VideoCommon: Work around Adreno TEV zero input miscompilation

Some Qualcomm Android drivers from version 700 onward miscompile
specialized TEV pixel shaders when color input A or B is TevColorArg::Zero.
Use konstLookup[8].rgb, which PixelShaderManager initializes to zero,
to keep these inputs dependent on a uniform value.

Use the existing driver bug gate for OpenGL and Vulkan. Other inputs and
unaffected drivers retain their original shader expressions.

4 files changed, 35 insertions, 7 deletions

diff --git a/Source/Core/VideoCommon/ConstantManager.h b/Source/Core/VideoCommon/ConstantManager.h
index 14c8732580..ec587411dc 100644
--- a/Source/Core/VideoCommon/ConstantManager.h
+++ b/Source/Core/VideoCommon/ConstantManager.h
@@ -33,7 +33,8 @@ struct alignas(16) PixelShaderConstants
   float4 zslope;
   std::array<float, 2> efbscale;  // .xy
 
-  // Constants from here onwards are only used in ubershaders, other than pack2.
+  // Constants from here onwards are mainly used in ubershaders. Specialized shaders also use
+  // pack2 and konst.
   u32 genmode;                  // .z
   u32 alphaTest;                // .w
   u32 fogParam3;                // .x
diff --git a/Source/Core/VideoCommon/DriverDetails.cpp b/Source/Core/VideoCommon/DriverDetails.cpp
index b657d7721c..3027aab154 100644
--- a/Source/Core/VideoCommon/DriverDetails.cpp
+++ b/Source/Core/VideoCommon/DriverDetails.cpp
@@ -161,6 +161,10 @@ constexpr BugInfo m_known_bugs[] = {
      BUG_BROKEN_DYNAMIC_SAMPLER_INDEXING, -1.0, -1.0, true},
     {API_VULKAN, OS_ANDROID, VENDOR_QUALCOMM, DRIVER_QUALCOMM, Family::UNKNOWN,
      BUG_SLOW_OPTIMAL_IMAGE_TO_BUFFER_COPY, -1.0, -1.0, true},
+    {API_OPENGL, OS_ANDROID, VENDOR_QUALCOMM, DRIVER_QUALCOMM, Family::UNKNOWN,
+     BUG_BROKEN_TEV_COLOR_ZERO, 700.0, -1.0, true},
+    {API_VULKAN, OS_ANDROID, VENDOR_QUALCOMM, DRIVER_QUALCOMM, Family::UNKNOWN,
+     BUG_BROKEN_TEV_COLOR_ZERO, 0x802bc000, -1.0, true},
     {API_VULKAN, OS_ALL, VENDOR_ATI, DRIVER_ATI, Family::UNKNOWN, BUG_BROKEN_DEPTH_CLAMP_CONTROL,
      -1.0, -1.0, true},
 };
@@ -312,6 +316,7 @@ static const char* to_string(Bug bug)
     case BUG_BROKEN_DISCARD_WITH_EARLY_Z:                return "broken-discard-with-early-z";
     case BUG_BROKEN_DYNAMIC_SAMPLER_INDEXING:            return "broken-dynamic-sampler-indexing";
     case BUG_SLOW_OPTIMAL_IMAGE_TO_BUFFER_COPY:          return "slow-optimal-image-to-buffer-copy";
+    case BUG_BROKEN_TEV_COLOR_ZERO:                      return "broken-tev-color-zero";
     case BUG_BROKEN_DEPTH_CLAMP_CONTROL:                 return "broken-depth-clamp-control";
   }
   return "Unknown";
diff --git a/Source/Core/VideoCommon/DriverDetails.h b/Source/Core/VideoCommon/DriverDetails.h
index 64ed656f3f..39eb86eb59 100644
--- a/Source/Core/VideoCommon/DriverDetails.h
+++ b/Source/Core/VideoCommon/DriverDetails.h
@@ -344,6 +344,14 @@ enum Bug
   // Ended Version: -1
   BUG_SLOW_OPTIMAL_IMAGE_TO_BUFFER_COPY,
 
+  // BUG: Some Qualcomm Android drivers miscompile specialized TEV pixel shaders when
+  // color input A or B is Zero.
+  // Use the konst lookup table to provide 0 at runtime and avoid the miscompile.
+  // Affected devices: Qualcomm drivers on Android.
+  // Started Version: 700 (0x802bc000 for Vulkan)
+  // Ended Version: -1
+  BUG_BROKEN_TEV_COLOR_ZERO,
+
   // BUG: Incorrect implementation of VK_EXT_depth_clamp_control causes incorrect depth values to
   // be written to the depth buffer.
   // Affected devices: Official AMD (RADV is unaffected)
diff --git a/Source/Core/VideoCommon/PixelShaderGen.cpp b/Source/Core/VideoCommon/PixelShaderGen.cpp
index 5ce120e9ff..669d80e317 100644
--- a/Source/Core/VideoCommon/PixelShaderGen.cpp
+++ b/Source/Core/VideoCommon/PixelShaderGen.cpp
@@ -1434,20 +1434,34 @@ static void WriteStage(ShaderCode& out, const pixel_shader_uid_data* uid_data, i
   if (ac.dest >= TevOutput::Color0)
     out.SetConstantsUsed(C_COLORS + u32(ac.dest.Value()), C_COLORS + u32(ac.dest.Value()));
 
+  const char* color_a = tev_c_input_table[cc.a];
+  const char* color_b = tev_c_input_table[cc.b];
+  if ((cc.a == TevColorArg::Zero || cc.b == TevColorArg::Zero) &&
+      DriverDetails::HasBug(DriverDetails::BUG_BROKEN_TEV_COLOR_ZERO))
+  {
+    // The Adreno shader compiler incorrectly optimizes the int3(0,0,0) constant, causing incorrect
+    // colors. To work around this, we trick the compiler by using konstLookup[8], which
+    // PixelShaderManager initializes to zero, to keep these inputs dependent on a uniform value.
+    // Slot 8 is safe because it is part of the invalid konst range (reads as zero on hardware).
+    // It is never written at runtime and therefore stays zero.
+    if (cc.a == TevColorArg::Zero)
+      color_a = "konstLookup[8].rgb";
+    if (cc.b == TevColorArg::Zero)
+      color_b = "konstLookup[8].rgb";
+  }
+
   if (DriverDetails::HasBug(DriverDetails::BUG_BROKEN_VECTOR_BITWISE_AND))
   {
-    out.Write("\ttevin_a = int4({} & 255, {} & 255);\n", tev_c_input_table[cc.a],
-              tev_a_input_table[ac.a]);
-    out.Write("\ttevin_b = int4({} & 255, {} & 255);\n", tev_c_input_table[cc.b],
-              tev_a_input_table[ac.b]);
+    out.Write("\ttevin_a = int4({} & 255, {} & 255);\n", color_a, tev_a_input_table[ac.a]);
+    out.Write("\ttevin_b = int4({} & 255, {} & 255);\n", color_b, tev_a_input_table[ac.b]);
     out.Write("\ttevin_c = int4({} & 255, {} & 255);\n", tev_c_input_table[cc.c],
               tev_a_input_table[ac.c]);
   }
   else
   {
-    out.Write("\ttevin_a = int4({}, {})&int4(255, 255, 255, 255);\n", tev_c_input_table[cc.a],
+    out.Write("\ttevin_a = int4({}, {})&int4(255, 255, 255, 255);\n", color_a,
               tev_a_input_table[ac.a]);
-    out.Write("\ttevin_b = int4({}, {})&int4(255, 255, 255, 255);\n", tev_c_input_table[cc.b],
+    out.Write("\ttevin_b = int4({}, {})&int4(255, 255, 255, 255);\n", color_b,
               tev_a_input_table[ac.b]);
     out.Write("\ttevin_c = int4({}, {})&int4(255, 255, 255, 255);\n", tev_c_input_table[cc.c],
               tev_a_input_table[ac.c]);