radeonsi: calculate ESGS_RING_ITEMSIZE in create_shader

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

diff --git a/src/gallium/drivers/radeonsi/si_shader.h b/src/gallium/drivers/radeonsi/si_shader.h
index 1dd25227e9d509bddcb3ca95317a2364917f47c3..b87fb715aac9d334c6b6c4428f787cc205254559 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_shader.h
+++ b/src/gallium/drivers/radeonsi/si_shader.h
@@ -202,6 +202,7 @@ struct si_shader_selector {
        bool            forces_persample_interp_for_persp;
        bool            forces_persample_interp_for_linear;
 
+       unsigned        esgs_itemsize;
        unsigned        gs_output_prim;
        unsigned        gs_max_out_vertices;
        unsigned        gs_num_invocations;

diff --git a/src/gallium/drivers/radeonsi/si_state_shaders.c b/src/gallium/drivers/radeonsi/si_state_shaders.c
index 0e403c492f527abc1002adcba99fa2e0ddc9e27e..04754a7e0ea4d8c6e3cc1b6a5ae6e4ed84360c4c 100644 (file)
--- a/src/gallium/drivers/radeonsi/si_state_shaders.c
+++ b/src/gallium/drivers/radeonsi/si_state_shaders.c
@@ -251,7 +251,7 @@ static void si_shader_gs(struct si_shader *shader)
        si_pm4_set_reg(pm4, R_028A68_VGT_GSVS_RING_OFFSET_3, gsvs_itemsize * ((max_stream >= 3) ? 3 : 1));
 
        si_pm4_set_reg(pm4, R_028AAC_VGT_ESGS_RING_ITEMSIZE,
-                      util_bitcount64(shader->selector->inputs_read) * (16 >> 2));
+                      shader->selector->esgs_itemsize / 4);
        si_pm4_set_reg(pm4, R_028AB0_VGT_GSVS_RING_ITEMSIZE, gsvs_itemsize * (max_stream + 1));
 
        si_pm4_set_reg(pm4, R_028B38_VGT_GS_MAX_VERT_OUT, gs_max_vert_out);
@@ -739,6 +739,7 @@ static void *si_create_shader_selector(struct pipe_context *ctx,
                                        1llu << si_shader_io_get_unique_index(name, index);
                        }
                }
+               sel->esgs_itemsize = util_last_bit64(sel->outputs_written) * 16;
                break;
        case PIPE_SHADER_FRAGMENT:
                for (i = 0; i < sel->info.num_outputs; i++) {