Go to the source code of this file.
|
template<typename dataType > |
DSPLIB_STATUS | DSPLIB_recip_init_ci (DSPLIB_kernelHandle handle, const DSPLIB_bufParams1D_t *bufParamsIn, const DSPLIB_bufParams1D_t *bufParamsOut, const DSPLIB_recip_InitArgs *pKerInitArgs) |
| This function is the initialization function for the C7x implementation of the kernel. The function declaration conforms to the declaration of DSPLIB_recip_init. More...
|
|
template DSPLIB_STATUS | DSPLIB_recip_init_ci< float > (DSPLIB_kernelHandle handle, const DSPLIB_bufParams1D_t *bufParamsIn, const DSPLIB_bufParams1D_t *bufParamsOut, const DSPLIB_recip_InitArgs *pKerInitArgs) |
|
template DSPLIB_STATUS | DSPLIB_recip_init_ci< double > (DSPLIB_kernelHandle handle, const DSPLIB_bufParams1D_t *bufParamsIn, const DSPLIB_bufParams1D_t *bufParamsOut, const DSPLIB_recip_InitArgs *pKerInitArgs) |
|
template<> |
DSPLIB_STATUS | DSPLIB_recip_init_ci< int16_t > (DSPLIB_kernelHandle handle, const DSPLIB_bufParams1D_t *bufParamsIn, const DSPLIB_bufParams1D_t *bufParamsOut, const DSPLIB_recip_InitArgs *pKerInitArgs) |
|
template<typename dataType , int32_t dataIn> |
DSPLIB_STATUS | DSPLIB_recip_exec_ci (DSPLIB_kernelHandle handle, void *restrict pIn, void *restrict pOut) |
| This function is the main execution function for the C7x implementation of the kernel. The function declaration conforms to the declaration of DSPLIB_recip_exec. More...
|
|
template DSPLIB_STATUS | DSPLIB_recip_exec_ci< float, DSPLIB_FLOAT32 > (DSPLIB_kernelHandle handle, void *restrict pIn, void *restrict pOut) |
|
template DSPLIB_STATUS | DSPLIB_recip_exec_ci< double, DSPLIB_FLOAT64 > (DSPLIB_kernelHandle handle, void *restrict pIn, void *restrict pOut) |
|
template<> |
DSPLIB_STATUS | DSPLIB_recip_exec_ci< int16_t, DSPLIB_INT16 > (DSPLIB_kernelHandle handle, void *restrict pIn, void *restrict pOut) |
|
◆ SE_PARAM_BASE
#define SE_PARAM_BASE (0x0000) |
◆ SE_SE0_PARAM_OFFSET
◆ SE_SA0_PARAM_OFFSET
◆ DSPLIB_recip_init_ci()
template<typename dataType >
This function is the initialization function for the C7x implementation of the kernel. The function declaration conforms to the declaration of DSPLIB_recip_init.
This function determines the configuration for the streaming engine and MMA hardware resources based on the function call parameters, and the configuration is saved in bufPBlock array. In the kernel call sequence, DSPLIB_recip_exec_ci would be called later independently by the application. When DSPLIB_recip_exec_ci runs, it merely retrieves the configuration from the bufPBlock and uses it to set up the hardware resources. This arrangement is so that DSPLIB_recip_exec_ci does not lose cycles to determine the hardware configuration.
- Parameters
-
[in] | handle | : Active handle to the kernel |
[in] | bufParamsIn | : Pointer to the structure containing dimensional information of input buffer |
[out] | bufParamsOut | : Pointer to the structure containing dimensional information of ouput buffer |
[in] | pKerInitArgs | : Pointer to the structure holding init parameters |
- Returns
- Status value indicating success or failure. Refer to DSPLIB_STATUS.
Definition at line 62 of file DSPLIB_recip_ci.cpp.
◆ DSPLIB_recip_init_ci< float >()
◆ DSPLIB_recip_init_ci< double >()
◆ DSPLIB_recip_init_ci< int16_t >()
◆ DSPLIB_recip_exec_ci()
template<typename dataType , int32_t dataIn>
This function is the main execution function for the C7x implementation of the kernel. The function declaration conforms to the declaration of DSPLIB_recip_exec.
The function uses MMA hardware accelerator to perform the convolution computation. Filter data is loaded into B panel of the MMA from memory using one streaming engine, while the input data is loaded into A vectors of the MMA using the other streaming engine. Result of the compute from MMA C panel is stored into memory using a stream sqrress generator.
- Parameters
-
[in] | handle | : Active handle to the kernel |
[in] | pIn | : Pointer to buffer holding the input data |
[out] | pOut | : Pointer to buffer holding the output data |
- Returns
- Status value indicating success or failure. Refer to DSPLIB_STATUS.
- Performance Considerations:
- For best performance,
- the input and output data buffers are expected to be in L2 memory
- the buffer pointers are assumed to be 64-byte aligned
Definition at line 190 of file DSPLIB_recip_ci.cpp.
◆ DSPLIB_recip_exec_ci< float, DSPLIB_FLOAT32 >()
◆ DSPLIB_recip_exec_ci< double, DSPLIB_FLOAT64 >()
◆ DSPLIB_recip_exec_ci< int16_t, DSPLIB_INT16 >()