[llvm] r239977 - [AsmPrinter] Make isRepeatedByteSequence smarter about odd integer types

[Benjamin Kramer]

Author: d0k
Date: Wed Jun 17 18:55:17 2015
New Revision: 239977

URL: http://llvm.org/viewvc/llvm-project?rev=239977&view=rev
Log:
[AsmPrinter] Make isRepeatedByteSequence smarter about odd integer types

- zext the value to alloc size first, then check if the value repeats
  with zero padding included. If so we can still emit a .space
- Do the checking with APInt.isSplat(8), which handles non-pow2 types
- Also handle large constants (bit width > 64)
- In a ConstantArray all elements have the same type, so it's sufficient
  to check the first constant recursively and then just compare if all
  following constants are the same by pointer compare

Added:
    llvm/trunk/test/CodeGen/X86/global-fill.ll
Modified:
    llvm/trunk/lib/CodeGen/AsmPrinter/AsmPrinter.cpp

Modified: llvm/trunk/lib/CodeGen/AsmPrinter/AsmPrinter.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/AsmPrinter/AsmPrinter.cpp?rev=239977&r1=239976&r2=239977&view=diff
==============================================================================
--- llvm/trunk/lib/CodeGen/AsmPrinter/AsmPrinter.cpp (original)
+++ llvm/trunk/lib/CodeGen/AsmPrinter/AsmPrinter.cpp Wed Jun 17 18:55:17 2015
@@ -1795,40 +1795,30 @@ static int isRepeatedByteSequence(const
 /// composed of a repeated sequence of identical bytes and return the
 /// byte value.  If it is not a repeated sequence, return -1.
 static int isRepeatedByteSequence(const Value *V, TargetMachine &TM) {
-
   if (const ConstantInt *CI = dyn_cast<ConstantInt>(V)) {
-    if (CI->getBitWidth() > 64) return -1;
-
-    uint64_t Size =
-        TM.getDataLayout()->getTypeAllocSize(V->getType());
-    uint64_t Value = CI->getZExtValue();
-
-    // Make sure the constant is at least 8 bits long and has a power
-    // of 2 bit width.  This guarantees the constant bit width is
-    // always a multiple of 8 bits, avoiding issues with padding out
-    // to Size and other such corner cases.
-    if (CI->getBitWidth() < 8 || !isPowerOf2_64(CI->getBitWidth())) return -1;
+    uint64_t Size = TM.getDataLayout()->getTypeAllocSizeInBits(V->getType());
+    assert(Size % 8 == 0);
 
-    uint8_t Byte = static_cast<uint8_t>(Value);
+    // Extend the element to take zero padding into account.
+    APInt Value = CI->getValue().zextOrSelf(Size);
+    if (!Value.isSplat(8))
+      return -1;
 
-    for (unsigned i = 1; i < Size; ++i) {
-      Value >>= 8;
-      if (static_cast<uint8_t>(Value) != Byte) return -1;
-    }
-    return Byte;
+    return Value.zextOrTrunc(8).getZExtValue();
   }
   if (const ConstantArray *CA = dyn_cast<ConstantArray>(V)) {
     // Make sure all array elements are sequences of the same repeated
     // byte.
     assert(CA->getNumOperands() != 0 && "Should be a CAZ");
-    int Byte = isRepeatedByteSequence(CA->getOperand(0), TM);
-    if (Byte == -1) return -1;
-
-    for (unsigned i = 1, e = CA->getNumOperands(); i != e; ++i) {
-      int ThisByte = isRepeatedByteSequence(CA->getOperand(i), TM);
-      if (ThisByte == -1) return -1;
-      if (Byte != ThisByte) return -1;
-    }
+    Constant *Op0 = CA->getOperand(0);
+    int Byte = isRepeatedByteSequence(Op0, TM);
+    if (Byte == -1)
+      return -1;
+
+    // All array elements must be equal.
+    for (unsigned i = 1, e = CA->getNumOperands(); i != e; ++i)
+      if (CA->getOperand(i) != Op0)
+        return -1;
     return Byte;
   }
 

Added: llvm/trunk/test/CodeGen/X86/global-fill.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/global-fill.ll?rev=239977&view=auto
==============================================================================
--- llvm/trunk/test/CodeGen/X86/global-fill.ll (added)
+++ llvm/trunk/test/CodeGen/X86/global-fill.ll Wed Jun 17 18:55:17 2015
@@ -0,0 +1,27 @@
+; RUN: llc -mtriple=x86_64-apple-darwin < %s | FileCheck %s
+
+ at test1 = global [2 x i24] [i24 -1, i24 -1]
+; CHECK-LABEL: test1:
+; CHECK-NEXT: .long	16777215
+; CHECK-NEXT: .long	16777215
+
+ at test2 = global [2 x i7] [i7 1, i7 1]
+; CHECK-LABEL: test2:
+; CHECK-NEXT: .space 2,1
+
+ at test3 = global [4 x i128] [i128 -1, i128 -1, i128 -1, i128 -1]
+; CHECK-LABEL: test3:
+; CHECK-NEXT: .space 64,255
+
+ at test4 = global [3 x i16] [i16 257, i16 257, i16 257]
+; CHECK-LABEL: test4:
+; CHECK-NEXT: .space 6,1
+
+ at test5 = global [2 x [2 x i16]] [[2 x i16] [i16 257, i16 257], [2 x i16] [i16 -1, i16 -1]]
+; CHECK-LABEL: test5:
+; CHECK-NEXT: .space 4,1
+; CHECK-NEXT: .space 4,255
+
+ at test6 = global [2 x [2 x i16]] [[2 x i16] [i16 257, i16 257], [2 x i16] [i16 257, i16 257]]
+; CHECK-LABEL: test6:
+; CHECK-NEXT: .space 8,1