https://bugs.freedesktop.org/show_bug.cgi?id=32912
The fix is to call update_derived_state before user buffer uploads.
I've also moved some code around.
Unfortunately, there are still some ZMASK-related bugs which cause
misrendering, i.e. flushing doesn't always work and glean/fbo fails.
The motivation behind this rework is to get some speed by reducing
CPU overhead. The performance increase depends on many factors,
but it's measurable (I think it's about 10% increase in Torcs).
This commit replaces libdrm's radeon_cs_gem with our own implemention.
It's optimized specifically for r300g, but r600g could use it as well.
Reloc writes and space checking are faster and simpler than their
counterparts in libdrm (the time complexity of all the functions
is O(1) in nearly all scenarios, thanks to hashing).
(libdrm's radeon_bo_gem is still being used in the driver.)
It works like this:
cs_add_reloc(cs, buf, read_domain, write_domain) adds a new relocation and
also adds the size of 'buf' to the used_gart and used_vram winsys variables
based on the domains, which are simply or'd for the accounting purposes.
The adding is skipped if the reloc is already present in the list, but it
accounts any newly-referenced domains.
cs_validate is then called, which just checks:
used_vram/gart < vram/gart_size * 0.8
The 0.8 number allows for some memory fragmentation. If the validation
fails, the pipe driver flushes CS and tries do the validation again,
i.e. it validates only that one operation. If it fails again, it drops
the operation on the floor and prints some nasty message to stderr.
cs_write_reloc(cs, buf) just writes a reloc that has been added using
cs_add_reloc. The read_domain and write_domain parameters have been removed,
because we already specify them in cs_add_reloc.
The space checking has been tested by putting small values in vram/gart_size
variables.
There really shouldn't be any difference between the two for us.
Fixes a bug where Z16 renderbuffers would be untiled on gen6, likely
leading to hangs.
By relying on just intel_span_supports_format, some formats that
aren't supported pre-gen4 were not reporting FBO incomplete. And we
also complained in stderr when it happened on i915 because draw_region
gets called before framebuffer completeness validation.
ARB_fbo no longer disallows mismatched width/height on attachments
(shouldn't be any problem), mixed format color attachments (we only
support 1), and L/A/LA/I color attachments (we already reject them on
965 too). It requires Gen'ed names (driver doesn't care), and adds
FramebufferTextureLayer (we don't do texture arrays). So it looks
like we're already in the position we need to be for this extension.
Bug #27468, #32381.
In general, we have to negate in immediate values we pass in because
the src1 negate field in the register description is in the bits3 slot
that the 32-bit value is loaded into, so it's ignored by the hardware.
However, the src0 negate field is in bits1, so after we'd negated the
immediate value loaded in, it would also get negated through the
register description. This broke this VP instruction in the position
calculation in civ4:
MAD TEMP[1], TEMP[1], CONST[256].zzzz, CONST[256].-y-y-y-y;
Bug #30156
This only uploads the [min_index, max_index] range instead of [0, userbuf size],
which greatly speeds up user buffer uploads.
This is also a prerequisite for atomizing vertex arrays in st/mesa.
Fix the error in uniform row calculating, it may alloc one line
more which may cause out of range on memory usage, sometimes program
aborted when free the memory.
NOTE: This is a candidate for 7.9 and 7.10 branches.
Signed-off-by: Brian Paul <brianp@vmware.com>
Christian König