LLVMValueRef vs_prim_id;
LLVMValueRef es2gs_offset;
- LLVMValueRef tcs_offchip_layout;
LLVMValueRef oc_lds;
LLVMValueRef merged_wave_info;
LLVMValueRef tess_factor_offset;
LLVMValueRef tes_u;
LLVMValueRef tes_v;
- LLVMValueRef gsvs_ring_stride;
- LLVMValueRef gsvs_num_entries;
LLVMValueRef gs2vs_offset;
LLVMValueRef gs_wave_id;
LLVMValueRef gs_vtx_offset[6];
unsigned gs_max_out_vertices;
unsigned tes_primitive_mode;
- uint64_t tess_outputs_written;
- uint64_t tess_patch_outputs_written;
uint32_t tcs_patch_outputs_read;
uint64_t tcs_outputs_read;
uint32_t tcs_vertices_per_patch;
uint32_t tcs_num_inputs;
uint32_t tcs_num_patches;
+ uint32_t max_gsvs_emit_size;
+ uint32_t gsvs_vertex_size;
};
enum radeon_llvm_calling_convention {
return num_patches;
}
+static unsigned
+calculate_tess_lds_size(struct radv_shader_context *ctx)
+{
+ unsigned num_tcs_input_cp = ctx->options->key.tcs.input_vertices;
+ unsigned num_tcs_output_cp;
+ unsigned num_tcs_outputs, num_tcs_patch_outputs;
+ unsigned input_vertex_size, output_vertex_size;
+ unsigned input_patch_size, output_patch_size;
+ unsigned pervertex_output_patch_size;
+ unsigned output_patch0_offset;
+ unsigned num_patches;
+ unsigned lds_size;
+
+ num_tcs_output_cp = ctx->tcs_vertices_per_patch;
+ num_tcs_outputs = util_last_bit64(ctx->shader_info->info.tcs.outputs_written);
+ num_tcs_patch_outputs = util_last_bit64(ctx->shader_info->info.tcs.patch_outputs_written);
+
+ input_vertex_size = ctx->tcs_num_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;
+
+ num_patches = ctx->tcs_num_patches;
+ output_patch0_offset = input_patch_size * num_patches;
+
+ lds_size = output_patch0_offset + output_patch_size * num_patches;
+ return lds_size;
+}
+
/* Tessellation shaders pass outputs to the next shader using LDS.
*
* LS outputs = TCS inputs
}
break;
case MESA_SHADER_TESS_EVAL:
- user_sgpr_info->sgpr_count += 1;
break;
case MESA_SHADER_GEOMETRY:
if (has_previous_stage) {
if (previous_stage == MESA_SHADER_VERTEX) {
user_sgpr_info->sgpr_count += count_vs_user_sgprs(ctx);
- } else {
- user_sgpr_info->sgpr_count++;
}
}
- user_sgpr_info->sgpr_count += 2;
break;
default:
break;
previous_stage, &user_sgpr_info,
&args, &desc_sets);
- add_arg(&args, ARG_SGPR, ctx->ac.i32, &ctx->tcs_offchip_layout);
if (needs_view_index)
add_arg(&args, ARG_SGPR, ctx->ac.i32,
&ctx->abi.view_index);
&user_sgpr_info, &args,
&desc_sets);
- if (previous_stage == MESA_SHADER_TESS_EVAL) {
- add_arg(&args, ARG_SGPR, ctx->ac.i32,
- &ctx->tcs_offchip_layout);
- } else {
+ if (previous_stage != MESA_SHADER_TESS_EVAL) {
declare_vs_specific_input_sgprs(ctx, stage,
has_previous_stage,
previous_stage,
&args);
}
- add_arg(&args, ARG_SGPR, ctx->ac.i32,
- &ctx->gsvs_ring_stride);
- add_arg(&args, ARG_SGPR, ctx->ac.i32,
- &ctx->gsvs_num_entries);
if (needs_view_index)
add_arg(&args, ARG_SGPR, ctx->ac.i32,
&ctx->abi.view_index);
&user_sgpr_info, &args,
&desc_sets);
- add_arg(&args, ARG_SGPR, ctx->ac.i32,
- &ctx->gsvs_ring_stride);
- add_arg(&args, ARG_SGPR, ctx->ac.i32,
- &ctx->gsvs_num_entries);
if (needs_view_index)
add_arg(&args, ARG_SGPR, ctx->ac.i32,
&ctx->abi.view_index);
set_loc_shader(ctx, AC_UD_VIEW_INDEX, &user_sgpr_idx, 1);
break;
case MESA_SHADER_TESS_EVAL:
- set_loc_shader(ctx, AC_UD_TES_OFFCHIP_LAYOUT, &user_sgpr_idx, 1);
if (ctx->abi.view_index)
set_loc_shader(ctx, AC_UD_VIEW_INDEX, &user_sgpr_idx, 1);
break;
has_previous_stage,
previous_stage,
&user_sgpr_idx);
- else
- set_loc_shader(ctx, AC_UD_TES_OFFCHIP_LAYOUT,
- &user_sgpr_idx, 1);
}
- set_loc_shader(ctx, AC_UD_GS_VS_RING_STRIDE_ENTRIES,
- &user_sgpr_idx, 2);
if (ctx->abi.view_index)
set_loc_shader(ctx, AC_UD_VIEW_INDEX, &user_sgpr_idx, 1);
break;
*/
static LLVMValueRef get_non_vertex_index_offset(struct radv_shader_context *ctx)
{
- if (ctx->stage == MESA_SHADER_TESS_CTRL) {
- uint32_t num_tcs_outputs = util_last_bit64(ctx->shader_info->info.tcs.outputs_written);
- uint32_t output_vertex_size = num_tcs_outputs * 16;
- uint32_t pervertex_output_patch_size = ctx->tcs_vertices_per_patch * output_vertex_size;
- uint32_t num_patches = ctx->tcs_num_patches;
+ uint32_t num_patches = ctx->tcs_num_patches;
+ uint32_t num_tcs_outputs;
+ if (ctx->stage == MESA_SHADER_TESS_CTRL)
+ num_tcs_outputs = util_last_bit64(ctx->shader_info->info.tcs.outputs_written);
+ else
+ num_tcs_outputs = ctx->options->key.tes.tcs_num_outputs;
- return LLVMConstInt(ctx->ac.i32, pervertex_output_patch_size * num_patches, false);
- } else
- return ac_unpack_param(&ctx->ac, ctx->tcs_offchip_layout, 16, 16);
+ uint32_t output_vertex_size = num_tcs_outputs * 16;
+ uint32_t pervertex_output_patch_size = ctx->tcs_vertices_per_patch * output_vertex_size;
+
+ return LLVMConstInt(ctx->ac.i32, pervertex_output_patch_size * num_patches, false);
}
static LLVMValueRef calc_param_stride(struct radv_shader_context *ctx,
LLVMValueRef vertex_index)
{
LLVMValueRef param_stride;
- if (ctx->stage == MESA_SHADER_TESS_CTRL) {
- if (vertex_index)
- param_stride = LLVMConstInt(ctx->ac.i32, ctx->tcs_vertices_per_patch * ctx->tcs_num_patches, false);
- else
- param_stride = LLVMConstInt(ctx->ac.i32, ctx->tcs_num_patches, false);
- } else {
- LLVMValueRef num_patches = ac_unpack_param(&ctx->ac, ctx->tcs_offchip_layout, 0, 9);
- LLVMValueRef vertices_per_patch = LLVMConstInt(ctx->ac.i32, ctx->tcs_vertices_per_patch, false);
- if (vertex_index)
- param_stride = LLVMBuildMul(ctx->ac.builder, vertices_per_patch,
- num_patches, "");
- else
- param_stride = num_patches;
- }
+ if (vertex_index)
+ param_stride = LLVMConstInt(ctx->ac.i32, ctx->tcs_vertices_per_patch * ctx->tcs_num_patches, false);
+ else
+ param_stride = LLVMConstInt(ctx->ac.i32, ctx->tcs_num_patches, false);
return param_stride;
}
return get_tcs_tes_buffer_address(ctx, vertex_index, param_index);
}
-static void
-mark_tess_output(struct radv_shader_context *ctx,
- bool is_patch, uint32_t param, int num_slots)
-
-{
- uint64_t slot_mask = (1ull << num_slots) - 1;
- if (is_patch) {
- ctx->tess_patch_outputs_written |= (slot_mask << param);
- } else
- ctx->tess_outputs_written |= (slot_mask << param);
-}
-
static LLVMValueRef
get_dw_address(struct radv_shader_context *ctx,
LLVMValueRef dw_addr,
const unsigned component = var->data.location_frac;
const bool is_patch = var->data.patch;
const bool is_compact = var->data.compact;
- const unsigned count = glsl_count_attribute_slots(var->type, false);
LLVMValueRef dw_addr;
LLVMValueRef stride = NULL;
LLVMValueRef buf_addr = NULL;
dw_addr = get_tcs_out_current_patch_data_offset(ctx);
}
- if (param_index)
- mark_tess_output(ctx, is_patch, param, count);
- else
- mark_tess_output(ctx, is_patch, param, 1);
-
dw_addr = get_dw_address(ctx, dw_addr, param, const_index, is_compact, vertex_index, stride,
param_index);
buf_addr = get_tcs_tes_buffer_address_params(ctx, param, const_index, is_compact,
static void
handle_vs_outputs_post(struct radv_shader_context *ctx,
- bool export_prim_id,
+ bool export_prim_id, bool export_layer_id,
struct radv_vs_output_info *outinfo)
{
uint32_t param_count = 0;
for (unsigned j = 1; j < 4; j++)
values[j] = ctx->ac.f32_0;
- radv_export_param(ctx, param_count, values, 0xf);
+ radv_export_param(ctx, param_count, values, 0x1);
outinfo->vs_output_param_offset[VARYING_SLOT_PRIMITIVE_ID] = param_count++;
outinfo->export_prim_id = true;
}
+ if (export_layer_id && layer_value) {
+ LLVMValueRef values[4];
+
+ values[0] = layer_value;
+ for (unsigned j = 1; j < 4; j++)
+ values[j] = ctx->ac.f32_0;
+
+ radv_export_param(ctx, param_count, values, 0x1);
+
+ outinfo->vs_output_param_offset[VARYING_SLOT_LAYER] = param_count++;
+ }
+
outinfo->pos_exports = num_pos_exports;
outinfo->param_exports = param_count;
}
if (i == VARYING_SLOT_CLIP_DIST0)
length = ctx->num_output_clips + ctx->num_output_culls;
int param = shader_io_get_unique_index(i);
- mark_tess_output(ctx, false, param, 1);
- if (length > 4)
- mark_tess_output(ctx, false, param + 1, 1);
LLVMValueRef dw_addr = LLVMBuildAdd(ctx->ac.builder, base_dw_addr,
LLVMConstInt(ctx->ac.i32, param * 4, false),
"");
if (inner_comps) {
tess_inner_index = shader_io_get_unique_index(VARYING_SLOT_TESS_LEVEL_INNER);
- mark_tess_output(ctx, true, tess_inner_index, 1);
lds_inner = LLVMBuildAdd(ctx->ac.builder, lds_base,
LLVMConstInt(ctx->ac.i32, tess_inner_index * 4, false), "");
}
tess_outer_index = shader_io_get_unique_index(VARYING_SLOT_TESS_LEVEL_OUTER);
- mark_tess_output(ctx, true, tess_outer_index, 1);
lds_outer = LLVMBuildAdd(ctx->ac.builder, lds_base,
LLVMConstInt(ctx->ac.i32, tess_outer_index * 4, false), "");
handle_es_outputs_post(ctx, &ctx->shader_info->vs.es_info);
else
handle_vs_outputs_post(ctx, ctx->options->key.vs.export_prim_id,
+ ctx->options->key.vs.export_layer_id,
&ctx->shader_info->vs.outinfo);
break;
case MESA_SHADER_FRAGMENT:
handle_es_outputs_post(ctx, &ctx->shader_info->tes.es_info);
else
handle_vs_outputs_post(ctx, ctx->options->key.tes.export_prim_id,
+ ctx->options->key.tes.export_layer_id,
&ctx->shader_info->tes.outinfo);
break;
default:
}
if (ctx->stage == MESA_SHADER_GEOMETRY) {
LLVMValueRef tmp;
+ uint32_t num_entries = 64;
+ LLVMValueRef gsvs_ring_stride = LLVMConstInt(ctx->ac.i32, ctx->max_gsvs_emit_size, false);
+ LLVMValueRef gsvs_ring_desc = LLVMConstInt(ctx->ac.i32, ctx->max_gsvs_emit_size << 16, false);
ctx->esgs_ring = ac_build_load_to_sgpr(&ctx->ac, ctx->ring_offsets, LLVMConstInt(ctx->ac.i32, RING_ESGS_GS, false));
ctx->gsvs_ring = ac_build_load_to_sgpr(&ctx->ac, ctx->ring_offsets, LLVMConstInt(ctx->ac.i32, RING_GSVS_GS, false));
ctx->gsvs_ring = LLVMBuildBitCast(ctx->ac.builder, ctx->gsvs_ring, ctx->ac.v4i32, "");
- ctx->gsvs_ring = LLVMBuildInsertElement(ctx->ac.builder, ctx->gsvs_ring, ctx->gsvs_num_entries, LLVMConstInt(ctx->ac.i32, 2, false), "");
+ tmp = LLVMConstInt(ctx->ac.i32, num_entries, false);
+ if (ctx->options->chip_class >= VI)
+ tmp = LLVMBuildMul(ctx->ac.builder, gsvs_ring_stride, tmp, "");
+ ctx->gsvs_ring = LLVMBuildInsertElement(ctx->ac.builder, ctx->gsvs_ring, tmp, LLVMConstInt(ctx->ac.i32, 2, false), "");
tmp = LLVMBuildExtractElement(ctx->ac.builder, ctx->gsvs_ring, ctx->ac.i32_1, "");
- tmp = LLVMBuildOr(ctx->ac.builder, tmp, ctx->gsvs_ring_stride, "");
+ tmp = LLVMBuildOr(ctx->ac.builder, tmp, gsvs_ring_desc, "");
ctx->gsvs_ring = LLVMBuildInsertElement(ctx->ac.builder, ctx->gsvs_ring, tmp, ctx->ac.i32_1, "");
}
for(int i = 0; i < shader_count; ++i) {
ctx.stage = shaders[i]->info.stage;
ctx.output_mask = 0;
- ctx.tess_outputs_written = 0;
ctx.num_output_clips = shaders[i]->info.clip_distance_array_size;
ctx.num_output_culls = shaders[i]->info.cull_distance_array_size;
ctx.abi.load_tess_coord = load_tess_coord;
ctx.abi.load_patch_vertices_in = load_patch_vertices_in;
ctx.tcs_vertices_per_patch = shaders[i]->info.tess.tcs_vertices_out;
+ ctx.tcs_num_patches = ctx.options->key.tes.num_patches;
} else if (shaders[i]->info.stage == MESA_SHADER_VERTEX) {
if (shader_info->info.vs.needs_instance_id) {
if (ctx.options->key.vs.as_ls) {
if (i)
ac_emit_barrier(&ctx.ac, ctx.stage);
+ nir_foreach_variable(variable, &shaders[i]->outputs)
+ scan_shader_output_decl(&ctx, variable, shaders[i], shaders[i]->info.stage);
+
+ if (shaders[i]->info.stage == MESA_SHADER_GEOMETRY) {
+ unsigned addclip = shaders[i]->info.clip_distance_array_size +
+ shaders[i]->info.cull_distance_array_size > 4;
+ ctx.gsvs_vertex_size = (util_bitcount64(ctx.output_mask) + addclip) * 16;
+ ctx.max_gsvs_emit_size = ctx.gsvs_vertex_size *
+ shaders[i]->info.gs.vertices_out;
+ }
+
ac_setup_rings(&ctx);
LLVMBasicBlockRef merge_block;
else if(shader_count >= 2 && shaders[i]->info.stage == MESA_SHADER_GEOMETRY)
prepare_gs_input_vgprs(&ctx);
- nir_foreach_variable(variable, &shaders[i]->outputs)
- scan_shader_output_decl(&ctx, variable, shaders[i], shaders[i]->info.stage);
-
ac_nir_translate(&ctx.ac, &ctx.abi, shaders[i]);
if (shader_count >= 2) {
}
if (shaders[i]->info.stage == MESA_SHADER_GEOMETRY) {
- unsigned addclip = shaders[i]->info.clip_distance_array_size +
- shaders[i]->info.cull_distance_array_size > 4;
- shader_info->gs.gsvs_vertex_size = (util_bitcount64(ctx.output_mask) + addclip) * 16;
- shader_info->gs.max_gsvs_emit_size = shader_info->gs.gsvs_vertex_size *
- shaders[i]->info.gs.vertices_out;
+ shader_info->gs.gsvs_vertex_size = ctx.gsvs_vertex_size;
+ shader_info->gs.max_gsvs_emit_size = ctx.max_gsvs_emit_size;
} else if (shaders[i]->info.stage == MESA_SHADER_TESS_CTRL) {
- shader_info->tcs.outputs_written = ctx.tess_outputs_written;
- shader_info->tcs.patch_outputs_written = ctx.tess_patch_outputs_written;
- assert(ctx.tess_outputs_written == ctx.shader_info->info.tcs.outputs_written);
- assert(ctx.tess_patch_outputs_written == ctx.shader_info->info.tcs.patch_outputs_written);
- } else if (shaders[i]->info.stage == MESA_SHADER_VERTEX && ctx.options->key.vs.as_ls) {
- shader_info->vs.outputs_written = ctx.tess_outputs_written;
- assert(ctx.tess_outputs_written == ctx.shader_info->info.vs.ls_outputs_written);
+ shader_info->tcs.num_patches = ctx.tcs_num_patches;
+ shader_info->tcs.lds_size = calculate_tess_lds_size(&ctx);
}
}
}
idx += slot_inc;
}
- handle_vs_outputs_post(ctx, false, &ctx->shader_info->vs.outinfo);
+ handle_vs_outputs_post(ctx, false, false, &ctx->shader_info->vs.outinfo);
}
void
projects / mesa.git / blobdiff