#include <stdio.h>
#include <stdlib.h>
#include <stdint.h>
#include <string.h>
#include <math.h>
#include <limits.h>
#include <time.h>

#include "nmglobal.h"
#include "randnum.h"


/************************
** BITFIELD OPERATIONS **
*************************/

static clock_t DoBitfieldIteration(unsigned long *bitarraybase,
		unsigned long *bitoparraybase,
		long bitoparraysize,
		unsigned long *nbitops);
static void ToggleBitRun(unsigned long *bitmap,
		unsigned long bit_addr,
		unsigned long nbits,
		unsigned int val);
static void FlipBitRun(unsigned long *bitmap,
		unsigned long bit_addr,
		unsigned long nbits);

/*************
** DoBitops **
**************
** Perform the bit operations test portion of the CPU
** benchmark.  Returns the iterations per second.
*/
void
DoBitops(void)
{
        const char*     context = "CPU:Bitfields";
        BitOpStruct*    locbitopstruct = &global_bitopstruct;
        clock_t         total_time = 0;
        int             iterations = 0;
        unsigned long*  bitarraybase = NULL;
        unsigned long*  bitoparraybase = NULL;
        unsigned long   nbitops;

        /*
        ** See if we need to run adjustment code.
        */
        if (locbitopstruct->adjust == FALSE) {
                locbitopstruct->adjust = TRUE;

                bitarraybase = realloc(bitarraybase, locbitopstruct->bitfieldarraysize * sizeof(unsigned long));
                if (!bitarraybase) {
                      fprintf(stderr, "Error in %s, could not allocate memory. Exitting...\n", context);
                      exit(1);
                }

	        /*
        	** Initialize bitfield operations array to [2,30] elements
                */
                locbitopstruct->bitoparraysize = 30L;

                while(1) {
                        bitoparraybase = malloc(locbitopstruct->bitoparraysize * 2 * sizeof(unsigned long));
                        if (!bitoparraybase) {
                                fprintf(stderr, "Error in %s, could not allocate memory. Exitting...\n", context);
                                free(bitarraybase);
                                exit(1);
                        }

                        /*
                        ** Do an iteration of the bitmap test.  If the
                        ** elapsed time is less than or equal to the permitted
                        ** minimum, then de-allocate the array, reallocate a
                        ** larger version, and try again.
                        */
                        if(DoBitfieldIteration(bitarraybase, bitoparraybase, locbitopstruct->bitoparraysize, &nbitops) > global_min_ticks) {
                                break;
                        }

                        locbitopstruct->bitoparraysize += 100L;
                }
        } else {
                /*
                ** Don't need to do self adjustment, just allocate
                ** the array space.
                */
                bitarraybase = malloc(locbitopstruct->bitfieldarraysize * sizeof(unsigned long));
                if (!bitarraybase) {
                        fprintf(stderr, "Error in %s, could not allocate memory. Exitting...\n", context);
                        exit(1);
                }

                bitoparraybase = malloc(locbitopstruct->bitoparraysize * 2 * sizeof(unsigned long));
                if (!bitoparraybase) {
                        fprintf(stderr, "Error in %s, could not allocate memory. Exitting...\n", context);
                        free(bitarraybase);
                        exit(1);
                }
        }

        /*
        ** All's well if we get here.  Repeatedly perform bitops until the
        ** accumulated elapsed time is greater than # of seconds requested.
        */
        do {
                total_time += DoBitfieldIteration(bitarraybase, bitoparraybase, locbitopstruct->bitoparraysize,&nbitops);
                iterations += nbitops;
        } while (total_time / CLOCKS_PER_SEC < locbitopstruct->request_secs);

        /*
        ** Clean up, calculate results, and go home.
        ** Also, set adjustment flag to show that we don't have
        ** to do self adjusting in the future.
        */
        free(bitarraybase);
        free(bitoparraybase);

        locbitopstruct->bitopspersec = (double)(iterations * CLOCKS_PER_SEC) / (double)total_time;
}

/************************
** DoBitfieldIteration **
*************************
** Perform a single iteration of the bitfield benchmark.
** Return the # of ticks accumulated by the operation.
*/
static clock_t
DoBitfieldIteration(unsigned long *bitarraybase, unsigned long *bitoparraybase, long bitoparraysize, unsigned long *nbitops)
{
        clock_t start, stop;
        long i;
        unsigned long bitoffset;

        /*
        ** Clear # bitops counter
        */
        *nbitops = 0L;

        /*
        ** Construct a set of bitmap offsets and run lengths.
        ** The offset can be any random number from 0 to the
        ** size of the bitmap (in bits).  The run length can
        ** be any random number from 1 to the number of bits
        ** between the offset and the end of the bitmap.
        ** Note that the bitmap has 8192 * 32 bits in it.
        ** (262,144 bits)
        */

        /*
        ** Reset random number generator so things repeat.
        ** Also reset the bit array we work on.
        ** added by Uwe F. Mayer
        */
        randnum(13);

        for (i = 0; i < global_bitopstruct.bitfieldarraysize; i++) {
#ifdef _LP64
                *(bitarraybase+i) = 0x5555555555555555UL;
#else
                *(bitarraybase+i) = 0x55555555UL;
#endif
        }

        randnum(13);

        for (i = 0; i < bitoparraysize; i++) {
                /* First item is offset */
                *(bitoparraybase + i + i) = bitoffset = abs_randwc((int32_t)262140);

                /* Next item is run length */
                *nbitops += *(bitoparraybase + i + i + 1L) = abs_randwc((int32_t)262140 - bitoffset);
        }

        start = clock();

        for(i = 0; i < bitoparraysize; i++) {
                switch(i % 3) {
                case 2: /* Complement run of bits */
                        FlipBitRun(bitarraybase,
                                *(bitoparraybase+i+i),
                                *(bitoparraybase+i+i+1));
                        break;
                default:
                        ToggleBitRun(bitarraybase,
                                *(bitoparraybase+i+i),
                                *(bitoparraybase+i+i+1),
                                !i);
                        break;
                }
        }

        stop = clock();

        return stop - start;
}

/***************
** FlipBitRun **
****************
** Complements a run of bits.
*/
static void
FlipBitRun(unsigned long *bitmap, unsigned long bit_addr, unsigned long nbits) 
{
        unsigned long bindex;   /* Index into array */
        unsigned long bitnumb;  /* Bit number */

        while (nbits--) {
#ifdef _LP64
                bindex=bit_addr>>6;     /* Index is number /64 */
                bitnumb=bit_addr % 64;  /* Bit number in longword */
#else
                bindex=bit_addr>>5;     /* Index is number /32 */
                bitnumb=bit_addr % 32;  /* Bit number in longword */
#endif
                bitmap[bindex]^=(1L<<bitnumb);
                bit_addr++;
        }
}

/*****************************
**     ToggleBitRun          *
******************************
** Set or clear a run of nbits starting at
** bit_addr in bitmap.
*/
void ToggleBitRun(unsigned long *bitmap, /* Bitmap */
		unsigned long bit_addr,         /* Address of bits to set */
		unsigned long nbits,            /* # of bits to set/clr */
		unsigned int val)               /* 1 or 0 */
{
	unsigned long bindex;   /* Index into array */
	unsigned long bitnumb;  /* Bit number */

	if (val) {
		for (; nbits != 0; nbits--) {
			bindex = bit_addr >> 6;
			bitnumb = bit_addr % 64;

			bitmap[bindex] |= (1L << bitnumb);

			bit_addr++;
		}
	} else {
		for (; nbits != 0; nbits--) {
			bindex = bit_addr >> 6;
			bitnumb = bit_addr % 64;

			bitmap[bindex] &= ~(1L << bitnumb);

			bit_addr++;
		}
	}	
}