[d3d9] Implement d3d9.longMad option

dxvk.conf

@@ -304,3 +304,14 @@
 
 # d3d9.forceSwapchainMSAA  = -1
 
+
+# Long Mad
+#
+# Should we make our Mads a FFma or do it the long way with an FMul and an FAdd?
+# This solves some rendering bugs in games that have z-pass shaders which
+# don't match entirely to the regular vertex shader in this way.
+#
+# Supported values:
+# - True/False
+
+# d3d9.longMad = False

src/d3d9/d3d9_options.cpp

@@ -69,6 +69,7 @@ namespace dxvk {
     this->allowDoNotWait        = config.getOption<bool>       ("d3d9.allowDoNotWait",     true);
     this->allowDiscard          = config.getOption<bool>       ("d3d9.allowDiscard",       true);
     this->enumerateByDisplays   = config.getOption<bool>       ("d3d9.enumerateByDisplays", true);
+    this->longMad               = config.getOption<bool>       ("d3d9.longMad",            false);
 
     // If we are not Nvidia, enable general hazards.
     this->generalHazards = adapter == nullptr || !adapter->matchesDriver(DxvkGpuVendor::Nvidia, VK_DRIVER_ID_NVIDIA_PROPRIETARY_KHR, 0, 0);

src/d3d9/d3d9_options.h

@@ -138,6 +138,11 @@ namespace dxvk {
 
     /// Enumerate adapters by displays
     bool enumerateByDisplays;
+
+    /// Should we make our Mads a FFma or do it the long way with an FMul and an FAdd?
+    /// This solves some rendering bugs in games that have z-pass shaders which
+    /// don't match entirely to the regular vertex shader in this way.
+    bool longMad;
   };
 
 }

src/dxso/dxso_compiler.cpp

@@ -1774,10 +1774,21 @@ namespace dxvk {
           emitRegisterLoad(src[1], mask).id);
         break;
       case DxsoOpcode::Mad:
-        result.id = m_module.opFFma(typeId,
-          emitRegisterLoad(src[0], mask).id,
-          emitRegisterLoad(src[1], mask).id,
-          emitRegisterLoad(src[2], mask).id);
+        if (!m_moduleInfo.options.longMad) {
+          result.id = m_module.opFFma(typeId,
+            emitRegisterLoad(src[0], mask).id,
+            emitRegisterLoad(src[1], mask).id,
+            emitRegisterLoad(src[2], mask).id);
+        }
+        else {
+          result.id = m_module.opFMul(typeId,
+            emitRegisterLoad(src[0], mask).id,
+            emitRegisterLoad(src[1], mask).id);
+
+          result.id = m_module.opFAdd(typeId,
+            result.id,
+            emitRegisterLoad(src[2], mask).id);
+        }
         break;
       case DxsoOpcode::Mul:
         result.id = m_module.opFMul(typeId,

src/dxso/dxso_options.cpp

@@ -45,6 +45,8 @@ namespace dxvk {
     forceSamplerTypeSpecConstants = options.forceSamplerTypeSpecConstants;
 
     vertexConstantBufferAsSSBO = pDevice->GetVertexConstantLayout().totalSize() > devInfo.core.properties.limits.maxUniformBufferRange;
+
+    longMad = options.longMad;
   }
 
 }

src/dxso/dxso_options.h

@@ -45,6 +45,11 @@ namespace dxvk {
 
     /// Should the VS constant buffer be an SSBO (swvp on NV)
     bool vertexConstantBufferAsSSBO;
+
+    /// Should we make our Mads a FFma or do it the long way with an FMul and an FAdd?
+    /// This solves some rendering bugs in games that have z-pass shaders which
+    /// don't match entirely to the regular vertex shader in this way.
+    bool longMad;
   };
 
 }