More optimizations using STk_C_make_list
#762
Conversation
STk_argv2list is only used twice in object.c, but still... It now conses much less.
Sligitly faster, conses much less: ``` (import (scheme list)) (define-macro (L i) `(list ,@(iota i))) (time (repeat 10000 (apply list (L 1000)))) ``` 740.866 ms 20000000 allocation calls 613.931 ms 10010000 allocation calls
Big improvement by using STk_C_make_list:
```
(import (scheme list))
(let ((a (iota 10000))
(b (map - (iota 10000))))
(time (repeat 10000 (append a b))))
```
Old code:
4097.964 ms
100000000 allocation calls
New code:
2899.566 ms
10000 allocation calls
Plus, the code doesn;'t use goto anymore, AND we report different
errors for cyclic lists, improper lists and no-lists. :)
``` (import (scheme vector @)) (let ((a (make-u64vector 5000 5))) (time (begin (repeat 10000 (%uvector->list a 7)) #t))) ``` Old code: 2492.513 ms 50000000 allocation calls New code: 1937.452 ms 10000 allocation calls But (scheme vector @) seems to not yet use this (it has Scheme code for turning the vectors into lists, we should be using %uvector->list instead...)
Using STk_C_make_list
|
A quick (Not counting single uses of |
|
For some reason, the code of Anyway, I have done some tests on the other patches, and as I said in #769, the gain of |
|
Hmmm. The nice thing about that is this part:
I'll see if we can implement the different-lists error reporting directly with the new code |
2 participants
The most interesting is APPEND. It really becomes faster, and conses much less.
Old code:
4097.964 ms
100000000 allocation calls
New code:
2899.566 ms
10000 allocation calls
Plus, the code doesn't use goto anymore, AND we report different errors for cyclic lists, improper lists and no-lists. :)