1/* 2 * Copyright © 2016 Intel Corporation 3 * 4 * Permission is hereby granted, free of charge, to any person obtaining a 5 * copy of this software and associated documentation files (the "Software"), 6 * to deal in the Software without restriction, including without limitation 7 * the rights to use, copy, modify, merge, publish, distribute, sublicense, 8 * and/or sell copies of the Software, and to permit persons to whom the 9 * Software is furnished to do so, subject to the following conditions: 10 * 11 * The above copyright notice and this permission notice (including the next 12 * paragraph) shall be included in all copies or substantial portions of the 13 * Software. 14 * 15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING 20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS 21 * IN THE SOFTWARE. 22 */ 23 24#include "nir.h" 25#include "nir_builder.h" 26 27static void 28build_constant_load(nir_builder *b, nir_deref_instr *deref, nir_constant *c) 29{ 30 if (glsl_type_is_vector_or_scalar(deref->type)) { 31 nir_load_const_instr *load = 32 nir_load_const_instr_create(b->shader, 33 glsl_get_vector_elements(deref->type), 34 glsl_get_bit_size(deref->type)); 35 memcpy(load->value, c->values, sizeof(*load->value) * load->def.num_components); 36 nir_builder_instr_insert(b, &load->instr); 37 nir_store_deref(b, deref, &load->def, ~0); 38 } else if (glsl_type_is_struct_or_ifc(deref->type)) { 39 unsigned len = glsl_get_length(deref->type); 40 for (unsigned i = 0; i < len; i++) { 41 build_constant_load(b, nir_build_deref_struct(b, deref, i), 42 c->elements[i]); 43 } 44 } else { 45 assert(glsl_type_is_array(deref->type) || 46 glsl_type_is_matrix(deref->type)); 47 unsigned len = glsl_get_length(deref->type); 48 for (unsigned i = 0; i < len; i++) { 49 build_constant_load(b, 50 nir_build_deref_array_imm(b, deref, i), 51 c->elements[i]); 52 } 53 } 54} 55 56static bool 57lower_const_initializer(struct nir_builder *b, struct exec_list *var_list, 58 nir_variable_mode modes) 59{ 60 bool progress = false; 61 62 b->cursor = nir_before_cf_list(&b->impl->body); 63 64 nir_foreach_variable_in_list(var, var_list) { 65 if (!(var->data.mode & modes)) 66 continue; 67 68 if (var->constant_initializer) { 69 build_constant_load(b, nir_build_deref_var(b, var), 70 var->constant_initializer); 71 72 progress = true; 73 var->constant_initializer = NULL; 74 } else if (var->pointer_initializer) { 75 nir_deref_instr *src_deref = nir_build_deref_var(b, var->pointer_initializer); 76 nir_deref_instr *dst_deref = nir_build_deref_var(b, var); 77 78 /* Note that this stores a pointer to src into dst */ 79 nir_store_deref(b, dst_deref, &src_deref->dest.ssa, ~0); 80 81 progress = true; 82 var->pointer_initializer = NULL; 83 } 84 85 } 86 87 return progress; 88} 89 90bool 91nir_lower_variable_initializers(nir_shader *shader, nir_variable_mode modes) 92{ 93 bool progress = false; 94 95 /* Only some variables have initializers that we want to lower. Others 96 * such as uniforms have initializers which are useful later during linking 97 * so we want to skip over those. Restrict to only variable types where 98 * initializers make sense so that callers can use nir_var_all. 99 */ 100 modes &= nir_var_shader_out | 101 nir_var_shader_temp | 102 nir_var_function_temp | 103 nir_var_system_value; 104 105 nir_foreach_function(function, shader) { 106 if (!function->impl) 107 continue; 108 109 bool impl_progress = false; 110 111 nir_builder builder; 112 nir_builder_init(&builder, function->impl); 113 114 if ((modes & ~nir_var_function_temp) && function->is_entrypoint) { 115 impl_progress |= lower_const_initializer(&builder, 116 &shader->variables, 117 modes); 118 } 119 120 if (modes & nir_var_function_temp) { 121 impl_progress |= lower_const_initializer(&builder, 122 &function->impl->locals, 123 nir_var_function_temp); 124 } 125 126 if (impl_progress) { 127 progress = true; 128 nir_metadata_preserve(function->impl, nir_metadata_block_index | 129 nir_metadata_dominance | 130 nir_metadata_live_ssa_defs); 131 } else { 132 nir_metadata_preserve(function->impl, nir_metadata_all); 133 } 134 } 135 136 return progress; 137} 138 139/* Zero initialize shared_size bytes of shared memory by splitting work writes 140 * of chunk_size bytes among the invocations. 141 * 142 * Used for implementing VK_KHR_zero_initialize_workgroup_memory. 143 */ 144bool 145nir_zero_initialize_shared_memory(nir_shader *shader, 146 const unsigned shared_size, 147 const unsigned chunk_size) 148{ 149 assert(shared_size > 0); 150 assert(chunk_size > 0); 151 assert(chunk_size % 4 == 0); 152 153 nir_builder b; 154 nir_builder_init(&b, nir_shader_get_entrypoint(shader)); 155 b.cursor = nir_before_cf_list(&b.impl->body); 156 157 assert(!shader->info.workgroup_size_variable); 158 const unsigned local_count = shader->info.workgroup_size[0] * 159 shader->info.workgroup_size[1] * 160 shader->info.workgroup_size[2]; 161 162 /* The initialization logic is simplified if we can always split the memory 163 * in full chunk_size units. 164 */ 165 assert(shared_size % chunk_size == 0); 166 167 const unsigned chunk_comps = chunk_size / 4; 168 169 nir_variable *it = nir_local_variable_create(b.impl, glsl_uint_type(), 170 "zero_init_iterator"); 171 nir_ssa_def *local_index = nir_load_local_invocation_index(&b); 172 nir_ssa_def *first_offset = nir_imul_imm(&b, local_index, chunk_size); 173 nir_store_var(&b, it, first_offset, 0x1); 174 175 nir_loop *loop = nir_push_loop(&b); 176 { 177 nir_ssa_def *offset = nir_load_var(&b, it); 178 179 nir_push_if(&b, nir_uge(&b, offset, nir_imm_int(&b, shared_size))); 180 { 181 nir_jump(&b, nir_jump_break); 182 } 183 nir_pop_if(&b, NULL); 184 185 nir_store_shared(&b, nir_imm_zero(&b, chunk_comps, 32), offset, 186 .align_mul=chunk_size, 187 .write_mask=((1 << chunk_comps) - 1)); 188 189 nir_ssa_def *new_offset = nir_iadd_imm(&b, offset, chunk_size * local_count); 190 nir_store_var(&b, it, new_offset, 0x1); 191 } 192 nir_pop_loop(&b, loop); 193 194 nir_scoped_barrier(&b, NIR_SCOPE_WORKGROUP, NIR_SCOPE_WORKGROUP, 195 NIR_MEMORY_ACQ_REL, nir_var_mem_shared); 196 197 nir_metadata_preserve(nir_shader_get_entrypoint(shader), nir_metadata_none); 198 199 return true; 200} 201