Fixes tests 'dEQP-GLES3.functional.texture.mipmap.*.generate.rgba5551*' on
Intel Broadwell 0x1616.
The GL 4.5 spec describes the algorithm of glGenerateMipmap as:
The contents of the derived images are computed by repeated, filtered
reduction of the level base image. [...] No particular filter algorithm is
required, though a box filter is recommended as the default filter.
Consider a texture for which all pixels are identical at level 0.
From the spec's description above, one may reasonably assume that the "filtered
reduction" of level 0 produces a new miplevel for which again all pixels are
identical. For any 2x2 subspan of identical pixels, it is difficult to see how
the "filtered reduction" of that subspan can produce a pixel that differs from
the source pixels.
Dithering during _mesa_meta_GenerateMipmap() violated that reasonable
assumption.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=99210
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: mesa-stable@lists.freedesktop.org
In its current state the unified i965 backend for
AMD_performance_monitor and INTEL_performance_query isn't able to report
meaningful Observation Architecture metrics since we haven't so far had
the necessary kernel support to fully configure the OA unit, nor the
corresponding support for normalizing the counters into a form that can
be usefully interpreted by application developers (as opposed to raw
values that may, for example, scale by the number of EUs there are).
So that we can focus on implementing just one of these extensions fully
and since we anticipate some significant backend changes as we look to
use a new kernel interface to configure the OA unit, this patch removes
the current backend. This will simplify our ability to update the
frontend infrastructure and backend interface before updating our
support for performance counters.
Signed-off-by: Robert Bragg <robert@sixbynine.org>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
The variable actually needs to be signed, otherwise converting it to a
float doesn't work as expected.
Fixes: https://bugs.freedesktop.org/show_bug.cgi?id=98914
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Nayan Deshmukh <nayan26deshmukh@gmail.com>
Cc: "13.0" <mesa-stable@lists.freedesktop.org>
Fixes: 1fb4179f92 ("vl: Fix trivial sign compare warnings")
handle the cases when vl_compositor_set_csc_matrix(),
vl_compositor_init_state() and vl_compositor_init() fail
Signed-off-by: Nayan Deshmukh <nayan26deshmukh@gmail.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
handle the cases when vl_compositor_set_csc_matrix(),
vl_compositor_init_state() and vl_compositor_init() fail
Signed-off-by: Nayan Deshmukh <nayan26deshmukh@gmail.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
pipe_buffer_map and pipe_buffer_create may return NULL
Signed-off-by: Nayan Deshmukh <nayan26deshmukh@gmail.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
FROM_DOUBLE opcodes are setup so that they use a dst register
with a size of 2 even if they only produce a single-precison
result (this is so that the opcode can use the larger register to
produce a 64-bit aligned intermediary result as required by the
hardware during the conversion process). This creates a problem for
spilling though, because when we attempt to emit a spill for the
dst we see a 32-bit destination and emit a scratch write that
allocates a single spill register, making the intermediary writes
go beyond the size of the allocation.
Prevent this by avoiding to spill the destination register of these
opcodes.
Alternatively, we can avoid this by splitting the opcode in two: one
that produces a 64-bit aligned result and one that takes the 64-bit
aligned result as input and produces a 32-bit result from it.
Reviewed-by: Matt Turner <mattst88@gmail.com>
When 64-bit registers are (un)spilled, we need to execute data shuffling
code before writing to or after reading from memory. If we have instructions
that operate on 64-bit data via 32-bit instructions, (un)spills for the
register produced by 32-bit instructions will not do data shuffling at all
(because we only see a normal 32-bit istruction seemingly operating on
32-bit data). This means that subsequent reads with that register using DF
access will unshuffle data read from memory that was never adequately
shuffled when it was written.
Fixing this would require to identify which 32-bit instructions write
64-bit data and emit spill instructions only when the full 64-bit
data has been written (by multiple 32-bit instructions writing to different
offsets of the same register) and always emit 64-bit unspills whenever
64-bit data is read, even when the instruction uses a 32-bit type to read
from them.
Reviewed-by: Matt Turner <mattst88@gmail.com>
The current spilling code can't spill vgrf allocations larger than 1
but SIMD4x2 doubles require 2 vgrfs, so we need to permit this case (which
is handled properly for DF data types by emitting 2 scratch messages and
doing data shuffling). We accomplish this by not auto-disabling spilling
for vgrf allocations with a size of 2, and then disable spilling on any
register with an offset != 0B (which indicates array access).
Disable spilling of partial DF reads/writes because these don't read/write
data for both logical threads and our scratch messages for 64-bit data
need data for both threads to be present.
Reviewed-by: Matt Turner <mattst88@gmail.com>
Spilling of 64-bit data requires data shuffling for the corresponding
scratch read/write messages. This produces unsupported swizzle regions
and writemasks that we need to scalarize.
Reviewed-by: Matt Turner <mattst88@gmail.com>
8-wide compressed DF operations are executed as two separate 4-wide
DF operations. In that scenario, we have to be careful when we allocate
register space for their operands to prevent the case where the first
half of the instruction overwrites the source of the second half.
To do this we mark compressed instructions as having hazards to make
sure that ther register allocators assigns a register regions for the
destination that does not overlap with the region assigned for any
of its source operands.
Reviewed-by: Matt Turner <mattst88@gmail.com>
By exploiting gen7's hardware decompression bug with vstride=0 we gain the
capacity to support additional swizzle combinations.
This also fixes ZW writes from X/Y channels like in:
mov r2.z:df r0.xxxx:df
Because DF regions use 2-wide rows with a vstride of 2, the region generated
for the source would be r0<2,2,1>.xyxy:DF, which is equivalent to r0.xxzz, so
we end up writing r0.z in r2.z instead of r0.x. Using a vertical stride of 0
in these cases we get to replicate the XX swizzle and write what we want.
Reviewed-by: Matt Turner <mattst88@gmail.com>
Certain swizzles like XYZW can be supported by translating only the first two
64-bit swizzle channels to 32-bit channels. This happens with swizzles such
that the first two logical components, when translated to 32-bit channels and
replicated across the second dvec2 row, select the same channels specified by
the 3rd and 4th logical swizzle components.
Notice that this opens up the possibility that some instructions are not
scalarized and can end up with XY or ZW 32-bit writemasks. Make sure we always
scalarize in such cases.
Reviewed-by: Matt Turner <mattst88@gmail.com>
Stages that use interleaved attributes generate regions with a vstride=0
that can hit the gen7 hardware decompression bug.
v2:
- Make static the function and fix indent (Matt)
Reviewed-by: Matt Turner <mattst88@gmail.com>
This came in handy when debugging the payload setup for Tess Eval,
since it prints correct subnr for attributes that can be loaded
in the second half of a register.
Reviewed-by: Matt Turner <mattst88@gmail.com>
Use a width of 2 with 64-bit attributes.
Also, if we have a dvec3/4 attribute that gets split across two registers
such that components XY are stored in the second half of a register and
components ZW are stored in the first half of the next, we need to fix
regioning for any instruction that reads components Z/W of the attribute.
Notice this also means that we can't support sources that read cross-dvec2
swizzles (like XZ for example).
v2: don't assert that we have a single channel swizzle in the case that we
have to fix up Z/W access on the first half of the next register. We
can handle any swizzle that does not cross dvec2 boundaries, which
the double scalarization pass should have prevented anyway.
Reviewed-by: Matt Turner <mattst88@gmail.com>
We need to shuffle the data before it is written to the URB. Also,
dvec3/4 need two vec4 slots.
v2: use byte_offset() instead of offset().
Reviewed-by: Matt Turner <mattst88@gmail.com>
This way callers don't need to know about 64-bit particularities and
we reuse some code.
v2:
- use byte_offset() instead of offset()
- only mark the surface as used once
Reviewed-by: Matt Turner <mattst88@gmail.com>
Mostly the same stuff as usual: we ned to shuffle the data before we
write and we need to emit two 32-bit write messages (with appropriate
32-bit writemask channels set) for a full dvec4 scratch write.
v2: use byte_offset() instead of offset().
Reviewed-by: Matt Turner <mattst88@gmail.com>
v2: Setup for a 64-bit scratch read by checking the type size of the
correct register
v3: Use byte_offset() instead of offset()
Reviewed-by: Matt Turner <mattst88@gmail.com>
A vec4 is 16 bytes and a dvec4 is 32 bytes so for doubles we have
to multiply the reladdr by 2. The reg_offset part is in units of 16
bytes and is used to select the low/high 16-byte chunk of a full
dvec4, so we don't want to multiply that part of the address.
Reviewed-by: Matt Turner <mattst88@gmail.com>
64-bit scratch read/writes require to shuffle data around so we need
to have access to the full 64-bit data. We will do the right thing
for these when we emit the messages.
Reviewed-by: Matt Turner <mattst88@gmail.com>
The BDW PRM says that it is not supported, but it seems that gen7 is also
affected, since doing DepCtrl on double-float instructions leads to
GPU hangs in some cases, which is probably not surprising knowing that
this is not supported in new hardware iterations. The SKL PRMs do not
mention this restriction, so it is probably fine.
Reviewed-by: Matt Turner <mattst88@gmail.com>
This means we would copy propagate partial reads or writes and that can affect
the result.
Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Otherwise we end up producing code that violates the register region
restriction that says that when execsize == width and hstride != 0
the vstride can't be 0.
Reviewed-by: Matt Turner <mattst88@gmail.com>
In gen < 8 instructions that write more than one register need to read
more than one register too. Make sure we don't break that restriction
by copy propagating from a uniform.
Reviewed-by: Matt Turner <mattst88@gmail.com>
Because the meaning of the swizzles and writemasks involved is different,
so replacing the source would lead to different semantics.
Reviewed-by: Matt Turner <mattst88@gmail.com>
v2: Also check if the instruction source target is 64-bit. (Samuel)
Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
In this case we need to shuffle the 64-bit data before we write it
to memory, source from reg_offset + 1 to write components Z and W
and consider that each DF channel is twice as big.
v2: use byte_offset() instead of offset().
Reviewed-by: Matt Turner <mattst88@gmail.com>
Same requirements as for UBO loads.
v2:
- use byte_offset() instead of offset() (Iago)
- keep the const. offset as an immediate like the original code did (Juan)
Reviewed-by: Matt Turner <mattst88@gmail.com>
We need to emit 2 32-bit load messages to load a full dvec4. If only
1 or 2 double components are needed dead-code-elimination will remove
the second one.
We also need to shuffle the result of the 32-bit messages to form
valid 64-bit SIMD4x2 data.
v2:
- use byte_offset() instead of offset() (Iago)
- keep the const. offset as an immediate like the original code did (Juan)
Reviewed-by: Matt Turner <mattst88@gmail.com>
SIMD4x2 64bit data is stored in register space like this:
r0.0:DF x0 y0 z0 w0
r1.0:DF x1 y1 z1 w1
When we need to write data such as this to memory using 32-bit write
messages we need to shuffle it in this fashion:
r0.0:DF x0 y0 x1 y1
r0.1:DF z0 w0 z1 w1
and emit two 32-bit write messages, one for r0.0 at base_offset
and another one for r0.1 at base_offset+16.
We also need to do the inverse operation when we read using 32-bit messages
to produce valid SIMD4x2 64bit data from the data read. We can achieve this
by aplying the exact same shuffling to the data read, although we need to
apply different channel enables since the layout of the data is reversed.
This helper implements the data shuffling logic and we will use it in
various places where we read and write 64bit data from/to memory.
v2 (Curro):
- Use the writemask helper and don't assert on the original writemask
being XYZW.
- Use the Vec4 IR builder to simplify the implementation.
v3 (Iago):
- Use byte_offset() instead of offset().
v3:
- Fix typo (Matt)
- Clarify the example and fix indention (Matt).
Reviewed-by: Matt Turner <mattst88@gmail.com>
The previous patch made sure that we do not generate MAD instructions
for any NIR's 64-bit ffma, but there is nothing preventing i965 from
producing MAD instructions as a result of lowerings or optimization
passes. This patch makes sure that any 64-bit MAD produced inside the
driver after translating from NIR is also converted to MUL+ADD before
we generate code.
v2:
- Use a copy constructor to copy all relevant instruction fields from
the original mad into the add and mul instructions
v3:
- Rename the lowering and fix commit log (Matt)
Signed-off-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
RepCtrl=1 does not work with 64-bit operands so we need to use RepCtrl=0.
In that situation, the regioning generated for the sources seems to be
equivalent to <4,4,1>:DF, so it will only work for components XY, which
means that we have to move any other swizzle to a temporary so that we can
source from channel X (or Y) in MAD and we also need to split the instruction
(we are already scalarizing DF instructions but there is room for
improvement and with MAD would be more restricted in that area)
Also, it seems that MAD operations like this only write proper output for
channels X and Y, so writes to Z and W also need to be done to a temporary
using channels X/Y and then move that to channels Z or W of the actual dst.
As a result the code we produce for native 64-bit MAD instructions is rather
bad, and much worse than just emitting MUL+ADD. For reference, a simple case
of a fully scalarized dvec4 MAD operation requires 15 instructions if we use
native MAD and 8 instructions if we emit ADD+MUL instead. There are some
improvements that we can do to the emission of MAD that might bring the
instruction count down in some cases, but it comes at the expense of a more
complex implementation so it does not seem worth it, at least initially.
This patch makes translation of NIR's 64-bit FMMA instructions produce MUL+ADD
instead of MAD. Currently, there is nothing else in the vec4 backend that emits
MAD instructions, so this is sufficient and it helps optimization passes see
MUL+ADD from the get go.
Reviewed-by: Matt Turner <mattst88@gmail.com>
We make scalar sources in 3src instructions use subnr instead of
swizzles because they don't really use swizzles.
With doubles it is more complicated because we use vstride=0 in
more scenarios in which they don't produce scalar regions. Also
RepCtrl=1 is not allowed with 64-bit operands, so we should avoid
this.
v2: Fix typo (Matt)
Reviewed-by: Matt Turner <mattst88@gmail.com>