[llvm] [Offload] Implement the remaining initial Offload API (PR #122106)

[Joseph Huber via llvm-commits]

================
@@ -0,0 +1,48 @@
+//===-- Memory.td - Memory definitions for Offload ---------*- tablegen -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// This file contains Offload API definitions related to memory allocations
+//
+//===----------------------------------------------------------------------===//
+
+def : Enum {
+  let name = "ol_alloc_type_t";
+  let desc = "Represents the type of allocation made with olMemAlloc.";
+  let etors = [
+    Etor<"HOST", "Host allocation">,
+    Etor<"DEVICE", "Device allocation">,
+    Etor<"SHARED", "Shared allocation">
+  ];
+}
+
+def : Function {
+  let name = "olMemAlloc";
+  let desc = "Creates a memory allocation on the specified device.";
+  let params = [
+    Param<"ol_device_handle_t", "Device", "handle of the device to allocate on", PARAM_IN>,
+    Param<"ol_alloc_type_t", "Type", "type of the allocation", PARAM_IN>,
+    Param<"size_t", "Size", "size of the allocation in bytes", PARAM_IN>,
+    Param<"void**", "AllocationOut", "output for the allocated pointer", PARAM_OUT>
+  ];
+  let returns = [
+    Return<"OL_ERRC_INVALID_SIZE", [
+      "`Size == 0`"
+    ]>
+  ];
+}
+
+def : Function {
+  let name = "olMemFree";
+  let desc = "Frees a memory allocation previously made by olMemAlloc.";
+  let params = [
+    Param<"ol_device_handle_t", "Device", "handle of the device to allocate on", PARAM_IN>,
----------------
jhuber6 wrote:

I'm not an expert, but I think in theory you could have HMM configured for the same memory region between separate GPUs. The HSA API thinks about thins in terms of memory pools and keeps some RB trees to see if a pointer is resident between some allocated ranged IIUC. The `free` function just takes a pointer and then determines the memory pool / device that owns it. They also have the concept of 'setting ownership'. I.e. when you allocate any memory you need to explicitly enable access to the device you want to touch it.

I think in general, we shouldn't force the device here since it'd get really complicated. We could pass the *type* like the CUDA APIs do if we really needed to, but ideally it'd just be a a pointer.

https://github.com/llvm/llvm-project/pull/122106