[PATCH v2 09/17] drm/vmwgfx: Add support for UA view commands

Roland Scheidegger (VMware) rscheidegger.oss at gmail.com
Mon Mar 23 23:12:30 UTC 2020


From: Deepak Rawat <drawat.floss at gmail.com>

Virtual device now support new commands to manage unordered access
views. Allow them as part of user-space command buffer. This involves
adding UA view cotable, binding tracker info, new view type and command
verifier functions.

v2: fix comment typo
v3: style fixes (don't use deprecated PTR_RET)

Signed-off-by: Deepak Rawat <drawat.floss at gmail.com>
Signed-off-by: Neha Bhende <bhenden at vmware.com>
Reviewed-by: Thomas Hellström (VMware) <thomas_os at shipmail.org>
Reviewed-by: Roland Scheidegger <sroland at vmware.com>
Signed-off-by: Roland Scheidegger <sroland at vmware.com>
---
 drivers/gpu/drm/vmwgfx/vmwgfx_binding.c | 129 ++++++++++++++++++++-
 drivers/gpu/drm/vmwgfx/vmwgfx_binding.h |  16 +++
 drivers/gpu/drm/vmwgfx/vmwgfx_context.c |  26 +++--
 drivers/gpu/drm/vmwgfx/vmwgfx_cotable.c |   4 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c | 144 +++++++++++++++++++++++-
 drivers/gpu/drm/vmwgfx/vmwgfx_so.c      |  12 +-
 drivers/gpu/drm/vmwgfx/vmwgfx_so.h      |   6 +
 7 files changed, 319 insertions(+), 18 deletions(-)

diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_binding.c b/drivers/gpu/drm/vmwgfx/vmwgfx_binding.c
index 5a7e8db3e826..f0a576698266 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_binding.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_binding.c
@@ -59,7 +59,9 @@
 #define VMW_BINDING_PS_BIT     1
 #define VMW_BINDING_SO_BIT     2
 #define VMW_BINDING_VB_BIT     3
-#define VMW_BINDING_NUM_BITS   4
+#define VMW_BINDING_UAV_BIT    4
+#define VMW_BINDING_CS_UAV_BIT 5
+#define VMW_BINDING_NUM_BITS   6
 
 #define VMW_BINDING_PS_SR_BIT  0
 
@@ -75,6 +77,7 @@
  * @vertex_buffers: Vertex buffer bindings.
  * @index_buffer: Index buffer binding.
  * @per_shader: Per shader-type bindings.
+ * @ua_views: UAV bindings.
  * @dirty: Bitmap tracking per binding-type changes that have not yet
  * been emitted to the device.
  * @dirty_vb: Bitmap tracking individual vertex buffer binding changes that
@@ -99,6 +102,7 @@ struct vmw_ctx_binding_state {
 	struct vmw_ctx_bindinfo_vb vertex_buffers[SVGA3D_DX_MAX_VERTEXBUFFERS];
 	struct vmw_ctx_bindinfo_ib index_buffer;
 	struct vmw_dx_shader_bindings per_shader[SVGA3D_NUM_SHADERTYPE];
+	struct vmw_ctx_bindinfo_uav ua_views[VMW_MAX_UAV_BIND_TYPE];
 
 	unsigned long dirty;
 	DECLARE_BITMAP(dirty_vb, SVGA3D_DX_MAX_VERTEXBUFFERS);
@@ -121,6 +125,9 @@ static int vmw_binding_scrub_dx_shader(struct vmw_ctx_bindinfo *bi,
 				       bool rebind);
 static int vmw_binding_scrub_ib(struct vmw_ctx_bindinfo *bi, bool rebind);
 static int vmw_binding_scrub_vb(struct vmw_ctx_bindinfo *bi, bool rebind);
+static int vmw_binding_scrub_uav(struct vmw_ctx_bindinfo *bi, bool rebind);
+static int vmw_binding_scrub_cs_uav(struct vmw_ctx_bindinfo *bi, bool rebind);
+
 static void vmw_binding_build_asserts(void) __attribute__ ((unused));
 
 typedef int (*vmw_scrub_func)(struct vmw_ctx_bindinfo *, bool);
@@ -189,6 +196,12 @@ static const size_t vmw_binding_vb_offsets[] = {
 static const size_t vmw_binding_ib_offsets[] = {
 	offsetof(struct vmw_ctx_binding_state, index_buffer),
 };
+static const size_t vmw_binding_uav_offsets[] = {
+	offsetof(struct vmw_ctx_binding_state, ua_views[0].views),
+};
+static const size_t vmw_binding_cs_uav_offsets[] = {
+	offsetof(struct vmw_ctx_binding_state, ua_views[1].views),
+};
 
 static const struct vmw_binding_info vmw_binding_infos[] = {
 	[vmw_ctx_binding_shader] = {
@@ -235,6 +248,14 @@ static const struct vmw_binding_info vmw_binding_infos[] = {
 		.size = sizeof(struct vmw_ctx_bindinfo_ib),
 		.offsets = vmw_binding_ib_offsets,
 		.scrub_func = vmw_binding_scrub_ib},
+	[vmw_ctx_binding_uav] = {
+		.size = sizeof(struct vmw_ctx_bindinfo_view),
+		.offsets = vmw_binding_uav_offsets,
+		.scrub_func = vmw_binding_scrub_uav},
+	[vmw_ctx_binding_cs_uav] = {
+		.size = sizeof(struct vmw_ctx_bindinfo_view),
+		.offsets = vmw_binding_cs_uav_offsets,
+		.scrub_func = vmw_binding_scrub_cs_uav},
 };
 
 /**
@@ -320,6 +341,18 @@ void vmw_binding_add(struct vmw_ctx_binding_state *cbs,
 	INIT_LIST_HEAD(&loc->res_list);
 }
 
+/**
+ * vmw_binding_add_uav_index - Add UAV index for tracking.
+ * @cbs: Pointer to the context binding state tracker.
+ * @slot: UAV type to which bind this index.
+ * @index: The splice index to track.
+ */
+void vmw_binding_add_uav_index(struct vmw_ctx_binding_state *cbs, uint32 slot,
+			       uint32 index)
+{
+	cbs->ua_views[slot].index = index;
+}
+
 /**
  * vmw_binding_transfer: Transfer a context binding tracking entry.
  *
@@ -459,6 +492,10 @@ void vmw_binding_state_commit(struct vmw_ctx_binding_state *to,
 		vmw_binding_transfer(to, from, entry);
 		vmw_binding_drop(entry);
 	}
+
+	/* Also transfer uav splice indices */
+	to->ua_views[0].index = from->ua_views[0].index;
+	to->ua_views[1].index = from->ua_views[1].index;
 }
 
 /**
@@ -1014,6 +1051,66 @@ static int vmw_emit_set_vb(struct vmw_ctx_binding_state *cbs)
 	return 0;
 }
 
+static int vmw_emit_set_uav(struct vmw_ctx_binding_state *cbs)
+{
+	const struct vmw_ctx_bindinfo *loc = &cbs->ua_views[0].views[0].bi;
+	struct {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdDXSetUAViews body;
+	} *cmd;
+	size_t cmd_size, view_id_size;
+	const struct vmw_resource *ctx = vmw_cbs_context(cbs);
+
+	vmw_collect_view_ids(cbs, loc, SVGA3D_MAX_UAVIEWS);
+	view_id_size = cbs->bind_cmd_count*sizeof(uint32);
+	cmd_size = sizeof(*cmd) + view_id_size;
+	cmd = VMW_FIFO_RESERVE_DX(ctx->dev_priv, cmd_size, ctx->id);
+	if (!cmd)
+		return -ENOMEM;
+
+	cmd->header.id = SVGA_3D_CMD_DX_SET_UA_VIEWS;
+	cmd->header.size = sizeof(cmd->body) + view_id_size;
+
+	/* Splice index is specified user-space   */
+	cmd->body.uavSpliceIndex = cbs->ua_views[0].index;
+
+	memcpy(&cmd[1], cbs->bind_cmd_buffer, view_id_size);
+
+	vmw_fifo_commit(ctx->dev_priv, cmd_size);
+
+	return 0;
+}
+
+static int vmw_emit_set_cs_uav(struct vmw_ctx_binding_state *cbs)
+{
+	const struct vmw_ctx_bindinfo *loc = &cbs->ua_views[1].views[0].bi;
+	struct {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdDXSetCSUAViews body;
+	} *cmd;
+	size_t cmd_size, view_id_size;
+	const struct vmw_resource *ctx = vmw_cbs_context(cbs);
+
+	vmw_collect_view_ids(cbs, loc, SVGA3D_MAX_UAVIEWS);
+	view_id_size = cbs->bind_cmd_count*sizeof(uint32);
+	cmd_size = sizeof(*cmd) + view_id_size;
+	cmd = VMW_FIFO_RESERVE_DX(ctx->dev_priv, cmd_size, ctx->id);
+	if (!cmd)
+		return -ENOMEM;
+
+	cmd->header.id = SVGA_3D_CMD_DX_SET_CS_UA_VIEWS;
+	cmd->header.size = sizeof(cmd->body) + view_id_size;
+
+	/* Start index is specified user-space */
+	cmd->body.startIndex = cbs->ua_views[1].index;
+
+	memcpy(&cmd[1], cbs->bind_cmd_buffer, view_id_size);
+
+	vmw_fifo_commit(ctx->dev_priv, cmd_size);
+
+	return 0;
+}
+
 /**
  * vmw_binding_emit_dirty - Issue delayed binding commands
  *
@@ -1045,6 +1142,12 @@ static int vmw_binding_emit_dirty(struct vmw_ctx_binding_state *cbs)
 		case VMW_BINDING_VB_BIT:
 			ret = vmw_emit_set_vb(cbs);
 			break;
+		case VMW_BINDING_UAV_BIT:
+			ret = vmw_emit_set_uav(cbs);
+			break;
+		case VMW_BINDING_CS_UAV_BIT:
+			ret = vmw_emit_set_cs_uav(cbs);
+			break;
 		default:
 			BUG();
 		}
@@ -1171,6 +1274,22 @@ static int vmw_binding_scrub_ib(struct vmw_ctx_bindinfo *bi, bool rebind)
 	return 0;
 }
 
+static int vmw_binding_scrub_uav(struct vmw_ctx_bindinfo *bi, bool rebind)
+{
+	struct vmw_ctx_binding_state *cbs = vmw_context_binding_state(bi->ctx);
+
+	__set_bit(VMW_BINDING_UAV_BIT, &cbs->dirty);
+	return 0;
+}
+
+static int vmw_binding_scrub_cs_uav(struct vmw_ctx_bindinfo *bi, bool rebind)
+{
+	struct vmw_ctx_binding_state *cbs = vmw_context_binding_state(bi->ctx);
+
+	__set_bit(VMW_BINDING_CS_UAV_BIT, &cbs->dirty);
+	return 0;
+}
+
 /**
  * vmw_binding_state_alloc - Allocate a struct vmw_ctx_binding_state with
  * memory accounting.
@@ -1257,8 +1376,8 @@ void vmw_binding_state_reset(struct vmw_ctx_binding_state *cbs)
  * Each time a resource is put on the validation list as the result of a
  * context binding referencing it, we need to determine whether that resource
  * will be dirtied (written to by the GPU) as a result of the corresponding
- * GPU operation. Currently rendertarget-, depth-stencil-, and
- * stream-output-target bindings are capable of dirtying its resource.
+ * GPU operation. Currently rendertarget-, depth-stencil-, stream-output-target
+ * and unordered access view bindings are capable of dirtying its resource.
  *
  * Return: Whether the binding type dirties the resource its binding points to.
  */
@@ -1269,10 +1388,12 @@ u32 vmw_binding_dirtying(enum vmw_ctx_binding_type binding_type)
 		[vmw_ctx_binding_dx_rt] = VMW_RES_DIRTY_SET,
 		[vmw_ctx_binding_ds] = VMW_RES_DIRTY_SET,
 		[vmw_ctx_binding_so] = VMW_RES_DIRTY_SET,
+		[vmw_ctx_binding_uav] = VMW_RES_DIRTY_SET,
+		[vmw_ctx_binding_cs_uav] = VMW_RES_DIRTY_SET,
 	};
 
 	/* Review this function as new bindings are added. */
-	BUILD_BUG_ON(vmw_ctx_binding_max != 11);
+	BUILD_BUG_ON(vmw_ctx_binding_max != 13);
 	return is_binding_dirtying[binding_type];
 }
 
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_binding.h b/drivers/gpu/drm/vmwgfx/vmwgfx_binding.h
index cd9805c045cb..da67cbc9cd73 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_binding.h
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_binding.h
@@ -33,6 +33,8 @@
 
 #define VMW_MAX_VIEW_BINDINGS 128
 
+#define VMW_MAX_UAV_BIND_TYPE 2
+
 struct vmw_private;
 struct vmw_ctx_binding_state;
 
@@ -51,6 +53,8 @@ enum vmw_ctx_binding_type {
 	vmw_ctx_binding_so,
 	vmw_ctx_binding_vb,
 	vmw_ctx_binding_ib,
+	vmw_ctx_binding_uav,
+	vmw_ctx_binding_cs_uav,
 	vmw_ctx_binding_max
 };
 
@@ -189,9 +193,21 @@ struct vmw_dx_shader_bindings {
 	unsigned long dirty;
 };
 
+/**
+ * struct vmw_ctx_bindinfo_uav - UAV context binding state.
+ * @views: UAV view bindings.
+ * @splice_index: The device splice index set by user-space.
+ */
+struct vmw_ctx_bindinfo_uav {
+	struct vmw_ctx_bindinfo_view views[SVGA3D_MAX_UAVIEWS];
+	uint32 index;
+};
+
 extern void vmw_binding_add(struct vmw_ctx_binding_state *cbs,
 			    const struct vmw_ctx_bindinfo *ci,
 			    u32 shader_slot, u32 slot);
+extern void vmw_binding_add_uav_index(struct vmw_ctx_binding_state *cbs,
+				      uint32 slot, uint32 splice_index);
 extern void
 vmw_binding_state_commit(struct vmw_ctx_binding_state *to,
 			 struct vmw_ctx_binding_state *from);
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_context.c b/drivers/gpu/drm/vmwgfx/vmwgfx_context.c
index 0477d9a74fe8..61c246335e66 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_context.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_context.c
@@ -36,7 +36,7 @@ struct vmw_user_context {
 	struct vmw_resource res;
 	struct vmw_ctx_binding_state *cbs;
 	struct vmw_cmdbuf_res_manager *man;
-	struct vmw_resource *cotables[SVGA_COTABLE_DX10_MAX];
+	struct vmw_resource *cotables[SVGA_COTABLE_MAX];
 	spinlock_t cotable_lock;
 	struct vmw_buffer_object *dx_query_mob;
 };
@@ -116,12 +116,15 @@ static const struct vmw_res_func vmw_dx_context_func = {
  * Context management:
  */
 
-static void vmw_context_cotables_unref(struct vmw_user_context *uctx)
+static void vmw_context_cotables_unref(struct vmw_private *dev_priv,
+				       struct vmw_user_context *uctx)
 {
 	struct vmw_resource *res;
 	int i;
+	u32 cotable_max = has_sm5_context(dev_priv) ?
+		SVGA_COTABLE_MAX : SVGA_COTABLE_DX10_MAX;
 
-	for (i = 0; i < SVGA_COTABLE_DX10_MAX; ++i) {
+	for (i = 0; i < cotable_max; ++i) {
 		spin_lock(&uctx->cotable_lock);
 		res = uctx->cotables[i];
 		uctx->cotables[i] = NULL;
@@ -155,7 +158,7 @@ static void vmw_hw_context_destroy(struct vmw_resource *res)
 		    !dev_priv->query_cid_valid)
 			__vmw_execbuf_release_pinned_bo(dev_priv, NULL);
 		mutex_unlock(&dev_priv->cmdbuf_mutex);
-		vmw_context_cotables_unref(uctx);
+		vmw_context_cotables_unref(dev_priv, uctx);
 		return;
 	}
 
@@ -208,7 +211,9 @@ static int vmw_gb_context_init(struct vmw_private *dev_priv,
 	spin_lock_init(&uctx->cotable_lock);
 
 	if (dx) {
-		for (i = 0; i < SVGA_COTABLE_DX10_MAX; ++i) {
+		u32 cotable_max = has_sm5_context(dev_priv) ?
+			SVGA_COTABLE_MAX : SVGA_COTABLE_DX10_MAX;
+		for (i = 0; i < cotable_max; ++i) {
 			uctx->cotables[i] = vmw_cotable_alloc(dev_priv,
 							      &uctx->res, i);
 			if (IS_ERR(uctx->cotables[i])) {
@@ -222,7 +227,7 @@ static int vmw_gb_context_init(struct vmw_private *dev_priv,
 	return 0;
 
 out_cotables:
-	vmw_context_cotables_unref(uctx);
+	vmw_context_cotables_unref(dev_priv, uctx);
 out_err:
 	if (res_free)
 		res_free(res);
@@ -545,10 +550,12 @@ void vmw_dx_context_scrub_cotables(struct vmw_resource *ctx,
 {
 	struct vmw_user_context *uctx =
 		container_of(ctx, struct vmw_user_context, res);
+	u32 cotable_max = has_sm5_context(ctx->dev_priv) ?
+		SVGA_COTABLE_MAX : SVGA_COTABLE_DX10_MAX;
 	int i;
 
 	vmw_binding_state_scrub(uctx->cbs);
-	for (i = 0; i < SVGA_COTABLE_DX10_MAX; ++i) {
+	for (i = 0; i < cotable_max; ++i) {
 		struct vmw_resource *res;
 
 		/* Avoid racing with ongoing cotable destruction. */
@@ -839,7 +846,10 @@ struct vmw_cmdbuf_res_manager *vmw_context_res_man(struct vmw_resource *ctx)
 struct vmw_resource *vmw_context_cotable(struct vmw_resource *ctx,
 					 SVGACOTableType cotable_type)
 {
-	if (cotable_type >= SVGA_COTABLE_DX10_MAX)
+	u32 cotable_max = has_sm5_context(ctx->dev_priv) ?
+		SVGA_COTABLE_MAX : SVGA_COTABLE_DX10_MAX;
+
+	if (cotable_type >= cotable_max)
 		return ERR_PTR(-EINVAL);
 
 	return container_of(ctx, struct vmw_user_context, res)->
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_cotable.c b/drivers/gpu/drm/vmwgfx/vmwgfx_cotable.c
index 3ca5cf375b01..08c5b2ccca98 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_cotable.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_cotable.c
@@ -82,7 +82,8 @@ static const struct vmw_cotable_info co_info[] = {
 	{1, sizeof(SVGACOTableDXSamplerEntry), NULL},
 	{1, sizeof(SVGACOTableDXStreamOutputEntry), NULL},
 	{1, sizeof(SVGACOTableDXQueryEntry), NULL},
-	{1, sizeof(SVGACOTableDXShaderEntry), &vmw_dx_shader_cotable_list_scrub}
+	{1, sizeof(SVGACOTableDXShaderEntry), &vmw_dx_shader_cotable_list_scrub},
+	{1, sizeof(SVGACOTableDXUAViewEntry), &vmw_view_cotable_list_destroy}
 };
 
 /*
@@ -102,6 +103,7 @@ const SVGACOTableType vmw_cotable_scrub_order[] = {
 	SVGA_COTABLE_SAMPLER,
 	SVGA_COTABLE_STREAMOUTPUT,
 	SVGA_COTABLE_DXQUERY,
+	SVGA_COTABLE_UAVIEW,
 };
 
 static int vmw_cotable_bind(struct vmw_resource *res,
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c b/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
index 4abed135c013..ed56d9918c6a 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
@@ -459,11 +459,13 @@ static int vmw_resource_context_res_add(struct vmw_private *dev_priv,
 	int ret = 0;
 	struct vmw_resource *res;
 	u32 i;
+	u32 cotable_max = has_sm5_context(ctx->dev_priv) ?
+		SVGA_COTABLE_MAX : SVGA_COTABLE_DX10_MAX;
 
 	/* Add all cotables to the validation list. */
 	if (has_sm4_context(dev_priv) &&
 	    vmw_res_type(ctx) == vmw_res_dx_context) {
-		for (i = 0; i < SVGA_COTABLE_DX10_MAX; ++i) {
+		for (i = 0; i < cotable_max; ++i) {
 			res = vmw_context_cotable(ctx, i);
 			if (IS_ERR(res))
 				continue;
@@ -2814,6 +2816,128 @@ static int vmw_cmd_intra_surface_copy(struct vmw_private *dev_priv,
 				 &cmd->body.surface.sid, NULL);
 }
 
+static int vmw_cmd_sm5_view_define(struct vmw_private *dev_priv,
+				   struct vmw_sw_context *sw_context,
+				   SVGA3dCmdHeader *header)
+{
+	if (!has_sm5_context(dev_priv))
+		return -EINVAL;
+
+	return vmw_cmd_dx_view_define(dev_priv, sw_context, header);
+}
+
+static int vmw_cmd_sm5_view_remove(struct vmw_private *dev_priv,
+				   struct vmw_sw_context *sw_context,
+				   SVGA3dCmdHeader *header)
+{
+	if (!has_sm5_context(dev_priv))
+		return -EINVAL;
+
+	return vmw_cmd_dx_view_remove(dev_priv, sw_context, header);
+}
+
+static int vmw_cmd_clear_uav_uint(struct vmw_private *dev_priv,
+				  struct vmw_sw_context *sw_context,
+				  SVGA3dCmdHeader *header)
+{
+	struct {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdDXClearUAViewUint body;
+	} *cmd = container_of(header, typeof(*cmd), header);
+	struct vmw_resource *ret;
+
+	if (!has_sm5_context(dev_priv))
+		return -EINVAL;
+
+	ret = vmw_view_id_val_add(sw_context, vmw_view_ua,
+				  cmd->body.uaViewId);
+
+	return PTR_ERR_OR_ZERO(ret);
+}
+
+static int vmw_cmd_clear_uav_float(struct vmw_private *dev_priv,
+				   struct vmw_sw_context *sw_context,
+				   SVGA3dCmdHeader *header)
+{
+	struct {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdDXClearUAViewFloat body;
+	} *cmd = container_of(header, typeof(*cmd), header);
+	struct vmw_resource *ret;
+
+	if (!has_sm5_context(dev_priv))
+		return -EINVAL;
+
+	ret = vmw_view_id_val_add(sw_context, vmw_view_ua,
+				  cmd->body.uaViewId);
+
+	return PTR_ERR_OR_ZERO(ret);
+}
+
+static int vmw_cmd_set_uav(struct vmw_private *dev_priv,
+			   struct vmw_sw_context *sw_context,
+			   SVGA3dCmdHeader *header)
+{
+	struct {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdDXSetUAViews body;
+	} *cmd = container_of(header, typeof(*cmd), header);
+	u32 num_uav = (cmd->header.size - sizeof(cmd->body)) /
+		sizeof(SVGA3dUAViewId);
+	int ret;
+
+	if (!has_sm5_context(dev_priv))
+		return -EINVAL;
+
+	if (num_uav > SVGA3D_MAX_UAVIEWS) {
+		VMW_DEBUG_USER("Invalid UAV binding.\n");
+		return -EINVAL;
+	}
+
+	ret = vmw_view_bindings_add(sw_context, vmw_view_ua,
+				    vmw_ctx_binding_uav, 0, (void *)&cmd[1],
+				    num_uav, 0);
+	if (ret)
+		return ret;
+
+	vmw_binding_add_uav_index(sw_context->dx_ctx_node->staged, 0,
+					 cmd->body.uavSpliceIndex);
+
+	return ret;
+}
+
+static int vmw_cmd_set_cs_uav(struct vmw_private *dev_priv,
+			      struct vmw_sw_context *sw_context,
+			      SVGA3dCmdHeader *header)
+{
+	struct {
+		SVGA3dCmdHeader header;
+		SVGA3dCmdDXSetCSUAViews body;
+	} *cmd = container_of(header, typeof(*cmd), header);
+	u32 num_uav = (cmd->header.size - sizeof(cmd->body)) /
+		sizeof(SVGA3dUAViewId);
+	int ret;
+
+	if (!has_sm5_context(dev_priv))
+		return -EINVAL;
+
+	if (num_uav > SVGA3D_MAX_UAVIEWS) {
+		VMW_DEBUG_USER("Invalid UAV binding.\n");
+		return -EINVAL;
+	}
+
+	ret = vmw_view_bindings_add(sw_context, vmw_view_ua,
+				    vmw_ctx_binding_cs_uav, 0, (void *)&cmd[1],
+				    num_uav, 0);
+	if (ret)
+		return ret;
+
+	vmw_binding_add_uav_index(sw_context->dx_ctx_node->staged, 1,
+				  cmd->body.startIndex);
+
+	return ret;
+}
+
 static int vmw_cmd_check_not_3d(struct vmw_private *dev_priv,
 				struct vmw_sw_context *sw_context,
 				void *buf, uint32_t *size)
@@ -3163,6 +3287,24 @@ static const struct vmw_cmd_entry vmw_cmd_entries[SVGA_3D_CMD_MAX] = {
 		    true, false, true),
 	VMW_CMD_DEF(SVGA_3D_CMD_INTRA_SURFACE_COPY, &vmw_cmd_intra_surface_copy,
 		    true, false, true),
+
+	/*
+	 * SM5 commands
+	 */
+	VMW_CMD_DEF(SVGA_3D_CMD_DX_DEFINE_UA_VIEW, &vmw_cmd_sm5_view_define,
+		    true, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_DX_DESTROY_UA_VIEW, &vmw_cmd_sm5_view_remove,
+		    true, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_DX_CLEAR_UA_VIEW_UINT, &vmw_cmd_clear_uav_uint,
+		    true, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_DX_CLEAR_UA_VIEW_FLOAT,
+		    &vmw_cmd_clear_uav_float, true, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_DX_COPY_STRUCTURE_COUNT, &vmw_cmd_invalid, true,
+		    false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_DX_SET_UA_VIEWS, &vmw_cmd_set_uav, true, false,
+		    true),
+	VMW_CMD_DEF(SVGA_3D_CMD_DX_SET_CS_UA_VIEWS, &vmw_cmd_set_cs_uav, true,
+		    false, true),
 };
 
 bool vmw_cmd_describe(const void *buf, u32 *size, char const **cmd)
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_so.c b/drivers/gpu/drm/vmwgfx/vmwgfx_so.c
index 63807361e16f..3f97b61dd5d8 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_so.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_so.c
@@ -319,7 +319,8 @@ int vmw_view_add(struct vmw_cmdbuf_res_manager *man,
 	static const size_t vmw_view_define_sizes[] = {
 		[vmw_view_sr] = sizeof(SVGA3dCmdDXDefineShaderResourceView),
 		[vmw_view_rt] = sizeof(SVGA3dCmdDXDefineRenderTargetView),
-		[vmw_view_ds] = sizeof(SVGA3dCmdDXDefineDepthStencilView)
+		[vmw_view_ds] = sizeof(SVGA3dCmdDXDefineDepthStencilView),
+		[vmw_view_ua] = sizeof(SVGA3dCmdDXDefineUAView)
 	};
 
 	struct vmw_private *dev_priv = ctx->dev_priv;
@@ -499,8 +500,8 @@ struct vmw_resource *vmw_view_lookup(struct vmw_cmdbuf_res_manager *man,
  * Each time a resource is put on the validation list as the result of a
  * view pointing to it, we need to determine whether that resource will
  * be dirtied (written to by the GPU) as a result of the corresponding
- * GPU operation. Currently only rendertarget- and depth-stencil views are
- * capable of dirtying its resource.
+ * GPU operation. Currently only rendertarget-, depth-stencil and unordered
+ * access views are capable of dirtying its resource.
  *
  * Return: Whether the view type of @res dirties the resource it points to.
  */
@@ -509,10 +510,11 @@ u32 vmw_view_dirtying(struct vmw_resource *res)
 	static u32 view_is_dirtying[vmw_view_max] = {
 		[vmw_view_rt] = VMW_RES_DIRTY_SET,
 		[vmw_view_ds] = VMW_RES_DIRTY_SET,
+		[vmw_view_ua] = VMW_RES_DIRTY_SET,
 	};
 
 	/* Update this function as we add more view types */
-	BUILD_BUG_ON(vmw_view_max != 3);
+	BUILD_BUG_ON(vmw_view_max != 4);
 	return view_is_dirtying[vmw_view(res)->view_type];
 }
 
@@ -520,12 +522,14 @@ const u32 vmw_view_destroy_cmds[] = {
 	[vmw_view_sr] = SVGA_3D_CMD_DX_DESTROY_SHADERRESOURCE_VIEW,
 	[vmw_view_rt] = SVGA_3D_CMD_DX_DESTROY_RENDERTARGET_VIEW,
 	[vmw_view_ds] = SVGA_3D_CMD_DX_DESTROY_DEPTHSTENCIL_VIEW,
+	[vmw_view_ua] = SVGA_3D_CMD_DX_DESTROY_UA_VIEW,
 };
 
 const SVGACOTableType vmw_view_cotables[] = {
 	[vmw_view_sr] = SVGA_COTABLE_SRVIEW,
 	[vmw_view_rt] = SVGA_COTABLE_RTVIEW,
 	[vmw_view_ds] = SVGA_COTABLE_DSVIEW,
+	[vmw_view_ua] = SVGA_COTABLE_UAVIEW,
 };
 
 const SVGACOTableType vmw_so_cotables[] = {
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_so.h b/drivers/gpu/drm/vmwgfx/vmwgfx_so.h
index 12565047bc55..22b4f5720908 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_so.h
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_so.h
@@ -30,6 +30,7 @@ enum vmw_view_type {
 	vmw_view_sr,
 	vmw_view_rt,
 	vmw_view_ds,
+	vmw_view_ua,
 	vmw_view_max,
 };
 
@@ -61,6 +62,7 @@ union vmw_view_destroy {
 	struct SVGA3dCmdDXDestroyRenderTargetView rtv;
 	struct SVGA3dCmdDXDestroyShaderResourceView srv;
 	struct SVGA3dCmdDXDestroyDepthStencilView dsv;
+	struct SVGA3dCmdDXDestroyUAView uav;
 	u32 view_id;
 };
 
@@ -87,6 +89,10 @@ static inline enum vmw_view_type vmw_view_cmd_to_type(u32 id)
 {
 	u32 tmp = (id - SVGA_3D_CMD_DX_DEFINE_SHADERRESOURCE_VIEW) / 2;
 
+	if (id == SVGA_3D_CMD_DX_DEFINE_UA_VIEW ||
+	    id == SVGA_3D_CMD_DX_DESTROY_UA_VIEW)
+		return vmw_view_ua;
+
 	if (tmp > (u32)vmw_view_max)
 		return vmw_view_max;
 
-- 
2.17.1



More information about the dri-devel mailing list