[PATCH] D47782: AMDGPU: Custom lower v2f16 fneg/fabs with illegal f16

[Matt Arsenault via Phabricator via llvm-commits]

arsenm added inline comments.


================
Comment at: test/CodeGen/AMDGPU/fabs.f16.ll:107
 ; CI: v_cvt_f32_f16_e32
-; CI: v_mul_f32_e64 v{{[0-9]+}}, |v{{[0-9]+}}|, v{{[0-9]+}}
+; CI: v_mul_f32_e32 v{{[0-9]+}}, v{{[0-9]+}}, v{{[0-9]+}}
 ; CI: v_cvt_f16_f32
----------------
rampitec wrote:
> Are there a lot of regressions like this?
> Any plans to address these?
I don't think so. I only really noticed this one because of the kind of perverse case of using the original and negated value resulting in doubling the number of conversions. We probably have to repeat the performFNeg/Fabs combines for the bit patterns used for lowered f16


https://reviews.llvm.org/D47782