r600g: build fetch shader from vertex elements

Vertex elements change are less frequent than draw call, those to
avoid rebuilding fetch shader to often build the fetch shader along
vertex elements. This also allow to move vertex buffer setup out
of draw path and make update to it less frequent.

Shader update can still be improved to only update SPI regs (based
on some rasterizer state like flat shading or point sprite ...).

Signed-off-by: Jerome Glisse <jglisse@redhat.com>

1 files changed, 19 insertions, 24 deletions

diff --git a/src/gallium/drivers/r600/r600_translate.c b/src/gallium/drivers/r600/r600_translate.c
index 2e082f1cff..d927f53398 100644
--- a/src/gallium/drivers/r600/r600_translate.c
+++ b/src/gallium/drivers/r600/r600_translate.c
@@ -41,6 +41,7 @@ void r600_begin_vertex_translate(struct r600_pipe_context *rctx)
 	struct pipe_transfer *vb_transfer[PIPE_MAX_ATTRIBS] = {0}, *out_transfer;
 	struct pipe_resource *out_buffer;
 	unsigned i, num_verts;
+	struct pipe_vertex_element new_velems[PIPE_MAX_ATTRIBS];
 
 	/* Initialize the translate key, i.e. the recipe how vertices should be
 	 * translated. */
@@ -51,9 +52,7 @@ void r600_begin_vertex_translate(struct r600_pipe_context *rctx)
 		unsigned output_format_size = ve->hw_format_size[i];
 
 		/* Check for support. */
-		if (ve->elements[i].src_format == ve->hw_format[i] &&
-		    (vb->buffer_offset + ve->elements[i].src_offset) % 4 == 0 &&
-		    vb->stride % 4 == 0) {
+		if (ve->elements[i].src_format == ve->hw_format[i]) {
 			continue;
 		}
 
@@ -147,29 +146,22 @@ void r600_begin_vertex_translate(struct r600_pipe_context *rctx)
 	}
 
 	/* Save and replace vertex elements. */
-	{
-		struct pipe_vertex_element new_velems[PIPE_MAX_ATTRIBS];
-
-		rctx->tran.saved_velems = rctx->vertex_elements;
-
-		for (i = 0; i < ve->count; i++) {
-			if (vb_translated[ve->elements[i].vertex_buffer_index]) {
-				te = &key.element[tr_elem_index[i]];
-				new_velems[i].instance_divisor = ve->elements[i].instance_divisor;
-				new_velems[i].src_format = te->output_format;
-				new_velems[i].src_offset = te->output_offset;
-				new_velems[i].vertex_buffer_index = rctx->tran.vb_slot;
-			} else {
-				memcpy(&new_velems[i], &ve->elements[i],
-				       sizeof(struct pipe_vertex_element));
-			}
+	for (i = 0; i < ve->count; i++) {
+		if (vb_translated[ve->elements[i].vertex_buffer_index]) {
+			te = &key.element[tr_elem_index[i]];
+			new_velems[i].instance_divisor = ve->elements[i].instance_divisor;
+			new_velems[i].src_format = te->output_format;
+			new_velems[i].src_offset = te->output_offset;
+			new_velems[i].vertex_buffer_index = rctx->tran.vb_slot;
+		} else {
+			memcpy(&new_velems[i], &ve->elements[i],
+					sizeof(struct pipe_vertex_element));
 		}
-
-		rctx->tran.new_velems =
-			pipe->create_vertex_elements_state(pipe, ve->count, new_velems);
-		pipe->bind_vertex_elements_state(pipe, rctx->tran.new_velems);
 	}
 
+	rctx->tran.new_velems = pipe->create_vertex_elements_state(pipe, ve->count, new_velems);
+	pipe->bind_vertex_elements_state(pipe, rctx->tran.new_velems);
+
 	pipe_resource_reference(&out_buffer, NULL);
 }
 
@@ -178,8 +170,11 @@ void r600_end_vertex_translate(struct r600_pipe_context *rctx)
 	struct pipe_context *pipe = &rctx->context;
 
 	/* Restore vertex elements. */
-	pipe->bind_vertex_elements_state(pipe, rctx->tran.saved_velems);
+	if (rctx->vertex_elements == rctx->tran.new_velems) {
+		pipe->bind_vertex_elements_state(pipe, NULL);
+	}
 	pipe->delete_vertex_elements_state(pipe, rctx->tran.new_velems);
+	rctx->tran.new_velems = NULL;
 
 	/* Delete the now-unused VBO. */
 	pipe_resource_reference(&rctx->vertex_buffer[rctx->tran.vb_slot].buffer,