QortalOS Brooklyn for Raspberry Pi 4
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
 
 
 
 
 
 

1003 lines
23 KiB

/* $Id: radiogatun.c 226 2010-06-16 17:28:08Z tp $ */
/*
* RadioGatun implementation.
*
* ==========================(LICENSE BEGIN)============================
*
* Copyright (c) 2007-2010 Projet RNRT SAPHIR
*
* Permission is hereby granted, free of charge, to any person obtaining
* a copy of this software and associated documentation files (the
* "Software"), to deal in the Software without restriction, including
* without limitation the rights to use, copy, modify, merge, publish,
* distribute, sublicense, and/or sell copies of the Software, and to
* permit persons to whom the Software is furnished to do so, subject to
* the following conditions:
*
* The above copyright notice and this permission notice shall be
* included in all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
* IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY
* CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
* TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
* SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
*
* ===========================(LICENSE END)=============================
*
* @author Thomas Pornin <[email protected]>
*/
#include <stddef.h>
#include <string.h>
#include "sph_radiogatun.h"
#if SPH_SMALL_FOOTPRINT && !defined SPH_SMALL_FOOTPRINT_RADIOGATUN
#define SPH_SMALL_FOOTPRINT_RADIOGATUN 1
#endif
/* ======================================================================= */
/*
* The core macros. We want to unroll 13 successive rounds so that the
* belt rotation becomes pure routing, solved at compilation time, with
* no unnecessary copying. We also wish all state variables to be
* independant local variables, so that the C compiler becomes free to
* map these on registers at it sees fit. This requires some heavy
* preprocessor trickeries, including a full addition macro modulo 13.
*
* These macros are size-independent. Some macros must be defined before
* use:
* WT evaluates to the type for a word (32-bit or 64-bit)
* T truncates a value to the proper word size
* ROR(x, n) right rotation of a word x, with explicit modular
* reduction of the rotation count n by the word size
* INW(i, j) input word j (0, 1, or 2) of block i (0 to 12)
*
* For INW, the input buffer is pointed to by "buf" which has type
* "const unsigned char *".
*/
#define MUL19(action) do { \
action(0); \
action(1); \
action(2); \
action(3); \
action(4); \
action(5); \
action(6); \
action(7); \
action(8); \
action(9); \
action(10); \
action(11); \
action(12); \
action(13); \
action(14); \
action(15); \
action(16); \
action(17); \
action(18); \
} while (0)
#define DECL19(b) b ## 0, b ## 1, b ## 2, b ## 3, b ## 4, b ## 5, \
b ## 6, b ## 7, b ## 8, b ## 9, b ## 10, b ## 11, \
b ## 12, b ## 13, b ## 14, b ## 15, b ## 16, \
b ## 17, b ## 18
#define M19_T7(i) M19_T7_(i)
#define M19_T7_(i) M19_T7_ ## i
#define M19_T7_0 0
#define M19_T7_1 7
#define M19_T7_2 14
#define M19_T7_3 2
#define M19_T7_4 9
#define M19_T7_5 16
#define M19_T7_6 4
#define M19_T7_7 11
#define M19_T7_8 18
#define M19_T7_9 6
#define M19_T7_10 13
#define M19_T7_11 1
#define M19_T7_12 8
#define M19_T7_13 15
#define M19_T7_14 3
#define M19_T7_15 10
#define M19_T7_16 17
#define M19_T7_17 5
#define M19_T7_18 12
#define M19_A1(i) M19_A1_(i)
#define M19_A1_(i) M19_A1_ ## i
#define M19_A1_0 1
#define M19_A1_1 2
#define M19_A1_2 3
#define M19_A1_3 4
#define M19_A1_4 5
#define M19_A1_5 6
#define M19_A1_6 7
#define M19_A1_7 8
#define M19_A1_8 9
#define M19_A1_9 10
#define M19_A1_10 11
#define M19_A1_11 12
#define M19_A1_12 13
#define M19_A1_13 14
#define M19_A1_14 15
#define M19_A1_15 16
#define M19_A1_16 17
#define M19_A1_17 18
#define M19_A1_18 0
#define M19_A2(i) M19_A2_(i)
#define M19_A2_(i) M19_A2_ ## i
#define M19_A2_0 2
#define M19_A2_1 3
#define M19_A2_2 4
#define M19_A2_3 5
#define M19_A2_4 6
#define M19_A2_5 7
#define M19_A2_6 8
#define M19_A2_7 9
#define M19_A2_8 10
#define M19_A2_9 11
#define M19_A2_10 12
#define M19_A2_11 13
#define M19_A2_12 14
#define M19_A2_13 15
#define M19_A2_14 16
#define M19_A2_15 17
#define M19_A2_16 18
#define M19_A2_17 0
#define M19_A2_18 1
#define M19_A4(i) M19_A4_(i)
#define M19_A4_(i) M19_A4_ ## i
#define M19_A4_0 4
#define M19_A4_1 5
#define M19_A4_2 6
#define M19_A4_3 7
#define M19_A4_4 8
#define M19_A4_5 9
#define M19_A4_6 10
#define M19_A4_7 11
#define M19_A4_8 12
#define M19_A4_9 13
#define M19_A4_10 14
#define M19_A4_11 15
#define M19_A4_12 16
#define M19_A4_13 17
#define M19_A4_14 18
#define M19_A4_15 0
#define M19_A4_16 1
#define M19_A4_17 2
#define M19_A4_18 3
#define ACC_a(i) ACC_a_(i)
#define ACC_a_(i) a ## i
#define ACC_atmp(i) ACC_atmp_(i)
#define ACC_atmp_(i) atmp ## i
#define MILL1(i) (atmp ## i = a ## i ^ T(ACC_a(M19_A1(i)) \
| ~ACC_a(M19_A2(i))))
#define MILL2(i) (a ## i = ROR(ACC_atmp(M19_T7(i)), ((i * (i + 1)) >> 1)))
#define MILL3(i) (atmp ## i = a ## i ^ ACC_a(M19_A1(i)) ^ ACC_a(M19_A4(i)))
#define MILL4(i) (a ## i = atmp ## i ^ (i == 0))
#define MILL do { \
WT DECL19(atmp); \
MUL19(MILL1); \
MUL19(MILL2); \
MUL19(MILL3); \
MUL19(MILL4); \
} while (0)
#define DECL13(b) b ## 0 ## _0, b ## 0 ## _1, b ## 0 ## _2, \
b ## 1 ## _0, b ## 1 ## _1, b ## 1 ## _2, \
b ## 2 ## _0, b ## 2 ## _1, b ## 2 ## _2, \
b ## 3 ## _0, b ## 3 ## _1, b ## 3 ## _2, \
b ## 4 ## _0, b ## 4 ## _1, b ## 4 ## _2, \
b ## 5 ## _0, b ## 5 ## _1, b ## 5 ## _2, \
b ## 6 ## _0, b ## 6 ## _1, b ## 6 ## _2, \
b ## 7 ## _0, b ## 7 ## _1, b ## 7 ## _2, \
b ## 8 ## _0, b ## 8 ## _1, b ## 8 ## _2, \
b ## 9 ## _0, b ## 9 ## _1, b ## 9 ## _2, \
b ## 10 ## _0, b ## 10 ## _1, b ## 10 ## _2, \
b ## 11 ## _0, b ## 11 ## _1, b ## 11 ## _2, \
b ## 12 ## _0, b ## 12 ## _1, b ## 12 ## _2
#define M13_A(i, j) M13_A_(i, j)
#define M13_A_(i, j) M13_A_ ## i ## _ ## j
#define M13_A_0_0 0
#define M13_A_0_1 1
#define M13_A_0_2 2
#define M13_A_0_3 3
#define M13_A_0_4 4
#define M13_A_0_5 5
#define M13_A_0_6 6
#define M13_A_0_7 7
#define M13_A_0_8 8
#define M13_A_0_9 9
#define M13_A_0_10 10
#define M13_A_0_11 11
#define M13_A_0_12 12
#define M13_A_1_0 1
#define M13_A_1_1 2
#define M13_A_1_2 3
#define M13_A_1_3 4
#define M13_A_1_4 5
#define M13_A_1_5 6
#define M13_A_1_6 7
#define M13_A_1_7 8
#define M13_A_1_8 9
#define M13_A_1_9 10
#define M13_A_1_10 11
#define M13_A_1_11 12
#define M13_A_1_12 0
#define M13_A_2_0 2
#define M13_A_2_1 3
#define M13_A_2_2 4
#define M13_A_2_3 5
#define M13_A_2_4 6
#define M13_A_2_5 7
#define M13_A_2_6 8
#define M13_A_2_7 9
#define M13_A_2_8 10
#define M13_A_2_9 11
#define M13_A_2_10 12
#define M13_A_2_11 0
#define M13_A_2_12 1
#define M13_A_3_0 3
#define M13_A_3_1 4
#define M13_A_3_2 5
#define M13_A_3_3 6
#define M13_A_3_4 7
#define M13_A_3_5 8
#define M13_A_3_6 9
#define M13_A_3_7 10
#define M13_A_3_8 11
#define M13_A_3_9 12
#define M13_A_3_10 0
#define M13_A_3_11 1
#define M13_A_3_12 2
#define M13_A_4_0 4
#define M13_A_4_1 5
#define M13_A_4_2 6
#define M13_A_4_3 7
#define M13_A_4_4 8
#define M13_A_4_5 9
#define M13_A_4_6 10
#define M13_A_4_7 11
#define M13_A_4_8 12
#define M13_A_4_9 0
#define M13_A_4_10 1
#define M13_A_4_11 2
#define M13_A_4_12 3
#define M13_A_5_0 5
#define M13_A_5_1 6
#define M13_A_5_2 7
#define M13_A_5_3 8
#define M13_A_5_4 9
#define M13_A_5_5 10
#define M13_A_5_6 11
#define M13_A_5_7 12
#define M13_A_5_8 0
#define M13_A_5_9 1
#define M13_A_5_10 2
#define M13_A_5_11 3
#define M13_A_5_12 4
#define M13_A_6_0 6
#define M13_A_6_1 7
#define M13_A_6_2 8
#define M13_A_6_3 9
#define M13_A_6_4 10
#define M13_A_6_5 11
#define M13_A_6_6 12
#define M13_A_6_7 0
#define M13_A_6_8 1
#define M13_A_6_9 2
#define M13_A_6_10 3
#define M13_A_6_11 4
#define M13_A_6_12 5
#define M13_A_7_0 7
#define M13_A_7_1 8
#define M13_A_7_2 9
#define M13_A_7_3 10
#define M13_A_7_4 11
#define M13_A_7_5 12
#define M13_A_7_6 0
#define M13_A_7_7 1
#define M13_A_7_8 2
#define M13_A_7_9 3
#define M13_A_7_10 4
#define M13_A_7_11 5
#define M13_A_7_12 6
#define M13_A_8_0 8
#define M13_A_8_1 9
#define M13_A_8_2 10
#define M13_A_8_3 11
#define M13_A_8_4 12
#define M13_A_8_5 0
#define M13_A_8_6 1
#define M13_A_8_7 2
#define M13_A_8_8 3
#define M13_A_8_9 4
#define M13_A_8_10 5
#define M13_A_8_11 6
#define M13_A_8_12 7
#define M13_A_9_0 9
#define M13_A_9_1 10
#define M13_A_9_2 11
#define M13_A_9_3 12
#define M13_A_9_4 0
#define M13_A_9_5 1
#define M13_A_9_6 2
#define M13_A_9_7 3
#define M13_A_9_8 4
#define M13_A_9_9 5
#define M13_A_9_10 6
#define M13_A_9_11 7
#define M13_A_9_12 8
#define M13_A_10_0 10
#define M13_A_10_1 11
#define M13_A_10_2 12
#define M13_A_10_3 0
#define M13_A_10_4 1
#define M13_A_10_5 2
#define M13_A_10_6 3
#define M13_A_10_7 4
#define M13_A_10_8 5
#define M13_A_10_9 6
#define M13_A_10_10 7
#define M13_A_10_11 8
#define M13_A_10_12 9
#define M13_A_11_0 11
#define M13_A_11_1 12
#define M13_A_11_2 0
#define M13_A_11_3 1
#define M13_A_11_4 2
#define M13_A_11_5 3
#define M13_A_11_6 4
#define M13_A_11_7 5
#define M13_A_11_8 6
#define M13_A_11_9 7
#define M13_A_11_10 8
#define M13_A_11_11 9
#define M13_A_11_12 10
#define M13_A_12_0 12
#define M13_A_12_1 0
#define M13_A_12_2 1
#define M13_A_12_3 2
#define M13_A_12_4 3
#define M13_A_12_5 4
#define M13_A_12_6 5
#define M13_A_12_7 6
#define M13_A_12_8 7
#define M13_A_12_9 8
#define M13_A_12_10 9
#define M13_A_12_11 10
#define M13_A_12_12 11
#define M13_N(i) M13_N_(i)
#define M13_N_(i) M13_N_ ## i
#define M13_N_0 12
#define M13_N_1 11
#define M13_N_2 10
#define M13_N_3 9
#define M13_N_4 8
#define M13_N_5 7
#define M13_N_6 6
#define M13_N_7 5
#define M13_N_8 4
#define M13_N_9 3
#define M13_N_10 2
#define M13_N_11 1
#define M13_N_12 0
#define ACC_b(i, k) ACC_b_(i, k)
#define ACC_b_(i, k) b ## i ## _ ## k
#define ROUND_ELT(k, s) do { \
if ((bj += 3) == 39) \
bj = 0; \
sc->b[bj + s] ^= a ## k; \
} while (0)
#define ROUND_SF(j) do { \
size_t bj = (j) * 3; \
ROUND_ELT(1, 0); \
ROUND_ELT(2, 1); \
ROUND_ELT(3, 2); \
ROUND_ELT(4, 0); \
ROUND_ELT(5, 1); \
ROUND_ELT(6, 2); \
ROUND_ELT(7, 0); \
ROUND_ELT(8, 1); \
ROUND_ELT(9, 2); \
ROUND_ELT(10, 0); \
ROUND_ELT(11, 1); \
ROUND_ELT(12, 2); \
MILL; \
bj = (j) * 3; \
a ## 13 ^= sc->b[bj + 0]; \
a ## 14 ^= sc->b[bj + 1]; \
a ## 15 ^= sc->b[bj + 2]; \
} while (0)
#define INPUT_SF(j, p0, p1, p2) do { \
size_t bj = ((j) + 1) * 3; \
if (bj == 39) \
bj = 0; \
sc->b[bj + 0] ^= (p0); \
sc->b[bj + 1] ^= (p1); \
sc->b[bj + 2] ^= (p2); \
a16 ^= (p0); \
a17 ^= (p1); \
a18 ^= (p2); \
} while (0)
#if SPH_SMALL_FOOTPRINT_RADIOGATUN
#define ROUND ROUND_SF
#define INPUT INPUT_SF
#else
/*
* Round function R, on base j. The value j is such that B[0] is actually
* b[j] after the initial rotation. On the 13-round macro, j has the
* successive values 12, 11, 10... 1, 0.
*/
#define ROUND(j) do { \
ACC_b(M13_A(1, j), 0) ^= a ## 1; \
ACC_b(M13_A(2, j), 1) ^= a ## 2; \
ACC_b(M13_A(3, j), 2) ^= a ## 3; \
ACC_b(M13_A(4, j), 0) ^= a ## 4; \
ACC_b(M13_A(5, j), 1) ^= a ## 5; \
ACC_b(M13_A(6, j), 2) ^= a ## 6; \
ACC_b(M13_A(7, j), 0) ^= a ## 7; \
ACC_b(M13_A(8, j), 1) ^= a ## 8; \
ACC_b(M13_A(9, j), 2) ^= a ## 9; \
ACC_b(M13_A(10, j), 0) ^= a ## 10; \
ACC_b(M13_A(11, j), 1) ^= a ## 11; \
ACC_b(M13_A(12, j), 2) ^= a ## 12; \
MILL; \
a ## 13 ^= ACC_b(j, 0); \
a ## 14 ^= ACC_b(j, 1); \
a ## 15 ^= ACC_b(j, 2); \
} while (0)
#define INPUT(j, p0, p1, p2) do { \
ACC_b(M13_A(1, j), 0) ^= (p0); \
ACC_b(M13_A(1, j), 1) ^= (p1); \
ACC_b(M13_A(1, j), 2) ^= (p2); \
a16 ^= (p0); \
a17 ^= (p1); \
a18 ^= (p2); \
} while (0)
#endif
#define MUL13(action) do { \
action(0); \
action(1); \
action(2); \
action(3); \
action(4); \
action(5); \
action(6); \
action(7); \
action(8); \
action(9); \
action(10); \
action(11); \
action(12); \
} while (0)
#define MILL_READ_ELT(i) do { \
a ## i = sc->a[i]; \
} while (0)
#define MILL_WRITE_ELT(i) do { \
sc->a[i] = a ## i; \
} while (0)
#define STATE_READ_SF do { \
MUL19(MILL_READ_ELT); \
} while (0)
#define STATE_WRITE_SF do { \
MUL19(MILL_WRITE_ELT); \
} while (0)
#define PUSH13_SF do { \
WT DECL19(a); \
const unsigned char *buf; \
\
buf = data; \
STATE_READ_SF; \
while (len >= sizeof sc->data) { \
size_t mk; \
for (mk = 13; mk > 0; mk --) { \
WT p0 = INW(0, 0); \
WT p1 = INW(0, 1); \
WT p2 = INW(0, 2); \
INPUT_SF(mk - 1, p0, p1, p2); \
ROUND_SF(mk - 1); \
buf += (sizeof sc->data) / 13; \
len -= (sizeof sc->data) / 13; \
} \
} \
STATE_WRITE_SF; \
return len; \
} while (0)
#if SPH_SMALL_FOOTPRINT_RADIOGATUN
#define STATE_READ STATE_READ_SF
#define STATE_WRITE STATE_WRITE_SF
#define PUSH13 PUSH13_SF
#else
#define BELT_READ_ELT(i) do { \
b ## i ## _0 = sc->b[3 * i + 0]; \
b ## i ## _1 = sc->b[3 * i + 1]; \
b ## i ## _2 = sc->b[3 * i + 2]; \
} while (0)
#define BELT_WRITE_ELT(i) do { \
sc->b[3 * i + 0] = b ## i ## _0; \
sc->b[3 * i + 1] = b ## i ## _1; \
sc->b[3 * i + 2] = b ## i ## _2; \
} while (0)
#define STATE_READ do { \
MUL13(BELT_READ_ELT); \
MUL19(MILL_READ_ELT); \
} while (0)
#define STATE_WRITE do { \
MUL13(BELT_WRITE_ELT); \
MUL19(MILL_WRITE_ELT); \
} while (0)
/*
* Input data by chunks of 13*3 blocks. This is the body of the
* radiogatun32_push13() and radiogatun64_push13() functions.
*/
#define PUSH13 do { \
WT DECL19(a), DECL13(b); \
const unsigned char *buf; \
\
buf = data; \
STATE_READ; \
while (len >= sizeof sc->data) { \
WT p0, p1, p2; \
MUL13(PUSH13_ELT); \
buf += sizeof sc->data; \
len -= sizeof sc->data; \
} \
STATE_WRITE; \
return len; \
} while (0)
#define PUSH13_ELT(k) do { \
p0 = INW(k, 0); \
p1 = INW(k, 1); \
p2 = INW(k, 2); \
INPUT(M13_N(k), p0, p1, p2); \
ROUND(M13_N(k)); \
} while (0)
#endif
#define BLANK13_SF do { \
size_t mk = 13; \
while (mk -- > 0) \
ROUND_SF(mk); \
} while (0)
#define BLANK1_SF do { \
WT tmp0, tmp1, tmp2; \
ROUND_SF(12); \
tmp0 = sc->b[36]; \
tmp1 = sc->b[37]; \
tmp2 = sc->b[38]; \
memmove(sc->b + 3, sc->b, 36 * sizeof sc->b[0]); \
sc->b[0] = tmp0; \
sc->b[1] = tmp1; \
sc->b[2] = tmp2; \
} while (0)
#if SPH_SMALL_FOOTPRINT_RADIOGATUN
#define BLANK13 BLANK13_SF
#define BLANK1 BLANK1_SF
#else
/*
* Run 13 blank rounds. This macro expects the "a" and "b" state variables
* to be alread declared.
*/
#define BLANK13 MUL13(BLANK13_ELT)
#define BLANK13_ELT(k) ROUND(M13_N(k))
#define MUL12(action) do { \
action(0); \
action(1); \
action(2); \
action(3); \
action(4); \
action(5); \
action(6); \
action(7); \
action(8); \
action(9); \
action(10); \
action(11); \
} while (0)
/*
* Run a single blank round, and physically rotate the belt. This is used
* for the last blank rounds, and the output rounds. This macro expects the
* "a" abd "b" state variables to be already declared.
*/
#define BLANK1 do { \
WT tmp0, tmp1, tmp2; \
ROUND(12); \
tmp0 = b0_0; \
tmp1 = b0_1; \
tmp2 = b0_2; \
MUL12(BLANK1_ELT); \
b1_0 = tmp0; \
b1_1 = tmp1; \
b1_2 = tmp2; \
} while (0)
#define BLANK1_ELT(i) do { \
ACC_b(M13_A(M13_N(i), 1), 0) = ACC_b(M13_N(i), 0); \
ACC_b(M13_A(M13_N(i), 1), 1) = ACC_b(M13_N(i), 1); \
ACC_b(M13_A(M13_N(i), 1), 2) = ACC_b(M13_N(i), 2); \
} while (0)
#endif
#define NO_TOKEN
/*
* Perform padding, then blank rounds, then output some words. This is
* the body of sph_radiogatun32_close() and sph_radiogatun64_close().
*/
#define CLOSE_SF(width) CLOSE_GEN(width, \
NO_TOKEN, STATE_READ_SF, BLANK1_SF, BLANK13_SF)
#if SPH_SMALL_FOOTPRINT_RADIOGATUN
#define CLOSE CLOSE_SF
#else
#define CLOSE(width) CLOSE_GEN(width, \
WT DECL13(b);, STATE_READ, BLANK1, BLANK13)
#endif
#define CLOSE_GEN(width, WTb13, state_read, blank1, blank13) do { \
unsigned ptr, num; \
unsigned char *out; \
WT DECL19(a); \
WTb13 \
\
ptr = sc->data_ptr; \
sc->data[ptr ++] = 0x01; \
memset(sc->data + ptr, 0, (sizeof sc->data) - ptr); \
radiogatun ## width ## _push13(sc, sc->data, sizeof sc->data); \
\
num = 17; \
for (;;) { \
ptr += 3 * (width >> 3); \
if (ptr > sizeof sc->data) \
break; \
num --; \
} \
\
state_read; \
if (num >= 13) { \
blank13; \
num -= 13; \
} \
while (num -- > 0) \
blank1; \
\
num = 0; \
out = dst; \
for (;;) { \
OUTW(out, a1); \
out += width >> 3; \
OUTW(out, a2); \
out += width >> 3; \
num += 2 * (width >> 3); \
if (num >= 32) \
break; \
blank1; \
} \
INIT; \
} while (0)
/*
* Initialize context structure.
*/
#if SPH_LITTLE_ENDIAN || SPH_BIG_ENDIAN
#define INIT do { \
memset(sc->a, 0, sizeof sc->a); \
memset(sc->b, 0, sizeof sc->b); \
sc->data_ptr = 0; \
} while (0)
#else
#define INIT do { \
size_t u; \
for (u = 0; u < 19; u ++) \
sc->a[u] = 0; \
for (u = 0; u < 39; u ++) \
sc->b[u] = 0; \
sc->data_ptr = 0; \
} while (0)
#endif
/* ======================================================================= */
/*
* RadioGatun[32].
*/
#if !SPH_NO_RG32
#undef WT
#define WT sph_u32
#undef T
#define T SPH_T32
#undef ROR
#define ROR(x, n) SPH_T32(((x) << ((32 - (n)) & 31)) | ((x) >> ((n) & 31)))
#undef INW
#define INW(i, j) sph_dec32le_aligned(buf + (4 * (3 * (i) + (j))))
#undef OUTW
#define OUTW(b, v) sph_enc32le(b, v)
/*
* Insert data by big chunks of 13*12 = 156 bytes. Returned value is the
* number of remaining bytes (between 0 and 155). This method assumes that
* the input data is suitably aligned.
*/
static size_t
radiogatun32_push13(sph_radiogatun32_context *sc, const void *data, size_t len)
{
PUSH13;
}
/* see sph_radiogatun.h */
void
sph_radiogatun32_init(void *cc)
{
sph_radiogatun32_context *sc;
sc = cc;
INIT;
}
#ifdef SPH_UPTR
static void
radiogatun32_short(void *cc, const void *data, size_t len)
#else
/* see sph_radiogatun.h */
void
sph_radiogatun32(void *cc, const void *data, size_t len)
#endif
{
sph_radiogatun32_context *sc;
unsigned ptr;
sc = cc;
ptr = sc->data_ptr;
while (len > 0) {
size_t clen;
clen = (sizeof sc->data) - ptr;
if (clen > len)
clen = len;
memcpy(sc->data + ptr, data, clen);
data = (const unsigned char *)data + clen;
len -= clen;
ptr += clen;
if (ptr == sizeof sc->data) {
radiogatun32_push13(sc, sc->data, sizeof sc->data);
ptr = 0;
}
}
sc->data_ptr = ptr;
}
#ifdef SPH_UPTR
/* see sph_radiogatun.h */
void
sph_radiogatun32(void *cc, const void *data, size_t len)
{
sph_radiogatun32_context *sc;
unsigned ptr;
size_t rlen;
if (len < (2 * sizeof sc->data)) {
radiogatun32_short(cc, data, len);
return;
}
sc = cc;
ptr = sc->data_ptr;
if (ptr > 0) {
unsigned t;
t = (sizeof sc->data) - ptr;
radiogatun32_short(sc, data, t);
data = (const unsigned char *)data + t;
len -= t;
}
#if !SPH_UNALIGNED
if (((SPH_UPTR)data & 3) != 0) {
radiogatun32_short(sc, data, len);
return;
}
#endif
rlen = radiogatun32_push13(sc, data, len);
memcpy(sc->data, (const unsigned char *)data + len - rlen, rlen);
sc->data_ptr = rlen;
}
#endif
/* see sph_radiogatun.h */
void
sph_radiogatun32_close(void *cc, void *dst)
{
sph_radiogatun32_context *sc;
sc = cc;
CLOSE(32);
}
#endif
/* ======================================================================= */
/*
* RadioGatun[64]. Compiled only if a 64-bit or more type is available.
*/
#if SPH_64
#if !SPH_NO_RG64
#undef WT
#define WT sph_u64
#undef T
#define T SPH_T64
#undef ROR
#define ROR(x, n) SPH_T64(((x) << ((64 - (n)) & 63)) | ((x) >> ((n) & 63)))
#undef INW
#define INW(i, j) sph_dec64le_aligned(buf + (8 * (3 * (i) + (j))))
#undef OUTW
#define OUTW(b, v) sph_enc64le(b, v)
/*
* On 32-bit x86, register pressure is such that using the small
* footprint version is a net gain (x2 speed), because that variant
* uses fewer local variables.
*/
#if SPH_I386_MSVC || SPH_I386_GCC || defined __i386__
#undef PUSH13
#define PUSH13 PUSH13_SF
#undef CLOSE
#define CLOSE CLOSE_SF
#endif
/*
* Insert data by big chunks of 13*24 = 312 bytes. Returned value is the
* number of remaining bytes (between 0 and 311). This method assumes that
* the input data is suitably aligned.
*/
static size_t
radiogatun64_push13(sph_radiogatun64_context *sc, const void *data, size_t len)
{
PUSH13;
}
/* see sph_radiogatun.h */
void
sph_radiogatun64_init(void *cc)
{
sph_radiogatun64_context *sc;
sc = cc;
INIT;
}
#ifdef SPH_UPTR
static void
radiogatun64_short(void *cc, const void *data, size_t len)
#else
/* see sph_radiogatun.h */
void
sph_radiogatun64(void *cc, const void *data, size_t len)
#endif
{
sph_radiogatun64_context *sc;
unsigned ptr;
sc = cc;
ptr = sc->data_ptr;
while (len > 0) {
size_t clen;
clen = (sizeof sc->data) - ptr;
if (clen > len)
clen = len;
memcpy(sc->data + ptr, data, clen);
data = (const unsigned char *)data + clen;
len -= clen;
ptr += clen;
if (ptr == sizeof sc->data) {
radiogatun64_push13(sc, sc->data, sizeof sc->data);
ptr = 0;
}
}
sc->data_ptr = ptr;
}
#ifdef SPH_UPTR
/* see sph_radiogatun.h */
void
sph_radiogatun64(void *cc, const void *data, size_t len)
{
sph_radiogatun64_context *sc;
unsigned ptr;
size_t rlen;
if (len < (2 * sizeof sc->data)) {
radiogatun64_short(cc, data, len);
return;
}
sc = cc;
ptr = sc->data_ptr;
if (ptr > 0) {
unsigned t;
t = (sizeof sc->data) - ptr;
radiogatun64_short(sc, data, t);
data = (const unsigned char *)data + t;
len -= t;
}
#if !SPH_UNALIGNED
if (((SPH_UPTR)data & 7) != 0) {
radiogatun64_short(sc, data, len);
return;
}
#endif
rlen = radiogatun64_push13(sc, data, len);
memcpy(sc->data, (const unsigned char *)data + len - rlen, rlen);
sc->data_ptr = rlen;
}
#endif
/* see sph_radiogatun.h */
void
sph_radiogatun64_close(void *cc, void *dst)
{
sph_radiogatun64_context *sc;
sc = cc;
CLOSE(64);
}
#endif
#endif