Age | Commit message (Collapse) | Author |
|
|
|
If GL context had e.g. tex0, tex2 and fog the VAPOutputCntl1 returned 0x104 instead of 0x124 - that meaned we're sending only 8 texcoords (instead of 12) which ended up in GPU hang.
|
|
We shouldn't use i variable for SWTCL_OVM_TEX because textures doesn't have to be enabled in "packed" order.
We could have tex1,tex3 and fog which would receive 7,9,8 OVM locations instead of 6,7,8.
|
|
|
|
|
|
RS_COL_FMT field is part of RS_IP_* reg not RS_INST_*
|
|
- when rewriting per component negate swizzle, first instruction should get not negated source
- KIL instruction ignores swizzles
TODO:
- tex instructions does not support saturation
- tex instructions cannot read from consant memory
|
|
|
|
Sending from VAP more texture coordinates than RS expects results in GPU hang.
Fixes BumpSelfShadow from DirectX8 SDK.
|
|
If the vertex program didn't write position attribute, the position invariant function would add necessary instructions, but the vertex position would be overwritten by artificial outputs insts added to satisfy fragment program requirements.
Fixes "whole screen is gray" problem for HW TCL path in sauerbraten when shaders are enabled, and whole slew of wine d3d9 tests.
|
|
|
|
|
|
If the vertex program wrote secondary color without primary color, the secondary color output register index would be 0 which resulted in overwriting vertex position in some cases.
|
|
|
|
|
|
|
|
Fixes glean depthStencil test.
|
|
vertices
When with memory manager we need to make sure the GPU won't try to access
beyond vertex buffer size, do so by enforcing that the maximun index is the
last vertex of the buffer.
|
|
The early Z stuff is supposed to be unsafe without some more work in the
enable/disable path (in particular, how do we want to get it disabled on
the way out to the X Server?), but at the moment is 6% in OA.
|
|
intel_miptree_pitch_align does this later on.
|
|
Y tiling is why the 965 check was there, but I wanted to experiment with Y
on pre-965 as well.
|
|
Thanks to Shuang He for catching this.
|
|
Fixes the render-to-texture test in progs/tests/getteximage.c
|
|
|
|
|
|
|
|
|
|
This gets the correct srclvl image map when uploading images to the new mipmap.
|
|
Only enabled for software drivers at this point.
Note that the gl_buffer_object::Access enum field has been replaced by
a gl_buffer_object::AccessFlags bitfield. The new field is a mask of
the GL_MAP_x_BIT flags which is a superset of the old GL_READ_ONLY,
GL_WRITE_ONLY and GL_READ_WRITE modes. When we query GL_BUFFER_ACCESS_ARB
we translate the bitfield into the conventional enum values.
|
|
|
|
Fixes GPU hangs in software TCL path
|
|
|
|
Signed-off-by: Maciej Cencora <m.cencora@gmail.com>
|
|
|
|
Also when index format is GL_UBYTE, convert it to GL_USHORT not GL_UINT.
Fix license header too.
Reported by: Nicolai Hähnle <nhaehnle@gmail.com>
|
|
|
|
software TCL path
|
|
|
|
|
|
|
|
|
|
Always allocate the vertex program input registers in the same order as the vertex attributes are passed in vertex arrays.
|
|
Even if we don't pass all 4 color components to vertex shader unit, the vertex program can generate the missing components.
|
|
autoconf had been designating the 8 bit libOSMesa as the default
standalone osmesa, but the Makefile expected it to be linked to libGL.
Fix up the osmesa Makefile so that it allows any of the combinations of
standalone and channel width to be built.
Fixes bug #21980.
(cherry picked from commit 7441dcd90b01df8351026af8bbb50e11bb86071a)
|
|
autoconf had been designating the 8 bit libOSMesa as the default
standalone osmesa, but the Makefile expected it to be linked to libGL.
Fix up the osmesa Makefile so that it allows any of the combinations of
standalone and channel width to be built.
Fixes bug #21980.
|
|
This is about a 30% performance win in OA with high settings on my GM45,
and experiments with 915GM indicate that it'll be around a 20% win there.
Currently, 915-class hardware is seriously hurt by the fact that we use
fence regs to control the tiling even for 3D instructions that could live
without them, so we spend a bunch of time waiting on previous rendering in
order to pull fences off. Thus, the texture_tiling driconf option defaults
off there for now.
|
|
We're on the way to telling the kernel about when we need fence regs on our
objects or not, and this will cut the number of places needing them.
|
|
|
|
|
|
|