[PATCH] D66050: Improve division estimation of floating points.

Qiu Chaofan via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Mon Aug 26 06:17:16 PDT 2019


qiucf marked 3 inline comments as done.
qiucf added inline comments.


================
Comment at: llvm/test/CodeGen/PowerPC/combine-fneg.ll:19
+; CHECK-NEXT:    xxswapd 4, 4
 ; CHECK-NEXT:    xvnmsubadp 1, 2, 0
+; CHECK-NEXT:    xxlor 3, 2, 2
----------------
hfinkel wrote:
> xbolva00 wrote:
> > Regression
> There's still an extra arithmetic instruction here?
It's as expected. Since `visitFDIV` calls `combineRepeatedFPDivisors` to transform the vector divisions before `BuildDivEstimate`.

But the extra instructions are really redundant here. If `visitFMA` folds `(fma (fneg a) (fneg b) c)` into `(fma a b c)` like what `visitFMUL` does, the extra instructions will get eliminated. This can be done in future patch.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D66050/new/

https://reviews.llvm.org/D66050





More information about the llvm-commits mailing list