commit d286d445416afe9b566e4fa29526aac0f2617e2a Author: David Sorber Date: Wed Oct 28 20:25:36 2015 -0400 Adding bit position test program used to benchmark various approaches and a couple of support scripts. diff --git a/software/misc/bit_position/bit_position b/software/misc/bit_position/bit_position new file mode 100755 index 0000000..d6513e9 Binary files /dev/null and b/software/misc/bit_position/bit_position differ diff --git a/software/misc/bit_position/bit_position.c b/software/misc/bit_position/bit_position.c new file mode 100644 index 0000000..22724f8 --- /dev/null +++ b/software/misc/bit_position/bit_position.c @@ -0,0 +1,225 @@ + +#define _XOPEN_SOURCE + +#include +#include +#include +#include +#include +#include + +#include + +// To compile: +// gcc -std=c99 -mlzcnt -O3 -mtune=native -o bit_position bit_position.c + +int timeval_subtract(struct timeval *result, struct timeval *x, struct timeval *y) +{ + /* Perform the carry for the later subtraction by updating y. */ + if (x->tv_usec < y->tv_usec) { + int nsec = (y->tv_usec - x->tv_usec) / 1000000 + 1; + y->tv_usec -= 1000000 * nsec; + y->tv_sec += nsec; + } + if (x->tv_usec - y->tv_usec > 1000000) { + int nsec = (x->tv_usec - y->tv_usec) / 1000000; + y->tv_usec += 1000000 * nsec; + y->tv_sec -= nsec; + } + /* Compute the time remaining to wait. + tv_usec is certainly positive. */ + result->tv_sec = x->tv_sec - y->tv_sec; + result->tv_usec = x->tv_usec - y->tv_usec; + /* Return 1 if result is negative. */ + return x->tv_sec < y->tv_sec; +} + +// Approach 0: naive version +void approach0(uint32_t count, uint64_t* vectors) +{ + for (uint32_t idx = 0; idx < count; idx++) { + uint64_t copy = vectors[idx]; + + printf("Index: %" PRIu32 "\n", idx); + + uint64_t compare = 1; + + for (uint32_t ctr = 0; ctr < 64; ctr++) + { + if (compare & copy) + { + printf("%" PRIu32 " ", ctr); + } + + compare = (compare << 1); + } + + printf("\n\n"); + } +} + +// Approach 1: use intrinsic and bitwise clear +void approach1(uint32_t count, uint64_t* vectors) +{ + for (uint32_t idx = 0; idx < count; idx++) { + uint64_t copy = vectors[idx]; + + printf("Index: %" PRIu32 "\n", idx); + + uint64_t position = 0; + while (copy) { + position = __builtin_ctzll(copy); + printf("%" PRIu64 " ", position); + copy &= ~((uint64_t)1 << position); + } + + printf("\n\n"); + } +} + +// Approach 2: use intrinsic and lookup table +void approach2(uint32_t count, uint64_t* vectors) +{ + uint64_t masks[64] = { + 0xFFFFFFFFFFFFFFFE, + 0xFFFFFFFFFFFFFFFD, + 0xFFFFFFFFFFFFFFFB, + 0xFFFFFFFFFFFFFFF7, + 0xFFFFFFFFFFFFFFEF, + 0xFFFFFFFFFFFFFFDF, + 0xFFFFFFFFFFFFFFBF, + 0xFFFFFFFFFFFFFF7F, + 0xFFFFFFFFFFFFFEFF, + 0xFFFFFFFFFFFFFDFF, + 0xFFFFFFFFFFFFFBFF, + 0xFFFFFFFFFFFFF7FF, + 0xFFFFFFFFFFFFEFFF, + 0xFFFFFFFFFFFFDFFF, + 0xFFFFFFFFFFFFBFFF, + 0xFFFFFFFFFFFF7FFF, + 0xFFFFFFFFFFFEFFFF, + 0xFFFFFFFFFFFDFFFF, + 0xFFFFFFFFFFFBFFFF, + 0xFFFFFFFFFFF7FFFF, + 0xFFFFFFFFFFEFFFFF, + 0xFFFFFFFFFFDFFFFF, + 0xFFFFFFFFFFBFFFFF, + 0xFFFFFFFFFF7FFFFF, + 0xFFFFFFFFFEFFFFFF, + 0xFFFFFFFFFDFFFFFF, + 0xFFFFFFFFFBFFFFFF, + 0xFFFFFFFFF7FFFFFF, + 0xFFFFFFFFEFFFFFFF, + 0xFFFFFFFFDFFFFFFF, + 0xFFFFFFFFBFFFFFFF, + 0xFFFFFFFF7FFFFFFF, + 0xFFFFFFFEFFFFFFFF, + 0xFFFFFFFDFFFFFFFF, + 0xFFFFFFFBFFFFFFFF, + 0xFFFFFFF7FFFFFFFF, + 0xFFFFFFEFFFFFFFFF, + 0xFFFFFFDFFFFFFFFF, + 0xFFFFFFBFFFFFFFFF, + 0xFFFFFF7FFFFFFFFF, + 0xFFFFFEFFFFFFFFFF, + 0xFFFFFDFFFFFFFFFF, + 0xFFFFFBFFFFFFFFFF, + 0xFFFFF7FFFFFFFFFF, + 0xFFFFEFFFFFFFFFFF, + 0xFFFFDFFFFFFFFFFF, + 0xFFFFBFFFFFFFFFFF, + 0xFFFF7FFFFFFFFFFF, + 0xFFFEFFFFFFFFFFFF, + 0xFFFDFFFFFFFFFFFF, + 0xFFFBFFFFFFFFFFFF, + 0xFFF7FFFFFFFFFFFF, + 0xFFEFFFFFFFFFFFFF, + 0xFFDFFFFFFFFFFFFF, + 0xFFBFFFFFFFFFFFFF, + 0xFF7FFFFFFFFFFFFF, + 0xFEFFFFFFFFFFFFFF, + 0xFDFFFFFFFFFFFFFF, + 0xFBFFFFFFFFFFFFFF, + 0xF7FFFFFFFFFFFFFF, + 0xEFFFFFFFFFFFFFFF, + 0xDFFFFFFFFFFFFFFF, + 0xBFFFFFFFFFFFFFFF, + 0x7FFFFFFFFFFFFFFF}; + + for (uint32_t idx = 0; idx < count; idx++) { + uint64_t copy = vectors[idx]; + + printf("Index: %" PRIu32 "\n", idx); + + uint64_t position = 0; + while (copy) { + position = __builtin_ctzll(copy); + printf("%" PRIu64 " ", position); + copy &= masks[position]; + } + + printf("\n\n"); + } +} + +int main(int argc, char** argv) +{ + // Make sure an argument, the file name of the file containing the vectors + // was specified + if (argc < 3) { + fprintf(stderr, "Please specify file containing vectors and approach " + "index (0-2)\n"); + return -1; + } + + // Loop over the file and convert the string vectors into uint64_t's + // stored in the "vectors" array + uint64_t* vectors = NULL; + uint32_t count = 0; + FILE* file = fopen(argv[1], "r"); + char line[256]; + + while (fgets(line, sizeof(line), file)) { + count++; + + if (vectors == NULL) { + vectors = (uint64_t*)malloc(sizeof(uint64_t)); + } else { + vectors = realloc(vectors, count * sizeof(uint64_t)); + } + + vectors[count - 1] = strtoull(line, NULL, 16); + //~ printf("%s", line); + } + + fclose(file); + + // Start time + struct timeval t0_time; + gettimeofday(&t0_time, NULL); + + // Decide which approach to use + switch (argv[2][0]) { + case '0': + approach0(count, vectors); + break; + case '1': + approach1(count, vectors); + break; + case '2': + approach2(count, vectors); + break; + default: + fprintf(stderr, "ERROR: invalid approach index\n"); + } + + // Grab stop time + struct timeval t1_time; + gettimeofday(&t1_time, NULL); + + struct timeval elapsed; + timeval_subtract(&elapsed, &t1_time, &t0_time); + printf("Elapsed time: %ld.%06ld s\n", elapsed.tv_sec, elapsed.tv_usec); + + return 0; +} diff --git a/software/misc/bit_position/make_vectors.py b/software/misc/bit_position/make_vectors.py new file mode 100644 index 0000000..1f25625 --- /dev/null +++ b/software/misc/bit_position/make_vectors.py @@ -0,0 +1,23 @@ +import sys + +def main(): + + if len(sys.argv) < 2: + print('ERROR: must specify number of output vectors') + return 1 + + num_vectors = 0 + try: + num_vectors = int(sys.argv[1]) + except ValueError: + print('ERROR: argument must be an integer') + return 1 + + with open('/dev/urandom', 'rb') as random_input: + for idx in range(num_vectors): + raw = random_input.read(8) + vector = int.from_bytes(raw, byteorder='little') + print('{:016X}'.format(vector)) + +if __name__ == '__main__': + sys.exit(main()) diff --git a/software/misc/walking_zeros.py b/software/misc/walking_zeros.py new file mode 100644 index 0000000..873fcf4 --- /dev/null +++ b/software/misc/walking_zeros.py @@ -0,0 +1,15 @@ +import sys + +def main(): + + # Print out ASCII hex walking zeros pattern + base = 0xffffffffffffffff + + for idx in range(64): + foo = base + foo &= ~(1 <