Rollup merge of #137256 - workingjubilee:untangle-vector-abi-assumptions, r=bjorn3,RalfJung

compiler: untangle SIMD alignment assumptions

There were a number of puzzling assumptions being made about SIMD types and their layout that I have corrected in this diff. These are mostly no-op edits in actual fact, but they do subtly alter a pair of checks in our invariant-checking and union layout computation that rested on those peculiar assumptions. Those unfortunately stand in the way of any further actual fixes. I submit this for review, even though it's not clearly motivated without its followups, because it should still be possible to independently conclude whether this is correct.

Author: matthiaskrgr

Diff for: crates/hir-ty/src/layout.rs

@@ -179,7 +179,7 @@ fn layout_of_simd_ty(
         .size
         .checked_mul(e_len, dl)
         .ok_or(LayoutError::BadCalc(LayoutCalculatorError::SizeOverflow))?;
-    let align = dl.vector_align(size);
+    let align = dl.llvmlike_vector_align(size);
     let size = size.align_to(align.abi);
 
     // Compute the placement of the vector fields: