[PATCH] D26855: New unsafe-fp-math implementation for X86 target

[Simon Pilgrim via Phabricator via llvm-commits]

RKSimon added a comment.

Does anyone else have comments on this please? There has been interest in moving more of such codegen to the MC so that we can use more realistic scheduling costs so I'm keen to see this go in.



================
Comment at: test/CodeGen/X86/recip-fastmath2.ll:191
+; SSE-NEXT:    rcpps %xmm0, %xmm3
+; SSE-NEXT:    movaps ${{\.LCPI.*}}, %xmm1
+; SSE-NEXT:    movaps %xmm0, %xmm4
----------------
Any ideas why we fail to get:
```
movaps {{.*#+}} xmm1 = [1.000000e+00,1.000000e+00,1.000000e+00,1.000000e+00]
```
There are other cases here as well.


https://reviews.llvm.org/D26855