[PATCH] D50036: [SLC] Expand the simplification of pow(x, 0.5) to sqrt(x)

[Evandro Menezes via Phabricator via llvm-commits]

evandro added a comment.

I'll commit the data type changes increasing type diversity on the side.



================
Comment at: llvm/test/Transforms/InstCombine/pow-sqrt.ll:8-9
 ;
-  %pow = call fast double @llvm.pow.f64(double %x, double 5.000000e-01)
-  ret double %pow
+  %pow = call fast float @llvm.pow.f32(float %x, float 5.0e-01)
+  ret float %pow
 }
----------------
spatel wrote:
> Why change the data types?
In order to increase test coverage.  Otherwise, it virtually tested for `double` only.


================
Comment at: llvm/test/Transforms/InstCombine/pow-sqrt.ll:20
 ;
   %pow = call afn <2 x double> @llvm.pow.v2f64(<2 x double> %x, <2 x double> <double 5.0e-01, double 5.0e-01>)
   ret <2 x double> %pow
----------------
spatel wrote:
> The minimal test for this case requires no FMF?
IIUC, not all of them, just AFN.


================
Comment at: llvm/test/Transforms/InstCombine/pow-sqrt.ll:30
 ;
-  %pow = call afn double @pow(double %x, double 5.0e-01)
-  ret double %pow
+  %pow = call ninf float @powf(float %x, float 5.0e-01)
+  ret float %pow
----------------
spatel wrote:
> Why change the data types? Same question for the later tests - the types should be irrelevant to the change in functionality, right?
> 
> Please make any IR changes to the tests that are necessary to show off the improved/fixed functionality as a preliminary step to this review.
Replacing AFN for NINF shows the improved functionality.


https://reviews.llvm.org/D50036