#include <locale.h>
#include <math.h>
#include "proxmark3.h"
+#include "comms.h"
#include "cmdmain.h"
#include "ui.h"
#include "util.h"
#include "crapto1/crapto1.h"
#include "parity.h"
#include "hardnested/hardnested_bruteforce.h"
+#include "hardnested/hardnested_bf_core.h"
#include "hardnested/hardnested_bitarray_core.h"
#include "zlib.h"
static void get_SIMD_instruction_set(char* instruction_set) {
-#if defined (__i386__) || defined (__x86_64__)
- #if !defined(__APPLE__) || (defined(__APPLE__) && (__clang_major__ > 8))
- #if (__GNUC__ >= 5) && (__GNUC__ > 5 || __GNUC_MINOR__ > 2)
- if (__builtin_cpu_supports("avx512f")) strcpy(instruction_set, "AVX512F");
- else if (__builtin_cpu_supports("avx2")) strcpy(instruction_set, "AVX2");
- #else
- if (__builtin_cpu_supports("avx2")) strcpy(instruction_set, "AVX2");
- #endif
- else if (__builtin_cpu_supports("avx")) strcpy(instruction_set, "AVX");
- else if (__builtin_cpu_supports("sse2")) strcpy(instruction_set, "SSE2");
- else if (__builtin_cpu_supports("mmx")) strcpy(instruction_set, "MMX");
- else
- #endif
-#endif
- strcpy(instruction_set, "no");
+ switch(GetSIMDInstrAuto()) {
+ case SIMD_AVX512:
+ strcpy(instruction_set, "AVX512F");
+ break;
+ case SIMD_AVX2:
+ strcpy(instruction_set, "AVX2");
+ break;
+ case SIMD_AVX:
+ strcpy(instruction_set, "AVX");
+ break;
+ case SIMD_SSE2:
+ strcpy(instruction_set, "SSE2");
+ break;
+ case SIMD_MMX:
+ strcpy(instruction_set, "MMX");
+ break;
+ default:
+ strcpy(instruction_set, "no");
+ break;
+ }
}
static void print_progress_header(void) {
char progress_text[80];
- char instr_set[12] = "";
+ char instr_set[12] = {0};
get_SIMD_instruction_set(instr_set);
sprintf(progress_text, "Start using %d threads and %s SIMD core", num_CPUs(), instr_set);
PrintAndLog("\n\n");
}
-static inline void clear_bit24(uint32_t *bitarray, uint32_t index)
-{
- bitarray[index>>5] &= ~(0x80000000>>(index&0x0000001f));
-}
-
-
static inline uint32_t test_bit24(uint32_t *bitarray, uint32_t index)
{
return bitarray[index>>5] & (0x80000000>>(index&0x0000001f));
}
-static inline uint32_t next_not_state(uint32_t *bitarray, uint32_t state)
-{
- if (++state == 1<<24) return 1<<24;
- uint32_t index = state >> 5;
- uint_fast8_t bit = state & 0x1f;
- uint32_t line = bitarray[index] << bit;
- while (bit <= 0x1f) {
- if ((line & 0x80000000) == 0) return state;
- state++;
- bit++;
- line <<= 1;
- }
- index++;
- while (bitarray[index] == 0xffffffff && state < 1<<24) {
- index++;
- state += 0x20;
- }
- if (state >= 1<<24) return 1<<24;
-#if defined __GNUC__
- return state + __builtin_clz(~bitarray[index]);
-#else
- bit = 0x00;
- line = bitarray[index];
- while (bit <= 0x1f) {
- if ((line & 0x80000000) == 0) return state;
- state++;
- bit++;
- line <<= 1;
- }
- return 1<<24;
-#endif
-}
-
-
#define BITFLIP_2ND_BYTE 0x0200
if (bytesread != filesize) {
printf("File read error with %s. Aborting...\n", state_file_name);
fclose(statesfile);
- inflateEnd(&compressed_stream);
exit(5);
}
fclose(statesfile);
}
-static void *check_for_BitFlipProperties_thread(void *args)
+static void
+#ifdef __has_attribute
+#if __has_attribute(force_align_arg_pointer)
+__attribute__((force_align_arg_pointer))
+#endif
+#endif
+*check_for_BitFlipProperties_thread(void *args)
{
uint8_t first_byte = ((uint8_t *)args)[0];
uint8_t last_byte = ((uint8_t *)args)[1];
}
-static void *generate_candidates_worker_thread(void *args)
+static void
+#ifdef __has_attribute
+#if __has_attribute(force_align_arg_pointer)
+__attribute__((force_align_arg_pointer))
+#endif
+#endif
+*generate_candidates_worker_thread(void *args)
{
uint16_t *sum_args = (uint16_t *)args;
uint16_t sum_a0 = sums[sum_args[0]];
int mfnestedhard(uint8_t blockNo, uint8_t keyType, uint8_t *key, uint8_t trgBlockNo, uint8_t trgKeyType, uint8_t *trgkey, bool nonce_file_read, bool nonce_file_write, bool slow, int tests)
{
char progress_text[80];
+
+ char instr_set[12] = {0};
+ get_SIMD_instruction_set(instr_set);
+ PrintAndLog("Using %s SIMD core.", instr_set);
srand((unsigned) time(NULL));
brute_force_per_second = brute_force_benchmark();