[IR]Add NumSrcElts param to is..Mask static function in ShuffleVectorInst.

[ABataev]

Summary

Need to add NumSrcElts param to is..Mask functions in
ShuffleVectorInstruction class for better mask analysis. Mask.size() not
always matches the sizes of the permuted vector(s). Allows to better
estimate the cost in SLP and fix uses of the functions in other cases.

[craig.topper]

llvm/lib/IR/Instructions.cpp

2178–2179

If Mask.size() == 4 and NumSrcElts == 6. Then this considers <5, 4, 3, 2> as a reverse?

[ABataev]

llvm/lib/IR/Instructions.cpp

2178–2179

Yes. Need to add a check that the operation does not change the size, just like non-static isReverse() does.

[ABataev]

Address comments

[RKSimon]

llvm/include/llvm/IR/Instructions.h

2070–2071

Update this comment

2165–2166

Update comment

2186–2187

Update comment

2259

Update comment

llvm/unittests/IR/ShuffleVectorInstTest.cpp

14

All of these tests need support for length changing shuffles

[ABataev]

Address comments

[ABataev]

Ping!

[craig.topper]

llvm/lib/IR/Instructions.cpp

2158

Why is the argument an int if we're going to cast it to unsigned?

[ABataev]

llvm/lib/IR/Instructions.cpp

2158

For conformance. All other functions have corresponding int parameter, though treat it as unsigned.

[ABataev]

Rebase, ping!

[ABataev]

Rebase, ping!

[RKSimon]

No objections to this, but I do think it'd make sense to split into 2 patches - (1) adding the NumSrcElts argument to the shuffle matching helpers + additional unit test coverage and (2) using it in SLP

[ABataev]

Rebase, address comments

[RKSimon]

llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp

6658

Maybe split this into a series of separate if () { return true/false; } to make it easier to grok?

[ABataev]

Address comment

[ABataev]

llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp

6658

Done

[RKSimon]

LGTM - cheers