search for: waligora

...x typically uses libgcc. If they're emitted in IR, the functions could potentially be inlined. For example if the size of the switch is known to be small so no binary search is done, could inlining the find_ function be smaller than setting up the call? On Tue, Feb 14, 2017 at 5:47 AM, Witold Waligora via llvm-dev <llvm-dev at lists.llvm.org> wrote: > I didn't answer compiler-rt and benchmarks question. > > compiler-rt would have to implement the decoding function(s) to undo > whatever the compiler did. This would be target-dependent. > Our target implements four of them...

...wering. > > Did you tried any benchmarks, please share if you have any numbers with performance improvements & code size reduction ? > > Regards, > Ashutosh > > -----Original Message----- > From: llvm-dev [mailto:llvm-dev-bounces at lists.llvm.org] On Behalf Of Witold Waligora via llvm-dev > Sent: Monday, February 13, 2017 6:41 PM > To: llvm-dev at lists.llvm.org > Subject: [llvm-dev] (RFC) JumpMaps: switch statement optimization > > Hi All, > > This is our first appearance on llvm-dev so a small introduction is in order. > We are a team of two...

Hi All, This is our first appearance on llvm-dev so a small introduction is in order. We are a team of two working for good few years now on an out-of-tree embedded back-end. Our primary goal is code size, secondary performance. We've recently implemented an optimization that we believe others could use. It's about time to contribute back.