Added LLVM metadata to generated IR to increase vectorization width

[emmettneyman]

Summary

Edited loop_proto_to_llvm to emit metadata at the end of the generated IR. This metadata will increase the vector width when the IR is optimized.

[morehouse]

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

142

I'm not sure kArraySize is what you want here. Does "width" refer to array size or the SIMD width?

Any problem with letting the vectorizer determine this automatically?

143

Does this metadata change coverage in the vectorizer?

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

18 ↗

(On Diff #159978)

This include should be in the cpp file.

[emmettneyman]

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

142

The "width" refers to how wide the vector types should be in the outputted IR. So for example, a width of 64 would produce IR instructions with the type <64 x i32>.

The problem with letting the vectorizer determine the width automatically is that since the length of the arrays (and so, the number of loop iterations) is passed to the function as a parameter, the vectorizer always chooses the default width of 4 when running the loop vectorize optimization pass.

143

I'm running the coverage now, but my guess is yes. Because this metadata overrides the default behavior of the loop vectorizer, I think it will increase the coverage.

[emmettneyman]

Switched the include statement to the cpp file

[morehouse]

Code LGTM, but let's make sure this actually helps before landing.

[emmettneyman]

Rebased and ready to land

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

 //
 // Implements functions for converting between protobufs and LLVM IR.
 //
 //
 //===----------------------------------------------------------------------===//
 
 #include "loop_proto_to_llvm.h"
 #include "cxx_loop_proto.pb.h"
+#include "../handle-llvm/input_arrays.h"
 
 // The following is needed to convert protos in human-readable form
 #include <google/protobuf/text_format.h>
 
 #include <ostream>
 #include <sstream>
 
 namespace clang_fuzzer {
return os << "target triple = \"x86_64-unknown-linux-gnu\"\n"
             << "br label %loop\n"
             << "end:\n"
             << "ret void\n"
             << "loop:\n"
             << " %ct   = phi i64 [ %ctnew, %loop ], [ 0, %start ]\n"
             << x.statements()
             << "%ctnew = add i64 %ct, 1\n"
             << "%j = icmp eq i64 %ctnew, %s\n"
-            << "br i1 %j, label %end, label %loop\n}\n";
+            << "br i1 %j, label %end, label %loop, !llvm.loop !0\n}\n"
+            << "!0 = distinct !{!0, !1, !2}\n"
+            << "!1 = !{!\"llvm.loop.vectorize.enable\", i1 true}\n"
+            << "!2 = !{!\"llvm.loop.vectorize.width\", i32 " << kArraySize
+            << "}\n";
 }
 
 // ---------------------------------
 
 std::string LoopFunctionToLLVMString(const LoopFunction &input) {
   std::ostringstream os;
   os << input;
   return os.str();