Make device_memory_resource::do_get_mem_info() and supports_get_mem_i…

…nfo() nonvirtual. Remove derived implementations and calls in RMM (#1430)

Closes #1426

As part of #1388, this PR contributes to deprecating and removing all `get_mem_info` functionality from memory resources.  This first PR makes these methods optional without deprecating them. 

 - Makes `rmm::mr::device_memory_resource::supports_get_mem_info()` nonvirtual (and always return false) 
 - Makes `rmm::mr::device_memory_resource::do_get_mem_info()` nonvirtual (and always return `{0, 0}`).
 - Removes all derived implementations of the above.
 - Removes all calls to the above.

Authors:
  - Mark Harris (https://github.com/harrism)

Approvers:
  - Bradley Dice (https://github.com/bdice)
  - Vyas Ramasubramani (https://github.com/vyasr)

URL: #1430

benchmarks/utilities/simulated_memory_resource.hpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2020-2021, NVIDIA CORPORATION.
+ * Copyright (c) 2020-2024, NVIDIA CORPORATION.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -59,13 +59,6 @@ class simulated_memory_resource final : public device_memory_resource {
    */
   [[nodiscard]] bool supports_streams() const noexcept override { return false; }
 
-  /**
-   * @brief Query whether the resource supports the get_mem_info API.
-   *
-   * @return false
-   */
-  [[nodiscard]] bool supports_get_mem_info() const noexcept override { return false; }
-
  private:
   /**
    * @brief Allocates memory of size at least `bytes`.
@@ -95,18 +88,6 @@ class simulated_memory_resource final : public device_memory_resource {
    */
   void do_deallocate(void* ptr, std::size_t, cuda_stream_view) override {}
 
-  /**
-   * @brief Get free and available memory for memory resource.
-   *
-   * @param stream to execute on.
-   * @return std::pair containing free_size and total_size of memory.
-   */
-  [[nodiscard]] std::pair<std::size_t, std::size_t> do_get_mem_info(
-    cuda_stream_view stream) const override
-  {
-    return std::make_pair(0, 0);
-  }
-
   char* begin_{};
   char* end_{};
 };

include/rmm/mr/device/aligned_resource_adaptor.hpp

@@ -96,16 +96,6 @@ class aligned_resource_adaptor final : public device_memory_resource {
     return upstream_->supports_streams();
   }
 
-  /**
-   * @brief Query whether the resource supports the get_mem_info API.
-   *
-   * @return bool true if the upstream resource supports get_mem_info, false otherwise.
-   */
-  [[nodiscard]] bool supports_get_mem_info() const noexcept override
-  {
-    return upstream_->supports_get_mem_info();
-  }
-
   /**
    * @brief The default alignment used by the adaptor.
    */
@@ -183,22 +173,6 @@ class aligned_resource_adaptor final : public device_memory_resource {
            alignment_ == cast->alignment_ && alignment_threshold_ == cast->alignment_threshold_;
   }
 
-  /**
-   * @brief Get free and available memory from upstream resource.
-   *
-   * The free size may not be fully allocatable because of alignment requirements.
-   *
-   * @throws rmm::cuda_error if unable to retrieve memory info.
-   *
-   * @param stream Stream on which to get the mem info.
-   * @return std::pair containing free_size and total_size of memory
-   */
-  [[nodiscard]] std::pair<std::size_t, std::size_t> do_get_mem_info(
-    cuda_stream_view stream) const override
-  {
-    return upstream_->get_mem_info(stream);
-  }
-
   /**
    * @brief Calculate the allocation size needed from upstream to account for alignments of both the
    * size and the base pointer.

include/rmm/mr/device/arena_memory_resource.hpp

@@ -118,13 +118,6 @@ class arena_memory_resource final : public device_memory_resource {
    */
   bool supports_streams() const noexcept override { return true; }
 
-  /**
-   * @brief Query whether the resource supports the get_mem_info API.
-   *
-   * @return bool false.
-   */
-  bool supports_get_mem_info() const noexcept override { return false; }
-
  private:
   using global_arena = rmm::mr::detail::arena::global_arena<Upstream>;
   using arena        = rmm::mr::detail::arena::arena<Upstream>;
@@ -312,18 +305,6 @@ class arena_memory_resource final : public device_memory_resource {
     }
   }
 
-  /**
-   * @brief Get free and available memory for memory resource.
-   *
-   * @param stream to execute on.
-   * @return std::pair containing free_size and total_size of memory.
-   */
-  std::pair<std::size_t, std::size_t> do_get_mem_info(
-    [[maybe_unused]] cuda_stream_view stream) const override
-  {
-    return std::make_pair(0, 0);
-  }
-
   /**
    * Dump memory to log.
    *

include/rmm/mr/device/binning_memory_resource.hpp

@@ -106,13 +106,6 @@ class binning_memory_resource final : public device_memory_resource {
    */
   [[nodiscard]] bool supports_streams() const noexcept override { return true; }
 
-  /**
-   * @brief Query whether the resource supports the get_mem_info API.
-   *
-   * @return false
-   */
-  [[nodiscard]] bool supports_get_mem_info() const noexcept override { return false; }
-
   /**
    * @brief Get the upstream memory_resource object.
    *
@@ -197,20 +190,6 @@ class binning_memory_resource final : public device_memory_resource {
     if (res != nullptr) { res->deallocate(ptr, bytes, stream); }
   }
 
-  /**
-   * @brief Get free and available memory for memory resource
-   *
-   * @throws std::runtime_error if we could not get free / total memory
-   *
-   * @param stream the stream being executed on
-   * @return std::pair with available and free memory for resource
-   */
-  [[nodiscard]] std::pair<std::size_t, std::size_t> do_get_mem_info(
-    [[maybe_unused]] cuda_stream_view stream) const override
-  {
-    return std::make_pair(0, 0);
-  }
-
   Upstream* upstream_mr_;  // The upstream memory_resource from which to allocate blocks.
 
   std::vector<std::unique_ptr<fixed_size_memory_resource<Upstream>>> owned_bin_resources_;

include/rmm/mr/device/callback_memory_resource.hpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2022, NVIDIA CORPORATION.
+ * Copyright (c) 2022-2024, NVIDIA CORPORATION.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -138,13 +138,7 @@ class callback_memory_resource final : public device_memory_resource {
     deallocate_callback_(ptr, bytes, stream, deallocate_callback_arg_);
   }
 
-  [[nodiscard]] std::pair<std::size_t, std::size_t> do_get_mem_info(cuda_stream_view) const override
-  {
-    throw std::runtime_error("cannot get free / total memory");
-  }
-
   [[nodiscard]] bool supports_streams() const noexcept override { return false; }
-  [[nodiscard]] bool supports_get_mem_info() const noexcept override { return false; }
 
   allocate_callback_t allocate_callback_;
   deallocate_callback_t deallocate_callback_;

include/rmm/mr/device/cuda_async_memory_resource.hpp

@@ -164,13 +164,6 @@ class cuda_async_memory_resource final : public device_memory_resource {
    */
   [[nodiscard]] bool supports_streams() const noexcept override { return true; }
 
-  /**
-   * @brief Query whether the resource supports the get_mem_info API.
-   *
-   * @return false
-   */
-  [[nodiscard]] bool supports_get_mem_info() const noexcept override { return false; }
-
  private:
 #ifdef RMM_CUDA_MALLOC_ASYNC_SUPPORT
   cuda_async_view_memory_resource pool_{};
@@ -232,19 +225,6 @@ class cuda_async_memory_resource final : public device_memory_resource {
     return async_mr != nullptr;
 #endif
   }
-
-  /**
-   * @brief Get free and available memory for memory resource
-   *
-   * @throws rmm::cuda_error if unable to retrieve memory info.
-   *
-   * @return std::pair contaiing free_size and total_size of memory
-   */
-  [[nodiscard]] std::pair<std::size_t, std::size_t> do_get_mem_info(
-    rmm::cuda_stream_view) const override
-  {
-    return std::make_pair(0, 0);
-  }
 };
 
 /** @} */  // end of group

include/rmm/mr/device/cuda_async_view_memory_resource.hpp

@@ -99,13 +99,6 @@ class cuda_async_view_memory_resource final : public device_memory_resource {
    */
   [[nodiscard]] bool supports_streams() const noexcept override { return true; }
 
-  /**
-   * @brief Query whether the resource supports the get_mem_info API.
-   *
-   * @return true
-   */
-  [[nodiscard]] bool supports_get_mem_info() const noexcept override { return false; }
-
  private:
 #ifdef RMM_CUDA_MALLOC_ASYNC_SUPPORT
   cudaMemPool_t cuda_pool_handle_{};
@@ -169,19 +162,6 @@ class cuda_async_view_memory_resource final : public device_memory_resource {
   {
     return dynamic_cast<cuda_async_view_memory_resource const*>(&other) != nullptr;
   }
-
-  /**
-   * @brief Get free and available memory for memory resource
-   *
-   * @throws rmm::cuda_error if unable to retrieve memory info.
-   *
-   * @return std::pair contaiing free_size and total_size of memory
-   */
-  [[nodiscard]] std::pair<std::size_t, std::size_t> do_get_mem_info(
-    rmm::cuda_stream_view) const override
-  {
-    return std::make_pair(0, 0);
-  }
 };
 
 /** @} */  // end of group

include/rmm/mr/device/cuda_memory_resource.hpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2019-2021, NVIDIA CORPORATION.
+ * Copyright (c) 2019-2024, NVIDIA CORPORATION.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -51,13 +51,6 @@ class cuda_memory_resource final : public device_memory_resource {
    */
   [[nodiscard]] bool supports_streams() const noexcept override { return false; }
 
-  /**
-   * @brief Query whether the resource supports the get_mem_info API.
-   *
-   * @return true
-   */
-  [[nodiscard]] bool supports_get_mem_info() const noexcept override { return true; }
-
  private:
   /**
    * @brief Allocates memory of size at least \p bytes.
@@ -108,21 +101,6 @@ class cuda_memory_resource final : public device_memory_resource {
   {
     return dynamic_cast<cuda_memory_resource const*>(&other) != nullptr;
   }
-
-  /**
-   * @brief Get free and available memory for memory resource
-   *
-   * @throws rmm::cuda_error if unable to retrieve memory info.
-   *
-   * @return std::pair contaiing free_size and total_size of memory
-   */
-  [[nodiscard]] std::pair<std::size_t, std::size_t> do_get_mem_info(cuda_stream_view) const override
-  {
-    std::size_t free_size{};
-    std::size_t total_size{};
-    RMM_CUDA_TRY(cudaMemGetInfo(&free_size, &total_size));
-    return std::make_pair(free_size, total_size);
-  }
 };
 /** @} */  // end of group
 }  // namespace rmm::mr

include/rmm/mr/device/device_memory_resource.hpp

@@ -306,7 +306,7 @@ class device_memory_resource {
    *
    * @return bool true if the resource supports get_mem_info, false otherwise.
    */
-  [[nodiscard]] virtual bool supports_get_mem_info() const noexcept = 0;
+  [[nodiscard]] virtual bool supports_get_mem_info() const noexcept { return false; };
 
   /**
    * @brief Queries the amount of free and total memory for the resource.
@@ -384,7 +384,10 @@ class device_memory_resource {
    * @return std::pair with available and free memory for resource
    */
   [[nodiscard]] virtual std::pair<std::size_t, std::size_t> do_get_mem_info(
-    cuda_stream_view stream) const = 0;
+    cuda_stream_view stream) const
+  {
+    return {0, 0};
+  }
 };
 static_assert(cuda::mr::async_resource_with<device_memory_resource, cuda::mr::device_accessible>);
 /** @} */  // end of group

include/rmm/mr/device/failure_callback_resource_adaptor.hpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2020-2021, NVIDIA CORPORATION.
+ * Copyright (c) 2020-2024, NVIDIA CORPORATION.
  *
  * Licensed under the Apache License, Version 2.0 (the "License");
  * you may not use this file except in compliance with the License.
@@ -135,16 +135,6 @@ class failure_callback_resource_adaptor final : public device_memory_resource {
     return upstream_->supports_streams();
   }
 
-  /**
-   * @brief Query whether the resource supports the get_mem_info API.
-   *
-   * @return bool true if the upstream resource supports get_mem_info, false otherwise.
-   */
-  [[nodiscard]] bool supports_get_mem_info() const noexcept override
-  {
-    return upstream_->supports_get_mem_info();
-  }
-
  private:
   /**
    * @brief Allocates memory of size at least `bytes` using the upstream
@@ -199,20 +189,6 @@ class failure_callback_resource_adaptor final : public device_memory_resource {
                            : upstream_->is_equal(other);
   }
 
-  /**
-   * @brief Get free and available memory from upstream resource.
-   *
-   * @throws rmm::cuda_error if unable to retrieve memory info.
-   *
-   * @param stream Stream on which to get the mem info.
-   * @return std::pair contaiing free_size and total_size of memory
-   */
-  [[nodiscard]] std::pair<std::size_t, std::size_t> do_get_mem_info(
-    cuda_stream_view stream) const override
-  {
-    return upstream_->get_mem_info(stream);
-  }
-
   Upstream* upstream_;  // the upstream resource used for satisfying allocation requests
   failure_callback_t callback_;
   void* callback_arg_;

include/rmm/mr/device/fixed_size_memory_resource.hpp

@@ -104,13 +104,6 @@ class fixed_size_memory_resource
    */
   [[nodiscard]] bool supports_streams() const noexcept override { return true; }
 
-  /**
-   * @brief Query whether the resource supports the get_mem_info API.
-   *
-   * @return false
-   */
-  [[nodiscard]] bool supports_get_mem_info() const noexcept override { return false; }
-
   /**
    * @brief Get the upstream memory_resource object.
    *
@@ -211,20 +204,6 @@ class fixed_size_memory_resource
     return block_type{ptr};
   }
 
-  /**
-   * @brief Get free and available memory for memory resource
-   *
-   * @throws std::runtime_error if we could not get free / total memory
-   *
-   * @param stream the stream being executed on
-   * @return std::pair with available and free memory for resource
-   */
-  [[nodiscard]] std::pair<std::size_t, std::size_t> do_get_mem_info(
-    [[maybe_unused]] cuda_stream_view stream) const override
-  {
-    return std::make_pair(0, 0);
-  }
-
   /**
    * @brief free all memory allocated using the upstream resource.
    *
@@ -244,7 +223,7 @@ class fixed_size_memory_resource
   {
     lock_guard lock(this->get_mutex());
 
-    auto const [free, total] = get_upstream()->get_mem_info(rmm::cuda_stream_default);
+    auto const [free, total] = rmm::available_device_memory();
     std::cout << "GPU free memory: " << free << " total: " << total << "\n";
 
     std::cout << "upstream_blocks: " << upstream_blocks_.size() << "\n";