- 第 2 页 -
If you dive deep into the use of floating point math you will find many buried areas that create round-off errors inside the FPU directly... these artifacts are the very reason Floating Point math is not a good choice for audio math... the audio seems to be continually degraded randomly along the way thru the engine... this is pretty noticably audible to many people that are HEARING the sound of digital... in my opinion.
I have tripped over examples in the 15 years of coding SAW where the same math calculation gives different and false results depending on the exact values of the variables in that calculation... one set of numbers results in a round-up of the result and another set results in a round-down... same equation, different numbers... therefore making it very difficult to actually have the code consistently interpret the results and in many cases branch to the wrong code segment based on a zero result not testing TRUE.
I have decided on integer math handled in a special way to preserve high orders of equivalent decimal data without these random issues... the results... much smoother audio quality... in many people's opinion... and definitely much higher performance.
It's a very tricky subject and does require you to dive deep in at the assembly level to see these issues... it seems many programmers do not code at this level anymore and many don't seem to question these issues at all, and simply make generic function calls to generic code libraries and accept the results they are handed... or simply repeat something they read in a magazine or programming book... I have tested and tested and made my decisions based on thousands of hours of trying many different techniques... SAWStudio is the result... but it is also well known that my views are not the generally accepted views of the industry... oh well...![]()
Bob L
I have tripped over examples in the 15 years of coding SAW where the same math calculation gives different and false results depending on the exact values of the variables in that calculation... one set of numbers results in a round-up of the result and another set results in a round-down... same equation, different numbers... therefore making it very difficult to actually have the code consistently interpret the results and in many cases branch to the wrong code segment based on a zero result not testing TRUE.
I have decided on integer math handled in a special way to preserve high orders of equivalent decimal data without these random issues... the results... much smoother audio quality... in many people's opinion... and definitely much higher performance.
It's a very tricky subject and does require you to dive deep in at the assembly level to see these issues... it seems many programmers do not code at this level anymore and many don't seem to question these issues at all, and simply make generic function calls to generic code libraries and accept the results they are handed... or simply repeat something they read in a magazine or programming book... I have tested and tested and made my decisions based on thousands of hours of trying many different techniques... SAWStudio is the result... but it is also well known that my views are not the generally accepted views of the industry... oh well...
Bob L
连bob都认为这个问题不太好解决了，还是内录吧。
现在出来了一个aiso驱动的多客户软件，这回方便了，原来用别的驱动老是不放心。
看这里：http://ch.midifan.com/m_New/DetailView.aspx?id=4570
现在出来了一个aiso驱动的多客户软件，这回方便了，原来用别的驱动老是不放心。
看这里：http://ch.midifan.com/m_New/DetailView.aspx?id=4570