[DAGCombine] Fix for PR35765

[samparker]

Summary

Remove the acceptance of ANY_EXTEND nodes while trying to move and nodes back to loads.

Bugzilla: https://bugs.llvm.org/show_bug.cgi?id=35765

[RKSimon]

test/CodeGen/X86/pr35765.ll

2

Please use the utils\update_llc_test_checks.py to regenerate the checks

[samparker]

Generated test.

[RKSimon]

LGTM thanks. If possible, please commit pr35765.ll before the patch with the current codegen, then commit the patch with the fixed codegen in pr35765.ll to show the change,

lib/CodeGen/SelectionDAG/DAGCombiner.cpp

         // Only add this load if we can make it more narrow.
         if (ExtVT.bitsLT(Load->getMemoryVT()))
           Loads.insert(Load);
         continue;
       }
       return false;
     }
     case ISD::ZERO_EXTEND:
-    case ISD::ANY_EXTEND:
     case ISD::AssertZext: {
       unsigned ActiveBits = Mask->getAPIntValue().countTrailingOnes();
       EVT ExtVT = EVT::getIntegerVT(*DAG.getContext(), ActiveBits);
       EVT VT = Op.getOpcode() == ISD::AssertZext ?
         cast<VTSDNode>(Op.getOperand(1))->getVT() :
         Op.getOperand(0).getValueType();
 
       // We can accept extending nodes if the mask is wider or an equal

test/CodeGen/X86/pr35765.ll

+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc -mtriple=x86_64-unknown-linux-gnu %s -o - | FileCheck %s
+
+@ll = local_unnamed_addr global i64 0, align 8
+@x = local_unnamed_addr global i64 2651237805702985558, align 8
+@s1 = local_unnamed_addr global { i8, i8 } { i8 123, i8 5 }, align 2
+@s2 = local_unnamed_addr global { i8, i8 } { i8 -122, i8 3 }, align 2
+
+define void @PR35765() {
+; CHECK-LABEL: PR35765:
+; CHECK:       # %bb.0: # %entry
+; CHECK-NEXT:    movzwl {{.*}}(%rip), %ecx
+; CHECK-NEXT:    addl $-1398, %ecx # imm = 0xFA8A
+; CHECK-NEXT:    movl $4, %eax
+; CHECK-NEXT:    # kill: def %cl killed %cl killed %ecx
+; CHECK-NEXT:    shll %cl, %eax
+; CHECK-NEXT:    movzwl {{.*}}(%rip), %ecx
+; CHECK-NEXT:    movzwl {{.*}}(%rip), %edx
+; CHECK-NEXT:    notl %edx
+; CHECK-NEXT:    orl $63488, %edx # imm = 0xF800
+; CHECK-NEXT:    movzwl %dx, %edx
+; CHECK-NEXT:    orl %ecx, %edx
+; CHECK-NEXT:    xorl %eax, %edx
+; CHECK-NEXT:    movslq %edx, %rax
+; CHECK-NEXT:    movq %rax, {{.*}}(%rip)
+; CHECK-NEXT:    retq
+entry:
+  %bf.load.i = load i16, i16* bitcast ({ i8, i8 }* @s1 to i16*), align 2
+  %bf.clear.i = and i16 %bf.load.i, 2047
+  %conv.i = zext i16 %bf.clear.i to i32
+  %sub.i = add nsw i32 %conv.i, -1398
+  %shl.i = shl i32 4, %sub.i
+  %0 = load i64, i64* @x, align 8
+  %bf.load1.i = load i16, i16* bitcast ({ i8, i8 }* @s2 to i16*), align 2
+  %bf.clear2.i = and i16 %bf.load1.i, 2047
+  %1 = xor i16 %bf.clear2.i, -1
+  %neg.i = zext i16 %1 to i64
+  %or.i = or i64 %0, %neg.i
+  %conv5.i = trunc i64 %or.i to i32
+  %conv6.i = and i32 %conv5.i, 65535
+  %xor.i = xor i32 %conv6.i, %shl.i
+  %conv7.i = sext i32 %xor.i to i64
+  store i64 %conv7.i, i64* @ll, align 8
+  ret void
+}