The SHBIN (SHader BINary) file is used to contain compiled and linked shader programs. These can include vertex shaders (typically compiled from .vsh files) and geometry shaders (typically compiled from .gsh files, though .asm have been observed). In commercial applications, SHBIN files can be found as standalone files with the extension .shbin, or contained within .bcsdr files. BCSDR files use CGFX as a container, but the underlying DVLB/DVLP/DVLE structure remains unchanged.
A SHBIN's structure starts with a generic header (DVLB), then a single program header (DVLP), then DVLE(s). The program header specifies the generic parts of the shader, i.e. the shader program data, the operand descriptor data, and a filename symbol table. The contextual details (entry point, constant values, debug symbols, etc) are specified in an executable header (DVLE). There may be multiple DVLE headers, so in this sense multiple shaders sharing the same program code can be stored in a single SHBIN. Hence for the following note the distinction between "program" and "executable".
For a description of the instruction set, see the following page : Shader Instruction Set
Offset
|
Size
|
Description
|
0x0
|
0x4
|
Magic "DVLB"
|
0x4
|
0x4
|
N = number of DVLEs in SHBIN
|
0x8
|
0x4*N
|
DVLE offset table; each offset is a u32 relative to the start of the DVLB section
|
The DVLP file comes directly after the header.
DVLP
Offset
|
Size
|
Description
|
0x0
|
0x4
|
Magic "DVLP"
|
0x4
|
0x4
|
? (Maybe a version number?)
|
0x8
|
0x4
|
Offset (relative to DVLP start) to the compiled shader binary blob
|
0xC
|
0x4
|
Size of compiled shader binary blob, in words
|
0x10
|
0x4
|
Offset (relative to DVLP start) to shader instruction extension table
|
0x14
|
0x4
|
Number of shader instruction extension table entries (each entry is 8-byte long)
|
0x18
|
0x4
|
Offset (relative to DVLP start) to filename symbol table
|
DVLE
Offset
|
Size
|
Description
|
0x0
|
0x4
|
Magic "DVLE"
|
0x6
|
0x1
|
Shader type (0x0 = vertex shader, 0x1 = geometry shader; might contain other flags)
|
0x8
|
0x4
|
Executable's main offset in binary blob (in words)
|
0xC
|
0x4
|
Executable's program's endmain offset in binary blob (in words)
|
0x18
|
0x4
|
Offset (relative to DVLE start) to constant table
|
0x1C
|
0x4
|
Number of entries in constant table (each entry is 0x14-byte long)
|
0x20
|
0x4
|
Offset (relative to DVLE start) to label table
|
0x24
|
0x4
|
Number of entries in label table (each entry is 0x10-byte long)
|
0x28
|
0x4
|
Offset (relative to DVLE start) to output register table
|
0x2C
|
0x4
|
Number of entries in output register table (each entry is 0x8-byte long)
|
0x30
|
0x4
|
Offset (relative to DVLE start) to uniform table
|
0x34
|
0x4
|
Number of entries in uniform table (each entry is 0x8-byte long)
|
0x38
|
0x4
|
Offset (relative to DVLE start) to symbol table
|
0x3C
|
0x4
|
Size of symbol table (in bytes)
|
Label Table Entry
Offset
|
Size
|
Description
|
0x0
|
0x1
|
Label ID
|
0x4
|
0x4
|
Offset (relative to shader program blob start) to label's location, in words
|
0x8
|
0x4
|
?
|
0xC
|
0x4
|
Offset (relative to DVLE symbol table start) to label's symbol
|
Constant Table Entry
Each executable's constants are stored as in constant uniform table. This information is used by ctrulib's SHDR framework to automatically send those values to the GPU when changing to a given program. An entry is constituted by a header and the constant data, the latter of which uses a format specific to the constant type.
Offset
|
Size
|
Description
|
0x0
|
0x1
|
Constant type (0=bool, 1=ivec4, 2=vec4)
|
0x2
|
0x1
|
Uniform ID
|
Corresponding constant entry formats:
Offset
|
Size
|
Description
|
0x0
|
0x1
|
0x0
|
0x2
|
0x1
|
Uniform bool ID
|
0x4
|
0x1
|
Value (boolean)
|
Offset
|
Size
|
Description
|
0x0
|
0x1
|
0x1
|
0x2
|
0x1
|
Uniform integer vector ID
|
0x4
|
0x1
|
x (u8)
|
0x5
|
0x1
|
y (u8)
|
0x6
|
0x1
|
z (u8)
|
0x7
|
0x1
|
w (u8)
|
Offset
|
Size
|
Description
|
0x0
|
0x1
|
0x2
|
0x2
|
0x1
|
Uniform vector ID
|
0x4
|
0x4
|
x (float24)
|
0x8
|
0x4
|
y (float24)
|
0xC
|
0x4
|
z (float24)
|
0x10
|
0x4
|
w (float24)
|
Output Table Entry
Bit
|
Description
|
0-3
|
Output type (see table below)
|
16-19
|
Register ID
|
32-35
|
Output attribute component mask (e.g. 5=xz)
|
Output types :
ID
|
Description
|
0x0
|
result.position
|
0x1
|
result.normalquat
|
0x2
|
result.color
|
0x3
|
result.texcoord0
|
0x4
|
result.texcoord0w
|
0x5
|
result.texcoord1
|
0x6
|
result.texcoord2
|
0x7
|
?
|
0x8
|
result.view
|
Uniform Table Entry
Offset
|
Size
|
Description
|
0x0
|
0x4
|
Offset (relative to DVLE symbol table start) to variable's symbol
|
0x4
|
0x2
|
Variable start register
|
0x6
|
0x2
|
Variable end register (equal to start register for non-arrays)
|
The register indices refer to a unified register space for non-output registers. The mapping of register index values to registers is the following:
Values
|
Registers
|
0x00-0x0F
|
v0-v15
|
0x10-0x6F
|
c0-c95
|
0x70-0x73
|
i0-i3
|
0x78-0x87
|
b0-b15
|
DVOJ
There is another file format for shaders which starts with the string "DVOJ". This format seems to be used for unlinked shader objects. It seems likely that one or multiple DVOJs can be linked to a DVLB file, similarly to the C compilation model.
Structurally, a DVOJ header captures all information there is about a single shader instance. It uses the same fields like the DVLB, DVLP, and DVLE structures, but also stores two unknown blocks of data. It seems that the entry point of a DVOJ is always the first shader instruction.
All offsets in the following table are given relative to the DVOJ start.
Offset
|
Size
|
Description
|
0x00
|
0x4
|
Magic "DVOJ"
|
0x04
|
0x4
|
Unknown. Seems to be related to the DVLE shader type.
|
0x08
|
0x4
|
Unknown.
|
0x0C
|
0x4
|
Padding? (usually 0xFFFFFFFF)
|
0x10
|
0x4
|
Offset to constant table
|
0x14
|
0x4
|
Number of entries in constant table (each entry is 0x14-byte long)
|
0x18
|
0x4
|
Offset to label table
|
0x1C
|
0x4
|
Number of entries in label table (each entry is 0x10-byte long)
|
0x20
|
0x4
|
Offset to the compiled shader binary blob
|
0x24
|
0x4
|
Size of compiled shader binary blob, in words
|
0x28
|
0x4
|
Offset (relative to DVLP start) to shader instruction extension table
|
0x2C
|
0x4
|
Number of shader instruction extension table entries (each entry is 8-byte long)
|
0x30
|
0x4
|
Offset to unknown block 1
|
0x34
|
0x4
|
Number of items in unknown block 1 (each item is 8-byte long). This seems to be equal to the total number of instructions.
|
0x38
|
0x4
|
Offset to unknown block 2
|
0x3C
|
0x4
|
Number of items in unknown block 2 (each item is 12-byte long). This seems to be equal to the number of instructions taking arguments (i.e. excluding NOP, END, ...)
|
0x40
|
0x4
|
Offset to output register table
|
0x44
|
0x4
|
Number of entries in output register table (each entry is 0x8-byte long)
|
0x48
|
0x4
|
Offset to uniform table
|
0x4C
|
0x4
|
Number of entries in uniform table (each entry is 0x8-byte long)
|
0x50
|
0x4
|
Offset to symbol table
|
0x54
|
0x4
|
Size of symbol table (in bytes)
|
Unknown Block 1 Item
A wild guess is that this denotes shader source line information. Take the information with a grain of salt, though, since it hasn't been backed by any empirical data so far.
The index N of the item within Unknown Block 1 corresponds to the Nth instruction in the shader binary.
Offset
|
Size
|
Description
|
0x0
|
0x4
|
Byte offset within symbol table pointing to a source shader filename.
|
0x4
|
0x4
|
Line number of the corresponding shader instruction within the shader source code.
|
Unknown Block 2 Item
Offset
|
Size
|
Description
|
0x0
|
0x4
|
This seems to be an index of some sort.
|
0x4
|
0x4
|
|
0x8
|
0x4
|
|