[Alignment] Change implementation of TargetCallingConv::OrigAlign

[gchatelet]

Summary

This patch makes sure that getOrigAlign() cannot return 0.
It now stores the OrigAlign value as a Log2 instead of Log2 + 1 (see Alignment.h encode and decodeMaybeAlign).
This change is not NFC but all the tests passes.

While reviewing the references to TargetCallingConv::getOrigAlign I stumbled upon this call which seems dubious:

• https://github.com/llvm/llvm-project/blob/2bf01dcbaa6723c9c41f8d6005a1f69818ddbd23/llvm/lib/Target/ARM/ARMCallingConv.cpp#L188
• https://github.com/llvm/llvm-project/blob/2bf01dcbaa6723c9c41f8d6005a1f69818ddbd23/llvm/include/llvm/CodeGen/CallingConvLower.h#L128
• https://github.com/llvm/llvm-project/blob/2bf01dcbaa6723c9c41f8d6005a1f69818ddbd23/llvm/include/llvm/CodeGen/CallingConvLower.h#L84

It looks like getOrigAlign (which is an alignment) is stored in CCValAssign::Loc (which is an offset or a Reg).
I traced it back to e95c5b3236cf @t.p.northover do you mind having a look?

This is patch is part of a series to introduce an Alignment type.
See this thread for context: http://lists.llvm.org/pipermail/llvm-dev/2019-July/133851.html
See this patch for the introduction of the type: https://reviews.llvm.org/D64790

[gchatelet]

It is possible that the current implementation reserves the zero value as an error, although it is currently not checked.
Maybe it is illegal to call getOrigAlign before setOrigAlign?

If this is the intended behavior then some code is broken, adding an assert(A && "OrigAlign must be set"); in getOrigAlignmakes the following tests fail:

LLVM :: CodeGen/ARM/2013-05-13-DAGCombiner-undef-mask.ll
LLVM :: CodeGen/ARM/GlobalISel/arm-unsupported.ll
LLVM :: CodeGen/ARM/arm-vld1.ll
LLVM :: CodeGen/ARM/arm-vlddup.ll
LLVM :: CodeGen/ARM/fp16-instructions.ll
LLVM :: CodeGen/ARM/fp16-no-condition.ll
LLVM :: CodeGen/ARM/fp16-vminmaxnm-safe.ll
LLVM :: CodeGen/ARM/fp16-vminmaxnm.ll
LLVM :: CodeGen/ARM/inc-of-add.ll
LLVM :: CodeGen/ARM/sub-of-not.ll
LLVM :: CodeGen/ARM/swift-return.ll
LLVM :: CodeGen/ARM/umulo-128-legalisation-lowering.ll
LLVM :: CodeGen/ARM/vld-vst-upgrade.ll
LLVM :: CodeGen/ARM/vldm-sched-a9.ll
LLVM :: CodeGen/ARM/vtrn.ll
LLVM :: CodeGen/ARM/vuzp.ll
LLVM :: CodeGen/ARM/vzip.ll
LLVM :: CodeGen/Mips/cconv/pr33883.ll
LLVM :: CodeGen/Mips/cconv/return-struct.ll
LLVM :: CodeGen/Mips/cconv/vector.ll
LLVM :: CodeGen/Mips/fp16-promote.ll
LLVM :: CodeGen/Mips/return-vector.ll
LLVM :: CodeGen/Thumb/umulo-128-legalisation-lowering.ll
LLVM :: CodeGen/Thumb2/mve-shuffle.ll
LLVM :: CodeGen/Thumb2/umulo-128-legalisation-lowering.ll

[gchatelet]

A gentle ping @t.p.northover

[gchatelet]

This will be implemented in D82307

llvm/include/llvm/CodeGen/TargetCallingConv.h

private:
     unsigned IsInAlloca : 1;   ///< Passed with inalloca
     unsigned IsSplitEnd : 1;   ///< Last part of a split
     unsigned IsSwiftSelf : 1;  ///< Swift self parameter
     unsigned IsSwiftError : 1; ///< Swift error parameter
     unsigned IsHva : 1;        ///< HVA field for
     unsigned IsHvaStart : 1;   ///< HVA structure start
     unsigned IsSecArgPass : 1; ///< Second argument
     unsigned ByValAlign : 4;   ///< Log 2 of byval alignment
-    unsigned OrigAlign : 5;    ///< Log 2 of original alignment
+    unsigned OrigLogAlign : 5; ///< Log 2 of original alignment
     unsigned IsInConsecutiveRegsLast : 1;
     unsigned IsInConsecutiveRegs : 1;
     unsigned IsCopyElisionCandidate : 1; ///< Argument copy elision candidate
     unsigned IsPointer : 1;
 
     unsigned ByValSize; ///< Byval struct size
 
     unsigned PointerAddrSpace; ///< Address space of pointer argument
 
   public:
     ArgFlagsTy()
         : IsZExt(0), IsSExt(0), IsInReg(0), IsSRet(0), IsByVal(0), IsNest(0),
           IsReturned(0), IsSplit(0), IsInAlloca(0), IsSplitEnd(0),
           IsSwiftSelf(0), IsSwiftError(0), IsHva(0), IsHvaStart(0),
-          IsSecArgPass(0), ByValAlign(0), OrigAlign(0),
+          IsSecArgPass(0), ByValAlign(0), OrigLogAlign(0),
           IsInConsecutiveRegsLast(0), IsInConsecutiveRegs(0),
           IsCopyElisionCandidate(0), IsPointer(0), ByValSize(0),
           PointerAddrSpace(0) {
       static_assert(sizeof(*this) == 3 * sizeof(unsigned), "flags are too big");
     }
 
     bool isZExt() const { return IsZExt; }
     void setZExt() { IsZExt = 1; }
unsigned getByValAlign() const {
       return A ? A->value() : 0;
     }
     void setByValAlign(unsigned A) {
       ByValAlign = encode(Align(A));
       assert(getByValAlign() == A && "bitfield overflow");
     }
 
     unsigned getOrigAlign() const {
-      MaybeAlign A = decodeMaybeAlign(OrigAlign);
-      return A ? A->value() : 0;
+      const Align A(1ULL << OrigLogAlign);
+      return A.value();
     }
     void setOrigAlign(Align A) {
-      OrigAlign = encode(A);
+      OrigLogAlign = Log2(A);
       assert(getOrigAlign() == A.value() && "bitfield overflow");
     }
 
     unsigned getByValSize() const { return ByValSize; }
     void setByValSize(unsigned S) { ByValSize = S; }
 
     unsigned getPointerAddrSpace() const { return PointerAddrSpace; }
     void setPointerAddrSpace(unsigned AS) { PointerAddrSpace = AS; }