[CIR][Dialect] Add FMaximumOp and FMinimumOp #1237

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Merged

bcardosolopes merged 2 commits into llvm:main from ghehg:m1Mac

Dec 17, 2024

Contributor

ghehg commented Dec 17, 2024

There are two sets of intrinsics regarding Min and Max operations for floating points
Maximum vs Maxnum
Minimum vs Minnum

The difference is whether NaN should be propagated when one of the inputs is NaN
Maxnum and Minnum would return number if one of inputs is NaN, and the other is a number,
But
Maximum and Minimum would return NaN (propagation of NaN)

And they are resolved to different ASM such as FMAX
vs FMAXNM

Both have user cases, we already implemented Maxnum and Minnum
But Maximum and Minimum has user cases in neon intrinsic
and __builtin_elementwise_maximum

ghehg added 2 commits

December 17, 2024 10:41


          change OP's name

7e50d21


          add FMaximumOp and FMinimumOp

27a1634

ghehg marked this pull request as ready for review

December 17, 2024 16:47

ghehg requested review from lanza and bcardosolopes as code owners

December 17, 2024 16:47

bcardosolopes approved these changes

View reviewed changes

Member

bcardosolopes left a comment

Awesome! thanks

bcardosolopes merged commit 8d649fb into llvm:main

9 checks passed

lanza pushed a commit that referenced this pull request


          [CIR][Dialect] Add FMaximumOp and FMinimumOp (#1237)

9d3c1cc

There are two sets of intrinsics regarding Min and Max operations for
floating points

[Maximum](https://mlir.llvm.org/docs/Dialects/LLVM/#llvmintrmaximum-llvmmaximumop)
vs
[Maxnum](https://mlir.llvm.org/docs/Dialects/LLVM/#llvmintrmaxnum-llvmmaxnumop)

[Minimum](https://mlir.llvm.org/docs/Dialects/LLVM/#llvmintrminimum-llvmminimumop)
vs
[Minnum](https://mlir.llvm.org/docs/Dialects/LLVM/#llvmintrminnum-llvmminnumop)

[The difference is whether NaN should be propagated when one of the
inputs is
NaN](https://llvm.org/docs/LangRef.html#llvm-maximumnum-intrinsic)
Maxnum and Minnum would return number if one of inputs is NaN, and the
other is a number,
But 
Maximum and Minimum would return NaN (propagation of NaN)

And they are resolved to different ASM such as
[FMAX](https://developer.arm.com/documentation/ddi0596/2021-03/SIMD-FP-Instructions/FMAX--vector---Floating-point-Maximum--vector--?lang=en)
vs
[FMAXNM](https://developer.arm.com/documentation/ddi0596/2021-03/SIMD-FP-Instructions/FMAXNM--vector---Floating-point-Maximum-Number--vector--?lang=en)

Both have user cases, we already implemented Maxnum and Minnum
But Maximum and Minimum has user cases in [neon intrinsic
](https://developer.arm.com/architectures/instruction-sets/intrinsics/vmax_f32
)
and [__builtin_elementwise_maximum
](https://github.com/llvm/clangir/blob/a989ecb2c55da1fe28e4072c31af025cba6c4f0f/clang/test/CodeGen/strictfp-elementwise-bulitins.cpp#L53)

terapines-osc-cir pushed a commit to Terapines/clangir that referenced this pull request


          [CIR][Dialect] Add FMaximumOp and FMinimumOp (llvm#1237)

4ba1a05

There are two sets of intrinsics regarding Min and Max operations for
floating points

[Maximum](https://mlir.llvm.org/docs/Dialects/LLVM/#llvmintrmaximum-llvmmaximumop)
vs
[Maxnum](https://mlir.llvm.org/docs/Dialects/LLVM/#llvmintrmaxnum-llvmmaxnumop)

[Minimum](https://mlir.llvm.org/docs/Dialects/LLVM/#llvmintrminimum-llvmminimumop)
vs
[Minnum](https://mlir.llvm.org/docs/Dialects/LLVM/#llvmintrminnum-llvmminnumop)

[The difference is whether NaN should be propagated when one of the
inputs is
NaN](https://llvm.org/docs/LangRef.html#llvm-maximumnum-intrinsic)
Maxnum and Minnum would return number if one of inputs is NaN, and the
other is a number,
But 
Maximum and Minimum would return NaN (propagation of NaN)

And they are resolved to different ASM such as
[FMAX](https://developer.arm.com/documentation/ddi0596/2021-03/SIMD-FP-Instructions/FMAX--vector---Floating-point-Maximum--vector--?lang=en)
vs
[FMAXNM](https://developer.arm.com/documentation/ddi0596/2021-03/SIMD-FP-Instructions/FMAXNM--vector---Floating-point-Maximum-Number--vector--?lang=en)

Both have user cases, we already implemented Maxnum and Minnum
But Maximum and Minimum has user cases in [neon intrinsic
](https://developer.arm.com/architectures/instruction-sets/intrinsics/vmax_f32
)
and [__builtin_elementwise_maximum
](https://github.com/llvm/clangir/blob/a989ecb2c55da1fe28e4072c31af025cba6c4f0f/clang/test/CodeGen/strictfp-elementwise-bulitins.cpp#L53)

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet