derek.holloway
7502bf64a4
CodeQL / Analyze (actions) (push) Canceled after 0s
CodeQL / Analyze (python) (push) Canceled after 0s
pre-commit / pre-commit (push) Canceled after 0s
Multi-Arch CI / setup (push) Canceled after 0s
Publish build_manylinux_x86_64 images / publish_build_manylinux_x86_64 (push) Canceled after 0s
Setup (Multi-Arch) / setup (push) Canceled after 0s
Multi-Arch CI / Linux::${{ fromJSON(needs.setup.outputs.linux_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish no_rocm_image_ubi10 image / publish_no_rocm_image_ubi10 (push) Canceled after 0s
Multi-Arch CI / Windows::${{ fromJSON(needs.setup.outputs.windows_build_config || '{}').build_variant_label || 'skip' }} (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Multi-Arch CI / Manifest Diff (push) Canceled after 0s
TheRock Manifest Diff Report / Generate Manifest Diff Report (push) Canceled after 0s
Multi-Arch CI / CI Summary (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04 (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04_ocl_rt images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04_rocgdb images / publish_no_rocm_image_ubuntu24_04_rocgdb (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_ocl_rt (push) Canceled after 0s
Publish TheRock Docker image / build-and-push-image (push) Canceled after 0s
Unit Tests / Unit Tests :: ubuntu-24.04 (push) Canceled after 0s
Unit Tests / Unit Tests :: windows-2022 (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_media (push) Canceled after 0s
Unit Tests / Unit Tests Summary (push) Canceled after 0s
Publish no_rocm_image_ubuntu24_04 images / publish_no_rocm_image_ubuntu24_04_openmpi (push) Canceled after 0s
467 lines
22 KiB
C
467 lines
22 KiB
C
/* ************************************************************************
|
|
* Copyright (C) 2016-2025 Advanced Micro Devices, Inc. All rights reserved.
|
|
*
|
|
* Permission is hereby granted, free of charge, to any person obtaining a copy
|
|
* of this software and associated documentation files (the "Software"), to deal
|
|
* in the Software without restriction, including without limitation the rights
|
|
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell cop-
|
|
* ies of the Software, and to permit persons to whom the Software is furnished
|
|
* to do so, subject to the following conditions:
|
|
*
|
|
* The above copyright notice and this permission notice shall be included in all
|
|
* copies or substantial portions of the Software.
|
|
*
|
|
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IM-
|
|
* PLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS
|
|
* FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR
|
|
* COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER
|
|
* IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNE-
|
|
* CTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
|
|
*
|
|
* ************************************************************************ */
|
|
|
|
#ifndef ROCBLAS_AUXILIARY_H
|
|
#define ROCBLAS_AUXILIARY_H
|
|
|
|
#include "rocblas-export.h"
|
|
#include "rocblas-macros.h"
|
|
#include "rocblas-types.h"
|
|
|
|
/*!\file
|
|
* \brief rocblas-auxiliary.h provides auxilary functions in rocblas
|
|
*/
|
|
|
|
#ifdef __cplusplus
|
|
extern "C" {
|
|
#endif
|
|
|
|
/*! \brief Create handle.
|
|
*/
|
|
ROCBLAS_EXPORT rocblas_status rocblas_create_handle(rocblas_handle* handle);
|
|
|
|
/*! \brief Destroy handle.
|
|
*/
|
|
ROCBLAS_EXPORT rocblas_status rocblas_destroy_handle(rocblas_handle handle);
|
|
|
|
/*! \brief Set stream for handle.
|
|
*/
|
|
ROCBLAS_EXPORT rocblas_status rocblas_set_stream(rocblas_handle handle, hipStream_t stream);
|
|
|
|
/*! \brief Get stream [0] from handle.
|
|
*/
|
|
ROCBLAS_EXPORT rocblas_status rocblas_get_stream(rocblas_handle handle, hipStream_t* stream);
|
|
|
|
/*! \brief Set ``rocblas_pointer_mode``.
|
|
*/
|
|
ROCBLAS_EXPORT rocblas_status rocblas_set_pointer_mode(rocblas_handle handle,
|
|
rocblas_pointer_mode pointer_mode);
|
|
/*! \brief Get ``rocblas_pointer_mode``.
|
|
*/
|
|
ROCBLAS_EXPORT rocblas_status rocblas_get_pointer_mode(rocblas_handle handle,
|
|
rocblas_pointer_mode* pointer_mode);
|
|
|
|
/*! \brief Set ``rocblas_atomics_mode``
|
|
* \details
|
|
* Some rocBLAS functions have implementations which use atomic operations to increase performance.
|
|
* By using atomic operations, results are not guaranteed to be identical between multiple runs.
|
|
* Results are accurate with or without atomic operations. Atomic operations in rocBLAS are turned
|
|
* off by default. They can be turned on or off on a per-handle basis by calling ``rocblas_set_atomics_mode``.
|
|
*/
|
|
ROCBLAS_EXPORT rocblas_status rocblas_set_atomics_mode(rocblas_handle handle,
|
|
rocblas_atomics_mode atomics_mode);
|
|
|
|
/*! \brief Get ``rocblas_atomics_mode``.
|
|
*/
|
|
ROCBLAS_EXPORT rocblas_status rocblas_get_atomics_mode(rocblas_handle handle,
|
|
rocblas_atomics_mode* atomics_mode);
|
|
|
|
/*! \brief Set alpha stride for limited set of batched and strided_batched functions to specify the stride for alpha between successive batch elements.
|
|
Only applies to rocblas_pointer_mode_device and thus device side allocations.
|
|
It enables interpretation of the alpha pointer for both batched and strided_batched functions as a pointer to a vector of values.
|
|
Default value is 0 which treats it as a pointer to a single scalar. Support is denoted with specific function documentation.
|
|
Warning this is a modal like state in the handle. Restore to value 0 if no longer applicable to later function calls.
|
|
*/
|
|
ROCBLAS_EXPORT rocblas_status rocblas_set_batch_alpha_stride(rocblas_handle handle,
|
|
rocblas_stride alpha_stride);
|
|
|
|
/*! \brief Get batch alpha stride from the handle.
|
|
*/
|
|
ROCBLAS_EXPORT rocblas_status rocblas_get_batch_alpha_stride(rocblas_handle handle,
|
|
rocblas_stride* alpha_stride);
|
|
|
|
/*! \brief Set beta stride for limited set of batched and strided_batched functions to specify the stride for beta between successive batch elements.
|
|
Only applies to rocblas_pointer_mode_device and thus device side allocations.
|
|
It enables interpretation of the beta pointer for both batched and strided_batched functions as a pointer to a vector of values.
|
|
Default value is 0 which treats it as a pointer to a single scalar. Support is denoted with specific function documentation.
|
|
Warning this is a modal like state in the handle. Restore to value 0 if no longer applicable to later function calls.
|
|
*/
|
|
ROCBLAS_EXPORT rocblas_status rocblas_set_batch_beta_stride(rocblas_handle handle,
|
|
rocblas_stride beta_stride);
|
|
|
|
/*! \brief Get batch beta stride from the handle.
|
|
*/
|
|
ROCBLAS_EXPORT rocblas_status rocblas_get_batch_beta_stride(rocblas_handle handle,
|
|
rocblas_stride* beta_stride);
|
|
|
|
/*! \brief Set ``rocblas_math_mode``.
|
|
*/
|
|
ROCBLAS_EXPORT rocblas_status rocblas_set_math_mode(rocblas_handle handle,
|
|
rocblas_math_mode math_mode);
|
|
|
|
/*! \brief Get ``rocblas_math_mode``.
|
|
*/
|
|
ROCBLAS_EXPORT rocblas_status rocblas_get_math_mode(rocblas_handle handle,
|
|
rocblas_math_mode* math_mode);
|
|
|
|
/*! \brief Indicates whether the pointer is on the host or device.
|
|
*/
|
|
ROCBLAS_EXPORT rocblas_pointer_mode rocblas_pointer_to_mode(void* ptr);
|
|
|
|
/*! \brief Copy vector from host to device.
|
|
@param[in]
|
|
n [rocblas_int]
|
|
number of elements in the vector
|
|
@param[in]
|
|
elem_size [rocblas_int]
|
|
number of bytes per element in the matrix
|
|
@param[in]
|
|
x pointer to vector on the host
|
|
@param[in]
|
|
incx [rocblas_int]
|
|
specifies the increment for the elements of the vector
|
|
@param[out]
|
|
y pointer to vector on the device
|
|
@param[in]
|
|
incy [rocblas_int]
|
|
specifies the increment for the elements of the vector
|
|
*/
|
|
ROCBLAS_EXPORT rocblas_status rocblas_set_vector(rocblas_int n,
|
|
rocblas_int elem_size,
|
|
const void* x,
|
|
rocblas_int incx,
|
|
void* y,
|
|
rocblas_int incy);
|
|
|
|
ROCBLAS_EXPORT rocblas_status rocblas_set_vector_64(
|
|
int64_t n, int64_t elem_size, const void* x, int64_t incx, void* y, int64_t incy);
|
|
|
|
/*! \brief Copy vector from device to host.
|
|
@param[in]
|
|
n [rocblas_int]
|
|
number of elements in the vector
|
|
@param[in]
|
|
elem_size [rocblas_int]
|
|
number of bytes per element in the matrix
|
|
@param[in]
|
|
x pointer to vector on the device
|
|
@param[in]
|
|
incx [rocblas_int]
|
|
specifies the increment for the elements of the vector
|
|
@param[out]
|
|
y pointer to vector on the host
|
|
@param[in]
|
|
incy [rocblas_int]
|
|
specifies the increment for the elements of the vector
|
|
*/
|
|
ROCBLAS_EXPORT rocblas_status rocblas_get_vector(rocblas_int n,
|
|
rocblas_int elem_size,
|
|
const void* x,
|
|
rocblas_int incx,
|
|
void* y,
|
|
rocblas_int incy);
|
|
|
|
ROCBLAS_EXPORT rocblas_status rocblas_get_vector_64(
|
|
int64_t n, int64_t elem_size, const void* x, int64_t incx, void* y, int64_t incy);
|
|
|
|
/*! \brief Copy matrix from host to device.
|
|
@param[in]
|
|
rows [rocblas_int]
|
|
number of rows in matrices
|
|
@param[in]
|
|
cols [rocblas_int]
|
|
number of columns in matrices
|
|
@param[in]
|
|
elem_size [rocblas_int]
|
|
number of bytes per element in the matrix
|
|
@param[in]
|
|
a pointer to matrix on the host
|
|
@param[in]
|
|
lda [rocblas_int]
|
|
specifies the leading dimension of A, lda >= rows
|
|
@param[out]
|
|
b pointer to matrix on the GPU
|
|
@param[in]
|
|
ldb [rocblas_int]
|
|
specifies the leading dimension of B, ldb >= rows
|
|
*/
|
|
ROCBLAS_EXPORT rocblas_status rocblas_set_matrix(rocblas_int rows,
|
|
rocblas_int cols,
|
|
rocblas_int elem_size,
|
|
const void* a,
|
|
rocblas_int lda,
|
|
void* b,
|
|
rocblas_int ldb);
|
|
|
|
ROCBLAS_EXPORT rocblas_status rocblas_set_matrix_64(int64_t rows,
|
|
int64_t cols,
|
|
int64_t elem_size,
|
|
const void* a,
|
|
int64_t lda,
|
|
void* b,
|
|
int64_t ldb);
|
|
|
|
/*! \brief Copy matrix from device to host.
|
|
@param[in]
|
|
rows [rocblas_int]
|
|
number of rows in matrices
|
|
@param[in]
|
|
cols [rocblas_int]
|
|
number of columns in matrices
|
|
@param[in]
|
|
elem_size [rocblas_int]
|
|
number of bytes per element in the matrix
|
|
@param[in]
|
|
a pointer to matrix on the GPU
|
|
@param[in]
|
|
lda [rocblas_int]
|
|
specifies the leading dimension of A, lda >= rows
|
|
@param[out]
|
|
b pointer to matrix on the host
|
|
@param[in]
|
|
ldb [rocblas_int]
|
|
specifies the leading dimension of B, ldb >= rows
|
|
*/
|
|
ROCBLAS_EXPORT rocblas_status rocblas_get_matrix(rocblas_int rows,
|
|
rocblas_int cols,
|
|
rocblas_int elem_size,
|
|
const void* a,
|
|
rocblas_int lda,
|
|
void* b,
|
|
rocblas_int ldb);
|
|
|
|
ROCBLAS_EXPORT rocblas_status rocblas_get_matrix_64(int64_t rows,
|
|
int64_t cols,
|
|
int64_t elem_size,
|
|
const void* a,
|
|
int64_t lda,
|
|
void* b,
|
|
int64_t ldb);
|
|
|
|
/*! \brief Asynchronously copy vector from host to device.
|
|
\details
|
|
``rocblas_set_vector_async`` copies a vector from pinned host memory to device memory asynchronously.
|
|
Memory on the host must be allocated with ``hipHostMalloc`` or the transfer will be synchronous.
|
|
@param[in]
|
|
n [rocblas_int]
|
|
number of elements in the vector
|
|
@param[in]
|
|
elem_size [rocblas_int]
|
|
number of bytes per element in the matrix
|
|
@param[in]
|
|
x pointer to vector on the host
|
|
@param[in]
|
|
incx [rocblas_int]
|
|
specifies the increment for the elements of the vector
|
|
@param[out]
|
|
y pointer to vector on the device
|
|
@param[in]
|
|
incy [rocblas_int]
|
|
specifies the increment for the elements of the vector
|
|
@param[in]
|
|
stream specifies the stream into which this transfer request is queued
|
|
********************************************************************/
|
|
ROCBLAS_EXPORT rocblas_status rocblas_set_vector_async(rocblas_int n,
|
|
rocblas_int elem_size,
|
|
const void* x,
|
|
rocblas_int incx,
|
|
void* y,
|
|
rocblas_int incy,
|
|
hipStream_t stream);
|
|
|
|
ROCBLAS_EXPORT rocblas_status rocblas_set_vector_async_64(int64_t n,
|
|
int64_t elem_size,
|
|
const void* x,
|
|
int64_t incx,
|
|
void* y,
|
|
int64_t incy,
|
|
hipStream_t stream);
|
|
|
|
/*! \brief Asynchronously copy vector from device to host.
|
|
\details
|
|
``rocblas_get_vector_async`` copies a vector from device memory to pinned host memory asynchronously.
|
|
Memory on the host must be allocated with ``hipHostMalloc`` or the transfer will be synchronous.
|
|
@param[in]
|
|
n [rocblas_int]
|
|
number of elements in the vector
|
|
@param[in]
|
|
elem_size [rocblas_int]
|
|
number of bytes per element in the matrix
|
|
@param[in]
|
|
x pointer to vector on the device
|
|
@param[in]
|
|
incx [rocblas_int]
|
|
specifies the increment for the elements of the vector
|
|
@param[out]
|
|
y pointer to vector on the host
|
|
@param[in]
|
|
incy [rocblas_int]
|
|
specifies the increment for the elements of the vector
|
|
@param[in]
|
|
stream specifies the stream into which this transfer request is queued
|
|
********************************************************************/
|
|
ROCBLAS_EXPORT rocblas_status rocblas_get_vector_async(rocblas_int n,
|
|
rocblas_int elem_size,
|
|
const void* x,
|
|
rocblas_int incx,
|
|
void* y,
|
|
rocblas_int incy,
|
|
hipStream_t stream);
|
|
|
|
ROCBLAS_EXPORT rocblas_status rocblas_get_vector_async_64(int64_t n,
|
|
int64_t elem_size,
|
|
const void* x,
|
|
int64_t incx,
|
|
void* y,
|
|
int64_t incy,
|
|
hipStream_t stream);
|
|
|
|
/*! \brief Asynchronously copy matrix from host to device.
|
|
\details
|
|
``rocblas_set_matrix_async`` copies a matrix from pinned host memory to device memory asynchronously.
|
|
Memory on the host must be allocated with ``hipHostMalloc`` or the transfer will be synchronous.
|
|
@param[in]
|
|
rows [rocblas_int]
|
|
number of rows in matrices
|
|
@param[in]
|
|
cols [rocblas_int]
|
|
number of columns in matrices
|
|
@param[in]
|
|
elem_size [rocblas_int]
|
|
number of bytes per element in the matrix
|
|
@param[in]
|
|
a pointer to matrix on the host
|
|
@param[in]
|
|
lda [rocblas_int]
|
|
specifies the leading dimension of A, lda >= rows
|
|
@param[out]
|
|
b pointer to matrix on the GPU
|
|
@param[in]
|
|
ldb [rocblas_int]
|
|
specifies the leading dimension of B, ldb >= rows
|
|
@param[in]
|
|
stream specifies the stream into which this transfer request is queued
|
|
********************************************************************/
|
|
ROCBLAS_EXPORT rocblas_status rocblas_set_matrix_async(rocblas_int rows,
|
|
rocblas_int cols,
|
|
rocblas_int elem_size,
|
|
const void* a,
|
|
rocblas_int lda,
|
|
void* b,
|
|
rocblas_int ldb,
|
|
hipStream_t stream);
|
|
|
|
ROCBLAS_EXPORT rocblas_status rocblas_set_matrix_async_64(int64_t rows,
|
|
int64_t cols,
|
|
int64_t elem_size,
|
|
const void* a,
|
|
int64_t lda,
|
|
void* b,
|
|
int64_t ldb,
|
|
hipStream_t stream);
|
|
|
|
/*! \brief Asynchronously copy matrix from device to host.
|
|
\details
|
|
``rocblas_get_matrix_async`` copies a matrix from device memory to pinned host memory asynchronously.
|
|
Memory on the host must be allocated with ``hipHostMalloc`` or the transfer will be synchronous.
|
|
@param[in]
|
|
rows [rocblas_int]
|
|
number of rows in matrices
|
|
@param[in]
|
|
cols [rocblas_int]
|
|
number of columns in matrices
|
|
@param[in]
|
|
elem_size [rocblas_int]
|
|
number of bytes per element in the matrix
|
|
@param[in]
|
|
a pointer to matrix on the GPU
|
|
@param[in]
|
|
lda [rocblas_int]
|
|
specifies the leading dimension of A, lda >= rows
|
|
@param[out]
|
|
b pointer to matrix on the host
|
|
@param[in]
|
|
ldb [rocblas_int]
|
|
specifies the leading dimension of B, ldb >= rows
|
|
@param[in]
|
|
stream specifies the stream into which this transfer request is queued
|
|
********************************************************************/
|
|
ROCBLAS_EXPORT rocblas_status rocblas_get_matrix_async(rocblas_int rows,
|
|
rocblas_int cols,
|
|
rocblas_int elem_size,
|
|
const void* a,
|
|
rocblas_int lda,
|
|
void* b,
|
|
rocblas_int ldb,
|
|
hipStream_t stream);
|
|
|
|
ROCBLAS_EXPORT rocblas_status rocblas_get_matrix_async_64(int64_t rows,
|
|
int64_t cols,
|
|
int64_t elem_size,
|
|
const void* a,
|
|
int64_t lda,
|
|
void* b,
|
|
int64_t ldb,
|
|
hipStream_t stream);
|
|
|
|
/*******************************************************************************
|
|
* Function to set start/stop event handlers (for internal use only)
|
|
******************************************************************************/
|
|
ROCBLAS_EXPORT rocblas_status rocblas_set_start_stop_events(rocblas_handle handle,
|
|
hipEvent_t startEvent,
|
|
hipEvent_t stopEvent);
|
|
#define ROCBLAS_INVOKE_START_STOP_EVENTS(handle, startEvent, stopEvent, call) \
|
|
do \
|
|
{ \
|
|
rocblas_handle tmp_h = (handle); \
|
|
rocblas_set_start_stop_events(tmp_h, (startEvent), (stopEvent)); \
|
|
call; \
|
|
rocblas_set_start_stop_events(tmp_h, (hipEvent_t)0, (hipEvent_t)0); \
|
|
} while(0)
|
|
|
|
// For testing solution selection fitness -- for internal testing only
|
|
ROCBLAS_EXPORT rocblas_status rocblas_set_solution_fitness_query(rocblas_handle handle,
|
|
double* fitness);
|
|
|
|
/*! \brief specifies the performance metric that solution selection uses
|
|
\details
|
|
Determines which performance metric will be used by Tensile when selecting the optimal solution
|
|
for gemm problems. If a valid solution benchmarked for this performance metric does not exist
|
|
for a problem, Tensile will default to a solution benchmarked for overall performance instead.
|
|
@param[in]
|
|
handle [rocblas_handle]
|
|
the handle of device
|
|
@param[in]
|
|
metric [rocblas_performance_metric]
|
|
the performance metric to be used
|
|
********************************************************************/
|
|
ROCBLAS_EXPORT rocblas_status rocblas_set_performance_metric(rocblas_handle handle,
|
|
rocblas_performance_metric metric);
|
|
/*! \brief returns the performance metric being used for solution selection
|
|
\details
|
|
Returns the performance metric used by Tensile to select the optimal solution for gemm problems.
|
|
@param[in]
|
|
handle [rocblas_handle]
|
|
the handle of device
|
|
@param[out]
|
|
metric [rocblas_performance_metric*]
|
|
pointer to where the metric will be stored
|
|
********************************************************************/
|
|
ROCBLAS_EXPORT rocblas_status rocblas_get_performance_metric(rocblas_handle handle,
|
|
rocblas_performance_metric* metric);
|
|
|
|
#ifdef __cplusplus
|
|
}
|
|
#endif
|
|
|
|
#endif /* ROCBLAS_AUXILIARY_H */
|