|
docc 0.8.3 (0d44999b)
C++ API Reference
|
#include <string>#include "sdfg/codegen/instrumentation/instrumentation_info.h"#include "sdfg/codegen/language_extension.h"#include "sdfg/codegen/utils.h"#include "sdfg/targets/gpu/gpu_offload_schedule_type.h"#include "sdfg/targets/gpu/gpu_schedule_type.h"#include "sdfg/targets/gpu/gpu_types.h"Go to the source code of this file.
Classes | |
| class | sdfg::rocm::ScheduleType_ROCM_Offload |
| ROCM schedule type inheriting shared GPU functionality Provides ROCM-specific value() and default block size (64 for wavefront size). More... | |
| class | sdfg::rocm::ScheduleType_ROCM |
| ROCM schedule type inheriting shared GPU functionality Provides ROCM-specific value() and default block size (64 for wavefront size). More... | |
Namespaces | |
| namespace | sdfg |
| namespace | sdfg::rocm |
Typedefs | |
| using | sdfg::rocm::ROCMDimension = gpu::GPUDimension |
Functions | |
| void | sdfg::rocm::rocm_error_checking (codegen::PrettyPrinter &stream, const codegen::LanguageExtension &language_extension, const std::string &status_variable) |
| bool | sdfg::rocm::do_rocm_error_checking () |
| int | sdfg::rocm::rocm_wavefront_size () |
| Wavefront width of the target device, queried once at codegen time. | |
| void | sdfg::rocm::check_rocm_kernel_launch_errors (codegen::PrettyPrinter &stream, const codegen::LanguageExtension &language_extension) |
Variables | |
| std::string | sdfg::rocm::ROCM_DEVICE_PREFIX = "__daisy_hip_" |
| constexpr int | sdfg::rocm::ROCM_WARP_SIZE = 64 |
| data_flow::ImplementationType | sdfg::rocm::ImplementationType_ROCM {"ROCM"} |
| Generic ROCM implementation. | |
| data_flow::ImplementationType | sdfg::rocm::ImplementationType_ROCMWithTransfers {"ROCMWithTransfers"} |
| ROCM implementation with automatic memory transfers Used for ROCm BLAS, memset, and other ROCm-accelerated library nodes. | |
| data_flow::ImplementationType | sdfg::rocm::ImplementationType_ROCMWithoutTransfers {"ROCMWithoutTransfers"} |
| ROCM implementation without memory transfers Used for ROCm BLAS, memset, and other ROCm-accelerated library nodes assuming data is already on GPU. | |
| codegen::TargetType | sdfg::rocm::TargetType_ROCM_Offload {ScheduleType_ROCM_Offload::value()} |
| codegen::TargetType | sdfg::rocm::TargetType_ROCM {ScheduleType_ROCM::value()} |