I hope I could convince you of various advantages introduced with the development of Intelligent Design. Compared against conventional programming techniques (as discussed in this document), Intelligent Design wins in all disciplines - be it sheer speed, high code density, simplified development or efficiency. Due to the strict renunciation of slow leave, pop and push instructions, coming along with many ill side-effects like an unpredictable stack pointer, these caveats of conventional software design easily are turned into some really welcome side-effects by replacing those slow instructions with simple movs. The most welcome side-effect surely is the fact that we get EBP back as a general purpose register. One additional register is quite a lot if only six of them are available, because EBP was abused as base pointer. Each additional register is an advantage, because we can pair reads to copy parameters from memory to registers and access these registers rather than reloading single registers with some frequently used parameters over and over again.
Another advantage is that we mov parameters to the right location rather than to push them onto the stack. While mov instructions can store changed parameters anywhere, a push always addresses the next lower stack element (a quite limited range). If we call a function repeatedly, and only the last parameter changes from call to call, we had to push all parameters after each call to update this topmost parameter with the current data. Moreover, the correction of ESP is mandatory after each call, blocking the following push for an entire clock cycle. Using mov instructions, all these problems vanish. We save many unnecessary instructions and keep the three execution pipes busy most of the time. Of course, there are a lot of cases where only one or two instructions are required between two calls, but this still is faster than a conventional push orgy.
One more advantage is the fact that we can mov parameters and other data onto the stack anywhere within our source file. Pairing multiple reads in groups, we can avoid direct dependencies completely. Pairing multiple writes, we trigger write combining. Keeping proper distances between reads and writes can eliminate most dependencies with few exceptions. Applying these tricks as often as possible speeds up code markably.
Putting it all together, Intelligent Design is an up-to-date alternative to old-fashioned conventional programming techniques. Conventional software design did not change too much in the last thirty years. We neither can create the most simple integrated circuit with stone age tools, nor are we able to create software making full use of the capabilities provided by modern processors with conventional code. While old fashioned code is spiced with mandatory brake pads and compilers generate tons of counter-productive dependencies, because the convention allows to abuse the most valuable resources - our registers - as garbage pile to save four instructions at the cost of multiple reloads, Intelligent Design replaces these obstacles with straight and clean rules, providing full support for features and improvements of recent processors. Intelligent Design is an up-to-date concept for the next generation of operating systems and applications, introducing a new quality to software design. Moreover, it is much easier to handle a single stack pointer than to manage two registers, a separate stack and base pointer, simultaneously. Getting rid of the institution called base pointer, we finally can say good bye to positive and negative offsets and error prone corrections of the stack pointer after each of the obligatory push orgies. Another positive side-effect is the return of a valuable resource (EBP) to the very sparse pool of general purpose registers - the worst conceptual flaw of LETNi's x86 architecture.
All in all, Intelligent Design points the way ahead to fastest code with high density, causing less head aches for stressed programmers. Best of all: This is not just a theoretical construction, floating around in the head of a weird, unwordly developer - it exists for real! ST-Open's libraries as well as DatTools and the SRE editor were ported to Intelligent Design, working as expected: Really fast and reliable.
Go to the next post (11- Rules)
Another advantage is that we mov parameters to the right location rather than to push them onto the stack. While mov instructions can store changed parameters anywhere, a push always addresses the next lower stack element (a quite limited range). If we call a function repeatedly, and only the last parameter changes from call to call, we had to push all parameters after each call to update this topmost parameter with the current data. Moreover, the correction of ESP is mandatory after each call, blocking the following push for an entire clock cycle. Using mov instructions, all these problems vanish. We save many unnecessary instructions and keep the three execution pipes busy most of the time. Of course, there are a lot of cases where only one or two instructions are required between two calls, but this still is faster than a conventional push orgy.
One more advantage is the fact that we can mov parameters and other data onto the stack anywhere within our source file. Pairing multiple reads in groups, we can avoid direct dependencies completely. Pairing multiple writes, we trigger write combining. Keeping proper distances between reads and writes can eliminate most dependencies with few exceptions. Applying these tricks as often as possible speeds up code markably.
Putting it all together, Intelligent Design is an up-to-date alternative to old-fashioned conventional programming techniques. Conventional software design did not change too much in the last thirty years. We neither can create the most simple integrated circuit with stone age tools, nor are we able to create software making full use of the capabilities provided by modern processors with conventional code. While old fashioned code is spiced with mandatory brake pads and compilers generate tons of counter-productive dependencies, because the convention allows to abuse the most valuable resources - our registers - as garbage pile to save four instructions at the cost of multiple reloads, Intelligent Design replaces these obstacles with straight and clean rules, providing full support for features and improvements of recent processors. Intelligent Design is an up-to-date concept for the next generation of operating systems and applications, introducing a new quality to software design. Moreover, it is much easier to handle a single stack pointer than to manage two registers, a separate stack and base pointer, simultaneously. Getting rid of the institution called base pointer, we finally can say good bye to positive and negative offsets and error prone corrections of the stack pointer after each of the obligatory push orgies. Another positive side-effect is the return of a valuable resource (EBP) to the very sparse pool of general purpose registers - the worst conceptual flaw of LETNi's x86 architecture.
All in all, Intelligent Design points the way ahead to fastest code with high density, causing less head aches for stressed programmers. Best of all: This is not just a theoretical construction, floating around in the head of a weird, unwordly developer - it exists for real! ST-Open's libraries as well as DatTools and the SRE editor were ported to Intelligent Design, working as expected: Really fast and reliable.
Go to the next post (11- Rules)
No comments:
Post a Comment