Updating LLVM patches names for easier maintenance

Change-Id: Ib1fe788e541b91e1cb880a729a52f0ad28dea84d

Author: AnupamaChandrasekhar

releases/4.0.0/patches_external/01-Adds-a-special-field-into-value-class.patch

@@ -0,0 +1,17 @@
+#    Description                  : stripPointerCastsAndOffsets should not process across AddrSpaceCast
+
+diff -Naur --strip-trailing-cr a/lib/IR/Value.cpp b/lib/IR/Value.cpp
+--- a/lib/IR/Value.cpp	2016-12-07 13:47:32.000000000 -0800
++++ b/lib/IR/Value.cpp	2018-03-12 11:25:42.752625100 -0700
+@@ -466,6 +466,11 @@
+       V = GEP->getPointerOperand();
+     } else if (Operator::getOpcode(V) == Instruction::BitCast ||
+                Operator::getOpcode(V) == Instruction::AddrSpaceCast) {
++      // Do not process across AddrSpaceCast, since its result could
++      // be a nullptr depending on the meaning of src and dest addrspace.
++      if (Operator::getOpcode(V) == Instruction::AddrSpaceCast) {
++         return V;
++      }
+       V = cast<Operator>(V)->getOperand(0);
+     } else if (GlobalAlias *GA = dyn_cast<GlobalAlias>(V)) {
+       if (StripKind == PSK_ZeroIndices || GA->isInterposable())

releases/4.0.0/patches_external/04-stripPointerCastsAndOffsets-should-not.patch

@@ -0,0 +1,19 @@
+#    Description                  : Enable aggressive (gep (gep base, idx0), idx1) combining
+
+diff -Naur --strip-trailing-cr a/lib/Transforms/InstCombine/InstructionCombining.cpp b/lib/Transforms/InstCombine/InstructionCombining.cpp
+--- a/lib/Transforms/InstCombine/InstructionCombining.cpp	2017-10-24 08:51:48.146973901 -0400
++++ b/lib/Transforms/InstCombine/InstructionCombining.cpp	2017-10-24 08:52:36.064972024 -0400
+@@ -1573,11 +1573,13 @@
+         // normalized.
+         if (SO1->getType() != GO1->getType())
+           return nullptr;
++#if 0 
+         // Only do the combine when GO1 and SO1 are both constants. Only in
+         // this case, we are sure the cost after the merge is never more than
+         // that before the merge.
+         if (!isa<Constant>(GO1) || !isa<Constant>(SO1))
+           return nullptr;
++#endif 
+         Sum = Builder->CreateAdd(SO1, GO1, PtrOp->getName()+".sum");
+       }
+ 

releases/4.0.0/patches_external/1-stripPointerCastsAndOffsets-should-not.patch

@@ -2,12 +2,6 @@
 #                                 This works around the issue with non-uniform constant-buffer
 #                                 index with an OGL test. The non-uniform buffer indexing is
 #                                 created due to code sinking.
-#    Type                       : Backport from LLVM 3.8
-#    P4V CL                     : 552325(3.8.0), 642679(4.0.0)
-#    Category                   : Custom
-#    Author                     : gangche1
-#    Open Source Classification : Internal only
-#    Status                     : Performance related, need not be ported to opensoure repository.
 
 diff -Naur --strip-trailing-cr a/lib/Transforms/InstCombine/InstructionCombining.cpp b/lib/Transforms/InstCombine/InstructionCombining.cpp
 --- a/lib/Transforms/InstCombine/InstructionCombining.cpp	2017-10-23 09:48:44.308224955 -0400

releases/4.0.0/patches_external/13-managed-static-mem-leak-fix.patch

@@ -1,17 +1,5 @@
 #    Description                  : SimplifyCFG::SinkThenElseCodeToEnd does not sink code similiar 
 #                                   to LLVM 3.8 
-#    When Fixed in Open Source    : Needed until we have resolved gen intrinsic handling with 
-#                                   LLVM 4.0 and resolved code sinking.
-#    P4V CL                       : 688188(4.0.0)
-#    Category                     : Optimization
-#    Author                       : pmistry
-#    Open Source Classification   : Internal Only
-#    Impact                       : Performance/Functional
-#    Failing Workload/Commandline : Ashes of Singularity Escalation.dx12-g1
-#    Notes                        : SimplifyCFG::SinkThenElseCodeToEnd does not sink code similiar to 
-#                                   LLVM 3.8 in certain cases. In case of GenIntrinsics it sinks and the 
-#                                   same is not handled properly by IGC passes later on. Benchmark that 
-#                                   exposed the issue was Ashes of Singularity Escalation.dx12-g1
 
 diff -Naur --strip-trailing-cr a/lib/Transforms/Utils/SimplifyCFG.cpp b/lib/Transforms/Utils/SimplifyCFG.cpp
 --- a/lib/Transforms/Utils/SimplifyCFG.cpp	2017-10-23 09:53:04.934214744 -0400

releases/4.0.0/patches_external/14-unwind-fix.patch

@@ -1,17 +1,6 @@
 #    Description                  : Added checks in SimplifyCFG, GVN, and EarlyCSE to prevent hoisting 
 #                                   or sinking convergent functions out of or into control flow (all 
 #                                   wave intrinsics are marked as convergent).
-#    When Fixed in Open Source    : Needed till a cleaner solution is worked out.
-#    P4V CL                       : 696122(4.0.0)
-#    Category                     : Custom
-#    Author                       : djwoo
-#    Open Source Classification   : Common
-#    Impact                       : Functional
-#    Failing Workload/Commandline : TE.exe D3DConf_12_Core.dll /enablewttlogging /appendwttlogging /name:DXILConfTest::WaveIntrinsics*
-#    Notes                        : Need further guidance from the community on whether the convergent 
-#                                   function attribute should cover this case or whether there needs to 
-#                                   be a new mechanism to handle it. The current suggestion of adding 
-#                                   inline asm at every relevant call site is hackish.
 
 diff -Naur --strip-trailing-cr a/lib/Transforms/Scalar/EarlyCSE.cpp b/lib/Transforms/Scalar/EarlyCSE.cpp
 --- a/lib/Transforms/Scalar/EarlyCSE.cpp	2017-10-23 14:47:17.025523189 -0400

releases/4.0.0/patches_external/2_1-enable-aggressive-combining.patch

@@ -1,14 +1,4 @@
 #    Description                  : Universal Driver Enabling
-#    When Fixed in Open Source    : Needed always
-#    P4V CL                       : 706504(4.0.0)
-#    Category                     : Build
-#    Author                       : juanrod2
-#    Open Source Classification   : Internal Only
-#    Impact                       : 
-#    Failing Workload/Commandline : 
-#    Notes                        : Process::GetArgumentVector for Windows has been 
-#                                   commented out since it is calling CommandLineToArgvW 
-#                                   which is not Universal API compliant.
 
 diff -Naur --strip-trailing-cr a/lib/Support/Windows/Process.inc b/lib/Support/Windows/Process.inc
 --- a/lib/Support/Windows/Process.inc	2017-10-27 09:51:01.756680018 -0400

releases/4.0.0/patches_external/03-completely-turn-off-code-sinking-in-InstructionCombining.patch releases/4.0.0/patches_external/2_2-completely-turn-off-code-sinking-in-InstructionCombining.patch

@@ -1,7 +1,4 @@
 #    Description                  : Adds AllowReassoc, AllowContract and ApproxFunc to FastMathFlags
-#    When Fixed in LLVM Core      : 22fddd859bee89a7950cb3dfbeb72cd6072656ee (03/29/2017),  
-#                                 : 00e900afdbd5dc97330de6bc0b8b09db1dcac9f7 (11/06/2017)
-#    Category                     : Enhanchement
 
 diff --git a/docs/LangRef.rst b/docs/LangRef.rst
 --- a/docs/LangRef.rst