You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: Libraries/hipBLAS/her/README.md
+3-3
Original file line number
Diff line number
Diff line change
@@ -35,7 +35,7 @@ The application provides the following optional command line arguments:
35
35
- hipBLAS is initialized by calling `hipblasCreate(hipblasHandle_t*)` and it is terminated by calling `hipblasDestroy(hipblasHandle_t)`.
36
36
- The _pointer mode_ controls whether scalar parameters must be allocated on the host (`HIPBLAS_POINTER_MODE_HOST`) or on the device (`HIPBLAS_POINTER_MODE_DEVICE`). It is controlled by `hipblasSetPointerMode`.
37
37
-`hipblasSetVector` and `hipblasSetMatrix` are helper functions provided by the hipBLAS API for writing data to the GPU, whereas `hipblasGetVector` and `hipblasGetMatrix` are intended for retrieving data from the GPU. Note that `hipMemcpy` can also be used to copy/get data to/from the GPU in the usual way.
38
-
-`hipblas[CZ]her2(handle, uplo, n, *alpha, *x, incx, *y, incy, *AP, lda)` computes a Hermitian rank-2 update. The character matched in `[CZ]` denotes the data type of the operation, and can be either `C` (complex float: `hipblasComplex`), or `Z` (complex double: `hipblasDoubleComplex`). The required arguments come as the following:
38
+
-`hipblas[CZ]her2(handle, uplo, n, *alpha, *x, incx, *y, incy, *AP, lda)` computes a Hermitian rank-2 update. The character matched in `[CZ]` denotes the data type of the operation, and can be either `C` (complex float: `hipComplex`), or `Z` (complex double: `hipDoubleComplex`). The required arguments come as the following:
39
39
-`handle`, the hipBLAS API handle.
40
40
-`uplo`. Because a Hermitian matrix is symmetric over the diagonal, except that the values in the upper triangle are the complex conjugate of the values in the lower triangle, the required work can be reduced by only updating a single half of the matrix. The part of the matrix to update is given by `uplo`: `HIPBLAS_FILL_MODE_UPPER` (used in this example) indicates that the upper triangle of $A$ should be updated, `HIPBLAS_FILL_MODE_LOWER` indicates that the lower triangle of $A$ should be updated and `HIPBLAS_FILL_MODE_FULL` indicates that the full matrix will be updated.
41
41
-`n` gives the dimensions of the vector and matrix inputs.
@@ -44,7 +44,7 @@ The application provides the following optional command line arguments:
44
44
-`AP` is the device pointer to matrix $A$ in device memory.
45
45
-`lda` is the _leading dimension_ of $A$, that is, the number of elements between the starts of the columns of $A$. Note that hipBLAS matrices are laid out in _column major_ ordering.
46
46
47
-
- If `ROCM_MATHLIBS_API_USE_HIP_COMPLEX` is defined (adding `#define ROCM_MATHLIBS_API_USE_HIP_COMPLEX` before `#include <hipblas/hipblas.h>`), the hipBLAS API is exposed as using the hip defined complex types. That is, `hipblasComplex` is a typedef of `hipFloatComplex` (also named `hipComplex`) and they can be used equivalently.
47
+
- If `ROCM_MATHLIBS_API_USE_HIP_COMPLEX` is defined (adding `#define ROCM_MATHLIBS_API_USE_HIP_COMPLEX` before `#include <hipblas/hipblas.h>`), the hipBLAS API is exposed as using the hip defined complex types. That is, `hipComplex` is a typedef of `hipFloatComplex` (also named `hipComplex`) and they can be used equivalently.
48
48
-`hipFloatComplex` and `std::complex<float>` have compatible memory layout, and performing a memory copy between values of these types will correctly perform the expected copy.
49
49
-`hipCaddf(a, b)` adds `hipFloatComplex` values `a` and `b` element-wise together. This function is from a family of host/device HIP functions which operate on complex values.
50
50
@@ -55,7 +55,7 @@ The application provides the following optional command line arguments:
Copy file name to clipboardExpand all lines: Libraries/hipBLAS/scal/README.md
+2-2
Original file line number
Diff line number
Diff line change
@@ -32,8 +32,8 @@ The application provides the following optional command line arguments:
32
32
-`hipblas[SDCZ]scal` multiplies each element of the vector by a scalar. Depending on the character matched in `[SDCZ]`, the scaling can be obtained with different precisions:
- The `hipblasPointerMode_t` type controls whether scalar parameters must be allocated on the host (`HIPBLAS_POINTER_MODE_HOST`) or on the device (`HIPBLAS_POINTER_MODE_DEVICE`). It is set by using `hipblasSetPointerMode`.
0 commit comments