Optimize Baseline interpreter by using padded code #315
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
chfast
force-pushed
the
baseline_padded_code
branch
3 times, most recently
from
1085606 to
915e7b5
Compare
Codecov Report
@@ Coverage Diff @@
## master #315 +/- ##
=======================================
Coverage 99.78% 99.78%
=======================================
Files 29 29
Lines 4108 4112 +4
=======================================
+ Hits 4099 4103 +4
Misses 9 9
Flags with carried forward coverage won't be shown. Click here to find out more.
|
chfast
force-pushed
the
baseline_padded_code
branch
from
915e7b5 to
fbf2d80
Compare
|
Haswell 4.4 GHz, clang-12 |
|
AMD Zen3, GCC-9 |
|
AMD Zen3, clang-12 |
gumb0
reviewed
Jun 1, 2021
lib/evmone/baseline.cpp
Outdated
| uint8_t buffer[Len]; | ||
| // This cannot overflow code buffer because code ends with valid STOP instruction. |
There was a problem hiding this comment.
I think here it's not relevant that it's a STOP?
Suggested change
| // This cannot overflow code buffer because code ends with valid STOP instruction. | |
| // This cannot overflow code buffer because code is padded with 0s. |
There was a problem hiding this comment.
I think the original comment is more correct, as the loop is looking for STOP
There was a problem hiding this comment.
I thought the comment here is supposed to explain why memcpy cannot overflow.
lib/evmone/baseline.cpp
Outdated
| auto pc = code; | ||
| while (pc != code_end) | ||
| while (true) // Guaranteed to terminate because code must end with STOP. |
There was a problem hiding this comment.
Suggested change
| while (true) // Guaranteed to terminate because code must end with STOP. | |
| while (true) // Guaranteed to terminate because padded code ends with STOP. |
chfast
force-pushed
the
baseline_padded_code
branch
3 times, most recently
from
ea36730 to
76a3fb9
Compare
gumb0
reviewed
Jun 3, 2021
lib/evmone/baseline.cpp
Outdated
| return CodeAnalysis{std::move(map)}; | ||
|
|
||
| // i is the needed code size including the last push data (can be bigger than code_size). | ||
| std::unique_ptr<uint8_t[]> padded_code{new uint8_t[i + 1]}; |
There was a problem hiding this comment.
make_unique would zero-initialize, as opposed to this?
There was a problem hiding this comment.
I would maybe add a comment that this leaves it unitialized.
gumb0
reviewed
Jun 3, 2021
gumb0
approved these changes
Jun 3, 2021
Guaranteed to terminate because padded code ends with STOP.
This cannot overflow code buffer because code ends with valid STOP instruction.
chfast
force-pushed
the
baseline_padded_code
branch
from
76a3fb9 to
6538e6a
Compare
Yes, this was suppose to be TODO, so I have added it now. This was my original plan, but there are additional complications:
|
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
During code analysis the code is copied and padded with 33 zero bytes. This guarantees that push data is always available in the code buffer and the code ends with STOP. This allows optimizing the interpreter loop and PUSH instructions.
Performance gains are up to ~5%.
This is also needed to enable other optimization: implementing interpreter with "computed goto" or "tail calls".