GPU partial scanout shenanigans, WIP

src/core/gpu.cpp

@@ -196,7 +196,7 @@ void GPU::Reset(bool clear_vram)
   m_command_tick_event.Deactivate();
 
   SoftReset();
-  UpdateDisplay();
+  UpdateDisplay(false, 0, 0);
 }
 
 void GPU::SoftReset()
@@ -373,7 +373,7 @@ bool GPU::DoState(StateWrapper& sw, GPUTexture** host_texture, bool update_displ
   {
     UpdateCRTCConfig();
     if (update_display)
-      UpdateDisplay();
+      UpdateDisplay(false, 0, 0);
 
     UpdateCommandTickEvent();
   }
@@ -950,6 +950,9 @@ void GPU::CRTCTickEvent(TickCount ticks)
         Timers::AddTicks(HBLANK_TIMER_INDEX, static_cast<TickCount>(hblank_timer_ticks));
     }
 
+    if (m_crtc_state.start_address_changed)
+      DoPartialScanout();
+
     UpdateCRTCTickEvent();
     return;
   }
@@ -995,6 +998,9 @@ void GPU::CRTCTickEvent(TickCount ticks)
       m_crtc_state.in_vblank = false;
     }
 
+    if (m_crtc_state.start_address_changed)
+      DoPartialScanout();
+
     const bool new_vblank = m_crtc_state.current_scanline < m_crtc_state.vertical_display_start ||
                             m_crtc_state.current_scanline >= m_crtc_state.vertical_display_end;
     if (m_crtc_state.in_vblank != new_vblank)
@@ -1003,10 +1009,14 @@ void GPU::CRTCTickEvent(TickCount ticks)
       {
         DEBUG_LOG("Now in v-blank");
 
-        // flush any pending draws and "scan out" the image
         // TODO: move present in here I guess
-        FlushRender();
+        if (m_crtc_state.last_scanout_line < m_crtc_state.display_vram_height)
-        UpdateDisplay();
+        {
+          UpdateDisplay(true, m_crtc_state.last_scanout_line, m_crtc_state.display_vram_height);
+          m_crtc_state.last_scanout_line = 0;
+          m_crtc_state.start_address_changed = false;
+        }
+
         TimingEvents::SetFrameDone();
 
         // switch fields early. this is needed so we draw to the correct one.
@@ -1070,6 +1080,20 @@ void GPU::CRTCTickEvent(TickCount ticks)
   UpdateCRTCTickEvent();
 }
 
+void GPU::DoPartialScanout()
+{
+  const u32 vram_first_line = m_crtc_state.vertical_visible_start + m_crtc_state.display_origin_top;
+  const u32 vram_line = (m_crtc_state.current_scanline < vram_first_line) ? 0 : std::min<u32>(m_crtc_state.current_scanline - vram_first_line, m_crtc_state.display_vram_height);
+  if (vram_line != m_crtc_state.last_scanout_line && m_crtc_state.start_address_changed)
+  {
+    if (m_crtc_state.last_scanout_line < m_crtc_state.display_vram_height)
+      UpdateDisplay(true, m_crtc_state.last_scanout_line, vram_line);
+
+    m_crtc_state.last_scanout_line = vram_line;
+    m_crtc_state.start_address_changed = false;
+  }
+}
+
 void GPU::CommandTickEvent(TickCount ticks)
 {
   m_pending_command_ticks -= SystemTicksToGPUTicks(ticks);
@@ -1286,8 +1310,16 @@ void GPU::WriteGP1(u32 value)
       System::IncrementInternalFrameNumber();
       if (m_crtc_state.regs.display_address_start != new_value)
       {
+        m_crtc_state.start_address_changed = true;
         SynchronizeCRTC();
         m_crtc_state.regs.display_address_start = new_value;
+
+        if (!m_crtc_state.in_vblank)
+        {
+          GL_INS_FMT("Display address start set to ({},{}) at scanline {}", m_crtc_state.regs.X.GetValue(),
+                     m_crtc_state.regs.Y.GetValue(), m_crtc_state.current_scanline);
+        }
+
         UpdateCRTCDisplayParameters();
         OnBufferSwapped();
       }

src/core/gpu.h

@@ -193,6 +193,7 @@ public:
   // Returns the number of visible lines.
   ALWAYS_INLINE u16 GetCRTCActiveStartLine() const { return m_crtc_state.vertical_display_start; }
   ALWAYS_INLINE u16 GetCRTCActiveEndLine() const { return m_crtc_state.vertical_display_end; }
+  ALWAYS_INLINE u32 GetCRTCCurrentScanline() const { return m_crtc_state.current_scanline; }
 
   // Returns the video clock frequency.
   TickCount GetCRTCFrequency() const;
@@ -258,7 +259,9 @@ protected:
   void UpdateCRTCDisplayParameters();
 
   // Update ticks for this execution slice
+public:
   void UpdateCRTCTickEvent();
+  protected:
   void UpdateCommandTickEvent();
 
   // Updates dynamic bits in GPUSTAT (ready to send VRAM/ready to receive DMA)
@@ -268,6 +271,7 @@ protected:
   // Ticks for hblank/vblank.
   void CRTCTickEvent(TickCount ticks);
   void CommandTickEvent(TickCount ticks);
+  void DoPartialScanout();
 
   /// Returns 0 if the currently-displayed field is on odd lines (1,3,5,...) or 1 if even (2,4,6,...).
   ALWAYS_INLINE u32 GetInterlacedDisplayField() const { return ZeroExtend32(m_crtc_state.interlaced_field); }
@@ -315,7 +319,7 @@ protected:
   virtual void CopyVRAM(u32 src_x, u32 src_y, u32 dst_x, u32 dst_y, u32 width, u32 height);
   virtual void DispatchRenderCommand() = 0;
   virtual void UpdateCLUT(GPUTexturePaletteReg reg, bool clut_is_8bit) = 0;
-  virtual void UpdateDisplay() = 0;
+  virtual void UpdateDisplay(bool partial, u32 start_line, u32 end_line) = 0;
   virtual void DrawRendererStats();
   virtual void OnBufferSwapped();
 
@@ -536,6 +540,9 @@ protected:
 
     TickCount fractional_dot_ticks; // only used when timer0 is enabled
 
+    u32 last_scanout_line;
+    bool start_address_changed;
+
     bool in_hblank;
     bool in_vblank;
 

src/core/gpu_hw.cpp

@@ -503,7 +503,7 @@ void GPU_HW::UpdateSettings(const Settings& old_settings)
     UpdateVRAM(0, 0, VRAM_WIDTH, VRAM_HEIGHT, g_vram, false, false);
     if (m_write_mask_as_depth)
       UpdateDepthBufferFromMaskBit();
-    UpdateDisplay();
+    UpdateDisplay(false, 0, 0);
   }
   else if (m_vram_depth_texture && depth_buffer_changed)
   {
@@ -3593,7 +3593,7 @@ void GPU_HW::FlushRender()
   }
 }
 
-void GPU_HW::UpdateDisplay()
+void GPU_HW::UpdateDisplay(bool partial, u32 start_line, u32 end_line)
 {
   FlushRender();
   DeactivateROV();

src/core/gpu_hw.h

@@ -71,7 +71,7 @@ public:
   std::tuple<u32, u32> GetEffectiveDisplayResolution(bool scaled = true) override;
   std::tuple<u32, u32> GetFullDisplayResolution(bool scaled = true) override;
 
-  void UpdateDisplay() override;
+  void UpdateDisplay(bool partial, u32 start_line, u32 end_line) override;
 
 private:
   enum : u32

src/core/gpu_sw.cpp

@@ -93,7 +93,7 @@ GPUTexture* GPU_SW::GetDisplayTexture(u32 width, u32 height, GPUTexture::Format
     ClearDisplayTexture();
     g_gpu_device->RecycleTexture(std::move(m_upload_texture));
     m_upload_texture =
-      g_gpu_device->FetchTexture(width, height, 1, 1, 1, GPUTexture::Type::DynamicTexture, format, nullptr, 0);
+      g_gpu_device->FetchTexture(width, height, 1, 1, 1, GPUTexture::Type::Texture, format, nullptr, 0);
     if (!m_upload_texture) [[unlikely]]
       ERROR_LOG("Failed to create {}x{} {} texture", width, height, static_cast<u32>(format));
   }
@@ -202,7 +202,7 @@ ALWAYS_INLINE void CopyOutRow16<GPUTexture::Format::BGRA8, u32>(const u16* src_p
 }
 
 template<GPUTexture::Format display_format>
-ALWAYS_INLINE_RELEASE bool GPU_SW::CopyOut15Bit(u32 src_x, u32 src_y, u32 width, u32 height, u32 line_skip)
+ALWAYS_INLINE_RELEASE bool GPU_SW::CopyOut15Bit(u32 src_x, u32 src_y, u32 skip_y, u32 width, u32 height, u32 line_skip)
 {
   using OutputPixelType =
     std::conditional_t<display_format == GPUTexture::Format::RGBA8 || display_format == GPUTexture::Format::BGRA8, u32,
@@ -212,9 +212,12 @@ ALWAYS_INLINE_RELEASE bool GPU_SW::CopyOut15Bit(u32 src_x, u32 src_y, u32 width,
   if (!texture) [[unlikely]]
     return false;
 
+  src_y += skip_y;
+  height -= skip_y;
+
   u32 dst_stride = width * sizeof(OutputPixelType);
   u8* dst_ptr = m_upload_buffer.data();
-  const bool mapped = texture->Map(reinterpret_cast<void**>(&dst_ptr), &dst_stride, 0, 0, width, height);
+  const bool mapped = (skip_y == 0 && texture->Map(reinterpret_cast<void**>(&dst_ptr), &dst_stride, 0, 0, width, height));
 
   // Fast path when not wrapping around.
   if ((src_x + width) <= VRAM_WIDTH && (src_y + height) <= VRAM_HEIGHT)
@@ -248,13 +251,14 @@ ALWAYS_INLINE_RELEASE bool GPU_SW::CopyOut15Bit(u32 src_x, u32 src_y, u32 width,
   if (mapped)
     texture->Unmap();
   else
-    texture->Update(0, 0, width, height, m_upload_buffer.data(), dst_stride);
+    texture->Update(0, skip_y, width, height, m_upload_buffer.data(), dst_stride);
 
   return true;
 }
 
 template<GPUTexture::Format display_format>
-ALWAYS_INLINE_RELEASE bool GPU_SW::CopyOut24Bit(u32 src_x, u32 src_y, u32 skip_x, u32 width, u32 height, u32 line_skip)
+ALWAYS_INLINE_RELEASE bool GPU_SW::CopyOut24Bit(u32 src_x, u32 src_y, u32 skip_x, u32 skip_y, u32 width, u32 height,
+                                                u32 line_skip)
 {
   using OutputPixelType =
     std::conditional_t<display_format == GPUTexture::Format::RGBA8 || display_format == GPUTexture::Format::BGRA8, u32,
@@ -374,7 +378,7 @@ ALWAYS_INLINE_RELEASE bool GPU_SW::CopyOut24Bit(u32 src_x, u32 src_y, u32 skip_x
   return true;
 }
 
-bool GPU_SW::CopyOut(u32 src_x, u32 src_y, u32 skip_x, u32 width, u32 height, u32 line_skip, bool is_24bit)
+bool GPU_SW::CopyOut(u32 src_x, u32 src_y, u32 skip_x, u32 skip_y, u32 width, u32 height, u32 line_skip, bool is_24bit)
 {
   if (!is_24bit)
   {
@@ -383,16 +387,16 @@ bool GPU_SW::CopyOut(u32 src_x, u32 src_y, u32 skip_x, u32 width, u32 height, u3
     switch (m_16bit_display_format)
     {
       case GPUTexture::Format::RGBA5551:
-        return CopyOut15Bit<GPUTexture::Format::RGBA5551>(src_x, src_y, width, height, line_skip);
+        return CopyOut15Bit<GPUTexture::Format::RGBA5551>(src_x, src_y, skip_y, width, height, line_skip);
 
       case GPUTexture::Format::RGB565:
-        return CopyOut15Bit<GPUTexture::Format::RGB565>(src_x, src_y, width, height, line_skip);
+        return CopyOut15Bit<GPUTexture::Format::RGB565>(src_x, src_y, skip_y, width, height, line_skip);
 
       case GPUTexture::Format::RGBA8:
-        return CopyOut15Bit<GPUTexture::Format::RGBA8>(src_x, src_y, width, height, line_skip);
+        return CopyOut15Bit<GPUTexture::Format::RGBA8>(src_x, src_y, skip_y, width, height, line_skip);
 
       case GPUTexture::Format::BGRA8:
-        return CopyOut15Bit<GPUTexture::Format::BGRA8>(src_x, src_y, width, height, line_skip);
+        return CopyOut15Bit<GPUTexture::Format::BGRA8>(src_x, src_y, skip_y, width, height, line_skip);
 
       default:
         UnreachableCode();
@@ -403,16 +407,16 @@ bool GPU_SW::CopyOut(u32 src_x, u32 src_y, u32 skip_x, u32 width, u32 height, u3
     switch (m_24bit_display_format)
     {
       case GPUTexture::Format::RGBA5551:
-        return CopyOut24Bit<GPUTexture::Format::RGBA5551>(src_x, src_y, skip_x, width, height, line_skip);
+        return CopyOut24Bit<GPUTexture::Format::RGBA5551>(src_x, src_y, skip_x, skip_y, width, height, line_skip);
 
       case GPUTexture::Format::RGB565:
-        return CopyOut24Bit<GPUTexture::Format::RGB565>(src_x, src_y, skip_x, width, height, line_skip);
+        return CopyOut24Bit<GPUTexture::Format::RGB565>(src_x, src_y, skip_x, skip_y, width, height, line_skip);
 
       case GPUTexture::Format::RGBA8:
-        return CopyOut24Bit<GPUTexture::Format::RGBA8>(src_x, src_y, skip_x, width, height, line_skip);
+        return CopyOut24Bit<GPUTexture::Format::RGBA8>(src_x, src_y, skip_x, skip_y, width, height, line_skip);
 
       case GPUTexture::Format::BGRA8:
-        return CopyOut24Bit<GPUTexture::Format::BGRA8>(src_x, src_y, skip_x, width, height, line_skip);
+        return CopyOut24Bit<GPUTexture::Format::BGRA8>(src_x, src_y, skip_x, skip_y, width, height, line_skip);
 
       default:
         UnreachableCode();
@@ -420,7 +424,7 @@ bool GPU_SW::CopyOut(u32 src_x, u32 src_y, u32 skip_x, u32 width, u32 height, u3
   }
 }
 
-void GPU_SW::UpdateDisplay()
+void GPU_SW::UpdateDisplay(bool partial, u32 start_line, u32 end_line)
 {
   // fill display texture
   m_backend.Sync(true);
@@ -440,13 +444,14 @@ void GPU_SW::UpdateDisplay()
     const u32 vram_offset_y =
       m_crtc_state.display_vram_top + ((interlaced && m_GPUSTAT.vertical_resolution) ? field : 0);
     const u32 skip_x = is_24bit ? (m_crtc_state.display_vram_left - m_crtc_state.regs.X) : 0;
+    const u32 skip_y = partial ? start_line : 0;
     const u32 read_width = m_crtc_state.display_vram_width;
     const u32 read_height = interlaced ? (m_crtc_state.display_vram_height / 2) : m_crtc_state.display_vram_height;
 
     if (IsInterlacedDisplayEnabled())
     {
       const u32 line_skip = m_GPUSTAT.vertical_resolution;
-      if (CopyOut(vram_offset_x, vram_offset_y, skip_x, read_width, read_height, line_skip, is_24bit))
+      if (CopyOut(vram_offset_x, vram_offset_y, skip_x, skip_y, read_width, read_height, line_skip, is_24bit))
       {
         SetDisplayTexture(m_upload_texture.get(), nullptr, 0, 0, read_width, read_height);
         if (is_24bit && g_settings.display_24bit_chroma_smoothing)
@@ -462,7 +467,7 @@ void GPU_SW::UpdateDisplay()
     }
     else
     {
-      if (CopyOut(vram_offset_x, vram_offset_y, skip_x, read_width, read_height, 0, is_24bit))
+      if (CopyOut(vram_offset_x, vram_offset_y, skip_x, skip_y, read_width, read_height, 0, is_24bit))
       {
         SetDisplayTexture(m_upload_texture.get(), nullptr, 0, 0, read_width, read_height);
         if (is_24bit && g_settings.display_24bit_chroma_smoothing)
@@ -472,7 +477,7 @@ void GPU_SW::UpdateDisplay()
   }
   else
   {
-    if (CopyOut(0, 0, 0, VRAM_WIDTH, VRAM_HEIGHT, 0, false))
+    if (CopyOut(0, 0, 0, 0, VRAM_WIDTH, VRAM_HEIGHT, 0, false))
       SetDisplayTexture(m_upload_texture.get(), nullptr, 0, 0, VRAM_WIDTH, VRAM_HEIGHT);
   }
 }

src/core/gpu_sw.h

@@ -44,14 +44,14 @@ protected:
   void UpdateCLUT(GPUTexturePaletteReg reg, bool clut_is_8bit) override;
 
   template<GPUTexture::Format display_format>
-  bool CopyOut15Bit(u32 src_x, u32 src_y, u32 width, u32 height, u32 line_skip);
+  bool CopyOut15Bit(u32 src_x, u32 src_y, u32 skip_y, u32 width, u32 height, u32 line_skip);
 
   template<GPUTexture::Format display_format>
-  bool CopyOut24Bit(u32 src_x, u32 src_y, u32 skip_x, u32 width, u32 height, u32 line_skip);
+  bool CopyOut24Bit(u32 src_x, u32 src_y, u32 skip_x, u32 skip_y, u32 width, u32 height, u32 line_skip);
 
-  bool CopyOut(u32 src_x, u32 src_y, u32 skip_x, u32 width, u32 height, u32 line_skip, bool is_24bit);
+  bool CopyOut(u32 src_x, u32 src_y, u32 skip_x, u32 skip_y, u32 width, u32 height, u32 line_skip, bool is_24bit);
 
-  void UpdateDisplay() override;
+  void UpdateDisplay(bool partial, u32 start_line, u32 end_line) override;
 
   void DispatchRenderCommand() override;
 

src/core/timers.cpp

@@ -254,7 +254,7 @@ void Timers::CheckForIRQ(u32 timer, u32 old_counter)
       if (!cs.irq_done || cs.mode.irq_repeat)
       {
         // this is actually low for a few cycles
-        DEBUG_LOG("Raising timer {} pulse IRQ", timer);
+        DEBUG_LOG("Raising timer {} pulse IRQ @ {}", timer, g_gpu->GetCRTCCurrentScanline());
         InterruptController::SetLineState(irqnum, false);
         InterruptController::SetLineState(irqnum, true);
       }
@@ -377,8 +377,8 @@ void Timers::WriteRegister(u32 offset, u32 value)
     case 0x00:
     {
       const u32 old_counter = cs.counter;
-      DEBUG_LOG("Timer {} write counter {}", timer_index, value);
+      DEBUG_LOG("Timer {} write counter {}", timer_index, ZeroExtend32(Truncate16(value)));
-      cs.counter = value & u32(0xFFFF);
+      cs.counter = ZeroExtend32(Truncate16(value));
       CheckForIRQ(timer_index, old_counter);
       if (timer_index == 2 || !cs.external_counting_enabled)
         UpdateSysClkEvent();
@@ -388,28 +388,42 @@ void Timers::WriteRegister(u32 offset, u32 value)
     case 0x04:
     {
       static constexpr u32 WRITE_MASK = 0b1110001111111111;
+      const bool prev_external_counting_enabled = cs.external_counting_enabled;
 
-      DEBUG_LOG("Timer {} write mode register 0x{:04X}", timer_index, value);
+      DEBUG_LOG("Timer {} write mode register 0x{:04X} @ scaline {}", timer_index, value, g_gpu->GetCRTCCurrentScanline());
       cs.mode.bits = (value & WRITE_MASK) | (cs.mode.bits & ~WRITE_MASK);
       cs.use_external_clock = (cs.mode.clock_source & (timer_index == 2 ? 2 : 1)) != 0;
+      UpdateCountingEnabled(cs);
+
+      // Need to re-sync GPU if ext counting changed, since we're resetting the counter.
+      if (timer_index < 2 && !prev_external_counting_enabled && cs.external_counting_enabled)
+      {
+        if (timer_index == 0 || g_gpu->IsCRTCScanlinePending())
+          g_gpu->SynchronizeCRTC();
+      }
+
       cs.counter = 0;
       cs.irq_done = false;
       InterruptController::SetLineState(
         static_cast<InterruptController::IRQ>(static_cast<u32>(InterruptController::IRQ::TMR0) + timer_index), false);
 
-      UpdateCountingEnabled(cs);
       CheckForIRQ(timer_index, cs.counter);
+      if (timer_index == 2 || !cs.external_counting_enabled)
         UpdateSysClkEvent();
+      else if (timer_index < 2 && cs.external_counting_enabled)
+        g_gpu->UpdateCRTCTickEvent();
     }
     break;
 
     case 0x08:
     {
-      DEBUG_LOG("Timer {} write target 0x{:04X}", timer_index, ZeroExtend32(Truncate16(value)));
+      DEBUG_LOG("Timer {} write target {} @ {}", timer_index, ZeroExtend32(Truncate16(value)), g_gpu->GetCRTCCurrentScanline());
-      cs.target = value & u32(0xFFFF);
+      cs.target = ZeroExtend32(Truncate16(value));
       CheckForIRQ(timer_index, cs.counter);
       if (timer_index == 2 || !cs.external_counting_enabled)
         UpdateSysClkEvent();
+      else if (timer_index < 2 && cs.external_counting_enabled)
+        g_gpu->UpdateCRTCTickEvent();
     }
     break;