Couple ispointer fixes #15312
Couple ispointer fixes #15312
Conversation
|
💯 |
|
The FixedSizeArrays code example from #15274 is also fixed with this. |
|
Nice. Is this testable in any more direct way than doing regular perf tracking? |
|
I'm not sure there is a reasonable way since we don't have first class llvm IR in the language and trying to parse it (or look at the size of the text ...) feels quite brittle |
|
I added a benchmark for this in JuliaCI/BaseBenchmarks.jl#7. If I did it correctly this PR should show a good performance gain for this. Should be something like |
|
If someone could run that for me :) |
|
|
|
Your benchmark job has completed - no performance regressions were detected. A full report can be found here. cc @jrevels |
Sorry, changes to that repo don't automatically deploy to @nanosoldier - I normally tag a version and do it manually, so that I can do a full test sweep of the CI cycle first. I didn't have time to do it yesterday, but I can deploy the change tomorrow. |
|
Okay, no worries. |
| // just compute the pointer and let user load it when necessary | ||
| Type *fty = julia_type_to_llvm(jt); | ||
| Value *addr = builder.CreateGEP(builder.CreatePointerCast(ptr, PointerType::get(fty,0)), idx); | ||
| jl_cgval_t fieldval = mark_julia_slot(addr, jt); |
Should: fix #15274, fix #13305, fix #15277
@KristofferC @simonster can you please check that it is all better ?
Also thanks to @vtjnash for the much nicer codegen.cpp now