Sync-change

Change-Id: I1854687d1a310f319718f8195848a241d6d22925

Author: sdolender

IGC/CMakeLists.txt

@@ -73,7 +73,6 @@ set(CMAKE_CXX_STANDARD_REQUIRED TRUE)
 
 # ================================================ Flags ===============================================
 
-
 # Escapes text for regular expressions.
 #
 # @param retValName Name of variable placeholder where result will be returned.

IGC/Compiler/CISACodeGen/DriverInfo.hpp

@@ -252,9 +252,6 @@ namespace IGC
         /// Allow aggressive vector value aliasing
         virtual bool EnableVecAliasing() const { return false; }
 
-        /// Prevent MAD optimization if result used in Floor
-        virtual bool PreventMadforRound() const { return false; }
-
 
 
     };

IGC/Compiler/CISACodeGen/PatternMatchPass.cpp

@@ -1797,7 +1797,7 @@ namespace IGC
             return I.getType()->isFloatingPointTy();
         };
 
-        if (isFpMad(I) && m_ctx->getModuleMetaData()->isPrecise)
+        if (isFpMad(I) && (m_ctx->getModuleMetaData()->isPrecise || m_ctx->getModuleMetaData()->compOpt.disableMathRefactoring))
         {
             return false;
         }
@@ -1819,12 +1819,6 @@ namespace IGC
             return false;
         }
 
-        bool preventMadForRound = false;
-        if (IGC_IS_FLAG_ENABLED(EnableMadRoundDepCheck) || m_ctx->m_DriverInfo.PreventMadforRound())
-        {
-            preventMadForRound = true;
-        }
-
         bool found = false;
         llvm::Value* sources[3];
         e_modifier src_mod[3];
@@ -1906,47 +1900,6 @@ namespace IGC
                 return false;
         }
 
-        if (found && preventMadForRound)
-        {
-            /*=============================================================================
-            This checks for pattern
-
-            mul x a b
-            add y x c
-            floor z y
-
-            ----or---
-
-            mul x a b
-            add y x c
-            floor z abs(y)
-
-            If the case falls in either of the two cases, Mad optimisation is skipped because
-            small precision difference between Mul+add and Mad can be extrapolated by floor
-            (currently just enabled for floor operation, if required add for ceiling in future)
-            =============================================================================*/
-
-            for (auto iter = I.user_begin(); iter != I.user_end(); iter++)
-            {
-                if (IntrinsicInst * source = dyn_cast<IntrinsicInst>(*iter))
-                {
-                    if (source->getIntrinsicID() == Intrinsic::fabs)
-                    {
-                        for (auto it = source->user_begin(); it != source->user_end(); it++)
-                        {
-                            IntrinsicInst* source_fabs = dyn_cast<IntrinsicInst>(*it);
-                            if (source_fabs && source_fabs->getIntrinsicID() == Intrinsic::floor)
-                                return false;
-                        }
-                    }
-                    if (source->getIntrinsicID() == Intrinsic::floor)
-                    {
-                        return false;
-                    }
-                }
-            }
-        }
-
         if (found)
         {
             MadPattern* pattern = new (m_allocator) MadPattern();

IGC/common/MDFrameWork.h

@@ -201,6 +201,7 @@ namespace IGC
         bool UniformWGS                                 = false;
         bool disableVertexComponentPacking              = false;
         bool PreferBindlessImages                       = false;
+        bool disableMathRefactoring                     = false;
     };
 
     struct ComputeShaderInfo

IGC/common/igc_flags.def

@@ -188,8 +188,6 @@ DECLARE_IGC_REGKEY(DWORD, RetryManagerFirstStateId,     0,     "For debugging pu
 DECLARE_IGC_REGKEY(bool, DisableSendSrcDstOverlapWA,    false, "Disable Send Source/destination overlap WA which is enabled for GEN10/GEN11 and whenever Wddm2Svm is set in WATable")
 DECLARE_IGC_REGKEY(debugString, DisablePassToggles,     0,     "Disable each IGC pass by setting the bit. HEXADECIMAL ONLY!. Ex: C0 is to disable pass 6 and pass 7.")
 DECLARE_IGC_REGKEY(bool, ForceStatelessForQueueT,       true,  "In OCL, force to use stateless memory to hold queue_t*. This is a legacy feature to be removed.")
-DECLARE_IGC_REGKEY(bool, EnableMadRoundDepCheck,        false,  "Enable Floor Dependency on Mad Check, Check was added since the floor was extrapolating the mad precision difference.")
-
 
 DECLARE_IGC_GROUP("Shader dumping")
 DECLARE_IGC_REGKEY(bool, EnableCosDump, false, "Enable cos dump")