Introduce the Cranelift IR instruction LoadSplat

It corresponds to WebAssembly's `load*_splat` operations, which
were previously represented as a combination of `Load` and `Splat`
instructions. However, there are architectures such as Armv8-A
that have a single machine instruction equivalent to the Wasm
operations. In order to generate it, it is necessary to merge the
`Load` and the `Splat` in the backend, which is not possible
because the load may have side effects. The new IR instruction
works around this limitation.

The AArch64 backend leverages the new instruction to improve code
generation.

Copyright (c) 2020, Arm Limited.

cranelift/codegen/meta/src/shared/instructions.rs

@@ -4409,5 +4409,24 @@ pub(crate) fn define(
         .other_side_effects(true),
     );
 
+    let Offset = &Operand::new("Offset", &imm.offset32).with_doc("Byte offset from base address");
+    let a = &Operand::new("a", TxN);
+
+    ig.push(
+        Inst::new(
+            "load_splat",
+            r#"
+        Load an element from memory at ``p + Offset`` and return a vector
+        whose lanes are all set to that element.
+
+        This is equivalent to ``load`` followed by ``splat``.
+        "#,
+            &formats.load,
+        )
+        .operands_in(vec![MemFlags, p, Offset])
+        .operands_out(vec![a])
+        .can_load(true),
+    );
+
     ig.build()
 }