missed fold, fcmp ogt (x - y), 0 => fcmp ogt x, y

[zhengyang92]

https://alive2.llvm.org/ce/z/azHAHq
https://godbolt.org/z/M8WbdYE8x

define i1 @src(float %0, float %1) {
%t2 = fsub float %0, %1
%t3 = fcmp ogt float %t2, 0.000000e+00
ret i1 %t3
}

define i1 @tgt(float %0, float %1) {
%olt = fcmp ogt float %0, %1
ret i1 %olt
}

@jayfoad @regehr

[llvmbot]

Hi!

This issue may be a good introductory issue for people new to working on LLVM. If you would like to work on this issue, your first steps are:

1. In the comments of the issue, request for it to be assigned to you.
2. Fix the issue locally.
3. Run the test suite locally. Remember that the subdirectories under test/ create fine-grained testing targets, so you can e.g. use make check-clang-ast to only run Clang's AST tests.
4. Create a Git commit.
5. Run git clang-format HEAD~1 to format your changes.
6. Open a pull request to the upstream repository on GitHub. Detailed instructions can be found in GitHub's documentation.

If you have any further questions about this issue, don't hesitate to ask via a comment in the thread below.

[llvmbot]

@llvm/issue-subscribers-good-first-issue

Author: Zhengyang Liu (zhengyang92)

https://alive2.llvm.org/ce/z/azHAHq https://godbolt.org/z/M8WbdYE8x

define i1 @<!-- -->src(float %0, float %1) {
%t2 = fsub float %0, %1
%t3 = fcmp ogt float %t2, 0.000000e+00
ret i1 %t3
}

define i1 @<!-- -->tgt(float %0, float %1) {
%olt = fcmp ogt float %0, %1
ret i1 %olt
}

@jayfoad @regehr

[SahilPatidar]

@dtcxzyw, I would like to take this one.

[jayfoad]

This only works for floating point formats that have IEEE-like denormals (does LLVM currently support any formats that do not have denormals?), and when denormals are not being flushed to zero. Otherwise you can have two different values x and y where x - y evaluates to 0.

[jayfoad]

As a follow up we could do the same kind of thing for plus, e.g. fcmp (x + y), 0 => fcmp x, -y, if either x or y can be negated for free.