Hi Benjamin, We are in the state, when we mainly have to concentrate on big improvements that we have to deliver till mid of March. I keep your ideas, but they are stuck at the end of my TODO list.
Thanks. Dmitry. On Tue, Jan 20, 2015 at 11:30 AM, Benjamin Coutu <ben.co...@zeyos.com> wrote: > Hi Dmitry, > > I was doing some code review of ext/standard/array.c and have recognized > some potential for a few performance improvements: > > === ARRAY_SLICE(..., preserve_keys=false) === > > array_slice() can always construct a packed array if preserve_keys is > false, restricting it to inputs with packed flag does not make much sense. > Removing the check for packed inputs on line 2376 would improve > performance if used on non-packed inputs with the default > preserve_keys=false. > Furthermore, ZEND_HASH_FOREACH_VAL should be used instead of > ZEND_HASH_FOREACH_NUM_KEY_VAL on line 2379. > It also think range() should use FAST_ZPP as it is a basic language > feature (other languages even have extra operators for it, e.g. [0..10]) > > === RANGE(...) === > > range() always returns a numerically indexed array [0..count-1]. The > resulting array therefore should be constructed as a packed array > (ZEND_HASH_FILL_PACKED+ZEND_HASH_FILL_ADD instead of > zend_hash_next_index_insert_new). > > === ARRAY_FILL(start_key=0, ...) > === > > Just like with range(), array_fill() always returns a numerically indexed > array [0..count-1] if start_key is 0. In this special but very common case > the resulting array can be constructed as a packed array > (ZEND_HASH_FILL_PACKED+ZEND_HASH_FILL_ADD instead of > zend_hash_next_index_insert_new). > Another common case is for start_key to be 1. One could refine the > proposed packed-array branch to just set the first bucket to undefined in > this case. > > === COUNT(...) === > > count() is so ubiquitous! Giving it an opcode and making it part of the VM > seams reasonable. > > Please let me know your thoughts. > > Cheers, > > Ben > > -- > > Benjamin Coutu > Zeyon Technologies Inc. > http://www.zeyos.com > >
