For something like

uint32_t hi(uint64_t res)
{
        uint_32t hi = res >> 32;
        return !hi;
}

llvm IR looks like this:
define i32 @hi(i64 %res) nounwind uwtable ssp {
entry:
  %lnot = icmp ult i64 %res, 4294967296
  %lnot.ext = zext i1 %lnot to i32
  ret i32 %lnot.ext
}

The optimizer has optimize away the right shift and truncate but the resulting
constant is too large to fit in the 32-bit immediate field. The resulting x86
code is worse as a result:
        movabsq $4294967296, %rax       ## imm = 0x100000000
        cmpq    %rax, %rdi
        sbbl    %eax, %eax
        andl    $1, %eax

This patch teaches the x86 lowering code to handle ult against a large immediate
with trailing zeros. It will issue a right shift and a truncate followed by
a comparison against a shifted immediate.
        shrq    $32, %rdi
        testl   %edi, %edi
        sete    %al
        movzbl  %al, %eax

It also handles a ugt comparison against a large immediate with trailing bits
set. i.e. X >  0x0ffffffff -> (X >> 32) >= 1

rdar://11866926


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@160312 91177308-0d34-0410-b5e6-96231b3b80d8

3 files changed, 94 insertions, 2 deletions

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index c55a1ef..1d72aad 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -3059,6 +3059,50 @@ static unsigned TranslateX86CC(ISD::CondCode SetCCOpcode, bool isFP,
         RHS = DAG.getConstant(0, RHS.getValueType());
         return X86::COND_LE;
       }
+      if (SetCCOpcode == ISD::SETULT || SetCCOpcode == ISD::SETUGE) {
+        unsigned TrailZeros = RHSC->getAPIntValue().countTrailingZeros();
+        if (TrailZeros >= 32) {
+          // The constant doesn't fit in cmp immediate field. Right shift LHS by
+          // the # of trailing zeros and truncate it to 32-bit. Then compare
+          // against shifted RHS.
+          assert(LHS.getValueType() == MVT::i64 && "Expecting a 64-bit cmp!");
+          DebugLoc dl = LHS.getDebugLoc();
+          LHS = DAG.getNode(ISD::TRUNCATE, dl, MVT::i32,
+                            DAG.getNode(ISD::SRL, dl, MVT::i64, LHS,
+                                        DAG.getConstant(TrailZeros, MVT::i8)));
+          uint64_t C = RHSC->getZExtValue() >> TrailZeros;
+
+          if (SetCCOpcode == ISD::SETULT) {
+            // X < 0x300000000 -> (X >> 32) < 3
+            // X < 0x100000000 -> (X >> 32) == 0
+            // X < 0x200000000 -> (X >> 33) == 0
+            if (C == 1) {
+              RHS = DAG.getConstant(0, MVT::i32);
+              return X86::COND_E;
+            }
+            RHS = DAG.getConstant(C, MVT::i32);
+            return X86::COND_B;
+          } else /* SetCCOpcode == ISD::SETUGE */ {
+            // X >= 0x100000000 -> (X >> 32) >= 1
+            RHS = DAG.getConstant(C, MVT::i32);
+            return X86::COND_AE;
+          }
+        }
+      }
+      if (SetCCOpcode == ISD::SETUGT) {
+        unsigned TrailOnes = RHSC->getAPIntValue().countTrailingOnes();
+        if (TrailOnes >= 32 && !RHSC->isAllOnesValue()) {
+          assert(LHS.getValueType() == MVT::i64 && "Expecting a 64-bit cmp!");
+          DebugLoc dl = LHS.getDebugLoc();
+          LHS = DAG.getNode(ISD::TRUNCATE, dl, MVT::i32,
+                            DAG.getNode(ISD::SRL, dl, MVT::i64, LHS,
+                                        DAG.getConstant(TrailOnes, MVT::i8)));
+          uint64_t C = (RHSC->getZExtValue()+1) >> TrailOnes;
+          // X >  0x0ffffffff -> (X >> 32) >= 1
+          RHS = DAG.getConstant(C, MVT::i32);
+          return X86::COND_AE;
+        }
+      }
     }
 
     switch (SetCCOpcode) {
diff --git a/test/CodeGen/X86/2011-04-13-SchedCmpJmp.ll b/test/CodeGen/X86/2011-04-13-SchedCmpJmp.ll
index c6f4b49..be10ad5 100644
--- a/test/CodeGen/X86/2011-04-13-SchedCmpJmp.ll
+++ b/test/CodeGen/X86/2011-04-13-SchedCmpJmp.ll
@@ -12,9 +12,9 @@ declare hidden fastcc void @_ZN3JSCL23returnToThrowTrampolineEPNS_12JSGlobalData
 
 ; Avoid hoisting the test above loads or copies
 ; CHECK: %entry
-; CHECK: cmpq
+; CHECK: test
 ; CHECK-NOT: mov
-; CHECK: jb
+; CHECK: je
 define i32 @cti_op_eq(i8** nocapture %args) nounwind ssp {
 entry:
   %0 = load i8** null, align 8
diff --git a/test/CodeGen/X86/cmp.ll b/test/CodeGen/X86/cmp.ll
index ef5e353..c9c85ab 100644
--- a/test/CodeGen/X86/cmp.ll
+++ b/test/CodeGen/X86/cmp.ll
@@ -90,3 +90,51 @@ F:
 ; CHECK: encoding: [0x48,0x83,0x7c,0x24,0xf8,0x00]
 }
 
+; rdar://11866926
+define i32 @test7(i64 %res) nounwind uwtable readnone ssp {
+entry:
+; CHECK: test7:
+; CHECK-NOT: movabsq
+; CHECK: shrq $32, %rdi
+; CHECK: testl %edi, %edi
+; CHECK: sete
+  %lnot = icmp ult i64 %res, 4294967296
+  %lnot.ext = zext i1 %lnot to i32
+  ret i32 %lnot.ext
+}
+
+define i32 @test8(i64 %res) nounwind uwtable readnone ssp {
+entry:
+; CHECK: test8:
+; CHECK-NOT: movabsq
+; CHECK: shrq $32, %rdi
+; CHECK: cmpl $3, %edi
+  %lnot = icmp ult i64 %res, 12884901888
+  %lnot.ext = zext i1 %lnot to i32
+  ret i32 %lnot.ext
+}
+
+define i32 @test9(i64 %res) nounwind uwtable readnone ssp {
+entry:
+; CHECK: test9:
+; CHECK-NOT: movabsq
+; CHECK: shrq $33, %rdi
+; CHECK: testl %edi, %edi
+; CHECK: sete
+  %lnot = icmp ult i64 %res, 8589934592
+  %lnot.ext = zext i1 %lnot to i32
+  ret i32 %lnot.ext
+}
+
+define i32 @test10(i64 %res) nounwind uwtable readnone ssp {
+entry:
+; CHECK: test10:
+; CHECK-NOT: movabsq
+; CHECK: shrq $32, %rdi
+; CHECK: cmpl $1, %edi
+; CHECK: setae
+  %lnot = icmp uge i64 %res, 4294967296
+  %lnot.ext = zext i1 %lnot to i32
+  ret i32 %lnot.ext
+}
+