UberShaderPixel: always run indirect stage logic

Hardware testing has confirmed that fb_addprev and wrapping both run even when the indirect stage is disabled.

2 files changed, 17 insertions, 22 deletions

diff --git a/Source/Core/VideoCommon/PixelShaderManager.cpp b/Source/Core/VideoCommon/PixelShaderManager.cpp
index 6d52d8ffc1..0ae1ced1e4 100644
--- a/Source/Core/VideoCommon/PixelShaderManager.cpp
+++ b/Source/Core/VideoCommon/PixelShaderManager.cpp
@@ -148,29 +148,18 @@ void PixelShaderManager::SetConstants()
 
     for (u32 i = 0; i < (bpmem.genMode.numtevstages + 1); ++i)
     {
+      // Note: a tevind of zero just happens to be a passthrough, so no need
+      // to set an extra bit.  Furthermore, wrap and add to previous apply even if there is no
+      // indirect stage.
+      constants.pack1[i][2] = bpmem.tevind[i].hex;
+
       u32 stage = bpmem.tevind[i].bt;
-      if (stage < bpmem.genMode.numindstages)
-      {
-        // We set some extra bits so the ubershader can quickly check if these
-        // features are in use.
-        if (bpmem.tevind[i].IsActive())
-          constants.pack1[stage][3] =
-              bpmem.tevindref.getTexCoord(stage) | bpmem.tevindref.getTexMap(stage) << 8 | 1 << 16;
-        // Note: a tevind of zero just happens to be a passthrough, so no need
-        // to set an extra bit.
-        constants.pack1[i][2] = bpmem.tevind[i].hex;  // TODO: This match shadergen, but videosw
-                                                      // will always wrap.
-
-        // The ubershader uses tevind != 0 as a condition whether to calculate texcoords,
-        // even when texture is disabled, instead of the stage < bpmem.genMode.numindstages.
-        // We set an unused bit here to indicate that the stage is active, even if it
-        // is just a pass-through.
-        constants.pack1[i][2] |= 0x80000000;
-      }
-      else
-      {
-        constants.pack1[i][2] = 0;
-      }
+
+      // We use an extra bit (1 << 16) to provide a fast way of testing if this feature is in use.
+      // Note also that this is indexed by indirect stage, not by TEV stage.
+      if (bpmem.tevind[i].IsActive() && stage < bpmem.genMode.numindstages)
+        constants.pack1[stage][3] =
+            bpmem.tevindref.getTexCoord(stage) | bpmem.tevindref.getTexMap(stage) << 8 | 1 << 16;
     }
 
     dirty = true;
diff --git a/Source/Core/VideoCommon/UberShaderPixel.cpp b/Source/Core/VideoCommon/UberShaderPixel.cpp
index a029f58812..642896a3dc 100644
--- a/Source/Core/VideoCommon/UberShaderPixel.cpp
+++ b/Source/Core/VideoCommon/UberShaderPixel.cpp
@@ -287,6 +287,8 @@ ShaderCode GenPixelShader(APIType ApiType, const ShaderHostConfig& host_config,
   // ======================
   const auto LookupIndirectTexture = [&out, stereo](std::string_view out_var_name,
                                                     std::string_view in_index_name) {
+    // in_index_name is the indirect stage, not the tev stage
+    // bpmem_iref is packed differently from RAS1_IREF
     out.Write("{{\n"
               "  uint iref = bpmem_iref({});\n"
               "  if ( iref != 0u)\n"
@@ -304,6 +306,10 @@ ShaderCode GenPixelShader(APIType ApiType, const ShaderHostConfig& host_config,
               "[texmap].xy, {})).abg;\n",
               in_index_name, in_index_name, in_index_name, in_index_name, out_var_name,
               stereo ? "float(layer)" : "0.0");
+    // There is always a bit set in bpmem_iref if the data is valid (matrix is not off, and the
+    // indirect texture stage is enabled). If the matrix is off, the result doesn't matter; if the
+    // indirect texture stage is disabled, the result is undefined (and produces a glitchy pattern
+    // on hardware, different from this).
     out.Write("  }}\n"
               "  else\n"
               "  {{\n"