Merge pull request #70315 from RandomShaper/d3d12_mesa

Direct3D 12 Rendering Driver (Mesa NIR approach)

.github/workflows/windows_builds.yml

@@ -7,7 +7,7 @@ on:
 env:
   # Used for the cache key. Add version suffix to force clean build.
   GODOT_BASE_BRANCH: master
-  SCONSFLAGS: verbose=yes warnings=extra werror=yes module_text_server_fb_enabled=yes
+  SCONSFLAGS: verbose=yes warnings=extra werror=yes module_text_server_fb_enabled=yes d3d12=yes "dxc_path=${{github.workspace}}/dxc" "mesa_libs=${{github.workspace}}/godot-nir-static"
   SCONS_CACHE_MSVC_CONFIG: true
 
 concurrency:
@@ -49,6 +49,28 @@ jobs:
       - name: Setup python and scons
         uses: ./.github/actions/godot-deps
 
+      - name: Download pre-built DirectX Shader Compiler
+        uses: dsaltares/fetch-gh-release-asset@1.0.0
+        with:
+          repo: Microsoft/DirectXShaderCompiler
+          version: tags/v1.7.2308
+          file: dxc_2023_08_14.zip
+          target: dxc.zip
+
+      - name: Extract pre-built DirectX Shader Compiler
+        run: 7z x dxc.zip -o${{github.workspace}}/dxc
+
+      - name: Download pre-built Mesa-NIR
+        uses: dsaltares/fetch-gh-release-asset@1.0.0
+        with:
+          repo: godotengine/godot-nir-static
+          version: tags/23.1.0-devel
+          file: godot-nir-23.1.0-1-devel.zip
+          target: godot-nir-static.zip
+
+      - name: Extract pre-built Mesa-NIR
+        run: 7z x godot-nir-static.zip -o${{github.workspace}}/godot-nir-static
+
       - name: Setup MSVC problem matcher
         uses: ammaraskar/msvc-problem-matcher@master
 

COPYRIGHT.txt

@@ -194,6 +194,16 @@ Copyright: 2018, Eric Lasota
  2018, Microsoft Corp.
 License: Expat
 
+Files: ./thirdparty/d3d12ma/
+Comment: D3D12 Memory Allocator
+Copyright: 2019-2022 Advanced Micro Devices, Inc.
+License: Expat
+
+Files: ./thirdparty/directx_headers/
+Comment: DirectX Headers
+Copyright: Microsoft Corporation
+License: Expat
+
 Files: ./thirdparty/doctest/
 Comment: doctest
 Copyright: 2016-2023, Viktor Kirilov

SConstruct

@@ -191,6 +191,7 @@ opts.Add(BoolVariable("brotli", "Enable Brotli for decompresson and WOFF2 fonts
 opts.Add(BoolVariable("xaudio2", "Enable the XAudio2 audio driver", False))
 opts.Add(BoolVariable("vulkan", "Enable the vulkan rendering driver", True))
 opts.Add(BoolVariable("opengl3", "Enable the OpenGL/GLES3 rendering driver", True))
+opts.Add(BoolVariable("d3d12", "Enable the Direct3D 12 rendering driver (Windows only)", False))
 opts.Add(BoolVariable("openxr", "Enable the OpenXR driver", True))
 opts.Add(BoolVariable("use_volk", "Use the volk library to load the Vulkan loader dynamically", True))
 opts.Add(BoolVariable("disable_exceptions", "Force disabling exception handling code", True))

core/config/project_settings.cpp

@@ -95,7 +95,7 @@ const PackedStringArray ProjectSettings::_get_supported_features() {
 	features.append(VERSION_FULL_CONFIG);
 	features.append(VERSION_FULL_BUILD);
 
-#ifdef VULKAN_ENABLED
+#if defined(VULKAN_ENABLED) || defined(D3D12_ENABLED)
 	features.append("Forward Plus");
 	features.append("Mobile");
 #endif
@@ -1399,6 +1399,12 @@ ProjectSettings::ProjectSettings() {
 	GLOBAL_DEF("rendering/rendering_device/staging_buffer/texture_upload_region_size_px", 64);
 	GLOBAL_DEF("rendering/rendering_device/pipeline_cache/save_chunk_size_mb", 3.0);
 	GLOBAL_DEF("rendering/rendering_device/vulkan/max_descriptors_per_pool", 64);
+	GLOBAL_DEF_RST("rendering/rendering_device/d3d12/max_resource_descriptors_per_frame", 16384);
+	custom_prop_info["rendering/rendering_device/d3d12/max_resource_descriptors_per_frame"] = PropertyInfo(Variant::INT, "rendering/rendering_device/d3d12/max_resource_descriptors_per_frame", PROPERTY_HINT_RANGE, "512,262144");
+	GLOBAL_DEF_RST("rendering/rendering_device/d3d12/max_sampler_descriptors_per_frame", 1024);
+	custom_prop_info["rendering/rendering_device/d3d12/max_sampler_descriptors_per_frame"] = PropertyInfo(Variant::INT, "rendering/rendering_device/d3d12/max_sampler_descriptors_per_frame", PROPERTY_HINT_RANGE, "256,2048");
+	GLOBAL_DEF_RST("rendering/rendering_device/d3d12/max_misc_descriptors_per_frame", 512);
+	custom_prop_info["rendering/rendering_device/d3d12/max_misc_descriptors_per_frame"] = PropertyInfo(Variant::INT, "rendering/rendering_device/d3d12/max_misc_descriptors_per_frame", PROPERTY_HINT_RANGE, "32,4096");
 
 	GLOBAL_DEF_BASIC(PropertyInfo(Variant::INT, "rendering/textures/canvas_textures/default_texture_filter", PROPERTY_HINT_ENUM, "Nearest,Linear,Linear Mipmap,Nearest Mipmap"), 1);
 	GLOBAL_DEF_BASIC(PropertyInfo(Variant::INT, "rendering/textures/canvas_textures/default_texture_repeat", PROPERTY_HINT_ENUM, "Disable,Enable,Mirror"), 0);

core/core_bind.cpp

@@ -662,6 +662,7 @@ void OS::_bind_methods() {
 
 	BIND_ENUM_CONSTANT(RENDERING_DRIVER_VULKAN);
 	BIND_ENUM_CONSTANT(RENDERING_DRIVER_OPENGL3);
+	BIND_ENUM_CONSTANT(RENDERING_DRIVER_D3D12);
 
 	BIND_ENUM_CONSTANT(SYSTEM_DIR_DESKTOP);
 	BIND_ENUM_CONSTANT(SYSTEM_DIR_DCIM);

core/core_bind.h

@@ -129,6 +129,7 @@ class OS : public Object {
 	enum RenderingDriver {
 		RENDERING_DRIVER_VULKAN,
 		RENDERING_DRIVER_OPENGL3,
+		RENDERING_DRIVER_D3D12,
 	};
 
 	virtual PackedStringArray get_connected_midi_inputs();

core/object/worker_thread_pool.cpp

@@ -535,6 +535,11 @@ void WorkerThreadPool::wait_for_group_task_completion(GroupID p_group) {
 	task_mutex.unlock();
 }
 
+int WorkerThreadPool::get_thread_index() {
+	Thread::ID tid = Thread::get_caller_id();
+	return singleton->thread_ids.has(tid) ? singleton->thread_ids[tid] : -1;
+}
+
 void WorkerThreadPool::init(int p_thread_count, bool p_use_native_threads_low_priority, float p_low_priority_task_ratio) {
 	ERR_FAIL_COND(threads.size() > 0);
 	if (p_thread_count < 0) {

core/object/worker_thread_pool.h

@@ -197,6 +197,8 @@ class WorkerThreadPool : public Object {
 	_FORCE_INLINE_ int get_thread_count() const { return threads.size(); }
 
 	static WorkerThreadPool *get_singleton() { return singleton; }
+	static int get_thread_index();
+
 	void init(int p_thread_count = -1, bool p_use_native_threads_low_priority = true, float p_low_priority_task_ratio = 0.3);
 	void finish();
 	WorkerThreadPool();

core/string/ustring.cpp

@@ -4842,6 +4842,7 @@ String String::sprintf(const Array &values, bool *error) const {
 	bool pad_with_zeros = false;
 	bool left_justified = false;
 	bool show_sign = false;
+	bool as_unsigned = false;
 
 	if (error) {
 		*error = true;
@@ -4882,16 +4883,27 @@ String String::sprintf(const Array &values, bool *error) const {
 						case 'x':
 							break;
 						case 'X':
-							base = 16;
 							capitalize = true;
 							break;
 					}
 					// Get basic number.
-					String str = String::num_int64(ABS(value), base, capitalize);
+					String str;
+					if (!as_unsigned) {
+						str = String::num_int64(ABS(value), base, capitalize);
+					} else {
+						uint64_t uvalue = *((uint64_t *)&value);
+						// In unsigned hex, if the value fits in 32 bits, trim it down to that.
+						if (base == 16 && value < 0 && value >= INT32_MIN) {
+							uvalue &= 0xffffffff;
+						}
+						str = String::num_uint64(uvalue, base, capitalize);
+					}
 					int number_len = str.length();
 
+					bool negative = value < 0 && !as_unsigned;
+
 					// Padding.
-					int pad_chars_count = (value < 0 || show_sign) ? min_chars - 1 : min_chars;
+					int pad_chars_count = (negative || show_sign) ? min_chars - 1 : min_chars;
 					String pad_char = pad_with_zeros ? String("0") : String(" ");
 					if (left_justified) {
 						str = str.rpad(pad_chars_count, pad_char);
@@ -4900,8 +4912,8 @@ String String::sprintf(const Array &values, bool *error) const {
 					}
 
 					// Sign.
-					if (show_sign || value < 0) {
-						String sign_char = value < 0 ? "-" : "+";
+					if (show_sign || negative) {
+						String sign_char = negative ? "-" : "+";
 						if (left_justified) {
 							str = str.insert(0, sign_char);
 						} else {
@@ -5094,6 +5106,10 @@ String String::sprintf(const Array &values, bool *error) const {
 					show_sign = true;
 					break;
 				}
+				case 'u': { // Treat as unsigned (for int/hex).
+					as_unsigned = true;
+					break;
+				}
 				case '0':
 				case '1':
 				case '2':

doc/classes/OS.xml

@@ -699,6 +699,9 @@
 		<constant name="RENDERING_DRIVER_OPENGL3" value="1" enum="RenderingDriver">
 			The OpenGL 3 rendering driver. It uses OpenGL 3.3 Core Profile on desktop platforms, OpenGL ES 3.0 on mobile devices, and WebGL 2.0 on Web.
 		</constant>
+		<constant name="RENDERING_DRIVER_D3D12" value="2" enum="RenderingDriver">
+			The Direct3D 12 rendering driver.
+		</constant>
 		<constant name="SYSTEM_DIR_DESKTOP" value="0" enum="SystemDir">
 			Desktop directory path.
 		</constant>

doc/classes/ProjectSettings.xml

@@ -2616,6 +2616,18 @@
 		<member name="rendering/renderer/rendering_method.web" type="String" setter="" getter="" default="&quot;gl_compatibility&quot;">
 			Override for [member rendering/renderer/rendering_method] on web.
 		</member>
+		<member name="rendering/rendering_device/d3d12/max_misc_descriptors_per_frame" type="int" setter="" getter="" default="512">
+			The number of entries in the miscellaneous descriptors heap the Direct3D 12 rendering driver uses each frame, used for various operations like clearing a texture.
+			Depending on the complexity of scenes, this value may be lowered or may need to be raised.
+		</member>
+		<member name="rendering/rendering_device/d3d12/max_resource_descriptors_per_frame" type="int" setter="" getter="" default="16384">
+			The number of entries in the resource descriptors heap the Direct3D 12 rendering driver uses each frame, used for most rendering operations.
+			Depending on the complexity of scenes, this value may be lowered or may need to be raised.
+		</member>
+		<member name="rendering/rendering_device/d3d12/max_sampler_descriptors_per_frame" type="int" setter="" getter="" default="1024">
+			The number of entries in the sampler descriptors heap the Direct3D 12 rendering driver uses each frame, used for most rendering operations.
+			Depending on the complexity of scenes, this value may be lowered or may need to be raised.
+		</member>
 		<member name="rendering/rendering_device/driver" type="String" setter="" getter="">
 			Sets the driver to be used by the renderer when using a RenderingDevice-based renderer like the clustered renderer or the mobile renderer. This property can not be edited directly, instead, set the driver using the platform-specific overrides.
 		</member>

drivers/SCsub

@@ -25,6 +25,8 @@ SConscript("winmidi/SCsub")
 # Graphics drivers
 if env["vulkan"]:
     SConscript("vulkan/SCsub")
+if env["d3d12"]:
+    SConscript("d3d12/SCsub")
 if env["opengl3"]:
     SConscript("gl_context/SCsub")
     SConscript("gles3/SCsub")

drivers/d3d12/SCsub

@@ -0,0 +1,151 @@
+#!/usr/bin/env python
+
+import os
+from pathlib import Path
+
+Import("env")
+
+env_d3d12_rd = env.Clone()
+
+thirdparty_obj = []
+
+
+# DirectX Headers (must take precedence over Windows SDK's).
+
+env.Prepend(CPPPATH=["#thirdparty/directx_headers"])
+env_d3d12_rd.Prepend(CPPPATH=["#thirdparty/directx_headers"])
+
+
+# Direct3D 12 Memory Allocator.
+
+env.Append(CPPPATH=["#thirdparty/d3d12ma"])
+env_d3d12_rd.Append(CPPPATH=["#thirdparty/d3d12ma"])
+
+
+# Agility SDK.
+
+if env["agility_sdk_path"] != "":
+    env_d3d12_rd.Append(CPPDEFINES=["AGILITY_SDK_ENABLED"])
+    if env["agility_sdk_multiarch"]:
+        env_d3d12_rd.Append(CPPDEFINES=["AGILITY_SDK_MULTIARCH_ENABLED"])
+
+
+# PIX.
+
+if env["pix_path"] != "":
+    env_d3d12_rd.Append(CPPDEFINES=["PIX_ENABLED"])
+    env_d3d12_rd.Append(CPPPATH=[env["pix_path"] + "/Include"])
+
+
+# Mesa (SPIR-V to DXIL functionality).
+
+mesa_dir = (env["mesa_libs"] + "/godot-mesa").replace("\\", "/")
+mesa_gen_dir = (env["mesa_libs"] + "/godot-mesa/generated").replace("\\", "/")
+mesa_absdir = Dir(mesa_dir).abspath
+mesa_gen_absdir = Dir(mesa_dir + "/generated").abspath
+
+custom_build_steps = [
+    [
+        "src/compiler",
+        "glsl/ir_expression_operation.py enum > %s/ir_expression_operation.h",
+        "ir_expression_operation.h",
+    ],
+    ["src/compiler/nir", "nir_builder_opcodes_h.py > %s/nir_builder_opcodes.h", "nir_builder_opcodes.h"],
+    ["src/compiler/nir", "nir_constant_expressions.py > %s/nir_constant_expressions.c", "nir_constant_expressions.c"],
+    ["src/compiler/nir", "nir_intrinsics_h.py --outdir %s", "nir_intrinsics.h"],
+    ["src/compiler/nir", "nir_intrinsics_c.py --outdir %s", "nir_intrinsics.c"],
+    ["src/compiler/nir", "nir_intrinsics_indices_h.py --outdir %s", "nir_intrinsics_indices.h"],
+    ["src/compiler/nir", "nir_opcodes_h.py > %s/nir_opcodes.h", "nir_opcodes.h"],
+    ["src/compiler/nir", "nir_opcodes_c.py > %s/nir_opcodes.c", "nir_opcodes.c"],
+    ["src/compiler/nir", "nir_opt_algebraic.py > %s/nir_opt_algebraic.c", "nir_opt_algebraic.c"],
+    ["src/compiler/spirv", "vtn_generator_ids_h.py spir-v.xml %s/vtn_generator_ids.h", "vtn_generator_ids.h"],
+    [
+        "src/microsoft/compiler",
+        "dxil_nir_algebraic.py -p ../../../src/compiler/nir > %s/dxil_nir_algebraic.c",
+        "dxil_nir_algebraic.c",
+    ],
+    ["src/util", "format_srgb.py > %s/format_srgb.c", "format_srgb.c"],
+    ["src/util/format", "u_format_table.py u_format.csv --header > %s/u_format_pack.h", "u_format_pack.h"],
+    ["src/util/format", "u_format_table.py u_format.csv > %s/u_format_table.c", "u_format_table.c"],
+]
+
+mesa_gen_include_paths = [mesa_gen_dir + "/src"]
+
+# See update_mesa.sh for explanation.
+for v in custom_build_steps:
+    subdir = v[0]
+    cmd = v[1]
+    gen_filename = v[2]
+
+    in_dir = str(Path(mesa_absdir + "/" + subdir))
+    out_full_path = mesa_dir + "/generated/" + subdir
+    out_file_full_path = out_full_path + "/" + gen_filename
+
+    if gen_filename.endswith(".h"):
+        mesa_gen_include_paths += [out_full_path.replace("\\", "/")]
+
+mesa_private_inc_paths = [v[0] for v in os.walk(mesa_absdir)]
+mesa_private_inc_paths = [v.replace(mesa_absdir, mesa_dir) for v in mesa_private_inc_paths]
+mesa_private_inc_paths = [v.replace("\\", "/") for v in mesa_private_inc_paths]
+# Avoid build results depending on if generated files already exist.
+mesa_private_inc_paths = [v for v in mesa_private_inc_paths if not v.startswith(mesa_gen_dir)]
+mesa_private_inc_paths.sort()
+# Include the list of the generated ones now, so out-of-the-box sources can include generated headers.
+mesa_private_inc_paths += mesa_gen_include_paths
+# We have to blacklist some because we are bindly adding every Mesa directory
+# to the include path and in some cases that causes the wrong header to be included.
+mesa_blacklist_inc_paths = [
+    "src/c11",
+]
+mesa_blacklist_inc_paths = [mesa_dir + "/" + v for v in mesa_blacklist_inc_paths]
+mesa_private_inc_paths = [v for v in mesa_private_inc_paths if v not in mesa_blacklist_inc_paths]
+
+# Added by ourselves.
+extra_defines = [
+    "WINDOWS_NO_FUTEX",
+]
+
+# These defines are inspired by the Meson build scripts in the original repo.
+extra_defines += [
+    "__STDC_CONSTANT_MACROS",
+    "__STDC_FORMAT_MACROS",
+    "__STDC_LIMIT_MACROS",
+    ("PACKAGE_VERSION", '\\"' + Path(mesa_absdir + "/VERSION").read_text().strip() + '\\"'),
+    ("PACKAGE_BUGREPORT", '\\"https://gitlab.freedesktop.org/mesa/mesa/-/issues\\"'),
+    "PIPE_SUBSYSTEM_WINDOWS_USER",
+    ("_Static_assert", "static_assert"),
+]
+
+if env.msvc:
+    extra_defines += [
+        "_USE_MATH_DEFINES",
+        "VC_EXTRALEAN",
+        "_CRT_SECURE_NO_WARNINGS",
+        "_CRT_SECURE_NO_DEPRECATE",
+        "_SCL_SECURE_NO_WARNINGS",
+        "_SCL_SECURE_NO_DEPRECATE",
+        "_ALLOW_KEYWORD_MACROS",
+        ("_HAS_EXCEPTIONS", 0),
+        "NOMINMAX",
+        "HAVE_STRUCT_TIMESPEC",
+    ]
+
+# This is needed since rendering_device_d3d12.cpp needs to include some Mesa internals.
+env_d3d12_rd.Prepend(CPPPATH=mesa_private_inc_paths)
+# For the same reason as above, the defines must be the same as in the 3rd-party code itself.
+env_d3d12_rd.Append(CPPDEFINES=extra_defines)
+
+
+# Add all.
+
+env.drivers_sources += thirdparty_obj
+
+
+# Godot source files.
+
+driver_obj = []
+env_d3d12_rd.add_source_files(driver_obj, "*.cpp")
+env.drivers_sources += driver_obj
+
+# Needed to force rebuilding the driver files when the thirdparty code is updated.
+env.Depends(driver_obj, thirdparty_obj)