Auto merge of #102750 - the8472:opt-field-order, r=wesleywiser

optimize field ordering by grouping m*2^n-sized fields with equivalently aligned ones

```rust
use std::ptr::addr_of;
use std::mem;

struct Foo {
    word: u32,
    byte: u8,
    ary: [u8; 4]
}

fn main() {
    let foo: Foo = unsafe { mem::zeroed() };

    println!("base: {:p}\nword: {:p}\nbyte: {:p}\nary:  {:p}", &foo, addr_of!(foo.word), addr_of!(foo.byte), addr_of!(foo.ary));
}
```

prints

```
base: 0x7fffc1a8a668
word: 0x7fffc1a8a668
byte: 0x7fffc1a8a66c
ary:  0x7fffc1a8a66d
```

I.e. the `u8` in the middle causes the array to sit at an odd offset, which might prevent optimizations, especially on architectures where unaligned loads are costly.

Note that this will make field ordering niche-dependent, i.e. a `Bar<T>` with `T=char` and `T=u32` may result in different field order, this may break some code that makes invalid assumptions about `repr(Rust)` types.

3 files changed, 5 insertions, 5 deletions

diff --git a/src/test/codegen/issue-37945.rs b/src/test/codegen/issue-37945.rs
index 12fa1e9e56b..fe54375bbf6 100644
--- a/src/test/codegen/issue-37945.rs
+++ b/src/test/codegen/issue-37945.rs
@@ -15,7 +15,7 @@ use std::slice::Iter;
 pub fn is_empty_1(xs: Iter<f32>) -> bool {
 // CHECK-LABEL: @is_empty_1(
 // CHECK-NEXT:  start:
-// CHECK-NEXT:    [[A:%.*]] = icmp ne {{i32\*|ptr}} %xs.1, null
+// CHECK-NEXT:    [[A:%.*]] = icmp ne {{i32\*|ptr}} {{%xs.0|%xs.1}}, null
 // CHECK-NEXT:    tail call void @llvm.assume(i1 [[A]])
 // The order between %xs.0 and %xs.1 on the next line doesn't matter
 // and different LLVM versions produce different order.
@@ -28,7 +28,7 @@ pub fn is_empty_1(xs: Iter<f32>) -> bool {
 pub fn is_empty_2(xs: Iter<f32>) -> bool {
 // CHECK-LABEL: @is_empty_2
 // CHECK-NEXT:  start:
-// CHECK-NEXT:    [[C:%.*]] = icmp ne {{i32\*|ptr}} %xs.1, null
+// CHECK-NEXT:    [[C:%.*]] = icmp ne {{i32\*|ptr}} {{%xs.0|%xs.1}}, null
 // CHECK-NEXT:    tail call void @llvm.assume(i1 [[C]])
 // The order between %xs.0 and %xs.1 on the next line doesn't matter
 // and different LLVM versions produce different order.
diff --git a/src/test/codegen/mem-replace-direct-memcpy.rs b/src/test/codegen/mem-replace-direct-memcpy.rs
index 4318e926e47..e8bbf0e1bbd 100644
--- a/src/test/codegen/mem-replace-direct-memcpy.rs
+++ b/src/test/codegen/mem-replace-direct-memcpy.rs
@@ -18,7 +18,7 @@ pub fn replace_byte(dst: &mut u8, src: u8) -> u8 {
 // CHECK-NOT: call void @llvm.memcpy
 // CHECK: ; core::mem::replace
 // CHECK-NOT: call void @llvm.memcpy
-// CHECK: call void @llvm.memcpy.{{.+}}({{i8\*|ptr}} align 1 %{{.*}}, {{i8\*|ptr}} align 1 %dest, i{{.*}} 1, i1 false)
+// CHECK: call void @llvm.memcpy.{{.+}}({{i8\*|ptr}} align 1 %{{.*}}, {{i8\*|ptr}} align 1 %{{.*}}, i{{.*}} 1, i1 false)
 // CHECK-NOT: call void @llvm.memcpy
-// CHECK: call void @llvm.memcpy.{{.+}}({{i8\*|ptr}} align 1 %dest, {{i8\*|ptr}} align 1 %src{{.*}}, i{{.*}} 1, i1 false)
+// CHECK: call void @llvm.memcpy.{{.+}}({{i8\*|ptr}} align 1 %{{.*}}, {{i8\*|ptr}} align 1 %{{.*}}, i{{.*}} 1, i1 false)
 // CHECK-NOT: call void @llvm.memcpy
diff --git a/src/test/codegen/slice-iter-len-eq-zero.rs b/src/test/codegen/slice-iter-len-eq-zero.rs
index 1124028253d..894b0ec3de4 100644
--- a/src/test/codegen/slice-iter-len-eq-zero.rs
+++ b/src/test/codegen/slice-iter-len-eq-zero.rs
@@ -9,7 +9,7 @@ type Demo = [u8; 3];
 #[no_mangle]
 pub fn slice_iter_len_eq_zero(y: std::slice::Iter<'_, Demo>) -> bool {
     // CHECK-NOT: sub
-    // CHECK: %2 = icmp eq {{i8\*|ptr}} %1, %0
+    // CHECK: %2 = icmp eq {{i8\*|ptr}} {{%1|%0}}, {{%1|%0}}
     // CHECK: ret i1 %2
     y.len() == 0
 }