[PATCH] D148624: [AArch64] Add sign bits for vector compare nodes

[Dave Green via Phabricator via llvm-commits]

dmgreen created this revision.
dmgreen added reviewers: lawben, efriedma, SjoerdMeijer, Sp00ph.
Herald added subscribers: hiraditya, kristof.beyls.
Herald added a project: All.
dmgreen requested review of this revision.
Herald added a project: LLVM.

This adds ComputeNumSignBits for the NEON vector comparison nodes, which all either return 0 or -1.

Unfortunately I couldn't find a good way to test this patch, and the only test case that changes isn't the best. Still may be useful.


https://reviews.llvm.org/D148624

Files:
  llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
  llvm/lib/Target/AArch64/AArch64ISelLowering.h
  llvm/test/CodeGen/AArch64/cmp-select-sign.ll


Index: llvm/test/CodeGen/AArch64/cmp-select-sign.ll
===================================================================
--- llvm/test/CodeGen/AArch64/cmp-select-sign.ll
+++ llvm/test/CodeGen/AArch64/cmp-select-sign.ll
@@ -71,7 +71,7 @@
 define i64 @not_sign_i64(i64 %a) {
 ; CHECK-LABEL: not_sign_i64:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    mov w8, #1
+; CHECK-NEXT:    mov w8, #1 // =0x1
 ; CHECK-NEXT:    cmp x0, #0
 ; CHECK-NEXT:    cneg x0, x8, le
 ; CHECK-NEXT:    ret
@@ -104,7 +104,7 @@
 define i64 @not_sign_i64_4(i64 %a) {
 ; CHECK-LABEL: not_sign_i64_4:
 ; CHECK:       // %bb.0:
-; CHECK-NEXT:    mov x0, #-1
+; CHECK-NEXT:    mov x0, #-1 // =0xffffffffffffffff
 ; CHECK-NEXT:    ret
   %c = icmp ugt i64 %a, -1
   %res = select i1 %c, i64 1, i64 -1
@@ -229,10 +229,10 @@
 ; CHECK:       // %bb.0:
 ; CHECK-NEXT:    movi v1.2d, #0xffffffffffffffff
 ; CHECK-NEXT:    adrp x8, .LCPI18_0
-; CHECK-NEXT:    movi v2.4s, #1
-; CHECK-NEXT:    ldr q3, [x8, :lo12:.LCPI18_0]
+; CHECK-NEXT:    ldr q2, [x8, :lo12:.LCPI18_0]
 ; CHECK-NEXT:    cmgt v0.4s, v0.4s, v1.4s
-; CHECK-NEXT:    bsl v0.16b, v2.16b, v3.16b
+; CHECK-NEXT:    bic v1.16b, v2.16b, v0.16b
+; CHECK-NEXT:    sub v0.4s, v1.4s, v0.4s
 ; CHECK-NEXT:    ret
   %c = icmp sgt <4 x i32> %a, <i32 -1, i32 -1, i32 -1, i32 -1>
   %res = select <4 x i1> %c, <4 x i32> <i32 1, i32 1, i32 1, i32 1>, <4 x i32> <i32 -1, i32 -1, i32 -1, i32 1>
Index: llvm/lib/Target/AArch64/AArch64ISelLowering.h
===================================================================
--- llvm/lib/Target/AArch64/AArch64ISelLowering.h
+++ llvm/lib/Target/AArch64/AArch64ISelLowering.h
@@ -535,6 +535,11 @@
                                      const SelectionDAG &DAG,
                                      unsigned Depth = 0) const override;
 
+  unsigned ComputeNumSignBitsForTargetNode(SDValue Op,
+                                           const APInt &DemandedElts,
+                                           const SelectionDAG &DAG,
+                                           unsigned Depth) const override;
+
   MVT getPointerTy(const DataLayout &DL, uint32_t AS = 0) const override {
     // Returning i64 unconditionally here (i.e. even for ILP32) means that the
     // *DAG* representation of pointers will always be 64-bits. They will be
Index: llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
===================================================================
--- llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
+++ llvm/lib/Target/AArch64/AArch64ISelLowering.cpp
@@ -2134,6 +2134,38 @@
   }
 }
 
+unsigned AArch64TargetLowering::ComputeNumSignBitsForTargetNode(
+    SDValue Op, const APInt &DemandedElts, const SelectionDAG &DAG,
+    unsigned Depth) const {
+  EVT VT = Op.getValueType();
+  unsigned VTBits = VT.getScalarSizeInBits();
+  unsigned Opcode = Op.getOpcode();
+  switch (Opcode) {
+    case AArch64ISD::CMEQ:
+    case AArch64ISD::CMGE:
+    case AArch64ISD::CMGT:
+    case AArch64ISD::CMHI:
+    case AArch64ISD::CMHS:
+    case AArch64ISD::FCMEQ:
+    case AArch64ISD::FCMGE:
+    case AArch64ISD::FCMGT:
+    case AArch64ISD::CMEQz:
+    case AArch64ISD::CMGEz:
+    case AArch64ISD::CMGTz:
+    case AArch64ISD::CMLEz:
+    case AArch64ISD::CMLTz:
+    case AArch64ISD::FCMEQz:
+    case AArch64ISD::FCMGEz:
+    case AArch64ISD::FCMGTz:
+    case AArch64ISD::FCMLEz:
+    case AArch64ISD::FCMLTz:
+      // Compares return either 0 or all-ones
+      return VTBits;
+  }
+
+  return 1;
+}
+
 MVT AArch64TargetLowering::getScalarShiftAmountTy(const DataLayout &DL,
                                                   EVT) const {
   return MVT::i64;


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D148624.514591.patch
Type: text/x-patch
Size: 3625 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20230418/df1b5aab/attachment.bin>