src/gallium/drivers/panfrost/pan_compute.c

   1 /*
   2  * Copyright (C) 2019 Collabora, Ltd.
   3  * Copyright (C) 2019 Red Hat Inc.
   4  *
   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:
  11  *
  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
  14  * Software.
  15  *
  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
  22  * SOFTWARE.
  23  *
  24  * Authors (Collabora):
  25  *   Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
  26  *
  27  */
  28
  29 #include "pan_context.h"
  30 #include "pan_cmdstream.h"
  31 #include "pan_bo.h"
  32 #include "util/u_memory.h"
  33 #include "nir_serialize.h"
  34
  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. */
  39
  40 static void *
  41 panfrost_create_compute_state(
  42         struct pipe_context *pctx,
  43         const struct pipe_compute_state *cso)
  44 {
  45         struct panfrost_context *ctx = pan_context(pctx);
  46
  47         struct panfrost_shader_variants *so = CALLOC_STRUCT(panfrost_shader_variants);
  48         so->cbase = *cso;
  49         so->is_compute = true;
  50
  51         struct panfrost_shader_state *v = calloc(1, sizeof(*v));
  52         so->variants = v;
  53
  54         so->variant_count = 1;
  55         so->active_variant = 0;
  56
  57         /* calloc, instead of malloc - to zero unused fields */
  58         v->tripipe = CALLOC_STRUCT(mali_shader_meta);
  59
  60         if (cso->ir_type == PIPE_SHADER_IR_NIR_SERIALIZED) {
  61                 struct blob_reader reader;
  62                 const struct pipe_binary_program_header *hdr = cso->prog;
  63
  64                 blob_reader_init(&reader, hdr->blob, hdr->num_bytes);
  65                 so->cbase.prog = nir_deserialize(NULL, &midgard_nir_options, &reader);
  66                 so->cbase.ir_type = PIPE_SHADER_IR_NIR;
  67         }
  68
  69         panfrost_shader_compile(ctx, v->tripipe,
  70                         so->cbase.ir_type, so->cbase.prog,
  71                         MESA_SHADER_COMPUTE, v, NULL);
  72
  73         return so;
  74 }
  75
  76 static void
  77 panfrost_bind_compute_state(struct pipe_context *pipe, void *cso)
  78 {
  79         struct panfrost_context *ctx = pan_context(pipe);
  80
  81         struct panfrost_shader_variants *variants =
  82                 (struct panfrost_shader_variants *) cso;
  83
  84         ctx->shader[PIPE_SHADER_COMPUTE] = variants;
  85 }
  86
  87 static void
  88 panfrost_delete_compute_state(struct pipe_context *pipe, void *cso)
  89 {
  90         free(cso);
  91 }
  92
  93 /* Launch grid is the compute equivalent of draw_vbo, so in this routine, we
  94  * construct the COMPUTE job and some of its payload.
  95  */
  96
  97 static void
  98 panfrost_launch_grid(struct pipe_context *pipe,
  99                 const struct pipe_grid_info *info)
 100 {
 101         struct panfrost_context *ctx = pan_context(pipe);
 102
 103         /* TODO: Do we want a special compute-only batch? */
 104         struct panfrost_batch *batch = panfrost_get_batch_for_fbo(ctx);
 105
 106         ctx->compute_grid = info;
 107
 108         /* TODO: Stub */
 109         struct midgard_payload_vertex_tiler *payload = &ctx->payloads[PIPE_SHADER_COMPUTE];
 110
 111         /* We implement OpenCL inputs as uniforms (or a UBO -- same thing), so
 112          * reuse the graphics path for this by lowering to Gallium */
 113
 114         struct pipe_constant_buffer ubuf = {
 115                 .buffer = NULL,
 116                 .buffer_offset = 0,
 117                 .buffer_size = ctx->shader[PIPE_SHADER_COMPUTE]->cbase.req_input_mem,
 118                 .user_buffer = info->input
 119         };
 120
 121         if (info->input)
 122                 pipe->set_constant_buffer(pipe, PIPE_SHADER_COMPUTE, 0, &ubuf);
 123
 124         panfrost_patch_shader_state(ctx, PIPE_SHADER_COMPUTE);
 125         panfrost_emit_shader_meta(batch, PIPE_SHADER_COMPUTE, payload);
 126         panfrost_emit_const_buf(batch, PIPE_SHADER_COMPUTE, payload);
 127         panfrost_emit_shared_memory(batch, info, payload);
 128
 129         /* Invoke according to the grid info */
 130
 131         panfrost_pack_work_groups_compute(&payload->prefix,
 132                         info->grid[0], info->grid[1], info->grid[2],
 133                         info->block[0], info->block[1], info->block[2], false);
 134
 135         panfrost_new_job(batch, JOB_TYPE_COMPUTE, true, 0, payload, sizeof(*payload), false);
 136         panfrost_flush_all_batches(ctx, true);
 137 }
 138
 139 static void
 140 panfrost_set_compute_resources(struct pipe_context *pctx,
 141                          unsigned start, unsigned count,
 142                          struct pipe_surface **resources)
 143 {
 144         /* TODO */
 145 }
 146
 147 static void
 148 panfrost_set_global_binding(struct pipe_context *pctx,
 149                       unsigned first, unsigned count,
 150                       struct pipe_resource **resources,
 151                       uint32_t **handles)
 152 {
 153         /* TODO */
 154 }
 155
 156 static void
 157 panfrost_memory_barrier(struct pipe_context *pctx, unsigned flags)
 158 {
 159         /* TODO */
 160 }
 161
 162 void
 163 panfrost_compute_context_init(struct pipe_context *pctx)
 164 {
 165         pctx->create_compute_state = panfrost_create_compute_state;
 166         pctx->bind_compute_state = panfrost_bind_compute_state;
 167         pctx->delete_compute_state = panfrost_delete_compute_state;
 168
 169         pctx->launch_grid = panfrost_launch_grid;
 170
 171         pctx->set_compute_resources = panfrost_set_compute_resources;
 172         pctx->set_global_binding = panfrost_set_global_binding;
 173
 174         pctx->memory_barrier = panfrost_memory_barrier;
 175 }