This patch puts a constraint on the attributes that will receive call base context.
Diff Detail
Event Timeline
llvm/include/llvm/Transforms/IPO/Attributor.h | ||
---|---|---|
1416 | Nit: For better or worse we use F and Fn as a variable name for functions usually. I read the comment and do not know what happens. We need to improve the wording, maybe elaborate. | |
llvm/lib/Transforms/IPO/Attributor.cpp | ||
96 | I'm confused by the description. Is the analysis invaliding values? | |
807 | Nit: Assertions always with messages. Can we have a explanation here or in the header what is happening here. The commit message should also contain more details. | |
815 | Style: No braces. Use a range loop or part of the constructor (or use append). |
llvm/lib/Transforms/IPO/Attributor.cpp | ||
---|---|---|
815 | That was my initial idea but since arg_begin() is a pointer the SmallVector was trying to converyt Argument to Value*. After investigating I found that there is something called pointer_iterator that would fix this issue. |
llvm/unittests/Transforms/IPO/AttributorTestBase.h | ||
---|---|---|
53 ↗ | (On Diff #281191) | contents BumpPtrAllocator will be freed after this block is dead. |
llvm/unittests/Transforms/IPO/AttributorTestBase.h | ||
---|---|---|
53 ↗ | (On Diff #281191) | Put the allocator in the same scope as the Attributor? |
llvm/unittests/Transforms/IPO/AttributorTestBase.h | ||
---|---|---|
53 ↗ | (On Diff #281191) | That's what I am planing to do. I just wanted to add a inline commend so I don't forget about this issue. |
Nit: For better or worse we use F and Fn as a variable name for functions usually.
I read the comment and do not know what happens. We need to improve the wording, maybe elaborate.