Size pixel to fetch, in bytes. Doesn't seem to be required, setting
it to 0x0 seems to work ok, but may be less optimal.
The pair of MEM_SIZE/ADDR registers get programmed
in sequence with the size/addr of each buffer.
aka clip_halfz
range of -8.0 to 8.0
range of -512.0 to 512.0
RENDER_MODE is RB_RESOLVE_PASS for gmem->mem, otherwise RB_RENDER_PASS
render targets - 1
Pitch (actually, appears to be pitch in bytes, so really is a stride)
in GMEM, so pitch of the current tile.
offset into GMEM (or system memory address in bypass mode)
actually, appears to be pitch in bytes, so really is a stride
Z_TEST_ENABLE bit is set for zfunc other than GL_ALWAYS or GL_NEVER
seems to be always set to 0x00000000
DEPTH_BASE is offset in GMEM to depth/stencil buffer, ie
bin_w * bin_h / 1024 (possible rounded up to multiple of
something?? ie. 39 becomes 40, 78 becomes 80.. 75 becomes
80.. so maybe it needs to be multiple of 8??
Pitch of depth buffer or combined depth+stencil buffer
in z24s8 cases.
seems to be always set to 0x00000000
Base address for stencil when not using interleaved depth/stencil
pitch of stencil buffer when not using interleaved depth/stencil
seems to be set to 0x00000002 during binning pass
X/Y offset of current bin
seems to be where firmware writes BIN_DATA_ADDR from
CP_SET_BIN_DATA packet.. probably should be called
PC_BIN_BASE (just using name from yamato for now)
probably should be PC_BIN_SIZE
SIZE is current pipe width * height (in tiles)
N is some sort of slot # between 0..(SIZE-1). In case
multiple tiles use same pipe, each tile gets unique slot #
STRIDE_IN_VPC: ALIGN(next_outloc - 8, 4) / 4
(but, in cases where you'd expect 1, the blob driver uses
2, so possibly 0 (no varying) or minimum of 2)
indexed by dimension
indexed by dimension, global_size / local_size
TOTALATTRTOVS is # of attributes to vertex shader, in register
slots (ie. vec4+vec3 -> 7)
STRMDECINSTRCNT is # of VFD_DECODE_INSTR registers valid
STRMFETCHINSTRCNT is # of VFD_FETCH_INSTR registers valid
MAXSTORAGE could be # of attributes/vbo's
SHIFTCNT appears to be size, ie. FLOAT_32_32_32 is 12, and BYTE_8 is 1
From register spec:
SP_FS_OBJ_OFFSET_REG.CONSTOBJECTSTARTOFFSET [16:24]: Constant object
start offset in on chip RAM,
128bit aligned
The full/half register footprint is in units of four components,
so if r0.x is used, that counts as all of r0.[xyzw] as used.
There are separate full/half register footprint values as the
full and half registers are independent (not overlapping).
Presumably the thread scheduler hardware allocates the full/half
register names from the actual physical register file and
handles the register renaming.
From regspec:
SP_FS_CTRL_REG0.FS_LENGTH [31:24]: FS length, unit = 256bits.
If bit31 is 1, it means overflow
or any long shader.
These seem to be offsets for storage of the varyings.
Always seems to start from 8, possibly loc 0 and 4
are for gl_Position and gl_PointSize?
SP_VS_OBJ_START_REG contains pointer to the vertex shader program,
immediately followed by the binning shader program (although I
guess that is probably just re-using the same gpu buffer)
The full/half register footprint is in units of four components,
so if r0.x is used, that counts as all of r0.[xyzw] as used.
There are separate full/half register footprint values as the
full and half registers are independent (not overlapping).
Presumably the thread scheduler hardware allocates the full/half
register names from the actual physical register file and
handles the register renaming.
From regspec:
SP_FS_CTRL_REG0.FS_LENGTH [31:24]: FS length, unit = 256bits.
If bit31 is 1, it means overflow
or any long shader.
SP_FS_OBJ_START_REG contains pointer to fragment shader program
seems to be one bit per scalar, '1' for flat, '0' for smooth
seems to be one bit per scalar, '1' for flat, '0' for smooth
render targets - 1
Configures the mapping between VSC_PIPE buffer and
bin, X/Y specify the bin index in the horiz/vert
direction (0,0 is upper left, 0,1 is leftmost bin
on second row, and so on). W/H specify the number
of bins assigned to this VSC_PIPE in the horiz/vert
dimension.
seems to be set to 0x00000001 during binning pass
seems to be always set to 0x00000001
seems to be always set to 0x00000001
seems to be always set to 0x00000001
seems to be always set to 0x00000003
seems to be always set to 0x00000001
Texture sampler dwords
Texture constant dwords
INDX is index of texture address(es) in MIPMAP state block
Pitch in bytes (so actually stride)
SWAP bit is set for BGRA instead of RGBA