Intel® DPC++ Compatibility Tool Developer Guide and Reference

Developer Guide and Reference

  • 2022.2
  • 09/08/2022
  • Public Content
Contents

DPCT1016

Message

The
<API name>
 was not migrated, because parameter(s)
<parameter name a>
 and/or
<parameter name b>
 could not be evaluated, or because
<parameter name a>
 is not equal to
<parameter name b>
. Rewrite this code manually.

Detailed Help

The
cublasSetMatrix
 can be migrated by the Intel® DPC++ Compatibility Tool only when
lda
 and
ldb
 have the same constant value. In cases where the migration did not occur, rewrite the code manually.
  • If the values of
    lda
     and
    ldb
     are the same, you can use the following code:
    dpct::dpct_memcpy((void*)(B), (void*)(A), lda*cols*elemSize, dpct::host_to_device); // For cublasGetMatrix, use dpct::device_to_host
  • Otherwise, you can copy the columns of the matrix or elements of the vector one by one.
    • The equivalent of
      cublasSetMatrix(rows, cols, elemSize, A, lda, B, ldb)
       is:
       auto A_backup = A;
 auto B_backup = B;
 A = A - lda;
 B = B - ldb;
 for (int c = 0; c < cols; ++c) {
   A = A + lda;
   B = B + ldb;
   dpct::dpct_memcpy(
       (void *)(B), (void *)(A), rows * elemSize,
       dpct::host_to_device); // For cublasGetMatrix, use dpct::device_to_host
 }
 A = A_backup;
 B = B_backup;
The Intel® DPC++ Compatibility Tool can migrate the
cublasSetVector
 only when
incx
 and
incy
 have the same constant value. In cases where the migration does not occur, rewrite the code manually.
  • If the values of
    incx
     and
    incy
     are the same, you can use the following code:
    dpct::dpct_memcpy((void*)(B), (void*)(A), n*incx*elemSize, dpct::host_to_device); // For cublasGetVector, use dpct::device_to_host
  • Otherwise, you can copy the elements of the vector one by one:
    • To replace
      cublasGetVector(n, elemSize, x, incx, y, incy)
      , use the following snippet:
       auto x_backup = x;
 auto y_backup = y;
 x = x - incx;
 y = y - incy;
 for (int c = 0; c < n; ++c) {
   x = x + incx;
   y = y + incy;
   dpct::dpct_memcpy(
       (void *)(y), (void *)(x), elemSize,
       dpct::device_to_host); // for cublasSetVector, use dpct::host_to_device
 }
 x = x_backup;
 y = y_backup;

Suggestions to Fix

Review the logic and adjust it.

Product and Performance Information

1

Performance varies by use, configuration and other factors. Learn more at www.Intel.com/PerformanceIndex.