fix julia

EnzymeAD · Oct 12, 2019 · ea74bd1 · ea74bd1
1 parent d04d3e4
commit ea74bd1
Show file tree

Hide file tree

Showing 8 changed files with 25 additions and 132 deletions.
diff --git a/.github/workflows/julia.yml b/.github/workflows/julia.yml
@@ -44,6 +44,8 @@ jobs:
         run: |
           mkdir build
           cd build
+          cmake --version
+          ls /home/runner/work/Enzyme/Enzyme/enzyme/Enzyme/
           cmake -DLLVM_DIR=../contrib/usr/lib/cmake/llvm -DLLVM_EXTERNAL_LIT=../contrib/usr/tools/lit/lit.py ../enzyme
           make -j
       - name: "Julia tests"

diff --git a/enzyme/Enzyme/ActiveVariable.cpp b/enzyme/Enzyme/ActiveVariable.cpp
@@ -27,6 +27,7 @@
 #include "llvm/IR/Type.h"
 #include "llvm/IR/Value.h"
 
+#include "llvm/Support/raw_ostream.h"
 
 #include "ActiveVariable.h"
 #include "Utils.h"

diff --git a/enzyme/Enzyme/CMakeLists.txt b/enzyme/Enzyme/CMakeLists.txt
@@ -2,12 +2,14 @@
 #  set(LLVM_LINK_COMPONENTS Core Support)
 #endif()
 
-file(GLOB ENZYME_SRC
+file(GLOB ENZYME_SRC RELATIVE ${CMAKE_CURRENT_SOURCE_DIR}
     "*.cpp"
 )
 
 list(APPEND ENZYME_SRC SCEV/ScalarEvolutionExpander.cpp)
 
+message("found enzyme sources " ${ENZYME_SRC})
+
 if (${LLVM_VERSION_MAJOR} LESS 8)
     add_llvm_loadable_module( LLVMEnzyme-${LLVM_VERSION_MAJOR}
         ${ENZYME_SRC}

diff --git a/enzyme/Enzyme/Enzyme.cpp b/enzyme/Enzyme/Enzyme.cpp
@@ -33,7 +33,6 @@
 
 #include "llvm/Support/Debug.h"
 #include "llvm/Transforms/Scalar.h"
-#include "llvm/Transforms/Utils.h"
 
 #include "llvm/Analysis/GlobalsModRef.h"
 #include "llvm/Analysis/ScalarEvolution.h"
@@ -222,7 +221,6 @@ class Enzyme : public FunctionPass {
     AU.addRequired<TargetLibraryInfoWrapperPass>();
     AU.addRequired<AAResultsWrapperPass>();
     AU.addRequired<GlobalsAAWrapperPass>();
-    AU.addRequiredID(LoopSimplifyID);
     //AU.addRequiredID(LCSSAID);
 
     AU.addRequired<LoopInfoWrapperPass>();

diff --git a/enzyme/Enzyme/LibraryFuncs.h b/enzyme/Enzyme/LibraryFuncs.h
@@ -20,114 +20,6 @@
 
 #include "llvm/Analysis/TargetLibraryInfo.h"
 
-/*
-//For updating below one should read MemoryBuiltins.cpp, TargetLibraryInfo.cpp
-static inline
-bool mapAllocationToDeallocation(const llvm::LibFunc libfunc, const llvm::TargetLibraryInfo &TLI) {
-	
-
-
-  // void operator delete[](void*);
-  case LibFunc_msvc_delete_array_ptr32:
-  // void operator delete[](void*);
-  case LibFunc_msvc_delete_array_ptr64:
-  // void operator delete(void*);
-  case LibFunc_msvc_delete_ptr32:
-  // void operator delete(void*);
-  case LibFunc_msvc_delete_ptr64:
-    return (NumParams == 1 && FTy.getParamType(0)->isPointerTy());
-
-  // void operator delete[](void*, unsigned long);
-  case LibFunc_ZdaPvm:
-  // void operator delete(void*, nothrow);
-  case LibFunc_ZdlPvRKSt9nothrow_t:
-  // void operator delete(void*, unsigned int);
-  case LibFunc_ZdlPvj:
-  // void operator delete(void*, unsigned long);
-  case LibFunc_ZdlPvm:
-  // void operator delete(void*, align_val_t)
-  case LibFunc_ZdlPvSt11align_val_t:
-  // void operator delete[](void*, align_val_t)
-  case LibFunc_ZdaPvSt11align_val_t:
-  // void operator delete[](void*, unsigned int);
-  case LibFunc_msvc_delete_array_ptr32_int:
-  // void operator delete[](void*, nothrow);
-  case LibFunc_msvc_delete_array_ptr32_nothrow:
-  // void operator delete[](void*, unsigned long long);
-  case LibFunc_msvc_delete_array_ptr64_longlong:
-  // void operator delete[](void*, nothrow);
-  case LibFunc_msvc_delete_array_ptr64_nothrow:
-  // void operator delete(void*, unsigned int);
-  case LibFunc_msvc_delete_ptr32_int:
-  // void operator delete(void*, nothrow);
-  case LibFunc_msvc_delete_ptr32_nothrow:
-  // void operator delete(void*, unsigned long long);
-  case LibFunc_msvc_delete_ptr64_longlong:
-  // void operator delete(void*, nothrow);
-  case LibFunc_msvc_delete_ptr64_nothrow:
-    return (NumParams == 2 && FTy.getParamType(0)->isPointerTy());
-
-  // void operator delete(void*, align_val_t, nothrow)
-  case LibFunc_ZdlPvSt11align_val_tRKSt9nothrow_t:
-  // void operator delete[](void*, align_val_t, nothrow)
-  case LibFunc_ZdaPvSt11align_val_tRKSt9nothrow_t:
-
-	switch (libfunc) {
-  		case LibFunc_malloc: // malloc(unsigned int);
-		case LibFunc_valloc: // valloc(unsigned int);
-			return LibFunc_free;
-
-		case LibFunc_Znwj: // new(unsigned int);
-			return LibFunc_ZdlPv; // void operator delete(void*);
-
-		case LibFunc_ZnwjRKSt9nothrow_t: // new(unsigned int, nothrow);
-		case LibFunc_ZnwjSt11align_val_t: // new(unsigned int, align_val_t)
-		case LibFunc_ZnwjSt11align_val_tRKSt9nothrow_t: // new(unsigned int, align_val_t, nothrow)
-
-		case LibFunc_Znwm: // new(unsigned long);
-		case LibFunc_ZnwmRKSt9nothrow_t: // new(unsigned long, nothrow);
-		case LibFunc_ZnwmSt11align_val_t: // new(unsigned long, align_val_t)
-		case LibFunc_ZnwmSt11align_val_tRKSt9nothrow_t: // new(unsigned long, align_val_t, nothrow)
-
-		case LibFunc_Znaj: // new[](unsigned int);
-  			return LibFunc_ZdaPv; // void operator delete[](void*);
-
-		case LibFunc_ZnajRKSt9nothrow_t: // new[](unsigned int, nothrow);
- 			return LibFunc_ZdaPvRKSt9nothrow_t; // void operator delete[](void*, nothrow);
- 
-		case LibFunc_ZnajSt11align_val_t: // new[](unsigned int, align_val_t)
-
-			return LibFunc_ZdaPvj; // void operator delete[](void*, unsigned int);
-
-		case LibFunc_ZnajSt11align_val_tRKSt9nothrow_t: // new[](unsigned int, align_val_t, nothrow)
-
-
-		case LibFunc_Znam: // new[](unsigned long);
-		case LibFunc_ZnamRKSt9nothrow_t: // new[](unsigned long, nothrow);
-		case LibFunc_ZnamSt11align_val_t: // new[](unsigned long, align_val_t)
-		case LibFunc_ZnamSt11align_val_tRKSt9nothrow_t: // new[](unsigned long, align_val_t, nothrow)
-
-		case LibFunc_msvc_new_int: // new(unsigned int);
-		case LibFunc_msvc_new_int_nothrow: // new(unsigned int, nothrow);
-		case LibFunc_msvc_new_longlong: // new(unsigned long long);
-		case LibFunc_msvc_new_longlong_nothrow: // new(unsigned long long, nothrow);
-		case LibFunc_msvc_new_array_int: // new[](unsigned int);
-		case LibFunc_msvc_new_array_int_nothrow: // new[](unsigned int, nothrow);
-		case LibFunc_msvc_new_array_longlong: // new[](unsigned long long);
-		case LibFunc_msvc_new_array_longlong_nothrow: // new[](unsigned long long, nothrow);
-
-		//TODO strdup, strndup
-
-		//TODO call, realloc, reallocf
-
-		//TODO (perhaps) posix_memalign
-			TLI.get
-			return true;
-		default:
-			llvm_unreachable("unknown allocation function to find deallocation function for");
-		  	return false;
-	}
-*/
 
 //For updating below one should read MemoryBuiltins.cpp, TargetLibraryInfo.cpp
 static inline
@@ -141,8 +33,10 @@ bool isAllocationFunction(const llvm::Function &F, const llvm::TargetLibraryInfo
 
 		case LibFunc_Znwj: // new(unsigned int);
 		case LibFunc_ZnwjRKSt9nothrow_t: // new(unsigned int, nothrow);
+#if LLVM_VERSION_MAJOR > 6
 		case LibFunc_ZnwjSt11align_val_t: // new(unsigned int, align_val_t)
 		case LibFunc_ZnwjSt11align_val_tRKSt9nothrow_t: // new(unsigned int, align_val_t, nothrow)
+#endif
 
 		case LibFunc_Znwm: // new(unsigned long);
 		case LibFunc_ZnwmRKSt9nothrow_t: // new(unsigned long, nothrow);
@@ -268,8 +162,10 @@ CallInst* freeKnownAllocation(llvm::IRBuilder <>& builder, llvm::Value* tofree,
 
 		case LibFunc_Znwj: // new(unsigned int);
 		case LibFunc_ZnwjRKSt9nothrow_t: // new(unsigned int, nothrow);
+#if LLVM_VERSION_MAJOR > 6
 		case LibFunc_ZnwjSt11align_val_t: // new(unsigned int, align_val_t)
 		case LibFunc_ZnwjSt11align_val_tRKSt9nothrow_t: // new(unsigned int, align_val_t, nothrow)
+#endif
 
 		case LibFunc_Znwm: // new(unsigned long);
 		case LibFunc_ZnwmRKSt9nothrow_t: // new(unsigned long, nothrow);
@@ -320,4 +216,4 @@ CallInst* freeKnownAllocation(llvm::IRBuilder <>& builder, llvm::Value* tofree,
   	return freecall;
 }
 
-#endif
+#endif
diff --git a/enzyme/Enzyme/Utils.h b/enzyme/Enzyme/Utils.h
@@ -18,17 +18,19 @@
 #ifndef ENZYME_UTILS_H
 #define ENZYME_UTILS_H
 
+#include "llvm/ADT/SmallPtrSet.h"
+
 #include "llvm/IR/Function.h"
-#include "llvm/IR/Operator.h"
+#include "llvm/IR/IntrinsicInst.h"
 #include "llvm/IR/Module.h"
+#include "llvm/IR/Operator.h"
 #include "llvm/IR/Type.h"
-#include "llvm/ADT/SmallPtrSet.h"
-#include "llvm/IR/IntrinsicInst.h"
 
 #include "llvm/IR/Function.h"
 #include "llvm/IR/IntrinsicInst.h"
 
 #include "llvm/Support/Casting.h"
+#include "llvm/Support/raw_ostream.h"
 
 static inline llvm::FastMathFlags getFast() {
     llvm::FastMathFlags f;

diff --git a/enzyme/test/Enzyme/memcpy-flt.ll b/enzyme/test/Enzyme/memcpy-flt.ll
@@ -66,7 +66,7 @@ attributes #3 = { nounwind }
 ; CHECK-NEXT:   br i1 %0, label %for.end, label %for.body
 
 ; CHECK: for.body:                                        
-; CHECK-NEXT:   %idx = phi i64 [ %idx.next, %for.body ], [ 0, %entry ]
+; CHECK-NEXT:   %idx = phi i64 [ 0, %entry ], [ %idx.next, %for.body ]
 ; CHECK-NEXT:   %dst.i = getelementptr double, double* %dst, i64 %idx
 ; CHECK-NEXT:   %dst.i.l = load double, double* %dst.i, align 8
 ; CHECK-NEXT:   store double 0.000000e+00, double* %dst.i, align 8

diff --git a/enzyme/test/Enzyme/sumwithbreak.ll b/enzyme/test/Enzyme/sumwithbreak.ll
@@ -1,4 +1,4 @@
-; RUN: opt < %s %loadEnzyme -enzyme -enzyme_preopt=false -mem2reg -instcombine -correlated-propagation -adce -instcombine -simplifycfg -early-cse -simplifycfg -loop-unroll -instcombine -simplifycfg -gvn -S | FileCheck %s
+; RUN: opt < %s %loadEnzyme -enzyme -enzyme_preopt=false -mem2reg -instcombine -correlated-propagation -adce -instcombine -simplifycfg -early-cse -simplifycfg -loop-unroll -instcombine -simplifycfg -gvn -jump-threading -instcombine -S | FileCheck %s
 
 ; Function Attrs: noinline nounwind uwtable
 define dso_local double @f(double* nocapture readonly %x, i64 %n) #0 {
@@ -50,51 +50,43 @@ attributes #0 = { noinline nounwind uwtable }
 ; CHECK-NEXT:   %iv = phi i64 [ %iv.next, %if.end ], [ 0, %entry ]
 ; CHECK-NEXT:   %data.016 = phi double [ %add5, %if.end ], [ 0.000000e+00, %entry ]
 ; CHECK-NEXT:   %cmp2 = fcmp fast ogt double %data.016, 1.000000e+01
-; CHECK-NEXT:   br i1 %cmp2, label %invertif.then, label %if.end
+; CHECK-NEXT:   br i1 %cmp2, label %[[invertifthen:.+]], label %if.end
 
 ; CHECK: if.end:                                           ; preds = %for.body
 ; CHECK-NEXT:   %iv.next = add nuw i64 %iv, 1
 ; CHECK-NEXT:   %arrayidx4 = getelementptr inbounds double, double* %x, i64 %iv
 ; CHECK-NEXT:   %0 = load double, double* %arrayidx4, align 8
 ; CHECK-NEXT:   %add5 = fadd fast double %0, %data.016
 ; CHECK-NEXT:   %cmp = icmp eq i64 %iv, %n
-; CHECK-NEXT:   br i1 %cmp, label %loopMerge.peel, label %for.body
+; CHECK-NEXT:   br i1 %cmp, label %invertif.end.peel, label %for.body
 
 ; CHECK: invertentry: 
 ; CHECK-NEXT:   ret {} undef
 
-; CHECK: invertif.then:
+; CHECK: [[invertifthen]]:
 ; CHECK-NEXT:   %"arrayidx'ipg" = getelementptr double, double* %"x'", i64 %n
 ; CHECK-NEXT:   %[[loadit:.+]] = load double, double* %"arrayidx'ipg", align 8
 ; CHECK-NEXT:   %[[tostoreit:.+]] = fadd fast double %[[loadit]], %differeturn
 ; CHECK-NEXT:   store double %[[tostoreit]], double* %"arrayidx'ipg", align 8
-; CHECK-NEXT:   br label %loopMerge.peel
-
-; CHECK: loopMerge.peel:    
-; CHECK-NEXT:   %"add5'de.1" = phi double [ 0.000000e+00, %invertif.then ], [ %differeturn, %if.end ]
-; CHECK-NEXT:   %"data.016'de.1" = phi double [ %differeturn, %invertif.then ], [ 0.000000e+00, %if.end ]
-; CHECK-NEXT:   br i1 %cmp2, label %invertfor.body.peel, label %invertif.end.peel
+; CHECK-NEXT:   br label %invertfor.body.peel
 
 ; CHECK: invertif.end.peel:  
-; CHECK-NEXT:   %[[diffehere:.+]] = fadd fast double %"data.016'de.1", %"add5'de.1"
 ; CHECK-NEXT:   %"arrayidx4'ipg.peel" = getelementptr double, double* %"x'", i64 %iv
 ; CHECK-NEXT:   %[[loaditp:.+]] = load double, double* %"arrayidx4'ipg.peel", align 8
-; CHECK-NEXT:   %[[tostoreitp:.+]] = fadd fast double %[[loaditp]], %"add5'de.1"
+; CHECK-NEXT:   %[[tostoreitp:.+]] = fadd fast double %[[loaditp]], %differeturn
 ; CHECK-NEXT:   store double %[[tostoreitp]], double* %"arrayidx4'ipg.peel", align 8
+; CHECK-NEXT:   br label %invertfor.body.peel
 
 ; CHECK: invertfor.body.peel:
-; CHECK-NEXT:   %"add5'de.0.peel" = phi double [ %"add5'de.1", %loopMerge.peel ], [ 0.000000e+00, %invertif.end.peel ]
-; CHECK-NEXT:   %"data.016'de.0.peel" = phi double [ %"data.016'de.1", %loopMerge.peel ], [ %[[diffehere]], %invertif.end.peel ]
 ; CHECK-NEXT:   %[[donecmp:.+]] = icmp eq i64 %iv, 0
-; CHECK-NEXT:   %[[dthere:.+]] = fadd fast double %"add5'de.0.peel", %"data.016'de.0.peel"
 ; CHECK-NEXT:   br i1 %[[donecmp]], label %invertentry, label %loopMerge
 
 ; CHECK: loopMerge:
 ; CHECK-NEXT:   %"iv'phi.in" = phi i64 [ %"iv'phi", %loopMerge ], [ %iv, %invertfor.body.peel ]
 ; CHECK-NEXT:   %"iv'phi" = add i64 %"iv'phi.in", -1
 ; CHECK-NEXT:   %"arrayidx4'ipg" = getelementptr double, double* %"x'", i64 %"iv'phi"
 ; CHECK-NEXT:   %[[ldhere:.+]] = load double, double* %"arrayidx4'ipg", align 8
-; CHECK-NEXT:   %[[tshere:.+]] = fadd fast double %[[ldhere]], %[[dthere]]
+; CHECK-NEXT:   %[[tshere:.+]] = fadd fast double %[[ldhere]], %differeturn
 ; CHECK-NEXT:   store double %[[tshere]], double* %"arrayidx4'ipg", align 8
 ; CHECK-NEXT:   %[[icmp:.+]] = icmp eq i64 %"iv'phi", 0
 ; CHECK-NEXT:   br i1 %[[icmp]], label %invertentry, label %loopMerge