[llvm] r345017 - Revert "[X86][BMI1] X86DAGToDAGISel: select BEXTR from x << (32 - y) >> (32 - y) pattern"

Roman Lebedev via llvm-commits llvm-commits at lists.llvm.org
Tue Oct 23 03:34:57 PDT 2018


Author: lebedevri
Date: Tue Oct 23 03:34:57 2018
New Revision: 345017

URL: http://llvm.org/viewvc/llvm-project?rev=345017&view=rev
Log:
Revert "[X86][BMI1] X86DAGToDAGISel: select BEXTR from x << (32 - y) >> (32 - y) pattern"

*Seems* to be breaking sanitizer-x86_64-linux-fast buildbot,
the ELF/relocatable-versioned.s test:

==17758==MemorySanitizer CHECK failed: /b/sanitizer-x86_64-linux-fast/build/llvm/projects/compiler-rt/lib/sanitizer_common/sanitizer_allocator.cc:191 "((kBlockMagic)) == ((((u64*)addr)[0]))" (0x6a6cb03abcebc041, 0x0)
    #0 0x59716b in MsanCheckFailed(char const*, int, char const*, unsigned long long, unsigned long long) /b/sanitizer-x86_64-linux-fast/build/llvm/projects/compiler-rt/lib/msan/msan.cc:393
    #1 0x586635 in __sanitizer::CheckFailed(char const*, int, char const*, unsigned long long, unsigned long long) /b/sanitizer-x86_64-linux-fast/build/llvm/projects/compiler-rt/lib/sanitizer_common/sanitizer_termination.cc:79
    #2 0x57d5ff in __sanitizer::InternalFree(void*, __sanitizer::SizeClassAllocatorLocalCache<__sanitizer::SizeClassAllocator32<__sanitizer::AP32> >*) /b/sanitizer-x86_64-linux-fast/build/llvm/projects/compiler-rt/lib/sanitizer_common/sanitizer_allocator.cc:191
    #3 0x7fc21b24193f  (/lib/x86_64-linux-gnu/libc.so.6+0x3593f)
    #4 0x7fc21b241999 in exit (/lib/x86_64-linux-gnu/libc.so.6+0x35999)
    #5 0x7fc21b22c2e7 in __libc_start_main (/lib/x86_64-linux-gnu/libc.so.6+0x202e7)
    #6 0x57c039 in _start (/b/sanitizer-x86_64-linux-fast/build/llvm_build_msan/bin/lld+0x57c039)

This reverts commit r345014.

Modified:
    llvm/trunk/lib/Target/X86/X86ISelDAGToDAG.cpp
    llvm/trunk/lib/Target/X86/X86InstrInfo.td
    llvm/trunk/test/CodeGen/X86/extract-bits.ll
    llvm/trunk/test/CodeGen/X86/extract-lowbits.ll

Modified: llvm/trunk/lib/Target/X86/X86ISelDAGToDAG.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86ISelDAGToDAG.cpp?rev=345017&r1=345016&r2=345017&view=diff
==============================================================================
--- llvm/trunk/lib/Target/X86/X86ISelDAGToDAG.cpp (original)
+++ llvm/trunk/lib/Target/X86/X86ISelDAGToDAG.cpp Tue Oct 23 03:34:57 2018
@@ -2688,10 +2688,6 @@ bool X86DAGToDAGISel::foldLoadStoreIntoM
 //   c) x &  (-1 >> (32 - y))
 //   d) x << (32 - y) >> (32 - y)
 bool X86DAGToDAGISel::matchBitExtract(SDNode *Node) {
-  assert(
-      (Node->getOpcode() == ISD::AND || Node->getOpcode() == ISD::SRL) &&
-      "Should be either an and-mask, or right-shift after clearing high bits.");
-
   // BEXTR is BMI instruction, BZHI is BMI2 instruction. We need at least one.
   if (!Subtarget->hasBMI() && !Subtarget->hasBMI2())
     return false;
@@ -2702,16 +2698,13 @@ bool X86DAGToDAGISel::matchBitExtract(SD
   if (NVT != MVT::i32 && NVT != MVT::i64)
     return false;
 
-  unsigned Size = NVT.getSizeInBits();
-
   SDValue NBits;
 
   // If we have BMI2's BZHI, we are ok with muti-use patterns.
   // Else, if we only have BMI1's BEXTR, we require one-use.
   const bool CanHaveExtraUses = Subtarget->hasBMI2();
-  auto checkOneUse = [CanHaveExtraUses](SDValue Op, unsigned NUses = 1) {
-    return CanHaveExtraUses ||
-           Op.getNode()->hasNUsesOfValue(NUses, Op.getResNo());
+  auto checkOneUse = [CanHaveExtraUses](SDValue Op) {
+    return CanHaveExtraUses || Op.hasOneUse();
   };
 
   // a) x & ((1 << nbits) + (-1))
@@ -2747,73 +2740,31 @@ bool X86DAGToDAGISel::matchBitExtract(SD
     return true;
   };
 
-  SDValue X;
-
-  // d) x << (32 - y) >> (32 - y)
-  auto matchPatternD = [&checkOneUse, Size, &X, &NBits](SDNode *Node) -> bool {
-    if (Node->getOpcode() != ISD::SRL)
-      return false;
-    SDValue N0 = Node->getOperand(0);
-    if (N0->getOpcode() != ISD::SHL || !checkOneUse(N0))
-      return false;
-    SDValue N1 = Node->getOperand(1);
-    SDValue N01 = N0->getOperand(1);
-    // Both of the shifts must be by the exact same value.
-    // There should not be any uses of the shift amount outside of the pattern.
-    if (N1 != N01 || !checkOneUse(N1, 2))
-      return false;
-    // Skip over a truncate of the shift amount.
-    if (N1->getOpcode() == ISD::TRUNCATE) {
-      N1 = N1->getOperand(0);
-      // The trunc should have been the only user of the real shift amount.
-      if (!checkOneUse(N1))
-        return false;
-    }
-    // Match the shift amount as: (bitwidth - y). It should go away, too.
-    if (N1.getOpcode() != ISD::SUB)
-      return false;
-    auto N10 = dyn_cast<ConstantSDNode>(N1.getOperand(0));
-    if (!N10 || N10->getZExtValue() != Size)
-      return false;
-    X = N0->getOperand(0);
-    NBits = N1.getOperand(1);
-    return true;
-  };
-
   auto matchLowBitMask = [&matchPatternA,
                           &matchPatternB](SDValue Mask) -> bool {
-    // FIXME: pattern c.
+    // FIXME: patterns c, d.
     return matchPatternA(Mask) || matchPatternB(Mask);
   };
 
-  if (Node->getOpcode() == ISD::AND) {
-    X = Node->getOperand(0);
-    SDValue Mask = Node->getOperand(1);
-
-    if (matchLowBitMask(Mask)) {
-      // Great.
-    } else {
-      std::swap(X, Mask);
-      if (!matchLowBitMask(Mask))
-        return false;
-    }
-  } else if (!matchPatternD(Node))
-    return false;
-
-  SDLoc DL(Node);
+  SDValue X = Node->getOperand(0);
+  SDValue Mask = Node->getOperand(1);
 
-  SDValue OrigNBits = NBits;
-  // Do we need to truncate the shift amount?
-  if (NBits.getValueType() != MVT::i8) {
-    NBits = CurDAG->getNode(ISD::TRUNCATE, DL, MVT::i8, NBits);
-    insertDAGNode(*CurDAG, OrigNBits, NBits);
+  if (matchLowBitMask(Mask)) {
+    // Great.
+  } else {
+    std::swap(X, Mask);
+    if (!matchLowBitMask(Mask))
+      return false;
   }
 
+  SDLoc DL(Node);
+
   // Insert 8-bit NBits into lowest 8 bits of NVT-sized (32 or 64-bit) register.
   // All the other bits are undefined, we do not care about them.
   SDValue ImplDef =
       SDValue(CurDAG->getMachineNode(TargetOpcode::IMPLICIT_DEF, DL, NVT), 0);
   insertDAGNode(*CurDAG, NBits, ImplDef);
+  SDValue OrigNBits = NBits;
   NBits = CurDAG->getTargetInsertSubreg(X86::sub_8bit, DL, NVT, ImplDef, NBits);
   insertDAGNode(*CurDAG, OrigNBits, NBits);
 
@@ -3012,8 +2963,17 @@ bool X86DAGToDAGISel::tryShiftAmountMod(
   if (ShiftAmt->getOpcode() == ISD::TRUNCATE)
     ShiftAmt = ShiftAmt->getOperand(0);
 
-  // This function is called after X86DAGToDAGISel::matchBitExtract(),
-  // so we are not afraid that we might mess up BZHI/BEXTR pattern.
+  // Special case to avoid messing up a BZHI pattern.
+  // Look for (srl (shl X, (size - y)), (size - y)
+  if (Subtarget->hasBMI2() && (VT == MVT::i32 || VT == MVT::i64) &&
+      N->getOpcode() == ISD::SRL && N->getOperand(0).getOpcode() == ISD::SHL &&
+      // Shift amounts the same?
+      N->getOperand(1) == N->getOperand(0).getOperand(1) &&
+      // Shift amounts size - y?
+      ShiftAmt.getOpcode() == ISD::SUB &&
+      isa<ConstantSDNode>(ShiftAmt.getOperand(0)) &&
+      cast<ConstantSDNode>(ShiftAmt.getOperand(0))->getZExtValue() == Size)
+    return false;
 
   SDValue NewShiftAmt;
   if (ShiftAmt->getOpcode() == ISD::ADD || ShiftAmt->getOpcode() == ISD::SUB) {
@@ -3212,9 +3172,6 @@ void X86DAGToDAGISel::Select(SDNode *Nod
   }
 
   case ISD::SRL:
-    if (matchBitExtract(Node))
-      return;
-    LLVM_FALLTHROUGH;
   case ISD::SRA:
   case ISD::SHL:
     if (tryShiftAmountMod(Node))

Modified: llvm/trunk/lib/Target/X86/X86InstrInfo.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86InstrInfo.td?rev=345017&r1=345016&r2=345017&view=diff
==============================================================================
--- llvm/trunk/lib/Target/X86/X86InstrInfo.td (original)
+++ llvm/trunk/lib/Target/X86/X86InstrInfo.td Tue Oct 23 03:34:57 2018
@@ -2519,6 +2519,14 @@ let Predicates = [HasBMI2] in {
                              (and (x86memop addr:$src),
                                   (srl -1, (sub bitwidth, GR8:$lz))),
                              RC, VT, DstInst, DstMemInst>;
+
+    // x << (bitwidth - y) >> (bitwidth - y)
+    defm : _bmi_bzhi_pattern<(srl (shl RC:$src, (sub bitwidth, GR8:$lz)),
+                                  (sub bitwidth, GR8:$lz)),
+                             (srl (shl (x86memop addr:$src),
+                                        (sub bitwidth, GR8:$lz)),
+                                  (sub bitwidth, GR8:$lz)),
+                             RC, VT, DstInst, DstMemInst>;
   }
 
   defm : bmi_bzhi_patterns<GR32, 32, i32, BZHI32rr, loadi32, BZHI32rm>;
@@ -2537,6 +2545,24 @@ let Predicates = [HasBMI2] in {
   def : Pat<(and (loadi64 addr:$src), (srl -1, (i8 (trunc (sub 64, GR32:$lz))))),
             (BZHI64rm addr:$src,
               (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GR32:$lz, sub_32bit))>;
+
+  // x << (32 - y) >> (32 - y)
+  def : Pat<(srl (shl GR32:$src, (i8 (trunc (sub 32, GR32:$lz)))),
+                 (i8 (trunc (sub 32, GR32:$lz)))),
+            (BZHI32rr GR32:$src, GR32:$lz)>;
+  def : Pat<(srl (shl (loadi32 addr:$src), (i8 (trunc (sub 32, GR32:$lz)))),
+                 (i8 (trunc (sub 32, GR32:$lz)))),
+            (BZHI32rm addr:$src, GR32:$lz)>;
+
+  // x << (64 - y) >> (64 - y)
+  def : Pat<(srl (shl GR64:$src, (i8 (trunc (sub 64, GR32:$lz)))),
+                 (i8 (trunc (sub 64, GR32:$lz)))),
+            (BZHI64rr GR64:$src,
+              (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GR32:$lz, sub_32bit))>;
+  def : Pat<(srl (shl (loadi64 addr:$src), (i8 (trunc (sub 64, GR32:$lz)))),
+                 (i8 (trunc (sub 64, GR32:$lz)))),
+            (BZHI64rm addr:$src,
+              (INSERT_SUBREG (i64 (IMPLICIT_DEF)), GR32:$lz, sub_32bit))>;
 } // HasBMI2
 
 multiclass bmi_pdep_pext<string mnemonic, RegisterClass RC,

Modified: llvm/trunk/test/CodeGen/X86/extract-bits.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/extract-bits.ll?rev=345017&r1=345016&r2=345017&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/extract-bits.ll (original)
+++ llvm/trunk/test/CodeGen/X86/extract-bits.ll Tue Oct 23 03:34:57 2018
@@ -2813,12 +2813,14 @@ define i32 @bextr32_c0(i32 %val, i32 %nu
 ;
 ; X86-BMI1NOTBM-LABEL: bextr32_c0:
 ; X86-BMI1NOTBM:       # %bb.0:
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
-; X86-BMI1NOTBM-NEXT:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, %edx, %eax
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    xorl %ecx, %ecx
+; X86-BMI1NOTBM-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bextr32_c0:
@@ -2844,10 +2846,13 @@ define i32 @bextr32_c0(i32 %val, i32 %nu
 ; X64-BMI1NOTBM-LABEL: bextr32_c0:
 ; X64-BMI1NOTBM:       # %bb.0:
 ; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movl %edi, %eax
 ; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
-; X64-BMI1NOTBM-NEXT:    shrl %cl, %edi
-; X64-BMI1NOTBM-NEXT:    shll $8, %edx
-; X64-BMI1NOTBM-NEXT:    bextrl %edx, %edi, %eax
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X64-BMI1NOTBM-NEXT:    negl %edx
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr32_c0:
@@ -2877,12 +2882,14 @@ define i32 @bextr32_c1_indexzext(i32 %va
 ;
 ; X86-BMI1NOTBM-LABEL: bextr32_c1_indexzext:
 ; X86-BMI1NOTBM:       # %bb.0:
-; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %al
 ; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
-; X86-BMI1NOTBM-NEXT:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, %edx, %eax
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    xorl %ecx, %ecx
+; X86-BMI1NOTBM-NEXT:    subb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bextr32_c1_indexzext:
@@ -2908,10 +2915,13 @@ define i32 @bextr32_c1_indexzext(i32 %va
 ; X64-BMI1NOTBM-LABEL: bextr32_c1_indexzext:
 ; X64-BMI1NOTBM:       # %bb.0:
 ; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movl %edi, %eax
 ; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
-; X64-BMI1NOTBM-NEXT:    shrl %cl, %edi
-; X64-BMI1NOTBM-NEXT:    shll $8, %edx
-; X64-BMI1NOTBM-NEXT:    bextrl %edx, %edi, %eax
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X64-BMI1NOTBM-NEXT:    negb %dl
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr32_c1_indexzext:
@@ -2944,13 +2954,15 @@ define i32 @bextr32_c2_load(i32* %w, i32
 ;
 ; X86-BMI1NOTBM-LABEL: bextr32_c2_load:
 ; X86-BMI1NOTBM:       # %bb.0:
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-BMI1NOTBM-NEXT:    movl (%edx), %edx
-; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
-; X86-BMI1NOTBM-NEXT:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, %edx, %eax
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    movl (%eax), %eax
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    xorl %ecx, %ecx
+; X86-BMI1NOTBM-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bextr32_c2_load:
@@ -2980,8 +2992,10 @@ define i32 @bextr32_c2_load(i32* %w, i32
 ; X64-BMI1NOTBM-NEXT:    movl (%rdi), %eax
 ; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
 ; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
-; X64-BMI1NOTBM-NEXT:    shll $8, %edx
-; X64-BMI1NOTBM-NEXT:    bextrl %edx, %eax, %eax
+; X64-BMI1NOTBM-NEXT:    negl %edx
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr32_c2_load:
@@ -3013,13 +3027,15 @@ define i32 @bextr32_c3_load_indexzext(i3
 ;
 ; X86-BMI1NOTBM-LABEL: bextr32_c3_load_indexzext:
 ; X86-BMI1NOTBM:       # %bb.0:
-; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %al
 ; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-BMI1NOTBM-NEXT:    movl (%edx), %edx
-; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
-; X86-BMI1NOTBM-NEXT:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, %edx, %eax
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    movl (%eax), %eax
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    xorl %ecx, %ecx
+; X86-BMI1NOTBM-NEXT:    subb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bextr32_c3_load_indexzext:
@@ -3049,8 +3065,10 @@ define i32 @bextr32_c3_load_indexzext(i3
 ; X64-BMI1NOTBM-NEXT:    movl (%rdi), %eax
 ; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
 ; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
-; X64-BMI1NOTBM-NEXT:    shll $8, %edx
-; X64-BMI1NOTBM-NEXT:    bextrl %edx, %eax, %eax
+; X64-BMI1NOTBM-NEXT:    negb %dl
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr32_c3_load_indexzext:
@@ -3083,12 +3101,14 @@ define i32 @bextr32_c4_commutative(i32 %
 ;
 ; X86-BMI1NOTBM-LABEL: bextr32_c4_commutative:
 ; X86-BMI1NOTBM:       # %bb.0:
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
-; X86-BMI1NOTBM-NEXT:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, %edx, %eax
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    xorl %ecx, %ecx
+; X86-BMI1NOTBM-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bextr32_c4_commutative:
@@ -3114,10 +3134,13 @@ define i32 @bextr32_c4_commutative(i32 %
 ; X64-BMI1NOTBM-LABEL: bextr32_c4_commutative:
 ; X64-BMI1NOTBM:       # %bb.0:
 ; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movl %edi, %eax
 ; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
-; X64-BMI1NOTBM-NEXT:    shrl %cl, %edi
-; X64-BMI1NOTBM-NEXT:    shll $8, %edx
-; X64-BMI1NOTBM-NEXT:    bextrl %edx, %edi, %eax
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X64-BMI1NOTBM-NEXT:    negl %edx
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr32_c4_commutative:
@@ -3157,13 +3180,16 @@ define i32 @bextr32_c5_skipextrauses(i32
 ; X86-BMI1NOTBM:       # %bb.0:
 ; X86-BMI1NOTBM-NEXT:    pushl %esi
 ; X86-BMI1NOTBM-NEXT:    subl $8, %esp
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %esi
 ; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
-; X86-BMI1NOTBM-NEXT:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, %edx, %esi
-; X86-BMI1NOTBM-NEXT:    movl %ecx, (%esp)
+; X86-BMI1NOTBM-NEXT:    movl %eax, %ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %esi
+; X86-BMI1NOTBM-NEXT:    xorl %ecx, %ecx
+; X86-BMI1NOTBM-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %esi
+; X86-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %esi
+; X86-BMI1NOTBM-NEXT:    movl %eax, (%esp)
 ; X86-BMI1NOTBM-NEXT:    calll use32
 ; X86-BMI1NOTBM-NEXT:    movl %esi, %eax
 ; X86-BMI1NOTBM-NEXT:    addl $8, %esp
@@ -3204,10 +3230,13 @@ define i32 @bextr32_c5_skipextrauses(i32
 ; X64-BMI1NOTBM-LABEL: bextr32_c5_skipextrauses:
 ; X64-BMI1NOTBM:       # %bb.0:
 ; X64-BMI1NOTBM-NEXT:    pushq %rbx
+; X64-BMI1NOTBM-NEXT:    movl %edi, %ebx
 ; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
-; X64-BMI1NOTBM-NEXT:    shrl %cl, %edi
-; X64-BMI1NOTBM-NEXT:    shll $8, %edx
-; X64-BMI1NOTBM-NEXT:    bextrl %edx, %edi, %ebx
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %ebx
+; X64-BMI1NOTBM-NEXT:    negl %edx
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %ebx
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %ebx
 ; X64-BMI1NOTBM-NEXT:    movl %esi, %edi
 ; X64-BMI1NOTBM-NEXT:    callq use32
 ; X64-BMI1NOTBM-NEXT:    movl %ebx, %eax
@@ -3350,10 +3379,13 @@ define i64 @bextr64_c0(i64 %val, i64 %nu
 ; X64-BMI1NOTBM-LABEL: bextr64_c0:
 ; X64-BMI1NOTBM:       # %bb.0:
 ; X64-BMI1NOTBM-NEXT:    movq %rsi, %rcx
+; X64-BMI1NOTBM-NEXT:    movq %rdi, %rax
 ; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $rcx
-; X64-BMI1NOTBM-NEXT:    shrq %cl, %rdi
-; X64-BMI1NOTBM-NEXT:    shlq $8, %rdx
-; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rdi, %rax
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    negl %edx
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr64_c0:
@@ -3483,12 +3515,14 @@ define i64 @bextr64_c1_indexzext(i64 %va
 ;
 ; X64-BMI1NOTBM-LABEL: bextr64_c1_indexzext:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    # kill: def $edx killed $edx def $rdx
 ; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movq %rdi, %rax
 ; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
-; X64-BMI1NOTBM-NEXT:    shrq %cl, %rdi
-; X64-BMI1NOTBM-NEXT:    shlq $8, %rdx
-; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rdi, %rax
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    negb %dl
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr64_c1_indexzext:
@@ -3629,8 +3663,10 @@ define i64 @bextr64_c2_load(i64* %w, i64
 ; X64-BMI1NOTBM-NEXT:    movq (%rdi), %rax
 ; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $rcx
 ; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
-; X64-BMI1NOTBM-NEXT:    shlq $8, %rdx
-; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rax, %rax
+; X64-BMI1NOTBM-NEXT:    negl %edx
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr64_c2_load:
@@ -3764,13 +3800,14 @@ define i64 @bextr64_c3_load_indexzext(i6
 ;
 ; X64-BMI1NOTBM-LABEL: bextr64_c3_load_indexzext:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    # kill: def $edx killed $edx def $rdx
 ; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
 ; X64-BMI1NOTBM-NEXT:    movq (%rdi), %rax
 ; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
 ; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
-; X64-BMI1NOTBM-NEXT:    shlq $8, %rdx
-; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rax, %rax
+; X64-BMI1NOTBM-NEXT:    negb %dl
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr64_c3_load_indexzext:
@@ -3906,10 +3943,13 @@ define i64 @bextr64_c4_commutative(i64 %
 ; X64-BMI1NOTBM-LABEL: bextr64_c4_commutative:
 ; X64-BMI1NOTBM:       # %bb.0:
 ; X64-BMI1NOTBM-NEXT:    movq %rsi, %rcx
+; X64-BMI1NOTBM-NEXT:    movq %rdi, %rax
 ; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $rcx
-; X64-BMI1NOTBM-NEXT:    shrq %cl, %rdi
-; X64-BMI1NOTBM-NEXT:    shlq $8, %rdx
-; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rdi, %rax
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    negl %edx
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr64_c4_commutative:
@@ -4089,10 +4129,13 @@ define i64 @bextr64_c5_skipextrauses(i64
 ; X64-BMI1NOTBM-LABEL: bextr64_c5_skipextrauses:
 ; X64-BMI1NOTBM:       # %bb.0:
 ; X64-BMI1NOTBM-NEXT:    pushq %rbx
-; X64-BMI1NOTBM-NEXT:    movq %rsi, %rcx
-; X64-BMI1NOTBM-NEXT:    shrq %cl, %rdi
-; X64-BMI1NOTBM-NEXT:    shlq $8, %rdx
-; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rdi, %rbx
+; X64-BMI1NOTBM-NEXT:    movq %rdi, %rbx
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rbx
+; X64-BMI1NOTBM-NEXT:    negl %edx
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rbx
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rbx
 ; X64-BMI1NOTBM-NEXT:    movq %rsi, %rdi
 ; X64-BMI1NOTBM-NEXT:    callq use64
 ; X64-BMI1NOTBM-NEXT:    movq %rbx, %rax
@@ -4136,12 +4179,14 @@ define i32 @bextr32_d0(i32 %val, i32 %nu
 ;
 ; X86-BMI1NOTBM-LABEL: bextr32_d0:
 ; X86-BMI1NOTBM:       # %bb.0:
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
-; X86-BMI1NOTBM-NEXT:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, %edx, %eax
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    xorl %ecx, %ecx
+; X86-BMI1NOTBM-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bextr32_d0:
@@ -4167,10 +4212,13 @@ define i32 @bextr32_d0(i32 %val, i32 %nu
 ; X64-BMI1NOTBM-LABEL: bextr32_d0:
 ; X64-BMI1NOTBM:       # %bb.0:
 ; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movl %edi, %eax
 ; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
-; X64-BMI1NOTBM-NEXT:    shrl %cl, %edi
-; X64-BMI1NOTBM-NEXT:    shll $8, %edx
-; X64-BMI1NOTBM-NEXT:    bextrl %edx, %edi, %eax
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X64-BMI1NOTBM-NEXT:    negl %edx
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr32_d0:
@@ -4200,12 +4248,14 @@ define i32 @bextr32_d1_indexzext(i32 %va
 ;
 ; X86-BMI1NOTBM-LABEL: bextr32_d1_indexzext:
 ; X86-BMI1NOTBM:       # %bb.0:
-; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %al
 ; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
-; X86-BMI1NOTBM-NEXT:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, %edx, %eax
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    xorl %ecx, %ecx
+; X86-BMI1NOTBM-NEXT:    subb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bextr32_d1_indexzext:
@@ -4231,10 +4281,13 @@ define i32 @bextr32_d1_indexzext(i32 %va
 ; X64-BMI1NOTBM-LABEL: bextr32_d1_indexzext:
 ; X64-BMI1NOTBM:       # %bb.0:
 ; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movl %edi, %eax
 ; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
-; X64-BMI1NOTBM-NEXT:    shrl %cl, %edi
-; X64-BMI1NOTBM-NEXT:    shll $8, %edx
-; X64-BMI1NOTBM-NEXT:    bextrl %edx, %edi, %eax
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X64-BMI1NOTBM-NEXT:    negb %dl
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr32_d1_indexzext:
@@ -4267,13 +4320,15 @@ define i32 @bextr32_d2_load(i32* %w, i32
 ;
 ; X86-BMI1NOTBM-LABEL: bextr32_d2_load:
 ; X86-BMI1NOTBM:       # %bb.0:
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-BMI1NOTBM-NEXT:    movl (%edx), %edx
-; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
-; X86-BMI1NOTBM-NEXT:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, %edx, %eax
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    movl (%eax), %eax
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    xorl %ecx, %ecx
+; X86-BMI1NOTBM-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bextr32_d2_load:
@@ -4303,8 +4358,10 @@ define i32 @bextr32_d2_load(i32* %w, i32
 ; X64-BMI1NOTBM-NEXT:    movl (%rdi), %eax
 ; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
 ; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
-; X64-BMI1NOTBM-NEXT:    shll $8, %edx
-; X64-BMI1NOTBM-NEXT:    bextrl %edx, %eax, %eax
+; X64-BMI1NOTBM-NEXT:    negl %edx
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr32_d2_load:
@@ -4336,13 +4393,15 @@ define i32 @bextr32_d3_load_indexzext(i3
 ;
 ; X86-BMI1NOTBM-LABEL: bextr32_d3_load_indexzext:
 ; X86-BMI1NOTBM:       # %bb.0:
-; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %al
 ; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-BMI1NOTBM-NEXT:    movl (%edx), %edx
-; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
-; X86-BMI1NOTBM-NEXT:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, %edx, %eax
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    movl (%eax), %eax
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
+; X86-BMI1NOTBM-NEXT:    xorl %ecx, %ecx
+; X86-BMI1NOTBM-NEXT:    subb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bextr32_d3_load_indexzext:
@@ -4372,8 +4431,10 @@ define i32 @bextr32_d3_load_indexzext(i3
 ; X64-BMI1NOTBM-NEXT:    movl (%rdi), %eax
 ; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
 ; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
-; X64-BMI1NOTBM-NEXT:    shll $8, %edx
-; X64-BMI1NOTBM-NEXT:    bextrl %edx, %eax, %eax
+; X64-BMI1NOTBM-NEXT:    negb %dl
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr32_d3_load_indexzext:
@@ -4416,13 +4477,16 @@ define i32 @bextr32_d5_skipextrauses(i32
 ; X86-BMI1NOTBM:       # %bb.0:
 ; X86-BMI1NOTBM-NEXT:    pushl %esi
 ; X86-BMI1NOTBM-NEXT:    subl $8, %esp
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %esi
 ; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %edx
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-BMI1NOTBM-NEXT:    shrl %cl, %edx
-; X86-BMI1NOTBM-NEXT:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, %edx, %esi
-; X86-BMI1NOTBM-NEXT:    movl %ecx, (%esp)
+; X86-BMI1NOTBM-NEXT:    movl %eax, %ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %esi
+; X86-BMI1NOTBM-NEXT:    xorl %ecx, %ecx
+; X86-BMI1NOTBM-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %esi
+; X86-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %esi
+; X86-BMI1NOTBM-NEXT:    movl %eax, (%esp)
 ; X86-BMI1NOTBM-NEXT:    calll use32
 ; X86-BMI1NOTBM-NEXT:    movl %esi, %eax
 ; X86-BMI1NOTBM-NEXT:    addl $8, %esp
@@ -4463,10 +4527,13 @@ define i32 @bextr32_d5_skipextrauses(i32
 ; X64-BMI1NOTBM-LABEL: bextr32_d5_skipextrauses:
 ; X64-BMI1NOTBM:       # %bb.0:
 ; X64-BMI1NOTBM-NEXT:    pushq %rbx
+; X64-BMI1NOTBM-NEXT:    movl %edi, %ebx
 ; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
-; X64-BMI1NOTBM-NEXT:    shrl %cl, %edi
-; X64-BMI1NOTBM-NEXT:    shll $8, %edx
-; X64-BMI1NOTBM-NEXT:    bextrl %edx, %edi, %ebx
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %ebx
+; X64-BMI1NOTBM-NEXT:    negl %edx
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %ebx
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %ebx
 ; X64-BMI1NOTBM-NEXT:    movl %esi, %edi
 ; X64-BMI1NOTBM-NEXT:    callq use32
 ; X64-BMI1NOTBM-NEXT:    movl %ebx, %eax
@@ -4646,10 +4713,13 @@ define i64 @bextr64_d0(i64 %val, i64 %nu
 ; X64-BMI1NOTBM-LABEL: bextr64_d0:
 ; X64-BMI1NOTBM:       # %bb.0:
 ; X64-BMI1NOTBM-NEXT:    movq %rsi, %rcx
+; X64-BMI1NOTBM-NEXT:    movq %rdi, %rax
 ; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $rcx
-; X64-BMI1NOTBM-NEXT:    shrq %cl, %rdi
-; X64-BMI1NOTBM-NEXT:    shlq $8, %rdx
-; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rdi, %rax
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    negl %edx
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr64_d0:
@@ -4816,12 +4886,14 @@ define i64 @bextr64_d1_indexzext(i64 %va
 ;
 ; X64-BMI1NOTBM-LABEL: bextr64_d1_indexzext:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    # kill: def $edx killed $edx def $rdx
 ; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movq %rdi, %rax
 ; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
-; X64-BMI1NOTBM-NEXT:    shrq %cl, %rdi
-; X64-BMI1NOTBM-NEXT:    shlq $8, %rdx
-; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rdi, %rax
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    negb %dl
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr64_d1_indexzext:
@@ -4999,8 +5071,10 @@ define i64 @bextr64_d2_load(i64* %w, i64
 ; X64-BMI1NOTBM-NEXT:    movq (%rdi), %rax
 ; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $rcx
 ; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
-; X64-BMI1NOTBM-NEXT:    shlq $8, %rdx
-; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rax, %rax
+; X64-BMI1NOTBM-NEXT:    negl %edx
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr64_d2_load:
@@ -5171,13 +5245,14 @@ define i64 @bextr64_d3_load_indexzext(i6
 ;
 ; X64-BMI1NOTBM-LABEL: bextr64_d3_load_indexzext:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    # kill: def $edx killed $edx def $rdx
 ; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
 ; X64-BMI1NOTBM-NEXT:    movq (%rdi), %rax
 ; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
 ; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
-; X64-BMI1NOTBM-NEXT:    shlq $8, %rdx
-; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rax, %rax
+; X64-BMI1NOTBM-NEXT:    negb %dl
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bextr64_d3_load_indexzext:
@@ -5391,10 +5466,13 @@ define i64 @bextr64_d5_skipextrauses(i64
 ; X64-BMI1NOTBM-LABEL: bextr64_d5_skipextrauses:
 ; X64-BMI1NOTBM:       # %bb.0:
 ; X64-BMI1NOTBM-NEXT:    pushq %rbx
-; X64-BMI1NOTBM-NEXT:    movq %rsi, %rcx
-; X64-BMI1NOTBM-NEXT:    shrq %cl, %rdi
-; X64-BMI1NOTBM-NEXT:    shlq $8, %rdx
-; X64-BMI1NOTBM-NEXT:    bextrq %rdx, %rdi, %rbx
+; X64-BMI1NOTBM-NEXT:    movq %rdi, %rbx
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rbx
+; X64-BMI1NOTBM-NEXT:    negl %edx
+; X64-BMI1NOTBM-NEXT:    movl %edx, %ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rbx
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rbx
 ; X64-BMI1NOTBM-NEXT:    movq %rsi, %rdi
 ; X64-BMI1NOTBM-NEXT:    callq use64
 ; X64-BMI1NOTBM-NEXT:    movq %rbx, %rax

Modified: llvm/trunk/test/CodeGen/X86/extract-lowbits.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/extract-lowbits.ll?rev=345017&r1=345016&r2=345017&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/X86/extract-lowbits.ll (original)
+++ llvm/trunk/test/CodeGen/X86/extract-lowbits.ll Tue Oct 23 03:34:57 2018
@@ -1442,8 +1442,11 @@ define i32 @bzhi32_c0(i32 %val, i32 %num
 ; X86-BMI1NOTBM-LABEL: bzhi32_c0:
 ; X86-BMI1NOTBM:       # %bb.0:
 ; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-BMI1NOTBM-NEXT:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    xorl %ecx, %ecx
+; X86-BMI1NOTBM-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bzhi32_c0:
@@ -1464,8 +1467,12 @@ define i32 @bzhi32_c0(i32 %val, i32 %num
 ;
 ; X64-BMI1NOTBM-LABEL: bzhi32_c0:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    shll $8, %esi
-; X64-BMI1NOTBM-NEXT:    bextrl %esi, %edi, %eax
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movl %edi, %eax
+; X64-BMI1NOTBM-NEXT:    negl %ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bzhi32_c0:
@@ -1491,9 +1498,12 @@ define i32 @bzhi32_c1_indexzext(i32 %val
 ;
 ; X86-BMI1NOTBM-LABEL: bzhi32_c1_indexzext:
 ; X86-BMI1NOTBM:       # %bb.0:
-; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %al
-; X86-BMI1NOTBM-NEXT:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    xorl %ecx, %ecx
+; X86-BMI1NOTBM-NEXT:    subb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bzhi32_c1_indexzext:
@@ -1514,8 +1524,12 @@ define i32 @bzhi32_c1_indexzext(i32 %val
 ;
 ; X64-BMI1NOTBM-LABEL: bzhi32_c1_indexzext:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    shll $8, %esi
-; X64-BMI1NOTBM-NEXT:    bextrl %esi, %edi, %eax
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movl %edi, %eax
+; X64-BMI1NOTBM-NEXT:    negb %cl
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bzhi32_c1_indexzext:
@@ -1544,16 +1558,19 @@ define i32 @bzhi32_c2_load(i32* %w, i32
 ; X86-BMI1NOTBM-LABEL: bzhi32_c2_load:
 ; X86-BMI1NOTBM:       # %bb.0:
 ; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-BMI1NOTBM-NEXT:    shll $8, %ecx
-; X86-BMI1NOTBM-NEXT:    bextrl %ecx, (%eax), %eax
+; X86-BMI1NOTBM-NEXT:    movl (%eax), %eax
+; X86-BMI1NOTBM-NEXT:    xorl %ecx, %ecx
+; X86-BMI1NOTBM-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bzhi32_c2_load:
 ; X86-BMI1BMI2:       # %bb.0:
 ; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-BMI1BMI2-NEXT:    bzhil %ecx, (%eax), %eax
+; X86-BMI1BMI2-NEXT:    bzhil %eax, (%ecx), %eax
 ; X86-BMI1BMI2-NEXT:    retl
 ;
 ; X64-NOBMI-LABEL: bzhi32_c2_load:
@@ -1568,8 +1585,12 @@ define i32 @bzhi32_c2_load(i32* %w, i32
 ;
 ; X64-BMI1NOTBM-LABEL: bzhi32_c2_load:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    shll $8, %esi
-; X64-BMI1NOTBM-NEXT:    bextrl %esi, (%rdi), %eax
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movl (%rdi), %eax
+; X64-BMI1NOTBM-NEXT:    negl %ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bzhi32_c2_load:
@@ -1598,9 +1619,12 @@ define i32 @bzhi32_c3_load_indexzext(i32
 ; X86-BMI1NOTBM-LABEL: bzhi32_c3_load_indexzext:
 ; X86-BMI1NOTBM:       # %bb.0:
 ; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
-; X86-BMI1NOTBM-NEXT:    shll $8, %ecx
-; X86-BMI1NOTBM-NEXT:    bextrl %ecx, (%eax), %eax
+; X86-BMI1NOTBM-NEXT:    movl (%eax), %eax
+; X86-BMI1NOTBM-NEXT:    xorl %ecx, %ecx
+; X86-BMI1NOTBM-NEXT:    subb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bzhi32_c3_load_indexzext:
@@ -1622,8 +1646,12 @@ define i32 @bzhi32_c3_load_indexzext(i32
 ;
 ; X64-BMI1NOTBM-LABEL: bzhi32_c3_load_indexzext:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    shll $8, %esi
-; X64-BMI1NOTBM-NEXT:    bextrl %esi, (%rdi), %eax
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movl (%rdi), %eax
+; X64-BMI1NOTBM-NEXT:    negb %cl
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bzhi32_c3_load_indexzext:
@@ -1652,8 +1680,11 @@ define i32 @bzhi32_c4_commutative(i32 %v
 ; X86-BMI1NOTBM-LABEL: bzhi32_c4_commutative:
 ; X86-BMI1NOTBM:       # %bb.0:
 ; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-BMI1NOTBM-NEXT:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    xorl %ecx, %ecx
+; X86-BMI1NOTBM-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bzhi32_c4_commutative:
@@ -1674,8 +1705,12 @@ define i32 @bzhi32_c4_commutative(i32 %v
 ;
 ; X64-BMI1NOTBM-LABEL: bzhi32_c4_commutative:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    shll $8, %esi
-; X64-BMI1NOTBM-NEXT:    bextrl %esi, %edi, %eax
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movl %edi, %eax
+; X64-BMI1NOTBM-NEXT:    negl %ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bzhi32_c4_commutative:
@@ -1756,8 +1791,12 @@ define i64 @bzhi64_c0(i64 %val, i64 %num
 ;
 ; X64-BMI1NOTBM-LABEL: bzhi64_c0:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    shlq $8, %rsi
-; X64-BMI1NOTBM-NEXT:    bextrq %rsi, %rdi, %rax
+; X64-BMI1NOTBM-NEXT:    movq %rsi, %rcx
+; X64-BMI1NOTBM-NEXT:    movq %rdi, %rax
+; X64-BMI1NOTBM-NEXT:    negl %ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $rcx
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bzhi64_c0:
@@ -1836,9 +1875,12 @@ define i64 @bzhi64_c1_indexzext(i64 %val
 ;
 ; X64-BMI1NOTBM-LABEL: bzhi64_c1_indexzext:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    # kill: def $esi killed $esi def $rsi
-; X64-BMI1NOTBM-NEXT:    shlq $8, %rsi
-; X64-BMI1NOTBM-NEXT:    bextrq %rsi, %rdi, %rax
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movq %rdi, %rax
+; X64-BMI1NOTBM-NEXT:    negb %cl
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bzhi64_c1_indexzext:
@@ -1928,8 +1970,12 @@ define i64 @bzhi64_c2_load(i64* %w, i64
 ;
 ; X64-BMI1NOTBM-LABEL: bzhi64_c2_load:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    shlq $8, %rsi
-; X64-BMI1NOTBM-NEXT:    bextrq %rsi, (%rdi), %rax
+; X64-BMI1NOTBM-NEXT:    movq %rsi, %rcx
+; X64-BMI1NOTBM-NEXT:    movq (%rdi), %rax
+; X64-BMI1NOTBM-NEXT:    negl %ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $rcx
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bzhi64_c2_load:
@@ -2018,9 +2064,12 @@ define i64 @bzhi64_c3_load_indexzext(i64
 ;
 ; X64-BMI1NOTBM-LABEL: bzhi64_c3_load_indexzext:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    # kill: def $esi killed $esi def $rsi
-; X64-BMI1NOTBM-NEXT:    shlq $8, %rsi
-; X64-BMI1NOTBM-NEXT:    bextrq %rsi, (%rdi), %rax
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movq (%rdi), %rax
+; X64-BMI1NOTBM-NEXT:    negb %cl
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bzhi64_c3_load_indexzext:
@@ -2102,8 +2151,12 @@ define i64 @bzhi64_c4_commutative(i64 %v
 ;
 ; X64-BMI1NOTBM-LABEL: bzhi64_c4_commutative:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    shlq $8, %rsi
-; X64-BMI1NOTBM-NEXT:    bextrq %rsi, %rdi, %rax
+; X64-BMI1NOTBM-NEXT:    movq %rsi, %rcx
+; X64-BMI1NOTBM-NEXT:    movq %rdi, %rax
+; X64-BMI1NOTBM-NEXT:    negl %ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $rcx
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bzhi64_c4_commutative:
@@ -2134,8 +2187,11 @@ define i32 @bzhi32_d0(i32 %val, i32 %num
 ; X86-BMI1NOTBM-LABEL: bzhi32_d0:
 ; X86-BMI1NOTBM:       # %bb.0:
 ; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-BMI1NOTBM-NEXT:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    xorl %ecx, %ecx
+; X86-BMI1NOTBM-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bzhi32_d0:
@@ -2156,8 +2212,12 @@ define i32 @bzhi32_d0(i32 %val, i32 %num
 ;
 ; X64-BMI1NOTBM-LABEL: bzhi32_d0:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    shll $8, %esi
-; X64-BMI1NOTBM-NEXT:    bextrl %esi, %edi, %eax
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movl %edi, %eax
+; X64-BMI1NOTBM-NEXT:    negl %ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bzhi32_d0:
@@ -2183,9 +2243,12 @@ define i32 @bzhi32_d1_indexzext(i32 %val
 ;
 ; X86-BMI1NOTBM-LABEL: bzhi32_d1_indexzext:
 ; X86-BMI1NOTBM:       # %bb.0:
-; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %al
-; X86-BMI1NOTBM-NEXT:    shll $8, %eax
-; X86-BMI1NOTBM-NEXT:    bextrl %eax, {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-BMI1NOTBM-NEXT:    xorl %ecx, %ecx
+; X86-BMI1NOTBM-NEXT:    subb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bzhi32_d1_indexzext:
@@ -2206,8 +2269,12 @@ define i32 @bzhi32_d1_indexzext(i32 %val
 ;
 ; X64-BMI1NOTBM-LABEL: bzhi32_d1_indexzext:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    shll $8, %esi
-; X64-BMI1NOTBM-NEXT:    bextrl %esi, %edi, %eax
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movl %edi, %eax
+; X64-BMI1NOTBM-NEXT:    negb %cl
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bzhi32_d1_indexzext:
@@ -2236,16 +2303,19 @@ define i32 @bzhi32_d2_load(i32* %w, i32
 ; X86-BMI1NOTBM-LABEL: bzhi32_d2_load:
 ; X86-BMI1NOTBM:       # %bb.0:
 ; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-BMI1NOTBM-NEXT:    shll $8, %ecx
-; X86-BMI1NOTBM-NEXT:    bextrl %ecx, (%eax), %eax
+; X86-BMI1NOTBM-NEXT:    movl (%eax), %eax
+; X86-BMI1NOTBM-NEXT:    xorl %ecx, %ecx
+; X86-BMI1NOTBM-NEXT:    subl {{[0-9]+}}(%esp), %ecx
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bzhi32_d2_load:
 ; X86-BMI1BMI2:       # %bb.0:
 ; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-BMI1BMI2-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-BMI1BMI2-NEXT:    bzhil %ecx, (%eax), %eax
+; X86-BMI1BMI2-NEXT:    bzhil %eax, (%ecx), %eax
 ; X86-BMI1BMI2-NEXT:    retl
 ;
 ; X64-NOBMI-LABEL: bzhi32_d2_load:
@@ -2260,8 +2330,12 @@ define i32 @bzhi32_d2_load(i32* %w, i32
 ;
 ; X64-BMI1NOTBM-LABEL: bzhi32_d2_load:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    shll $8, %esi
-; X64-BMI1NOTBM-NEXT:    bextrl %esi, (%rdi), %eax
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movl (%rdi), %eax
+; X64-BMI1NOTBM-NEXT:    negl %ecx
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bzhi32_d2_load:
@@ -2290,9 +2364,12 @@ define i32 @bzhi32_d3_load_indexzext(i32
 ; X86-BMI1NOTBM-LABEL: bzhi32_d3_load_indexzext:
 ; X86-BMI1NOTBM:       # %bb.0:
 ; X86-BMI1NOTBM-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-BMI1NOTBM-NEXT:    movb {{[0-9]+}}(%esp), %cl
-; X86-BMI1NOTBM-NEXT:    shll $8, %ecx
-; X86-BMI1NOTBM-NEXT:    bextrl %ecx, (%eax), %eax
+; X86-BMI1NOTBM-NEXT:    movl (%eax), %eax
+; X86-BMI1NOTBM-NEXT:    xorl %ecx, %ecx
+; X86-BMI1NOTBM-NEXT:    subb {{[0-9]+}}(%esp), %cl
+; X86-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X86-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X86-BMI1NOTBM-NEXT:    shrl %cl, %eax
 ; X86-BMI1NOTBM-NEXT:    retl
 ;
 ; X86-BMI1BMI2-LABEL: bzhi32_d3_load_indexzext:
@@ -2314,8 +2391,12 @@ define i32 @bzhi32_d3_load_indexzext(i32
 ;
 ; X64-BMI1NOTBM-LABEL: bzhi32_d3_load_indexzext:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    shll $8, %esi
-; X64-BMI1NOTBM-NEXT:    bextrl %esi, (%rdi), %eax
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movl (%rdi), %eax
+; X64-BMI1NOTBM-NEXT:    negb %cl
+; X64-BMI1NOTBM-NEXT:    shll %cl, %eax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shrl %cl, %eax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bzhi32_d3_load_indexzext:
@@ -2455,8 +2536,12 @@ define i64 @bzhi64_d0(i64 %val, i64 %num
 ;
 ; X64-BMI1NOTBM-LABEL: bzhi64_d0:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    shlq $8, %rsi
-; X64-BMI1NOTBM-NEXT:    bextrq %rsi, %rdi, %rax
+; X64-BMI1NOTBM-NEXT:    movq %rsi, %rcx
+; X64-BMI1NOTBM-NEXT:    movq %rdi, %rax
+; X64-BMI1NOTBM-NEXT:    negl %ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $rcx
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bzhi64_d0:
@@ -2592,9 +2677,12 @@ define i64 @bzhi64_d1_indexzext(i64 %val
 ;
 ; X64-BMI1NOTBM-LABEL: bzhi64_d1_indexzext:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    # kill: def $esi killed $esi def $rsi
-; X64-BMI1NOTBM-NEXT:    shlq $8, %rsi
-; X64-BMI1NOTBM-NEXT:    bextrq %rsi, %rdi, %rax
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movq %rdi, %rax
+; X64-BMI1NOTBM-NEXT:    negb %cl
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bzhi64_d1_indexzext:
@@ -2735,8 +2823,12 @@ define i64 @bzhi64_d2_load(i64* %w, i64
 ;
 ; X64-BMI1NOTBM-LABEL: bzhi64_d2_load:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    shlq $8, %rsi
-; X64-BMI1NOTBM-NEXT:    bextrq %rsi, (%rdi), %rax
+; X64-BMI1NOTBM-NEXT:    movq %rsi, %rcx
+; X64-BMI1NOTBM-NEXT:    movq (%rdi), %rax
+; X64-BMI1NOTBM-NEXT:    negl %ecx
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $rcx
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bzhi64_d2_load:
@@ -2876,9 +2968,12 @@ define i64 @bzhi64_d3_load_indexzext(i64
 ;
 ; X64-BMI1NOTBM-LABEL: bzhi64_d3_load_indexzext:
 ; X64-BMI1NOTBM:       # %bb.0:
-; X64-BMI1NOTBM-NEXT:    # kill: def $esi killed $esi def $rsi
-; X64-BMI1NOTBM-NEXT:    shlq $8, %rsi
-; X64-BMI1NOTBM-NEXT:    bextrq %rsi, (%rdi), %rax
+; X64-BMI1NOTBM-NEXT:    movl %esi, %ecx
+; X64-BMI1NOTBM-NEXT:    movq (%rdi), %rax
+; X64-BMI1NOTBM-NEXT:    negb %cl
+; X64-BMI1NOTBM-NEXT:    shlq %cl, %rax
+; X64-BMI1NOTBM-NEXT:    # kill: def $cl killed $cl killed $ecx
+; X64-BMI1NOTBM-NEXT:    shrq %cl, %rax
 ; X64-BMI1NOTBM-NEXT:    retq
 ;
 ; X64-BMI1BMI2-LABEL: bzhi64_d3_load_indexzext:




More information about the llvm-commits mailing list