+static void si_multiwave_lds_size_workaround(struct radv_device *device,
+ unsigned *lds_size)
+{
+ /* SPI barrier management bug:
+ * Make sure we have at least 4k of LDS in use to avoid the bug.
+ * It applies to workgroup sizes of more than one wavefront.
+ */
+ if (device->physical_device->rad_info.family == CHIP_BONAIRE ||
+ device->physical_device->rad_info.family == CHIP_KABINI ||
+ device->physical_device->rad_info.family == CHIP_MULLINS)
+ *lds_size = MAX2(*lds_size, 8);
+}
+
+static void
+calculate_tess_state(struct radv_pipeline *pipeline,
+ const VkGraphicsPipelineCreateInfo *pCreateInfo)
+{
+ unsigned num_tcs_input_cp = pCreateInfo->pTessellationState->patchControlPoints;
+ unsigned num_tcs_output_cp, num_tcs_inputs, num_tcs_outputs;
+ unsigned num_tcs_patch_outputs;
+ unsigned input_vertex_size, output_vertex_size, pervertex_output_patch_size;
+ unsigned input_patch_size, output_patch_size, output_patch0_offset;
+ unsigned lds_size, hardware_lds_size;
+ unsigned perpatch_output_offset;
+ unsigned num_patches;
+ struct radv_tessellation_state *tess = &pipeline->graphics.tess;
+
+ /* This calculates how shader inputs and outputs among VS, TCS, and TES
+ * are laid out in LDS. */
+ num_tcs_inputs = util_last_bit64(pipeline->shaders[MESA_SHADER_VERTEX]->info.vs.outputs_written);
+
+ num_tcs_outputs = util_last_bit64(pipeline->shaders[MESA_SHADER_TESS_CTRL]->info.tcs.outputs_written); //tcs->outputs_written
+ num_tcs_output_cp = pipeline->shaders[MESA_SHADER_TESS_CTRL]->info.tcs.tcs_vertices_out; //TCS VERTICES OUT
+ num_tcs_patch_outputs = util_last_bit64(pipeline->shaders[MESA_SHADER_TESS_CTRL]->info.tcs.patch_outputs_written);
+
+ /* Ensure that we only need one wave per SIMD so we don't need to check
+ * resource usage. Also ensures that the number of tcs in and out
+ * vertices per threadgroup are at most 256.
+ */
+ input_vertex_size = num_tcs_inputs * 16;
+ output_vertex_size = num_tcs_outputs * 16;
+
+ input_patch_size = num_tcs_input_cp * input_vertex_size;
+
+ pervertex_output_patch_size = num_tcs_output_cp * output_vertex_size;
+ output_patch_size = pervertex_output_patch_size + num_tcs_patch_outputs * 16;
+ /* Ensure that we only need one wave per SIMD so we don't need to check
+ * resource usage. Also ensures that the number of tcs in and out
+ * vertices per threadgroup are at most 256.
+ */
+ num_patches = 64 / MAX2(num_tcs_input_cp, num_tcs_output_cp) * 4;
+
+ /* Make sure that the data fits in LDS. This assumes the shaders only
+ * use LDS for the inputs and outputs.
+ */
+ hardware_lds_size = pipeline->device->physical_device->rad_info.chip_class >= CIK ? 65536 : 32768;
+ num_patches = MIN2(num_patches, hardware_lds_size / (input_patch_size + output_patch_size));
+
+ /* Make sure the output data fits in the offchip buffer */
+ num_patches = MIN2(num_patches,
+ (pipeline->device->tess_offchip_block_dw_size * 4) /
+ output_patch_size);
+
+ /* Not necessary for correctness, but improves performance. The
+ * specific value is taken from the proprietary driver.
+ */
+ num_patches = MIN2(num_patches, 40);
+
+ /* SI bug workaround - limit LS-HS threadgroups to only one wave. */
+ if (pipeline->device->physical_device->rad_info.chip_class == SI) {
+ unsigned one_wave = 64 / MAX2(num_tcs_input_cp, num_tcs_output_cp);
+ num_patches = MIN2(num_patches, one_wave);
+ }
+
+ output_patch0_offset = input_patch_size * num_patches;
+ perpatch_output_offset = output_patch0_offset + pervertex_output_patch_size;
+
+ lds_size = output_patch0_offset + output_patch_size * num_patches;
+
+ if (pipeline->device->physical_device->rad_info.chip_class >= CIK) {
+ assert(lds_size <= 65536);
+ lds_size = align(lds_size, 512) / 512;
+ } else {
+ assert(lds_size <= 32768);
+ lds_size = align(lds_size, 256) / 256;
+ }
+ si_multiwave_lds_size_workaround(pipeline->device, &lds_size);
+
+ tess->lds_size = lds_size;
+
+ tess->tcs_in_layout = (input_patch_size / 4) |
+ ((input_vertex_size / 4) << 13);
+ tess->tcs_out_layout = (output_patch_size / 4) |
+ ((output_vertex_size / 4) << 13);
+ tess->tcs_out_offsets = (output_patch0_offset / 16) |
+ ((perpatch_output_offset / 16) << 16);
+ tess->offchip_layout = (pervertex_output_patch_size * num_patches << 16) |
+ (num_tcs_output_cp << 9) | num_patches;
+
+ tess->ls_hs_config = S_028B58_NUM_PATCHES(num_patches) |
+ S_028B58_HS_NUM_INPUT_CP(num_tcs_input_cp) |
+ S_028B58_HS_NUM_OUTPUT_CP(num_tcs_output_cp);
+ tess->num_patches = num_patches;
+ tess->num_tcs_input_cp = num_tcs_input_cp;
+
+ struct radv_shader_variant *tes = pipeline->shaders[MESA_SHADER_TESS_EVAL];
+ unsigned type = 0, partitioning = 0, topology = 0, distribution_mode = 0;
+
+ switch (tes->info.tes.primitive_mode) {
+ case GL_TRIANGLES:
+ type = V_028B6C_TESS_TRIANGLE;
+ break;
+ case GL_QUADS:
+ type = V_028B6C_TESS_QUAD;
+ break;
+ case GL_ISOLINES:
+ type = V_028B6C_TESS_ISOLINE;
+ break;
+ }
+
+ switch (tes->info.tes.spacing) {
+ case TESS_SPACING_EQUAL:
+ partitioning = V_028B6C_PART_INTEGER;
+ break;
+ case TESS_SPACING_FRACTIONAL_ODD:
+ partitioning = V_028B6C_PART_FRAC_ODD;
+ break;
+ case TESS_SPACING_FRACTIONAL_EVEN:
+ partitioning = V_028B6C_PART_FRAC_EVEN;
+ break;
+ default:
+ break;
+ }
+
+ if (tes->info.tes.point_mode)
+ topology = V_028B6C_OUTPUT_POINT;
+ else if (tes->info.tes.primitive_mode == GL_ISOLINES)
+ topology = V_028B6C_OUTPUT_LINE;
+ else if (tes->info.tes.ccw)
+ topology = V_028B6C_OUTPUT_TRIANGLE_CW;
+ else
+ topology = V_028B6C_OUTPUT_TRIANGLE_CCW;
+
+ if (pipeline->device->has_distributed_tess) {
+ if (pipeline->device->physical_device->rad_info.family == CHIP_FIJI ||
+ pipeline->device->physical_device->rad_info.family >= CHIP_POLARIS10)
+ distribution_mode = V_028B6C_DISTRIBUTION_MODE_TRAPEZOIDS;
+ else
+ distribution_mode = V_028B6C_DISTRIBUTION_MODE_DONUTS;
+ } else
+ distribution_mode = V_028B6C_DISTRIBUTION_MODE_NO_DIST;
+
+ tess->tf_param = S_028B6C_TYPE(type) |
+ S_028B6C_PARTITIONING(partitioning) |
+ S_028B6C_TOPOLOGY(topology) |
+ S_028B6C_DISTRIBUTION_MODE(distribution_mode);
+}
+
+static const struct radv_prim_vertex_count prim_size_table[] = {
+ [V_008958_DI_PT_NONE] = {0, 0},
+ [V_008958_DI_PT_POINTLIST] = {1, 1},
+ [V_008958_DI_PT_LINELIST] = {2, 2},
+ [V_008958_DI_PT_LINESTRIP] = {2, 1},
+ [V_008958_DI_PT_TRILIST] = {3, 3},
+ [V_008958_DI_PT_TRIFAN] = {3, 1},
+ [V_008958_DI_PT_TRISTRIP] = {3, 1},
+ [V_008958_DI_PT_LINELIST_ADJ] = {4, 4},
+ [V_008958_DI_PT_LINESTRIP_ADJ] = {4, 1},
+ [V_008958_DI_PT_TRILIST_ADJ] = {6, 6},
+ [V_008958_DI_PT_TRISTRIP_ADJ] = {6, 2},
+ [V_008958_DI_PT_RECTLIST] = {3, 3},
+ [V_008958_DI_PT_LINELOOP] = {2, 1},
+ [V_008958_DI_PT_POLYGON] = {3, 1},
+ [V_008958_DI_PT_2D_TRI_STRIP] = {0, 0},
+};
+
+static uint32_t si_vgt_gs_mode(struct radv_shader_variant *gs)
+{
+ unsigned gs_max_vert_out = gs->info.gs.vertices_out;
+ unsigned cut_mode;
+
+ if (gs_max_vert_out <= 128) {
+ cut_mode = V_028A40_GS_CUT_128;
+ } else if (gs_max_vert_out <= 256) {
+ cut_mode = V_028A40_GS_CUT_256;
+ } else if (gs_max_vert_out <= 512) {
+ cut_mode = V_028A40_GS_CUT_512;
+ } else {
+ assert(gs_max_vert_out <= 1024);
+ cut_mode = V_028A40_GS_CUT_1024;
+ }
+
+ return S_028A40_MODE(V_028A40_GS_SCENARIO_G) |
+ S_028A40_CUT_MODE(cut_mode)|
+ S_028A40_ES_WRITE_OPTIMIZE(1) |
+ S_028A40_GS_WRITE_OPTIMIZE(1);
+}
+
+static void calculate_vgt_gs_mode(struct radv_pipeline *pipeline)
+{
+ struct radv_shader_variant *vs;
+ vs = radv_pipeline_has_gs(pipeline) ? pipeline->gs_copy_shader : (radv_pipeline_has_tess(pipeline) ? pipeline->shaders[MESA_SHADER_TESS_EVAL] : pipeline->shaders[MESA_SHADER_VERTEX]);
+
+ struct ac_vs_output_info *outinfo = &vs->info.vs.outinfo;
+
+ pipeline->graphics.vgt_primitiveid_en = false;
+ pipeline->graphics.vgt_gs_mode = 0;
+
+ if (radv_pipeline_has_gs(pipeline)) {
+ pipeline->graphics.vgt_gs_mode = si_vgt_gs_mode(pipeline->shaders[MESA_SHADER_GEOMETRY]);
+ } else if (outinfo->export_prim_id) {
+ pipeline->graphics.vgt_gs_mode = S_028A40_MODE(V_028A40_GS_SCENARIO_A);
+ pipeline->graphics.vgt_primitiveid_en = true;
+ }
+}
+
+static void calculate_pa_cl_vs_out_cntl(struct radv_pipeline *pipeline)
+{
+ struct radv_shader_variant *vs;
+ vs = radv_pipeline_has_gs(pipeline) ? pipeline->gs_copy_shader : (radv_pipeline_has_tess(pipeline) ? pipeline->shaders[MESA_SHADER_TESS_EVAL] : pipeline->shaders[MESA_SHADER_VERTEX]);
+
+ struct ac_vs_output_info *outinfo = &vs->info.vs.outinfo;
+
+ unsigned clip_dist_mask, cull_dist_mask, total_mask;
+ clip_dist_mask = outinfo->clip_dist_mask;
+ cull_dist_mask = outinfo->cull_dist_mask;
+ total_mask = clip_dist_mask | cull_dist_mask;
+
+ bool misc_vec_ena = outinfo->writes_pointsize ||
+ outinfo->writes_layer ||
+ outinfo->writes_viewport_index;
+ pipeline->graphics.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) |
+ S_02881C_VS_OUT_MISC_VEC_ENA(misc_vec_ena) |
+ S_02881C_VS_OUT_MISC_SIDE_BUS_ENA(misc_vec_ena) |
+ S_02881C_VS_OUT_CCDIST0_VEC_ENA((total_mask & 0x0f) != 0) |
+ S_02881C_VS_OUT_CCDIST1_VEC_ENA((total_mask & 0xf0) != 0) |
+ cull_dist_mask << 8 |
+ clip_dist_mask;
+
+}
+
+static uint32_t offset_to_ps_input(uint32_t offset, bool flat_shade)
+{
+ uint32_t ps_input_cntl;
+ if (offset <= AC_EXP_PARAM_OFFSET_31) {
+ ps_input_cntl = S_028644_OFFSET(offset);
+ if (flat_shade)
+ ps_input_cntl |= S_028644_FLAT_SHADE(1);
+ } else {
+ /* The input is a DEFAULT_VAL constant. */
+ assert(offset >= AC_EXP_PARAM_DEFAULT_VAL_0000 &&
+ offset <= AC_EXP_PARAM_DEFAULT_VAL_1111);
+ offset -= AC_EXP_PARAM_DEFAULT_VAL_0000;
+ ps_input_cntl = S_028644_OFFSET(0x20) |
+ S_028644_DEFAULT_VAL(offset);
+ }
+ return ps_input_cntl;
+}
+
+static void calculate_ps_inputs(struct radv_pipeline *pipeline)
+{
+ struct radv_shader_variant *ps, *vs;
+ struct ac_vs_output_info *outinfo;
+
+ ps = pipeline->shaders[MESA_SHADER_FRAGMENT];
+ vs = radv_pipeline_has_gs(pipeline) ? pipeline->gs_copy_shader : (radv_pipeline_has_tess(pipeline) ? pipeline->shaders[MESA_SHADER_TESS_EVAL] : pipeline->shaders[MESA_SHADER_VERTEX]);
+
+ outinfo = &vs->info.vs.outinfo;
+
+ unsigned ps_offset = 0;
+
+ if (ps->info.fs.prim_id_input) {
+ unsigned vs_offset = outinfo->vs_output_param_offset[VARYING_SLOT_PRIMITIVE_ID];
+ if (vs_offset != AC_EXP_PARAM_UNDEFINED) {
+ pipeline->graphics.ps_input_cntl[ps_offset] = offset_to_ps_input(vs_offset, true);
+ ++ps_offset;
+ }
+ }
+
+ if (ps->info.fs.layer_input) {
+ unsigned vs_offset = outinfo->vs_output_param_offset[VARYING_SLOT_LAYER];
+ if (vs_offset != AC_EXP_PARAM_UNDEFINED) {
+ pipeline->graphics.ps_input_cntl[ps_offset] = offset_to_ps_input(vs_offset, true);
+ ++ps_offset;
+ }
+ }
+
+ if (ps->info.fs.has_pcoord) {
+ unsigned val;
+ val = S_028644_PT_SPRITE_TEX(1) | S_028644_OFFSET(0x20);
+ pipeline->graphics.ps_input_cntl[ps_offset] = val;
+ ps_offset++;
+ }
+
+ for (unsigned i = 0; i < 32 && (1u << i) <= ps->info.fs.input_mask; ++i) {
+ unsigned vs_offset;
+ bool flat_shade;
+ if (!(ps->info.fs.input_mask & (1u << i)))
+ continue;
+
+ vs_offset = outinfo->vs_output_param_offset[VARYING_SLOT_VAR0 + i];
+ if (vs_offset == AC_EXP_PARAM_UNDEFINED) {
+ pipeline->graphics.ps_input_cntl[ps_offset] = S_028644_OFFSET(0x20);
+ ++ps_offset;
+ continue;
+ }
+
+ flat_shade = !!(ps->info.fs.flat_shaded_mask & (1u << ps_offset));
+
+ pipeline->graphics.ps_input_cntl[ps_offset] = offset_to_ps_input(vs_offset, flat_shade);
+ ++ps_offset;
+ }
+
+ pipeline->graphics.ps_input_cntl_num = ps_offset;
+}
+