radv: move calculating vs out info regs into pipeline.

This moves some calculations of register values into the pipeline
construction, it saves looking at outinfo in the cmd buffer emit.

Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>

diff --git a/src/amd/vulkan/radv_cmd_buffer.c b/src/amd/vulkan/radv_cmd_buffer.c
index 9a191cda1c559e0efc0dbc4ba49bfe2b47ef14fe..b1dbb04eaad5fdb2d508c78f5c9528772f3028bf 100644 (file)
--- a/src/amd/vulkan/radv_cmd_buffer.c
+++ b/src/amd/vulkan/radv_cmd_buffer.c
@@ -698,28 +698,15 @@ radv_emit_shaders_prefetch(struct radv_cmd_buffer *cmd_buffer,
 static void
 radv_emit_hw_vs(struct radv_cmd_buffer *cmd_buffer,
                struct radv_pipeline *pipeline,
-               struct radv_shader_variant *shader,
-               struct ac_vs_output_info *outinfo)
+               struct radv_shader_variant *shader)
 {
        uint64_t va = radv_buffer_get_va(shader->bo) + shader->bo_offset;
-       unsigned export_count;
 
-       export_count = MAX2(1, outinfo->param_exports);
        radeon_set_context_reg(cmd_buffer->cs, R_0286C4_SPI_VS_OUT_CONFIG,
-                              S_0286C4_VS_EXPORT_COUNT(export_count - 1));
+                              pipeline->graphics.vs.spi_vs_out_config);
 
        radeon_set_context_reg(cmd_buffer->cs, R_02870C_SPI_SHADER_POS_FORMAT,
-                              S_02870C_POS0_EXPORT_FORMAT(V_02870C_SPI_SHADER_4COMP) |
-                              S_02870C_POS1_EXPORT_FORMAT(outinfo->pos_exports > 1 ?
-                                                          V_02870C_SPI_SHADER_4COMP :
-                                                          V_02870C_SPI_SHADER_NONE) |
-                              S_02870C_POS2_EXPORT_FORMAT(outinfo->pos_exports > 2 ?
-                                                          V_02870C_SPI_SHADER_4COMP :
-                                                          V_02870C_SPI_SHADER_NONE) |
-                              S_02870C_POS3_EXPORT_FORMAT(outinfo->pos_exports > 3 ?
-                                                          V_02870C_SPI_SHADER_4COMP :
-                                                          V_02870C_SPI_SHADER_NONE));
-
+                              pipeline->graphics.vs.spi_shader_pos_format);
 
        radeon_set_sh_reg_seq(cmd_buffer->cs, R_00B120_SPI_SHADER_PGM_LO_VS, 4);
        radeon_emit(cmd_buffer->cs, va >> 8);
@@ -735,11 +722,11 @@ radv_emit_hw_vs(struct radv_cmd_buffer *cmd_buffer,
 
 
        radeon_set_context_reg(cmd_buffer->cs, R_02881C_PA_CL_VS_OUT_CNTL,
-                              pipeline->graphics.pa_cl_vs_out_cntl);
+                              pipeline->graphics.vs.pa_cl_vs_out_cntl);
 
        if (cmd_buffer->device->physical_device->rad_info.chip_class <= VI)
                radeon_set_context_reg(cmd_buffer->cs, R_028AB4_VGT_REUSE_OFF,
-                                      S_028AB4_REUSE_OFF(outinfo->writes_viewport_index));
+                                      pipeline->graphics.vs.vgt_reuse_off);
 }
 
 static void
@@ -821,7 +808,7 @@ radv_emit_vertex_shader(struct radv_cmd_buffer *cmd_buffer,
        else if (vs->info.vs.as_es)
                radv_emit_hw_es(cmd_buffer, vs, &vs->info.vs.es_info);
        else
-               radv_emit_hw_vs(cmd_buffer, pipeline, vs, &vs->info.vs.outinfo);
+               radv_emit_hw_vs(cmd_buffer, pipeline, vs);
 }
 
 
@@ -841,7 +828,7 @@ radv_emit_tess_shaders(struct radv_cmd_buffer *cmd_buffer,
                if (tes->info.tes.as_es)
                        radv_emit_hw_es(cmd_buffer, tes, &tes->info.tes.es_info);
                else
-                       radv_emit_hw_vs(cmd_buffer, pipeline, tes, &tes->info.tes.outinfo);
+                       radv_emit_hw_vs(cmd_buffer, pipeline, tes);
        }
 
        radv_emit_hw_hs(cmd_buffer, tcs);
@@ -951,7 +938,7 @@ radv_emit_geometry_shader(struct radv_cmd_buffer *cmd_buffer,
                radeon_emit(cmd_buffer->cs, gs->rsrc2);
        }
 
-       radv_emit_hw_vs(cmd_buffer, pipeline, pipeline->gs_copy_shader, &pipeline->gs_copy_shader->info.vs.outinfo);
+       radv_emit_hw_vs(cmd_buffer, pipeline, pipeline->gs_copy_shader);
 
        struct ac_userdata_info *loc = radv_lookup_user_sgpr(cmd_buffer->state.pipeline, MESA_SHADER_GEOMETRY,
                                                             AC_UD_GS_VS_RING_STRIDE_ENTRIES);

diff --git a/src/amd/vulkan/radv_pipeline.c b/src/amd/vulkan/radv_pipeline.c
index f62c8b4be882fdc3e1687ad5b065d6c676477112..d74e05d6c821b1fdd4dd95272b10482c99de9df1 100644 (file)
--- a/src/amd/vulkan/radv_pipeline.c
+++ b/src/amd/vulkan/radv_pipeline.c
@@ -1545,7 +1545,7 @@ static void calculate_vgt_gs_mode(struct radv_pipeline *pipeline)
        }
 }
 
-static void calculate_pa_cl_vs_out_cntl(struct radv_pipeline *pipeline)
+static void calculate_vs_outinfo(struct radv_pipeline *pipeline)
 {
        struct ac_vs_output_info *outinfo = get_vs_output_info(pipeline);
 
@@ -1557,7 +1557,7 @@ static void calculate_pa_cl_vs_out_cntl(struct radv_pipeline *pipeline)
        bool misc_vec_ena = outinfo->writes_pointsize ||
                outinfo->writes_layer ||
                outinfo->writes_viewport_index;
-       pipeline->graphics.pa_cl_vs_out_cntl =
+       pipeline->graphics.vs.pa_cl_vs_out_cntl =
                S_02881C_USE_VTX_POINT_SIZE(outinfo->writes_pointsize) |
                S_02881C_USE_VTX_RENDER_TARGET_INDX(outinfo->writes_layer) |
                S_02881C_USE_VTX_VIEWPORT_INDX(outinfo->writes_viewport_index) |
@@ -1568,6 +1568,21 @@ static void calculate_pa_cl_vs_out_cntl(struct radv_pipeline *pipeline)
                cull_dist_mask << 8 |
                clip_dist_mask;
 
+       pipeline->graphics.vs.spi_shader_pos_format =
+               S_02870C_POS0_EXPORT_FORMAT(V_02870C_SPI_SHADER_4COMP) |
+               S_02870C_POS1_EXPORT_FORMAT(outinfo->pos_exports > 1 ?
+                                           V_02870C_SPI_SHADER_4COMP :
+                                           V_02870C_SPI_SHADER_NONE) |
+               S_02870C_POS2_EXPORT_FORMAT(outinfo->pos_exports > 2 ?
+                                           V_02870C_SPI_SHADER_4COMP :
+                                           V_02870C_SPI_SHADER_NONE) |
+               S_02870C_POS3_EXPORT_FORMAT(outinfo->pos_exports > 3 ?
+                                           V_02870C_SPI_SHADER_4COMP :
+                                           V_02870C_SPI_SHADER_NONE);
+
+       pipeline->graphics.vs.spi_vs_out_config = S_0286C4_VS_EXPORT_COUNT(MAX2(1, outinfo->param_exports) - 1);
+       /* only emitted on pre-VI */
+       pipeline->graphics.vs.vgt_reuse_off = S_028AB4_REUSE_OFF(outinfo->writes_viewport_index);
 }
 
 static uint32_t offset_to_ps_input(uint32_t offset, bool flat_shade)
@@ -2093,7 +2108,7 @@ radv_pipeline_init(struct radv_pipeline *pipeline,
                V_028710_SPI_SHADER_ZERO;
 
        calculate_vgt_gs_mode(pipeline);
-       calculate_pa_cl_vs_out_cntl(pipeline);
+       calculate_vs_outinfo(pipeline);
        calculate_ps_inputs(pipeline);
 
        for (unsigned i = 0; i < MESA_SHADER_STAGES; i++) {

diff --git a/src/amd/vulkan/radv_private.h b/src/amd/vulkan/radv_private.h
index 0f48681c941d3ff787ecc7506346091c1f0b9e3d..58b9f469048855beb0846ced1fd6616d7a91cded 100644 (file)
--- a/src/amd/vulkan/radv_private.h
+++ b/src/amd/vulkan/radv_private.h
@@ -1110,6 +1110,13 @@ struct radv_vertex_elements_info {
        uint32_t count;
 };
 
+struct radv_vs_state {
+       uint32_t pa_cl_vs_out_cntl;
+       uint32_t spi_shader_pos_format;
+       uint32_t spi_vs_out_config;
+       uint32_t vgt_reuse_off;
+};
+
 #define SI_GS_PER_ES 128
 
 struct radv_pipeline {
@@ -1137,6 +1144,7 @@ struct radv_pipeline {
                        struct radv_multisample_state ms;
                        struct radv_tessellation_state tess;
                        struct radv_gs_state gs;
+                       struct radv_vs_state vs;
                        uint32_t db_shader_control;
                        uint32_t shader_z_format;
                        unsigned prim;
@@ -1150,7 +1158,6 @@ struct radv_pipeline {
                        unsigned gsvs_ring_size;
                        uint32_t ps_input_cntl[32];
                        uint32_t ps_input_cntl_num;
-                       uint32_t pa_cl_vs_out_cntl;
                        uint32_t vgt_shader_stages_en;
                        uint32_t vtx_base_sgpr;
                        uint32_t base_ia_multi_vgt_param;