Tensor changes in master graph and ring buffer (#17)

* Add changes in master graph.h and master_graph.cpp

Changed all instances from Image to Tensor
Remove the cycle idx loop in output routine
Removed video output routine
Removed copy out API
Remove API to get output width, height from info
Remove output tensor allocation

* Remove instances of output tensor in MasterGraph

* Add get_output_tensors API in MasterGraph

* Remove IBS and associated changes from MasterGraph

* Add changes in ring buffer

Remove master buffer and intrduce sub buffer for HOST

* Remove processing on HIP and OCL check from master graph

* Remove sub buffer count variable from ring buffer init

* Remove use of vx_tensor for the output tensors

* Code clean up

* Changed output ptr type to auto

rocAL/rocAL/include/pipeline/context.h

@@ -35,7 +35,6 @@ struct Context
     {
         LOG("Processing on " + STR(((affinity == RocalAffinity::CPU)?" CPU": " GPU")))
         master_graph = std::make_shared<MasterGraph>(batch_size, affinity, gpu_id, prefetch_queue_depth, output_tensor_type);
-        _internal_batch_size = master_graph->internal_batch_size();
     }
     ~Context()
     {
@@ -52,10 +51,8 @@ struct Context
         return master_graph->timing();
     }
     size_t user_batch_size() { return _user_batch_size; }
-    size_t internal_batch_size() { return _internal_batch_size; }
 private:
     void clear_errors() { error = "";}
     std::string error;
     size_t _user_batch_size;
-    size_t _internal_batch_size;
-};
+};

rocAL/rocAL/include/pipeline/ring_buffer.h

@@ -45,10 +45,10 @@ class RingBuffer
     ///\param dev
     ///\param sub_buffer_size
     ///\param sub_buffer_count
-    void init(RocalMemType mem_type, void *dev, unsigned sub_buffer_size, unsigned sub_buffer_count);
+    void init(RocalMemType mem_type, void *dev, std::vector<size_t> &sub_buffer_size);
     void initBoxEncoderMetaData(RocalMemType mem_type, size_t encoded_bbox_size, size_t encoded_labels_size);
     void release_gpu_res();
-    std::vector<void*> get_read_buffers() ;
+    std::vector<void*> get_read_buffers();
     void* get_host_master_read_buffer();
     std::vector<void*> get_write_buffers();
     std::pair<void*, void*> get_box_encode_write_buffers();
@@ -72,13 +72,12 @@ class RingBuffer
     void increment_write_ptr();
     bool full();
     const unsigned BUFF_DEPTH;
-    unsigned _sub_buffer_size;
+    std::vector<size_t> _sub_buffer_size;
     unsigned _sub_buffer_count;
     std::mutex _lock;
     std::condition_variable _wait_for_load;
     std::condition_variable _wait_for_unload;
     std::vector<std::vector<void*>> _dev_sub_buffer;
-    std::vector<void*> _host_master_buffers;
     std::vector<std::vector<void*>> _host_sub_buffers;
     std::vector<void *> _dev_bbox_buffer;
     std::vector<void *> _dev_labels_buffer;

rocAL/rocAL/source/api/rocal_api_data_loaders.cpp

@@ -301,7 +301,6 @@ rocalSequenceReader(
         // Set sequence batch size and batch ratio in master graph as it varies according to sequence length
         context->master_graph->set_sequence_reader_output();
         context->master_graph->set_sequence_batch_size(sequence_length);
-        context->master_graph->set_sequence_batch_ratio();
         bool decoder_keep_original = true;
 
         // This has been introduced to support variable width and height video frames in future.
@@ -384,7 +383,6 @@ rocalSequenceReaderSingleShard(
         // Set sequence batch size and batch ratio in master graph as it varies according to sequence length
         context->master_graph->set_sequence_reader_output();
         context->master_graph->set_sequence_batch_size(sequence_length);
-        context->master_graph->set_sequence_batch_ratio();
         bool decoder_keep_original = true;
 
         // This has been introduced to support variable width and height video frames in future.