Merge remote-tracking branch 'b/main'

build_files/cmake/platform/platform_win32.cmake

@@ -58,6 +58,22 @@ else()
   endif()
 endif()
 
+set(WINDOWS_ARM64_MIN_VSCMD_VER 17.12.3)
+# We have a minimum version of VSCMD for ARM64 (ie, the version the libs were compiled against)
+# This checks for the version on initial run, and caches it, so users do not have to run the VS CMD window every time
+if(CMAKE_SYSTEM_PROCESSOR STREQUAL "ARM64")
+  set(VC_VSCMD_VER $ENV{VSCMD_VER} CACHE STRING "Version of the VSCMD initially run from")
+  mark_as_advanced(VC_VSCMD_VER)
+  set(VSCMD_VER ${VC_VSCMD_VER})
+  if(DEFINED VSCMD_VER)
+    if(VSCMD_VER VERSION_LESS WINDOWS_ARM64_MIN_VSCMD_VER)
+      message(FATAL_ERROR "Windows ARM64 requires VS2022 version ${WINDOWS_ARM64_MIN_VSCMD_VER} or greater - please update your VS2022 install!")
+    endif()
+  else()
+    message(FATAL_ERROR "Unable to detect the Visual Studio CMD version, try running from the visual studio developer prompt.")
+  endif()
+endif()
+
 if(WITH_BLENDER AND NOT WITH_PYTHON_MODULE)
   set_property(DIRECTORY PROPERTY VS_STARTUP_PROJECT blender)
 endif()

intern/audaspace/intern/AUD_PyInit.cpp

@@ -12,9 +12,7 @@
 #include <python/PyAPI.h>
 #include <python/PySound.h>
 
-extern "C" {
 extern void *BKE_sound_get_factory(void *sound);
-}
 
 static PyObject *AUD_getSoundFromPointer(PyObject *self, PyObject *args)
 {

intern/cycles/device/cuda/device_impl.cpp

@@ -537,7 +537,7 @@ void CUDADevice::free_device(void *device_pointer)
   cuda_assert(cuMemFree((CUdeviceptr)device_pointer));
 }
 
-bool CUDADevice::alloc_host(void *&shared_pointer, const size_t size)
+bool CUDADevice::shared_alloc(void *&shared_pointer, const size_t size)
 {
   CUDAContextScope scope(this);
 
@@ -546,14 +546,14 @@ bool CUDADevice::alloc_host(void *&shared_pointer, const size_t size)
   return mem_alloc_result == CUDA_SUCCESS;
 }
 
-void CUDADevice::free_host(void *shared_pointer)
+void CUDADevice::shared_free(void *shared_pointer)
 {
   CUDAContextScope scope(this);
 
   cuMemFreeHost(shared_pointer);
 }
 
-void *CUDADevice::transform_host_to_device_pointer(const void *shared_pointer)
+void *CUDADevice::shared_to_device_pointer(const void *shared_pointer)
 {
   CUDAContextScope scope(this);
   void *device_pointer = nullptr;
@@ -646,7 +646,7 @@ void CUDADevice::mem_zero(device_memory &mem)
     return;
   }
 
-  if (!(mem.is_host_mapped(this) && mem.host_pointer == mem.shared_pointer)) {
+  if (!(mem.is_shared(this) && mem.host_pointer == mem.shared_pointer)) {
     const CUDAContextScope scope(this);
     cuda_assert(cuMemsetD8((CUdeviceptr)mem.device_pointer, 0, mem.memory_size()));
   }

intern/cycles/device/cuda/device_impl.h

@@ -59,42 +59,43 @@ class CUDADevice : public GPUDevice {
                         bool force_ptx = false);
 
   bool load_kernels(const uint kernel_features) override;
-
   void reserve_local_memory(const uint kernel_features);
 
-  void get_device_memory_info(size_t &total, size_t &free) override;
-  bool alloc_device(void *&device_pointer, const size_t size) override;
-  void free_device(void *device_pointer) override;
-  bool alloc_host(void *&shared_pointer, const size_t size) override;
-  void free_host(void *shared_pointer) override;
-  void *transform_host_to_device_pointer(const void *shared_pointer) override;
-  void copy_host_to_device(void *device_pointer, void *host_pointer, const size_t size) override;
-
+  /* All memory types. */
   void mem_alloc(device_memory &mem) override;
-
   void mem_copy_to(device_memory &mem) override;
-
   void mem_move_to_host(device_memory &mem) override;
-
   void mem_copy_from(
       device_memory &mem, const size_t y, size_t w, const size_t h, size_t elem) override;
-
   void mem_zero(device_memory &mem) override;
-
   void mem_free(device_memory &mem) override;
 
   device_ptr mem_alloc_sub_ptr(device_memory &mem, const size_t offset, size_t /*size*/) override;
 
-  void const_copy_to(const char *name, void *host, const size_t size) override;
-
+  /* Global memory. */
   void global_alloc(device_memory &mem);
   void global_copy_to(device_memory &mem);
   void global_free(device_memory &mem);
 
+  /* Texture memory. */
   void tex_alloc(device_texture &mem);
   void tex_copy_to(device_texture &mem);
   void tex_free(device_texture &mem);
 
+  /* Device side memory. */
+  void get_device_memory_info(size_t &total, size_t &free) override;
+  bool alloc_device(void *&device_pointer, const size_t size) override;
+  void free_device(void *device_pointer) override;
+
+  /* Shared memory. */
+  bool shared_alloc(void *&shared_pointer, const size_t size) override;
+  void shared_free(void *shared_pointer) override;
+  void *shared_to_device_pointer(const void *shared_pointer) override;
+
+  /* Memory copy. */
+  void copy_host_to_device(void *device_pointer, void *host_pointer, const size_t size) override;
+  void const_copy_to(const char *name, void *host, const size_t size) override;
+
   bool should_use_graphics_interop() override;
 
   unique_ptr<DeviceQueue> gpu_queue_create() override;

intern/cycles/device/device.cpp

@@ -496,6 +496,16 @@ OSLGlobals *Device::get_cpu_osl_memory()
   return nullptr;
 }
 
+void *Device::host_alloc(const MemoryType /*type*/, const size_t size)
+{
+  return util_aligned_malloc(size, MIN_ALIGNMENT_CPU_DATA_TYPES);
+}
+
+void Device::host_free(const MemoryType /*type*/, void *host_pointer, const size_t size)
+{
+  util_aligned_free(host_pointer, size);
+}
+
 GPUDevice::~GPUDevice() noexcept(false) = default;
 
 bool GPUDevice::load_texture_info()
@@ -572,7 +582,7 @@ void GPUDevice::move_textures_to_host(size_t size, const size_t headroom, const
 
       /* Can only move textures allocated on this device (and not those from peer devices).
        * And need to ignore memory that is already on the host. */
-      if (!mem.is_resident(this) || mem.is_host_mapped(this)) {
+      if (!mem.is_resident(this) || mem.is_shared(this)) {
         continue;
       }
 
@@ -677,14 +687,14 @@ GPUDevice::Mem *GPUDevice::generic_alloc(device_memory &mem, const size_t pitch_
     }
     else if (map_host_used + size < map_host_limit) {
       /* Allocate host memory ourselves. */
-      mem_alloc_result = alloc_host(shared_pointer, size);
+      mem_alloc_result = shared_alloc(shared_pointer, size);
 
       assert((mem_alloc_result && shared_pointer != nullptr) ||
              (!mem_alloc_result && shared_pointer == nullptr));
     }
 
     if (mem_alloc_result) {
-      device_pointer = transform_host_to_device_pointer(shared_pointer);
+      device_pointer = shared_to_device_pointer(shared_pointer);
       map_host_used += size;
       status = " in host memory";
     }
@@ -728,7 +738,7 @@ GPUDevice::Mem *GPUDevice::generic_alloc(device_memory &mem, const size_t pitch_
         mem.host_pointer != shared_pointer)
     {
       memcpy(shared_pointer, mem.host_pointer, size);
-      util_aligned_free(mem.host_pointer, mem.memory_size());
+      host_free(mem.type, mem.host_pointer, mem.memory_size());
       mem.host_pointer = shared_pointer;
     }
     mem.shared_pointer = shared_pointer;
@@ -752,7 +762,7 @@ void GPUDevice::generic_free(device_memory &mem)
   DCHECK(device_mem_map.find(&mem) != device_mem_map.end());
 
   /* For host mapped memory, reference counting is used to safely free it. */
-  if (mem.is_host_mapped(this)) {
+  if (mem.is_shared(this)) {
     assert(mem.shared_counter > 0);
     if (--mem.shared_counter == 0) {
       if (mem.host_pointer == mem.shared_pointer) {
@@ -764,7 +774,7 @@ void GPUDevice::generic_free(device_memory &mem)
         mem.host_pointer = mem.host_alloc(size);
         memcpy(mem.host_pointer, mem.shared_pointer, size);
       }
-      free_host(mem.shared_pointer);
+      shared_free(mem.shared_pointer);
       mem.shared_pointer = nullptr;
     }
     map_host_used -= mem.device_size;
@@ -791,17 +801,17 @@ void GPUDevice::generic_copy_to(device_memory &mem)
   /* If not host mapped, the current device only uses device memory allocated by backend
    * device allocation regardless of mem.host_pointer and mem.shared_pointer, and should
    * copy data from mem.host_pointer. */
-  if (!(mem.is_host_mapped(this) && mem.host_pointer == mem.shared_pointer)) {
+  if (!(mem.is_shared(this) && mem.host_pointer == mem.shared_pointer)) {
     copy_host_to_device((void *)mem.device_pointer, mem.host_pointer, mem.memory_size());
   }
 }
 
-bool GPUDevice::is_host_mapped(const void *shared_pointer,
-                               const device_ptr device_pointer,
-                               Device * /*sub_device*/)
+bool GPUDevice::is_shared(const void *shared_pointer,
+                          const device_ptr device_pointer,
+                          Device * /*sub_device*/)
 {
   return (shared_pointer && device_pointer &&
-          (device_ptr)transform_host_to_device_pointer(shared_pointer) == device_pointer);
+          (device_ptr)shared_to_device_pointer(shared_pointer) == device_pointer);
 }
 
 /* DeviceInfo */

intern/cycles/device/device.h

@@ -247,9 +247,9 @@ class Device {
     return false;
   }
 
-  virtual bool is_host_mapped(const void * /*shared_pointer*/,
-                              const device_ptr /*device_pointer*/,
-                              Device * /*sub_device*/)
+  virtual bool is_shared(const void * /*shared_pointer*/,
+                         const device_ptr /*device_pointer*/,
+                         Device * /*sub_device*/)
   {
     return false;
   }
@@ -320,6 +320,9 @@ class Device {
   friend class DeviceServer;
   friend class device_memory;
 
+  virtual void *host_alloc(const MemoryType type, const size_t size);
+  virtual void host_free(const MemoryType type, void *host_pointer, const size_t size);
+
   virtual void mem_alloc(device_memory &mem) = 0;
   virtual void mem_copy_to(device_memory &mem) = 0;
   virtual void mem_move_to_host(device_memory &mem) = 0;
@@ -398,22 +401,21 @@ class GPUDevice : public Device {
   /* total - amount of device memory, free - amount of available device memory */
   virtual void get_device_memory_info(size_t &total, size_t &free) = 0;
 
+  /* Device side memory. */
   virtual bool alloc_device(void *&device_pointer, const size_t size) = 0;
-
   virtual void free_device(void *device_pointer) = 0;
 
-  virtual bool alloc_host(void *&shared_pointer, const size_t size) = 0;
-
-  virtual void free_host(void *shared_pointer) = 0;
-
-  bool is_host_mapped(const void *shared_pointer,
-                      const device_ptr device_pointer,
-                      Device *sub_device) override;
-
+  /* Shared memory. */
+  virtual bool shared_alloc(void *&shared_pointer, const size_t size) = 0;
+  virtual void shared_free(void *shared_pointer) = 0;
+  bool is_shared(const void *shared_pointer,
+                 const device_ptr device_pointer,
+                 Device *sub_device) override;
   /* This function should return device pointer corresponding to shared pointer, which
-   * is host buffer, allocated in `alloc_host`. */
-  virtual void *transform_host_to_device_pointer(const void *shared_pointer) = 0;
+   * is host buffer, allocated in `shared_alloc`. */
+  virtual void *shared_to_device_pointer(const void *shared_pointer) = 0;
 
+  /* Memory copy. */
   virtual void copy_host_to_device(void *device_pointer,
                                    void *host_pointer,
                                    const size_t size) = 0;

intern/cycles/device/hip/device_impl.cpp

@@ -498,7 +498,7 @@ void HIPDevice::free_device(void *device_pointer)
   hip_assert(hipFree((hipDeviceptr_t)device_pointer));
 }
 
-bool HIPDevice::alloc_host(void *&shared_pointer, const size_t size)
+bool HIPDevice::shared_alloc(void *&shared_pointer, const size_t size)
 {
   HIPContextScope scope(this);
 
@@ -508,14 +508,14 @@ bool HIPDevice::alloc_host(void *&shared_pointer, const size_t size)
   return mem_alloc_result == hipSuccess;
 }
 
-void HIPDevice::free_host(void *shared_pointer)
+void HIPDevice::shared_free(void *shared_pointer)
 {
   HIPContextScope scope(this);
 
   hipHostFree(shared_pointer);
 }
 
-void *HIPDevice::transform_host_to_device_pointer(const void *shared_pointer)
+void *HIPDevice::shared_to_device_pointer(const void *shared_pointer)
 {
   HIPContextScope scope(this);
   void *device_pointer = nullptr;
@@ -608,7 +608,7 @@ void HIPDevice::mem_zero(device_memory &mem)
     return;
   }
 
-  if (!(mem.is_host_mapped(this) && mem.host_pointer == mem.shared_pointer)) {
+  if (!(mem.is_shared(this) && mem.host_pointer == mem.shared_pointer)) {
     const HIPContextScope scope(this);
     hip_assert(hipMemsetD8((hipDeviceptr_t)mem.device_pointer, 0, mem.memory_size()));
   }

intern/cycles/device/hip/device_impl.h

@@ -60,39 +60,41 @@ class HIPDevice : public GPUDevice {
   bool load_kernels(const uint kernel_features) override;
   void reserve_local_memory(const uint kernel_features);
 
-  void get_device_memory_info(size_t &total, size_t &free) override;
-  bool alloc_device(void *&device_pointer, const size_t size) override;
-  void free_device(void *device_pointer) override;
-  bool alloc_host(void *&shared_pointer, const size_t size) override;
-  void free_host(void *shared_pointer) override;
-  void *transform_host_to_device_pointer(const void *shared_pointer) override;
-  void copy_host_to_device(void *device_pointer, void *host_pointer, const size_t size) override;
-
+  /* All memory types. */
   void mem_alloc(device_memory &mem) override;
-
   void mem_copy_to(device_memory &mem) override;
-
   void mem_move_to_host(device_memory &mem) override;
-
   void mem_copy_from(
       device_memory &mem, const size_t y, size_t w, const size_t h, size_t elem) override;
-
   void mem_zero(device_memory &mem) override;
-
   void mem_free(device_memory &mem) override;
 
   device_ptr mem_alloc_sub_ptr(device_memory &mem, const size_t offset, size_t /*size*/) override;
 
-  void const_copy_to(const char *name, void *host, const size_t size) override;
-
+  /* Global memory. */
   void global_alloc(device_memory &mem);
   void global_copy_to(device_memory &mem);
   void global_free(device_memory &mem);
 
+  /* Texture memory. */
   void tex_alloc(device_texture &mem);
   void tex_copy_to(device_texture &mem);
   void tex_free(device_texture &mem);
 
+  /* Device side memory. */
+  void get_device_memory_info(size_t &total, size_t &free) override;
+  bool alloc_device(void *&device_pointer, const size_t size) override;
+  void free_device(void *device_pointer) override;
+
+  /* Shared memory. */
+  bool shared_alloc(void *&shared_pointer, const size_t size) override;
+  void shared_free(void *shared_pointer) override;
+  void *shared_to_device_pointer(const void *shared_pointer) override;
+
+  /* Memory copy. */
+  void copy_host_to_device(void *device_pointer, void *host_pointer, const size_t size) override;
+  void const_copy_to(const char *name, void *host, const size_t size) override;
+
   /* Graphics resources interoperability. */
   bool should_use_graphics_interop() override;