#
# June 2017.
#
-# This is KECCAK_1X variant (see keccak1600.c) with bit interleaving.
-# How does it compare to Keccak Code Package? It's as fast, but several
-# times smaller, and is endian- and ISA-neutral. ISA neutrality means
-# that minimum ISA requirement is ARMv4, yet it can be assembled even
-# as ARMv7 Thumb-2.
+# Non-NEON code is KECCAK_1X variant (see sha/keccak1600.c) with bit
+# interleaving. How does it compare to Keccak Code Package? It's as
+# fast, but several times smaller, and is endian- and ISA-neutral. ISA
+# neutrality means that minimum ISA requirement is ARMv4, yet it can
+# be assembled even as Thumb-2. NEON code path is KECCAK_1X_ALT with
+# register layout taken from Keccak Code Package. It's also as fast,
+# in fact faster by 10-15% on some processors, and endian-neutral.
#
########################################################################
-# Numbers are cycles per processed byte accounting even for input bit
-# interleaving.
+# Numbers are cycles per processed byte. Non-NEON results account even
+# for input bit interleaving.
#
-# r=1600(*) r=1024
+# r=1600(*),NEON r=1088(**),NEON
#
-# Cortex-A7 71/+180% 103
-# Cortex-A8 48/+290% 69
-# Cortex-A15 34/+210% 49
+# Cortex-A5 67/+130%, 24 96, 36
+# Cortex-A7 60/+90%, 23 87, 34
+# Cortex-A8 39/+220%, 20 56, 30
+# Cortex-A9 41/+160%, 17 58, 26
+# Cortex-A15 30/+65%, 12 41, 18
+# Snapdragon S4 35/+120%, 16 50, 24
#
-# (*) Not used in real life, meaningful as estimate for single sponge
-# operation performance. Numbers after slash are improvement over
-# compiler-generated KECCAK_1X reference code.
+# (*) Not used in real life, meaningful as estimate for single absorb
+# operation performance. Percentage after slash is improvement
+# over compiler-generated KECCAK_1X reference code.
+# (**) Corresponds to SHA3-256, 8KB message size.
my @C = map("r$_",(0..9));
my @E = map("r$_",(10..12,14));
.code 32
#endif
-.type iotas,%object
+.type iotas32, %object
.align 5
-iotas:
+iotas32:
.long 0x00000001, 0x00000000
.long 0x00000000, 0x00000089
.long 0x00000000, 0x8000008b
.long 0x00000000, 0x80000088
.long 0x00000001, 0x00008000
.long 0x00000000, 0x80008082
-.size iotas,.-iotas
+.size iotas32,.-iotas32
.type KeccakF1600_int, %function
.align 5
ldr @E[2],[sp,#324] @ load counter
eor @C[2],@C[2],@E[0]
ror @C[4],@E[3],#32-22
- adr @E[3],iotas
+ adr @E[3],iotas32
eor @C[3],@C[3],@E[1]
ror @C[2],@C[2],#32-22 @ C[1] = ROL64(A[1][1] ^ E[0], rhotates[1][1]); /* D[1] */
add @E[3],@E[3],@E[2]
ldmia sp!,{r4-r11,pc}
.size KeccakF1600,.-KeccakF1600
___
-{ my ($hi,$lo,$i,$A_flat, $len,$bsz,$inp) = map("r$_",(5..8, 10..12));
+{ my ($A_flat,$inp,$len,$bsz) = map("r$_",(10..12,14));
########################################################################
# Stack layout
# | ... |
# | ... |
# +336->+-----------------------+
-# | uint64_t *A |
+# | 0x55555555 |
# +340->+-----------------------+
-# | const void *inp |
+# | 0x33333333 |
# +344->+-----------------------+
-# | size_t len |
+# | 0x0f0f0f0f |
# +348->+-----------------------+
-# | size_t bs |
+# | 0x00ff00ff |
# +352->+-----------------------+
+# | uint64_t *A |
+# +356->+-----------------------+
+# | const void *inp |
+# +360->+-----------------------+
+# | size_t len |
+# +364->+-----------------------+
+# | size_t bs |
+# +368->+-----------------------+
# | ....
$code.=<<___;
.align 5
SHA3_absorb:
stmdb sp!,{r0-r12,lr}
- sub sp,sp,#320+16
+ sub sp,sp,#336+16
- mov r12,r0
- add r14,sp,#0
+ add $A_flat,r0,#$A[1][0]
+ @ mov $inp,r1
mov $len,r2
mov $bsz,r3
+ cmp r2,r3
+ blo .Labsorb_abort
- ldmia r12!,{@C[0]-@C[9]} @ copy A[5][5] to stack
- stmia r14!,{@C[0]-@C[9]}
- ldmia r12!,{@C[0]-@C[9]}
- stmia r14!,{@C[0]-@C[9]}
- ldmia r12!,{@C[0]-@C[9]}
- stmia r14!,{@C[0]-@C[9]}
- ldmia r12!,{@C[0]-@C[9]}
- stmia r14!,{@C[0]-@C[9]}
- ldmia r12, {@C[0]-@C[9]}
- stmia r14, {@C[0]-@C[9]}
+ add $inp,sp,#0
+ ldmia r0, {@C[0]-@C[9]} @ copy A[5][5] to stack
+ stmia $inp!, {@C[0]-@C[9]}
+ ldmia $A_flat!,{@C[0]-@C[9]}
+ stmia $inp!, {@C[0]-@C[9]}
+ ldmia $A_flat!,{@C[0]-@C[9]}
+ stmia $inp!, {@C[0]-@C[9]}
+ ldmia $A_flat!,{@C[0]-@C[9]}
+ stmia $inp!, {@C[0]-@C[9]}
+ ldmia $A_flat!,{@C[0]-@C[9]}
+ stmia $inp, {@C[0]-@C[9]}
- ldr $inp,[sp,#340]
+ ldr $inp,[sp,#356] @ restore $inp
+#ifdef __thumb2__
+ mov r9,#0x00ff00ff
+ mov r8,#0x0f0f0f0f
+ mov r7,#0x33333333
+ mov r6,#0x55555555
+#else
+ mov r6,#0x11 @ compose constants
+ mov r8,#0x0f
+ mov r9,#0xff
+ orr r6,r6,r6,lsl#8
+ orr r8,r8,r8,lsl#8
+ orr r6,r6,r6,lsl#16 @ 0x11111111
+ orr r9,r9,r9,lsl#16 @ 0x00ff00ff
+ orr r8,r8,r8,lsl#16 @ 0x0f0f0f0f
+ orr r7,r6,r6,lsl#1 @ 0x33333333
+ orr r6,r6,r6,lsl#2 @ 0x55555555
+#endif
+ str r9,[sp,#348]
+ str r8,[sp,#344]
+ str r7,[sp,#340]
+ str r6,[sp,#336]
+ b .Loop_absorb
+.align 4
.Loop_absorb:
subs r0,$len,$bsz
blo .Labsorbed
add $A_flat,sp,#0
- str r0,[sp,#344] @ save len - bsz
+ str r0,[sp,#360] @ save len - bsz
+.align 4
.Loop_block:
- ldmia $A_flat,{r2-r3} @ A_flat[i]
- ldrb r0,[$inp,#7]! @ inp[7]
- mov $i,#8
-
-.Lane_loop:
- subs $i,$i,#1
- lsl r1,r0,#24
- blo .Lane_done
-#ifdef __thumb2__
- it ne
- ldrbne r0,[$inp,#-1]!
-#else
- ldrneb r0,[$inp,#-1]!
-#endif
- adds r1,r1,r1 @ sip through carry flag
- adc $hi,$hi,$hi
- adds r1,r1,r1
- adc $lo,$lo,$lo
- adds r1,r1,r1
- adc $hi,$hi,$hi
- adds r1,r1,r1
- adc $lo,$lo,$lo
- adds r1,r1,r1
- adc $hi,$hi,$hi
- adds r1,r1,r1
- adc $lo,$lo,$lo
- adds r1,r1,r1
- adc $hi,$hi,$hi
- adds r1,r1,r1
- adc $lo,$lo,$lo
- b .Lane_loop
-
-.Lane_done:
- eor r2,r2,$lo
- eor r3,r3,$hi
- add $inp,$inp,#8
- stmia $A_flat!,{r2-r3} @ A_flat[i++] ^= BitInterleave(inp[0..7])
+ ldrb r0,[$inp],#1
+ ldrb r1,[$inp],#1
+ ldrb r2,[$inp],#1
+ ldrb r3,[$inp],#1
+ ldrb r4,[$inp],#1
+ orr r0,r0,r1,lsl#8
+ ldrb r1,[$inp],#1
+ orr r0,r0,r2,lsl#16
+ ldrb r2,[$inp],#1
+ orr r0,r0,r3,lsl#24 @ lo
+ ldrb r3,[$inp],#1
+ orr r1,r4,r1,lsl#8
+ orr r1,r1,r2,lsl#16
+ orr r1,r1,r3,lsl#24 @ hi
+
+ and r2,r0,r6 @ &=0x55555555
+ and r0,r0,r6,lsl#1 @ &=0xaaaaaaaa
+ and r3,r1,r6 @ &=0x55555555
+ and r1,r1,r6,lsl#1 @ &=0xaaaaaaaa
+ orr r2,r2,r2,lsr#1
+ orr r0,r0,r0,lsl#1
+ orr r3,r3,r3,lsr#1
+ orr r1,r1,r1,lsl#1
+ and r2,r2,r7 @ &=0x33333333
+ and r0,r0,r7,lsl#2 @ &=0xcccccccc
+ and r3,r3,r7 @ &=0x33333333
+ and r1,r1,r7,lsl#2 @ &=0xcccccccc
+ orr r2,r2,r2,lsr#2
+ orr r0,r0,r0,lsl#2
+ orr r3,r3,r3,lsr#2
+ orr r1,r1,r1,lsl#2
+ and r2,r2,r8 @ &=0x0f0f0f0f
+ and r0,r0,r8,lsl#4 @ &=0xf0f0f0f0
+ and r3,r3,r8 @ &=0x0f0f0f0f
+ and r1,r1,r8,lsl#4 @ &=0xf0f0f0f0
+ ldmia $A_flat,{r4-r5} @ A_flat[i]
+ orr r2,r2,r2,lsr#4
+ orr r0,r0,r0,lsl#4
+ orr r3,r3,r3,lsr#4
+ orr r1,r1,r1,lsl#4
+ and r2,r2,r9 @ &=0x00ff00ff
+ and r0,r0,r9,lsl#8 @ &=0xff00ff00
+ and r3,r3,r9 @ &=0x00ff00ff
+ and r1,r1,r9,lsl#8 @ &=0xff00ff00
+ orr r2,r2,r2,lsr#8
+ orr r0,r0,r0,lsl#8
+ orr r3,r3,r3,lsr#8
+ orr r1,r1,r1,lsl#8
+
+ lsl r2,r2,#16
+ lsr r1,r1,#16
+ eor r4,r4,r3,lsl#16
+ eor r5,r5,r0,lsr#16
+ eor r4,r4,r2,lsr#16
+ eor r5,r5,r1,lsl#16
+ stmia $A_flat!,{r4-r5} @ A_flat[i++] ^= BitInterleave(inp[0..7])
+
subs $bsz,$bsz,#8
bhi .Loop_block
- str $inp,[sp,#340]
+ str $inp,[sp,#356]
bl KeccakF1600_int
- ldr $inp,[sp,#340]
- ldr $len,[sp,#344]
- ldr $bsz,[sp,#348]
+ add r14,sp,#336
+ ldmia r14,{r6-r12,r14} @ restore constants and variables
b .Loop_absorb
.align 4
.Labsorbed:
- add r12,sp,#$A[1][0]
- ldr r14, [sp,#336] @ pull pointer to A[5][5]
- ldmia sp, {@C[0]-@C[9]}
- stmia r14!,{@C[0]-@C[9]} @ return A[5][5]
- ldmia r12!,{@C[0]-@C[9]}
- stmia r14!,{@C[0]-@C[9]}
- ldmia r12!,{@C[0]-@C[9]}
- stmia r14!,{@C[0]-@C[9]}
- ldmia r12!,{@C[0]-@C[9]}
- stmia r14!,{@C[0]-@C[9]}
- ldmia r12, {@C[0]-@C[9]}
- stmia r14, {@C[0]-@C[9]}
-
- add sp,sp,#320+32
+ add $inp,sp,#$A[1][0]
+ ldmia sp, {@C[0]-@C[9]}
+ stmia $A_flat!,{@C[0]-@C[9]} @ return A[5][5]
+ ldmia $inp!, {@C[0]-@C[9]}
+ stmia $A_flat!,{@C[0]-@C[9]}
+ ldmia $inp!, {@C[0]-@C[9]}
+ stmia $A_flat!,{@C[0]-@C[9]}
+ ldmia $inp!, {@C[0]-@C[9]}
+ stmia $A_flat!,{@C[0]-@C[9]}
+ ldmia $inp, {@C[0]-@C[9]}
+ stmia $A_flat, {@C[0]-@C[9]}
+
+.Labsorb_abort:
+ add sp,sp,#336+32
mov r0,$len @ return value
ldmia sp!,{r4-r12,pc}
.size SHA3_absorb,.-SHA3_absorb
___
}
-{ my ($A_flat,$out,$len,$bsz, $byte,$shl) = map("r$_", (4..9));
+{ my ($out,$len,$A_flat,$bsz) = map("r$_", (4,5,10,12));
$code.=<<___;
.global SHA3_squeeze
.type SHA3_squeeze,%function
.align 5
SHA3_squeeze:
- stmdb sp!,{r4-r10,lr}
- mov r12,r0
+ stmdb sp!,{r0,r3-r10,lr}
+
mov $A_flat,r0
mov $out,r1
mov $len,r2
mov $bsz,r3
- mov r14,r3
+
+#ifdef __thumb2__
+ mov r9,#0x00ff00ff
+ mov r8,#0x0f0f0f0f
+ mov r7,#0x33333333
+ mov r6,#0x55555555
+#else
+ mov r6,#0x11 @ compose constants
+ mov r8,#0x0f
+ mov r9,#0xff
+ orr r6,r6,r6,lsl#8
+ orr r8,r8,r8,lsl#8
+ orr r6,r6,r6,lsl#16 @ 0x11111111
+ orr r9,r9,r9,lsl#16 @ 0x00ff00ff
+ orr r8,r8,r8,lsl#16 @ 0x0f0f0f0f
+ orr r7,r6,r6,lsl#1 @ 0x33333333
+ orr r6,r6,r6,lsl#2 @ 0x55555555
+#endif
+ stmdb sp!,{r6-r9}
+
+ mov r14,$A_flat
b .Loop_squeeze
.align 4
.Loop_squeeze:
- ldmia r12!,{r0,r1} @ A_flat[i++]
- mov $shl,#28
-
-.Lane_squeeze:
- lsl r2,r0,$shl
- lsl r3,r1,$shl
- eor $byte,$byte,$byte
- adds r3,r3,r3 @ sip through carry flag
- adc $byte,$byte,$byte
- adds r2,r2,r2
- adc $byte,$byte,$byte
- adds r3,r3,r3
- adc $byte,$byte,$byte
- adds r2,r2,r2
- adc $byte,$byte,$byte
- adds r3,r3,r3
- adc $byte,$byte,$byte
- adds r2,r2,r2
- adc $byte,$byte,$byte
- adds r3,r3,r3
- adc $byte,$byte,$byte
- adds r2,r2,r2
- adc $byte,$byte,$byte
- subs $len,$len,#1 @ len -= 1
- str $byte,[$out],#1
+ ldmia $A_flat!,{r0,r1} @ A_flat[i++]
+
+ lsl r2,r0,#16
+ lsl r3,r1,#16 @ r3 = r1 << 16
+ lsr r2,r2,#16 @ r2 = r0 & 0x0000ffff
+ lsr r1,r1,#16
+ lsr r0,r0,#16 @ r0 = r0 >> 16
+ lsl r1,r1,#16 @ r1 = r1 & 0xffff0000
+
+ orr r2,r2,r2,lsl#8
+ orr r3,r3,r3,lsr#8
+ orr r0,r0,r0,lsl#8
+ orr r1,r1,r1,lsr#8
+ and r2,r2,r9 @ &=0x00ff00ff
+ and r3,r3,r9,lsl#8 @ &=0xff00ff00
+ and r0,r0,r9 @ &=0x00ff00ff
+ and r1,r1,r9,lsl#8 @ &=0xff00ff00
+ orr r2,r2,r2,lsl#4
+ orr r3,r3,r3,lsr#4
+ orr r0,r0,r0,lsl#4
+ orr r1,r1,r1,lsr#4
+ and r2,r2,r8 @ &=0x0f0f0f0f
+ and r3,r3,r8,lsl#4 @ &=0xf0f0f0f0
+ and r0,r0,r8 @ &=0x0f0f0f0f
+ and r1,r1,r8,lsl#4 @ &=0xf0f0f0f0
+ orr r2,r2,r2,lsl#2
+ orr r3,r3,r3,lsr#2
+ orr r0,r0,r0,lsl#2
+ orr r1,r1,r1,lsr#2
+ and r2,r2,r7 @ &=0x33333333
+ and r3,r3,r7,lsl#2 @ &=0xcccccccc
+ and r0,r0,r7 @ &=0x33333333
+ and r1,r1,r7,lsl#2 @ &=0xcccccccc
+ orr r2,r2,r2,lsl#1
+ orr r3,r3,r3,lsr#1
+ orr r0,r0,r0,lsl#1
+ orr r1,r1,r1,lsr#1
+ and r2,r2,r6 @ &=0x55555555
+ and r3,r3,r6,lsl#1 @ &=0xaaaaaaaa
+ and r0,r0,r6 @ &=0x55555555
+ and r1,r1,r6,lsl#1 @ &=0xaaaaaaaa
+
+ orr r2,r2,r3
+ orr r0,r0,r1
+
+ cmp $len,#8
+ blo .Lsqueeze_tail
+ lsr r1,r2,#8
+ strb r2,[$out],#1
+ lsr r3,r2,#16
+ strb r1,[$out],#1
+ lsr r2,r2,#24
+ strb r3,[$out],#1
+ strb r2,[$out],#1
+
+ lsr r1,r0,#8
+ strb r0,[$out],#1
+ lsr r3,r0,#16
+ strb r1,[$out],#1
+ lsr r0,r0,#24
+ strb r3,[$out],#1
+ strb r0,[$out],#1
+ subs $len,$len,#8
beq .Lsqueeze_done
- subs $shl,$shl,#4
- bhs .Lane_squeeze
- subs r14,r14,#8 @ bsz -= 8
+ subs $bsz,$bsz,#8 @ bsz -= 8
bhi .Loop_squeeze
- mov r0,$A_flat
+ mov r0,r14 @ original $A_flat
bl KeccakF1600
- mov r12,$A_flat
- mov r14,$bsz
+ ldmia sp,{r6-r10,r12} @ restore constants and variables
+ mov r14,$A_flat
b .Loop_squeeze
+.align 4
+.Lsqueeze_tail:
+ strb r2,[$out],#1
+ lsr r2,r2,#8
+ subs $len,$len,#1
+ beq .Lsqueeze_done
+ strb r2,[$out],#1
+ lsr r2,r2,#8
+ subs $len,$len,#1
+ beq .Lsqueeze_done
+ strb r2,[$out],#1
+ lsr r2,r2,#8
+ subs $len,$len,#1
+ beq .Lsqueeze_done
+ strb r2,[$out],#1
+ subs $len,$len,#1
+ beq .Lsqueeze_done
+
+ strb r0,[$out],#1
+ lsr r0,r0,#8
+ subs $len,$len,#1
+ beq .Lsqueeze_done
+ strb r0,[$out],#1
+ lsr r0,r0,#8
+ subs $len,$len,#1
+ beq .Lsqueeze_done
+ strb r0,[$out]
+ b .Lsqueeze_done
+
+.align 4
.Lsqueeze_done:
+ add sp,sp,#24
ldmia sp!,{r4-r10,pc}
.size SHA3_squeeze,.-SHA3_squeeze
-.asciz "Keccak-1600 absorb and squeeze for ARMv4, CRYPTOGAMS by <appro\@openssl.org>"
-.align 2
___
}
+$code.=<<___;
+.fpu neon
+
+.type iotas64, %object
+.align 5
+iotas64:
+ .quad 0x0000000000000001
+ .quad 0x0000000000008082
+ .quad 0x800000000000808a
+ .quad 0x8000000080008000
+ .quad 0x000000000000808b
+ .quad 0x0000000080000001
+ .quad 0x8000000080008081
+ .quad 0x8000000000008009
+ .quad 0x000000000000008a
+ .quad 0x0000000000000088
+ .quad 0x0000000080008009
+ .quad 0x000000008000000a
+ .quad 0x000000008000808b
+ .quad 0x800000000000008b
+ .quad 0x8000000000008089
+ .quad 0x8000000000008003
+ .quad 0x8000000000008002
+ .quad 0x8000000000000080
+ .quad 0x000000000000800a
+ .quad 0x800000008000000a
+ .quad 0x8000000080008081
+ .quad 0x8000000000008080
+ .quad 0x0000000080000001
+ .quad 0x8000000080008008
+.size iotas64,.-iotas64
+
+.type KeccakF1600_neon, %function
+.align 5
+KeccakF1600_neon:
+ add r1, r0, #16
+ adr r2, iotas64
+ mov r3, #24 @ loop counter
+ b .Loop_neon
+
+.align 4
+.Loop_neon:
+ @ Theta
+ vst1.64 {q4}, [r0:64] @ offload A[0..1][4]
+ veor q13, q0, q5 @ A[0..1][0]^A[2..3][0]
+ vst1.64 {d18}, [r1:64] @ offload A[2][4]
+ veor q14, q1, q6 @ A[0..1][1]^A[2..3][1]
+ veor q15, q2, q7 @ A[0..1][2]^A[2..3][2]
+ veor d26, d26, d27 @ C[0]=A[0][0]^A[1][0]^A[2][0]^A[3][0]
+ veor d27, d28, d29 @ C[1]=A[0][1]^A[1][1]^A[2][1]^A[3][1]
+ veor q14, q3, q8 @ A[0..1][3]^A[2..3][3]
+ veor q4, q4, q9 @ A[0..1][4]^A[2..3][4]
+ veor d30, d30, d31 @ C[2]=A[0][2]^A[1][2]^A[2][2]^A[3][2]
+ veor d31, d28, d29 @ C[3]=A[0][3]^A[1][3]^A[2][3]^A[3][3]
+ veor d25, d8, d9 @ C[4]=A[0][4]^A[1][4]^A[2][4]^A[3][4]
+ veor q13, q13, q10 @ C[0..1]^=A[4][0..1]
+ veor q14, q15, q11 @ C[2..3]^=A[4][2..3]
+ veor d25, d25, d24 @ C[4]^=A[4][4]
+
+ vadd.u64 q4, q13, q13 @ C[0..1]<<1
+ vadd.u64 q15, q14, q14 @ C[2..3]<<1
+ vadd.u64 d18, d25, d25 @ C[4]<<1
+ vsri.u64 q4, q13, #63 @ ROL64(C[0..1],1)
+ vsri.u64 q15, q14, #63 @ ROL64(C[2..3],1)
+ vsri.u64 d18, d25, #63 @ ROL64(C[4],1)
+ veor d25, d25, d9 @ D[0] = C[4] ^= ROL64(C[1],1)
+ veor q13, q13, q15 @ D[1..2] = C[0..1] ^ ROL64(C[2..3],1)
+ veor d28, d28, d18 @ D[3] = C[2] ^= ROL64(C[4],1)
+ veor d29, d29, d8 @ D[4] = C[3] ^= ROL64(C[0],1)
+
+ veor d0, d0, d25 @ A[0][0] ^= C[4]
+ veor d1, d1, d25 @ A[1][0] ^= C[4]
+ veor d10, d10, d25 @ A[2][0] ^= C[4]
+ veor d11, d11, d25 @ A[3][0] ^= C[4]
+ veor d20, d20, d25 @ A[4][0] ^= C[4]
+
+ veor d2, d2, d26 @ A[0][1] ^= D[1]
+ veor d3, d3, d26 @ A[1][1] ^= D[1]
+ veor d12, d12, d26 @ A[2][1] ^= D[1]
+ veor d13, d13, d26 @ A[3][1] ^= D[1]
+ veor d21, d21, d26 @ A[4][1] ^= D[1]
+ vmov d26, d27
+
+ veor d6, d6, d28 @ A[0][3] ^= C[2]
+ veor d7, d7, d28 @ A[1][3] ^= C[2]
+ veor d16, d16, d28 @ A[2][3] ^= C[2]
+ veor d17, d17, d28 @ A[3][3] ^= C[2]
+ veor d23, d23, d28 @ A[4][3] ^= C[2]
+ vld1.64 {q4}, [r0:64] @ restore A[0..1][4]
+ vmov d28, d29
+
+ vld1.64 {d18}, [r1:64] @ restore A[2][4]
+ veor q2, q2, q13 @ A[0..1][2] ^= D[2]
+ veor q7, q7, q13 @ A[2..3][2] ^= D[2]
+ veor d22, d22, d27 @ A[4][2] ^= D[2]
+
+ veor q4, q4, q14 @ A[0..1][4] ^= C[3]
+ veor q9, q9, q14 @ A[2..3][4] ^= C[3]
+ veor d24, d24, d29 @ A[4][4] ^= C[3]
+
+ @ Rho + Pi
+ vmov d26, d2 @ C[1] = A[0][1]
+ vshl.u64 d2, d3, #44
+ vmov d27, d4 @ C[2] = A[0][2]
+ vshl.u64 d4, d14, #43
+ vmov d28, d6 @ C[3] = A[0][3]
+ vshl.u64 d6, d17, #21
+ vmov d29, d8 @ C[4] = A[0][4]
+ vshl.u64 d8, d24, #14
+ vsri.u64 d2, d3, #64-44 @ A[0][1] = ROL64(A[1][1], rhotates[1][1])
+ vsri.u64 d4, d14, #64-43 @ A[0][2] = ROL64(A[2][2], rhotates[2][2])
+ vsri.u64 d6, d17, #64-21 @ A[0][3] = ROL64(A[3][3], rhotates[3][3])
+ vsri.u64 d8, d24, #64-14 @ A[0][4] = ROL64(A[4][4], rhotates[4][4])
+
+ vshl.u64 d3, d9, #20
+ vshl.u64 d14, d16, #25
+ vshl.u64 d17, d15, #15
+ vshl.u64 d24, d21, #2
+ vsri.u64 d3, d9, #64-20 @ A[1][1] = ROL64(A[1][4], rhotates[1][4])
+ vsri.u64 d14, d16, #64-25 @ A[2][2] = ROL64(A[2][3], rhotates[2][3])
+ vsri.u64 d17, d15, #64-15 @ A[3][3] = ROL64(A[3][2], rhotates[3][2])
+ vsri.u64 d24, d21, #64-2 @ A[4][4] = ROL64(A[4][1], rhotates[4][1])
+
+ vshl.u64 d9, d22, #61
+ @ vshl.u64 d16, d19, #8
+ vshl.u64 d15, d12, #10
+ vshl.u64 d21, d7, #55
+ vsri.u64 d9, d22, #64-61 @ A[1][4] = ROL64(A[4][2], rhotates[4][2])
+ vext.8 d16, d19, d19, #8-1 @ A[2][3] = ROL64(A[3][4], rhotates[3][4])
+ vsri.u64 d15, d12, #64-10 @ A[3][2] = ROL64(A[2][1], rhotates[2][1])
+ vsri.u64 d21, d7, #64-55 @ A[4][1] = ROL64(A[1][3], rhotates[1][3])
+
+ vshl.u64 d22, d18, #39
+ @ vshl.u64 d19, d23, #56
+ vshl.u64 d12, d5, #6
+ vshl.u64 d7, d13, #45
+ vsri.u64 d22, d18, #64-39 @ A[4][2] = ROL64(A[2][4], rhotates[2][4])
+ vext.8 d19, d23, d23, #8-7 @ A[3][4] = ROL64(A[4][3], rhotates[4][3])
+ vsri.u64 d12, d5, #64-6 @ A[2][1] = ROL64(A[1][2], rhotates[1][2])
+ vsri.u64 d7, d13, #64-45 @ A[1][3] = ROL64(A[3][1], rhotates[3][1])
+
+ vshl.u64 d18, d20, #18
+ vshl.u64 d23, d11, #41
+ vshl.u64 d5, d10, #3
+ vshl.u64 d13, d1, #36
+ vsri.u64 d18, d20, #64-18 @ A[2][4] = ROL64(A[4][0], rhotates[4][0])
+ vsri.u64 d23, d11, #64-41 @ A[4][3] = ROL64(A[3][0], rhotates[3][0])
+ vsri.u64 d5, d10, #64-3 @ A[1][2] = ROL64(A[2][0], rhotates[2][0])
+ vsri.u64 d13, d1, #64-36 @ A[3][1] = ROL64(A[1][0], rhotates[1][0])
+
+ vshl.u64 d1, d28, #28
+ vshl.u64 d10, d26, #1
+ vshl.u64 d11, d29, #27
+ vshl.u64 d20, d27, #62
+ vsri.u64 d1, d28, #64-28 @ A[1][0] = ROL64(C[3], rhotates[0][3])
+ vsri.u64 d10, d26, #64-1 @ A[2][0] = ROL64(C[1], rhotates[0][1])
+ vsri.u64 d11, d29, #64-27 @ A[3][0] = ROL64(C[4], rhotates[0][4])
+ vsri.u64 d20, d27, #64-62 @ A[4][0] = ROL64(C[2], rhotates[0][2])
+
+ @ Chi + Iota
+ vbic q13, q2, q1
+ vbic q14, q3, q2
+ vbic q15, q4, q3
+ veor q13, q13, q0 @ A[0..1][0] ^ (~A[0..1][1] & A[0..1][2])
+ veor q14, q14, q1 @ A[0..1][1] ^ (~A[0..1][2] & A[0..1][3])
+ veor q2, q2, q15 @ A[0..1][2] ^= (~A[0..1][3] & A[0..1][4])
+ vst1.64 {q13}, [r0:64] @ offload A[0..1][0]
+ vbic q13, q0, q4
+ vbic q15, q1, q0
+ vmov q1, q14 @ A[0..1][1]
+ veor q3, q3, q13 @ A[0..1][3] ^= (~A[0..1][4] & A[0..1][0])
+ veor q4, q4, q15 @ A[0..1][4] ^= (~A[0..1][0] & A[0..1][1])
+
+ vbic q13, q7, q6
+ vmov q0, q5 @ A[2..3][0]
+ vbic q14, q8, q7
+ vmov q15, q6 @ A[2..3][1]
+ veor q5, q5, q13 @ A[2..3][0] ^= (~A[2..3][1] & A[2..3][2])
+ vbic q13, q9, q8
+ veor q6, q6, q14 @ A[2..3][1] ^= (~A[2..3][2] & A[2..3][3])
+ vbic q14, q0, q9
+ veor q7, q7, q13 @ A[2..3][2] ^= (~A[2..3][3] & A[2..3][4])
+ vbic q13, q15, q0
+ veor q8, q8, q14 @ A[2..3][3] ^= (~A[2..3][4] & A[2..3][0])
+ vmov q14, q10 @ A[4][0..1]
+ veor q9, q9, q13 @ A[2..3][4] ^= (~A[2..3][0] & A[2..3][1])
+
+ vld1.64 d25, [r2:64]! @ Iota[i++]
+ vbic d26, d22, d21
+ vbic d27, d23, d22
+ vld1.64 {q0}, [r0:64] @ restore A[0..1][0]
+ veor d20, d20, d26 @ A[4][0] ^= (~A[4][1] & A[4][2])
+ vbic d26, d24, d23
+ veor d21, d21, d27 @ A[4][1] ^= (~A[4][2] & A[4][3])
+ vbic d27, d28, d24
+ veor d22, d22, d26 @ A[4][2] ^= (~A[4][3] & A[4][4])
+ vbic d26, d29, d28
+ veor d23, d23, d27 @ A[4][3] ^= (~A[4][4] & A[4][0])
+ veor d0, d0, d25 @ A[0][0] ^= Iota[i]
+ veor d24, d24, d26 @ A[4][4] ^= (~A[4][0] & A[4][1])
+
+ subs r3, r3, #1
+ bne .Loop_neon
+
+ bx lr
+.size KeccakF1600_neon,.-KeccakF1600_neon
+
+.global SHA3_absorb_neon
+.type SHA3_absorb_neon, %function
+.align 5
+SHA3_absorb_neon:
+ stmdb sp!, {r4-r6,lr}
+ vstmdb sp!, {d8-d15}
+
+ mov r4, r1 @ inp
+ mov r5, r2 @ len
+ mov r6, r3 @ bsz
+
+ vld1.32 {d0}, [r0:64]! @ A[0][0]
+ vld1.32 {d2}, [r0:64]! @ A[0][1]
+ vld1.32 {d4}, [r0:64]! @ A[0][2]
+ vld1.32 {d6}, [r0:64]! @ A[0][3]
+ vld1.32 {d8}, [r0:64]! @ A[0][4]
+
+ vld1.32 {d1}, [r0:64]! @ A[1][0]
+ vld1.32 {d3}, [r0:64]! @ A[1][1]
+ vld1.32 {d5}, [r0:64]! @ A[1][2]
+ vld1.32 {d7}, [r0:64]! @ A[1][3]
+ vld1.32 {d9}, [r0:64]! @ A[1][4]
+
+ vld1.32 {d10}, [r0:64]! @ A[2][0]
+ vld1.32 {d12}, [r0:64]! @ A[2][1]
+ vld1.32 {d14}, [r0:64]! @ A[2][2]
+ vld1.32 {d16}, [r0:64]! @ A[2][3]
+ vld1.32 {d18}, [r0:64]! @ A[2][4]
+
+ vld1.32 {d11}, [r0:64]! @ A[3][0]
+ vld1.32 {d13}, [r0:64]! @ A[3][1]
+ vld1.32 {d15}, [r0:64]! @ A[3][2]
+ vld1.32 {d17}, [r0:64]! @ A[3][3]
+ vld1.32 {d19}, [r0:64]! @ A[3][4]
+
+ vld1.32 {d20-d23}, [r0:64]! @ A[4][0..3]
+ vld1.32 {d24}, [r0:64] @ A[4][4]
+ sub r0, r0, #24*8 @ rewind
+ b .Loop_absorb_neon
+
+.align 4
+.Loop_absorb_neon:
+ subs r12, r5, r6 @ len - bsz
+ blo .Labsorbed_neon
+ mov r5, r12
+
+ vld1.8 {d31}, [r4]! @ endian-neutral loads...
+ cmp r6, #8*2
+ veor d0, d0, d31 @ A[0][0] ^= *inp++
+ blo .Lprocess_neon
+ vld1.8 {d31}, [r4]!
+ veor d2, d2, d31 @ A[0][1] ^= *inp++
+ beq .Lprocess_neon
+ vld1.8 {d31}, [r4]!
+ cmp r6, #8*4
+ veor d4, d4, d31 @ A[0][2] ^= *inp++
+ blo .Lprocess_neon
+ vld1.8 {d31}, [r4]!
+ veor d6, d6, d31 @ A[0][3] ^= *inp++
+ beq .Lprocess_neon
+ vld1.8 {d31},[r4]!
+ cmp r6, #8*6
+ veor d8, d8, d31 @ A[0][4] ^= *inp++
+ blo .Lprocess_neon
+
+ vld1.8 {d31}, [r4]!
+ veor d1, d1, d31 @ A[1][0] ^= *inp++
+ beq .Lprocess_neon
+ vld1.8 {d31}, [r4]!
+ cmp r6, #8*8
+ veor d3, d3, d31 @ A[1][1] ^= *inp++
+ blo .Lprocess_neon
+ vld1.8 {d31}, [r4]!
+ veor d5, d5, d31 @ A[1][2] ^= *inp++
+ beq .Lprocess_neon
+ vld1.8 {d31}, [r4]!
+ cmp r6, #8*10
+ veor d7, d7, d31 @ A[1][3] ^= *inp++
+ blo .Lprocess_neon
+ vld1.8 {d31}, [r4]!
+ veor d9, d9, d31 @ A[1][4] ^= *inp++
+ beq .Lprocess_neon
+
+ vld1.8 {d31}, [r4]!
+ cmp r6, #8*12
+ veor d10, d10, d31 @ A[2][0] ^= *inp++
+ blo .Lprocess_neon
+ vld1.8 {d31}, [r4]!
+ veor d12, d12, d31 @ A[2][1] ^= *inp++
+ beq .Lprocess_neon
+ vld1.8 {d31}, [r4]!
+ cmp r6, #8*14
+ veor d14, d14, d31 @ A[2][2] ^= *inp++
+ blo .Lprocess_neon
+ vld1.8 {d31}, [r4]!
+ veor d16, d16, d31 @ A[2][3] ^= *inp++
+ beq .Lprocess_neon
+ vld1.8 {d31}, [r4]!
+ cmp r6, #8*16
+ veor d18, d18, d31 @ A[2][4] ^= *inp++
+ blo .Lprocess_neon
+
+ vld1.8 {d31}, [r4]!
+ veor d11, d11, d31 @ A[3][0] ^= *inp++
+ beq .Lprocess_neon
+ vld1.8 {d31}, [r4]!
+ cmp r6, #8*18
+ veor d13, d13, d31 @ A[3][1] ^= *inp++
+ blo .Lprocess_neon
+ vld1.8 {d31}, [r4]!
+ veor d15, d15, d31 @ A[3][2] ^= *inp++
+ beq .Lprocess_neon
+ vld1.8 {d31}, [r4]!
+ cmp r6, #8*20
+ veor d17, d17, d31 @ A[3][3] ^= *inp++
+ blo .Lprocess_neon
+ vld1.8 {d31}, [r4]!
+ veor d19, d19, d31 @ A[3][4] ^= *inp++
+ beq .Lprocess_neon
+
+ vld1.8 {d31}, [r4]!
+ cmp r6, #8*22
+ veor d20, d20, d31 @ A[4][0] ^= *inp++
+ blo .Lprocess_neon
+ vld1.8 {d31}, [r4]!
+ veor d21, d21, d31 @ A[4][1] ^= *inp++
+ beq .Lprocess_neon
+ vld1.8 {d31}, [r4]!
+ cmp r6, #8*24
+ veor d22, d22, d31 @ A[4][2] ^= *inp++
+ blo .Lprocess_neon
+ vld1.8 {d31}, [r4]!
+ veor d23, d23, d31 @ A[4][3] ^= *inp++
+ beq .Lprocess_neon
+ vld1.8 {d31}, [r4]!
+ veor d24, d24, d31 @ A[4][4] ^= *inp++
+
+.Lprocess_neon:
+ bl KeccakF1600_neon
+ b .Loop_absorb_neon
+
+.align 4
+.Labsorbed_neon:
+ vst1.32 {d0}, [r0:64]! @ A[0][0..4]
+ vst1.32 {d2}, [r0:64]!
+ vst1.32 {d4}, [r0:64]!
+ vst1.32 {d6}, [r0:64]!
+ vst1.32 {d8}, [r0:64]!
+
+ vst1.32 {d1}, [r0:64]! @ A[1][0..4]
+ vst1.32 {d3}, [r0:64]!
+ vst1.32 {d5}, [r0:64]!
+ vst1.32 {d7}, [r0:64]!
+ vst1.32 {d9}, [r0:64]!
+
+ vst1.32 {d10}, [r0:64]! @ A[2][0..4]
+ vst1.32 {d12}, [r0:64]!
+ vst1.32 {d14}, [r0:64]!
+ vst1.32 {d16}, [r0:64]!
+ vst1.32 {d18}, [r0:64]!
+
+ vst1.32 {d11}, [r0:64]! @ A[3][0..4]
+ vst1.32 {d13}, [r0:64]!
+ vst1.32 {d15}, [r0:64]!
+ vst1.32 {d17}, [r0:64]!
+ vst1.32 {d19}, [r0:64]!
+
+ vst1.32 {d20-d23}, [r0:64]! @ A[4][0..4]
+ vst1.32 {d24}, [r0:64]
+
+ mov r0, r5 @ return value
+ vldmia sp!, {d8-d15}
+ ldmia sp!, {r4-r6,pc}
+.size SHA3_absorb_neon,.-SHA3_absorb_neon
+
+.global SHA3_squeeze_neon
+.type SHA3_squeeze_neon, %function
+.align 5
+SHA3_squeeze_neon:
+ stmdb sp!, {r4-r6,lr}
+
+ mov r4, r1 @ out
+ mov r5, r2 @ len
+ mov r6, r3 @ bsz
+ mov r12, r0 @ A_flat
+ mov r14, r3 @ bsz
+ b .Loop_squeeze_neon
+
+.align 4
+.Loop_squeeze_neon:
+ cmp r5, #8
+ blo .Lsqueeze_neon_tail
+ vld1.32 {d0}, [r12]!
+ vst1.8 {d0}, [r4]! @ endian-neutral store
+
+ subs r5, r5, #8 @ len -= 8
+ beq .Lsqueeze_neon_done
+
+ subs r14, r14, #8 @ bsz -= 8
+ bhi .Loop_squeeze_neon
+
+ vstmdb sp!, {d8-d15}
+
+ vld1.32 {d0}, [r0:64]! @ A[0][0..4]
+ vld1.32 {d2}, [r0:64]!
+ vld1.32 {d4}, [r0:64]!
+ vld1.32 {d6}, [r0:64]!
+ vld1.32 {d8}, [r0:64]!
+
+ vld1.32 {d1}, [r0:64]! @ A[1][0..4]
+ vld1.32 {d3}, [r0:64]!
+ vld1.32 {d5}, [r0:64]!
+ vld1.32 {d7}, [r0:64]!
+ vld1.32 {d9}, [r0:64]!
+
+ vld1.32 {d10}, [r0:64]! @ A[2][0..4]
+ vld1.32 {d12}, [r0:64]!
+ vld1.32 {d14}, [r0:64]!
+ vld1.32 {d16}, [r0:64]!
+ vld1.32 {d18}, [r0:64]!
+
+ vld1.32 {d11}, [r0:64]! @ A[3][0..4]
+ vld1.32 {d13}, [r0:64]!
+ vld1.32 {d15}, [r0:64]!
+ vld1.32 {d17}, [r0:64]!
+ vld1.32 {d19}, [r0:64]!
+
+ vld1.32 {d20-d23}, [r0:64]! @ A[4][0..4]
+ vld1.32 {d24}, [r0:64]
+ sub r0, r0, #24*8 @ rewind
+
+ bl KeccakF1600_neon
+
+ mov r12, r0 @ A_flat
+ vst1.32 {d0}, [r0:64]! @ A[0][0..4]
+ vst1.32 {d2}, [r0:64]!
+ vst1.32 {d4}, [r0:64]!
+ vst1.32 {d6}, [r0:64]!
+ vst1.32 {d8}, [r0:64]!
+
+ vst1.32 {d1}, [r0:64]! @ A[1][0..4]
+ vst1.32 {d3}, [r0:64]!
+ vst1.32 {d5}, [r0:64]!
+ vst1.32 {d7}, [r0:64]!
+ vst1.32 {d9}, [r0:64]!
+
+ vst1.32 {d10}, [r0:64]! @ A[2][0..4]
+ vst1.32 {d12}, [r0:64]!
+ vst1.32 {d14}, [r0:64]!
+ vst1.32 {d16}, [r0:64]!
+ vst1.32 {d18}, [r0:64]!
+
+ vst1.32 {d11}, [r0:64]! @ A[3][0..4]
+ vst1.32 {d13}, [r0:64]!
+ vst1.32 {d15}, [r0:64]!
+ vst1.32 {d17}, [r0:64]!
+ vst1.32 {d19}, [r0:64]!
+
+ vst1.32 {d20-d23}, [r0:64]! @ A[4][0..4]
+ mov r14, r6 @ bsz
+ vst1.32 {d24}, [r0:64]
+ mov r0, r12 @ rewind
+
+ vldmia sp!, {d8-d15}
+ b .Loop_squeeze_neon
+
+.align 4
+.Lsqueeze_neon_tail:
+ ldmia r12, {r2,r3}
+ cmp r5, #2
+ strb r2, [r4],#1 @ endian-neutral store
+ lsr r2, r2, #8
+ blo .Lsqueeze_neon_done
+ strb r2, [r4], #1
+ lsr r2, r2, #8
+ beq .Lsqueeze_neon_done
+ strb r2, [r4], #1
+ lsr r2, r2, #8
+ cmp r5, #4
+ blo .Lsqueeze_neon_done
+ strb r2, [r4], #1
+ beq .Lsqueeze_neon_done
+
+ strb r3, [r4], #1
+ lsr r3, r3, #8
+ cmp r5, #6
+ blo .Lsqueeze_neon_done
+ strb r3, [r4], #1
+ lsr r3, r3, #8
+ beq .Lsqueeze_neon_done
+ strb r3, [r4], #1
+
+.Lsqueeze_neon_done:
+ ldmia sp!, {r4-r6,pc}
+.size SHA3_squeeze_neon,.-SHA3_squeeze_neon
+.asciz "Keccak-1600 absorb and squeeze for ARMv4/NEON, CRYPTOGAMS by <appro\@openssl.org>"
+.align 2
+___
+
print $code;
close STDOUT; # enforce flush
projects / openssl.git / blobdiff