AVX-512: Fixed a bug in comparison of i1 vectors.

cmp eq should give kxnor instruction
cmp neq should give kxor

https://llvm.org/bugs/show_bug.cgi?id=23631

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@239460 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 72f18cc114358eb18c73b1fc4b231b1b8b1ee509..f4e8579d67cb30e832f7cafb009fe4df02023edc 100644 (file)
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -13239,13 +13239,13 @@ static SDValue LowerBoolVSETCC_AVX512(SDValue Op, SelectionDAG &DAG) {
                                DAG.getConstant(-1, dl, VT));
   switch (SetCCOpcode) {
   default: llvm_unreachable("Unexpected SETCC condition");
-  case ISD::SETNE:
-    // (x != y) -> ~(x ^ y)
+  case ISD::SETEQ:
+    // (x == y) -> ~(x ^ y)
     return DAG.getNode(ISD::XOR, dl, VT,
                        DAG.getNode(ISD::XOR, dl, VT, Op0, Op1),
                        DAG.getConstant(-1, dl, VT));
-  case ISD::SETEQ:
-    // (x == y) -> (x ^ y)
+  case ISD::SETNE:
+    // (x != y) -> (x ^ y)
     return DAG.getNode(ISD::XOR, dl, VT, Op0, Op1);
   case ISD::SETUGT:
   case ISD::SETGT:

diff --git a/lib/Target/X86/X86InstrAVX512.td b/lib/Target/X86/X86InstrAVX512.td
index c1d0aef07118a4520233af16bdd485deea1f3e07..b0e1880ab51341c62de964afd49f099ee8b94fef 100644 (file)
--- a/lib/Target/X86/X86InstrAVX512.td
+++ b/lib/Target/X86/X86InstrAVX512.td
@@ -2044,11 +2044,11 @@ defm : avx512_binop_pat<xor,  KXORWrr>;
 def : Pat<(xor (xor VK16:$src1, VK16:$src2), (v16i1 immAllOnesV)),
           (KXNORWrr VK16:$src1, VK16:$src2)>;
 def : Pat<(xor (xor VK8:$src1, VK8:$src2), (v8i1 immAllOnesV)),
-          (KXNORBrr VK8:$src1, VK8:$src2)>;
+          (KXNORBrr VK8:$src1, VK8:$src2)>, Requires<[HasDQI]>;
 def : Pat<(xor (xor VK32:$src1, VK32:$src2), (v32i1 immAllOnesV)),
-          (KXNORDrr VK32:$src1, VK32:$src2)>;
+          (KXNORDrr VK32:$src1, VK32:$src2)>, Requires<[HasBWI]>;
 def : Pat<(xor (xor VK64:$src1, VK64:$src2), (v64i1 immAllOnesV)),
-          (KXNORQrr VK64:$src1, VK64:$src2)>;
+          (KXNORQrr VK64:$src1, VK64:$src2)>, Requires<[HasBWI]>;
 
 let Predicates = [NoDQI] in
 def : Pat<(xor (xor VK8:$src1, VK8:$src2), (v8i1 immAllOnesV)),

diff --git a/test/CodeGen/X86/avx512-vec-cmp.ll b/test/CodeGen/X86/avx512-vec-cmp.ll
index 04028a1da510ddff7bea3d4f6a1e32f21618d7a4..6a4a3aa7e371d26d3c439f013166b5bff5abb528 100644 (file)
--- a/test/CodeGen/X86/avx512-vec-cmp.ll
+++ b/test/CodeGen/X86/avx512-vec-cmp.ll
@@ -394,7 +394,7 @@ define <8 x i64> @test27(<8 x i64> %x, i64* %yb.ptr, <8 x i64> %x1, <8 x i64> %y
 ; KNL-LABEL: test28
 ; KNL: vpcmpgtq
 ; KNL: vpcmpgtq
-; KNL: kxorw
+; KNL: kxnorw
 define <8 x i32>@test28(<8 x i64> %x, <8 x i64> %y, <8 x i64> %x1, <8 x i64> %y1) {
   %x_gt_y = icmp sgt <8 x i64> %x, %y
   %x1_gt_y1 = icmp sgt <8 x i64> %x1, %y1
@@ -406,7 +406,7 @@ define <8 x i32>@test28(<8 x i64> %x, <8 x i64> %y, <8 x i64> %x1, <8 x i64> %y1
 ; KNL-LABEL: test29
 ; KNL: vpcmpgtd
 ; KNL: vpcmpgtd
-; KNL: kxnorw
+; KNL: kxorw
 define <16 x i8>@test29(<16 x i32> %x, <16 x i32> %y, <16 x i32> %x1, <16 x i32> %y1) {
   %x_gt_y = icmp sgt <16 x i32> %x, %y
   %x1_gt_y1 = icmp sgt <16 x i32> %x1, %y1