bsaes-armv7.pl: add bsaes_cbc_encrypt and bsaes_ctr32_encrypt_blocks.
authorAndy Polyakov <appro@openssl.org>
Tue, 23 Apr 2013 15:52:14 +0000 (17:52 +0200)
committerAndy Polyakov <appro@openssl.org>
Tue, 23 Apr 2013 15:52:14 +0000 (17:52 +0200)
Submitted by: Ard Biesheuvel <ard.biesheuvel@linaro.org>

Contributor claims ~50% improvement in CTR and ~9% in CBC decrypt
on Cortex-A15.

crypto/aes/asm/bsaes-armv7.pl
crypto/evp/e_aes.c

index 71aae8d..b350b32 100644 (file)
@@ -715,6 +715,8 @@ _bsaes_const:
        .quad   0x0304090e00050a0f, 0x01060b0c0207080d
 .LM0:
        .quad   0x02060a0e03070b0f, 0x0004080c0105090d
+.LREVM0SR:
+       .quad   0x090d02060c030708, 0x00040b0f050a0e01
 .asciz "Bit-sliced AES for NEON, CRYPTOGAMS by <appro\@openssl.org>"
 .align 6
 .size  _bsaes_const,.-_bsaes_const
@@ -727,6 +729,7 @@ _bsaes_encrypt8:
        sub     $const,$const,#_bsaes_encrypt8-.LM0SR
 
        vldmia  $const!, {@XMM[8]}              @ .LM0SR
+_bsaes_encrypt8_alt:
        veor    @XMM[10], @XMM[0], @XMM[9]      @ xor with round0 key
        veor    @XMM[11], @XMM[1], @XMM[9]
         vtbl.8 `&Dlo(@XMM[0])`, {@XMM[10]}, `&Dlo(@XMM[8])`
@@ -879,7 +882,7 @@ _bsaes_key_convert:
 ___
 }
 
-if (1) {               # following four functions are unsupported interface
+if (0) {               # following four functions are unsupported interface
                        # used for benchmarking...
 $code.=<<___;
 .globl bsaes_enc_key_convert
@@ -981,6 +984,432 @@ bsaes_decrypt_128:
 .size  bsaes_decrypt_128,.-bsaes_decrypt_128
 ___
 }
+{
+my ($inp,$out,$len,$key, $ivp,$fp,$rounds)=map("r$_",(0..3,8..10));
+my ($keysched)=("sp");
+
+$code.=<<___;
+.extern AES_cbc_encrypt
+.extern AES_decrypt
+
+.global        bsaes_cbc_encrypt
+.type  bsaes_cbc_encrypt,%function
+.align 5
+bsaes_cbc_encrypt:
+       cmp     $len, #128
+       blo     AES_cbc_encrypt
+
+       @ it is up to the caller to make sure we are called with enc == 0
+
+       stmdb   sp!, {r4-r10, lr}
+       vstmdb  sp!, {d8-d15}                   @ ABI specification says so
+       ldr     $ivp, [sp, #0x60]               @ IV is 1st arg on the stack
+       mov     $len, $len, lsr#4               @ len in 16 byte blocks
+       sub     sp, #0x10                       @ scratch space to carry over the IV
+       mov     $fp, sp                         @ save sp
+
+       @ allocate the key schedule on the stack
+       ldr     $rounds, [$key, #240]           @ get # of rounds
+       sub     sp, sp, $rounds, lsl#7          @ 128 bytes per inner round key
+       add     sp, sp, #`128-32`               @ size of bit-sliced key schedule
+
+       @ populate the key schedule
+       mov     r4, $key                        @ pass key
+       mov     r5, $rounds                     @ pass # of rounds
+       mov     r12, $keysched                  @ pass key schedule
+       bl      _bsaes_key_convert
+       vldmia  $keysched, {@XMM[6]}
+       vstmia  r12,  {@XMM[15]}                @ save last round key
+       veor    @XMM[7], @XMM[7], @XMM[6]       @ fix up round 0 key
+       vstmia  $keysched, {@XMM[7]}
+
+       vld1.8  {@XMM[15]}, [$ivp]              @ load IV
+       b       .Lcbc_dec_loop
+
+.align 4
+.Lcbc_dec_loop:
+       subs    $len, $len, #0x8
+       bmi     .Lcbc_dec_loop_finish
+
+       vld1.8  {@XMM[0]-@XMM[1]}, [$inp]!      @ load input
+       vld1.8  {@XMM[2]-@XMM[3]}, [$inp]!
+       mov     r4, $keysched                   @ pass the key
+       vld1.8  {@XMM[4]-@XMM[5]}, [$inp]!
+       mov     r5, $rounds
+       vld1.8  {@XMM[6]-@XMM[7]}, [$inp]
+       sub     $inp, $inp, #0x60
+       vstmia  $fp, {@XMM[15]}                 @ put aside IV
+
+       bl      _bsaes_decrypt8
+
+       vldmia  $fp, {@XMM[14]}                 @ reload IV
+       vld1.8  {@XMM[8]-@XMM[9]}, [$inp]!      @ reload input
+       veor    @XMM[0], @XMM[0], @XMM[14]      @ ^= IV
+       vld1.8  {@XMM[10]-@XMM[11]}, [$inp]!
+       veor    @XMM[1], @XMM[1], @XMM[8]
+       veor    @XMM[6], @XMM[6], @XMM[9]
+       vld1.8  {@XMM[12]-@XMM[13]}, [$inp]!
+       veor    @XMM[4], @XMM[4], @XMM[10]
+       veor    @XMM[2], @XMM[2], @XMM[11]
+       vld1.8  {@XMM[14]-@XMM[15]}, [$inp]!
+       veor    @XMM[7], @XMM[7], @XMM[12]
+       vst1.8  {@XMM[0]-@XMM[1]}, [$out]!      @ write output
+       veor    @XMM[3], @XMM[3], @XMM[13]
+       vst1.8  {@XMM[6]}, [$out]!
+       veor    @XMM[5], @XMM[5], @XMM[14]
+       vst1.8  {@XMM[4]}, [$out]!
+       vst1.8  {@XMM[2]}, [$out]!
+       vst1.8  {@XMM[7]}, [$out]!
+       vst1.8  {@XMM[3]}, [$out]!
+       vst1.8  {@XMM[5]}, [$out]!
+
+       b       .Lcbc_dec_loop
+
+.Lcbc_dec_loop_finish:
+       adds    $len, $len, #8
+       beq     .Lcbc_dec_done
+
+       vld1.8  {@XMM[0]}, [$inp]!              @ load input
+       cmp     $len, #2
+       blo     .Lcbc_dec_one
+       vld1.8  {@XMM[1]}, [$inp]!
+       mov     r4, $keysched                   @ pass the key
+       mov     r5, $rounds
+       vstmia  $fp, {@XMM[15]}                 @ put aside IV
+       beq     .Lcbc_dec_two
+       vld1.8  {@XMM[2]}, [$inp]!
+       cmp     $len, #4
+       blo     .Lcbc_dec_three
+       vld1.8  {@XMM[3]}, [$inp]!
+       beq     .Lcbc_dec_four
+       vld1.8  {@XMM[4]}, [$inp]!
+       cmp     $len, #6
+       blo     .Lcbc_dec_five
+       vld1.8  {@XMM[5]}, [$inp]!
+       beq     .Lcbc_dec_six
+       vld1.8  {@XMM[6]}, [$inp]!
+       sub     $inp, $inp, #0x70
+
+       bl      _bsaes_decrypt8
+
+       vldmia  $fp, {@XMM[14]}                 @ reload IV
+       vld1.8  {@XMM[8]-@XMM[9]}, [$inp]!      @ reload input
+       veor    @XMM[0], @XMM[0], @XMM[14]      @ ^= IV
+       vld1.8  {@XMM[10]-@XMM[11]}, [$inp]!
+       veor    @XMM[1], @XMM[1], @XMM[8]
+       veor    @XMM[6], @XMM[6], @XMM[9]
+       vld1.8  {@XMM[12]-@XMM[13]}, [$inp]!
+       veor    @XMM[4], @XMM[4], @XMM[10]
+       veor    @XMM[2], @XMM[2], @XMM[11]
+       vld1.8  {@XMM[15]}, [$inp]!
+       veor    @XMM[7], @XMM[7], @XMM[12]
+       vst1.8  {@XMM[0]-@XMM[1]}, [$out]!      @ write output
+       veor    @XMM[3], @XMM[3], @XMM[13]
+       vst1.8  {@XMM[6]}, [$out]!
+       vst1.8  {@XMM[4]}, [$out]!
+       vst1.8  {@XMM[2]}, [$out]!
+       vst1.8  {@XMM[7]}, [$out]!
+       vst1.8  {@XMM[3]}, [$out]!
+       b       .Lcbc_dec_done
+.align 4
+.Lcbc_dec_six:
+       sub     $inp, $inp, #0x60
+       bl      _bsaes_decrypt8
+       vldmia  $fp,{@XMM[14]}                  @ reload IV
+       vld1.8  {@XMM[8]-@XMM[9]}, [$inp]!      @ reload input
+       veor    @XMM[0], @XMM[0], @XMM[14]      @ ^= IV
+       vld1.8  {@XMM[10]-@XMM[11]}, [$inp]!
+       veor    @XMM[1], @XMM[1], @XMM[8]
+       veor    @XMM[6], @XMM[6], @XMM[9]
+       vld1.8  {@XMM[12]}, [$inp]!
+       veor    @XMM[4], @XMM[4], @XMM[10]
+       veor    @XMM[2], @XMM[2], @XMM[11]
+       vld1.8  {@XMM[15]}, [$inp]!
+       veor    @XMM[7], @XMM[7], @XMM[12]
+       vst1.8  {@XMM[0]-@XMM[1]}, [$out]!      @ write output
+       vst1.8  {@XMM[6]}, [$out]!
+       vst1.8  {@XMM[4]}, [$out]!
+       vst1.8  {@XMM[2]}, [$out]!
+       vst1.8  {@XMM[7]}, [$out]!
+       b       .Lcbc_dec_done
+.align 4
+.Lcbc_dec_five:
+       sub     $inp, $inp, #0x50
+       bl      _bsaes_decrypt8
+       vldmia  $fp, {@XMM[14]}                 @ reload IV
+       vld1.8  {@XMM[8]-@XMM[9]}, [$inp]!      @ reload input
+       veor    @XMM[0], @XMM[0], @XMM[14]      @ ^= IV
+       vld1.8  {@XMM[10]-@XMM[11]}, [$inp]!
+       veor    @XMM[1], @XMM[1], @XMM[8]
+       veor    @XMM[6], @XMM[6], @XMM[9]
+       vld1.8  {@XMM[15]}, [$inp]!
+       veor    @XMM[4], @XMM[4], @XMM[10]
+       vst1.8  {@XMM[0]-@XMM[1]}, [$out]!      @ write output
+       veor    @XMM[2], @XMM[2], @XMM[11]
+       vst1.8  {@XMM[6]}, [$out]!
+       vst1.8  {@XMM[4]}, [$out]!
+       vst1.8  {@XMM[2]}, [$out]!
+       b       .Lcbc_dec_done
+.align 4
+.Lcbc_dec_four:
+       sub     $inp, $inp, #0x40
+       bl      _bsaes_decrypt8
+       vldmia  $fp, {@XMM[14]}                 @ reload IV
+       vld1.8  {@XMM[8]-@XMM[9]}, [$inp]!      @ reload input
+       veor    @XMM[0], @XMM[0], @XMM[14]      @ ^= IV
+       vld1.8  {@XMM[10]}, [$inp]!
+       veor    @XMM[1], @XMM[1], @XMM[8]
+       veor    @XMM[6], @XMM[6], @XMM[9]
+       vld1.8  {@XMM[15]}, [$inp]!
+       veor    @XMM[4], @XMM[4], @XMM[10]
+       vst1.8  {@XMM[0]-@XMM[1]}, [$out]!      @ write output
+       vst1.8  {@XMM[6]}, [$out]!
+       vst1.8  {@XMM[4]}, [$out]!
+       b       .Lcbc_dec_done
+.align 4
+.Lcbc_dec_three:
+       sub     $inp, $inp, #0x30
+       bl      _bsaes_decrypt8
+       vldmia  $fp, {@XMM[14]}                 @ reload IV
+       vld1.8  {@XMM[8]-@XMM[9]}, [$inp]!      @ reload input
+       veor    @XMM[0], @XMM[0], @XMM[14]      @ ^= IV
+       vld1.8  {@XMM[15]}, [$inp]!
+       veor    @XMM[1], @XMM[1], @XMM[8]
+       veor    @XMM[6], @XMM[6], @XMM[9]
+       vst1.8  {@XMM[0]-@XMM[1]}, [$out]!      @ write output
+       vst1.8  {@XMM[6]}, [$out]!
+       b       .Lcbc_dec_done
+.align 4
+.Lcbc_dec_two:
+       sub     $inp, $inp, #0x20
+       bl      _bsaes_decrypt8
+       vldmia  $fp, {@XMM[14]}                 @ reload IV
+       vld1.8  {@XMM[8]}, [$inp]!              @ reload input
+       veor    @XMM[0], @XMM[0], @XMM[14]      @ ^= IV
+       vld1.8  {@XMM[15]}, [$inp]!             @ reload input
+       veor    @XMM[1], @XMM[1], @XMM[8]
+       vst1.8  {@XMM[0]-@XMM[1]}, [$out]!      @ write output
+       b       .Lcbc_dec_done
+.align 4
+.Lcbc_dec_one:
+       sub     $inp, $inp, #0x10
+       mov     $rounds, $out                   @ save original out pointer
+       mov     $out, $fp                       @ use the iv scratch space as out buffer
+       mov     r2, $key
+       vmov    @XMM[4],@XMM[15]                @ just in case ensure that IV
+       vmov    @XMM[5],@XMM[0]                 @ and input are preserved
+       bl      AES_decrypt
+       vld1.8  {@XMM[0]}, [$fp,:64]            @ load result
+       veor    @XMM[0], @XMM[0], @XMM[4]       @ ^= IV
+       vmov    @XMM[15], @XMM[5]               @ @XMM[5] holds input
+       vst1.8  {@XMM[0]}, [$rounds]            @ write output
+
+.Lcbc_dec_done:
+       vmov.i32        q0, #0
+       vmov.i32        q1, #0
+.Lcbc_dec_bzero:                               @ wipe key schedule [if any]
+       vstmia          $keysched!, {q0-q1}
+       teq             $keysched, $fp
+       bne             .Lcbc_dec_bzero
+
+       add     sp, $fp, #0x10
+       vst1.8  {@XMM[15]}, [$ivp]              @ return IV
+       vldmia  sp!, {d8-d15}
+       ldmia   sp!, {r4-r10, pc}
+.size  bsaes_cbc_encrypt,.-bsaes_cbc_encrypt
+___
+}
+{
+my ($inp,$out,$len,$key, $ctr,$fp,$rounds)=(map("r$_",(0..3,8..10)));
+my $const = "r6";      # shared with _bsaes_encrypt8_alt
+my $keysched = "sp";
+
+$code.=<<___;
+.extern        AES_encrypt
+.global        bsaes_ctr32_encrypt_blocks
+.type  bsaes_ctr32_encrypt_blocks,%function
+.align 5
+bsaes_ctr32_encrypt_blocks:
+       cmp     $len, #8                        @ use plain AES for
+       blo     .Lctr_enc_short                 @ small sizes
+
+       stmdb   sp!, {r4-r10, lr}
+       vstmdb  sp!, {d8-d15}                   @ ABI specification says so
+       ldr     $ctr, [sp, #0x60]               @ ctr is 1st arg on the stack
+       sub     sp, sp, #0x10                   @ scratch space to carry over the ctr
+       mov     $fp, sp                         @ save sp
+
+       @ allocate the key schedule on the stack
+       ldr     $rounds, [$key, #240]           @ get # of rounds
+       sub     sp, sp, $rounds, lsl#7          @ 128 bytes per inner round key
+       add     sp, sp, #`128-32`               @ size of bit-sliced key schedule
+
+       @ populate the key schedule
+       mov     r4, $key                        @ pass key
+       mov     r5, $rounds                     @ pass # of rounds
+       mov     r12, $keysched                  @ pass key schedule
+       bl      _bsaes_key_convert
+       veor    @XMM[7],@XMM[7],@XMM[15]        @ fix up last round key
+       vstmia  r12, {@XMM[7]}                  @ save last round key
+
+       vld1.8  {@XMM[0]}, [$ctr]               @ load counter
+       add     $ctr, $const, #.LREVM0SR-.LM0   @ borrow $ctr
+       vldmia  $keysched, {@XMM[4]}            @ load round0 key
+
+       vmov.i32        `&Dhi("@XMM[8]")`,#1    @ compose 1<<96
+       vmov.i32        `&Dlo("@XMM[8]")`,#0
+       vrev32.8        `&Dhi("@XMM[0]")`,`&Dhi("@XMM[0]")`
+       vshl.u64        `&Dhi("@XMM[8]")`,#32
+       vrev32.8        `&Dhi("@XMM[4]")`,`&Dhi("@XMM[4]")`
+       vadd.u32        @XMM[9],@XMM[8],@XMM[8] @ compose 2<<96
+       vstmia  $keysched, {@XMM[4]}            @ save adjusted round0 key
+       b       .Lctr_enc_loop
+
+.align 4
+.Lctr_enc_loop:
+       vadd.u32        @XMM[10], @XMM[8], @XMM[9]      @ compose 3<<96
+       vadd.u32        @XMM[1], @XMM[0], @XMM[8]       @ +1
+       vadd.u32        @XMM[2], @XMM[0], @XMM[9]       @ +2
+       vadd.u32        @XMM[3], @XMM[0], @XMM[10]      @ +3
+       vadd.u32        @XMM[4], @XMM[1], @XMM[10]
+       vadd.u32        @XMM[5], @XMM[2], @XMM[10]
+       vadd.u32        @XMM[6], @XMM[3], @XMM[10]
+       vadd.u32        @XMM[7], @XMM[4], @XMM[10]
+       vadd.u32        @XMM[10], @XMM[5], @XMM[10]     @ next counter
+
+       @ Borrow prologue from _bsaes_encrypt8 to use the opportunity
+       @ to flip byte order in 32-bit counter
+
+       vldmia          $keysched, {@XMM[9]}            @ load round0 key
+       add             r4, $keysched, #0x10            @ pass next round key
+       vldmia          $ctr, {@XMM[8]}                 @ .LREVM0SR
+       mov             r5, $rounds                     @ pass rounds
+       vstmia          $fp, {@XMM[10]}                 @ save next counter
+       sub             $const, $ctr, #.LREVM0SR-.LSR   @ pass constants
+
+       bl              _bsaes_encrypt8_alt
+
+       subs            $len, $len, #8
+       blo             .Lctr_enc_loop_done
+
+       vld1.8          {@XMM[8]-@XMM[9]}, [$inp]!      @ load input
+       vld1.8          {@XMM[10]-@XMM[11]}, [$inp]!
+       veor            @XMM[0], @XMM[8]
+       veor            @XMM[1], @XMM[9]
+       vld1.8          {@XMM[12]-@XMM[13]}, [$inp]!
+       veor            @XMM[4], @XMM[10]
+       veor            @XMM[6], @XMM[11]
+       vld1.8          {@XMM[14]-@XMM[15]}, [$inp]!
+       veor            @XMM[3], @XMM[12]
+       vst1.8          {@XMM[0]-@XMM[1]}, [$out]!      @ write output
+       veor            @XMM[7], @XMM[13]
+       veor            @XMM[2], @XMM[14]
+       vst1.8          {@XMM[4]}, [$out]!
+       veor            @XMM[5], @XMM[15]
+       vst1.8          {@XMM[6]}, [$out]!
+       vmov.i32        `&Dhi("@XMM[8]")`,#1            @ compose 1<<96
+       vst1.8          {@XMM[3]}, [$out]!
+       vmov.i32        `&Dlo("@XMM[8]")`,#0
+       vst1.8          {@XMM[7]}, [$out]!
+       vshl.u64        `&Dhi("@XMM[8]")`,#32
+       vst1.8          {@XMM[2]}, [$out]!
+       vadd.u32        @XMM[9],@XMM[8],@XMM[8]         @ compose 2<<96
+       vst1.8          {@XMM[5]}, [$out]!
+       vldmia          $fp, {@XMM[0]}                  @ load counter
+
+       bne             .Lctr_enc_loop
+       b               .Lctr_enc_done
+
+.align 4
+.Lctr_enc_loop_done:
+       add             $len, $len, #8
+       vld1.8          {@XMM[8]}, [$inp]!      @ load input
+       veor            @XMM[0], @XMM[8]
+       vst1.8          {@XMM[0]}, [$out]!      @ write output
+       cmp             $len, #2
+       blo             .Lctr_enc_done
+       vld1.8          {@XMM[9]}, [$inp]!
+       veor            @XMM[1], @XMM[9]
+       vst1.8          {@XMM[1]}, [$out]!
+       beq             .Lctr_enc_done
+       vld1.8          {@XMM[10]}, [$inp]!
+       veor            @XMM[4], @XMM[10]
+       vst1.8          {@XMM[4]}, [$out]!
+       cmp             $len, #4
+       blo             .Lctr_enc_done
+       vld1.8          {@XMM[11]}, [$inp]!
+       veor            @XMM[6], @XMM[11]
+       vst1.8          {@XMM[6]}, [$out]!
+       beq             .Lctr_enc_done
+       vld1.8          {@XMM[12]}, [$inp]!
+       veor            @XMM[3], @XMM[12]
+       vst1.8          {@XMM[3]}, [$out]!
+       cmp             $len, #6
+       blo             .Lctr_enc_done
+       vld1.8          {@XMM[13]}, [$inp]!
+       veor            @XMM[7], @XMM[13]
+       vst1.8          {@XMM[7]}, [$out]!
+       beq             .Lctr_enc_done
+       vld1.8          {@XMM[14]}, [$inp]
+       veor            @XMM[2], @XMM[14]
+       vst1.8          {@XMM[2]}, [$out]!
+
+.Lctr_enc_done:
+       vmov.i32        q0, #0
+       vmov.i32        q1, #0
+.Lctr_enc_bzero:                       @ wipe key schedule [if any]
+       vstmia          $keysched!, {q0-q1}
+       teq             $keysched, $fp
+       bne             .Lctr_enc_bzero
+
+       add     sp, $fp, #0x10
+       vldmia  sp!, {d8-d15}
+       ldmia   sp!, {r4-r10, pc}       @ return
+
+.align 4
+.Lctr_enc_short:
+       ldr     ip, [sp]                @ ctr pointer is passed on stack
+       stmdb   sp!, {r4-r8, lr}
+
+       mov     r4, $inp                @ copy arguments
+       mov     r5, $out
+       mov     r6, $len
+       mov     r7, $key
+       ldr     r8, [ip, #12]           @ load counter LSW
+       vld1.8  {@XMM[1]}, [ip]         @ load whole counter value
+#ifdef __ARMEL__
+       rev     r8, r8
+#endif
+       sub     sp, sp, #0x10
+       vst1.8  {@XMM[1]}, [sp,:64]     @ copy counter value
+       sub     sp, sp, #0x10
+
+.Lctr_enc_short_loop:
+       add     r0, sp, #0x10           @ input counter value
+       mov     r1, sp                  @ output on the stack
+       mov     r2, r7                  @ key
+
+       bl      AES_encrypt
+
+       vld1.8  {@XMM[0]}, [r4]!        @ load input
+       vld1.8  {@XMM[1]}, [sp,:64]     @ load encrypted counter
+       add     r8, r8, #1
+#ifdef __ARMEL__
+       rev     r0, r8
+       str     r0, [sp, #0x1c]         @ next counter value
+#else
+       str     r8, [sp, #0x1c]         @ next counter value
+#endif
+       veor    @XMM[0],@XMM[0],@XMM[1]
+       vst1.8  {@XMM[0]}, [r5]!        @ store output
+       subs    r6, r6, #1
+       bne     .Lctr_enc_short_loop
+
+       add     sp, sp, #0x20
+       ldmia   sp!, {r4-r8, pc}
+.size  bsaes_ctr32_encrypt_blocks,.-bsaes_ctr32_encrypt_blocks
+___
+}
 $code.=<<___;
 #endif
 ___
index a7d4ce4..c0fd031 100644 (file)
@@ -892,6 +892,14 @@ static const EVP_CIPHER aes_##keylen##_##mode = { \
        NULL,NULL,aes_##mode##_ctrl,NULL }; \
 const EVP_CIPHER *EVP_aes_##keylen##_##mode(void) \
 { return &aes_##keylen##_##mode; }
+
+#endif
+
+#if defined(AES_ASM) && defined(BSAES_ASM) && (defined(__arm__) || defined(__arm))
+#include "arm_arch.h"
+#if __ARM_ARCH__>=7
+#define BSAES_CAPABLE  (OPENSSL_armcap_P & ARMV7_NEON)
+#endif
 #endif
 
 #define BLOCK_CIPHER_generic_pack(nid,keylen,flags)            \
@@ -1624,11 +1632,13 @@ static int aes_xts_init_key(EVP_CIPHER_CTX *ctx, const unsigned char *key,
                xctx->stream = NULL;
 #endif
                /* key_len is two AES keys */
+#if !(defined(__arm__) || defined(__arm))      /* not yet? */
 #ifdef BSAES_CAPABLE
                if (BSAES_CAPABLE)
                        xctx->stream = enc ? bsaes_xts_encrypt : bsaes_xts_decrypt;
                else
 #endif
+#endif
 #ifdef VPAES_CAPABLE
                if (VPAES_CAPABLE)
                    {