diff --git a/examples/crc/crc_combine_example.c b/examples/crc/crc_combine_example.c index ccabe04..a454c82 100644 --- a/examples/crc/crc_combine_example.c +++ b/examples/crc/crc_combine_example.c @@ -47,396 +47,412 @@ #include #include -int verbose; // Global for tests +int verbose; // Global for tests -#if defined (_MSC_VER) -# define __builtin_parity(x) (__popcnt64(x) & 1) +#if defined(_MSC_VER) +#define __builtin_parity(x) (__popcnt64(x) & 1) #endif -#if defined (__GNUC__) || defined (__clang__) -# define ATTRIBUTE_TARGET(x) __attribute__((target(x))) +#if defined(__GNUC__) || defined(__clang__) +#define ATTRIBUTE_TARGET(x) __attribute__((target(x))) #else -# define ATTRIBUTE_TARGET(x) +#define ATTRIBUTE_TARGET(x) #endif struct crc64_desc { - uint64_t poly; - uint64_t k5; - uint64_t k7; - uint64_t k8; + uint64_t poly; + uint64_t k5; + uint64_t k7; + uint64_t k8; }; -void gen_crc64_refl_consts(uint64_t poly, struct crc64_desc *c) +void +gen_crc64_refl_consts(uint64_t poly, struct crc64_desc *c) { - uint64_t quotienth = 0; - uint64_t div; - uint64_t rem = 1ull; - int i; + uint64_t quotienth = 0; + uint64_t div; + uint64_t rem = 1ull; + int i; - for (i = 0; i < 64; i++) { - div = (rem & 1ull) != 0; - quotienth = (quotienth >> 1) | (div ? 0x8000000000000000ull : 0); - rem = (div ? poly : 0) ^ (rem >> 1); - } - c->k5 = rem; - c->poly = poly; - c->k7 = quotienth; - c->k8 = poly << 1; + for (i = 0; i < 64; i++) { + div = (rem & 1ull) != 0; + quotienth = (quotienth >> 1) | (div ? 0x8000000000000000ull : 0); + rem = (div ? poly : 0) ^ (rem >> 1); + } + c->k5 = rem; + c->poly = poly; + c->k7 = quotienth; + c->k8 = poly << 1; } -void gen_crc64_norm_consts(uint64_t poly, struct crc64_desc *c) +void +gen_crc64_norm_consts(uint64_t poly, struct crc64_desc *c) { - uint64_t quotientl = 0; - uint64_t div; - uint64_t rem = 1ull << 63; - int i; + uint64_t quotientl = 0; + uint64_t div; + uint64_t rem = 1ull << 63; + int i; - for (i = 0; i < 65; i++) { - div = (rem & 0x8000000000000000ull) != 0; - quotientl = (quotientl << 1) | div; - rem = (div ? poly : 0) ^ (rem << 1); - } + for (i = 0; i < 65; i++) { + div = (rem & 0x8000000000000000ull) != 0; + quotientl = (quotientl << 1) | div; + rem = (div ? poly : 0) ^ (rem << 1); + } - c->poly = poly; - c->k5 = rem; - c->k7 = quotientl; - c->k8 = poly; + c->poly = poly; + c->k5 = rem; + c->k7 = quotientl; + c->k8 = poly; } -uint32_t calc_xi_mod(int n) +uint32_t +calc_xi_mod(int n) { - uint32_t rem = 0x1ul; - int i, j; + uint32_t rem = 0x1ul; + int i, j; - const uint32_t poly = 0x82f63b78; + const uint32_t poly = 0x82f63b78; - if (n < 16) - return 0; + if (n < 16) + return 0; - for (i = 0; i < n - 8; i++) - for (j = 0; j < 8; j++) - rem = (rem & 0x1ul) ? (rem >> 1) ^ poly : (rem >> 1); + for (i = 0; i < n - 8; i++) + for (j = 0; j < 8; j++) + rem = (rem & 0x1ul) ? (rem >> 1) ^ poly : (rem >> 1); - return rem; + return rem; } -uint64_t calc64_refl_xi_mod(int n, struct crc64_desc *c) +uint64_t +calc64_refl_xi_mod(int n, struct crc64_desc *c) { - uint64_t rem = 1ull; - int i, j; + uint64_t rem = 1ull; + int i, j; - const uint64_t poly = c->poly; + const uint64_t poly = c->poly; - if (n < 32) - return 0; + if (n < 32) + return 0; - for (i = 0; i < n - 16; i++) - for (j = 0; j < 8; j++) - rem = (rem & 0x1ull) ? (rem >> 1) ^ poly : (rem >> 1); + for (i = 0; i < n - 16; i++) + for (j = 0; j < 8; j++) + rem = (rem & 0x1ull) ? (rem >> 1) ^ poly : (rem >> 1); - return rem; + return rem; } -uint64_t calc64_norm_xi_mod(int n, struct crc64_desc *c) +uint64_t +calc64_norm_xi_mod(int n, struct crc64_desc *c) { - uint64_t rem = 1ull; - int i, j; + uint64_t rem = 1ull; + int i, j; - const uint64_t poly = c->poly; + const uint64_t poly = c->poly; - if (n < 32) - return 0; + if (n < 32) + return 0; - for (i = 0; i < n - 8; i++) - for (j = 0; j < 8; j++) - rem = (rem & 0x8000000000000000ull ? poly : 0) ^ (rem << 1); + for (i = 0; i < n - 8; i++) + for (j = 0; j < 8; j++) + rem = (rem & 0x8000000000000000ull ? poly : 0) ^ (rem << 1); - return rem; + return rem; } // Base function for crc32_iscsi_shiftx() if c++ multi-function versioning #ifdef __cplusplus -static inline uint32_t bit_reverse32(uint32_t x) +static inline uint32_t +bit_reverse32(uint32_t x) { - x = (((x & 0xaaaaaaaa) >> 1) | ((x & 0x55555555) << 1)); - x = (((x & 0xcccccccc) >> 2) | ((x & 0x33333333) << 2)); - x = (((x & 0xf0f0f0f0) >> 4) | ((x & 0x0f0f0f0f) << 4)); - x = (((x & 0xff00ff00) >> 8) | ((x & 0x00ff00ff) << 8)); - return ((x >> 16) | (x << 16)); + x = (((x & 0xaaaaaaaa) >> 1) | ((x & 0x55555555) << 1)); + x = (((x & 0xcccccccc) >> 2) | ((x & 0x33333333) << 2)); + x = (((x & 0xf0f0f0f0) >> 4) | ((x & 0x0f0f0f0f) << 4)); + x = (((x & 0xff00ff00) >> 8) | ((x & 0x00ff00ff) << 8)); + return ((x >> 16) | (x << 16)); } // Base function for crc32_iscsi_shiftx without clmul optimizations ATTRIBUTE_TARGET("default") -uint32_t crc32_iscsi_shiftx(uint32_t crc1, uint32_t x) +uint32_t +crc32_iscsi_shiftx(uint32_t crc1, uint32_t x) { - int i; - uint64_t xrev, q = 0; - union { - uint8_t a[8]; - uint64_t q; - } qu; + int i; + uint64_t xrev, q = 0; + union { + uint8_t a[8]; + uint64_t q; + } qu; - xrev = bit_reverse32(x); - xrev <<= 32; + xrev = bit_reverse32(x); + xrev <<= 32; - for (i = 0; i < 64; i++, xrev >>= 1) - q = (q << 1) | __builtin_parity(crc1 & xrev); + for (i = 0; i < 64; i++, xrev >>= 1) + q = (q << 1) | __builtin_parity(crc1 & xrev); - qu.q = q; - return crc32_iscsi(qu.a, 8, 0); + qu.q = q; + return crc32_iscsi(qu.a, 8, 0); } #endif // cplusplus ATTRIBUTE_TARGET("pclmul,sse4.2") -uint32_t crc32_iscsi_shiftx(uint32_t crc1, uint32_t x) +uint32_t +crc32_iscsi_shiftx(uint32_t crc1, uint32_t x) { - __m128i crc1x, constx; - uint64_t crc64; + __m128i crc1x, constx; + uint64_t crc64; - crc1x = _mm_setr_epi32(crc1, 0, 0, 0); - constx = _mm_setr_epi32(x, 0, 0, 0); - crc1x = _mm_clmulepi64_si128(crc1x, constx, 0); - crc64 = _mm_cvtsi128_si64(crc1x); - crc64 = _mm_crc32_u64(0, crc64); - return crc64 & 0xffffffff; + crc1x = _mm_setr_epi32(crc1, 0, 0, 0); + constx = _mm_setr_epi32(x, 0, 0, 0); + crc1x = _mm_clmulepi64_si128(crc1x, constx, 0); + crc64 = _mm_cvtsi128_si64(crc1x); + crc64 = _mm_crc32_u64(0, crc64); + return crc64 & 0xffffffff; } ATTRIBUTE_TARGET("pclmul,sse4.2") -uint64_t crc64_refl_shiftx(uint64_t crc1, uint64_t x, struct crc64_desc *c) +uint64_t +crc64_refl_shiftx(uint64_t crc1, uint64_t x, struct crc64_desc *c) { - __m128i crc1x, crc2x, crc3x, constx; - const __m128i rk5 = _mm_loadu_si64(&c->k5); - const __m128i rk7 = _mm_loadu_si128((__m128i *) & c->k7); + __m128i crc1x, crc2x, crc3x, constx; + const __m128i rk5 = _mm_loadu_si64(&c->k5); + const __m128i rk7 = _mm_loadu_si128((__m128i *) &c->k7); - crc1x = _mm_cvtsi64_si128(crc1); - constx = _mm_cvtsi64_si128(x); - crc1x = _mm_clmulepi64_si128(crc1x, constx, 0x00); + crc1x = _mm_cvtsi64_si128(crc1); + constx = _mm_cvtsi64_si128(x); + crc1x = _mm_clmulepi64_si128(crc1x, constx, 0x00); - // Fold to 64b - crc2x = _mm_clmulepi64_si128(crc1x, rk5, 0x00); - crc3x = _mm_bsrli_si128(crc1x, 8); - crc1x = _mm_xor_si128(crc2x, crc3x); + // Fold to 64b + crc2x = _mm_clmulepi64_si128(crc1x, rk5, 0x00); + crc3x = _mm_bsrli_si128(crc1x, 8); + crc1x = _mm_xor_si128(crc2x, crc3x); - // Reduce - crc2x = _mm_clmulepi64_si128(crc1x, rk7, 0x00); - crc3x = _mm_clmulepi64_si128(crc2x, rk7, 0x10); - crc2x = _mm_bslli_si128(crc2x, 8); - crc1x = _mm_xor_si128(crc1x, crc2x); - crc1x = _mm_xor_si128(crc1x, crc3x); - return _mm_extract_epi64(crc1x, 1); + // Reduce + crc2x = _mm_clmulepi64_si128(crc1x, rk7, 0x00); + crc3x = _mm_clmulepi64_si128(crc2x, rk7, 0x10); + crc2x = _mm_bslli_si128(crc2x, 8); + crc1x = _mm_xor_si128(crc1x, crc2x); + crc1x = _mm_xor_si128(crc1x, crc3x); + return _mm_extract_epi64(crc1x, 1); } ATTRIBUTE_TARGET("pclmul,sse4.2") -uint64_t crc64_norm_shiftx(uint64_t crc1, uint64_t x, struct crc64_desc *c) +uint64_t +crc64_norm_shiftx(uint64_t crc1, uint64_t x, struct crc64_desc *c) { - __m128i crc1x, crc2x, crc3x, constx; - const __m128i rk5 = _mm_loadu_si64(&c->k5); - const __m128i rk7 = _mm_loadu_si128((__m128i *) & c->k7); + __m128i crc1x, crc2x, crc3x, constx; + const __m128i rk5 = _mm_loadu_si64(&c->k5); + const __m128i rk7 = _mm_loadu_si128((__m128i *) &c->k7); - crc1x = _mm_cvtsi64_si128(crc1); - constx = _mm_cvtsi64_si128(x); - crc1x = _mm_clmulepi64_si128(crc1x, constx, 0x00); + crc1x = _mm_cvtsi64_si128(crc1); + constx = _mm_cvtsi64_si128(x); + crc1x = _mm_clmulepi64_si128(crc1x, constx, 0x00); - // Fold to 64b - crc2x = _mm_clmulepi64_si128(crc1x, rk5, 0x01); - crc3x = _mm_bslli_si128(crc1x, 8); - crc1x = _mm_xor_si128(crc2x, crc3x); + // Fold to 64b + crc2x = _mm_clmulepi64_si128(crc1x, rk5, 0x01); + crc3x = _mm_bslli_si128(crc1x, 8); + crc1x = _mm_xor_si128(crc2x, crc3x); - // Reduce - crc2x = _mm_clmulepi64_si128(crc1x, rk7, 0x01); - crc2x = _mm_xor_si128(crc1x, crc2x); - crc3x = _mm_clmulepi64_si128(crc2x, rk7, 0x11); - crc1x = _mm_xor_si128(crc1x, crc3x); - return _mm_extract_epi64(crc1x, 0); + // Reduce + crc2x = _mm_clmulepi64_si128(crc1x, rk7, 0x01); + crc2x = _mm_xor_si128(crc1x, crc2x); + crc3x = _mm_clmulepi64_si128(crc2x, rk7, 0x11); + crc1x = _mm_xor_si128(crc1x, crc3x); + return _mm_extract_epi64(crc1x, 0); } -uint32_t crc32_iscsi_combine_4k(uint32_t * crc_array, int n) +uint32_t +crc32_iscsi_combine_4k(uint32_t *crc_array, int n) { - const uint32_t cn4k = 0x82f89c77; //calc_xi_mod(4*1024); - int i; + const uint32_t cn4k = 0x82f89c77; // calc_xi_mod(4*1024); + int i; - if (n < 1) - return 0; + if (n < 1) + return 0; - uint32_t crc = crc_array[0]; + uint32_t crc = crc_array[0]; - for (i = 1; i < n; i++) - crc = crc32_iscsi_shiftx(crc, cn4k) ^ crc_array[i]; + for (i = 1; i < n; i++) + crc = crc32_iscsi_shiftx(crc, cn4k) ^ crc_array[i]; - return crc; + return crc; } // Tests -#define printv(...) {if (verbose) printf(__VA_ARGS__); else printf(".");} +#define printv(...) \ + { \ + if (verbose) \ + printf(__VA_ARGS__); \ + else \ + printf("."); \ + } -uint64_t test_combine64(uint8_t * inp, size_t len, uint64_t poly, int reflected, - uint64_t(*func) (uint64_t, const uint8_t *, uint64_t)) +uint64_t +test_combine64(uint8_t *inp, size_t len, uint64_t poly, int reflected, + uint64_t (*func)(uint64_t, const uint8_t *, uint64_t)) { - uint64_t crc64_init, crc64, crc64a, crc64b; - uint64_t crc64_1, crc64_2, crc64_3, crc64_n, err = 0; - uint64_t xi_mod; - struct crc64_desc crc64_c; - size_t l1, l2, l3; + uint64_t crc64_init, crc64, crc64a, crc64b; + uint64_t crc64_1, crc64_2, crc64_3, crc64_n, err = 0; + uint64_t xi_mod; + struct crc64_desc crc64_c; + size_t l1, l2, l3; - l1 = len / 2; - l2 = len - l1; + l1 = len / 2; + l2 = len - l1; - crc64_init = rand(); - crc64 = func(crc64_init, inp, len); - printv("\ncrc64 all = 0x%" PRIx64 "\n", crc64); + crc64_init = rand(); + crc64 = func(crc64_init, inp, len); + printv("\ncrc64 all = 0x%" PRIx64 "\n", crc64); - // Do a sequential crc update - crc64a = func(crc64_init, &inp[0], l1); - crc64b = func(crc64a, &inp[l1], l2); - printv("crc64 seq = 0x%" PRIx64 "\n", crc64b); + // Do a sequential crc update + crc64a = func(crc64_init, &inp[0], l1); + crc64b = func(crc64a, &inp[l1], l2); + printv("crc64 seq = 0x%" PRIx64 "\n", crc64b); - // Split into 2 independent crc calc and combine - crc64_1 = func(crc64_init, &inp[0], l1); - crc64_2 = func(0, &inp[l1], l2); + // Split into 2 independent crc calc and combine + crc64_1 = func(crc64_init, &inp[0], l1); + crc64_2 = func(0, &inp[l1], l2); - if (reflected) { - gen_crc64_refl_consts(poly, &crc64_c); - xi_mod = calc64_refl_xi_mod(l1, &crc64_c); - crc64_1 = crc64_refl_shiftx(crc64_1, xi_mod, &crc64_c); - } else { - gen_crc64_norm_consts(poly, &crc64_c); - xi_mod = calc64_norm_xi_mod(l1, &crc64_c); - crc64_1 = crc64_norm_shiftx(crc64_1, xi_mod, &crc64_c); - } - crc64_n = crc64_1 ^ crc64_2; + if (reflected) { + gen_crc64_refl_consts(poly, &crc64_c); + xi_mod = calc64_refl_xi_mod(l1, &crc64_c); + crc64_1 = crc64_refl_shiftx(crc64_1, xi_mod, &crc64_c); + } else { + gen_crc64_norm_consts(poly, &crc64_c); + xi_mod = calc64_norm_xi_mod(l1, &crc64_c); + crc64_1 = crc64_norm_shiftx(crc64_1, xi_mod, &crc64_c); + } + crc64_n = crc64_1 ^ crc64_2; - printv("crc64 combined2 = 0x%" PRIx64 "\n", crc64_n); - err |= crc64_n ^ crc64; - if (err) - return err; + printv("crc64 combined2 = 0x%" PRIx64 "\n", crc64_n); + err |= crc64_n ^ crc64; + if (err) + return err; - // Split into 3 uneven segments and combine - l1 = len / 3; - l2 = (len / 3) - 3; - l3 = len - l2 - l1; - crc64_1 = func(crc64_init, &inp[0], l1); - crc64_2 = func(0, &inp[l1], l2); - crc64_3 = func(0, &inp[l1 + l2], l3); - if (reflected) { - xi_mod = calc64_refl_xi_mod(l3, &crc64_c); - crc64_2 = crc64_refl_shiftx(crc64_2, xi_mod, &crc64_c); - xi_mod = calc64_refl_xi_mod(len - l1, &crc64_c); - crc64_1 = crc64_refl_shiftx(crc64_1, xi_mod, &crc64_c); - } else { - xi_mod = calc64_norm_xi_mod(l3, &crc64_c); - crc64_2 = crc64_norm_shiftx(crc64_2, xi_mod, &crc64_c); - xi_mod = calc64_norm_xi_mod(len - l1, &crc64_c); - crc64_1 = crc64_norm_shiftx(crc64_1, xi_mod, &crc64_c); - } - crc64_n = crc64_1 ^ crc64_2 ^ crc64_3; + // Split into 3 uneven segments and combine + l1 = len / 3; + l2 = (len / 3) - 3; + l3 = len - l2 - l1; + crc64_1 = func(crc64_init, &inp[0], l1); + crc64_2 = func(0, &inp[l1], l2); + crc64_3 = func(0, &inp[l1 + l2], l3); + if (reflected) { + xi_mod = calc64_refl_xi_mod(l3, &crc64_c); + crc64_2 = crc64_refl_shiftx(crc64_2, xi_mod, &crc64_c); + xi_mod = calc64_refl_xi_mod(len - l1, &crc64_c); + crc64_1 = crc64_refl_shiftx(crc64_1, xi_mod, &crc64_c); + } else { + xi_mod = calc64_norm_xi_mod(l3, &crc64_c); + crc64_2 = crc64_norm_shiftx(crc64_2, xi_mod, &crc64_c); + xi_mod = calc64_norm_xi_mod(len - l1, &crc64_c); + crc64_1 = crc64_norm_shiftx(crc64_1, xi_mod, &crc64_c); + } + crc64_n = crc64_1 ^ crc64_2 ^ crc64_3; - printv("crc64 combined3 = 0x%" PRIx64 "\n", crc64_n); - err |= crc64_n ^ crc64; + printv("crc64 combined3 = 0x%" PRIx64 "\n", crc64_n); + err |= crc64_n ^ crc64; - return err; + return err; } -#define N (1024) -#define B (2*N) -#define T (3*N) -#define N4k (4*1024) -#define NMAX 32 +#define N (1024) +#define B (2 * N) +#define T (3 * N) +#define N4k (4 * 1024) +#define NMAX 32 #define NMAX_SIZE (NMAX * N4k) -int main(int argc, char *argv[]) +int +main(int argc, char *argv[]) { - int i; - uint32_t crc, crca, crcb, crc1, crc2, crc3, crcn; - uint32_t crc_init = rand(); - uint32_t err = 0; - uint8_t *inp = (uint8_t *) malloc(NMAX_SIZE); - verbose = argc - 1; + int i; + uint32_t crc, crca, crcb, crc1, crc2, crc3, crcn; + uint32_t crc_init = rand(); + uint32_t err = 0; + uint8_t *inp = (uint8_t *) malloc(NMAX_SIZE); + verbose = argc - 1; - if (NULL == inp) - return -1; + if (NULL == inp) + return -1; - for (int i = 0; i < NMAX_SIZE; i++) - inp[i] = rand(); + for (int i = 0; i < NMAX_SIZE; i++) + inp[i] = rand(); - printf("crc_combine_test:"); + printf("crc_combine_test:"); - // Calc crc all at once - crc = crc32_iscsi(inp, B, crc_init); - printv("\ncrcB all = 0x%" PRIx32 "\n", crc); + // Calc crc all at once + crc = crc32_iscsi(inp, B, crc_init); + printv("\ncrcB all = 0x%" PRIx32 "\n", crc); - // Do a sequential crc update - crca = crc32_iscsi(&inp[0], N, crc_init); - crcb = crc32_iscsi(&inp[N], N, crca); - printv("crcB seq = 0x%" PRIx32 "\n", crcb); + // Do a sequential crc update + crca = crc32_iscsi(&inp[0], N, crc_init); + crcb = crc32_iscsi(&inp[N], N, crca); + printv("crcB seq = 0x%" PRIx32 "\n", crcb); - // Split into 2 independent crc calc and combine - crc1 = crc32_iscsi(&inp[0], N, crc_init); - crc2 = crc32_iscsi(&inp[N], N, 0); - crcn = crc32_iscsi_shiftx(crc1, calc_xi_mod(N)) ^ crc2; - printv("crcB combined2 = 0x%" PRIx32 "\n", crcn); - err |= crcn ^ crc; + // Split into 2 independent crc calc and combine + crc1 = crc32_iscsi(&inp[0], N, crc_init); + crc2 = crc32_iscsi(&inp[N], N, 0); + crcn = crc32_iscsi_shiftx(crc1, calc_xi_mod(N)) ^ crc2; + printv("crcB combined2 = 0x%" PRIx32 "\n", crcn); + err |= crcn ^ crc; - // Split into 3 uneven segments and combine - crc1 = crc32_iscsi(&inp[0], 100, crc_init); - crc2 = crc32_iscsi(&inp[100], 100, 0); - crc3 = crc32_iscsi(&inp[200], B - 200, 0); - crcn = crc3 ^ - crc32_iscsi_shiftx(crc2, calc_xi_mod(B - 200)) ^ - crc32_iscsi_shiftx(crc1, calc_xi_mod(B - 100)); - printv("crcB combined3 = 0x%" PRIx32 "\n\n", crcn); - err |= crcn ^ crc; + // Split into 3 uneven segments and combine + crc1 = crc32_iscsi(&inp[0], 100, crc_init); + crc2 = crc32_iscsi(&inp[100], 100, 0); + crc3 = crc32_iscsi(&inp[200], B - 200, 0); + crcn = crc3 ^ crc32_iscsi_shiftx(crc2, calc_xi_mod(B - 200)) ^ + crc32_iscsi_shiftx(crc1, calc_xi_mod(B - 100)); + printv("crcB combined3 = 0x%" PRIx32 "\n\n", crcn); + err |= crcn ^ crc; - // Call all size T at once - crc = crc32_iscsi(inp, T, crc_init); - printv("crcT all = 0x%" PRIx32 "\n", crc); + // Call all size T at once + crc = crc32_iscsi(inp, T, crc_init); + printv("crcT all = 0x%" PRIx32 "\n", crc); - // Split into 3 segments and combine with 2 consts - crc1 = crc32_iscsi(&inp[0], N, crc_init); - crc2 = crc32_iscsi(&inp[N], N, 0); - crc3 = crc32_iscsi(&inp[2 * N], N, 0); - crcn = crc3 ^ - crc32_iscsi_shiftx(crc2, calc_xi_mod(N)) ^ - crc32_iscsi_shiftx(crc1, calc_xi_mod(2 * N)); - printv("crcT combined3 = 0x%" PRIx32 "\n", crcn); - err |= crcn ^ crc; + // Split into 3 segments and combine with 2 consts + crc1 = crc32_iscsi(&inp[0], N, crc_init); + crc2 = crc32_iscsi(&inp[N], N, 0); + crc3 = crc32_iscsi(&inp[2 * N], N, 0); + crcn = crc3 ^ crc32_iscsi_shiftx(crc2, calc_xi_mod(N)) ^ + crc32_iscsi_shiftx(crc1, calc_xi_mod(2 * N)); + printv("crcT combined3 = 0x%" PRIx32 "\n", crcn); + err |= crcn ^ crc; - // Combine 3 segments with one const by sequential shift - uint32_t xi_mod_n = calc_xi_mod(N); - crcn = crc3 ^ crc32_iscsi_shiftx(crc32_iscsi_shiftx(crc1, xi_mod_n) - ^ crc2, xi_mod_n); - printv("crcT comb3 seq = 0x%" PRIx32 "\n\n", crcn); - err |= crcn ^ crc; + // Combine 3 segments with one const by sequential shift + uint32_t xi_mod_n = calc_xi_mod(N); + crcn = crc3 ^ crc32_iscsi_shiftx(crc32_iscsi_shiftx(crc1, xi_mod_n) ^ crc2, xi_mod_n); + printv("crcT comb3 seq = 0x%" PRIx32 "\n\n", crcn); + err |= crcn ^ crc; - // Test 4k array function - crc = crc32_iscsi(inp, NMAX_SIZE, crc_init); - printv("crc 4k x n all = 0x%" PRIx32 "\n", crc); + // Test 4k array function + crc = crc32_iscsi(inp, NMAX_SIZE, crc_init); + printv("crc 4k x n all = 0x%" PRIx32 "\n", crc); - // Test crc 4k array combine function - uint32_t crcs[NMAX]; - crcs[0] = crc32_iscsi(&inp[0], N4k, crc_init); - for (i = 1; i < NMAX; i++) - crcs[i] = crc32_iscsi(&inp[i * N4k], N4k, 0); + // Test crc 4k array combine function + uint32_t crcs[NMAX]; + crcs[0] = crc32_iscsi(&inp[0], N4k, crc_init); + for (i = 1; i < NMAX; i++) + crcs[i] = crc32_iscsi(&inp[i * N4k], N4k, 0); - crcn = crc32_iscsi_combine_4k(crcs, NMAX); - printv("crc4k_array = 0x%" PRIx32 "\n", crcn); - err |= crcn ^ crc; + crcn = crc32_iscsi_combine_4k(crcs, NMAX); + printv("crc4k_array = 0x%" PRIx32 "\n", crcn); + err |= crcn ^ crc; - // CRC64 generic poly tests - reflected - uint64_t len = NMAX_SIZE; - err |= test_combine64(inp, len, 0xc96c5795d7870f42ull, 1, crc64_ecma_refl); - err |= test_combine64(inp, len, 0xd800000000000000ull, 1, crc64_iso_refl); - err |= test_combine64(inp, len, 0x95ac9329ac4bc9b5ull, 1, crc64_jones_refl); + // CRC64 generic poly tests - reflected + uint64_t len = NMAX_SIZE; + err |= test_combine64(inp, len, 0xc96c5795d7870f42ull, 1, crc64_ecma_refl); + err |= test_combine64(inp, len, 0xd800000000000000ull, 1, crc64_iso_refl); + err |= test_combine64(inp, len, 0x95ac9329ac4bc9b5ull, 1, crc64_jones_refl); - // CRC64 non-reflected polynomial tests - err |= test_combine64(inp, len, 0x42f0e1eba9ea3693ull, 0, crc64_ecma_norm); - err |= test_combine64(inp, len, 0x000000000000001bull, 0, crc64_iso_norm); - err |= test_combine64(inp, len, 0xad93d23594c935a9ull, 0, crc64_jones_norm); + // CRC64 non-reflected polynomial tests + err |= test_combine64(inp, len, 0x42f0e1eba9ea3693ull, 0, crc64_ecma_norm); + err |= test_combine64(inp, len, 0x000000000000001bull, 0, crc64_iso_norm); + err |= test_combine64(inp, len, 0xad93d23594c935a9ull, 0, crc64_jones_norm); - printf(err == 0 ? "pass\n" : "fail\n"); - free(inp); - return err; + printf(err == 0 ? "pass\n" : "fail\n"); + free(inp); + return err; } diff --git a/examples/ec/ec_piggyback_example.c b/examples/ec/ec_piggyback_example.c index 6b9c11b..9fb27eb 100644 --- a/examples/ec/ec_piggyback_example.c +++ b/examples/ec/ec_piggyback_example.c @@ -31,7 +31,7 @@ #include #include #include -#include "erasure_code.h" // use instead when linking against installed +#include "erasure_code.h" // use instead when linking against installed #include "test.h" #define MMAX 255 @@ -40,467 +40,463 @@ typedef unsigned char u8; int verbose = 0; -int usage(void) +int +usage(void) { - fprintf(stderr, - "Usage: ec_piggyback_example [options]\n" - " -h Help\n" - " -k Number of source fragments\n" - " -p Number of parity fragments\n" - " -l Length of fragments\n" - " -e Simulate erasure on frag index val. Zero based. Can be repeated.\n" - " -v Verbose\n" - " -b Run timed benchmark\n" - " -s Toggle use of sparse matrix opt\n" - " -r Pick random (k, p) with seed\n"); - exit(0); + fprintf(stderr, + "Usage: ec_piggyback_example [options]\n" + " -h Help\n" + " -k Number of source fragments\n" + " -p Number of parity fragments\n" + " -l Length of fragments\n" + " -e Simulate erasure on frag index val. Zero based. Can be repeated.\n" + " -v Verbose\n" + " -b Run timed benchmark\n" + " -s Toggle use of sparse matrix opt\n" + " -r Pick random (k, p) with seed\n"); + exit(0); } // Cauchy-based matrix -void gf_gen_full_pb_cauchy_matrix(u8 * a, int m, int k) +void +gf_gen_full_pb_cauchy_matrix(u8 *a, int m, int k) { - int i, j, p = m - k; + int i, j, p = m - k; - // Identity matrix in top k x k to indicate a symmetric code - memset(a, 0, k * m); - for (i = 0; i < k; i++) - a[k * i + i] = 1; + // Identity matrix in top k x k to indicate a symmetric code + memset(a, 0, k * m); + for (i = 0; i < k; i++) + a[k * i + i] = 1; - for (i = k; i < (k + p / 2); i++) { - for (j = 0; j < k / 2; j++) - a[k * i + j] = gf_inv(i ^ j); - for (; j < k; j++) - a[k * i + j] = 0; - } - for (; i < m; i++) { - for (j = 0; j < k / 2; j++) - a[k * i + j] = 0; - for (; j < k; j++) - a[k * i + j] = gf_inv((i - p / 2) ^ (j - k / 2)); - } + for (i = k; i < (k + p / 2); i++) { + for (j = 0; j < k / 2; j++) + a[k * i + j] = gf_inv(i ^ j); + for (; j < k; j++) + a[k * i + j] = 0; + } + for (; i < m; i++) { + for (j = 0; j < k / 2; j++) + a[k * i + j] = 0; + for (; j < k; j++) + a[k * i + j] = gf_inv((i - p / 2) ^ (j - k / 2)); + } - // Fill in mixture of B parity depending on a few localized A sources - int r = 0, c = 0; - int repeat_len = k / (p - 2); - int parity_rows = p / 2; + // Fill in mixture of B parity depending on a few localized A sources + int r = 0, c = 0; + int repeat_len = k / (p - 2); + int parity_rows = p / 2; - for (i = 1 + k + parity_rows; i < m; i++, r++) { - if (r == (parity_rows - 1) - ((k / 2 % (parity_rows - 1)))) - repeat_len++; + for (i = 1 + k + parity_rows; i < m; i++, r++) { + if (r == (parity_rows - 1) - ((k / 2 % (parity_rows - 1)))) + repeat_len++; - for (j = 0; j < repeat_len; j++, c++) - a[k * i + c] = gf_inv((k + 1) ^ c); - } + for (j = 0; j < repeat_len; j++, c++) + a[k * i + c] = gf_inv((k + 1) ^ c); + } } // Vandermonde based matrix - not recommended due to limits when invertable -void gf_gen_full_pb_vand_matrix(u8 * a, int m, int k) +void +gf_gen_full_pb_vand_matrix(u8 *a, int m, int k) { - int i, j, p = m - k; - unsigned char q, gen = 1; + int i, j, p = m - k; + unsigned char q, gen = 1; - // Identity matrix in top k x k to indicate a symmetric code - memset(a, 0, k * m); - for (i = 0; i < k; i++) - a[k * i + i] = 1; + // Identity matrix in top k x k to indicate a symmetric code + memset(a, 0, k * m); + for (i = 0; i < k; i++) + a[k * i + i] = 1; - for (i = k; i < (k + (p / 2)); i++) { - q = 1; - for (j = 0; j < k / 2; j++) { - a[k * i + j] = q; - q = gf_mul(q, gen); - } - for (; j < k; j++) - a[k * i + j] = 0; - gen = gf_mul(gen, 2); - } - gen = 1; - for (; i < m; i++) { - q = 1; - for (j = 0; j < k / 2; j++) { - a[k * i + j] = 0; - } - for (; j < k; j++) { - a[k * i + j] = q; - q = gf_mul(q, gen); - } - gen = gf_mul(gen, 2); - } + for (i = k; i < (k + (p / 2)); i++) { + q = 1; + for (j = 0; j < k / 2; j++) { + a[k * i + j] = q; + q = gf_mul(q, gen); + } + for (; j < k; j++) + a[k * i + j] = 0; + gen = gf_mul(gen, 2); + } + gen = 1; + for (; i < m; i++) { + q = 1; + for (j = 0; j < k / 2; j++) { + a[k * i + j] = 0; + } + for (; j < k; j++) { + a[k * i + j] = q; + q = gf_mul(q, gen); + } + gen = gf_mul(gen, 2); + } - // Fill in mixture of B parity depending on a few localized A sources - int r = 0, c = 0; - int repeat_len = k / (p - 2); - int parity_rows = p / 2; + // Fill in mixture of B parity depending on a few localized A sources + int r = 0, c = 0; + int repeat_len = k / (p - 2); + int parity_rows = p / 2; - for (i = 1 + k + parity_rows; i < m; i++, r++) { - if (r == (parity_rows - 1) - ((k / 2 % (parity_rows - 1)))) - repeat_len++; + for (i = 1 + k + parity_rows; i < m; i++, r++) { + if (r == (parity_rows - 1) - ((k / 2 % (parity_rows - 1)))) + repeat_len++; - for (j = 0; j < repeat_len; j++) - a[k * i + c++] = 1; - } + for (j = 0; j < repeat_len; j++) + a[k * i + c++] = 1; + } } -void print_matrix(int m, int k, unsigned char *s, const char *msg) +void +print_matrix(int m, int k, unsigned char *s, const char *msg) { - int i, j; + int i, j; - printf("%s:\n", msg); - for (i = 0; i < m; i++) { - printf("%3d- ", i); - for (j = 0; j < k; j++) { - printf(" %2x", 0xff & s[j + (i * k)]); - } - printf("\n"); - } - printf("\n"); + printf("%s:\n", msg); + for (i = 0; i < m; i++) { + printf("%3d- ", i); + for (j = 0; j < k; j++) { + printf(" %2x", 0xff & s[j + (i * k)]); + } + printf("\n"); + } + printf("\n"); } -void print_list(int n, unsigned char *s, const char *msg) +void +print_list(int n, unsigned char *s, const char *msg) { - int i; - if (!verbose) - return; + int i; + if (!verbose) + return; - printf("%s: ", msg); - for (i = 0; i < n; i++) - printf(" %d", s[i]); - printf("\n"); + printf("%s: ", msg); + for (i = 0; i < n; i++) + printf(" %d", s[i]); + printf("\n"); } -static int gf_gen_decode_matrix(u8 * encode_matrix, - u8 * decode_matrix, - u8 * invert_matrix, - u8 * temp_matrix, - u8 * decode_index, - u8 * frag_err_list, int nerrs, int k, int m); +static int +gf_gen_decode_matrix(u8 *encode_matrix, u8 *decode_matrix, u8 *invert_matrix, u8 *temp_matrix, + u8 *decode_index, u8 *frag_err_list, int nerrs, int k, int m); -int main(int argc, char *argv[]) +int +main(int argc, char *argv[]) { - int i, j, m, c, e, ret; - int k = 10, p = 4, len = 8 * 1024; // Default params - int nerrs = 0; - int benchmark = 0; - int sparse_matrix_opt = 1; + int i, j, m, c, e, ret; + int k = 10, p = 4, len = 8 * 1024; // Default params + int nerrs = 0; + int benchmark = 0; + int sparse_matrix_opt = 1; - // Fragment buffer pointers - u8 *frag_ptrs[MMAX]; - u8 *parity_ptrs[KMAX]; - u8 *recover_srcs[KMAX]; - u8 *recover_outp[KMAX]; - u8 frag_err_list[MMAX]; + // Fragment buffer pointers + u8 *frag_ptrs[MMAX]; + u8 *parity_ptrs[KMAX]; + u8 *recover_srcs[KMAX]; + u8 *recover_outp[KMAX]; + u8 frag_err_list[MMAX]; - // Coefficient matrices - u8 *encode_matrix, *decode_matrix; - u8 *invert_matrix, *temp_matrix; - u8 *g_tbls; - u8 decode_index[MMAX]; + // Coefficient matrices + u8 *encode_matrix, *decode_matrix; + u8 *invert_matrix, *temp_matrix; + u8 *g_tbls; + u8 decode_index[MMAX]; - if (argc == 1) - for (i = 0; i < p; i++) - frag_err_list[nerrs++] = rand() % (k + p); + if (argc == 1) + for (i = 0; i < p; i++) + frag_err_list[nerrs++] = rand() % (k + p); - while ((c = getopt(argc, argv, "k:p:l:e:r:hvbs")) != -1) { - switch (c) { - case 'k': - k = atoi(optarg); - break; - case 'p': - p = atoi(optarg); - break; - case 'l': - len = atoi(optarg); - if (len < 0) - usage(); - break; - case 'e': - e = atoi(optarg); - frag_err_list[nerrs++] = e; - break; - case 'r': - srand(atoi(optarg)); - k = (rand() % MMAX) / 4; - k = (k < 2) ? 2 : k; - p = (rand() % (MMAX - k)) / 4; - p = (p < 2) ? 2 : p; - for (i = 0; i < k && nerrs < p; i++) - if (rand() & 1) - frag_err_list[nerrs++] = i; - break; - case 'v': - verbose++; - break; - case 'b': - benchmark = 1; - break; - case 's': - sparse_matrix_opt = !sparse_matrix_opt; - break; - case 'h': - default: - usage(); - break; - } - } - m = k + p; + while ((c = getopt(argc, argv, "k:p:l:e:r:hvbs")) != -1) { + switch (c) { + case 'k': + k = atoi(optarg); + break; + case 'p': + p = atoi(optarg); + break; + case 'l': + len = atoi(optarg); + if (len < 0) + usage(); + break; + case 'e': + e = atoi(optarg); + frag_err_list[nerrs++] = e; + break; + case 'r': + srand(atoi(optarg)); + k = (rand() % MMAX) / 4; + k = (k < 2) ? 2 : k; + p = (rand() % (MMAX - k)) / 4; + p = (p < 2) ? 2 : p; + for (i = 0; i < k && nerrs < p; i++) + if (rand() & 1) + frag_err_list[nerrs++] = i; + break; + case 'v': + verbose++; + break; + case 'b': + benchmark = 1; + break; + case 's': + sparse_matrix_opt = !sparse_matrix_opt; + break; + case 'h': + default: + usage(); + break; + } + } + m = k + p; - // Check for valid parameters - if (m > (MMAX / 2) || k > (KMAX / 2) || m < 0 || p < 2 || k < 1) { - printf(" Input test parameter error m=%d, k=%d, p=%d, erasures=%d\n", - m, k, p, nerrs); - usage(); - } - if (nerrs > p) { - printf(" Number of erasures chosen exceeds power of code erasures=%d p=%d\n", - nerrs, p); - } - for (i = 0; i < nerrs; i++) { - if (frag_err_list[i] >= m) - printf(" fragment %d not in range\n", frag_err_list[i]); - } + // Check for valid parameters + if (m > (MMAX / 2) || k > (KMAX / 2) || m < 0 || p < 2 || k < 1) { + printf(" Input test parameter error m=%d, k=%d, p=%d, erasures=%d\n", m, k, p, + nerrs); + usage(); + } + if (nerrs > p) { + printf(" Number of erasures chosen exceeds power of code erasures=%d p=%d\n", nerrs, + p); + } + for (i = 0; i < nerrs; i++) { + if (frag_err_list[i] >= m) + printf(" fragment %d not in range\n", frag_err_list[i]); + } - printf("ec_piggyback_example:\n"); + printf("ec_piggyback_example:\n"); - /* - * One simple way to implement piggyback codes is to keep a 2x wide matrix - * that covers the how each parity is related to both A and B sources. This - * keeps it easy to generalize in parameters m,k and the resulting sparse - * matrix multiplication can be optimized by pre-removal of zero items. - */ + /* + * One simple way to implement piggyback codes is to keep a 2x wide matrix + * that covers the how each parity is related to both A and B sources. This + * keeps it easy to generalize in parameters m,k and the resulting sparse + * matrix multiplication can be optimized by pre-removal of zero items. + */ - int k2 = 2 * k; - int p2 = 2 * p; - int m2 = k2 + p2; - int nerrs2 = nerrs; + int k2 = 2 * k; + int p2 = 2 * p; + int m2 = k2 + p2; + int nerrs2 = nerrs; - encode_matrix = malloc(m2 * k2); - decode_matrix = malloc(m2 * k2); - invert_matrix = malloc(m2 * k2); - temp_matrix = malloc(m2 * k2); - g_tbls = malloc(k2 * p2 * 32); + encode_matrix = malloc(m2 * k2); + decode_matrix = malloc(m2 * k2); + invert_matrix = malloc(m2 * k2); + temp_matrix = malloc(m2 * k2); + g_tbls = malloc(k2 * p2 * 32); - if (encode_matrix == NULL || decode_matrix == NULL - || invert_matrix == NULL || temp_matrix == NULL || g_tbls == NULL) { - printf("Test failure! Error with malloc\n"); - return -1; - } - // Allocate the src fragments - for (i = 0; i < k; i++) { - if (NULL == (frag_ptrs[i] = malloc(len))) { - printf("alloc error: Fail\n"); - return -1; - } - } - // Allocate the parity fragments - for (i = 0; i < p2; i++) { - if (NULL == (parity_ptrs[i] = malloc(len / 2))) { - printf("alloc error: Fail\n"); - return -1; - } - } + if (encode_matrix == NULL || decode_matrix == NULL || invert_matrix == NULL || + temp_matrix == NULL || g_tbls == NULL) { + printf("Test failure! Error with malloc\n"); + return -1; + } + // Allocate the src fragments + for (i = 0; i < k; i++) { + if (NULL == (frag_ptrs[i] = malloc(len))) { + printf("alloc error: Fail\n"); + return -1; + } + } + // Allocate the parity fragments + for (i = 0; i < p2; i++) { + if (NULL == (parity_ptrs[i] = malloc(len / 2))) { + printf("alloc error: Fail\n"); + return -1; + } + } - // Allocate buffers for recovered data - for (i = 0; i < p2; i++) { - if (NULL == (recover_outp[i] = malloc(len / 2))) { - printf("alloc error: Fail\n"); - return -1; - } - } + // Allocate buffers for recovered data + for (i = 0; i < p2; i++) { + if (NULL == (recover_outp[i] = malloc(len / 2))) { + printf("alloc error: Fail\n"); + return -1; + } + } - // Fill sources with random data - for (i = 0; i < k; i++) - for (j = 0; j < len; j++) - frag_ptrs[i][j] = rand(); + // Fill sources with random data + for (i = 0; i < k; i++) + for (j = 0; j < len; j++) + frag_ptrs[i][j] = rand(); - printf(" encode (m,k,p)=(%d,%d,%d) len=%d\n", m, k, p, len); + printf(" encode (m,k,p)=(%d,%d,%d) len=%d\n", m, k, p, len); - // Pick an encode matrix. - gf_gen_full_pb_cauchy_matrix(encode_matrix, m2, k2); + // Pick an encode matrix. + gf_gen_full_pb_cauchy_matrix(encode_matrix, m2, k2); - if (verbose) - print_matrix(m2, k2, encode_matrix, "encode matrix"); + if (verbose) + print_matrix(m2, k2, encode_matrix, "encode matrix"); - // Initialize g_tbls from encode matrix - ec_init_tables(k2, p2, &encode_matrix[k2 * k2], g_tbls); + // Initialize g_tbls from encode matrix + ec_init_tables(k2, p2, &encode_matrix[k2 * k2], g_tbls); - // Fold A and B into single list of fragments - for (i = 0; i < k; i++) - frag_ptrs[i + k] = &frag_ptrs[i][len / 2]; + // Fold A and B into single list of fragments + for (i = 0; i < k; i++) + frag_ptrs[i + k] = &frag_ptrs[i][len / 2]; - if (!sparse_matrix_opt) { - // Standard encode using no assumptions on the encode matrix + if (!sparse_matrix_opt) { + // Standard encode using no assumptions on the encode matrix - // Generate EC parity blocks from sources - ec_encode_data(len / 2, k2, p2, g_tbls, frag_ptrs, parity_ptrs); + // Generate EC parity blocks from sources + ec_encode_data(len / 2, k2, p2, g_tbls, frag_ptrs, parity_ptrs); - if (benchmark) { - struct perf start; - BENCHMARK(&start, BENCHMARK_TIME, - ec_encode_data(len / 2, k2, p2, g_tbls, frag_ptrs, - parity_ptrs)); - printf("ec_piggyback_encode_std: "); - perf_print(start, m2 * len / 2); - } - } else { - // Sparse matrix optimization - use fact that input matrix is sparse + if (benchmark) { + struct perf start; + BENCHMARK(&start, BENCHMARK_TIME, + ec_encode_data(len / 2, k2, p2, g_tbls, frag_ptrs, parity_ptrs)); + printf("ec_piggyback_encode_std: "); + perf_print(start, m2 * len / 2); + } + } else { + // Sparse matrix optimization - use fact that input matrix is sparse - // Keep an encode matrix with some zero elements removed - u8 *encode_matrix_faster, *g_tbls_faster; - encode_matrix_faster = malloc(m * k); - g_tbls_faster = malloc(k * p * 32); - if (encode_matrix_faster == NULL || g_tbls_faster == NULL) { - printf("Test failure! Error with malloc\n"); - return -1; - } + // Keep an encode matrix with some zero elements removed + u8 *encode_matrix_faster, *g_tbls_faster; + encode_matrix_faster = malloc(m * k); + g_tbls_faster = malloc(k * p * 32); + if (encode_matrix_faster == NULL || g_tbls_faster == NULL) { + printf("Test failure! Error with malloc\n"); + return -1; + } - /* - * Pack with only the part that we know are non-zero. Alternatively - * we could search and keep track of non-zero elements but for - * simplicity we just skip the lower quadrant. - */ - for (i = k, j = k2; i < m; i++, j++) - memcpy(&encode_matrix_faster[k * i], &encode_matrix[k2 * j], k); + /* + * Pack with only the part that we know are non-zero. Alternatively + * we could search and keep track of non-zero elements but for + * simplicity we just skip the lower quadrant. + */ + for (i = k, j = k2; i < m; i++, j++) + memcpy(&encode_matrix_faster[k * i], &encode_matrix[k2 * j], k); - if (verbose) { - print_matrix(p, k, &encode_matrix_faster[k * k], - "encode via sparse-opt"); - print_matrix(p2 / 2, k2, &encode_matrix[(k2 + p2 / 2) * k2], - "encode via sparse-opt"); - } - // Initialize g_tbls from encode matrix - ec_init_tables(k, p, &encode_matrix_faster[k * k], g_tbls_faster); + if (verbose) { + print_matrix(p, k, &encode_matrix_faster[k * k], "encode via sparse-opt"); + print_matrix(p2 / 2, k2, &encode_matrix[(k2 + p2 / 2) * k2], + "encode via sparse-opt"); + } + // Initialize g_tbls from encode matrix + ec_init_tables(k, p, &encode_matrix_faster[k * k], g_tbls_faster); - // Generate EC parity blocks from sources - ec_encode_data(len / 2, k, p, g_tbls_faster, frag_ptrs, parity_ptrs); - ec_encode_data(len / 2, k2, p, &g_tbls[k2 * p * 32], frag_ptrs, - &parity_ptrs[p]); + // Generate EC parity blocks from sources + ec_encode_data(len / 2, k, p, g_tbls_faster, frag_ptrs, parity_ptrs); + ec_encode_data(len / 2, k2, p, &g_tbls[k2 * p * 32], frag_ptrs, &parity_ptrs[p]); - if (benchmark) { - struct perf start; - BENCHMARK(&start, BENCHMARK_TIME, - ec_encode_data(len / 2, k, p, g_tbls_faster, frag_ptrs, - parity_ptrs); - ec_encode_data(len / 2, k2, p, &g_tbls[k2 * p * 32], - frag_ptrs, &parity_ptrs[p])); - printf("ec_piggyback_encode_sparse: "); - perf_print(start, m2 * len / 2); - } - } + if (benchmark) { + struct perf start; + BENCHMARK(&start, BENCHMARK_TIME, + ec_encode_data(len / 2, k, p, g_tbls_faster, frag_ptrs, + parity_ptrs); + ec_encode_data(len / 2, k2, p, &g_tbls[k2 * p * 32], frag_ptrs, + &parity_ptrs[p])); + printf("ec_piggyback_encode_sparse: "); + perf_print(start, m2 * len / 2); + } + } - if (nerrs <= 0) - return 0; + if (nerrs <= 0) + return 0; - printf(" recover %d fragments\n", nerrs); + printf(" recover %d fragments\n", nerrs); - // Set frag pointers to correspond to parity - for (i = k2; i < m2; i++) - frag_ptrs[i] = parity_ptrs[i - k2]; + // Set frag pointers to correspond to parity + for (i = k2; i < m2; i++) + frag_ptrs[i] = parity_ptrs[i - k2]; - print_list(nerrs2, frag_err_list, " frag err list"); + print_list(nerrs2, frag_err_list, " frag err list"); - // Find a decode matrix to regenerate all erasures from remaining frags - ret = gf_gen_decode_matrix(encode_matrix, decode_matrix, - invert_matrix, temp_matrix, decode_index, frag_err_list, - nerrs2, k2, m2); + // Find a decode matrix to regenerate all erasures from remaining frags + ret = gf_gen_decode_matrix(encode_matrix, decode_matrix, invert_matrix, temp_matrix, + decode_index, frag_err_list, nerrs2, k2, m2); - if (ret != 0) { - printf("Fail on generate decode matrix\n"); - return -1; - } - // Pack recovery array pointers as list of valid fragments - for (i = 0; i < k2; i++) - if (decode_index[i] < k2) - recover_srcs[i] = frag_ptrs[decode_index[i]]; - else - recover_srcs[i] = parity_ptrs[decode_index[i] - k2]; + if (ret != 0) { + printf("Fail on generate decode matrix\n"); + return -1; + } + // Pack recovery array pointers as list of valid fragments + for (i = 0; i < k2; i++) + if (decode_index[i] < k2) + recover_srcs[i] = frag_ptrs[decode_index[i]]; + else + recover_srcs[i] = parity_ptrs[decode_index[i] - k2]; - print_list(k2, decode_index, " decode index"); + print_list(k2, decode_index, " decode index"); - // Recover data - ec_init_tables(k2, nerrs2, decode_matrix, g_tbls); - ec_encode_data(len / 2, k2, nerrs2, g_tbls, recover_srcs, recover_outp); + // Recover data + ec_init_tables(k2, nerrs2, decode_matrix, g_tbls); + ec_encode_data(len / 2, k2, nerrs2, g_tbls, recover_srcs, recover_outp); - if (benchmark) { - struct perf start; - BENCHMARK(&start, BENCHMARK_TIME, - ec_encode_data(len / 2, k2, nerrs2, g_tbls, recover_srcs, - recover_outp)); - printf("ec_piggyback_decode: "); - perf_print(start, (k2 + nerrs2) * len / 2); - } - // Check that recovered buffers are the same as original - printf(" check recovery of block {"); - for (i = 0; i < nerrs2; i++) { - printf(" %d", frag_err_list[i]); - if (memcmp(recover_outp[i], frag_ptrs[frag_err_list[i]], len / 2)) { - printf(" Fail erasure recovery %d, frag %d\n", i, frag_err_list[i]); - return -1; - } - } - printf(" } done all: Pass\n"); + if (benchmark) { + struct perf start; + BENCHMARK(&start, BENCHMARK_TIME, + ec_encode_data(len / 2, k2, nerrs2, g_tbls, recover_srcs, recover_outp)); + printf("ec_piggyback_decode: "); + perf_print(start, (k2 + nerrs2) * len / 2); + } + // Check that recovered buffers are the same as original + printf(" check recovery of block {"); + for (i = 0; i < nerrs2; i++) { + printf(" %d", frag_err_list[i]); + if (memcmp(recover_outp[i], frag_ptrs[frag_err_list[i]], len / 2)) { + printf(" Fail erasure recovery %d, frag %d\n", i, frag_err_list[i]); + return -1; + } + } + printf(" } done all: Pass\n"); - return 0; + return 0; } // Generate decode matrix from encode matrix and erasure list -static int gf_gen_decode_matrix(u8 * encode_matrix, - u8 * decode_matrix, - u8 * invert_matrix, - u8 * temp_matrix, - u8 * decode_index, u8 * frag_err_list, int nerrs, int k, int m) +static int +gf_gen_decode_matrix(u8 *encode_matrix, u8 *decode_matrix, u8 *invert_matrix, u8 *temp_matrix, + u8 *decode_index, u8 *frag_err_list, int nerrs, int k, int m) { - int i, j, p, r; - int nsrcerrs = 0; - u8 s, *b = temp_matrix; - u8 frag_in_err[MMAX]; + int i, j, p, r; + int nsrcerrs = 0; + u8 s, *b = temp_matrix; + u8 frag_in_err[MMAX]; - memset(frag_in_err, 0, sizeof(frag_in_err)); + memset(frag_in_err, 0, sizeof(frag_in_err)); - // Order the fragments in erasure for easier sorting - for (i = 0; i < nerrs; i++) { - if (frag_err_list[i] < k) - nsrcerrs++; - frag_in_err[frag_err_list[i]] = 1; - } + // Order the fragments in erasure for easier sorting + for (i = 0; i < nerrs; i++) { + if (frag_err_list[i] < k) + nsrcerrs++; + frag_in_err[frag_err_list[i]] = 1; + } - // Construct b (matrix that encoded remaining frags) by removing erased rows - for (i = 0, r = 0; i < k; i++, r++) { - while (frag_in_err[r]) - r++; - for (j = 0; j < k; j++) - b[k * i + j] = encode_matrix[k * r + j]; - decode_index[i] = r; - } - if (verbose > 1) - print_matrix(k, k, b, "matrix to invert"); + // Construct b (matrix that encoded remaining frags) by removing erased rows + for (i = 0, r = 0; i < k; i++, r++) { + while (frag_in_err[r]) + r++; + for (j = 0; j < k; j++) + b[k * i + j] = encode_matrix[k * r + j]; + decode_index[i] = r; + } + if (verbose > 1) + print_matrix(k, k, b, "matrix to invert"); - // Invert matrix to get recovery matrix - if (gf_invert_matrix(b, invert_matrix, k) < 0) - return -1; + // Invert matrix to get recovery matrix + if (gf_invert_matrix(b, invert_matrix, k) < 0) + return -1; - if (verbose > 2) - print_matrix(k, k, invert_matrix, "matrix inverted"); + if (verbose > 2) + print_matrix(k, k, invert_matrix, "matrix inverted"); - // Get decode matrix with only wanted recovery rows - for (i = 0; i < nsrcerrs; i++) { - for (j = 0; j < k; j++) { - decode_matrix[k * i + j] = invert_matrix[k * frag_err_list[i] + j]; - } - } + // Get decode matrix with only wanted recovery rows + for (i = 0; i < nsrcerrs; i++) { + for (j = 0; j < k; j++) { + decode_matrix[k * i + j] = invert_matrix[k * frag_err_list[i] + j]; + } + } - // For non-src (parity) erasures need to multiply encode matrix * invert - for (p = nsrcerrs; p < nerrs; p++) { - for (i = 0; i < k; i++) { - s = 0; - for (j = 0; j < k; j++) - s ^= gf_mul(invert_matrix[j * k + i], - encode_matrix[k * frag_err_list[p] + j]); + // For non-src (parity) erasures need to multiply encode matrix * invert + for (p = nsrcerrs; p < nerrs; p++) { + for (i = 0; i < k; i++) { + s = 0; + for (j = 0; j < k; j++) + s ^= gf_mul(invert_matrix[j * k + i], + encode_matrix[k * frag_err_list[p] + j]); - decode_matrix[k * p + i] = s; - } - } - if (verbose > 1) - print_matrix(nerrs, k, decode_matrix, "decode matrix"); - return 0; + decode_matrix[k * p + i] = s; + } + } + if (verbose > 1) + print_matrix(nerrs, k, decode_matrix, "decode matrix"); + return 0; } diff --git a/examples/ec/ec_simple_example.c b/examples/ec/ec_simple_example.c index 82efa6b..e93b3a3 100644 --- a/examples/ec/ec_simple_example.c +++ b/examples/ec/ec_simple_example.c @@ -31,187 +31,186 @@ #include #include #include -#include "erasure_code.h" // use instead when linking against installed +#include "erasure_code.h" // use instead when linking against installed #define MMAX 255 #define KMAX 255 typedef unsigned char u8; -int usage(void) +int +usage(void) { - fprintf(stderr, - "Usage: ec_simple_example [options]\n" - " -h Help\n" - " -k Number of source fragments\n" - " -p Number of parity fragments\n" - " -l Length of fragments\n" - " -e Simulate erasure on frag index val. Zero based. Can be repeated.\n" - " -r Pick random (k, p) with seed\n"); - exit(0); + fprintf(stderr, + "Usage: ec_simple_example [options]\n" + " -h Help\n" + " -k Number of source fragments\n" + " -p Number of parity fragments\n" + " -l Length of fragments\n" + " -e Simulate erasure on frag index val. Zero based. Can be repeated.\n" + " -r Pick random (k, p) with seed\n"); + exit(0); } -static int gf_gen_decode_matrix_simple(u8 * encode_matrix, - u8 * decode_matrix, - u8 * invert_matrix, - u8 * temp_matrix, - u8 * decode_index, - u8 * frag_err_list, int nerrs, int k, int m); +static int +gf_gen_decode_matrix_simple(u8 *encode_matrix, u8 *decode_matrix, u8 *invert_matrix, + u8 *temp_matrix, u8 *decode_index, u8 *frag_err_list, int nerrs, int k, + int m); -int main(int argc, char *argv[]) +int +main(int argc, char *argv[]) { - int i, j, m, c, e, ret; - int k = 10, p = 4, len = 8 * 1024; // Default params - int nerrs = 0; + int i, j, m, c, e, ret; + int k = 10, p = 4, len = 8 * 1024; // Default params + int nerrs = 0; - // Fragment buffer pointers - u8 *frag_ptrs[MMAX]; - u8 *recover_srcs[KMAX]; - u8 *recover_outp[KMAX]; - u8 frag_err_list[MMAX]; + // Fragment buffer pointers + u8 *frag_ptrs[MMAX]; + u8 *recover_srcs[KMAX]; + u8 *recover_outp[KMAX]; + u8 frag_err_list[MMAX]; - // Coefficient matrices - u8 *encode_matrix, *decode_matrix; - u8 *invert_matrix, *temp_matrix; - u8 *g_tbls; - u8 decode_index[MMAX]; + // Coefficient matrices + u8 *encode_matrix, *decode_matrix; + u8 *invert_matrix, *temp_matrix; + u8 *g_tbls; + u8 decode_index[MMAX]; - if (argc == 1) - for (i = 0; i < p; i++) - frag_err_list[nerrs++] = rand() % (k + p); + if (argc == 1) + for (i = 0; i < p; i++) + frag_err_list[nerrs++] = rand() % (k + p); - while ((c = getopt(argc, argv, "k:p:l:e:r:h")) != -1) { - switch (c) { - case 'k': - k = atoi(optarg); - break; - case 'p': - p = atoi(optarg); - break; - case 'l': - len = atoi(optarg); - if (len < 0) - usage(); - break; - case 'e': - e = atoi(optarg); - frag_err_list[nerrs++] = e; - break; - case 'r': - srand(atoi(optarg)); - k = (rand() % (MMAX - 1)) + 1; // Pick k {1 to MMAX - 1} - p = (rand() % (MMAX - k)) + 1; // Pick p {1 to MMAX - k} + while ((c = getopt(argc, argv, "k:p:l:e:r:h")) != -1) { + switch (c) { + case 'k': + k = atoi(optarg); + break; + case 'p': + p = atoi(optarg); + break; + case 'l': + len = atoi(optarg); + if (len < 0) + usage(); + break; + case 'e': + e = atoi(optarg); + frag_err_list[nerrs++] = e; + break; + case 'r': + srand(atoi(optarg)); + k = (rand() % (MMAX - 1)) + 1; // Pick k {1 to MMAX - 1} + p = (rand() % (MMAX - k)) + 1; // Pick p {1 to MMAX - k} - for (i = 0; i < k + p && nerrs < p; i++) - if (rand() & 1) - frag_err_list[nerrs++] = i; - break; - case 'h': - default: - usage(); - break; - } - } - m = k + p; + for (i = 0; i < k + p && nerrs < p; i++) + if (rand() & 1) + frag_err_list[nerrs++] = i; + break; + case 'h': + default: + usage(); + break; + } + } + m = k + p; - // Check for valid parameters - if (m > MMAX || k > KMAX || m < 0 || p < 1 || k < 1) { - printf(" Input test parameter error m=%d, k=%d, p=%d, erasures=%d\n", - m, k, p, nerrs); - usage(); - } - if (nerrs > p) { - printf(" Number of erasures chosen exceeds power of code erasures=%d p=%d\n", - nerrs, p); - usage(); - } - for (i = 0; i < nerrs; i++) { - if (frag_err_list[i] >= m) { - printf(" fragment %d not in range\n", frag_err_list[i]); - usage(); - } - } + // Check for valid parameters + if (m > MMAX || k > KMAX || m < 0 || p < 1 || k < 1) { + printf(" Input test parameter error m=%d, k=%d, p=%d, erasures=%d\n", m, k, p, + nerrs); + usage(); + } + if (nerrs > p) { + printf(" Number of erasures chosen exceeds power of code erasures=%d p=%d\n", nerrs, + p); + usage(); + } + for (i = 0; i < nerrs; i++) { + if (frag_err_list[i] >= m) { + printf(" fragment %d not in range\n", frag_err_list[i]); + usage(); + } + } - printf("ec_simple_example:\n"); + printf("ec_simple_example:\n"); - // Allocate coding matrices - encode_matrix = malloc(m * k); - decode_matrix = malloc(m * k); - invert_matrix = malloc(m * k); - temp_matrix = malloc(m * k); - g_tbls = malloc(k * p * 32); + // Allocate coding matrices + encode_matrix = malloc(m * k); + decode_matrix = malloc(m * k); + invert_matrix = malloc(m * k); + temp_matrix = malloc(m * k); + g_tbls = malloc(k * p * 32); - if (encode_matrix == NULL || decode_matrix == NULL - || invert_matrix == NULL || temp_matrix == NULL || g_tbls == NULL) { - printf("Test failure! Error with malloc\n"); - return -1; - } - // Allocate the src & parity buffers - for (i = 0; i < m; i++) { - if (NULL == (frag_ptrs[i] = malloc(len))) { - printf("alloc error: Fail\n"); - return -1; - } - } + if (encode_matrix == NULL || decode_matrix == NULL || invert_matrix == NULL || + temp_matrix == NULL || g_tbls == NULL) { + printf("Test failure! Error with malloc\n"); + return -1; + } + // Allocate the src & parity buffers + for (i = 0; i < m; i++) { + if (NULL == (frag_ptrs[i] = malloc(len))) { + printf("alloc error: Fail\n"); + return -1; + } + } - // Allocate buffers for recovered data - for (i = 0; i < p; i++) { - if (NULL == (recover_outp[i] = malloc(len))) { - printf("alloc error: Fail\n"); - return -1; - } - } + // Allocate buffers for recovered data + for (i = 0; i < p; i++) { + if (NULL == (recover_outp[i] = malloc(len))) { + printf("alloc error: Fail\n"); + return -1; + } + } - // Fill sources with random data - for (i = 0; i < k; i++) - for (j = 0; j < len; j++) - frag_ptrs[i][j] = rand(); + // Fill sources with random data + for (i = 0; i < k; i++) + for (j = 0; j < len; j++) + frag_ptrs[i][j] = rand(); - printf(" encode (m,k,p)=(%d,%d,%d) len=%d\n", m, k, p, len); + printf(" encode (m,k,p)=(%d,%d,%d) len=%d\n", m, k, p, len); - // Pick an encode matrix. A Cauchy matrix is a good choice as even - // large k are always invertable keeping the recovery rule simple. - gf_gen_cauchy1_matrix(encode_matrix, m, k); + // Pick an encode matrix. A Cauchy matrix is a good choice as even + // large k are always invertable keeping the recovery rule simple. + gf_gen_cauchy1_matrix(encode_matrix, m, k); - // Initialize g_tbls from encode matrix - ec_init_tables(k, p, &encode_matrix[k * k], g_tbls); + // Initialize g_tbls from encode matrix + ec_init_tables(k, p, &encode_matrix[k * k], g_tbls); - // Generate EC parity blocks from sources - ec_encode_data(len, k, p, g_tbls, frag_ptrs, &frag_ptrs[k]); + // Generate EC parity blocks from sources + ec_encode_data(len, k, p, g_tbls, frag_ptrs, &frag_ptrs[k]); - if (nerrs <= 0) - return 0; + if (nerrs <= 0) + return 0; - printf(" recover %d fragments\n", nerrs); + printf(" recover %d fragments\n", nerrs); - // Find a decode matrix to regenerate all erasures from remaining frags - ret = gf_gen_decode_matrix_simple(encode_matrix, decode_matrix, - invert_matrix, temp_matrix, decode_index, - frag_err_list, nerrs, k, m); - if (ret != 0) { - printf("Fail on generate decode matrix\n"); - return -1; - } - // Pack recovery array pointers as list of valid fragments - for (i = 0; i < k; i++) - recover_srcs[i] = frag_ptrs[decode_index[i]]; + // Find a decode matrix to regenerate all erasures from remaining frags + ret = gf_gen_decode_matrix_simple(encode_matrix, decode_matrix, invert_matrix, temp_matrix, + decode_index, frag_err_list, nerrs, k, m); + if (ret != 0) { + printf("Fail on generate decode matrix\n"); + return -1; + } + // Pack recovery array pointers as list of valid fragments + for (i = 0; i < k; i++) + recover_srcs[i] = frag_ptrs[decode_index[i]]; - // Recover data - ec_init_tables(k, nerrs, decode_matrix, g_tbls); - ec_encode_data(len, k, nerrs, g_tbls, recover_srcs, recover_outp); + // Recover data + ec_init_tables(k, nerrs, decode_matrix, g_tbls); + ec_encode_data(len, k, nerrs, g_tbls, recover_srcs, recover_outp); - // Check that recovered buffers are the same as original - printf(" check recovery of block {"); - for (i = 0; i < nerrs; i++) { - printf(" %d", frag_err_list[i]); - if (memcmp(recover_outp[i], frag_ptrs[frag_err_list[i]], len)) { - printf(" Fail erasure recovery %d, frag %d\n", i, frag_err_list[i]); - return -1; - } - } + // Check that recovered buffers are the same as original + printf(" check recovery of block {"); + for (i = 0; i < nerrs; i++) { + printf(" %d", frag_err_list[i]); + if (memcmp(recover_outp[i], frag_ptrs[frag_err_list[i]], len)) { + printf(" Fail erasure recovery %d, frag %d\n", i, frag_err_list[i]); + return -1; + } + } - printf(" } done all: Pass\n"); - return 0; + printf(" } done all: Pass\n"); + return 0; } /* @@ -219,59 +218,56 @@ int main(int argc, char *argv[]) * */ -static int gf_gen_decode_matrix_simple(u8 * encode_matrix, - u8 * decode_matrix, - u8 * invert_matrix, - u8 * temp_matrix, - u8 * decode_index, u8 * frag_err_list, int nerrs, int k, - int m) +static int +gf_gen_decode_matrix_simple(u8 *encode_matrix, u8 *decode_matrix, u8 *invert_matrix, + u8 *temp_matrix, u8 *decode_index, u8 *frag_err_list, int nerrs, int k, + int m) { - int i, j, p, r; - int nsrcerrs = 0; - u8 s, *b = temp_matrix; - u8 frag_in_err[MMAX]; + int i, j, p, r; + int nsrcerrs = 0; + u8 s, *b = temp_matrix; + u8 frag_in_err[MMAX]; - memset(frag_in_err, 0, sizeof(frag_in_err)); + memset(frag_in_err, 0, sizeof(frag_in_err)); - // Order the fragments in erasure for easier sorting - for (i = 0; i < nerrs; i++) { - if (frag_err_list[i] < k) - nsrcerrs++; - frag_in_err[frag_err_list[i]] = 1; - } + // Order the fragments in erasure for easier sorting + for (i = 0; i < nerrs; i++) { + if (frag_err_list[i] < k) + nsrcerrs++; + frag_in_err[frag_err_list[i]] = 1; + } - // Construct b (matrix that encoded remaining frags) by removing erased rows - for (i = 0, r = 0; i < k; i++, r++) { - while (frag_in_err[r]) - r++; - for (j = 0; j < k; j++) - b[k * i + j] = encode_matrix[k * r + j]; - decode_index[i] = r; - } + // Construct b (matrix that encoded remaining frags) by removing erased rows + for (i = 0, r = 0; i < k; i++, r++) { + while (frag_in_err[r]) + r++; + for (j = 0; j < k; j++) + b[k * i + j] = encode_matrix[k * r + j]; + decode_index[i] = r; + } - // Invert matrix to get recovery matrix - if (gf_invert_matrix(b, invert_matrix, k) < 0) - return -1; + // Invert matrix to get recovery matrix + if (gf_invert_matrix(b, invert_matrix, k) < 0) + return -1; - // Get decode matrix with only wanted recovery rows - for (i = 0; i < nerrs; i++) { - if (frag_err_list[i] < k) // A src err - for (j = 0; j < k; j++) - decode_matrix[k * i + j] = - invert_matrix[k * frag_err_list[i] + j]; - } + // Get decode matrix with only wanted recovery rows + for (i = 0; i < nerrs; i++) { + if (frag_err_list[i] < k) // A src err + for (j = 0; j < k; j++) + decode_matrix[k * i + j] = invert_matrix[k * frag_err_list[i] + j]; + } - // For non-src (parity) erasures need to multiply encode matrix * invert - for (p = 0; p < nerrs; p++) { - if (frag_err_list[p] >= k) { // A parity err - for (i = 0; i < k; i++) { - s = 0; - for (j = 0; j < k; j++) - s ^= gf_mul(invert_matrix[j * k + i], - encode_matrix[k * frag_err_list[p] + j]); - decode_matrix[k * p + i] = s; - } - } - } - return 0; + // For non-src (parity) erasures need to multiply encode matrix * invert + for (p = 0; p < nerrs; p++) { + if (frag_err_list[p] >= k) { // A parity err + for (i = 0; i < k; i++) { + s = 0; + for (j = 0; j < k; j++) + s ^= gf_mul(invert_matrix[j * k + i], + encode_matrix[k * frag_err_list[p] + j]); + decode_matrix[k * p + i] = s; + } + } + } + return 0; }