r600g: optimise state setting in r600_draw_vbo.

This drop the r600_draw_vbo CPU usage on a run of nexuiz from 1.40% to 0.72%
in sysprof for me on my Fusion APU.

Signed-off-by: Dave Airlie <airlied@redhat.com>

diff --git a/src/gallium/drivers/r600/r600_pipe.h b/src/gallium/drivers/r600/r600_pipe.h
index 0e4cfeb5b8032633918a7ad31c5961360668c5d6..1102c5c72ebfc8dda806cf9a1fb5f8380d15bff2 100644 (file)
--- a/src/gallium/drivers/r600/r600_pipe.h
+++ b/src/gallium/drivers/r600/r600_pipe.h
@@ -188,6 +188,7 @@ struct r600_pipe_context {
        struct r600_pipe_state          ps_const_buffer;
        struct r600_pipe_state          ps_const_buffer_resource[R600_MAX_CONST_BUFFERS];
        struct r600_pipe_rasterizer     *rasterizer;
+       struct r600_pipe_state          vgt;
        /* shader information */
        unsigned                        sprite_coord_enable;
        bool                            flatshade;

diff --git a/src/gallium/drivers/r600/r600_state_common.c b/src/gallium/drivers/r600/r600_state_common.c
index de1b811ce8975caf0924466bd2eba4109bbcd2df..4619d291029b0b88891e0e21c3254c5949c6a736 100644 (file)
--- a/src/gallium/drivers/r600/r600_state_common.c
+++ b/src/gallium/drivers/r600/r600_state_common.c
@@ -504,7 +504,6 @@ void r600_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info *info)
        struct r600_resource *rbuffer;
        u32 vgt_dma_index_type, vgt_dma_swap_mode, vgt_draw_initiator, mask;
        struct r600_draw rdraw;
-       struct r600_pipe_state vgt;
        struct r600_drawl draw = {};
        unsigned prim;
 
@@ -583,16 +582,26 @@ void r600_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info *info)
                mask |= (0xF << (i * 4));
        }
 
-       vgt.id = R600_PIPE_STATE_VGT;
-       vgt.nregs = 0;
-       r600_pipe_state_add_reg(&vgt, R_008958_VGT_PRIMITIVE_TYPE, prim, 0xFFFFFFFF, NULL);
-       r600_pipe_state_add_reg(&vgt, R_028408_VGT_INDX_OFFSET, draw.info.index_bias, 0xFFFFFFFF, NULL);
-       r600_pipe_state_add_reg(&vgt, R_028400_VGT_MAX_VTX_INDX, draw.info.max_index, 0xFFFFFFFF, NULL);
-       r600_pipe_state_add_reg(&vgt, R_028404_VGT_MIN_VTX_INDX, draw.info.min_index, 0xFFFFFFFF, NULL);
-       r600_pipe_state_add_reg(&vgt, R_028238_CB_TARGET_MASK, rctx->cb_target_mask & mask, 0xFFFFFFFF, NULL);
-       r600_pipe_state_add_reg(&vgt, R_03CFF0_SQ_VTX_BASE_VTX_LOC, 0, 0xFFFFFFFF, NULL);
-       r600_pipe_state_add_reg(&vgt, R_03CFF4_SQ_VTX_START_INST_LOC, draw.info.start_instance, 0xFFFFFFFF, NULL);
-       r600_context_pipe_state_set(&rctx->ctx, &vgt);
+       if (rctx->vgt.id != R600_PIPE_STATE_VGT) {
+               rctx->vgt.id = R600_PIPE_STATE_VGT;
+               rctx->vgt.nregs = 0;
+               r600_pipe_state_add_reg(&rctx->vgt, R_008958_VGT_PRIMITIVE_TYPE, prim, 0xFFFFFFFF, NULL);
+               r600_pipe_state_add_reg(&rctx->vgt, R_028238_CB_TARGET_MASK, rctx->cb_target_mask & mask, 0xFFFFFFFF, NULL);
+               r600_pipe_state_add_reg(&rctx->vgt, R_028400_VGT_MAX_VTX_INDX, draw.info.max_index, 0xFFFFFFFF, NULL);
+               r600_pipe_state_add_reg(&rctx->vgt, R_028404_VGT_MIN_VTX_INDX, draw.info.min_index, 0xFFFFFFFF, NULL);
+               r600_pipe_state_add_reg(&rctx->vgt, R_028408_VGT_INDX_OFFSET, draw.info.index_bias, 0xFFFFFFFF, NULL);
+               r600_pipe_state_add_reg(&rctx->vgt, R_03CFF0_SQ_VTX_BASE_VTX_LOC, 0, 0xFFFFFFFF, NULL);
+               r600_pipe_state_add_reg(&rctx->vgt, R_03CFF4_SQ_VTX_START_INST_LOC, draw.info.start_instance, 0xFFFFFFFF, NULL);
+       } else {
+               rctx->vgt.regs[0].value = prim;
+               rctx->vgt.regs[1].value = rctx->cb_target_mask & mask;
+               rctx->vgt.regs[2].value = draw.info.max_index;
+               rctx->vgt.regs[3].value = draw.info.min_index;
+               rctx->vgt.regs[4].value = draw.info.index_bias;
+               rctx->vgt.regs[6].value = draw.info.start_instance;
+       }
+
+       r600_context_pipe_state_set(&rctx->ctx, &rctx->vgt);
 
        rdraw.vgt_num_indices = draw.info.count;
        rdraw.vgt_num_instances = draw.info.instance_count;