From 10df7ba3195e730059462c356664bd0118296007 Mon Sep 17 00:00:00 2001
From: Stenzek <stenzek@gmail.com>
Date: Thu, 23 May 2024 14:30:18 +1000
Subject: [PATCH] GPU/HW: Ensure CLUT cache is synced when using
 SW-for-readbacks

---
 src/core/gpu.cpp    |  5 +++++
 src/core/gpu.h      |  1 +
 src/core/gpu_hw.cpp | 24 ++++++++++++++++++++----
 3 files changed, 26 insertions(+), 4 deletions(-)

diff --git a/src/core/gpu.cpp b/src/core/gpu.cpp
index 14a3eebafa..8c61a88c55 100644
--- a/src/core/gpu.cpp
+++ b/src/core/gpu.cpp
@@ -1481,6 +1481,11 @@ void GPU::InvalidateCLUT()
   m_current_clut_is_8bit = false;
 }
 
+bool GPU::IsCLUTValid() const
+{
+  return (m_current_clut_reg_bits != std::numeric_limits<decltype(m_current_clut_reg_bits)>::max());
+}
+
 void GPU::ClearDisplay()
 {
   ClearDisplayTexture();
diff --git a/src/core/gpu.h b/src/core/gpu.h
index 1ee2c3b1fc..cede79e8eb 100644
--- a/src/core/gpu.h
+++ b/src/core/gpu.h
@@ -311,6 +311,7 @@ class GPU
   void HandleGetGPUInfoCommand(u32 value);
   void UpdateCLUTIfNeeded(GPUTextureMode texmode, GPUTexturePaletteReg clut);
   void InvalidateCLUT();
+  bool IsCLUTValid() const;
 
   // Rendering in the backend
   virtual void ReadVRAM(u32 x, u32 y, u32 width, u32 height);
diff --git a/src/core/gpu_hw.cpp b/src/core/gpu_hw.cpp
index 45c4ff96ea..9dfd6252bc 100644
--- a/src/core/gpu_hw.cpp
+++ b/src/core/gpu_hw.cpp
@@ -2594,10 +2594,16 @@ void GPU_HW::UpdateSoftwareRenderer(bool copy_vram_from_hw)
     FlushRender();
     ReadVRAM(0, 0, VRAM_WIDTH, VRAM_HEIGHT);
 
-    // Sync the drawing area.
-    GPUBackendSetDrawingAreaCommand* cmd = sw_renderer->NewSetDrawingAreaCommand();
-    cmd->new_area = m_drawing_area;
-    sw_renderer->PushCommand(cmd);
+    // Sync the drawing area and CLUT.
+    GPUBackendSetDrawingAreaCommand* clip_cmd = sw_renderer->NewSetDrawingAreaCommand();
+    clip_cmd->new_area = m_drawing_area;
+    sw_renderer->PushCommand(clip_cmd);
+
+    GPUBackendUpdateCLUTCommand* clut_cmd = sw_renderer->NewUpdateCLUTCommand();
+    FillBackendCommandParameters(clut_cmd);
+    clut_cmd->reg.bits = m_current_clut_reg_bits;
+    clut_cmd->clut_is_8bit = m_current_clut_is_8bit;
+    sw_renderer->PushCommand(clut_cmd);
   }
 
   m_sw_renderer = std::move(sw_renderer);
@@ -3115,6 +3121,16 @@ void GPU_HW::UpdateCLUT(GPUTexturePaletteReg reg, bool clut_is_8bit)
   // Not done in HW
   GL_INS_FMT("Reloading CLUT from {},{}, {} not implemented", reg.GetXBase(), reg.GetYBase(),
              clut_is_8bit ? "8-bit" : "4-bit");
+
+  // But need to forward through to SW if using that for readbacks
+  if (m_sw_renderer)
+  {
+    GPUBackendUpdateCLUTCommand* cmd = m_sw_renderer->NewUpdateCLUTCommand();
+    FillBackendCommandParameters(cmd);
+    cmd->reg.bits = reg.bits;
+    cmd->clut_is_8bit = clut_is_8bit;
+    m_sw_renderer->PushCommand(cmd);
+  }
 }
 
 void GPU_HW::FlushRender()