[X86] Remove unnecessary custom lowering of vXi1 SADDSAT/SSUBSAT/UADDSAT/USUBSAT

As discussed on D97478. The removal of the custom tag causes some changes in the add/sub-overflow expansion as it no longer expands to sat-arith codegen.

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index b35d9bb..86a4190 100644 (file)
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -1467,13 +1467,8 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM,
       setOperationAction(ISD::ANY_EXTEND,  VT, Custom);
     }
 
-    for (auto VT : { MVT::v1i1, MVT::v2i1, MVT::v4i1, MVT::v8i1, MVT::v16i1 }) {
-      setOperationAction(ISD::UADDSAT,          VT, Custom);
-      setOperationAction(ISD::SADDSAT,          VT, Custom);
-      setOperationAction(ISD::USUBSAT,          VT, Custom);
-      setOperationAction(ISD::SSUBSAT,          VT, Custom);
-      setOperationAction(ISD::VSELECT,          VT,  Expand);
-    }
+    for (auto VT : { MVT::v1i1, MVT::v2i1, MVT::v4i1, MVT::v8i1, MVT::v16i1 })
+      setOperationAction(ISD::VSELECT,          VT, Expand);
 
     for (auto VT : { MVT::v2i1, MVT::v4i1, MVT::v8i1, MVT::v16i1 }) {
       setOperationAction(ISD::SETCC,            VT, Custom);
@@ -1851,11 +1846,6 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM,
 
     for (auto VT : { MVT::v32i1, MVT::v64i1 }) {
       setOperationAction(ISD::VSELECT,            VT, Expand);
-      setOperationAction(ISD::UADDSAT,            VT, Custom);
-      setOperationAction(ISD::SADDSAT,            VT, Custom);
-      setOperationAction(ISD::USUBSAT,            VT, Custom);
-      setOperationAction(ISD::SSUBSAT,            VT, Custom);
-
       setOperationAction(ISD::TRUNCATE,           VT, Custom);
       setOperationAction(ISD::SETCC,              VT, Custom);
       setOperationAction(ISD::EXTRACT_VECTOR_ELT, VT, Custom);
@@ -27162,20 +27152,6 @@ static SDValue LowerADDSAT_SUBSAT(SDValue Op, SelectionDAG &DAG,
   unsigned Opcode = Op.getOpcode();
   SDLoc DL(Op);
 
-  if (VT.getScalarType() == MVT::i1) {
-    switch (Opcode) {
-    default: llvm_unreachable("Expected saturated arithmetic opcode");
-    case ISD::UADDSAT:
-    case ISD::SADDSAT:
-      // *addsat i1 X, Y --> X | Y
-      return DAG.getNode(ISD::OR, DL, VT, X, Y);
-    case ISD::USUBSAT:
-    case ISD::SSUBSAT:
-      // *subsat i1 X, Y --> X & ~Y
-      return DAG.getNode(ISD::AND, DL, VT, X, DAG.getNOT(DL, Y, VT));
-    }
-  }
-
   if (VT == MVT::v32i16 || VT == MVT::v64i8 ||
       (VT.is256BitVector() && !Subtarget.hasInt256())) {
     assert(Op.getSimpleValueType().isInteger() &&

diff --git a/llvm/test/CodeGen/X86/vec_saddo.ll b/llvm/test/CodeGen/X86/vec_saddo.ll
index cba1057..c341db6 100644 (file)
--- a/llvm/test/CodeGen/X86/vec_saddo.ll
+++ b/llvm/test/CodeGen/X86/vec_saddo.ll
@@ -1023,17 +1023,16 @@ define <4 x i32> @saddo_v4i1(<4 x i1> %a0, <4 x i1> %a1, <4 x i1>* %p2) nounwind
 ;
 ; AVX512-LABEL: saddo_v4i1:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vpslld $31, %xmm1, %xmm1
 ; AVX512-NEXT:    vpslld $31, %xmm0, %xmm0
-; AVX512-NEXT:    vpor %xmm1, %xmm0, %xmm2
 ; AVX512-NEXT:    vptestmd %xmm0, %xmm0, %k0
+; AVX512-NEXT:    vpslld $31, %xmm1, %xmm1
 ; AVX512-NEXT:    vptestmd %xmm1, %xmm1, %k1
-; AVX512-NEXT:    vptestmd %xmm2, %xmm2, %k2
-; AVX512-NEXT:    kxorw %k1, %k0, %k0
-; AVX512-NEXT:    kxorw %k2, %k0, %k1
+; AVX512-NEXT:    kxorw %k1, %k0, %k2
+; AVX512-NEXT:    vptestnmd %xmm0, %xmm0, %k0 {%k2}
+; AVX512-NEXT:    kxorw %k0, %k1, %k1
 ; AVX512-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
 ; AVX512-NEXT:    vmovdqa32 %xmm0, %xmm0 {%k1} {z}
-; AVX512-NEXT:    kshiftlw $12, %k0, %k0
+; AVX512-NEXT:    kshiftlw $12, %k2, %k0
 ; AVX512-NEXT:    kshiftrw $12, %k0, %k0
 ; AVX512-NEXT:    kmovd %k0, %eax
 ; AVX512-NEXT:    movb %al, (%rdi)

diff --git a/llvm/test/CodeGen/X86/vec_ssubo.ll b/llvm/test/CodeGen/X86/vec_ssubo.ll
index 3e6b135..20fca27 100644 (file)
--- a/llvm/test/CodeGen/X86/vec_ssubo.ll
+++ b/llvm/test/CodeGen/X86/vec_ssubo.ll
@@ -1032,16 +1032,15 @@ define <4 x i32> @ssubo_v4i1(<4 x i1> %a0, <4 x i1> %a1, <4 x i1>* %p2) nounwind
 ;
 ; AVX512-LABEL: ssubo_v4i1:
 ; AVX512:       # %bb.0:
-; AVX512-NEXT:    vpslld $31, %xmm1, %xmm1
-; AVX512-NEXT:    vptestmd %xmm1, %xmm1, %k0
 ; AVX512-NEXT:    vpslld $31, %xmm0, %xmm0
-; AVX512-NEXT:    vptestmd %xmm0, %xmm0, %k1
-; AVX512-NEXT:    vptestnmd %xmm1, %xmm1, %k2 {%k1}
-; AVX512-NEXT:    kxorw %k0, %k1, %k0
-; AVX512-NEXT:    kxorw %k2, %k0, %k1
+; AVX512-NEXT:    vptestmd %xmm0, %xmm0, %k0
+; AVX512-NEXT:    vpslld $31, %xmm1, %xmm1
+; AVX512-NEXT:    vptestmd %xmm1, %xmm1, %k1
+; AVX512-NEXT:    kxorw %k1, %k0, %k1
+; AVX512-NEXT:    vptestnmd %xmm0, %xmm0, %k2 {%k1}
 ; AVX512-NEXT:    vpcmpeqd %xmm0, %xmm0, %xmm0
-; AVX512-NEXT:    vmovdqa32 %xmm0, %xmm0 {%k1} {z}
-; AVX512-NEXT:    kshiftlw $12, %k0, %k0
+; AVX512-NEXT:    vmovdqa32 %xmm0, %xmm0 {%k2} {z}
+; AVX512-NEXT:    kshiftlw $12, %k1, %k0
 ; AVX512-NEXT:    kshiftrw $12, %k0, %k0
 ; AVX512-NEXT:    kmovd %k0, %eax
 ; AVX512-NEXT:    movb %al, (%rdi)