legalizer/split.rs: simplify_branch_arguments: use SmallVec instead of Vec

This function is responsible for 8.5% of all heap allocation (calls) in CL.
This change avoids almost all of them by using a SmallVec::<[Value; 32]>
instead.  Dynamic instruction count falls by 0.25%.  The fixed size of 32 was
arrived at after profiling with fixed sizes of 1, 2, 4, 8, 16, 32, 64 and 128.
32 is as high as I can push it without the instruction count starting to creep
up again, and gets almost all the block-reduction win of 64 and 128.

cranelift/codegen/src/legalizer/split.rs

@@ -68,6 +68,7 @@ use crate::cursor::{Cursor, CursorPosition, FuncCursor};
 use crate::flowgraph::{BasicBlock, ControlFlowGraph};
 use crate::ir::{self, Ebb, Inst, InstBuilder, InstructionData, Opcode, Type, Value, ValueDef};
 use core::iter;
+use smallvec::SmallVec;
 use std::vec::Vec;
 
 /// Split `value` into two values using the `isplit` semantics. Do this by reusing existing values
@@ -373,7 +374,7 @@ fn resolve_splits(dfg: &ir::DataFlowGraph, value: Value) -> Value {
 /// After legalizing the instructions computing the value that was split, it is likely that we can
 /// avoid depending on the split instruction. Its input probably comes from a concatenation.
 pub fn simplify_branch_arguments(dfg: &mut ir::DataFlowGraph, branch: Inst) {
-    let mut new_args = Vec::new();
+    let mut new_args = SmallVec::<[Value; 32]>::new();
 
     for &arg in dfg.inst_args(branch) {
         let new_arg = resolve_splits(dfg, arg);