| Age | Commit message (Collapse) | Author |
|---|
|
This existed to get the icache flushed. However, GEM handles this for us
now for sure, and we had disabled it prematurely anyway.
|
|
The GEM flags are much more descriptive for what we need. Since this makes
bufmgr_fake rather device-specific, move it to the intel common directory.
We've wanted to do device-specific stuff to it before.
|
|
Mostly some pedantic changes such that `make clean' always ignores
errors. Also changed the top clean target to do the `touch
configs/current' dance instead of realclean.
|
|
|
|
|
|
Currently, there is a single path in src/mesa/Makefile to install that
has a few conditionals in it. This commit changes install to act like
default where we loop over $(DRIVER_DIRS), deciding what to do.
A new target, install-headers, has been broken out to accomodate
installing a standalone OSMesa where neither libGL or gl.pc are wanted.
|
|
There were a couple spots left where a recursive make could fail in a
chain of commands without stopping.
|
|
|
|
|
|
FP inputs are now counted and mapped correctly, and temps
are allocated tightly and correctly.
|
|
DP3/DP4 only takes two arguments, but tried to load three, causing
a false fallback to the dumb shader.
|
|
This is not the same as r3xx indexing. It only tries to protect inputs on
the pixel stack from getting clobbered by temps or texs.
Texs don't need special treatment since they read from special input regs
and write to the same temp regs as ALU/FC instructions.
|
|
Use _mesa_reference_vert/fragprog() wherever we assign program pointers.
Fixes a memory corruption bug found with glean/api2 test.
|
|
Doesn't deal with local modifications yet (need new kernel set_domain ioctl
for that to work). Also, guesses what domains are affected based on the
read/write bits set in the flags. Works for 915, probably not so much for
965.
|
|
Tex lookup with biased LOD. Should magically work.
|
|
We can't really do anything like emit_alu, so we're doing emit_mov instead.
|
|
|
|
The default target in src/mesa/Makefile calls a recursive $(MAKE). With
parallel jobs, this causes makedepend to run twice. Instead, block on
the first make until depend has been created.
|
|
Consistently cleanup the depend.bak files created by makedepend. Also,
realclean has been changed to use a single find command, which speeds it
up considerably.
|
|
|
|
This should make TEX/TXP work right. (Note: "Should" is not "does.")
|
|
This roughly doubles the speed of glxgears (GINAB) by allowing
more pixels to run concurrently.
|
|
When changing directories and running a sub-make, ensure that both the
cd and make commands propagate errors to the parent make.
|
|
Gotta be more careful with my cut'n'paste, lawl.
|
|
|
|
Accessing tiled surfaces without using the fence registers requires that
software deal with the address swizzling itself.
|
|
In classic mode googleearth triggered a case where vbos weren't getting accounted properly.
|
|
|
|
in insert_wpos.
fixes bug 15447
|
|
|
|
|
|
|
|
small integers are much prettier, and let me correlate to DRM debug output.
|
|
This is just cosmetic, to produce less scary values when the ioctl fails and
doesn't return values there.
|
|
|
|
|
|
Instead of attempting to fix these for GEM, just disable until GEM is
working.
|
|
based on patch by Dan Nicholson <dbn.lists@gmail.com>
|
|
you cannot change a tex into an output so this means we have to actually
do another instruction after this one to mov if its an output
|
|
|
|
This makes constant work which are 32-bit on r500 unlike r300.
Switch MOV to using MAD no idea if we might have negative things MAX 0,-5 is
likely to do the wrong thing..
|
|
We don't appear to update max_temp_idx yet anywhere though
|
|
We really need to update the shader state so the texrect parameters work.
This should fix compiz looking crappy on rs480 and rs690
|
|
Negative value means other errors, not aperture overflow. fix bug #15752
|
|
We really need to update the shader state so the texrect parameters work.
This should fix compiz looking crappy on rs480 and rs690
(cherry picked from commit 66a5562ce2906fbf5b96d1cee18f9a31a78c4360)
|
|
(cherry picked from commit 026ef8111a94f6449dfa5e5cc0ae91fca4e68c0c)
|
|
|
|
adapted from Markus' patch on bug 15386 with updates for non-TCL
and R500.
|
|
Thanks to dli in IRC for pointing this out.
|
|
For some reason, FGLRX doesn't actually set R500_US_INST_TEX.
Let us not make that same mistake.
|
