JIT: Switch remainder of lowering to new ABI info

[jakobbotsch]

This required introducing some new information in ABIPassingSegment for stack-passed segments about whether they own the full stack slot or not. Apple arm64 notoriously has cases where multiple parameters are packed in the same stack slot.

[dotnet-policy-service]

Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch
See info in area-owners.md if you want to be subscribed.

[jakobbotsch]

I decided to remove this; the backend is already be doing similar validation for IR shapes depending on the ABI information.

[jakobbotsch]

PTAL @dotnet/jit-contrib

No diffs. Some MinOpts TP regressions in the 0.1-0.2% range (and some larger outliers in collections with very few MinOpts contexts).

Detailed TP diff for MinOpts contexts only looks like:

Base: 198188647104, Diff: 198652714462, +0.2342%

503226159  : NA      : 49.19% : +0.2539% : private: struct GenTree * __cdecl Lowering::NewPutArg(struct GenTreeCall *, struct GenTree *, class CallArg *, enum var_types)
149634454  : NA      : 14.63% : +0.0755% : private: unsigned int __cdecl StackLevelSetter::PopArgumentsFromCall(struct GenTreeCall *)                                    
73920858   : +4.24%  : 7.23%  : +0.0373% : public: void __cdecl CallArgs::AddFinalArgsAndDetermineABIInfo(class Compiler *, struct GenTreeCall *)                        
10015964   : +2.21%  : 0.98%  : +0.0051% : protected: void __cdecl CodeGen::genCallPlaceRegArgs(struct GenTreeCall *)                                                    
6685913    : +8.72%  : 0.65%  : +0.0034% : public: void __cdecl Compiler::lvaClassifyParameterABI<class WinX64Classifier>(class WinX64Classifier &)                      
-22497456  : -2.32%  : 2.20%  : -0.0114% : private: void __cdecl StackLevelSetter::ProcessBlock(struct BasicBlock *)                                                     
-256862005 : -33.27% : 25.11% : -0.1296% : private: void __cdecl Lowering::LowerArg(struct GenTreeCall *, class CallArg *, bool)                                         

MSVC no longer inlines NewPutArg, so hopefully some of that will be resolved by PGO data. Also #112561 is paying off all of this cost and more (and will also make PopArgumentsFromCall completely disappear outside x86).

[amanasifkhalid]

LGTM, thanks!

[amanasifkhalid]

Suggested change

	// stack, and which does not gets its own full stack slot.
	// stack, and which does not get its own full stack slot.

[jakobbotsch]

I will fix this as part of a follow up