[llvm-commits] CVS: llvm-gcc/gcc/llvm-expand.c

[Chris Lattner]

Changes in directory llvm-gcc/gcc:

llvm-expand.c updated: 1.44 -> 1.45

---
Log message:

Fix PR377: http://llvm.cs.uiuc.edu/PR377 : [llvmgcc] miscompilation of staticly initialized unsigned bitfields

There were two bugs in this code.  First, llvm_type_get_size returns #bytes, not
#bits.  Second, we must use 1LL instead of 1 or else all bits get shifted out.

Testcase here: SingleSource/UnitTests/2004-06-20-StaticBitfieldInit.c



---
Diffs of the changes:  (+3 -3)

Index: llvm-gcc/gcc/llvm-expand.c
diff -u llvm-gcc/gcc/llvm-expand.c:1.44 llvm-gcc/gcc/llvm-expand.c:1.45
--- llvm-gcc/gcc/llvm-expand.c:1.44	Fri Jun 18 22:35:20 2004
+++ llvm-gcc/gcc/llvm-expand.c	Sun Jun 20 13:59:59 2004
@@ -3814,8 +3814,8 @@
           
           /* If there is already a value set for this field, mask out any bits
            * that are part of the current field (in case the current field is
-           * multiply initialized to different values).  The LLVM stuff should
-           * constant propagate these operations.
+           * multiply initialized to different values).  LLVM will constant fold
+           * all of the constant expressions generated.
            */
           if (Result[FieldIndexVal]) {
             long long MaskVal = ~(((1 << FieldSize)-1) << FieldOffset);
@@ -3825,7 +3825,7 @@
             /* If this is an unsigned type, mask off bits outside the range of
                the data type in question... */
             if (!llvm_type_is_signed(FieldType))
-              MaskVal &= (1 << llvm_type_get_size(FieldType))-1;
+              MaskVal &= (1LL << (llvm_type_get_size(FieldType)*8))-1;
 
             Mask = llvm_constant_new_integral(FieldType, MaskVal);
             Result[FieldIndexVal] =