drt: Flip map keys in graph init #6161
Conversation
|
clang-tidy review says "All clean, LGTM! 👍" |
|
@kbieganski you might want to push something to restart the CI it looks like the C++20 action might have had a problem unrelated to this PR |
kbieganski
force-pushed
the
drt-graph-init-flip-keys
branch
from
e685c4f to
96295ad
Compare
|
clang-tidy review says "All clean, LGTM! 👍" |
|
What motived flipping the keys? Since we only have data for flipping + flat_map it is hard to see what benefit was had from each. |
|
|
kbieganski
force-pushed
the
drt-graph-init-flip-keys
branch
2 times, most recently
from
a38fb86 to
b5aa75d
Compare
|
clang-tidy review says "All clean, LGTM! 👍" |
|
Please merge master into this PR. I suspect that will solve the pr-head issue. |
kbieganski
force-pushed
the
drt-graph-init-flip-keys
branch
from
b5aa75d to
ef1c920
Compare
|
clang-tidy review says "All clean, LGTM! 👍" |
Signed-off-by: Krzysztof Bieganski <[email protected]>
Signed-off-by: Krzysztof Bieganski <[email protected]>
Signed-off-by: Krzysztof Bieganski <[email protected]>
Signed-off-by: Krzysztof Bieganski <[email protected]>
Signed-off-by: Krzysztof Bieganski <[email protected]>
Signed-off-by: Krzysztof Bieganski <[email protected]>
Signed-off-by: Krzysztof Bieganski <[email protected]>
Signed-off-by: Krzysztof Bieganski <[email protected]>
Signed-off-by: Krzysztof Bieganski <[email protected]>
Signed-off-by: Krzysztof Bieganski <[email protected]>
kbieganski
force-pushed
the
drt-graph-init-flip-keys
branch
from
ef1c920 to
1aaeafd
Compare
|
clang-tidy review says "All clean, LGTM! 👍" |
Signed-off-by: Krzysztof Bieganski <[email protected]>
kbieganski
force-pushed
the
drt-graph-init-flip-keys
branch
from
1aaeafd to
6dbb628
Compare
|
clang-tidy review says "All clean, LGTM! 👍" |
| // add edge for preferred direction | ||
| if (dir == dbTechLayerDir::HORIZONTAL && yFound) { | ||
| auto& yLayerMap = yMap[layerNum]; | ||
| auto& yLayer2Map = yMap[layerNum + 2]; |
There was a problem hiding this comment.
What if layerNum is the top routing layer? Wouldn't that insert layerNum+2 as a new entry to the flatmap and possible reallocate memory?
There was a problem hiding this comment.
Yes, but it's cheap:
- it only happens once at most,
- it may not reallocate,
- even if it does, it's on the order of 10 elements,
- and all the inner maps are just moved so they themselves don't reallocate.
In the original code this can happen for each coordinate.
|
@osamahammad21 Any further feedback? |
Also use Boost's
flat_mapinstead ofstd::map.aes-asap7-5_2_route- 8 thread(s)drt-graph-init-flip-keysibex-asap7-5_2_route- 8 thread(s)drt-graph-init-flip-keysibex-nangate45-5_2_route- 8 thread(s)drt-graph-init-flip-keysblack_parrot-nangate45-5_2_route- 8 thread(s)drt-graph-init-flip-keysI observed up to 10% speedup on some designs that I cannot share. I believe this scales nicely for bigger designs.