Différences

Ci-dessous, les différences entre deux révisions de la page.

--- back2root:archives:denthor:part-08 [2021/09/05 14:05] – [C] frater
+++ back2root:archives:denthor:part-08 [2021/09/18 20:39] (Version actuelle) – [Optimisation] frater
@@ Ligne 24: / Ligne 24: @@
 ==== Optimisation ====
-Before I begin with the note on 3-D, I would like to stress that many of these routines, and probably most of your own, could be sped up quite a bit with a little optimisation. One must realise, however, that you must
+Before I begin with the note on 3-D, I would like to stress that many of these routines, and probably most of your own, could be sped up quite a bit with a little optimisation.
-take a look at WHAT to optimise ... converting a routine that is only called once at startup into a tightly coded assembler routine may show off your merits as a coder, but does absolutely nothing to speed up your program. Something that is called often per frame is something that needs to be as fast as possible. For some, a much used procedure is the PutPixel procedure. Here is the putpixel procedure I gave you last week:
+One must realise, however, that you must take a look at WHAT to optimise ... converting a routine that is only called once at startup into a tightly coded assembler routine may show off your merits as a coder, but does absolutely nothing to speed up your program.
+Something that is called often per frame is something that needs to be as fast as possible. For some, a much used procedure is the PutPixel procedure. Here is the putpixel procedure I gave you last week:
 <code pascal>
@@ Ligne 66: / Ligne 69: @@
 </WRAP>
-Right, now for some optimising. Firstly, if you have 286 instructions turned on, you may replace the 6 shl,1 with shl,6. Secondly, the Pascal compiler automatically pushes and pops ES, so those two lines may be removed. DS:[SI] is not altered in this procedure, so we may remove those too. Also, instead of moving COL into ah, we move it into AL and call stosb (es:[di]:=al; inc di). Let's have a look at the routine now :
+Right, now for some optimising. Firstly, if you have 286 instructions turned on, you may replace the 6 shl,1 with shl,6.
+Secondly, the Pascal compiler automatically pushes and pops ES, so those two lines may be removed. DS:[SI] is not altered in this procedure, so we may remove those too. Also, instead of moving COL into ah, we move it into AL and call stosb (es:[di]:=al; inc di). Let's have a look at the routine now :
 <code pascal>
@@ Ligne 92: / Ligne 97: @@
 Total = 95 clock ticks
-Now, let us move the value of BX directly into DI, thereby removing a costly push and pop. The MOV and the XOR of DX can be replaced by it's equivalent, SHL DX,8
+Now, let us move the value of BX directly into DI, thereby removing a costly push and pop. The MOV and the XOR of DX can be replaced by it's equivalent, **SHL DX,8**
 <code pascal>
@@ Ligne 113: / Ligne 118: @@
 Total = 71 clock ticks
-As you can see, we have brought the clock ticks down from 153 ticks to 71 ticks ... quite an improvement. (The current ASPHYXIA putpixel takes 48 clock ticks) . As you can see, by going through your routines a few times, you can spot and remove unnecessary instructions, thereby greatly increasing the speed of your program.
+As you can see, we have brought the clock ticks down from 153 ticks to 71 ticks ... quite an improvement. (The current ASPHYXIA putpixel takes 48 clock ticks).
+As you can see, by going through your routines a few times, you can spot and remove unnecessary instructions, thereby greatly increasing the speed of your program.
 ==== Defining a 3-D object ====

Nox-Rhea WiKi

Outils pour utilisateurs

Outils du site

Différences

Outils de la page