+ 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
+ }
+ }
+}
+
+/// \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,
+ SmallVectorImpl<int> &ShuffleMask) {
+ unsigned NumLanes = VT.getSizeInBits() / 128;
+ unsigned NumElementsInLane = 128 / VT.getScalarSizeInBits();
+ 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,
+ SmallVectorImpl<int> &ShuffleMask) {
+ unsigned HalfSize = VT.getVectorNumElements() / 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);
+ }
+}
+
+void DecodePSHUFBMask(const Constant *C, SmallVectorImpl<int> &ShuffleMask) {
+ Type *MaskTy = C->getType();
+ // It is not an error for the PSHUFB mask to not be a vector of i8 because the
+ // constant pool uniques constants by their bit representation.
+ // e.g. the following take up the same space in the constant pool:
+ // i128 -170141183420855150465331762880109871104
+ //
+ // <2 x i64> <i64 -9223372034707292160, i64 -9223372034707292160>
+ //
+ // <4 x i32> <i32 -2147483648, i32 -2147483648,
+ // i32 -2147483648, i32 -2147483648>
+
+ unsigned MaskTySize = MaskTy->getPrimitiveSizeInBits();
+ assert(MaskTySize == 128 || MaskTySize == 256 || MaskTySize == 512);
+
+ // This is a straightforward byte vector.
+ if (MaskTy->isVectorTy() && MaskTy->getVectorElementType()->isIntegerTy(8)) {
+ int NumElements = MaskTy->getVectorNumElements();
+ ShuffleMask.reserve(NumElements);
+
+ for (int i = 0; i < NumElements; ++i) {
+ // For AVX vectors with 32 bytes the base of the shuffle is the 16-byte
+ // lane of the vector we're inside.
+ int Base = i & ~0xf;
+ Constant *COp = C->getAggregateElement(i);
+ if (!COp) {
+ ShuffleMask.clear();
+ return;
+ } else if (isa<UndefValue>(COp)) {
+ ShuffleMask.push_back(SM_SentinelUndef);
+ continue;
+ }
+ uint64_t Element = cast<ConstantInt>(COp)->getZExtValue();
+ // If the high bit (7) of the byte is set, the element is zeroed.
+ if (Element & (1 << 7))
+ ShuffleMask.push_back(SM_SentinelZero);
+ else {
+ // Only the least significant 4 bits of the byte are used.
+ int Index = Base + (Element & 0xf);
+ ShuffleMask.push_back(Index);
+ }
+ }
+ }
+ // TODO: Handle funny-looking vectors too.
+}
+
+void DecodePSHUFBMask(ArrayRef<uint64_t> RawMask,
+ SmallVectorImpl<int> &ShuffleMask) {
+ for (int i = 0, e = RawMask.size(); i < e; ++i) {
+ uint64_t M = RawMask[i];
+ if (M == (uint64_t)SM_SentinelUndef) {
+ ShuffleMask.push_back(M);
+ continue;
+ }
+ // For AVX vectors with 32 bytes the base of the shuffle is the half of
+ // the vector we're inside.
+ int Base = i < 16 ? 0 : 16;
+ // If the high bit (7) of the byte is set, the element is zeroed.
+ if (M & (1 << 7))
+ ShuffleMask.push_back(SM_SentinelZero);
+ 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) {
+ // 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);
+ }
+}
+
+/// DecodeVPERMMask - this decodes the shuffle masks for VPERMQ/VPERMPD.
+/// No VT provided since it only works on 256-bit, 4 element vectors.
+void DecodeVPERMMask(unsigned Imm, SmallVectorImpl<int> &ShuffleMask) {
+ for (unsigned i = 0; i != 4; ++i) {
+ ShuffleMask.push_back((Imm >> (2 * i)) & 3);
+ }
+}
+
+void DecodeVPERMILPMask(const Constant *C, unsigned ElSize,
+ SmallVectorImpl<int> &ShuffleMask) {
+ Type *MaskTy = C->getType();
+ // It is not an error for the PSHUFB mask to not be a vector of i8 because the
+ // constant pool uniques constants by their bit representation.
+ // e.g. the following take up the same space in the constant pool:
+ // i128 -170141183420855150465331762880109871104
+ //
+ // <2 x i64> <i64 -9223372034707292160, i64 -9223372034707292160>
+ //
+ // <4 x i32> <i32 -2147483648, i32 -2147483648,
+ // i32 -2147483648, i32 -2147483648>
+
+ unsigned MaskTySize = MaskTy->getPrimitiveSizeInBits();
+
+ if (MaskTySize != 128 && MaskTySize != 256) // FIXME: Add support for AVX-512.
+ return;
+
+ // Only support vector types.
+ if (!MaskTy->isVectorTy())
+ return;
+
+ // Make sure its an integer type.
+ Type *VecEltTy = MaskTy->getVectorElementType();
+ if (!VecEltTy->isIntegerTy())
+ return;
+
+ // Support any element type from byte up to element size.
+ // This is necesary primarily because 64-bit elements get split to 32-bit
+ // in the constant pool on 32-bit target.
+ unsigned EltTySize = VecEltTy->getIntegerBitWidth();
+ if (EltTySize < 8 || EltTySize > ElSize)
+ return;
+
+ unsigned NumElements = MaskTySize / ElSize;
+ assert((NumElements == 2 || NumElements == 4 || NumElements == 8) &&
+ "Unexpected number of vector elements.");
+ ShuffleMask.reserve(NumElements);
+ unsigned NumElementsPerLane = 128 / ElSize;
+ unsigned Factor = ElSize / EltTySize;
+
+ for (unsigned i = 0; i < NumElements; ++i) {
+ Constant *COp = C->getAggregateElement(i * Factor);
+ if (!COp) {
+ ShuffleMask.clear();
+ return;
+ } else if (isa<UndefValue>(COp)) {
+ ShuffleMask.push_back(SM_SentinelUndef);
+ continue;