diff options
author | Greg Clayton <gclayton@apple.com> | 2014-07-31 18:19:28 +0000 |
---|---|---|
committer | Greg Clayton <gclayton@apple.com> | 2014-07-31 18:19:28 +0000 |
commit | ac583610479c63056a64356a1d3bd3b40814f60d (patch) | |
tree | c718602c5422dcd57385517f84424b52b32b5947 /llvm/lib/MC/MCAnalysis/MCObjectDisassembler.cpp | |
parent | 0f4fbf17f57f4697cbda98f544920cfbcda22762 (diff) | |
download | bcm5719-llvm-ac583610479c63056a64356a1d3bd3b40814f60d.tar.gz bcm5719-llvm-ac583610479c63056a64356a1d3bd3b40814f60d.zip |
Rewrote the initial DW_OP_piece support to be able to support opcodes like:
DW_OP_fbreg(N) DW_OP_piece(4) DW_OP_fbreg(M) DW_OP_piece(8)
DW_OP_fbreg(N) DW_OP_piece(4) DW_OP_piece(8)
The first grabs 4 bytes from FP+N followed by 8 bytes from FP+M, the second grabs 4 bytes from FP+N followed by zero filling 8 bytes which are unavailable. Of course regiters are stuff supported:
DW_OP_reg3 DW_OP_piece(4) DW_OP_reg8 DW_OP_piece(8)
The fix does the following:
1 - don't push the full piece value onto the stack, keep it on the side
2 - fill zeros for DW_OP_piece(N) opcodes that have nothing on the stack (instead of previously consuming the full piece that was pushed onto the stack)
3 - simplify the logic
<rdar://problem/16930524>
llvm-svn: 214415
Diffstat (limited to 'llvm/lib/MC/MCAnalysis/MCObjectDisassembler.cpp')
0 files changed, 0 insertions, 0 deletions