x64: Fix codegen for the `i8x16.swizzle` instruction (#4318)

This commit fixes a mistake in the `Swizzle` opcode implementation in
the x64 backend of Cranelift. Previously an input register was casted to
a writable register and then modified, which I believe instructions are
not supposed to do. This was discovered as part of my investigation
into #4315.

cranelift/codegen/src/isa/x64/lower.rs

@@ -2554,17 +2554,18 @@ fn lower_insn_to_regs<C: LowerCtx<I = Inst>>(
             ctx.emit(Inst::xmm_load_const(constant, zero_mask, ty));
 
             // Use the `zero_mask` on a writable `swizzle_mask`.
-            let swizzle_mask = Writable::from_reg(swizzle_mask);
+            let swizzle_mask_tmp = ctx.alloc_tmp(types::I8X16).only_reg().unwrap();
+            ctx.emit(Inst::gen_move(swizzle_mask_tmp, swizzle_mask, ty));
             ctx.emit(Inst::xmm_rm_r(
                 SseOpcode::Paddusb,
                 RegMem::from(zero_mask),
-                swizzle_mask,
+                swizzle_mask_tmp,
             ));
 
             // Shuffle `dst` using the fixed-up `swizzle_mask`.
             ctx.emit(Inst::xmm_rm_r(
                 SseOpcode::Pshufb,
-                RegMem::from(swizzle_mask),
+                RegMem::from(swizzle_mask_tmp),
                 dst,
             ));
         }