Vectorization Post: 302895149

Sponsored Content

Special Forums UNIX and Linux Applications High Performance Computing Vectorization Post 302895149 by a3mlord on Friday 28th of March 2014 05:03:17 PM

03-28-2014

Registered User

Vectorization

Hi,

I have the following vectorized code:

Code:

long valor = 0, i=0;

 __m128i vsum, vecPi, vecCi, vecQCi;

 vsum = _mm_set1_epi32(0);

 int32_t * const pA = A->data;
 int32_t * const pB = B->data;

 int sumDot[1];

 for( ; i<SIZE-3 ;i+=4){

 vecPi = _mm_loadu_si128((__m128i *)&(pA)[i] );
 vecCi = _mm_loadu_si128((__m128i *)&(pB)[i] );
 vecQCi = _mm_mullo_epi32(vecPi,vecCi);
 vsum = _mm_add_epi32(vsum,vecQCi);

 } 
 vsum = _mm_hadd_epi32(vsum, vsum);
 vsum = _mm_hadd_epi32(vsum, vsum);
 _mm_storeu_si128((__m128i *)&(sumDot), vsum);

 for( ; i<SIZE; i++)
 valor += A->data[i] * B->data[i];   valor += sumDot[0];

However, as I get overflows, I need to handle those cases. Could you please help me with that?

Thanks

Last edited by bartus11; 03-28-2014 at 06:06 PM.. Reason: Please use [code][/code] tags.

a3mlord

View Public Profile for a3mlord

Find all posts by a3mlord

LEARN ABOUT OPENSOLARIS

ntp_adjtime

ntp_adjtime(2)							   System Calls 						    ntp_adjtime(2)

NAME

       ntp_adjtime - adjust local clock parameters

SYNOPSIS

       #include <sys/timex.h>

       int ntp_adjtime(struct timex *tptr);

DESCRIPTION

       The  ntp_adjtime()  function adjusts the parameters used to discipline the local clock, according to the values in the struct timex pointed
       to by  tptr. Before returning, it fills in the structure with the most recent values kept in the kernel.

       The adjustment is effected in part by speeding up or slowing down the clock, as necessary, and in part by  phase-locking  onto  a  once-per
       second pulse (PPS) provided by a driver, if available.

	 struct timex {
	     uint32_t modes;	    /* clock mode bits (w) */
	     int32_t  offset;	    /* time offset (us) (rw) */
	     int32_t  freq;	    /* frequency offset (scaled ppm) (rw) */
	     int32_t  maxerror;     /* maximum error (us) (rw) */
	     int32_t  esterror;     /* estimated error (us) (rw) */
	     int32_t  status;	    /* clock status bits (rw) */
	     int32_t  constant;     /* pll time constant (rw) */
	     int32_t  precision;    /* clock precision (us) (r) */
	     int32_t  tolerance;    /* clock frequency tolerance
				       (scaled ppm) (r) */
	     int32_t  ppsfreq;	    /* pps frequency (scaled ppm) (r) */
	     int32_t  jitter;	    /* pps jitter (us) (r) */
	     int32_t  shift;	    /* interval duration (s) (shift) (r) */
	     int32_t  stabil;	    /* pps stability (scaled ppm) (r) */
	     int32_t  jitcnt;	    /* jitter limit exceeded (r) */
	     int32_t  calcnt;	    /* calibration intervals (r) */
	     int32_t  errcnt;	    /* calibration errors (r) */
	     int32_t  stbcnt;	    /* stability limit exceeded (r) */
	 };

RETURN VALUES

       Upon  successful  completion, ntp_adjtime() returns the current clock state (see <sys/timex.h>). Otherwise, it returns -1 and sets errno to
       indicate the error.

ERRORS

       The ntp_adjtime() function will fail if:

       EFAULT	 The  tptr argument is an invalid pointer.

       EINVAL	 The constant member of the structure pointed to by tptr is less than 0 or greater than 30.

       EPERM	 The {PRIV_SYS_TIME} privilege is not asserted in the effective set of the calling process.

SEE ALSO

       xntpd(1M), ntp_gettime(2), privileges(5)

SunOS 5.11							    1 Feb 2003							    ntp_adjtime(2)