This is basically a re-write of the slab allocator into a design where
multiple child pools are linked to a parent pool. The intention is that
every (GL, pipe) context has its own child pool, while the corresponding
parent pool is held by the winsys or screen, or possibly the GL share group.
The fast path is still used when objects are freed by the same child pool
that allocated them. However, it is now also possible to free an object in a
different pool, as long as they belong to the same parent. Objects also
survive the destruction of the (child) pool from which they were allocated.
The slow path will return freed objects to the child pool from which they
were originally allocated. If that child pool was destroyed, the corresponding
page is considered an orphan and will be freed once all objects in it have
been freed.
This allocation pattern is required for pipe_transfers that correspond to
(GL) buffer object mappings when the mapping is created in one context
which is later destroyed while other contexts of the same share group live
on -- see the bug report referenced below.
Note that individual drivers do need to migrate to the new interface in
order to benefit and fix the bug.
v2: use singly-linked lists everywhere
v3: use p_atomic_set for page->u.num_remaining
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=97894
This is motivated by the fact that p_atomic_read and p_atomic_set may
somewhat surprisingly not do the right thing in the old version: while
stores and loads are de facto atomic at least on x86, the compiler may
apply re-ordering and speculation quite liberally. Basically, the old
version uses the "relaxed" memory ordering.
The new ordering always uses acquire/release ordering. This is the
strongest possible memory ordering that doesn't require additional
fence instructions on x86. (And the only stronger ordering is
"sequentially consistent", which is usually more than you need anyway.)
I would feel more comfortable if p_atomic_set/read in the old
implementation were at least using volatile loads and stores, but I
don't see a way to get there without typeof (which we cannot use here
since the code is compiled with -std=c99).
Eventually, we should really just move to something that is based on
the atomics in C11 / C++11.
Acked-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Anv programs the hardware to use L3 data cache if we use either SSBOs or
images in the shaders, we can program i965 the same way.
gl_shader_program has a bit of a confusing named field with
'NumAtomicBuffers'. It doesn't tell how many buffers are accessed by the
shader in an atomic way but instead the number of atomic counters
manipulated by the shader.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
ES3-CTS.functional.negative_api.buffer.framebuffer_texture2d expects
glFramebufferTexture[123]D to raise GL_INVALID_ENUM when
supplied a completely bogus textarget parameter (i.e. 0xffffffff).
This is at odds with the spec. GLES 3.1 says:
"An INVALID_OPERATION error is generated if texture is not zero and
textarget is not one of TEXTURE_2D, TEXTURE_2D_MULTISAMPLE, or one
of the cube map face targets from table 8.21."
(and GLES 3.0 and GL 4.5 both have similar text). However, GL has a
general guideline that says:
"If a command that requires an enumerated value is passed a symbolic
constant that is not one of those specified as allowable for that
command, an INVALID_ENUM error is generated."
Apparently other vendors reconcile these two rules as follows: GL should
raise INVALID_OPERATION for actual texture target enumeration values
which are not allowed for this particular glFramebufferTexture*D call.
Any value that is not a texture target should result in GL_INVALID_ENUM.
For example, glFramebufferTexture2D with GL_TEXTURE_1D would result in
INVALID_OPERATION because it is a real texture target, but not allowed
for the 2D version of the function. But calling it with GL_FRONT would
result in INVALID_ENUM, as that isn't even a texture target.
Fixes:
- {ES3-CTS,dEQP-GLES3}.functional.negative_api.buffer.framebuffer_texture2d
- {ES31-CTS,ES32-CTS,dEQP-GLES31}.functional.debug.negative_coverage.get_error.buffer.framebuffer_texture2d
References: https://gitlab.khronos.org/opengl/cts/merge_requests/387
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Having one top-level switch statement covering all known texture targets
will make the next change easier to implement.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
From the less man page:
"Warning: when the -r option is used, less cannot keep track of the
actual appearance of the screen (since this depends on how the
screen responds to each type of control character). Thus, various
display problems may result, such as long lines being split in the
wrong place."
Lines which are too long to fit in the terminal would be word wrapped,
but unfortunately less would get confused about which line it was on,
and text would be drawn on top of other text. The most noticable case
was shader assembly, which is frequently too wide for an 80 character
terminal, and thus would be drawn on top of the following state packets,
making them completely unreadable.
Using -R instead of -r fixes this problem by only allowing color escape
sequences. (Notably, Git's implicit pager invocation uses -R.)
Unfortunately, it means our "clear to the end of the line" hack for
extending the blue bar headers won't work anymore.
Word wrapping usually isn't terribly readable, anyway, so we also add
the -S option (chop long lines) to restrict it to the terminal width.
(You can hit the left and right arrow keys to scroll sideways.)
Then, for a new blue bar hack, we can use a printf specifier to pad
the command packet names to be 80 characters long (arbitrarily), which
extends them "far enough" to look good, and doesn't require us to use
ioctls to determine the terminal width.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Sirisha Gandikota <sirisha.gandikota@intel.com>
The atom that uploads push constants listens to _NEW_TRANSFORM for
legacy clip plane handling. On Sandybridge, the gen6_vs_state atom
emits 3DSTATE_CONSTANT_VS as well as 3DSTATE_VS, so it needs to listen
to the same set of conditions.
However, it looks like Gen7 doesn't need this. The push constant atom
emits 3DSTATE_CONSTANT_VS directly, and the gen7_vs_state atom that
emits 3DSTATE_VS doesn't have a dependency on ctx->Transform.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
We need to free prog_data for TCS/TES too.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Timothy Arceri <timothy.arcero@collabora.com>
CACHE_NEW_CS_PROG hasn't existed in quite a long time...the old
comment was there, but not the actual bit.
Cc: mesa-stable@lists.freedesktop.org
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Needed for user clip plane enables. Broken since this code was
introduced.
Cc: mesa-stable@lists.freedesktop.org
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
This will make it easier to add more operations.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
v2: Delete some stray debug code notice by Iago.
v3: Massive rebase on new ir_function_signature::intrinsic_id mechanism.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com> [v1]
Acked-by: Ilia Mirkin <imirkin@alum.mit.edu>
Otherwise grepping for where atomic_counter_inc and friends are defined
is a very frustrating experience.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Acked-by: Ilia Mirkin <imirkin@alum.mit.edu>
Just generate an __intrinsic_atomic_add with a negated parameter.
Some background on the non-obvious reasons for the the big change to
builtin_builder::call()... this is cribbed from some discussion with
Ilia on mesa-dev.
Why change builtin_builder::call() to allow taking dereferences and
create them here rather than just feeding in the ir_variables directly?
The problem is the neg_data ir_variable node would have to be in two
lists at the same time: the instruction stream and parameters. The
ir_variable node is automatically added to the instruction stream by the
call to make_temp. Restructuring the code so that the ir_variables
could be in parameters then move them to the instruction stream would
have been pretty terrible.
ir_call in the instruction stream has an exec_list that contains
ir_dereference_variable nodes.
The builtin_builder::call method previously took an exec_list of
ir_variables and created a list of ir_dereference_variable. All of the
original users of that method wanted to make a function call using
exactly the set of parameters passed to the built-in function (i.e.,
call __intrinsic_atomic_add using the parameters to atomicAdd). For
these users, the list of ir_variables already existed: the list of
parameters in the built-in function signature.
This new caller doesn't do that. It wants to call a function with a
parameter from the function and a value calculated in the function. So,
I changed builtin_builder::call to take a list that could either be a
list of ir_variable or a list of ir_dereference_variable. In the former
case it behaves just as it previously did. In the latter case, it uses
(and removes from the input list) the ir_dereference_variable nodes
instead of creating new ones.
text data bss dec hex filename
6036395 283160 28608 6348163 60dd83 lib64/i965_dri.so before
6036923 283160 28608 6348691 60df93 lib64/i965_dri.so after
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Acked-by: Ilia Mirkin <imirkin@alum.mit.edu>
text data bss dec hex filename
6036491 283160 28608 6348259 60dde3 lib64/i965_dri.so before
6036395 283160 28608 6348163 60dd83 lib64/i965_dri.so after
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Acked-by: Ilia Mirkin <imirkin@alum.mit.edu>
This necessetated renaming the is_intrinsic field to _is_intrinsic. The
next commit will remove the field.
text data bss dec hex filename
6036507 283160 28608 6348275 60ddf3 lib64/i965_dri.so before
6036491 283160 28608 6348259 60dde3 lib64/i965_dri.so after
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Acked-by: Ilia Mirkin <imirkin@alum.mit.edu>
text data bss dec hex filename
6038043 283160 28608 6349811 60e3f3 lib64/i965_dri.so before
6036507 283160 28608 6348275 60ddf3 lib64/i965_dri.so after
v2: s/ir_intrinsic_atomic_sub/ir_intrinsic_atomic_counter_sub/. Noticed
by Ilia.
v3: Silence unhandled enum in switch warnings in st_glsl_to_tgsi.
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Acked-by: Ilia Mirkin <imirkin@alum.mit.edu>
text data bss dec hex filename
6037483 283160 28608 6349251 60e1c3 lib64/i965_dri.so before
6038043 283160 28608 6349811 60e3f3 lib64/i965_dri.so after
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Acked-by: Ilia Mirkin <imirkin@alum.mit.edu>
st_glsl_to_tgsi only calls lower_instructions once (instead of in a
loop), so the ir_binop_carry generated would not get lowered. Fixes
assertion failure
state_tracker/st_glsl_to_tgsi.cpp:2265: void glsl_to_tgsi_visitor::visit_expression(ir_expression*, st_src_reg*): Assertion `!"Invalid ir opcode in glsl_to_tgsi_visitor::visit()"' failed.
on softpipe in 16 piglit tests:
mesa_shader_integer_functions/execution/built-in-functions/fs-imulExtended-nonuniform.shader_test
mesa_shader_integer_functions/execution/built-in-functions/fs-imulExtended-only-msb-nonuniform.shader_test
mesa_shader_integer_functions/execution/built-in-functions/fs-imulExtended-only-msb.shader_test
mesa_shader_integer_functions/execution/built-in-functions/fs-imulExtended.shader_test
mesa_shader_integer_functions/execution/built-in-functions/fs-umulExtended-nonuniform.shader_test
mesa_shader_integer_functions/execution/built-in-functions/fs-umulExtended-only-msb-nonuniform.shader_test
mesa_shader_integer_functions/execution/built-in-functions/fs-umulExtended-only-msb.shader_test
mesa_shader_integer_functions/execution/built-in-functions/fs-umulExtended.shader_test
mesa_shader_integer_functions/execution/built-in-functions/vs-imulExtended-nonuniform.shader_test
mesa_shader_integer_functions/execution/built-in-functions/vs-imulExtended-only-msb-nonuniform.shader_test
mesa_shader_integer_functions/execution/built-in-functions/vs-imulExtended-only-msb.shader_test
mesa_shader_integer_functions/execution/built-in-functions/vs-imulExtended.shader_test
mesa_shader_integer_functions/execution/built-in-functions/vs-umulExtended-nonuniform.shader_test
mesa_shader_integer_functions/execution/built-in-functions/vs-umulExtended-only-msb-nonuniform.shader_test
mesa_shader_integer_functions/execution/built-in-functions/vs-umulExtended-only-msb.shader_test
mesa_shader_integer_functions/execution/built-in-functions/vs-umulExtended.shader_test
Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
This function is only ever used by an assert() this fixes an
unused function warning in release builds.
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
v2: Make sure that with num_lods > 1 and min_filter != mag_filter we
still enter the splitting path. So this case would still use 4-wide aos
path (as a side note, the 4-wide aos sampling path could actually be
improved quite a bit if we have avx2, by just doing the filtering with
256bit vectors).
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
_eglInitSync checked that the display supported the sync type (such as
EGL_SYNC_FENCE), and did it wrong. When the check failed it emitted
EGL_BAD_ATTRIBUTE, but sometimes EGL_BAD_PARAMETER is needed.
_eglCreateSync already does the error checking, and it does it right.
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
When the function encountered an error, it effectively returned
immediately. However, it did so indirectly by breaking out of a loop.
Replace the loop breakout with a explicit 'return'.
Do the same for _eglParseSyncAttribList64 too.
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
This function converts an attribute list from EGLint[] to EGLAttrib[].
Will be used in following patches to cleanup EGLSync attribute parsing.
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
When the user called eglCreateSync64KHR on a display without
EGL_KHR_cl_event2 (the only extension that exposes it), we returned
EGL_NO_SYNC but did not update the error code.
We also did the same for eglCreateSync on a display without EGL 1.5.
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
The function stores EGLAttrib values in EGLint variables. On 64-bit
systems, this truncated the values.
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
On the error path, eglGetSyncAttribKHR neglected to unlock the
EGLDisplay before returning.
Fixes deadlock in dEQP-EGL.functional.fence_sync.invalid.get_invalid_value.
Cc: mesa-stable@lists.freedesktop.org
Cc: Mark Janes <mark.a.janes@intel.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
This patch causes 2 regressions in khronos' gles cts tests
on various intel platforms.
Failing tests:
ES3-CTS.functional.state_query.integers.viewport_getinteger
ES3-CTS.functional.state_query.integers.viewport_getfloat
Here is an explanation of what's causing the failures:
CTS tests are not clamping the x, y location of the viewport's
bottom-left corner as recommended by ARB_viewport_array and
OES_viewport_array:
"The location of the viewport's bottom-left corner, given by (x,y), are
clamped to be within the implementation-dependent viewport bounds range.
The viewport bounds range [min, max] tuple may be determined by
calling GetFloatv with the symbolic constant VIEWPORT_BOUNDS_RANGE_OES"
Khronos CTS merge request to fix the test case:
https://gitlab.khronos.org/opengl/cts/merge_requests/399
V2: Initialize the relevant variables for GL_OES_viewport_array on gen8+
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Nicolai Hähnle
Lionel Landwerlin
Kenneth Graunke
Ian Romanick
Timothy Arceri
Jose Fonseca
Roland Scheidegger
Chad Versace
Anuj Phogat