DriverDetails: Fix broken vector bitwise AND on Mali drivers

dolphin-emu · Apr 11, 2021 · fcaea7d · fcaea7d
1 parent 5322256
commit fcaea7d
Show file tree

Hide file tree

Showing 3 changed files with 31 additions and 6 deletions.
diff --git a/Source/Core/VideoCommon/DriverDetails.cpp b/Source/Core/VideoCommon/DriverDetails.cpp
@@ -122,6 +122,10 @@ constexpr BugInfo m_known_bugs[] = {
      -1.0, -1.0, true},
     {API_VULKAN, OS_ALL, VENDOR_QUALCOMM, DRIVER_QUALCOMM, Family::UNKNOWN,
      BUG_SLOW_CACHED_READBACK_MEMORY, -1.0, -1.0, true},
+    {API_OPENGL, OS_ALL, VENDOR_ARM, DRIVER_ARM, Family::UNKNOWN, BUG_BROKEN_VECTOR_BITWISE_AND,
+     -1.0, -1.0, true},
+    {API_VULKAN, OS_ALL, VENDOR_ARM, DRIVER_ARM, Family::UNKNOWN, BUG_BROKEN_VECTOR_BITWISE_AND,
+     -1.0, -1.0, true},
 };
 
 static std::map<Bug, BugInfo> m_bugs;

diff --git a/Source/Core/VideoCommon/DriverDetails.h b/Source/Core/VideoCommon/DriverDetails.h
@@ -288,6 +288,16 @@ enum Bug
   // Mali Vulkan driver, causing high CPU usage in the __pi___inval_cache_range kernel
   // function. This flag causes readback buffers to select the coherent type.
   BUG_SLOW_CACHED_READBACK_MEMORY,
+
+  // BUG: Apparently ARM Mali GLSL compiler managed to break bitwise AND operations between
+  // two integers vectors, when one of them is non-constant (though the exact cases of when
+  // this occurs are still unclear to me). The resulting vector from the operation will be
+  // the constant vector.
+  // Easy enough to fix, just do the bitwise AND operation on the vector components first and
+  // then construct the final vector.
+  // Started version: -1
+  // Ended version: -1
+  BUG_BROKEN_VECTOR_BITWISE_AND,
 };
 
 // Initializes our internal vendor, device family, and driver version

diff --git a/Source/Core/VideoCommon/PixelShaderGen.cpp b/Source/Core/VideoCommon/PixelShaderGen.cpp
@@ -1252,12 +1252,23 @@ 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()));
 
-  out.Write("\ttevin_a = int4({}, {})&int4(255, 255, 255, 255);\n",
-            tev_c_input_table[u32(cc.a.Value())], tev_a_input_table[u32(ac.a.Value())]);
-  out.Write("\ttevin_b = int4({}, {})&int4(255, 255, 255, 255);\n",
-            tev_c_input_table[u32(cc.b.Value())], tev_a_input_table[u32(ac.b.Value())]);
-  out.Write("\ttevin_c = int4({}, {})&int4(255, 255, 255, 255);\n",
-            tev_c_input_table[u32(cc.c.Value())], tev_a_input_table[u32(ac.c.Value())]);
+  if (DriverDetails::HasBug(DriverDetails::BUG_BROKEN_VECTOR_BITWISE_AND)) {
+    out.Write("\ttevin_a = int4({} & 255, {} & 255);\n",
+              tev_c_input_table[u32(cc.a.Value())], tev_a_input_table[u32(ac.a.Value())]);
+    out.Write("\ttevin_b = int4({} & 255, {} & 255);\n",
+              tev_c_input_table[u32(cc.b.Value())], tev_a_input_table[u32(ac.b.Value())]);
+    out.Write("\ttevin_c = int4({} & 255, {} & 255);\n",
+              tev_c_input_table[u32(cc.c.Value())], tev_a_input_table[u32(ac.c.Value())]);
+  }
+  else
+  {
+    out.Write("\ttevin_a = int4({}, {})&int4(255, 255, 255, 255);\n",
+              tev_c_input_table[u32(cc.a.Value())], tev_a_input_table[u32(ac.a.Value())]);
+    out.Write("\ttevin_b = int4({}, {})&int4(255, 255, 255, 255);\n",
+              tev_c_input_table[u32(cc.b.Value())], tev_a_input_table[u32(ac.b.Value())]);
+    out.Write("\ttevin_c = int4({}, {})&int4(255, 255, 255, 255);\n",
+              tev_c_input_table[u32(cc.c.Value())], tev_a_input_table[u32(ac.c.Value())]);
+  }
   out.Write("\ttevin_d = int4({}, {});\n", tev_c_input_table[u32(cc.d.Value())],
             tev_a_input_table[u32(ac.d.Value())]);