| Age | Commit message (Collapse) | Author |
|---|
|
At last it's possible to turn on tgsi dumps and other debugging in the
state tracker without modifying sources...
|
|
Always test for PIPE_TRANSFER_READ/WRITE using the bit-wise and operator, and
add a pipe_transfer_buffer_flags() helper for getting the buffer usage flags
corresponding to them.
|
|
The existing implementation was already implemented on software, but relied
on the pipe driver to always support the R16G16B16A16_SNORM format. This
patch eliminates that, without prejudice against a future hardware-only
implementation.
It also avoids some of the short <-> float conversions, and only does a read
transfer of the color buffer on GL_RETURN if absolutely necessary.
|
|
There are two usage types of buffer CPU accesses:
One where we try to use the buffer contents for multiple draw commands in
a batch. (batch := sequence of commands that are flushed together),
like incrementally adding bitmaps to a bitmap texture that is reallocated
on flush.
And one where we assume we can safely overwrite the old buffer contexts, like
glTexSubImage. In this case we need to make sure all old drawing commands
referencing the buffer are flushed before we map the buffer.
This is easily forgotten.
Add wrappers for the most common of these operations. The first type is
prefixed with "st_no_flush" and the second type is prefixed with
"st_cond_flush", where "cond" indicates that we attmpt to only flush
if there is indeed unflushed draw commands referencing the buffer.
Prefixed functions are
screen::get_tex_transfer
pipe_buffer_write
pipe_buffer_read
pipe_buffer_map
Please use the wrappers whenever possible.
Signed-off-by: Thomas Hellstrom <thellstrom-at-vmware-dot-com>
|
|
Also implement context member functions to optimize away those
flushes whenever possible.
Signed-off-by: Thomas Hellstrom <thellstrom-at-vmware-dot-com>
|
|
The core reference counting code is centralized in p_refcnt.h.
This has some consequences related to struct pipe_buffer:
* The screen member of struct pipe_buffer must be initialized, or
pipe_buffer_reference() will crash trying to destroy a buffer with reference
count 0. u_simple_screen takes care of this, but I may have missed some of
the drivers not using it.
* Except for rare exceptions deep in winsys code, buffers must always be
allocated via pipe_buffer_create() or via screen->*buffer_create() rather
than via winsys->*buffer_create().
|
|
It looks like I resolved the merge conflicts but did not save my emacs
buffers before committing...
|
|
Conflicts:
src/mesa/state_tracker/st_cb_accum.c
src/mesa/state_tracker/st_cb_drawpixels.c
|
|
|
|
Use _mesa_malloc(), _mesa_free(), etc everywhere, not malloc(), free(), etc.
Still using CALLOC_STRUCT() at this point.
|
|
Instead, a new pipe_transfer object has to be created and mapped for
transferring data between the CPU and a texture. This gives the driver more
flexibility for textures in address spaces that aren't CPU accessible.
This is a first pass; softpipe/xlib builds and runs glxgears, but it only shows
a black window. Looks like something's off related to the Z buffer, so the
depth test always fails.
|
|
Quite a few util modules were maintaining a single vertex buffer over multiple
frames, and potentially reusing it in subsequent frames. Unfortunately that
would force us into syncrhonous rendering as the buffer manager would be
forced to wait for the previous rendering to complete prior to allowing the
map.
This resolves that issue, but requires the state tracker to issue a few new
flush() calls at the end of each frame.
|
|
Also, rename p_tile.[ch] to u_tile.[ch]
|
|
|
|
You don't need a pipe_context * for this, and all other necessary info is
already inside pipe_surface.
|
|
|
|
The chars-per-pixel concept falls apart with compressed and yuv images,
where more than one pixel are coded in a single data block.
|
|
Create different temporary surfaces for CPU_READ/WRITE when needed (such as
for glReadPixels, glAccum, some glCopy/DrawPixels, glCopyTexSubImage, etc).
|
|
|
|
|
|
Instead of get/put_tile(), map the buffer and operate on values directly.
|
|
If the driver can't create a PIPE_FORMAT_R16G16B16A16_SNORM surface, create
an accum surface using a shallower format and taller height. Since only the
accum buffer code accesses the surface the actual format doesn't really
matter, just that there's enough memory.
|
|
The cell driver still uses an internal CELL_FLUSH_WAIT flag, in the long run
proper fencing should be implemented for it.
|
|
Use the code in cso_context.c rather than st_cache.c.
Basically, binding of state objects now goes through the CSO module.
But Vertex/fragment shaders go through pipe->bind_fs/vs_state() since they're
not cached by the CSO module at this time.
Also, update softpipe driver to handle NULL state objects in various places.
This happens during context destruction. May need to update other drivers...
|
|
Update the Makefiles and includes for the new paths.
Note that there hasn't been no separation of the Makefiles yet, and make is
jumping all over the place. That will be taken care shortly. But for now, make
should work. It was tested with linux and linux-dri. Linux-cell and linux-llvm
might require some minor tweaks.
|
|
pipe_get/put_tile_rgba() now use pipe->get/put_tile internally.
Also simplify the <format>_get/put_tile_rgba() helper functions and clean up
some inconsitencies in them.
|
|
It's now the responsibility of surface users to keep track of their mappings.
|
|
|
|
|
|
Directly use struct pipe_buffer_handle for storage and struct pipe_surface for
(un)mapping.
|
|
The _rgba versions are temporary until the state tracker is updated.
|
|
|
|
|
|
doesn't handle negative color values)
Also, remove unneeded clamping in the accum ops.
|
|
|
|
|
