GPU: Let the GPU drive DMA transfers for large time slices

Fixes FMV breakage... again.

src/core/gpu.cpp

@@ -65,8 +65,6 @@ void GPU::SoftReset()
   m_draw_mode.SetTextureWindow(0);
   UpdateDMARequest();
   UpdateCRTCConfig();
-
-  m_tick_event->Deactivate();
   UpdateSliceTicks();
 }
 
@@ -291,8 +289,13 @@ void GPU::DMAWrite(const u32* words, u32 word_count)
       {
         Assert(m_blitter_ticks == 0);
         m_blitter_ticks = GetPendingGPUTicks() + word_count;
+
+        // reschedule GPU tick event
+        const TickCount sysclk_ticks = GPUTicksToSystemTicks(word_count);
+        if (m_tick_event->GetTicksUntilNextExecution() > sysclk_ticks)
+          m_tick_event->Schedule(sysclk_ticks);
+
         UpdateDMARequest();
-        UpdateSliceTicks();
       }
     }
     break;
@@ -434,12 +437,6 @@ void GPU::UpdateCRTCConfig()
   UpdateSliceTicks();
 }
 
-static TickCount GPUTicksToSystemTicks(TickCount gpu_ticks)
-{
-  // convert to master clock, rounding up as we want to overshoot not undershoot
-  return static_cast<TickCount>((static_cast<u32>(gpu_ticks) * 7u + 10u) / 11u);
-}
-
 TickCount GPU::GetPendingGPUTicks() const
 {
   const TickCount pending_sysclk_ticks = m_tick_event->GetTicksSinceLastExecution();
@@ -462,7 +459,6 @@ void GPU::UpdateSliceTicks()
 
   m_tick_event->Schedule(
     GPUTicksToSystemTicks((m_blitter_ticks > 0) ? std::min(m_blitter_ticks, ticks_until_vblank) : ticks_until_vblank));
-  m_tick_event->SetPeriod(GPUTicksToSystemTicks(ticks_until_hblank));
 }
 
 bool GPU::IsRasterScanlinePending() const
@@ -479,16 +475,16 @@ void GPU::Execute(TickCount ticks)
     m_crtc_state.fractional_ticks = ticks_mul_11 % 7;
     m_crtc_state.current_tick_in_scanline += gpu_ticks;
 
-    if (m_blitter_ticks > 0)
+    // handle blits
+    TickCount blit_ticks_remaining = gpu_ticks;
+    while (m_blitter_ticks > 0 && blit_ticks_remaining > 0)
     {
-      m_blitter_ticks -= gpu_ticks;
+      const TickCount slice = std::min(blit_ticks_remaining, m_blitter_ticks);
-      if (m_blitter_ticks <= 0)
+      m_blitter_ticks -= slice;
-      {
+      blit_ticks_remaining -= slice;
-        m_blitter_ticks = 0;
       UpdateDMARequest();
     }
   }
-  }
 
   if (m_crtc_state.current_tick_in_scanline < m_crtc_state.horizontal_total)
   {
@@ -636,6 +632,7 @@ void GPU::WriteGP1(u32 value)
     case 0x00: // Reset GPU
     {
       Log_DebugPrintf("GP1 reset GPU");
+      Synchronize();
       SoftReset();
     }
     break;
@@ -647,6 +644,7 @@ void GPU::WriteGP1(u32 value)
       m_command_total_words = 0;
       m_vram_transfer = {};
       m_GP0_buffer.clear();
+      m_blitter_ticks = 0;
       UpdateDMARequest();
     }
     break;
@@ -662,7 +660,7 @@ void GPU::WriteGP1(u32 value)
     {
       const bool disable = ConvertToBoolUnchecked(value & 0x01);
       Log_DebugPrintf("Display %s", disable ? "disabled" : "enabled");
-      m_tick_event->InvokeEarly();
+      Synchronize();
       m_GPUSTAT.display_disable = disable;
     }
     break;
@@ -690,7 +688,7 @@ void GPU::WriteGP1(u32 value)
 
       if (m_crtc_state.regs.horizontal_display_range != new_value)
       {
-        m_tick_event->InvokeEarly(true);
+        Synchronize();
         m_crtc_state.regs.horizontal_display_range = new_value;
         UpdateCRTCConfig();
       }
@@ -704,7 +702,7 @@ void GPU::WriteGP1(u32 value)
 
       if (m_crtc_state.regs.vertical_display_range != new_value)
       {
-        m_tick_event->InvokeEarly(true);
+        Synchronize();
         m_crtc_state.regs.vertical_display_range = new_value;
         UpdateCRTCConfig();
       }
@@ -739,7 +737,7 @@ void GPU::WriteGP1(u32 value)
 
       if (m_GPUSTAT.bits != new_GPUSTAT.bits)
       {
-        m_tick_event->InvokeEarly(true);
+        Synchronize();
         m_GPUSTAT.bits = new_GPUSTAT.bits;
         UpdateCRTCConfig();
       }

src/core/gpu.h

@@ -133,6 +133,12 @@ public:
   void DMARead(u32* words, u32 word_count);
   void DMAWrite(const u32* words, u32 word_count);
 
+  /// Returns the number of pending GPU ticks.
+  TickCount GetPendingGPUTicks() const;
+
+  /// Returns true if enough ticks have passed for the raster to be on the next line.
+  bool IsRasterScanlinePending() const;
+
   // Synchronizes the CRTC, updating the hblank timer.
   void Synchronize();
 
@@ -152,6 +158,12 @@ public:
   static std::unique_ptr<GPU> CreateSoftwareRenderer();
 
 protected:
+  static TickCount GPUTicksToSystemTicks(TickCount gpu_ticks)
+  {
+    // convert to master clock, rounding up as we want to overshoot not undershoot
+    return static_cast<TickCount>((static_cast<u32>(gpu_ticks) * 7u + 10u) / 11u);
+  }
+
   // Helper/format conversion functions.
   static constexpr u8 Convert5To8(u8 x5) { return (x5 << 3) | (x5 & 7); }
   static constexpr u8 Convert8To5(u8 x8) { return (x8 >> 3); }
@@ -311,12 +323,6 @@ protected:
   // Ticks for hblank/vblank.
   void Execute(TickCount ticks);
 
-  /// Returns the number of pending GPU ticks.
-  TickCount GetPendingGPUTicks() const;
-
-  /// Returns true if enough ticks have passed for the raster to be on the next line.
-  bool IsRasterScanlinePending() const;
-
   /// Returns true if scanout should be interlaced.
   bool IsDisplayInterlaced() const { return !m_force_progressive_scan && m_GPUSTAT.In480iMode(); }
 

src/core/timers.cpp

@@ -166,10 +166,10 @@ u32 Timers::ReadRegister(u32 offset)
   {
     case 0x00:
     {
-      if (timer_index < 2)
+      if (timer_index < 2 && cs.external_counting_enabled)
       {
         // timers 0/1 depend on the GPU
-        if (cs.external_counting_enabled)
+        if (timer_index == 0 || m_gpu->IsRasterScanlinePending())
           m_gpu->Synchronize();
       }
 
@@ -180,10 +180,10 @@ u32 Timers::ReadRegister(u32 offset)
 
     case 0x04:
     {
-      if (timer_index < 2)
+      if (timer_index < 2 && cs.external_counting_enabled)
       {
         // timers 0/1 depend on the GPU
-        if (cs.external_counting_enabled)
+        if (timer_index == 0 || m_gpu->IsRasterScanlinePending())
           m_gpu->Synchronize();
       }
 
@@ -211,8 +211,13 @@ void Timers::WriteRegister(u32 offset, u32 value)
 
   CounterState& cs = m_states[timer_index];
 
-  if (timer_index < 2)
+  if (timer_index < 2 && cs.external_counting_enabled)
+  {
+    // timers 0/1 depend on the GPU
+    if (timer_index == 0 || m_gpu->IsRasterScanlinePending())
       m_gpu->Synchronize();
+  }
+
   m_sysclk_event->InvokeEarly();
 
   switch (port_offset)

src/core/timing_event.cpp

@@ -74,7 +74,7 @@ void TimingEvent::InvokeEarly(bool force /* = false */)
   if (!m_active)
     return;
 
-  const TickCount pending_ticks = m_system->m_running_events ? 0 : m_system->m_cpu->GetPendingTicks();
+  const TickCount pending_ticks = m_system->m_cpu->GetPendingTicks();
   const TickCount ticks_to_execute = m_time_since_last_run + pending_ticks;
   if (!force && ticks_to_execute < m_period)
     return;
@@ -93,7 +93,7 @@ void TimingEvent::Activate()
     return;
 
   // leave the downcount intact
-  const TickCount pending_ticks = m_system->m_running_events ? 0 : m_system->m_cpu->GetPendingTicks();
+  const TickCount pending_ticks = m_system->m_cpu->GetPendingTicks();
   m_downcount += pending_ticks;
   m_time_since_last_run -= pending_ticks;
 
@@ -106,7 +106,7 @@ void TimingEvent::Deactivate()
   if (!m_active)
     return;
 
-  const TickCount pending_ticks = m_system->m_running_events ? 0 : m_system->m_cpu->GetPendingTicks();
+  const TickCount pending_ticks = m_system->m_cpu->GetPendingTicks();
   m_downcount -= pending_ticks;
   m_time_since_last_run += pending_ticks;