Make pnacl-bccompress add abbreviations for obvious constants.

lib/Bitcode/NaCl/Reader/NaClBitCodes.cpp

Index: lib/Bitcode/NaCl/Reader/NaClBitCodes.cpp
diff --git a/lib/Bitcode/NaCl/Reader/NaClBitCodes.cpp b/lib/Bitcode/NaCl/Reader/NaClBitCodes.cpp
new file mode 100644
index 0000000000000000000000000000000000000000..a0e31f16d68d55fa6551167ebceceea005495dd7
--- /dev/null
+++ b/lib/Bitcode/NaCl/Reader/NaClBitCodes.cpp
@@ -0,0 +1,95 @@
+//===- NaClBitcodeHeader.cpp ----------------------------------------------===//
+//     PNaCl bitcode header reader.
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// Implementation of Bitcode abbrevations.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/Bitcode/NaCl/NaClBitCodes.h"
+#include "llvm/Support/raw_ostream.h"
+
+using namespace llvm;
+
+void NaClBitCodeAbbrevOp::Print(raw_ostream& Stream) const {
+  if (isLiteral()) {
+    Stream << getLiteralValue();
+  } else if (isEncoding()) {
+    switch (getEncoding()) {
+    case Fixed:
+      Stream << "Fixed(" << getEncodingData() << ")";
+      break;
+    case VBR:
+      Stream << "VBR(" << getEncodingData() << ")";
+      break;
+    case Array:
+      Stream << "Array";
+      break;
+    case Char6:
+      Stream << "Char6";
+      break;
+    case Blob:
+      Stream << "Blob";
+      break;
+    default:
+      assert(false);

[jvoung (off chromium), 2014/02/07 21:34:10]

llvm_unreachable("...") instead ?

[Karl, 2014/02/07 23:26:04]

Done.

+      Stream << "??";  // In case asserts are turned off.
+      break;
+    }
+  } else {
+    assert(false);
+    Stream << "??";  // In case asserts are turned off.
+  }
+}
+
+static void PrintExpression(raw_ostream &Stream,
+                            const NaClBitCodeAbbrev *Abbrev,
+                            unsigned &Index) {
+  // Bail out early, in case we are incrementally building the
+  // expression and the argument is not available yet.
+  if (Index >= Abbrev->getNumOperandInfos()) return;
+
+  const NaClBitCodeAbbrevOp &Op = Abbrev->getOperandInfo(Index);
+  Op.Print(Stream);
+  if (unsigned NumArgs = Op.NumArguments()) {
+    Stream << "(";
+    for (unsigned i = 0; i < NumArgs; ++i) {
+      ++Index;
+      if (i > 0) Stream << ",";
+      PrintExpression(Stream, Abbrev, Index);
+    }
+    Stream << ")";
+  }
+}
+
+void NaClBitCodeAbbrev::Print(raw_ostream &Stream) const {
+  Stream << "[";
+  for (unsigned i = 0; i < getNumOperandInfos(); ++i) {
+    if (i > 0) Stream << ", ";
+    PrintExpression(Stream, this, i);
+  }
+  Stream << "]\n";
+}
+
+NaClBitCodeAbbrev *NaClBitCodeAbbrev::Simplify() const {
+  NaClBitCodeAbbrev *Abbrev = new NaClBitCodeAbbrev();
+  for (unsigned i = 0; i < OperandList.size(); ++i) {
+    const NaClBitCodeAbbrevOp &Op = OperandList[i];
+    // Simplify if possible.  Currently, the only simplification known
+    // is to remove unnecessary operands appearing immediately before an
+    // array operator. That is, apply the simplification:
+    //    Op Array(Op) -> Array(Op)

[jvoung (off chromium), 2014/02/07 21:34:10]

perhaps check if that the abbrev is well formed:

assert(!Op.isArrayOp() || i == (OperandList.size()-2))

or something like that?

[Karl, 2014/02/07 23:26:04]

Done.

+    while (Op.isArrayOp() && !Abbrev->OperandList.empty() &&
+           Abbrev->OperandList.back() == OperandList[i+1]) {
+      Abbrev->OperandList.pop_back();
+    }
+    Abbrev->OperandList.push_back(Op);
+  }
+  return Abbrev;
+}