src/gallium/drivers/r600/sfn/sfn_shader_geometry.cpp

   1 /* -*- mesa-c++  -*-
   2  *
   3  * Copyright (c) 2018 Collabora LTD
   4  *
   5  * Author: Gert Wollny <gert.wollny@collabora.com>
   6  *
   7  * Permission is hereby granted, free of charge, to any person obtaining a
   8  * copy of this software and associated documentation files (the "Software"),
   9  * to deal in the Software without restriction, including without limitation
  10  * on the rights to use, copy, modify, merge, publish, distribute, sub
  11  * license, and/or sell copies of the Software, and to permit persons to whom
  12  * the Software is furnished to do so, subject to the following conditions:
  13  *
  14  * The above copyright notice and this permission notice (including the next
  15  * paragraph) shall be included in all copies or substantial portions of the
  16  * Software.
  17  *
  18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  19  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  20  * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
  21  * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
  22  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
  23  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
  24  * USE OR OTHER DEALINGS IN THE SOFTWARE.
  25  */
  26
  27 #include "sfn_shader_geometry.h"
  28 #include "sfn_instruction_misc.h"
  29 #include "sfn_instruction_fetch.h"
  30 #include "sfn_shaderio.h"
  31
  32 namespace r600 {
  33
  34 GeometryShaderFromNir::GeometryShaderFromNir(r600_pipe_shader *sh,
  35                                              r600_pipe_shader_selector &sel,
  36                                              const r600_shader_key &key,
  37                                              enum chip_class chip_class):
  38    VertexStage(PIPE_SHADER_GEOMETRY, sel, sh->shader,
  39                sh->scratch_space_needed, chip_class),
  40    m_pipe_shader(sh),
  41    m_so_info(&sel.so),
  42    m_first_vertex_emitted(false),
  43    m_offset(0),
  44    m_next_input_ring_offset(0),
  45    m_key(key),
  46    m_num_clip_dist(0),
  47    m_cur_ring_output(0),
  48    m_gs_tri_strip_adj_fix(false)
  49 {
  50    sh_info().atomic_base = key.gs.first_atomic_counter;
  51 }
  52
  53 bool GeometryShaderFromNir::do_emit_load_deref(UNUSED const nir_variable *in_var, UNUSED nir_intrinsic_instr* instr)
  54 {
  55    return false;
  56 }
  57
  58 bool GeometryShaderFromNir::do_emit_store_deref(const nir_variable *out_var, nir_intrinsic_instr* instr)
  59 {
  60    uint32_t write_mask = nir_intrinsic_write_mask(instr);
  61    GPRVector::Swizzle swz = swizzle_from_mask(write_mask);
  62    auto out_value = vec_from_nir_with_fetch_constant(instr->src[1], write_mask, swz, true);
  63
  64    sh_info().output[out_var->data.driver_location].write_mask = write_mask;
  65
  66    auto ir = new MemRingOutIntruction(cf_mem_ring, mem_write_ind, out_value,
  67                                       4 * out_var->data.driver_location,
  68                                       instr->num_components, m_export_base);
  69    emit_instruction(ir);
  70
  71    return true;
  72 }
  73
  74 bool GeometryShaderFromNir::scan_sysvalue_access(UNUSED nir_instr *instr)
  75 {
  76    return true;
  77 }
  78
  79 bool GeometryShaderFromNir::do_process_inputs(nir_variable *input)
  80 {
  81
  82    if (input->data.location == VARYING_SLOT_POS ||
  83        input->data.location == VARYING_SLOT_PSIZ ||
  84        input->data.location == VARYING_SLOT_CLIP_VERTEX ||
  85        input->data.location == VARYING_SLOT_CLIP_DIST0 ||
  86        input->data.location == VARYING_SLOT_CLIP_DIST1 ||
  87        input->data.location == VARYING_SLOT_COL0 ||
  88        input->data.location == VARYING_SLOT_COL1 ||
  89        (input->data.location >= VARYING_SLOT_VAR0 &&
  90        input->data.location <= VARYING_SLOT_VAR31) ||
  91        (input->data.location >= VARYING_SLOT_TEX0 &&
  92        input->data.location <= VARYING_SLOT_TEX7)) {
  93
  94       r600_shader_io& io = sh_info().input[input->data.driver_location];
  95       auto semantic = r600_get_varying_semantic(input->data.location);
  96       io.name = semantic.first;
  97       io.sid = semantic.second;
  98
  99       io.ring_offset = 16 * input->data.driver_location;
 100       ++sh_info().ninput;
 101       m_next_input_ring_offset += 16;
 102       return true;
 103    }
 104
 105    return false;
 106 }
 107
 108 bool GeometryShaderFromNir::do_process_outputs(nir_variable *output)
 109 {
 110    if (output->data.location == VARYING_SLOT_COL0 ||
 111        output->data.location == VARYING_SLOT_COL1 ||
 112        (output->data.location >= VARYING_SLOT_VAR0 &&
 113        output->data.location <= VARYING_SLOT_VAR31) ||
 114        (output->data.location >= VARYING_SLOT_TEX0 &&
 115        output->data.location <= VARYING_SLOT_TEX7) ||
 116        output->data.location == VARYING_SLOT_BFC0 ||
 117        output->data.location == VARYING_SLOT_BFC1 ||
 118        output->data.location == VARYING_SLOT_CLIP_VERTEX ||
 119        output->data.location == VARYING_SLOT_CLIP_DIST0 ||
 120        output->data.location == VARYING_SLOT_CLIP_DIST1 ||
 121        output->data.location == VARYING_SLOT_PRIMITIVE_ID ||
 122        output->data.location == VARYING_SLOT_POS ||
 123        output->data.location == VARYING_SLOT_PSIZ ||
 124        output->data.location == VARYING_SLOT_LAYER ||
 125        output->data.location == VARYING_SLOT_VIEWPORT ||
 126        output->data.location == VARYING_SLOT_FOGC) {
 127       r600_shader_io& io = sh_info().output[output->data.driver_location];
 128
 129       auto semantic = r600_get_varying_semantic(output->data.location);
 130       io.name = semantic.first;
 131       io.sid = semantic.second;
 132
 133       evaluate_spi_sid(io);
 134       ++sh_info().noutput;
 135
 136       if (output->data.location == VARYING_SLOT_CLIP_DIST0 ||
 137           output->data.location == VARYING_SLOT_CLIP_DIST1) {
 138          m_num_clip_dist += 4;
 139       }
 140
 141       if (output->data.location == VARYING_SLOT_VIEWPORT) {
 142          sh_info().vs_out_viewport = 1;
 143          sh_info().vs_out_misc_write = 1;
 144       }
 145       return true;
 146    }
 147    return false;
 148 }
 149
 150
 151 bool GeometryShaderFromNir::do_allocate_reserved_registers()
 152 {
 153    const int sel[6] = {0, 0 ,0, 1, 1, 1};
 154    const int chan[6] = {0, 1 ,3, 0, 1, 2};
 155
 156    increment_reserved_registers();
 157    increment_reserved_registers();
 158
 159    /* Reserve registers used by the shaders (should check how many
 160     * components are actually used */
 161    for (int i = 0; i < 6; ++i) {
 162       auto reg = new GPRValue(sel[i], chan[i]);
 163       reg->set_as_input();
 164       m_per_vertex_offsets[i].reset(reg);
 165       inject_register(sel[i], chan[i], m_per_vertex_offsets[i], false);
 166    }
 167    auto reg = new GPRValue(0, 2);
 168    reg->set_as_input();
 169    m_primitive_id.reset(reg);
 170    inject_register(0, 2, m_primitive_id, false);
 171
 172    reg = new GPRValue(1, 3);
 173    reg->set_as_input();
 174    m_invocation_id.reset(reg);
 175    inject_register(1, 3, m_invocation_id, false);
 176
 177    m_export_base = get_temp_register();
 178    emit_instruction(new AluInstruction(op1_mov, m_export_base, Value::zero, {alu_write, alu_last_instr}));
 179
 180    sh_info().ring_item_sizes[0] = m_next_input_ring_offset;
 181
 182    if (m_key.gs.tri_strip_adj_fix)
 183       emit_adj_fix();
 184
 185    return true;
 186 }
 187
 188 void GeometryShaderFromNir::emit_adj_fix()
 189 {
 190    PValue adjhelp0(new  GPRValue(m_export_base->sel(), 1));
 191    emit_instruction(op2_and_int, adjhelp0, {m_primitive_id, Value::one_i}, {alu_write, alu_last_instr});
 192
 193    int help2 = allocate_temp_register();
 194    int reg_indices[6];
 195    int reg_chanels[6] = {0, 1, 2, 3, 2, 3};
 196
 197    int rotate_indices[6] = {4, 5, 0, 1, 2, 3};
 198
 199    reg_indices[0] = reg_indices[1] = reg_indices[2] = reg_indices[3] = help2;
 200    reg_indices[4] = reg_indices[5] = m_export_base->sel();
 201
 202    std::array<PValue, 6> adjhelp;
 203
 204    AluInstruction *ir = nullptr;
 205    for (int i = 0; i < 6; i++) {
 206       adjhelp[i].reset(new GPRValue(reg_indices[i], reg_chanels[i]));
 207       ir = new AluInstruction(op3_cnde_int, adjhelp[i],
 208                              {adjhelp0, m_per_vertex_offsets[i],
 209                               m_per_vertex_offsets[rotate_indices[i]]},
 210                              {alu_write});
 211       if (i == 3)
 212          ir->set_flag(alu_last_instr);
 213       emit_instruction(ir);
 214    }
 215    ir->set_flag(alu_last_instr);
 216
 217    for (int i = 0; i < 6; i++)
 218       m_per_vertex_offsets[i] = adjhelp[i];
 219 }
 220
 221 bool GeometryShaderFromNir::emit_deref_instruction_override(nir_deref_instr* instr)
 222 {
 223    if (instr->deref_type  == nir_deref_type_array) {
 224       auto var = get_deref_location(instr->parent);
 225       ArrayDeref ad = {var, &instr->arr.index};
 226       assert(instr->dest.is_ssa);
 227       m_in_array_deref[instr->dest.ssa.index] = ad;
 228
 229       /* Problem: nir_intrinsice_load_deref tries to lookup the
 230        * variable, and will not find it, need to override that too */
 231       return true;
 232    }
 233    return false;
 234 }
 235
 236 bool GeometryShaderFromNir::emit_intrinsic_instruction_override(nir_intrinsic_instr* instr)
 237 {
 238    switch (instr->intrinsic) {
 239    case nir_intrinsic_load_deref: {
 240       auto& src = instr->src[0];
 241       assert(src.is_ssa);
 242       auto array = m_in_array_deref.find(src.ssa->index);
 243       if (array != m_in_array_deref.end())
 244          return emit_load_from_array(instr, array->second);
 245    } break;
 246    case nir_intrinsic_emit_vertex:
 247       return emit_vertex(instr, false);
 248    case nir_intrinsic_end_primitive:
 249       return emit_vertex(instr, true);
 250    case nir_intrinsic_load_primitive_id:
 251       return load_preloaded_value(instr->dest, 0, m_primitive_id);
 252    case nir_intrinsic_load_invocation_id:
 253       return load_preloaded_value(instr->dest, 0, m_invocation_id);
 254    default:
 255       ;
 256    }
 257    return false;
 258 }
 259
 260 bool GeometryShaderFromNir::emit_vertex(nir_intrinsic_instr* instr, bool cut)
 261 {
 262    int stream = nir_intrinsic_stream_id(instr);
 263    assert(stream < 4);
 264
 265    emit_instruction(new EmitVertex(stream, cut));
 266
 267    if (!cut)
 268       emit_instruction(new AluInstruction(op2_add_int, m_export_base, m_export_base,
 269                                           PValue(new LiteralValue(sh_info().noutput)),
 270                                           {alu_write, alu_last_instr}));
 271
 272    return true;
 273 }
 274
 275 bool GeometryShaderFromNir::emit_load_from_array(nir_intrinsic_instr* instr,
 276                                                  const ArrayDeref& array_deref)
 277 {
 278    auto dest = vec_from_nir(instr->dest, instr->num_components);
 279
 280    const nir_load_const_instr* literal_index = nullptr;
 281
 282    if (array_deref.index->is_ssa)
 283       literal_index = get_literal_constant(array_deref.index->ssa->index);
 284
 285    if (!literal_index) {
 286       sfn_log << SfnLog::err << "GS: Indirect input addressing not (yet) supported\n";
 287       return false;
 288    }
 289    assert(literal_index->value[0].u32 < 6);
 290    PValue addr = m_per_vertex_offsets[literal_index->value[0].u32];
 291
 292    auto fetch = new FetchInstruction(vc_fetch, no_index_offset, dest, addr,
 293                                      16 * array_deref.var->data.driver_location,
 294                                      R600_GS_RING_CONST_BUFFER, PValue(), bim_none, true);
 295    emit_instruction(fetch);
 296    return true;
 297 }
 298
 299 void GeometryShaderFromNir::do_finalize()
 300 {
 301    if (m_num_clip_dist) {
 302       sh_info().cc_dist_mask = (1 << m_num_clip_dist) - 1;
 303       sh_info().clip_dist_write = (1 << m_num_clip_dist) - 1;
 304    }
 305 }
 306
 307 }