From 9982f1be6f11c6d341197487867239b822e91350 Mon Sep 17 00:00:00 2001
From: psucien <bad_cast@protonmail.com>
Date: Thu, 27 Jun 2024 22:10:55 +0200
Subject: [PATCH] renderer_vulkan: skip redundant FCE and FMask decompression
 passes

---
 src/video_core/amdgpu/liverpool.h             | 18 ++++++++++++++++-
 .../renderer_vulkan/vk_pipeline_cache.cpp     | 20 ++++++++++++++++++-
 .../renderer_vulkan/vk_rasterizer.cpp         |  8 ++++++++
 3 files changed, 44 insertions(+), 2 deletions(-)

diff --git a/src/video_core/amdgpu/liverpool.h b/src/video_core/amdgpu/liverpool.h
index a170a30cf..68f7dc37f 100644
--- a/src/video_core/amdgpu/liverpool.h
+++ b/src/video_core/amdgpu/liverpool.h
@@ -606,6 +606,20 @@ struct Liverpool {
         BitField<30, 1, u32> enable;
     };
 
+    union ColorControl {
+        enum class OperationMode : u32 {
+            Disable = 0u,
+            Normal = 1u,
+            EliminateFastClear = 2u,
+            Resolve = 3u,
+            FmaskDecompress = 5u,
+        };
+
+        BitField<3, 1, u32> degamma_enable;
+        BitField<4, 3, OperationMode> mode;
+        BitField<16, 8, u32> rop3;
+    };
+
     struct ColorBuffer {
         enum class EndianSwap : u32 {
             None = 0,
@@ -825,7 +839,8 @@ struct Liverpool {
             u32 draw_initiator;
             INSERT_PADDING_WORDS(0xA200 - 0xA1F9 - 4);
             DepthControl depth_control;
-            INSERT_PADDING_WORDS(2);
+            INSERT_PADDING_WORDS(1);
+            ColorControl color_control;
             DepthBufferControl depth_buffer_control;
             ClipperControl clipper_control;
             PolygonControl polygon_control;
@@ -995,6 +1010,7 @@ static_assert(GFX6_3D_REG_INDEX(blend_control) == 0xA1E0);
 static_assert(GFX6_3D_REG_INDEX(index_base_address) == 0xA1F9);
 static_assert(GFX6_3D_REG_INDEX(draw_initiator) == 0xA1FC);
 static_assert(GFX6_3D_REG_INDEX(depth_control) == 0xA200);
+static_assert(GFX6_3D_REG_INDEX(color_control) == 0xA202);
 static_assert(GFX6_3D_REG_INDEX(clipper_control) == 0xA204);
 static_assert(GFX6_3D_REG_INDEX(viewport_control) == 0xA206);
 static_assert(GFX6_3D_REG_INDEX(vs_output_control) == 0xA207);
diff --git a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
index cb72edafb..96e731956 100644
--- a/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
+++ b/src/video_core/renderer_vulkan/vk_pipeline_cache.cpp
@@ -120,6 +120,11 @@ void PipelineCache::RefreshGraphicsKey() {
         key.depth_format = LiverpoolToVK::DepthFormat(db.z_info.format, db.stencil_info.format);
         key.depth.depth_enable.Assign(key.depth_format != vk::Format::eUndefined);
     }
+
+    // TODO: Should be a check for `OperationMode::Disable` once we emulate HW state init packet
+    // sent by system software.
+    const auto skip_cb_binding = false;
+
     // `RenderingInfo` is assumed to be initialized with a contiguous array of valid color
     // attachments. This might be not a case as HW color buffers can be bound in an arbitrary order.
     // We need to do some arrays compaction at this stage
@@ -129,7 +134,7 @@ void PipelineCache::RefreshGraphicsKey() {
     int remapped_cb{};
     for (auto cb = 0u; cb < Liverpool::NumColorBuffers; ++cb) {
         auto const& col_buf = regs.color_buffers[cb];
-        if (!col_buf) {
+        if (!col_buf || skip_cb_binding) {
             continue;
         }
         const auto base_format =
@@ -160,6 +165,19 @@ void PipelineCache::RefreshGraphicsKey() {
 std::unique_ptr<GraphicsPipeline> PipelineCache::CreateGraphicsPipeline() {
     const auto& regs = liverpool->regs;
 
+    // There are several cases (e.g. FCE, FMask/HTile decompression) where we don't need to do an
+    // actual draw hence can skip pipeline creation.
+    if (regs.color_control.mode == Liverpool::ColorControl::OperationMode::EliminateFastClear) {
+        LOG_TRACE(Render_Vulkan, "FCE pass skipped");
+        return {};
+    }
+
+    if (regs.color_control.mode == Liverpool::ColorControl::OperationMode::FmaskDecompress) {
+        // TODO: check for a valid MRT1 to promote the draw to the resolve pass.
+        LOG_TRACE(Render_Vulkan, "FMask decompression pass skipped");
+        return {};
+    }
+
     u32 binding{};
     std::array<Shader::IR::Program, MaxShaderStages> programs;
     std::array<const Shader::Info*, MaxShaderStages> infos{};
diff --git a/src/video_core/renderer_vulkan/vk_rasterizer.cpp b/src/video_core/renderer_vulkan/vk_rasterizer.cpp
index 70c574366..ef937025b 100644
--- a/src/video_core/renderer_vulkan/vk_rasterizer.cpp
+++ b/src/video_core/renderer_vulkan/vk_rasterizer.cpp
@@ -40,6 +40,10 @@ void Rasterizer::Draw(bool is_indexed, u32 index_offset) {
     const auto& regs = liverpool->regs;
     const u32 num_indices = SetupIndexBuffer(is_indexed, index_offset);
     const GraphicsPipeline* pipeline = pipeline_cache.GetGraphicsPipeline();
+    if (!pipeline) {
+        return;
+    }
+
     pipeline->BindResources(memory, vertex_index_buffer, texture_cache);
 
     boost::container::static_vector<vk::RenderingAttachmentInfo, Liverpool::NumColorBuffers>
@@ -117,6 +121,10 @@ void Rasterizer::DispatchDirect() {
     const auto cmdbuf = scheduler.CommandBuffer();
     const auto& cs_program = liverpool->regs.cs_program;
     const ComputePipeline* pipeline = pipeline_cache.GetComputePipeline();
+    if (!pipeline) {
+        return;
+    }
+
     pipeline->BindResources(memory, vertex_index_buffer, texture_cache);
 
     cmdbuf.bindPipeline(vk::PipelineBindPoint::eCompute, pipeline->Handle());