Rollup merge of #95961 - RalfJung:gather-scatter, r=workingjubilee

implement SIMD gather/scatter via vector getelementptr

Fixes https://github.com/rust-lang/portable-simd/issues/271

However, I don't *really* know what I am doing here... Cc ``@workingjubilee`` ``@calebzulawski``

I didn't do anything for cranelift -- ``@bjorn3`` not sure if it's okay for that backend to temporarily break. I'm happy to cherry-pick a patch that adds cranelift support. :)

src/test/codegen/simd_arith_offset.rs

@@ -0,0 +1,26 @@
+// compile-flags: -C no-prepopulate-passes
+// only-64bit (because the LLVM type of i64 for usize shows up)
+//
+
+#![crate_type = "lib"]
+#![feature(repr_simd, platform_intrinsics)]
+
+extern "platform-intrinsic" {
+    pub(crate) fn simd_arith_offset<T, U>(ptrs: T, offsets: U) -> T;
+}
+
+/// A vector of *const T.
+#[derive(Debug, Copy, Clone)]
+#[repr(simd)]
+pub struct SimdConstPtr<T, const LANES: usize>([*const T; LANES]);
+
+#[derive(Debug, Copy, Clone)]
+#[repr(simd)]
+pub struct Simd<T, const LANES: usize>([T; LANES]);
+
+// CHECK-LABEL: smoke
+#[no_mangle]
+pub fn smoke(ptrs: SimdConstPtr<u8, 8>, offsets: Simd<usize, 8>) -> SimdConstPtr<u8, 8> {
+    // CHECK: getelementptr i8, <8 x i8*> %_3, <8 x i64> %_4
+    unsafe { simd_arith_offset(ptrs, offsets) }
+}