[chore] add pinned mem buffer

hpcaitech · Oct 29, 2024 · 4d19a44 · 4d19a44
1 parent 0ed8faa
commit 4d19a44
Show file tree

Hide file tree

Showing 12 changed files with 37 additions and 23 deletions.
diff --git a/csrc/aio.cpp b/csrc/aio.cpp
@@ -1,5 +1,3 @@
-#include <stdexcept>
-#include <memory>
 #include "aio.h"
 
 AIOAsyncIO::AIOAsyncIO(unsigned int n_entries)
@@ -128,9 +126,14 @@ void AIOAsyncIO::readv(int fd, const iovec *iov, unsigned int iovcnt, unsigned l
     this->n_read_events++;
 }
 
-void AIOAsyncIO::write_tensor(int fd, torch::Tensor t, unsigned long long offset, callback_t callback) {
+void AIOAsyncIO::write_tensor(int fd, torch::Tensor t, unsigned long long offset, callback_t callback, std::optional<torch::Tensor> pinned) {
     if (t.is_cuda()) {
-        t = t.to(torch::kCPU);
+        if (pinned.has_value()) {
+            pinned.value().copy_(t);
+            t = pinned.value();
+        } else {
+            t = t.to(torch::kCPU);
+        }
     }
     void *buffer = t.data_ptr();
     size_t n_bytes = t.numel() * t.element_size();

diff --git a/csrc/async_file_io.cpp b/csrc/async_file_io.cpp
@@ -8,8 +8,8 @@ void AsyncFileWriter::write(size_t buffer, size_t n_bytes, unsigned long long of
     this->aio->write(this->fd, ptr, n_bytes, offset, callback);
 }
 
-void AsyncFileWriter::write_tensor(torch::Tensor tensor, unsigned long long offset, callback_t callback) {
-    this->aio->write_tensor(this->fd, tensor, offset, callback);
+void AsyncFileWriter::write_tensor(torch::Tensor tensor, unsigned long long offset, callback_t callback, std::optional<torch::Tensor> pinned) {
+    this->aio->write_tensor(this->fd, tensor, offset, callback, pinned);
 }
 
 

diff --git a/csrc/pthread_backend.cpp b/csrc/pthread_backend.cpp
@@ -80,14 +80,17 @@ void PthreadAsyncIO::synchronize() {
 
 void PthreadAsyncIO::register_file(int fd) {}
 
-void PthreadAsyncIO::write_tensor(int fd, torch::Tensor t, unsigned long long offset, callback_t callback) {
+void PthreadAsyncIO::write_tensor(int fd, torch::Tensor t, unsigned long long offset, callback_t callback, std::optional<torch::Tensor> pinned) {
     auto fut = this->pool.submit_task(
-        [fd, t, offset] {
+        [fd, t, offset, pinned] {
             torch::Tensor cpu_tensor;
             if (t.is_cuda()) {
-                cpu_tensor = t.to(torch::kCPU);
-            } else {
-                cpu_tensor = t;
+                if (pinned.has_value()) {
+                    pinned.value().copy_(t);
+                    cpu_tensor = pinned.value();
+                } else {
+                    cpu_tensor = t.to(torch::kCPU);
+                }
             }
             void *buf = cpu_tensor.data_ptr();
             size_t n_bytes = cpu_tensor.numel() * cpu_tensor.element_size();

diff --git a/csrc/py_api.cpp b/csrc/py_api.cpp
@@ -29,6 +29,6 @@ PYBIND11_MODULE(TORCH_EXTENSION_NAME, m)
     py::class_<AsyncFileWriter>(m, "AsyncFileWriter")
         .def(py::init<int, unsigned int, const std::string &>(), py::arg("fd"), py::arg("n_entries"), py::arg("backend") = "aio")
         .def("write", &AsyncFileWriter::write, py::arg("buffer"), py::arg("n_bytes"), py::arg("offset"), py::arg("callback") = py::none())
-        .def("write_tensor", &AsyncFileWriter::write_tensor, py::arg("tensor"), py::arg("offset"), py::arg("callback") = py::none())
+        .def("write_tensor", &AsyncFileWriter::write_tensor, py::arg("tensor"), py::arg("offset"), py::arg("callback") = py::none(), py::arg("pinned") = py::none())
         .def("synchronize", &AsyncFileWriter::synchronize);
 }
diff --git a/csrc/uring.cpp b/csrc/uring.cpp
@@ -99,9 +99,14 @@ void UringAsyncIO::readv(int fd, const iovec *iov, unsigned int iovcnt, unsigned
     this->n_read_events++;
 }
 
-void UringAsyncIO::write_tensor(int fd, torch::Tensor t, unsigned long long offset, callback_t callback) {
+void UringAsyncIO::write_tensor(int fd, torch::Tensor t, unsigned long long offset, callback_t callback, std::optional<torch::Tensor> pinned) {
     if (t.is_cuda()) {
-        t = t.to(torch::kCPU);
+        if (pinned.has_value()) {
+            pinned.value().copy_(t);
+            t = pinned.value();
+        } else {
+            t = t.to(torch::kCPU);
+        }
     }
     void *buffer = t.data_ptr<float>();
     size_t n_bytes = t.numel() * t.element_size();

diff --git a/include/aio.h b/include/aio.h
@@ -2,6 +2,8 @@
 
 #include <libaio.h>
 #include <torch/torch.h>
+#include <stdexcept>
+#include <memory>
 #include "asyncio.h"
 
 class AIOAsyncIO : public AsyncIO
@@ -30,5 +32,5 @@ class AIOAsyncIO : public AsyncIO
     void synchronize();
 
     void register_file(int fd);
-    void write_tensor(int fd, torch::Tensor t, unsigned long long offset, callback_t callback);
+    void write_tensor(int fd, torch::Tensor t, unsigned long long offset, callback_t callback, std::optional<torch::Tensor> pinned);
 };
diff --git a/include/async_file_io.h b/include/async_file_io.h
@@ -1,6 +1,7 @@
 #pragma once
 #include <string>
 #include <torch/torch.h>
+#include <optional>
 
 #include "asyncio.h"
 #include "backend.h"
@@ -18,7 +19,7 @@ class AsyncFileWriter
 public:
     AsyncFileWriter(int fd, unsigned int n_entries, const std::string &backend);
     void write(size_t buffer, size_t n_bytes, unsigned long long offset, callback_t callback);
-    void write_tensor(torch::Tensor tensor, unsigned long long offset, callback_t callback);
+    void write_tensor(torch::Tensor tensor, unsigned long long offset, callback_t callback, std::optional<torch::Tensor> pinned);
     void synchronize();
     ~AsyncFileWriter();
 

diff --git a/include/asyncio.h b/include/asyncio.h
@@ -48,5 +48,5 @@ class AsyncIO
     virtual void synchronize() = 0;
 
     virtual void register_file(int fd) = 0;
-    virtual void write_tensor(int fd, torch::Tensor t, unsigned long long offset, callback_t callback) = 0;
+    virtual void write_tensor(int fd, torch::Tensor t, unsigned long long offset, callback_t callback, std::optional<torch::Tensor> pinned) = 0;
 };
diff --git a/include/pthread_backend.h b/include/pthread_backend.h
@@ -39,5 +39,5 @@ class PthreadAsyncIO : public AsyncIO
 
     void register_file(int fd);
 
-    void write_tensor(int fd, torch::Tensor t, unsigned long long offset, callback_t callback);
+    void write_tensor(int fd, torch::Tensor t, unsigned long long offset, callback_t callback, std::optional<torch::Tensor> pinned);
 };
diff --git a/include/uring.h b/include/uring.h
@@ -26,5 +26,5 @@ class UringAsyncIO : public AsyncIO
     void synchronize();
 
     void register_file(int fd);
-    void write_tensor(int fd, torch::Tensor t, unsigned long long offset, callback_t callback);
+    void write_tensor(int fd, torch::Tensor t, unsigned long long offset, callback_t callback, std::optional<torch::Tensor> pinned);
 };
diff --git a/tensornvme/_C/__init__.pyi b/tensornvme/_C/__init__.pyi
@@ -22,5 +22,5 @@ def probe_backend(backend: str) -> bool: ...
 class AsyncFileWriter:
     def __init__(self, fd: int, n_entries: int, backend: str = "aio") -> None: ...
     def write(self, buffer: int, n_bytes: int, offset: int, callback: Optional[Callable[[], None]] = None) -> None: ...
-    def write_tensor(self, tensor: Tensor, offset: int, callback: Optional[Callable[[], None]] = None) -> None: ...
+    def write_tensor(self, tensor: Tensor, offset: int, callback: Optional[Callable[[], None]] = None, pinned: Optional[Tensor] = None) -> None: ...
     def synchronize(self) -> None: ...
diff --git a/tensornvme/async_file_io.py b/tensornvme/async_file_io.py
@@ -1,7 +1,7 @@
 import ctypes
 from functools import partial
 from torch import Tensor
-from typing import List
+from typing import List, Optional
 from io import IOBase
 from tensornvme._C import AsyncFileWriter as AsyncFileWriterC
 
@@ -31,9 +31,9 @@ def write_raw(self, py_ref: object, buffer: int, n_bytes: int, offset: int) -> N
         self.io.write(buffer, n_bytes, offset, partial(AsyncFileWriter.gc_callback, self.buffers, len(self.buffers) - 1))
         self.offset += n_bytes
 
-    def write_tensor(self, tensor: Tensor) -> None:
+    def write_tensor(self, tensor: Tensor, pinned: Optional[Tensor] = None) -> None:
         self.buffers.append(tensor)  # append before callback is called
-        self.io.write_tensor(tensor, self.offset, partial(AsyncFileWriter.gc_callback, self.buffers, len(self.buffers) - 1))
+        self.io.write_tensor(tensor, self.offset, partial(AsyncFileWriter.gc_callback, self.buffers, len(self.buffers) - 1), pinned)
         self.offset += tensor.numel() * tensor.element_size()
 
     @staticmethod