rocfft: include batch dimension in MPI sample bricks (#1785)

## Motivation

Fix the rocFFT MPI example.  Currently it fails to create the FFT plan.

## Technical Details

An N-dimensional distributed FFT is specified to rocFFT as a set of
bricks. However, the bricks require N+1 indexes for strides and brick
coordinates. The bricks were incorrectly specified with N indexes.

## Test Plan

Ran the example manually with the fix. MPI unit tests cover additional
cases in the correct (N+1) way.

## Test Result

Manual run was successful.

## Submission Checklist

- [x] Look over the contributing guidelines at
https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.

Author: evetsso

clients/samples/mpi/rocfft_mpi_example.cpp

@@ -126,6 +126,11 @@ int main(int argc, char** argv)
     if(fftrc != rocfft_status_success)
         throw std::runtime_error("failed to create description");
 
+    // This example is unbatched, so the batch stride is not used
+    // for anything.  For batched examples, this would be
+    // distance in elements between consecutive batches.
+    const size_t batch_stride = 0;
+
     if(mpi_rank == 0)
     {
         std::cout << "input data decomposition:\n";
@@ -135,14 +140,14 @@ int main(int argc, char** argv)
         rocfft_field infield = nullptr;
         rocfft_field_create(&infield);
 
-        std::vector<size_t> inbrick_stride  = {1, length[1]};
+        std::vector<size_t> inbrick_stride  = {1, length[1], batch_stride};
         const size_t        inbrick_length1 = length[1] / (size_t)mpi_size
                                        + ((size_t)mpi_rank < length[1] % (size_t)mpi_size ? 1 : 0);
         const size_t inbrick_lower1
             = mpi_rank * (length[1] / mpi_size) + std::min((size_t)mpi_rank, length[1] % mpi_size);
         const size_t        inbrick_upper1 = inbrick_lower1 + inbrick_length1;
-        std::vector<size_t> inbrick_lower  = {0, inbrick_lower1};
-        std::vector<size_t> inbrick_upper  = {length[0], inbrick_upper1};
+        std::vector<size_t> inbrick_lower  = {0, inbrick_lower1, 0};
+        std::vector<size_t> inbrick_upper  = {length[0], inbrick_upper1, 1};
 
         rocfft_brick inbrick = nullptr;
         rocfft_brick_create(&inbrick,
@@ -219,15 +224,15 @@ int main(int argc, char** argv)
     std::vector<void*>  gpu_out = {nullptr};
     std::vector<size_t> outbrick_lower;
     std::vector<size_t> outbrick_upper;
-    std::vector<size_t> outbrick_stride = {1, length[1]};
+    std::vector<size_t> outbrick_stride = {1, length[1], batch_stride};
     {
         const size_t outbrick_length1 = length[1] / (size_t)mpi_size
                                         + ((size_t)mpi_rank < length[1] % (size_t)mpi_size ? 1 : 0);
         const size_t outbrick_lower1
             = mpi_rank * (length[1] / mpi_size) + std::min((size_t)mpi_rank, length[1] % mpi_size);
         const size_t outbrick_upper1 = outbrick_lower1 + outbrick_length1;
-        outbrick_lower               = {0, outbrick_lower1};
-        outbrick_upper               = {length[0], outbrick_upper1};
+        outbrick_lower               = {0, outbrick_lower1, 0};
+        outbrick_upper               = {length[0], outbrick_upper1, 1};
 
         const size_t memSize = length[0] * outbrick_length1 * sizeof(std::complex<double>);
         for(int irank = 0; irank < mpi_size; ++irank)
@@ -254,8 +259,8 @@ int main(int argc, char** argv)
         rocfft_field_create(&outfield);
 
         rocfft_brick outbrick = nullptr;
-        outbrick_lower        = {0, outbrick_lower1};
-        outbrick_upper        = {length[0], outbrick_lower1 + outbrick_length1};
+        outbrick_lower        = {0, outbrick_lower1, 0};
+        outbrick_upper        = {length[0], outbrick_lower1 + outbrick_length1, 1};
         rocfft_brick_create(&outbrick,
                             outbrick_lower.data(),
                             outbrick_upper.data(),