drm/vmwgfx: add support for updating only offsets of constant buffers

This adds support for the
SVGA_3D_CMD_DX_SET_VS/PS/GS/HS/DS/CS_CONSTANT_BUFFER_OFFSET commands (which only update
the offset, but don't rebind the buffer), which saves some overhead.

Signed-off-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Charmaine Lee <charmainel@vmware.com>
Reviewed-by: Martin Krastev <krastevm@vmware.com>
Signed-off-by: Zack Rusin <zackr@vmware.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20211206172620.3139754-11-zack@kde.org

3 files changed, 79 insertions, 0 deletions

diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_binding.c b/drivers/gpu/drm/vmwgfx/vmwgfx_binding.c
index 9aa69ba85670..ae2de914eb89 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_binding.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_binding.c
@@ -354,6 +354,27 @@ void vmw_binding_add(struct vmw_ctx_binding_state *cbs,
 }
 
 /**
+ * vmw_binding_cb_offset_update: Update the offset of a cb binding
+ *
+ * @cbs: Pointer to the context binding state tracker.
+ * @shader_slot: The shader slot of the binding.
+ * @slot: The slot of the binding.
+ * @offsetInBytes: The new offset of the binding.
+ *
+ * Updates the offset of an existing cb binding in the context binding
+ * state structure @cbs.
+ */
+void vmw_binding_cb_offset_update(struct vmw_ctx_binding_state *cbs,
+				  u32 shader_slot, u32 slot, u32 offsetInBytes)
+{
+	struct vmw_ctx_bindinfo *loc =
+		vmw_binding_loc(cbs, vmw_ctx_binding_cb, shader_slot, slot);
+	struct vmw_ctx_bindinfo_cb *loc_cb =
+		(struct vmw_ctx_bindinfo_cb *)((u8 *) loc);
+	loc_cb->offset = offsetInBytes;
+}
+
+/**
  * 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.
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_binding.h b/drivers/gpu/drm/vmwgfx/vmwgfx_binding.h
index 6b1b234d12a1..85b90f7d398d 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_binding.h
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_binding.h
@@ -217,6 +217,8 @@ struct vmw_ctx_bindinfo_so {
 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_cb_offset_update(struct vmw_ctx_binding_state *cbs,
+					 u32 shader_slot, u32 slot, u32 offsetInBytes);
 extern void vmw_binding_add_uav_index(struct vmw_ctx_binding_state *cbs,
 				      uint32 slot, uint32 splice_index);
 extern void
diff --git a/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c b/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
index fd204fe2c68f..44ca23b0ea4e 100644
--- a/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_execbuf.c
@@ -2164,6 +2164,44 @@ vmw_cmd_dx_set_single_constant_buffer(struct vmw_private *dev_priv,
 }
 
 /**
+ * vmw_cmd_dx_set_constant_buffer_offset - Validate
+ * SVGA_3D_CMD_DX_SET_VS/PS/GS/HS/DS/CS_CONSTANT_BUFFER_OFFSET command.
+ *
+ * @dev_priv: Pointer to a device private struct.
+ * @sw_context: The software context being used for this batch.
+ * @header: Pointer to the command header in the command stream.
+ */
+static int
+vmw_cmd_dx_set_constant_buffer_offset(struct vmw_private *dev_priv,
+				      struct vmw_sw_context *sw_context,
+				      SVGA3dCmdHeader *header)
+{
+	VMW_DECLARE_CMD_VAR(*cmd, SVGA3dCmdDXSetConstantBufferOffset);
+
+	struct vmw_ctx_validation_info *ctx_node = VMW_GET_CTX_NODE(sw_context);
+	u32 shader_slot;
+
+	if (!has_sm5_context(dev_priv))
+		return -EINVAL;
+
+	if (!ctx_node)
+		return -EINVAL;
+
+	cmd = container_of(header, typeof(*cmd), header);
+	if (cmd->body.slot >= SVGA3D_DX_MAX_CONSTBUFFERS) {
+		VMW_DEBUG_USER("Illegal const buffer slot %u.\n",
+			       (unsigned int) cmd->body.slot);
+		return -EINVAL;
+	}
+
+	shader_slot = cmd->header.id - SVGA_3D_CMD_DX_SET_VS_CONSTANT_BUFFER_OFFSET;
+	vmw_binding_cb_offset_update(ctx_node->staged, shader_slot,
+				     cmd->body.slot, cmd->body.offsetInBytes);
+
+	return 0;
+}
+
+/**
  * vmw_cmd_dx_set_shader_res - Validate SVGA_3D_CMD_DX_SET_SHADER_RESOURCES
  * command
  *
@@ -3526,6 +3564,24 @@ static const struct vmw_cmd_entry vmw_cmd_entries[SVGA_3D_CMD_MAX] = {
 	VMW_CMD_DEF(SVGA_3D_CMD_DX_TRANSFER_FROM_BUFFER,
 		    &vmw_cmd_dx_transfer_from_buffer,
 		    true, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_DX_SET_VS_CONSTANT_BUFFER_OFFSET,
+		    &vmw_cmd_dx_set_constant_buffer_offset,
+		    true, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_DX_SET_PS_CONSTANT_BUFFER_OFFSET,
+		    &vmw_cmd_dx_set_constant_buffer_offset,
+		    true, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_DX_SET_GS_CONSTANT_BUFFER_OFFSET,
+		    &vmw_cmd_dx_set_constant_buffer_offset,
+		    true, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_DX_SET_HS_CONSTANT_BUFFER_OFFSET,
+		    &vmw_cmd_dx_set_constant_buffer_offset,
+		    true, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_DX_SET_DS_CONSTANT_BUFFER_OFFSET,
+		    &vmw_cmd_dx_set_constant_buffer_offset,
+		    true, false, true),
+	VMW_CMD_DEF(SVGA_3D_CMD_DX_SET_CS_CONSTANT_BUFFER_OFFSET,
+		    &vmw_cmd_dx_set_constant_buffer_offset,
+		    true, false, true),
 	VMW_CMD_DEF(SVGA_3D_CMD_INTRA_SURFACE_COPY, &vmw_cmd_intra_surface_copy,
 		    true, false, true),