[X86] Remove use of MVT class from the ShuffleDecode library.

[craig.topper]

Summary

MVT belongs to the CodeGen layer, but ShuffleDecode is used by the X86 InstPrinter which is part of the MC layer. This only works because MVT is completely implemented in a header file with no other library dependencies.

[craig.topper]

Fix EXTRQI/INSERTQI

[RKSimon]

LGTM - thanks

llvm/trunk/lib/Target/X86/InstPrinter/X86InstComments.cpp

 // an output stream for -fverbose-asm.
 //
 //===----------------------------------------------------------------------===//
 
 #include "X86InstComments.h"
 #include "MCTargetDesc/X86BaseInfo.h"
 #include "MCTargetDesc/X86MCTargetDesc.h"
 #include "Utils/X86ShuffleDecode.h"
-#include "llvm/CodeGen/MachineValueType.h"
 #include "llvm/MC/MCInst.h"
 #include "llvm/MC/MCInstrInfo.h"
 #include "llvm/Support/raw_ostream.h"
 
 using namespace llvm;
 
 #define CASE_SSE_INS_COMMON(Inst, src)            \
   case X86::Inst##src:
static unsigned getVectorRegSize(unsigned RegNo) {
   if (X86::XMM0 <= RegNo && RegNo <= X86::XMM31)
     return 128;
   if (X86::MM0 <= RegNo && RegNo <= X86::MM7)
     return 64;
 
   llvm_unreachable("Unknown vector reg!");
 }
 
-static MVT getRegOperandVectorVT(const MCInst *MI, const MVT &ScalarVT,
+static unsigned getRegOperandNumElts(const MCInst *MI, unsigned ScalarSize,
                                      unsigned OperandIndex) {
   unsigned OpReg = MI->getOperand(OperandIndex).getReg();
-  return MVT::getVectorVT(ScalarVT,
-                          getVectorRegSize(OpReg)/ScalarVT.getSizeInBits());
-}
-
-/// \brief Extracts the dst type for a given zero extension instruction.
-static MVT getZeroExtensionResultType(const MCInst *MI) {
-  switch (MI->getOpcode()) {
-  default:
-    llvm_unreachable("Unknown zero extension instruction");
-  // zero extension to i16
-  CASE_PMOVZX(PMOVZXBW, m)
-  CASE_PMOVZX(PMOVZXBW, r)
-    return getRegOperandVectorVT(MI, MVT::i16, 0);
-  // zero extension to i32
-  CASE_PMOVZX(PMOVZXBD, m)
-  CASE_PMOVZX(PMOVZXBD, r)
-  CASE_PMOVZX(PMOVZXWD, m)
-  CASE_PMOVZX(PMOVZXWD, r)
-    return getRegOperandVectorVT(MI, MVT::i32, 0);
-  // zero extension to i64
-  CASE_PMOVZX(PMOVZXBQ, m)
-  CASE_PMOVZX(PMOVZXBQ, r)
-  CASE_PMOVZX(PMOVZXWQ, m)
-  CASE_PMOVZX(PMOVZXWQ, r)
-  CASE_PMOVZX(PMOVZXDQ, m)
-  CASE_PMOVZX(PMOVZXDQ, r)
-    return getRegOperandVectorVT(MI, MVT::i64, 0);
-  }
+  return getVectorRegSize(OpReg) / ScalarSize;
 }
 
 /// Wraps the destination register name with AVX512 mask/maskz filtering.
 static void printMasking(raw_ostream &OS, const MCInst *MI,
                          const MCInstrInfo &MCII,
                          const char *(*getRegName)(unsigned)) {
   const MCInstrDesc &Desc = MCII.get(MI->getOpcode());
   uint64_t TSFlags = Desc.TSFlags;
bool llvm::EmitAnyX86InstComments(const MCInst *MI, raw_ostream &OS,
   case X86::VBLENDPDrri:
   case X86::VBLENDPDYrri:
     Src2Name = getRegName(MI->getOperand(2).getReg());
     LLVM_FALLTHROUGH;
   case X86::BLENDPDrmi:
   case X86::VBLENDPDrmi:
   case X86::VBLENDPDYrmi:
     if (MI->getOperand(NumOperands - 1).isImm())
-      DecodeBLENDMask(getRegOperandVectorVT(MI, MVT::f64, 0),
+      DecodeBLENDMask(getRegOperandNumElts(MI, 64, 0),
                       MI->getOperand(NumOperands - 1).getImm(),
                       ShuffleMask);
     Src1Name = getRegName(MI->getOperand(1).getReg());
     DestName = getRegName(MI->getOperand(0).getReg());
     break;
 
   case X86::BLENDPSrri:
   case X86::VBLENDPSrri:
   case X86::VBLENDPSYrri:
     Src2Name = getRegName(MI->getOperand(2).getReg());
     LLVM_FALLTHROUGH;
   case X86::BLENDPSrmi:
   case X86::VBLENDPSrmi:
   case X86::VBLENDPSYrmi:
     if (MI->getOperand(NumOperands - 1).isImm())
-      DecodeBLENDMask(getRegOperandVectorVT(MI, MVT::f32, 0),
+      DecodeBLENDMask(getRegOperandNumElts(MI, 32, 0),
                       MI->getOperand(NumOperands - 1).getImm(),
                       ShuffleMask);
     Src1Name = getRegName(MI->getOperand(1).getReg());
     DestName = getRegName(MI->getOperand(0).getReg());
     break;
 
   case X86::PBLENDWrri:
   case X86::VPBLENDWrri:
   case X86::VPBLENDWYrri:
     Src2Name = getRegName(MI->getOperand(2).getReg());
     LLVM_FALLTHROUGH;
   case X86::PBLENDWrmi:
   case X86::VPBLENDWrmi:
   case X86::VPBLENDWYrmi:
     if (MI->getOperand(NumOperands - 1).isImm())
-      DecodeBLENDMask(getRegOperandVectorVT(MI, MVT::i16, 0),
+      DecodeBLENDMask(getRegOperandNumElts(MI, 16, 0),
                       MI->getOperand(NumOperands - 1).getImm(),
                       ShuffleMask);
     Src1Name = getRegName(MI->getOperand(1).getReg());
     DestName = getRegName(MI->getOperand(0).getReg());
     break;
 
   case X86::VPBLENDDrri:
   case X86::VPBLENDDYrri:
     Src2Name = getRegName(MI->getOperand(2).getReg());
     LLVM_FALLTHROUGH;
   case X86::VPBLENDDrmi:
   case X86::VPBLENDDYrmi:
     if (MI->getOperand(NumOperands - 1).isImm())
-      DecodeBLENDMask(getRegOperandVectorVT(MI, MVT::i32, 0),
+      DecodeBLENDMask(getRegOperandNumElts(MI, 32, 0),
                       MI->getOperand(NumOperands - 1).getImm(),
                       ShuffleMask);
     Src1Name = getRegName(MI->getOperand(1).getReg());
     DestName = getRegName(MI->getOperand(0).getReg());
     break;
 
   case X86::INSERTPSrr:
   case X86::VINSERTPSrr:
case X86::VMOVHLPSZrr:
     DecodeMOVHLPSMask(2, ShuffleMask);
     break;
 
   case X86::MOVHPDrm:
   case X86::VMOVHPDrm:
   case X86::VMOVHPDZ128rm:
     Src1Name = getRegName(MI->getOperand(1).getReg());
     DestName = getRegName(MI->getOperand(0).getReg());
-    DecodeInsertElementMask(MVT::v2f64, 1, 1, ShuffleMask);
+    DecodeInsertElementMask(2, 1, 1, ShuffleMask);
     break;
 
   case X86::MOVHPSrm:
   case X86::VMOVHPSrm:
   case X86::VMOVHPSZ128rm:
     Src1Name = getRegName(MI->getOperand(1).getReg());
     DestName = getRegName(MI->getOperand(0).getReg());
-    DecodeInsertElementMask(MVT::v4f32, 2, 2, ShuffleMask);
+    DecodeInsertElementMask(4, 2, 2, ShuffleMask);
     break;
 
   case X86::MOVLPDrm:
   case X86::VMOVLPDrm:
   case X86::VMOVLPDZ128rm:
     Src1Name = getRegName(MI->getOperand(1).getReg());
     DestName = getRegName(MI->getOperand(0).getReg());
-    DecodeInsertElementMask(MVT::v2f64, 0, 1, ShuffleMask);
+    DecodeInsertElementMask(2, 0, 1, ShuffleMask);
     break;
 
   case X86::MOVLPSrm:
   case X86::VMOVLPSrm:
   case X86::VMOVLPSZ128rm:
     Src1Name = getRegName(MI->getOperand(1).getReg());
     DestName = getRegName(MI->getOperand(0).getReg());
-    DecodeInsertElementMask(MVT::v4f32, 0, 2, ShuffleMask);
+    DecodeInsertElementMask(4, 0, 2, ShuffleMask);
     break;
 
   CASE_MOVDUP(MOVSLDUP, r)
     Src1Name = getRegName(MI->getOperand(NumOperands - 1).getReg());
     LLVM_FALLTHROUGH;
 
   CASE_MOVDUP(MOVSLDUP, m)
     DestName = getRegName(MI->getOperand(0).getReg());
-    DecodeMOVSLDUPMask(getRegOperandVectorVT(MI, MVT::f32, 0), ShuffleMask);
+    DecodeMOVSLDUPMask(getRegOperandNumElts(MI, 32, 0), ShuffleMask);
     break;
 
   CASE_MOVDUP(MOVSHDUP, r)
     Src1Name = getRegName(MI->getOperand(NumOperands - 1).getReg());
     LLVM_FALLTHROUGH;
 
   CASE_MOVDUP(MOVSHDUP, m)
     DestName = getRegName(MI->getOperand(0).getReg());
-    DecodeMOVSHDUPMask(getRegOperandVectorVT(MI, MVT::f32, 0), ShuffleMask);
+    DecodeMOVSHDUPMask(getRegOperandNumElts(MI, 32, 0), ShuffleMask);
     break;
 
   CASE_MOVDUP(MOVDDUP, r)
     Src1Name = getRegName(MI->getOperand(NumOperands - 1).getReg());
     LLVM_FALLTHROUGH;
 
   CASE_MOVDUP(MOVDDUP, m)
     DestName = getRegName(MI->getOperand(0).getReg());
-    DecodeMOVDDUPMask(getRegOperandVectorVT(MI, MVT::f64, 0), ShuffleMask);
+    DecodeMOVDDUPMask(getRegOperandNumElts(MI, 64, 0), ShuffleMask);
     break;
 
   case X86::PSLLDQri:
   case X86::VPSLLDQri:
   case X86::VPSLLDQYri:
   case X86::VPSLLDQZ128rr:
   case X86::VPSLLDQZ256rr:
   case X86::VPSLLDQZrr:
     Src1Name = getRegName(MI->getOperand(1).getReg());
     LLVM_FALLTHROUGH;
   case X86::VPSLLDQZ128rm:
   case X86::VPSLLDQZ256rm:
   case X86::VPSLLDQZrm:
     DestName = getRegName(MI->getOperand(0).getReg());
     if (MI->getOperand(NumOperands - 1).isImm())
-      DecodePSLLDQMask(getRegOperandVectorVT(MI, MVT::i8, 0),
+      DecodePSLLDQMask(getRegOperandNumElts(MI, 8, 0),
                        MI->getOperand(NumOperands - 1).getImm(),
                        ShuffleMask);
     break;
 
   case X86::PSRLDQri:
   case X86::VPSRLDQri:
   case X86::VPSRLDQYri:
   case X86::VPSRLDQZ128rr:
   case X86::VPSRLDQZ256rr:
   case X86::VPSRLDQZrr:
     Src1Name = getRegName(MI->getOperand(1).getReg());
     LLVM_FALLTHROUGH;
   case X86::VPSRLDQZ128rm:
   case X86::VPSRLDQZ256rm:
   case X86::VPSRLDQZrm:
     DestName = getRegName(MI->getOperand(0).getReg());
     if (MI->getOperand(NumOperands - 1).isImm())
-      DecodePSRLDQMask(getRegOperandVectorVT(MI, MVT::i8, 0),
+      DecodePSRLDQMask(getRegOperandNumElts(MI, 8, 0),
                        MI->getOperand(NumOperands - 1).getImm(),
                        ShuffleMask);
     break;
 
   CASE_SHUF(PALIGNR, rri)
     Src1Name = getRegName(MI->getOperand(NumOperands - 2).getReg());
     RegForm = true;
     LLVM_FALLTHROUGH;
 
   CASE_SHUF(PALIGNR, rmi)
     Src2Name = getRegName(MI->getOperand(NumOperands-(RegForm?3:7)).getReg());
     DestName = getRegName(MI->getOperand(0).getReg());
     if (MI->getOperand(NumOperands - 1).isImm())
-      DecodePALIGNRMask(getRegOperandVectorVT(MI, MVT::i8, 0),
+      DecodePALIGNRMask(getRegOperandNumElts(MI, 8, 0),
                         MI->getOperand(NumOperands - 1).getImm(),
                         ShuffleMask);
     break;
 
   CASE_AVX512_INS_COMMON(ALIGNQ, Z, rri)
   CASE_AVX512_INS_COMMON(ALIGNQ, Z256, rri)
   CASE_AVX512_INS_COMMON(ALIGNQ, Z128, rri)
     Src1Name = getRegName(MI->getOperand(NumOperands - 2).getReg());
     RegForm = true;
     LLVM_FALLTHROUGH;
 
   CASE_AVX512_INS_COMMON(ALIGNQ, Z, rmi)
   CASE_AVX512_INS_COMMON(ALIGNQ, Z256, rmi)
   CASE_AVX512_INS_COMMON(ALIGNQ, Z128, rmi)
     Src2Name = getRegName(MI->getOperand(NumOperands-(RegForm?3:7)).getReg());
     DestName = getRegName(MI->getOperand(0).getReg());
     if (MI->getOperand(NumOperands - 1).isImm())
-      DecodeVALIGNMask(getRegOperandVectorVT(MI, MVT::i64, 0),
+      DecodeVALIGNMask(getRegOperandNumElts(MI, 64, 0),
                        MI->getOperand(NumOperands - 1).getImm(),
                        ShuffleMask);
     break;
 
   CASE_AVX512_INS_COMMON(ALIGND, Z, rri)
   CASE_AVX512_INS_COMMON(ALIGND, Z256, rri)
   CASE_AVX512_INS_COMMON(ALIGND, Z128, rri)
     Src1Name = getRegName(MI->getOperand(NumOperands - 2).getReg());
     RegForm = true;
     LLVM_FALLTHROUGH;
 
   CASE_AVX512_INS_COMMON(ALIGND, Z, rmi)
   CASE_AVX512_INS_COMMON(ALIGND, Z256, rmi)
   CASE_AVX512_INS_COMMON(ALIGND, Z128, rmi)
     Src2Name = getRegName(MI->getOperand(NumOperands-(RegForm?3:7)).getReg());
     DestName = getRegName(MI->getOperand(0).getReg());
     if (MI->getOperand(NumOperands - 1).isImm())
-      DecodeVALIGNMask(getRegOperandVectorVT(MI, MVT::i32, 0),
+      DecodeVALIGNMask(getRegOperandNumElts(MI, 32, 0),
                        MI->getOperand(NumOperands - 1).getImm(),
                        ShuffleMask);
     break;
 
   CASE_SHUF(PSHUFD, ri)
     Src1Name = getRegName(MI->getOperand(NumOperands - 2).getReg());
     LLVM_FALLTHROUGH;
 
   CASE_SHUF(PSHUFD, mi)
     DestName = getRegName(MI->getOperand(0).getReg());
     if (MI->getOperand(NumOperands - 1).isImm())
-      DecodePSHUFMask(getRegOperandVectorVT(MI, MVT::i32, 0),
+      DecodePSHUFMask(getRegOperandNumElts(MI, 32, 0), 32,
                       MI->getOperand(NumOperands - 1).getImm(),
                       ShuffleMask);
     break;
 
   CASE_SHUF(PSHUFHW, ri)
     Src1Name = getRegName(MI->getOperand(NumOperands - 2).getReg());
     LLVM_FALLTHROUGH;
 
   CASE_SHUF(PSHUFHW, mi)
     DestName = getRegName(MI->getOperand(0).getReg());
     if (MI->getOperand(NumOperands - 1).isImm())
-      DecodePSHUFHWMask(getRegOperandVectorVT(MI, MVT::i16, 0),
+      DecodePSHUFHWMask(getRegOperandNumElts(MI, 16, 0),
                         MI->getOperand(NumOperands - 1).getImm(),
                         ShuffleMask);
     break;
 
   CASE_SHUF(PSHUFLW, ri)
     Src1Name = getRegName(MI->getOperand(NumOperands - 2).getReg());
     LLVM_FALLTHROUGH;
 
   CASE_SHUF(PSHUFLW, mi)
     DestName = getRegName(MI->getOperand(0).getReg());
     if (MI->getOperand(NumOperands - 1).isImm())
-      DecodePSHUFLWMask(getRegOperandVectorVT(MI, MVT::i16, 0),
+      DecodePSHUFLWMask(getRegOperandNumElts(MI, 16, 0),
                         MI->getOperand(NumOperands - 1).getImm(),
                         ShuffleMask);
     break;
 
   case X86::MMX_PSHUFWri:
     Src1Name = getRegName(MI->getOperand(1).getReg());
     LLVM_FALLTHROUGH;
 
   case X86::MMX_PSHUFWmi:
     DestName = getRegName(MI->getOperand(0).getReg());
     if (MI->getOperand(NumOperands - 1).isImm())
-      DecodePSHUFMask(MVT::v4i16,
-                      MI->getOperand(NumOperands - 1).getImm(),
+      DecodePSHUFMask(4, 16, MI->getOperand(NumOperands - 1).getImm(),
                       ShuffleMask);
     break;
 
   case X86::PSWAPDrr:
     Src1Name = getRegName(MI->getOperand(1).getReg());
     LLVM_FALLTHROUGH;
 
   case X86::PSWAPDrm:
     DestName = getRegName(MI->getOperand(0).getReg());
-    DecodePSWAPMask(MVT::v2i32, ShuffleMask);
+    DecodePSWAPMask(2, ShuffleMask);
     break;
 
   CASE_UNPCK(PUNPCKHBW, r)
   case X86::MMX_PUNPCKHBWirr:
     Src2Name = getRegName(MI->getOperand(NumOperands - 1).getReg());
     RegForm = true;
     LLVM_FALLTHROUGH;
 
   CASE_UNPCK(PUNPCKHBW, m)
   case X86::MMX_PUNPCKHBWirm:
     Src1Name = getRegName(MI->getOperand(NumOperands-(RegForm?2:6)).getReg());
     DestName = getRegName(MI->getOperand(0).getReg());
-    DecodeUNPCKHMask(getRegOperandVectorVT(MI, MVT::i8, 0), ShuffleMask);
+    DecodeUNPCKHMask(getRegOperandNumElts(MI, 8, 0), 8, ShuffleMask);
     break;
 
   CASE_UNPCK(PUNPCKHWD, r)
   case X86::MMX_PUNPCKHWDirr:
     Src2Name = getRegName(MI->getOperand(NumOperands - 1).getReg());
     RegForm = true;
     LLVM_FALLTHROUGH;
 
   CASE_UNPCK(PUNPCKHWD, m)
   case X86::MMX_PUNPCKHWDirm:
     Src1Name = getRegName(MI->getOperand(NumOperands-(RegForm?2:6)).getReg());
     DestName = getRegName(MI->getOperand(0).getReg());
-    DecodeUNPCKHMask(getRegOperandVectorVT(MI, MVT::i16, 0), ShuffleMask);
+    DecodeUNPCKHMask(getRegOperandNumElts(MI, 16, 0), 16, ShuffleMask);
     break;
 
   CASE_UNPCK(PUNPCKHDQ, r)
   case X86::MMX_PUNPCKHDQirr:
     Src2Name = getRegName(MI->getOperand(NumOperands - 1).getReg());
     RegForm = true;
     LLVM_FALLTHROUGH;
 
   CASE_UNPCK(PUNPCKHDQ, m)
   case X86::MMX_PUNPCKHDQirm:
     Src1Name = getRegName(MI->getOperand(NumOperands-(RegForm?2:6)).getReg());
     DestName = getRegName(MI->getOperand(0).getReg());
-    DecodeUNPCKHMask(getRegOperandVectorVT(MI, MVT::i32, 0), ShuffleMask);
+    DecodeUNPCKHMask(getRegOperandNumElts(MI, 32, 0), 32, ShuffleMask);
     break;
 
   CASE_UNPCK(PUNPCKHQDQ, r)
     Src2Name = getRegName(MI->getOperand(NumOperands - 1).getReg());
     RegForm = true;
     LLVM_FALLTHROUGH;
 
   CASE_UNPCK(PUNPCKHQDQ, m)
     Src1Name = getRegName(MI->getOperand(NumOperands-(RegForm?2:6)).getReg());
     DestName = getRegName(MI->getOperand(0).getReg());
-    DecodeUNPCKHMask(getRegOperandVectorVT(MI, MVT::i64, 0), ShuffleMask);
+    DecodeUNPCKHMask(getRegOperandNumElts(MI, 64, 0), 64, ShuffleMask);
     break;
 
   CASE_UNPCK(PUNPCKLBW, r)
   case X86::MMX_PUNPCKLBWirr:
     Src2Name = getRegName(MI->getOperand(NumOperands - 1).getReg());
     RegForm = true;
     LLVM_FALLTHROUGH;
 
   CASE_UNPCK(PUNPCKLBW, m)
   case X86::MMX_PUNPCKLBWirm:
     Src1Name = getRegName(MI->getOperand(NumOperands-(RegForm?2:6)).getReg());
     DestName = getRegName(MI->getOperand(0).getReg());
-    DecodeUNPCKLMask(getRegOperandVectorVT(MI, MVT::i8, 0), ShuffleMask);
+    DecodeUNPCKLMask(getRegOperandNumElts(MI, 8, 0), 8, ShuffleMask);
     break;
 
   CASE_UNPCK(PUNPCKLWD, r)
   case X86::MMX_PUNPCKLWDirr:
     Src2Name = getRegName(MI->getOperand(NumOperands - 1).getReg());
     RegForm = true;
     LLVM_FALLTHROUGH;
 
   CASE_UNPCK(PUNPCKLWD, m)
   case X86::MMX_PUNPCKLWDirm:
     Src1Name = getRegName(MI->getOperand(NumOperands-(RegForm?2:6)).getReg());
     DestName = getRegName(MI->getOperand(0).getReg());
-    DecodeUNPCKLMask(getRegOperandVectorVT(MI, MVT::i16, 0), ShuffleMask);
+    DecodeUNPCKLMask(getRegOperandNumElts(MI, 16, 0), 16, ShuffleMask);
     break;
 
   CASE_UNPCK(PUNPCKLDQ, r)
   case X86::MMX_PUNPCKLDQirr:
     Src2Name = getRegName(MI->getOperand(NumOperands - 1).getReg());
     RegForm = true;
     LLVM_FALLTHROUGH;
 
   CASE_UNPCK(PUNPCKLDQ, m)
   case X86::MMX_PUNPCKLDQirm:
     Src1Name = getRegName(MI->getOperand(NumOperands-(RegForm?2:6)).getReg());
     DestName = getRegName(MI->getOperand(0).getReg());
-    DecodeUNPCKLMask(getRegOperandVectorVT(MI, MVT::i32, 0), ShuffleMask);
+    DecodeUNPCKLMask(getRegOperandNumElts(MI, 32, 0), 32, ShuffleMask);
     break;
 
   CASE_UNPCK(PUNPCKLQDQ, r)
     Src2Name = getRegName(MI->getOperand(NumOperands - 1).getReg());
     RegForm = true;
     LLVM_FALLTHROUGH;
 
   CASE_UNPCK(PUNPCKLQDQ, m)
     Src1Name = getRegName(MI->getOperand(NumOperands-(RegForm?2:6)).getReg());
     DestName = getRegName(MI->getOperand(0).getReg());
-    DecodeUNPCKLMask(getRegOperandVectorVT(MI, MVT::i64, 0), ShuffleMask);
+    DecodeUNPCKLMask(getRegOperandNumElts(MI, 64, 0), 64, ShuffleMask);
     break;
 
   CASE_SHUF(SHUFPD, rri)
     Src2Name = getRegName(MI->getOperand(NumOperands - 2).getReg());
     RegForm = true;
     LLVM_FALLTHROUGH;
 
   CASE_SHUF(SHUFPD, rmi)
     if (MI->getOperand(NumOperands - 1).isImm())
-      DecodeSHUFPMask(getRegOperandVectorVT(MI, MVT::f64, 0),
-                      MI->getOperand(NumOperands - 1).getImm(),
-                      ShuffleMask);
+      DecodeSHUFPMask(getRegOperandNumElts(MI, 64, 0), 64,
+                      MI->getOperand(NumOperands - 1).getImm(), ShuffleMask);
     Src1Name = getRegName(MI->getOperand(NumOperands-(RegForm?3:7)).getReg());
     DestName = getRegName(MI->getOperand(0).getReg());
     break;
 
   CASE_SHUF(SHUFPS, rri)
     Src2Name = getRegName(MI->getOperand(NumOperands - 2).getReg());
     RegForm = true;
     LLVM_FALLTHROUGH;
 
   CASE_SHUF(SHUFPS, rmi)
     if (MI->getOperand(NumOperands - 1).isImm())
-      DecodeSHUFPMask(getRegOperandVectorVT(MI, MVT::f32, 0),
+      DecodeSHUFPMask(getRegOperandNumElts(MI, 32, 0), 32,
                       MI->getOperand(NumOperands - 1).getImm(),
                       ShuffleMask);
     Src1Name = getRegName(MI->getOperand(NumOperands-(RegForm?3:7)).getReg());
     DestName = getRegName(MI->getOperand(0).getReg());
     break;
 
   CASE_VSHUF(64X2, r)
     Src2Name = getRegName(MI->getOperand(NumOperands - 2).getReg());
     RegForm = true;
     LLVM_FALLTHROUGH;
 
   CASE_VSHUF(64X2, m)
-    decodeVSHUF64x2FamilyMask(getRegOperandVectorVT(MI, MVT::i64, 0),
+    decodeVSHUF64x2FamilyMask(getRegOperandNumElts(MI, 64, 0), 64,
                               MI->getOperand(NumOperands - 1).getImm(),
                               ShuffleMask);
     Src1Name = getRegName(MI->getOperand(NumOperands-(RegForm?3:7)).getReg());
     DestName = getRegName(MI->getOperand(0).getReg());
     break;
 
   CASE_VSHUF(32X4, r)
     Src2Name = getRegName(MI->getOperand(NumOperands - 2).getReg());
     RegForm = true;
     LLVM_FALLTHROUGH;
 
   CASE_VSHUF(32X4, m)
-    decodeVSHUF64x2FamilyMask(getRegOperandVectorVT(MI, MVT::i32, 0),
+    decodeVSHUF64x2FamilyMask(getRegOperandNumElts(MI, 32, 0), 32,
                               MI->getOperand(NumOperands - 1).getImm(),
                               ShuffleMask);
     Src1Name = getRegName(MI->getOperand(NumOperands-(RegForm?3:7)).getReg());
     DestName = getRegName(MI->getOperand(0).getReg());
     break;
 
   CASE_UNPCK(UNPCKLPD, r)
     Src2Name = getRegName(MI->getOperand(NumOperands - 1).getReg());
     RegForm = true;
     LLVM_FALLTHROUGH;
 
   CASE_UNPCK(UNPCKLPD, m)
-    DecodeUNPCKLMask(getRegOperandVectorVT(MI, MVT::f64, 0), ShuffleMask);
+    DecodeUNPCKLMask(getRegOperandNumElts(MI, 64, 0), 64, ShuffleMask);
     Src1Name = getRegName(MI->getOperand(NumOperands-(RegForm?2:6)).getReg());
     DestName = getRegName(MI->getOperand(0).getReg());
     break;
 
   CASE_UNPCK(UNPCKLPS, r)
     Src2Name = getRegName(MI->getOperand(NumOperands - 1).getReg());
     RegForm = true;
     LLVM_FALLTHROUGH;
 
   CASE_UNPCK(UNPCKLPS, m)
-    DecodeUNPCKLMask(getRegOperandVectorVT(MI, MVT::f32, 0), ShuffleMask);
+    DecodeUNPCKLMask(getRegOperandNumElts(MI, 32, 0), 32, ShuffleMask);
     Src1Name = getRegName(MI->getOperand(NumOperands-(RegForm?2:6)).getReg());
     DestName = getRegName(MI->getOperand(0).getReg());
     break;
 
   CASE_UNPCK(UNPCKHPD, r)
     Src2Name = getRegName(MI->getOperand(NumOperands - 1).getReg());
     RegForm = true;
     LLVM_FALLTHROUGH;
 
   CASE_UNPCK(UNPCKHPD, m)
-    DecodeUNPCKHMask(getRegOperandVectorVT(MI, MVT::f64, 0), ShuffleMask);
+    DecodeUNPCKHMask(getRegOperandNumElts(MI, 64, 0), 64, ShuffleMask);
     Src1Name = getRegName(MI->getOperand(NumOperands-(RegForm?2:6)).getReg());
     DestName = getRegName(MI->getOperand(0).getReg());
     break;
 
   CASE_UNPCK(UNPCKHPS, r)
     Src2Name = getRegName(MI->getOperand(NumOperands - 1).getReg());
     RegForm = true;
     LLVM_FALLTHROUGH;
 
   CASE_UNPCK(UNPCKHPS, m)
-    DecodeUNPCKHMask(getRegOperandVectorVT(MI, MVT::f32, 0), ShuffleMask);
+    DecodeUNPCKHMask(getRegOperandNumElts(MI, 32, 0), 32, ShuffleMask);
     Src1Name = getRegName(MI->getOperand(NumOperands-(RegForm?2:6)).getReg());
     DestName = getRegName(MI->getOperand(0).getReg());
     break;
 
   CASE_VPERMILPI(PERMILPS, r)
     Src1Name = getRegName(MI->getOperand(NumOperands - 2).getReg());
     LLVM_FALLTHROUGH;
 
   CASE_VPERMILPI(PERMILPS, m)
     if (MI->getOperand(NumOperands - 1).isImm())
-      DecodePSHUFMask(getRegOperandVectorVT(MI, MVT::f32, 0),
+      DecodePSHUFMask(getRegOperandNumElts(MI, 32, 0), 32,
                       MI->getOperand(NumOperands - 1).getImm(),
                       ShuffleMask);
     DestName = getRegName(MI->getOperand(0).getReg());
     break;
 
   CASE_VPERMILPI(PERMILPD, r)
     Src1Name = getRegName(MI->getOperand(NumOperands - 2).getReg());
     LLVM_FALLTHROUGH;
 
   CASE_VPERMILPI(PERMILPD, m)
     if (MI->getOperand(NumOperands - 1).isImm())
-      DecodePSHUFMask(getRegOperandVectorVT(MI, MVT::f64, 0),
+      DecodePSHUFMask(getRegOperandNumElts(MI, 64, 0), 64,
                       MI->getOperand(NumOperands - 1).getImm(),
                       ShuffleMask);
     DestName = getRegName(MI->getOperand(0).getReg());
     break;
 
   case X86::VPERM2F128rr:
   case X86::VPERM2I128rr:
     Src2Name = getRegName(MI->getOperand(2).getReg());
     LLVM_FALLTHROUGH;
 
   case X86::VPERM2F128rm:
   case X86::VPERM2I128rm:
     // For instruction comments purpose, assume the 256-bit vector is v4i64.
     if (MI->getOperand(NumOperands - 1).isImm())
-      DecodeVPERM2X128Mask(MVT::v4i64,
-                           MI->getOperand(NumOperands - 1).getImm(),
+      DecodeVPERM2X128Mask(4, MI->getOperand(NumOperands - 1).getImm(),
                            ShuffleMask);
     Src1Name = getRegName(MI->getOperand(1).getReg());
     DestName = getRegName(MI->getOperand(0).getReg());
     break;
 
   CASE_VPERM(PERMPD, r)
     Src1Name = getRegName(MI->getOperand(NumOperands - 2).getReg());
     LLVM_FALLTHROUGH;
 
   CASE_VPERM(PERMPD, m)
     if (MI->getOperand(NumOperands - 1).isImm())
-      DecodeVPERMMask(getRegOperandVectorVT(MI, MVT::f64, 0),
+      DecodeVPERMMask(getRegOperandNumElts(MI, 64, 0),
                       MI->getOperand(NumOperands - 1).getImm(),
                       ShuffleMask);
     DestName = getRegName(MI->getOperand(0).getReg());
     break;
 
   CASE_VPERM(PERMQ, r)
     Src1Name = getRegName(MI->getOperand(NumOperands - 2).getReg());
     LLVM_FALLTHROUGH;
 
   CASE_VPERM(PERMQ, m)
     if (MI->getOperand(NumOperands - 1).isImm())
-      DecodeVPERMMask(getRegOperandVectorVT(MI, MVT::i64, 0),
+      DecodeVPERMMask(getRegOperandNumElts(MI, 64, 0),
                       MI->getOperand(NumOperands - 1).getImm(),
                       ShuffleMask);
     DestName = getRegName(MI->getOperand(0).getReg());
     break;
 
   case X86::MOVSDrr:
   case X86::VMOVSDrr:
   case X86::VMOVSDZrr:
     Src2Name = getRegName(MI->getOperand(2).getReg());
     Src1Name = getRegName(MI->getOperand(1).getReg());
     LLVM_FALLTHROUGH;
 
   case X86::MOVSDrm:
   case X86::VMOVSDrm:
   case X86::VMOVSDZrm:
-    DecodeScalarMoveMask(MVT::v2f64, nullptr == Src2Name, ShuffleMask);
+    DecodeScalarMoveMask(2, nullptr == Src2Name, ShuffleMask);
     DestName = getRegName(MI->getOperand(0).getReg());
     break;
 
   case X86::MOVSSrr:
   case X86::VMOVSSrr:
   case X86::VMOVSSZrr:
     Src2Name = getRegName(MI->getOperand(2).getReg());
     Src1Name = getRegName(MI->getOperand(1).getReg());
     LLVM_FALLTHROUGH;
 
   case X86::MOVSSrm:
   case X86::VMOVSSrm:
   case X86::VMOVSSZrm:
-    DecodeScalarMoveMask(MVT::v4f32, nullptr == Src2Name, ShuffleMask);
+    DecodeScalarMoveMask(4, nullptr == Src2Name, ShuffleMask);
     DestName = getRegName(MI->getOperand(0).getReg());
     break;
 
   case X86::MOVPQI2QIrr:
   case X86::MOVZPQILo2PQIrr:
   case X86::VMOVPQI2QIrr:
   case X86::VMOVPQI2QIZrr:
   case X86::VMOVZPQILo2PQIrr:
   case X86::VMOVZPQILo2PQIZrr:
     Src1Name = getRegName(MI->getOperand(1).getReg());
     LLVM_FALLTHROUGH;
 
   case X86::MOVQI2PQIrm:
   case X86::VMOVQI2PQIrm:
   case X86::VMOVQI2PQIZrm:
-    DecodeZeroMoveLowMask(MVT::v2i64, ShuffleMask);
+    DecodeZeroMoveLowMask(2, ShuffleMask);
     DestName = getRegName(MI->getOperand(0).getReg());
     break;
 
   case X86::MOVDI2PDIrm:
   case X86::VMOVDI2PDIrm:
   case X86::VMOVDI2PDIZrm:
-    DecodeZeroMoveLowMask(MVT::v4i32, ShuffleMask);
+    DecodeZeroMoveLowMask(4, ShuffleMask);
     DestName = getRegName(MI->getOperand(0).getReg());
     break;
 
   case X86::EXTRQI:
     if (MI->getOperand(2).isImm() &&
         MI->getOperand(3).isImm())
-      DecodeEXTRQIMask(MVT::v16i8, MI->getOperand(2).getImm(),
-                       MI->getOperand(3).getImm(),
-                       ShuffleMask);
+      DecodeEXTRQIMask(16, 8, MI->getOperand(2).getImm(),
+                       MI->getOperand(3).getImm(), ShuffleMask);
 
     DestName = getRegName(MI->getOperand(0).getReg());
     Src1Name = getRegName(MI->getOperand(1).getReg());
     break;
 
   case X86::INSERTQI:
     if (MI->getOperand(3).isImm() &&
         MI->getOperand(4).isImm())
-      DecodeINSERTQIMask(MVT::v16i8, MI->getOperand(3).getImm(),
-                         MI->getOperand(4).getImm(),
-                         ShuffleMask);
+      DecodeINSERTQIMask(16, 8, MI->getOperand(3).getImm(),
+                         MI->getOperand(4).getImm(), ShuffleMask);
 
     DestName = getRegName(MI->getOperand(0).getReg());
     Src1Name = getRegName(MI->getOperand(1).getReg());
     Src2Name = getRegName(MI->getOperand(2).getReg());
     break;
 
   case X86::VBROADCASTF128:
   case X86::VBROADCASTI128:
   CASE_AVX512_INS_COMMON(BROADCASTF64X2, Z128, rm)
   CASE_AVX512_INS_COMMON(BROADCASTI64X2, Z128, rm)
-    DecodeSubVectorBroadcast(MVT::v4f64, MVT::v2f64, ShuffleMask);
+    DecodeSubVectorBroadcast(4, 2, ShuffleMask);
     DestName = getRegName(MI->getOperand(0).getReg());
     break;
   CASE_AVX512_INS_COMMON(BROADCASTF64X2, , rm)
   CASE_AVX512_INS_COMMON(BROADCASTI64X2, , rm)
-    DecodeSubVectorBroadcast(MVT::v8f64, MVT::v2f64, ShuffleMask);
+    DecodeSubVectorBroadcast(8, 2, ShuffleMask);
     DestName = getRegName(MI->getOperand(0).getReg());
     break;
   CASE_AVX512_INS_COMMON(BROADCASTF64X4, , rm)
   CASE_AVX512_INS_COMMON(BROADCASTI64X4, , rm)
-    DecodeSubVectorBroadcast(MVT::v8f64, MVT::v4f64, ShuffleMask);
+    DecodeSubVectorBroadcast(8, 4, ShuffleMask);
     DestName = getRegName(MI->getOperand(0).getReg());
     break;
   CASE_AVX512_INS_COMMON(BROADCASTF32X4, Z256, rm)
   CASE_AVX512_INS_COMMON(BROADCASTI32X4, Z256, rm)
-    DecodeSubVectorBroadcast(MVT::v8f32, MVT::v4f32, ShuffleMask);
+    DecodeSubVectorBroadcast(8, 4, ShuffleMask);
     DestName = getRegName(MI->getOperand(0).getReg());
     break;
   CASE_AVX512_INS_COMMON(BROADCASTF32X4, , rm)
   CASE_AVX512_INS_COMMON(BROADCASTI32X4, , rm)
-    DecodeSubVectorBroadcast(MVT::v16f32, MVT::v4f32, ShuffleMask);
+    DecodeSubVectorBroadcast(16, 4, ShuffleMask);
     DestName = getRegName(MI->getOperand(0).getReg());
     break;
   CASE_AVX512_INS_COMMON(BROADCASTF32X8, , rm)
   CASE_AVX512_INS_COMMON(BROADCASTI32X8, , rm)
-    DecodeSubVectorBroadcast(MVT::v16f32, MVT::v8f32, ShuffleMask);
+    DecodeSubVectorBroadcast(16, 8, ShuffleMask);
     DestName = getRegName(MI->getOperand(0).getReg());
     break;
   CASE_AVX512_INS_COMMON(BROADCASTI32X2, Z128, r)
     Src1Name = getRegName(MI->getOperand(NumOperands - 1).getReg());
     LLVM_FALLTHROUGH;
   CASE_AVX512_INS_COMMON(BROADCASTI32X2, Z128, m)
-    DecodeSubVectorBroadcast(MVT::v4f32, MVT::v2f32, ShuffleMask);
+    DecodeSubVectorBroadcast(4, 2, ShuffleMask);
     DestName = getRegName(MI->getOperand(0).getReg());
     break;
   CASE_AVX512_INS_COMMON(BROADCASTF32X2, Z256, r)
   CASE_AVX512_INS_COMMON(BROADCASTI32X2, Z256, r)
     Src1Name = getRegName(MI->getOperand(NumOperands - 1).getReg());
     LLVM_FALLTHROUGH;
   CASE_AVX512_INS_COMMON(BROADCASTF32X2, Z256, m)
   CASE_AVX512_INS_COMMON(BROADCASTI32X2, Z256, m)
-    DecodeSubVectorBroadcast(MVT::v8f32, MVT::v2f32, ShuffleMask);
+    DecodeSubVectorBroadcast(8, 2, ShuffleMask);
     DestName = getRegName(MI->getOperand(0).getReg());
     break;
   CASE_AVX512_INS_COMMON(BROADCASTF32X2, Z, r)
   CASE_AVX512_INS_COMMON(BROADCASTI32X2, Z, r)
     Src1Name = getRegName(MI->getOperand(NumOperands - 1).getReg());
     LLVM_FALLTHROUGH;
   CASE_AVX512_INS_COMMON(BROADCASTF32X2, Z, m)
   CASE_AVX512_INS_COMMON(BROADCASTI32X2, Z, m)
-    DecodeSubVectorBroadcast(MVT::v16f32, MVT::v2f32, ShuffleMask);
+    DecodeSubVectorBroadcast(16, 2, ShuffleMask);
     DestName = getRegName(MI->getOperand(0).getReg());
     break;
 
   CASE_PMOVZX(PMOVZXBW, r)
-  CASE_PMOVZX(PMOVZXBD, r)
-  CASE_PMOVZX(PMOVZXBQ, r)
     Src1Name = getRegName(MI->getOperand(NumOperands - 1).getReg());
     LLVM_FALLTHROUGH;
-
   CASE_PMOVZX(PMOVZXBW, m)
+    DecodeZeroExtendMask(8, 16, getRegOperandNumElts(MI, 16, 0), ShuffleMask);
+    DestName = getRegName(MI->getOperand(0).getReg());
+    break;
+
+  CASE_PMOVZX(PMOVZXBD, r)
+    Src1Name = getRegName(MI->getOperand(NumOperands - 1).getReg());
+    LLVM_FALLTHROUGH;
   CASE_PMOVZX(PMOVZXBD, m)
+    DecodeZeroExtendMask(8, 32, getRegOperandNumElts(MI, 32, 0), ShuffleMask);
+    DestName = getRegName(MI->getOperand(0).getReg());
+    break;
+
+  CASE_PMOVZX(PMOVZXBQ, r)
+    Src1Name = getRegName(MI->getOperand(NumOperands - 1).getReg());
+    LLVM_FALLTHROUGH;
   CASE_PMOVZX(PMOVZXBQ, m)
-    DecodeZeroExtendMask(MVT::i8, getZeroExtensionResultType(MI), ShuffleMask);
+    DecodeZeroExtendMask(8, 64, getRegOperandNumElts(MI, 64, 0), ShuffleMask);
     DestName = getRegName(MI->getOperand(0).getReg());
     break;
 
   CASE_PMOVZX(PMOVZXWD, r)
-  CASE_PMOVZX(PMOVZXWQ, r)
     Src1Name = getRegName(MI->getOperand(NumOperands - 1).getReg());
     LLVM_FALLTHROUGH;
-
   CASE_PMOVZX(PMOVZXWD, m)
+    DecodeZeroExtendMask(16, 32, getRegOperandNumElts(MI, 32, 0), ShuffleMask);
+    DestName = getRegName(MI->getOperand(0).getReg());
+    break;
+
+  CASE_PMOVZX(PMOVZXWQ, r)
+    Src1Name = getRegName(MI->getOperand(NumOperands - 1).getReg());
+    LLVM_FALLTHROUGH;
   CASE_PMOVZX(PMOVZXWQ, m)
-    DecodeZeroExtendMask(MVT::i16, getZeroExtensionResultType(MI), ShuffleMask);
+    DecodeZeroExtendMask(16, 64, getRegOperandNumElts(MI, 64, 0), ShuffleMask);
     DestName = getRegName(MI->getOperand(0).getReg());
     break;
 
   CASE_PMOVZX(PMOVZXDQ, r)
     Src1Name = getRegName(MI->getOperand(NumOperands - 1).getReg());
     LLVM_FALLTHROUGH;
-
   CASE_PMOVZX(PMOVZXDQ, m)
-    DecodeZeroExtendMask(MVT::i32, getZeroExtensionResultType(MI), ShuffleMask);
+    DecodeZeroExtendMask(32, 64, getRegOperandNumElts(MI, 64, 0), ShuffleMask);
     DestName = getRegName(MI->getOperand(0).getReg());
     break;
   }
 
   // The only comments we decode are shuffles, so give up if we were unable to
   // decode a shuffle mask.
   if (ShuffleMask.empty())
     return false;

llvm/trunk/lib/Target/X86/Utils/X86ShuffleDecode.h

 #include "llvm/ADT/SmallVector.h"
 
 //===----------------------------------------------------------------------===//
 //  Vector Mask Decoding
 //===----------------------------------------------------------------------===//
 
 namespace llvm {
 template <typename T> class ArrayRef;
-class MVT;
 
 enum { SM_SentinelUndef = -1, SM_SentinelZero = -2 };
 
 /// Decode a 128-bit INSERTPS instruction as a v4f32 shuffle mask.
 void DecodeINSERTPSMask(unsigned Imm, SmallVectorImpl<int> &ShuffleMask);
 
 // Insert the bottom Len elements from a second source into a vector starting at
 // element Idx.
-void DecodeInsertElementMask(MVT VT, unsigned Idx, unsigned Len,
+void DecodeInsertElementMask(unsigned NumElts, unsigned Idx, unsigned Len,
                              SmallVectorImpl<int> &ShuffleMask);
 
 /// Decode a MOVHLPS instruction as a v2f64/v4f32 shuffle mask.
 /// i.e. <3,1> or <6,7,2,3>
 void DecodeMOVHLPSMask(unsigned NElts, SmallVectorImpl<int> &ShuffleMask);
 
 /// Decode a MOVLHPS instruction as a v2f64/v4f32 shuffle mask.
 /// i.e. <0,2> or <0,1,4,5>
 void DecodeMOVLHPSMask(unsigned NElts, SmallVectorImpl<int> &ShuffleMask);
 
-void DecodeMOVSLDUPMask(MVT VT, SmallVectorImpl<int> &ShuffleMask);
+void DecodeMOVSLDUPMask(unsigned NumElts, SmallVectorImpl<int> &ShuffleMask);
 
-void DecodeMOVSHDUPMask(MVT VT, SmallVectorImpl<int> &ShuffleMask);
+void DecodeMOVSHDUPMask(unsigned NumElts, SmallVectorImpl<int> &ShuffleMask);
 
-void DecodeMOVDDUPMask(MVT VT, SmallVectorImpl<int> &ShuffleMask);
+void DecodeMOVDDUPMask(unsigned NumElts, SmallVectorImpl<int> &ShuffleMask);
 
-void DecodePSLLDQMask(MVT VT, unsigned Imm, SmallVectorImpl<int> &ShuffleMask);
+void DecodePSLLDQMask(unsigned NumElts, unsigned Imm,
+                      SmallVectorImpl<int> &ShuffleMask);
 
-void DecodePSRLDQMask(MVT VT, unsigned Imm, SmallVectorImpl<int> &ShuffleMask);
+void DecodePSRLDQMask(unsigned NumElts, unsigned Imm,
+                      SmallVectorImpl<int> &ShuffleMask);
 
-void DecodePALIGNRMask(MVT VT, unsigned Imm, SmallVectorImpl<int> &ShuffleMask);
+void DecodePALIGNRMask(unsigned NumElts, unsigned Imm,
+                       SmallVectorImpl<int> &ShuffleMask);
 
-void DecodeVALIGNMask(MVT VT, unsigned Imm, SmallVectorImpl<int> &ShuffleMask);
+void DecodeVALIGNMask(unsigned NumElts, unsigned Imm,
+                      SmallVectorImpl<int> &ShuffleMask);
 
 /// Decodes the shuffle masks for pshufd/pshufw/vpermilpd/vpermilps.
 /// VT indicates the type of the vector allowing it to handle different
 /// datatypes and vector widths.
-void DecodePSHUFMask(MVT VT, unsigned Imm, SmallVectorImpl<int> &ShuffleMask);
+void DecodePSHUFMask(unsigned NumElts, unsigned ScalarBits, unsigned Imm,
+                     SmallVectorImpl<int> &ShuffleMask);
 
 /// Decodes the shuffle masks for pshufhw.
 /// VT indicates the type of the vector allowing it to handle different
 /// datatypes and vector widths.
-void DecodePSHUFHWMask(MVT VT, unsigned Imm, SmallVectorImpl<int> &ShuffleMask);
+void DecodePSHUFHWMask(unsigned NumElts, unsigned Imm,
+                       SmallVectorImpl<int> &ShuffleMask);
 
 /// Decodes the shuffle masks for pshuflw.
 /// VT indicates the type of the vector allowing it to handle different
 /// datatypes and vector widths.
-void DecodePSHUFLWMask(MVT VT, unsigned Imm, SmallVectorImpl<int> &ShuffleMask);
+void DecodePSHUFLWMask(unsigned NumElts, unsigned Imm,
+                       SmallVectorImpl<int> &ShuffleMask);
 
 /// Decodes a PSWAPD 3DNow! instruction.
-void DecodePSWAPMask(MVT VT, SmallVectorImpl<int> &ShuffleMask);
+void DecodePSWAPMask(unsigned NumElts, SmallVectorImpl<int> &ShuffleMask);
 
 /// Decodes the shuffle masks for shufp*.
 /// VT indicates the type of the vector allowing it to handle different
 /// datatypes and vector widths.
-void DecodeSHUFPMask(MVT VT, unsigned Imm, SmallVectorImpl<int> &ShuffleMask);
+void DecodeSHUFPMask(unsigned NumElts, unsigned ScalarBits, unsigned Imm,
+                     SmallVectorImpl<int> &ShuffleMask);
 
 /// Decodes the shuffle masks for unpckhps/unpckhpd and punpckh*.
 /// VT indicates the type of the vector allowing it to handle different
 /// datatypes and vector widths.
-void DecodeUNPCKHMask(MVT VT, SmallVectorImpl<int> &ShuffleMask);
+void DecodeUNPCKHMask(unsigned NumElts, unsigned ScalarBits,
+                      SmallVectorImpl<int> &ShuffleMask);
 
 /// Decodes the shuffle masks for unpcklps/unpcklpd and punpckl*.
 /// VT indicates the type of the vector allowing it to handle different
 /// datatypes and vector widths.
-void DecodeUNPCKLMask(MVT VT, SmallVectorImpl<int> &ShuffleMask);
+void DecodeUNPCKLMask(unsigned NumElts, unsigned ScalarBits,
+                      SmallVectorImpl<int> &ShuffleMask);
 
 /// Decodes a broadcast of the first element of a vector.
-void DecodeVectorBroadcast(MVT DstVT, SmallVectorImpl<int> &ShuffleMask);
+void DecodeVectorBroadcast(unsigned NumElts, SmallVectorImpl<int> &ShuffleMask);
 
 /// Decodes a broadcast of a subvector to a larger vector type.
-void DecodeSubVectorBroadcast(MVT DstVT, MVT SrcVT,
+void DecodeSubVectorBroadcast(unsigned DstNumElts, unsigned SrcNumElts,
                               SmallVectorImpl<int> &ShuffleMask);
 
 /// Decode a PSHUFB mask from a raw array of constants such as from
 /// BUILD_VECTOR.
 void DecodePSHUFBMask(ArrayRef<uint64_t> RawMask,
                       SmallVectorImpl<int> &ShuffleMask);
 
 /// Decode a BLEND immediate mask into a shuffle mask.
-void DecodeBLENDMask(MVT VT, unsigned Imm, SmallVectorImpl<int> &ShuffleMask);
+void DecodeBLENDMask(unsigned NumElts, unsigned Imm,
+                     SmallVectorImpl<int> &ShuffleMask);
 
-void DecodeVPERM2X128Mask(MVT VT, unsigned Imm,
+void DecodeVPERM2X128Mask(unsigned NumElts, unsigned Imm,
                           SmallVectorImpl<int> &ShuffleMask);
 
 /// Decode a shuffle packed values at 128-bit granularity
 /// immediate mask into a shuffle mask.
-void decodeVSHUF64x2FamilyMask(MVT VT, unsigned Imm,
-                               SmallVectorImpl<int> &ShuffleMask);
+void decodeVSHUF64x2FamilyMask(unsigned NumElts, unsigned ScalarSize,
+                               unsigned Imm, SmallVectorImpl<int> &ShuffleMask);
 
 /// Decodes the shuffle masks for VPERMQ/VPERMPD.
-void DecodeVPERMMask(MVT VT, unsigned Imm, SmallVectorImpl<int> &ShuffleMask);
+void DecodeVPERMMask(unsigned NumElts, unsigned Imm,
+                     SmallVectorImpl<int> &ShuffleMask);
 
 /// Decode a VPPERM mask from a raw array of constants such as from
 /// BUILD_VECTOR.
 /// This can only basic masks (permutes + zeros), not any of the other
 /// operations that VPPERM can perform.
 void DecodeVPPERMMask(ArrayRef<uint64_t> RawMask,
                       SmallVectorImpl<int> &ShuffleMask);
 
 /// Decode a zero extension instruction as a shuffle mask.
-void DecodeZeroExtendMask(MVT SrcScalarVT, MVT DstVT,
+void DecodeZeroExtendMask(unsigned SrcScalarBits, unsigned DstScalarBits,
+                          unsigned NumDstElts,
                           SmallVectorImpl<int> &ShuffleMask);
 
 /// Decode a move lower and zero upper instruction as a shuffle mask.
-void DecodeZeroMoveLowMask(MVT VT, SmallVectorImpl<int> &ShuffleMask);
+void DecodeZeroMoveLowMask(unsigned NumElts, SmallVectorImpl<int> &ShuffleMask);
 
 /// Decode a scalar float move instruction as a shuffle mask.
-void DecodeScalarMoveMask(MVT VT, bool IsLoad,
+void DecodeScalarMoveMask(unsigned NumElts, bool IsLoad,
                           SmallVectorImpl<int> &ShuffleMask);
 
 /// Decode a SSE4A EXTRQ instruction as a shuffle mask.
-void DecodeEXTRQIMask(MVT VT, int Len, int Idx,
+void DecodeEXTRQIMask(unsigned NumElts, unsigned EltSize, int Len, int Idx,
                       SmallVectorImpl<int> &ShuffleMask);
 
 /// Decode a SSE4A INSERTQ instruction as a shuffle mask.
-void DecodeINSERTQIMask(MVT VT, int Len, int Idx,
+void DecodeINSERTQIMask(unsigned NumElts, unsigned EltSize, int Len, int Idx,
                         SmallVectorImpl<int> &ShuffleMask);
 
 /// Decode a VPERMILPD/VPERMILPS variable mask from a raw array of constants.
-void DecodeVPERMILPMask(MVT VT, ArrayRef<uint64_t> RawMask,
+void DecodeVPERMILPMask(unsigned NumElts, unsigned ScalarBits,
+                        ArrayRef<uint64_t> RawMask,
                         SmallVectorImpl<int> &ShuffleMask);
 
 /// Decode a VPERMIL2PD/VPERMIL2PS variable mask from a raw array of constants.
-void DecodeVPERMIL2PMask(MVT VT, unsigned M2Z, ArrayRef<uint64_t> RawMask,
+void DecodeVPERMIL2PMask(unsigned NumElts, unsigned ScalarBits, unsigned M2Z,
+                         ArrayRef<uint64_t> RawMask,
                          SmallVectorImpl<int> &ShuffleMask);
 
 /// Decode a VPERM W/D/Q/PS/PD mask from a raw array of constants.
 void DecodeVPERMVMask(ArrayRef<uint64_t> RawMask,
                       SmallVectorImpl<int> &ShuffleMask);
 
 /// Decode a VPERMT2 W/D/Q/PS/PD mask from a raw array of constants.
 void DecodeVPERMV3Mask(ArrayRef<uint64_t> RawMask,
                       SmallVectorImpl<int> &ShuffleMask);
 } // llvm namespace
 
 #endif

llvm/trunk/lib/Target/X86/Utils/X86ShuffleDecode.cpp

 //===-- X86ShuffleDecode.cpp - X86 shuffle decode logic -------------------===//
 //
 //                     The LLVM Compiler Infrastructure
 //
 // This file is distributed under the University of Illinois Open Source
 // License. See LICENSE.TXT for details.
 //
 //===----------------------------------------------------------------------===//
 //
 // Define several functions to decode x86 specific shuffle semantics into a
 // generic vector mask.
 //
 //===----------------------------------------------------------------------===//
 
 #include "X86ShuffleDecode.h"
 #include "llvm/ADT/ArrayRef.h"
-#include "llvm/CodeGen/MachineValueType.h"
 
 //===----------------------------------------------------------------------===//
 //  Vector Mask Decoding
 //===----------------------------------------------------------------------===//
 
 namespace llvm {
 
 void DecodeINSERTPSMask(unsigned Imm, SmallVectorImpl<int> &ShuffleMask) {
void DecodeINSERTPSMask(unsigned Imm, SmallVectorImpl<int> &ShuffleMask) {
   ShuffleMask[CountD] = InVal;
   // ZMask zaps values, potentially overriding the CountD elt.
   if (ZMask & 1) ShuffleMask[0] = SM_SentinelZero;
   if (ZMask & 2) ShuffleMask[1] = SM_SentinelZero;
   if (ZMask & 4) ShuffleMask[2] = SM_SentinelZero;
   if (ZMask & 8) ShuffleMask[3] = SM_SentinelZero;
 }
 
-void DecodeInsertElementMask(MVT VT, unsigned Idx, unsigned Len,
+void DecodeInsertElementMask(unsigned NumElts, unsigned Idx, unsigned Len,
                              SmallVectorImpl<int> &ShuffleMask) {
-  unsigned NumElts = VT.getVectorNumElements();
   assert((Idx + Len) <= NumElts && "Insertion out of range");
 
   for (unsigned i = 0; i != NumElts; ++i)
     ShuffleMask.push_back(i);
   for (unsigned i = 0; i != Len; ++i)
     ShuffleMask[Idx + i] = NumElts + i;
 }
 

 void DecodeMOVLHPSMask(unsigned NElts, SmallVectorImpl<int> &ShuffleMask) {
   for (unsigned i = 0; i != NElts / 2; ++i)
     ShuffleMask.push_back(i);
 
   for (unsigned i = 0; i != NElts / 2; ++i)
     ShuffleMask.push_back(NElts + i);
 }
 
-void DecodeMOVSLDUPMask(MVT VT, SmallVectorImpl<int> &ShuffleMask) {
-  unsigned NumElts = VT.getVectorNumElements();
+void DecodeMOVSLDUPMask(unsigned NumElts, SmallVectorImpl<int> &ShuffleMask) {
   for (int i = 0, e = NumElts / 2; i < e; ++i) {
     ShuffleMask.push_back(2 * i);
     ShuffleMask.push_back(2 * i);
   }
 }
 
-void DecodeMOVSHDUPMask(MVT VT, SmallVectorImpl<int> &ShuffleMask) {
-  unsigned NumElts = VT.getVectorNumElements();
+void DecodeMOVSHDUPMask(unsigned NumElts, SmallVectorImpl<int> &ShuffleMask) {
   for (int i = 0, e = NumElts / 2; i < e; ++i) {
     ShuffleMask.push_back(2 * i + 1);
     ShuffleMask.push_back(2 * i + 1);
   }
 }
 
-void DecodeMOVDDUPMask(MVT VT, SmallVectorImpl<int> &ShuffleMask) {
-  unsigned VectorSizeInBits = VT.getSizeInBits();
-  unsigned ScalarSizeInBits = VT.getScalarSizeInBits();
-  unsigned NumElts = VT.getVectorNumElements();
-  unsigned NumLanes = VectorSizeInBits / 128;
-  unsigned NumLaneElts = NumElts / NumLanes;
-  unsigned NumLaneSubElts = 64 / ScalarSizeInBits;
+void DecodeMOVDDUPMask(unsigned NumElts, SmallVectorImpl<int> &ShuffleMask) {
+  const unsigned NumLaneElts = 2;
 
   for (unsigned l = 0; l < NumElts; l += NumLaneElts)
-    for (unsigned i = 0; i < NumLaneElts; i += NumLaneSubElts)
-      for (unsigned s = 0; s != NumLaneSubElts; s++)
-        ShuffleMask.push_back(l + s);
+    for (unsigned i = 0; i < NumLaneElts; ++i)
+      ShuffleMask.push_back(l);
 }
 
-void DecodePSLLDQMask(MVT VT, unsigned Imm, SmallVectorImpl<int> &ShuffleMask) {
-  unsigned VectorSizeInBits = VT.getSizeInBits();
-  unsigned NumElts = VectorSizeInBits / 8;
-  unsigned NumLanes = VectorSizeInBits / 128;
-  unsigned NumLaneElts = NumElts / NumLanes;
+void DecodePSLLDQMask(unsigned NumElts, unsigned Imm,
+                      SmallVectorImpl<int> &ShuffleMask) {
+  const unsigned NumLaneElts = 16;
 
   for (unsigned l = 0; l < NumElts; l += NumLaneElts)
     for (unsigned i = 0; i < NumLaneElts; ++i) {
       int M = SM_SentinelZero;
       if (i >= Imm) M = i - Imm + l;
       ShuffleMask.push_back(M);
     }
 }
 
-void DecodePSRLDQMask(MVT VT, unsigned Imm, SmallVectorImpl<int> &ShuffleMask) {
-  unsigned VectorSizeInBits = VT.getSizeInBits();
-  unsigned NumElts = VectorSizeInBits / 8;
-  unsigned NumLanes = VectorSizeInBits / 128;
-  unsigned NumLaneElts = NumElts / NumLanes;
+void DecodePSRLDQMask(unsigned NumElts, unsigned Imm,
+                      SmallVectorImpl<int> &ShuffleMask) {
+  const unsigned NumLaneElts = 16;
 
   for (unsigned l = 0; l < NumElts; l += NumLaneElts)
     for (unsigned i = 0; i < NumLaneElts; ++i) {
       unsigned Base = i + Imm;
       int M = Base + l;
       if (Base >= NumLaneElts) M = SM_SentinelZero;
       ShuffleMask.push_back(M);
     }
 }
 
-void DecodePALIGNRMask(MVT VT, unsigned Imm,
+void DecodePALIGNRMask(unsigned NumElts, unsigned Imm,
                        SmallVectorImpl<int> &ShuffleMask) {
-  unsigned NumElts = VT.getVectorNumElements();
-  unsigned Offset = Imm * (VT.getScalarSizeInBits() / 8);
-
-  unsigned NumLanes = VT.getSizeInBits() / 128;
-  unsigned NumLaneElts = NumElts / NumLanes;
+  const unsigned NumLaneElts = 16;
 
   for (unsigned l = 0; l != NumElts; l += NumLaneElts) {
     for (unsigned i = 0; i != NumLaneElts; ++i) {
-      unsigned Base = i + Offset;
-      // if i+offset is out of this lane then we actually need the other source
+      unsigned Base = i + Imm;
+      // if i+imm is out of this lane then we actually need the other source
       if (Base >= NumLaneElts) Base += NumElts - NumLaneElts;
       ShuffleMask.push_back(Base + l);
     }
   }
 }
 
-void DecodeVALIGNMask(MVT VT, unsigned Imm,
+void DecodeVALIGNMask(unsigned NumElts, unsigned Imm,
                       SmallVectorImpl<int> &ShuffleMask) {
-  int NumElts = VT.getVectorNumElements();
   // Not all bits of the immediate are used so mask it.
   assert(isPowerOf2_32(NumElts) && "NumElts should be power of 2");
   Imm = Imm & (NumElts - 1);
-  for (int i = 0; i != NumElts; ++i)
+  for (unsigned i = 0; i != NumElts; ++i)
     ShuffleMask.push_back(i + Imm);
 }
 
 /// DecodePSHUFMask - This decodes the shuffle masks for pshufw, pshufd, and vpermilp*.
 /// VT indicates the type of the vector allowing it to handle different
 /// datatypes and vector widths.
-void DecodePSHUFMask(MVT VT, unsigned Imm, SmallVectorImpl<int> &ShuffleMask) {
-  unsigned NumElts = VT.getVectorNumElements();
-
-  unsigned NumLanes = VT.getSizeInBits() / 128;
+void DecodePSHUFMask(unsigned NumElts, unsigned ScalarBits, unsigned Imm,
+                     SmallVectorImpl<int> &ShuffleMask) {
+  unsigned Size = NumElts * ScalarBits;
+  unsigned NumLanes = Size / 128;
   if (NumLanes == 0) NumLanes = 1;  // Handle MMX
   unsigned NumLaneElts = NumElts / NumLanes;
 
   unsigned NewImm = Imm;
   for (unsigned l = 0; l != NumElts; l += NumLaneElts) {
     for (unsigned i = 0; i != NumLaneElts; ++i) {
       ShuffleMask.push_back(NewImm % NumLaneElts + l);
       NewImm /= NumLaneElts;
     }
     if (NumLaneElts == 4) NewImm = Imm; // reload imm
   }
 }
 
-void DecodePSHUFHWMask(MVT VT, unsigned Imm,
+void DecodePSHUFHWMask(unsigned NumElts, unsigned Imm,
                        SmallVectorImpl<int> &ShuffleMask) {
-  unsigned NumElts = VT.getVectorNumElements();
-
   for (unsigned l = 0; l != NumElts; l += 8) {
     unsigned NewImm = Imm;
     for (unsigned i = 0, e = 4; i != e; ++i) {
       ShuffleMask.push_back(l + i);
     }
     for (unsigned i = 4, e = 8; i != e; ++i) {
       ShuffleMask.push_back(l + 4 + (NewImm & 3));
       NewImm >>= 2;
     }
   }
 }
 
-void DecodePSHUFLWMask(MVT VT, unsigned Imm,
+void DecodePSHUFLWMask(unsigned NumElts, unsigned Imm,
                        SmallVectorImpl<int> &ShuffleMask) {
-  unsigned NumElts = VT.getVectorNumElements();
-
   for (unsigned l = 0; l != NumElts; l += 8) {
     unsigned NewImm = Imm;
     for (unsigned i = 0, e = 4; i != e; ++i) {
       ShuffleMask.push_back(l + (NewImm & 3));
       NewImm >>= 2;
     }
     for (unsigned i = 4, e = 8; i != e; ++i) {
       ShuffleMask.push_back(l + i);
     }
   }
 }
 
-void DecodePSWAPMask(MVT VT, SmallVectorImpl<int> &ShuffleMask) {
-  unsigned NumElts = VT.getVectorNumElements();
+void DecodePSWAPMask(unsigned NumElts, SmallVectorImpl<int> &ShuffleMask) {
   unsigned NumHalfElts = NumElts / 2;
 
   for (unsigned l = 0; l != NumHalfElts; ++l)
     ShuffleMask.push_back(l + NumHalfElts);
   for (unsigned h = 0; h != NumHalfElts; ++h)
     ShuffleMask.push_back(h);
 }
 
 /// DecodeSHUFPMask - This decodes the shuffle masks for shufp*. VT indicates
 /// the type of the vector allowing it to handle different datatypes and vector
 /// widths.
-void DecodeSHUFPMask(MVT VT, unsigned Imm, SmallVectorImpl<int> &ShuffleMask) {
-  unsigned NumElts = VT.getVectorNumElements();
-
-  unsigned NumLanes = VT.getSizeInBits() / 128;
-  unsigned NumLaneElts = NumElts / NumLanes;
+void DecodeSHUFPMask(unsigned NumElts, unsigned ScalarBits,
+                     unsigned Imm, SmallVectorImpl<int> &ShuffleMask) {
+  unsigned NumLaneElts = 128 / ScalarBits;
 
   unsigned NewImm = Imm;
   for (unsigned l = 0; l != NumElts; l += NumLaneElts) {
     // each half of a lane comes from different source
     for (unsigned s = 0; s != NumElts * 2; s += NumElts) {
       for (unsigned i = 0; i != NumLaneElts / 2; ++i) {
         ShuffleMask.push_back(NewImm % NumLaneElts + s + l);
         NewImm /= NumLaneElts;
       }
     }
     if (NumLaneElts == 4) NewImm = Imm; // reload imm
   }
 }
 
 /// DecodeUNPCKHMask - This decodes the shuffle masks for unpckhps/unpckhpd
 /// and punpckh*. VT indicates the type of the vector allowing it to handle
 /// different datatypes and vector widths.
-void DecodeUNPCKHMask(MVT VT, SmallVectorImpl<int> &ShuffleMask) {
-  unsigned NumElts = VT.getVectorNumElements();
-
+void DecodeUNPCKHMask(unsigned NumElts, unsigned ScalarBits,
+                      SmallVectorImpl<int> &ShuffleMask) {
   // Handle 128 and 256-bit vector lengths. AVX defines UNPCK* to operate
   // independently on 128-bit lanes.
-  unsigned NumLanes = VT.getSizeInBits() / 128;
+  unsigned NumLanes = (NumElts * ScalarBits) / 128;
   if (NumLanes == 0) NumLanes = 1;  // Handle MMX
   unsigned NumLaneElts = NumElts / NumLanes;
 
   for (unsigned l = 0; l != NumElts; l += NumLaneElts) {
     for (unsigned i = l + NumLaneElts / 2, e = l + NumLaneElts; i != e; ++i) {
       ShuffleMask.push_back(i);           // Reads from dest/src1
       ShuffleMask.push_back(i + NumElts); // Reads from src/src2
     }
   }
 }
 
 /// DecodeUNPCKLMask - This decodes the shuffle masks for unpcklps/unpcklpd
 /// and punpckl*. VT indicates the type of the vector allowing it to handle
 /// different datatypes and vector widths.
-void DecodeUNPCKLMask(MVT VT, SmallVectorImpl<int> &ShuffleMask) {
-  unsigned NumElts = VT.getVectorNumElements();
-
+void DecodeUNPCKLMask(unsigned NumElts, unsigned ScalarBits,
+                      SmallVectorImpl<int> &ShuffleMask) {
   // Handle 128 and 256-bit vector lengths. AVX defines UNPCK* to operate
   // independently on 128-bit lanes.
-  unsigned NumLanes = VT.getSizeInBits() / 128;
+  unsigned NumLanes = (NumElts * ScalarBits) / 128;
   if (NumLanes == 0 ) NumLanes = 1;  // Handle MMX
   unsigned NumLaneElts = NumElts / NumLanes;
 
   for (unsigned l = 0; l != NumElts; l += NumLaneElts) {
     for (unsigned i = l, e = l + NumLaneElts / 2; i != e; ++i) {
       ShuffleMask.push_back(i);           // Reads from dest/src1
       ShuffleMask.push_back(i + NumElts); // Reads from src/src2
     }
   }
 }
 
 /// Decodes a broadcast of the first element of a vector.
-void DecodeVectorBroadcast(MVT DstVT, SmallVectorImpl<int> &ShuffleMask) {
-  unsigned NumElts = DstVT.getVectorNumElements();
+void DecodeVectorBroadcast(unsigned NumElts,
+                           SmallVectorImpl<int> &ShuffleMask) {
   ShuffleMask.append(NumElts, 0);
 }
 
 /// Decodes a broadcast of a subvector to a larger vector type.
-void DecodeSubVectorBroadcast(MVT DstVT, MVT SrcVT,
+void DecodeSubVectorBroadcast(unsigned DstNumElts, unsigned SrcNumElts,
                               SmallVectorImpl<int> &ShuffleMask) {
-  assert(SrcVT.getScalarType() == DstVT.getScalarType() &&
-         "Non matching vector element types");
-  unsigned NumElts = SrcVT.getVectorNumElements();
-  unsigned Scale = DstVT.getSizeInBits() / SrcVT.getSizeInBits();
+  unsigned Scale = DstNumElts / SrcNumElts;
 
   for (unsigned i = 0; i != Scale; ++i)
-    for (unsigned j = 0; j != NumElts; ++j)
+    for (unsigned j = 0; j != SrcNumElts; ++j)
       ShuffleMask.push_back(j);
 }
 
 /// \brief Decode a shuffle packed values at 128-bit granularity
 /// (SHUFF32x4/SHUFF64x2/SHUFI32x4/SHUFI64x2)
 /// immediate mask into a shuffle mask.
-void decodeVSHUF64x2FamilyMask(MVT VT, unsigned Imm,
+void decodeVSHUF64x2FamilyMask(unsigned NumElts, unsigned ScalarSize,
+                               unsigned Imm,
                                SmallVectorImpl<int> &ShuffleMask) {
-  unsigned NumLanes = VT.getSizeInBits() / 128;
-  unsigned NumElementsInLane = 128 / VT.getScalarSizeInBits();
+  unsigned NumElementsInLane = 128 / ScalarSize;
+  unsigned NumLanes = NumElts / NumElementsInLane;
   unsigned ControlBitsMask = NumLanes - 1;
   unsigned NumControlBits  = NumLanes / 2;
 
   for (unsigned l = 0; l != NumLanes; ++l) {
     unsigned LaneMask = (Imm >> (l * NumControlBits)) & ControlBitsMask;
     // We actually need the other source.
     if (l >= NumLanes / 2)
       LaneMask += NumLanes;
     for (unsigned i = 0; i != NumElementsInLane; ++i)
       ShuffleMask.push_back(LaneMask * NumElementsInLane + i);
   }
 }
 
-void DecodeVPERM2X128Mask(MVT VT, unsigned Imm,
+void DecodeVPERM2X128Mask(unsigned NumElts, unsigned Imm,
                           SmallVectorImpl<int> &ShuffleMask) {
-  unsigned HalfSize = VT.getVectorNumElements() / 2;
+  unsigned HalfSize = NumElts / 2;
 
   for (unsigned l = 0; l != 2; ++l) {
     unsigned HalfMask = Imm >> (l * 4);
     unsigned HalfBegin = (HalfMask & 0x3) * HalfSize;
     for (unsigned i = HalfBegin, e = HalfBegin + HalfSize; i != e; ++i)
       ShuffleMask.push_back(HalfMask & 8 ? SM_SentinelZero : (int)i);
   }
 }
for (int i = 0, e = RawMask.size(); i < e; ++i) {
     else {
       // Only the least significant 4 bits of the byte are used.
       int Index = Base + (M & 0xf);
       ShuffleMask.push_back(Index);
     }
   }
 }
 
-void DecodeBLENDMask(MVT VT, unsigned Imm, SmallVectorImpl<int> &ShuffleMask) {
-  int ElementBits = VT.getScalarSizeInBits();
-  int NumElements = VT.getVectorNumElements();
-  for (int i = 0; i < NumElements; ++i) {
+void DecodeBLENDMask(unsigned NumElts, unsigned Imm,
+                     SmallVectorImpl<int> &ShuffleMask) {
+  for (unsigned i = 0; i < NumElts; ++i) {
     // If there are more than 8 elements in the vector, then any immediate blend
-    // mask applies to each 128-bit lane. There can never be more than
-    // 8 elements in a 128-bit lane with an immediate blend.
-    int Bit = NumElements > 8 ? i % (128 / ElementBits) : i;
-    assert(Bit < 8 &&
-           "Immediate blends only operate over 8 elements at a time!");
-    ShuffleMask.push_back(((Imm >> Bit) & 1) ? NumElements + i : i);
+    // mask wraps around.
+    unsigned Bit = i % 8;
+    ShuffleMask.push_back(((Imm >> Bit) & 1) ? NumElts + i : i);
   }
 }
 
 void DecodeVPPERMMask(ArrayRef<uint64_t> RawMask,
                       SmallVectorImpl<int> &ShuffleMask) {
   assert(RawMask.size() == 16 && "Illegal VPPERM shuffle mask size");
 
   // VPPERM Operation
for (int i = 0, e = RawMask.size(); i < e; ++i) {
     }
 
     uint64_t Index = M & 0x1F;
     ShuffleMask.push_back((int)Index);
   }
 }
 
 /// DecodeVPERMMask - this decodes the shuffle masks for VPERMQ/VPERMPD.
-void DecodeVPERMMask(MVT VT, unsigned Imm, SmallVectorImpl<int> &ShuffleMask) {
-  assert((VT.is256BitVector() || VT.is512BitVector()) &&
-         (VT.getScalarSizeInBits() == 64) && "Unexpected vector value type");
-  unsigned NumElts = VT.getVectorNumElements();
+void DecodeVPERMMask(unsigned NumElts, unsigned Imm,
+                     SmallVectorImpl<int> &ShuffleMask) {
   for (unsigned l = 0; l != NumElts; l += 4)
     for (unsigned i = 0; i != 4; ++i)
       ShuffleMask.push_back(l + ((Imm >> (2 * i)) & 3));
 }
 
-void DecodeZeroExtendMask(MVT SrcScalarVT, MVT DstVT, SmallVectorImpl<int> &Mask) {
-  unsigned NumDstElts = DstVT.getVectorNumElements();
-  unsigned SrcScalarBits = SrcScalarVT.getSizeInBits();
-  unsigned DstScalarBits = DstVT.getScalarSizeInBits();
+void DecodeZeroExtendMask(unsigned SrcScalarBits, unsigned DstScalarBits,
+                          unsigned NumDstElts, SmallVectorImpl<int> &Mask) {
   unsigned Scale = DstScalarBits / SrcScalarBits;
   assert(SrcScalarBits < DstScalarBits &&
          "Expected zero extension mask to increase scalar size");
 
   for (unsigned i = 0; i != NumDstElts; i++) {
     Mask.push_back(i);
     for (unsigned j = 1; j != Scale; j++)
       Mask.push_back(SM_SentinelZero);
   }
 }
 
-void DecodeZeroMoveLowMask(MVT VT, SmallVectorImpl<int> &ShuffleMask) {
-  unsigned NumElts = VT.getVectorNumElements();
+void DecodeZeroMoveLowMask(unsigned NumElts,
+                           SmallVectorImpl<int> &ShuffleMask) {
   ShuffleMask.push_back(0);
   for (unsigned i = 1; i < NumElts; i++)
     ShuffleMask.push_back(SM_SentinelZero);
 }
 
-void DecodeScalarMoveMask(MVT VT, bool IsLoad, SmallVectorImpl<int> &Mask) {
+void DecodeScalarMoveMask(unsigned NumElts, bool IsLoad,
+                          SmallVectorImpl<int> &Mask) {
   // First element comes from the first element of second source.
   // Remaining elements: Load zero extends / Move copies from first source.
-  unsigned NumElts = VT.getVectorNumElements();
   Mask.push_back(NumElts);
   for (unsigned i = 1; i < NumElts; i++)
     Mask.push_back(IsLoad ? static_cast<int>(SM_SentinelZero) : i);
 }
 
-void DecodeEXTRQIMask(MVT VT, int Len, int Idx,
+void DecodeEXTRQIMask(unsigned NumElts, unsigned EltSize, int Len, int Idx,
                       SmallVectorImpl<int> &ShuffleMask) {
-  assert(VT.is128BitVector() && "Expected 128-bit vector");
-  unsigned NumElts = VT.getVectorNumElements();
-  unsigned EltSize = VT.getScalarSizeInBits();
   unsigned HalfElts = NumElts / 2;
 
   // Only the bottom 6 bits are valid for each immediate.
   Len &= 0x3F;
   Idx &= 0x3F;
 
   // We can only decode this bit extraction instruction as a shuffle if both the
   // length and index work with whole elements.
void DecodeEXTRQIMask(unsigned NumElts, unsigned EltSize, int Len, int Idx,
   for (int i = 0; i != Len; ++i)
     ShuffleMask.push_back(i + Idx);
   for (int i = Len; i != (int)HalfElts; ++i)
     ShuffleMask.push_back(SM_SentinelZero);
   for (int i = HalfElts; i != (int)NumElts; ++i)
     ShuffleMask.push_back(SM_SentinelUndef);
 }
 
-void DecodeINSERTQIMask(MVT VT, int Len, int Idx,
+void DecodeINSERTQIMask(unsigned NumElts, unsigned EltSize, int Len, int Idx,
                         SmallVectorImpl<int> &ShuffleMask) {
-  assert(VT.is128BitVector() && "Expected 128-bit vector");
-  unsigned NumElts = VT.getVectorNumElements();
-  unsigned EltSize = VT.getScalarSizeInBits();
   unsigned HalfElts = NumElts / 2;
 
   // Only the bottom 6 bits are valid for each immediate.
   Len &= 0x3F;
   Idx &= 0x3F;
 
   // We can only decode this bit insertion instruction as a shuffle if both the
   // length and index work with whole elements.
void DecodeINSERTQIMask(unsigned NumElts, unsigned EltSize, int Len, int Idx,
   for (int i = 0; i != Len; ++i)
     ShuffleMask.push_back(i + NumElts);
   for (int i = Idx + Len; i != (int)HalfElts; ++i)
     ShuffleMask.push_back(i);
   for (int i = HalfElts; i != (int)NumElts; ++i)
     ShuffleMask.push_back(SM_SentinelUndef);
 }
 
-void DecodeVPERMILPMask(MVT VT, ArrayRef<uint64_t> RawMask,
+void DecodeVPERMILPMask(unsigned NumElts, unsigned ScalarBits,
+                        ArrayRef<uint64_t> RawMask,
                         SmallVectorImpl<int> &ShuffleMask) {
-  unsigned VecSize = VT.getSizeInBits();
-  unsigned EltSize = VT.getScalarSizeInBits();
+  unsigned VecSize = NumElts * ScalarBits;
   unsigned NumLanes = VecSize / 128;
-  unsigned NumEltsPerLane = VT.getVectorNumElements() / NumLanes;
+  unsigned NumEltsPerLane = NumElts / NumLanes;
   assert((VecSize == 128 || VecSize == 256 || VecSize == 512) &&
          "Unexpected vector size");
-  assert((EltSize == 32 || EltSize == 64) && "Unexpected element size");
+  assert((ScalarBits == 32 || ScalarBits == 64) && "Unexpected element size");
 
   for (unsigned i = 0, e = RawMask.size(); i < e; ++i) {
     uint64_t M = RawMask[i];
-    M = (EltSize == 64 ? ((M >> 1) & 0x1) : (M & 0x3));
+    M = (ScalarBits == 64 ? ((M >> 1) & 0x1) : (M & 0x3));
     unsigned LaneOffset = i & ~(NumEltsPerLane - 1);
     ShuffleMask.push_back((int)(LaneOffset + M));
   }
 }
 
-void DecodeVPERMIL2PMask(MVT VT, unsigned M2Z, ArrayRef<uint64_t> RawMask,
+void DecodeVPERMIL2PMask(unsigned NumElts, unsigned ScalarBits, unsigned M2Z,
+                         ArrayRef<uint64_t> RawMask,
                          SmallVectorImpl<int> &ShuffleMask) {
-  unsigned VecSize = VT.getSizeInBits();
-  unsigned EltSize = VT.getScalarSizeInBits();
+  unsigned VecSize = NumElts * ScalarBits;
   unsigned NumLanes = VecSize / 128;
-  unsigned NumElts = VT.getVectorNumElements();
   unsigned NumEltsPerLane = NumElts / NumLanes;
   assert((VecSize == 128 || VecSize == 256) && "Unexpected vector size");
-  assert((EltSize == 32 || EltSize == 64) && "Unexpected element size");
+  assert((ScalarBits == 32 || ScalarBits == 64) && "Unexpected element size");
   assert((NumElts == RawMask.size()) && "Unexpected mask size");
 
   for (unsigned i = 0, e = RawMask.size(); i < e; ++i) {
     // VPERMIL2 Operation.
     // Bits[3] - Match Bit.
     // Bits[2:1] - (Per Lane) PD Shuffle Mask.
     // Bits[2:0] - (Per Lane) PS Shuffle Mask.
     uint64_t Selector = RawMask[i];
     unsigned MatchBit = (Selector >> 3) & 0x1;
 
     // M2Z[0:1]     MatchBit
     //   0Xb           X        Source selected by Selector index.
     //   10b           0        Source selected by Selector index.
     //   10b           1        Zero.
     //   11b           0        Zero.
     //   11b           1        Source selected by Selector index.
     if ((M2Z & 0x2) != 0 && MatchBit != (M2Z & 0x1)) {
       ShuffleMask.push_back(SM_SentinelZero);
       continue;
     }
 
     int Index = i & ~(NumEltsPerLane - 1);
-    if (EltSize == 64)
+    if (ScalarBits == 64)
       Index += (Selector >> 1) & 0x1;
     else
       Index += Selector & 0x3;
 
     int Src = (Selector >> 2) & 0x1;
     Index += Src * NumElts;
     ShuffleMask.push_back(Index);
   }

llvm/trunk/lib/Target/X86/X86ISelLowering.cpp

static bool getTargetShuffleMask(SDNode *N, MVT VT, bool AllowSentinelZero,
 
   IsUnary = false;
   bool IsFakeUnary = false;
   switch(N->getOpcode()) {
   case X86ISD::BLENDI:
     assert(N->getOperand(0).getValueType() == VT && "Unexpected value type");
     assert(N->getOperand(1).getValueType() == VT && "Unexpected value type");
     ImmN = N->getOperand(N->getNumOperands()-1);
-    DecodeBLENDMask(VT, cast<ConstantSDNode>(ImmN)->getZExtValue(), Mask);
+    DecodeBLENDMask(NumElems, cast<ConstantSDNode>(ImmN)->getZExtValue(), Mask);
     IsUnary = IsFakeUnary = N->getOperand(0) == N->getOperand(1);
     break;
   case X86ISD::SHUFP:
     assert(N->getOperand(0).getValueType() == VT && "Unexpected value type");
     assert(N->getOperand(1).getValueType() == VT && "Unexpected value type");
     ImmN = N->getOperand(N->getNumOperands()-1);
-    DecodeSHUFPMask(VT, cast<ConstantSDNode>(ImmN)->getZExtValue(), Mask);
+    DecodeSHUFPMask(NumElems, VT.getScalarSizeInBits(),
+                    cast<ConstantSDNode>(ImmN)->getZExtValue(), Mask);
     IsUnary = IsFakeUnary = N->getOperand(0) == N->getOperand(1);
     break;
   case X86ISD::INSERTPS:
     assert(N->getOperand(0).getValueType() == VT && "Unexpected value type");
     assert(N->getOperand(1).getValueType() == VT && "Unexpected value type");
     ImmN = N->getOperand(N->getNumOperands()-1);
     DecodeINSERTPSMask(cast<ConstantSDNode>(ImmN)->getZExtValue(), Mask);
     IsUnary = IsFakeUnary = N->getOperand(0) == N->getOperand(1);
     break;
   case X86ISD::EXTRQI:
     assert(N->getOperand(0).getValueType() == VT && "Unexpected value type");
     if (isa<ConstantSDNode>(N->getOperand(1)) &&
         isa<ConstantSDNode>(N->getOperand(2))) {
       int BitLen = N->getConstantOperandVal(1);
       int BitIdx = N->getConstantOperandVal(2);
-      DecodeEXTRQIMask(VT, BitLen, BitIdx, Mask);
+      DecodeEXTRQIMask(NumElems, VT.getScalarSizeInBits(), BitLen, BitIdx,
+                       Mask);
       IsUnary = true;
     }
     break;
   case X86ISD::INSERTQI:
     assert(N->getOperand(0).getValueType() == VT && "Unexpected value type");
     assert(N->getOperand(1).getValueType() == VT && "Unexpected value type");
     if (isa<ConstantSDNode>(N->getOperand(2)) &&
         isa<ConstantSDNode>(N->getOperand(3))) {
       int BitLen = N->getConstantOperandVal(2);
       int BitIdx = N->getConstantOperandVal(3);
-      DecodeINSERTQIMask(VT, BitLen, BitIdx, Mask);
+      DecodeINSERTQIMask(NumElems, VT.getScalarSizeInBits(), BitLen, BitIdx,
+                         Mask);
       IsUnary = IsFakeUnary = N->getOperand(0) == N->getOperand(1);
     }
     break;
   case X86ISD::UNPCKH:
     assert(N->getOperand(0).getValueType() == VT && "Unexpected value type");
     assert(N->getOperand(1).getValueType() == VT && "Unexpected value type");
-    DecodeUNPCKHMask(VT, Mask);
+    DecodeUNPCKHMask(NumElems, VT.getScalarSizeInBits(), Mask);
     IsUnary = IsFakeUnary = N->getOperand(0) == N->getOperand(1);
     break;
   case X86ISD::UNPCKL:
     assert(N->getOperand(0).getValueType() == VT && "Unexpected value type");
     assert(N->getOperand(1).getValueType() == VT && "Unexpected value type");
-    DecodeUNPCKLMask(VT, Mask);
+    DecodeUNPCKLMask(NumElems, VT.getScalarSizeInBits(), Mask);
     IsUnary = IsFakeUnary = N->getOperand(0) == N->getOperand(1);
     break;
   case X86ISD::MOVHLPS:
     assert(N->getOperand(0).getValueType() == VT && "Unexpected value type");
     assert(N->getOperand(1).getValueType() == VT && "Unexpected value type");
     DecodeMOVHLPSMask(NumElems, Mask);
     IsUnary = IsFakeUnary = N->getOperand(0) == N->getOperand(1);
     break;
   case X86ISD::MOVLHPS:
     assert(N->getOperand(0).getValueType() == VT && "Unexpected value type");
     assert(N->getOperand(1).getValueType() == VT && "Unexpected value type");
     DecodeMOVLHPSMask(NumElems, Mask);
     IsUnary = IsFakeUnary = N->getOperand(0) == N->getOperand(1);
     break;
   case X86ISD::PALIGNR:
     assert(VT.getScalarType() == MVT::i8 && "Byte vector expected");
     assert(N->getOperand(0).getValueType() == VT && "Unexpected value type");
     assert(N->getOperand(1).getValueType() == VT && "Unexpected value type");
     ImmN = N->getOperand(N->getNumOperands()-1);
-    DecodePALIGNRMask(VT, cast<ConstantSDNode>(ImmN)->getZExtValue(), Mask);
+    DecodePALIGNRMask(NumElems, cast<ConstantSDNode>(ImmN)->getZExtValue(),
+                      Mask);
     IsUnary = IsFakeUnary = N->getOperand(0) == N->getOperand(1);
     Ops.push_back(N->getOperand(1));
     Ops.push_back(N->getOperand(0));
     break;
   case X86ISD::VSHLDQ:
     assert(VT.getScalarType() == MVT::i8 && "Byte vector expected");
     assert(N->getOperand(0).getValueType() == VT && "Unexpected value type");
     ImmN = N->getOperand(N->getNumOperands() - 1);
-    DecodePSLLDQMask(VT, cast<ConstantSDNode>(ImmN)->getZExtValue(), Mask);
+    DecodePSLLDQMask(NumElems, cast<ConstantSDNode>(ImmN)->getZExtValue(),
+                     Mask);
     IsUnary = true;
     break;
   case X86ISD::VSRLDQ:
     assert(VT.getScalarType() == MVT::i8 && "Byte vector expected");
     assert(N->getOperand(0).getValueType() == VT && "Unexpected value type");
     ImmN = N->getOperand(N->getNumOperands() - 1);
-    DecodePSRLDQMask(VT, cast<ConstantSDNode>(ImmN)->getZExtValue(), Mask);
+    DecodePSRLDQMask(NumElems, cast<ConstantSDNode>(ImmN)->getZExtValue(),
+                     Mask);
     IsUnary = true;
     break;
   case X86ISD::PSHUFD:
   case X86ISD::VPERMILPI:
     assert(N->getOperand(0).getValueType() == VT && "Unexpected value type");
     ImmN = N->getOperand(N->getNumOperands()-1);
-    DecodePSHUFMask(VT, cast<ConstantSDNode>(ImmN)->getZExtValue(), Mask);
+    DecodePSHUFMask(NumElems, VT.getScalarSizeInBits(),
+                    cast<ConstantSDNode>(ImmN)->getZExtValue(), Mask);
     IsUnary = true;
     break;
   case X86ISD::PSHUFHW:
     assert(N->getOperand(0).getValueType() == VT && "Unexpected value type");
     ImmN = N->getOperand(N->getNumOperands()-1);
-    DecodePSHUFHWMask(VT, cast<ConstantSDNode>(ImmN)->getZExtValue(), Mask);
+    DecodePSHUFHWMask(NumElems, cast<ConstantSDNode>(ImmN)->getZExtValue(),
+                      Mask);
     IsUnary = true;
     break;
   case X86ISD::PSHUFLW:
     assert(N->getOperand(0).getValueType() == VT && "Unexpected value type");
     ImmN = N->getOperand(N->getNumOperands()-1);
-    DecodePSHUFLWMask(VT, cast<ConstantSDNode>(ImmN)->getZExtValue(), Mask);
+    DecodePSHUFLWMask(NumElems, cast<ConstantSDNode>(ImmN)->getZExtValue(),
+                      Mask);
     IsUnary = true;
     break;
   case X86ISD::VZEXT_MOVL:
     assert(N->getOperand(0).getValueType() == VT && "Unexpected value type");
-    DecodeZeroMoveLowMask(VT, Mask);
+    DecodeZeroMoveLowMask(NumElems, Mask);
     IsUnary = true;
     break;
   case X86ISD::VBROADCAST: {
     SDValue N0 = N->getOperand(0);
     // See if we're broadcasting from index 0 of an EXTRACT_SUBVECTOR. If so,
     // add the pre-extracted value to the Ops vector.
     if (N0.getOpcode() == ISD::EXTRACT_SUBVECTOR &&
         N0.getOperand(0).getValueType() == VT &&
         N0.getConstantOperandVal(1) == 0)
       Ops.push_back(N0.getOperand(0));
 
     // We only decode broadcasts of same-sized vectors, unless the broadcast
     // came from an extract from the original width. If we found one, we
     // pushed it the Ops vector above.
     if (N0.getValueType() == VT || !Ops.empty()) {
-      DecodeVectorBroadcast(VT, Mask);
+      DecodeVectorBroadcast(NumElems, Mask);
       IsUnary = true;
       break;
     }
     return false;
   }
   case X86ISD::VPERMILPV: {
     assert(N->getOperand(0).getValueType() == VT && "Unexpected value type");
     IsUnary = true;
     SDValue MaskNode = N->getOperand(1);
     unsigned MaskEltSize = VT.getScalarSizeInBits();
     SmallVector<uint64_t, 32> RawMask;
     if (getTargetShuffleMaskIndices(MaskNode, MaskEltSize, RawMask)) {
-      DecodeVPERMILPMask(VT, RawMask, Mask);
+      DecodeVPERMILPMask(NumElems, VT.getScalarSizeInBits(), RawMask, Mask);
       break;
     }
     if (auto *C = getTargetConstantFromNode(MaskNode)) {
       DecodeVPERMILPMask(C, MaskEltSize, Mask);
       break;
     }
     return false;
   }
if (auto *C = getTargetConstantFromNode(MaskNode)) {
       DecodePSHUFBMask(C, Mask);
       break;
     }
     return false;
   }
   case X86ISD::VPERMI:
     assert(N->getOperand(0).getValueType() == VT && "Unexpected value type");
     ImmN = N->getOperand(N->getNumOperands()-1);
-    DecodeVPERMMask(VT, cast<ConstantSDNode>(ImmN)->getZExtValue(), Mask);
+    DecodeVPERMMask(NumElems, cast<ConstantSDNode>(ImmN)->getZExtValue(), Mask);
     IsUnary = true;
     break;
   case X86ISD::MOVSS:
   case X86ISD::MOVSD:
     assert(N->getOperand(0).getValueType() == VT && "Unexpected value type");
     assert(N->getOperand(1).getValueType() == VT && "Unexpected value type");
-    DecodeScalarMoveMask(VT, /* IsLoad */ false, Mask);
+    DecodeScalarMoveMask(NumElems, /* IsLoad */ false, Mask);
     break;
   case X86ISD::VPERM2X128:
     assert(N->getOperand(0).getValueType() == VT && "Unexpected value type");
     assert(N->getOperand(1).getValueType() == VT && "Unexpected value type");
     ImmN = N->getOperand(N->getNumOperands()-1);
-    DecodeVPERM2X128Mask(VT, cast<ConstantSDNode>(ImmN)->getZExtValue(), Mask);
+    DecodeVPERM2X128Mask(NumElems, cast<ConstantSDNode>(ImmN)->getZExtValue(),
+                         Mask);
     IsUnary = IsFakeUnary = N->getOperand(0) == N->getOperand(1);
     break;
   case X86ISD::MOVSLDUP:
     assert(N->getOperand(0).getValueType() == VT && "Unexpected value type");
-    DecodeMOVSLDUPMask(VT, Mask);
+    DecodeMOVSLDUPMask(NumElems, Mask);
     IsUnary = true;
     break;
   case X86ISD::MOVSHDUP:
     assert(N->getOperand(0).getValueType() == VT && "Unexpected value type");
-    DecodeMOVSHDUPMask(VT, Mask);
+    DecodeMOVSHDUPMask(NumElems, Mask);
     IsUnary = true;
     break;
   case X86ISD::MOVDDUP:
     assert(N->getOperand(0).getValueType() == VT && "Unexpected value type");
-    DecodeMOVDDUPMask(VT, Mask);
+    DecodeMOVDDUPMask(NumElems, Mask);
     IsUnary = true;
     break;
   case X86ISD::MOVLPD:
   case X86ISD::MOVLPS:
     // Not yet implemented
     return false;
   case X86ISD::VPERMIL2: {
     assert(N->getOperand(0).getValueType() == VT && "Unexpected value type");
     assert(N->getOperand(1).getValueType() == VT && "Unexpected value type");
     IsUnary = IsFakeUnary = N->getOperand(0) == N->getOperand(1);
     unsigned MaskEltSize = VT.getScalarSizeInBits();
     SDValue MaskNode = N->getOperand(2);
     SDValue CtrlNode = N->getOperand(3);
     if (ConstantSDNode *CtrlOp = dyn_cast<ConstantSDNode>(CtrlNode)) {
       unsigned CtrlImm = CtrlOp->getZExtValue();
       SmallVector<uint64_t, 32> RawMask;
       if (getTargetShuffleMaskIndices(MaskNode, MaskEltSize, RawMask)) {
-        DecodeVPERMIL2PMask(VT, CtrlImm, RawMask, Mask);
+        DecodeVPERMIL2PMask(NumElems, VT.getScalarSizeInBits(), CtrlImm,
+                            RawMask, Mask);
         break;
       }
       if (auto *C = getTargetConstantFromNode(MaskNode)) {
         DecodeVPERMIL2PMask(C, CtrlImm, MaskEltSize, Mask);
         break;
       }
     }
     return false;
static bool getFauxShuffleMask(SDValue N, SmallVectorImpl<int> &Mask,
   }
   case ISD::ZERO_EXTEND_VECTOR_INREG:
   case X86ISD::VZEXT: {
     // TODO - add support for VPMOVZX with smaller input vector types.
     SDValue Src = N.getOperand(0);
     MVT SrcVT = Src.getSimpleValueType();
     if (NumSizeInBits != SrcVT.getSizeInBits())
       break;
-    DecodeZeroExtendMask(SrcVT.getScalarType(), VT, Mask);
+    DecodeZeroExtendMask(SrcVT.getScalarSizeInBits(), VT.getScalarSizeInBits(),
+                         VT.getVectorNumElements(), Mask);
     Ops.push_back(Src);
     return true;
   }
   }
 
   return false;
 }