[libc-commits] [PATCH] D123335: [libc] Add support for x86-64 targets that do not have FMA instructions.

[Tue Ly via Phabricator via libc-commits]

lntue added inline comments.


================
Comment at: libc/src/math/generic/CMakeLists.txt:517
     -O3
     -mfma
 )
----------------
michaelrj wrote:
> On windows, removing the `-mfma` options in this cmake file with this patch doesn't break any of the tests except for this one. The `expm1f` tests fail with a ULP error, copied below.
> 
> 
> ```
> [ RUN      ] LlvmLibcExpm1fTest.InFloatRange
> C:/src/llvm-project/libc/test/src/math/expm1f_test.cpp:119: FAILURE
> Failed to match __llvm_libc::expm1f(x) against __llvm_libc::testing::mpfr::get_mpfr_matcher<mpfr::Operation::Expm1>( x, __llvm_libc::expm1f(x), 0.5, mpfr::RoundingMode::TowardZero).
> Match value not within tolerance value of MPFR result:
>   Input decimal: -0.00000000000000000000000000000000000001175976677754
>      Input bits: 0x80800d72, (S | E | M) = (1 | 0x01 | 0x000d72)
> 
>   Match decimal: -0.00000000000000000000000000000000000001175976677754
>      Match bits: 0x80800d72, (S | E | M) = (1 | 0x01 | 0x000d72)
> 
>     MPFR result: -0.00000000000000000000000000000000000001175976537624
>    MPFR rounded: 0x80800d71, (S | E | M) = (1 | 0x01 | 0x000d71)
> 
>       ULP error: 1.000000
> [  FAILED  ] LlvmLibcExpm1fTest.InFloatRange
> Ran 5 tests.  PASS: 4  FAIL: 1
> ```
This is expected.  There will be followup changes to take care of different exceptional values from math functions when FMA instructions are not supported.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D123335/new/

https://reviews.llvm.org/D123335