NVIDIA
diff --git a/‎cuda_core/cuda/core/_cpp/rt/DESIGN.md‎
Lines changed: 5 additions & 0 deletions b/‎cuda_core/cuda/core/_cpp/rt/DESIGN.md‎
Lines changed: 5 additions & 0 deletions
diff --git a/‎cuda_core/cuda/core/_cpp/rt/VMM_DESIGN.md‎
Lines changed: 269 additions & 0 deletions b/‎cuda_core/cuda/core/_cpp/rt/VMM_DESIGN.md‎
Lines changed: 269 additions & 0 deletions
diff --git a/‎cuda_core/cuda/core/_cpp/rt/api.hpp‎
Lines changed: 66 additions & 0 deletions b/‎cuda_core/cuda/core/_cpp/rt/api.hpp‎
Lines changed: 66 additions & 0 deletions
diff --git a/‎cuda_core/cuda/core/_cpp/rt/driver_api.hpp‎
Lines changed: 13 additions & 1 deletion b/‎cuda_core/cuda/core/_cpp/rt/driver_api.hpp‎
Lines changed: 13 additions & 1 deletion
diff --git a/‎cuda_core/cuda/core/_cpp/rt/internal.hpp‎
Lines changed: 9 additions & 0 deletions b/‎cuda_core/cuda/core/_cpp/rt/internal.hpp‎
Lines changed: 9 additions & 0 deletions
diff --git a/‎cuda_core/cuda/core/_cpp/rt/memory.cpp‎
Lines changed: 46 additions & 2 deletions b/‎cuda_core/cuda/core/_cpp/rt/memory.cpp‎
Lines changed: 46 additions & 2 deletions
diff --git a/‎cuda_core/cuda/core/_cpp/rt/py.hpp‎
Lines changed: 51 additions & 0 deletions b/‎cuda_core/cuda/core/_cpp/rt/py.hpp‎
Lines changed: 51 additions & 0 deletions
diff --git a/‎cuda_core/cuda/core/_cpp/rt/py_driver_fns.cpp‎
Lines changed: 0 additions & 33 deletions b/‎cuda_core/cuda/core/_cpp/rt/py_driver_fns.cpp‎
Lines changed: 0 additions & 33 deletions
diff --git a/‎cuda_core/cuda/core/_cpp/rt/py_report.cpp‎
Lines changed: 1 addition & 11 deletions b/‎cuda_core/cuda/core/_cpp/rt/py_report.cpp‎
Lines changed: 1 addition & 11 deletions
@@ -63,6 +63,11 @@ Internally, handles use **shared pointer aliasing**: the actual managed object i
 "box" containing the resource, its dependencies, and any state needed for destruction.
 The public handle points only to the raw resource field, keeping the API minimal.
 
+The virtual memory resource adds `MemAllocationHandle`, `VaReservationHandle` and
+`VaMappingHandle`. Their values are `TaggedHandle<T, N>` wrappers, because
+`CUmemGenericAllocationHandle` and `CUdeviceptr` are both `unsigned long long` and the
+accessor overloads must stay distinct. See [VMM_DESIGN.md](VMM_DESIGN.md).
+
 ### Why shared_ptr?
 
 - **Automatic reference counting**: Resources are released when the last reference
 
@@ -5,6 +5,7 @@
 #pragma once
 
 #include "types.hpp"
+#include <vector>
 #include <cuda.h>
 #include <nvrtc.h>
 #include <cstddef>
@@ -232,6 +233,7 @@ DevicePtrHandle deviceptr_import_ipc(
 
 // Access the deallocation stream for a device pointer handle (read-only).
 // For non-owning handles, the stream is not used but can still be accessed.
+// Returns an empty handle for an empty device pointer handle.
 StreamHandle deallocation_stream(const DevicePtrHandle& h) noexcept;
 
 // Set the deallocation stream for a device pointer handle.
@@ -240,6 +242,70 @@ StreamHandle deallocation_stream(const DevicePtrHandle& h) noexcept;
 CUresult set_deallocation_stream(
     const DevicePtrHandle& h, const StreamHandle& h_stream) noexcept;
 
+// ============================================================================
+// Virtual memory management (VMM_DESIGN.md)
+//
+// A VirtualMemoryResource buffer is a range of mappings. Each mapping holds
+// one physical allocation and one address reservation; the mapping deleter
+// unmaps, then the allocation is released and the reservation freed as their
+// last references go. A buffer's DevicePtrHandle owns the range. Ranges are
+// immutable: a grow builds a new range for its result.
+// ============================================================================
+
+// Create a physical allocation via cuMemCreate. The access descriptors are
+// applied to every mapping of this allocation. When the last reference is
+// released, cuMemRelease is called; the memory is freed once no mapping
+// remains. Returns empty handle on error (caller must check).
+MemAllocationHandle create_mem_allocation_handle(size_t size, const CUmemAllocationProp& prop,
+                                                 const CUmemAccessDesc* descs, size_t count);
+
+// Size of the allocation; the only size cuMemMap accepts for it.
+size_t mem_allocation_size(const MemAllocationHandle& h) noexcept;
+
+// Reserve an address range via cuMemAddressReserve. Pass alignment 0 for the
+// driver default. When the last reference is released, cuMemAddressFree is
+// called with the exact reserved pair. Returns empty handle on error.
+VaReservationHandle create_va_reservation_handle(size_t size, size_t alignment, CUdeviceptr hint);
+
+// Size of the reservation.
+size_t va_reservation_size(const VaReservationHandle& h) noexcept;
+
+// Map the whole allocation at ptr inside the reservation via cuMemMap and
+// apply the allocation's access descriptors. The mapping structurally depends
+// on both handles. When the last reference is released, cuMemUnmap is called
+// first. Returns empty handle on error, including a range outside the
+// reservation; a failed cuMemSetAccess unmaps before returning.
+VaMappingHandle create_va_mapping_handle(CUdeviceptr ptr, const MemAllocationHandle& h_alloc,
+                                         const VaReservationHandle& h_res);
+
+// Mapping accessors.
+size_t va_mapping_size(const VaMappingHandle& h) noexcept;
+MemAllocationHandle va_mapping_allocation(const VaMappingHandle& h) noexcept;
+
+// Build an immutable range from mappings in ascending, contiguous order. A
+// grow builds a new range for its result and never changes the input's.
+// May throw std::bad_alloc.
+VmmRangeHandle create_vmm_range(const std::vector<VaMappingHandle>& mappings);
+
+// The range of a device pointer handle created by deviceptr_create_vmm. Only
+// for such handles: the caller (VirtualMemoryBuffer) guarantees the origin.
+// Empty for an empty handle.
+VmmRangeHandle vmm_range(const DevicePtrHandle& h) noexcept;
+
+// Range accessors: a copy of the mapping list, which a grow extends and turns
+// into a new range, and the range total. Reads of an immutable range need no
+// synchronization.
+std::vector<VaMappingHandle> vmm_range_mappings(const VmmRangeHandle& range);  // may throw
+size_t vmm_range_total(const VmmRangeHandle& range) noexcept;
+
+// Create a device pointer handle whose box owns a range (an empty range for
+// a size-zero buffer). The box records no deallocation stream; set one with
+// set_deallocation_stream. When the last reference is released, the recorded
+// stream is synchronized (skipped with a report when that would disturb a
+// capture) and the box freed, which unmaps every mapping this buffer was the
+// last to hold. Returns empty handle for a null range handle.
+DevicePtrHandle deviceptr_create_vmm(CUdeviceptr base, const VmmRangeHandle& range);
+
 // ============================================================================
 // Library handle functions
 // ============================================================================
 
@@ -126,7 +126,19 @@ namespace cuda_core::rt {
     X(cuTexObjectCreate, 5000)                   \
     X(cuTexObjectDestroy, 5000)                  \
     X(cuSurfObjectCreate, 5000)                  \
-    X(cuSurfObjectDestroy, 5000)
+    X(cuSurfObjectDestroy, 5000)                 \
+    /* Virtual memory management (VMM_DESIGN.md) */ \
+    X(cuMemCreate, 10020)                        \
+    X(cuMemRelease, 10020)                       \
+    X(cuMemAddressReserve, 10020)                \
+    X(cuMemAddressFree, 10020)                   \
+    X(cuMemMap, 10020)                           \
+    X(cuMemUnmap, 10020)                         \
+    X(cuMemSetAccess, 10020)                     \
+    /* cuda-bindings requests 7000 (PTDS) or 2000 (legacy) */ \
+    X(cuStreamSynchronize, 7000)                 \
+    X(cuStreamIsCapturing, 10000)                \
+    X(cuThreadExchangeStreamCaptureMode, 10010)
 
 #define CUDA_CORE_DECLARE_DRIVER_FN(name, introduced) extern decltype(&name) p_##name;
 CUDA_CORE_DRIVER_FUNCTIONS(CUDA_CORE_DECLARE_DRIVER_FN)
 
@@ -62,6 +62,12 @@ ContextHandle deallocation_context(const DeallocationStream& stream) noexcept;
 // Implemented in stream.cpp
 bool make_deallocation_stream(const StreamHandle& h, DeallocationStream& out) noexcept;
 
+// Implemented in virtual_memory.cpp. Synchronize the stream a VMM buffer
+// recorded before its mappings are released. Skips the sync, with a report
+// when a capture is the reason, if no stream was recorded, the interpreter is
+// finalizing, or the sync would disturb a graph capture (VMM_DESIGN.md).
+void vmm_sync_before_release(const DeallocationStream& stream) noexcept;
+
 // Decorate a status-returning cleanup call to report whenever it fails. CUDA
 // calls (CUresult) are reported with the error name and description; NVRTC,
 // NVVM and nvJitLink calls (integer status codes) with the raw code.
@@ -126,6 +132,9 @@ const WarnOnFailure<p_cuSurfObjectDestroy> pw_cuSurfObjectDestroy{"cuSurfObjectD
 const WarnOnFailure<p_cuGreenCtxDestroy> pw_cuGreenCtxDestroy{"cuGreenCtxDestroy"};
 const WarnOnFailure<p_cuMemPoolDestroy> pw_cuMemPoolDestroy{"cuMemPoolDestroy"};
 const WarnOnFailure<p_cuMemFreeHost> pw_cuMemFreeHost{"cuMemFreeHost"};
+const WarnOnFailure<p_cuMemRelease> pw_cuMemRelease{"cuMemRelease"};
+const WarnOnFailure<p_cuMemUnmap> pw_cuMemUnmap{"cuMemUnmap"};
+const WarnOnFailure<p_cuMemAddressFree> pw_cuMemAddressFree{"cuMemAddressFree"};
 const WarnOnFailure<p_cuGraphDestroy> pw_cuGraphDestroy{"cuGraphDestroy"};
 const WarnOnFailure<p_cuGraphExecDestroy> pw_cuGraphExecDestroy{"cuGraphExecDestroy"};
 const WarnOnFailure<p_cuGraphicsUnregisterResource> pw_cuGraphicsUnregisterResource{"cuGraphicsUnregisterResource"};
 
@@ -123,6 +123,15 @@ struct DevicePtrBox {
     // stream tokens carry a bound context.
     mutable DeallocationStream deallocation;
 };
+
+// The box behind a VirtualMemoryResource buffer: the range of mappings it
+// owns (VMM_DESIGN.md). Every handle on a VirtualMemoryBuffer comes from
+// deviceptr_create_vmm, so vmm_range() downcasts without a tag; there are no
+// virtual functions, so other boxes pay nothing. A size-zero buffer has an
+// empty range.
+struct VmmDevicePtrBox : DevicePtrBox {
+    VmmRangeHandle range;
+};
 }  // namespace
 
 // Recovers the owning DevicePtrBox from the aliased CUdeviceptr pointer.
@@ -137,9 +146,10 @@ static DevicePtrBox* get_box(const DevicePtrHandle& h) {
     );
 }
 
-// Return the stream that orders a device pointer's deallocation.
+// Return the stream that orders a device pointer's deallocation; empty for an
+// empty handle, as set_deallocation_stream rejects one.
 StreamHandle deallocation_stream(const DevicePtrHandle& h) noexcept {
-    return get_box(h)->deallocation.h_stream;
+    return h ? get_box(h)->deallocation.h_stream : StreamHandle{};
 }
 
 // Replace the stream that orders a device pointer's deallocation.
@@ -300,6 +310,36 @@ DevicePtrHandle deviceptr_create_mapped_graphics(
     return DevicePtrHandle(box, &box->resource);
 }
 
+// ============================================================================
+// Virtual memory ranges (VMM_DESIGN.md)
+// ============================================================================
+
+DevicePtrHandle deviceptr_create_vmm(CUdeviceptr base, const VmmRangeHandle& range) {
+    if (!range) {  // a null handle; an empty range is valid
+        err = CUDA_ERROR_INVALID_VALUE;
+        return {};
+    }
+    auto box = std::shared_ptr<VmmDevicePtrBox>(
+        new VmmDevicePtrBox{{base, DeallocationStream{}}, range},
+        [](VmmDevicePtrBox* b) {
+            GILReleaseGuard gil;
+            // Order the release on this buffer's recorded stream, then free
+            // the box, which drops the range: every mapping this buffer was
+            // the last to hold unmaps, and its reservation and allocation
+            // follow.
+            vmm_sync_before_release(b->deallocation);
+            delete b;
+        }
+    );
+    return DevicePtrHandle(box, &box->resource);
+}
+
+VmmRangeHandle vmm_range(const DevicePtrHandle& h) noexcept {
+    // Only for handles from deviceptr_create_vmm; the VirtualMemoryBuffer
+    // class guarantees that for its callers.
+    return h ? static_cast<VmmDevicePtrBox*>(get_box(h))->range : VmmRangeHandle{};
+}
+
 // ============================================================================
 // MemoryResource-owned Device Pointer Handles
 // ============================================================================
@@ -325,6 +365,10 @@ DevicePtrHandle deviceptr_create_with_mr(CUdeviceptr ptr, size_t size, PyObject*
         [mr, size](DevicePtrBox* b) {
             GILAcquireGuard gil;
             if (gil.acquired()) {
+                // The last reference may go while an exception propagates
+                // through the releasing caller; deallocate() must run with a
+                // clean error state and leave that exception in place.
+                PendingExceptionGuard pending;
                 if (mr_dealloc_cb) {
                     const DeallocationStream& stream = b->deallocation;
                     cleanup_in_context(
 
@@ -94,6 +94,45 @@ class GILAcquireGuard {
     bool acquired_;
 };
 
+// Save the Python exception in flight and restore it on scope exit, dropping
+// anything the scope itself raised. A deleter may run while an exception is
+// propagating through the caller that released the last reference; Python
+// code it calls (a MemoryResource's deallocate, the warnings machinery) must
+// start with a clean error state and must not leave that exception cleared,
+// or the caller returns an error with no exception set. Allocates nothing.
+// Construct with the GIL held and destroy before releasing it.
+class PendingExceptionGuard {
+public:
+    PendingExceptionGuard() noexcept {
+#if PY_VERSION_HEX >= 0x030C0000
+        pending_ = PyErr_GetRaisedException();
+#else
+        PyErr_Fetch(&type_, &value_, &tb_);
+#endif
+    }
+
+    ~PendingExceptionGuard() {
+        PyErr_Clear();
+#if PY_VERSION_HEX >= 0x030C0000
+        PyErr_SetRaisedException(pending_);
+#else
+        PyErr_Restore(type_, value_, tb_);
+#endif
+    }
+
+    PendingExceptionGuard(const PendingExceptionGuard&) = delete;
+    PendingExceptionGuard& operator=(const PendingExceptionGuard&) = delete;
+
+private:
+#if PY_VERSION_HEX >= 0x030C0000
+    PyObject* pending_ = nullptr;
+#else
+    PyObject* type_ = nullptr;
+    PyObject* value_ = nullptr;
+    PyObject* tb_ = nullptr;
+#endif
+};
+
 // as_py() - convert handle to Python wrapper object (returns new reference)
 namespace detail {
 // n.b. class lookup is not cached to avoid deadlock hazard, see DESIGN.md
@@ -205,6 +244,18 @@ inline PyObject* as_py(const SurfObjectHandle& h) noexcept {
     return detail::make_py("cuda.bindings.driver", "CUsurfObject", as_intptr(h));
 }
 
+inline PyObject* as_py(const MemAllocationHandle& h) noexcept {
+    return detail::make_py("cuda.bindings.driver", "CUmemGenericAllocationHandle", as_intptr(h));
+}
+
+inline PyObject* as_py(const VaReservationHandle& h) noexcept {
+    return detail::make_py("cuda.bindings.driver", "CUdeviceptr", as_intptr(h));
+}
+
+inline PyObject* as_py(const VaMappingHandle& h) noexcept {
+    return detail::make_py("cuda.bindings.driver", "CUdeviceptr", as_intptr(h));
+}
+
 // ============================================================================
 // Python-coupled API: the prototypes that take or return PyObject*
 // ============================================================================
 
@@ -48,39 +48,6 @@ std::atomic<bool> unavailable_reported[kTables];
 std::mutex fill_mutex;
 char fill_error[kTables][512] = {};  // guarded by fill_mutex
 
-// Saves the pending Python exception on construction and restores it on
-// destruction. The Python calls in between start from a clean error state, and
-// the caller's exception survives. Requires the GIL.
-class PendingExceptionGuard {
-public:
-    PendingExceptionGuard() noexcept {
-#if PY_VERSION_HEX >= 0x030C0000
-        exc_ = PyErr_GetRaisedException();
-#else
-        PyErr_Fetch(&type_, &value_, &traceback_);
-#endif
-    }
-    ~PendingExceptionGuard() {
-        PyErr_Clear();  // drop anything the guarded calls left set
-#if PY_VERSION_HEX >= 0x030C0000
-        PyErr_SetRaisedException(exc_);
-#else
-        PyErr_Restore(type_, value_, traceback_);
-#endif
-    }
-    PendingExceptionGuard(const PendingExceptionGuard&) = delete;
-    PendingExceptionGuard& operator=(const PendingExceptionGuard&) = delete;
-
-private:
-#if PY_VERSION_HEX >= 0x030C0000
-    PyObject* exc_ = nullptr;
-#else
-    PyObject* type_ = nullptr;
-    PyObject* value_ = nullptr;
-    PyObject* traceback_ = nullptr;
-#endif
-};
-
 std::size_t index_of(FnTable table) noexcept { return static_cast<std::size_t>(table); }
 
 const char* module_name(FnTable table) noexcept {
 
@@ -34,12 +34,7 @@ void report_message(const char* message) noexcept {
         GILAcquireGuard gil;
         if (gil.acquired()) {
             // Deleters can run while a Python exception is propagating; keep it.
-#if PY_VERSION_HEX >= 0x030C0000
-            PyObject* pending = PyErr_GetRaisedException();
-#else
-            PyObject *pending_type, *pending_value, *pending_tb;
-            PyErr_Fetch(&pending_type, &pending_value, &pending_tb);
-#endif
+            PendingExceptionGuard pending;
             bool interrupted = false;
             if (PyErr_WarnEx(category, message, 1) != 0) {
                 interrupted = PyErr_ExceptionMatches(PyExc_KeyboardInterrupt);
@@ -52,11 +47,6 @@ void report_message(const char* message) noexcept {
                     Py_XDECREF(subject);
                 }
             }
-#if PY_VERSION_HEX >= 0x030C0000
-            PyErr_SetRaisedException(pending);
-#else
-            PyErr_Restore(pending_type, pending_value, pending_tb);
-#endif
             if (interrupted) {
                 PyErr_SetInterrupt();
             }