// Registers used to hold message words temporarily. There aren't
// enough ARM registers to hold the whole message block, so we have to
// load the words on-demand.
M_0 .req r12
M_1 .req r14
// Execute a quarter-round of BLAKE2s by mixing two columns or two diagonals.
// (a0, b0, c0, d0) and (a1, b1, c1, d1) give the registers containing the two
// columns/diagonals. s0-s1 are the word offsets to the message words the first
// column/diagonal needs, and likewise s2-s3 for the second column/diagonal.
// M_0 and M_1 are free to use, and the message block can be found at sp + 32.
//
// Note that to save instructions, the rotations don't happen when the
// pseudocode says they should, but rather they are delayed until the values are
// used. See the comment above _blake2s_round().
.macro _blake2s_quarterround a0, b0, c0, d0, a1, b1, c1, d1, s0, s1, s2, s3
// Execute one round of BLAKE2s by updating the state matrix v[0..15]. v[0..9]
// are in r0..r9. The stack pointer points to 8 bytes of scratch space for
// spilling v[8..9], then to v[9..15], then to the message block. r10-r12 and
// r14 are free to use. The macro arguments s0-s15 give the order in which the
// message words are used in this round.
//
// All rotates are performed using the implicit rotate operand accepted by the
// 'add' and 'eor' instructions. This is faster than using explicit rotate
// instructions. To make this work, we allow the values in the second and last
// rows of the BLAKE2s state matrix (rows 'b' and 'd') to temporarily have the
// wrong rotation amount. The rotation amount is then fixed up just in time
// when the values are used. 'brot' is the number of bits the values in row 'b'
// need to be rotated right to arrive at the correct values, and 'drot'
// similarly for row 'd'. (brot, drot) start out as (0, 0) but we make it such
// that they end up as (7, 8) after every round.
.macro _blake2s_round s0, s1, s2, s3, s4, s5, s6, s7, \
s8, s9, s10, s11, s12, s13, s14, s15
// Mix second two columns:
// (v[2], v[6], v[10], v[14]) and (v[3], v[7], v[11], v[15]).
__ldrd r8, r9, sp, 8 // load v[10] and v[11]
__ldrd r10, r11, sp, 24 // load v[14] and v[15]
_blake2s_quarterround r2, r6, r8, r10, r3, r7, r9, r11, \
\s4, \s5, \s6, \s7
str r10, [sp, #24] // store v[14]
// v[10], v[11], and v[15] are used below, so no need to store them yet.
//
// void blake2s_compress(struct blake2s_state *state,
// const u8 *block, size_t nblocks, u32 inc);
//
// Only the first three fields of struct blake2s_state are used:
// u32 h[8]; (inout)
// u32 t[2]; (inout)
// u32 f[2]; (in)
//
.align 5
ENTRY(blake2s_compress)
push {r0-r2,r4-r11,lr} // keep this an even number
.Lnext_block:
// r0 is 'state'
// r1 is 'block'
// r3 is 'inc'
// _blake2s_round is very short on registers, so copy the message block
// to the stack to save a register during the rounds. This also has the
// advantage that misalignment only needs to be dealt with in one place. sub sp, sp, #64
mov r12, sp
tst r1, #3
bne .Lcopy_block_misaligned
ldmia r1!, {r2-r9}
_le32_bswap_8x r2, r3, r4, r5, r6, r7, r8, r9, r14
stmia r12!, {r2-r9}
ldmia r1!, {r2-r9}
_le32_bswap_8x r2, r3, r4, r5, r6, r7, r8, r9, r14
stmia r12, {r2-r9}
.Lcopy_block_done:
str r1, [sp, #68] // Update message pointer
// Advance to the next block, if there is one. Note that if there are
// multiple blocks, then 'inc' (the counter increment amount) must be
// 64. So we can simply set it to 64 without re-loading it.
ldm sp, {r0, r1, r2} // load (state, block, nblocks)
mov r3, #64 // set'inc'
subs r2, r2, #1 // nblocks--
str r2, [sp, #8]
bne .Lnext_block // nblocks != 0?
pop {r0-r2,r4-r11,pc}
// The next message block (pointed to by r1) isn't 4-byte aligned, so it
// can't be loaded using ldmia. Copy it to the stack buffer (pointed to
// by r12) using an alternative method. r2-r9 are free to use.
.Lcopy_block_misaligned:
mov r2, #64
1:
#ifdef CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS
ldr r3, [r1], #4
_le32_bswap r3, r4
#else
ldrb r3, [r1, #0]
ldrb r4, [r1, #1]
ldrb r5, [r1, #2]
ldrb r6, [r1, #3]
add r1, r1, #4
orr r3, r3, r4, lsl #8
orr r3, r3, r5, lsl #16
orr r3, r3, r6, lsl #24
#endif
subs r2, r2, #4
str r3, [r12], #4
bne 1b
b .Lcopy_block_done
ENDPROC(blake2s_compress)
Messung V0.5 in Prozent
¤ Dauer der Verarbeitung: 0.16 Sekunden
(vorverarbeitet am 2026-04-29)
¤
Die Informationen auf dieser Webseite wurden
nach bestem Wissen sorgfältig zusammengestellt. Es wird jedoch weder Vollständigkeit, noch Richtigkeit,
noch Qualität der bereit gestellten Informationen zugesichert.
Bemerkung:
Die farbliche Syntaxdarstellung und die Messung sind noch experimentell.