[ConstantFPRange] Implement `ConstantFPRange::makeExactFCmpRegion` (#111490)

Note: The current implementation doesn't return optimal result for `fcmp
one/une x, +/-inf` since we don't handle this c

[ConstantFPRange] Implement `ConstantFPRange::makeExactFCmpRegion` (#111490)

Note: The current implementation doesn't return optimal result for `fcmp
one/une x, +/-inf` since we don't handle this case in
https://github.com/llvm/llvm-project/pull/110891. Maybe we can make it
optimal after seeing some real-world cases.

show more ...

[ConstantFPRange] Implement `ConstantFPRange::makeSatisfyingFCmpRegion` (#110891)

This patch adds support for `ConstantFPRange::makeSatisfyingFCmpRegion`.
We only check the optimality for cases whe

[ConstantFPRange] Implement `ConstantFPRange::makeSatisfyingFCmpRegion` (#110891)

This patch adds support for `ConstantFPRange::makeSatisfyingFCmpRegion`.
We only check the optimality for cases where the result can be
represented by a ConstantFPRange.

This patch also adds some tests for `ConstantFPRange::fcmp` because it
depends on `makeSatisfyingFCmpRegion`. Unfortunately we cannot
exhaustively test this function due to time limit. I just pick some
interesting ranges instead.

show more ...

[ConstantFPRange][UnitTests] Guard exhaustive checks by `EXPENSIVE_CHECKS` (#111276)

Addresses comment
https://github.com/llvm/llvm-project/pull/111056#issuecomment-2393951499.

[ConstantFPRange][UnitTest] Ignore NaN payloads when enumerating values in a range (#111083)

NaN payloads can be ignored because they are unrelated with
ConstantFPRange (except the conversion from

[ConstantFPRange][UnitTest] Ignore NaN payloads when enumerating values in a range (#111083)

NaN payloads can be ignored because they are unrelated with
ConstantFPRange (except the conversion from ConstantFPRange to
KnownBits). This patch just enumerates `+/-[S/Q]NaN` to avoid
enumerating 32 NaN values in all ranges which contain NaN values.
Addresses comment
https://github.com/llvm/llvm-project/pull/110082#issuecomment-2392037204.
This patch reduces the execution time for unittests from 30.37s to
10.59s with an optimized build.

show more ...

[ConstantFPRange] Implement `ConstantFPRange::makeAllowedFCmpRegion` (#110082)

This patch adds `makeAllowedFCmpRegion` support for `ConstantFPRange`.

[ConstantFPRange] Address review comments. NFC. (#110793)

1. Address post-commit review comments
https://github.com/llvm/llvm-project/pull/86483#discussion_r1782305961.
2. Move some NFC changes fr

[ConstantFPRange] Address review comments. NFC. (#110793)

1. Address post-commit review comments
https://github.com/llvm/llvm-project/pull/86483#discussion_r1782305961.
2. Move some NFC changes from
https://github.com/llvm/llvm-project/pull/110082 to this patch.

show more ...

[ConstantFPRange] Remove `ConstantFPRange::toKnownFPClass` (#109960)

Addresses comment
https://github.com/llvm/llvm-project/pull/86483#pullrequestreview-2327710679.

[ConstantFPRange] Suppress unused result warnings. NFC. (#109921)

Fixes warnings `error: ignoring return value of function declared with
'nodiscard' attribute [-Werror,-Wunused-result]`.

[LLVM][IR] Add constant range support for floating-point types (#86483)

This patch adds basic constant range support for floating-point types to enable range-based optimizations.