Merge pull request #12642 from jordan-woyak/state-write-osd-errors

Core/State: Display messages on state write failures.

Data/Sys/Shaders/PerceptualHDR.glsl

@@ -0,0 +1,109 @@
+/*
+[configuration]
+
+[OptionRangeFloat]
+GUIName = Amplificiation
+OptionName = AMPLIFICATION
+MinValue = 1.0
+MaxValue = 6.0
+StepAmount = 0.25
+DefaultValue = 2.5
+
+[/configuration]
+*/
+
+// ICtCP Colorspace as defined by Dolby here:
+// https://professional.dolby.com/siteassets/pdfs/ictcp_dolbywhitepaper_v071.pdf
+
+/***** Transfer Function *****/
+
+const float4 m_1 = float4(2610.0 / 16384.0);
+const float4 m_2 = float4(128.0 * 2523.0 / 4096.0);
+const float4 m_1_inv = float4(16384.0 / 2610.0);
+const float4 m_2_inv = float4(4096.0 / (128.0 * 2523.0));
+
+const float4 c_1 = float4(3424.0 / 4096.0);
+const float4 c_2 = float4(2413.0 / 4096.0 * 32.0);
+const float4 c_3 = float4(2392.0 / 4096.0 * 32.0);
+
+float4 EOTF_inv(float4 lms) {
+    float4 y = pow(lms, m_1);
+    return pow((c_1 + c_2 * y) / (1.0 + c_3 * y), m_2);
+}
+
+float4 EOTF(float4 lms) {
+    float4 x = pow(lms, m_2_inv);
+    return pow(-(x - c_1) / (c_3 * x - c_2), m_1_inv);
+}
+
+// This is required as scaling in EOTF space is not linear.
+float EOTF_AMPLIFICATION = EOTF_inv(float4(AMPLIFICATION)).x;
+
+/***** Linear <--> ICtCp *****/
+
+const mat4 RGBtoLMS = mat4(
+    1688.0, 683.0,  99.0,   0.0,
+    2146.0, 2951.0, 309.0,  0.0,
+    262.0,  462.0,  3688.0, 0.0,
+    0.0,    0.0,    0.0,    4096.0) / 4096.0;
+
+const mat4 LMStoICtCp = mat4(
+    +2048.0, +6610.0,  +17933.0, 0.0,
+    +2048.0, -13613.0, -17390.0, 0.0,
+    +0.0,    +7003.0,  -543.0,   0.0,
+    +0.0,    +0.0,     +0.0,     4096.0) / 4096.0;
+
+float4 LinearRGBToICtCP(float4 c)
+{
+    return LMStoICtCp * EOTF_inv(RGBtoLMS * c);
+}
+
+/***** ICtCp <--> Linear *****/
+
+mat4 ICtCptoLMS = inverse(LMStoICtCp);
+mat4 LMStoRGB = inverse(RGBtoLMS);
+
+float4 ICtCpToLinearRGB(float4 c)
+{
+    return LMStoRGB * EOTF(ICtCptoLMS * c);
+}
+
+void main()
+{
+    float4 color = Sample();
+
+    // Nothing to do here, we are in SDR
+    if (!OptionEnabled(hdr_output) || !OptionEnabled(linear_space_output)) {
+        SetOutput(color);
+        return;
+    }
+
+    // Renormalize Color to be in [0.0 - 1.0] SDR Space. We will revert this later.
+    const float hdr_paper_white = hdr_paper_white_nits / hdr_sdr_white_nits;
+    color.rgb /= hdr_paper_white;
+
+    // Convert Color to Perceptual Color Space. This will allow us to do perceptual
+    // scaling while also being able to use the luminance channel.
+    float4 ictcp_color = LinearRGBToICtCP(color);
+
+    // Scale the color in perceptual space depending on the percieved luminance.
+    //
+    // At low luminances, ~0.0, pow(EOTF_AMPLIFICATION, ~0.0) ~= 1.0, so the
+    // color will appear to be unchanged. This is important as we don't want to
+    // over expose dark colors which would not have otherwise been seen.
+    //
+    // At high luminances, ~1.0, pow(EOTF_AMPLIFICATION, ~1.0) ~= EOTF_AMPLIFICATION,
+    // which is equivilant to scaling the color by EOTF_AMPLIFICATION. This is
+    // important as we want to get the most out of the display, and we want to
+    // get bright colors to hit their target brightness.
+    //
+    // For more information, see this desmos demonstrating this scaling process:
+    // https://www.desmos.com/calculator/syjyrjsj5c
+    const float luminance = ictcp_color.x;
+    ictcp_color *= pow(EOTF_AMPLIFICATION, luminance);
+
+    // Convert back to Linear RGB and output the color to the display.
+    // We use hdr_paper_white to renormalize the color to the comfortable
+    // SDR viewing range.
+    SetOutput(hdr_paper_white * ICtCpToLinearRGB(ictcp_color));
+}

Source/Core/Core/Core.cpp

@@ -210,11 +210,6 @@ bool IsRunningAndStarted()
   return s_is_started && !s_is_stopping;
 }
 
-bool IsRunningInCurrentThread()
-{
-  return IsRunning() && IsCPUThread();
-}
-
 bool IsCPUThread()
 {
   return tls_is_cpu_thread;

Source/Core/Core/Core.h

@@ -139,7 +139,6 @@ std::string StopMessage(bool main_thread, std::string_view message);
 
 bool IsRunning();
 bool IsRunningAndStarted();  // is running and the CPU loop has been entered
-bool IsRunningInCurrentThread();  // this tells us whether we are running in the CPU thread.
 bool IsCPUThread();          // this tells us whether we are the CPU thread.
 bool IsGPUThread();
 bool IsHostThread();

Source/Core/Core/State.cpp

@@ -402,7 +402,7 @@ static void CompressAndDumpState(CompressAndDumpState_args& save_args)
   File::IOFile f(temp_filename, "wb");
   if (!f)
   {
-    Core::DisplayMessage("Could not save state", 2000);
+    Core::DisplayMessage("Failed to create state file", 2000);
     return;
   }
 
@@ -413,6 +413,9 @@ static void CompressAndDumpState(CompressAndDumpState_args& save_args)
   else
     f.WriteBytes(buffer_data, buffer_size);
 
+  if (!f.IsGood())
+    Core::DisplayMessage("Failed to write state file", 2000);
+
   const std::string last_state_filename = File::GetUserPath(D_STATESAVES_IDX) + "lastState.sav";
   const std::string last_state_dtmname = last_state_filename + ".dtm";
   const std::string dtmname = filename + ".dtm";
@@ -448,12 +451,20 @@ static void CompressAndDumpState(CompressAndDumpState_args& save_args)
     // Move written state to final location.
     // TODO: This should also be atomic. This is possible on all systems, but needs a special
     // implementation of IOFile on Windows.
-    f.Close();
+    if (!f.Close())
-    File::Rename(temp_filename, filename);
+      Core::DisplayMessage("Failed to close state file", 2000);
+
+    if (!File::Rename(temp_filename, filename))
+    {
+      Core::DisplayMessage("Failed to rename state file", 2000);
+    }
+    else
+    {
+      const std::filesystem::path temp_path(filename);
+      Core::DisplayMessage(fmt::format("Saved State to {}", temp_path.filename().string()), 2000);
+    }
   }
 
-  std::filesystem::path tempfilename(filename);
-  Core::DisplayMessage(fmt::format("Saved State to {}", tempfilename.filename().string()), 2000);
   Host_UpdateMainFrame();
 }
 

Source/Core/DolphinQt/Config/ToolTipControls/BalloonTip.cpp

@@ -315,10 +315,12 @@ void BalloonTip::UpdateBoundsAndRedraw(const QPoint& target_arrow_tip_position,
   // Place the arrow tip at the target position whether the arrow tip is drawn or not
   const int target_balloontip_global_x =
       target_arrow_tip_position.x() - static_cast<int>(arrow_tip_x);
-  const int rightmost_valid_balloontip_global_x = screen_rect.width() - size_hint.width();
+  const int rightmost_valid_balloontip_global_x =
+      screen_rect.left() + screen_rect.width() - size_hint.width();
   // If the balloon would extend off the screen, push it left or right until it's not
   const int actual_balloontip_global_x =
-      std::max(0, std::min(rightmost_valid_balloontip_global_x, target_balloontip_global_x));
+      std::max(screen_rect.left(),
+               std::min(rightmost_valid_balloontip_global_x, target_balloontip_global_x));
   // The tip pixel should be in the middle of the control, and arrow_tip_exterior_y is at the bottom
   // of that pixel. When arrow_at_bottom is true the arrow is above arrow_tip_exterior_y and so the
   // tip pixel is in the right place, but when it's false the arrow is below arrow_tip_exterior_y

Source/Core/VideoBackends/Metal/MTLMain.mm

@@ -164,8 +164,23 @@ void Metal::VideoBackend::PrepareWindow(WindowSystemInfo& wsi)
     return;
   NSView* view = static_cast<NSView*>(wsi.render_surface);
   CAMetalLayer* layer = [CAMetalLayer layer];
+
+  Util::PopulateBackendInfo(&g_Config);
+
+  if (g_Config.backend_info.bSupportsHDROutput && g_Config.bHDR)
+  {
+    [layer setWantsExtendedDynamicRangeContent:YES];
+    [layer setPixelFormat:MTLPixelFormatRGBA16Float];
+
+    const CFStringRef name = kCGColorSpaceExtendedLinearSRGB;
+    CGColorSpaceRef colorspace = CGColorSpaceCreateWithName(name);
+    [layer setColorspace:colorspace];
+    CGColorSpaceRelease(colorspace);
+  }
+
   [view setWantsLayer:YES];
   [view setLayer:layer];
+
   wsi.render_surface = layer;
 #endif
 }

Source/Core/VideoBackends/Metal/MTLUtil.h

@@ -3,7 +3,9 @@
 
 #pragma once
 
+#include <AppKit/NSScreen.h>
 #include <Metal/Metal.h>
+
 #include <vector>
 
 #include "VideoCommon/AbstractShader.h"

Source/Core/VideoBackends/Metal/MTLUtil.mm

@@ -77,6 +77,8 @@ void Metal::Util::PopulateBackendInfo(VideoConfig* config)
   config->backend_info.bSupportsPartialMultisampleResolve = false;
   config->backend_info.bSupportsDynamicVertexLoader = true;
   config->backend_info.bSupportsVSLinePointExpand = true;
+  config->backend_info.bSupportsHDROutput =
+      1.0 < [[NSScreen deepestScreen] maximumPotentialExtendedDynamicRangeColorComponentValue];
 }
 
 void Metal::Util::PopulateBackendInfoAdapters(VideoConfig* config,

Source/Core/VideoCommon/VideoBackendBase.cpp

@@ -232,18 +232,17 @@ const std::vector<std::unique_ptr<VideoBackendBase>>& VideoBackendBase::GetAvail
   static auto s_available_backends = [] {
     std::vector<std::unique_ptr<VideoBackendBase>> backends;
 
-    // OGL > D3D11 > D3D12 > Vulkan > SW > Null
-    // On macOS, we prefer Vulkan over OpenGL due to OpenGL support being deprecated by Apple.
-#ifdef HAS_OPENGL
-    backends.push_back(std::make_unique<OGL::VideoBackend>());
-#endif
 #ifdef _WIN32
     backends.push_back(std::make_unique<DX11::VideoBackend>());
     backends.push_back(std::make_unique<DX12::VideoBackend>());
 #endif
+#ifdef HAS_OPENGL
+    backends.push_back(std::make_unique<OGL::VideoBackend>());
+#endif
 #ifdef HAS_VULKAN
 #ifdef __APPLE__
     // Emplace the Vulkan backend at the beginning so it takes precedence over OpenGL.
+    // On macOS, we prefer Vulkan over OpenGL due to OpenGL support being deprecated by Apple.
     backends.emplace(backends.begin(), std::make_unique<Vulkan::VideoBackend>());
 #else
     backends.push_back(std::make_unique<Vulkan::VideoBackend>());