Identifiers follow the same rules as C identifiers.
Labels consist of an identifier plus a colon. E.g.
```
myLabel:
mov r0, r1
```
Procedures are delimited using the `.proc` and `.end` directives. E.g.
```
.proc normalize3
dp4 r15, r8, r8
rsq r15, r15
mul r8, r15, r8
.end
```
Instructions consist of an opcode name and a comma-delimited list of arguments.
Directives are special statements that start with a period and control certain aspects of `picasso`'s code emission; such as defining procedures, uniforms, constants and more.
PICA200 registers are often used as arguments to instructions. There exist the following registers:
All registers contain floating point vectors (it is currently unknown whether they are 24-bit or 32-bit); except for integer vector uniforms (containing 8-bit integers) and boolean uniforms. Vectors have 4 components: x, y, z and w. The components may alternatively be referred to as r, g, b and a (respectively); or s, t, p and q (respectively). Uniforms are special registers that are writable by the CPU; thus they are used to pass configuration parameters to the shader such as transformation matrices. Sometimes they are preloaded with constant values that may be used in the logic of the shader.
In most situations, vectors may be [swizzled](http://en.wikipedia.org/wiki/Swizzling_%28computer_graphics%29), that is; their components may be rearranged. Register arguments support specifying a swizzling mask: `r0.wwxy`. The swizzling mask usually has 4 components (but not more), if it has less the last component is repeated to fill the mask. The default mask applied to registers is `xyzw`; that is, identity (no effect).
Output parameters have an output mask instead of a swizzling mask. This allows the shader to write to some components of a register without affecting the others. In `picasso`, the output mask is parsed exactly the same way as the swizzling mask, enabling write access for the components that are used in it. By default it is also `xyzw`; that is, writing to all components.
Registers may also be assigned additional names in order to make the code more legible. These additional names are called aliases. Aliases may also contain a swizzling mask; if a swizzling mask is applied to an alias afterwards the masks are combined. For example, provided that `someAlias` is an alias for `c0.wyxz`, `someAlias.xxww` would be equivalent to `c0.wwzz`. Aliases may be created by several directives which reserve certain kinds of registers.
For convenience, registers may be addressed using an offset from a known register. This is called indexing. For example, `c8[4]` is equivalent to `c12`; and `r4[-2]` is equivalent to `r2`. Indexing is useful for addressing arrays of registers (such as matrices).
Some source operands of instructions (called SRC1) support relative addressing. This means that it is possible to use one of the three built-in indexing registers (`a0`, `a1` and `a2` aka `lcnt`) to address a register, e.g. `someArray[lcnt]`. Adding an offset is also supported, e.g. `someArray[lcnt+2]`. This is useful in FOR loops.
In geometry shaders, `b15` is automatically set to true *after* each execution of the geometry shader. This can be useful to detect whether program state should be initialized - GPU management code usually resets all unused boolean uniforms to false when setting up the PICA200's shader processing units.
`picasso` takes one or more source code files, and assembles them into a single `.shbin` file. A DVLE object is generated for each source code file, unless the `.nodvle` directive is used (see below). Procedures are shared amongst all source code files, and they may be defined and called wherever. Uniform space for vertex shaders is also shared, that is, if two vertex shader source code files declare the same uniform, they are assigned the same location. Geometry shaders however do not share uniforms, and each geometry shader source code file will have its own uniform allocation map. On the other hand, constants are never shared, and the same space is reused for the constants of each DVLE. Outputs and aliases are, by necessity, never shared either.
Uniforms that start with the underscore (`_`) character are not exposed in the DVLE table of uniforms. This allows for creating private uniforms that can be internally used to configure the behaviour of shared procedures.
**Note**: Older versions of `picasso` handled geometry shaders in a different way. Specifically, uniform space was shared with vertex shaders and it was possible to use `.gsh` without parameters or `setemit` to flag a DVLE as a geometry shader. For backwards compatibility purposes this functionality has been retained, however its use is not recommended.
## PICA200 Caveats & Errata
The PICA200's shader units have numerous implementation caveats and errata that should be taken into account when designing and writing shader code. Some of these include:
- Certain flow of control statements may not work at the end of another block, including the closing of other nested blocks. picasso detects these situations and automatically inserts padding NOP instructions (unless the `--no-nop` command line flag is used).
- The `mova` instruction is finicky and for instance two consecutive `mova` instructions will freeze the PICA200.
- Only a single input register is able to be referenced reliabily at a time in the source registers of an operand. That is, while specifying the same input register in one or more source registers will behave correctly, specifying different input registers will produce incorrect results. picasso detects this situation and displays an error message.
Allocates new boolean uniforms (or arrays of uniforms) and creates aliases for them that point to the allocated registers. Example:
```
.bool useLight[4]
.bool useRawVertexColor
```
### .constf
```
.constf constName(x, y, z, w)
```
Reserves a new floating-point vector uniform to be preloaded with the specified constant; creates an alias for it that points to the allocated register. Example:
```
.constf floatConsts(0.0, 1.0, -1.0, 3.14159)
```
### .consti
```
.consti constName(x, y, z, w)
```
Reserves a new integer vector uniform to be preloaded with the specified constant; creates an alias for it that points to the allocated register. Example:
Reserves a new array of floating-point vector uniforms to be preloaded with the specified constants; creates an alias for it that points to the first element. Example:
```
; Create an array of two elements
.constfa myArray[]
.constfa (1.0, 2.0, 3.0, 4.0)
.constfa (5.0, 6.0, 7.0, 8.0)
.end
```
Optionally the size of the array may be specified. If a number of elements less than the size is specified, the missing elements are initialized to zero. Example:
```
.constfa myArray[4]
.constfa (1.0, 2.0, 3.0, 4.0)
.constfa (5.0, 6.0, 7.0, 8.0)
; The remaining two elements are vectors full of zeroes.
Wires an output register to a certain output property and (optionally) creates an alias for it called `outName` (specify a dash in order not to create the alias). If no output register is specified it is automatically allocated. The following property names are supported:
-`normalquat` (or `nquat`): Used in fragment lighting, this represents the quaternion associated to the normal vector of the vertex.
-`color` (or `clr`): Represents the color of the outputted vertex. Its format is (R, G, B, xx) where R,G,B are values ranging from 0.0 to 1.0. The fourth component isn't used.
-`texcoord0` (or `tcoord0`): Represents the first texture coordinate, which is always fed to the Texture Unit 0. Only the first two components are used.
-`texcoord0w` (or `tcoord0w`): Represents the third component of the first texture coordinate, used for 3D/cube textures.
-`texcoord1` (or `tcoord1`): Similarly to `texcoord0`, this is the second texture coordinate, which is usually but not always fed to Texture Unit 1.
-`texcoord2` (or `tcoord2`): Similarly `texcoord0`, this is the third texture coordinate, which is usually but not always fed to Texture Unit 2.
-`view`: Used in fragment lighting, this represents the view vector associated to the vertex. The fourth component is not used.
-`dummy`: Used in vertex shaders to pass generic semanticless parameters to the geometry shader, and in geometry shaders to use the appropriate property type from the output map of the vertex shader, thus 'merging' the output maps.
An output mask that specifies to which components of the output register should the property be wired to is also accepted. If the output register is explicitly specified, it attaches to it (e.g. `o2.xy`); otherwise it attaches to the property name (e.g. `texcoord0.xy`).
Specifies the name of the procedure to use as the entrypoint of the current DVLE. If this directive is not used, `main` is assumed.
### .nodvle
```
.nodvle
```
This directive tells `picasso` not to generate a DVLE for the source code file that is being processed. This allows for writing files that contain shared procedures to be used by other files.
This directive flags the current DVLE as a geometry shader and specifies the geometry shader operation mode, which can be one of the following:
-`point` mode: In this mode the geometry shader is called according to the input stride and input permutation configured by the user. On entry, the data is stored starting at the `v0` register. This type of geometry shader can be used with both array-drawing mode (aka `C3D_DrawArrays`) and element-drawing mode (aka `C3D_DrawElements`).
-`variable` mode (also called `subdivision` mode): In this mode the geometry shader processes variable-sized primitives, which are required to have `vtxNum` vertices for which full attribute information will be stored, and **one or more** additional vertices for which only position information will be stored. On entry the register `c0` stores in all its components the total number of vertices of the primitive, and subsequent registers store vertex information in order. This type of geometry shader can only used with element-drawing mode - inside the index array each primitive is prefixed with the number of vertices in it.
-`fixed` mode (also called `particle` mode): In this mode the geometry shader processes fixed-size primitives, which always have `vtxNum` vertices. On entry, the array of vertex information will be stored starting at the float uniform register `arrayStartReg`. This type of geometry shader can only used with element-drawing mode.
The `firstReg` parameter specifies the first float uniform register that is available for use in float uniform register allocation (this is especially useful in variable and fixed mode).
Similar to `.constf`, this directive adds a DVLE constant entry for the specified floating-point vector uniform register to be loaded with the specified value. This is useful in order to instantiate a generalized shared procedure with the specified parameters.
### .seti
```
.seti register(x, y, z, w)
```
Similar to `.consti`, this directive adds a DVLE constant entry for the specified integer vector uniform register to be loaded with the specified value. This is useful in order to instantiate a generalized shared procedure with the specified parameters.
### .setb
```
.setb register value
```
This directive adds a DVLE constant entry for the specified boolean uniform register to be loaded with the specified value (which may be `true`, `false`, `on`, `off`, `1` or `0`). This is useful in order to control the flow of a generalized shared procedure.
-`mad`, which takes three source operands, has two forms: the first is narrow-wide-narrow, and the second is narrow-narrow-wide. This is also detected automatically.
-`idxReg`: Represents an indexing register to write to using the mova instruction. Can be `a0`, `a1` or `a01` (the latter writes to both `a0` and `a1`).
-`opx` and `opy`: They represent a conditional operator that is applied to the source registers and whose result is stored in the appropriate flag (`cmp.x` and `cmp.y` respectively). Supported values include:
-`condExp`: Represents a conditional expression, which uses the conditional flags `cmp.x` and `cmp.y` set by the CMP instruction. These flags may be negated using the `!` symbol, e.g. `!cmp.x`. The conditional expression can take any of the following forms:
-`flag1`: It tests a single flag.
-`flag1 && flag2`: It performs AND between the two flags. Optionally, a single `&` may be specified.
-`flag1 || flag2`: It performs OR between the two flags. Optionally, a single `|` may be specified.
