1bf215546Sopenharmony_ci/* 2bf215546Sopenharmony_ci * Copyright (C) 2022 Collabora Ltd. 3bf215546Sopenharmony_ci * 4bf215546Sopenharmony_ci * Permission is hereby granted, free of charge, to any person obtaining a 5bf215546Sopenharmony_ci * copy of this software and associated documentation files (the "Software"), 6bf215546Sopenharmony_ci * to deal in the Software without restriction, including without limitation 7bf215546Sopenharmony_ci * the rights to use, copy, modify, merge, publish, distribute, sublicense, 8bf215546Sopenharmony_ci * and/or sell copies of the Software, and to permit persons to whom the 9bf215546Sopenharmony_ci * Software is furnished to do so, subject to the following conditions: 10bf215546Sopenharmony_ci * 11bf215546Sopenharmony_ci * The above copyright notice and this permission notice (including the next 12bf215546Sopenharmony_ci * paragraph) shall be included in all copies or substantial portions of the 13bf215546Sopenharmony_ci * Software. 14bf215546Sopenharmony_ci * 15bf215546Sopenharmony_ci * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 16bf215546Sopenharmony_ci * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 17bf215546Sopenharmony_ci * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 18bf215546Sopenharmony_ci * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 19bf215546Sopenharmony_ci * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 20bf215546Sopenharmony_ci * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 21bf215546Sopenharmony_ci * SOFTWARE. 22bf215546Sopenharmony_ci */ 23bf215546Sopenharmony_ci 24bf215546Sopenharmony_ci 25bf215546Sopenharmony_ci#include "pan_ir.h" 26bf215546Sopenharmony_ci#include "compiler/nir/nir_builder.h" 27bf215546Sopenharmony_ci 28bf215546Sopenharmony_cistatic void 29bf215546Sopenharmony_cilower_xfb_output(nir_builder *b, nir_intrinsic_instr *intr, 30bf215546Sopenharmony_ci unsigned start_component, unsigned num_components, 31bf215546Sopenharmony_ci unsigned buffer, unsigned offset_words) 32bf215546Sopenharmony_ci{ 33bf215546Sopenharmony_ci assert(buffer < MAX_XFB_BUFFERS); 34bf215546Sopenharmony_ci assert(nir_intrinsic_component(intr) == 0); // TODO 35bf215546Sopenharmony_ci 36bf215546Sopenharmony_ci /* Transform feedback info in units of words, convert to bytes. */ 37bf215546Sopenharmony_ci uint16_t stride = b->shader->info.xfb_stride[buffer] * 4; 38bf215546Sopenharmony_ci assert(stride != 0); 39bf215546Sopenharmony_ci 40bf215546Sopenharmony_ci uint16_t offset = offset_words * 4; 41bf215546Sopenharmony_ci 42bf215546Sopenharmony_ci nir_ssa_def *index = nir_iadd(b, 43bf215546Sopenharmony_ci nir_imul(b, nir_load_instance_id(b), 44bf215546Sopenharmony_ci nir_load_num_vertices(b)), 45bf215546Sopenharmony_ci nir_load_vertex_id_zero_base(b)); 46bf215546Sopenharmony_ci 47bf215546Sopenharmony_ci BITSET_SET(b->shader->info.system_values_read, SYSTEM_VALUE_VERTEX_ID_ZERO_BASE); 48bf215546Sopenharmony_ci BITSET_SET(b->shader->info.system_values_read, SYSTEM_VALUE_INSTANCE_ID); 49bf215546Sopenharmony_ci 50bf215546Sopenharmony_ci nir_ssa_def *buf = nir_load_xfb_address(b, 64, .base = buffer); 51bf215546Sopenharmony_ci nir_ssa_def *addr = 52bf215546Sopenharmony_ci nir_iadd(b, buf, nir_u2u64(b, 53bf215546Sopenharmony_ci nir_iadd_imm(b, 54bf215546Sopenharmony_ci nir_imul_imm(b, index, stride), 55bf215546Sopenharmony_ci offset))); 56bf215546Sopenharmony_ci 57bf215546Sopenharmony_ci assert(intr->src[0].is_ssa && "must lower XFB before lowering SSA"); 58bf215546Sopenharmony_ci nir_ssa_def *src = intr->src[0].ssa; 59bf215546Sopenharmony_ci nir_ssa_def *value = nir_channels(b, src, BITFIELD_MASK(num_components) << start_component); 60bf215546Sopenharmony_ci nir_store_global(b, addr, 4, value, BITFIELD_MASK(num_components)); 61bf215546Sopenharmony_ci} 62bf215546Sopenharmony_ci 63bf215546Sopenharmony_cistatic bool 64bf215546Sopenharmony_cilower_xfb(nir_builder *b, nir_instr *instr, UNUSED void *data) 65bf215546Sopenharmony_ci{ 66bf215546Sopenharmony_ci if (instr->type != nir_instr_type_intrinsic) 67bf215546Sopenharmony_ci return false; 68bf215546Sopenharmony_ci 69bf215546Sopenharmony_ci nir_intrinsic_instr *intr = nir_instr_as_intrinsic(instr); 70bf215546Sopenharmony_ci if (intr->intrinsic != nir_intrinsic_store_output) 71bf215546Sopenharmony_ci return false; 72bf215546Sopenharmony_ci 73bf215546Sopenharmony_ci bool progress = false; 74bf215546Sopenharmony_ci 75bf215546Sopenharmony_ci b->cursor = nir_before_instr(&intr->instr); 76bf215546Sopenharmony_ci 77bf215546Sopenharmony_ci for (unsigned i = 0; i < 2; ++i) { 78bf215546Sopenharmony_ci nir_io_xfb xfb = i ? nir_intrinsic_io_xfb2(intr) : nir_intrinsic_io_xfb(intr); 79bf215546Sopenharmony_ci for (unsigned j = 0; j < 2; ++j) { 80bf215546Sopenharmony_ci if (!xfb.out[j].num_components) continue; 81bf215546Sopenharmony_ci 82bf215546Sopenharmony_ci lower_xfb_output(b, intr, i*2 + j, 83bf215546Sopenharmony_ci xfb.out[j].num_components, 84bf215546Sopenharmony_ci xfb.out[j].buffer, 85bf215546Sopenharmony_ci xfb.out[j].offset); 86bf215546Sopenharmony_ci progress = true; 87bf215546Sopenharmony_ci } 88bf215546Sopenharmony_ci } 89bf215546Sopenharmony_ci 90bf215546Sopenharmony_ci nir_instr_remove(instr); 91bf215546Sopenharmony_ci return progress; 92bf215546Sopenharmony_ci} 93bf215546Sopenharmony_ci 94bf215546Sopenharmony_cibool 95bf215546Sopenharmony_cipan_lower_xfb(nir_shader *nir) 96bf215546Sopenharmony_ci{ 97bf215546Sopenharmony_ci return nir_shader_instructions_pass(nir, lower_xfb, 98bf215546Sopenharmony_ci nir_metadata_block_index | 99bf215546Sopenharmony_ci nir_metadata_dominance, NULL); 100bf215546Sopenharmony_ci} 101bf215546Sopenharmony_ci 102