1#! /usr/bin/env perl
2# Copyright 2011-2020 The OpenSSL Project Authors. All Rights Reserved.
3#
4# Licensed under the OpenSSL license (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# May 2011
18#
19# The module implements bn_GF2m_mul_2x2 polynomial multiplication used
20# in bn_gf2m.c. It's kind of low-hanging mechanical port from C for
21# the time being... Except that it has two code paths: code suitable
22# for any x86_64 CPU and PCLMULQDQ one suitable for Westmere and
23# later. Improvement varies from one benchmark and µ-arch to another.
24# Vanilla code path is at most 20% faster than compiler-generated code
25# [not very impressive], while PCLMULQDQ - whole 85%-160% better on
26# 163- and 571-bit ECDH benchmarks on Intel CPUs. Keep in mind that
27# these coefficients are not ones for bn_GF2m_mul_2x2 itself, as not
28# all CPU time is burnt in it...
29
30$flavour = shift;
31$output  = shift;
32if ($flavour =~ /\./) { $output = $flavour; undef $flavour; }
33
34$win64=0; $win64=1 if ($flavour =~ /[nm]asm|mingw64/ || $output =~ /\.asm$/);
35
36$0 =~ m/(.*[\/\\])[^\/\\]+$/; $dir=$1;
37( $xlate="${dir}x86_64-xlate.pl" and -f $xlate ) or
38( $xlate="${dir}../../perlasm/x86_64-xlate.pl" and -f $xlate) or
39die "can't locate x86_64-xlate.pl";
40
41open OUT,"| \"$^X\" \"$xlate\" $flavour \"$output\"";
42*STDOUT=*OUT;
43
44($lo,$hi)=("%rax","%rdx");	$a=$lo;
45($i0,$i1)=("%rsi","%rdi");
46($t0,$t1)=("%rbx","%rcx");
47($b,$mask)=("%rbp","%r8");
48($a1,$a2,$a4,$a8,$a12,$a48)=map("%r$_",(9..15));
49($R,$Tx)=("%xmm0","%xmm1");
50
51$code.=<<___;
52.text
53
54.type	_mul_1x1,\@abi-omnipotent
55.align	16
56_mul_1x1:
57.cfi_startproc
58	sub	\$128+8,%rsp
59.cfi_adjust_cfa_offset	128+8
60	mov	\$-1,$a1
61	lea	($a,$a),$i0
62	shr	\$3,$a1
63	lea	(,$a,4),$i1
64	and	$a,$a1			# a1=a&0x1fffffffffffffff
65	lea	(,$a,8),$a8
66	sar	\$63,$a			# broadcast 63rd bit
67	lea	($a1,$a1),$a2
68	sar	\$63,$i0		# broadcast 62nd bit
69	lea	(,$a1,4),$a4
70	and	$b,$a
71	sar	\$63,$i1		# broadcast 61st bit
72	mov	$a,$hi			# $a is $lo
73	shl	\$63,$lo
74	and	$b,$i0
75	shr	\$1,$hi
76	mov	$i0,$t1
77	shl	\$62,$i0
78	and	$b,$i1
79	shr	\$2,$t1
80	xor	$i0,$lo
81	mov	$i1,$t0
82	shl	\$61,$i1
83	xor	$t1,$hi
84	shr	\$3,$t0
85	xor	$i1,$lo
86	xor	$t0,$hi
87
88	mov	$a1,$a12
89	movq	\$0,0(%rsp)		# tab[0]=0
90	xor	$a2,$a12		# a1^a2
91	mov	$a1,8(%rsp)		# tab[1]=a1
92	 mov	$a4,$a48
93	mov	$a2,16(%rsp)		# tab[2]=a2
94	 xor	$a8,$a48		# a4^a8
95	mov	$a12,24(%rsp)		# tab[3]=a1^a2
96
97	xor	$a4,$a1
98	mov	$a4,32(%rsp)		# tab[4]=a4
99	xor	$a4,$a2
100	mov	$a1,40(%rsp)		# tab[5]=a1^a4
101	xor	$a4,$a12
102	mov	$a2,48(%rsp)		# tab[6]=a2^a4
103	 xor	$a48,$a1		# a1^a4^a4^a8=a1^a8
104	mov	$a12,56(%rsp)		# tab[7]=a1^a2^a4
105	 xor	$a48,$a2		# a2^a4^a4^a8=a1^a8
106
107	mov	$a8,64(%rsp)		# tab[8]=a8
108	xor	$a48,$a12		# a1^a2^a4^a4^a8=a1^a2^a8
109	mov	$a1,72(%rsp)		# tab[9]=a1^a8
110	 xor	$a4,$a1			# a1^a8^a4
111	mov	$a2,80(%rsp)		# tab[10]=a2^a8
112	 xor	$a4,$a2			# a2^a8^a4
113	mov	$a12,88(%rsp)		# tab[11]=a1^a2^a8
114
115	xor	$a4,$a12		# a1^a2^a8^a4
116	mov	$a48,96(%rsp)		# tab[12]=a4^a8
117	 mov	$mask,$i0
118	mov	$a1,104(%rsp)		# tab[13]=a1^a4^a8
119	 and	$b,$i0
120	mov	$a2,112(%rsp)		# tab[14]=a2^a4^a8
121	 shr	\$4,$b
122	mov	$a12,120(%rsp)		# tab[15]=a1^a2^a4^a8
123	 mov	$mask,$i1
124	 and	$b,$i1
125	 shr	\$4,$b
126
127	movq	(%rsp,$i0,8),$R		# half of calculations is done in SSE2
128	mov	$mask,$i0
129	and	$b,$i0
130	shr	\$4,$b
131___
132    for ($n=1;$n<8;$n++) {
133	$code.=<<___;
134	mov	(%rsp,$i1,8),$t1
135	mov	$mask,$i1
136	mov	$t1,$t0
137	shl	\$`8*$n-4`,$t1
138	and	$b,$i1
139	 movq	(%rsp,$i0,8),$Tx
140	shr	\$`64-(8*$n-4)`,$t0
141	xor	$t1,$lo
142	 pslldq	\$$n,$Tx
143	 mov	$mask,$i0
144	shr	\$4,$b
145	xor	$t0,$hi
146	 and	$b,$i0
147	 shr	\$4,$b
148	 pxor	$Tx,$R
149___
150    }
151$code.=<<___;
152	mov	(%rsp,$i1,8),$t1
153	mov	$t1,$t0
154	shl	\$`8*$n-4`,$t1
155	movq	$R,$i0
156	shr	\$`64-(8*$n-4)`,$t0
157	xor	$t1,$lo
158	psrldq	\$8,$R
159	xor	$t0,$hi
160	movq	$R,$i1
161	xor	$i0,$lo
162	xor	$i1,$hi
163
164	add	\$128+8,%rsp
165.cfi_adjust_cfa_offset	-128-8
166	ret
167.Lend_mul_1x1:
168.cfi_endproc
169.size	_mul_1x1,.-_mul_1x1
170___
171
172($rp,$a1,$a0,$b1,$b0) = $win64?	("%rcx","%rdx","%r8", "%r9","%r10") :	# Win64 order
173				("%rdi","%rsi","%rdx","%rcx","%r8");	# Unix order
174
175$code.=<<___;
176.extern	OPENSSL_ia32cap_P
177.globl	bn_GF2m_mul_2x2
178.type	bn_GF2m_mul_2x2,\@abi-omnipotent
179.align	16
180bn_GF2m_mul_2x2:
181.cfi_startproc
182	mov	%rsp,%rax
183	mov	OPENSSL_ia32cap_P(%rip),%r10
184	bt	\$33,%r10
185	jnc	.Lvanilla_mul_2x2
186
187	movq		$a1,%xmm0
188	movq		$b1,%xmm1
189	movq		$a0,%xmm2
190___
191$code.=<<___ if ($win64);
192	movq		40(%rsp),%xmm3
193___
194$code.=<<___ if (!$win64);
195	movq		$b0,%xmm3
196___
197$code.=<<___;
198	movdqa		%xmm0,%xmm4
199	movdqa		%xmm1,%xmm5
200	pclmulqdq	\$0,%xmm1,%xmm0	# a1·b1
201	pxor		%xmm2,%xmm4
202	pxor		%xmm3,%xmm5
203	pclmulqdq	\$0,%xmm3,%xmm2	# a0·b0
204	pclmulqdq	\$0,%xmm5,%xmm4	# (a0+a1)·(b0+b1)
205	xorps		%xmm0,%xmm4
206	xorps		%xmm2,%xmm4	# (a0+a1)·(b0+b1)-a0·b0-a1·b1
207	movdqa		%xmm4,%xmm5
208	pslldq		\$8,%xmm4
209	psrldq		\$8,%xmm5
210	pxor		%xmm4,%xmm2
211	pxor		%xmm5,%xmm0
212	movdqu		%xmm2,0($rp)
213	movdqu		%xmm0,16($rp)
214	ret
215
216.align	16
217.Lvanilla_mul_2x2:
218	lea	-8*17(%rsp),%rsp
219.cfi_adjust_cfa_offset	8*17
220___
221$code.=<<___ if ($win64);
222	mov	`8*17+40`(%rsp),$b0
223	mov	%rdi,8*15(%rsp)
224	mov	%rsi,8*16(%rsp)
225___
226$code.=<<___;
227	mov	%r14,8*10(%rsp)
228.cfi_rel_offset	%r14,8*10
229	mov	%r13,8*11(%rsp)
230.cfi_rel_offset	%r13,8*11
231	mov	%r12,8*12(%rsp)
232.cfi_rel_offset	%r12,8*12
233	mov	%rbp,8*13(%rsp)
234.cfi_rel_offset	%rbp,8*13
235	mov	%rbx,8*14(%rsp)
236.cfi_rel_offset	%rbx,8*14
237.Lbody_mul_2x2:
238	mov	$rp,32(%rsp)		# save the arguments
239	mov	$a1,40(%rsp)
240	mov	$a0,48(%rsp)
241	mov	$b1,56(%rsp)
242	mov	$b0,64(%rsp)
243
244	mov	\$0xf,$mask
245	mov	$a1,$a
246	mov	$b1,$b
247	call	_mul_1x1		# a1·b1
248	mov	$lo,16(%rsp)
249	mov	$hi,24(%rsp)
250
251	mov	48(%rsp),$a
252	mov	64(%rsp),$b
253	call	_mul_1x1		# a0·b0
254	mov	$lo,0(%rsp)
255	mov	$hi,8(%rsp)
256
257	mov	40(%rsp),$a
258	mov	56(%rsp),$b
259	xor	48(%rsp),$a
260	xor	64(%rsp),$b
261	call	_mul_1x1		# (a0+a1)·(b0+b1)
262___
263	@r=("%rbx","%rcx","%rdi","%rsi");
264$code.=<<___;
265	mov	0(%rsp),@r[0]
266	mov	8(%rsp),@r[1]
267	mov	16(%rsp),@r[2]
268	mov	24(%rsp),@r[3]
269	mov	32(%rsp),%rbp
270
271	xor	$hi,$lo
272	xor	@r[1],$hi
273	xor	@r[0],$lo
274	mov	@r[0],0(%rbp)
275	xor	@r[2],$hi
276	mov	@r[3],24(%rbp)
277	xor	@r[3],$lo
278	xor	@r[3],$hi
279	xor	$hi,$lo
280	mov	$hi,16(%rbp)
281	mov	$lo,8(%rbp)
282
283	mov	8*10(%rsp),%r14
284.cfi_restore	%r14
285	mov	8*11(%rsp),%r13
286.cfi_restore	%r13
287	mov	8*12(%rsp),%r12
288.cfi_restore	%r12
289	mov	8*13(%rsp),%rbp
290.cfi_restore	%rbp
291	mov	8*14(%rsp),%rbx
292.cfi_restore	%rbx
293___
294$code.=<<___ if ($win64);
295	mov	8*15(%rsp),%rdi
296	mov	8*16(%rsp),%rsi
297___
298$code.=<<___;
299	lea	8*17(%rsp),%rsp
300.cfi_adjust_cfa_offset	-8*17
301.Lepilogue_mul_2x2:
302	ret
303.Lend_mul_2x2:
304.cfi_endproc
305.size	bn_GF2m_mul_2x2,.-bn_GF2m_mul_2x2
306.asciz	"GF(2^m) Multiplication for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
307.align	16
308___
309
310# EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame,
311#               CONTEXT *context,DISPATCHER_CONTEXT *disp)
312if ($win64) {
313$rec="%rcx";
314$frame="%rdx";
315$context="%r8";
316$disp="%r9";
317
318$code.=<<___;
319.extern __imp_RtlVirtualUnwind
320
321.type	se_handler,\@abi-omnipotent
322.align	16
323se_handler:
324	push	%rsi
325	push	%rdi
326	push	%rbx
327	push	%rbp
328	push	%r12
329	push	%r13
330	push	%r14
331	push	%r15
332	pushfq
333	sub	\$64,%rsp
334
335	mov	120($context),%rax	# pull context->Rax
336	mov	248($context),%rbx	# pull context->Rip
337
338	lea	.Lbody_mul_2x2(%rip),%r10
339	cmp	%r10,%rbx		# context->Rip<"prologue" label
340	jb	.Lin_prologue
341
342	mov	152($context),%rax	# pull context->Rsp
343
344	lea	.Lepilogue_mul_2x2(%rip),%r10
345	cmp	%r10,%rbx		# context->Rip>="epilogue" label
346	jae	.Lin_prologue
347
348	mov	8*10(%rax),%r14		# mimic epilogue
349	mov	8*11(%rax),%r13
350	mov	8*12(%rax),%r12
351	mov	8*13(%rax),%rbp
352	mov	8*14(%rax),%rbx
353	mov	8*15(%rax),%rdi
354	mov	8*16(%rax),%rsi
355
356	mov	%rbx,144($context)	# restore context->Rbx
357	mov	%rbp,160($context)	# restore context->Rbp
358	mov	%rsi,168($context)	# restore context->Rsi
359	mov	%rdi,176($context)	# restore context->Rdi
360	mov	%r12,216($context)	# restore context->R12
361	mov	%r13,224($context)	# restore context->R13
362	mov	%r14,232($context)	# restore context->R14
363
364	lea	8*17(%rax),%rax
365
366.Lin_prologue:
367	mov	%rax,152($context)	# restore context->Rsp
368
369	mov	40($disp),%rdi		# disp->ContextRecord
370	mov	$context,%rsi		# context
371	mov	\$154,%ecx		# sizeof(CONTEXT)
372	.long	0xa548f3fc		# cld; rep movsq
373
374	mov	$disp,%rsi
375	xor	%rcx,%rcx		# arg1, UNW_FLAG_NHANDLER
376	mov	8(%rsi),%rdx		# arg2, disp->ImageBase
377	mov	0(%rsi),%r8		# arg3, disp->ControlPc
378	mov	16(%rsi),%r9		# arg4, disp->FunctionEntry
379	mov	40(%rsi),%r10		# disp->ContextRecord
380	lea	56(%rsi),%r11		# &disp->HandlerData
381	lea	24(%rsi),%r12		# &disp->EstablisherFrame
382	mov	%r10,32(%rsp)		# arg5
383	mov	%r11,40(%rsp)		# arg6
384	mov	%r12,48(%rsp)		# arg7
385	mov	%rcx,56(%rsp)		# arg8, (NULL)
386	call	*__imp_RtlVirtualUnwind(%rip)
387
388	mov	\$1,%eax		# ExceptionContinueSearch
389	add	\$64,%rsp
390	popfq
391	pop	%r15
392	pop	%r14
393	pop	%r13
394	pop	%r12
395	pop	%rbp
396	pop	%rbx
397	pop	%rdi
398	pop	%rsi
399	ret
400.size	se_handler,.-se_handler
401
402.section	.pdata
403.align	4
404	.rva	_mul_1x1
405	.rva	.Lend_mul_1x1
406	.rva	.LSEH_info_1x1
407
408	.rva	.Lvanilla_mul_2x2
409	.rva	.Lend_mul_2x2
410	.rva	.LSEH_info_2x2
411.section	.xdata
412.align	8
413.LSEH_info_1x1:
414	.byte	0x01,0x07,0x02,0x00
415	.byte	0x07,0x01,0x11,0x00	# sub rsp,128+8
416.LSEH_info_2x2:
417	.byte	9,0,0,0
418	.rva	se_handler
419___
420}
421
422$code =~ s/\`([^\`]*)\`/eval($1)/gem;
423print $code;
424close STDOUT or die "error closing STDOUT: $!";
425