Replaced zeroinitializer to poison in second operand for shufflevector.

Moved tests to <2 x > types.

Updated tests.

Changing tests.

Changed the second test case with suggestions from comment.

Rebased, Resolved comments.

Ping!

Removed double not in AArch64TargetLowering::preferScalarizeSplat() function.

Addressed remarks.

Adjust according to comments, fixed error with the same mask but different type size by forbid to delete the store.

Rebasing.

Gentle ping.

Replaced to TypeSize::isKnownLE() usage for all constant true case. Added few tests.

Restoring the previous revision.

Combined two ifs

Fixed typo in test funtion name.

Addressed comments.

I found an error in my implementation: we could not remove the store if the chained store is a fixed type and the store we consider to remove is a scalable type, since we don't know scalable type size in the runtime. fixed.

Resolved comments.

Use TypeSize::isKnownLE only if base pointers are equal.

Addressed remakrs, using now TypeSize::isKnownLE.

Dropped (STBitSize <= ChainBitSize) checks, it is not required as STBase.contains() ignores those to handle.

Fixed case where ST1 type size is bigger than in ST.

Missed to update sve-redundant-store.ll test after change, fixed.

Renamed two functions in sve-splat-one-and-ptrue.ll

Addressed comments.

Added combine for AArch64ISD::REINTERPRET_CAST nodes, addressed remarks.

Addressed coments.

Addressed comments.

Changed proposed test in order to remove -O3 out of llc parameters.

Removed unnecessary cast to CallInst.

Addressed comments.

Rebased, Addressed remarks.

Fixed remarks.

Fixed remarks.

Removed the Multiplier functionality. If a type is not representable on hardware then allow the legilizer to deduce the cost and in case a type is representable on hardware then assume basic SVE cost with 128-bit vector operation.

Missed two comments, addressed those as well.

Addressed suggestions.

Added overflow check for the increment.

Added tests for signed/unsigned overflow.

Addressed remarks, avoided to check overflow while adding one to the range for equal or same conditions.

Addressed comments, added signed overflow handling while subtraction/addintion.

Unified handling of WHILEop.

Avoid to look at ConvertCostTableLookup table after switching to SVE registers, but call to AArch64TTIImpl::getCastInstrCost() recursively instead.

I found an error in the previous revision where we estimated cost for SVE based operation using NEON cost, added tests for target without NEON operations and SVE 128-bit size registers.

Fixed typo.

Fixed error in converting from fixed to a scalable type by assuming to represent fixed vector type to SVE chunk 128-bit register.

Addressed remarks.

Corrected calculation for a number of elements in a vector for SVE type selection for cost approximation. No cost changes in test/Analysis/CostModel/AArch64/cast.ll with this revision against the previous.

Add support for any range not just from 0.

Rebased, addressed comments.

Rebased, further improved cost estimation.

Fixed remarks.