const struct operator_s* fft_create2(unsigned int D, const long dimensions[D], unsigned long flags, const long ostrides[D], complex float* dst, const long istrides[D], const complex float* src, bool backwards)
{
PTR_ALLOC(struct fft_plan_s, plan);
SET_TYPEID(fft_plan_s, plan);
plan->fftw = fft_fftwf_plan(D, dimensions, flags, ostrides, dst, istrides, src, backwards, false);
#ifdef USE_CUDA
plan->cuplan = NULL;
#ifndef LAZY_CUDA
if (cuda_ondevice(src))
plan->cuplan = fft_cuda_plan(D, dimensions, flags, ostrides, istrides, backwards);
#else
plan->D = D;
plan->flags = flags;
plan->backwards = backwards;
PTR_ALLOC(long[D], dims);
md_copy_dims(D, *dims, dimensions);
plan->dims = *PTR_PASS(dims);
PTR_ALLOC(long[D], istrs);
md_copy_strides(D, *istrs, istrides);
plan->istrs = *PTR_PASS(istrs);
PTR_ALLOC(long[D], ostrs);
md_copy_strides(D, *ostrs, ostrides);
plan->ostrs = *PTR_PASS(ostrs);
#endif
#endif
return operator_create2(D, dimensions, ostrides, D, dimensions, istrides, CAST_UP(PTR_PASS(plan)), fft_apply, fft_free_plan);
}
const struct operator_s* fft_create2(unsigned int D, const long dimensions[D], unsigned long flags, const long ostrides[D], complex float* dst, const long istrides[D], const complex float* src, bool backwards)
{
PTR_ALLOC(struct fft_plan_s, plan);
plan->fftw = fft_fftwf_plan(D, dimensions, flags, ostrides, dst, istrides, src, backwards);
#ifdef USE_CUDA
plan->cuplan = NULL;
if (cuda_ondevice(src))
plan->cuplan = fft_cuda_plan(D, dimensions, flags, ostrides, istrides, backwards);
#endif
return operator_create2(D, dimensions, ostrides, D, dimensions, istrides, &PTR_PASS(plan)->base, fft_apply, fft_free_plan);
}
const struct operator_s* fft_measure_create(unsigned int D, const long dimensions[D], unsigned long flags, bool inplace, bool backwards)
{
PTR_ALLOC(struct fft_plan_s, plan);
SET_TYPEID(fft_plan_s, plan);
complex float* src = md_alloc(D, dimensions, CFL_SIZE);
complex float* dst = inplace ? src : md_alloc(D, dimensions, CFL_SIZE);
long strides[D];
md_calc_strides(D, strides, dimensions, CFL_SIZE);
plan->fftw = fft_fftwf_plan(D, dimensions, flags, strides, dst, strides, src, backwards, true);
md_free(src);
if (!inplace)
md_free(dst);
#ifdef USE_CUDA
plan->cuplan = NULL;
#ifndef LAZY_CUDA
if (cuda_ondevice(src))
plan->cuplan = fft_cuda_plan(D, dimensions, flags, strides, strides, backwards);
#else
plan->D = D;
plan->flags = flags;
plan->backwards = backwards;
PTR_ALLOC(long[D], dims);
md_copy_dims(D, *dims, dimensions);
plan->dims = *PTR_PASS(dims);
PTR_ALLOC(long[D], istrs);
md_copy_strides(D, *istrs, strides);
plan->istrs = *PTR_PASS(istrs);
PTR_ALLOC(long[D], ostrs);
md_copy_strides(D, *ostrs, strides);
plan->ostrs = *PTR_PASS(ostrs);
#endif
#endif
return operator_create2(D, dimensions, strides, D, dimensions, strides, CAST_UP(PTR_PASS(plan)), fft_apply, fft_free_plan);
}
static void fft_apply(const operator_data_t* _plan, unsigned int N, void* args[N])
{
complex float* dst = args[0];
const complex float* src = args[1];
const struct fft_plan_s* plan = CAST_DOWN(fft_plan_s, _plan);
assert(2 == N);
#ifdef USE_CUDA
if (cuda_ondevice(src)) {
#ifdef LAZY_CUDA
if (NULL == plan->cuplan)
((struct fft_plan_s*)plan)->cuplan = fft_cuda_plan(plan->D, plan->dims, plan->flags, plan->ostrs, plan->istrs, plan->backwards);
#endif
assert(NULL != plan->cuplan);
fft_cuda_exec(plan->cuplan, dst, src);
} else
#endif
{
assert(NULL != plan->fftw);
fftwf_execute_dft(plan->fftw, (complex float*)src, dst);
}
}
