/* * Copyright (C) 2019 Collabora, Ltd. * * Permission is hereby granted, free of charge, to any person obtaining a * copy of this software and associated documentation files (the "Software"), * to deal in the Software without restriction, including without limitation * the rights to use, copy, modify, merge, publish, distribute, sublicense, * and/or sell copies of the Software, and to permit persons to whom the * Software is furnished to do so, subject to the following conditions: * * The above copyright notice and this permission notice (including the next * paragraph) shall be included in all copies or substantial portions of the * Software. * * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE * SOFTWARE. * * Authors (Collabora): * Alyssa Rosenzweig */ #include "compiler.h" /* Midgard's generic load/store instructions, particularly to implement SSBOs * and globals, have support for address arithmetic natively. In particularly, * they take two indirect arguments A, B and two immediates #s, #c, calculating * the address: * * A + (zext?(B) << #s) + #c * * This allows for fast indexing into arrays. This file tries to pattern match * the offset in NIR with this form to reduce pressure on the ALU pipe. */ struct mir_address { nir_scalar A; nir_scalar B; midgard_index_address_format type; unsigned shift; unsigned bias; }; static bool mir_args_ssa(nir_scalar s, unsigned count) { nir_alu_instr *alu = nir_instr_as_alu(s.def->parent_instr); if (count > nir_op_infos[alu->op].num_inputs) return false; return true; } /* Matches a constant in either slot and moves it to the bias */ static void mir_match_constant(struct mir_address *address) { if (address->A.def && nir_scalar_is_const(address->A)) { address->bias += nir_scalar_as_uint(address->A); address->A.def = NULL; } if (address->B.def && nir_scalar_is_const(address->B)) { address->bias += nir_scalar_as_uint(address->B); address->B.def = NULL; } } /* Matches an iadd when there is a free slot or constant */ /* The offset field is a 18-bit signed integer */ #define MAX_POSITIVE_OFFSET ((1 << 17) - 1) static void mir_match_iadd(struct mir_address *address, bool first_free) { if (!address->B.def || !nir_scalar_is_alu(address->B)) return; if (!mir_args_ssa(address->B, 2)) return; nir_op op = nir_scalar_alu_op(address->B); if (op != nir_op_iadd) return; nir_scalar op1 = nir_scalar_chase_alu_src(address->B, 0); nir_scalar op2 = nir_scalar_chase_alu_src(address->B, 1); if (nir_scalar_is_const(op1) && nir_scalar_as_uint(op1) <= MAX_POSITIVE_OFFSET) { address->bias += nir_scalar_as_uint(op1); address->B = op2; } else if (nir_scalar_is_const(op2) && nir_scalar_as_uint(op2) <= MAX_POSITIVE_OFFSET) { address->bias += nir_scalar_as_uint(op2); address->B = op1; } else if (!nir_scalar_is_const(op1) && !nir_scalar_is_const(op2) && first_free && !address->A.def) { address->A = op1; address->B = op2; } } /* Matches u2u64 and sets type */ static void mir_match_u2u64(struct mir_address *address) { if (!address->B.def || !nir_scalar_is_alu(address->B)) return; if (!mir_args_ssa(address->B, 1)) return; nir_op op = nir_scalar_alu_op(address->B); if (op != nir_op_u2u64) return; nir_scalar arg = nir_scalar_chase_alu_src(address->B, 0); address->B = arg; address->type = midgard_index_address_u32; } /* Matches i2i64 and sets type */ static void mir_match_i2i64(struct mir_address *address) { if (!address->B.def || !nir_scalar_is_alu(address->B)) return; if (!mir_args_ssa(address->B, 1)) return; nir_op op = nir_scalar_alu_op(address->B); if (op != nir_op_i2i64) return; nir_scalar arg = nir_scalar_chase_alu_src(address->B, 0); address->B = arg; address->type = midgard_index_address_s32; } /* Matches ishl to shift */ static void mir_match_ishl(struct mir_address *address) { if (!address->B.def || !nir_scalar_is_alu(address->B)) return; if (!mir_args_ssa(address->B, 2)) return; nir_op op = nir_scalar_alu_op(address->B); if (op != nir_op_ishl) return; nir_scalar op1 = nir_scalar_chase_alu_src(address->B, 0); nir_scalar op2 = nir_scalar_chase_alu_src(address->B, 1); if (!nir_scalar_is_const(op2)) return; unsigned shift = nir_scalar_as_uint(op2); if (shift > 0x7) return; address->B = op1; address->shift = shift; } /* Strings through mov which can happen from NIR vectorization */ static void mir_match_mov(struct mir_address *address) { if (address->A.def && nir_scalar_is_alu(address->A)) { nir_op op = nir_scalar_alu_op(address->A); if (op == nir_op_mov && mir_args_ssa(address->A, 1)) address->A = nir_scalar_chase_alu_src(address->A, 0); } if (address->B.def && nir_scalar_is_alu(address->B)) { nir_op op = nir_scalar_alu_op(address->B); if (op == nir_op_mov && mir_args_ssa(address->B, 1)) address->B = nir_scalar_chase_alu_src(address->B, 0); } } /* Tries to pattern match into mir_address */ static struct mir_address mir_match_offset(nir_def *offset, bool first_free, bool extend) { struct mir_address address = { .B = {.def = offset}, .type = extend ? midgard_index_address_u64 : midgard_index_address_u32, }; mir_match_mov(&address); mir_match_constant(&address); mir_match_mov(&address); mir_match_iadd(&address, first_free); mir_match_mov(&address); if (extend) { mir_match_u2u64(&address); mir_match_i2i64(&address); mir_match_mov(&address); } mir_match_ishl(&address); return address; } void mir_set_offset(compiler_context *ctx, midgard_instruction *ins, nir_src *offset, unsigned seg) { for (unsigned i = 0; i < 16; ++i) { ins->swizzle[1][i] = 0; ins->swizzle[2][i] = 0; } /* Sign extend instead of zero extend in case the address is something * like `base + offset + 20`, where offset could be negative. */ bool force_sext = (nir_src_bit_size(*offset) < 64); bool first_free = (seg == LDST_GLOBAL); struct mir_address match = mir_match_offset(offset->ssa, first_free, true); if (match.A.def) { unsigned bitsize = match.A.def->bit_size; assert(bitsize == 32 || bitsize == 64); ins->src[1] = nir_ssa_index(match.A.def); ins->swizzle[1][0] = match.A.comp; ins->src_types[1] = nir_type_uint | bitsize; ins->load_store.bitsize_toggle = (bitsize == 64); } else { ins->load_store.bitsize_toggle = true; ins->load_store.arg_comp = seg & 0x3; ins->load_store.arg_reg = (seg >> 2) & 0x7; } if (match.B.def) { ins->src[2] = nir_ssa_index(match.B.def); ins->swizzle[2][0] = match.B.comp; ins->src_types[2] = nir_type_uint | match.B.def->bit_size; } else ins->load_store.index_reg = REGISTER_LDST_ZERO; if (force_sext) match.type = midgard_index_address_s32; ins->load_store.index_format = match.type; assert(match.shift <= 7); ins->load_store.index_shift = match.shift; ins->constants.u32[0] = match.bias; } void mir_set_ubo_offset(midgard_instruction *ins, nir_src *src, unsigned bias) { struct mir_address match = mir_match_offset(src->ssa, false, false); if (match.B.def) { ins->src[2] = nir_ssa_index(match.B.def); for (unsigned i = 0; i < ARRAY_SIZE(ins->swizzle[2]); ++i) ins->swizzle[2][i] = match.B.comp; } ins->load_store.index_shift = match.shift; ins->constants.u32[0] = match.bias + bias; }