Csound6 Released!

We are delighted to announce that Csound6.00 is now released. Currently only a few builds are on SourceForge (OSX1.8, ioS, Android); but the others will arrive soon. Sources are there as well. Note Csound6 is internally fairly different but it should have maintained backward compatability with pieces; the API is changed. There are many new facilities such as:

  • arrays
  • on-the-fly compilation
  • sample accuracy

Release Notes are below
- John ffitch
Csound version 6 provides significant advances in functionality over version 5, not only for the Csound application, but also for the Csound library and application programming interface (API). Complete backward compatibility has been retained for all orc/sco/csd based works, but changes in the API are not backwardly compatible.

The major changes for Csound 6 are:

  • sample-accurate option provides sample-accurate timing of score events. This breaks backward compatibility for plugin opcodes, which require code changes for tracking current sample frame position in audio type variables (see below for an example).
  • realtime option, which can also be enabled by setting the CSOUND_PARAMS field realtime_mode to 1, provides asynchronous execution of file reading and writing, and of init-time orchestra code. This should make real-time operation of Csound more resistant to dropouts.
  • New opcodes and syntax support arrays of arbitrary dimensionality.
    Arrays are created (usually) with the init opcode or with fillarray:
    k1[] init 4
    generates a k-rate 1-dimensional array of length 4. Similarly,
    a2[][] init 4, 4
    creates a square, 4 x 4 a-rate array. Or,
    k2[] fillarray 1, 2, 3, 4
    creates a 4-element vector filled with {1, 2, 3, 4}, and implicitly
    defines a length. Array elements are indexed with brackets [] such as k1[2] or
    a2[2][3]. One dimensional arrays replace tvars, and can be used in opcodes like
    maxtab/maxarray, mintab/minarray and sumtab/sumarray (see below).
    Array setting can also be done on the left-hand side of opcodes, e.g.:
    aSigs[0] vco2 .1, 440
    aSigs[1] vco2 .1, 880
  • Multicore support has been completely rewritten using an improved algorithm for task dispatching, which should use less memory and fewer locks. Multicore benchmarks are now much faster than before. Multithreading often provides significant speedups over single-threaded rendering.
  • Compilation of instruments and orchestra code can be done at any stage; new instruments can be added or can replace old instruments with the same name/number even while Csound is running.
  • The Csound API has been cleaned up and extended to support new functionality.

New opcodes:
o faustgen
o array — many new or revised opcodes — see below under orchestra.
o compileorc — takes a filename containing a collection if instrument definitions and compiles them, replacing existing versions. It returns 0 on success.
o compilestr — like compileorc, but takes a string of orchestra code.
o readscore — runs the score preprocessor on a string and then schedules new events via the RT event mechanism, returning 0 if successful.

New Gen and Macros:
o Note events can start and end in mid-kcycle. As this is an incompatible change it is only invoked with the command-line option –sample-accurate is specified. Note that this does not work for tied notes, and skipping initialisation is probably not useful.
o Instruments can run at local values of ksmps using setksmps iksmps as in Csound 5 UDOs.
o Compilation can be done at any stage; new instruments can be added or can replace old instruments with the same name/number. Running instances of old instrument definitions are not affected. The only limitation is that header constants in instr 0 are read only once at the time of the first compilation. Init-time code can be placed outside instruments in the global space, but this also will be executed only once following the compilation. In this way, score event generation can be completely replaced by orchestra code. See also the new opcodes compileorc and compilestr.
o New syntax operators +=, -=, *= and /=. These are more than syntactic sugar; please use += and -= for accumulating reverbs as it gives better multicore behaviour.
o The opcodes add, sub, mul and div have been deleted; use the forms + – * /. Not many people were aware of these opcodes.
o Any opcode with zero outputs or one output can be used as a function. Some opcodes might require type annotation to resolve ambiguities; see the the new syntax page in the Csound 6.00 manual.
o A statement can be broken across lines after a , = or any arithmetic operation.
o There are a range of new or recoded operations on k-valued arrays, most restricted to 1 dimensional arrays (vectors):
kans minarray ktab returns the smallest value in the
(possibly) multidimensional array
kans maxarray ktab returns its largest value
kabs sumarray ktab returns sum of all values in the array
ktab genarray imin, imax[, inc]
generates vector of values from imin
to imax by increments of inc (default 1)
ktab2 maparray ktab1, “sin” maps the k-rate 1-arg function in
the string to every element of the vector
ktab2 maparray_i ktab1, “sin” maps the i-rate 1-arg function
in the string to every element of the vector
ktab2 slicearray ktab1, istart, iend
returns a slice of ktab1 from ktab1[istart]
to ktab1[iend]
copyf2array ktab, kfn copies data from an ftable to a vector
copya2ftab ktab, kfn copies data from a vector to an ftable.
o Arithmetic on arrays is allowed. In particular addition, subtraction, multiplication, and division are provided in arithmetic syntax on vectors. Similar operations between arrays and scalars are also allowed.
o Each instance of every instrument has a scratchpad of 4 values that persists after turnoff; this allows values to carry to next use of the instrument; this may be useful for legato etc.
o If a table number is given as -1 then an internal sine wave equivalent to “f. 0 16384 10 1″ is used. Attempts to write to this table will give unpredictable results, but is not policed. The 16384 can be change by command line option –sine-size=# where the # is rounded up to a power of two.
o A number of oscil opcodes now have the f-table parameter as optional, defaulting to the internal sine wave. (oscil1, oscil1i, oscil, oscil3, oscili, foscil, foscil1, loscil, loscil3, poscil, poscil3)
o Score lines can have multiple strings.
o Change to escape characters in score strings — they do not happen.
o Also see readscore opcode

Modified Opcodes and Gens:
o The k() function can take an a-rate argument, in which case it invokes a call to the downsamp opcode.

o Hetro/adsyn analysis files can be machine byte-order independent if created with -X. The down side is a longer file and a little slower loading. The het_export utility will create the independent format from the old, and het_import is no longer necessary.
o cvanal and lpanal will produce machine independent files if the -X option is used. The convolve and lpread etc. opcodes will accept either format. You are encouraged to use the machine independent form. Analysis files produced with -X can be used on other systems.

o CsoundQt has been ported to Csound 6.
o CsoundVST has been ported to Csound 6.
o A new Csound6 app for Android has been developed, extending the Csound 5 app with:
(1) An improved user interface, including text editing, scrolling performance messages, a built-in User Guide, and links to the Csound manual and Web site.
(2) Plugin opcodes, including the signal flow graph opcodes, the FluidSynth opcodes, and the Lua opcodes, which enable executing arbitrary Lua coded in Csound using the very fast LuaJIT engine, even writing opcodes in Lua, and calling any public C function from Csound using LuaJIT’s FFI.

General usage:
o The environment variables for module directories have been renamed OPCODE6DIR64 or OPCODE6DIR (note the 6) so they can co-exist with directories for Csound 5 modules.
o Similarly, .csoundrc has been renamed .csound6rc.

Bugs fixed:
o Too many to record!

New configuration/parameter setting functions:
PUBLIC int csoundSetOption(CSOUND *csound, char *option);
PUBLIC void csoundSetParams(CSOUND *csound, CSOUND_PARAMS *p);
PUBLIC void csoundGetParams(CSOUND *csound, CSOUND_PARAMS *p);
PUBLIC void csoundSetOutput(CSOUND *csound, char *name, char *type, char *format);
PUBLIC void csoundSetInput(CSOUND *csound, char *name);
PUBLIC void csoundSetMIDIInput(CSOUND *csound, char *name);
PUBLIC void csoundSetMIDIFileInput(CSOUND *csound, char *name);
PUBLIC void csoundSetMIDIOutput(CSOUND *csound, char *name);
PUBLIC void csoundSetMIDIFileOutput(CSOUND *csound, char *name);

New parsing/compilation functions:
PUBLIC TREE *csoundParseOrc(CSOUND *csound, char *str);
PUBLIC int csoundCompileTree(CSOUND *csound, TREE *root);
PUBLIC int csoundCompileOrc(CSOUND *csound, const char *str);
PUBLIC int csoundReadScore(CSOUND *csound, char *str);
PUBLIC int csoundCompileArgs(CSOUND *, int argc, char **argv);

New function for starting csound after the first compilation:
PUBLIC int csoundStart(CSOUND *csound);

New threadsafe getters/setters for the software bus:
PUBLIC MYFLT csoundGetControlChannel(CSOUND *csound, const char *name);
PUBLIC void csoundSetControlChannel(CSOUND *csound, const char *name, MYFLT val);
PUBLIC void csoundGetAudioChannel(CSOUND *csound, const char *name, MYFLT *samples);
PUBLIC void csoundSetAudioChannel(CSOUND *csound, const char *name, MYFLT *samples);
PUBLIC void csoundSetStringChannel(CSOUND *csound, const char *name, char *string);
PUBLIC void csoundGetStringChannel(CSOUND *csound, const char *name, char *string);

New threadsafe copy functions for tables:
PUBLIC void csoundTableCopyOut(CSOUND *csound, int table, MYFLT *dest);
PUBLIC void csoundTableCopyIn(CSOUND *csound, int table, MYFLT *src);

The entire Csound API has been made threadsafe so that performance and control can occur in separate threads (after a call to csoundStart()
or csoundCompile()). Thread safety has been implemented by:

o Using atomic operations for reading and writing in control channels,
o Using spinlocks in audio and string channels,
o Using mutexes to protect compilation, score events, and table access.

The Csound API and SDK for iOS has been changed to use Automatic Resource Counting (ARC). All projects using Csound 6 for iOS will require the use of ARC. The Csound for iOS Examples project shows how to use ARC compliant code with Csound. Users with existing Csound 5-based iOS projects can use the “Edit->Refactor->Convert to Objective-C ARC…” wizard to help with updating their projects to use ARC.

For more information about ARC, consult the “Transitioning to ARC Release Notes” on Apple’s Developer website.

The Csound API and SDK for Android has been changed to use the new Csound API.
System changes:
o in Linux and OSX, locales are now thread-safe and local.

Internal changes:
o The build system now uses CMake (instead of SCons as in Csound 5).
o A number of table access opcodes have been rewritten, but should behave the same.
o Similarly diskin and diskin2 now use the same code, so that diskin should be more stable
o The old parser has been completely removed.
o New internal functions in CSOUND:
void (*FlushCircularBuffer)(CSOUND *, void *);
void *(*FileOpenAsync)(CSOUND *, void *, int, const char *, void *,
const char *, int, int, int);
unsigned int (*ReadAsync)(CSOUND *, void *, MYFLT *, int);
unsigned int (*WriteAsync)(CSOUND *, void *, MYFLT *, int);
int (*FSeekAsync)(CSOUND *, void *, int, int);
char *(*GetString)(CSOUND *, MYFLT);
Extract a string originating from a score-event argument.
o Functions removed from CSOUND:
void *(*FileOpen)(CSOUND *,
void*, int, const char*, void*, const char*);
o The “private” parts of the API have been changed considerably.
Also structures like EVTBLK have been changed.
o The LINKAGE1/FLINKAGE1 macros have been renamed to
The template for arate perf-pass opcodes is:
int perf_myopcode(CSOUND *csound, MYOPCODE *p)
uint32_t offset = p->h.insdshead->ksmps_offset;
uint32_t early = p->h.insdshead->ksmps_no_end;
uint32_t nsmps = CS_KSMPS;

if (UNLIKELY(offset)) memset(p->res, ”, offset*sizeof(MYFLT));
if (UNLIKELY(early)) {
nsmps -= early;
memset(&p->res[nsmps], ”, early*sizeof(MYFLT));
for (n=offset; nres[n] = ….
return OK;

Csound’s string variable type has been re-implemented.

The OENTRY structure has been changed. It has a new dependency field. This field is now required for multicore semantics. Setting this field to -1 safely disables parallel execution, but it is recommended to enable parallel execution by using the flags defined in include/interlocks.h

If your opcode reads or writes to zak space, use ZR or ZW, or ZB for both. Similarly, if it reads or writes tables, use TR, TW or TB. If it reads or writes to channels, use CR, CW and CB. If it uses the stack, use SK. If it outputs text, then use WR to stop mixing of output. The _QQ flag marks an opcode as deprecated. Other flags may be added later.

All opcodes that touch audio should take note of sample-accurate code. A number of previous API functions have been removed. OpenFile and OpenFile2 have both been replaced by new OpenFile2 with an additional argument. Additions have been made for arg type specifications for opcodes.
o Any-types have been added, as follows:
* ‘.’ signifies a required arg of any-type
* ‘?’ signifies an optional arg of any-type
* ‘*’ signifies a var-arg list of any-type
o Arrays are now specified using “[x]” where x is a type-specifier. The type-specifier can be any of the of the current specifiers, including any-types. See Opcodes/arrays.c for examples of usage.

A new type system has been added to Csound6, and significant changes have been made to the compiler. The previous system for handling types involved reading the first letter of a variable’s name every time the variable was used to determine its type. This meant there was much re-checking of types. Also, adding new types was difficult, because much custom code had to be updated to check for new type letters.

In Csound6, a separate system of types has been added. Types are defined as CS_TYPEs. The creation of variables from types and the initialisation of memory has been encapsulated within the CS_TYPEs. This change allows easier addition of new types, as well as generic calculations of memory pools, amongst other things.

The compiler has been changed to use the new type system in its semantic checking phase. Variables are now registered into a CS_VAR_POOL when they are first defined, with the CS_VARIABLE having a reference to its CS_TYPE. After first time definition within the pool, the type information is looked up in consequent variable uses, rather than re-calculated from the variable name. This opens up possibilities for new variable naming and typing strategies, i.e. using “myVar:K” to denote a k-rate arg. It also opens up possibilities for user-defined types, such as “data myType kval, aval”, then using “myVar:myType” to define a var of that type. (The previous is speculative, and is not an active proposal at this time.)

The addition of the type system has formalised the static type system that has existed in Csound prior to Csound6. It has, arguably, simplified the code-base in terms of type handling, as well as laid groundwork for future type-related research to be integrated into Csound.