GPU: Implement PGXP for lines

stenzek · Dec 22, 2024 · 0b4e302 · 0b4e302
1 parent b81287e
commit 0b4e302
Show file tree

Hide file tree

Showing 13 changed files with 312 additions and 87 deletions.
diff --git a/src/common/gsvector_neon.h b/src/common/gsvector_neon.h
@@ -3131,6 +3131,11 @@ class alignas(16) GSVector4
 
   ALWAYS_INLINE GSVector2 zw() const { return GSVector2(vget_high_s32(v4s)); }
 
+  ALWAYS_INLINE static GSVector4 xyxy(const GSVector2& l, const GSVector2& h)
+  {
+    return GSVector4(vcombine_f32(l.v2s, h.v2s));
+  }
+
 #define VECTOR4_SHUFFLE_4(xs, xn, ys, yn, zs, zn, ws, wn)                                                              \
   ALWAYS_INLINE GSVector4 xs##ys##zs##ws() const                                                                       \
   {                                                                                                                    \

diff --git a/src/common/gsvector_nosimd.h b/src/common/gsvector_nosimd.h
@@ -2316,6 +2316,11 @@ class alignas(16) GSVector4
 
   ALWAYS_INLINE GSVector2 zw() const { return GSVector2(z, w); }
 
+  ALWAYS_INLINE static GSVector4 xyxy(const GSVector2& l, const GSVector2& h)
+  {
+    return GSVector4(l.x, l.y, h.x, h.y);
+  }
+
 #define VECTOR4_SHUFFLE_4(xs, xn, ys, yn, zs, zn, ws, wn)                                                              \
   ALWAYS_INLINE GSVector4 xs##ys##zs##ws() const { return GSVector4(F32[xn], F32[yn], F32[zn], F32[wn]); }
 

diff --git a/src/core/gpu.h b/src/core/gpu.h
@@ -279,7 +279,7 @@ class GPU final
   /// Returns the number of vertices in the buffered poly-line.
   ALWAYS_INLINE u32 GetPolyLineVertexCount() const
   {
-    return (static_cast<u32>(m_blit_buffer.size()) + BoolToUInt32(m_render_command.shading_enable)) >>
+    return (static_cast<u32>(m_polyline_buffer.size()) + BoolToUInt32(m_render_command.shading_enable)) >>
            BoolToUInt8(m_render_command.shading_enable);
   }
 
@@ -520,20 +520,20 @@ class GPU final
     u16 row;
   } m_vram_transfer = {};
 
+  std::unique_ptr<GPUDump::Recorder> m_gpu_dump;
+
   HeapFIFOQueue<u64, MAX_FIFO_SIZE> m_fifo;
-  std::vector<u32> m_blit_buffer;
+  TickCount m_max_run_ahead = 128;
+  u32 m_fifo_size = 128;
   u32 m_blit_remaining_words;
   GPURenderCommand m_render_command{};
-
-  std::unique_ptr<GPUDump::Recorder> m_gpu_dump;
+  std::vector<u32> m_blit_buffer;
+  std::vector<u64> m_polyline_buffer;
 
   ALWAYS_INLINE u32 FifoPop() { return Truncate32(m_fifo.Pop()); }
   ALWAYS_INLINE u32 FifoPeek() { return Truncate32(m_fifo.Peek()); }
   ALWAYS_INLINE u32 FifoPeek(u32 i) { return Truncate32(m_fifo.Peek(i)); }
 
-  TickCount m_max_run_ahead = 128;
-  u32 m_fifo_size = 128;
-
 private:
   using GP0CommandHandler = bool (GPU::*)();
   using GP0CommandHandlerTable = std::array<GP0CommandHandler, 256>;

diff --git a/src/core/gpu_backend.cpp b/src/core/gpu_backend.cpp
@@ -249,6 +249,16 @@ GPUBackendDrawLineCommand* GPUBackend::NewDrawLineCommand(u32 num_vertices)
   return cmd;
 }
 
+GPUBackendDrawPreciseLineCommand* GPUBackend::NewDrawPreciseLineCommand(u32 num_vertices)
+{
+  const u32 size =
+    sizeof(GPUBackendDrawPreciseLineCommand) + (num_vertices * sizeof(GPUBackendDrawPreciseLineCommand::Vertex));
+  GPUBackendDrawPreciseLineCommand* cmd = static_cast<GPUBackendDrawPreciseLineCommand*>(
+    GPUThread::AllocateCommand(GPUBackendCommandType::DrawPreciseLine, size));
+  cmd->num_vertices = Truncate16(num_vertices);
+  return cmd;
+}
+
 void GPUBackend::PushCommand(GPUThreadCommand* cmd)
 {
   GPUThread::PushCommand(cmd);
@@ -489,6 +499,15 @@ void GPUBackend::HandleCommand(const GPUThreadCommand* cmd)
       s_counters.num_primitives += ccmd->num_vertices / 2;
       DrawLine(ccmd);
     }
+    break;
+
+    case GPUBackendCommandType::DrawPreciseLine:
+    {
+      const GPUBackendDrawPreciseLineCommand* ccmd = static_cast<const GPUBackendDrawPreciseLineCommand*>(cmd);
+      s_counters.num_vertices += ccmd->num_vertices;
+      s_counters.num_primitives += ccmd->num_vertices / 2;
+      DrawPreciseLine(ccmd);
+    }
     break;
 
       DefaultCaseIsUnreachable();

diff --git a/src/core/gpu_backend.h b/src/core/gpu_backend.h
@@ -51,6 +51,7 @@ class GPUBackend
   static GPUBackendDrawPrecisePolygonCommand* NewDrawPrecisePolygonCommand(u32 num_vertices);
   static GPUBackendDrawRectangleCommand* NewDrawRectangleCommand();
   static GPUBackendDrawLineCommand* NewDrawLineCommand(u32 num_vertices);
+  static GPUBackendDrawPreciseLineCommand* NewDrawPreciseLineCommand(u32 num_vertices);
   static void PushCommand(GPUThreadCommand* cmd);
   static void PushCommandAndWakeThread(GPUThreadCommand* cmd);
   static void PushCommandAndSync(GPUThreadCommand* cmd, bool spin);
@@ -125,6 +126,7 @@ class GPUBackend
   virtual void DrawPrecisePolygon(const GPUBackendDrawPrecisePolygonCommand* cmd) = 0;
   virtual void DrawSprite(const GPUBackendDrawRectangleCommand* cmd) = 0;
   virtual void DrawLine(const GPUBackendDrawLineCommand* cmd) = 0;
+  virtual void DrawPreciseLine(const GPUBackendDrawPreciseLineCommand* cmd) = 0;
 
   virtual void DrawingAreaChanged() = 0;
   virtual void UpdateCLUT(GPUTexturePaletteReg reg, bool clut_is_8bit) = 0;