Details
Diff Detail
- Repository
- rG LLVM Github Monorepo
Event Timeline
llvm/test/Analysis/ValueTracking/knownbits-and-or-xor-lowbit.ll | ||
---|---|---|
16 | Ideally, input IR and output IR for test cases are the same (before the patch) to the degree that this is possible. Otherwise it gives the impression that a test covers a codepath it doesn't actually use -- for example, the test is written as if this checks the sub X, Odd case, but it really is a convoluted way to write an add X, Odd test. (It looks like a lot of tests here end up testing add X, 1 in the end -- was that intentional?) |
llvm/test/Analysis/ValueTracking/knownbits-and-or-xor-lowbit.ll | ||
---|---|---|
16 |
Okay, will update to the preexisting generated IR (unless its a transform that loses information and breaks the test).
Yes, if you do and X, 1 it commutes and simplifies. add X, 1 can't commute w.o the known bits so its better for testing the exact case. |
llvm/test/Analysis/ValueTracking/knownbits-and-or-xor-lowbit.ll | ||
---|---|---|
33 | Will remove this, sorry had thought I had dropped all the tests that already evaluated. |
llvm/test/Analysis/ValueTracking/knownbits-and-or-xor-lowbit.ll | ||
---|---|---|
16 |
Dramatically simplified tests. Think they are all relatively clear now (maybe too many failure cases?) |
Ideally, input IR and output IR for test cases are the same (before the patch) to the degree that this is possible. Otherwise it gives the impression that a test covers a codepath it doesn't actually use -- for example, the test is written as if this checks the sub X, Odd case, but it really is a convoluted way to write an add X, Odd test.
(It looks like a lot of tests here end up testing add X, 1 in the end -- was that intentional?)