/* * Copyright (C) 2018-2019 Alyssa Rosenzweig * * 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. */ #include #include #include #include #include #include #include #include #include "main/mtypes.h" #include "compiler/glsl/glsl_to_nir.h" #include "compiler/nir_types.h" #include "util/imports.h" #include "compiler/nir/nir_builder.h" #include "util/half_float.h" #include "util/u_math.h" #include "util/u_debug.h" #include "util/u_dynarray.h" #include "util/list.h" #include "main/mtypes.h" #include "midgard.h" #include "midgard_nir.h" #include "midgard_compile.h" #include "midgard_ops.h" #include "helpers.h" #include "compiler.h" #include "midgard_quirks.h" #include "disassemble.h" static const struct debug_named_value debug_options[] = { {"msgs", MIDGARD_DBG_MSGS, "Print debug messages"}, {"shaders", MIDGARD_DBG_SHADERS, "Dump shaders in NIR and MIR"}, {"shaderdb", MIDGARD_DBG_SHADERDB, "Prints shader-db statistics"}, DEBUG_NAMED_VALUE_END }; DEBUG_GET_ONCE_FLAGS_OPTION(midgard_debug, "MIDGARD_MESA_DEBUG", debug_options, 0) unsigned SHADER_DB_COUNT = 0; int midgard_debug = 0; #define DBG(fmt, ...) \ do { if (midgard_debug & MIDGARD_DBG_MSGS) \ fprintf(stderr, "%s:%d: "fmt, \ __FUNCTION__, __LINE__, ##__VA_ARGS__); } while (0) static midgard_block * create_empty_block(compiler_context *ctx) { midgard_block *blk = rzalloc(ctx, midgard_block); blk->base.predecessors = _mesa_set_create(blk, _mesa_hash_pointer, _mesa_key_pointer_equal); blk->base.name = ctx->block_source_count++; return blk; } static void schedule_barrier(compiler_context *ctx) { midgard_block *temp = ctx->after_block; ctx->after_block = create_empty_block(ctx); ctx->block_count++; list_addtail(&ctx->after_block->base.link, &ctx->blocks); list_inithead(&ctx->after_block->base.instructions); pan_block_add_successor(&ctx->current_block->base, &ctx->after_block->base); ctx->current_block = ctx->after_block; ctx->after_block = temp; } /* Helpers to generate midgard_instruction's using macro magic, since every * driver seems to do it that way */ #define EMIT(op, ...) emit_mir_instruction(ctx, v_##op(__VA_ARGS__)); #define M_LOAD_STORE(name, store) \ static midgard_instruction m_##name(unsigned ssa, unsigned address) { \ midgard_instruction i = { \ .type = TAG_LOAD_STORE_4, \ .mask = 0xF, \ .dest = ~0, \ .src = { ~0, ~0, ~0, ~0 }, \ .swizzle = SWIZZLE_IDENTITY_4, \ .load_store = { \ .op = midgard_op_##name, \ .address = address \ } \ }; \ \ if (store) \ i.src[0] = ssa; \ else \ i.dest = ssa; \ \ return i; \ } #define M_LOAD(name) M_LOAD_STORE(name, false) #define M_STORE(name) M_LOAD_STORE(name, true) /* Inputs a NIR ALU source, with modifiers attached if necessary, and outputs * the corresponding Midgard source */ static midgard_vector_alu_src vector_alu_modifiers(nir_alu_src *src, bool is_int, unsigned broadcast_count, bool half, bool sext) { /* Figure out how many components there are so we can adjust. * Specifically we want to broadcast the last channel so things like * ball2/3 work. */ if (broadcast_count && src) { uint8_t last_component = src->swizzle[broadcast_count - 1]; for (unsigned c = broadcast_count; c < NIR_MAX_VEC_COMPONENTS; ++c) { src->swizzle[c] = last_component; } } midgard_vector_alu_src alu_src = { .rep_low = 0, .rep_high = 0, .half = half }; if (is_int) { alu_src.mod = midgard_int_normal; /* Sign/zero-extend if needed */ if (half) { alu_src.mod = sext ? midgard_int_sign_extend : midgard_int_zero_extend; } /* These should have been lowered away */ if (src) assert(!(src->abs || src->negate)); } else { if (src) alu_src.mod = (src->abs << 0) | (src->negate << 1); } return alu_src; } /* load/store instructions have both 32-bit and 16-bit variants, depending on * whether we are using vectors composed of highp or mediump. At the moment, we * don't support half-floats -- this requires changes in other parts of the * compiler -- therefore the 16-bit versions are commented out. */ //M_LOAD(ld_attr_16); M_LOAD(ld_attr_32); //M_LOAD(ld_vary_16); M_LOAD(ld_vary_32); M_LOAD(ld_ubo_int4); M_LOAD(ld_int4); M_STORE(st_int4); M_LOAD(ld_color_buffer_32u); //M_STORE(st_vary_16); M_STORE(st_vary_32); M_LOAD(ld_cubemap_coords); M_LOAD(ld_compute_id); static midgard_instruction v_branch(bool conditional, bool invert) { midgard_instruction ins = { .type = TAG_ALU_4, .unit = ALU_ENAB_BRANCH, .compact_branch = true, .branch = { .conditional = conditional, .invert_conditional = invert }, .dest = ~0, .src = { ~0, ~0, ~0, ~0 }, }; return ins; } static midgard_branch_extended midgard_create_branch_extended( midgard_condition cond, midgard_jmp_writeout_op op, unsigned dest_tag, signed quadword_offset) { /* The condition code is actually a LUT describing a function to * combine multiple condition codes. However, we only support a single * condition code at the moment, so we just duplicate over a bunch of * times. */ uint16_t duplicated_cond = (cond << 14) | (cond << 12) | (cond << 10) | (cond << 8) | (cond << 6) | (cond << 4) | (cond << 2) | (cond << 0); midgard_branch_extended branch = { .op = op, .dest_tag = dest_tag, .offset = quadword_offset, .cond = duplicated_cond }; return branch; } static void attach_constants(compiler_context *ctx, midgard_instruction *ins, void *constants, int name) { ins->has_constants = true; memcpy(&ins->constants, constants, 16); } static int glsl_type_size(const struct glsl_type *type, bool bindless) { return glsl_count_attribute_slots(type, false); } /* Lower fdot2 to a vector multiplication followed by channel addition */ static void midgard_nir_lower_fdot2_body(nir_builder *b, nir_alu_instr *alu) { if (alu->op != nir_op_fdot2) return; b->cursor = nir_before_instr(&alu->instr); nir_ssa_def *src0 = nir_ssa_for_alu_src(b, alu, 0); nir_ssa_def *src1 = nir_ssa_for_alu_src(b, alu, 1); nir_ssa_def *product = nir_fmul(b, src0, src1); nir_ssa_def *sum = nir_fadd(b, nir_channel(b, product, 0), nir_channel(b, product, 1)); /* Replace the fdot2 with this sum */ nir_ssa_def_rewrite_uses(&alu->dest.dest.ssa, nir_src_for_ssa(sum)); } static bool midgard_nir_lower_fdot2(nir_shader *shader) { bool progress = false; nir_foreach_function(function, shader) { if (!function->impl) continue; nir_builder _b; nir_builder *b = &_b; nir_builder_init(b, function->impl); nir_foreach_block(block, function->impl) { nir_foreach_instr_safe(instr, block) { if (instr->type != nir_instr_type_alu) continue; nir_alu_instr *alu = nir_instr_as_alu(instr); midgard_nir_lower_fdot2_body(b, alu); progress |= true; } } nir_metadata_preserve(function->impl, nir_metadata_block_index | nir_metadata_dominance); } return progress; } /* Midgard can't write depth and stencil separately. It has to happen in a * single store operation containing both. Let's add a panfrost specific * intrinsic and turn all depth/stencil stores into a packed depth+stencil * one. */ static bool midgard_nir_lower_zs_store(nir_shader *nir) { if (nir->info.stage != MESA_SHADER_FRAGMENT) return false; nir_variable *z_var = NULL, *s_var = NULL; nir_foreach_variable(var, &nir->outputs) { if (var->data.location == FRAG_RESULT_DEPTH) z_var = var; else if (var->data.location == FRAG_RESULT_STENCIL) s_var = var; } if (!z_var && !s_var) return false; bool progress = false; nir_foreach_function(function, nir) { if (!function->impl) continue; nir_intrinsic_instr *z_store = NULL, *s_store = NULL, *last_store = NULL; nir_foreach_block(block, function->impl) { nir_foreach_instr_safe(instr, block) { if (instr->type != nir_instr_type_intrinsic) continue; nir_intrinsic_instr *intr = nir_instr_as_intrinsic(instr); if (intr->intrinsic != nir_intrinsic_store_output) continue; if (z_var && nir_intrinsic_base(intr) == z_var->data.driver_location) { assert(!z_store); z_store = intr; last_store = intr; } if (s_var && nir_intrinsic_base(intr) == s_var->data.driver_location) { assert(!s_store); s_store = intr; last_store = intr; } } } if (!z_store && !s_store) continue; nir_builder b; nir_builder_init(&b, function->impl); b.cursor = nir_before_instr(&last_store->instr); nir_ssa_def *zs_store_src; if (z_store && s_store) { nir_ssa_def *srcs[2] = { nir_ssa_for_src(&b, z_store->src[0], 1), nir_ssa_for_src(&b, s_store->src[0], 1), }; zs_store_src = nir_vec(&b, srcs, 2); } else { zs_store_src = nir_ssa_for_src(&b, last_store->src[0], 1); } nir_intrinsic_instr *zs_store; zs_store = nir_intrinsic_instr_create(b.shader, nir_intrinsic_store_zs_output_pan); zs_store->src[0] = nir_src_for_ssa(zs_store_src); zs_store->num_components = z_store && s_store ? 2 : 1; nir_intrinsic_set_component(zs_store, z_store ? 0 : 1); /* Replace the Z and S store by a ZS store */ nir_builder_instr_insert(&b, &zs_store->instr); if (z_store) nir_instr_remove(&z_store->instr); if (s_store) nir_instr_remove(&s_store->instr); nir_metadata_preserve(function->impl, nir_metadata_block_index | nir_metadata_dominance); progress = true; } return progress; } /* Flushes undefined values to zero */ static void optimise_nir(nir_shader *nir, unsigned quirks) { bool progress; unsigned lower_flrp = (nir->options->lower_flrp16 ? 16 : 0) | (nir->options->lower_flrp32 ? 32 : 0) | (nir->options->lower_flrp64 ? 64 : 0); NIR_PASS(progress, nir, nir_lower_regs_to_ssa); NIR_PASS(progress, nir, nir_lower_idiv, nir_lower_idiv_fast); nir_lower_tex_options lower_tex_options = { .lower_txs_lod = true, .lower_txp = ~0, .lower_tex_without_implicit_lod = (quirks & MIDGARD_EXPLICIT_LOD), /* TODO: we have native gradient.. */ .lower_txd = true, }; NIR_PASS(progress, nir, nir_lower_tex, &lower_tex_options); /* Must lower fdot2 after tex is lowered */ NIR_PASS(progress, nir, midgard_nir_lower_fdot2); /* T720 is broken. */ if (quirks & MIDGARD_BROKEN_LOD) NIR_PASS_V(nir, midgard_nir_lod_errata); do { progress = false; NIR_PASS(progress, nir, nir_lower_var_copies); NIR_PASS(progress, nir, nir_lower_vars_to_ssa); NIR_PASS(progress, nir, nir_copy_prop); NIR_PASS(progress, nir, nir_opt_remove_phis); NIR_PASS(progress, nir, nir_opt_dce); NIR_PASS(progress, nir, nir_opt_dead_cf); NIR_PASS(progress, nir, nir_opt_cse); NIR_PASS(progress, nir, nir_opt_peephole_select, 64, false, true); NIR_PASS(progress, nir, nir_opt_algebraic); NIR_PASS(progress, nir, nir_opt_constant_folding); if (lower_flrp != 0) { bool lower_flrp_progress = false; NIR_PASS(lower_flrp_progress, nir, nir_lower_flrp, lower_flrp, false /* always_precise */, nir->options->lower_ffma); if (lower_flrp_progress) { NIR_PASS(progress, nir, nir_opt_constant_folding); progress = true; } /* Nothing should rematerialize any flrps, so we only * need to do this lowering once. */ lower_flrp = 0; } NIR_PASS(progress, nir, nir_opt_undef); NIR_PASS(progress, nir, nir_undef_to_zero); NIR_PASS(progress, nir, nir_opt_loop_unroll, nir_var_shader_in | nir_var_shader_out | nir_var_function_temp); NIR_PASS(progress, nir, nir_opt_vectorize); } while (progress); /* Must be run at the end to prevent creation of fsin/fcos ops */ NIR_PASS(progress, nir, midgard_nir_scale_trig); do { progress = false; NIR_PASS(progress, nir, nir_opt_dce); NIR_PASS(progress, nir, nir_opt_algebraic); NIR_PASS(progress, nir, nir_opt_constant_folding); NIR_PASS(progress, nir, nir_copy_prop); } while (progress); NIR_PASS(progress, nir, nir_opt_algebraic_late); /* We implement booleans as 32-bit 0/~0 */ NIR_PASS(progress, nir, nir_lower_bool_to_int32); /* Now that booleans are lowered, we can run out late opts */ NIR_PASS(progress, nir, midgard_nir_lower_algebraic_late); /* Lower mods for float ops only. Integer ops don't support modifiers * (saturate doesn't make sense on integers, neg/abs require dedicated * instructions) */ NIR_PASS(progress, nir, nir_lower_to_source_mods, nir_lower_float_source_mods); NIR_PASS(progress, nir, nir_copy_prop); NIR_PASS(progress, nir, nir_opt_dce); /* Take us out of SSA */ NIR_PASS(progress, nir, nir_lower_locals_to_regs); NIR_PASS(progress, nir, nir_convert_from_ssa, true); /* We are a vector architecture; write combine where possible */ NIR_PASS(progress, nir, nir_move_vec_src_uses_to_dest); NIR_PASS(progress, nir, nir_lower_vec_to_movs); NIR_PASS(progress, nir, nir_opt_dce); } /* Do not actually emit a load; instead, cache the constant for inlining */ static void emit_load_const(compiler_context *ctx, nir_load_const_instr *instr) { nir_ssa_def def = instr->def; midgard_constants *consts = rzalloc(NULL, midgard_constants); assert(instr->def.num_components * instr->def.bit_size <= sizeof(*consts) * 8); #define RAW_CONST_COPY(bits) \ nir_const_value_to_array(consts->u##bits, instr->value, \ instr->def.num_components, u##bits) switch (instr->def.bit_size) { case 64: RAW_CONST_COPY(64); break; case 32: RAW_CONST_COPY(32); break; case 16: RAW_CONST_COPY(16); break; case 8: RAW_CONST_COPY(8); break; default: unreachable("Invalid bit_size for load_const instruction\n"); } /* Shifted for SSA, +1 for off-by-one */ _mesa_hash_table_u64_insert(ctx->ssa_constants, (def.index << 1) + 1, consts); } /* Normally constants are embedded implicitly, but for I/O and such we have to * explicitly emit a move with the constant source */ static void emit_explicit_constant(compiler_context *ctx, unsigned node, unsigned to) { void *constant_value = _mesa_hash_table_u64_search(ctx->ssa_constants, node + 1); if (constant_value) { midgard_instruction ins = v_mov(SSA_FIXED_REGISTER(REGISTER_CONSTANT), to); attach_constants(ctx, &ins, constant_value, node + 1); emit_mir_instruction(ctx, ins); } } static bool nir_is_non_scalar_swizzle(nir_alu_src *src, unsigned nr_components) { unsigned comp = src->swizzle[0]; for (unsigned c = 1; c < nr_components; ++c) { if (src->swizzle[c] != comp) return true; } return false; } #define ALU_CASE(nir, _op) \ case nir_op_##nir: \ op = midgard_alu_op_##_op; \ assert(src_bitsize == dst_bitsize); \ break; #define ALU_CASE_BCAST(nir, _op, count) \ case nir_op_##nir: \ op = midgard_alu_op_##_op; \ broadcast_swizzle = count; \ assert(src_bitsize == dst_bitsize); \ break; static bool nir_is_fzero_constant(nir_src src) { if (!nir_src_is_const(src)) return false; for (unsigned c = 0; c < nir_src_num_components(src); ++c) { if (nir_src_comp_as_float(src, c) != 0.0) return false; } return true; } /* Analyze the sizes of the inputs to determine which reg mode. Ops needed * special treatment override this anyway. */ static midgard_reg_mode reg_mode_for_nir(nir_alu_instr *instr) { unsigned src_bitsize = nir_src_bit_size(instr->src[0].src); switch (src_bitsize) { case 8: return midgard_reg_mode_8; case 16: return midgard_reg_mode_16; case 32: return midgard_reg_mode_32; case 64: return midgard_reg_mode_64; default: unreachable("Invalid bit size"); } } static void emit_alu(compiler_context *ctx, nir_alu_instr *instr) { /* Derivatives end up emitted on the texture pipe, not the ALUs. This * is handled elsewhere */ if (instr->op == nir_op_fddx || instr->op == nir_op_fddy) { midgard_emit_derivatives(ctx, instr); return; } bool is_ssa = instr->dest.dest.is_ssa; unsigned dest = nir_dest_index(&instr->dest.dest); unsigned nr_components = nir_dest_num_components(instr->dest.dest); unsigned nr_inputs = nir_op_infos[instr->op].num_inputs; /* Most Midgard ALU ops have a 1:1 correspondance to NIR ops; these are * supported. A few do not and are commented for now. Also, there are a * number of NIR ops which Midgard does not support and need to be * lowered, also TODO. This switch block emits the opcode and calling * convention of the Midgard instruction; actual packing is done in * emit_alu below */ unsigned op; /* Number of components valid to check for the instruction (the rest * will be forced to the last), or 0 to use as-is. Relevant as * ball-type instructions have a channel count in NIR but are all vec4 * in Midgard */ unsigned broadcast_swizzle = 0; /* What register mode should we operate in? */ midgard_reg_mode reg_mode = reg_mode_for_nir(instr); /* Do we need a destination override? Used for inline * type conversion */ midgard_dest_override dest_override = midgard_dest_override_none; /* Should we use a smaller respective source and sign-extend? */ bool half_1 = false, sext_1 = false; bool half_2 = false, sext_2 = false; unsigned src_bitsize = nir_src_bit_size(instr->src[0].src); unsigned dst_bitsize = nir_dest_bit_size(instr->dest.dest); switch (instr->op) { ALU_CASE(fadd, fadd); ALU_CASE(fmul, fmul); ALU_CASE(fmin, fmin); ALU_CASE(fmax, fmax); ALU_CASE(imin, imin); ALU_CASE(imax, imax); ALU_CASE(umin, umin); ALU_CASE(umax, umax); ALU_CASE(ffloor, ffloor); ALU_CASE(fround_even, froundeven); ALU_CASE(ftrunc, ftrunc); ALU_CASE(fceil, fceil); ALU_CASE(fdot3, fdot3); ALU_CASE(fdot4, fdot4); ALU_CASE(iadd, iadd); ALU_CASE(isub, isub); ALU_CASE(imul, imul); /* Zero shoved as second-arg */ ALU_CASE(iabs, iabsdiff); ALU_CASE(mov, imov); ALU_CASE(feq32, feq); ALU_CASE(fne32, fne); ALU_CASE(flt32, flt); ALU_CASE(ieq32, ieq); ALU_CASE(ine32, ine); ALU_CASE(ilt32, ilt); ALU_CASE(ult32, ult); /* We don't have a native b2f32 instruction. Instead, like many * GPUs, we exploit booleans as 0/~0 for false/true, and * correspondingly AND * by 1.0 to do the type conversion. For the moment, prime us * to emit: * * iand [whatever], #0 * * At the end of emit_alu (as MIR), we'll fix-up the constant */ ALU_CASE(b2f32, iand); ALU_CASE(b2i32, iand); /* Likewise, we don't have a dedicated f2b32 instruction, but * we can do a "not equal to 0.0" test. */ ALU_CASE(f2b32, fne); ALU_CASE(i2b32, ine); ALU_CASE(frcp, frcp); ALU_CASE(frsq, frsqrt); ALU_CASE(fsqrt, fsqrt); ALU_CASE(fexp2, fexp2); ALU_CASE(flog2, flog2); ALU_CASE(f2i64, f2i_rtz); ALU_CASE(f2u64, f2u_rtz); ALU_CASE(i2f64, i2f_rtz); ALU_CASE(u2f64, u2f_rtz); ALU_CASE(f2i32, f2i_rtz); ALU_CASE(f2u32, f2u_rtz); ALU_CASE(i2f32, i2f_rtz); ALU_CASE(u2f32, u2f_rtz); ALU_CASE(f2i16, f2i_rtz); ALU_CASE(f2u16, f2u_rtz); ALU_CASE(i2f16, i2f_rtz); ALU_CASE(u2f16, u2f_rtz); ALU_CASE(fsin, fsin); ALU_CASE(fcos, fcos); /* We'll set invert */ ALU_CASE(inot, imov); ALU_CASE(iand, iand); ALU_CASE(ior, ior); ALU_CASE(ixor, ixor); ALU_CASE(ishl, ishl); ALU_CASE(ishr, iasr); ALU_CASE(ushr, ilsr); ALU_CASE_BCAST(b32all_fequal2, fball_eq, 2); ALU_CASE_BCAST(b32all_fequal3, fball_eq, 3); ALU_CASE(b32all_fequal4, fball_eq); ALU_CASE_BCAST(b32any_fnequal2, fbany_neq, 2); ALU_CASE_BCAST(b32any_fnequal3, fbany_neq, 3); ALU_CASE(b32any_fnequal4, fbany_neq); ALU_CASE_BCAST(b32all_iequal2, iball_eq, 2); ALU_CASE_BCAST(b32all_iequal3, iball_eq, 3); ALU_CASE(b32all_iequal4, iball_eq); ALU_CASE_BCAST(b32any_inequal2, ibany_neq, 2); ALU_CASE_BCAST(b32any_inequal3, ibany_neq, 3); ALU_CASE(b32any_inequal4, ibany_neq); /* Source mods will be shoved in later */ ALU_CASE(fabs, fmov); ALU_CASE(fneg, fmov); ALU_CASE(fsat, fmov); /* For size conversion, we use a move. Ideally though we would squash * these ops together; maybe that has to happen after in NIR as part of * propagation...? An earlier algebraic pass ensured we step down by * only / exactly one size. If stepping down, we use a dest override to * reduce the size; if stepping up, we use a larger-sized move with a * half source and a sign/zero-extension modifier */ case nir_op_i2i8: case nir_op_i2i16: case nir_op_i2i32: case nir_op_i2i64: /* If we end up upscale, we'll need a sign-extend on the * operand (the second argument) */ sext_2 = true; /* fallthrough */ case nir_op_u2u8: case nir_op_u2u16: case nir_op_u2u32: case nir_op_u2u64: case nir_op_f2f16: case nir_op_f2f32: case nir_op_f2f64: { if (instr->op == nir_op_f2f16 || instr->op == nir_op_f2f32 || instr->op == nir_op_f2f64) op = midgard_alu_op_fmov; else op = midgard_alu_op_imov; if (dst_bitsize == (src_bitsize * 2)) { /* Converting up */ half_2 = true; /* Use a greater register mode */ reg_mode++; } else if (src_bitsize == (dst_bitsize * 2)) { /* Converting down */ dest_override = midgard_dest_override_lower; } break; } /* For greater-or-equal, we lower to less-or-equal and flip the * arguments */ case nir_op_fge: case nir_op_fge32: case nir_op_ige32: case nir_op_uge32: { op = instr->op == nir_op_fge ? midgard_alu_op_fle : instr->op == nir_op_fge32 ? midgard_alu_op_fle : instr->op == nir_op_ige32 ? midgard_alu_op_ile : instr->op == nir_op_uge32 ? midgard_alu_op_ule : 0; /* Swap via temporary */ nir_alu_src temp = instr->src[1]; instr->src[1] = instr->src[0]; instr->src[0] = temp; break; } case nir_op_b32csel: { /* Midgard features both fcsel and icsel, depending on * the type of the arguments/output. However, as long * as we're careful we can _always_ use icsel and * _never_ need fcsel, since the latter does additional * floating-point-specific processing whereas the * former just moves bits on the wire. It's not obvious * why these are separate opcodes, save for the ability * to do things like sat/pos/abs/neg for free */ bool mixed = nir_is_non_scalar_swizzle(&instr->src[0], nr_components); op = mixed ? midgard_alu_op_icsel_v : midgard_alu_op_icsel; /* The condition is the first argument; move the other * arguments up one to be a binary instruction for * Midgard with the condition last */ nir_alu_src temp = instr->src[2]; instr->src[2] = instr->src[0]; instr->src[0] = instr->src[1]; instr->src[1] = temp; break; } default: DBG("Unhandled ALU op %s\n", nir_op_infos[instr->op].name); assert(0); return; } /* Midgard can perform certain modifiers on output of an ALU op */ unsigned outmod; if (midgard_is_integer_out_op(op)) { outmod = midgard_outmod_int_wrap; } else { bool sat = instr->dest.saturate || instr->op == nir_op_fsat; outmod = sat ? midgard_outmod_sat : midgard_outmod_none; } /* fmax(a, 0.0) can turn into a .pos modifier as an optimization */ if (instr->op == nir_op_fmax) { if (nir_is_fzero_constant(instr->src[0].src)) { op = midgard_alu_op_fmov; nr_inputs = 1; outmod = midgard_outmod_pos; instr->src[0] = instr->src[1]; } else if (nir_is_fzero_constant(instr->src[1].src)) { op = midgard_alu_op_fmov; nr_inputs = 1; outmod = midgard_outmod_pos; } } /* Fetch unit, quirks, etc information */ unsigned opcode_props = alu_opcode_props[op].props; bool quirk_flipped_r24 = opcode_props & QUIRK_FLIPPED_R24; /* src0 will always exist afaik, but src1 will not for 1-argument * instructions. The latter can only be fetched if the instruction * needs it, or else we may segfault. */ unsigned src0 = nir_alu_src_index(ctx, &instr->src[0]); unsigned src1 = nr_inputs >= 2 ? nir_alu_src_index(ctx, &instr->src[1]) : ~0; unsigned src2 = nr_inputs == 3 ? nir_alu_src_index(ctx, &instr->src[2]) : ~0; assert(nr_inputs <= 3); /* Rather than use the instruction generation helpers, we do it * ourselves here to avoid the mess */ midgard_instruction ins = { .type = TAG_ALU_4, .src = { quirk_flipped_r24 ? ~0 : src0, quirk_flipped_r24 ? src0 : src1, src2, ~0 }, .dest = dest, }; nir_alu_src *nirmods[3] = { NULL }; if (nr_inputs >= 2) { nirmods[0] = &instr->src[0]; nirmods[1] = &instr->src[1]; } else if (nr_inputs == 1) { nirmods[quirk_flipped_r24] = &instr->src[0]; } else { assert(0); } if (nr_inputs == 3) nirmods[2] = &instr->src[2]; /* These were lowered to a move, so apply the corresponding mod */ if (instr->op == nir_op_fneg || instr->op == nir_op_fabs) { nir_alu_src *s = nirmods[quirk_flipped_r24]; if (instr->op == nir_op_fneg) s->negate = !s->negate; if (instr->op == nir_op_fabs) s->abs = !s->abs; } bool is_int = midgard_is_integer_op(op); ins.mask = mask_of(nr_components); midgard_vector_alu alu = { .op = op, .reg_mode = reg_mode, .dest_override = dest_override, .outmod = outmod, .src1 = vector_alu_srco_unsigned(vector_alu_modifiers(nirmods[0], is_int, broadcast_swizzle, half_1, sext_1)), .src2 = vector_alu_srco_unsigned(vector_alu_modifiers(nirmods[1], is_int, broadcast_swizzle, half_2, sext_2)), }; /* Apply writemask if non-SSA, keeping in mind that we can't write to components that don't exist */ if (!is_ssa) ins.mask &= instr->dest.write_mask; for (unsigned m = 0; m < 3; ++m) { if (!nirmods[m]) continue; for (unsigned c = 0; c < NIR_MAX_VEC_COMPONENTS; ++c) ins.swizzle[m][c] = nirmods[m]->swizzle[c]; /* Replicate. TODO: remove when vec16 lands */ for (unsigned c = NIR_MAX_VEC_COMPONENTS; c < MIR_VEC_COMPONENTS; ++c) ins.swizzle[m][c] = nirmods[m]->swizzle[NIR_MAX_VEC_COMPONENTS - 1]; } if (nr_inputs == 3) { /* Conditions can't have mods */ assert(!nirmods[2]->abs); assert(!nirmods[2]->negate); } ins.alu = alu; /* Late fixup for emulated instructions */ if (instr->op == nir_op_b2f32 || instr->op == nir_op_b2i32) { /* Presently, our second argument is an inline #0 constant. * Switch over to an embedded 1.0 constant (that can't fit * inline, since we're 32-bit, not 16-bit like the inline * constants) */ ins.has_inline_constant = false; ins.src[1] = SSA_FIXED_REGISTER(REGISTER_CONSTANT); ins.has_constants = true; if (instr->op == nir_op_b2f32) ins.constants.f32[0] = 1.0f; else ins.constants.i32[0] = 1; for (unsigned c = 0; c < 16; ++c) ins.swizzle[1][c] = 0; } else if (nr_inputs == 1 && !quirk_flipped_r24) { /* Lots of instructions need a 0 plonked in */ ins.has_inline_constant = false; ins.src[1] = SSA_FIXED_REGISTER(REGISTER_CONSTANT); ins.has_constants = true; ins.constants.u32[0] = 0; for (unsigned c = 0; c < 16; ++c) ins.swizzle[1][c] = 0; } else if (instr->op == nir_op_inot) { ins.invert = true; } if ((opcode_props & UNITS_ALL) == UNIT_VLUT) { /* To avoid duplicating the lookup tables (probably), true LUT * instructions can only operate as if they were scalars. Lower * them here by changing the component. */ unsigned orig_mask = ins.mask; for (int i = 0; i < nr_components; ++i) { /* Mask the associated component, dropping the * instruction if needed */ ins.mask = 1 << i; ins.mask &= orig_mask; if (!ins.mask) continue; for (unsigned j = 0; j < MIR_VEC_COMPONENTS; ++j) ins.swizzle[0][j] = nirmods[0]->swizzle[i]; /* Pull from the correct component */ emit_mir_instruction(ctx, ins); } } else { emit_mir_instruction(ctx, ins); } } #undef ALU_CASE static void mir_set_intr_mask(nir_instr *instr, midgard_instruction *ins, bool is_read) { nir_intrinsic_instr *intr = nir_instr_as_intrinsic(instr); unsigned nir_mask = 0; unsigned dsize = 0; if (is_read) { nir_mask = mask_of(nir_intrinsic_dest_components(intr)); dsize = nir_dest_bit_size(intr->dest); } else { nir_mask = nir_intrinsic_write_mask(intr); dsize = 32; } /* Once we have the NIR mask, we need to normalize to work in 32-bit space */ unsigned bytemask = pan_to_bytemask(dsize, nir_mask); mir_set_bytemask(ins, bytemask); if (dsize == 64) ins->load_64 = true; } /* Uniforms and UBOs use a shared code path, as uniforms are just (slightly * optimized) versions of UBO #0 */ static midgard_instruction * emit_ubo_read( compiler_context *ctx, nir_instr *instr, unsigned dest, unsigned offset, nir_src *indirect_offset, unsigned indirect_shift, unsigned index) { /* TODO: half-floats */ midgard_instruction ins = m_ld_ubo_int4(dest, 0); ins.constants.u32[0] = offset; if (instr->type == nir_instr_type_intrinsic) mir_set_intr_mask(instr, &ins, true); if (indirect_offset) { ins.src[2] = nir_src_index(ctx, indirect_offset); ins.load_store.arg_2 = (indirect_shift << 5); } else { ins.load_store.arg_2 = 0x1E; } ins.load_store.arg_1 = index; return emit_mir_instruction(ctx, ins); } /* Globals are like UBOs if you squint. And shared memory is like globals if * you squint even harder */ static void emit_global( compiler_context *ctx, nir_instr *instr, bool is_read, unsigned srcdest, nir_src *offset, bool is_shared) { /* TODO: types */ midgard_instruction ins; if (is_read) ins = m_ld_int4(srcdest, 0); else ins = m_st_int4(srcdest, 0); mir_set_offset(ctx, &ins, offset, is_shared); mir_set_intr_mask(instr, &ins, is_read); emit_mir_instruction(ctx, ins); } static void emit_varying_read( compiler_context *ctx, unsigned dest, unsigned offset, unsigned nr_comp, unsigned component, nir_src *indirect_offset, nir_alu_type type, bool flat) { /* XXX: Half-floats? */ /* TODO: swizzle, mask */ midgard_instruction ins = m_ld_vary_32(dest, offset); ins.mask = mask_of(nr_comp); for (unsigned i = 0; i < ARRAY_SIZE(ins.swizzle[0]); ++i) ins.swizzle[0][i] = MIN2(i + component, COMPONENT_W); midgard_varying_parameter p = { .is_varying = 1, .interpolation = midgard_interp_default, .flat = flat, }; unsigned u; memcpy(&u, &p, sizeof(p)); ins.load_store.varying_parameters = u; if (indirect_offset) ins.src[2] = nir_src_index(ctx, indirect_offset); else ins.load_store.arg_2 = 0x1E; ins.load_store.arg_1 = 0x9E; /* Use the type appropriate load */ switch (type) { case nir_type_uint: case nir_type_bool: ins.load_store.op = midgard_op_ld_vary_32u; break; case nir_type_int: ins.load_store.op = midgard_op_ld_vary_32i; break; case nir_type_float: ins.load_store.op = midgard_op_ld_vary_32; break; default: unreachable("Attempted to load unknown type"); break; } emit_mir_instruction(ctx, ins); } static void emit_attr_read( compiler_context *ctx, unsigned dest, unsigned offset, unsigned nr_comp, nir_alu_type t) { midgard_instruction ins = m_ld_attr_32(dest, offset); ins.load_store.arg_1 = 0x1E; ins.load_store.arg_2 = 0x1E; ins.mask = mask_of(nr_comp); /* Use the type appropriate load */ switch (t) { case nir_type_uint: case nir_type_bool: ins.load_store.op = midgard_op_ld_attr_32u; break; case nir_type_int: ins.load_store.op = midgard_op_ld_attr_32i; break; case nir_type_float: ins.load_store.op = midgard_op_ld_attr_32; break; default: unreachable("Attempted to load unknown type"); break; } emit_mir_instruction(ctx, ins); } static void emit_sysval_read(compiler_context *ctx, nir_instr *instr, unsigned nr_components, unsigned offset) { nir_dest nir_dest; /* Figure out which uniform this is */ int sysval = panfrost_sysval_for_instr(instr, &nir_dest); void *val = _mesa_hash_table_u64_search(ctx->sysvals.sysval_to_id, sysval); unsigned dest = nir_dest_index(&nir_dest); /* Sysvals are prefix uniforms */ unsigned uniform = ((uintptr_t) val) - 1; /* Emit the read itself -- this is never indirect */ midgard_instruction *ins = emit_ubo_read(ctx, instr, dest, (uniform * 16) + offset, NULL, 0, 0); ins->mask = mask_of(nr_components); } static unsigned compute_builtin_arg(nir_op op) { switch (op) { case nir_intrinsic_load_work_group_id: return 0x14; case nir_intrinsic_load_local_invocation_id: return 0x10; default: unreachable("Invalid compute paramater loaded"); } } static void emit_fragment_store(compiler_context *ctx, unsigned src, enum midgard_rt_id rt) { assert(rt < ARRAY_SIZE(ctx->writeout_branch)); midgard_instruction *br = ctx->writeout_branch[rt]; assert(!br); emit_explicit_constant(ctx, src, src); struct midgard_instruction ins = v_branch(false, false); ins.writeout = true; /* Add dependencies */ ins.src[0] = src; ins.constants.u32[0] = rt == MIDGARD_ZS_RT ? 0xFF : (rt - MIDGARD_COLOR_RT0) * 0x100; /* Emit the branch */ br = emit_mir_instruction(ctx, ins); schedule_barrier(ctx); ctx->writeout_branch[rt] = br; /* Push our current location = current block count - 1 = where we'll * jump to. Maybe a bit too clever for my own good */ br->branch.target_block = ctx->block_count - 1; } static void emit_compute_builtin(compiler_context *ctx, nir_intrinsic_instr *instr) { unsigned reg = nir_dest_index(&instr->dest); midgard_instruction ins = m_ld_compute_id(reg, 0); ins.mask = mask_of(3); ins.swizzle[0][3] = COMPONENT_X; /* xyzx */ ins.load_store.arg_1 = compute_builtin_arg(instr->intrinsic); emit_mir_instruction(ctx, ins); } static unsigned vertex_builtin_arg(nir_op op) { switch (op) { case nir_intrinsic_load_vertex_id: return PAN_VERTEX_ID; case nir_intrinsic_load_instance_id: return PAN_INSTANCE_ID; default: unreachable("Invalid vertex builtin"); } } static void emit_vertex_builtin(compiler_context *ctx, nir_intrinsic_instr *instr) { unsigned reg = nir_dest_index(&instr->dest); emit_attr_read(ctx, reg, vertex_builtin_arg(instr->intrinsic), 1, nir_type_int); } static void emit_control_barrier(compiler_context *ctx) { midgard_instruction ins = { .type = TAG_TEXTURE_4, .src = { ~0, ~0, ~0, ~0 }, .texture = { .op = TEXTURE_OP_BARRIER, /* TODO: optimize */ .barrier_buffer = 1, .barrier_shared = 1 } }; emit_mir_instruction(ctx, ins); } static const nir_variable * search_var(struct exec_list *vars, unsigned driver_loc) { nir_foreach_variable(var, vars) { if (var->data.driver_location == driver_loc) return var; } return NULL; } static void emit_intrinsic(compiler_context *ctx, nir_intrinsic_instr *instr) { unsigned offset = 0, reg; switch (instr->intrinsic) { case nir_intrinsic_discard_if: case nir_intrinsic_discard: { bool conditional = instr->intrinsic == nir_intrinsic_discard_if; struct midgard_instruction discard = v_branch(conditional, false); discard.branch.target_type = TARGET_DISCARD; if (conditional) discard.src[0] = nir_src_index(ctx, &instr->src[0]); emit_mir_instruction(ctx, discard); schedule_barrier(ctx); break; } case nir_intrinsic_load_uniform: case nir_intrinsic_load_ubo: case nir_intrinsic_load_global: case nir_intrinsic_load_shared: case nir_intrinsic_load_input: case nir_intrinsic_load_interpolated_input: { bool is_uniform = instr->intrinsic == nir_intrinsic_load_uniform; bool is_ubo = instr->intrinsic == nir_intrinsic_load_ubo; bool is_global = instr->intrinsic == nir_intrinsic_load_global; bool is_shared = instr->intrinsic == nir_intrinsic_load_shared; bool is_flat = instr->intrinsic == nir_intrinsic_load_input; bool is_interp = instr->intrinsic == nir_intrinsic_load_interpolated_input; /* Get the base type of the intrinsic */ /* TODO: Infer type? Does it matter? */ nir_alu_type t = (is_ubo || is_global || is_shared) ? nir_type_uint : (is_interp) ? nir_type_float : nir_intrinsic_type(instr); t = nir_alu_type_get_base_type(t); if (!(is_ubo || is_global)) { offset = nir_intrinsic_base(instr); } unsigned nr_comp = nir_intrinsic_dest_components(instr); nir_src *src_offset = nir_get_io_offset_src(instr); bool direct = nir_src_is_const(*src_offset); nir_src *indirect_offset = direct ? NULL : src_offset; if (direct) offset += nir_src_as_uint(*src_offset); /* We may need to apply a fractional offset */ int component = (is_flat || is_interp) ? nir_intrinsic_component(instr) : 0; reg = nir_dest_index(&instr->dest); if (is_uniform && !ctx->is_blend) { emit_ubo_read(ctx, &instr->instr, reg, (ctx->sysvals.sysval_count + offset) * 16, indirect_offset, 4, 0); } else if (is_ubo) { nir_src index = instr->src[0]; /* TODO: Is indirect block number possible? */ assert(nir_src_is_const(index)); uint32_t uindex = nir_src_as_uint(index) + 1; emit_ubo_read(ctx, &instr->instr, reg, offset, indirect_offset, 0, uindex); } else if (is_global || is_shared) { emit_global(ctx, &instr->instr, true, reg, src_offset, is_shared); } else if (ctx->stage == MESA_SHADER_FRAGMENT && !ctx->is_blend) { emit_varying_read(ctx, reg, offset, nr_comp, component, indirect_offset, t, is_flat); } else if (ctx->is_blend) { /* For blend shaders, load the input color, which is * preloaded to r0 */ midgard_instruction move = v_mov(SSA_FIXED_REGISTER(0), reg); emit_mir_instruction(ctx, move); schedule_barrier(ctx); } else if (ctx->stage == MESA_SHADER_VERTEX) { emit_attr_read(ctx, reg, offset, nr_comp, t); } else { DBG("Unknown load\n"); assert(0); } break; } /* Artefact of load_interpolated_input. TODO: other barycentric modes */ case nir_intrinsic_load_barycentric_pixel: case nir_intrinsic_load_barycentric_centroid: break; /* Reads 128-bit value raw off the tilebuffer during blending, tasty */ case nir_intrinsic_load_raw_output_pan: case nir_intrinsic_load_output_u8_as_fp16_pan: reg = nir_dest_index(&instr->dest); assert(ctx->is_blend); /* T720 and below use different blend opcodes with slightly * different semantics than T760 and up */ midgard_instruction ld = m_ld_color_buffer_32u(reg, 0); bool old_blend = ctx->quirks & MIDGARD_OLD_BLEND; if (instr->intrinsic == nir_intrinsic_load_output_u8_as_fp16_pan) { ld.load_store.op = old_blend ? midgard_op_ld_color_buffer_u8_as_fp16_old : midgard_op_ld_color_buffer_u8_as_fp16; if (old_blend) { ld.load_store.address = 1; ld.load_store.arg_2 = 0x1E; } for (unsigned c = 2; c < 16; ++c) ld.swizzle[0][c] = 0; } emit_mir_instruction(ctx, ld); break; case nir_intrinsic_load_blend_const_color_rgba: { assert(ctx->is_blend); reg = nir_dest_index(&instr->dest); /* Blend constants are embedded directly in the shader and * patched in, so we use some magic routing */ midgard_instruction ins = v_mov(SSA_FIXED_REGISTER(REGISTER_CONSTANT), reg); ins.has_constants = true; ins.has_blend_constant = true; emit_mir_instruction(ctx, ins); break; } case nir_intrinsic_store_zs_output_pan: { assert(ctx->stage == MESA_SHADER_FRAGMENT); emit_fragment_store(ctx, nir_src_index(ctx, &instr->src[0]), MIDGARD_ZS_RT); midgard_instruction *br = ctx->writeout_branch[MIDGARD_ZS_RT]; if (!nir_intrinsic_component(instr)) br->writeout_depth = true; if (nir_intrinsic_component(instr) || instr->num_components) br->writeout_stencil = true; assert(br->writeout_depth | br->writeout_stencil); break; } case nir_intrinsic_store_output: assert(nir_src_is_const(instr->src[1]) && "no indirect outputs"); offset = nir_intrinsic_base(instr) + nir_src_as_uint(instr->src[1]); reg = nir_src_index(ctx, &instr->src[0]); if (ctx->stage == MESA_SHADER_FRAGMENT) { const nir_variable *var; enum midgard_rt_id rt; var = search_var(&ctx->nir->outputs, nir_intrinsic_base(instr)); assert(var); if (var->data.location == FRAG_RESULT_COLOR) rt = MIDGARD_COLOR_RT0; else if (var->data.location >= FRAG_RESULT_DATA0) rt = MIDGARD_COLOR_RT0 + var->data.location - FRAG_RESULT_DATA0; else assert(0); emit_fragment_store(ctx, reg, rt); } else if (ctx->stage == MESA_SHADER_VERTEX) { /* We should have been vectorized, though we don't * currently check that st_vary is emitted only once * per slot (this is relevant, since there's not a mask * parameter available on the store [set to 0 by the * blob]). We do respect the component by adjusting the * swizzle. If this is a constant source, we'll need to * emit that explicitly. */ emit_explicit_constant(ctx, reg, reg); unsigned dst_component = nir_intrinsic_component(instr); unsigned nr_comp = nir_src_num_components(instr->src[0]); midgard_instruction st = m_st_vary_32(reg, offset); st.load_store.arg_1 = 0x9E; st.load_store.arg_2 = 0x1E; switch (nir_alu_type_get_base_type(nir_intrinsic_type(instr))) { case nir_type_uint: case nir_type_bool: st.load_store.op = midgard_op_st_vary_32u; break; case nir_type_int: st.load_store.op = midgard_op_st_vary_32i; break; case nir_type_float: st.load_store.op = midgard_op_st_vary_32; break; default: unreachable("Attempted to store unknown type"); break; } /* nir_intrinsic_component(store_intr) encodes the * destination component start. Source component offset * adjustment is taken care of in * install_registers_instr(), when offset_swizzle() is * called. */ unsigned src_component = COMPONENT_X; assert(nr_comp > 0); for (unsigned i = 0; i < ARRAY_SIZE(st.swizzle); ++i) { st.swizzle[0][i] = src_component; if (i >= dst_component && i < dst_component + nr_comp - 1) src_component++; } emit_mir_instruction(ctx, st); } else { DBG("Unknown store\n"); assert(0); } break; /* Special case of store_output for lowered blend shaders */ case nir_intrinsic_store_raw_output_pan: assert (ctx->stage == MESA_SHADER_FRAGMENT); reg = nir_src_index(ctx, &instr->src[0]); if (ctx->quirks & MIDGARD_OLD_BLEND) { /* Suppose reg = qr0.xyzw. That means 4 8-bit ---> 1 32-bit. So * reg = r0.x. We want to splatter. So we can do a 32-bit move * of: * * imov r0.xyzw, r0.xxxx */ unsigned expanded = make_compiler_temp(ctx); midgard_instruction splatter = v_mov(reg, expanded); for (unsigned c = 0; c < 16; ++c) splatter.swizzle[1][c] = 0; emit_mir_instruction(ctx, splatter); emit_fragment_store(ctx, expanded, ctx->blend_rt); } else emit_fragment_store(ctx, reg, ctx->blend_rt); break; case nir_intrinsic_store_global: case nir_intrinsic_store_shared: reg = nir_src_index(ctx, &instr->src[0]); emit_explicit_constant(ctx, reg, reg); emit_global(ctx, &instr->instr, false, reg, &instr->src[1], instr->intrinsic == nir_intrinsic_store_shared); break; case nir_intrinsic_load_ssbo_address: emit_sysval_read(ctx, &instr->instr, 1, 0); break; case nir_intrinsic_get_buffer_size: emit_sysval_read(ctx, &instr->instr, 1, 8); break; case nir_intrinsic_load_viewport_scale: case nir_intrinsic_load_viewport_offset: case nir_intrinsic_load_num_work_groups: case nir_intrinsic_load_sampler_lod_parameters_pan: emit_sysval_read(ctx, &instr->instr, 3, 0); break; case nir_intrinsic_load_work_group_id: case nir_intrinsic_load_local_invocation_id: emit_compute_builtin(ctx, instr); break; case nir_intrinsic_load_vertex_id: case nir_intrinsic_load_instance_id: emit_vertex_builtin(ctx, instr); break; case nir_intrinsic_memory_barrier_buffer: case nir_intrinsic_memory_barrier_shared: break; case nir_intrinsic_control_barrier: schedule_barrier(ctx); emit_control_barrier(ctx); schedule_barrier(ctx); break; default: fprintf(stderr, "Unhandled intrinsic %s\n", nir_intrinsic_infos[instr->intrinsic].name); assert(0); break; } } static unsigned midgard_tex_format(enum glsl_sampler_dim dim) { switch (dim) { case GLSL_SAMPLER_DIM_1D: case GLSL_SAMPLER_DIM_BUF: return MALI_TEX_1D; case GLSL_SAMPLER_DIM_2D: case GLSL_SAMPLER_DIM_EXTERNAL: case GLSL_SAMPLER_DIM_RECT: return MALI_TEX_2D; case GLSL_SAMPLER_DIM_3D: return MALI_TEX_3D; case GLSL_SAMPLER_DIM_CUBE: return MALI_TEX_CUBE; default: DBG("Unknown sampler dim type\n"); assert(0); return 0; } } /* Tries to attach an explicit LOD / bias as a constant. Returns whether this * was successful */ static bool pan_attach_constant_bias( compiler_context *ctx, nir_src lod, midgard_texture_word *word) { /* To attach as constant, it has to *be* constant */ if (!nir_src_is_const(lod)) return false; float f = nir_src_as_float(lod); /* Break into fixed-point */ signed lod_int = f; float lod_frac = f - lod_int; /* Carry over negative fractions */ if (lod_frac < 0.0) { lod_int--; lod_frac += 1.0; } /* Encode */ word->bias = float_to_ubyte(lod_frac); word->bias_int = lod_int; return true; } static enum mali_sampler_type midgard_sampler_type(nir_alu_type t) { switch (nir_alu_type_get_base_type(t)) { case nir_type_float: return MALI_SAMPLER_FLOAT; case nir_type_int: return MALI_SAMPLER_SIGNED; case nir_type_uint: return MALI_SAMPLER_UNSIGNED; default: unreachable("Unknown sampler type"); } } static void emit_texop_native(compiler_context *ctx, nir_tex_instr *instr, unsigned midgard_texop) { /* TODO */ //assert (!instr->sampler); int texture_index = instr->texture_index; int sampler_index = texture_index; /* No helper to build texture words -- we do it all here */ midgard_instruction ins = { .type = TAG_TEXTURE_4, .mask = 0xF, .dest = nir_dest_index(&instr->dest), .src = { ~0, ~0, ~0, ~0 }, .swizzle = SWIZZLE_IDENTITY_4, .texture = { .op = midgard_texop, .format = midgard_tex_format(instr->sampler_dim), .texture_handle = texture_index, .sampler_handle = sampler_index, /* TODO: half */ .in_reg_full = 1, .out_full = 1, .sampler_type = midgard_sampler_type(instr->dest_type), .shadow = instr->is_shadow, } }; /* We may need a temporary for the coordinate */ bool needs_temp_coord = (midgard_texop == TEXTURE_OP_TEXEL_FETCH) || (instr->sampler_dim == GLSL_SAMPLER_DIM_CUBE) || (instr->is_shadow); unsigned coords = needs_temp_coord ? make_compiler_temp_reg(ctx) : 0; for (unsigned i = 0; i < instr->num_srcs; ++i) { int index = nir_src_index(ctx, &instr->src[i].src); unsigned nr_components = nir_src_num_components(instr->src[i].src); switch (instr->src[i].src_type) { case nir_tex_src_coord: { emit_explicit_constant(ctx, index, index); unsigned coord_mask = mask_of(instr->coord_components); bool flip_zw = (instr->sampler_dim == GLSL_SAMPLER_DIM_2D) && (coord_mask & (1 << COMPONENT_Z)); if (flip_zw) coord_mask ^= ((1 << COMPONENT_Z) | (1 << COMPONENT_W)); if (instr->sampler_dim == GLSL_SAMPLER_DIM_CUBE) { /* texelFetch is undefined on samplerCube */ assert(midgard_texop != TEXTURE_OP_TEXEL_FETCH); /* For cubemaps, we use a special ld/st op to * select the face and copy the xy into the * texture register */ midgard_instruction ld = m_ld_cubemap_coords(coords, 0); ld.src[1] = index; ld.mask = 0x3; /* xy */ ld.load_store.arg_1 = 0x20; ld.swizzle[1][3] = COMPONENT_X; emit_mir_instruction(ctx, ld); /* xyzw -> xyxx */ ins.swizzle[1][2] = instr->is_shadow ? COMPONENT_Z : COMPONENT_X; ins.swizzle[1][3] = COMPONENT_X; } else if (needs_temp_coord) { /* mov coord_temp, coords */ midgard_instruction mov = v_mov(index, coords); mov.mask = coord_mask; if (flip_zw) mov.swizzle[1][COMPONENT_W] = COMPONENT_Z; emit_mir_instruction(ctx, mov); } else { coords = index; } ins.src[1] = coords; /* Texelfetch coordinates uses all four elements * (xyz/index) regardless of texture dimensionality, * which means it's necessary to zero the unused * components to keep everything happy */ if (midgard_texop == TEXTURE_OP_TEXEL_FETCH) { /* mov index.zw, #0, or generalized */ midgard_instruction mov = v_mov(SSA_FIXED_REGISTER(REGISTER_CONSTANT), coords); mov.has_constants = true; mov.mask = coord_mask ^ 0xF; emit_mir_instruction(ctx, mov); } if (instr->sampler_dim == GLSL_SAMPLER_DIM_2D) { /* Array component in w but NIR wants it in z, * but if we have a temp coord we already fixed * that up */ if (nr_components == 3) { ins.swizzle[1][2] = COMPONENT_Z; ins.swizzle[1][3] = needs_temp_coord ? COMPONENT_W : COMPONENT_Z; } else if (nr_components == 2) { ins.swizzle[1][2] = instr->is_shadow ? COMPONENT_Z : COMPONENT_X; ins.swizzle[1][3] = COMPONENT_X; } else unreachable("Invalid texture 2D components"); } if (midgard_texop == TEXTURE_OP_TEXEL_FETCH) { /* We zeroed */ ins.swizzle[1][2] = COMPONENT_Z; ins.swizzle[1][3] = COMPONENT_W; } break; } case nir_tex_src_bias: case nir_tex_src_lod: { /* Try as a constant if we can */ bool is_txf = midgard_texop == TEXTURE_OP_TEXEL_FETCH; if (!is_txf && pan_attach_constant_bias(ctx, instr->src[i].src, &ins.texture)) break; ins.texture.lod_register = true; ins.src[2] = index; for (unsigned c = 0; c < MIR_VEC_COMPONENTS; ++c) ins.swizzle[2][c] = COMPONENT_X; emit_explicit_constant(ctx, index, index); break; }; case nir_tex_src_offset: { ins.texture.offset_register = true; ins.src[3] = index; for (unsigned c = 0; c < MIR_VEC_COMPONENTS; ++c) ins.swizzle[3][c] = (c > COMPONENT_Z) ? 0 : c; emit_explicit_constant(ctx, index, index); break; }; case nir_tex_src_comparator: { unsigned comp = COMPONENT_Z; /* mov coord_temp.foo, coords */ midgard_instruction mov = v_mov(index, coords); mov.mask = 1 << comp; for (unsigned i = 0; i < MIR_VEC_COMPONENTS; ++i) mov.swizzle[1][i] = COMPONENT_X; emit_mir_instruction(ctx, mov); break; } default: { fprintf(stderr, "Unknown texture source type: %d\n", instr->src[i].src_type); assert(0); } } } emit_mir_instruction(ctx, ins); /* Used for .cont and .last hinting */ ctx->texture_op_count++; } static void emit_tex(compiler_context *ctx, nir_tex_instr *instr) { switch (instr->op) { case nir_texop_tex: case nir_texop_txb: emit_texop_native(ctx, instr, TEXTURE_OP_NORMAL); break; case nir_texop_txl: emit_texop_native(ctx, instr, TEXTURE_OP_LOD); break; case nir_texop_txf: emit_texop_native(ctx, instr, TEXTURE_OP_TEXEL_FETCH); break; case nir_texop_txs: emit_sysval_read(ctx, &instr->instr, 4, 0); break; default: { fprintf(stderr, "Unhandled texture op: %d\n", instr->op); assert(0); } } } static void emit_jump(compiler_context *ctx, nir_jump_instr *instr) { switch (instr->type) { case nir_jump_break: { /* Emit a branch out of the loop */ struct midgard_instruction br = v_branch(false, false); br.branch.target_type = TARGET_BREAK; br.branch.target_break = ctx->current_loop_depth; emit_mir_instruction(ctx, br); break; } default: DBG("Unknown jump type %d\n", instr->type); break; } } static void emit_instr(compiler_context *ctx, struct nir_instr *instr) { switch (instr->type) { case nir_instr_type_load_const: emit_load_const(ctx, nir_instr_as_load_const(instr)); break; case nir_instr_type_intrinsic: emit_intrinsic(ctx, nir_instr_as_intrinsic(instr)); break; case nir_instr_type_alu: emit_alu(ctx, nir_instr_as_alu(instr)); break; case nir_instr_type_tex: emit_tex(ctx, nir_instr_as_tex(instr)); break; case nir_instr_type_jump: emit_jump(ctx, nir_instr_as_jump(instr)); break; case nir_instr_type_ssa_undef: /* Spurious */ break; default: DBG("Unhandled instruction type\n"); break; } } /* ALU instructions can inline or embed constants, which decreases register * pressure and saves space. */ #define CONDITIONAL_ATTACH(idx) { \ void *entry = _mesa_hash_table_u64_search(ctx->ssa_constants, alu->src[idx] + 1); \ \ if (entry) { \ attach_constants(ctx, alu, entry, alu->src[idx] + 1); \ alu->src[idx] = SSA_FIXED_REGISTER(REGISTER_CONSTANT); \ } \ } static void inline_alu_constants(compiler_context *ctx, midgard_block *block) { mir_foreach_instr_in_block(block, alu) { /* Other instructions cannot inline constants */ if (alu->type != TAG_ALU_4) continue; if (alu->compact_branch) continue; /* If there is already a constant here, we can do nothing */ if (alu->has_constants) continue; CONDITIONAL_ATTACH(0); if (!alu->has_constants) { CONDITIONAL_ATTACH(1) } else if (!alu->inline_constant) { /* Corner case: _two_ vec4 constants, for instance with a * csel. For this case, we can only use a constant * register for one, we'll have to emit a move for the * other. Note, if both arguments are constants, then * necessarily neither argument depends on the value of * any particular register. As the destination register * will be wiped, that means we can spill the constant * to the destination register. */ void *entry = _mesa_hash_table_u64_search(ctx->ssa_constants, alu->src[1] + 1); unsigned scratch = alu->dest; if (entry) { midgard_instruction ins = v_mov(SSA_FIXED_REGISTER(REGISTER_CONSTANT), scratch); attach_constants(ctx, &ins, entry, alu->src[1] + 1); /* Set the source */ alu->src[1] = scratch; /* Inject us -before- the last instruction which set r31 */ mir_insert_instruction_before(ctx, mir_prev_op(alu), ins); } } } } /* Being a little silly with the names, but returns the op that is the bitwise * inverse of the op with the argument switched. I.e. (f and g are * contrapositives): * * f(a, b) = ~g(b, a) * * Corollary: if g is the contrapositve of f, f is the contrapositive of g: * * f(a, b) = ~g(b, a) * ~f(a, b) = g(b, a) * ~f(a, b) = ~h(a, b) where h is the contrapositive of g * f(a, b) = h(a, b) * * Thus we define this function in pairs. */ static inline midgard_alu_op mir_contrapositive(midgard_alu_op op) { switch (op) { case midgard_alu_op_flt: return midgard_alu_op_fle; case midgard_alu_op_fle: return midgard_alu_op_flt; case midgard_alu_op_ilt: return midgard_alu_op_ile; case midgard_alu_op_ile: return midgard_alu_op_ilt; default: unreachable("No known contrapositive"); } } /* Midgard supports two types of constants, embedded constants (128-bit) and * inline constants (16-bit). Sometimes, especially with scalar ops, embedded * constants can be demoted to inline constants, for space savings and * sometimes a performance boost */ static void embedded_to_inline_constant(compiler_context *ctx, midgard_block *block) { mir_foreach_instr_in_block(block, ins) { if (!ins->has_constants) continue; if (ins->has_inline_constant) continue; /* Blend constants must not be inlined by definition */ if (ins->has_blend_constant) continue; /* We can inline 32-bit (sometimes) or 16-bit (usually) */ bool is_16 = ins->alu.reg_mode == midgard_reg_mode_16; bool is_32 = ins->alu.reg_mode == midgard_reg_mode_32; if (!(is_16 || is_32)) continue; /* src1 cannot be an inline constant due to encoding * restrictions. So, if possible we try to flip the arguments * in that case */ int op = ins->alu.op; if (ins->src[0] == SSA_FIXED_REGISTER(REGISTER_CONSTANT)) { bool flip = alu_opcode_props[op].props & OP_COMMUTES; switch (op) { /* Conditionals can be inverted */ case midgard_alu_op_flt: case midgard_alu_op_ilt: case midgard_alu_op_fle: case midgard_alu_op_ile: ins->alu.op = mir_contrapositive(ins->alu.op); ins->invert = true; flip = true; break; case midgard_alu_op_fcsel: case midgard_alu_op_icsel: DBG("Missed non-commutative flip (%s)\n", alu_opcode_props[op].name); default: break; } if (flip) mir_flip(ins); } if (ins->src[1] == SSA_FIXED_REGISTER(REGISTER_CONSTANT)) { /* Extract the source information */ midgard_vector_alu_src *src; int q = ins->alu.src2; midgard_vector_alu_src *m = (midgard_vector_alu_src *) &q; src = m; /* Component is from the swizzle. Take a nonzero component */ assert(ins->mask); unsigned first_comp = ffs(ins->mask) - 1; unsigned component = ins->swizzle[1][first_comp]; /* Scale constant appropriately, if we can legally */ uint16_t scaled_constant = 0; if (is_16) { scaled_constant = ins->constants.u16[component]; } else if (midgard_is_integer_op(op)) { scaled_constant = ins->constants.u32[component]; /* Constant overflow after resize */ if (scaled_constant != ins->constants.u32[component]) continue; } else { float original = ins->constants.f32[component]; scaled_constant = _mesa_float_to_half(original); /* Check for loss of precision. If this is * mediump, we don't care, but for a highp * shader, we need to pay attention. NIR * doesn't yet tell us which mode we're in! * Practically this prevents most constants * from being inlined, sadly. */ float fp32 = _mesa_half_to_float(scaled_constant); if (fp32 != original) continue; } /* We don't know how to handle these with a constant */ if (mir_nontrivial_source2_mod_simple(ins) || src->rep_low || src->rep_high) { DBG("Bailing inline constant...\n"); continue; } /* Make sure that the constant is not itself a vector * by checking if all accessed values are the same. */ const midgard_constants *cons = &ins->constants; uint32_t value = is_16 ? cons->u16[component] : cons->u32[component]; bool is_vector = false; unsigned mask = effective_writemask(&ins->alu, ins->mask); for (unsigned c = 0; c < MIR_VEC_COMPONENTS; ++c) { /* We only care if this component is actually used */ if (!(mask & (1 << c))) continue; uint32_t test = is_16 ? cons->u16[ins->swizzle[1][c]] : cons->u32[ins->swizzle[1][c]]; if (test != value) { is_vector = true; break; } } if (is_vector) continue; /* Get rid of the embedded constant */ ins->has_constants = false; ins->src[1] = ~0; ins->has_inline_constant = true; ins->inline_constant = scaled_constant; } } } /* Dead code elimination for branches at the end of a block - only one branch * per block is legal semantically */ static void midgard_opt_cull_dead_branch(compiler_context *ctx, midgard_block *block) { bool branched = false; mir_foreach_instr_in_block_safe(block, ins) { if (!midgard_is_branch_unit(ins->unit)) continue; if (branched) mir_remove_instruction(ins); branched = true; } } /* fmov.pos is an idiom for fpos. Propoagate the .pos up to the source, so then * the move can be propagated away entirely */ static bool mir_compose_float_outmod(midgard_outmod_float *outmod, midgard_outmod_float comp) { /* Nothing to do */ if (comp == midgard_outmod_none) return true; if (*outmod == midgard_outmod_none) { *outmod = comp; return true; } /* TODO: Compose rules */ return false; } static bool midgard_opt_pos_propagate(compiler_context *ctx, midgard_block *block) { bool progress = false; mir_foreach_instr_in_block_safe(block, ins) { if (ins->type != TAG_ALU_4) continue; if (ins->alu.op != midgard_alu_op_fmov) continue; if (ins->alu.outmod != midgard_outmod_pos) continue; /* TODO: Registers? */ unsigned src = ins->src[1]; if (src & IS_REG) continue; /* There might be a source modifier, too */ if (mir_nontrivial_source2_mod(ins)) continue; /* Backpropagate the modifier */ mir_foreach_instr_in_block_from_rev(block, v, mir_prev_op(ins)) { if (v->type != TAG_ALU_4) continue; if (v->dest != src) continue; /* Can we even take a float outmod? */ if (midgard_is_integer_out_op(v->alu.op)) continue; midgard_outmod_float temp = v->alu.outmod; progress |= mir_compose_float_outmod(&temp, ins->alu.outmod); /* Throw in the towel.. */ if (!progress) break; /* Otherwise, transfer the modifier */ v->alu.outmod = temp; ins->alu.outmod = midgard_outmod_none; break; } } return progress; } static unsigned emit_fragment_epilogue(compiler_context *ctx, unsigned rt) { /* Loop to ourselves */ midgard_instruction *br = ctx->writeout_branch[rt]; struct midgard_instruction ins = v_branch(false, false); ins.writeout = true; ins.writeout_depth = br->writeout_depth; ins.writeout_stencil = br->writeout_stencil; ins.branch.target_block = ctx->block_count - 1; ins.constants.u32[0] = br->constants.u32[0]; emit_mir_instruction(ctx, ins); ctx->current_block->epilogue = true; schedule_barrier(ctx); return ins.branch.target_block; } static midgard_block * emit_block(compiler_context *ctx, nir_block *block) { midgard_block *this_block = ctx->after_block; ctx->after_block = NULL; if (!this_block) this_block = create_empty_block(ctx); list_addtail(&this_block->base.link, &ctx->blocks); this_block->scheduled = false; ++ctx->block_count; /* Set up current block */ list_inithead(&this_block->base.instructions); ctx->current_block = this_block; nir_foreach_instr(instr, block) { emit_instr(ctx, instr); ++ctx->instruction_count; } return this_block; } static midgard_block *emit_cf_list(struct compiler_context *ctx, struct exec_list *list); static void emit_if(struct compiler_context *ctx, nir_if *nif) { midgard_block *before_block = ctx->current_block; /* Speculatively emit the branch, but we can't fill it in until later */ EMIT(branch, true, true); midgard_instruction *then_branch = mir_last_in_block(ctx->current_block); then_branch->src[0] = nir_src_index(ctx, &nif->condition); /* Emit the two subblocks. */ midgard_block *then_block = emit_cf_list(ctx, &nif->then_list); midgard_block *end_then_block = ctx->current_block; /* Emit a jump from the end of the then block to the end of the else */ EMIT(branch, false, false); midgard_instruction *then_exit = mir_last_in_block(ctx->current_block); /* Emit second block, and check if it's empty */ int else_idx = ctx->block_count; int count_in = ctx->instruction_count; midgard_block *else_block = emit_cf_list(ctx, &nif->else_list); midgard_block *end_else_block = ctx->current_block; int after_else_idx = ctx->block_count; /* Now that we have the subblocks emitted, fix up the branches */ assert(then_block); assert(else_block); if (ctx->instruction_count == count_in) { /* The else block is empty, so don't emit an exit jump */ mir_remove_instruction(then_exit); then_branch->branch.target_block = after_else_idx; } else { then_branch->branch.target_block = else_idx; then_exit->branch.target_block = after_else_idx; } /* Wire up the successors */ ctx->after_block = create_empty_block(ctx); pan_block_add_successor(&before_block->base, &then_block->base); pan_block_add_successor(&before_block->base, &else_block->base); pan_block_add_successor(&end_then_block->base, &ctx->after_block->base); pan_block_add_successor(&end_else_block->base, &ctx->after_block->base); } static void emit_loop(struct compiler_context *ctx, nir_loop *nloop) { /* Remember where we are */ midgard_block *start_block = ctx->current_block; /* Allocate a loop number, growing the current inner loop depth */ int loop_idx = ++ctx->current_loop_depth; /* Get index from before the body so we can loop back later */ int start_idx = ctx->block_count; /* Emit the body itself */ midgard_block *loop_block = emit_cf_list(ctx, &nloop->body); /* Branch back to loop back */ struct midgard_instruction br_back = v_branch(false, false); br_back.branch.target_block = start_idx; emit_mir_instruction(ctx, br_back); /* Mark down that branch in the graph. */ pan_block_add_successor(&start_block->base, &loop_block->base); pan_block_add_successor(&ctx->current_block->base, &loop_block->base); /* Find the index of the block about to follow us (note: we don't add * one; blocks are 0-indexed so we get a fencepost problem) */ int break_block_idx = ctx->block_count; /* Fix up the break statements we emitted to point to the right place, * now that we can allocate a block number for them */ ctx->after_block = create_empty_block(ctx); mir_foreach_block_from(ctx, start_block, _block) { mir_foreach_instr_in_block(((midgard_block *) _block), ins) { if (ins->type != TAG_ALU_4) continue; if (!ins->compact_branch) continue; /* We found a branch -- check the type to see if we need to do anything */ if (ins->branch.target_type != TARGET_BREAK) continue; /* It's a break! Check if it's our break */ if (ins->branch.target_break != loop_idx) continue; /* Okay, cool, we're breaking out of this loop. * Rewrite from a break to a goto */ ins->branch.target_type = TARGET_GOTO; ins->branch.target_block = break_block_idx; pan_block_add_successor(_block, &ctx->after_block->base); } } /* Now that we've finished emitting the loop, free up the depth again * so we play nice with recursion amid nested loops */ --ctx->current_loop_depth; /* Dump loop stats */ ++ctx->loop_count; } static midgard_block * emit_cf_list(struct compiler_context *ctx, struct exec_list *list) { midgard_block *start_block = NULL; foreach_list_typed(nir_cf_node, node, node, list) { switch (node->type) { case nir_cf_node_block: { midgard_block *block = emit_block(ctx, nir_cf_node_as_block(node)); if (!start_block) start_block = block; break; } case nir_cf_node_if: emit_if(ctx, nir_cf_node_as_if(node)); break; case nir_cf_node_loop: emit_loop(ctx, nir_cf_node_as_loop(node)); break; case nir_cf_node_function: assert(0); break; } } return start_block; } /* Due to lookahead, we need to report the first tag executed in the command * stream and in branch targets. An initial block might be empty, so iterate * until we find one that 'works' */ static unsigned midgard_get_first_tag_from_block(compiler_context *ctx, unsigned block_idx) { midgard_block *initial_block = mir_get_block(ctx, block_idx); mir_foreach_block_from(ctx, initial_block, _v) { midgard_block *v = (midgard_block *) _v; if (v->quadword_count) { midgard_bundle *initial_bundle = util_dynarray_element(&v->bundles, midgard_bundle, 0); return initial_bundle->tag; } } /* Default to a tag 1 which will break from the shader, in case we jump * to the exit block (i.e. `return` in a compute shader) */ return 1; } /* For each fragment writeout instruction, generate a writeout loop to * associate with it */ static void mir_add_writeout_loops(compiler_context *ctx) { for (unsigned rt = 0; rt < ARRAY_SIZE(ctx->writeout_branch); ++rt) { midgard_instruction *br = ctx->writeout_branch[rt]; if (!br) continue; unsigned popped = br->branch.target_block; pan_block_add_successor(&(mir_get_block(ctx, popped - 1)->base), &ctx->current_block->base); br->branch.target_block = emit_fragment_epilogue(ctx, rt); /* If we have more RTs, we'll need to restore back after our * loop terminates */ if ((rt + 1) < ARRAY_SIZE(ctx->writeout_branch) && ctx->writeout_branch[rt + 1]) { midgard_instruction uncond = v_branch(false, false); uncond.branch.target_block = popped; emit_mir_instruction(ctx, uncond); pan_block_add_successor(&ctx->current_block->base, &(mir_get_block(ctx, popped)->base)); schedule_barrier(ctx); } else { /* We're last, so we can terminate here */ br->last_writeout = true; } } } int midgard_compile_shader_nir(nir_shader *nir, panfrost_program *program, bool is_blend, unsigned blend_rt, unsigned gpu_id, bool shaderdb) { struct util_dynarray *compiled = &program->compiled; midgard_debug = debug_get_option_midgard_debug(); /* TODO: Bound against what? */ compiler_context *ctx = rzalloc(NULL, compiler_context); ctx->nir = nir; ctx->stage = nir->info.stage; ctx->is_blend = is_blend; ctx->alpha_ref = program->alpha_ref; ctx->blend_rt = MIDGARD_COLOR_RT0 + blend_rt; ctx->quirks = midgard_get_quirks(gpu_id); /* Start off with a safe cutoff, allowing usage of all 16 work * registers. Later, we'll promote uniform reads to uniform registers * if we determine it is beneficial to do so */ ctx->uniform_cutoff = 8; /* Initialize at a global (not block) level hash tables */ ctx->ssa_constants = _mesa_hash_table_u64_create(NULL); ctx->hash_to_temp = _mesa_hash_table_u64_create(NULL); /* Lower gl_Position pre-optimisation, but after lowering vars to ssa * (so we don't accidentally duplicate the epilogue since mesa/st has * messed with our I/O quite a bit already) */ NIR_PASS_V(nir, nir_lower_vars_to_ssa); if (ctx->stage == MESA_SHADER_VERTEX) { NIR_PASS_V(nir, nir_lower_viewport_transform); NIR_PASS_V(nir, nir_lower_point_size, 1.0, 1024.0); } NIR_PASS_V(nir, nir_lower_var_copies); NIR_PASS_V(nir, nir_lower_vars_to_ssa); NIR_PASS_V(nir, nir_split_var_copies); NIR_PASS_V(nir, nir_lower_var_copies); NIR_PASS_V(nir, nir_lower_global_vars_to_local); NIR_PASS_V(nir, nir_lower_var_copies); NIR_PASS_V(nir, nir_lower_vars_to_ssa); NIR_PASS_V(nir, nir_lower_io, nir_var_all, glsl_type_size, 0); NIR_PASS_V(nir, nir_lower_ssbo); NIR_PASS_V(nir, midgard_nir_lower_zs_store); /* Optimisation passes */ optimise_nir(nir, ctx->quirks); if (midgard_debug & MIDGARD_DBG_SHADERS) { nir_print_shader(nir, stdout); } /* Assign sysvals and counts, now that we're sure * (post-optimisation) */ panfrost_nir_assign_sysvals(&ctx->sysvals, nir); program->sysval_count = ctx->sysvals.sysval_count; memcpy(program->sysvals, ctx->sysvals.sysvals, sizeof(ctx->sysvals.sysvals[0]) * ctx->sysvals.sysval_count); nir_foreach_function(func, nir) { if (!func->impl) continue; list_inithead(&ctx->blocks); ctx->block_count = 0; ctx->func = func; emit_cf_list(ctx, &func->impl->body); break; /* TODO: Multi-function shaders */ } util_dynarray_init(compiled, NULL); /* Per-block lowering before opts */ mir_foreach_block(ctx, _block) { midgard_block *block = (midgard_block *) _block; inline_alu_constants(ctx, block); midgard_opt_promote_fmov(ctx, block); embedded_to_inline_constant(ctx, block); } /* MIR-level optimizations */ bool progress = false; do { progress = false; mir_foreach_block(ctx, _block) { midgard_block *block = (midgard_block *) _block; progress |= midgard_opt_pos_propagate(ctx, block); progress |= midgard_opt_copy_prop(ctx, block); progress |= midgard_opt_dead_code_eliminate(ctx, block); progress |= midgard_opt_combine_projection(ctx, block); progress |= midgard_opt_varying_projection(ctx, block); progress |= midgard_opt_not_propagate(ctx, block); progress |= midgard_opt_fuse_src_invert(ctx, block); progress |= midgard_opt_fuse_dest_invert(ctx, block); progress |= midgard_opt_csel_invert(ctx, block); progress |= midgard_opt_drop_cmp_invert(ctx, block); progress |= midgard_opt_invert_branch(ctx, block); } } while (progress); mir_foreach_block(ctx, _block) { midgard_block *block = (midgard_block *) _block; midgard_lower_invert(ctx, block); midgard_lower_derivatives(ctx, block); } /* Nested control-flow can result in dead branches at the end of the * block. This messes with our analysis and is just dead code, so cull * them */ mir_foreach_block(ctx, _block) { midgard_block *block = (midgard_block *) _block; midgard_opt_cull_dead_branch(ctx, block); } /* Ensure we were lowered */ mir_foreach_instr_global(ctx, ins) { assert(!ins->invert); } if (ctx->stage == MESA_SHADER_FRAGMENT) mir_add_writeout_loops(ctx); /* Schedule! */ midgard_schedule_program(ctx); mir_ra(ctx); /* Now that all the bundles are scheduled and we can calculate block * sizes, emit actual branch instructions rather than placeholders */ int br_block_idx = 0; mir_foreach_block(ctx, _block) { midgard_block *block = (midgard_block *) _block; util_dynarray_foreach(&block->bundles, midgard_bundle, bundle) { for (int c = 0; c < bundle->instruction_count; ++c) { midgard_instruction *ins = bundle->instructions[c]; if (!midgard_is_branch_unit(ins->unit)) continue; /* Parse some basic branch info */ bool is_compact = ins->unit == ALU_ENAB_BR_COMPACT; bool is_conditional = ins->branch.conditional; bool is_inverted = ins->branch.invert_conditional; bool is_discard = ins->branch.target_type == TARGET_DISCARD; bool is_writeout = ins->writeout; /* Determine the block we're jumping to */ int target_number = ins->branch.target_block; /* Report the destination tag */ int dest_tag = is_discard ? 0 : midgard_get_first_tag_from_block(ctx, target_number); /* Count up the number of quadwords we're * jumping over = number of quadwords until * (br_block_idx, target_number) */ int quadword_offset = 0; if (is_discard) { /* Ignored */ } else if (target_number > br_block_idx) { /* Jump forward */ for (int idx = br_block_idx + 1; idx < target_number; ++idx) { midgard_block *blk = mir_get_block(ctx, idx); assert(blk); quadword_offset += blk->quadword_count; } } else { /* Jump backwards */ for (int idx = br_block_idx; idx >= target_number; --idx) { midgard_block *blk = mir_get_block(ctx, idx); assert(blk); quadword_offset -= blk->quadword_count; } } /* Unconditional extended branches (far jumps) * have issues, so we always use a conditional * branch, setting the condition to always for * unconditional. For compact unconditional * branches, cond isn't used so it doesn't * matter what we pick. */ midgard_condition cond = !is_conditional ? midgard_condition_always : is_inverted ? midgard_condition_false : midgard_condition_true; midgard_jmp_writeout_op op = is_discard ? midgard_jmp_writeout_op_discard : is_writeout ? midgard_jmp_writeout_op_writeout : (is_compact && !is_conditional) ? midgard_jmp_writeout_op_branch_uncond : midgard_jmp_writeout_op_branch_cond; if (!is_compact) { midgard_branch_extended branch = midgard_create_branch_extended( cond, op, dest_tag, quadword_offset); memcpy(&ins->branch_extended, &branch, sizeof(branch)); } else if (is_conditional || is_discard) { midgard_branch_cond branch = { .op = op, .dest_tag = dest_tag, .offset = quadword_offset, .cond = cond }; assert(branch.offset == quadword_offset); memcpy(&ins->br_compact, &branch, sizeof(branch)); } else { assert(op == midgard_jmp_writeout_op_branch_uncond); midgard_branch_uncond branch = { .op = op, .dest_tag = dest_tag, .offset = quadword_offset, .unknown = 1 }; assert(branch.offset == quadword_offset); memcpy(&ins->br_compact, &branch, sizeof(branch)); } } } ++br_block_idx; } /* Emit flat binary from the instruction arrays. Iterate each block in * sequence. Save instruction boundaries such that lookahead tags can * be assigned easily */ /* Cache _all_ bundles in source order for lookahead across failed branches */ int bundle_count = 0; mir_foreach_block(ctx, _block) { midgard_block *block = (midgard_block *) _block; bundle_count += block->bundles.size / sizeof(midgard_bundle); } midgard_bundle **source_order_bundles = malloc(sizeof(midgard_bundle *) * bundle_count); int bundle_idx = 0; mir_foreach_block(ctx, _block) { midgard_block *block = (midgard_block *) _block; util_dynarray_foreach(&block->bundles, midgard_bundle, bundle) { source_order_bundles[bundle_idx++] = bundle; } } int current_bundle = 0; /* Midgard prefetches instruction types, so during emission we * need to lookahead. Unless this is the last instruction, in * which we return 1. */ mir_foreach_block(ctx, _block) { midgard_block *block = (midgard_block *) _block; mir_foreach_bundle_in_block(block, bundle) { int lookahead = 1; if (!bundle->last_writeout && (current_bundle + 1 < bundle_count)) lookahead = source_order_bundles[current_bundle + 1]->tag; emit_binary_bundle(ctx, bundle, compiled, lookahead); ++current_bundle; } /* TODO: Free deeper */ //util_dynarray_fini(&block->instructions); } free(source_order_bundles); /* Report the very first tag executed */ program->first_tag = midgard_get_first_tag_from_block(ctx, 0); /* Deal with off-by-one related to the fencepost problem */ program->work_register_count = ctx->work_registers + 1; program->uniform_cutoff = ctx->uniform_cutoff; program->blend_patch_offset = ctx->blend_constant_offset; program->tls_size = ctx->tls_size; if (midgard_debug & MIDGARD_DBG_SHADERS) disassemble_midgard(stdout, program->compiled.data, program->compiled.size, gpu_id, ctx->stage); if (midgard_debug & MIDGARD_DBG_SHADERDB || shaderdb) { unsigned nr_bundles = 0, nr_ins = 0; /* Count instructions and bundles */ mir_foreach_block(ctx, _block) { midgard_block *block = (midgard_block *) _block; nr_bundles += util_dynarray_num_elements( &block->bundles, midgard_bundle); mir_foreach_bundle_in_block(block, bun) nr_ins += bun->instruction_count; } /* Calculate thread count. There are certain cutoffs by * register count for thread count */ unsigned nr_registers = program->work_register_count; unsigned nr_threads = (nr_registers <= 4) ? 4 : (nr_registers <= 8) ? 2 : 1; /* Dump stats */ fprintf(stderr, "shader%d - %s shader: " "%u inst, %u bundles, %u quadwords, " "%u registers, %u threads, %u loops, " "%u:%u spills:fills\n", SHADER_DB_COUNT++, gl_shader_stage_name(ctx->stage), nr_ins, nr_bundles, ctx->quadword_count, nr_registers, nr_threads, ctx->loop_count, ctx->spills, ctx->fills); } ralloc_free(ctx); return 0; }