[llvm] a3e38b4 - [DAG] SimplifyDemandedVectorElts - if every and/mul element-pair has a zero/undef then just constant fold to zero

[Simon Pilgrim via llvm-commits]

Author: Simon Pilgrim
Date: 2022-07-24T12:00:31+01:00
New Revision: a3e38b4a206b07077f964e8e0a32c4c39714ac42

URL: https://github.com/llvm/llvm-project/commit/a3e38b4a206b07077f964e8e0a32c4c39714ac42
DIFF: https://github.com/llvm/llvm-project/commit/a3e38b4a206b07077f964e8e0a32c4c39714ac42.diff

LOG: [DAG] SimplifyDemandedVectorElts - if every and/mul element-pair has a zero/undef then just constant fold to zero

Added: 
    

Modified: 
    llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
    llvm/test/CodeGen/X86/vector-partial-undef.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
index 8f71b9f49ad47..cd4f0ae42bcdb 100644
--- a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
@@ -3364,6 +3364,12 @@ bool TargetLowering::SimplifyDemandedVectorElts(
                                    TLO, Depth + 1))
       return true;
 
+    // If every element pair has a zero/undef then just fold to zero.
+    // fold (and x, undef) -> 0  /  (and x, 0) -> 0
+    // fold (mul x, undef) -> 0  /  (mul x, 0) -> 0
+    if (DemandedElts.isSubsetOf(SrcZero | KnownZero | SrcUndef | KnownUndef))
+      return TLO.CombineTo(Op, TLO.DAG.getConstant(0, SDLoc(Op), VT));
+
     // If either side has a zero element, then the result element is zero, even
     // if the other is an UNDEF.
     // TODO: Extend getKnownUndefForVectorBinop to also deal with known zeros
@@ -3373,7 +3379,6 @@ bool TargetLowering::SimplifyDemandedVectorElts(
     KnownUndef &= ~KnownZero;
 
     // Attempt to avoid multi-use ops if we don't need anything from them.
-    // TODO - use KnownUndef to relax the demandedelts?
     if (!DemandedElts.isAllOnes())
       if (SimplifyDemandedVectorEltsBinOp(Op0, Op1))
         return true;

diff  --git a/llvm/test/CodeGen/X86/vector-partial-undef.ll b/llvm/test/CodeGen/X86/vector-partial-undef.ll
index cd4a77e3d4717..ffd95b38dc31b 100644
--- a/llvm/test/CodeGen/X86/vector-partial-undef.ll
+++ b/llvm/test/CodeGen/X86/vector-partial-undef.ll
@@ -85,9 +85,7 @@ define <4 x i64> @and_undef_elts(<2 x i64> %x) {
 ;
 ; AVX-LABEL: and_undef_elts:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
-; AVX-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[3,0,1,2]
-; AVX-NEXT:    vandps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
+; AVX-NEXT:    vxorps %xmm0, %xmm0, %xmm0
 ; AVX-NEXT:    retq
   %extend = shufflevector <2 x i64> %x, <2 x i64> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef>
   %bogus_bo = and <4 x i64> %extend, <i64 undef, i64 undef, i64 42, i64 43>