[llvm-bugs] [Bug 43682] New: Invalid DWARF expression for long double on x64

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=43682

            Bug ID: 43682
           Summary: Invalid DWARF expression for long double on x64
           Product: clang
           Version: unspecified
          Hardware: PC
                OS: Linux
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: C
          Assignee: unassignedclangbugs at nondot.org
          Reporter: brettw at gmail.com
                CC: blitzrakete at gmail.com, dgregor at apple.com,
                    erik.pilkington at gmail.com, llvm-bugs at lists.llvm.org,
                    richard-llvm at metafoo.co.uk

Created attachment 22675
  --> https://bugs.llvm.org/attachment.cgi?id=22675&action=edit
Repro case

I was experimenting with long doubles on x64 (80 bits which force x87
instructions). Clang seems to generate corrupt DWARF expressions. For this:

  long double ld = 2.38;
  ld += argc;

(see attached source file), it produces this DWARF entry:

  DW_TAG_variable
    DW_AT_location      (0x0000005a
      [0x0000000000201714,  0x0000000000201722): DW_OP_constu
0x9851eb851eb85000, DW_OP_stack_value, DW_OP_piece 0x8, DW_OP_constu 0x4000,
DW_OP_stack_value, DW_OP_bit_piece 0x10 0x40, DW_OP_stack_value)
    DW_AT_name  ("ld")
    DW_AT_decl_file     ("float.cc")
    DW_AT_decl_line     (4)
    DW_AT_type  (0x00000090 "long double")

This expression makes no sense. It seems to be three different ones
concatenated together, as the three DW_OP_stack_value instructions should each
indicate the end of the expression.

Additionally this variable should have two distinct values while these symbols
only indicate one.

In that expression, I used "argc" to force it to not optimize the math.
Removing that generates a reasonable looking constant value:

  DW_TAG_variable
    DW_AT_const_value   (<0x0a> 00 50 b8 1e 85 eb 51 98 00 40 )
    DW_AT_name  ("ld")
    DW_AT_decl_file     ("float.cc")
    DW_AT_decl_line     (4)
    DW_AT_type  (0x00000097 "long double")

If the type is changed to a "double", the DWARF is reasonable, showing a
constant value and then the result of the math operation in a register:

  DW_TAG_variable
    DW_AT_location      (0x00000046
      [0x0000000000201711,  0x000000000020171d): DW_OP_constu
0x40030a3d70a3d70a, DW_OP_stack_value
      [0x000000000020171d,  0x0000000000201729): DW_OP_reg17 XMM0)
    DW_AT_name  ("ld")
    DW_AT_decl_file     ("float.cc")
    DW_AT_decl_line     (4)
    DW_AT_type  (0x0000002a "double")

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20191015/dfff0a0b/attachment.html>