Ouch, an e-mail wasn't sent to the mailing list because I sent it direct.
I rewrite the wrong part and send it again.
On 2019/06/05 22:24, John Benediktsson wrote:
> Or clear the cache each time, so you don’t have to be lower level code...
>> On Jun 5, 2019, at 4:24 AM, Alexander Ilin <ajs...@yandex.ru>wrote: I think
>> the pairs of {font, string} are cached and only drawn once by the GUI
>> framework. I think you need to call the drawing code more directly to get
>> real performance data.
Thank you, Alexander and John.
I decided to measure only internal processing time.
On my machine, my code took 9 to 10 times longer to make a text image with a
transparent background
than it did with an opaque background.
! Copyright (C) 2019 KUSUMOTO Norio.
! See http://factorcode.org/license.txt for BSD license.
USING: kernel accessors locals math formatting ui ui.gadgets.labels
ui.commands ui.gadgets ui.gadgets.toolbar ui.gadgets.tracks ui.gestures
fonts colors tools.time memory math.ranges namespaces arrays sequences
windows.uniscribe windows.uniscribe.private ;
IN: uni-bench
TUPLE: uni-bench-gadget < track
opaque-font
transparent-font
opaque-time
transparent-time ;
: <uni-bench-gadget> ( -- gadget )
vertical uni-bench-gadget new-track
<font>
"monospace" >>name
T{ rgba f 0.0 0.0 0.0 1.0 } >>foreground
T{ rgba f 0.0 0.0 1.0 1.0 } >>background
36 >>size
>>opaque-font
<font>
"monospace" >>name
T{ rgba f 0.0 0.0 0.0 1.0 } >>foreground
T{ rgba f 0.0 0.0 1.0 0.0 } >>background
36 >>size
>>transparent-font
{
" script-string>image "
""
" Press any of the keys listed below"
""
" 1: Opaque background \"9999\" (1000 times)"
" 2: Transparent background \"9999\" (1000 times) "
""
} <label> f track-add
"" <label> >>opaque-time
"" <label> >>transparent-time
dup opaque-time>> " Opaque background:" label-on-left f track-add
dup transparent-time>> " Transparent:background" label-on-left f track-add ;
:: com-benchmark-opaque-script-string>image ( gadget -- )
gadget opaque-font>> :> test-font
gc
[
1000 [
test-font "9999" <script-string> script-string>image drop
] times
] benchmark
gadget opaque-time>>
swap 1000000 / "%d ms" sprintf >>text relayout-1 ;
:: com-benchmark-transparent-script-string>image ( gadget -- )
gadget transparent-font>> :> test-font
gc
[
1000 [
test-font "9999" <script-string> script-string>image drop
] times
] benchmark
gadget transparent-time>>
swap 1000000 / "%d ms" sprintf >>text relayout-1 ;
uni-bench-gadget "gestures" f {
{ T{ key-down { sym "1" } } com-benchmark-opaque-script-string>image }
{ T{ key-down { sym "2" } } com-benchmark-transparent-script-string>image }
} define-command-map
MAIN-WINDOW: uni-bench { { title "Uniscribe benchmark" } }
<uni-bench-gadget> >>gadgets ;
--
KUSUMOTO Norio
> 2019/06/05 22:24、John Benediktsson <mrj...@gmail.com> のメール：
>
> Or clear the cache each time, so you don’t have to be lower level code...
>
>
>> On Jun 5, 2019, at 4:24 AM, Alexander Ilin <ajs...@yandex.ru> wrote:
>>
>> I think the pairs of {font, string} are cached and only drawn once by the
>> GUI framework.
>> I think you need to call the drawing code more directly to get real
>> performance data.
>>
>> 04.06.2019, 07:39, "KUSUMOTO, Norio" <kusum...@na.rim.or.jp>:
>>> My code was uploaded.
>>>
>>> https://github.com/kusumotonorio/factor/blob/win-transparent-background-text/basis/windows/uniscribe/uniscribe.factor
>>>
>>> I want to know how slow the code when it draws transparent background
>>> texts. So I am writing
>>> a test app to know it. But it has an issue.
>>> I try to measure time by actually drawing a label with a transparent
>>> background over and over again, and that happens very quickly.
>>> Creating an image from text takes a long time (3: and 4:) , so it should
>>> actually take longer.
>>> What's wrong with my code?
>>>
>>> ! Copyright (C) 2019 KUSUMOTO Norio.
>>> ! See http://factorcode.org/license.txt for BSD license.
>>> USING: kernel accessors locals math formatting ui ui.gadgets.labels
>>> ui.commands ui.gadgets ui.gadgets.toolbar ui.gadgets.tracks ui.gestures
>>> fonts colors tools.time memory math.ranges namespaces arrays sequences
>>> windows.uniscribe windows.uniscribe.private ;
>>> IN: uni-bench
>>>
>>> TUPLE: uni-bench-gadget < track
>>> opaque-label
>>> transparent-label
>>> opaque-time
>>> transparent-time ;
>>>
>>> SYMBOL: test-strings
>>> V{ } clone test-strings set-global
>>> 1 5000 [a,b] [
>>> "%04d" sprintf test-strings get push
>>> ] each
>>> test-strings get >array test-strings set-global
>>>
>>> : <uni-bench-gadget> ( -- gadget )
>>> vertical uni-bench-gadget new-track
>>> {
>>> " Press any of the keys listed below"
>>> ""
>>> " 1: Opaque background label (5000 times)"
>>> " 2: Transparent background label (5000 times)"
>>> " 3: Opaque script-string>image (5000 times)"
>>> " 4: Transparent script-string>image (5000 times)"
>>> ""
>>> } <label> f track-add
>>>
>>> "0000" <label> >>opaque-label
>>> "0000" <label> >>transparent-label
>>> "" <label> >>opaque-time
>>> "" <label> >>transparent-time
>>>
>>> dup opaque-label>>
>>> <font>
>>> "monospace" >>name
>>> T{ rgba f 0.0 0.0 0.0 1.0 } >>foreground
>>> T{ rgba f 0.0 0.0 1.0 1.0 } >>background
>>> 36 >>size
>>>>> font f track-add
>>>
>>> dup transparent-label>>
>>> <font>
>>> "monospace" >>name
>>> T{ rgba f 0.0 0.0 0.0 1.0 } >>foreground
>>> T{ rgba f 0.0 0.0 1.0 0.0 } >>background
>>> 36 >>size
>>>>> font f track-add
>>>
>>> dup opaque-time>> " Opaque:" label-on-left f track-add
>>> dup transparent-time>> " Transparent:" label-on-left f track-add ;
>>>
>>> :: com-benchmark-opaque ( gadget -- )
>>> gadget opaque-label>> :> test-label
>>> gc
>>> [
>>> test-strings get-global [
>>> test-label swap >>text relayout-1
>>> ] each
>>> ] benchmark
>>> gadget opaque-time>>
>>> swap 1000000 / "%d ms" sprintf >>text relayout-1 ;
>>>
>>> :: com-benchmark-transparent ( gadget -- )
>>> gadget transparent-label>> :> test-label
>>> gc
>>> [
>>> test-strings get-global [
>>> test-label swap >>text relayout-1
>>> ] each
>>> ] benchmark
>>> gadget transparent-time>>
>>> swap 1000000 / "%d ms" sprintf >>text relayout-1 ;
>>>
>>> :: com-benchmark-opaque-script-string>image ( gadget -- )
>>> gadget opaque-label>> font>> :> test-font
>>> gc
>>> [
>>> test-strings get-global [
>>> test-font swap <script-string> script-string>image drop
>>> ] each
>>> ] benchmark
>>> gadget opaque-time>>
>>> swap 1000000 / "%d ms" sprintf >>text relayout-1 ;
>>>
>>> :: com-benchmark-transparent-script-string>image ( gadget -- )
>>> gadget transparent-label>> font>> :> test-font
>>> gc
>>> [
>>> test-strings get-global [
>>> test-font swap <script-string> script-string>image drop
>>> ] each
>>> ] benchmark
>>> gadget transparent-time>>
>>> swap 1000000 / "%d ms" sprintf >>text relayout-1 ;
>>>
>>> uni-bench-gadget "gestures" f {
>>> { T{ key-down { sym "1" } } com-benchmark-opaque }
>>> { T{ key-down { sym "2" } } com-benchmark-transparent }
>>> { T{ key-down { sym "3" } } com-benchmark-opaque-script-string>image }
>>> { T{ key-down { sym "4" } }
>>> com-benchmark-transparent-script-string>image }
>>> } define-command-map
>>>
>>> M: uni-bench-gadget pref-dim* drop { 300 250 } ;
>>>
>>> MAIN-WINDOW: uni-bench { { title "Uniscribe benchmark" } }
>>> <uni-bench-gadget> >>gadgets ;
>>>
>>>> On 2019/05/30 21:17, KUSUMOTO Norio wrote:
>>>> Thank you, Alexander! I will read it and study. I tried the idea I
>>>> wrote earlier. I've only tried a little, but it seems to work well.
>>>> Yay! <https://pbs.twimg.com/media/D70GEGkVUAUxgWb.png> -- KUSUMOTO Norio
>>>>> 2019/05/30 20:19、Alexander Ilin <ajs...@yandex.ru>のメール: Hello! I
>>>>> think I have figured out the algorithm for the image processing. I'm
>>>>> not sure if it's the same one that you came up with.
>>>>> https://github.com/factor/factor/issues/152#issuecomment-497292323
>>>>> Also, I don't know if there are standard WinApi functions to perform
>>>>> the necessary operation, namely copying a color channel into the
>>>>> alpha channel, but I suspect there might be, in which case we won't
>>>>> lose performance there. 30.05.2019, 03:50, "KUSUMOTO Norio"
>>>>> <kusum...@na.rim.or.jp>:
>>>>>> Although it seems that to transfer the appropriate data to the
>>>>>> device context is the 'right' solution, I have a hunch that it will
>>>>>> be a difficult task. So I am beginning to think that we should take
>>>>>> another, a cheat, approach. It's like a chroma key. Uniscribe draws
>>>>>> text with a color background instead of transparency, and a word
>>>>>> replaces the color with transparency when converting from the bitmap
>>>>>> to an Factor's image. It may be slow, but we don't have many chances
>>>>>> to draw characters on a transparent background. And for example, on
>>>>>> a button label, once Factor creates an image with such characters,
>>>>>> Factor use the image again, so I think there are few problems. This
>>>>>> approach can localize changes. I can't define the word for image
>>>>>> conversion to do such a special action, but I think that it's not a
>>>>>> difficult task for someone familiar with image processing words.
>>>>> ---=====--- Александр _______________________________________________
>>>>> Factor-talk mailing list Factor-talk@lists.sourceforge.net
>>>>> https://lists.sourceforge.net/lists/listinfo/factor-talk
>>>> _______________________________________________ Factor-talk mailing
>>>> list Factor-talk@lists.sourceforge.net
>>>> https://lists.sourceforge.net/lists/listinfo/factor-talk
>>>
>>> _______________________________________________
>>> Factor-talk mailing list
>>> Factor-talk@lists.sourceforge.net
>>> https://lists.sourceforge.net/lists/listinfo/factor-talk
>>
>> ---=====---
>> Александр
>>
>>
>>
>> _______________________________________________
>> Factor-talk mailing list
>> Factor-talk@lists.sourceforge.net
>> https://lists.sourceforge.net/lists/listinfo/factor-talk
>
>
> _______________________________________________
> Factor-talk mailing list
> Factor-talk@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/factor-talk
_______________________________________________
Factor-talk mailing list
Factor-talk@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/factor-talk
