[Mesa-dev] [PATCH] gallium: remove set_shader_resources, add set_shader_buffers for untyped buffers
Roland Scheidegger
sroland at vmware.com
Thu Jan 8 09:45:12 PST 2015
Yes, I'm just trying to figure out if there's a design which easily
works both for d3d and opengl. Apparently, being per stage or global is
a difference which we just need to live with, so I don't have any
objections of making this more suited to GL. Using different types for
images and buffers is also something which helps GL I guess.
(The initial count issue is probably something which can be dealt with
later.)
Roland
Am 08.01.2015 um 00:26 schrieb Marek Olšák:
> Hardware doesn't work like that actually, but it's flexible enough
> that it can do anything. However, some solutions work better than
> others. We also have to take OpenGL into account, which has per-shader
> slots.
>
> RadeonSI shaders don't have any fixed slots for resources and sampler
> states. Instead, they get pointers to arrays of slots (resource
> descriptions), which are in memory. The hardware has to fetch the
> description from memory first (e.g. constant buffer #1) and use that
> description as a parameter of a load instruction. Therefore, we can
> support as many slots as we want. We can share them between shaders,
> etc. (you want to fetch vertex buffers in a pixel shader? Sure I can
> do that) However, we are constrained by what OpenGL requires, so we
> have to design things around that too, especially when it gets ahead
> of Direct3D in terms of features.
>
> Marek
>
> On Wed, Jan 7, 2015 at 6:44 PM, Roland Scheidegger <sroland at vmware.com> wrote:
>> Hmm I'm not quite sure what to think of it. Apparently
>> set_shader_resources was a closer match to what d3d does (seems to use
>> UAVs for everything as far as I can tell, plus they are global and not
>> per stage - I guess another reason why they were set together with RTs).
>> I guess set_shader_buffers would then cover what you can access in d3d
>> as StructuredBuffer and RWStructuredBuffer whereas set_shader_images
>> would cover what can be accessed declared as RWBuffer/RWTextureXD? In
>> that case I guess this would be manageable for translation, though a
>> driver would need to figure out what to set with
>> set_shader_buffers/set_shader_images on their own based on the actual
>> shaders. But if hardware works like that I don't really oppose that.
>> As for the slot numbers though d3d11.1 seems to support 64 UAVs -
>> globally but no further restrictions as far as I can tell (so all could
>> be atomic counters, for instance).
>> d3d also has some initial count feature to set the current offset (or
>> set to -1 to keep current value). I guess this works similarly to how
>> setting offset for streamout buffers work (a major pita to deal with).
>>
>> And I think too that the compute interface should match - I guess though
>> since you now can set this per shader stage you don't need a separate
>> interface at all (for the record I found how you set that with d3d in a
>> compute shader, instead of OMSetRenderTargetsAndUnorderedAccessViews()
>> you use CSSetUnorderedAccessViews()).
>>
>> Roland
>>
>>
>> Am 07.01.2015 um 11:56 schrieb Marek Olšák:
>>> From: Marek Olšák <marek.olsak at amd.com>
>>>
>>> set_shader_resources is unused.
>>>
>>> set_shader_buffers should support shader atomic counter buffers and shader
>>> storage buffers from OpenGL.
>>>
>>> The plan is to use slots 0..15 for atomic counters and slots 16..31
>>> for storage buffers. Atomic counters are planned to be supported first.
>>>
>>> This doesn't add any interface for images. The documentation is added
>>> for future reference.
>>> ---
>>>
>>> This is the interface only. I don't plan to do anything else for now.
>>> Comments welcome.
>>>
>>> src/gallium/docs/source/context.rst | 16 ++++++++--------
>>> src/gallium/docs/source/screen.rst | 4 ++--
>>> src/gallium/drivers/galahad/glhd_context.c | 2 +-
>>> src/gallium/drivers/ilo/ilo_state.c | 2 +-
>>> src/gallium/drivers/nouveau/nouveau_buffer.c | 2 +-
>>> src/gallium/drivers/nouveau/nouveau_screen.c | 2 +-
>>> src/gallium/drivers/nouveau/nv50/nv50_formats.c | 2 +-
>>> src/gallium/drivers/nouveau/nvc0/nvc0_state.c | 2 +-
>>> src/gallium/include/pipe/p_context.h | 20 +++++++++++---------
>>> src/gallium/include/pipe/p_defines.h | 2 +-
>>> src/gallium/include/pipe/p_state.h | 10 ++++++++++
>>> 11 files changed, 38 insertions(+), 26 deletions(-)
>>>
>>> diff --git a/src/gallium/docs/source/context.rst b/src/gallium/docs/source/context.rst
>>> index 5861f46..73fd35f 100644
>>> --- a/src/gallium/docs/source/context.rst
>>> +++ b/src/gallium/docs/source/context.rst
>>> @@ -126,14 +126,14 @@ from a shader without an associated sampler. This means that they
>>> have no support for floating point coordinates, address wrap modes or
>>> filtering.
>>>
>>> -Shader resources are specified for all the shader stages at once using
>>> -the ``set_shader_resources`` method. When binding texture resources,
>>> -the ``level``, ``first_layer`` and ``last_layer`` pipe_surface fields
>>> -specify the mipmap level and the range of layers the texture will be
>>> -constrained to. In the case of buffers, ``first_element`` and
>>> -``last_element`` specify the range within the buffer that will be used
>>> -by the shader resource. Writes to a shader resource are only allowed
>>> -when the ``writable`` flag is set.
>>> +There are 2 types of shader resources: buffers and images.
>>> +
>>> +Buffers are specified using the ``set_shader_buffers`` method.
>>> +
>>> +Images are specified using the ``set_shader_images`` method. When binding
>>> +images, the ``level``, ``first_layer`` and ``last_layer`` pipe_image_view
>>> +fields specify the mipmap level and the range of layers the image will be
>>> +constrained to.
>>>
>>> Surfaces
>>> ^^^^^^^^
>>> diff --git a/src/gallium/docs/source/screen.rst b/src/gallium/docs/source/screen.rst
>>> index 55d114c..c81ad66 100644
>>> --- a/src/gallium/docs/source/screen.rst
>>> +++ b/src/gallium/docs/source/screen.rst
>>> @@ -403,8 +403,8 @@ resources might be created and handled quite differently.
>>> process.
>>> * ``PIPE_BIND_GLOBAL``: A buffer that can be mapped into the global
>>> address space of a compute program.
>>> -* ``PIPE_BIND_SHADER_RESOURCE``: A buffer or texture that can be
>>> - bound to the graphics pipeline as a shader resource.
>>> +* ``PIPE_BIND_SHADER_BUFFER``: A buffer that can be bound to a shader where
>>> + it should support reads, writes, and atomics.
>>> * ``PIPE_BIND_COMPUTE_RESOURCE``: A buffer or texture that can be
>>> bound to the compute program as a shader resource.
>>> * ``PIPE_BIND_COMMAND_ARGS_BUFFER``: A buffer that may be sourced by the
>>> diff --git a/src/gallium/drivers/galahad/glhd_context.c b/src/gallium/drivers/galahad/glhd_context.c
>>> index 37ea170..383d76c 100644
>>> --- a/src/gallium/drivers/galahad/glhd_context.c
>>> +++ b/src/gallium/drivers/galahad/glhd_context.c
>>> @@ -1017,7 +1017,7 @@ galahad_context_create(struct pipe_screen *_screen, struct pipe_context *pipe)
>>> GLHD_PIPE_INIT(set_scissor_states);
>>> GLHD_PIPE_INIT(set_viewport_states);
>>> GLHD_PIPE_INIT(set_sampler_views);
>>> - //GLHD_PIPE_INIT(set_shader_resources);
>>> + //GLHD_PIPE_INIT(set_shader_buffers);
>>> GLHD_PIPE_INIT(set_vertex_buffers);
>>> GLHD_PIPE_INIT(set_index_buffer);
>>> GLHD_PIPE_INIT(create_stream_output_target);
>>> diff --git a/src/gallium/drivers/ilo/ilo_state.c b/src/gallium/drivers/ilo/ilo_state.c
>>> index b852f9f..09209ec 100644
>>> --- a/src/gallium/drivers/ilo/ilo_state.c
>>> +++ b/src/gallium/drivers/ilo/ilo_state.c
>>> @@ -1267,7 +1267,7 @@ ilo_init_state_functions(struct ilo_context *ilo)
>>> ilo->base.set_scissor_states = ilo_set_scissor_states;
>>> ilo->base.set_viewport_states = ilo_set_viewport_states;
>>> ilo->base.set_sampler_views = ilo_set_sampler_views;
>>> - ilo->base.set_shader_resources = ilo_set_shader_resources;
>>> + //ilo->base.set_shader_resources = ilo_set_shader_resources;
>>> ilo->base.set_vertex_buffers = ilo_set_vertex_buffers;
>>> ilo->base.set_index_buffer = ilo_set_index_buffer;
>>>
>>> diff --git a/src/gallium/drivers/nouveau/nouveau_buffer.c b/src/gallium/drivers/nouveau/nouveau_buffer.c
>>> index 49ff100..722c516 100644
>>> --- a/src/gallium/drivers/nouveau/nouveau_buffer.c
>>> +++ b/src/gallium/drivers/nouveau/nouveau_buffer.c
>>> @@ -44,7 +44,7 @@ nouveau_buffer_allocate(struct nouveau_screen *screen,
>>>
>>> if (buf->base.bind & (PIPE_BIND_CONSTANT_BUFFER |
>>> PIPE_BIND_COMPUTE_RESOURCE |
>>> - PIPE_BIND_SHADER_RESOURCE))
>>> + PIPE_BIND_SHADER_BUFFER))
>>> size = align(size, 0x100);
>>>
>>> if (domain == NOUVEAU_BO_VRAM) {
>>> diff --git a/src/gallium/drivers/nouveau/nouveau_screen.c b/src/gallium/drivers/nouveau/nouveau_screen.c
>>> index 517978d..68ab672 100644
>>> --- a/src/gallium/drivers/nouveau/nouveau_screen.c
>>> +++ b/src/gallium/drivers/nouveau/nouveau_screen.c
>>> @@ -197,7 +197,7 @@ nouveau_screen_init(struct nouveau_screen *screen, struct nouveau_device *dev)
>>> PIPE_BIND_DISPLAY_TARGET | PIPE_BIND_SCANOUT |
>>> PIPE_BIND_CURSOR |
>>> PIPE_BIND_SAMPLER_VIEW |
>>> - PIPE_BIND_SHADER_RESOURCE | PIPE_BIND_COMPUTE_RESOURCE |
>>> + PIPE_BIND_SHADER_BUFFER | PIPE_BIND_COMPUTE_RESOURCE |
>>> PIPE_BIND_GLOBAL;
>>> screen->sysmem_bindings =
>>> PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_STREAM_OUTPUT |
>>> diff --git a/src/gallium/drivers/nouveau/nv50/nv50_formats.c b/src/gallium/drivers/nouveau/nv50/nv50_formats.c
>>> index d394015..4fc8380 100644
>>> --- a/src/gallium/drivers/nouveau/nv50/nv50_formats.c
>>> +++ b/src/gallium/drivers/nouveau/nv50/nv50_formats.c
>>> @@ -44,7 +44,7 @@
>>> */
>>> #define U_V PIPE_BIND_VERTEX_BUFFER
>>> #define U_T PIPE_BIND_SAMPLER_VIEW
>>> -#define U_I PIPE_BIND_SHADER_RESOURCE | PIPE_BIND_COMPUTE_RESOURCE
>>> +#define U_I PIPE_BIND_SHADER_BUFFER | PIPE_BIND_COMPUTE_RESOURCE
>>> #define U_TR PIPE_BIND_RENDER_TARGET | U_T
>>> #define U_IR U_TR | U_I
>>> #define U_TB PIPE_BIND_BLENDABLE | U_TR
>>> diff --git a/src/gallium/drivers/nouveau/nvc0/nvc0_state.c b/src/gallium/drivers/nouveau/nvc0/nvc0_state.c
>>> index 728618f..f8fb955 100644
>>> --- a/src/gallium/drivers/nouveau/nvc0/nvc0_state.c
>>> +++ b/src/gallium/drivers/nouveau/nvc0/nvc0_state.c
>>> @@ -1269,7 +1269,7 @@ nvc0_init_state_functions(struct nvc0_context *nvc0)
>>>
>>> pipe->set_global_binding = nvc0_set_global_bindings;
>>> pipe->set_compute_resources = nvc0_set_compute_resources;
>>> - pipe->set_shader_resources = nvc0_set_shader_resources;
>>> + //pipe->set_shader_resources = nvc0_set_shader_resources;
>>>
>>> nvc0->sample_mask = ~0;
>>> nvc0->min_samples = 1;
>>> diff --git a/src/gallium/include/pipe/p_context.h b/src/gallium/include/pipe/p_context.h
>>> index af5674f..641b93a 100644
>>> --- a/src/gallium/include/pipe/p_context.h
>>> +++ b/src/gallium/include/pipe/p_context.h
>>> @@ -57,6 +57,7 @@ struct pipe_resource;
>>> struct pipe_sampler_state;
>>> struct pipe_sampler_view;
>>> struct pipe_scissor_state;
>>> +struct pipe_shader_buffer;
>>> struct pipe_shader_state;
>>> struct pipe_stencil_ref;
>>> struct pipe_stream_output_target;
>>> @@ -222,20 +223,21 @@ struct pipe_context {
>>> struct pipe_sampler_view **);
>>>
>>> /**
>>> - * Bind an array of shader resources that will be used by the
>>> - * graphics pipeline. Any resources that were previously bound to
>>> - * the specified range will be unbound after this call.
>>> + * Bind an array of shader buffers that will be used by a shader.
>>> + * Any resources that were previously bound to the specified range
>>> + * will be unbound.
>>> *
>>> - * \param start first resource to bind.
>>> - * \param count number of consecutive resources to bind.
>>> - * \param resources array of pointers to the resources to bind, it
>>> + * \param shader shader stage where the buffers will be bound.
>>> + * \param start_slot first buffer slot to bind.
>>> + * \param count number of consecutive buffers to bind.
>>> + * \param buffers array of pointers to the buffers to bind, it
>>> * should contain at least \a count elements
>>> * unless it's NULL, in which case no new
>>> * resources will be bound.
>>> */
>>> - void (*set_shader_resources)(struct pipe_context *,
>>> - unsigned start, unsigned count,
>>> - struct pipe_surface **resources);
>>> + void (*set_shader_buffers)(struct pipe_context *, unsigned shader,
>>> + unsigned start_slot, unsigned count,
>>> + struct pipe_shader_buffer *buffers);
>>>
>>> void (*set_vertex_buffers)( struct pipe_context *,
>>> unsigned start_slot,
>>> diff --git a/src/gallium/include/pipe/p_defines.h b/src/gallium/include/pipe/p_defines.h
>>> index 6c5703a..b964fd6 100644
>>> --- a/src/gallium/include/pipe/p_defines.h
>>> +++ b/src/gallium/include/pipe/p_defines.h
>>> @@ -348,7 +348,7 @@ enum pipe_flush_flags {
>>> #define PIPE_BIND_CURSOR (1 << 16) /* mouse cursor */
>>> #define PIPE_BIND_CUSTOM (1 << 17) /* state-tracker/winsys usages */
>>> #define PIPE_BIND_GLOBAL (1 << 18) /* set_global_binding */
>>> -#define PIPE_BIND_SHADER_RESOURCE (1 << 19) /* set_shader_resources */
>>> +#define PIPE_BIND_SHADER_BUFFER (1 << 19) /* set_shader_buffers */
>>> #define PIPE_BIND_COMPUTE_RESOURCE (1 << 20) /* set_compute_resources */
>>> #define PIPE_BIND_COMMAND_ARGS_BUFFER (1 << 21) /* pipe_draw_info.indirect */
>>>
>>> diff --git a/src/gallium/include/pipe/p_state.h b/src/gallium/include/pipe/p_state.h
>>> index 43bc48b..450a270 100644
>>> --- a/src/gallium/include/pipe/p_state.h
>>> +++ b/src/gallium/include/pipe/p_state.h
>>> @@ -464,6 +464,16 @@ struct pipe_constant_buffer {
>>>
>>>
>>> /**
>>> + * An untyped shader buffer supporting loads, stores, and atomics.
>>> + */
>>> +struct pipe_shader_buffer {
>>> + struct pipe_resource *buffer; /**< the actual buffer */
>>> + unsigned buffer_offset; /**< offset to start of data in buffer, in bytes */
>>> + unsigned buffer_size; /**< how much data can be read in shader */
>>> +};
>>> +
>>> +
>>> +/**
>>> * A stream output target. The structure specifies the range vertices can
>>> * be written to.
>>> *
>>>
>>
More information about the mesa-dev
mailing list