Merge pull request #10290 from OatmealDome/m1-earlyz-bug

DriverDetails: Add broken discard with early-Z bug on Apple Silicon GPUs

1 files changed, 37 insertions, 14 deletions

diff --git a/Source/Core/VideoCommon/PixelShaderGen.cpp b/Source/Core/VideoCommon/PixelShaderGen.cpp
index 82de68a290..4a9e9105a8 100644
--- a/Source/Core/VideoCommon/PixelShaderGen.cpp
+++ b/Source/Core/VideoCommon/PixelShaderGen.cpp
@@ -321,7 +321,9 @@ PixelShaderUid GetPixelShaderUid()
   BlendingState state = {};
   state.Generate(bpmem);
 
-  if (state.usedualsrc && state.dstalpha && g_ActiveConfig.backend_info.bSupportsFramebufferFetch &&
+  if (((state.usedualsrc && state.dstalpha) ||
+       DriverDetails::HasBug(DriverDetails::BUG_BROKEN_DISCARD_WITH_EARLY_Z)) &&
+      g_ActiveConfig.backend_info.bSupportsFramebufferFetch &&
       !g_ActiveConfig.backend_info.bSupportsDualSourceBlend)
   {
     uid_data->blend_enable = state.blendenable;
@@ -944,10 +946,15 @@ ShaderCode GeneratePixelShaderCode(APIType api_type, const ShaderHostConfig& hos
       (!DriverDetails::HasBug(DriverDetails::BUG_BROKEN_DUAL_SOURCE_BLENDING) ||
        uid_data->useDstAlpha);
   const bool use_shader_blend =
-      !use_dual_source && uid_data->useDstAlpha && host_config.backend_shader_framebuffer_fetch;
+      !use_dual_source &&
+      (uid_data->useDstAlpha ||
+       DriverDetails::HasBug(DriverDetails::BUG_BROKEN_DISCARD_WITH_EARLY_Z)) &&
+      host_config.backend_shader_framebuffer_fetch;
   const bool use_shader_logic_op = !host_config.backend_logic_op && uid_data->logic_op_enable &&
                                    host_config.backend_shader_framebuffer_fetch;
-  const bool use_framebuffer_fetch = use_shader_blend || use_shader_logic_op;
+  const bool use_framebuffer_fetch =
+      use_shader_blend || use_shader_logic_op ||
+      DriverDetails::HasBug(DriverDetails::BUG_BROKEN_DISCARD_WITH_EARLY_Z);
 
   if (api_type == APIType::OpenGL || api_type == APIType::Vulkan)
   {
@@ -1869,9 +1876,23 @@ static void WriteAlphaTest(ShaderCode& out, const pixel_shader_uid_data* uid_dat
   // ZCOMPLOC HACK:
   if (!uid_data->alpha_test_use_zcomploc_hack)
   {
-    out.Write("\t\tdiscard;\n");
-    if (api_type == APIType::D3D)
+#ifdef __APPLE__
+    if (uid_data->forced_early_z &&
+        DriverDetails::HasBug(DriverDetails::BUG_BROKEN_DISCARD_WITH_EARLY_Z))
+    {
+      // Instead of using discard, fetch the framebuffer's color value and use it as the output
+      // for this fragment.
+      out.Write("\t\t{} = float4(initial_ocol0.xyz, 1.0);\n",
+                use_dual_source ? "real_ocol0" : "ocol0");
       out.Write("\t\treturn;\n");
+    }
+    else
+#endif
+    {
+      out.Write("\t\tdiscard;\n");
+      if (api_type == APIType::D3D)
+        out.Write("\t\treturn;\n");
+    }
   }
 
   out.Write("\t}}\n");
@@ -2018,8 +2039,8 @@ static void WriteBlend(ShaderCode& out, const pixel_shader_uid_data* uid_data)
         "float3(1,1,1);",                      // ONE
         "initial_ocol0.rgb;",                  // DSTCLR
         "float3(1,1,1) - initial_ocol0.rgb;",  // INVDSTCLR
-        "ocol1.aaa;",                          // SRCALPHA
-        "float3(1,1,1) - ocol1.aaa;",          // INVSRCALPHA
+        "src_color.aaa;",                      // SRCALPHA
+        "float3(1,1,1) - src_color.aaa;",      // INVSRCALPHA
         "initial_ocol0.aaa;",                  // DSTALPHA
         "float3(1,1,1) - initial_ocol0.aaa;",  // INVDSTALPHA
     };
@@ -2028,8 +2049,8 @@ static void WriteBlend(ShaderCode& out, const pixel_shader_uid_data* uid_data)
         "1.0;",                    // ONE
         "initial_ocol0.a;",        // DSTCLR
         "1.0 - initial_ocol0.a;",  // INVDSTCLR
-        "ocol1.a;",                // SRCALPHA
-        "1.0 - ocol1.a;",          // INVSRCALPHA
+        "src_color.a;",            // SRCALPHA
+        "1.0 - src_color.a;",      // INVSRCALPHA
         "initial_ocol0.a;",        // DSTALPHA
         "1.0 - initial_ocol0.a;",  // INVDSTALPHA
     };
@@ -2038,8 +2059,8 @@ static void WriteBlend(ShaderCode& out, const pixel_shader_uid_data* uid_data)
         "float3(1,1,1);",                      // ONE
         "ocol0.rgb;",                          // SRCCLR
         "float3(1,1,1) - ocol0.rgb;",          // INVSRCCLR
-        "ocol1.aaa;",                          // SRCALHA
-        "float3(1,1,1) - ocol1.aaa;",          // INVSRCALPHA
+        "src_color.aaa;",                      // SRCALHA
+        "float3(1,1,1) - src_color.aaa;",      // INVSRCALPHA
         "initial_ocol0.aaa;",                  // DSTALPHA
         "float3(1,1,1) - initial_ocol0.aaa;",  // INVDSTALPHA
     };
@@ -2048,12 +2069,14 @@ static void WriteBlend(ShaderCode& out, const pixel_shader_uid_data* uid_data)
         "1.0;",                    // ONE
         "ocol0.a;",                // SRCCLR
         "1.0 - ocol0.a;",          // INVSRCCLR
-        "ocol1.a;",                // SRCALPHA
-        "1.0 - ocol1.a;",          // INVSRCALPHA
+        "src_color.a;",            // SRCALPHA
+        "1.0 - src_color.a;",      // INVSRCALPHA
         "initial_ocol0.a;",        // DSTALPHA
         "1.0 - initial_ocol0.a;",  // INVDSTALPHA
     };
-    out.Write("\tfloat4 blend_src;\n");
+    out.Write("\tfloat4 src_color = {};\n"
+              "\tfloat4 blend_src;",
+              uid_data->useDstAlpha ? "ocol1" : "ocol0");
     out.Write("\tblend_src.rgb = {}\n", blend_src_factor[uid_data->blend_src_factor]);
     out.Write("\tblend_src.a = {}\n", blend_src_factor_alpha[uid_data->blend_src_factor_alpha]);
     out.Write("\tfloat4 blend_dst;\n");