Changed how LLVM IR was generated to increase vectorization

[emmettneyman]

Summary

Changed the structure of the generated IR to make it easier to vectorize

[morehouse]

clang/tools/clang-fuzzer/proto-to-llvm/loop_proto_to_llvm.cpp

125

What does pc mean in this triple? I'm used to seeing x86_64-unknown-linux-gnu.

126

Does removing the variable names really make this easier to vectorize?

129

Does removing branch names really make this easier to vectorize?

[emmettneyman]

clang/tools/clang-fuzzer/proto-to-llvm/loop_proto_to_llvm.cpp

125

It's probably generating pc since I'm using my desktop. I'll change it to unknown so it's not platform-specific.

126

It doesn't, I just thought it was cleaner and produced slightly smaller IR. It also more closely mimics the behavior of the -emit-llvm flag.

129

Same answer as above. Should I change these back?

[emmettneyman]

Changed pc to unknown

[morehouse]

clang/tools/clang-fuzzer/proto-to-llvm/loop_proto_to_llvm.cpp

129

I think the previous labels were more human-readable.

[emmettneyman]

Added back more descriptive variable and loop names

[morehouse]

clang/tools/clang-fuzzer/proto-to-llvm/loop_proto_to_llvm.cpp

126

Should %s be signed? Do we want unsigned compare here?

133

Seems like the endloop label is unnecessary. Does this help vectorize? If not, lets get rid of unconditional jumps to the next line.

138–140

This will make overflow undefined... Isn't that the opposite of what we want? That will permit LLVM to assume overflow never happens and modify the code in new ways based on that assumption.

[emmettneyman]

clang/tools/clang-fuzzer/proto-to-llvm/loop_proto_to_llvm.cpp

126

I don't think it matters since %s will always be positive (it will always be equal to
kArraySize from input_arrays.h).

133

I initially had it this way since that's how -emit-llvm generated the code. But it doesn't seem like it's necessary for loops to be vectorized upon further inspection.

138–140

That's a good point. It probably won't matter since this addition will never overflow (we're just starting at zero and incrementing by one). This is what was generated by -emit-llvm. I'll remove it to reduce confusion.

[emmettneyman]

Some small fixes to improve simplicity of generated IR

[emmettneyman]

Rebased and ready to land

clang/tools/clang-fuzzer/proto-to-llvm/loop_proto_to_llvm.cpp

 
 std::string ConstToString(const Const &x) {
   return std::to_string(x.val());
 }
 std::string VarRefToString(std::ostream &os, const VarRef &x) {
   std::string arr;
   switch(x.arr()) {
   case VarRef::ARR_A:
-    arr = "%a";
+    arr = "%0";
     break;
   case VarRef::ARR_B:
-    arr = "%b";
+    arr = "%1";
     break;
   case VarRef::ARR_C:
-    arr = "%c";
+    arr = "%2";
     break;
   }
   std::string ptr_var = get_var();
-  os << ptr_var << " = getelementptr i32, i32* " << arr << ", i64 %ct\n";
+  os << ptr_var << " = getelementptr inbounds i32, i32* " << arr << ", i64 %ct\n";
   return ptr_var;
 }
 std::string RvalueToString(std::ostream &os, const Rvalue &x) {
   if(x.has_cons())
     return ConstToString(x.cons());
   if(x.has_binop())
     return BinopToString(os, x.binop());
   if(x.has_varref()) {

 }
 std::ostream &operator<<(std::ostream &os, const StatementSeq &x) {
   for (auto &st : x.statements()) {
     os << st;
   }
   return os;
 }
 std::ostream &operator<<(std::ostream &os, const LoopFunction &x) {
-  return os << "define void @foo(i32* %a, i32* %b, i32* noalias %c, i64 %s) {\n"
-            << "%i = alloca i64\n"
-            << "store i64 0, i64* %i\n"
-            << "br label %loop\n\n"
-            << "loop:\n"
-            << "%ct = load i64, i64* %i\n"
-            << "%comp = icmp eq i64 %ct, %s\n"
-            << "br i1 %comp, label %endloop, label %body\n\n"
-            << "body:\n"
+  return os << "target triple = \"x86_64-pc-linux-gnu\"\n"
+            << "define void @foo(i32*, i32*, i32*, i64) {\n"
+            << "%5 = icmp sgt i64 %3, 0\n"
+            << "br i1 %5, label %6, label %8\n"
+            << "; <label>:6:\n"
+            << "br label %9\n"
+            << "; <label>:7:\n"
+            << "br label %8\n"
+            << "; <label>:8:\n"
+            << "ret void\n"
+            << "; <label>:9:\n"
+            << " %ct   = phi i64 [ %10, %9 ], [ 0, %6 ]\n"
             << x.statements()
-            << "%z = add i64 1, %ct\n"
-            << "store i64 %z, i64* %i\n"
-            << "br label %loop\n\n"
-            << "endloop:\n"
-            << "ret void\n}\n";
+            << "%10 = add nuw nsw i64 %ct, 1\n"
+            << "%11 = icmp eq i64 %10, %3\n"
+            << "br i1 %11, label %7, label %9\n}\n";
 }
 
 // ---------------------------------
 
 std::string LoopFunctionToLLVMString(const LoopFunction &input) {
   std::ostringstream os;
   os << input;
   return os.str();