X-Git-Url: http://git.tuebingen.mpg.de/?p=paraslash.git;a=blobdiff_plain;f=fec.c;h=2301cc8d2d4653b8bc82a8bcc46b5867b5d7af14;hp=d097b77de836248d79142348f254dd4c3ef57817;hb=3685a9093ae12ff9ce02fc58e607eb9b63894443;hpb=fb62100d23baea388cec78e572b0f36d051a61c0 diff --git a/fec.c b/fec.c index d097b77d..2301cc8d 100644 --- a/fec.c +++ b/fec.c @@ -41,24 +41,40 @@ #include "string.h" #include "fec.h" -#define GF_BITS 8 /* code over GF(256) */ +/** Code over GF(256). */ +#define GF_BITS 8 +/** The largest number in GF(256) */ #define GF_SIZE ((1 << GF_BITS) - 1) /* * To speed up computations, we have tables for logarithm, exponent and inverse - * of a number. We use a table for multiplication as well (it takes 64K, no big - * deal even on a PDA, especially because it can be pre-initialized an put into - * a ROM!). The macro gf_mul(x,y) takes care of multiplications. + * of a number. + */ + +/** Index->poly form conversion table. */ +static unsigned char gf_exp[2 * GF_SIZE]; + +/** Poly->index form conversion table. */ +static int gf_log[GF_SIZE + 1]; + +/** Inverse of a field element. */ +static unsigned char inverse[GF_SIZE + 1]; + +/** + * The multiplication table. + * + * We use a table for multiplication as well. It takes 64K, no big deal even on + * a PDA, especially because it can be pre-initialized and put into a ROM. + * + * \sa \ref gf_mul. */ -static unsigned char gf_exp[2 * GF_SIZE]; /* index->poly form conversion table */ -static int gf_log[GF_SIZE + 1]; /* Poly->index form conversion table */ -static unsigned char inverse[GF_SIZE + 1]; /* inverse of field elem. */ static unsigned char gf_mul_table[GF_SIZE + 1][GF_SIZE + 1]; -/* Multiply two numbers. */ + +/** Multiply two GF numbers. */ #define gf_mul(x,y) gf_mul_table[x][y] /* Compute x % GF_SIZE without a slow divide. */ -static inline unsigned char modnn(int x) +__a_const static inline unsigned char modnn(int x) { while (x >= GF_SIZE) { x -= GF_SIZE; @@ -154,21 +170,27 @@ static void generate_gf(void) inverse[i] = gf_exp[GF_SIZE - gf_log[i]]; } +/** How often the loop is unrolled. */ +#define UNROLL 16 + /* * Compute dst[] = dst[] + c * src[] * * This is used often, so better optimize it! Currently the loop is unrolled 16 * times. The case c=0 is also optimized, whereas c=1 is not. */ -#define UNROLL 16 -static void addmul(unsigned char *dst1, const unsigned char const *src1, +static void addmul(unsigned char *dst1, const unsigned char *src1, unsigned char c, int sz) { + unsigned char *dst, *lim, *col; + const unsigned char *src = src1; + if (c == 0) return; - unsigned char *dst = dst1, *lim = &dst[sz - UNROLL + 1], - *col = gf_mul_table[c]; - const unsigned char const *src = src1; + + dst = dst1; + lim = &dst[sz - UNROLL + 1]; + col = gf_mul_table[c]; for (; dst < lim; dst += UNROLL, src += UNROLL) { dst[0] ^= col[src[0]]; @@ -211,6 +233,7 @@ static void matmul(unsigned char *a, unsigned char *b, unsigned char *c, } } +/** Swap two numbers. */ #define FEC_SWAP(a,b) {typeof(a) tmp = a; a = b; b = tmp;} /*