panvk: Implement vkCmdDispatch()

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15248>

src/panfrost/ci/deqp-panfrost-g52-vk.toml

@@ -5,4 +5,12 @@ renderer_check = "Mali-G52"
 include = [
     "dEQP-VK.pipeline.blend.*",
     "dEQP-VK.api.copy_and_blit.core.*",
+    "dEQP-VK.compute.builtin_var.*",
+    "dEQP-VK.image.load_store.with_format.1d.*",
+    "dEQP-VK.image.load_store.with_format.1d_array.*",
+    "dEQP-VK.image.load_store.with_format.2d.*",
+    "dEQP-VK.image.load_store.with_format.2d_array.*",
+    "dEQP-VK.image.load_store.with_format.3d.*",
+    "dEQP-VK.image.load_store.with_format.cube.*",
+    "dEQP-VK.ssbo.layout.single_basic_type.*",
 ]

src/panfrost/vulkan/panvk_cmd_buffer.c

@@ -573,15 +573,6 @@ panvk_CmdDispatchBase(VkCommandBuffer commandBuffer,
    panvk_stub();
 }
 
-void
-panvk_CmdDispatch(VkCommandBuffer commandBuffer,
-                  uint32_t x,
-                  uint32_t y,
-                  uint32_t z)
-{
-   panvk_stub();
-}
-
 void
 panvk_CmdDispatchIndirect(VkCommandBuffer commandBuffer,
                           VkBuffer _buffer,

src/panfrost/vulkan/panvk_private.h

@@ -554,6 +554,17 @@ struct panvk_draw_info {
    } jobs;
 };
 
+struct panvk_dispatch_info {
+   struct pan_compute_dim wg_count;
+   mali_ptr attributes;
+   mali_ptr attribute_bufs;
+   mali_ptr tsd;
+   mali_ptr ubos;
+   mali_ptr push_uniforms;
+   mali_ptr textures;
+   mali_ptr samplers;
+};
+
 struct panvk_attrib_info {
    unsigned buf;
    unsigned offset;
@@ -630,6 +641,10 @@ struct panvk_cmd_state {
       bool crc_valid[MAX_RTS];
    } fb;
 
+   struct {
+      struct pan_compute_dim wg_count;
+   } compute;
+
    const struct panvk_render_pass *pass;
    const struct panvk_subpass *subpass;
    const struct panvk_framebuffer *framebuffer;

src/panfrost/vulkan/panvk_vX_cmd_buffer.c

@@ -349,6 +349,16 @@ panvk_cmd_upload_sysval(struct panvk_cmd_buffer *cmdbuf,
        */
       panvk_sysval_upload_ssbo_info(cmdbuf, PAN_SYSVAL_ID(id), bind_point_state, data);
       break;
+   case PAN_SYSVAL_NUM_WORK_GROUPS:
+      data->u32[0] = cmdbuf->state.compute.wg_count.x;
+      data->u32[1] = cmdbuf->state.compute.wg_count.y;
+      data->u32[2] = cmdbuf->state.compute.wg_count.z;
+      break;
+   case PAN_SYSVAL_LOCAL_GROUP_SIZE:
+      data->u32[0] = bind_point_state->pipeline->cs.local_size.x;
+      data->u32[1] = bind_point_state->pipeline->cs.local_size.y;
+      data->u32[2] = bind_point_state->pipeline->cs.local_size.z;
+      break;
    default:
       unreachable("Invalid static sysval");
    }
@@ -1368,3 +1378,59 @@ panvk_per_arch(TrimCommandPool)(VkDevice device,
                             &pool->free_cmd_buffers, pool_link)
       panvk_destroy_cmdbuf(cmdbuf);
 }
+
+void
+panvk_per_arch(CmdDispatch)(VkCommandBuffer commandBuffer,
+                            uint32_t x,
+                            uint32_t y,
+                            uint32_t z)
+{
+   VK_FROM_HANDLE(panvk_cmd_buffer, cmdbuf, commandBuffer);
+   const struct panfrost_device *pdev =
+      &cmdbuf->device->physical_device->pdev;
+   struct panvk_dispatch_info dispatch = {
+      .wg_count = { x, y, z },
+   };
+
+   panvk_per_arch(cmd_close_batch)(cmdbuf);
+   struct panvk_batch *batch = panvk_cmd_open_batch(cmdbuf);
+
+   struct panvk_cmd_bind_point_state *bind_point_state =
+      panvk_cmd_get_bind_point_state(cmdbuf, COMPUTE);
+   struct panvk_descriptor_state *desc_state = &bind_point_state->desc_state;
+   const struct panvk_pipeline *pipeline = bind_point_state->pipeline;
+   struct panfrost_ptr job =
+      pan_pool_alloc_desc(&cmdbuf->desc_pool.base, COMPUTE_JOB);
+
+   cmdbuf->state.compute.wg_count = dispatch.wg_count;
+   panvk_per_arch(cmd_alloc_tls_desc)(cmdbuf, false);
+   dispatch.tsd = batch->tls.gpu;
+
+   panvk_prepare_non_vs_attribs(cmdbuf, bind_point_state);
+   dispatch.attributes = desc_state->non_vs_attribs;
+   dispatch.attribute_bufs = desc_state->non_vs_attrib_bufs;
+
+   panvk_cmd_prepare_ubos(cmdbuf, bind_point_state);
+   dispatch.ubos = desc_state->ubos;
+
+   panvk_cmd_prepare_textures(cmdbuf, bind_point_state);
+   dispatch.textures = desc_state->textures;
+
+   panvk_cmd_prepare_samplers(cmdbuf, bind_point_state);
+   dispatch.samplers = desc_state->samplers;
+
+   panvk_per_arch(emit_compute_job)(pipeline, &dispatch, job.cpu);
+   panfrost_add_job(&cmdbuf->desc_pool.base, &batch->scoreboard,
+                    MALI_JOB_TYPE_COMPUTE, false, false, 0, 0,
+                    &job, false);
+
+   batch->tlsinfo.tls.size = pipeline->tls_size;
+   batch->tlsinfo.wls.size = pipeline->wls_size;
+   if (batch->tlsinfo.wls.size) {
+      batch->wls_total_size =
+         pan_wls_mem_size(pdev, &dispatch.wg_count, batch->tlsinfo.wls.size);
+   }
+
+   panvk_per_arch(cmd_close_batch)(cmdbuf);
+   desc_state->dirty = 0;
+}

src/panfrost/vulkan/panvk_vX_cs.c

@@ -416,6 +416,39 @@ panvk_per_arch(emit_vertex_job)(const struct panvk_pipeline *pipeline,
    }
 }
 
+void
+panvk_per_arch(emit_compute_job)(const struct panvk_pipeline *pipeline,
+                                 const struct panvk_dispatch_info *dispatch,
+                                 void *job)
+{
+   panfrost_pack_work_groups_compute(pan_section_ptr(job, COMPUTE_JOB, INVOCATION),
+                                     dispatch->wg_count.x,
+                                     dispatch->wg_count.y,
+                                     dispatch->wg_count.z,
+                                     pipeline->cs.local_size.x,
+                                     pipeline->cs.local_size.y,
+                                     pipeline->cs.local_size.z,
+                                     false, false);
+
+   pan_section_pack(job, COMPUTE_JOB, PARAMETERS, cfg) {
+      cfg.job_task_split =
+         util_logbase2_ceil(pipeline->cs.local_size.x + 1) +
+         util_logbase2_ceil(pipeline->cs.local_size.y + 1) +
+         util_logbase2_ceil(pipeline->cs.local_size.z + 1);
+   }
+
+   pan_section_pack(job, COMPUTE_JOB, DRAW, cfg) {
+      cfg.state = pipeline->rsds[MESA_SHADER_COMPUTE];
+      cfg.attributes = dispatch->attributes;
+      cfg.attribute_buffers = dispatch->attribute_bufs;
+      cfg.thread_storage = dispatch->tsd;
+      cfg.uniform_buffers = dispatch->ubos;
+      cfg.push_uniforms = dispatch->push_uniforms;
+      cfg.textures = dispatch->textures;
+      cfg.samplers = dispatch->samplers;
+   }
+}
+
 static void
 panvk_emit_tiler_primitive(const struct panvk_pipeline *pipeline,
                            const struct panvk_draw_info *draw,

src/panfrost/vulkan/panvk_vX_cs.h

@@ -79,6 +79,11 @@ panvk_per_arch(emit_vertex_job)(const struct panvk_pipeline *pipeline,
                                 const struct panvk_draw_info *draw,
                                 void *job);
 
+void
+panvk_per_arch(emit_compute_job)(const struct panvk_pipeline *pipeline,
+                                 const struct panvk_dispatch_info *dispatch,
+                                 void *job);
+
 void
 panvk_per_arch(emit_tiler_job)(const struct panvk_pipeline *pipeline,
                                const struct panvk_draw_info *draw,