From ea038d66538975319858f792052af1d0fa997fe3 Mon Sep 17 00:00:00 2001
From: ReinUsesLisp <reinuseslisp@airmail.cc>
Date: Sat, 12 Jun 2021 05:07:52 -0300
Subject: vulkan: Add VK_EXT_vertex_input_dynamic_state support

Reduces the number of total pipelines generated on Vulkan.
Tested on Super Smash Bros. Ultimate.
---
 .../renderer_vulkan/vk_state_tracker.cpp           | 50 ++++++++++++++--------
 1 file changed, 32 insertions(+), 18 deletions(-)

(limited to 'src/video_core/renderer_vulkan/vk_state_tracker.cpp')

diff --git a/src/video_core/renderer_vulkan/vk_state_tracker.cpp b/src/video_core/renderer_vulkan/vk_state_tracker.cpp
index 956f86845..0ebe0473f 100644
--- a/src/video_core/renderer_vulkan/vk_state_tracker.cpp
+++ b/src/video_core/renderer_vulkan/vk_state_tracker.cpp
@@ -29,9 +29,10 @@ using Flags = Maxwell3D::DirtyState::Flags;
 
 Flags MakeInvalidationFlags() {
     static constexpr int INVALIDATION_FLAGS[]{
-        Viewports,         Scissors,  DepthBias,         BlendConstants,    DepthBounds,
-        StencilProperties, CullMode,  DepthBoundsEnable, DepthTestEnable,   DepthWriteEnable,
-        DepthCompareOp,    FrontFace, StencilOp,         StencilTestEnable, VertexBuffers,
+        Viewports,       Scissors,          DepthBias,      BlendConstants,
+        DepthBounds,     StencilProperties, CullMode,       DepthBoundsEnable,
+        DepthTestEnable, DepthWriteEnable,  DepthCompareOp, FrontFace,
+        StencilOp,       StencilTestEnable, VertexBuffers,  VertexInput,
     };
     Flags flags{};
     for (const int flag : INVALIDATION_FLAGS) {
@@ -40,6 +41,12 @@ Flags MakeInvalidationFlags() {
     for (int index = VertexBuffer0; index <= VertexBuffer31; ++index) {
         flags[index] = true;
     }
+    for (int index = VertexAttribute0; index <= VertexAttribute31; ++index) {
+        flags[index] = true;
+    }
+    for (int index = VertexBinding0; index <= VertexBinding31; ++index) {
+        flags[index] = true;
+    }
     return flags;
 }
 
@@ -134,31 +141,38 @@ void SetupDirtyBlending(Tables& tables) {
     FillBlock(tables[0], OFF(independent_blend), NUM(independent_blend), Blending);
 }
 
-void SetupDirtyInstanceDivisors(Tables& tables) {
-    static constexpr size_t divisor_offset = 3;
-    for (size_t index = 0; index < Regs::NumVertexArrays; ++index) {
-        tables[0][OFF(instanced_arrays) + index] = InstanceDivisors;
-        tables[0][OFF(vertex_array) + index * NUM(vertex_array[0]) + divisor_offset] =
-            InstanceDivisors;
+void SetupDirtyViewportSwizzles(Tables& tables) {
+    static constexpr size_t swizzle_offset = 6;
+    for (size_t index = 0; index < Regs::NumViewports; ++index) {
+        tables[0][OFF(viewport_transform) + index * NUM(viewport_transform[0]) + swizzle_offset] =
+            ViewportSwizzles;
     }
 }
 
 void SetupDirtyVertexAttributes(Tables& tables) {
-    FillBlock(tables[0], OFF(vertex_attrib_format), NUM(vertex_attrib_format), VertexAttributes);
+    for (size_t i = 0; i < Regs::NumVertexAttributes; ++i) {
+        const size_t offset = OFF(vertex_attrib_format) + i * NUM(vertex_attrib_format[0]);
+        FillBlock(tables[0], offset, NUM(vertex_attrib_format[0]), VertexAttribute0 + i);
+    }
+    FillBlock(tables[1], OFF(vertex_attrib_format), Regs::NumVertexAttributes, VertexInput);
 }
 
-void SetupDirtyViewportSwizzles(Tables& tables) {
-    static constexpr size_t swizzle_offset = 6;
-    for (size_t index = 0; index < Regs::NumViewports; ++index) {
-        tables[0][OFF(viewport_transform) + index * NUM(viewport_transform[0]) + swizzle_offset] =
-            ViewportSwizzles;
+void SetupDirtyVertexBindings(Tables& tables) {
+    // Do NOT include stride here, it's implicit in VertexBuffer
+    static constexpr size_t divisor_offset = 3;
+    for (size_t i = 0; i < Regs::NumVertexArrays; ++i) {
+        const u8 flag = static_cast<u8>(VertexBinding0 + i);
+        tables[0][OFF(instanced_arrays) + i] = VertexInput;
+        tables[1][OFF(instanced_arrays) + i] = flag;
+        tables[0][OFF(vertex_array) + i * NUM(vertex_array[0]) + divisor_offset] = VertexInput;
+        tables[1][OFF(vertex_array) + i * NUM(vertex_array[0]) + divisor_offset] = flag;
     }
 }
 } // Anonymous namespace
 
 StateTracker::StateTracker(Tegra::GPU& gpu)
     : flags{gpu.Maxwell3D().dirty.flags}, invalidation_flags{MakeInvalidationFlags()} {
-    auto& tables = gpu.Maxwell3D().dirty.tables;
+    auto& tables{gpu.Maxwell3D().dirty.tables};
     SetupDirtyFlags(tables);
     SetupDirtyViewports(tables);
     SetupDirtyScissors(tables);
@@ -175,9 +189,9 @@ StateTracker::StateTracker(Tegra::GPU& gpu)
     SetupDirtyStencilOp(tables);
     SetupDirtyStencilTestEnable(tables);
     SetupDirtyBlending(tables);
-    SetupDirtyInstanceDivisors(tables);
-    SetupDirtyVertexAttributes(tables);
     SetupDirtyViewportSwizzles(tables);
+    SetupDirtyVertexAttributes(tables);
+    SetupDirtyVertexBindings(tables);
 }
 
 } // namespace Vulkan
-- 
cgit v1.2.3