forked from ROCm/MIVisionX
-
Notifications
You must be signed in to change notification settings - Fork 0
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Adding openvx changes for downmix node
- Loading branch information
1 parent
1e89c02
commit d53f81d
Showing
7 changed files
with
261 additions
and
7 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
222 changes: 222 additions & 0 deletions
222
amd_openvx_extensions/amd_rpp/source/tensor/Downmix.cpp
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,222 @@ | ||
| /* | ||
| Copyright (c) 2024 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 | ||
| copies 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 | ||
| IMPLIED, 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 CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN | ||
| THE SOFTWARE. | ||
| */ | ||
|
|
||
| #include "internal_publishKernels.h" | ||
| #include "vx_ext_amd.h" | ||
|
|
||
| struct DownmixLocalData | ||
| { | ||
| vxRppHandle *handle; | ||
| Rpp32u device_type; | ||
| RppPtr_t pSrc; | ||
| RppPtr_t pDst; | ||
| vx_int32 *pSamples; | ||
| vx_int32 *pChannels; | ||
| RpptDescPtr pSrcDesc; | ||
| RpptDescPtr pDstDesc; | ||
| RpptDesc srcDesc; | ||
| RpptDesc dstDesc; | ||
| size_t inputTensorDims[RPP_MAX_TENSOR_DIMS]; | ||
| size_t outputTensorDims[RPP_MAX_TENSOR_DIMS]; | ||
| }; | ||
|
|
||
| static vx_status VX_CALLBACK refreshDownmix(vx_node node, const vx_reference *parameters, vx_uint32 num, DownmixLocalData *data) { | ||
| vx_status status = VX_SUCCESS; | ||
| void *roi_tensor_ptr_src; | ||
| if (data->device_type == AGO_TARGET_AFFINITY_GPU) | ||
| { | ||
| #if ENABLE_HIP | ||
| return VX_ERROR_NOT_IMPLEMENTED; | ||
| } | ||
| #endif | ||
| if (data->device_type == AGO_TARGET_AFFINITY_CPU) | ||
| { | ||
| STATUS_ERROR_CHECK(vxQueryTensor((vx_tensor)parameters[0], VX_TENSOR_BUFFER_HOST, &data->pSrc, sizeof(data->pSrc))); | ||
| STATUS_ERROR_CHECK(vxQueryTensor((vx_tensor)parameters[1], VX_TENSOR_BUFFER_HOST, &data->pDst, sizeof(data->pDst))); | ||
| STATUS_ERROR_CHECK(vxQueryTensor((vx_tensor)parameters[2], VX_TENSOR_BUFFER_HOST, &roi_tensor_ptr_src, sizeof(roi_tensor_ptr_src))); | ||
| } | ||
| RpptROI *src_roi = reinterpret_cast<RpptROI *>(roi_tensor_ptr_src); | ||
| for(int n = 0; n < data->inputTensorDims[0] ; n++) { | ||
| data->pSamples[n] = src_roi[n].xywhROI.xy.x; | ||
| data->pChannels[n] = src_roi[n].xywhROI.xy.y; | ||
| } | ||
| return status; | ||
| } | ||
|
|
||
| static vx_status VX_CALLBACK validateDownmix(vx_node node, const vx_reference parameters[], vx_uint32 num, vx_meta_format metas[]) { | ||
| vx_status status = VX_SUCCESS; | ||
| vx_enum scalar_type; | ||
| STATUS_ERROR_CHECK(vxQueryScalar((vx_scalar)parameters[3], VX_SCALAR_TYPE, &scalar_type, sizeof(scalar_type))); | ||
| if (scalar_type != VX_TYPE_UINT32) | ||
| return ERRMSG(VX_ERROR_INVALID_TYPE, "validate: Paramter: #4 type=%d (must be size)\n", scalar_type); | ||
|
|
||
| // Check for input parameters | ||
| size_t num_tensor_dims; | ||
| STATUS_ERROR_CHECK(vxQueryTensor((vx_tensor)parameters[0], VX_TENSOR_NUMBER_OF_DIMS, &num_tensor_dims, sizeof(num_tensor_dims))); | ||
| if(num_tensor_dims < 3) return ERRMSG(VX_ERROR_INVALID_DIMENSION, "validate: Downmix: tensor: #0 dimensions=%lu (must be greater than or equal to 3)\n", num_tensor_dims); | ||
|
|
||
| // Check for output parameters | ||
| STATUS_ERROR_CHECK(vxQueryTensor((vx_tensor)parameters[1], VX_TENSOR_NUMBER_OF_DIMS, &num_tensor_dims, sizeof(num_tensor_dims))); | ||
| if(num_tensor_dims < 3) return ERRMSG(VX_ERROR_INVALID_DIMENSION, "validate: Downmix: tensor: #1 dimensions=%lu (must be greater than or equal to 3)\n", num_tensor_dims); | ||
|
|
||
| vx_uint8 tensor_fixed_point_position; | ||
| size_t tensor_dims[RPP_MAX_TENSOR_DIMS]; | ||
| vx_enum tensor_datatype; | ||
|
|
||
| STATUS_ERROR_CHECK(vxQueryTensor((vx_tensor)parameters[1], VX_TENSOR_DIMS, &tensor_dims, sizeof(tensor_dims))); | ||
| STATUS_ERROR_CHECK(vxQueryTensor((vx_tensor)parameters[1], VX_TENSOR_DATA_TYPE, &tensor_datatype, sizeof(tensor_datatype))); | ||
| STATUS_ERROR_CHECK(vxQueryTensor((vx_tensor)parameters[1], VX_TENSOR_FIXED_POINT_POSITION, &tensor_fixed_point_position, sizeof(tensor_fixed_point_position))); | ||
| STATUS_ERROR_CHECK(vxSetMetaFormatAttribute(metas[1], VX_TENSOR_NUMBER_OF_DIMS, &num_tensor_dims, sizeof(num_tensor_dims))); | ||
| STATUS_ERROR_CHECK(vxSetMetaFormatAttribute(metas[1], VX_TENSOR_DIMS, &tensor_dims, sizeof(tensor_dims))); | ||
| STATUS_ERROR_CHECK(vxSetMetaFormatAttribute(metas[1], VX_TENSOR_DATA_TYPE, &tensor_datatype, sizeof(tensor_datatype))); | ||
| STATUS_ERROR_CHECK(vxSetMetaFormatAttribute(metas[1], VX_TENSOR_FIXED_POINT_POSITION, &tensor_fixed_point_position, sizeof(tensor_fixed_point_position))); | ||
| return status; | ||
| } | ||
|
|
||
| static vx_status VX_CALLBACK processDownmix(vx_node node, const vx_reference *parameters, vx_uint32 num) { | ||
| RppStatus rpp_status = RPP_SUCCESS; | ||
| vx_status return_status = VX_SUCCESS; | ||
| DownmixLocalData *data = NULL; | ||
| STATUS_ERROR_CHECK(vxQueryNode(node, VX_NODE_LOCAL_DATA_PTR, &data, sizeof(data))); | ||
| if (data->device_type == AGO_TARGET_AFFINITY_GPU) | ||
| { | ||
| #if ENABLE_HIP | ||
| return_status = VX_ERROR_NOT_IMPLEMENTED; | ||
| } | ||
| #endif | ||
| if (data->device_type == AGO_TARGET_AFFINITY_CPU) | ||
| { | ||
| refreshDownmix(node, parameters, num, data); | ||
| rpp_status = rppt_down_mixing_host((float *)data->pSrc, data->pSrcDesc, (float *)data->pDst, data->pDstDesc, data->pSamples, data->pChannels, false, data->handle->rppHandle); | ||
| return_status = (rpp_status == RPP_SUCCESS) ? VX_SUCCESS : VX_FAILURE; | ||
| } | ||
| return return_status; | ||
| } | ||
|
|
||
| static vx_status VX_CALLBACK initializeDownmix(vx_node node, const vx_reference *parameters, vx_uint32 num) { | ||
| DownmixLocalData *data = new DownmixLocalData; | ||
| memset(data, 0, sizeof(DownmixLocalData)); | ||
|
|
||
| vx_enum input_tensor_datatype, output_tensor_datatype; | ||
|
|
||
| STATUS_ERROR_CHECK(vxCopyScalar((vx_scalar)parameters[3], &data->device_type, VX_READ_ONLY, VX_MEMORY_TYPE_HOST)); | ||
|
|
||
| // Querying for input tensor | ||
| data->pSrcDesc = new RpptDesc; | ||
| STATUS_ERROR_CHECK(vxQueryTensor((vx_tensor)parameters[0], VX_TENSOR_NUMBER_OF_DIMS, &data->pSrcDesc->numDims, sizeof(data->pSrcDesc->numDims))); | ||
| STATUS_ERROR_CHECK(vxQueryTensor((vx_tensor)parameters[0], VX_TENSOR_DIMS, &data->inputTensorDims, sizeof(vx_size) * data->pSrcDesc->numDims)); | ||
| STATUS_ERROR_CHECK(vxQueryTensor((vx_tensor)parameters[0], VX_TENSOR_DATA_TYPE, &input_tensor_datatype, sizeof(input_tensor_datatype))); | ||
| data->pSrcDesc->dataType = getRpptDataType(input_tensor_datatype); | ||
| data->pSrcDesc->offsetInBytes = 0; | ||
| fillAudioDescriptionPtrFromDims(data->pSrcDesc, data->inputTensorDims); | ||
|
|
||
| // Querying for output tensor | ||
| data->pDstDesc = new RpptDesc; | ||
| STATUS_ERROR_CHECK(vxQueryTensor((vx_tensor)parameters[1], VX_TENSOR_NUMBER_OF_DIMS, &data->pDstDesc->numDims, sizeof(data->pDstDesc->numDims))); | ||
| STATUS_ERROR_CHECK(vxQueryTensor((vx_tensor)parameters[1], VX_TENSOR_DIMS, &data->outputTensorDims, sizeof(vx_size) * data->pDstDesc->numDims)); | ||
| STATUS_ERROR_CHECK(vxQueryTensor((vx_tensor)parameters[1],VX_TENSOR_DATA_TYPE, &output_tensor_datatype, sizeof(output_tensor_datatype))); | ||
| data->pDstDesc->dataType = getRpptDataType(output_tensor_datatype); | ||
| data->pDstDesc->offsetInBytes = 0; | ||
| fillAudioDescriptionPtrFromDims(data->pDstDesc, data->outputTensorDims); | ||
|
|
||
| data->pSamples = new vx_int32[data->pSrcDesc->n]; | ||
| data->pChannels = new vx_int32[data->pSrcDesc->n]; | ||
|
|
||
| refreshDownmix(node, parameters, num, data); | ||
| STATUS_ERROR_CHECK(createRPPHandle(node, &data->handle, data->pSrcDesc->n, data->device_type)); | ||
|
|
||
| STATUS_ERROR_CHECK(vxSetNodeAttribute(node, VX_NODE_LOCAL_DATA_PTR, &data, sizeof(data))); | ||
| return VX_SUCCESS; | ||
| } | ||
|
|
||
| static vx_status VX_CALLBACK uninitializeDownmix(vx_node node, const vx_reference *parameters, vx_uint32 num) { | ||
| DownmixLocalData *data; | ||
| STATUS_ERROR_CHECK(vxQueryNode(node, VX_NODE_LOCAL_DATA_PTR, &data, sizeof(data))); | ||
| STATUS_ERROR_CHECK(releaseRPPHandle(node, data->handle, data->device_type)); | ||
| delete(data->pSamples); | ||
| delete(data->pChannels); | ||
| delete(data); | ||
| return VX_SUCCESS; | ||
| } | ||
|
|
||
| //! \brief The kernel target support callback. | ||
| // TODO::currently the node is setting the same affinity as context. This needs to change when we have hubrid modes in the same graph | ||
| static vx_status VX_CALLBACK query_target_support(vx_graph graph, vx_node node, | ||
| vx_bool use_opencl_1_2, // [input] false: OpenCL driver is 2.0+; true: OpenCL driver is 1.2 | ||
| vx_uint32 &supported_target_affinity // [output] must be set to AGO_TARGET_AFFINITY_CPU or AGO_TARGET_AFFINITY_GPU or (AGO_TARGET_AFFINITY_CPU | AGO_TARGET_AFFINITY_GPU) | ||
| ) { | ||
| vx_context context = vxGetContext((vx_reference)graph); | ||
| AgoTargetAffinityInfo affinity; | ||
| vxQueryContext(context, VX_CONTEXT_ATTRIBUTE_AMD_AFFINITY, &affinity, sizeof(affinity)); | ||
| if (affinity.device_type == AGO_TARGET_AFFINITY_GPU) | ||
| supported_target_affinity = AGO_TARGET_AFFINITY_GPU; | ||
| else | ||
| supported_target_affinity = AGO_TARGET_AFFINITY_CPU; | ||
|
|
||
| // hardcode the affinity to CPU for OpenCL backend to avoid VerifyGraph failure since there is no codegen callback for amd_rpp nodes | ||
|
|
||
| return VX_SUCCESS; | ||
| } | ||
|
|
||
| vx_status Downmix_Register(vx_context context) { | ||
| vx_status status = VX_SUCCESS; | ||
| // Add kernel to the context with callbacks | ||
| vx_kernel kernel = vxAddUserKernel(context, "org.rpp.Downmix", | ||
| VX_KERNEL_RPP_DOWNMIX, | ||
| processDownmix, | ||
| 4, | ||
| validateDownmix, | ||
| initializeDownmix, | ||
| uninitializeDownmix); | ||
| ERROR_CHECK_OBJECT(kernel); | ||
| AgoTargetAffinityInfo affinity; | ||
| vxQueryContext(context, VX_CONTEXT_ATTRIBUTE_AMD_AFFINITY, &affinity, sizeof(affinity)); | ||
| #if ENABLE_HIP | ||
| // enable OpenCL buffer access since the kernel_f callback uses OpenCL buffers instead of host accessible buffers | ||
| vx_bool enableBufferAccess = vx_true_e; | ||
| if (affinity.device_type == AGO_TARGET_AFFINITY_GPU) | ||
| STATUS_ERROR_CHECK(vxSetKernelAttribute(kernel, VX_KERNEL_ATTRIBUTE_AMD_GPU_BUFFER_ACCESS_ENABLE, &enableBufferAccess, sizeof(enableBufferAccess))); | ||
| #else | ||
| vx_bool enableBufferAccess = vx_false_e; | ||
| #endif | ||
| amd_kernel_query_target_support_f query_target_support_f = query_target_support; | ||
|
|
||
| if (kernel) | ||
| { | ||
| STATUS_ERROR_CHECK(vxSetKernelAttribute(kernel, VX_KERNEL_ATTRIBUTE_AMD_QUERY_TARGET_SUPPORT, &query_target_support_f, sizeof(query_target_support_f))); | ||
| PARAM_ERROR_CHECK(vxAddParameterToKernel(kernel, 0, VX_INPUT, VX_TYPE_TENSOR, VX_PARAMETER_STATE_REQUIRED)); | ||
| PARAM_ERROR_CHECK(vxAddParameterToKernel(kernel, 1, VX_OUTPUT, VX_TYPE_TENSOR, VX_PARAMETER_STATE_REQUIRED)); | ||
| PARAM_ERROR_CHECK(vxAddParameterToKernel(kernel, 2, VX_INPUT, VX_TYPE_TENSOR, VX_PARAMETER_STATE_REQUIRED)); | ||
| // PARAM_ERROR_CHECK(vxAddParameterToKernel(kernel, 3, VX_INPUT, VX_TYPE_ARRAY, VX_PARAMETER_STATE_REQUIRED)); | ||
| PARAM_ERROR_CHECK(vxAddParameterToKernel(kernel, 3, VX_INPUT, VX_TYPE_SCALAR, VX_PARAMETER_STATE_REQUIRED)); | ||
| PARAM_ERROR_CHECK(vxFinalizeKernel(kernel)); | ||
| } | ||
| if (status != VX_SUCCESS) | ||
| { | ||
| exit: | ||
| vxRemoveKernel(kernel); | ||
| return VX_FAILURE; | ||
| } | ||
|
|
||
| return status; | ||
| } | ||
|
|