3 # ====================================================================
4 # Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
5 # project. The module is, however, dual licensed under OpenSSL and
6 # CRYPTOGAMS licenses depending on where you obtain it. For further
7 # details see http://www.openssl.org/~appro/cryptogams/.
8 # ====================================================================
12 # The module implements GCM GHASH function and underlying single
13 # multiplication operation in GF(2^128). Even though subroutines
14 # have _4bit suffix, they are not using any tables, but rely on
15 # hardware Galois Field Multiply support. Streamed GHASH processes
16 # byte in ~7 cycles, which is >6x faster than "4-bit" table-driven
17 # code compiled with TI's cl6x 6.0 with -mv6400+ -o2 flags. We are
18 # comparing apples vs. oranges, but compiler surely could have done
19 # better, because theoretical [though not necessarily achievable]
20 # estimate for "4-bit" table-driven implementation is ~12 cycles.
22 while (($output=shift) && ($output!~/\w[\w\-]*\.\w+$/)) {}
23 open STDOUT,">$output";
25 ($Xip,$Htable,$inp,$len)=("A4","B4","A6","B6"); # arguments
27 ($Z0,$Z1,$Z2,$Z3, $H0, $H1, $H2, $H3,
28 $H0x,$H1x,$H2x,$H3x)=map("A$_",(16..27));
29 ($H01u,$H01y,$H2u,$H3u, $H0y,$H1y,$H2y,$H3y,
30 $H0z,$H1z,$H2z,$H3z)=map("B$_",(16..27));
31 ($FF000000,$E10000)=("B30","B31");
32 ($xip,$x0,$x1,$xib)=map("B$_",(6..9)); # $xip zaps $len
34 ($rem,$res)=("B4","B5"); # $rem zaps $Htable
39 .if .ASSEMBLER_VERSION<7000000
43 .asg gcm_gmult_1bit,_gcm_gmult_1bit
44 .asg gcm_gmult_4bit,_gcm_gmult_4bit
45 .asg gcm_ghash_4bit,_gcm_ghash_4bit
51 .global _gcm_gmult_1bit
53 ADDAD $Htable,2,$Htable
55 .global _gcm_gmult_4bit
58 LDDW *${Htable}[-1],$H1:$H0 ; H.lo
59 LDDW *${Htable}[-2],$H3:$H2 ; H.hi
60 || MV $Xip,${xip} ; reassign Xi
61 || MVK 15,B1 ; SPLOOPD constant
64 || LDBU *++${xip}[15],$x1 ; Xi[15]
66 || LDBU *--${xip},$x0 ; Xi[14]
67 SHL $E10000,16,$E10000 ; [pre-shifted] reduction polynomial
68 SHL $FF000000,24,$FF000000 ; upper byte mask
70 || MVK 1,B0 ; take a single spin
72 PACKH2 $H0,$H1,$xia ; pack H0' and H1's upper bytes
73 AND $H2,$FF000000,$H2u ; H2's upper byte
74 AND $H3,$FF000000,$H3u ; H3's upper byte
82 .global _gcm_ghash_4bit
85 LDDW *${Htable}[-1],$H1:$H0 ; H.lo
86 || SHRU $len,4,B0 ; reassign len
87 LDDW *${Htable}[-2],$H3:$H2 ; H.hi
88 || MV $Xip,${xip} ; reassign Xi
89 || MVK 15,B1 ; SPLOOPD constant
92 || [B0] LDNDW *${inp}[1],$H1x:$H0x
94 || [B0] LDNDW *${inp}++[2],$H3x:$H2x
95 SHL $E10000,16,$E10000 ; [pre-shifted] reduction polynomial
96 || LDDW *${xip}[1],$Z1:$Z0
97 SHL $FF000000,24,$FF000000 ; upper byte mask
98 || LDDW *${xip}[0],$Z3:$Z2
100 PACKH2 $H0,$H1,$xia ; pack H0' and H1's upper bytes
101 AND $H2,$FF000000,$H2u ; H2's upper byte
102 AND $H3,$FF000000,$H3u ; H3's upper byte
107 || [B0] XOR $H0x,$Z0,$Z0 ; Xi^=inp
108 || [B0] XOR $H1x,$Z1,$Z1
110 [B0] XOR $H2x,$Z2,$Z2
111 || [B0] XOR $H3x,$Z3,$Z3
112 || [B0] SHRU $Z1,24,$xia ; Xi[15], avoid cross-path stall
113 STDW $Z1:$Z0,*${xip}[1]
114 || [B0] SHRU $Z1,16,$x0 ; Xi[14]
117 [B0] XOR $H2x,$Z2,$Z2
118 || [B0] XOR $H3x,$Z3,$Z3
119 || [B0] MV $Z0,$xia ; Xi[15], avoid cross-path stall
120 STDW $Z1:$Z0,*${xip}[1]
121 || [B0] SHRU $Z0,8,$x0 ; Xi[14]
124 STDW $Z3:$Z2,*${xip}[0]
134 || ADD $x1,$x1,$xib ; SHL $x1,1,$xib
138 ########____________________________
144 # 5 S2 S1x L1 D2 L2 |____________________________
145 # 6/0 L1 S1 L2 S2x |D2. M1 M2 |
146 # 7/1 L1 S1 D1x S2 M2 | M1 |
147 # 8/2 S1 L1x S2 | M1 M2 |
148 # 9/3 S1 L1x | D1. M1 M2 |
149 # 10/4 D1x | S1. L1 |
150 # 11/5 |S2 S1x L1 D2 L2 |____________
151 # 12/6/0 D1x __| L1 S1 L2 S2x |D2. ....
152 # 7/1 L1 S1 D1x S2 M2 | ....
153 # 8/2 S1 L1x S2 | ....
154 #####... ................|............
156 XORMPY $H0,$xia,$H0x ; 0 ; H·(Xi[i]<<1)
157 || XORMPY $H01u,$xib,$H01y
158 || [A0] LDBU *--${xip},$x0
159 XORMPY $H1,$xia,$H1x ; 1
160 XORMPY $H2,$xia,$H2x ; 2
161 || XORMPY $H2u,$xib,$H2y
162 XORMPY $H3,$xia,$H3x ; 3
163 || XORMPY $H3u,$xib,$H3y
164 ||[!A0] MVK.D 15,A0 ; *--${xip} counter
165 XOR.L $H0x,$Z0,$Z0 ; 4 ; Z^=H·(Xi[i]<<1)
166 || [A0] SUB.S A0,1,A0
167 XOR.L $H1x,$Z1,$Z1 ; 5
168 || AND.D $H01y,$FF000000,$H0z
169 || SWAP2.L $H01y,$H1y ; ; SHL $H01y,16,$H1y
173 XOR.L $H2x,$Z2,$Z2 ; 6/0 ; [0,0] in epilogue
174 || SHL $Z0,1,$rem ; ; rem=Z<<1
175 || SHRMB.S $Z1,$Z0,$Z0 ; ; Z>>=8
176 || AND.L $H1y,$FF000000,$H1z
177 XOR.L $H3x,$Z3,$Z3 ; 7/1
178 || SHRMB.S $Z2,$Z1,$Z1
179 || XOR.D $H0z,$Z0,$Z0 ; merge upper byte products
180 || AND.S $H2y,$FF000000,$H2z
181 || XORMPY $E10000,$rem,$res ; ; implicit rem&0x1FE
182 XOR.L $H1z,$Z1,$Z1 ; 8/2
183 || SHRMB.S $Z3,$Z2,$Z2
184 || AND.S $H3y,$FF000000,$H3z
185 XOR.L $H2z,$Z2,$Z2 ; 9/3
187 XOR.D $H3z,$Z3,$Z3 ; 10/4
191 || XOR.D $res,$Z3,$Z3 ; 12/6/0; Z^=res
193 ; input pre-fetch is possible where D1 slot is available...
194 [B0] LDNDW *${inp}[1],$H1x:$H0x ; 8/-
195 [B0] LDNDW *${inp}++[2],$H3x:$H2x ; 9/-
207 || [B0] BNOP ghash_loop?
208 [B0] XOR $H0x,$Z0,$Z0 ; Xi^=inp
209 || [B0] XOR $H1x,$Z1,$Z1
210 [B0] XOR $H2x,$Z2,$Z2
211 || [B0] XOR $H3x,$Z3,$Z3
212 || [B0] SHRU $Z1,24,$xia ; Xi[15], avoid cross-path stall
213 STDW $Z1:$Z0,*${xip}[1]
214 || [B0] SHRU $Z1,16,$x0 ; Xi[14]
218 [B0] BNOP ghash_loop? ; 12/-
219 [B0] XOR $H0x,$Z0,$Z0 ; Xi^=inp
220 || [B0] XOR $H1x,$Z1,$Z1
221 [B0] XOR $H2x,$Z2,$Z2
222 || [B0] XOR $H3x,$Z3,$Z3
223 || [B0] MV $Z0,$xia ; Xi[15], avoid cross-path stall
224 STDW $Z1:$Z0,*${xip}[1]
225 || [B0] SHRU $Z0,8,$x0 ; Xi[14]
228 STDW $Z3:$Z2,*${xip}[0]
235 .cstring "GHASH for C64x+, CRYPTOGAMS by <appro\@openssl.org>"