[llvm] 7e4c6e9 - [SystemZ] Bugfix in getDemandedSrcElements(). (#88623)

[via llvm-commits]

Author: Jonas Paulsson
Date: 2024-04-15T16:32:14+02:00
New Revision: 7e4c6e98fa05f5c3bf14f96365ae74a8d12c6257

URL: https://github.com/llvm/llvm-project/commit/7e4c6e98fa05f5c3bf14f96365ae74a8d12c6257
DIFF: https://github.com/llvm/llvm-project/commit/7e4c6e98fa05f5c3bf14f96365ae74a8d12c6257.diff

LOG: [SystemZ] Bugfix in getDemandedSrcElements(). (#88623)

For the intrinsic s390_vperm, all of the elements are demanded, so use
an APInt with the value of '-1' for them (not '1').

Fixes https://github.com/llvm/llvm-project/issues/88397

Added: 
    

Modified: 
    llvm/lib/Target/SystemZ/SystemZISelLowering.cpp
    llvm/test/CodeGen/SystemZ/knownbits-intrinsics-binop.ll

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/SystemZ/SystemZISelLowering.cpp b/llvm/lib/Target/SystemZ/SystemZISelLowering.cpp
index 3b3057f8fe9645..5c2579f3bf1878 100644
--- a/llvm/lib/Target/SystemZ/SystemZISelLowering.cpp
+++ b/llvm/lib/Target/SystemZ/SystemZISelLowering.cpp
@@ -7818,7 +7818,7 @@ static APInt getDemandedSrcElements(SDValue Op, const APInt &DemandedElts,
       break;
     }
     case Intrinsic::s390_vperm:
-      SrcDemE = APInt(NumElts, 1);
+      SrcDemE = APInt(NumElts, -1);
       break;
     default:
       llvm_unreachable("Unhandled intrinsic.");

diff  --git a/llvm/test/CodeGen/SystemZ/knownbits-intrinsics-binop.ll b/llvm/test/CodeGen/SystemZ/knownbits-intrinsics-binop.ll
index 3bcbbb45581f9c..b855d019347820 100644
--- a/llvm/test/CodeGen/SystemZ/knownbits-intrinsics-binop.ll
+++ b/llvm/test/CodeGen/SystemZ/knownbits-intrinsics-binop.ll
@@ -458,3 +458,22 @@ define <16 x i8> @f30() {
                                i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1, i8 1>
   ret <16 x i8> %res
 }
+
+; Test VPERM with various constant operands.
+define i32 @f31() {
+; CHECK-LABEL: f31:
+; CHECK-LABEL: # %bb.0:
+; CHECK-NEXT: larl %r1, .LCPI31_0
+; CHECK-NEXT: vl %v0, 0(%r1), 3
+; CHECK-NEXT: larl %r1, .LCPI31_1
+; CHECK-NEXT: vl %v1, 0(%r1), 3
+; CHECK-NEXT: vperm %v0, %v1, %v1, %v0
+; CHECK-NEXT: vlgvb %r2, %v0, 0
+; CHECK-NEXT: nilf %r2, 7
+; CHECK-NEXT:                           # kill: def $r2l killed $r2l killed $r2d
+; CHECK-NEXT: br %r14
+  %P = tail call <16 x i8> @llvm.s390.vperm(<16 x i8> <i8 0, i8 1, i8 1, i8 2, i8 1, i8 2, i8 2, i8 3, i8 1, i8 2, i8 2, i8 3, i8 2, i8 3, i8 3, i8 4>, <16 x i8> <i8 0, i8 1, i8 1, i8 2, i8 1, i8 2, i8 2, i8 3, i8 1, i8 2, i8 2, i8 3, i8 2, i8 3, i8 3, i8 4>, <16 x i8> <i8 1, i8 2, i8 3, i8 4, i8 5, i8 6, i8 7, i8 8, i8 9, i8 10, i8 11, i8 12, i8 13, i8 14, i8 15, i8 0>)
+  %E = extractelement <16 x i8> %P, i64 0
+  %res = zext i8 %E to i32
+  ret i32 %res
+}