#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <math.h>
#include <mkl.h>
#include <mkl_dfti.h>
#include <immintrin.h>
#include<time.h>

float *input;
float *output;

#define nFFT (98304)

DFTI_DESCRIPTOR_HANDLE HandlerIdft;
MKL_LONG status = 0;
int main(void)
{
__m512i in;
__m512i out;
input = (float *)malloc(nFFT*2*sizeof(float));
output = (float *)malloc(nFFT*2*sizeof(float));
memset(output,0,nFFT*2);
/********************print input*********************/
printf("input:  ");
for(int i = 0; i < nFFT*2; i++)
{
        input[i] = i+1;
	//printf(" %f ",*(input+i));
}
printf("\n");
/****************************************************/
float para = 1.0f / nFFT;
/*
Matlab IFFT scale: para = 1.0f / nFFT
x86 MKL DftBackward scale: para = 1.0f / sqrt(nFFT) 
*/

int irc = mkl_enable_instructions(MKL_ENABLE_AVX512);
printf("irc-%d\n", irc);
status = DftiCreateDescriptor(&HandlerIdft, DFTI_SINGLE, DFTI_COMPLEX, 1, nFFT);
/* HandlerIdft: Descriptor Handler
   DFTI_SINGLE: Precision of the computation - DFTI_SINGLE or DFTI_DOUBLE
   DFTI_COMPLEX: Type of the transform - DFTI_COMPLEX or DFTI_REAL
   1(DFTI_DIMENSION): Integer
   DFTI_LENGTHS: DFTI_LENGTHS
*/
status = DftiSetValue(HandlerIdft, DFTI_BACKWARD_SCALE, para);
/* DFTI_FORWARD_SCALE/DFTI_BACKWARD_SCALE:
	Floating-point scalar
	Scale factor for the forward transform/backward transform
*/
status = DftiSetValue(HandlerIdft, DFTI_PLACEMENT, DFTI_NOT_INPLACE);
/* DFTI_PLACEMENT: Named constant DFTI_INPLACE or DFTI_NOT_INPLACE
	Defines whether the result overwrites the input data.
	Default value: DFTI_INPLACE.
*/
//status = DftiSetValue(HandlerIdft, DFTI_THREAD_LIMIT, 1);
status = DftiCommitDescriptor(HandlerIdft);
/* ================================================ */
/*              INTRINSICS TO BE TESTED             */
/* ================================================ */
struct timespec CrcSendtime;
struct timespec endtime;
clock_gettime(CLOCK_REALTIME, &CrcSendtime);				

status = DftiComputeBackward(HandlerIdft, input, output);

clock_gettime(CLOCK_REALTIME, &endtime);
printf("msec [%f]\n", (float)(endtime.tv_nsec - CrcSendtime.tv_nsec)/1000000);
printf("status[%d]\n ", *(int *)(&status));
/********************print output********************/
printf("output: ");
for(int i = 0; i < 66; i++)
{
	//printf(" %f ",*(output+i));
}
printf("\n");
/****************************************************/

status = DftiFreeDescriptor(&HandlerIdft);
return 0;
}