search for: opentcp

...At this point mechanism is applied to two optimizations: Inliner and Loop Invariant Code Motion. Supported platform is MIPS32r2. Despite small number of instrumented points we got improvements on some test files from CSiBE benchmark. Here are the promising results: jpeg-6b/jidctflt.c -10.2% OpenTCP-1.0.4/system.c -6.7% libmspack/test/chmd_md5.c -5.8% flex-2.5.31/regex.c -4.8% We plan to apply this framework to more compiler phases and to some other target architectures. It is expected to give better results on some non orthogonal architectures like microMIPS. Another direction is to...

...p > > Invariant Code Motion. Supported platform is MIPS32r2. > > > > Despite small number of instrumented points we got improvements on > > some test > > files from CSiBE benchmark. Here are the promising results: > > > > jpeg-6b/jidctflt.c -10.2% > > OpenTCP-1.0.4/system.c -6.7% > > libmspack/test/chmd_md5.c -5.8% > > flex-2.5.31/regex.c -4.8% > > Very interesting. Out of curiosity, any slowdowns? jidctflt.c is optimized due to the fact that LICM moves some constant definitions out of the loop and increases the register pressure so...

...wo optimizations: Inliner and > Loop > Invariant Code Motion. Supported platform is MIPS32r2. > > Despite small number of instrumented points we got improvements on > some test > files from CSiBE benchmark. Here are the promising results: > > jpeg-6b/jidctflt.c -10.2% > OpenTCP-1.0.4/system.c -6.7% > libmspack/test/chmd_md5.c -5.8% > flex-2.5.31/regex.c -4.8% Very interesting. Out of curiosity, any slowdowns? > > We plan to apply this framework to more compiler phases and to some > other > target architectures. It is expected to give better results on...

...Supported platform is MIPS32r2. > > > > > > Despite small number of instrumented points we got improvements on > > > some test > > > files from CSiBE benchmark. Here are the promising results: > > > > > > jpeg-6b/jidctflt.c -10.2% > > > OpenTCP-1.0.4/system.c -6.7% > > > libmspack/test/chmd_md5.c -5.8% > > > flex-2.5.31/regex.c -4.8% > > > > Very interesting. Out of curiosity, any slowdowns? > > jidctflt.c is optimized due to the fact that LICM moves some > constant definitions out of the loop and in...