gpu::TensorPool Struct Reference

Represents a pool of tensors to manage GPU resources. The pool is responsible for managing the lifetime of the tensors and freeing them when the pool is destroyed. More...

#include <gpu.h>

Public Member Functions
	TensorPool (Context *ctx)
	~TensorPool ()
	Destructor for TensorPool which frees all tensors in the pool.

Public Attributes
Context *	ctx
std::unordered_map< WGPUBuffer, Tensor >	data

Detailed Description

Represents a pool of tensors to manage GPU resources. The pool is responsible for managing the lifetime of the tensors and freeing them when the pool is destroyed.

Most users do not need to interact with the TensorPool type, as there is a member instance in the Context struct to simplify lifetime management of GPU resources.

Definition at line 176 of file gpu.h.

Constructor & Destructor Documentation

TensorPool()

gpu::TensorPool::TensorPool ( Context * ctx )

inline

Definition at line 177 of file gpu.h.

: ctx(ctx), data() {};

~TensorPool()

gpu::TensorPool::~TensorPool ( )

inline

Destructor for TensorPool which frees all tensors in the pool.

Definition at line 624 of file gpu.h.

                               {
  // Need to get keys in a separate iteration, otherwise iterator is getting
  // invalidated during erase.
  std::vector<WGPUBuffer> keys;
  for (auto &pair : data) {
    keys.push_back(pair.first);
  }
  for (auto &key : keys) {
    FreeTensor(*this, data[key]);
    LOG(kDefLog, kTrace, "Freed tensor");
  }
}

Member Data Documentation

ctx

Context* gpu::TensorPool::ctx

Definition at line 178 of file gpu.h.

data

std::unordered_map<WGPUBuffer, Tensor> gpu::TensorPool::data

Definition at line 179 of file gpu.h.

---

The documentation for this struct was generated from the following file:

• gpu.h