Avoid parsing field name twice when matching named instruction in get_json_object kernel
#2471
Conversation
Signed-off-by: Nghia Truong <[email protected]>
Signed-off-by: Nghia Truong <[email protected]>
This reverts commit 59ac5bf.
Signed-off-by: Nghia Truong <[email protected]>
Signed-off-by: Nghia Truong <[email protected]>
Signed-off-by: Nghia Truong <[email protected]>
ttnghia
force-pushed
the
improve_get_json_obj
branch
from
94ea0ca to
bd5645c
Compare
Signed-off-by: Nghia Truong <[email protected]>
Signed-off-by: Nghia Truong <[email protected]>
Signed-off-by: Nghia Truong <[email protected]>
This reverts commit a8d563c.
Signed-off-by: Nghia Truong <[email protected]>
ttnghia
force-pushed
the
improve_get_json_obj
branch
from
ec4e2f9 to
b86615a
Compare
Signed-off-by: Nghia Truong <[email protected]>
|
This seems not providing much performance improvement. Probably because when parsing the field name a second time we jump back just a short distance to the previous data which is already in L2 cache and still not being evicted. Anyway, this is considered as a cleanup PR. |
|
build |
| // check match if enabled | ||
| if (!try_match_char(to_match, str.current_char())) { return std::make_pair(false, 0); } | ||
|
|
||
| matched_field_name = matched_field_name && (to_match.is_null() || to_match.is_empty()); |
There was a problem hiding this comment.
Why do this self assignment? matched_field_name = matched_field_name
There was a problem hiding this comment.
This is an expression on boolean variable. In C++, such expressions can be short-circuit: the left operand is evaluated first, then the right operand is evaluated only if the left is true. So this means we can avoid the overhead of evaluating the right side (to_match.is_null() ||...) if matched_field_name is already false.
Because of such short-circuit functionality, there no exists the operators like &&=, ||= etc.
|
LGTM |
This enhances
get_json_objectkernel by parsing each field name only once, doing both verifying its validity and checking if it is matched with a given string name at the same time. By doing so, it avoids parsing the field name twice as in the current implementation, reducing the kernel runtime to some extent.The PR also does some cleanup to the code. Not much performance improvement is observed.
Closes #2413.