2 * Copyright (C) 2019 Collabora, Ltd.
3 * Copyright (C) 2019 Red Hat Inc.
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9 * and/or sell copies of the Software, and to permit persons to whom the
10 * Software is furnished to do so, subject to the following conditions:
12 * The above copyright notice and this permission notice (including the next
13 * paragraph) shall be included in all copies or substantial portions of the
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
24 * Authors (Collabora):
25 * Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
29 #include "pan_context.h"
30 #include "pan_cmdstream.h"
32 #include "util/u_memory.h"
33 #include "nir_serialize.h"
35 /* Compute CSOs are tracked like graphics shader CSOs, but are
36 * considerably simpler. We do not implement multiple
37 * variants/keying. So the CSO create function just goes ahead and
38 * compiles the thing. */
41 panfrost_create_compute_state(
42 struct pipe_context
*pctx
,
43 const struct pipe_compute_state
*cso
)
45 struct panfrost_context
*ctx
= pan_context(pctx
);
47 struct panfrost_shader_variants
*so
= CALLOC_STRUCT(panfrost_shader_variants
);
49 so
->is_compute
= true;
51 struct panfrost_shader_state
*v
= calloc(1, sizeof(*v
));
54 so
->variant_count
= 1;
55 so
->active_variant
= 0;
57 if (cso
->ir_type
== PIPE_SHADER_IR_NIR_SERIALIZED
) {
58 struct blob_reader reader
;
59 const struct pipe_binary_program_header
*hdr
= cso
->prog
;
61 blob_reader_init(&reader
, hdr
->blob
, hdr
->num_bytes
);
62 so
->cbase
.prog
= nir_deserialize(NULL
, &midgard_nir_options
, &reader
);
63 so
->cbase
.ir_type
= PIPE_SHADER_IR_NIR
;
66 panfrost_shader_compile(ctx
, so
->cbase
.ir_type
, so
->cbase
.prog
,
67 MESA_SHADER_COMPUTE
, v
, NULL
);
73 panfrost_bind_compute_state(struct pipe_context
*pipe
, void *cso
)
75 struct panfrost_context
*ctx
= pan_context(pipe
);
77 struct panfrost_shader_variants
*variants
=
78 (struct panfrost_shader_variants
*) cso
;
80 ctx
->shader
[PIPE_SHADER_COMPUTE
] = variants
;
84 panfrost_delete_compute_state(struct pipe_context
*pipe
, void *cso
)
89 /* Launch grid is the compute equivalent of draw_vbo, so in this routine, we
90 * construct the COMPUTE job and some of its payload.
94 panfrost_launch_grid(struct pipe_context
*pipe
,
95 const struct pipe_grid_info
*info
)
97 struct panfrost_context
*ctx
= pan_context(pipe
);
99 /* TODO: Do we want a special compute-only batch? */
100 struct panfrost_batch
*batch
= panfrost_get_batch_for_fbo(ctx
);
102 ctx
->compute_grid
= info
;
105 struct midgard_payload_vertex_tiler payload
;
107 /* We implement OpenCL inputs as uniforms (or a UBO -- same thing), so
108 * reuse the graphics path for this by lowering to Gallium */
110 struct pipe_constant_buffer ubuf
= {
113 .buffer_size
= ctx
->shader
[PIPE_SHADER_COMPUTE
]->cbase
.req_input_mem
,
114 .user_buffer
= info
->input
118 pipe
->set_constant_buffer(pipe
, PIPE_SHADER_COMPUTE
, 0, &ubuf
);
120 panfrost_vt_init(ctx
, PIPE_SHADER_COMPUTE
, &payload
.prefix
, &payload
.postfix
);
122 panfrost_emit_shader_meta(batch
, PIPE_SHADER_COMPUTE
, &payload
.postfix
);
123 panfrost_emit_const_buf(batch
, PIPE_SHADER_COMPUTE
, &payload
.postfix
);
124 panfrost_emit_shared_memory(batch
, info
, &payload
);
126 /* Invoke according to the grid info */
128 panfrost_pack_work_groups_compute(&payload
.prefix
,
129 info
->grid
[0], info
->grid
[1],
131 info
->block
[0], info
->block
[1],
135 panfrost_new_job(batch
, JOB_TYPE_COMPUTE
, true, 0, &payload
,
136 sizeof(payload
), false);
137 panfrost_flush_all_batches(ctx
, true);
141 panfrost_set_compute_resources(struct pipe_context
*pctx
,
142 unsigned start
, unsigned count
,
143 struct pipe_surface
**resources
)
149 panfrost_set_global_binding(struct pipe_context
*pctx
,
150 unsigned first
, unsigned count
,
151 struct pipe_resource
**resources
,
158 panfrost_memory_barrier(struct pipe_context
*pctx
, unsigned flags
)
164 panfrost_compute_context_init(struct pipe_context
*pctx
)
166 pctx
->create_compute_state
= panfrost_create_compute_state
;
167 pctx
->bind_compute_state
= panfrost_bind_compute_state
;
168 pctx
->delete_compute_state
= panfrost_delete_compute_state
;
170 pctx
->launch_grid
= panfrost_launch_grid
;
172 pctx
->set_compute_resources
= panfrost_set_compute_resources
;
173 pctx
->set_global_binding
= panfrost_set_global_binding
;
175 pctx
->memory_barrier
= panfrost_memory_barrier
;