[PATCH] D20019: [PPC] initial exploitation of xs[min,max]cdp
Nemanja Ivanovic via llvm-commits
llvm-commits at lists.llvm.org
Tue May 10 06:21:56 PDT 2016
nemanjai added inline comments.
================
Comment at: lib/Target/PowerPC/PPCISelDAGToDAG.cpp:461
@@ -460,1 +460,3 @@
+static bool isDoubleLegalMinOrMax(SDNode *N) {
+
----------------
Why the restriction to double precision values? The ISA document mentions this instruction can be used for both single and double precision operands.
================
Comment at: lib/Target/PowerPC/PPCInstrVSX.td:2069
@@ -2065,1 +2068,3 @@
+ def : Pat <(f64 (selectcc f64:$XB, f64:$XA, f64:$XA, f64:$XB, SETOLT)),
+ (XSMAXCDP $XA, $XB)>;
----------------
Patterns for f32?
================
Comment at: test/CodeGen/PowerPC/vsx-p9.ll:3
@@ +2,3 @@
+
+define double @max_test1(double %x, double %y) #0 {
+
----------------
And if you add f32 above, test cases for float as well.
http://reviews.llvm.org/D20019
More information about the llvm-commits
mailing list