[PATCH] D101017: [NewPM] Make GlobalsAA available earlier in the pipeline

aeubanks abandoned this revision. aeubanks added a comment. turns out it was an issue with GlobalsAA not being recalculated and being too conservative anyway, I don't think adding GlobalsAA early on really does much, adding it to the inliner pipeline seems good enough https://reviews.llvm.org/D

[PATCH] D101017: [NewPM] Make GlobalsAA available earlier in the pipeline

aeubanks added a comment. ah the issue is LoopIdiomRecognize can't recognize the following as a memcpy: @__const.f.c = private unnamed_addr constant [20 x double] [double 1.051000e-01, double 1.57e-02, double 1.85e-02, double 0x3F823A29C779A6B5, double 2.19e-02, double 1.41e-0

[PATCH] D101017: [NewPM] Make GlobalsAA available earlier in the pipeline

aeubanks added a comment. Actually, the InstCombine transform allows LoopFullUnrollPass to unroll the copy loop which results in a bunch of stores. Previously during LoopFullUnrollPass we wouldn't see the global within the loop. Repository: rG LLVM Github Monorepo CHANGES SINCE LAST ACTION

[PATCH] D101017: [NewPM] Make GlobalsAA available earlier in the pipeline

aeubanks added a comment. Looking at one of the examples in llvm-test-suite that regressed, this seems to be regressing some very simple code: void f(double *a) { double c[20] = { 0.1051, 0.0157, 0.0185, 0.0089, 0.0219, 0.0141, 0.0097, 0.0758, 0.0168, 0.1188, 0.1635, 0.0112

[PATCH] D101017: [NewPM] Make GlobalsAA available earlier in the pipeline

ychen added a comment. That sounds like less powerful optimizations due to the GlobalsAA movement. If adding a GlobalsAA plus D100917 is a win overall, that should work. Repository: rG LLVM Github Monorepo CHANGES SINCE LAST ACTION https://reviews.llvm.o

[PATCH] D101017: [NewPM] Make GlobalsAA available earlier in the pipeline

nikic added a comment. @aeubanks It seems like this has non-trivial impact on code size (http://llvm-compile-time-tracker.com/compare.php?from=a62cbd9a0211d08bace8794b435996890feb44d4&to=7805d7f72c337bfcc2fbc2dc9d2b9ac23474d5d9&stat=size-text), so probably the compile-time change is a side-effec

[PATCH] D101017: [NewPM] Make GlobalsAA available earlier in the pipeline

aeubanks added a comment. this seems to have some compile time improvements? http://llvm-compile-time-tracker.com/compare.php?from=a62cbd9a0211d08bace8794b435996890feb44d4&to=7805d7f72c337bfcc2fbc2dc9d2b9ac23474d5d9&stat=instructions Repository: rG LLVM Github Monorepo CHANGES SINCE LAST ACTI

[PATCH] D101017: [NewPM] Make GlobalsAA available earlier in the pipeline

aeubanks created this revision. Herald added subscribers: wenlei, steven_wu, hiraditya. aeubanks requested review of this revision. Herald added projects: clang, LLVM. Herald added subscribers: llvm-commits, cfe-commits. A future change will cause fewer analyses to be invalidated. Currently, the p