- 14 Jun, 2022 2 commits
-
-
Igor Torrente authoredvrend_sampler_view_reference, vrend_shader_state_reference and vrend_resource_reference have some obscure address calculation that cause confusion for some that don't fully understand what is happening there. Adds new comments to explain what is happening in these codes and why we aren't really de-referencing any NULL pointer. Signed-off-by:
Igor Torrente <igor.torrente@collabora.com> -
Signed-off-by:
Italo Nicola <italonicola@collabora.com> Part-of: <virgl/virglrenderer!829>
-
- 13 Jun, 2022 2 commits
-
-
Signed-off-by:
Yiwei Zhang <zzyiwei@chromium.org> Reviewed-by:
Ryan Neph <ryanneph@google.com> Part-of: <virgl/virglrenderer!821>
-
Signed-off-by:
-
- 09 Jun, 2022 1 commit
-
-
Corentin Noël authored
Signed-off-by:
Corentin Noël <corentin.noel@collabora.com> Part-of: <virgl/virglrenderer!827>
-
- 07 Jun, 2022 1 commit
-
-
Signed-off-by:
Dawn Han <dawnhan@google.com> Part-of: <virgl/virglrenderer!826>
-
- 02 Jun, 2022 6 commits
-
-
Signed-off-by:
Akihiko Odaki <akihiko.odaki@gmail.com> Part-of: <virgl/virglrenderer!822>
-
Signed-off-by:
-
Signed-off-by:
Gert Wollny <gert.wollny@collabora.com> Part-of: <virgl/virglrenderer!792>
-
The reason this condition exists is to avoid an extra call to glTexParameteri when the depth texture mode changes. But this is hardly ever true, and even when it is, we already call glTexParameteri multiple times anyway, so an extra call here doesn't really matter and just makes the code slightly more complex by adding more state to keep track of. Signed-off-by:
-
Signed-off-by:
-
Corentin Noël authored
This is not used anymore. Signed-off-by:
-
- 01 Jun, 2022 1 commit
-
-
Signed-off-by:
-
- 27 May, 2022 8 commits
-
-
Gert Wollny authored
Signed-off-by:
-
Gert Wollny authored
The necessary info is already in the shader key, so use it. Signed-off-by:
-
Gert Wollny authored
It doesn't make sense to add this info to the shader if it is not the last vertex stage. Signed-off-by:
-
Gert Wollny authored
With the new handling of overlapping IO ranges we no longer emit enhanced layouts, therefore the code that matches the layouts can be removed. Signed-off-by:
-
小田喜陽彦 authored
Before this change, it tried to fill interpolants for tessellation evaluation shader inputs, which have no interpolation, resulting in incorrect "mismatch in number of interps" warning. Fill interpolants only for fragment shader since it is the only case where interpolants are valid. Signed-off-by:
-
小田喜陽彦 authored
The information of fragment shader is needed by vrend_sync_shader_io for later shaders. Signed-off-by:
-
小田喜陽彦 authored
Signed-off-by:
-
小田喜陽彦 authored
Signed-off-by:
-
- 23 May, 2022 2 commits
-
-
Signed-off-by:
Yonggang Luo <luoyonggang@gmail.com> Reviewed-by:
-
Gert Wollny authored
This fixes a memory leak. v2: Use ARRAY_SIZE (Corientin) Signed-off-by:
-
- 20 May, 2022 1 commit
-
-
Gert Wollny authored
v2: Fix range evaluation and format Fixes: spec@arb_texture_buffer_object@texture-buffer-size-clamp Signed-off-by:
-
- 19 May, 2022 2 commits
-
-
Fixes a crash with piglits: spec@arb_texture_buffer_object@texture-buffer-size-clamp Signed-off-by:
-
Corentin Noël authored
Avoids to recalculate the need for uniform values over and over and make sure to always declare it when used. Signed-off-by:
-
- 18 May, 2022 10 commits
-
-
same justification as the earlier commit for vrend. Signed-off-by:
-
same justification as the earlier commit for vrend. Signed-off-by:
-
same justification as the previous commit for vrend. Signed-off-by:
-
vrend has no reason to store the fence id as a void* fence_cookie. If the user wants to continue stashing opaque data in a fence, they may cast their pointer to uint64_t before calling into virglrenderer (as vtest currently does). This way, the behavior is identical on both 32- and 64-bit systems. Signed-off-by:
-
uint64_t fence_id is now the standard for the virglrenderer abstraction, forcing each context to decide what to do with it on their own. For now each context casts their void* back to uintptr_t. There shouldn't be any functional change from this, since contexts previously got a void* directly from the VMM. Signed-off-by:
-
makes uint64_t fence_id the standard for the virglrenderer abstraction, forcing each context to decide what to do with it on their own. For now all contexts immediately cast it back to (void*) and treat as a fence_cookie (the old behavior). On 32-bit systems, this already had the undesirable effect of truncating the 64-bit fence id passed by the virtio-gpu device so nothing changes. Part-of: <virgl/virglrenderer!806>
-
Signed-off-by:
-
Signed-off-by:
-
Signed-off-by:
-
Copy-blits between externally-allocated (GBM) surface resources that use glCopyImageSubData trigger a fatal GL error since strict error checking was enabled in commit 8d452da8 . The issue has since been fixed in mesa 20.x, but a short-term relaxation is needed to restore functionality to old guests until a guest driver uprev can be carried out. Signed-off-by:
-
- 13 May, 2022 3 commits
-
-
And fix a few warnings that crop up as a result. Signed-off-by:
Rob Clark <robdclark@chromium.org> Part-of: <virgl/virglrenderer!804>
-
We were setting the COMM param twice, the second one was intended to be the CMDLINE param. Signed-off-by:
-
Since we no longer emit enhanced layouts IO slots can no longer overlap and we don't have to check the usage mask when searching for the IO slot. This fixes picking up stale IO info that may be passed in with the shader key when it is initialized using an unrelated prev shader stage that might still be bound. Fixes: 028542a2 shader: factor out input search and simplify code Signed-off-by:
-
- 12 May, 2022 1 commit
-
-
Corentin Noël authored
Notably do not automatically run the non-merge Marge pipelines. Removed many unused rules. Part-of: <virgl/virglrenderer!803>
-
