Reduce code size by sharing code for clamps

[charles-cooper]

What I did

See title

How I did it

All clamps have the following code:

(assume condition is on the stack)
ISZERO _revert JUMPI # if not condition goto _revert
PUSH1 0 DUP REVERT
_revert
JUMPDEST

Since clamps are super common this code gets repeated. This reduces code
size by adding the PUSH1 0 DUP REVERT to each program's postamble, and
then clamps are simplified to the following

(assume condition is on the stack)
_revert JUMPI # if not condition goto _revert
-- cool now 5 opcodes are skipped

Then at the end of the contract, the following "postamble" is inserted:

_revert
JUMPDEST
PUSH1 0 DUP REVERT # repeated code is shared, yay

This commit also optimizes the following common "truthy" sequences:

ISZERO ISZERO ISZERO -> ISZERO
ISZERO ISZERO _jumpdest JUMPI -> _jumpdest JUMPI

How to verify it

Tests pass. I compared with master (ba0676b) and compiling a couple large contracts resulted in nearly 10% codesize reduction:

vyper -f bytecode_runtime

	master (ba0676b)	This branch (revert_codesize)
curve-contract/StableSwapIB.vy	23792 bytes	21629 bytes
yearn-vaults/Vault.vy	21816 bytes	20243 bytes

Description for the changelog

Reduce code size by ~10% by sharing clamp code

Cute Animal Picture

[codecov-commenter]

Codecov Report

Merging #2387 (82c5641) into master (ba0676b) will increase coverage by 0.09%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master    #2387      +/-   ##
==========================================
+ Coverage   85.74%   85.84%   +0.09%     
==========================================
  Files          91       91              
  Lines        8995     9021      +26     
  Branches     2145     2152       +7     
==========================================
+ Hits         7713     7744      +31     
+ Misses        788      785       -3     
+ Partials      494      492       -2     

Impacted Files	Coverage Δ
vyper/compile_lll.py	93.88% <100.00%> (+1.02%)	⬆️
vyper/functions/functions.py	88.74% <0.00%> (+0.38%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update ba0676b...82c5641. Read the comment docs.

[iamdefinitelyahuman]

Tried testing this by compiling a Curve contract and running the related tests, it's failing in Brownie during the compilation phase because of an error in the source map.

[iamdefinitelyahuman]

Could this change instead be applied during LLL optimizations in optimizer.py ?

[charles-cooper]

@iamdefinitelyahuman it would require adding a conditional jump macro to LLL. Otherwise, we will have sequences like the following

(if condition (goto fail))

Which would generate an extra 3 instruction JUMP block like this

condition
PUSH _escape
JUMPI
PUSH _fail_block
JUMP
_escape
JUMPDEST

instead of

not_condition
PUSH _fail_block
JUMPI

But the way I have it also causes optimization problems being generated so late, for instance the jumpdest label can't be optimized into a dup.

[charles-cooper]

@iamdefinitelyahuman source map error should be fixed in 82c5641

(I still have reservations about the approach of doing this in the opcode gen instead of as an LLL optimization step)

EDIT: After sleeping on it I decided I'm OK with doing it in the opcode gen since I plan to rewrite all of LLL anyway.