#7684: cgrun071 segfaults ----------------------------------+---------------------------------- Reporter: tibbe | Owner: Type: bug | Status: new Priority: normal | Milestone: 7.8.1 Component: Compiler | Version: 7.7 Resolution: | Keywords: Operating System: MacOS X | Architecture: x86_64 (amd64) Type of failure: Runtime crash | Difficulty: Unknown Test Case: cgrun071 | Blocked By: Blocking: | Related Tickets: ----------------------------------+---------------------------------- Comment (by rwbarton): I don't understand how the first 8 or so lines of gdb output are consistent with the rest, so I'm going to assume that gdb was temporarily confused. The `info registers` output looks much more plausible. Replying to [comment:15 simonmar]:
The native code generator is supposed to emit code to mask out the extra bits. If it isn't doing that, that's a bug. See commit 085e8145f63c8f42d8bc19cd3cff52b8cd5b6455
Going by leroux's `info registers` output, it sure looks like it isn't masking them out, since the argument register `rdi` is `0x98b7fa5e6c84f828`. From the assembly diff I see that gcc 4.8 outputs code to do the masking in the callee (`movzbl %dil, %edi`) where gcc 4.2 does not. Oddly the System V x86_64 ABI doesn't seem to specify whether arguments in registers that are smaller than the register size should be masked/sign-extended by the caller, but I gather that we are working under the assumption that they should be (which is consistent with the 386 ABI). leroux, could you attach the entire `cgrun071.S` disassembly output? Or if it's very large, at least the disassembly of the function which calls `hs_popcnt8`? -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/7684#comment:29 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler