BDW adds the following restriction: "When multiplying DW x DW, the dst
cannot be accumulator."
Cc: "11.1,11.0" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
The cleaning up was quite a performance hog (making pipe_resource_reference
the number two in profilers on the vertex path, and 3rd overall, with its
cousin pipe_reference_described not far behind) if there were lots
of tiny draw calls (ipers). Now the reason was really that it was blindly
calling this for all potential shader views (so 32 each for vs and gs) even
though the app never touched a single one which could have been fixed,
however I can't come up with a good reason why we refcount these. We've got
references, of course, in the sampler views, which should be quite sufficient
as we do all vertex and geometry shader execution fully synchronous.
(Calling prepare_shader_sampling for all draw calls even if there were no
changes looks quite suboptimal too, but generally we don't really expect vs/gs
shader sampling to be used much with llvmpipe, and there's even an early exit
if there aren't any views to avoid the "null loop" albeit it's now no longer
always trying to loop through all 32 slots. Maybe improve another time...).
Of course, if we manage to make vertex loads run asynchronously some day,
we need references again, but adding that back would be the least of the
problems...
Also only set LP_NEW_SAMPLER_VIEW for fragment sampler views. Nothing on the
vertex side depends on it (I suppose we'd really wanted a separate flag in
any case).
(Good for a 3% improvement or so in ipers under the right conditions.)
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
This was not really a leak per se, but we were referencing the textures for
longer than intended. If textures were set via llvmpipe_set_sampler_views()
(for fs) and then picked up by lp_setup_set_fragment_sampler_views(), they
were referenced in the setup state. However, the only way to unreference them
was by replacing them with another texture, and not when the texture slot
was replaced with a NULL sampler view. (They were then further also referenced
by the scene too which might have additional minor side effects as we limit
the memory size which is allowed to be referenced by a scene in a rather crude
way.) Only setup destruction (at context destruction time) then finally would
get rid of the references.
Fix this by noting the number of textures the last time, and unreference
things if the new view is NULL (avoiding having to unreference things
always up to PIPE_MAX_SHADER_SAMPLER_VIEWS which would also have worked).
Found by code inspection, no test...
v2: rename var
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Only modify interpolation type for integer-based varyings or when the
consumer is known and different than fragment shader.
If we are linking separate shader programs and the consumer is unknown,
the consumer could be added later and be a fragment shader. If we
modify the interpolation type in this case, we could read wrong
values in the fragment shader inputs, as shown in bug 93320.
Fixes the following CTS test:
ES31-CTS.vertex_attrib_binding.advanced-bindingUpdate
Fixes the following dEQP tests:
dEQP-GLES31.functional.separate_shader.random.102
dEQP-GLES31.functional.separate_shader.random.111
dEQP-GLES31.functional.separate_shader.random.115
dEQP-GLES31.functional.separate_shader.random.17
dEQP-GLES31.functional.separate_shader.random.22
dEQP-GLES31.functional.separate_shader.random.23
dEQP-GLES31.functional.separate_shader.random.3
dEQP-GLES31.functional.separate_shader.random.32
dEQP-GLES31.functional.separate_shader.random.39
dEQP-GLES31.functional.separate_shader.random.64
dEQP-GLES31.functional.separate_shader.random.73
dEQP-GLES31.functional.separate_shader.random.91
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=93320
Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
This shouldn't hurt anything, and I'm about to introduce a pass that
will want it.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
This makes it a pass, hiding the parameter structs and block callbacks
so it's simpler to work with.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Shorter than compiler->scalar_stage[MESA_SHADER_VERTEX], which can
help with line-wrapping.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
nir_build_ivec4 is more readable and succinct than using nir_build_imm
directly, even if you have C99.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
If shader declares uniform explicit location in one stage but
implicit in another, explicit location should be used. Patch marks
implicit uniforms as explicit if they were explicit in previous stage.
This makes sure that we don't treat them implicit later when assigning
locations.
Fixes following CTS test:
ES31-CTS.explicit_uniform_location.uniform-loc-implicit-in-some-stages3
v2: move check to cross_validate_globals (Timothy)
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
The RS and hardware binding tables are only supported on the 3D
pipeline and can lead to corruption if left enabled during a GPGPU
workload. Disable it when switching to the GPGPU (or media) pipeline
and re-enable it when switching back to the 3D pipeline.
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Abdiel Janulgue <abdiel.janulgue@linux.intel.com>
This hardware bug can supposedly lead to a hang on IVB and VLV.
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
AFAIK brw_emit_select_pipeline() is only called once during context
init on Gen4-5, at which point the pipeline is likely to be already
idle so it may just happen to work by luck regardless of the MI_FLUSH.
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Switching the current pipeline while it's not completely idle or the
read and write caches aren't flushed can lead to corruption. Fixes
misrendering of at least the following Khronos CTS test:
ES31-CTS.shader_image_load_store.basic-allTargets-store-fs
The stall and flushes are no longer required on Gen8+.
v2: Emit PIPE_CONTROL with non-zero post-sync op before the write
cache flush on SNB due to hardware bug. (Ken)
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=93323
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This hardware bug can cause a hang on context restore while the
current pipeline is set to GPGPU (BDWGFX HSD 1909593). In addition to
clearing the valid bit, mark the CC state as dirty to make sure that
the CC indirect state pointer is re-emitted when we switch back to the
3D pipeline.
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
This will be used on Gen8+ to make sure that the color calculator
state pointers are re-emitted when switching back to the 3D pipeline
after some GPGPU workload due to a hardware workaround. There are
other state bits already defined that could be used to achieve the
same effect but they all cause a ton of unrelated state to be
re-emitted (e.g. BRW_NEW_STATE_BASE_ADDRESS), so just define a new
one, state bits are cheap.
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reduces local memory usage in a lot of Metro 2033 Redux and a few KSP
shaders:
total local used in shared programs : 54116 -> 30372 (-43.88%)
Probably modest advantage to execution, but it's an imporant
prerequisite to dropping some of the TGSI optimizations done by the
state tracker.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Previously we were treating any indirect temp array usage to mean that
everything should end up in lmem. The MemoryOpt pass would clean a lot
of that up later, but in the meanwhile we would lose a lot of
opportunity for optimization.
This helps a lot of Metro 2033 Redux and a handful of KSP shaders:
total instructions in shared programs : 6288373 -> 6261517 (-0.43%)
total gprs used in shared programs : 944051 -> 945131 (0.11%)
total local used in shared programs : 54116 -> 54116 (0.00%)
A typical case is for register usage to double and for instructions to
halve. A future commit can also optimize local memory usage size to be
reduced with better packing.
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Indirect constbuf indexing works by using very large offsets. However if
an indirect constbuf index load is const-propagated, it becomes a very
large const offset. Take that into account when legalizing the SSA by
moving the high parts of that offset into the file index. Also disallow
very large (or small) indices on most other instructions.
This fixes regressions in ubo_array_indexing/*-two-arrays piglit tests.
Fixes: abd326e81b (nv50/ir: propagate indirect loads into instructions)
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
I'm not sure about the consequences of this bug, but it's definitely
dangerous.
This applies to SI, CIK, VI.
Cc: 11.0 11.1 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
For those formats that support hw mipmap generation, use the
DXGenMips command. Otherwise fallback to the mipmap generation utility.
Tested with piglit, OpenGL apps (Heaven, Turbine, Cinebench)
v2: make sure the texture surface was created with the render target bind flag
set relocation flag to SVGA_RELOC_WRITE for the texture surface
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
The actual increment of the num-generate-mipmap counter will be done
in a subsequent patch when hw generate mipmap is supported.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
This patch adds a new interface to support hardware mipmap generation.
PIPE_CAP_GENERATE_MIPMAP is added to allow a driver to specify
if this new interface is supported; if not supported, the state tracker will
fallback to mipmap generation by rendering/texturing.
v2: add PIPE_CAP_GENERATE_MIPMAP to the disabled section for all drivers
v3: add format to the generate_mipmap interface to allow mipmap generation
using a format other than the resource format
v4: fix return type of trace_context_generate_mipmap()
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
The OpenGL specifications for bitfieldExtract() says:
The result will be undefined if <offset> or <bits> is negative, or if
the sum of <offset> and <bits> is greater than the number of bits
used to store the operand.
Therefore passing bits=32, offset=0 is legal and defined in GLSL.
But the earlier SM5 ubfe/ibfe opcodes are specified to accept a bitfield width
ranging from 0-31. As such, Intel and AMD instructions read only the low 5 bits
of the width operand, making them not able to implement the GLSL-specified
behavior directly.
This commit adds ubfe/ibfe operations from SM5 and a lowering pass for
bitfield_extract to to handle the trivial case of <bits> = 32 as
bitfieldExtract:
bits > 31 ? value : bfe(value, offset, bits)
Fixes:
ES31-CTS.shader_bitfield_operation.bitfieldExtract.uvec3_0
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92595
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Tested-by: Marta Lofstedt <marta.lofstedt@intel.com>
The OpenGL specifications for bitfieldInsert() says:
The result will be undefined if <offset> or <bits> is negative, or if
the sum of <offset> and <bits> is greater than the number of bits
used to store the operand.
Therefore passing bits=32, offset=0 is legal and defined in GLSL.
But the earlier SM5 bfi opcode is specified to accept a bitfield width
ranging from 0-31. As such, Intel and AMD instructions read only the low
5 bits of the width operand, making them not able to implement the
GLSL-specified behavior directly.
This commit fixes the lowering of bitfield_insert to handle the trivial
case of <bits> = 32 as
bitfieldInsert:
bits > 31 ? insert : bfi(bfm(bits, offset), insert, base)
Fixes:
ES31-CTS.shader_bitfield_operation.bitfieldInsert.uint_2
ES31-CTS.shader_bitfield_operation.bitfieldInsert.uvec4_3
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92595
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Tested-by: Marta Lofstedt <marta.lofstedt@intel.com>
We check that a bunch of raster operations are disabled in
blit_copy_pixels(). We also need to check that color logicop is
disabled.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
The whole point of AMD_pinned_memory is that applications don't have to map
buffers via OpenGL - but they're still allowed to, so make sure we don't break
the link between buffer object and user memory unless explicitly instructed
to.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
This accomodates a streaming pattern where the discard flag is set when the
application wraps back to the beginning of the buffer.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
It makes sense to re-use pipe->invalidate_resource for the purpose of
glInvalidateBufferData, but this function is already implemented in vc4
where it doesn't have the expected behavior. So add a capability flag
to indicate that the driver supports the expected behavior.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Change the check to be in line with what the quoted spec fragment says.
I have sent out a piglit test for this as well.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
The internal Mesa format used for a texture might not match the one
requested in the internalFormat when the texture was created, for
example if the driver is internally remapping RGB textures to RGBA.
Otherwise it can cause false positives for completeness if one mipmap
image is created as RGBA and the other as RGB because they would both
have an RGBA Mesa format. If we check the InternalFormat instead then
we are directly checking the API usage which I think better matches
the intention of the check.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=93700
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
I spotted this while looking for what needs updating in future platforms.
I'm too lazy to go through the git logs, but it was probably missed by Jason
when all the brw refactoring happened.
Signed-off-by: Ben Widawsky <benjamin.widawsky@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Intel/AMD's hardware instructions do not handle arguments of 32.
Constant evaluation should not produce a result different from the
hardware instruction.
The s/1ull/1u/ change is intentional: previously we wanted defined
behavior for the "1 << 32" case, but we're making this case undefined so
we can make it 1u and save ourselves a 64-bit operation.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Shifting into the sign bit is undefined, as is shifting by 32.
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
If a Python codegen script failed, it would write a zero-byte file,
which on subsequent invocations of make would trick it into thinking the
file was appropriately generated.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
We would like to be able to combine
result.x = bitfieldExtract(src0.x, src1.x, src2.x);
result.y = bitfieldExtract(src0.y, src1.y, src2.y);
result.z = bitfieldExtract(src0.z, src1.z, src2.z);
result.w = bitfieldExtract(src0.w, src1.w, src2.w);
into a single ivec4 bitfieldInsert operation. This should be possible
with most drivers.
This patch changes the offset and bits parameters from scalar ints
to ivecN or uvecN. The type of all three operands will be the same,
for simplicity.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
We would like to be able to combine
result.x = bitfieldInsert(src0.x, src1.x, src2.x, src3.x);
result.y = bitfieldInsert(src0.y, src1.y, src2.y, src3.y);
result.z = bitfieldInsert(src0.z, src1.z, src2.z, src3.z);
result.w = bitfieldInsert(src0.w, src1.w, src2.w, src3.w);
into a single ivec4 bitfieldInsert operation. This should be possible
with most drivers.
This patch changes the offset and bits parameters from scalar ints
to ivecN or uvecN. The type of all four operands will be the same,
for simplicity.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
TGSI doesn't use these - it just translates ir_quadop_bitfield_insert
directly. NIR can handle ir_quadop_bitfield_insert as well.
These opcodes were only used for i965, and with Jason's recent patches,
we can do this lowering in NIR (which also gains us SPIR-V handling).
So there's not much point to retaining this GLSL IR lowering code.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
NIR's bfm, like Intel/AMD's hardware instructions and GLSL IR's
ir_binop_bfm takes <bits> as src0 and <offset> as src1.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
A shader in Unreal4 uses the result of divide by zero in its color
output, producing NaN and triggering this assertion since NaN is not
equal to itself.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=93560
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>