Optimize send calls

Similar to the bmethod optimization, this avoids using
CALLER_ARG_SPLAT if not necessary.  As long as the method argument
can be shifted off, other arguments are passed through as-is.

This optimizes the following types of calls:

* send(meth, arg) ~5%
* send(meth, *args) ~75% for args.length == 200
* send(meth, *args, **kw) ~50% for args.length == 200
* send(meth, **kw) ~25%
* send(meth, kw: 1) ~115%

Note that empty argument splats do get slower with this approach,
by about 20%.  This is probably because iseq argument setup is
slower for empty argument splats than CALLER_SETUP_ARG is. Other
than non-empty argument splats, other argument splats are faster,
with the speedup depending on the number of arguments.

The following types of calls are not optimized:

* send(*args)
* send(*args, **kw)

This is because the you cannot shift the method argument off
without first splatting the arg.

1 files changed, 77 insertions, 0 deletions

diff --git a/benchmark/vm_call_send_iseq.yml b/benchmark/vm_call_send_iseq.yml
new file mode 100644
index 0000000000..60ff23c475
--- /dev/null
+++ b/benchmark/vm_call_send_iseq.yml
@@ -0,0 +1,77 @@
+prelude: |
+  def a0; end
+  def a1(a) a; end
+  def s(*a) a; end
+  def b(kw: 1) kw end
+  def sb(*a, kw: 1) kw end
+
+  t0 = 0.times.to_a
+  t1 = 1.times.to_a
+  t10 = 10.times.to_a
+  t200 = 200.times.to_a
+
+  a0_t0 = [:a0, *t0]
+  a1_t1 = [:a1, *t1]
+  s_t0 = [:s, *t0]
+  s_t1 = [:s, *t1]
+  s_t10 = [:s, *t10]
+  s_t200 = [:s, *t200]
+  sb_t0 = [:sb, *t0]
+  sb_t1 = [:sb, *t1]
+  sb_t10 = [:sb, *t10]
+  sb_t200 = [:sb, *t200]
+  kw = {kw: 2}
+benchmark:
+  send_simple_0: |
+    send(:a0)
+  send_simple_1: |
+    send(:a1, 1)
+  send_simple_0_splat: |
+    send(:a0, *t0)
+  send_simple_1_splat: |
+    send(:a1, *t1)
+  send_simple_0_splat_comb: |
+    send(*a0_t0)
+  send_simple_1_splat_comb: |
+    send(*a1_t1)
+  send_no_splat: |
+    send(:s)
+  send_0_splat: |
+    send(:s, *t0)
+  send_1_splat: |
+    send(:s, *t1)
+  send_10_splat: |
+    send(:s, *t10)
+  send_200_splat: |
+    send(:s, *t200)
+  send_0_splat_comb: |
+    send(*s_t0)
+  send_1_splat_comb: |
+    send(*s_t1)
+  send_10_splat_comb: |
+    send(*s_t10)
+  send_200_splat_comb: |
+    send(*s_t200)
+  send_kw: |
+    send(:b, kw: 1)
+  send_no_kw: |
+    send(:b)
+  send_kw_splat: |
+    send(:b, **kw)
+  send_0_splat_kw: |
+    send(:sb, *t0, **kw)
+  send_1_splat_kw: |
+    send(:sb, *t1, **kw)
+  send_10_splat_kw: |
+    send(:sb, *t10, **kw)
+  send_200_splat_kw: |
+    send(:sb, *t200, **kw)
+  send_0_splat_comb_kw: |
+    send(*sb_t0, **kw)
+  send_1_splat_comb_kw: |
+    send(*sb_t1, **kw)
+  send_10_splat_comb_kw: |
+    send(*sb_t10, **kw)
+  send_200_splat_comb_kw: |
+    send(*sb_t200, **kw)
+loop_count: 3000000