Diff Detail
Event Timeline
| llvm/include/llvm/IR/IRBuilder.h | ||
|---|---|---|
| 651 | But ObjId is uint64_t? | |
| 654 | Here and elsewhere - currently llvm variable naming scheme did not change, all variables start with Uppercase Letter. | |
| 663 | Should this be uint32_t ? | |
| 671 | Should this be uint32_t ? | |
| llvm/include/llvm/IR/IntrinsicInst.h | ||
| 889–890 | assert(lhs->getIntrinsicID() == rhs->getIntrinsicID() && ;hs->getIntrinsicID() == Intrinsic::side_noalias && "Can only check noalias compatibility of side_noalias intrinsics"); | |
| 891–896 | std::tie(<...>) == std::tie(<...>) ? | |
| llvm/lib/IR/IRBuilder.cpp | ||
| 514 | Should this be uint64_t? | |
| 565–566 | Ops.insert(Ops.end(), MDNodes.begin(), MDNodes.end()) ? | |
| 569 | same | |
| 573–577 | This should be in the parent patch that added that | |
| llvm/lib/IR/Verifier.cpp | ||
| 4760–4761 | } else if (auto *PHI = dyn_cast<PHINode>(V)) { | |
| 4762–4763 | Worklist.insert() ? | |
| llvm/include/llvm/IR/IRBuilder.h | ||
|---|---|---|
| 651 | I am preparing an update where I use uint64_t more consistenly. | |
| llvm/include/llvm/IR/IntrinsicInst.h | ||
| 891–896 | You mean: std::forward_as_tuple ? | |
| llvm/lib/IR/IRBuilder.cpp | ||
| 514 | No. It should be int64_t. | |
| 565–566 | That won't work. | |
| 569 | But here it will. | |
| 573–577 | Maybe. I tried to not do that kind of changes in the rebased versions. | |
Treat objId as uint64_t everywhere; treat indices as int64_t; use std::forward_as_tuple; .use insert instead of for-loop+push_back.
| llvm/include/llvm/IR/IntrinsicInst.h | ||
|---|---|---|
| 891–896 | Doesn't that loose the short-circuit behavior of operator&&? Detail: IMHO it doesn't really improve readability: (a == b) && (c == d) versus std::tie(a, c) == std::tie(b, d); | |
Yes, using forward_as_tuple might result in slower code... I am reverting that change.
But ObjId is uint64_t?