Auto merge of #95837 - scottmcm:ptr-offset-from-unsigned, r=oli-obk

Add `sub_ptr` on pointers (the `usize` version of `offset_from`)

We have `add`/`sub` which are the `usize` versions of `offset`, this adds the `usize` equivalent of `offset_from`.  Like how `.add(d)` replaced a whole bunch of `.offset(d as isize)`, you can see from the changes here that it's fairly common that code actually knows the order between the pointers and *wants* a `usize`, not an `isize`.

As a bonus, this can do `sub nuw`+`udiv exact`, rather than `sub`+`sdiv exact`, which can be optimized slightly better because it doesn't have to worry about negatives.  That's why the slice iterators weren't using `offset_from`, though I haven't updated that code in this PR because slices are so perf-critical that I'll do it as its own change.

This is an intrinsic, like `offset_from`, so that it can eventually be allowed in CTFE.  It also allows checking the extra safety condition -- see the test confirming that CTFE catches it if you pass the pointers in the wrong order.

1 files changed, 14 insertions, 7 deletions

diff --git a/compiler/rustc_codegen_ssa/src/mir/intrinsic.rs b/compiler/rustc_codegen_ssa/src/mir/intrinsic.rs
index f15c469ae57..6d6d3ae01f4 100644
--- a/compiler/rustc_codegen_ssa/src/mir/intrinsic.rs
+++ b/compiler/rustc_codegen_ssa/src/mir/intrinsic.rs
@@ -555,21 +555,28 @@ impl<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> {
                 }
             }
 
-            sym::ptr_offset_from => {
+            sym::ptr_offset_from | sym::ptr_offset_from_unsigned => {
                 let ty = substs.type_at(0);
                 let pointee_size = bx.layout_of(ty).size;
 
-                // This is the same sequence that Clang emits for pointer subtraction.
-                // It can be neither `nsw` nor `nuw` because the input is treated as
-                // unsigned but then the output is treated as signed, so neither works.
                 let a = args[0].immediate();
                 let b = args[1].immediate();
                 let a = bx.ptrtoint(a, bx.type_isize());
                 let b = bx.ptrtoint(b, bx.type_isize());
-                let d = bx.sub(a, b);
                 let pointee_size = bx.const_usize(pointee_size.bytes());
-                // this is where the signed magic happens (notice the `s` in `exactsdiv`)
-                bx.exactsdiv(d, pointee_size)
+                if name == sym::ptr_offset_from {
+                    // This is the same sequence that Clang emits for pointer subtraction.
+                    // It can be neither `nsw` nor `nuw` because the input is treated as
+                    // unsigned but then the output is treated as signed, so neither works.
+                    let d = bx.sub(a, b);
+                    // this is where the signed magic happens (notice the `s` in `exactsdiv`)
+                    bx.exactsdiv(d, pointee_size)
+                } else {
+                    // The `_unsigned` version knows the relative ordering of the pointers,
+                    // so can use `sub nuw` and `udiv exact` instead of dealing in signed.
+                    let d = bx.unchecked_usub(a, b);
+                    bx.exactudiv(d, pointee_size)
+                }
             }
 
             _ => {