1364     // We directly match byte blends in the backend as they match the VSELECT  in X86TargetLowering()
9378 // designed to handle arbitrary vector shuffles and blends, gracefully
10668     assert(Subtarget.hasAVX2() && "256-bit integer blends require AVX2!");  in lowerShuffleAsBlend()
10672     assert(Subtarget.hasAVX() && "256-bit float blends require AVX!");  in lowerShuffleAsBlend()
10679     assert(Subtarget.hasSSE41() && "128-bit blends require SSE41!");  in lowerShuffleAsBlend()
10683     assert(Subtarget.hasAVX2() && "v16i16 blends require AVX2!");  in lowerShuffleAsBlend()
10712     assert(Subtarget.hasAVX2() && "256-bit byte-blends require AVX2!");  in lowerShuffleAsBlend()
10715     assert(Subtarget.hasSSE41() && "128-bit byte-blends require SSE41!");  in lowerShuffleAsBlend()
10830   // If only immediate blends, then bail if the blend mask can't be widened to  in lowerShuffleAsBlendAndPermute()
11150 /// blends and permutes.
11155 /// blends. For vXi8/vXi16 shuffles we may use unpack instead of blend.
11220     // Only prefer immediate blends to unpack/rotate.  in lowerShuffleAsDecomposedShuffleMerge()
11238     // Unpack/rotate failed - try again with variable blends.  in lowerShuffleAsDecomposedShuffleMerge()
12048 /// can to emit an efficient lowering. It handles both blends with all-zero
12871 /// the integer unit to minimize domain crossing penalties. However, for blends
12953   // If we have direct support for blends, we should lower by decomposing into  in lowerV2I64Shuffle()
13128   // There are special ways we can lower some single-element blends. However, we  in lowerV4F32Shuffle()
13129   // have custom ways we can lower more complex single-element blends below that  in lowerV4F32Shuffle()
13169 /// blends we use the floating point domain blend instructions.
13233   // There are special ways we can lower some single-element blends.  in lowerV4I32Shuffle()
13272     // If we have direct support for blends, we should lower by decomposing into  in lowerV4I32Shuffle()
13858 /// This handles both single-input shuffles and combined shuffle/blends with
13862 /// The blends are lowered in one of three fundamental ways. If there are few
13942   // There are special ways we can lower some single-element blends.  in lowerV8I16Shuffle()
14066   // decompose into single-input permutes and blends/unpacks.  in lowerV8I16Shuffle()
14145 /// the existing lowering for v8i16 blends on each half, finally PACK-ing them
14323   // blends but after all of the single-input lowerings. If the single input  in lowerV16I8Shuffle()
14334   // FIXME: The only exceptions to the above are blends which are exact  in lowerV16I8Shuffle()
14345     // do so. This avoids using them to handle blends-with-zero which is  in lowerV16I8Shuffle()
14386   // There are special ways we can lower some single-element blends.  in lowerV16I8Shuffle()
14568   // Now create two 4-way blends of these half-width vectors.  in splitAndLowerShuffle()
14987   // Blends are faster and handle all the non-lane-crossing cases.  in lowerV2X128Shuffle()
16112     // have already handled any direct blends.  in lowerV8F32Shuffle()
17767   // shuffles and re-use the shuffle lowering path for blends.  in lowerVSELECTtoVectorShuffle()
17810   // Variable blends are only legal from SSE4.1 onward.  in LowerVSELECT()
17867     // Most of the vector types have blends past SSE4.1.  in LowerVSELECT()
17871     // The byte blends for AVX vectors were introduced only in AVX2.  in LowerVSELECT()
18390         // than an insertps. Blends are simpler operations in hardware and so  in LowerINSERT_VECTOR_ELT()
38693     // Prefer blends for sequential shuffles unless we are optimizing for size.  in combineX86ShuffleChain()
38712       // Prefer blends to X86ISD::VPERM2X128.  in combineX86ShuffleChain()
41759   // If we have legalized the vector types, look for blends of FADD and FSUB  in combineShuffle()
45665 /// condition so that blends can use the high (sign) bit of each element.
45694   // FIXME: We don't support i16-element blends currently. We could and  in combineVSelectToBLENDV()
45702   // Byte blends are only available in AVX2  in combineVSelectToBLENDV()