vulkan: Use dynamic vertex buffer strides when dynamic bindings unava…

…ilable. (shadps4-emu#1164)

CMakeLists.txt

@@ -8,7 +8,7 @@ set(CMAKE_CXX_STANDARD_REQUIRED True)
 
 if(APPLE)
     enable_language(OBJC)
-    set(CMAKE_OSX_DEPLOYMENT_TARGET 11)
+    set(CMAKE_OSX_DEPLOYMENT_TARGET 14)
 endif()
 
 if (NOT CMAKE_BUILD_TYPE)

README.md

@@ -72,7 +72,7 @@ Check the build instructions for [**Linux**](https://github.com/shadps4-emu/shad
 Check the build instructions for [**macOS**](https://github.com/shadps4-emu/shadPS4/blob/main/documents/building-macos.md).
 
 > [!IMPORTANT]
-> macOS users need at least macOS 15 on Apple Silicon-based Mac devices and at least macOS 11 on Intel-based Mac devices.
+> macOS users need at least macOS 15 on Apple Silicon-based Mac devices and at least macOS 14 on Intel-based Mac devices.
 
 # Debugging and reporting issues
 

src/video_core/buffer_cache/buffer_cache.cpp

@@ -114,6 +114,8 @@ bool BufferCache::BindVertexBuffers(const Shader::Info& vs_info) {
 
     std::array<vk::Buffer, NumVertexBuffers> host_buffers;
     std::array<vk::DeviceSize, NumVertexBuffers> host_offsets;
+    std::array<vk::DeviceSize, NumVertexBuffers> host_sizes;
+    std::array<vk::DeviceSize, NumVertexBuffers> host_strides;
     boost::container::static_vector<AmdGpu::Buffer, NumVertexBuffers> guest_buffers;
 
     struct BufferRange {
@@ -193,11 +195,18 @@ bool BufferCache::BindVertexBuffers(const Shader::Info& vs_info) {
 
         host_buffers[i] = host_buffer->vk_buffer;
         host_offsets[i] = host_buffer->offset + buffer.base_address - host_buffer->base_address;
+        host_sizes[i] = buffer.GetSize();
+        host_strides[i] = buffer.GetStride();
     }
 
     if (num_buffers > 0) {
         const auto cmdbuf = scheduler.CommandBuffer();
-        cmdbuf.bindVertexBuffers(0, num_buffers, host_buffers.data(), host_offsets.data());
+        if (instance.IsVertexInputDynamicState()) {
+            cmdbuf.bindVertexBuffers(0, num_buffers, host_buffers.data(), host_offsets.data());
+        } else {
+            cmdbuf.bindVertexBuffers2EXT(0, num_buffers, host_buffers.data(), host_offsets.data(),
+                                         host_sizes.data(), host_strides.data());
+        }
     }
 
     return has_step_rate;

src/video_core/renderer_vulkan/vk_graphics_pipeline.cpp

@@ -46,28 +46,34 @@ GraphicsPipeline::GraphicsPipeline(const Instance& instance_, Scheduler& schedul
 
     boost::container::static_vector<vk::VertexInputBindingDescription, 32> vertex_bindings;
     boost::container::static_vector<vk::VertexInputAttributeDescription, 32> vertex_attributes;
-    const auto& vs_info = stages[u32(Shader::Stage::Vertex)];
-    for (const auto& input : vs_info->vs_inputs) {
-        if (input.instance_step_rate == Shader::Info::VsInput::InstanceIdType::OverStepRate0 ||
-            input.instance_step_rate == Shader::Info::VsInput::InstanceIdType::OverStepRate1) {
-            // Skip attribute binding as the data will be pulled by shader
-            continue;
-        }
+    if (!instance.IsVertexInputDynamicState()) {
+        const auto& vs_info = stages[u32(Shader::Stage::Vertex)];
+        for (const auto& input : vs_info->vs_inputs) {
+            if (input.instance_step_rate == Shader::Info::VsInput::InstanceIdType::OverStepRate0 ||
+                input.instance_step_rate == Shader::Info::VsInput::InstanceIdType::OverStepRate1) {
+                // Skip attribute binding as the data will be pulled by shader
+                continue;
+            }
 
-        const auto buffer = vs_info->ReadUd<AmdGpu::Buffer>(input.sgpr_base, input.dword_offset);
-        vertex_attributes.push_back({
-            .location = input.binding,
-            .binding = input.binding,
-            .format = LiverpoolToVK::SurfaceFormat(buffer.GetDataFmt(), buffer.GetNumberFmt()),
-            .offset = 0,
-        });
-        vertex_bindings.push_back({
-            .binding = input.binding,
-            .stride = buffer.GetStride(),
-            .inputRate = input.instance_step_rate == Shader::Info::VsInput::None
-                             ? vk::VertexInputRate::eVertex
-                             : vk::VertexInputRate::eInstance,
-        });
+            const auto buffer =
+                vs_info->ReadUd<AmdGpu::Buffer>(input.sgpr_base, input.dword_offset);
+            if (buffer.GetSize() == 0) {
+                continue;
+            }
+            vertex_attributes.push_back({
+                .location = input.binding,
+                .binding = input.binding,
+                .format = LiverpoolToVK::SurfaceFormat(buffer.GetDataFmt(), buffer.GetNumberFmt()),
+                .offset = 0,
+            });
+            vertex_bindings.push_back({
+                .binding = input.binding,
+                .stride = buffer.GetStride(),
+                .inputRate = input.instance_step_rate == Shader::Info::VsInput::None
+                                 ? vk::VertexInputRate::eVertex
+                                 : vk::VertexInputRate::eInstance,
+            });
+        }
     }
 
     const vk::PipelineVertexInputStateCreateInfo vertex_input_info = {
@@ -147,6 +153,8 @@ GraphicsPipeline::GraphicsPipeline(const Instance& instance_, Scheduler& schedul
     }
     if (instance.IsVertexInputDynamicState()) {
         dynamic_states.push_back(vk::DynamicState::eVertexInputEXT);
+    } else {
+        dynamic_states.push_back(vk::DynamicState::eVertexInputBindingStrideEXT);
     }
 
     const vk::PipelineDynamicStateCreateInfo dynamic_info = {
@@ -273,7 +281,7 @@ GraphicsPipeline::GraphicsPipeline(const Instance& instance_, Scheduler& schedul
         .pNext = &pipeline_rendering_ci,
         .stageCount = static_cast<u32>(shader_stages.size()),
         .pStages = shader_stages.data(),
-        .pVertexInputState = &vertex_input_info,
+        .pVertexInputState = !instance.IsVertexInputDynamicState() ? &vertex_input_info : nullptr,
         .pInputAssemblyState = &input_assembly,
         .pViewportState = &viewport_info,
         .pRasterizationState = &raster_state,

src/video_core/renderer_vulkan/vk_graphics_pipeline.h

@@ -45,6 +45,7 @@ struct GraphicsPipelineKey {
     Liverpool::ColorBufferMask cb_shader_mask;
     std::array<Liverpool::BlendControl, Liverpool::NumColorBuffers> blend_controls;
     std::array<vk::ColorComponentFlags, Liverpool::NumColorBuffers> write_masks;
+    std::array<vk::Format, MaxVertexBufferCount> vertex_buffer_formats;
 
     bool operator==(const GraphicsPipelineKey& key) const noexcept {
         return std::memcmp(this, &key, sizeof(key)) == 0;

src/video_core/renderer_vulkan/vk_instance.cpp

@@ -272,6 +272,7 @@ bool Instance::CreateDevice() {
     add_extension(VK_KHR_DYNAMIC_RENDERING_EXTENSION_NAME);
     add_extension(VK_EXT_SHADER_DEMOTE_TO_HELPER_INVOCATION_EXTENSION_NAME);
     add_extension(VK_KHR_SYNCHRONIZATION_2_EXTENSION_NAME);
+    add_extension(VK_EXT_EXTENDED_DYNAMIC_STATE_EXTENSION_NAME);
 
 #ifdef __APPLE__
     // Required by Vulkan spec if supported.

src/video_core/renderer_vulkan/vk_pipeline_cache.cpp

@@ -247,6 +247,7 @@ bool PipelineCache::RefreshGraphicsKey() {
     key.blend_controls.fill({});
     key.write_masks.fill({});
     key.mrt_swizzles.fill(Liverpool::ColorBuffer::SwapMode::Standard);
+    key.vertex_buffer_formats.fill(vk::Format::eUndefined);
 
     // First pass of bindings check to idenitfy formats and swizzles and pass them to rhe shader
     // recompiler.
@@ -310,7 +311,26 @@ bool PipelineCache::RefreshGraphicsKey() {
         std::tie(infos[i], modules[i], key.stage_hashes[i]) = GetProgram(stage, params, binding);
     }
 
-    const auto* fs_info = infos[u32(Shader::Stage::Fragment)];
+    const auto* vs_info = infos[static_cast<u32>(Shader::Stage::Vertex)];
+    if (vs_info && !instance.IsVertexInputDynamicState()) {
+        u32 vertex_binding = 0;
+        for (const auto& input : vs_info->vs_inputs) {
+            if (input.instance_step_rate == Shader::Info::VsInput::InstanceIdType::OverStepRate0 ||
+                input.instance_step_rate == Shader::Info::VsInput::InstanceIdType::OverStepRate1) {
+                continue;
+            }
+            const auto& buffer =
+                vs_info->ReadUd<AmdGpu::Buffer>(input.sgpr_base, input.dword_offset);
+            if (buffer.GetSize() == 0) {
+                continue;
+            }
+            ASSERT(vertex_binding < MaxVertexBufferCount);
+            key.vertex_buffer_formats[vertex_binding++] =
+                Vulkan::LiverpoolToVK::SurfaceFormat(buffer.GetDataFmt(), buffer.GetNumberFmt());
+        }
+    }
+
+    const auto* fs_info = infos[static_cast<u32>(Shader::Stage::Fragment)];
     key.mrt_mask = fs_info ? fs_info->mrt_mask : 0u;
 
     // Second pass to fill remain CB pipeline key data