VM2 Investigate Contract Code Size

[miloszm]

Describe what you want implemented In VM2 contract size grew in comparison with VM1. Gather more data for proper comparison and try to explain what is causing the growth.

Describe "Why" this is needed We need it to be able to influence and control bytecode sizes for VM2.

Describe alternatives you've considered There is really no alternative.

Additional context Until now we focused on gas usage measurements, but code size is also important.

[ureeves]

At the time of writing - https://github.com/dusk-network/rusk-vm/commit/e2fa3209589303c246b14f9cb5ce571de55f266a - here is a comparison test contract sizes. The first column is the size with the code unaltered, while the second is the code with the following modifications:

• Removing std dependency of map, stack, and string_argument contracts.

• Introducing the following language items in rusk-uplink:

  • alloc_error_handler
  • panic_handler
  • wee_alloc as global_allocator

Test contract	VM1	VM1 with build opts	VM2	VM2 with mods	VM2 with mods and build opts
block_height	13K	5K	27K	18K	11K
callee_1	23K	16K	45K	37K	24K
callee_2	19K	12K	40K	31K	19K
caller	20K	12K	45K	36K	23K
counter	17K	9K	34K	25K	17K
delegator	19K	11K	32K	24K	15K
events	N/A	N/A	30K	22K	14K
fibonacci	16K	7K	30K	21K	13K
gas_consumed	15K	7K	30K	22K	14K
gas_context	22K	14K	50K	42K	26K
map	37K	28K	66K	57K	44K
minimal_counter	N/A	N/A	28K	20K	12K
register	N/A	N/A	71K	63K	48K
self_snapshot	25K	19K	38K	30K	22K
stack	39K	30K	73K	65K	49K
string_argument	N/A	N/A	28K	20K	12K
tx_vec	20K	12K	49K	40K	18K

As one can see the sizes are still significantly larger - sometimes double. The experiment can be found in the size-reduction-374 branch.

EDIT: the last column adds some build optimization options to the contracts, and makes them much smaller. It does make some tests fail - specifically the ones using the stack contract.

EDIT: added another column adding the same optimization options to VM1. As expected, sizes are significantly smaller as in VM2. Some tests fail here too - specifically panicking_nested and self_snapshot. Experiment in the size-reduction-vm1-374 branch.

[miloszm]

VM1 vs VM2 gas usage comparison results: (VM1 test run on 06 May 2022)

Test	VM1	VM2	VM2 with mods
counter	5,493	9,564	9,532
stack single push/pop	31,937	13,426	13,726
stack multiple push/pop (65536)	806,802,898	255,004,445	254,352,221
stack multiple transactions push (8192)	1,290,562,425	221,023,128	230,148,192
hamt single insert/get	41,472	19,860	20,492
hamt multiple insert/get (8192)	1,606,679,210	161,104,066	166,281,410

the same tests with refreshed VM1 data: (run on 23 June 2022)

Test	VM1	VM2	VM2 with mods
counter	5,426	9,564	9,532
stack single push/pop	19,015	13,426	13,726
stack multiple push/pop (65536)	594,745,327	255,004,445	254,352,221
stack multiple transactions push (8192)	791,263,208	221,023,128	230,148,192
hamt single insert/get	25,088	19,860	20,492
hamt multiple insert/get (8192)	1,048,506,722	161,104,066	166,281,410

tested VM1 code is in branch master-vm1-gas-usage 1ebdc25162 tested VM2 code is in branch master e2fa320958 tested VM2 with mods code is in branch size-reduction-374 dd1ec63887

[miloszm]

Bench run comparison between VM1 and VM2:

Bench	VM1	VM2
Fibonacci 15	165.51 ms	211.29 ms
Stack 64	6.734 ms	8.889 ms