^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 1) #!/usr/bin/env perl
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 2) # SPDX-License-Identifier: GPL-2.0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 3)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 4) # This code is taken from the OpenSSL project but the author (Andy Polyakov)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 5) # has relicensed it under the GPLv2. Therefore this program is free software;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 6) # you can redistribute it and/or modify it under the terms of the GNU General
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 7) # Public License version 2 as published by the Free Software Foundation.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 8) #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 9) # The original headers, including the original license headers, are
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 10) # included below for completeness.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 11)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 12) # ====================================================================
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 13) # Written by Andy Polyakov <appro@openssl.org> for the OpenSSL
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 14) # project. The module is, however, dual licensed under OpenSSL and
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 15) # CRYPTOGAMS licenses depending on where you obtain it. For further
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 16) # details see https://www.openssl.org/~appro/cryptogams/.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 17) # ====================================================================
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 18)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 19) # SHA512 block procedure for ARMv4. September 2007.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 20)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 21) # This code is ~4.5 (four and a half) times faster than code generated
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 22) # by gcc 3.4 and it spends ~72 clock cycles per byte [on single-issue
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 23) # Xscale PXA250 core].
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 24) #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 25) # July 2010.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 26) #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 27) # Rescheduling for dual-issue pipeline resulted in 6% improvement on
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 28) # Cortex A8 core and ~40 cycles per processed byte.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 29)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 30) # February 2011.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 31) #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 32) # Profiler-assisted and platform-specific optimization resulted in 7%
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 33) # improvement on Coxtex A8 core and ~38 cycles per byte.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 34)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 35) # March 2011.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 36) #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 37) # Add NEON implementation. On Cortex A8 it was measured to process
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 38) # one byte in 23.3 cycles or ~60% faster than integer-only code.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 39)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 40) # August 2012.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 41) #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 42) # Improve NEON performance by 12% on Snapdragon S4. In absolute
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 43) # terms it's 22.6 cycles per byte, which is disappointing result.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 44) # Technical writers asserted that 3-way S4 pipeline can sustain
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 45) # multiple NEON instructions per cycle, but dual NEON issue could
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 46) # not be observed, see https://www.openssl.org/~appro/Snapdragon-S4.html
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 47) # for further details. On side note Cortex-A15 processes one byte in
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 48) # 16 cycles.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 49)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 50) # Byte order [in]dependence. =========================================
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 51) #
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 52) # Originally caller was expected to maintain specific *dword* order in
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 53) # h[0-7], namely with most significant dword at *lower* address, which
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 54) # was reflected in below two parameters as 0 and 4. Now caller is
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 55) # expected to maintain native byte order for whole 64-bit values.
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 56) $hi="HI";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 57) $lo="LO";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 58) # ====================================================================
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 59)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 60) while (($output=shift) && ($output!~/^\w[\w\-]*\.\w+$/)) {}
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 61) open STDOUT,">$output";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 62)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 63) $ctx="r0"; # parameter block
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 64) $inp="r1";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 65) $len="r2";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 66)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 67) $Tlo="r3";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 68) $Thi="r4";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 69) $Alo="r5";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 70) $Ahi="r6";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 71) $Elo="r7";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 72) $Ehi="r8";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 73) $t0="r9";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 74) $t1="r10";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 75) $t2="r11";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 76) $t3="r12";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 77) ############ r13 is stack pointer
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 78) $Ktbl="r14";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 79) ############ r15 is program counter
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 80)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 81) $Aoff=8*0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 82) $Boff=8*1;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 83) $Coff=8*2;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 84) $Doff=8*3;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 85) $Eoff=8*4;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 86) $Foff=8*5;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 87) $Goff=8*6;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 88) $Hoff=8*7;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 89) $Xoff=8*8;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 90)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 91) sub BODY_00_15() {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 92) my $magic = shift;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 93) $code.=<<___;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 94) @ Sigma1(x) (ROTR((x),14) ^ ROTR((x),18) ^ ROTR((x),41))
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 95) @ LO lo>>14^hi<<18 ^ lo>>18^hi<<14 ^ hi>>9^lo<<23
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 96) @ HI hi>>14^lo<<18 ^ hi>>18^lo<<14 ^ lo>>9^hi<<23
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 97) mov $t0,$Elo,lsr#14
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 98) str $Tlo,[sp,#$Xoff+0]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 99) mov $t1,$Ehi,lsr#14
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 100) str $Thi,[sp,#$Xoff+4]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 101) eor $t0,$t0,$Ehi,lsl#18
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 102) ldr $t2,[sp,#$Hoff+0] @ h.lo
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 103) eor $t1,$t1,$Elo,lsl#18
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 104) ldr $t3,[sp,#$Hoff+4] @ h.hi
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 105) eor $t0,$t0,$Elo,lsr#18
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 106) eor $t1,$t1,$Ehi,lsr#18
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 107) eor $t0,$t0,$Ehi,lsl#14
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 108) eor $t1,$t1,$Elo,lsl#14
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 109) eor $t0,$t0,$Ehi,lsr#9
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 110) eor $t1,$t1,$Elo,lsr#9
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 111) eor $t0,$t0,$Elo,lsl#23
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 112) eor $t1,$t1,$Ehi,lsl#23 @ Sigma1(e)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 113) adds $Tlo,$Tlo,$t0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 114) ldr $t0,[sp,#$Foff+0] @ f.lo
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 115) adc $Thi,$Thi,$t1 @ T += Sigma1(e)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 116) ldr $t1,[sp,#$Foff+4] @ f.hi
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 117) adds $Tlo,$Tlo,$t2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 118) ldr $t2,[sp,#$Goff+0] @ g.lo
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 119) adc $Thi,$Thi,$t3 @ T += h
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 120) ldr $t3,[sp,#$Goff+4] @ g.hi
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 121)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 122) eor $t0,$t0,$t2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 123) str $Elo,[sp,#$Eoff+0]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 124) eor $t1,$t1,$t3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 125) str $Ehi,[sp,#$Eoff+4]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 126) and $t0,$t0,$Elo
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 127) str $Alo,[sp,#$Aoff+0]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 128) and $t1,$t1,$Ehi
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 129) str $Ahi,[sp,#$Aoff+4]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 130) eor $t0,$t0,$t2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 131) ldr $t2,[$Ktbl,#$lo] @ K[i].lo
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 132) eor $t1,$t1,$t3 @ Ch(e,f,g)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 133) ldr $t3,[$Ktbl,#$hi] @ K[i].hi
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 134)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 135) adds $Tlo,$Tlo,$t0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 136) ldr $Elo,[sp,#$Doff+0] @ d.lo
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 137) adc $Thi,$Thi,$t1 @ T += Ch(e,f,g)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 138) ldr $Ehi,[sp,#$Doff+4] @ d.hi
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 139) adds $Tlo,$Tlo,$t2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 140) and $t0,$t2,#0xff
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 141) adc $Thi,$Thi,$t3 @ T += K[i]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 142) adds $Elo,$Elo,$Tlo
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 143) ldr $t2,[sp,#$Boff+0] @ b.lo
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 144) adc $Ehi,$Ehi,$Thi @ d += T
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 145) teq $t0,#$magic
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 146)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 147) ldr $t3,[sp,#$Coff+0] @ c.lo
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 148) #if __ARM_ARCH__>=7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 149) it eq @ Thumb2 thing, sanity check in ARM
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 150) #endif
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 151) orreq $Ktbl,$Ktbl,#1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 152) @ Sigma0(x) (ROTR((x),28) ^ ROTR((x),34) ^ ROTR((x),39))
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 153) @ LO lo>>28^hi<<4 ^ hi>>2^lo<<30 ^ hi>>7^lo<<25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 154) @ HI hi>>28^lo<<4 ^ lo>>2^hi<<30 ^ lo>>7^hi<<25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 155) mov $t0,$Alo,lsr#28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 156) mov $t1,$Ahi,lsr#28
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 157) eor $t0,$t0,$Ahi,lsl#4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 158) eor $t1,$t1,$Alo,lsl#4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 159) eor $t0,$t0,$Ahi,lsr#2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 160) eor $t1,$t1,$Alo,lsr#2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 161) eor $t0,$t0,$Alo,lsl#30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 162) eor $t1,$t1,$Ahi,lsl#30
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 163) eor $t0,$t0,$Ahi,lsr#7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 164) eor $t1,$t1,$Alo,lsr#7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 165) eor $t0,$t0,$Alo,lsl#25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 166) eor $t1,$t1,$Ahi,lsl#25 @ Sigma0(a)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 167) adds $Tlo,$Tlo,$t0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 168) and $t0,$Alo,$t2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 169) adc $Thi,$Thi,$t1 @ T += Sigma0(a)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 170)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 171) ldr $t1,[sp,#$Boff+4] @ b.hi
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 172) orr $Alo,$Alo,$t2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 173) ldr $t2,[sp,#$Coff+4] @ c.hi
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 174) and $Alo,$Alo,$t3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 175) and $t3,$Ahi,$t1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 176) orr $Ahi,$Ahi,$t1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 177) orr $Alo,$Alo,$t0 @ Maj(a,b,c).lo
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 178) and $Ahi,$Ahi,$t2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 179) adds $Alo,$Alo,$Tlo
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 180) orr $Ahi,$Ahi,$t3 @ Maj(a,b,c).hi
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 181) sub sp,sp,#8
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 182) adc $Ahi,$Ahi,$Thi @ h += T
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 183) tst $Ktbl,#1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 184) add $Ktbl,$Ktbl,#8
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 185) ___
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 186) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 187) $code=<<___;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 188) #ifndef __KERNEL__
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 189) # include "arm_arch.h"
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 190) # define VFP_ABI_PUSH vstmdb sp!,{d8-d15}
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 191) # define VFP_ABI_POP vldmia sp!,{d8-d15}
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 192) #else
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 193) # define __ARM_ARCH__ __LINUX_ARM_ARCH__
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 194) # define __ARM_MAX_ARCH__ 7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 195) # define VFP_ABI_PUSH
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 196) # define VFP_ABI_POP
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 197) #endif
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 198)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 199) #ifdef __ARMEL__
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 200) # define LO 0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 201) # define HI 4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 202) # define WORD64(hi0,lo0,hi1,lo1) .word lo0,hi0, lo1,hi1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 203) #else
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 204) # define HI 0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 205) # define LO 4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 206) # define WORD64(hi0,lo0,hi1,lo1) .word hi0,lo0, hi1,lo1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 207) #endif
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 208)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 209) .text
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 210) #if __ARM_ARCH__<7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 211) .code 32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 212) #else
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 213) .syntax unified
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 214) # ifdef __thumb2__
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 215) .thumb
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 216) # else
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 217) .code 32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 218) # endif
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 219) #endif
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 220)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 221) .type K512,%object
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 222) .align 5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 223) K512:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 224) WORD64(0x428a2f98,0xd728ae22, 0x71374491,0x23ef65cd)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 225) WORD64(0xb5c0fbcf,0xec4d3b2f, 0xe9b5dba5,0x8189dbbc)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 226) WORD64(0x3956c25b,0xf348b538, 0x59f111f1,0xb605d019)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 227) WORD64(0x923f82a4,0xaf194f9b, 0xab1c5ed5,0xda6d8118)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 228) WORD64(0xd807aa98,0xa3030242, 0x12835b01,0x45706fbe)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 229) WORD64(0x243185be,0x4ee4b28c, 0x550c7dc3,0xd5ffb4e2)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 230) WORD64(0x72be5d74,0xf27b896f, 0x80deb1fe,0x3b1696b1)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 231) WORD64(0x9bdc06a7,0x25c71235, 0xc19bf174,0xcf692694)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 232) WORD64(0xe49b69c1,0x9ef14ad2, 0xefbe4786,0x384f25e3)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 233) WORD64(0x0fc19dc6,0x8b8cd5b5, 0x240ca1cc,0x77ac9c65)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 234) WORD64(0x2de92c6f,0x592b0275, 0x4a7484aa,0x6ea6e483)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 235) WORD64(0x5cb0a9dc,0xbd41fbd4, 0x76f988da,0x831153b5)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 236) WORD64(0x983e5152,0xee66dfab, 0xa831c66d,0x2db43210)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 237) WORD64(0xb00327c8,0x98fb213f, 0xbf597fc7,0xbeef0ee4)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 238) WORD64(0xc6e00bf3,0x3da88fc2, 0xd5a79147,0x930aa725)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 239) WORD64(0x06ca6351,0xe003826f, 0x14292967,0x0a0e6e70)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 240) WORD64(0x27b70a85,0x46d22ffc, 0x2e1b2138,0x5c26c926)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 241) WORD64(0x4d2c6dfc,0x5ac42aed, 0x53380d13,0x9d95b3df)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 242) WORD64(0x650a7354,0x8baf63de, 0x766a0abb,0x3c77b2a8)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 243) WORD64(0x81c2c92e,0x47edaee6, 0x92722c85,0x1482353b)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 244) WORD64(0xa2bfe8a1,0x4cf10364, 0xa81a664b,0xbc423001)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 245) WORD64(0xc24b8b70,0xd0f89791, 0xc76c51a3,0x0654be30)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 246) WORD64(0xd192e819,0xd6ef5218, 0xd6990624,0x5565a910)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 247) WORD64(0xf40e3585,0x5771202a, 0x106aa070,0x32bbd1b8)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 248) WORD64(0x19a4c116,0xb8d2d0c8, 0x1e376c08,0x5141ab53)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 249) WORD64(0x2748774c,0xdf8eeb99, 0x34b0bcb5,0xe19b48a8)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 250) WORD64(0x391c0cb3,0xc5c95a63, 0x4ed8aa4a,0xe3418acb)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 251) WORD64(0x5b9cca4f,0x7763e373, 0x682e6ff3,0xd6b2b8a3)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 252) WORD64(0x748f82ee,0x5defb2fc, 0x78a5636f,0x43172f60)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 253) WORD64(0x84c87814,0xa1f0ab72, 0x8cc70208,0x1a6439ec)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 254) WORD64(0x90befffa,0x23631e28, 0xa4506ceb,0xde82bde9)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 255) WORD64(0xbef9a3f7,0xb2c67915, 0xc67178f2,0xe372532b)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 256) WORD64(0xca273ece,0xea26619c, 0xd186b8c7,0x21c0c207)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 257) WORD64(0xeada7dd6,0xcde0eb1e, 0xf57d4f7f,0xee6ed178)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 258) WORD64(0x06f067aa,0x72176fba, 0x0a637dc5,0xa2c898a6)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 259) WORD64(0x113f9804,0xbef90dae, 0x1b710b35,0x131c471b)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 260) WORD64(0x28db77f5,0x23047d84, 0x32caab7b,0x40c72493)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 261) WORD64(0x3c9ebe0a,0x15c9bebc, 0x431d67c4,0x9c100d4c)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 262) WORD64(0x4cc5d4be,0xcb3e42b6, 0x597f299c,0xfc657e2a)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 263) WORD64(0x5fcb6fab,0x3ad6faec, 0x6c44198c,0x4a475817)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 264) .size K512,.-K512
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 265) #if __ARM_MAX_ARCH__>=7 && !defined(__KERNEL__)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 266) .LOPENSSL_armcap:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 267) .word OPENSSL_armcap_P-sha512_block_data_order
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 268) .skip 32-4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 269) #else
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 270) .skip 32
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 271) #endif
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 272)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 273) .global sha512_block_data_order
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 274) .type sha512_block_data_order,%function
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 275) sha512_block_data_order:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 276) .Lsha512_block_data_order:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 277) #if __ARM_ARCH__<7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 278) sub r3,pc,#8 @ sha512_block_data_order
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 279) #else
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 280) adr r3,.Lsha512_block_data_order
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 281) #endif
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 282) #if __ARM_MAX_ARCH__>=7 && !defined(__KERNEL__)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 283) ldr r12,.LOPENSSL_armcap
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 284) ldr r12,[r3,r12] @ OPENSSL_armcap_P
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 285) tst r12,#1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 286) bne .LNEON
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 287) #endif
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 288) add $len,$inp,$len,lsl#7 @ len to point at the end of inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 289) stmdb sp!,{r4-r12,lr}
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 290) sub $Ktbl,r3,#672 @ K512
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 291) sub sp,sp,#9*8
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 292)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 293) ldr $Elo,[$ctx,#$Eoff+$lo]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 294) ldr $Ehi,[$ctx,#$Eoff+$hi]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 295) ldr $t0, [$ctx,#$Goff+$lo]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 296) ldr $t1, [$ctx,#$Goff+$hi]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 297) ldr $t2, [$ctx,#$Hoff+$lo]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 298) ldr $t3, [$ctx,#$Hoff+$hi]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 299) .Loop:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 300) str $t0, [sp,#$Goff+0]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 301) str $t1, [sp,#$Goff+4]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 302) str $t2, [sp,#$Hoff+0]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 303) str $t3, [sp,#$Hoff+4]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 304) ldr $Alo,[$ctx,#$Aoff+$lo]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 305) ldr $Ahi,[$ctx,#$Aoff+$hi]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 306) ldr $Tlo,[$ctx,#$Boff+$lo]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 307) ldr $Thi,[$ctx,#$Boff+$hi]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 308) ldr $t0, [$ctx,#$Coff+$lo]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 309) ldr $t1, [$ctx,#$Coff+$hi]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 310) ldr $t2, [$ctx,#$Doff+$lo]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 311) ldr $t3, [$ctx,#$Doff+$hi]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 312) str $Tlo,[sp,#$Boff+0]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 313) str $Thi,[sp,#$Boff+4]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 314) str $t0, [sp,#$Coff+0]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 315) str $t1, [sp,#$Coff+4]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 316) str $t2, [sp,#$Doff+0]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 317) str $t3, [sp,#$Doff+4]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 318) ldr $Tlo,[$ctx,#$Foff+$lo]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 319) ldr $Thi,[$ctx,#$Foff+$hi]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 320) str $Tlo,[sp,#$Foff+0]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 321) str $Thi,[sp,#$Foff+4]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 322)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 323) .L00_15:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 324) #if __ARM_ARCH__<7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 325) ldrb $Tlo,[$inp,#7]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 326) ldrb $t0, [$inp,#6]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 327) ldrb $t1, [$inp,#5]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 328) ldrb $t2, [$inp,#4]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 329) ldrb $Thi,[$inp,#3]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 330) ldrb $t3, [$inp,#2]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 331) orr $Tlo,$Tlo,$t0,lsl#8
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 332) ldrb $t0, [$inp,#1]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 333) orr $Tlo,$Tlo,$t1,lsl#16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 334) ldrb $t1, [$inp],#8
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 335) orr $Tlo,$Tlo,$t2,lsl#24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 336) orr $Thi,$Thi,$t3,lsl#8
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 337) orr $Thi,$Thi,$t0,lsl#16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 338) orr $Thi,$Thi,$t1,lsl#24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 339) #else
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 340) ldr $Tlo,[$inp,#4]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 341) ldr $Thi,[$inp],#8
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 342) #ifdef __ARMEL__
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 343) rev $Tlo,$Tlo
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 344) rev $Thi,$Thi
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 345) #endif
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 346) #endif
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 347) ___
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 348) &BODY_00_15(0x94);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 349) $code.=<<___;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 350) tst $Ktbl,#1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 351) beq .L00_15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 352) ldr $t0,[sp,#`$Xoff+8*(16-1)`+0]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 353) ldr $t1,[sp,#`$Xoff+8*(16-1)`+4]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 354) bic $Ktbl,$Ktbl,#1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 355) .L16_79:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 356) @ sigma0(x) (ROTR((x),1) ^ ROTR((x),8) ^ ((x)>>7))
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 357) @ LO lo>>1^hi<<31 ^ lo>>8^hi<<24 ^ lo>>7^hi<<25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 358) @ HI hi>>1^lo<<31 ^ hi>>8^lo<<24 ^ hi>>7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 359) mov $Tlo,$t0,lsr#1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 360) ldr $t2,[sp,#`$Xoff+8*(16-14)`+0]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 361) mov $Thi,$t1,lsr#1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 362) ldr $t3,[sp,#`$Xoff+8*(16-14)`+4]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 363) eor $Tlo,$Tlo,$t1,lsl#31
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 364) eor $Thi,$Thi,$t0,lsl#31
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 365) eor $Tlo,$Tlo,$t0,lsr#8
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 366) eor $Thi,$Thi,$t1,lsr#8
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 367) eor $Tlo,$Tlo,$t1,lsl#24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 368) eor $Thi,$Thi,$t0,lsl#24
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 369) eor $Tlo,$Tlo,$t0,lsr#7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 370) eor $Thi,$Thi,$t1,lsr#7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 371) eor $Tlo,$Tlo,$t1,lsl#25
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 372)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 373) @ sigma1(x) (ROTR((x),19) ^ ROTR((x),61) ^ ((x)>>6))
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 374) @ LO lo>>19^hi<<13 ^ hi>>29^lo<<3 ^ lo>>6^hi<<26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 375) @ HI hi>>19^lo<<13 ^ lo>>29^hi<<3 ^ hi>>6
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 376) mov $t0,$t2,lsr#19
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 377) mov $t1,$t3,lsr#19
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 378) eor $t0,$t0,$t3,lsl#13
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 379) eor $t1,$t1,$t2,lsl#13
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 380) eor $t0,$t0,$t3,lsr#29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 381) eor $t1,$t1,$t2,lsr#29
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 382) eor $t0,$t0,$t2,lsl#3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 383) eor $t1,$t1,$t3,lsl#3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 384) eor $t0,$t0,$t2,lsr#6
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 385) eor $t1,$t1,$t3,lsr#6
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 386) ldr $t2,[sp,#`$Xoff+8*(16-9)`+0]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 387) eor $t0,$t0,$t3,lsl#26
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 388)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 389) ldr $t3,[sp,#`$Xoff+8*(16-9)`+4]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 390) adds $Tlo,$Tlo,$t0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 391) ldr $t0,[sp,#`$Xoff+8*16`+0]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 392) adc $Thi,$Thi,$t1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 393)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 394) ldr $t1,[sp,#`$Xoff+8*16`+4]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 395) adds $Tlo,$Tlo,$t2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 396) adc $Thi,$Thi,$t3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 397) adds $Tlo,$Tlo,$t0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 398) adc $Thi,$Thi,$t1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 399) ___
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 400) &BODY_00_15(0x17);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 401) $code.=<<___;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 402) #if __ARM_ARCH__>=7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 403) ittt eq @ Thumb2 thing, sanity check in ARM
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 404) #endif
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 405) ldreq $t0,[sp,#`$Xoff+8*(16-1)`+0]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 406) ldreq $t1,[sp,#`$Xoff+8*(16-1)`+4]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 407) beq .L16_79
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 408) bic $Ktbl,$Ktbl,#1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 409)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 410) ldr $Tlo,[sp,#$Boff+0]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 411) ldr $Thi,[sp,#$Boff+4]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 412) ldr $t0, [$ctx,#$Aoff+$lo]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 413) ldr $t1, [$ctx,#$Aoff+$hi]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 414) ldr $t2, [$ctx,#$Boff+$lo]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 415) ldr $t3, [$ctx,#$Boff+$hi]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 416) adds $t0,$Alo,$t0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 417) str $t0, [$ctx,#$Aoff+$lo]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 418) adc $t1,$Ahi,$t1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 419) str $t1, [$ctx,#$Aoff+$hi]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 420) adds $t2,$Tlo,$t2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 421) str $t2, [$ctx,#$Boff+$lo]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 422) adc $t3,$Thi,$t3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 423) str $t3, [$ctx,#$Boff+$hi]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 424)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 425) ldr $Alo,[sp,#$Coff+0]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 426) ldr $Ahi,[sp,#$Coff+4]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 427) ldr $Tlo,[sp,#$Doff+0]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 428) ldr $Thi,[sp,#$Doff+4]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 429) ldr $t0, [$ctx,#$Coff+$lo]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 430) ldr $t1, [$ctx,#$Coff+$hi]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 431) ldr $t2, [$ctx,#$Doff+$lo]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 432) ldr $t3, [$ctx,#$Doff+$hi]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 433) adds $t0,$Alo,$t0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 434) str $t0, [$ctx,#$Coff+$lo]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 435) adc $t1,$Ahi,$t1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 436) str $t1, [$ctx,#$Coff+$hi]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 437) adds $t2,$Tlo,$t2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 438) str $t2, [$ctx,#$Doff+$lo]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 439) adc $t3,$Thi,$t3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 440) str $t3, [$ctx,#$Doff+$hi]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 441)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 442) ldr $Tlo,[sp,#$Foff+0]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 443) ldr $Thi,[sp,#$Foff+4]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 444) ldr $t0, [$ctx,#$Eoff+$lo]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 445) ldr $t1, [$ctx,#$Eoff+$hi]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 446) ldr $t2, [$ctx,#$Foff+$lo]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 447) ldr $t3, [$ctx,#$Foff+$hi]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 448) adds $Elo,$Elo,$t0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 449) str $Elo,[$ctx,#$Eoff+$lo]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 450) adc $Ehi,$Ehi,$t1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 451) str $Ehi,[$ctx,#$Eoff+$hi]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 452) adds $t2,$Tlo,$t2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 453) str $t2, [$ctx,#$Foff+$lo]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 454) adc $t3,$Thi,$t3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 455) str $t3, [$ctx,#$Foff+$hi]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 456)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 457) ldr $Alo,[sp,#$Goff+0]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 458) ldr $Ahi,[sp,#$Goff+4]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 459) ldr $Tlo,[sp,#$Hoff+0]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 460) ldr $Thi,[sp,#$Hoff+4]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 461) ldr $t0, [$ctx,#$Goff+$lo]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 462) ldr $t1, [$ctx,#$Goff+$hi]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 463) ldr $t2, [$ctx,#$Hoff+$lo]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 464) ldr $t3, [$ctx,#$Hoff+$hi]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 465) adds $t0,$Alo,$t0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 466) str $t0, [$ctx,#$Goff+$lo]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 467) adc $t1,$Ahi,$t1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 468) str $t1, [$ctx,#$Goff+$hi]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 469) adds $t2,$Tlo,$t2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 470) str $t2, [$ctx,#$Hoff+$lo]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 471) adc $t3,$Thi,$t3
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 472) str $t3, [$ctx,#$Hoff+$hi]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 473)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 474) add sp,sp,#640
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 475) sub $Ktbl,$Ktbl,#640
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 476)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 477) teq $inp,$len
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 478) bne .Loop
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 479)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 480) add sp,sp,#8*9 @ destroy frame
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 481) #if __ARM_ARCH__>=5
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 482) ldmia sp!,{r4-r12,pc}
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 483) #else
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 484) ldmia sp!,{r4-r12,lr}
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 485) tst lr,#1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 486) moveq pc,lr @ be binary compatible with V4, yet
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 487) bx lr @ interoperable with Thumb ISA:-)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 488) #endif
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 489) .size sha512_block_data_order,.-sha512_block_data_order
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 490) ___
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 491)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 492) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 493) my @Sigma0=(28,34,39);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 494) my @Sigma1=(14,18,41);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 495) my @sigma0=(1, 8, 7);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 496) my @sigma1=(19,61,6);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 497)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 498) my $Ktbl="r3";
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 499) my $cnt="r12"; # volatile register known as ip, intra-procedure-call scratch
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 500)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 501) my @X=map("d$_",(0..15));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 502) my @V=($A,$B,$C,$D,$E,$F,$G,$H)=map("d$_",(16..23));
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 503)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 504) sub NEON_00_15() {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 505) my $i=shift;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 506) my ($a,$b,$c,$d,$e,$f,$g,$h)=@_;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 507) my ($t0,$t1,$t2,$T1,$K,$Ch,$Maj)=map("d$_",(24..31)); # temps
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 508)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 509) $code.=<<___ if ($i<16 || $i&1);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 510) vshr.u64 $t0,$e,#@Sigma1[0] @ $i
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 511) #if $i<16
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 512) vld1.64 {@X[$i%16]},[$inp]! @ handles unaligned
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 513) #endif
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 514) vshr.u64 $t1,$e,#@Sigma1[1]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 515) #if $i>0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 516) vadd.i64 $a,$Maj @ h+=Maj from the past
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 517) #endif
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 518) vshr.u64 $t2,$e,#@Sigma1[2]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 519) ___
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 520) $code.=<<___;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 521) vld1.64 {$K},[$Ktbl,:64]! @ K[i++]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 522) vsli.64 $t0,$e,#`64-@Sigma1[0]`
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 523) vsli.64 $t1,$e,#`64-@Sigma1[1]`
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 524) vmov $Ch,$e
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 525) vsli.64 $t2,$e,#`64-@Sigma1[2]`
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 526) #if $i<16 && defined(__ARMEL__)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 527) vrev64.8 @X[$i],@X[$i]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 528) #endif
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 529) veor $t1,$t0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 530) vbsl $Ch,$f,$g @ Ch(e,f,g)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 531) vshr.u64 $t0,$a,#@Sigma0[0]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 532) veor $t2,$t1 @ Sigma1(e)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 533) vadd.i64 $T1,$Ch,$h
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 534) vshr.u64 $t1,$a,#@Sigma0[1]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 535) vsli.64 $t0,$a,#`64-@Sigma0[0]`
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 536) vadd.i64 $T1,$t2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 537) vshr.u64 $t2,$a,#@Sigma0[2]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 538) vadd.i64 $K,@X[$i%16]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 539) vsli.64 $t1,$a,#`64-@Sigma0[1]`
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 540) veor $Maj,$a,$b
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 541) vsli.64 $t2,$a,#`64-@Sigma0[2]`
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 542) veor $h,$t0,$t1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 543) vadd.i64 $T1,$K
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 544) vbsl $Maj,$c,$b @ Maj(a,b,c)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 545) veor $h,$t2 @ Sigma0(a)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 546) vadd.i64 $d,$T1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 547) vadd.i64 $Maj,$T1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 548) @ vadd.i64 $h,$Maj
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 549) ___
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 550) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 551)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 552) sub NEON_16_79() {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 553) my $i=shift;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 554)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 555) if ($i&1) { &NEON_00_15($i,@_); return; }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 556)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 557) # 2x-vectorized, therefore runs every 2nd round
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 558) my @X=map("q$_",(0..7)); # view @X as 128-bit vector
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 559) my ($t0,$t1,$s0,$s1) = map("q$_",(12..15)); # temps
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 560) my ($d0,$d1,$d2) = map("d$_",(24..26)); # temps from NEON_00_15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 561) my $e=@_[4]; # $e from NEON_00_15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 562) $i /= 2;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 563) $code.=<<___;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 564) vshr.u64 $t0,@X[($i+7)%8],#@sigma1[0]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 565) vshr.u64 $t1,@X[($i+7)%8],#@sigma1[1]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 566) vadd.i64 @_[0],d30 @ h+=Maj from the past
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 567) vshr.u64 $s1,@X[($i+7)%8],#@sigma1[2]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 568) vsli.64 $t0,@X[($i+7)%8],#`64-@sigma1[0]`
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 569) vext.8 $s0,@X[$i%8],@X[($i+1)%8],#8 @ X[i+1]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 570) vsli.64 $t1,@X[($i+7)%8],#`64-@sigma1[1]`
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 571) veor $s1,$t0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 572) vshr.u64 $t0,$s0,#@sigma0[0]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 573) veor $s1,$t1 @ sigma1(X[i+14])
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 574) vshr.u64 $t1,$s0,#@sigma0[1]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 575) vadd.i64 @X[$i%8],$s1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 576) vshr.u64 $s1,$s0,#@sigma0[2]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 577) vsli.64 $t0,$s0,#`64-@sigma0[0]`
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 578) vsli.64 $t1,$s0,#`64-@sigma0[1]`
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 579) vext.8 $s0,@X[($i+4)%8],@X[($i+5)%8],#8 @ X[i+9]
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 580) veor $s1,$t0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 581) vshr.u64 $d0,$e,#@Sigma1[0] @ from NEON_00_15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 582) vadd.i64 @X[$i%8],$s0
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 583) vshr.u64 $d1,$e,#@Sigma1[1] @ from NEON_00_15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 584) veor $s1,$t1 @ sigma0(X[i+1])
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 585) vshr.u64 $d2,$e,#@Sigma1[2] @ from NEON_00_15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 586) vadd.i64 @X[$i%8],$s1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 587) ___
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 588) &NEON_00_15(2*$i,@_);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 589) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 590)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 591) $code.=<<___;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 592) #if __ARM_MAX_ARCH__>=7
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 593) .arch armv7-a
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 594) .fpu neon
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 595)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 596) .global sha512_block_data_order_neon
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 597) .type sha512_block_data_order_neon,%function
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 598) .align 4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 599) sha512_block_data_order_neon:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 600) .LNEON:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 601) dmb @ errata #451034 on early Cortex A8
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 602) add $len,$inp,$len,lsl#7 @ len to point at the end of inp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 603) VFP_ABI_PUSH
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 604) adr $Ktbl,.Lsha512_block_data_order
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 605) sub $Ktbl,$Ktbl,.Lsha512_block_data_order-K512
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 606) vldmia $ctx,{$A-$H} @ load context
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 607) .Loop_neon:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 608) ___
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 609) for($i=0;$i<16;$i++) { &NEON_00_15($i,@V); unshift(@V,pop(@V)); }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 610) $code.=<<___;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 611) mov $cnt,#4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 612) .L16_79_neon:
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 613) subs $cnt,#1
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 614) ___
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 615) for(;$i<32;$i++) { &NEON_16_79($i,@V); unshift(@V,pop(@V)); }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 616) $code.=<<___;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 617) bne .L16_79_neon
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 618)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 619) vadd.i64 $A,d30 @ h+=Maj from the past
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 620) vldmia $ctx,{d24-d31} @ load context to temp
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 621) vadd.i64 q8,q12 @ vectorized accumulate
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 622) vadd.i64 q9,q13
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 623) vadd.i64 q10,q14
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 624) vadd.i64 q11,q15
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 625) vstmia $ctx,{$A-$H} @ save context
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 626) teq $inp,$len
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 627) sub $Ktbl,#640 @ rewind K512
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 628) bne .Loop_neon
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 629)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 630) VFP_ABI_POP
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 631) ret @ bx lr
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 632) .size sha512_block_data_order_neon,.-sha512_block_data_order_neon
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 633) #endif
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 634) ___
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 635) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 636) $code.=<<___;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 637) .asciz "SHA512 block transform for ARMv4/NEON, CRYPTOGAMS by <appro\@openssl.org>"
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 638) .align 2
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 639) #if __ARM_MAX_ARCH__>=7 && !defined(__KERNEL__)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 640) .comm OPENSSL_armcap_P,4,4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 641) #endif
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 642) ___
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 643)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 644) $code =~ s/\`([^\`]*)\`/eval $1/gem;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 645) $code =~ s/\bbx\s+lr\b/.word\t0xe12fff1e/gm; # make it possible to compile with -march=armv4
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 646) $code =~ s/\bret\b/bx lr/gm;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 647)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 648) open SELF,$0;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 649) while(<SELF>) {
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 650) next if (/^#!/);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 651) last if (!s/^#/@/ and !/^$/);
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 652) print;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 653) }
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 654) close SELF;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 655)
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 656) print $code;
^8f3ce5b39 (kx 2023-10-28 12:00:06 +0300 657) close STDOUT; # enforce flush
Orange Pi5 kernel
Deprecated Linux kernel 5.10.110 for OrangePi 5/5B/5+ boards
3 Commits
0 Branches
0 Tags