Is there a flag to set in Tablegen source so that all function calls act as instruction barriers. I.e. no instruction can be schedule around a function call. Thanks, Micah -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20100601/0dea390a/attachment.html>
On Tue, Jun 1, 2010 at 5:13 PM, Villmow, Micah <Micah.Villmow at amd.com> wrote:> Is there a flag to set in Tablegen source so that all function calls act as > instruction barriers. I.e. no instruction can be schedule around a function > call.Do you mean "don't schedule instructions which appear before the call in the source after the call, and vice versa?" There is no such flag. The normal schedulers doesn't pay attention to source order at all; -pre-RA-sched=source was recently implemented which does pay attention to source order. -Eli
Thanks Eli, that seems to do the trick. Micah -----Original Message----- From: Eli Friedman [mailto:eli.friedman at gmail.com] Sent: Tuesday, June 01, 2010 5:42 PM To: Villmow, Micah Cc: llvmdev at cs.uiuc.edu List Subject: Re: [LLVMdev] Treating function calls as barriers On Tue, Jun 1, 2010 at 5:13 PM, Villmow, Micah <Micah.Villmow at amd.com> wrote:> Is there a flag to set in Tablegen source so that all function calls act as > instruction barriers. I.e. no instruction can be schedule around a function > call.Do you mean "don't schedule instructions which appear before the call in the source after the call, and vice versa?" There is no such flag. The normal schedulers doesn't pay attention to source order at all; -pre-RA-sched=source was recently implemented which does pay attention to source order. -Eli
Reasonably Related Threads
- [LLVMdev] Treating function calls as barriers
- [LLVMdev] Instructions that cannot be duplicated
- [LLVMdev] [cfe-dev] SPIR Portability Discussion
- [LLVMdev] [cfe-dev] SPIR Portability Discussion
- [LLVMdev] FW: RFC: Supporting different sized address space arithmetic