diff options
| -rw-r--r-- | src/glsl/nir/glsl_to_nir.cpp | 29 | ||||
| -rw-r--r-- | src/glsl/nir/nir_intrinsics.h | 1 |
2 files changed, 30 insertions, 0 deletions
diff --git a/src/glsl/nir/glsl_to_nir.cpp b/src/glsl/nir/glsl_to_nir.cpp index 45d045cd4d6..38f980465b6 100644 --- a/src/glsl/nir/glsl_to_nir.cpp +++ b/src/glsl/nir/glsl_to_nir.cpp @@ -727,6 +727,8 @@ nir_visitor::visit(ir_call *ir) op = nir_intrinsic_memory_barrier_image; } else if (strcmp(ir->callee_name(), "__intrinsic_memory_barrier_shared") == 0) { op = nir_intrinsic_memory_barrier_shared; + } else if (strcmp(ir->callee_name(), "__intrinsic_load_shared") == 0) { + op = nir_intrinsic_load_shared; } else { unreachable("not reached"); } @@ -974,6 +976,33 @@ nir_visitor::visit(ir_call *ir) nir_builder_instr_insert(&b, &instr->instr); break; } + case nir_intrinsic_load_shared: { + exec_node *param = ir->actual_parameters.get_head(); + ir_rvalue *offset = ((ir_instruction *)param)->as_rvalue(); + + /* Check if we need the indirect version */ + ir_constant *const_offset = offset->as_constant(); + if (!const_offset) { + op = nir_intrinsic_load_shared_indirect; + ralloc_free(instr); + instr = nir_intrinsic_instr_create(shader, op); + instr->src[0] = nir_src_for_ssa(evaluate_rvalue(offset)); + instr->const_index[0] = 0; + dest = &instr->dest; + } else { + instr->const_index[0] = const_offset->value.u[0]; + } + + const glsl_type *type = ir->return_deref->var->type; + instr->num_components = type->vector_elements; + + /* Setup destination register */ + nir_ssa_dest_init(&instr->instr, &instr->dest, + type->vector_elements, NULL); + + nir_builder_instr_insert(&b, &instr->instr); + break; + } default: unreachable("not reached"); } diff --git a/src/glsl/nir/nir_intrinsics.h b/src/glsl/nir/nir_intrinsics.h index b2565c54b20..011b95db9d6 100644 --- a/src/glsl/nir/nir_intrinsics.h +++ b/src/glsl/nir/nir_intrinsics.h @@ -258,6 +258,7 @@ LOAD(per_vertex_input, 1, 1, NIR_INTRINSIC_CAN_ELIMINATE | NIR_INTRINSIC_CAN_REO LOAD(ssbo, 1, 1, NIR_INTRINSIC_CAN_ELIMINATE) LOAD(output, 0, 1, NIR_INTRINSIC_CAN_ELIMINATE) LOAD(per_vertex_output, 1, 1, NIR_INTRINSIC_CAN_ELIMINATE) +LOAD(shared, 0, 1, NIR_INTRINSIC_CAN_ELIMINATE) /* * Stores work the same way as loads, except now the first register input is |