The polygon stipple pattern, like the viewport and the
polygon face orientation, must be inverted on the i965
when rendering to a FBO (which itself has an inverted pixel
coordinate system compared to raw Mesa).
In addition, the polygon stipple offset, which orients
the stipple to the window system, disappears when rendering
to an FBO (because the window system offset doesn't apply,
and there's no associated FBO offset).
With these fixes, the conform triangle and polygon stipple
tests pass when rendering to texture.
(cherry picked from commit 29309b45b0)
In the i965, the FBO coordinate system is inverted from the standard
OpenGL/Mesa coordinate system; that means that the viewport and the
polygon face orientation have to be inverted if rendering to a FBO.
The viewport was already being handled correctly; but polygon face
was not. This caused a conform failure when rendering to texture with
two-sided lighting enabled.
This fixes the problem in the i965 driver, and adds to the comment about
the gl_framebuffer "Name" field so that this isn't a surprise to other
driver writers.
(cherry picked from commit 6dceeb2eb8)
The MAX-based function can produce values that are non-monotonic for a span
which causes glitches in texture filtering. The sqrt-based one avoids that.
This is perhaps slightly slower than before, but the difference
probably isn't noticable given we're doing software mipmap filtering.
Issue reported by Nir Radian <nirr@horizonsemi.com>
(cherry picked from master, commit c334ce273e)
Before this change we would up emitting instructions with invalid register
numbers. This typically (but not always) hung the GPU. For now, just
prevent emitting bad instructions to avoid hangs. Still need to do some
kind of proper error recovery.
(cherry picked from commit e60b3067d8)
Instructions such as RCP, RSQ, LOG must smear the result of the function
across the dest register's X, Y, Z and W channels (subject to write masking).
Before this change, only the X component was getting written.
Among other things, this fixes cube map texture sampling in GLSL shaders
(since cube lookups involve normalizing the texcoord).
(cherry picked from commit 3485801978)
Previously, the prog_instruction::Data field was used to map original Mesa
instructions to brw instructions in order to resolve subroutine calls. This
was a rather tangled mess. Plus it's an obstacle to implementing dynamic
allocation/growing of the instruction buffer (it's still a fixed size).
Mesa's GLSL compiler emits a label for each subroutine and CAL instruction.
Now we use those labels to patch the subroutine calls after code generation
has been done. We just keep a list of all CAL instructions that needs patching
and a list of all subroutine labels. It's a simple matter to resolve them.
This also consolidates some redundant post-emit code between brw_vs_emit.c and
brw_wm_glsl.c and removes some loops that cleared the prog_instruction::Data
fields at the end.
Plus, a bunch of new comments.
(cherry picked from commit c51c822ee0)
This was causing hangs in cairogears, as we would blit to the 8bpp target
(A8 texture) as 16bpp, and stomp over state objects.
(cherry picked from commit 19e134051c)
The i965 hardware cannot do GL_CLAMP behavior on textures; an earlier
commit forced a software fallback if strict conformance was required
(i.e. the INTEL_STRICT_CONFORMANCE environment variable was set) and
2D textures were used, but it was somewhat flawed - it could trigger
the software fallback even if 2D textures weren't enabled, as long
as one texture unit was enabled.
This fixes that, and adds software fallback for GL_CLAMP behavior with
1D and 3D textures.
It also adds support for a particular setting of the INTEL_STRICT_CONFORMANCE
environment variable, which forces software fallbacks to be taken *all*
the time. This is helpful with debugging. The value is:
export INTEL_STRICT_CONFORMANCE=2
(cherry picked from commit 3468315087)
For regular GL, we must have vertex positions in order to draw. But ES2
doesn't have that requirement (positions can be computed from any array
of data).
See bug 19911.
(cherry picked from commit 97dd2ddbd9)
i965 doesn't natively support GL_CLAMP; it treats it like
GL_CLAMP_TO_EDGE, which fails conformance tests.
This fix adds a clause to the check_fallbacks() test to check
whether GL_CLAMP is in use on any enabled 2D texture. If so,
and if strict conformance is required (via INTEL_STRICT_CONFORMANCE),
a software fallback is mandated.
In addition, validate textures *before* checking for fallbacks,
rather than after; otherwise, the texture state is never validated
and can't be trusted. (In particular, if texturing is enabled and
the sampler would access any level beyond level 0 of a texture, the
sampler will segfault, because texture validation sets the firstLevel
and lastLevel fields of a texture object so that the valid levels
will be mapped and accessed correctly. If texture validation doesn't
occur, only level 0 is accessed correctly, and that only because
firstLevel and lastLevel happen to be set to 0.)
(cherry picked from commit 17c7852bf9)
Replace Default1D/2D/3D/Cube/etc with DefaultTex[TEXTURE_x_INDEX].
The same should be done with the Current1D/2D/3D/etc pointers...
(cherry picked from commit 4d24b639d1)
While running conform with render-to-texture:
conform -d 33 -v 2 -t -direct
the i965 driver failed this assertion:
intel_clear.c:77: intel_clear_tris: Assertion `(mask & ~((1 << BUFFER_BACK_LEFT) | (1 << BUFFER_FRONT_LEFT) | (1 << BUFFER_DEPTH) | (1 << BUFFER_STENCIL))) == 0' failed.
The problem is that intel_clear_tris() is called by intelClear() to
clear any and all of the available color buffers, but intel_clear_tris()
actually only handles the back left and front left color buffers; so
the assertion fails as soon as you try to clear a non-standard color
buffer.
The fix is to have intelClear() only call intel_clear_tris() with
buffers that intel_clear_tris() can support. intelClear() already backs
down to _swrast_Clear() for all buffers that aren't handled explicitly.
(cherry picked from commit 0ccbc3c905)
This lets us avoid allocing new buffers for renderbuffers, finalized miptrees,
and PBO-uploaded textures when there's an unreferenced but still active one
cached, while also avoiding CPU waits for batchbuffers and CPU-uploaded
textures. The size of BOs allocated for a desktop running current GL
cairogears on i915 is cut in half with this.
Note that this means we require libdrm 2.4.5.
(cherry picked from commit 40dd024be6)
We were asking for something illegal (write_domain != 0 && read_domains !=
write_domain) because at the time of writing the region surfaces were used
for texturing occasionally as well, and we weren't really clear on the model
GEM was going to use.
This reliably triggered a kernel bug with domain handling, resulting in
oglconform mustpass.c failure. Of course, it only became visible after
01bc4d441f cleaned up some gratuitous flushing.
(cherry picked from commit 078e8a61b2)
It is possible that an object whose vertices all are outside of a
view plane is passed to clip thread due to the RHW workaround. This
object should be rejected by clip thread. Fix bug #19879
(cherry picked from commit 68915fd6fa)
The default for EmitCondCodes got flipped when gallium-0.2 was merged.
This fixes GLSL if/else/endif regressions.
Drivers that use GLSL should always explicitly set the flag to be safe.
(cherry picked from commit a9e753c84c)
