all-in-one makefile to check speed of sqrt

Makefile

bench: sqrttest
	./sqrttest

sqrttest: sqrttest.c tst.c
	rm -f sqrttest.o tst.o $@
	mpicc -std=gnu99 -march=native -Ofast -g3 -fopenmp -fno-lto -c sqrttest.c
	mpicc -std=gnu99 -march=native -Ofast -g3 -fopenmp -fno-lto -c tst.c
	mpicc -O0 -g3 -fopenmp -fno-lto sqrttest.o tst.o -o $@ -lm

define SQRTTEST_BODY
#include <math.h>
#include <mpi.h>
#include <stdio.h>
#include <stdlib.h>
#include <stddef.h>

#define SZ (1UL<<26)
#define ALIGN 64

static double *a; /* input */
static double *b; /* output */

extern void tst1(double * restrict a, double * restrict b, const size_t sz);
extern void tst2(double * restrict a, double * restrict b, const size_t sz);

int main(void) {
  posix_memalign((void**)&a, ALIGN, sizeof(*a)*SZ);
  posix_memalign((void**)&b, ALIGN, sizeof(*b)*SZ);
  for(size_t i = 0 ; i < SZ ; i++) {
    a[i] = 1.234567890; /* something non-zero to make sqrt do work */
    b[i] = 0.;          /* touch destination to get memory pages */
  }
  {
    const double start = MPI_Wtime();
    tst1(a, b, SZ);
    const double end = MPI_Wtime();
    double sum = 0.;
    for(size_t i = 0 ; i < SZ ; i++) {
      sum += b[i];
    }
    printf("sqrt took %gms for %g\n", (end-start)*1e3, sum);
  }
  {
    const double start = MPI_Wtime();
    tst2(a, b, SZ);
    const double end = MPI_Wtime();
    double sum = 0.;
    for(size_t i = 0 ; i < SZ ; i++) {
      sum += b[i];
    }
    printf("copy took %gms for %g\n", (end-start)*1e3, sum);
  }
  return 0;
}
endef
export SQRTTEST_BODY
sqrttest.c: Makefile
	echo "$$SQRTTEST_BODY" >$@

define TST_BODY
#include <math.h>
#include <string.h>
#include <stddef.h>

#define ALIGN 64

void tst1(double * restrict a, double * restrict b, const size_t sz)
{
  a = __builtin_assume_aligned(a, ALIGN);
  b = __builtin_assume_aligned(b, ALIGN);
#pragma omd simd
  for(size_t i = 0 ; i < sz ; i++) {
    b[i] = sqrt(a[i]);
  }
}

void tst2(double * restrict a, double * restrict b, const size_t sz)
{
  a = __builtin_assume_aligned(a, ALIGN);
  b = __builtin_assume_aligned(b, ALIGN);
#pragma omd simd
  for(size_t i = 0 ; i < sz ; i++) {
    b[i] = a[i];
  }
}
endef
export TST_BODY

tst.c: Makefile
	echo "$$TST_BODY" >$@