| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |
|
|
|
|
|
|
| |
Now that loop_controls no longer creates normatively bound loops,
there is no need for ir_loop::normative_bound or the
lower_bounded_loops pass.
Reviewed-by: Ian Romanick <[email protected]>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Previously, when loop_controls analyzed a loop and found that it had a
fixed bound (known at compile time), it would remove all of the loop
terminators and instead set the loop's normative_bound field to force
the loop to execute the correct number of times.
This made loop unrolling easy, but it had a serious disadvantage.
Since most GPU's don't have a native mechanism for executing a loop a
fixed number of times, in order to implement the normative bound, the
back-ends would have to synthesize a new loop induction variable. As
a result, many loops wound up having two induction variables instead
of one. This caused extra register pressure and unnecessary
instructions.
This patch modifies loop_controls so that it doesn't set the loop's
normative_bound anymore. Instead it leaves one of the terminators in
the loop (the limiting terminator), so the back-end doesn't have to go
to any extra work to ensure the loop terminates at the right time.
This complicates loop unrolling slightly: when deciding whether a loop
can be unrolled, we have to account for the presence of the limiting
terminator. And when we do unroll the loop, we have to remove the
limiting terminator first.
For an example of how this results in more efficient back end code,
consider the loop:
for (int i = 0; i < 100; i++) {
total += i;
}
Previous to this patch, on i965, this loop would compile down to this
(vec4) native code:
mov(8) g4<1>.xD 0D
mov(8) g8<1>.xD 0D
loop:
cmp.ge.f0(8) null g8<4;4,1>.xD 100D
(+f0) if(8)
break(8)
endif(8)
add(8) g5<1>.xD g5<4;4,1>.xD g4<4;4,1>.xD
add(8) g8<1>.xD g8<4;4,1>.xD 1D
add(8) g4<1>.xD g4<4;4,1>.xD 1D
while(8) loop
(notice that both g8 and g4 are loop induction variables; one is used
to terminate the loop, and the other is used to accumulate the total).
After this patch, the same loop compiles to:
mov(8) g4<1>.xD 0D
loop:
cmp.ge.f0(8) null g4<4;4,1>.xD 100D
(+f0) if(8)
break(8)
endif(8)
add(8) g5<1>.xD g5<4;4,1>.xD g4<4;4,1>.xD
add(8) g4<1>.xD g4<4;4,1>.xD 1D
while(8) loop
Reviewed-by: Ian Romanick <[email protected]>
|
| |
|
|
|
|
|
|
| |
This value is now redundant with
loop_variable_state::limiting_terminator->iterations and
ir_loop::normative_bound.
Reviewed-by: Ian Romanick <[email protected]>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The old logic of loop_unroll_visitor::visit_leave(ir_loop *) was:
heuristics to skip unrolling in various circumstances;
if (loop contains more than one jump)
return;
else if (loop contains one jump) {
if (the jump is an unconditional "break" at the end of the loop) {
remove the break and set iteration count to 1;
fall through to simple loop unrolling code;
} else {
for (each "if" statement in the loop body)
see if the jump is a "break" at the end of one of its forks;
if (the "break" wasn't found)
return;
splice the remainder of the loop into the other fork of the "if";
remove the "break";
complex loop unrolling code;
return;
}
}
simple loop unrolling code;
return;
These tasks have been moved to their own functions:
- splice the remainder of the loop into the other fork of the "if"
- simple loop unrolling code
- complex loop unrolling code
And the logic has been flattened to:
heuristics to skip unrolling in various circumstances;
if (loop contains more than one jump)
return;
if (loop contains no jumps) {
simple loop unroll;
return;
}
if (the jump is an unconditional "break" at the end of the loop) {
remove the break;
simple loop unroll with iteration count of 1;
return;
}
for (each "if" statement in the loop body) {
if (the jump is a "break" at the end of one of its forks) {
splice the remainder of the loop into the other fork of the "if";
remove the "break";
complex loop unroll;
return;
}
}
This will make it easier to modify the loop unrolling algorithm in a
future patch.
Reviewed-by: Ian Romanick <[email protected]>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Previously, the sole responsibility of loop_analysis was to find all
the variables referenced in the loop that are either loop constant or
induction variables, and find all of the simple if statements that
might terminate the loop. The remainder of the analysis necessary to
determine how many times a loop executed was performed by
loop_controls.
This patch makes loop_analysis also responsible for determining the
number of iterations after which each loop terminator will terminate
the loop, and for figuring out which terminator will terminate the
loop first (I'm calling this the "limiting terminator").
This will allow loop unrolling to make use of information that was
previously only visible from loop_controls, namely the identity of the
limiting terminator.
Reviewed-by: Ian Romanick <[email protected]>
|
| |
|
|
|
|
|
|
| |
Patches to follow will introduce code into the loop_terminator
constructor. Allocating loop_terminator using new(mem_ctx) syntax
will ensure that the constructor runs.
Reviewed-by: Ian Romanick <[email protected]>
|
| |
|
|
|
|
|
|
|
|
|
| |
When loop_control_visitor::visit_leave(ir_loop *) is analyzing a loop
terminator that acts on a certain ir_variable, it doesn't need to walk
the list of induction variables to find the loop_variable entry
corresponding to the variable. It can just look it up in the
loop_variable_state hashtable and verify that the loop_variable entry
represents an induction variable.
Reviewed-by: Ian Romanick <[email protected]>
|
| |
|
|
|
|
|
|
|
| |
These fields were part of some planned optimizations that never
materialized. Remove them for now to simplify things; if we ever get
round to adding the optimizations that would require them, we can
always re-introduce them.
Reviewed-by: Ian Romanick <[email protected]>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
This patch replaces the ir_loop fields "from", "to", "increment",
"counter", and "cmp" with a single integer ("normative_bound") that
serves the same purpose.
I've used the name "normative_bound" to emphasize the fact that the
back-end is required to emit code to prevent the loop from running
more than normative_bound times. (By contrast, an "informative" bound
would be a bound that is informational only).
Reviewed-by: Jordan Justen <[email protected]>
Reviewed-by: Ian Romanick <[email protected]>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Previously, all of the back-ends (ir_to_mesa, st_glsl_to_tgsi, and the
i965 fs and vec4 visitors) had nearly identical logic for handling
bounded loops. This replaces the duplicate logic with an equivalent
lowering pass that is used by all the back-ends.
Note: on i965, there is a slight increase in instruction count. For
example, a loop like this:
for (int i = 0; i < 100; i++) {
total += i;
}
would previously compile down to this (vec4) native code:
mov(8) g4<1>.xD 0D
mov(8) g8<1>.xD 0D
loop:
cmp.ge.f0(8) null g8<4;4,1>.xD 100D
(+f0) break(8)
add(8) g5<1>.xD g5<4;4,1>.xD g4<4;4,1>.xD
add(8) g8<1>.xD g8<4;4,1>.xD 1D
add(8) g4<1>.xD g4<4;4,1>.xD 1D
while(8) loop
After this patch, the "(+f0) break(8)" turns into:
(+f0) if(8)
break(8)
endif(8)
because the back-end isn't smart enough to recognize that "if
(condition) break;" can be done using a conditional break instruction.
However, it should be relatively easy for a future peephole
optimization to properly optimize this.
Reviewed-by: Jordan Justen <[email protected]>
Reviewed-by: Ian Romanick <[email protected]>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Previously, loop analysis would set
this->conditional_or_nested_assignment based on the most recently
visited assignment to the variable. As a result, if a vaiable was
assigned to more than once in a loop, the flag might be set
incorrectly. For example, in a loop like this:
int x;
for (int i = 0; i < 3; i++) {
if (i == 0)
x = 10;
...
x = 20;
...
}
loop analysis would have incorrectly concluded that all assignments to
x were unconditional.
In practice this was a benign bug, because
conditional_or_nested_assignment is only used to disqualify variables
from being considered as loop induction variables or loop constant
variables, and having multiple assignments also disqualifies a
variable from being considered as either of those things.
Still, we should get the analysis correct to avoid future confusion.
Reviewed-by: Jordan Justen <[email protected]>
Reviewed-by: Ian Romanick <[email protected]>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Previously, when visiting an ir_call, loop analysis would only mark
the innermost enclosing loop as containing a call. As a result, when
encountering a loop like this:
for (i = 0; i < 3; i++) {
for (int j = 0; j < 3; j++) {
foo();
}
}
it would incorrectly conclude that the outer loop ran three times.
(This is not certain; if foo() modifies i, then the outer loop might
run more or fewer times).
Fixes piglit test "vs-call-in-nested-loop.shader_test".
Reviewed-by: Jordan Justen <[email protected]>
Reviewed-by: Ian Romanick <[email protected]>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Previously, when visiting a variable dereference, loop analysis would
only consider its effect on the innermost enclosing loop. As a
result, when encountering a loop like this:
for (int i = 0; i < 3; i++) {
for (int j = 0; j < 3; j++) {
...
i = 2;
}
}
it would incorrectly conclude that the outer loop ran three times.
Fixes piglit test "vs-inner-loop-modifies-outer-loop-var.shader_test".
Reviewed-by: Jordan Justen <[email protected]>
Reviewed-by: Ian Romanick <[email protected]>
|
| |
|
|
|
|
|
| |
This function is about to get more complex.
Reviewed-by: Jordan Justen <[email protected]>
Reviewed-by: Ian Romanick <[email protected]>
|
| |
|
|
|
|
|
|
|
|
| |
Fast color clears of MSAA buffers work just like fast color clears
with non-MSAA buffers, except that the alignment and scaledown
requirements are different.
Reviewed-by: Kenneth Graunke <[email protected]>
Reviewed-by: Chad Versace <[email protected]>
Reviewed-by: Anuj Phogat <[email protected]>
|
| |
|
|
|
|
|
|
|
| |
This will make it easier to add fast color clear support to MSAA
buffers, since they have different alignment and scaling requirements.
Reviewed-by: Chad Versace <[email protected]>
Reviewed-by: Anuj Phogat <[email protected]>
Reviewed-by: Kenneth Graunke <[email protected]>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Previously, we didn't do multisample blorp clears because we couldn't
figure out how to get them to work. The reason for this was because
we weren't setting the brw_blorp_params num_samples field consistently
with dst.num_samples. Now that those two fields have been collapsed
down into one, we can do multisample blorp clears.
However, we need to do a few other pieces of bookkeeping to make them
work correctly in all circumstances:
- Since blorp clears may now operate on multisampled window system
framebuffers, they need to call
intel_renderbuffer_set_needs_downsample() to ensure that a
downsample happens before buffer swap (or glReadPixels()).
- When clearing a layered multisample buffer attachment using UMS or
CMS layout, we need to advance layer by multiples of num_samples
(since each logical layer is associated with num_samples physical
layers).
Note: we still don't do multisample fast color clears; more work needs
to be done to enable those.
Reviewed-by: Chad Versace <[email protected]>
Reviewed-by: Anuj Phogat <[email protected]>
Reviewed-by: Kenneth Graunke <[email protected]>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Previously, brw_blorp_params contained two fields for determining
sample count: num_samples (which determined the multisample
configuration of the rendering pipeline) and dst.num_samples (which
determined the multisample configuration of the render target
surface). This was redundant, since both fields had to be set to the
same value to avoid rendering errors.
This patch eliminates num_samples to avoid future confusion.
Reviewed-by: Chad Versace <[email protected]>
Reviewed-by: Anuj Phogat <[email protected]>
Reviewed-by: Kenneth Graunke <[email protected]>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
This patch renames the enum that's used to keep track of fast clear
state from "mcs_state" to "fast_clear_state", and it removes the enum
value INTEL_MCS_STATE_MSAA (which previously meant, "this is an MSAA
buffer, so we're not keeping track of fast clear state"). The only
real purpose that enum value was serving was to prevent us from trying
to do fast clear resolves on MSAA buffers, and it's just as easy to
prevent that by checking the buffer's msaa_layout.
This paves the way for implementing fast clears of MSAA buffers.
Reviewed-by: Chad Versace <[email protected]>
Reviewed-by: Anuj Phogat <[email protected]>
Reviewed-by: Kenneth Graunke <[email protected]>
|
| |
|
|
|
|
|
|
|
| |
The hardware blitter doesn't understand multisampled layouts, so
there's no way this could possibly succeed.
Reviewed-by: Chad Versace <[email protected]>
Reviewed-by: Anuj Phogat <[email protected]>
Reviewed-by: Kenneth Graunke <[email protected]>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The "layer" parameters used in blorp, and the
intel_renderbuffer::mt_layer field, represent a physical layer rather
than a logical layer. This is important for 2D multisample arrays on
Gen7+ because the UMS and CMS multisample layouts use N physical
layers to represent each logical layer, where N is the number of
samples.
Also add an assertion to blorp to help catch bugs if we fail to follow
these conventions.
Reviewed-by: Chad Versace <[email protected]>
Reviewed-by: Anuj Phogat <[email protected]>
Reviewed-by: Kenneth Graunke <[email protected]>
|
| |
|
|
|
|
|
|
| |
Clarify the fact that we only optimize full buffer clears using fast
color clear, and why.
Reviewed-by: Chad Versace <[email protected]>
Reviewed-by: Kenneth Graunke <[email protected]>
|
| |
|
|
|
|
| |
Reviewed-by: Alex Deucher <[email protected]>
CC: "9.2" "10.0" <[email protected]>
|
| |
|
|
|
|
| |
Reviewed-by: Alex Deucher <[email protected]>
CC: "9.2" "10.0" <[email protected]>
|
| |
|
|
|
|
|
|
|
|
|
| |
Based on comments by Benjamin Morris <[email protected]> in
http://lists.freedesktop.org/archives/nouveau/2013-December/015328.html
This adds setting of is_long_term, and updates a few field names we were
unclear about.
Signed-off-by: Ilia Mirkin <[email protected]>
Cc: "10.0" <[email protected]>
|
| |
|
|
|
| |
Signed-off-by: Ilia Mirkin <[email protected]>
Cc: "10.0" <[email protected]>
|
| |
|
|
|
|
|
|
|
|
| |
Create the ref_bo without any storage type flags set for now. The issue
probably arises from our use of the additional buffer space at the end
of the ref_bo. It should probably be split up in the future.
Signed-off-by: Ilia Mirkin <[email protected]>
Tested-by: Martin Peres <[email protected]>
Cc: "10.0" <[email protected]>
|
| |
|
|
| |
Signed-off-by: Ilia Mirkin <[email protected]>
|
| |
|
|
|
|
| |
Fixes the texelFetch piglit tests
Signed-off-by: Ilia Mirkin <[email protected]>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
With this patch, generate_fs_loop will clamp any fragment shader depth writes
to the viewport's min and max depth values. Viewport selection is determined
by the geometry shader output for the viewport array index. If no index is
specified, then the default viewport index is zero. Semantics for this path
can be found in draw_clamp_viewport_idx and lp_clamp_viewport_idx.
lp_jit_viewport was created to store viewport information visible to JIT code,
and is validated when the LP_NEW_VIEWPORT dirty flag is set.
lp_rast_shader_inputs is responsible for passing the viewport_index through
the rasterizer stage to fragment stage (via lp_jit_thread_data).
Reviewed-by: Roland Scheidegger <[email protected]>
Reviewed-by: José Fonseca <[email protected]>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
The Wayland EGL platform now respects the eglSwapInterval value. The value is
clamped to either 0 or 1 because it is difficult (and probably not useful) to
sync to more than 1 redraw.
The main change is that if the swap interval is 0 then Mesa won't install a
frame callback so that eglSwapBuffers can be executed as often as necessary.
Instead it will do a sync request after the swap buffers. It will block for
sync complete event in get_back_bo instead of the frame callback. The
compositor is likely to send a release event while processing the new buffer
attach and this makes sure we will receive that before deciding whether to
allocate a new buffer.
If there are no buffers available then instead of returning with an error,
get_back_bo will now poll the compositor by repeatedly sending sync requests
every 10ms. This is a last resort and in theory this shouldn't happen because
there should be no reason for the compositor to hold on to more than three
buffers. That means whenever we attach the fourth buffer we should always get
an immediate release event which should come in with the notification for the
first sync request that we are throttled to.
When the compositor is directly scanning out from the application's buffer it
may end up holding on to three buffers. These are the one that is is currently
scanning out from, one that has been given to DRM as the next buffer to flip
to, and one that has been attached and will be given to DRM as soon as the
previous flip completes. When we attach a fourth buffer to the compositor it
should replace that third buffer so we should get a release event immediately
after that. This patch therefore also changes the number of buffer slots to 4
so that we can accomodate that situation.
If DRM eventually gets a way to cancel a pending page flip then the compositors
can be changed to only need to hold on to two buffers and this value can be
put back to 3.
This also moves the vblank configuration defines from platform_x11.c to the
common egl_dri2.h header so they can be shared by both platforms.
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Consider a typical game-style main loop which might be like this:
while (1) {
draw_something();
eglSwapBuffers();
}
In this case the game is relying on eglSwapBuffers to throttle to a sensible
frame rate. Previously this game would end up using three buffers even though
it should only need two. This is because Mesa decides whether to allocate a
new buffer in get_back_bo which would be before it has tried to read any
events from the compositor so it wouldn't have seen any buffer release events
yet.
This patch just moves the block for the frame callback to get_back_bo.
Typically the compositor will send a release event immediately after one of
the attaches so if we block for the frame callback here then we can be sure to
have completed at least one roundtrip and received that release event after
attaching the previous buffer before deciding whether to allocate a new one.
dri2_swap_buffers always calls get_back_bo so even if the client doesn't
render anything we will still be sure to block to the frame callback. The code
to create the new frame callback has been moved to after this call so that we
can be sure to have cleared the previous frame callback before requesting a
new one.
|
| |
|
|
|
|
|
|
|
|
| |
This patch fixes this MinGW build error.
CC glapi_gentable.lo
glapi_gentable.c:47:19: fatal error: dlfcn.h: No such file or directory
Signed-off-by: Vinson Lee <[email protected]>
Reviewed-by: Brian Paul <[email protected]>
|
| | |
|
| |
|
|
| |
Signed-off-by: Rob Clark <[email protected]>
|
| |
|
|
| |
Signed-off-by: Rob Clark <[email protected]>
|
| |
|
|
|
| |
Signed-off-by: Chris Forbes <[email protected]>
Reviewed-by: Francisco Jerez <[email protected]>
|
| |
|
|
|
| |
Signed-off-by: Chris Forbes <[email protected]>
Reviewed-by: Francisco Jerez <[email protected]>
|
| |
|
|
|
|
|
|
| |
Drivers will need to look at this to decide if they need to do
per-sample fragment shader dispatch.
Signed-off-by: Chris Forbes <[email protected]>
Reviewed-by: Francisco Jerez <[email protected]>
|
| |
|
|
|
| |
Signed-off-by: Chris Forbes <[email protected]>
Reviewed-by: Francisco Jerez <[email protected]>
|
| |
|
|
|
| |
Signed-off-by: Chris Forbes <[email protected]>
Reviewed-by: Francisco Jerez <[email protected]>
|
| |
|
|
|
| |
Signed-off-by: Chris Forbes <[email protected]>
Reviewed-by: Francisco Jerez <[email protected]>
|
| |
|
|
|
|
|
|
|
|
|
| |
My understanding is that Broadwell retains the same SCS mechanism
that Haswell has, so even if the underlying issue with this format
is not fixed, the w/a will be applied in SCS rather than needing
shader code.
Signed-off-by: Chris Forbes <[email protected]>
Cc: Kenneth Graunke <[email protected]>
Reviewed-by: Kenneth Graunke <[email protected]>
|
| |
|
|
|
|
|
|
|
| |
Now that all the pieces are in place, this should provide
a nice performance boost for apps using multisample textures.
Signed-off-by: Chris Forbes <[email protected]>
Reviewed-by: Paul Berry <[email protected]>
Reviewed-by: Kenneth Graunke <[email protected]>
|
| |
|
|
|
| |
Signed-off-by: Chris Forbes <[email protected]>
Reviewed-by: Paul Berry <[email protected]>
|
| |
|
|
|
| |
Signed-off-by: Chris Forbes <[email protected]>
Reviewed-by: Paul Berry <[email protected]>
|
| |
|
|
|
|
| |
Signed-off-by: Chris Forbes <[email protected]>
Reviewed-by: Paul Berry <[email protected]>
Reviewed-by: Kenneth Graunke <[email protected]>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
We need to emit extra shader code in this case to sample the
MCS surface first; we can't just blindly do this all the time
since IVB will sometimes try to access the MCS surface even if
disabled.
V3: Use actual MSAA layout from the texture's mt, rather
then computing what would have been used based on the format.
This is simpler and less fragile - there's at least one case where
we might want to have a texture's MSAA layout change based on what
the app does (CMS SINT falling back to UMS if the app ever attempts
to render to it with a channel disabled.)
This also obsoletes V2's 1/10 -- compute_msaa_layout can now remain
an implementation detail of the miptree code.
Signed-off-by: Chris Forbes <[email protected]>
Reviewed-by: Paul Berry <[email protected]>
|
| |
|
|
|
|
|
|
|
|
| |
This gives us correct behavior for both renderbuffers (which previously
worked) and multisample textures (which would never get an MCS surface
allocated, even if CMS layout was selected)
Signed-off-by: Chris Forbes <[email protected]>
Reviewed-by: Paul Berry <[email protected]>
Reviewed-by: Kenneth Graunke <[email protected]>
|
| |
|
|
|
|
|
|
| |
Previously this was only done for render targets.
Signed-off-by: Chris Forbes <[email protected]>
Reviewed-by: Paul Berry <[email protected]>
Reviewed-by: Kenneth Graunke <[email protected]>
|
