mirror/sointu
1
0
Fork 0
mirror of https://github.com/vsariola/sointu.git synced 2025-06-04 01:28:45 -04:00
Code Issues Packages Projects Releases Wiki Activity

Fix builds and tests to pass on Linux.

Browse Source 
Builds on both 32-bit and 64-bit executables and all tests (except gm.dls stuff obviously, which was excluded) pass on 64-bit Linux. Cannot test the 32-bit executables, as WSL does not support running 32-bit.
This commit is contained in:
Veikko 2020-05-26 21:27:53 +03:00 committed by Veikko Sariola
parent b64723323f
commit 5e05057240
4 changed files with 87 additions and 62 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

113
README.md
View File

@ -5,28 +5,28 @@ A cross-platform modular software synthesizer for small intros, forked from
Summary Summary
------- -------
Sointu is work-in-progress. It is a fork and an evolution of [4klang]( Sointu is work-in-progress. It is a fork and an evolution of [4klang](
https://github.com/hzdgopher/4klang), a modular software synthesizer intended https://github.com/hzdgopher/4klang), a modular software synthesizer intended
to easily produce music for 4k intros-small executables with a maximum to easily produce music for 4k intros-small executables with a maximum
filesize of 4096 bytes containing realtime audio and visuals. Like 4klang, the filesize of 4096 bytes containing realtime audio and visuals. Like 4klang, the
sound is produced by a virtual machine that executes small bytecode to sound is produced by a virtual machine that executes small bytecode to
produce the audio; however, by now the internal virtual machine has been produce the audio; however, by now the internal virtual machine has been
heavily rewritten and extended to make the code more maintainable, possibly heavily rewritten and extended to make the code more maintainable, possibly
even saving some bytes in the process. even saving some bytes in the process.
New features since fork New features since fork
----------------------- -----------------------
- **Per instrument polyphonism**. An instrument has the possibility to - **Per instrument polyphonism**. An instrument has the possibility to
have any number of voices, meaning in practice that multiple voices can have any number of voices, meaning in practice that multiple voices can
reuse the same opcodes. Done, see [here](tests/test_polyphony.asm) for an reuse the same opcodes. Done, see [here](tests/test_polyphony.asm) for an
example and [here](src/opcodes/flowcontrol.asm) for the implementation. The example and [here](src/opcodes/flowcontrol.asm) for the implementation. The
maximum total number of voices will be 32: you can have 32 monophonic maximum total number of voices will be 32: you can have 32 monophonic
instruments or any combination of polyphonic instruments adding up to 32. instruments or any combination of polyphonic instruments adding up to 32.
- **Any number of voices per track**. For example, a polyphonic instrument of - **Any number of voices per track**. For example, a polyphonic instrument of
3 voices can be triggered by 3 parallel tracks, to produce chords. But one 3 voices can be triggered by 3 parallel tracks, to produce chords. But one
track can also trigger 3 voices, for example when using arpeggio. A track track can also trigger 3 voices, for example when using arpeggio. A track
can even trigger 2 voices of different instruments, alternating between can even trigger 2 voices of different instruments, alternating between
these two; maybe useful for example as an easy way to alternate between an these two; maybe useful for example as an easy way to alternate between an
open and a closed hihat. open and a closed hihat.
- **Easily extensible**. Instead of %ifdef hell, the primary extension - **Easily extensible**. Instead of %ifdef hell, the primary extension
mechanism will be through new opcodes for the virtual machine. Only the mechanism will be through new opcodes for the virtual machine. Only the
@ -41,32 +41,32 @@ New features since fork
in future, there will be no need for binary format to save patches: the .asm in future, there will be no need for binary format to save patches: the .asm
is easy enough to be imported into / exported from the GUI. Being a text is easy enough to be imported into / exported from the GUI. Being a text
format, the .asm based patch definitions play nicely with source control. format, the .asm based patch definitions play nicely with source control.
- **Harmonized support for stereo signals**. Every opcode supports a stereo - **Harmonized support for stereo signals**. Every opcode supports a stereo
variant: the stereo bit is hidden in the least significant bit of the variant: the stereo bit is hidden in the least significant bit of the
command stream and passed in carry to the opcode. This has several nice command stream and passed in carry to the opcode. This has several nice
advantages: 1) the opcodes that don't need any parameters do not need an advantages: 1) the opcodes that don't need any parameters do not need an
entire byte in the value stream to define whether it is stereo; 2) stereo entire byte in the value stream to define whether it is stereo; 2) stereo
variants of opcodes can be implemented rather efficiently; in many cases, variants of opcodes can be implemented rather efficiently; in many cases,
the extra cost of stereo variant is only 7 bytes, of which 4 are zeros, so the extra cost of stereo variant is only 7 bytes, of which 4 are zeros, so
should compress quite nicely. 3) Since stereo opcodes usually follow stereo should compress quite nicely. 3) Since stereo opcodes usually follow stereo
opcodes (and mono opcodes follow mono opcodes), the stereo bits of the opcodes (and mono opcodes follow mono opcodes), the stereo bits of the
command bytes will be highly correlated and if crinkler or any other command bytes will be highly correlated and if crinkler or any other
modeling compressor is doing its job, that should make them highly modeling compressor is doing its job, that should make them highly
predictable i.e. highly compressably. Done. predictable i.e. highly compressably. Done.
- **Test-driven development**. Given that 4klang was already a mature project, - **Test-driven development**. Given that 4klang was already a mature project,
the first thing actually implemented was a set of regression tests to avoid the first thing actually implemented was a set of regression tests to avoid
breaking everything beyond any hope of repair. Done, using CTest. breaking everything beyond any hope of repair. Done, using CTest.
- **New units**. Bit-crusher, gain, inverse gain, clip, modulate bpm - **New units**. Bit-crusher, gain, inverse gain, clip, modulate bpm
(proper triplets!), compressor (can be used for side-chaining)... As (proper triplets!), compressor (can be used for side-chaining)... As
always, if you don't use them, they won't be compiled into the code. always, if you don't use them, they won't be compiled into the code.
- **Arbitrary signal routing**. SEND (used to be called FST) opcode normally - **Arbitrary signal routing**. SEND (used to be called FST) opcode normally
sends the signal as a modulation to another opcode. But with the new sends the signal as a modulation to another opcode. But with the new
RECEIVE opcode, you just receive the plain signal there. So you can connect RECEIVE opcode, you just receive the plain signal there. So you can connect
signals in an arbitrary way. Actually, 4klang could already do this but in signals in an arbitrary way. Actually, 4klang could already do this but in
a very awkward way: it had FLD (load value) opcode that could be modulated; a very awkward way: it had FLD (load value) opcode that could be modulated;
FLD 0 with modulation basically achieved what RECEIVE does, except that FLD 0 with modulation basically achieved what RECEIVE does, except that
RECEIVE can also handle stereo signals. Additionally, we have OUTAUX, AUX RECEIVE can also handle stereo signals. Additionally, we have OUTAUX, AUX
and IN opcodes, which route the signals through global main or aux ports, and IN opcodes, which route the signals through global main or aux ports,
more closer to how 4klang does. But this time we have 8 mono ports / 4 more closer to how 4klang does. But this time we have 8 mono ports / 4
stereo ports, so even this method of routing is unlikely to run out of ports stereo ports, so even this method of routing is unlikely to run out of ports
in small intros. in small intros.
@ -74,25 +74,28 @@ New features since fork
- **Sample-based oscillators, with samples imported from gm.dls**. Reading - **Sample-based oscillators, with samples imported from gm.dls**. Reading
gm.dls is obviously Windows only, but the sample mechanism can be used also gm.dls is obviously Windows only, but the sample mechanism can be used also
without it, in case you are working on a 64k and have some kilobytes to without it, in case you are working on a 64k and have some kilobytes to
spare. See [this example](tests/test_oscillat_sample.asm), and this Python spare. See [this example](tests/test_oscillat_sample.asm), and this Python
[script](scripts/parse_gmdls.py) parses the gm.dls file and dumps the [script](scripts/parse_gmdls.py) parses the gm.dls file and dumps the
sample offsets from it. sample offsets from it.
- **Unison oscillators**. Multiple copies of the oscillator running sligthly - **Unison oscillators**. Multiple copies of the oscillator running sligthly
detuned and added up to together. Great for trance leads (supersaw). Unison detuned and added up to together. Great for trance leads (supersaw). Unison
of up to 4, or 8 if you make stereo unison oscillator and add up both left of up to 4, or 8 if you make stereo unison oscillator and add up both left
and right channels. See [this example](tests/test_oscillat_unison.asm). and right channels. See [this example](tests/test_oscillat_unison.asm).
- **Supports 32 and 64 bit builds**. The 64-bit version is done with minimal - **Supports 32 and 64 bit builds**. The 64-bit version is done with minimal
changes to get it work, mainly for the future prospect of running the MIDI changes to get it work, mainly for the future prospect of running the MIDI
instrument in 64-bit mode. All the tests are passing so it seems to work. instrument in 64-bit mode. All the tests are passing so it seems to work.
- **Supports both Windows and Linux**. Currently, all the tests are compiling
on Windows and Linux, both 32-bit and 64-bit, and the tests are passing on
64-bit Linux, tested on WSL. 32-bit executables don't run on WSL, so those
remain to be tested.
Future goals Future goals
------------ ------------
- **Cross-platform support for win / mac / linux**. The build is already based - **Support for mac**. This should be rather easy, as all the macros are
on CMake and compiles on Windows. Cross-platform NASM/YASM macros have been designed for cross-platform support, but as I don't have a mac, I cannot
drafted and remain to be tested. Once the project is more mature, I will test this.
try compiling on other platforms. - **Find a more general solution for skipping opcodes / early outs**. It's
- **Find a more general solution for skipping opcodes / early outs**. It's
probably a new opcode "skip" that skips from the opcode to the next out in probably a new opcode "skip" that skips from the opcode to the next out in
case the signal entering skip and the signal leaving out are both close to case the signal entering skip and the signal leaving out are both close to
zero. zero.
@ -102,22 +105,22 @@ Future goals
on pretty much any computer. The only thing needed is to be able to on pretty much any computer. The only thing needed is to be able to
communicate with the platform specific synth; for this, the best communicate with the platform specific synth; for this, the best
option seems to be to run the synth inside a tiny websocket server that option seems to be to run the synth inside a tiny websocket server that
receives messages from browser and streams the audio to the browser. receives messages from browser and streams the audio to the browser.
The feasibility of the approach is proven (localhost websocket calls The feasibility of the approach is proven (localhost websocket calls
have 1 ms range of latency), but nothing more is done yet. have 1 ms range of latency), but nothing more is done yet.
Nice-to-have ideas Nice-to-have ideas
------------------ ------------------
- **Tracker**. If the list of primary goals is ever exhausted, a browser-based - **Tracker**. If the list of primary goals is ever exhausted, a browser-based
tracker would be nice to take advantage of all the features. tracker would be nice to take advantage of all the features.
Anti-goals Anti-goals
---------- ----------
- **Ability to run Sointu as a DAW plugin (VSTi, AU, LADSPA and DSSI...)**. - **Ability to run Sointu as a DAW plugin (VSTi, AU, LADSPA and DSSI...)**.
None of these plugin technologies are cross-platform and they are full of None of these plugin technologies are cross-platform and they are full of
proprietary technologies. In particular, since Sointu was initiated after proprietary technologies. In particular, since Sointu was initiated after
Steinberg ceased to give out VSTi2 licenses, there is currently no legal or Steinberg ceased to give out VSTi2 licenses, there is currently no legal or
easy way to compile it as a VSTi2 plugin. I downloaded the VSTi3 API and, easy way to compile it as a VSTi2 plugin. I downloaded the VSTi3 API and,
nope, sorry, I don't want to spend my time on it. And Renoise supports only nope, sorry, I don't want to spend my time on it. And Renoise supports only
VSTi2... There is [JUCE](https://juce.com/), but it is again a mammoth and VSTi2... There is [JUCE](https://juce.com/), but it is again a mammoth and
@ -130,17 +133,17 @@ Design philosophy
- Try to avoid %ifdef hell as much as possible. If needed, try to include all - Try to avoid %ifdef hell as much as possible. If needed, try to include all
code toggled by a define in one block. code toggled by a define in one block.
- Instead of prematurely adding %ifdef toggles to optimize away unused - Instead of prematurely adding %ifdef toggles to optimize away unused
features, start with the most advanced featureset and see if you can features, start with the most advanced featureset and see if you can
implement it in a generalized way. For example, all the modulations are implement it in a generalized way. For example, all the modulations are
now added into the values when they are converted from integers, in a now added into the values when they are converted from integers, in a
standardized way. This got rid of most of the %ifdefs in 4klang. Also, with standardized way. This got rid of most of the %ifdefs in 4klang. Also, with
no %ifdefs cluttering the view, many opportunities to shave away no %ifdefs cluttering the view, many opportunities to shave away
instructions became apparent. Also, by making the most advanced synth instructions became apparent. Also, by making the most advanced synth
cheaply available to the scene, we promote better music in future 4ks :) cheaply available to the scene, we promote better music in future 4ks :)
- Size first, speed second. Speed will only considered if the situation - Size first, speed second. Speed will only considered if the situation
becomes untolerable. becomes untolerable.
- Benchmark optimizations. Compression results are sometimes slightly - Benchmark optimizations. Compression results are sometimes slightly
nonintuitive so alternative implementations should always be benchmarked nonintuitive so alternative implementations should always be benchmarked
e.g. by compiling and linking a real-world song with [Leviathan](https://github.com/armak/Leviathan-2.0) e.g. by compiling and linking a real-world song with [Leviathan](https://github.com/armak/Leviathan-2.0)
and observing how the optimizations and observing how the optimizations
@ -149,24 +152,24 @@ Design philosophy
Background and history Background and history
---------------------- ----------------------
[4klang](https://github.com/hzdgopher/4klang) development was started in 2007 [4klang](https://github.com/hzdgopher/4klang) development was started in 2007
by Dominik Ries (gopher) and Paul Kraus (pOWL) of Alcatraz. The [write-up]( by Dominik Ries (gopher) and Paul Kraus (pOWL) of Alcatraz. The [write-up](
http://zine.bitfellas.org/article.php?zine=14&id=35) will still be helpful for http://zine.bitfellas.org/article.php?zine=14&id=35) will still be helpful for
anyone looking to understand how 4klang and Sointu use the FPU stack to anyone looking to understand how 4klang and Sointu use the FPU stack to
manipulate the signals. Since then, 4klang has been used in countless of scene manipulate the signals. Since then, 4klang has been used in countless of scene
productions and people use it even today. productions and people use it even today.
However, 4klang is pretty deep in the [%ifdef hell](https://www.cqse.eu/en/blog/living-in-the-ifdef-hell/), However, 4klang is pretty deep in the [%ifdef hell](https://www.cqse.eu/en/blog/living-in-the-ifdef-hell/),
and the polyphonism was never implemented in a very well engineered way (you and the polyphonism was never implemented in a very well engineered way (you
can have exactly 2 voices per instrument if you enable it). Also, reading can have exactly 2 voices per instrument if you enable it). Also, reading
through the code, I spotted several avenues to squeeze away more bytes. These through the code, I spotted several avenues to squeeze away more bytes. These
observations triggered project Sointu. That, and I just wanted to learn x86 observations triggered project Sointu. That, and I just wanted to learn x86
assembly, and needed a real-world project to work on. assembly, and needed a real-world project to work on.
Credits Credits
------- -------
The original 4klang was developed by Dominik Ries ([gopher](https://github.com/hzdgopher/4klang)) and Paul Kraus The original 4klang was developed by Dominik Ries ([gopher](https://github.com/hzdgopher/4klang)) and Paul Kraus
(pOWL) of Alcatraz. (pOWL) of Alcatraz.
Sointu was initiated by Veikko Sariola (pestis/bC!). Sointu was initiated by Veikko Sariola (pestis/bC!).

9
src/sointu.inc
View File

@ -27,11 +27,18 @@
%define MANGLE_DATA(d) d %define MANGLE_DATA(d) d
%endif %endif
%ifidn __OUTPUT_FORMAT__,elf32 %ifidn __OUTPUT_FORMAT__,elf
; on linux, function f with n parameters is mangled as "f" ; on linux, function f with n parameters is mangled as "f"
%define MANGLE_FUNC(f,n) f %define MANGLE_FUNC(f,n) f
; On linux, data label d is mangled as "d" ; On linux, data label d is mangled as "d"
%define MANGLE_DATA(d) d %define MANGLE_DATA(d) d
%assign BITS 32
%endif
%ifidn __OUTPUT_FORMAT__,elf64
%define MANGLE_FUNC(f,n) f
%define MANGLE_DATA(d) d
%assign BITS 64
%endif %endif
%ifidn __OUTPUT_FORMAT__,macho32 %ifidn __OUTPUT_FORMAT__,macho32

6
tests/CMakeLists.txt
View File

@ -91,8 +91,10 @@ regression_test(test_oscillat_trisaw ENVELOPE)
regression_test(test_oscillat_pulse ENVELOPE VCO_PULSE) regression_test(test_oscillat_pulse ENVELOPE VCO_PULSE)
regression_test(test_oscillat_gate ENVELOPE) regression_test(test_oscillat_gate ENVELOPE)
regression_test(test_oscillat_stereo ENVELOPE) regression_test(test_oscillat_stereo ENVELOPE)
regression_test(test_oscillat_sample ENVELOPE) if(WIN32) # The samples are currently only GMDLs based, and thus require Windows.
regression_test(test_oscillat_sample_stereo ENVELOPE) regression_test(test_oscillat_sample ENVELOPE)
regression_test(test_oscillat_sample_stereo ENVELOPE)
endif()
regression_test(test_oscillat_unison ENVELOPE) regression_test(test_oscillat_unison ENVELOPE)
regression_test(test_oscillat_unison_stereo ENVELOPE) regression_test(test_oscillat_unison_stereo ENVELOPE)
regression_test(test_oscillat_lfo "ENVELOPE;VCO_SINE;VCO_PULSE;FOP_MULP2") regression_test(test_oscillat_lfo "ENVELOPE;VCO_SINE;VCO_PULSE;FOP_MULP2")

21
tests/test_renderer.c
View File

@ -2,6 +2,7 @@
#include <stdlib.h> #include <stdlib.h>
#include <sys/types.h> #include <sys/types.h>
#include <sys/stat.h> #include <sys/stat.h>
#include <stdbool.h>
#if defined (_WIN32) #if defined (_WIN32)
#include <windows.h> #include <windows.h>
@ -12,7 +13,19 @@
#include <math.h> #include <math.h>
extern void __stdcall su_render(); #include <stdint.h>
#if UINTPTR_MAX == 0xffffffff // are we 32-bit?
#if defined(__clang__) || defined(__GNUC__)
#define CALLCONV __attribute__ ((stdcall))
#elif defined(_WIN32)
#define CALLCONV __stdcall // on 32-bit platforms, we just use stdcall, as all know it
#endif
#else // 64-bit
#define CALLCONV // the asm will use honor honor correct x64 ABI on all 64-bit platforms
#endif
extern void CALLCONV su_render(void *);
extern int su_max_samples; extern int su_max_samples;
int main(int argc, char* argv[]) { int main(int argc, char* argv[]) {
@ -25,7 +38,7 @@ int main(int argc, char* argv[]) {
char actual_output_folder[] = "actual_output/"; char actual_output_folder[] = "actual_output/";
long fsize; long fsize;
long bufsize; long bufsize;
boolean small_difference; bool small_difference;
double diff; double diff;
#ifndef SU_USE_16BIT_OUTPUT #ifndef SU_USE_16BIT_OUTPUT
float* buf = NULL; float* buf = NULL;
@ -84,7 +97,7 @@ int main(int argc, char* argv[]) {
fread((void*)filebuf, su_max_samples * 2, sizeof(*filebuf), f); fread((void*)filebuf, su_max_samples * 2, sizeof(*filebuf), f);
small_difference = FALSE; small_difference = false;
for (n = 0; n < su_max_samples * 2; n++) { for (n = 0; n < su_max_samples * 2; n++) {
diff = (double)(buf[n]) - (double)(filebuf[n]); diff = (double)(buf[n]) - (double)(filebuf[n]);
@ -97,7 +110,7 @@ int main(int argc, char* argv[]) {
goto fail; goto fail;
} }
else if (diff > 0.0f) { else if (diff > 0.0f) {
small_difference = TRUE; small_difference = true;
} }
} }