perf(codegen): Eliminate size_of_val == 0 for DSTs with Non-zero-sized Prefix via NUW and Assume

compiler/rustc_codegen_ssa/src/size_of_val.rs

@@ -159,7 +159,8 @@ pub fn size_and_align_of_dst<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>>(
             // Furthermore, `align >= unsized_align`, and therefore we only need to do:
             // let full_size = (unsized_offset_unadjusted + unsized_size).align_to(full_align);
 
-            let full_size = bx.add(unsized_offset_unadjusted, unsized_size);
+            // total <= isize::MAX, so nuw+nsw.
+            let unrounded_size = bx.unchecked_suadd(unsized_offset_unadjusted, unsized_size);
 
             // Issue #27023: must add any necessary padding to `size`
             // (to make it a multiple of `align`) before returning it.
@@ -173,10 +174,15 @@ pub fn size_and_align_of_dst<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>>(
             //   `(size + (align-1)) & -align`
             let one = bx.const_usize(1);
             let addend = bx.sub(full_align, one);
-            let add = bx.add(full_size, addend);
+            let add = bx.add(unrounded_size, addend);
             let neg = bx.neg(full_align);
             let full_size = bx.and(add, neg);
 
+            // round_up(x, a) >= x for pow2 a; with nuw above LLVM deduces
+            // full_size >= unrounded_size >= offset > 0 (#152788).
+            let size_ge = bx.icmp(IntPredicate::IntUGE, full_size, unrounded_size);
+            bx.assume(size_ge);
+
             (full_size, full_align)
         }
         _ => bug!("size_and_align_of_dst: {t} not supported"),

tests/codegen-llvm/dst-size-of-val-not-zst.rs

@@ -0,0 +1,31 @@
+//@ compile-flags: -Copt-level=3 -Z merge-functions=disabled
+//@ min-llvm-version: 21
+//@ needs-deterministic-layouts
+
+#![crate_type = "lib"]
+
+// Regression test for #152788: `size_of_val(p) == 0` folds to `false` for
+// DSTs with a non-zero prefix (nuw+nsw on offset+tail, assume on rounding).
+
+pub struct Foo<T: ?Sized>(pub [u32; 3], pub T);
+
+// CHECK-LABEL: @size_of_val_dyn_not_zero
+#[no_mangle]
+pub fn size_of_val_dyn_not_zero(p: &Foo<dyn std::fmt::Debug>) -> bool {
+    // CHECK: ret i1 false
+    std::mem::size_of_val(p) == 0
+}
+
+// CHECK-LABEL: @size_of_val_slice_u8_not_zero
+#[no_mangle]
+pub fn size_of_val_slice_u8_not_zero(p: &Foo<[u8]>) -> bool {
+    // CHECK: ret i1 false
+    std::mem::size_of_val(p) == 0
+}
+
+// CHECK-LABEL: @size_of_val_slice_i32_not_zero
+#[no_mangle]
+pub fn size_of_val_slice_i32_not_zero(p: &Foo<[i32]>) -> bool {
+    // CHECK: ret i1 false
+    std::mem::size_of_val(p) == 0
+}

tests/codegen-llvm/dst-vtable-align-nonzero.rs

@@ -1,4 +1,7 @@
 //@ compile-flags: -Copt-level=3 -Z merge-functions=disabled
+//@ revisions: LLVM20 CURRENT
+//@ [LLVM20] max-llvm-major-version: 20
+//@ [CURRENT] min-llvm-version: 21
 
 #![crate_type = "lib"]
 #![feature(core_intrinsics)]
@@ -30,10 +33,18 @@ pub struct Struct<W: ?Sized> {
 pub fn eliminates_runtime_check_when_align_1(
     x: &Struct<WrapperWithAlign1<dyn Trait>>,
 ) -> &WrapperWithAlign1<dyn Trait> {
-    // CHECK: load [[USIZE:i[0-9]+]], {{.+}} !range [[RANGE_META:![0-9]+]]
+    // LLVM20: load [[USIZE:i[0-9]+]], {{.+}} !range {{![0-9]+}}
+    // LLVM20: load [[USIZE]], {{.+}} !range [[RANGE_META:![0-9]+]]
+    // CURRENT: load [[USIZE:i[0-9]+]], {{.+}} !range [[RANGE_META:![0-9]+]]
     // CHECK-NOT: llvm.umax
-    // CHECK-NOT: icmp
     // CHECK-NOT: select
+    // CURRENT-NOT: icmp
+    // LLVM20-NOT: icmp
+    // LLVM20: [[DOES_NOT_SHRINK:%.+]] = icmp ug{{[et]}}
+    // LLVM20-NEXT: call void @llvm.assume(i1 [[DOES_NOT_SHRINK]])
+    // LLVM20-NOT: llvm.umax
+    // LLVM20-NOT: icmp
+    // LLVM20-NOT: select
     // CHECK: ret
     &x.dst
 }