zink: add set_global_binding

This requires VK_KHR_buffer_device_address which is the only sane way of
implementing CL global memory.

Signed-off-by: Karol Herbst <kherbst@redhat.com>
Reviewed-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24839>

diff --git a/src/gallium/drivers/zink/zink_context.c b/src/gallium/drivers/zink/zink_context.c
index 99a5c3bf7a3e17b40a5766f030a005908a1406da..08d70ce26f480241d0280293d4d8eb67ba4c9002 100644 (file)
--- a/src/gallium/drivers/zink/zink_context.c
+++ b/src/gallium/drivers/zink/zink_context.c
@@ -2471,6 +2471,44 @@ zink_make_image_handle_resident(struct pipe_context *pctx, uint64_t handle, unsi
    ctx->di.bindless_dirty[1] = true;
 }
 
+static void
+zink_set_global_binding(struct pipe_context *pctx,
+                        unsigned first, unsigned count,
+                        struct pipe_resource **resources,
+                        uint32_t **handles)
+{
+   struct zink_context *ctx = zink_context(pctx);
+
+   size_t size = ctx->di.global_bindings.capacity;
+   if (!util_dynarray_resize(&ctx->di.global_bindings, struct pipe_resource*, first + count + 8))
+      unreachable("zink: out of memory somehow");
+   if (size != ctx->di.global_bindings.capacity) {
+      uint8_t *data = ctx->di.global_bindings.data;
+      memset(data + size, 0, ctx->di.global_bindings.capacity - size);
+   }
+
+   struct pipe_resource **globals = ctx->di.global_bindings.data;
+   for (unsigned i = 0; i < count; i++) {
+      if (resources && resources[i]) {
+         struct zink_resource *res = zink_resource(resources[i]);
+
+         util_range_add(&res->base.b, &res->valid_buffer_range, 0, res->base.b.width0);
+         pipe_resource_reference(&globals[first + i], resources[i]);
+
+         uint64_t addr = 0;
+         memcpy(&addr, handles[i], sizeof(addr));
+         addr += zink_resource_get_address(zink_screen(pctx->screen), res);
+         memcpy(handles[i], &addr, sizeof(addr));
+         zink_resource_usage_set(res, ctx->batch.state, true);
+         res->obj->unordered_read = res->obj->unordered_write = false;
+         zink_screen(ctx->base.screen)->buffer_barrier(ctx, res, VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT);
+      } else if (globals[i]) {
+         zink_batch_reference_resource(&ctx->batch, zink_resource(globals[first + i]));
+         pipe_resource_reference(&globals[first + i], NULL);
+      }
+   }
+}
+
 static void
 zink_set_stencil_ref(struct pipe_context *pctx,
                      const struct pipe_stencil_ref ref)
@@ -3203,6 +3241,16 @@ zink_update_descriptor_refs(struct zink_context *ctx, bool compute)
          }
       }
    }
+
+   unsigned global_count = util_dynarray_num_elements(&ctx->di.global_bindings, struct zink_resource*);
+   struct zink_resource **globals = ctx->di.global_bindings.data;
+   for (unsigned i = 0; i < global_count; i++) {
+      struct zink_resource *res = globals[i];
+      if (!res)
+         continue;
+      zink_batch_resource_usage_set(batch, res, true, true);
+      res->obj->unordered_read = res->obj->unordered_write = false;
+   }
 }
 
 static void
@@ -5208,6 +5256,8 @@ zink_context_create(struct pipe_screen *pscreen, void *priv, unsigned flags)
 
    ctx->base.set_stream_output_targets = zink_set_stream_output_targets;
    ctx->base.flush_resource = zink_flush_resource;
+   if (screen->info.have_KHR_buffer_device_address)
+      ctx->base.set_global_binding = zink_set_global_binding;
 
    ctx->base.emit_string_marker = zink_emit_string_marker;
 

diff --git a/src/gallium/drivers/zink/zink_types.h b/src/gallium/drivers/zink/zink_types.h
index 7a2c79ccf0ba27346cb9069661454d11309e7a56..455be335c1c39ca8bcd12db8471356ac9674b362 100644 (file)
--- a/src/gallium/drivers/zink/zink_types.h
+++ b/src/gallium/drivers/zink/zink_types.h
@@ -1937,6 +1937,7 @@ struct zink_context {
       uint8_t num_ubos[MESA_SHADER_STAGES];
 
       uint8_t num_ssbos[MESA_SHADER_STAGES];
+      struct util_dynarray global_bindings;
 
       VkDescriptorImageInfo textures[MESA_SHADER_STAGES][PIPE_MAX_SAMPLERS];
       uint32_t emulate_nonseamless[MESA_SHADER_STAGES];