new cufftPlanMany() in CUFFT When shall we use it?

Accelerated Computing CUDA CUDA Programming and Performance
#1

Hi folks,

I’m interested in this new function that seems to be documented only in the release notes.
What’s the aim of this new kind of batched transformed? Are they launched concurrently?
What will be the difference between a regular call to transform 100 lines and the new batched call on 100 batches?

Thanks a lot for your help!

#2

Where do you see that in the release notes?

#3

Look at the release notes from the win64 Cuda Toolkit installer

#4

I assume cufftPlanMany is stream-enabled and is more efficient than looping through a large set of fixed element samples.

I am trying to implement this function, but I run into the following compiling error:

[codebox]

//lines 352 and 353 of test.cu

cufftHandle batchPlan;

cufftPlanMany(&batchPlan, 2, { 128, 256 }, NULL, 1, 0, NULL, 1, 0, CUFFT_C2C, 1000);

[/codebox]

test.cu(353): error: expected an expression

Any ideas? Thanks in advance.

#5

you can read README.txt in CUDA3.0 doc:
" o Batched 2D & 3D transforms are now supported in CUFFT, using the new
cufftPlanMany() API. This is defined in cufft.h, as follows:

cufftResult CUFFTAPI cufftPlanMany(cufftHandle *plan,
int rank,
int *n,
int *inembed, // Unused: pass NULL
int istride, // Unused: pass 1
int idist, // Unused: pass 0
int *onembed, // Unused: pass NULL
int ostride, // Unused: pass 1
int odist, // Unused: pass 0
cufftType type,
int batch);

The arguments are:
*plan - The plan is returned here, as for other cufft calls
rank - The dimensionality of the transform (1, 2 or 3)
*n - An array of size [rank], describing the size of each
dimension
type - Transform type (e.g. CUFFT_C2C), as per other cufft calls
batch - Batch size for this transform

Return values are as for all other cufftPlanXxx functions. Thus to plan
a batch of 1000, 2D, double-precision, complex-to-complex transforms of
size (128, 256), you would do:

   cufftHandle *myplan;
   cufftPlanMany(myplan, 2, { 128, 256 }, NULL, 1, 0, NULL, 1, 0, CUFFT_Z2Z, 1000);

Note that for CUFFT 3.0, the layout of batched data must be side-by-side
and not interleaved. The inembed, istride, idist, onembed, ostride and
odist parameters are for enabling data windowing and interleaving in a
future version.
"

#6

There is a bug in example:
cufftHandle *myplan;
cufftPlanMany(myplan, 2, { 128, 256 }, NULL, 1, 0, NULL, 1, 0, CUFFT_Z2Z, 1000);
should be changed as follows:
int num[2];
num[0]=128; num[1]=256];
…
cufftPlanMany(myplan, 2, num, NULL, 1, 0, NULL, 1, 0, CUFFT_Z2Z, 1000);