[X86] Don't allow optimizeCompareInstr to replace a CMP with BEXTR if the sign flag...

The BEXTR instruction documents the SF bit as undefined.

The TBM BEXTR instruction has the same issue, but I'm not sure how to test it. With the control being an immediate we can determine the sign bit is 0 or the BEXTR would have been removed.

Fixes PR40060

Differential Revision: https://reviews.llvm.org/D55807

llvm-svn: 349956

diff --git a/llvm/lib/Target/X86/X86InstrInfo.cpp b/llvm/lib/Target/X86/X86InstrInfo.cpp
index 40de049..ab14ee7 100644 (file)
--- a/llvm/lib/Target/X86/X86InstrInfo.cpp
+++ b/llvm/lib/Target/X86/X86InstrInfo.cpp
@@ -3455,7 +3455,9 @@ inline static bool isRedundantFlagInstr(const MachineInstr &FlagI,
 
 /// Check whether the definition can be converted
 /// to remove a comparison against zero.
-inline static bool isDefConvertible(const MachineInstr &MI) {
+inline static bool isDefConvertible(const MachineInstr &MI, bool &NoSignFlag) {
+  NoSignFlag = false;
+
   switch (MI.getOpcode()) {
   default: return false;
 
@@ -3520,8 +3522,6 @@ inline static bool isDefConvertible(const MachineInstr &MI) {
   case X86::SHL8r1:    case X86::SHL16r1:  case X86::SHL32r1:case X86::SHL64r1:
   case X86::ANDN32rr:  case X86::ANDN32rm:
   case X86::ANDN64rr:  case X86::ANDN64rm:
-  case X86::BEXTR32rr: case X86::BEXTR64rr:
-  case X86::BEXTR32rm: case X86::BEXTR64rm:
   case X86::BLSI32rr:  case X86::BLSI32rm:
   case X86::BLSI64rr:  case X86::BLSI64rm:
   case X86::BLSMSK32rr:case X86::BLSMSK32rm:
@@ -3539,8 +3539,6 @@ inline static bool isDefConvertible(const MachineInstr &MI) {
   case X86::TZCNT16rr: case X86::TZCNT16rm:
   case X86::TZCNT32rr: case X86::TZCNT32rm:
   case X86::TZCNT64rr: case X86::TZCNT64rm:
-  case X86::BEXTRI32ri:  case X86::BEXTRI32mi:
-  case X86::BEXTRI64ri:  case X86::BEXTRI64mi:
   case X86::BLCFILL32rr: case X86::BLCFILL32rm:
   case X86::BLCFILL64rr: case X86::BLCFILL64rm:
   case X86::BLCI32rr:    case X86::BLCI32rm:
@@ -3560,6 +3558,13 @@ inline static bool isDefConvertible(const MachineInstr &MI) {
   case X86::TZMSK32rr:   case X86::TZMSK32rm:
   case X86::TZMSK64rr:   case X86::TZMSK64rm:
     return true;
+  case X86::BEXTR32rr:   case X86::BEXTR64rr:
+  case X86::BEXTR32rm:   case X86::BEXTR64rm:
+  case X86::BEXTRI32ri:  case X86::BEXTRI32mi:
+  case X86::BEXTRI64ri:  case X86::BEXTRI64mi:
+    // BEXTR doesn't update the sign flag so we can't use it.
+    NoSignFlag = true;
+    return true;
   }
 }
 
@@ -3662,8 +3667,9 @@ bool X86InstrInfo::optimizeCompareInstr(MachineInstr &CmpInstr, unsigned SrcReg,
   // instruction we can eliminate the compare iff the use sets EFLAGS in the
   // right way.
   bool ShouldUpdateCC = false;
+  bool NoSignFlag = false;
   X86::CondCode NewCC = X86::COND_INVALID;
-  if (IsCmpZero && !isDefConvertible(*MI)) {
+  if (IsCmpZero && !isDefConvertible(*MI, NoSignFlag)) {
     // Scan forward from the use until we hit the use we're looking for or the
     // compare instruction.
     for (MachineBasicBlock::iterator J = MI;; ++J) {
@@ -3782,6 +3788,12 @@ bool X86InstrInfo::optimizeCompareInstr(MachineInstr &CmpInstr, unsigned SrcReg,
       case X86::COND_O: case X86::COND_NO:
         // CF and OF are used, we can't perform this optimization.
         return false;
+      case X86::COND_S: case X86::COND_NS:
+        // If SF is used, but the instruction doesn't update the SF, then we
+        // can't do the optimization.
+        if (NoSignFlag)
+          return false;
+        break;
       }
 
       // If we're updating the condition code check if we have to reverse the

diff --git a/llvm/test/CodeGen/X86/bmi.ll b/llvm/test/CodeGen/X86/bmi.ll
index 03d6df0..ee7cf5f 100644 (file)
--- a/llvm/test/CodeGen/X86/bmi.ll
+++ b/llvm/test/CodeGen/X86/bmi.ll
@@ -1032,6 +1032,7 @@ define void @pr40060(i32, i32) {
 ; X86:       # %bb.0:
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NEXT:    bextrl %eax, {{[0-9]+}}(%esp), %eax
+; X86-NEXT:    testl %eax, %eax
 ; X86-NEXT:    js .LBB45_1
 ; X86-NEXT:  # %bb.2:
 ; X86-NEXT:    jmp bar # TAILCALL
@@ -1041,6 +1042,7 @@ define void @pr40060(i32, i32) {
 ; X64-LABEL: pr40060:
 ; X64:       # %bb.0:
 ; X64-NEXT:    bextrl %esi, %edi, %eax
+; X64-NEXT:    testl %eax, %eax
 ; X64-NEXT:    js .LBB45_1
 ; X64-NEXT:  # %bb.2:
 ; X64-NEXT:    jmp bar # TAILCALL