Turn on MICROPY_OPT_COMPUTED_GOTO for 5x CPU-bound speedup

[dhalbert]

Turning this on gives about a 5x speedup (!) for CPU-bound code.

I expect the first submission of this PR to fail because it won't fit on some boards. I'll adjust with further commits.

[shazz]

Tested, works on the pygamer.

[tannewt]

I'm concerned that the inline knob is breaking the code.

Here is a sample disassembly with 25:

(gdb) disassemble audioio_mixer_obj_get_sample_rate
Dump of assembler code for function audioio_mixer_obj_get_sample_rate:
   0x0000ead0 <+0>:	push	{r4, lr}
   0x0000ead2 <+2>:	movs	r4, r0
   0x0000ead4 <+4>:	ldr	r0, [r0, #4]
   0x0000ead6 <+6>:	negs	r3, r0
   0x0000ead8 <+8>:	adcs	r0, r3
   0x0000eada <+10>:	uxtb	r0, r0
   0x0000eadc <+12>:	bl	0xe364 <raise_error_if_deinited>
   0x0000eae0 <+16>:	movs	r0, #1
   0x0000eae2 <+18>:	ldr	r3, [r4, #20]
   0x0000eae4 <+20>:	lsls	r3, r3, #1
   0x0000eae6 <+22>:	orrs	r0, r3
   0x0000eae8 <+24>:	pop	{r4, pc}

With 23:

(gdb) disassemble audioio_mixer_obj_get_sample_rate
Dump of assembler code for function audioio_mixer_obj_get_sample_rate:
   0x0000e85e <+0>:	ldr	r3, [r0, #20]
   0x0000e860 <+2>:	movs	r0, #1
   0x0000e862 <+4>:	lsls	r3, r3, #1
   0x0000e864 <+6>:	orrs	r0, r3
   0x0000e866 <+8>:	bx	lr

[tannewt]

I did some size improvements here: tannewt@5399f02

[dhalbert]

That jogged my memory: I saw breakage a long time ago when I set the inline limit really low: #322. I tested it briefly with 23 but it was really just a smoke test

[dhalbert]

USB_MIDI is turned on for all builds. We could turn it off for small builds. Saves 728 bytes.

I looked at the pinyin translation in hopes of trimming it a bit. It's not that verbose in terms of character count: it's the double-byte characters that really push it up.

[dhalbert]

Using regular -Os instead of -O3 (SUPEROPT) for gc.c saves 2144 bytes.

Using regular -Os instead of -O3 (SUPEROPT) for vm.c only saves 524 bytes.

My inclination is to slow down gc for CIRCUITPY_SMALL_BUILD and/or for the overflowing builds, as you suggested. Then we can dial way back towards normal on CFLAGS_INLINE_LIMIT. I'll work on setting up compilation flags for this.

[tannewt]

Looks great! Thank you so much!