Merge pull request #114746 from akien-mga/4.5-cherrypicks-rendering

[4.5] Cherry-picks for the 4.5 branch (future 4.5.2) - 1st batch (rendering)

Author: akien-mga

drivers/d3d12/rendering_device_driver_d3d12.cpp

@@ -1573,12 +1573,6 @@ RDD::TextureID RenderingDeviceDriverD3D12::texture_create(const TextureFormat &p
 	tex_info->view_descs.srv = srv_desc;
 	tex_info->view_descs.uav = uav_desc;
 
-	if (!barrier_capabilities.enhanced_barriers_supported && (p_format.usage_bits & (TEXTURE_USAGE_STORAGE_BIT | TEXTURE_USAGE_COLOR_ATTACHMENT_BIT))) {
-		// Fallback to clear resources when they're first used in a uniform set. Not necessary if enhanced barriers
-		// are supported, as the discard flag will be used instead when transitioning from an undefined layout.
-		textures_pending_clear.add(&tex_info->pending_clear);
-	}
-
 	return TextureID(tex_info);
 }
 
@@ -3055,7 +3049,7 @@ D3D12_UNORDERED_ACCESS_VIEW_DESC RenderingDeviceDriverD3D12::_make_ranged_uav_fo
 		} break;
 		case D3D12_UAV_DIMENSION_TEXTURE3D: {
 			uav_desc.Texture3D.MipSlice = mip;
-			uav_desc.Texture3D.WSize >>= p_mipmap_offset;
+			uav_desc.Texture3D.WSize = MAX(uav_desc.Texture3D.WSize >> p_mipmap_offset, 1U);
 		} break;
 		default:
 			break;
@@ -3643,21 +3637,6 @@ void RenderingDeviceDriverD3D12::command_uniform_set_prepare_for_use(CommandBuff
 		return;
 	}
 
-	// Perform pending blackouts.
-	{
-		SelfList<TextureInfo> *E = textures_pending_clear.first();
-		while (E) {
-			TextureSubresourceRange subresources;
-			subresources.layer_count = E->self()->layers;
-			subresources.mipmap_count = E->self()->mipmaps;
-			command_clear_color_texture(p_cmd_buffer, TextureID(E->self()), TEXTURE_LAYOUT_UNDEFINED, Color(), subresources);
-
-			SelfList<TextureInfo> *next = E->next();
-			E->remove_from_list();
-			E = next;
-		}
-	}
-
 	CommandBufferInfo *cmd_buf_info = (CommandBufferInfo *)p_cmd_buffer.id;
 	const UniformSetInfo *uniform_set_info = (const UniformSetInfo *)p_uniform_set.id;
 	const ShaderInfo *shader_info_in = (const ShaderInfo *)p_shader.id;
@@ -4571,8 +4550,7 @@ void RenderingDeviceDriverD3D12::command_begin_render_pass(CommandBufferID p_cmd
 			p_rect.position.y,
 			p_rect.position.x + p_rect.size.x,
 			p_rect.position.y + p_rect.size.y);
-	cmd_buf_info->render_pass_state.region_is_all = !(
-			cmd_buf_info->render_pass_state.region_rect.left == 0 &&
+	cmd_buf_info->render_pass_state.region_is_all = (cmd_buf_info->render_pass_state.region_rect.left == 0 &&
 			cmd_buf_info->render_pass_state.region_rect.top == 0 &&
 			cmd_buf_info->render_pass_state.region_rect.right == fb_info->size.x &&
 			cmd_buf_info->render_pass_state.region_rect.bottom == fb_info->size.y);
@@ -4616,7 +4594,6 @@ void RenderingDeviceDriverD3D12::command_begin_render_pass(CommandBufferID p_cmd
 			if (pass_info->attachments[i].load_op == ATTACHMENT_LOAD_OP_CLEAR) {
 				clear.aspect.set_flag(TEXTURE_ASPECT_COLOR_BIT);
 				clear.color_attachment = i;
-				tex_info->pending_clear.remove_from_list();
 			}
 		} else if ((tex_info->desc.Flags & D3D12_RESOURCE_FLAG_ALLOW_DEPTH_STENCIL)) {
 			if (pass_info->attachments[i].load_op == ATTACHMENT_LOAD_OP_CLEAR) {
@@ -5799,6 +5776,8 @@ uint64_t RenderingDeviceDriverD3D12::api_trait_get(ApiTrait p_trait) {
 			return true;
 		case API_TRAIT_BUFFERS_REQUIRE_TRANSITIONS:
 			return !barrier_capabilities.enhanced_barriers_supported;
+		case API_TRAIT_TEXTURE_OUTPUTS_REQUIRE_CLEARS:
+			return true;
 		default:
 			return RenderingDeviceDriver::api_trait_get(p_trait);
 	}
@@ -5807,7 +5786,7 @@ uint64_t RenderingDeviceDriverD3D12::api_trait_get(ApiTrait p_trait) {
 bool RenderingDeviceDriverD3D12::has_feature(Features p_feature) {
 	switch (p_feature) {
 		case SUPPORTS_HALF_FLOAT:
-			return shader_capabilities.native_16bit_ops && storage_buffer_capabilities.storage_buffer_16_bit_access_is_supported;
+			return shader_capabilities.native_16bit_ops;
 		case SUPPORTS_FRAGMENT_SHADER_WITH_ONLY_SIDE_EFFECTS:
 			return true;
 		case SUPPORTS_BUFFER_DEVICE_ADDRESS:
@@ -6037,7 +6016,6 @@ Error RenderingDeviceDriverD3D12::_check_capabilities() {
 	subgroup_capabilities.wave_ops_supported = false;
 	shader_capabilities.shader_model = (D3D_SHADER_MODEL)0;
 	shader_capabilities.native_16bit_ops = false;
-	storage_buffer_capabilities.storage_buffer_16_bit_access_is_supported = false;
 	format_capabilities.relaxed_casting_supported = false;
 
 	{
@@ -6078,9 +6056,8 @@ Error RenderingDeviceDriverD3D12::_check_capabilities() {
 
 	D3D12_FEATURE_DATA_D3D12_OPTIONS options = {};
 	res = device->CheckFeatureSupport(D3D12_FEATURE_D3D12_OPTIONS, &options, sizeof(options));
-	if (SUCCEEDED(res)) {
-		storage_buffer_capabilities.storage_buffer_16_bit_access_is_supported = options.TypedUAVLoadAdditionalFormats;
-	}
+	ERR_FAIL_COND_V_MSG(!SUCCEEDED(res), ERR_UNAVAILABLE, "CheckFeatureSupport failed with error " + vformat("0x%08ux", (uint64_t)res) + ".");
+	ERR_FAIL_COND_V_MSG(!options.TypedUAVLoadAdditionalFormats, ERR_UNAVAILABLE, "No support for Typed UAV Load Additional Formats has been found.");
 
 	D3D12_FEATURE_DATA_D3D12_OPTIONS1 options1 = {};
 	res = device->CheckFeatureSupport(D3D12_FEATURE_D3D12_OPTIONS1, &options1, sizeof(options1));

drivers/d3d12/rendering_device_driver_d3d12.h

@@ -114,10 +114,6 @@ class RenderingDeviceDriverD3D12 : public RenderingDeviceDriver {
 		bool native_16bit_ops = false;
 	};
 
-	struct StorageBufferCapabilities {
-		bool storage_buffer_16_bit_access_is_supported = false;
-	};
-
 	struct FormatCapabilities {
 		bool relaxed_casting_supported = false;
 	};
@@ -143,7 +139,6 @@ class RenderingDeviceDriverD3D12 : public RenderingDeviceDriver {
 	FragmentShadingRateCapabilities fsr_capabilities;
 	FragmentDensityMapCapabilities fdm_capabilities;
 	ShaderCapabilities shader_capabilities;
-	StorageBufferCapabilities storage_buffer_capabilities;
 	FormatCapabilities format_capabilities;
 	BarrierCapabilities barrier_capabilities;
 	MiscFeaturesSupport misc_features_support;
@@ -360,12 +355,10 @@ class RenderingDeviceDriverD3D12 : public RenderingDeviceDriver {
 		TextureInfo *main_texture = nullptr;
 
 		UINT mapped_subresource = UINT_MAX;
-		SelfList<TextureInfo> pending_clear{ this };
 #ifdef DEBUG_ENABLED
 		bool created_from_extension = false;
 #endif
 	};
-	SelfList<TextureInfo>::List textures_pending_clear;
 
 	HashMap<DXGI_FORMAT, uint32_t> format_sample_counts_mask_cache;
 	Mutex format_sample_counts_mask_cache_mutex;

drivers/d3d12/rendering_shader_container_d3d12.cpp

@@ -99,28 +99,26 @@ uint32_t RenderingDXIL::patch_specialization_constant(
 		const uint64_t (&p_stages_bit_offsets)[D3D12_BITCODE_OFFSETS_NUM_STAGES],
 		HashMap<RenderingDeviceCommons::ShaderStage, Vector<uint8_t>> &r_stages_bytecodes,
 		bool p_is_first_patch) {
-	uint32_t patch_val = 0;
+	int64_t patch_val = 0;
 	switch (p_type) {
 		case RenderingDeviceCommons::PIPELINE_SPECIALIZATION_CONSTANT_TYPE_INT: {
-			uint32_t int_value = *((const int *)p_value);
-			ERR_FAIL_COND_V(int_value & (1 << 31), 0);
-			patch_val = int_value;
+			patch_val = *((const int32_t *)p_value);
 		} break;
 		case RenderingDeviceCommons::PIPELINE_SPECIALIZATION_CONSTANT_TYPE_BOOL: {
 			bool bool_value = *((const bool *)p_value);
-			patch_val = (uint32_t)bool_value;
+			patch_val = (int32_t)bool_value;
 		} break;
 		case RenderingDeviceCommons::PIPELINE_SPECIALIZATION_CONSTANT_TYPE_FLOAT: {
-			uint32_t int_value = *((const int *)p_value);
-			ERR_FAIL_COND_V(int_value & (1 << 31), 0);
-			patch_val = (int_value >> 1);
+			patch_val = *((const int32_t *)p_value);
 		} break;
 	}
-	// For VBR encoding to encode the number of bits we expect (32), we need to set the MSB unconditionally.
-	// However, signed VBR moves the MSB to the LSB, so setting the MSB to 1 wouldn't help. Therefore,
-	// the bit we set to 1 is the one at index 30.
-	patch_val |= (1 << 30);
-	patch_val <<= 1; // What signed VBR does.
+
+	// Encode to signed VBR.
+	if (patch_val >= 0) {
+		patch_val <<= 1;
+	} else {
+		patch_val = ((-patch_val) << 1) | 1;
+	}
 
 	auto tamper_bits = [](uint8_t *p_start, uint64_t p_bit_offset, uint64_t p_tb_value) -> uint64_t {
 		uint64_t original = 0;
@@ -174,13 +172,13 @@ uint32_t RenderingDXIL::patch_specialization_constant(
 
 		Vector<uint8_t> &bytecode = r_stages_bytecodes[(RenderingDeviceCommons::ShaderStage)stage];
 #ifdef DEV_ENABLED
-		uint64_t orig_patch_val = tamper_bits(bytecode.ptrw(), offset, patch_val);
+		uint64_t orig_patch_val = tamper_bits(bytecode.ptrw(), offset, (uint64_t)patch_val);
 		// Checking against the value the NIR patch should have set.
 		DEV_ASSERT(!p_is_first_patch || ((orig_patch_val >> 1) & GODOT_NIR_SC_SENTINEL_MAGIC_MASK) == GODOT_NIR_SC_SENTINEL_MAGIC);
-		uint64_t readback_patch_val = tamper_bits(bytecode.ptrw(), offset, patch_val);
-		DEV_ASSERT(readback_patch_val == patch_val);
+		uint64_t readback_patch_val = tamper_bits(bytecode.ptrw(), offset, (uint64_t)patch_val);
+		DEV_ASSERT(readback_patch_val == (uint64_t)patch_val);
 #else
-		tamper_bits(bytecode.ptrw(), offset, patch_val);
+		tamper_bits(bytecode.ptrw(), offset, (uint64_t)patch_val);
 #endif
 
 		stages_patched_mask |= (1 << stage);
@@ -319,10 +317,6 @@ bool RenderingShaderContainerD3D12::_convert_spirv_to_nir(const Vector<Rendering
 
 		ERR_FAIL_NULL_V_MSG(shader, false, "Shader translation (step 1) at stage " + String(RenderingDeviceCommons::SHADER_STAGE_NAMES[stage]) + " failed.");
 
-#ifdef DEV_ENABLED
-		nir_validate_shader(shader, "Validate before feeding NIR to the DXIL compiler");
-#endif
-
 		if (stage == RenderingDeviceCommons::SHADER_STAGE_VERTEX) {
 			dxil_runtime_conf.yz_flip.y_mask = 0xffff;
 			dxil_runtime_conf.yz_flip.mode = DXIL_SPIRV_Y_FLIP_UNCONDITIONAL;
@@ -621,9 +615,7 @@ bool RenderingShaderContainerD3D12::_generate_root_signature(BitField<RenderingD
 	D3D12_ROOT_SIGNATURE_FLAGS root_sig_flags =
 			D3D12_ROOT_SIGNATURE_FLAG_DENY_HULL_SHADER_ROOT_ACCESS |
 			D3D12_ROOT_SIGNATURE_FLAG_DENY_DOMAIN_SHADER_ROOT_ACCESS |
-			D3D12_ROOT_SIGNATURE_FLAG_DENY_GEOMETRY_SHADER_ROOT_ACCESS |
-			D3D12_ROOT_SIGNATURE_FLAG_DENY_AMPLIFICATION_SHADER_ROOT_ACCESS |
-			D3D12_ROOT_SIGNATURE_FLAG_DENY_MESH_SHADER_ROOT_ACCESS;
+			D3D12_ROOT_SIGNATURE_FLAG_DENY_GEOMETRY_SHADER_ROOT_ACCESS;
 
 	if (!p_stages_processed.has_flag(RenderingDeviceCommons::SHADER_STAGE_VERTEX_BIT)) {
 		root_sig_flags |= D3D12_ROOT_SIGNATURE_FLAG_DENY_VERTEX_SHADER_ROOT_ACCESS;

drivers/gles3/rasterizer_gles3.cpp

@@ -410,7 +410,7 @@ void RasterizerGLES3::_blit_render_target_to_screen(DisplayServer::WindowID p_sc
 	}
 #endif
 
-	glBindFramebuffer(GL_DRAW_FRAMEBUFFER, GLES3::TextureStorage::system_fbo);
+	glBindFramebuffer(GL_FRAMEBUFFER, GLES3::TextureStorage::system_fbo);
 
 	if (p_first) {
 		if (p_blit.dst_rect.position != Vector2() || p_blit.dst_rect.size != rt->size) {

drivers/gles3/shaders/scene.glsl

@@ -1022,12 +1022,6 @@ uniform highp mat4 world_transform;
 uniform highp uint instance_offset;
 uniform highp uint model_flags;
 
-/* clang-format off */
-
-#GLOBALS
-
-/* clang-format on */
-
 #define LIGHT_BAKE_DISABLED 0u
 #define LIGHT_BAKE_STATIC 1u
 #define LIGHT_BAKE_DYNAMIC 2u
@@ -1268,6 +1262,12 @@ layout(location = 0) out vec4 frag_color;
 
 #endif // !RENDER_MATERIAL
 
+/* clang-format off */
+
+#GLOBALS
+
+/* clang-format on */
+
 vec3 F0(float metallic, float specular, vec3 albedo) {
 	float dielectric = 0.16 * specular * specular;
 	// use albedo * metallic as colored specular reflectance at 0 angle for metallic materials;

drivers/gles3/storage/config.cpp

@@ -232,7 +232,7 @@ Config::Config() {
 			//https://github.com/godotengine/godot/issues/92662#issuecomment-2161199477
 			//disable_particles_workaround = false;
 		}
-	} else if (rendering_device_name == "PowerVR Rogue GE8320") {
+	} else if (rendering_device_name.contains("PowerVR")) {
 		disable_transform_feedback_shader_cache = true;
 	}
 

drivers/gles3/storage/light_storage.cpp

@@ -350,7 +350,14 @@ AABB LightStorage::light_get_aabb(RID p_light) const {
 	switch (light->type) {
 		case RS::LIGHT_SPOT: {
 			float len = light->param[RS::LIGHT_PARAM_RANGE];
-			float size = Math::tan(Math::deg_to_rad(light->param[RS::LIGHT_PARAM_SPOT_ANGLE])) * len;
+			float angle = Math::deg_to_rad(light->param[RS::LIGHT_PARAM_SPOT_ANGLE]);
+
+			if (angle > Math::PI * 0.5) {
+				// Light casts backwards as well.
+				return AABB(Vector3(-1, -1, -1) * len, Vector3(2, 2, 2) * len);
+			}
+
+			float size = Math::sin(angle) * len;
 			return AABB(Vector3(-size, -size, -len), Vector3(size * 2, size * 2, len));
 		};
 		case RS::LIGHT_OMNI: {
@@ -807,6 +814,9 @@ bool LightStorage::reflection_probe_instance_begin_render(RID p_instance, RID p_
 
 	ERR_FAIL_NULL_V(atlas, false);
 
+	ERR_FAIL_COND_V_MSG(atlas->size < 4, false, "Attempted to render to a reflection atlas of invalid resolution.");
+	ERR_FAIL_COND_V_MSG(atlas->count < 1, false, "Attempted to render to a reflection atlas of size < 1.");
+
 	ReflectionProbeInstance *rpi = reflection_probe_instance_owner.get_or_null(p_instance);
 	ERR_FAIL_NULL_V(rpi, false);
 

drivers/gles3/storage/material_storage.cpp

@@ -2265,6 +2265,7 @@ void MaterialStorage::shader_set_code(RID p_shader, const String &p_code) {
 	}
 
 	if (shader->data) {
+		shader->data->set_path_hint(shader->path_hint);
 		shader->data->set_code(p_code);
 	}
 

drivers/gles3/storage/render_scene_buffers_gles3.cpp

@@ -200,6 +200,10 @@ void RenderSceneBuffersGLES3::_check_render_buffers() {
 	uint32_t depth_format_size = 4;
 	bool use_multiview = view_count > 1;
 
+	if (!use_internal_buffer && internal3d.color != 0) {
+		_clear_intermediate_buffers();
+	}
+
 	if ((!use_internal_buffer || internal3d.color != 0) && (msaa3d.mode == RS::VIEWPORT_MSAA_DISABLED || msaa3d.color != 0)) {
 		// already setup!
 		return;