[clang] [HLSL][Sema] Fix Struct Size Calculation containing 16/32 bit scalars (PR #128086)

[Ashley Coleman via cfe-commits]

================
@@ -172,6 +172,27 @@ Decl *SemaHLSL::ActOnStartBuffer(Scope *BufferScope, bool CBuffer,
   return Result;
 }
 
+static unsigned calculateLegacyCbufferFieldAlign(const ASTContext &Context,
+                                                 QualType T) {
+  // Aggregate types are always aligned to new buffer rows
+  if (T->isAggregateType())
+    return 16;
+
+  assert(Context.getTypeSize(T) <= 64 &&
+         "Scalar bit widths larger than 64 not supported");
+
+  // 64 bit types such as double and uint64_t align to 8 bytes
+  if (Context.getTypeSize(T) == 64)
+    return 8;
+
+  // Half types align to 2 bytes only if native half is available
+  if (T->isHalfType() && Context.getLangOpts().NativeHalfType)
----------------
V-FEXrt wrote:

True, I can probably make this function a lot simpler by just looking at the bitwidth.  Are there any types where the bitwidth is 32 but the alignment is not 4 bytes? 

I think I read somewhere that "everything is 4 byte align except these X exceptions" which is a weird way to write it unless some non-32 bitwide type is also 4 byte aligned. Otherwise its clearer to say "everything is aligned to its bytewidth except these X exceptions"

https://github.com/llvm/llvm-project/pull/128086