Without an explicit volatile, aarch64 gcc will copy queue->tail into a
register outside the loop and then continuously compare two registers
against each other. The *(volatile LONG *)& forces gcc to re-read the
memory every iteration. Therefore, queue_is_empty as it is written will
only work from the CS thread.
Signed-off-by: Stefan Dösinger <stefan@codeweavers.com>
Signed-off-by: Henri Verbeet <hverbeet@codeweavers.com>
Signed-off-by: Alexandre Julliard <julliard@winehq.org>
ARB_pipeline_statistics_query is included in OpenGL 4.6.
Signed-off-by: Józef Kucia <jkucia@codeweavers.com>
Signed-off-by: Henri Verbeet <hverbeet@codeweavers.com>
Signed-off-by: Alexandre Julliard <julliard@winehq.org>
Makes sure that we do not overwrite source parameters for consecutive
GLSL function calls when the instruction is translated into multiple
lines.
Signed-off-by: Józef Kucia <jkucia@codeweavers.com>
Signed-off-by: Henri Verbeet <hverbeet@codeweavers.com>
Signed-off-by: Alexandre Julliard <julliard@winehq.org>
We still want to avoid invalidating the system memory copy for these surfaces,
but we've become a lot better at preserving the current location for
resources, to the point that the special case actually hurts. In particular,
if the resource wasn't already current on the CPU, using the CPU blitter would
cause us to read it back, which is what we're trying to avoid.
Signed-off-by: Henri Verbeet <hverbeet@codeweavers.com>
Signed-off-by: Alexandre Julliard <julliard@winehq.org>
Since that would imply reading them back, with associated format conversion.
Signed-off-by: Henri Verbeet <hverbeet@codeweavers.com>
Signed-off-by: Alexandre Julliard <julliard@winehq.org>
The issue isn't so much stencil clears as clearing either depth or stencil on
formats that contain both.
Signed-off-by: Henri Verbeet <hverbeet@codeweavers.com>
Signed-off-by: Alexandre Julliard <julliard@winehq.org>
Use fences for the APPLESYNC codepath, and implement event queries
on top of fences.
Signed-off-by: Józef Kucia <jkucia@codeweavers.com>
Signed-off-by: Henri Verbeet <hverbeet@codeweavers.com>
Signed-off-by: Alexandre Julliard <julliard@winehq.org>
Fixes a regression introduced by commit
be20ddc38b.
Signed-off-by: Józef Kucia <jkucia@codeweavers.com>
Signed-off-by: Henri Verbeet <hverbeet@codeweavers.com>
Signed-off-by: Alexandre Julliard <julliard@winehq.org>
Fixes a regression introduced in ee17d7ba1b.
Signed-off-by: Sebastian Lackner <sebastian@fds-team.de>
Signed-off-by: Henri Verbeet <hverbeet@codeweavers.com>
Signed-off-by: Alexandre Julliard <julliard@winehq.org>
Fixes a regression introduced in cad4badbcf.
Signed-off-by: Sebastian Lackner <sebastian@fds-team.de>
Signed-off-by: Henri Verbeet <hverbeet@codeweavers.com>
Signed-off-by: Alexandre Julliard <julliard@winehq.org>
Fixes a regression introduced in 62ca4f3826.
Signed-off-by: Sebastian Lackner <sebastian@fds-team.de>
Signed-off-by: Henri Verbeet <hverbeet@codeweavers.com>
Signed-off-by: Alexandre Julliard <julliard@winehq.org>
This is a workaround for a bug in Mesa GLSL linker. The bug is triggered
by GLSL generated for SM3 shaders when shader inputs are not read and
interface blocks are used for shader inputs/outputs (core profile code
path).
Declarations for unused shader inputs are not generated for pixel
shaders. We still declare shader outputs in the previous shader stage,
but it's allowed by the GLSL 1.50 spec:
"Only output variables that are read by the subsequent pipeline stage
need to be written; it is allowed to have superfluous declarations of
output variables."
The table in section 4.3.4 (Input Variables) of the GLSL 4.20 spec even
more explicitly allows shader output variables with no matching input
variables declared in the consuming shader.
Signed-off-by: Józef Kucia <jkucia@codeweavers.com>
Signed-off-by: Henri Verbeet <hverbeet@codeweavers.com>
Signed-off-by: Alexandre Julliard <julliard@winehq.org>
The existing checks should be sufficient, but the GL_VERSION one is
potentially more reliable.
Signed-off-by: Henri Verbeet <hverbeet@codeweavers.com>
Signed-off-by: Alexandre Julliard <julliard@winehq.org>
For D3D10+.
When WINED3D_RTYPE_NONE is passed the wined3d_check_device_format()
function checks if a format is supported for all reasonable resource
types for a given usage. For now render target and depth stencil support
can be queried using this method.
Signed-off-by: Józef Kucia <jkucia@codeweavers.com>
Signed-off-by: Henri Verbeet <hverbeet@codeweavers.com>
Signed-off-by: Alexandre Julliard <julliard@winehq.org>
Matteo Bruni