]=[r10],4*$SZ
233(p14)	br.cond.dpnt.many	.L6byte	};;
234{ .mmb;	$LDW	X[ 3]=[input],4*$SZ
235	$LDW	X[ 2]=[r8],4*$SZ
236(p15)	br.cond.dpnt.many	.L7byte	};;
237{ .mmb;	$LDW	X[ 1]=[r9],4*$SZ
238	$LDW	X[ 0]=[r10],4*$SZ	}
239{ .mib;	mov	r8=0
240	mux1	X[15]=X[15],\@rev		// eliminated on big-endian
241	br.many	.L_first16		};;
242.L1byte:
243{ .mmi;	$LDW	X[13]=[r9],4*$SZ
244	$LDW	X[12]=[r10],4*$SZ
245	shrp	X[15]=X[15],X[14],56	};;
246{ .mmi;	$LDW	X[11]=[input],4*$SZ
247	$LDW	X[10]=[r8],4*$SZ
248	shrp	X[14]=X[14],X[13],56	}
249{ .mmi;	$LDW	X[ 9]=[r9],4*$SZ
250	$LDW	X[ 8]=[r10],4*$SZ
251	shrp	X[13]=X[13],X[12],56	};;
252{ .mmi;	$LDW	X[ 7]=[input],4*$SZ
253	$LDW	X[ 6]=[r8],4*$SZ
254	shrp	X[12]=X[12],X[11],56	}
255{ .mmi;	$LDW	X[ 5]=[r9],4*$SZ
256	$LDW	X[ 4]=[r10],4*$SZ
257	shrp	X[11]=X[11],X[10],56	};;
258{ .mmi;	$LDW	X[ 3]=[input],4*$SZ
259	$LDW	X[ 2]=[r8],4*$SZ
260	shrp	X[10]=X[10],X[ 9],56	}
261{ .mmi;	$LDW	X[ 1]=[r9],4*$SZ
262	$LDW	X[ 0]=[r10],4*$SZ
263	shrp	X[ 9]=X[ 9],X[ 8],56	};;
264{ .mii;	$LDW	T1=[input]
265	shrp	X[ 8]=X[ 8],X[ 7],56
266	shrp	X[ 7]=X[ 7],X[ 6],56	}
267{ .mii;	shrp	X[ 6]=X[ 6],X[ 5],56
268	shrp	X[ 5]=X[ 5],X[ 4],56	};;
269{ .mii;	shrp	X[ 4]=X[ 4],X[ 3],56
270	shrp	X[ 3]=X[ 3],X[ 2],56	}
271{ .mii;	shrp	X[ 2]=X[ 2],X[ 1],56
272	shrp	X[ 1]=X[ 1],X[ 0],56	}
273{ .mib;	shrp	X[ 0]=X[ 0],T1,56	}
274{ .mib;	mov	r8=0
275	mux1	X[15]=X[15],\@rev		// eliminated on big-endian
276	br.many	.L_first16		};;
277.L2byte:
278{ .mmi;	$LDW	X[11]=[input],4*$SZ
279	$LDW	X[10]=[r8],4*$SZ
280	shrp	X[15]=X[15],X[14],48	}
281{ .mmi;	$LDW	X[ 9]=[r9],4*$SZ
282	$LDW	X[ 8]=[r10],4*$SZ
283	shrp	X[14]=X[14],X[13],48	};;
284{ .mmi;	$LDW	X[ 7]=[input],4*$SZ
285	$LDW	X[ 6]=[r8],4*$SZ
286	shrp	X[13]=X[13],X[12],48	}
287{ .mmi;	$LDW	X[ 5]=[r9],4*$SZ
288	$LDW	X[ 4]=[r10],4*$SZ
289	shrp	X[12]=X[12],X[11],48	};;
290{ .mmi;	$LDW	X[ 3]=[input],4*$SZ
291	$LDW	X[ 2]=[r8],4*$SZ
292	shrp	X[11]=X[11],X[10],48	}
293{ .mmi;	$LDW	X[ 1]=[r9],4*$SZ
294	$LDW	X[ 0]=[r10],4*$SZ
295	shrp	X[10]=X[10],X[ 9],48	};;
296{ .mii;	$LDW	T1=[input]
297	shrp	X[ 9]=X[ 9],X[ 8],48
298	shrp	X[ 8]=X[ 8],X[ 7],48	}
299{ .mii;	shrp	X[ 7]=X[ 7],X[ 6],48
300	shrp	X[ 6]=X[ 6],X[ 5],48	};;
301{ .mii;	shrp	X[ 5]=X[ 5],X[ 4],48
302	shrp	X[ 4]=X[ 4],X[ 3],48	}
303{ .mii;	shrp	X[ 3]=X[ 3],X[ 2],48
304	shrp	X[ 2]=X[ 2],X[ 1],48	}
305{ .mii;	shrp	X[ 1]=X[ 1],X[ 0],48
306	shrp	X[ 0]=X[ 0],T1,48	}
307{ .mib;	mov	r8=0
308	mux1	X[15]=X[15],\@rev		// eliminated on big-endian
309	br.many	.L_first16		};;
310.L3byte:
311{ .mmi;	$LDW	X[ 9]=[r9],4*$SZ
312	$LDW	X[ 8]=[r10],4*$SZ
313	shrp	X[15]=X[15],X[14],40	};;
314{ .mmi;	$LDW	X[ 7]=[input],4*$SZ
315	$LDW	X[ 6]=[r8],4*$SZ
316	shrp	X[14]=X[14],X[13],40	}
317{ .mmi;	$LDW	X[ 5]=[r9],4*$SZ
318	$LDW	X[ 4]=[r10],4*$SZ
319	shrp	X[13]=X[13],X[12],40	};;
320{ .mmi;	$LDW	X[ 3]=[input],4*$SZ
321	$LDW	X[ 2]=[r8],4*$SZ
322	shrp	X[12]=X[12],X[11],40	}
323{ .mmi;	$LDW	X[ 1]=[r9],4*$SZ
324	$LDW	X[ 0]=[r10],4*$SZ
325	shrp	X[11]=X[11],X[10],40	};;
326{ .mii;	$LDW	T1=[input]
327	shrp	X[10]=X[10],X[ 9],40
328	shrp	X[ 9]=X[ 9],X[ 8],40	}
329{ .mii;	shrp	X[ 8]=X[ 8],X[ 7],40
330	shrp	X[ 7]=X[ 7],X[ 6],40	};;
331{ .mii;	shrp	X[ 6]=X[ 6],X[ 5],40
332	shrp	X[ 5]=X[ 5],X[ 4],40	}
333{ .mii;	shrp	X[ 4]=X[ 4],X[ 3],40
334	shrp	X[ 3]=X[ 3],X[ 2],40	}
335{ .mii;	shrp	X[ 2]=X[ 2],X[ 1],40
336	shrp	X[ 1]=X[ 1],X[ 0],40	}
337{ .mib;	shrp	X[ 0]=X[ 0],T1,40	}
338{ .mib;	mov	r8=0
339	mux1	X[15]=X[15],\@rev		// eliminated on big-endian
340	br.many	.L_first16		};;
341.L4byte:
342{ .mmi;	$LDW	X[ 7]=[input],4*$SZ
343	$LDW	X[ 6]=[r8],4*$SZ
344	shrp	X[15]=X[15],X[14],32	}
345{ .mmi;	$LDW	X[ 5]=[r9],4*$SZ
346	$LDW	X[ 4]=[r10],4*$SZ
347	shrp	X[14]=X[14],X[13],32	};;
348{ .mmi;	$LDW	X[ 3]=[input],4*$SZ
349	$LDW	X[ 2]=[r8],4*$SZ
350	shrp	X[13]=X[13],X[12],32	}
351{ .mmi;	$LDW	X[ 1]=[r9],4*$SZ
352	$LDW	X[ 0]=[r10],4*$SZ
353	shrp	X[12]=X[12],X[11],32	};;
354{ .mii;	$LDW	T1=[input]
355	shrp	X[11]=X[11],X[10],32
356	shrp	X[10]=X[10],X[ 9],32	}
357{ .mii;	shrp	X[ 9]=X[ 9],X[ 8],32
358	shrp	X[ 8]=X[ 8],X[ 7],32	};;
359{ .mii;	shrp	X[ 7]=X[ 7],X[ 6],32
360	shrp	X[ 6]=X[ 6],X[ 5],32	}
361{ .mii;	shrp	X[ 5]=X[ 5],X[ 4],32
362	shrp	X[ 4]=X[ 4],X[ 3],32	}
363{ .mii;	shrp	X[ 3]=X[ 3],X[ 2],32
364	shrp	X[ 2]=X[ 2],X[ 1],32	}
365{ .mii;	shrp	X[ 1]=X[ 1],X[ 0],32
366	shrp	X[ 0]=X[ 0],T1,32	}
367{ .mib;	mov	r8=0
368	mux1	X[15]=X[15],\@rev		// eliminated on big-endian
369	br.many	.L_first16		};;
370.L5byte:
371{ .mmi;	$LDW	X[ 5]=[r9],4*$SZ
372	$LDW	X[ 4]=[r10],4*$SZ
373	shrp	X[15]=X[15],X[14],24	};;
374{ .mmi;	$LDW	X[ 3]=[input],4*$SZ
375	$LDW	X[ 2]=[r8],4*$SZ
376	shrp	X[14]=X[14],X[13],24	}
377{ .mmi;	$LDW	X[ 1]=[r9],4*$SZ
378	$LDW	X[ 0]=[r10],4*$SZ
379	shrp	X[13]=X[13],X[12],24	};;
380{ .mii;	$LDW	T1=[input]
381	shrp	X[12]=X[12],X[11],24
382	shrp	X[11]=X[11],X[10],24	}
383{ .mii;	shrp	X[10]=X[10],X[ 9],24
384	shrp	X[ 9]=X[ 9],X[ 8],24	};;
385{ .mii;	shrp	X[ 8]=X[ 8],X[ 7],24
386	shrp	X[ 7]=X[ 7],X[ 6],24	}
387{ .mii;	shrp	X[ 6]=X[ 6],X[ 5],24
388	shrp	X[ 5]=X[ 5],X[ 4],24	}
389{ .mii;	shrp	X[ 4]=X[ 4],X[ 3],24
390	shrp	X[ 3]=X[ 3],X[ 2],24	}
391{ .mii;	shrp	X[ 2]=X[ 2],X[ 1],24
392	shrp	X[ 1]=X[ 1],X[ 0],24	}
393{ .mib;	shrp	X[ 0]=X[ 0],T1,24	}
394{ .mib;	mov	r8=0
395	mux1	X[15]=X[15],\@rev		// eliminated on big-endian
396	br.many	.L_first16		};;
397.L6byte:
398{ .mmi;	$LDW	X[ 3]=[input],4*$SZ
399	$LDW	X[ 2]=[r8],4*$SZ
400	shrp	X[15]=X[15],X[14],16	}
401{ .mmi;	$LDW	X[ 1]=[r9],4*$SZ
402	$LDW	X[ 0]=[r10],4*$SZ
403	shrp	X[14]=X[14],X[13],16	};;
404{ .mii;	$LDW	T1=[input]
405	shrp	X[13]=X[13],X[12],16
406	shrp	X[12]=X[12],X[11],16	}
407{ .mii;	shrp	X[11]=X[11],X[10],16
408	shrp	X[10]=X[10],X[ 9],16	};;
409{ .mii;	shrp	X[ 9]=X[ 9],X[ 8],16
410	shrp	X[ 8]=X[ 8],X[ 7],16	}
411{ .mii;	shrp	X[ 7]=X[ 7],X[ 6],16
412	shrp	X[ 6]=X[ 6],X[ 5],16	}
413{ .mii;	shrp	X[ 5]=X[ 5],X[ 4],16
414	shrp	X[ 4]=X[ 4],X[ 3],16	}
415{ .mii;	shrp	X[ 3]=X[ 3],X[ 2],16
416	shrp	X[ 2]=X[ 2],X[ 1],16	}
417{ .mii;	shrp	X[ 1]=X[ 1],X[ 0],16
418	shrp	X[ 0]=X[ 0],T1,16	}
419{ .mib;	mov	r8=0
420	mux1	X[15]=X[15],\@rev		// eliminated on big-endian
421	br.many	.L_first16		};;
422.L7byte:
423{ .mmi;	$LDW	X[ 1]=[r9],4*$SZ
424	$LDW	X[ 0]=[r10],4*$SZ
425	shrp	X[15]=X[15],X[14],8	};;
426{ .mii;	$LDW	T1=[input]
427	shrp	X[14]=X[14],X[13],8
428	shrp	X[13]=X[13],X[12],8	}
429{ .mii;	shrp	X[12]=X[12],X[11],8
430	shrp	X[11]=X[11],X[10],8	};;
431{ .mii;	shrp	X[10]=X[10],X[ 9],8
432	shrp	X[ 9]=X[ 9],X[ 8],8	}
433{ .mii;	shrp	X[ 8]=X[ 8],X[ 7],8
434	shrp	X[ 7]=X[ 7],X[ 6],8	}
435{ .mii;	shrp	X[ 6]=X[ 6],X[ 5],8
436	shrp	X[ 5]=X[ 5],X[ 4],8	}
437{ .mii;	shrp	X[ 4]=X[ 4],X[ 3],8
438	shrp	X[ 3]=X[ 3],X[ 2],8	}
439{ .mii;	shrp	X[ 2]=X[ 2],X[ 1],8
440	shrp	X[ 1]=X[ 1],X[ 0],8	}
441{ .mib;	shrp	X[ 0]=X[ 0],T1,8	}
442{ .mib;	mov	r8=0
443	mux1	X[15]=X[15],\@rev	};;	// eliminated on big-endian
444
445.align	32
446.L_first16:
447{ .mmi;		$LDW	K=[Ktbl],$SZ
448		add	A=A,r8			// H+=Sigma(0) from the past
449		_rotr	r10=$t1,$Sigma1[0]  }	// ROTR(e,14)
450{ .mmi;		and	T1=F,E
451		andcm	r8=G,E
452	(p16)	mux1	X[14]=X[14],\@rev   };;	// eliminated on big-endian
453{ .mmi;		and	T2=A,B
454		and	r9=A,C
455		_rotr	r11=$t1,$Sigma1[1]  }	// ROTR(e,41)
456{ .mmi;		xor	T1=T1,r8		// T1=((e & f) ^ (~e & g))
457		and	r8=B,C		    };;
458___
459$t0="t0", $t1="t1", $code.=<<___ if ($BITS==32);
460.align	32
461.L_first16:
462{ .mmi;		add	A=A,r8			// H+=Sigma(0) from the past
463		add	r10=2-$SZ,input
464		add	r11=3-$SZ,input	};;
465{ .mmi;		ld1	r9=[r9]
466		ld1	r10=[r10]
467		dep.z	$t1=E,32,32	}
468{ .mmi;		ld1	r11=[r11]
469		$LDW	K=[Ktbl],$SZ
470		zxt4	E=E		};;
471{ .mii;		or	$t1=$t1,E
472		dep	X[15]=X[15],r9,8,8
473		mux2	$t0=A,0x44	};;	// copy lower half to upper
474{ .mmi;		and	T1=F,E
475		andcm	r8=G,E
476		dep	r11=r10,r11,8,8	};;
477{ .mmi;		and	T2=A,B
478		and	r9=A,C
479		dep	X[15]=X[15],r11,16,16	};;
480{ .mmi;	(p16)	ld1	X[15-1]=[input],$SZ	// prefetch
481		xor	T1=T1,r8		// T1=((e & f) ^ (~e & g))
482		_rotr	r10=$t1,$Sigma1[0] }	// ROTR(e,14)
483{ .mmi;		and	r8=B,C
484		_rotr	r11=$t1,$Sigma1[1] };;	// ROTR(e,18)
485___
486$code.=<<___;
487{ .mmi;		add	T1=T1,H			// T1=Ch(e,f,g)+h
488		xor	r10=r10,r11
489		_rotr	r11=$t1,$Sigma1[2]  }	// ROTR(e,41)
490{ .mmi;		xor	T2=T2,r9
491		add	K=K,X[15]	    };;
492{ .mmi;		add	T1=T1,K			// T1+=K[i]+X[i]
493		xor	T2=T2,r8		// T2=((a & b) ^ (a & c) ^ (b & c))
494		_rotr	r8=$t0,$Sigma0[0]   }	// ROTR(a,28)
495{ .mmi;		xor	r11=r11,r10		// Sigma1(e)
496		_rotr	r9=$t0,$Sigma0[1]   };;	// ROTR(a,34)
497{ .mmi;		add	T1=T1,r11		// T+=Sigma1(e)
498		xor	r8=r8,r9
499		_rotr	r9=$t0,$Sigma0[2]   };;	// ROTR(a,39)
500{ .mmi;		xor	r8=r8,r9		// Sigma0(a)
501		add	D=D,T1
502		mux2	H=X[15],0x44	    }	// mov H=X[15] in sha512
503{ .mib;	(p16)	add	r9=1-$SZ,input		// not used in sha512
504		add	X[15]=T1,T2		// H=T1+Maj(a,b,c)
505	br.ctop.sptk	.L_first16	    };;
506.L_first16_end:
507
508{ .mib;	mov	ar.lc=$rounds-17
509	brp.loop.imp	.L_rest,.L_rest_end-16		}
510{ .mib;	mov	ar.ec=1
511	br.many	.L_rest			};;
512
513.align	32
514.L_rest:
515{ .mmi;		$LDW	K=[Ktbl],$SZ
516		add	A=A,r8			// H+=Sigma0(a) from the past
517		_rotr	r8=X[15-1],$sigma0[0] }	// ROTR(s0,1)
518{ .mmi; 	add	X[15]=X[15],X[15-9]	// X[i&0xF]+=X[(i+9)&0xF]
519		$SHRU	s0=X[15-1],sgm0	    };;	// s0=X[(i+1)&0xF]>>7
520{ .mib;		and	T1=F,E
521		_rotr	r9=X[15-1],$sigma0[1] }	// ROTR(s0,8)
522{ .mib;		andcm	r10=G,E
523		$SHRU	s1=X[15-14],sgm1    };;	// s1=X[(i+14)&0xF]>>6
524// Pair of mmi; splits on Itanium 1 and prevents pipeline flush
525// upon $SHRU output usage
526{ .mmi;		xor	T1=T1,r10		// T1=((e & f) ^ (~e & g))
527		xor	r9=r8,r9
528		_rotr	r10=X[15-14],$sigma1[0] }// ROTR(s1,19)
529{ .mmi;		and	T2=A,B
530		and	r8=A,C
531		_rotr	r11=X[15-14],$sigma1[1] };;// ROTR(s1,61)
532___
533$t0="t0", $t1="t1", $code.=<<___ if ($BITS==32);
534{ .mib;		xor	s0=s0,r9		// s0=sigma0(X[(i+1)&0xF])
535		dep.z	$t1=E,32,32	    }
536{ .mib;		xor	r10=r11,r10
537		zxt4	E=E		    };;
538{ .mii;		xor	s1=s1,r10		// s1=sigma1(X[(i+14)&0xF])
539		shrp	r9=E,$t1,32+$Sigma1[0]	// ROTR(e,14)
540		mux2	$t0=A,0x44	    };;	// copy lower half to upper
541// Pair of mmi; splits on Itanium 1 and prevents pipeline flush
542// upon mux2 output usage
543{ .mmi;		xor	T2=T2,r8
544		shrp	r8=E,$t1,32+$Sigma1[1]}	// ROTR(e,18)
545{ .mmi;		and	r10=B,C
546		add	T1=T1,H			// T1=Ch(e,f,g)+h
547		or	$t1=$t1,E   	    };;
548___
549$t0="A", $t1="E", $code.=<<___ if ($BITS==64);
550{ .mib;		xor	s0=s0,r9		// s0=sigma0(X[(i+1)&0xF])
551		_rotr	r9=$t1,$Sigma1[0]   }	// ROTR(e,14)
552{ .mib;		xor	r10=r11,r10
553		xor	T2=T2,r8	    };;
554{ .mib;		xor	s1=s1,r10		// s1=sigma1(X[(i+14)&0xF])
555		_rotr	r8=$t1,$Sigma1[1]   }	// ROTR(e,18)
556{ .mib;		and	r10=B,C
557		add	T1=T1,H		    };;	// T1+=H
558___
559$code.=<<___;
560{ .mib;		xor	r9=r9,r8
561		_rotr	r8=$t1,$Sigma1[2]   }	// ROTR(e,41)
562{ .mib;		xor	T2=T2,r10		// T2=((a & b) ^ (a & c) ^ (b & c))
563		add	X[15]=X[15],s0	    };;	// X[i]+=sigma0(X[i+1])
564{ .mmi;		xor	r9=r9,r8		// Sigma1(e)
565		add	X[15]=X[15],s1		// X[i]+=sigma0(X[i+14])
566		_rotr	r8=$t0,$Sigma0[0]   };;	// ROTR(a,28)
567{ .mmi;		add	K=K,X[15]
568		add	T1=T1,r9		// T1+=Sigma1(e)
569		_rotr	r9=$t0,$Sigma0[1]   };;	// ROTR(a,34)
570{ .mmi;		add	T1=T1,K			// T1+=K[i]+X[i]
571		xor	r8=r8,r9
572		_rotr	r9=$t0,$Sigma0[2]   };;	// ROTR(a,39)
573{ .mib;		add	D=D,T1
574		mux2	H=X[15],0x44	    }	// mov H=X[15] in sha512
575{ .mib;		xor	r8=r8,r9		// Sigma0(a)
576		add	X[15]=T1,T2		// H=T1+Maj(a,b,c)
577	br.ctop.sptk	.L_rest		    };;
578.L_rest_end:
579
580{ .mmi;	add	A=A,r8			};;	// H+=Sigma0(a) from the past
581{ .mmi;	add	A_=A_,A
582	add	B_=B_,B
583	add	C_=C_,C			}
584{ .mmi;	add	D_=D_,D
585	add	E_=E_,E
586	cmp.ltu	p16,p0=1,num		};;
587{ .mmi;	add	F_=F_,F
588	add	G_=G_,G
589	add	H_=H_,H			}
590{ .mmb;	add	Ktbl=-$SZ*$rounds,Ktbl
591(p16)	add	num=-1,num
592(p16)	br.dptk.many	.L_outer	};;
593
594{ .mib;	add	r8=0*$SZ,ctx
595	add	r9=1*$SZ,ctx		}
596{ .mib;	add	r10=2*$SZ,ctx
597	add	r11=3*$SZ,ctx		};;
598{ .mmi;	$STW	[r8]=A_,4*$SZ
599	$STW	[r9]=B_,4*$SZ
600	mov	ar.lc=lcsave		}
601{ .mmi;	$STW	[r10]=C_,4*$SZ
602	$STW	[r11]=D_,4*$SZ
603	mov	pr=prsave,0x1ffff	};;
604{ .mmb;	$STW	[r8]=E_
605	$STW	[r9]=F_			}
606{ .mmb;	$STW	[r10]=G_
607	$STW	[r11]=H_
608	br.ret.sptk.many	b0	};;
609.endp	$func#
610___
611
612foreach(split($/,$code)) {
613    s/\`([^\`]*)\`/eval $1/gem;
614    s/_rotr(\s+)([^=]+)=([^,]+),([0-9]+)/shrp$1$2=$3,$3,$4/gm;
615    if ($BITS==64) {
616	s/mux2(\s+)([^=]+)=([^,]+),\S+/mov$1 $2=$3/gm;
617	s/mux1(\s+)\S+/nop.i$1 0x0/gm	if ($big_endian);
618	s/(shrp\s+X\[[^=]+)=([^,]+),([^,]+),([1-9]+)/$1=$3,$2,64-$4/gm
619    						if (!$big_endian);
620	s/ld1(\s+)X\[\S+/nop.m$1 0x0/gm;
621    }
622
623    print $_,"\n";
624}
625
626print<<___ if ($BITS==32);
627.align	64
628.type	K256#,\@object
629K256:	data4	0x428a2f98,0x71374491,0xb5c0fbcf,0xe9b5dba5
630	data4	0x3956c25b,0x59f111f1,0x923f82a4,0xab1c5ed5
631	data4	0xd807aa98,0x12835b01,0x243185be,0x550c7dc3
632	data4	0x72be5d74,0x80deb1fe,0x9bdc06a7,0xc19bf174
633	data4	0xe49b69c1,0xefbe4786,0x0fc19dc6,0x240ca1cc
634	data4	0x2de92c6f,0x4a7484aa,0x5cb0a9dc,0x76f988da
635	data4	0x983e5152,0xa831c66d,0xb00327c8,0xbf597fc7
636	data4	0xc6e00bf3,0xd5a79147,0x06ca6351,0x14292967
637	data4	0x27b70a85,0x2e1b2138,0x4d2c6dfc,0x53380d13
638	data4	0x650a7354,0x766a0abb,0x81c2c92e,0x92722c85
639	data4	0xa2bfe8a1,0xa81a664b,0xc24b8b70,0xc76c51a3
640	data4	0xd192e819,0xd6990624,0xf40e3585,0x106aa070
641	data4	0x19a4c116,0x1e376c08,0x2748774c,0x34b0bcb5
642	data4	0x391c0cb3,0x4ed8aa4a,0x5b9cca4f,0x682e6ff3
643	data4	0x748f82ee,0x78a5636f,0x84c87814,0x8cc70208
644	data4	0x90befffa,0xa4506ceb,0xbef9a3f7,0xc67178f2
645.size	K256#,$SZ*$rounds
646stringz	"SHA256 block transform for IA64, CRYPTOGAMS by <appro\@openssl.org>"
647___
648print<<___ if ($BITS==64);
649.align	64
650.type	K512#,\@object
651K512:	data8	0x428a2f98d728ae22,0x7137449123ef65cd
652	data8	0xb5c0fbcfec4d3b2f,0xe9b5dba58189dbbc
653	data8	0x3956c25bf348b538,0x59f111f1b605d019
654	data8	0x923f82a4af194f9b,0xab1c5ed5da6d8118
655	data8	0xd807aa98a3030242,0x12835b0145706fbe
656	data8	0x243185be4ee4b28c,0x550c7dc3d5ffb4e2
657	data8	0x72be5d74f27b896f,0x80deb1fe3b1696b1
658	data8	0x9bdc06a725c71235,0xc19bf174cf692694
659	data8	0xe49b69c19ef14ad2,0xefbe4786384f25e3
660	data8	0x0fc19dc68b8cd5b5,0x240ca1cc77ac9c65
661	data8	0x2de92c6f592b0275,0x4a7484aa6ea6e483
662	data8	0x5cb0a9dcbd41fbd4,0x76f988da831153b5
663	data8	0x983e5152ee66dfab,0xa831c66d2db43210
664	data8	0xb00327c898fb213f,0xbf597fc7beef0ee4
665	data8	0xc6e00bf33da88fc2,0xd5a79147930aa725
666	data8	0x06ca6351e003826f,0x142929670a0e6e70
667	data8	0x27b70a8546d22ffc,0x2e1b21385c26c926
668	data8	0x4d2c6dfc5ac42aed,0x53380d139d95b3df
669	data8	0x650a73548baf63de,0x766a0abb3c77b2a8
670	data8	0x81c2c92e47edaee6,0x92722c851482353b
671	data8	0xa2bfe8a14cf10364,0xa81a664bbc423001
672	data8	0xc24b8b70d0f89791,0xc76c51a30654be30
673	data8	0xd192e819d6ef5218,0xd69906245565a910
674	data8	0xf40e35855771202a,0x106aa07032bbd1b8
675	data8	0x19a4c116b8d2d0c8,0x1e376c085141ab53
676	data8	0x2748774cdf8eeb99,0x34b0bcb5e19b48a8
677	data8	0x391c0cb3c5c95a63,0x4ed8aa4ae3418acb
678	data8	0x5b9cca4f7763e373,0x682e6ff3d6b2b8a3
679	data8	0x748f82ee5defb2fc,0x78a5636f43172f60
680	data8	0x84c87814a1f0ab72,0x8cc702081a6439ec
681	data8	0x90befffa23631e28,0xa4506cebde82bde9
682	data8	0xbef9a3f7b2c67915,0xc67178f2e372532b
683	data8	0xca273eceea26619c,0xd186b8c721c0c207
684	data8	0xeada7dd6cde0eb1e,0xf57d4f7fee6ed178
685	data8	0x06f067aa72176fba,0x0a637dc5a2c898a6
686	data8	0x113f9804bef90dae,0x1b710b35131c471b
687	data8	0x28db77f523047d84,0x32caab7b40c72493
688	data8	0x3c9ebe0a15c9bebc,0x431d67c49c100d4c
689	data8	0x4cc5d4becb3e42b6,0x597f299cfc657e2a
690	data8	0x5fcb6fab3ad6faec,0x6c44198c4a475817
691.size	K512#,$SZ*$rounds
692stringz	"SHA512 block transform for IA64, CRYPTOGAMS by <appro\@openssl.org>"
693___
694