/* * Copyright 2018 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 * on the rights to use, copy, modify, merge, publish, distribute, sub * license, 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 NON-INFRINGEMENT. IN NO EVENT SHALL * THE AUTHOR(S) AND/OR THEIR SUPPLIERS 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 "nir_to_spirv.h" #include "spirv_builder.h" #include "nir.h" #include "pipe/p_state.h" #include "util/u_math.h" #include "util/u_memory.h" #include "util/hash_table.h" #define SLOT_UNSET ((unsigned char) -1) struct ntv_context { void *mem_ctx; /* SPIR-V 1.4 and later requires entrypoints to list all global * variables in the interface. */ bool spirv_1_4_interfaces; bool explicit_lod; //whether to set lod=0 for texture() struct spirv_builder builder; nir_shader *nir; struct hash_table *glsl_types; struct hash_table *bo_struct_types; struct hash_table *bo_array_types; SpvId GLSL_std_450; gl_shader_stage stage; const struct zink_shader_info *sinfo; SpvId ubos[PIPE_MAX_CONSTANT_BUFFERS][5]; //8, 16, 32, unused, 64 nir_variable *ubo_vars[PIPE_MAX_CONSTANT_BUFFERS]; SpvId ssbos[5]; //8, 16, 32, unused, 64 nir_variable *ssbo_vars; SpvId images[PIPE_MAX_SHADER_IMAGES]; struct hash_table image_types; SpvId samplers[PIPE_MAX_SHADER_SAMPLER_VIEWS]; SpvId bindless_samplers[2]; nir_variable *sampler_var[PIPE_MAX_SHADER_SAMPLER_VIEWS]; /* driver_location -> variable */ nir_variable *bindless_sampler_var[2]; unsigned last_sampler; unsigned bindless_set_idx; nir_variable *image_var[PIPE_MAX_SHADER_IMAGES]; /* driver_location -> variable */ SpvId entry_ifaces[PIPE_MAX_SHADER_INPUTS * 4 + PIPE_MAX_SHADER_OUTPUTS * 4]; size_t num_entry_ifaces; SpvId *defs; nir_alu_type *def_types; SpvId *resident_defs; size_t num_defs; struct hash_table *vars; /* nir_variable -> SpvId */ const SpvId *block_ids; size_t num_blocks; bool block_started; SpvId loop_break, loop_cont; SpvId shared_block_var[5]; //8, 16, 32, unused, 64 SpvId shared_block_arr_type[5]; //8, 16, 32, unused, 64 SpvId scratch_block_var[5]; //8, 16, 32, unused, 64 SpvId front_face_var, instance_id_var, vertex_id_var, primitive_id_var, invocation_id_var, // geometry sample_mask_type, sample_id_var, sample_pos_var, sample_mask_in_var, tess_patch_vertices_in, tess_coord_var, // tess push_const_var, point_coord_var, workgroup_id_var, num_workgroups_var, local_invocation_id_var, global_invocation_id_var, local_invocation_index_var, helper_invocation_var, local_group_size_var, view_index_var, base_vertex_var, base_instance_var, draw_id_var; SpvId shared_mem_size; SpvId subgroup_eq_mask_var, subgroup_ge_mask_var, subgroup_gt_mask_var, subgroup_id_var, subgroup_invocation_var, subgroup_le_mask_var, subgroup_lt_mask_var, subgroup_size_var; SpvId discard_func; SpvId float_array_type[2]; }; static SpvId get_fvec_constant(struct ntv_context *ctx, unsigned bit_size, unsigned num_components, double value); static SpvId get_ivec_constant(struct ntv_context *ctx, unsigned bit_size, unsigned num_components, int64_t value); static SpvId emit_unop(struct ntv_context *ctx, SpvOp op, SpvId type, SpvId src); static SpvId emit_binop(struct ntv_context *ctx, SpvOp op, SpvId type, SpvId src0, SpvId src1); static SpvId emit_triop(struct ntv_context *ctx, SpvOp op, SpvId type, SpvId src0, SpvId src1, SpvId src2); static bool alu_op_is_typeless(nir_op op) { switch (op) { case nir_op_mov: case nir_op_vec16: case nir_op_vec2: case nir_op_vec3: case nir_op_vec4: case nir_op_vec5: case nir_op_vec8: case nir_op_bcsel: return true; default: break; } return false; } static nir_alu_type get_nir_alu_type(const struct glsl_type *type) { return nir_alu_type_get_base_type(nir_get_nir_type_for_glsl_base_type(glsl_get_base_type(glsl_without_array_or_matrix(type)))); } static nir_alu_type infer_nir_alu_type_from_uses_ssa(nir_def *ssa); static nir_alu_type infer_nir_alu_type_from_use(nir_src *src) { nir_instr *instr = nir_src_parent_instr(src); nir_alu_type atype = nir_type_invalid; switch (instr->type) { case nir_instr_type_alu: { nir_alu_instr *alu = nir_instr_as_alu(instr); if (alu->op == nir_op_bcsel) { if (nir_srcs_equal(alu->src[0].src, *src)) { /* special case: the first src in bcsel is always bool */ return nir_type_bool; } } /* ignore typeless ops */ if (alu_op_is_typeless(alu->op)) { atype = infer_nir_alu_type_from_uses_ssa(&alu->def); break; } for (unsigned i = 0; i < nir_op_infos[alu->op].num_inputs; i++) { if (!nir_srcs_equal(alu->src[i].src, *src)) continue; atype = nir_op_infos[alu->op].input_types[i]; break; } break; } case nir_instr_type_tex: { nir_tex_instr *tex = nir_instr_as_tex(instr); for (unsigned i = 0; i < tex->num_srcs; i++) { if (!nir_srcs_equal(tex->src[i].src, *src)) continue; switch (tex->src[i].src_type) { case nir_tex_src_coord: case nir_tex_src_lod: if (tex->op == nir_texop_txf || tex->op == nir_texop_txf_ms || tex->op == nir_texop_txs) atype = nir_type_int; else atype = nir_type_float; break; case nir_tex_src_projector: case nir_tex_src_bias: case nir_tex_src_min_lod: case nir_tex_src_comparator: case nir_tex_src_ddx: case nir_tex_src_ddy: atype = nir_type_float; break; case nir_tex_src_offset: case nir_tex_src_ms_index: case nir_tex_src_texture_offset: case nir_tex_src_sampler_offset: case nir_tex_src_sampler_handle: case nir_tex_src_texture_handle: atype = nir_type_int; break; default: break; } break; } break; } case nir_instr_type_intrinsic: { if (nir_instr_as_intrinsic(instr)->intrinsic == nir_intrinsic_load_deref) { atype = get_nir_alu_type(nir_instr_as_deref(instr)->type); } else if (nir_instr_as_intrinsic(instr)->intrinsic == nir_intrinsic_store_deref) { atype = get_nir_alu_type(nir_src_as_deref(nir_instr_as_intrinsic(instr)->src[0])->type); } break; } default: break; } return nir_alu_type_get_base_type(atype); } static nir_alu_type infer_nir_alu_type_from_uses_ssa(nir_def *ssa) { nir_alu_type atype = nir_type_invalid; /* try to infer a type: if it's wrong then whatever, but at least we tried */ nir_foreach_use_including_if(src, ssa) { if (nir_src_is_if(src)) return nir_type_bool; atype = infer_nir_alu_type_from_use(src); if (atype) break; } return atype ? atype : nir_type_uint; } static SpvId get_bvec_type(struct ntv_context *ctx, int num_components) { SpvId bool_type = spirv_builder_type_bool(&ctx->builder); if (num_components > 1) return spirv_builder_type_vector(&ctx->builder, bool_type, num_components); assert(num_components == 1); return bool_type; } static SpvId find_image_type(struct ntv_context *ctx, nir_variable *var) { struct hash_entry *he = _mesa_hash_table_search(&ctx->image_types, var); return he ? (intptr_t)he->data : 0; } static SpvScope get_scope(mesa_scope scope) { SpvScope conv[] = { [SCOPE_NONE] = 0, [SCOPE_INVOCATION] = SpvScopeInvocation, [SCOPE_SUBGROUP] = SpvScopeSubgroup, [SCOPE_SHADER_CALL] = SpvScopeShaderCallKHR, [SCOPE_WORKGROUP] = SpvScopeWorkgroup, [SCOPE_QUEUE_FAMILY] = SpvScopeQueueFamily, [SCOPE_DEVICE] = SpvScopeDevice, }; return conv[scope]; } static SpvId block_label(struct ntv_context *ctx, nir_block *block) { assert(block->index < ctx->num_blocks); return ctx->block_ids[block->index]; } static void emit_access_decorations(struct ntv_context *ctx, nir_variable *var, SpvId var_id) { u_foreach_bit(bit, var->data.access) { switch (1 << bit) { case ACCESS_COHERENT: /* SpvDecorationCoherent can't be used with vulkan memory model */ break; case ACCESS_RESTRICT: spirv_builder_emit_decoration(&ctx->builder, var_id, SpvDecorationRestrict); break; case ACCESS_VOLATILE: /* SpvDecorationVolatile can't be used with vulkan memory model */ break; case ACCESS_NON_READABLE: spirv_builder_emit_decoration(&ctx->builder, var_id, SpvDecorationNonReadable); break; case ACCESS_NON_WRITEABLE: spirv_builder_emit_decoration(&ctx->builder, var_id, SpvDecorationNonWritable); break; case ACCESS_NON_UNIFORM: spirv_builder_emit_decoration(&ctx->builder, var_id, SpvDecorationNonUniform); break; case ACCESS_CAN_REORDER: case ACCESS_NON_TEMPORAL: /* no equivalent */ break; default: unreachable("unknown access bit"); } } /* The Simple, GLSL, and Vulkan memory models can assume that aliasing is generally * not present between the memory object declarations. Specifically, the consumer * is free to assume aliasing is not present between memory object declarations, * unless the memory object declarations explicitly indicate they alias. * ... * Applying Restrict is allowed, but has no effect. * ... * Only those memory object declarations decorated with Aliased or AliasedPointer may alias each other. * * - SPIRV 2.18.2 Aliasing * * thus if the variable isn't marked restrict, assume it may alias */ if (!(var->data.access & ACCESS_RESTRICT)) spirv_builder_emit_decoration(&ctx->builder, var_id, SpvDecorationAliased); } static SpvOp get_atomic_op(struct ntv_context *ctx, unsigned bit_size, nir_atomic_op op) { switch (op) { #define ATOMIC_FCAP(NAME) \ do {\ if (bit_size == 16) \ spirv_builder_emit_cap(&ctx->builder, SpvCapabilityAtomicFloat16##NAME##EXT); \ if (bit_size == 32) \ spirv_builder_emit_cap(&ctx->builder, SpvCapabilityAtomicFloat32##NAME##EXT); \ if (bit_size == 64) \ spirv_builder_emit_cap(&ctx->builder, SpvCapabilityAtomicFloat64##NAME##EXT); \ } while (0) case nir_atomic_op_fadd: ATOMIC_FCAP(Add); if (bit_size == 16) spirv_builder_emit_extension(&ctx->builder, "SPV_EXT_shader_atomic_float16_add"); else spirv_builder_emit_extension(&ctx->builder, "SPV_EXT_shader_atomic_float_add"); return SpvOpAtomicFAddEXT; case nir_atomic_op_fmax: ATOMIC_FCAP(MinMax); spirv_builder_emit_extension(&ctx->builder, "SPV_EXT_shader_atomic_float_min_max"); return SpvOpAtomicFMaxEXT; case nir_atomic_op_fmin: ATOMIC_FCAP(MinMax); spirv_builder_emit_extension(&ctx->builder, "SPV_EXT_shader_atomic_float_min_max"); return SpvOpAtomicFMinEXT; case nir_atomic_op_iadd: return SpvOpAtomicIAdd; case nir_atomic_op_umin: return SpvOpAtomicUMin; case nir_atomic_op_imin: return SpvOpAtomicSMin; case nir_atomic_op_umax: return SpvOpAtomicUMax; case nir_atomic_op_imax: return SpvOpAtomicSMax; case nir_atomic_op_iand: return SpvOpAtomicAnd; case nir_atomic_op_ior: return SpvOpAtomicOr; case nir_atomic_op_ixor: return SpvOpAtomicXor; case nir_atomic_op_xchg: return SpvOpAtomicExchange; case nir_atomic_op_cmpxchg: return SpvOpAtomicCompareExchange; default: debug_printf("%s - ", nir_intrinsic_infos[op].name); unreachable("unhandled atomic op"); } return 0; } static SpvId emit_float_const(struct ntv_context *ctx, int bit_size, double value) { assert(bit_size == 16 || bit_size == 32 || bit_size == 64); return spirv_builder_const_float(&ctx->builder, bit_size, value); } static SpvId emit_uint_const(struct ntv_context *ctx, int bit_size, uint64_t value) { assert(bit_size == 8 || bit_size == 16 || bit_size == 32 || bit_size == 64); return spirv_builder_const_uint(&ctx->builder, bit_size, value); } static SpvId emit_int_const(struct ntv_context *ctx, int bit_size, int64_t value) { assert(bit_size == 8 || bit_size == 16 || bit_size == 32 || bit_size == 64); return spirv_builder_const_int(&ctx->builder, bit_size, value); } static SpvId get_fvec_type(struct ntv_context *ctx, unsigned bit_size, unsigned num_components) { assert(bit_size == 16 || bit_size == 32 || bit_size == 64); SpvId float_type = spirv_builder_type_float(&ctx->builder, bit_size); if (num_components > 1) return spirv_builder_type_vector(&ctx->builder, float_type, num_components); assert(num_components == 1); return float_type; } static SpvId get_ivec_type(struct ntv_context *ctx, unsigned bit_size, unsigned num_components) { assert(bit_size == 8 || bit_size == 16 || bit_size == 32 || bit_size == 64); SpvId int_type = spirv_builder_type_int(&ctx->builder, bit_size); if (num_components > 1) return spirv_builder_type_vector(&ctx->builder, int_type, num_components); assert(num_components == 1); return int_type; } static SpvId get_uvec_type(struct ntv_context *ctx, unsigned bit_size, unsigned num_components) { assert(bit_size == 8 || bit_size == 16 || bit_size == 32 || bit_size == 64); SpvId uint_type = spirv_builder_type_uint(&ctx->builder, bit_size); if (num_components > 1) return spirv_builder_type_vector(&ctx->builder, uint_type, num_components); assert(num_components == 1); return uint_type; } static SpvId get_alu_type(struct ntv_context *ctx, nir_alu_type type, unsigned num_components, unsigned bit_size) { if (bit_size == 1) return get_bvec_type(ctx, num_components); type = nir_alu_type_get_base_type(type); switch (nir_alu_type_get_base_type(type)) { case nir_type_bool: return get_bvec_type(ctx, num_components); case nir_type_int: return get_ivec_type(ctx, bit_size, num_components); case nir_type_uint: return get_uvec_type(ctx, bit_size, num_components); case nir_type_float: return get_fvec_type(ctx, bit_size, num_components); default: unreachable("unsupported nir_alu_type"); } } static SpvStorageClass get_storage_class(struct nir_variable *var) { switch (var->data.mode) { case nir_var_function_temp: return SpvStorageClassFunction; case nir_var_mem_push_const: return SpvStorageClassPushConstant; case nir_var_shader_in: return SpvStorageClassInput; case nir_var_shader_out: return SpvStorageClassOutput; case nir_var_uniform: case nir_var_image: return SpvStorageClassUniformConstant; case nir_var_mem_ubo: return SpvStorageClassUniform; case nir_var_mem_ssbo: return SpvStorageClassStorageBuffer; default: unreachable("Unsupported nir_variable_mode"); } return 0; } static SpvId get_def_uvec_type(struct ntv_context *ctx, nir_def *def) { unsigned bit_size = def->bit_size; return get_uvec_type(ctx, bit_size, def->num_components); } static SpvId get_glsl_basetype(struct ntv_context *ctx, enum glsl_base_type type) { switch (type) { case GLSL_TYPE_BOOL: return spirv_builder_type_bool(&ctx->builder); case GLSL_TYPE_FLOAT16: return spirv_builder_type_float(&ctx->builder, 16); case GLSL_TYPE_FLOAT: return spirv_builder_type_float(&ctx->builder, 32); case GLSL_TYPE_INT: return spirv_builder_type_int(&ctx->builder, 32); case GLSL_TYPE_UINT: return spirv_builder_type_uint(&ctx->builder, 32); case GLSL_TYPE_DOUBLE: return spirv_builder_type_float(&ctx->builder, 64); case GLSL_TYPE_INT64: return spirv_builder_type_int(&ctx->builder, 64); case GLSL_TYPE_UINT64: return spirv_builder_type_uint(&ctx->builder, 64); case GLSL_TYPE_UINT16: return spirv_builder_type_uint(&ctx->builder, 16); case GLSL_TYPE_INT16: return spirv_builder_type_int(&ctx->builder, 16); case GLSL_TYPE_INT8: return spirv_builder_type_int(&ctx->builder, 8); case GLSL_TYPE_UINT8: return spirv_builder_type_uint(&ctx->builder, 8); default: unreachable("unknown GLSL type"); } } static SpvId get_glsl_type(struct ntv_context *ctx, const struct glsl_type *type) { assert(type); if (glsl_type_is_scalar(type)) return get_glsl_basetype(ctx, glsl_get_base_type(type)); if (glsl_type_is_vector(type)) return spirv_builder_type_vector(&ctx->builder, get_glsl_basetype(ctx, glsl_get_base_type(type)), glsl_get_vector_elements(type)); if (glsl_type_is_matrix(type)) return spirv_builder_type_matrix(&ctx->builder, spirv_builder_type_vector(&ctx->builder, get_glsl_basetype(ctx, glsl_get_base_type(type)), glsl_get_vector_elements(type)), glsl_get_matrix_columns(type)); /* Aggregate types aren't cached in spirv_builder, so let's cache * them here instead. */ struct hash_entry *entry = _mesa_hash_table_search(ctx->glsl_types, type); if (entry) return (SpvId)(uintptr_t)entry->data; SpvId ret; if (glsl_type_is_array(type)) { SpvId element_type = get_glsl_type(ctx, glsl_get_array_element(type)); if (glsl_type_is_unsized_array(type)) ret = spirv_builder_type_runtime_array(&ctx->builder, element_type); else ret = spirv_builder_type_array(&ctx->builder, element_type, emit_uint_const(ctx, 32, glsl_get_length(type))); uint32_t stride = glsl_get_explicit_stride(type); if (!stride && glsl_type_is_scalar(glsl_get_array_element(type))) { stride = MAX2(glsl_get_bit_size(glsl_get_array_element(type)) / 8, 1); } if (stride) spirv_builder_emit_array_stride(&ctx->builder, ret, stride); } else if (glsl_type_is_struct_or_ifc(type)) { const unsigned length = glsl_get_length(type); /* allocate some SpvId on the stack, falling back to the heap if the array is too long */ SpvId *types, types_stack[16]; if (length <= ARRAY_SIZE(types_stack)) { types = types_stack; } else { types = ralloc_array_size(ctx->mem_ctx, sizeof(SpvId), length); assert(types != NULL); } for (unsigned i = 0; i < glsl_get_length(type); i++) types[i] = get_glsl_type(ctx, glsl_get_struct_field(type, i)); ret = spirv_builder_type_struct(&ctx->builder, types, glsl_get_length(type)); for (unsigned i = 0; i < glsl_get_length(type); i++) { int32_t offset = glsl_get_struct_field_offset(type, i); if (offset >= 0) spirv_builder_emit_member_offset(&ctx->builder, ret, i, offset); } } else unreachable("Unhandled GLSL type"); _mesa_hash_table_insert(ctx->glsl_types, type, (void *)(uintptr_t)ret); return ret; } static void create_scratch_block(struct ntv_context *ctx, unsigned scratch_size, unsigned bit_size) { unsigned idx = bit_size >> 4; SpvId type = spirv_builder_type_uint(&ctx->builder, bit_size); unsigned block_size = scratch_size / (bit_size / 8); assert(block_size); SpvId array = spirv_builder_type_array(&ctx->builder, type, emit_uint_const(ctx, 32, block_size)); spirv_builder_emit_array_stride(&ctx->builder, array, bit_size / 8); SpvId ptr_type = spirv_builder_type_pointer(&ctx->builder, SpvStorageClassPrivate, array); ctx->scratch_block_var[idx] = spirv_builder_emit_var(&ctx->builder, ptr_type, SpvStorageClassPrivate); if (ctx->spirv_1_4_interfaces) { assert(ctx->num_entry_ifaces < ARRAY_SIZE(ctx->entry_ifaces)); ctx->entry_ifaces[ctx->num_entry_ifaces++] = ctx->scratch_block_var[idx]; } } static SpvId get_scratch_block(struct ntv_context *ctx, unsigned bit_size) { unsigned idx = bit_size >> 4; if (!ctx->scratch_block_var[idx]) create_scratch_block(ctx, ctx->nir->scratch_size, bit_size); return ctx->scratch_block_var[idx]; } static void create_shared_block(struct ntv_context *ctx, unsigned bit_size) { unsigned idx = bit_size >> 4; SpvId type = spirv_builder_type_uint(&ctx->builder, bit_size); SpvId array; assert(gl_shader_stage_is_compute(ctx->nir->info.stage)); if (ctx->nir->info.cs.has_variable_shared_mem) { assert(ctx->shared_mem_size); SpvId const_shared_size = emit_uint_const(ctx, 32, ctx->nir->info.shared_size); SpvId shared_mem_size = spirv_builder_emit_triop(&ctx->builder, SpvOpSpecConstantOp, spirv_builder_type_uint(&ctx->builder, 32), SpvOpIAdd, const_shared_size, ctx->shared_mem_size); shared_mem_size = spirv_builder_emit_triop(&ctx->builder, SpvOpSpecConstantOp, spirv_builder_type_uint(&ctx->builder, 32), SpvOpUDiv, shared_mem_size, emit_uint_const(ctx, 32, bit_size / 8)); array = spirv_builder_type_array(&ctx->builder, type, shared_mem_size); } else { unsigned block_size = ctx->nir->info.shared_size / (bit_size / 8); assert(block_size); array = spirv_builder_type_array(&ctx->builder, type, emit_uint_const(ctx, 32, block_size)); } ctx->shared_block_arr_type[idx] = array; spirv_builder_emit_array_stride(&ctx->builder, array, bit_size / 8); /* Create wrapper struct for Block, Offset and Aliased decorations. */ SpvId block = spirv_builder_type_struct(&ctx->builder, &array, 1); SpvId ptr_type = spirv_builder_type_pointer(&ctx->builder, SpvStorageClassWorkgroup, block); ctx->shared_block_var[idx] = spirv_builder_emit_var(&ctx->builder, ptr_type, SpvStorageClassWorkgroup); if (ctx->spirv_1_4_interfaces) { assert(ctx->num_entry_ifaces < ARRAY_SIZE(ctx->entry_ifaces)); ctx->entry_ifaces[ctx->num_entry_ifaces++] = ctx->shared_block_var[idx]; } /* Alias our shared memory blocks */ if (ctx->sinfo->have_workgroup_memory_explicit_layout) { spirv_builder_emit_member_offset(&ctx->builder, block, 0, 0); spirv_builder_emit_decoration(&ctx->builder, block, SpvDecorationBlock); spirv_builder_emit_decoration(&ctx->builder, ctx->shared_block_var[idx], SpvDecorationAliased); } } static SpvId get_shared_block(struct ntv_context *ctx, unsigned bit_size) { unsigned idx = bit_size >> 4; if (!ctx->shared_block_var[idx]) create_shared_block(ctx, bit_size); if (ctx->sinfo->have_workgroup_memory_explicit_layout) { spirv_builder_emit_extension(&ctx->builder, "SPV_KHR_workgroup_memory_explicit_layout"); spirv_builder_emit_cap(&ctx->builder, SpvCapabilityWorkgroupMemoryExplicitLayoutKHR); if (ctx->shared_block_var[0]) spirv_builder_emit_cap(&ctx->builder, SpvCapabilityWorkgroupMemoryExplicitLayout8BitAccessKHR); if (ctx->shared_block_var[1]) spirv_builder_emit_cap(&ctx->builder, SpvCapabilityWorkgroupMemoryExplicitLayout16BitAccessKHR); } SpvId ptr_type = spirv_builder_type_pointer(&ctx->builder, SpvStorageClassWorkgroup, ctx->shared_block_arr_type[idx]); SpvId zero = emit_uint_const(ctx, 32, 0); return spirv_builder_emit_access_chain(&ctx->builder, ptr_type, ctx->shared_block_var[idx], &zero, 1); } #define HANDLE_EMIT_BUILTIN(SLOT, BUILTIN) \ case VARYING_SLOT_##SLOT: \ spirv_builder_emit_builtin(&ctx->builder, var_id, SpvBuiltIn##BUILTIN); \ break static SpvId input_var_init(struct ntv_context *ctx, struct nir_variable *var) { SpvId var_type = get_glsl_type(ctx, var->type); SpvStorageClass sc = get_storage_class(var); if (sc == SpvStorageClassPushConstant) spirv_builder_emit_decoration(&ctx->builder, var_type, SpvDecorationBlock); SpvId pointer_type = spirv_builder_type_pointer(&ctx->builder, sc, var_type); SpvId var_id = spirv_builder_emit_var(&ctx->builder, pointer_type, sc); if (var->name) spirv_builder_emit_name(&ctx->builder, var_id, var->name); if (var->data.mode == nir_var_mem_push_const) { ctx->push_const_var = var_id; if (ctx->spirv_1_4_interfaces) { assert(ctx->num_entry_ifaces < ARRAY_SIZE(ctx->entry_ifaces)); ctx->entry_ifaces[ctx->num_entry_ifaces++] = var_id; } } return var_id; } static void emit_interpolation(struct ntv_context *ctx, SpvId var_id, enum glsl_interp_mode mode) { switch (mode) { case INTERP_MODE_NONE: case INTERP_MODE_SMOOTH: /* XXX spirv doesn't seem to have anything for this */ break; case INTERP_MODE_FLAT: spirv_builder_emit_decoration(&ctx->builder, var_id, SpvDecorationFlat); break; case INTERP_MODE_EXPLICIT: spirv_builder_emit_decoration(&ctx->builder, var_id, SpvDecorationExplicitInterpAMD); break; case INTERP_MODE_NOPERSPECTIVE: spirv_builder_emit_decoration(&ctx->builder, var_id, SpvDecorationNoPerspective); break; default: unreachable("unknown interpolation value"); } } static void emit_input(struct ntv_context *ctx, struct nir_variable *var) { SpvId var_id = input_var_init(ctx, var); if (ctx->stage == MESA_SHADER_VERTEX) spirv_builder_emit_location(&ctx->builder, var_id, var->data.driver_location); else if (ctx->stage == MESA_SHADER_FRAGMENT) { switch (var->data.location) { HANDLE_EMIT_BUILTIN(POS, FragCoord); HANDLE_EMIT_BUILTIN(LAYER, Layer); HANDLE_EMIT_BUILTIN(PRIMITIVE_ID, PrimitiveId); HANDLE_EMIT_BUILTIN(CLIP_DIST0, ClipDistance); HANDLE_EMIT_BUILTIN(CULL_DIST0, CullDistance); HANDLE_EMIT_BUILTIN(VIEWPORT, ViewportIndex); HANDLE_EMIT_BUILTIN(FACE, FrontFacing); default: spirv_builder_emit_location(&ctx->builder, var_id, var->data.driver_location); } if (var->data.centroid) spirv_builder_emit_decoration(&ctx->builder, var_id, SpvDecorationCentroid); else if (var->data.sample) spirv_builder_emit_decoration(&ctx->builder, var_id, SpvDecorationSample); emit_interpolation(ctx, var_id, var->data.interpolation); } else if (ctx->stage < MESA_SHADER_FRAGMENT) { switch (var->data.location) { HANDLE_EMIT_BUILTIN(POS, Position); HANDLE_EMIT_BUILTIN(PSIZ, PointSize); HANDLE_EMIT_BUILTIN(LAYER, Layer); HANDLE_EMIT_BUILTIN(PRIMITIVE_ID, PrimitiveId); HANDLE_EMIT_BUILTIN(CULL_DIST0, CullDistance); HANDLE_EMIT_BUILTIN(VIEWPORT, ViewportIndex); HANDLE_EMIT_BUILTIN(TESS_LEVEL_OUTER, TessLevelOuter); HANDLE_EMIT_BUILTIN(TESS_LEVEL_INNER, TessLevelInner); case VARYING_SLOT_CLIP_DIST0: assert(glsl_type_is_array(var->type)); spirv_builder_emit_builtin(&ctx->builder, var_id, SpvBuiltInClipDistance); break; default: spirv_builder_emit_location(&ctx->builder, var_id, var->data.driver_location); } } if (var->data.location_frac) spirv_builder_emit_component(&ctx->builder, var_id, var->data.location_frac); if (var->data.patch) spirv_builder_emit_decoration(&ctx->builder, var_id, SpvDecorationPatch); _mesa_hash_table_insert(ctx->vars, var, (void *)(intptr_t)var_id); assert(ctx->num_entry_ifaces < ARRAY_SIZE(ctx->entry_ifaces)); ctx->entry_ifaces[ctx->num_entry_ifaces++] = var_id; } static void emit_output(struct ntv_context *ctx, struct nir_variable *var) { SpvId var_type = get_glsl_type(ctx, var->type); /* SampleMask is always an array in spirv */ if (ctx->stage == MESA_SHADER_FRAGMENT && var->data.location == FRAG_RESULT_SAMPLE_MASK) ctx->sample_mask_type = var_type = spirv_builder_type_array(&ctx->builder, var_type, emit_uint_const(ctx, 32, 1)); SpvId pointer_type = spirv_builder_type_pointer(&ctx->builder, SpvStorageClassOutput, var_type); SpvId var_id = spirv_builder_emit_var(&ctx->builder, pointer_type, SpvStorageClassOutput); if (var->name) spirv_builder_emit_name(&ctx->builder, var_id, var->name); if (var->data.precision == GLSL_PRECISION_MEDIUM || var->data.precision == GLSL_PRECISION_LOW) { spirv_builder_emit_decoration(&ctx->builder, var_id, SpvDecorationRelaxedPrecision); } if (ctx->stage != MESA_SHADER_FRAGMENT) { switch (var->data.location) { HANDLE_EMIT_BUILTIN(POS, Position); HANDLE_EMIT_BUILTIN(PSIZ, PointSize); HANDLE_EMIT_BUILTIN(LAYER, Layer); HANDLE_EMIT_BUILTIN(PRIMITIVE_ID, PrimitiveId); HANDLE_EMIT_BUILTIN(CLIP_DIST0, ClipDistance); HANDLE_EMIT_BUILTIN(CULL_DIST0, CullDistance); HANDLE_EMIT_BUILTIN(VIEWPORT, ViewportIndex); HANDLE_EMIT_BUILTIN(TESS_LEVEL_OUTER, TessLevelOuter); HANDLE_EMIT_BUILTIN(TESS_LEVEL_INNER, TessLevelInner); default: /* non-xfb psiz output will have location -1 */ if (var->data.location >= 0) spirv_builder_emit_location(&ctx->builder, var_id, var->data.driver_location); } emit_interpolation(ctx, var_id, var->data.interpolation); } else { if (var->data.location >= FRAG_RESULT_DATA0) { spirv_builder_emit_location(&ctx->builder, var_id, var->data.location - FRAG_RESULT_DATA0); spirv_builder_emit_index(&ctx->builder, var_id, var->data.index); } else { switch (var->data.location) { case FRAG_RESULT_COLOR: unreachable("gl_FragColor should be lowered by now"); case FRAG_RESULT_DEPTH: spirv_builder_emit_builtin(&ctx->builder, var_id, SpvBuiltInFragDepth); break; case FRAG_RESULT_SAMPLE_MASK: spirv_builder_emit_builtin(&ctx->builder, var_id, SpvBuiltInSampleMask); break; case FRAG_RESULT_STENCIL: spirv_builder_emit_builtin(&ctx->builder, var_id, SpvBuiltInFragStencilRefEXT); break; default: spirv_builder_emit_location(&ctx->builder, var_id, var->data.location); spirv_builder_emit_index(&ctx->builder, var_id, var->data.index); } } if (var->data.sample) spirv_builder_emit_decoration(&ctx->builder, var_id, SpvDecorationSample); } if (var->data.location_frac) spirv_builder_emit_component(&ctx->builder, var_id, var->data.location_frac); if (var->data.patch) spirv_builder_emit_decoration(&ctx->builder, var_id, SpvDecorationPatch); if (var->data.explicit_xfb_buffer && ctx->nir->xfb_info) { spirv_builder_emit_offset(&ctx->builder, var_id, var->data.offset); spirv_builder_emit_xfb_buffer(&ctx->builder, var_id, var->data.xfb.buffer); spirv_builder_emit_xfb_stride(&ctx->builder, var_id, var->data.xfb.stride); if (var->data.stream) spirv_builder_emit_stream(&ctx->builder, var_id, var->data.stream); } _mesa_hash_table_insert(ctx->vars, var, (void *)(intptr_t)var_id); assert(ctx->num_entry_ifaces < ARRAY_SIZE(ctx->entry_ifaces)); ctx->entry_ifaces[ctx->num_entry_ifaces++] = var_id; } static void emit_shader_temp(struct ntv_context *ctx, struct nir_variable *var) { SpvId var_type = get_glsl_type(ctx, var->type); SpvId pointer_type = spirv_builder_type_pointer(&ctx->builder, SpvStorageClassPrivate, var_type); SpvId var_id = spirv_builder_emit_var(&ctx->builder, pointer_type, SpvStorageClassPrivate); if (var->name) spirv_builder_emit_name(&ctx->builder, var_id, var->name); _mesa_hash_table_insert(ctx->vars, var, (void *)(intptr_t)var_id); assert(ctx->num_entry_ifaces < ARRAY_SIZE(ctx->entry_ifaces)); ctx->entry_ifaces[ctx->num_entry_ifaces++] = var_id; } static void emit_temp(struct ntv_context *ctx, struct nir_variable *var) { SpvId var_type = get_glsl_type(ctx, var->type); SpvId pointer_type = spirv_builder_type_pointer(&ctx->builder, SpvStorageClassFunction, var_type); SpvId var_id = spirv_builder_emit_var(&ctx->builder, pointer_type, SpvStorageClassFunction); if (var->name) spirv_builder_emit_name(&ctx->builder, var_id, var->name); _mesa_hash_table_insert(ctx->vars, var, (void *)(intptr_t)var_id); } static SpvDim type_to_dim(enum glsl_sampler_dim gdim, bool *is_ms) { *is_ms = false; switch (gdim) { case GLSL_SAMPLER_DIM_1D: return SpvDim1D; case GLSL_SAMPLER_DIM_2D: return SpvDim2D; case GLSL_SAMPLER_DIM_3D: return SpvDim3D; case GLSL_SAMPLER_DIM_CUBE: return SpvDimCube; case GLSL_SAMPLER_DIM_RECT: return SpvDim2D; case GLSL_SAMPLER_DIM_BUF: return SpvDimBuffer; case GLSL_SAMPLER_DIM_EXTERNAL: return SpvDim2D; /* seems dodgy... */ case GLSL_SAMPLER_DIM_MS: *is_ms = true; return SpvDim2D; case GLSL_SAMPLER_DIM_SUBPASS_MS: *is_ms = true; return SpvDimSubpassData; case GLSL_SAMPLER_DIM_SUBPASS: return SpvDimSubpassData; default: fprintf(stderr, "unknown sampler type %d\n", gdim); break; } return SpvDim2D; } static inline SpvImageFormat get_shader_image_format(enum pipe_format format) { switch (format) { case PIPE_FORMAT_R32G32B32A32_FLOAT: return SpvImageFormatRgba32f; case PIPE_FORMAT_R16G16B16A16_FLOAT: return SpvImageFormatRgba16f; case PIPE_FORMAT_R32_FLOAT: return SpvImageFormatR32f; case PIPE_FORMAT_R8G8B8A8_UNORM: return SpvImageFormatRgba8; case PIPE_FORMAT_R8G8B8A8_SNORM: return SpvImageFormatRgba8Snorm; case PIPE_FORMAT_R32G32B32A32_SINT: return SpvImageFormatRgba32i; case PIPE_FORMAT_R16G16B16A16_SINT: return SpvImageFormatRgba16i; case PIPE_FORMAT_R8G8B8A8_SINT: return SpvImageFormatRgba8i; case PIPE_FORMAT_R32_SINT: return SpvImageFormatR32i; case PIPE_FORMAT_R32G32B32A32_UINT: return SpvImageFormatRgba32ui; case PIPE_FORMAT_R16G16B16A16_UINT: return SpvImageFormatRgba16ui; case PIPE_FORMAT_R8G8B8A8_UINT: return SpvImageFormatRgba8ui; case PIPE_FORMAT_R32_UINT: return SpvImageFormatR32ui; default: return SpvImageFormatUnknown; } } static inline SpvImageFormat get_extended_image_format(enum pipe_format format) { switch (format) { case PIPE_FORMAT_R32G32_FLOAT: return SpvImageFormatRg32f; case PIPE_FORMAT_R16G16_FLOAT: return SpvImageFormatRg16f; case PIPE_FORMAT_R11G11B10_FLOAT: return SpvImageFormatR11fG11fB10f; case PIPE_FORMAT_R16_FLOAT: return SpvImageFormatR16f; case PIPE_FORMAT_R16G16B16A16_UNORM: return SpvImageFormatRgba16; case PIPE_FORMAT_R10G10B10A2_UNORM: return SpvImageFormatRgb10A2; case PIPE_FORMAT_R16G16_UNORM: return SpvImageFormatRg16; case PIPE_FORMAT_R8G8_UNORM: return SpvImageFormatRg8; case PIPE_FORMAT_R16_UNORM: return SpvImageFormatR16; case PIPE_FORMAT_R8_UNORM: return SpvImageFormatR8; case PIPE_FORMAT_R16G16B16A16_SNORM: return SpvImageFormatRgba16Snorm; case PIPE_FORMAT_R16G16_SNORM: return SpvImageFormatRg16Snorm; case PIPE_FORMAT_R8G8_SNORM: return SpvImageFormatRg8Snorm; case PIPE_FORMAT_R16_SNORM: return SpvImageFormatR16Snorm; case PIPE_FORMAT_R8_SNORM: return SpvImageFormatR8Snorm; case PIPE_FORMAT_R32G32_SINT: return SpvImageFormatRg32i; case PIPE_FORMAT_R16G16_SINT: return SpvImageFormatRg16i; case PIPE_FORMAT_R8G8_SINT: return SpvImageFormatRg8i; case PIPE_FORMAT_R16_SINT: return SpvImageFormatR16i; case PIPE_FORMAT_R8_SINT: return SpvImageFormatR8i; case PIPE_FORMAT_R10G10B10A2_UINT: return SpvImageFormatRgb10a2ui; case PIPE_FORMAT_R32G32_UINT: return SpvImageFormatRg32ui; case PIPE_FORMAT_R16G16_UINT: return SpvImageFormatRg16ui; case PIPE_FORMAT_R8G8_UINT: return SpvImageFormatRg8ui; case PIPE_FORMAT_R16_UINT: return SpvImageFormatR16ui; case PIPE_FORMAT_R8_UINT: return SpvImageFormatR8ui; default: return SpvImageFormatUnknown; } } static inline SpvImageFormat get_image_format(struct ntv_context *ctx, enum pipe_format format) { /* always supported */ if (format == PIPE_FORMAT_NONE) return SpvImageFormatUnknown; SpvImageFormat ret = get_shader_image_format(format); if (ret != SpvImageFormatUnknown) { /* requires the shader-cap, but we already emit that */ return ret; } ret = get_extended_image_format(format); assert(ret != SpvImageFormatUnknown); spirv_builder_emit_cap(&ctx->builder, SpvCapabilityStorageImageExtendedFormats); return ret; } static SpvId get_bare_image_type(struct ntv_context *ctx, struct nir_variable *var, bool is_sampler) { const struct glsl_type *type = glsl_without_array(var->type); bool is_ms; if (var->data.fb_fetch_output) { spirv_builder_emit_cap(&ctx->builder, SpvCapabilityInputAttachment); } else if (!is_sampler && !var->data.image.format) { if (!(var->data.access & ACCESS_NON_WRITEABLE)) spirv_builder_emit_cap(&ctx->builder, SpvCapabilityStorageImageWriteWithoutFormat); if (!(var->data.access & ACCESS_NON_READABLE)) spirv_builder_emit_cap(&ctx->builder, SpvCapabilityStorageImageReadWithoutFormat); } SpvDim dimension = type_to_dim(glsl_get_sampler_dim(type), &is_ms); if (dimension == SpvDim1D) { if (is_sampler) spirv_builder_emit_cap(&ctx->builder, SpvCapabilitySampled1D); else spirv_builder_emit_cap(&ctx->builder, SpvCapabilityImage1D); } if (dimension == SpvDimBuffer) { if (is_sampler) spirv_builder_emit_cap(&ctx->builder, SpvCapabilitySampledBuffer); else spirv_builder_emit_cap(&ctx->builder, SpvCapabilityImageBuffer); } bool arrayed = glsl_sampler_type_is_array(type); if (dimension == SpvDimCube && arrayed) spirv_builder_emit_cap(&ctx->builder, SpvCapabilityImageCubeArray); if (arrayed && !is_sampler && is_ms) spirv_builder_emit_cap(&ctx->builder, SpvCapabilityImageMSArray); SpvId result_type = get_glsl_basetype(ctx, glsl_get_sampler_result_type(type)); return spirv_builder_type_image(&ctx->builder, result_type, dimension, false, arrayed, is_ms, is_sampler ? 1 : 2, get_image_format(ctx, var->data.image.format)); } static SpvId get_image_type(struct ntv_context *ctx, struct nir_variable *var, bool is_sampler, bool is_buffer) { SpvId image_type = get_bare_image_type(ctx, var, is_sampler); return is_sampler && ctx->stage != MESA_SHADER_KERNEL && !is_buffer ? spirv_builder_type_sampled_image(&ctx->builder, image_type) : image_type; } static SpvId emit_image(struct ntv_context *ctx, struct nir_variable *var, SpvId image_type) { if (var->data.bindless) return 0; const struct glsl_type *type = glsl_without_array(var->type); bool is_sampler = glsl_type_is_sampler(type); bool is_buffer = glsl_get_sampler_dim(type) == GLSL_SAMPLER_DIM_BUF; SpvId var_type = is_sampler && ctx->stage != MESA_SHADER_KERNEL && !is_buffer ? spirv_builder_type_sampled_image(&ctx->builder, image_type) : image_type; bool mediump = (var->data.precision == GLSL_PRECISION_MEDIUM || var->data.precision == GLSL_PRECISION_LOW); int index = var->data.driver_location; assert(!find_image_type(ctx, var)); if (glsl_type_is_array(var->type)) { var_type = spirv_builder_type_array(&ctx->builder, var_type, emit_uint_const(ctx, 32, glsl_get_aoa_size(var->type))); spirv_builder_emit_array_stride(&ctx->builder, var_type, sizeof(void*)); } SpvId pointer_type = spirv_builder_type_pointer(&ctx->builder, SpvStorageClassUniformConstant, var_type); SpvId var_id = spirv_builder_emit_var(&ctx->builder, pointer_type, SpvStorageClassUniformConstant); if (mediump) { spirv_builder_emit_decoration(&ctx->builder, var_id, SpvDecorationRelaxedPrecision); } if (var->name) spirv_builder_emit_name(&ctx->builder, var_id, var->name); if (var->data.fb_fetch_output) spirv_builder_emit_input_attachment_index(&ctx->builder, var_id, var->data.index); _mesa_hash_table_insert(ctx->vars, var, (void *)(intptr_t)var_id); if (is_sampler) { if (var->data.descriptor_set == ctx->bindless_set_idx) { assert(!ctx->bindless_samplers[index]); ctx->bindless_samplers[index] = var_id; } else { assert(!ctx->samplers[index]); ctx->samplers[index] = var_id; } } else { assert(!ctx->images[index]); ctx->images[index] = var_id; emit_access_decorations(ctx, var, var_id); } _mesa_hash_table_insert(&ctx->image_types, var, (void *)(intptr_t)image_type); if (ctx->spirv_1_4_interfaces) { assert(ctx->num_entry_ifaces < ARRAY_SIZE(ctx->entry_ifaces)); ctx->entry_ifaces[ctx->num_entry_ifaces++] = var_id; } spirv_builder_emit_descriptor_set(&ctx->builder, var_id, var->data.descriptor_set); spirv_builder_emit_binding(&ctx->builder, var_id, var->data.binding); return var_id; } static void emit_sampler(struct ntv_context *ctx, nir_variable *var) { SpvId type = spirv_builder_type_sampler(&ctx->builder); SpvId pointer_type = spirv_builder_type_pointer(&ctx->builder, SpvStorageClassUniformConstant, type); SpvId var_id = spirv_builder_emit_var(&ctx->builder, pointer_type, SpvStorageClassUniformConstant); char buf[128]; snprintf(buf, sizeof(buf), "sampler_%u", var->data.driver_location); spirv_builder_emit_name(&ctx->builder, var_id, buf); spirv_builder_emit_descriptor_set(&ctx->builder, var_id, var->data.descriptor_set); spirv_builder_emit_binding(&ctx->builder, var_id, var->data.driver_location); _mesa_hash_table_insert(ctx->vars, var, (void *)(intptr_t)var_id); if (ctx->spirv_1_4_interfaces) { assert(ctx->num_entry_ifaces < ARRAY_SIZE(ctx->entry_ifaces)); ctx->entry_ifaces[ctx->num_entry_ifaces++] = var_id; } } static SpvId get_sized_uint_array_type(struct ntv_context *ctx, unsigned array_size, unsigned bitsize) { SpvId array_length = emit_uint_const(ctx, 32, array_size); SpvId array_type = spirv_builder_type_array(&ctx->builder, get_uvec_type(ctx, bitsize, 1), array_length); spirv_builder_emit_array_stride(&ctx->builder, array_type, bitsize / 8); return array_type; } /* get array */ static SpvId get_bo_array_type(struct ntv_context *ctx, struct nir_variable *var) { struct hash_entry *he = _mesa_hash_table_search(ctx->bo_array_types, var); if (he) return (SpvId)(uintptr_t)he->data; unsigned bitsize = glsl_get_bit_size(glsl_get_array_element(glsl_get_struct_field(glsl_without_array(var->type), 0))); assert(bitsize); SpvId array_type; const struct glsl_type *type = glsl_without_array(var->type); const struct glsl_type *first_type = glsl_get_struct_field(type, 0); if (!glsl_type_is_unsized_array(first_type)) { uint32_t array_size = glsl_get_length(first_type); assert(array_size); return get_sized_uint_array_type(ctx, array_size, bitsize); } SpvId uint_type = spirv_builder_type_uint(&ctx->builder, bitsize); array_type = spirv_builder_type_runtime_array(&ctx->builder, uint_type); spirv_builder_emit_array_stride(&ctx->builder, array_type, bitsize / 8); return array_type; } /* get array */ static SpvId get_bo_struct_type(struct ntv_context *ctx, struct nir_variable *var) { struct hash_entry *he = _mesa_hash_table_search(ctx->bo_struct_types, var); if (he) return (SpvId)(uintptr_t)he->data; const struct glsl_type *bare_type = glsl_without_array(var->type); unsigned bitsize = glsl_get_bit_size(glsl_get_array_element(glsl_get_struct_field(bare_type, 0))); SpvId array_type = get_bo_array_type(ctx, var); _mesa_hash_table_insert(ctx->bo_array_types, var, (void *)(uintptr_t)array_type); bool ssbo = var->data.mode == nir_var_mem_ssbo; // wrap UBO-array in a struct SpvId runtime_array = 0; if (ssbo && glsl_get_length(bare_type) > 1) { const struct glsl_type *last_member = glsl_get_struct_field(bare_type, glsl_get_length(bare_type) - 1); if (glsl_type_is_unsized_array(last_member)) { runtime_array = spirv_builder_type_runtime_array(&ctx->builder, get_uvec_type(ctx, bitsize, 1)); spirv_builder_emit_array_stride(&ctx->builder, runtime_array, glsl_get_explicit_stride(last_member)); } } SpvId types[] = {array_type, runtime_array}; SpvId struct_type = spirv_builder_type_struct(&ctx->builder, types, 1 + !!runtime_array); if (var->name) { char struct_name[100]; snprintf(struct_name, sizeof(struct_name), "struct_%s", var->name); spirv_builder_emit_name(&ctx->builder, struct_type, struct_name); } spirv_builder_emit_decoration(&ctx->builder, struct_type, SpvDecorationBlock); spirv_builder_emit_member_offset(&ctx->builder, struct_type, 0, 0); if (runtime_array) spirv_builder_emit_member_offset(&ctx->builder, struct_type, 1, 0); return struct_type; } static void emit_bo(struct ntv_context *ctx, struct nir_variable *var, bool aliased) { unsigned bitsize = glsl_get_bit_size(glsl_get_array_element(glsl_get_struct_field(glsl_without_array(var->type), 0))); bool ssbo = var->data.mode == nir_var_mem_ssbo; SpvId struct_type = get_bo_struct_type(ctx, var); _mesa_hash_table_insert(ctx->bo_struct_types, var, (void *)(uintptr_t)struct_type); SpvId array_length = emit_uint_const(ctx, 32, glsl_get_length(var->type)); SpvId array_type = spirv_builder_type_array(&ctx->builder, struct_type, array_length); SpvId pointer_type = spirv_builder_type_pointer(&ctx->builder, ssbo ? SpvStorageClassStorageBuffer : SpvStorageClassUniform, array_type); SpvId var_id = spirv_builder_emit_var(&ctx->builder, pointer_type, ssbo ? SpvStorageClassStorageBuffer : SpvStorageClassUniform); if (var->name) spirv_builder_emit_name(&ctx->builder, var_id, var->name); if (aliased) spirv_builder_emit_decoration(&ctx->builder, var_id, SpvDecorationAliased); unsigned idx = bitsize >> 4; assert(idx < ARRAY_SIZE(ctx->ssbos)); if (ssbo) { assert(!ctx->ssbos[idx]); ctx->ssbos[idx] = var_id; if (bitsize == 32) ctx->ssbo_vars = var; } else { assert(!ctx->ubos[var->data.driver_location][idx]); ctx->ubos[var->data.driver_location][idx] = var_id; ctx->ubo_vars[var->data.driver_location] = var; } if (ctx->spirv_1_4_interfaces) { assert(ctx->num_entry_ifaces < ARRAY_SIZE(ctx->entry_ifaces)); ctx->entry_ifaces[ctx->num_entry_ifaces++] = var_id; } _mesa_hash_table_insert(ctx->vars, var, (void *)(intptr_t)var_id); spirv_builder_emit_descriptor_set(&ctx->builder, var_id, var->data.descriptor_set); spirv_builder_emit_binding(&ctx->builder, var_id, var->data.binding); } static SpvId get_vec_from_bit_size(struct ntv_context *ctx, uint32_t bit_size, uint32_t num_components) { if (bit_size == 1) return get_bvec_type(ctx, num_components); return get_uvec_type(ctx, bit_size, num_components); } static SpvId get_src_ssa(struct ntv_context *ctx, const nir_def *ssa, nir_alu_type *atype) { assert(ssa->index < ctx->num_defs); assert(ctx->defs[ssa->index] != 0); *atype = ctx->def_types[ssa->index]; return ctx->defs[ssa->index]; } static void init_reg(struct ntv_context *ctx, nir_intrinsic_instr *decl, nir_alu_type atype) { unsigned index = decl->def.index; unsigned num_components = nir_intrinsic_num_components(decl); unsigned bit_size = nir_intrinsic_bit_size(decl); if (ctx->defs[index]) return; SpvId type = get_alu_type(ctx, atype, num_components, bit_size); SpvId pointer_type = spirv_builder_type_pointer(&ctx->builder, SpvStorageClassFunction, type); SpvId var = spirv_builder_emit_var(&ctx->builder, pointer_type, SpvStorageClassFunction); ctx->defs[index] = var; ctx->def_types[index] = nir_alu_type_get_base_type(atype); } static SpvId get_src(struct ntv_context *ctx, nir_src *src, nir_alu_type *atype) { return get_src_ssa(ctx, src->ssa, atype); } static SpvId get_alu_src_raw(struct ntv_context *ctx, nir_alu_instr *alu, unsigned src, nir_alu_type *atype) { SpvId def = get_src(ctx, &alu->src[src].src, atype); unsigned used_channels = 0; bool need_swizzle = false; for (unsigned i = 0; i < NIR_MAX_VEC_COMPONENTS; i++) { if (!nir_alu_instr_channel_used(alu, src, i)) continue; used_channels++; if (alu->src[src].swizzle[i] != i) need_swizzle = true; } assert(used_channels != 0); unsigned live_channels = nir_src_num_components(alu->src[src].src); if (used_channels != live_channels) need_swizzle = true; if (!need_swizzle) return def; int bit_size = nir_src_bit_size(alu->src[src].src); SpvId raw_type = get_alu_type(ctx, *atype, 1, bit_size); if (used_channels == 1) { uint32_t indices[] = { alu->src[src].swizzle[0] }; return spirv_builder_emit_composite_extract(&ctx->builder, raw_type, def, indices, ARRAY_SIZE(indices)); } else if (live_channels == 1) { SpvId raw_vec_type = spirv_builder_type_vector(&ctx->builder, raw_type, used_channels); SpvId constituents[NIR_MAX_VEC_COMPONENTS] = {0}; for (unsigned i = 0; i < used_channels; ++i) constituents[i] = def; return spirv_builder_emit_composite_construct(&ctx->builder, raw_vec_type, constituents, used_channels); } else { SpvId raw_vec_type = spirv_builder_type_vector(&ctx->builder, raw_type, used_channels); uint32_t components[NIR_MAX_VEC_COMPONENTS] = {0}; size_t num_components = 0; for (unsigned i = 0; i < NIR_MAX_VEC_COMPONENTS; i++) { if (!nir_alu_instr_channel_used(alu, src, i)) continue; components[num_components++] = alu->src[src].swizzle[i]; } return spirv_builder_emit_vector_shuffle(&ctx->builder, raw_vec_type, def, def, components, num_components); } } static SpvId emit_select(struct ntv_context *ctx, SpvId type, SpvId cond, SpvId if_true, SpvId if_false) { return emit_triop(ctx, SpvOpSelect, type, cond, if_true, if_false); } static SpvId emit_bitcast(struct ntv_context *ctx, SpvId type, SpvId value) { return emit_unop(ctx, SpvOpBitcast, type, value); } static SpvId bitcast_to_uvec(struct ntv_context *ctx, SpvId value, unsigned bit_size, unsigned num_components) { SpvId type = get_uvec_type(ctx, bit_size, num_components); return emit_bitcast(ctx, type, value); } static SpvId bitcast_to_ivec(struct ntv_context *ctx, SpvId value, unsigned bit_size, unsigned num_components) { SpvId type = get_ivec_type(ctx, bit_size, num_components); return emit_bitcast(ctx, type, value); } static SpvId bitcast_to_fvec(struct ntv_context *ctx, SpvId value, unsigned bit_size, unsigned num_components) { SpvId type = get_fvec_type(ctx, bit_size, num_components); return emit_bitcast(ctx, type, value); } static SpvId cast_src_to_type(struct ntv_context *ctx, SpvId value, nir_src src, nir_alu_type atype) { atype = nir_alu_type_get_base_type(atype); unsigned num_components = nir_src_num_components(src); unsigned bit_size = nir_src_bit_size(src); return emit_bitcast(ctx, get_alu_type(ctx, atype, num_components, bit_size), value); } static void store_def(struct ntv_context *ctx, unsigned def_index, SpvId result, nir_alu_type type) { assert(result != 0); assert(def_index < ctx->num_defs); ctx->def_types[def_index] = nir_alu_type_get_base_type(type); ctx->defs[def_index] = result; } static SpvId emit_unop(struct ntv_context *ctx, SpvOp op, SpvId type, SpvId src) { return spirv_builder_emit_unop(&ctx->builder, op, type, src); } static SpvId emit_atomic(struct ntv_context *ctx, SpvId op, SpvId type, SpvId src0, SpvId src1, SpvId src2) { if (op == SpvOpAtomicLoad) return spirv_builder_emit_triop(&ctx->builder, op, type, src0, emit_uint_const(ctx, 32, SpvScopeDevice), emit_uint_const(ctx, 32, 0)); if (op == SpvOpAtomicCompareExchange) return spirv_builder_emit_hexop(&ctx->builder, op, type, src0, emit_uint_const(ctx, 32, SpvScopeDevice), emit_uint_const(ctx, 32, 0), emit_uint_const(ctx, 32, 0), /* these params are intentionally swapped */ src2, src1); return spirv_builder_emit_quadop(&ctx->builder, op, type, src0, emit_uint_const(ctx, 32, SpvScopeDevice), emit_uint_const(ctx, 32, 0), src1); } static SpvId emit_binop(struct ntv_context *ctx, SpvOp op, SpvId type, SpvId src0, SpvId src1) { return spirv_builder_emit_binop(&ctx->builder, op, type, src0, src1); } static SpvId emit_triop(struct ntv_context *ctx, SpvOp op, SpvId type, SpvId src0, SpvId src1, SpvId src2) { return spirv_builder_emit_triop(&ctx->builder, op, type, src0, src1, src2); } static SpvId emit_builtin_unop(struct ntv_context *ctx, enum GLSLstd450 op, SpvId type, SpvId src) { SpvId args[] = { src }; return spirv_builder_emit_ext_inst(&ctx->builder, type, ctx->GLSL_std_450, op, args, ARRAY_SIZE(args)); } static SpvId emit_builtin_binop(struct ntv_context *ctx, enum GLSLstd450 op, SpvId type, SpvId src0, SpvId src1) { SpvId args[] = { src0, src1 }; return spirv_builder_emit_ext_inst(&ctx->builder, type, ctx->GLSL_std_450, op, args, ARRAY_SIZE(args)); } static SpvId emit_builtin_triop(struct ntv_context *ctx, enum GLSLstd450 op, SpvId type, SpvId src0, SpvId src1, SpvId src2) { SpvId args[] = { src0, src1, src2 }; return spirv_builder_emit_ext_inst(&ctx->builder, type, ctx->GLSL_std_450, op, args, ARRAY_SIZE(args)); } static SpvId get_fvec_constant(struct ntv_context *ctx, unsigned bit_size, unsigned num_components, double value) { assert(bit_size == 16 || bit_size == 32 || bit_size == 64); SpvId result = emit_float_const(ctx, bit_size, value); if (num_components == 1) return result; assert(num_components > 1); SpvId components[NIR_MAX_VEC_COMPONENTS]; for (int i = 0; i < num_components; i++) components[i] = result; SpvId type = get_fvec_type(ctx, bit_size, num_components); return spirv_builder_const_composite(&ctx->builder, type, components, num_components); } static SpvId get_ivec_constant(struct ntv_context *ctx, unsigned bit_size, unsigned num_components, int64_t value) { assert(bit_size == 8 || bit_size == 16 || bit_size == 32 || bit_size == 64); SpvId result = emit_int_const(ctx, bit_size, value); if (num_components == 1) return result; assert(num_components > 1); SpvId components[NIR_MAX_VEC_COMPONENTS]; for (int i = 0; i < num_components; i++) components[i] = result; SpvId type = get_ivec_type(ctx, bit_size, num_components); return spirv_builder_const_composite(&ctx->builder, type, components, num_components); } static inline unsigned alu_instr_src_components(const nir_alu_instr *instr, unsigned src) { if (nir_op_infos[instr->op].input_sizes[src] > 0) return nir_op_infos[instr->op].input_sizes[src]; return instr->def.num_components; } static SpvId get_alu_src(struct ntv_context *ctx, nir_alu_instr *alu, unsigned src, SpvId *raw_value, nir_alu_type *atype) { *raw_value = get_alu_src_raw(ctx, alu, src, atype); unsigned num_components = alu_instr_src_components(alu, src); unsigned bit_size = nir_src_bit_size(alu->src[src].src); nir_alu_type type = alu_op_is_typeless(alu->op) ? *atype : nir_op_infos[alu->op].input_types[src]; type = nir_alu_type_get_base_type(type); if (type == *atype) return *raw_value; if (bit_size == 1) return *raw_value; else { switch (nir_alu_type_get_base_type(type)) { case nir_type_bool: unreachable("bool should have bit-size 1"); case nir_type_int: return bitcast_to_ivec(ctx, *raw_value, bit_size, num_components); case nir_type_uint: return bitcast_to_uvec(ctx, *raw_value, bit_size, num_components); case nir_type_float: return bitcast_to_fvec(ctx, *raw_value, bit_size, num_components); default: unreachable("unknown nir_alu_type"); } } } static void store_alu_result(struct ntv_context *ctx, nir_alu_instr *alu, SpvId result, nir_alu_type atype) { store_def(ctx, alu->def.index, result, atype); } static SpvId get_def_type(struct ntv_context *ctx, nir_def *def, nir_alu_type type) { return get_alu_type(ctx, type, def->num_components, def->bit_size); } static void emit_alu(struct ntv_context *ctx, nir_alu_instr *alu) { bool is_bcsel = alu->op == nir_op_bcsel; nir_alu_type stype[NIR_MAX_VEC_COMPONENTS] = {0}; SpvId src[NIR_MAX_VEC_COMPONENTS]; SpvId raw_src[NIR_MAX_VEC_COMPONENTS]; for (unsigned i = 0; i < nir_op_infos[alu->op].num_inputs; i++) src[i] = get_alu_src(ctx, alu, i, &raw_src[i], &stype[i]); nir_alu_type typeless_type = stype[is_bcsel]; if (nir_op_infos[alu->op].num_inputs > 1 && alu_op_is_typeless(alu->op) && nir_src_bit_size(alu->src[is_bcsel].src) != 1) { unsigned uint_count = 0; unsigned int_count = 0; unsigned float_count = 0; for (unsigned i = is_bcsel; i < nir_op_infos[alu->op].num_inputs; i++) { if (stype[i] == nir_type_bool) break; switch (stype[i]) { case nir_type_uint: uint_count++; break; case nir_type_int: int_count++; break; case nir_type_float: float_count++; break; default: unreachable("this shouldn't happen"); } } if (uint_count > int_count && uint_count > float_count) typeless_type = nir_type_uint; else if (int_count > uint_count && int_count > float_count) typeless_type = nir_type_int; else if (float_count > uint_count && float_count > int_count) typeless_type = nir_type_float; else if (float_count == uint_count || uint_count == int_count) typeless_type = nir_type_uint; else if (float_count == int_count) typeless_type = nir_type_float; else typeless_type = nir_type_uint; assert(typeless_type != nir_type_bool); for (unsigned i = is_bcsel; i < nir_op_infos[alu->op].num_inputs; i++) { unsigned num_components = alu_instr_src_components(alu, i); unsigned bit_size = nir_src_bit_size(alu->src[i].src); SpvId type = get_alu_type(ctx, typeless_type, num_components, bit_size); if (stype[i] != typeless_type) { src[i] = emit_bitcast(ctx, type, src[i]); } } } unsigned bit_size = alu->def.bit_size; unsigned num_components = alu->def.num_components; nir_alu_type atype = bit_size == 1 ? nir_type_bool : (alu_op_is_typeless(alu->op) ? typeless_type : nir_op_infos[alu->op].output_type); SpvId dest_type = get_def_type(ctx, &alu->def, atype); SpvId result = 0; switch (alu->op) { case nir_op_mov: assert(nir_op_infos[alu->op].num_inputs == 1); result = src[0]; break; #define UNOP(nir_op, spirv_op) \ case nir_op: \ assert(nir_op_infos[alu->op].num_inputs == 1); \ result = emit_unop(ctx, spirv_op, dest_type, src[0]); \ break; UNOP(nir_op_ineg, SpvOpSNegate) UNOP(nir_op_fneg, SpvOpFNegate) UNOP(nir_op_f2i8, SpvOpConvertFToS) UNOP(nir_op_f2u8, SpvOpConvertFToU) UNOP(nir_op_f2i16, SpvOpConvertFToS) UNOP(nir_op_f2u16, SpvOpConvertFToU) UNOP(nir_op_f2i32, SpvOpConvertFToS) UNOP(nir_op_f2u32, SpvOpConvertFToU) UNOP(nir_op_i2f16, SpvOpConvertSToF) UNOP(nir_op_i2f32, SpvOpConvertSToF) UNOP(nir_op_u2f16, SpvOpConvertUToF) UNOP(nir_op_u2f32, SpvOpConvertUToF) UNOP(nir_op_i2i8, SpvOpSConvert) UNOP(nir_op_i2i16, SpvOpSConvert) UNOP(nir_op_i2i32, SpvOpSConvert) UNOP(nir_op_u2u8, SpvOpUConvert) UNOP(nir_op_u2u16, SpvOpUConvert) UNOP(nir_op_u2u32, SpvOpUConvert) UNOP(nir_op_f2f16, SpvOpFConvert) UNOP(nir_op_f2f32, SpvOpFConvert) UNOP(nir_op_f2i64, SpvOpConvertFToS) UNOP(nir_op_f2u64, SpvOpConvertFToU) UNOP(nir_op_u2f64, SpvOpConvertUToF) UNOP(nir_op_i2f64, SpvOpConvertSToF) UNOP(nir_op_i2i64, SpvOpSConvert) UNOP(nir_op_u2u64, SpvOpUConvert) UNOP(nir_op_f2f64, SpvOpFConvert) UNOP(nir_op_bitfield_reverse, SpvOpBitReverse) UNOP(nir_op_bit_count, SpvOpBitCount) #undef UNOP case nir_op_f2f16_rtz: assert(nir_op_infos[alu->op].num_inputs == 1); result = emit_unop(ctx, SpvOpFConvert, dest_type, src[0]); spirv_builder_emit_rounding_mode(&ctx->builder, result, SpvFPRoundingModeRTZ); break; case nir_op_inot: if (bit_size == 1) result = emit_unop(ctx, SpvOpLogicalNot, dest_type, src[0]); else result = emit_unop(ctx, SpvOpNot, dest_type, src[0]); break; case nir_op_b2i8: case nir_op_b2i16: case nir_op_b2i32: case nir_op_b2i64: assert(nir_op_infos[alu->op].num_inputs == 1); result = emit_select(ctx, dest_type, src[0], get_ivec_constant(ctx, bit_size, num_components, 1), get_ivec_constant(ctx, bit_size, num_components, 0)); break; case nir_op_b2f16: case nir_op_b2f32: case nir_op_b2f64: assert(nir_op_infos[alu->op].num_inputs == 1); result = emit_select(ctx, dest_type, src[0], get_fvec_constant(ctx, bit_size, num_components, 1), get_fvec_constant(ctx, bit_size, num_components, 0)); break; case nir_op_uclz: assert(nir_op_infos[alu->op].num_inputs == 1); result = emit_unop(ctx, SpvOpUCountLeadingZerosINTEL, dest_type, src[0]); spirv_builder_emit_cap(&ctx->builder, SpvCapabilityIntegerFunctions2INTEL); spirv_builder_emit_extension(&ctx->builder, "SPV_INTEL_shader_integer_functions2"); break; #define BUILTIN_UNOP(nir_op, spirv_op) \ case nir_op: \ assert(nir_op_infos[alu->op].num_inputs == 1); \ result = emit_builtin_unop(ctx, spirv_op, dest_type, src[0]); \ break; #define BUILTIN_UNOPF(nir_op, spirv_op) \ case nir_op: \ assert(nir_op_infos[alu->op].num_inputs == 1); \ result = emit_builtin_unop(ctx, spirv_op, get_def_type(ctx, &alu->def, nir_type_float), src[0]); \ atype = nir_type_float; \ break; BUILTIN_UNOP(nir_op_iabs, GLSLstd450SAbs) BUILTIN_UNOP(nir_op_fabs, GLSLstd450FAbs) BUILTIN_UNOP(nir_op_fsqrt, GLSLstd450Sqrt) BUILTIN_UNOP(nir_op_frsq, GLSLstd450InverseSqrt) BUILTIN_UNOP(nir_op_flog2, GLSLstd450Log2) BUILTIN_UNOP(nir_op_fexp2, GLSLstd450Exp2) BUILTIN_UNOP(nir_op_ffract, GLSLstd450Fract) BUILTIN_UNOP(nir_op_ffloor, GLSLstd450Floor) BUILTIN_UNOP(nir_op_fceil, GLSLstd450Ceil) BUILTIN_UNOP(nir_op_ftrunc, GLSLstd450Trunc) BUILTIN_UNOP(nir_op_fround_even, GLSLstd450RoundEven) BUILTIN_UNOP(nir_op_fsign, GLSLstd450FSign) BUILTIN_UNOP(nir_op_isign, GLSLstd450SSign) BUILTIN_UNOP(nir_op_fsin, GLSLstd450Sin) BUILTIN_UNOP(nir_op_fcos, GLSLstd450Cos) BUILTIN_UNOP(nir_op_ufind_msb, GLSLstd450FindUMsb) BUILTIN_UNOP(nir_op_find_lsb, GLSLstd450FindILsb) BUILTIN_UNOP(nir_op_ifind_msb, GLSLstd450FindSMsb) case nir_op_pack_half_2x16: assert(nir_op_infos[alu->op].num_inputs == 1); result = emit_builtin_unop(ctx, GLSLstd450PackHalf2x16, get_def_type(ctx, &alu->def, nir_type_uint), src[0]); break; BUILTIN_UNOPF(nir_op_unpack_half_2x16, GLSLstd450UnpackHalf2x16) #undef BUILTIN_UNOP #undef BUILTIN_UNOPF case nir_op_frcp: assert(nir_op_infos[alu->op].num_inputs == 1); result = emit_binop(ctx, SpvOpFDiv, dest_type, get_fvec_constant(ctx, bit_size, num_components, 1), src[0]); break; #define BINOP(nir_op, spirv_op) \ case nir_op: \ assert(nir_op_infos[alu->op].num_inputs == 2); \ result = emit_binop(ctx, spirv_op, dest_type, src[0], src[1]); \ break; BINOP(nir_op_iadd, SpvOpIAdd) BINOP(nir_op_isub, SpvOpISub) BINOP(nir_op_imul, SpvOpIMul) BINOP(nir_op_idiv, SpvOpSDiv) BINOP(nir_op_udiv, SpvOpUDiv) BINOP(nir_op_umod, SpvOpUMod) BINOP(nir_op_imod, SpvOpSMod) BINOP(nir_op_irem, SpvOpSRem) BINOP(nir_op_fadd, SpvOpFAdd) BINOP(nir_op_fsub, SpvOpFSub) BINOP(nir_op_fmul, SpvOpFMul) BINOP(nir_op_fdiv, SpvOpFDiv) BINOP(nir_op_fmod, SpvOpFMod) BINOP(nir_op_ilt, SpvOpSLessThan) BINOP(nir_op_ige, SpvOpSGreaterThanEqual) BINOP(nir_op_ult, SpvOpULessThan) BINOP(nir_op_uge, SpvOpUGreaterThanEqual) BINOP(nir_op_flt, SpvOpFOrdLessThan) BINOP(nir_op_fge, SpvOpFOrdGreaterThanEqual) BINOP(nir_op_frem, SpvOpFRem) #undef BINOP #define BINOP_LOG(nir_op, spv_op, spv_log_op) \ case nir_op: \ assert(nir_op_infos[alu->op].num_inputs == 2); \ if (nir_src_bit_size(alu->src[0].src) == 1) \ result = emit_binop(ctx, spv_log_op, dest_type, src[0], src[1]); \ else \ result = emit_binop(ctx, spv_op, dest_type, src[0], src[1]); \ break; BINOP_LOG(nir_op_iand, SpvOpBitwiseAnd, SpvOpLogicalAnd) BINOP_LOG(nir_op_ior, SpvOpBitwiseOr, SpvOpLogicalOr) BINOP_LOG(nir_op_ieq, SpvOpIEqual, SpvOpLogicalEqual) BINOP_LOG(nir_op_ine, SpvOpINotEqual, SpvOpLogicalNotEqual) BINOP_LOG(nir_op_ixor, SpvOpBitwiseXor, SpvOpLogicalNotEqual) #undef BINOP_LOG #define BINOP_SHIFT(nir_op, spirv_op) \ case nir_op: { \ assert(nir_op_infos[alu->op].num_inputs == 2); \ int shift_bit_size = nir_src_bit_size(alu->src[1].src); \ nir_alu_type shift_nir_type = nir_alu_type_get_base_type(nir_op_infos[alu->op].input_types[1]); \ SpvId shift_type = get_alu_type(ctx, shift_nir_type, num_components, shift_bit_size); \ SpvId shift_mask = get_ivec_constant(ctx, shift_bit_size, num_components, bit_size - 1); \ SpvId shift_count = emit_binop(ctx, SpvOpBitwiseAnd, shift_type, src[1], shift_mask); \ result = emit_binop(ctx, spirv_op, dest_type, src[0], shift_count); \ break; \ } BINOP_SHIFT(nir_op_ishl, SpvOpShiftLeftLogical) BINOP_SHIFT(nir_op_ishr, SpvOpShiftRightArithmetic) BINOP_SHIFT(nir_op_ushr, SpvOpShiftRightLogical) #undef BINOP_SHIFT #define BUILTIN_BINOP(nir_op, spirv_op) \ case nir_op: \ assert(nir_op_infos[alu->op].num_inputs == 2); \ result = emit_builtin_binop(ctx, spirv_op, dest_type, src[0], src[1]); \ break; BUILTIN_BINOP(nir_op_fmin, GLSLstd450FMin) BUILTIN_BINOP(nir_op_fmax, GLSLstd450FMax) BUILTIN_BINOP(nir_op_imin, GLSLstd450SMin) BUILTIN_BINOP(nir_op_imax, GLSLstd450SMax) BUILTIN_BINOP(nir_op_umin, GLSLstd450UMin) BUILTIN_BINOP(nir_op_umax, GLSLstd450UMax) BUILTIN_BINOP(nir_op_ldexp, GLSLstd450Ldexp) BUILTIN_BINOP(nir_op_fpow, GLSLstd450Pow) #undef BUILTIN_BINOP #define INTEL_BINOP(nir_op, spirv_op) \ case nir_op: \ assert(nir_op_infos[alu->op].num_inputs == 2); \ result = emit_binop(ctx, spirv_op, dest_type, src[0], src[1]); \ spirv_builder_emit_cap(&ctx->builder, SpvCapabilityIntegerFunctions2INTEL); \ spirv_builder_emit_extension(&ctx->builder, "SPV_INTEL_shader_integer_functions2"); \ break; INTEL_BINOP(nir_op_uabs_isub, SpvOpAbsISubINTEL) INTEL_BINOP(nir_op_uabs_usub, SpvOpAbsUSubINTEL) INTEL_BINOP(nir_op_iadd_sat, SpvOpIAddSatINTEL) INTEL_BINOP(nir_op_uadd_sat, SpvOpUAddSatINTEL) INTEL_BINOP(nir_op_ihadd, SpvOpIAverageINTEL) INTEL_BINOP(nir_op_uhadd, SpvOpUAverageINTEL) INTEL_BINOP(nir_op_irhadd, SpvOpIAverageRoundedINTEL) INTEL_BINOP(nir_op_urhadd, SpvOpUAverageRoundedINTEL) INTEL_BINOP(nir_op_isub_sat, SpvOpISubSatINTEL) INTEL_BINOP(nir_op_usub_sat, SpvOpUSubSatINTEL) INTEL_BINOP(nir_op_imul_32x16, SpvOpIMul32x16INTEL) INTEL_BINOP(nir_op_umul_32x16, SpvOpUMul32x16INTEL) #undef INTEL_BINOP case nir_op_fdot2: case nir_op_fdot3: case nir_op_fdot4: assert(nir_op_infos[alu->op].num_inputs == 2); result = emit_binop(ctx, SpvOpDot, dest_type, src[0], src[1]); break; case nir_op_fdph: case nir_op_seq: case nir_op_sne: case nir_op_slt: case nir_op_sge: unreachable("should already be lowered away"); case nir_op_fneu: assert(nir_op_infos[alu->op].num_inputs == 2); if (raw_src[0] == raw_src[1]) result = emit_unop(ctx, SpvOpIsNan, dest_type, src[0]); else result = emit_binop(ctx, SpvOpFUnordNotEqual, dest_type, src[0], src[1]); break; case nir_op_feq: assert(nir_op_infos[alu->op].num_inputs == 2); if (raw_src[0] == raw_src[1]) result = emit_unop(ctx, SpvOpLogicalNot, dest_type, emit_unop(ctx, SpvOpIsNan, dest_type, src[0])); else result = emit_binop(ctx, SpvOpFOrdEqual, dest_type, src[0], src[1]); break; case nir_op_flrp: assert(nir_op_infos[alu->op].num_inputs == 3); result = emit_builtin_triop(ctx, GLSLstd450FMix, dest_type, src[0], src[1], src[2]); break; case nir_op_bcsel: assert(nir_op_infos[alu->op].num_inputs == 3); result = emit_select(ctx, dest_type, src[0], src[1], src[2]); break; case nir_op_pack_half_2x16_split: { SpvId fvec = spirv_builder_emit_composite_construct(&ctx->builder, get_fvec_type(ctx, 32, 2), src, 2); result = emit_builtin_unop(ctx, GLSLstd450PackHalf2x16, dest_type, fvec); break; } case nir_op_vec2: case nir_op_vec3: case nir_op_vec4: { int num_inputs = nir_op_infos[alu->op].num_inputs; assert(2 <= num_inputs && num_inputs <= 4); result = spirv_builder_emit_composite_construct(&ctx->builder, dest_type, src, num_inputs); } break; case nir_op_ubitfield_extract: assert(nir_op_infos[alu->op].num_inputs == 3); result = emit_triop(ctx, SpvOpBitFieldUExtract, dest_type, src[0], src[1], src[2]); break; case nir_op_ibitfield_extract: assert(nir_op_infos[alu->op].num_inputs == 3); result = emit_triop(ctx, SpvOpBitFieldSExtract, dest_type, src[0], src[1], src[2]); break; case nir_op_bitfield_insert: assert(nir_op_infos[alu->op].num_inputs == 4); result = spirv_builder_emit_quadop(&ctx->builder, SpvOpBitFieldInsert, dest_type, src[0], src[1], src[2], src[3]); break; /* those are all simple bitcasts, we could do better, but it doesn't matter */ case nir_op_pack_32_4x8: case nir_op_pack_32_2x16: case nir_op_pack_64_2x32: case nir_op_pack_64_4x16: case nir_op_unpack_32_4x8: case nir_op_unpack_32_2x16: case nir_op_unpack_64_2x32: case nir_op_unpack_64_4x16: { result = emit_bitcast(ctx, dest_type, src[0]); break; } case nir_op_pack_32_2x16_split: case nir_op_pack_64_2x32_split: { nir_alu_type type = nir_alu_type_get_base_type(nir_op_infos[alu->op].input_types[0]); if (num_components <= 2) { SpvId components[] = {src[0], src[1]}; SpvId vec_type = get_alu_type(ctx, type, num_components * 2, nir_src_bit_size(alu->src[0].src)); result = spirv_builder_emit_composite_construct(&ctx->builder, vec_type, components, 2); result = emit_bitcast(ctx, dest_type, result); } else { SpvId components[NIR_MAX_VEC_COMPONENTS]; SpvId conv_type = get_alu_type(ctx, type, 1, nir_src_bit_size(alu->src[0].src)); SpvId vec_type = get_alu_type(ctx, type, 2, nir_src_bit_size(alu->src[0].src)); SpvId dest_scalar_type = get_alu_type(ctx, nir_op_infos[alu->op].output_type, 1, bit_size); for (unsigned i = 0; i < nir_src_num_components(alu->src[0].src); i++) { SpvId conv[2]; conv[0] = spirv_builder_emit_composite_extract(&ctx->builder, conv_type, src[0], &i, 1); conv[1] = spirv_builder_emit_composite_extract(&ctx->builder, conv_type, src[1], &i, 1); SpvId vec = spirv_builder_emit_composite_construct(&ctx->builder, vec_type, conv, 2); components[i] = emit_bitcast(ctx, dest_scalar_type, vec); } result = spirv_builder_emit_composite_construct(&ctx->builder, dest_type, components, num_components); } break; } case nir_op_unpack_32_2x16_split_x: case nir_op_unpack_64_2x32_split_x: { nir_alu_type type = nir_alu_type_get_base_type(nir_op_infos[alu->op].input_types[0]); SpvId vec_type = get_alu_type(ctx, type, 2, bit_size); unsigned idx = 0; if (num_components == 1) { SpvId vec = emit_bitcast(ctx, vec_type, src[0]); result = spirv_builder_emit_composite_extract(&ctx->builder, dest_type, vec, &idx, 1); } else { SpvId components[NIR_MAX_VEC_COMPONENTS]; for (unsigned i = 0; i < nir_src_num_components(alu->src[0].src); i++) { SpvId conv = spirv_builder_emit_composite_extract(&ctx->builder, get_alu_type(ctx, type, 1, nir_src_bit_size(alu->src[0].src)), src[0], &i, 1); conv = emit_bitcast(ctx, vec_type, conv); SpvId conv_type = get_alu_type(ctx, type, 1, bit_size); components[i] = spirv_builder_emit_composite_extract(&ctx->builder, conv_type, conv, &idx, 1); } result = spirv_builder_emit_composite_construct(&ctx->builder, dest_type, components, num_components); } break; } case nir_op_unpack_32_2x16_split_y: case nir_op_unpack_64_2x32_split_y: { nir_alu_type type = nir_alu_type_get_base_type(nir_op_infos[alu->op].input_types[0]); SpvId vec_type = get_alu_type(ctx, type, 2, bit_size); unsigned idx = 1; if (num_components == 1) { SpvId vec = emit_bitcast(ctx, vec_type, src[0]); result = spirv_builder_emit_composite_extract(&ctx->builder, dest_type, vec, &idx, 1); } else { SpvId components[NIR_MAX_VEC_COMPONENTS]; for (unsigned i = 0; i < nir_src_num_components(alu->src[0].src); i++) { SpvId conv = spirv_builder_emit_composite_extract(&ctx->builder, get_alu_type(ctx, type, 1, nir_src_bit_size(alu->src[0].src)), src[0], &i, 1); conv = emit_bitcast(ctx, vec_type, conv); SpvId conv_type = get_alu_type(ctx, type, 1, bit_size); components[i] = spirv_builder_emit_composite_extract(&ctx->builder, conv_type, conv, &idx, 1); } result = spirv_builder_emit_composite_construct(&ctx->builder, dest_type, components, num_components); } break; } default: fprintf(stderr, "emit_alu: not implemented (%s)\n", nir_op_infos[alu->op].name); unreachable("unsupported opcode"); return; } if (alu->exact) spirv_builder_emit_decoration(&ctx->builder, result, SpvDecorationNoContraction); store_alu_result(ctx, alu, result, atype); } static void emit_load_const(struct ntv_context *ctx, nir_load_const_instr *load_const) { unsigned bit_size = load_const->def.bit_size; unsigned num_components = load_const->def.num_components; SpvId components[NIR_MAX_VEC_COMPONENTS]; nir_alu_type atype; if (bit_size == 1) { atype = nir_type_bool; for (int i = 0; i < num_components; i++) components[i] = spirv_builder_const_bool(&ctx->builder, load_const->value[i].b); } else { atype = infer_nir_alu_type_from_uses_ssa(&load_const->def); for (int i = 0; i < num_components; i++) { switch (atype) { case nir_type_uint: { uint64_t tmp = nir_const_value_as_uint(load_const->value[i], bit_size); components[i] = emit_uint_const(ctx, bit_size, tmp); break; } case nir_type_int: { int64_t tmp = nir_const_value_as_int(load_const->value[i], bit_size); components[i] = emit_int_const(ctx, bit_size, tmp); break; } case nir_type_float: { double tmp = nir_const_value_as_float(load_const->value[i], bit_size); components[i] = emit_float_const(ctx, bit_size, tmp); break; } default: unreachable("this shouldn't happen!"); } } } if (num_components > 1) { SpvId type = get_alu_type(ctx, atype, num_components, bit_size); SpvId value = spirv_builder_const_composite(&ctx->builder, type, components, num_components); store_def(ctx, load_const->def.index, value, atype); } else { assert(num_components == 1); store_def(ctx, load_const->def.index, components[0], atype); } } static void emit_discard(struct ntv_context *ctx, nir_intrinsic_instr *intr) { assert(ctx->discard_func); SpvId type_void = spirv_builder_type_void(&ctx->builder); spirv_builder_function_call(&ctx->builder, type_void, ctx->discard_func, NULL, 0); } static void emit_load_deref(struct ntv_context *ctx, nir_intrinsic_instr *intr) { nir_alu_type atype; SpvId ptr = get_src(ctx, intr->src, &atype); nir_deref_instr *deref = nir_src_as_deref(intr->src[0]); SpvId type; if (glsl_type_is_image(deref->type)) { nir_variable *var = nir_deref_instr_get_variable(deref); const struct glsl_type *gtype = glsl_without_array(var->type); type = get_image_type(ctx, var, glsl_type_is_sampler(gtype), glsl_get_sampler_dim(gtype) == GLSL_SAMPLER_DIM_BUF); atype = nir_get_nir_type_for_glsl_base_type(glsl_get_sampler_result_type(gtype)); } else { type = get_glsl_type(ctx, deref->type); atype = get_nir_alu_type(deref->type); } SpvId result; if (nir_intrinsic_access(intr) & ACCESS_COHERENT) result = emit_atomic(ctx, SpvOpAtomicLoad, type, ptr, 0, 0); else result = spirv_builder_emit_load(&ctx->builder, type, ptr); store_def(ctx, intr->def.index, result, atype); } static void emit_store_deref(struct ntv_context *ctx, nir_intrinsic_instr *intr) { nir_alu_type ptype, stype; SpvId ptr = get_src(ctx, &intr->src[0], &ptype); SpvId src = get_src(ctx, &intr->src[1], &stype); const struct glsl_type *gtype = nir_src_as_deref(intr->src[0])->type; SpvId type = get_glsl_type(ctx, gtype); nir_variable *var = nir_intrinsic_get_var(intr, 0); unsigned wrmask = nir_intrinsic_write_mask(intr); if (!glsl_type_is_scalar(gtype) && wrmask != BITFIELD_MASK(glsl_type_is_array(gtype) ? glsl_get_aoa_size(gtype) : glsl_get_vector_elements(gtype))) { /* no idea what we do if this fails */ assert(glsl_type_is_array(gtype) || glsl_type_is_vector(gtype)); /* this is a partial write, so we have to loop and do a per-component write */ SpvId result_type; SpvId member_type; if (glsl_type_is_vector(gtype)) { result_type = get_glsl_basetype(ctx, glsl_get_base_type(gtype)); member_type = get_alu_type(ctx, stype, 1, glsl_get_bit_size(gtype)); } else member_type = result_type = get_glsl_type(ctx, glsl_get_array_element(gtype)); SpvId ptr_type = spirv_builder_type_pointer(&ctx->builder, get_storage_class(var), result_type); for (unsigned i = 0; i < 4; i++) if (wrmask & BITFIELD_BIT(i)) { SpvId idx = emit_uint_const(ctx, 32, i); SpvId val = spirv_builder_emit_composite_extract(&ctx->builder, member_type, src, &i, 1); if (stype != ptype) val = emit_bitcast(ctx, result_type, val); SpvId member = spirv_builder_emit_access_chain(&ctx->builder, ptr_type, ptr, &idx, 1); spirv_builder_emit_store(&ctx->builder, member, val); } return; } SpvId result; if (ctx->stage == MESA_SHADER_FRAGMENT && var->data.mode == nir_var_shader_out && var->data.location == FRAG_RESULT_SAMPLE_MASK) { src = emit_bitcast(ctx, type, src); /* SampleMask is always an array in spirv, so we need to construct it into one */ result = spirv_builder_emit_composite_construct(&ctx->builder, ctx->sample_mask_type, &src, 1); } else { if (ptype == stype) result = src; else result = emit_bitcast(ctx, type, src); } if (nir_intrinsic_access(intr) & ACCESS_COHERENT) spirv_builder_emit_atomic_store(&ctx->builder, ptr, SpvScopeDevice, 0, result); else spirv_builder_emit_store(&ctx->builder, ptr, result); } static void emit_load_shared(struct ntv_context *ctx, nir_intrinsic_instr *intr) { SpvId dest_type = get_def_type(ctx, &intr->def, nir_type_uint); unsigned num_components = intr->def.num_components; unsigned bit_size = intr->def.bit_size; SpvId uint_type = get_uvec_type(ctx, bit_size, 1); SpvId ptr_type = spirv_builder_type_pointer(&ctx->builder, SpvStorageClassWorkgroup, uint_type); nir_alu_type atype; SpvId offset = get_src(ctx, &intr->src[0], &atype); if (atype == nir_type_float) offset = bitcast_to_uvec(ctx, offset, nir_src_bit_size(intr->src[0]), 1); SpvId constituents[NIR_MAX_VEC_COMPONENTS]; SpvId shared_block = get_shared_block(ctx, bit_size); /* need to convert array -> vec */ for (unsigned i = 0; i < num_components; i++) { SpvId member = spirv_builder_emit_access_chain(&ctx->builder, ptr_type, shared_block, &offset, 1); constituents[i] = spirv_builder_emit_load(&ctx->builder, uint_type, member); offset = emit_binop(ctx, SpvOpIAdd, spirv_builder_type_uint(&ctx->builder, 32), offset, emit_uint_const(ctx, 32, 1)); } SpvId result; if (num_components > 1) result = spirv_builder_emit_composite_construct(&ctx->builder, dest_type, constituents, num_components); else result = constituents[0]; store_def(ctx, intr->def.index, result, nir_type_uint); } static void emit_store_shared(struct ntv_context *ctx, nir_intrinsic_instr *intr) { nir_alu_type atype; SpvId src = get_src(ctx, &intr->src[0], &atype); unsigned wrmask = nir_intrinsic_write_mask(intr); unsigned bit_size = nir_src_bit_size(intr->src[0]); SpvId uint_type = get_uvec_type(ctx, bit_size, 1); SpvId ptr_type = spirv_builder_type_pointer(&ctx->builder, SpvStorageClassWorkgroup, uint_type); nir_alu_type otype; SpvId offset = get_src(ctx, &intr->src[1], &otype); if (otype == nir_type_float) offset = bitcast_to_uvec(ctx, offset, nir_src_bit_size(intr->src[0]), 1); SpvId shared_block = get_shared_block(ctx, bit_size); /* this is a partial write, so we have to loop and do a per-component write */ u_foreach_bit(i, wrmask) { SpvId shared_offset = emit_binop(ctx, SpvOpIAdd, spirv_builder_type_uint(&ctx->builder, 32), offset, emit_uint_const(ctx, 32, i)); SpvId val = src; if (nir_src_num_components(intr->src[0]) != 1) val = spirv_builder_emit_composite_extract(&ctx->builder, uint_type, src, &i, 1); if (atype != nir_type_uint) val = emit_bitcast(ctx, get_alu_type(ctx, nir_type_uint, 1, bit_size), val); SpvId member = spirv_builder_emit_access_chain(&ctx->builder, ptr_type, shared_block, &shared_offset, 1); spirv_builder_emit_store(&ctx->builder, member, val); } } static void emit_load_scratch(struct ntv_context *ctx, nir_intrinsic_instr *intr) { SpvId dest_type = get_def_type(ctx, &intr->def, nir_type_uint); unsigned num_components = intr->def.num_components; unsigned bit_size = intr->def.bit_size; SpvId uint_type = get_uvec_type(ctx, bit_size, 1); SpvId ptr_type = spirv_builder_type_pointer(&ctx->builder, SpvStorageClassPrivate, uint_type); nir_alu_type atype; SpvId offset = get_src(ctx, &intr->src[0], &atype); if (atype != nir_type_uint) offset = bitcast_to_uvec(ctx, offset, nir_src_bit_size(intr->src[0]), 1); SpvId constituents[NIR_MAX_VEC_COMPONENTS]; SpvId scratch_block = get_scratch_block(ctx, bit_size); /* need to convert array -> vec */ for (unsigned i = 0; i < num_components; i++) { SpvId member = spirv_builder_emit_access_chain(&ctx->builder, ptr_type, scratch_block, &offset, 1); constituents[i] = spirv_builder_emit_load(&ctx->builder, uint_type, member); offset = emit_binop(ctx, SpvOpIAdd, spirv_builder_type_uint(&ctx->builder, 32), offset, emit_uint_const(ctx, 32, 1)); } SpvId result; if (num_components > 1) result = spirv_builder_emit_composite_construct(&ctx->builder, dest_type, constituents, num_components); else result = constituents[0]; store_def(ctx, intr->def.index, result, nir_type_uint); } static void emit_store_scratch(struct ntv_context *ctx, nir_intrinsic_instr *intr) { nir_alu_type atype; SpvId src = get_src(ctx, &intr->src[0], &atype); unsigned wrmask = nir_intrinsic_write_mask(intr); unsigned bit_size = nir_src_bit_size(intr->src[0]); SpvId uint_type = get_uvec_type(ctx, bit_size, 1); SpvId ptr_type = spirv_builder_type_pointer(&ctx->builder, SpvStorageClassPrivate, uint_type); nir_alu_type otype; SpvId offset = get_src(ctx, &intr->src[1], &otype); if (otype != nir_type_uint) offset = bitcast_to_uvec(ctx, offset, nir_src_bit_size(intr->src[1]), 1); SpvId scratch_block = get_scratch_block(ctx, bit_size); /* this is a partial write, so we have to loop and do a per-component write */ u_foreach_bit(i, wrmask) { SpvId scratch_offset = emit_binop(ctx, SpvOpIAdd, spirv_builder_type_uint(&ctx->builder, 32), offset, emit_uint_const(ctx, 32, i)); SpvId val = src; if (nir_src_num_components(intr->src[0]) != 1) val = spirv_builder_emit_composite_extract(&ctx->builder, uint_type, src, &i, 1); if (atype != nir_type_uint) val = emit_bitcast(ctx, get_alu_type(ctx, nir_type_uint, 1, bit_size), val); SpvId member = spirv_builder_emit_access_chain(&ctx->builder, ptr_type, scratch_block, &scratch_offset, 1); spirv_builder_emit_store(&ctx->builder, member, val); } } static void emit_load_push_const(struct ntv_context *ctx, nir_intrinsic_instr *intr) { SpvId uint_type = get_uvec_type(ctx, 32, 1); SpvId load_type = get_uvec_type(ctx, 32, 1); /* number of components being loaded */ unsigned num_components = intr->def.num_components; SpvId constituents[NIR_MAX_VEC_COMPONENTS * 2]; SpvId result; /* destination type for the load */ SpvId type = get_def_uvec_type(ctx, &intr->def); SpvId one = emit_uint_const(ctx, 32, 1); /* we grab a single array member at a time, so it's a pointer to a uint */ SpvId pointer_type = spirv_builder_type_pointer(&ctx->builder, SpvStorageClassPushConstant, load_type); nir_alu_type atype; SpvId member = get_src(ctx, &intr->src[0], &atype); if (atype == nir_type_float) member = bitcast_to_uvec(ctx, member, nir_src_bit_size(intr->src[0]), 1); /* reuse the offset from ZINK_PUSH_CONST_OFFSET */ SpvId offset = emit_uint_const(ctx, 32, nir_intrinsic_component(intr)); /* OpAccessChain takes an array of indices that drill into a hierarchy based on the type: * index 0 is accessing 'base' * index 1 is accessing 'base[index 1]' * */ for (unsigned i = 0; i < num_components; i++) { SpvId indices[2] = { member, offset }; SpvId ptr = spirv_builder_emit_access_chain(&ctx->builder, pointer_type, ctx->push_const_var, indices, ARRAY_SIZE(indices)); /* load a single value into the constituents array */ constituents[i] = spirv_builder_emit_load(&ctx->builder, load_type, ptr); /* increment to the next vec4 member index for the next load */ offset = emit_binop(ctx, SpvOpIAdd, uint_type, offset, one); } /* if loading more than 1 value, reassemble the results into the desired type, * otherwise just use the loaded result */ if (num_components > 1) { result = spirv_builder_emit_composite_construct(&ctx->builder, type, constituents, num_components); } else result = constituents[0]; store_def(ctx, intr->def.index, result, nir_type_uint); } static void emit_load_global(struct ntv_context *ctx, nir_intrinsic_instr *intr) { bool coherent = ctx->sinfo->have_vulkan_memory_model && nir_intrinsic_access(intr) & ACCESS_COHERENT; spirv_builder_emit_cap(&ctx->builder, SpvCapabilityPhysicalStorageBufferAddresses); SpvId dest_type = get_def_type(ctx, &intr->def, nir_type_uint); SpvId pointer_type = spirv_builder_type_pointer(&ctx->builder, SpvStorageClassPhysicalStorageBuffer, dest_type); nir_alu_type atype; SpvId ptr = emit_bitcast(ctx, pointer_type, get_src(ctx, &intr->src[0], &atype)); SpvId result = spirv_builder_emit_load_aligned(&ctx->builder, dest_type, ptr, intr->def.bit_size / 8, coherent); store_def(ctx, intr->def.index, result, nir_type_uint); } static void emit_store_global(struct ntv_context *ctx, nir_intrinsic_instr *intr) { bool coherent = ctx->sinfo->have_vulkan_memory_model && nir_intrinsic_access(intr) & ACCESS_COHERENT; spirv_builder_emit_cap(&ctx->builder, SpvCapabilityPhysicalStorageBufferAddresses); unsigned bit_size = nir_src_bit_size(intr->src[0]); SpvId dest_type = get_uvec_type(ctx, bit_size, 1); SpvId pointer_type = spirv_builder_type_pointer(&ctx->builder, SpvStorageClassPhysicalStorageBuffer, dest_type); nir_alu_type atype; SpvId param = get_src(ctx, &intr->src[0], &atype); if (atype != nir_type_uint) param = emit_bitcast(ctx, dest_type, param); SpvId ptr = emit_bitcast(ctx, pointer_type, get_src(ctx, &intr->src[1], &atype)); spirv_builder_emit_store_aligned(&ctx->builder, ptr, param, bit_size / 8, coherent); } static void emit_load_reg(struct ntv_context *ctx, nir_intrinsic_instr *intr) { assert(nir_intrinsic_base(intr) == 0 && "no array registers"); nir_intrinsic_instr *decl = nir_reg_get_decl(intr->src[0].ssa); unsigned num_components = nir_intrinsic_num_components(decl); unsigned bit_size = nir_intrinsic_bit_size(decl); unsigned index = decl->def.index; assert(index < ctx->num_defs); init_reg(ctx, decl, nir_type_uint); assert(ctx->defs[index] != 0); nir_alu_type atype = ctx->def_types[index]; SpvId var = ctx->defs[index]; SpvId type = get_alu_type(ctx, atype, num_components, bit_size); SpvId result = spirv_builder_emit_load(&ctx->builder, type, var); store_def(ctx, intr->def.index, result, atype); } static void emit_store_reg(struct ntv_context *ctx, nir_intrinsic_instr *intr) { nir_alu_type atype; SpvId param = get_src(ctx, &intr->src[0], &atype); nir_intrinsic_instr *decl = nir_reg_get_decl(intr->src[1].ssa); unsigned index = decl->def.index; unsigned num_components = nir_intrinsic_num_components(decl); unsigned bit_size = nir_intrinsic_bit_size(decl); atype = nir_alu_type_get_base_type(atype); init_reg(ctx, decl, atype); SpvId var = ctx->defs[index]; nir_alu_type vtype = ctx->def_types[index]; if (atype != vtype) { assert(vtype != nir_type_bool); param = emit_bitcast(ctx, get_alu_type(ctx, vtype, num_components, bit_size), param); } assert(var); spirv_builder_emit_store(&ctx->builder, var, param); } static SpvId create_builtin_var(struct ntv_context *ctx, SpvId var_type, SpvStorageClass storage_class, const char *name, SpvBuiltIn builtin) { SpvId pointer_type = spirv_builder_type_pointer(&ctx->builder, storage_class, var_type); SpvId var = spirv_builder_emit_var(&ctx->builder, pointer_type, storage_class); spirv_builder_emit_name(&ctx->builder, var, name); spirv_builder_emit_builtin(&ctx->builder, var, builtin); if (ctx->stage == MESA_SHADER_FRAGMENT) { switch (builtin) { case SpvBuiltInSampleId: case SpvBuiltInSubgroupLocalInvocationId: spirv_builder_emit_decoration(&ctx->builder, var, SpvDecorationFlat); break; default: break; } } assert(ctx->num_entry_ifaces < ARRAY_SIZE(ctx->entry_ifaces)); ctx->entry_ifaces[ctx->num_entry_ifaces++] = var; return var; } static void emit_load_front_face(struct ntv_context *ctx, nir_intrinsic_instr *intr) { SpvId var_type = spirv_builder_type_bool(&ctx->builder); if (!ctx->front_face_var) ctx->front_face_var = create_builtin_var(ctx, var_type, SpvStorageClassInput, "gl_FrontFacing", SpvBuiltInFrontFacing); SpvId result = spirv_builder_emit_load(&ctx->builder, var_type, ctx->front_face_var); assert(1 == intr->def.num_components); store_def(ctx, intr->def.index, result, nir_type_bool); } static void emit_load_view_index(struct ntv_context *ctx, nir_intrinsic_instr *intr) { SpvId var_type = spirv_builder_type_uint(&ctx->builder, 32); spirv_builder_emit_extension(&ctx->builder, "SPV_KHR_multiview"); spirv_builder_emit_cap(&ctx->builder, SpvCapabilityMultiView); if (!ctx->view_index_var) ctx->view_index_var = create_builtin_var(ctx, var_type, SpvStorageClassInput, "gl_ViewIndex", SpvBuiltInViewIndex); SpvId result = spirv_builder_emit_load(&ctx->builder, var_type, ctx->view_index_var); assert(1 == intr->def.num_components); store_def(ctx, intr->def.index, result, nir_type_uint); } static void emit_load_uint_input(struct ntv_context *ctx, nir_intrinsic_instr *intr, SpvId *var_id, const char *var_name, SpvBuiltIn builtin) { SpvId var_type = spirv_builder_type_uint(&ctx->builder, 32); if (!*var_id) { if (builtin == SpvBuiltInSampleMask) { /* gl_SampleMaskIn is an array[1] in spirv... */ var_type = spirv_builder_type_array(&ctx->builder, var_type, emit_uint_const(ctx, 32, 1)); spirv_builder_emit_array_stride(&ctx->builder, var_type, sizeof(uint32_t)); } *var_id = create_builtin_var(ctx, var_type, SpvStorageClassInput, var_name, builtin); } SpvId load_var = *var_id; if (builtin == SpvBuiltInSampleMask) { SpvId zero = emit_uint_const(ctx, 32, 0); var_type = spirv_builder_type_uint(&ctx->builder, 32); SpvId pointer_type = spirv_builder_type_pointer(&ctx->builder, SpvStorageClassInput, var_type); load_var = spirv_builder_emit_access_chain(&ctx->builder, pointer_type, load_var, &zero, 1); } SpvId result = spirv_builder_emit_load(&ctx->builder, var_type, load_var); assert(1 == intr->def.num_components); store_def(ctx, intr->def.index, result, nir_type_uint); } static void emit_load_vec_input(struct ntv_context *ctx, nir_intrinsic_instr *intr, SpvId *var_id, const char *var_name, SpvBuiltIn builtin, nir_alu_type type) { SpvId var_type; switch (type) { case nir_type_bool: var_type = get_bvec_type(ctx, intr->def.num_components); break; case nir_type_int: var_type = get_ivec_type(ctx, intr->def.bit_size, intr->def.num_components); break; case nir_type_uint: var_type = get_uvec_type(ctx, intr->def.bit_size, intr->def.num_components); break; case nir_type_float: var_type = get_fvec_type(ctx, intr->def.bit_size, intr->def.num_components); break; default: unreachable("unknown type passed"); } if (!*var_id) *var_id = create_builtin_var(ctx, var_type, SpvStorageClassInput, var_name, builtin); SpvId result = spirv_builder_emit_load(&ctx->builder, var_type, *var_id); store_def(ctx, intr->def.index, result, type); } static void emit_interpolate(struct ntv_context *ctx, nir_intrinsic_instr *intr) { SpvId op; spirv_builder_emit_cap(&ctx->builder, SpvCapabilityInterpolationFunction); SpvId src1 = 0; nir_alu_type atype; switch (intr->intrinsic) { case nir_intrinsic_interp_deref_at_centroid: op = GLSLstd450InterpolateAtCentroid; break; case nir_intrinsic_interp_deref_at_sample: op = GLSLstd450InterpolateAtSample; src1 = get_src(ctx, &intr->src[1], &atype); if (atype != nir_type_int) src1 = emit_bitcast(ctx, get_ivec_type(ctx, 32, 1), src1); break; case nir_intrinsic_interp_deref_at_offset: op = GLSLstd450InterpolateAtOffset; src1 = get_src(ctx, &intr->src[1], &atype); /* The offset operand must be a vector of 2 components of 32-bit floating-point type. - InterpolateAtOffset spec */ if (atype != nir_type_float) src1 = emit_bitcast(ctx, get_fvec_type(ctx, 32, 2), src1); break; default: unreachable("unknown interp op"); } nir_alu_type ptype; SpvId ptr = get_src(ctx, &intr->src[0], &ptype); SpvId result; const struct glsl_type *gtype = nir_src_as_deref(intr->src[0])->type; assert(glsl_get_vector_elements(gtype) == intr->num_components); assert(ptype == get_nir_alu_type(gtype)); if (intr->intrinsic == nir_intrinsic_interp_deref_at_centroid) result = emit_builtin_unop(ctx, op, get_glsl_type(ctx, gtype), ptr); else result = emit_builtin_binop(ctx, op, get_glsl_type(ctx, gtype), ptr, src1); store_def(ctx, intr->def.index, result, ptype); } static void handle_atomic_op(struct ntv_context *ctx, nir_intrinsic_instr *intr, SpvId ptr, SpvId param, SpvId param2, nir_alu_type type) { SpvId dest_type = get_def_type(ctx, &intr->def, type); SpvId result = emit_atomic(ctx, get_atomic_op(ctx, intr->def.bit_size, nir_intrinsic_atomic_op(intr)), dest_type, ptr, param, param2); assert(result); store_def(ctx, intr->def.index, result, type); } static void emit_deref_atomic_intrinsic(struct ntv_context *ctx, nir_intrinsic_instr *intr) { nir_alu_type atype; nir_alu_type ret_type = nir_atomic_op_type(nir_intrinsic_atomic_op(intr)) == nir_type_float ? nir_type_float : nir_type_uint; SpvId ptr = get_src(ctx, &intr->src[0], &atype); if (atype != ret_type && ret_type == nir_type_float) { unsigned bit_size = nir_src_bit_size(intr->src[0]); SpvId *float_array_type = &ctx->float_array_type[bit_size == 32 ? 0 : 1]; if (!*float_array_type) { *float_array_type = spirv_builder_type_pointer(&ctx->builder, SpvStorageClassStorageBuffer, spirv_builder_type_float(&ctx->builder, bit_size)); } ptr = emit_unop(ctx, SpvOpBitcast, *float_array_type, ptr); } SpvId param = get_src(ctx, &intr->src[1], &atype); if (atype != ret_type) param = cast_src_to_type(ctx, param, intr->src[1], ret_type); SpvId param2 = 0; if (nir_src_bit_size(intr->src[1]) == 64) spirv_builder_emit_cap(&ctx->builder, SpvCapabilityInt64Atomics); if (intr->intrinsic == nir_intrinsic_deref_atomic_swap) { param2 = get_src(ctx, &intr->src[2], &atype); if (atype != ret_type) param2 = cast_src_to_type(ctx, param2, intr->src[2], ret_type); } handle_atomic_op(ctx, intr, ptr, param, param2, ret_type); } static void emit_shared_atomic_intrinsic(struct ntv_context *ctx, nir_intrinsic_instr *intr) { unsigned bit_size = nir_src_bit_size(intr->src[1]); SpvId dest_type = get_def_type(ctx, &intr->def, nir_type_uint); nir_alu_type atype; nir_alu_type ret_type = nir_atomic_op_type(nir_intrinsic_atomic_op(intr)) == nir_type_float ? nir_type_float : nir_type_uint; SpvId param = get_src(ctx, &intr->src[1], &atype); if (atype != ret_type) param = cast_src_to_type(ctx, param, intr->src[1], ret_type); SpvId pointer_type = spirv_builder_type_pointer(&ctx->builder, SpvStorageClassWorkgroup, dest_type); SpvId offset = get_src(ctx, &intr->src[0], &atype); if (atype != nir_type_uint) offset = cast_src_to_type(ctx, offset, intr->src[0], nir_type_uint); offset = emit_binop(ctx, SpvOpUDiv, get_uvec_type(ctx, 32, 1), offset, emit_uint_const(ctx, 32, bit_size / 8)); SpvId shared_block = get_shared_block(ctx, bit_size); SpvId ptr = spirv_builder_emit_access_chain(&ctx->builder, pointer_type, shared_block, &offset, 1); if (nir_src_bit_size(intr->src[1]) == 64) spirv_builder_emit_cap(&ctx->builder, SpvCapabilityInt64Atomics); SpvId param2 = 0; if (intr->intrinsic == nir_intrinsic_shared_atomic_swap) { param2 = get_src(ctx, &intr->src[2], &atype); if (atype != ret_type) param2 = cast_src_to_type(ctx, param2, intr->src[2], ret_type); } handle_atomic_op(ctx, intr, ptr, param, param2, ret_type); } static void emit_global_atomic_intrinsic(struct ntv_context *ctx, nir_intrinsic_instr *intr) { unsigned bit_size = nir_src_bit_size(intr->src[1]); SpvId dest_type = get_def_type(ctx, &intr->def, nir_type_uint); nir_alu_type atype; nir_alu_type ret_type = nir_atomic_op_type(nir_intrinsic_atomic_op(intr)) == nir_type_float ? nir_type_float : nir_type_uint; SpvId param = get_src(ctx, &intr->src[1], &atype); spirv_builder_emit_cap(&ctx->builder, SpvCapabilityPhysicalStorageBufferAddresses); SpvId pointer_type = spirv_builder_type_pointer(&ctx->builder, SpvStorageClassPhysicalStorageBuffer, dest_type); SpvId ptr = emit_bitcast(ctx, pointer_type, get_src(ctx, &intr->src[0], &atype)); if (bit_size == 64) spirv_builder_emit_cap(&ctx->builder, SpvCapabilityInt64Atomics); SpvId param2 = 0; if (intr->intrinsic == nir_intrinsic_global_atomic_swap) param2 = get_src(ctx, &intr->src[2], &atype); handle_atomic_op(ctx, intr, ptr, param, param2, ret_type); } static void emit_get_ssbo_size(struct ntv_context *ctx, nir_intrinsic_instr *intr) { SpvId uint_type = get_uvec_type(ctx, 32, 1); nir_variable *var = ctx->ssbo_vars; const struct glsl_type *bare_type = glsl_without_array(var->type); unsigned last_member_idx = glsl_get_length(bare_type) - 1; SpvId pointer_type = spirv_builder_type_pointer(&ctx->builder, SpvStorageClassStorageBuffer, get_bo_struct_type(ctx, var)); nir_alu_type atype; SpvId bo = get_src(ctx, &intr->src[0], &atype); if (atype == nir_type_float) bo = bitcast_to_uvec(ctx, bo, nir_src_bit_size(intr->src[0]), 1); SpvId indices[] = { bo }; SpvId ptr = spirv_builder_emit_access_chain(&ctx->builder, pointer_type, ctx->ssbos[2], indices, ARRAY_SIZE(indices)); SpvId result = spirv_builder_emit_binop(&ctx->builder, SpvOpArrayLength, uint_type, ptr, last_member_idx); /* this is going to be converted by nir to: length = (buffer_size - offset) / stride * so we need to un-convert it to avoid having the calculation performed twice */ const struct glsl_type *last_member = glsl_get_struct_field(bare_type, last_member_idx); /* multiply by stride */ result = emit_binop(ctx, SpvOpIMul, uint_type, result, emit_uint_const(ctx, 32, glsl_get_explicit_stride(last_member))); /* get total ssbo size by adding offset */ result = emit_binop(ctx, SpvOpIAdd, uint_type, result, emit_uint_const(ctx, 32, glsl_get_struct_field_offset(bare_type, last_member_idx))); store_def(ctx, intr->def.index, result, nir_type_uint); } static SpvId get_image_coords(struct ntv_context *ctx, const struct glsl_type *type, nir_src *src) { uint32_t num_coords = glsl_get_sampler_coordinate_components(type); uint32_t src_components = nir_src_num_components(*src); nir_alu_type atype; SpvId spv = get_src(ctx, src, &atype); if (num_coords == src_components) return spv; /* need to extract the coord dimensions that the image can use */ SpvId vec_type = get_alu_type(ctx, atype, num_coords, 32); if (num_coords == 1) return spirv_builder_emit_vector_extract(&ctx->builder, vec_type, spv, 0); uint32_t constituents[4]; SpvId zero = atype == nir_type_uint ? emit_uint_const(ctx, nir_src_bit_size(*src), 0) : emit_float_const(ctx, nir_src_bit_size(*src), 0); assert(num_coords < ARRAY_SIZE(constituents)); for (unsigned i = 0; i < num_coords; i++) constituents[i] = i < src_components ? i : zero; return spirv_builder_emit_vector_shuffle(&ctx->builder, vec_type, spv, spv, constituents, num_coords); } static void emit_image_deref_store(struct ntv_context *ctx, nir_intrinsic_instr *intr) { nir_alu_type atype; SpvId img_var = get_src(ctx, &intr->src[0], &atype); nir_deref_instr *deref = nir_src_as_deref(intr->src[0]); nir_variable *var = nir_deref_instr_get_variable(deref); SpvId img_type = find_image_type(ctx, var); const struct glsl_type *type = glsl_without_array(var->type); SpvId base_type = get_glsl_basetype(ctx, glsl_get_sampler_result_type(type)); SpvId img = spirv_builder_emit_load(&ctx->builder, img_type, img_var); SpvId coord = get_image_coords(ctx, type, &intr->src[1]); SpvId texel = get_src(ctx, &intr->src[3], &atype); /* texel type must match image type */ if (atype != nir_get_nir_type_for_glsl_base_type(glsl_get_sampler_result_type(type))) texel = emit_bitcast(ctx, spirv_builder_type_vector(&ctx->builder, base_type, 4), texel); bool use_sample = glsl_get_sampler_dim(type) == GLSL_SAMPLER_DIM_MS || glsl_get_sampler_dim(type) == GLSL_SAMPLER_DIM_SUBPASS_MS; SpvId sample = use_sample ? get_src(ctx, &intr->src[2], &atype) : 0; assert(nir_src_bit_size(intr->src[3]) == glsl_base_type_bit_size(glsl_get_sampler_result_type(type))); spirv_builder_emit_image_write(&ctx->builder, img, coord, texel, 0, sample, 0); } static SpvId extract_sparse_load(struct ntv_context *ctx, SpvId result, SpvId dest_type, nir_def *def) { /* Result Type must be an OpTypeStruct with two members. * The first member’s type must be an integer type scalar. * It holds a Residency Code that can be passed to OpImageSparseTexelsResident * - OpImageSparseRead spec */ uint32_t idx = 0; SpvId resident = spirv_builder_emit_composite_extract(&ctx->builder, spirv_builder_type_uint(&ctx->builder, 32), result, &idx, 1); idx = 1; /* normal vec4 return */ if (def->num_components == 4) result = spirv_builder_emit_composite_extract(&ctx->builder, dest_type, result, &idx, 1); else { /* shadow */ assert(def->num_components == 1); SpvId type = spirv_builder_type_float(&ctx->builder, def->bit_size); SpvId val[2]; /* pad to 2 components: the upcoming is_sparse_texels_resident instr will always use the * separate residency value, but the shader still expects this return to be a vec2, * so give it a vec2 */ val[0] = spirv_builder_emit_composite_extract(&ctx->builder, type, result, &idx, 1); val[1] = emit_float_const(ctx, def->bit_size, 0); result = spirv_builder_emit_composite_construct(&ctx->builder, get_fvec_type(ctx, def->bit_size, 2), val, 2); } assert(resident != 0); assert(def->index < ctx->num_defs); ctx->resident_defs[def->index] = resident; return result; } static void emit_image_deref_load(struct ntv_context *ctx, nir_intrinsic_instr *intr) { bool sparse = intr->intrinsic == nir_intrinsic_image_deref_sparse_load; nir_alu_type atype; SpvId img_var = get_src(ctx, &intr->src[0], &atype); nir_deref_instr *deref = nir_src_as_deref(intr->src[0]); nir_variable *var = nir_deref_instr_get_variable(deref); bool mediump = (var->data.precision == GLSL_PRECISION_MEDIUM || var->data.precision == GLSL_PRECISION_LOW); SpvId img_type = find_image_type(ctx, var); const struct glsl_type *type = glsl_without_array(var->type); SpvId base_type = get_glsl_basetype(ctx, glsl_get_sampler_result_type(type)); SpvId img = spirv_builder_emit_load(&ctx->builder, img_type, img_var); SpvId coord = get_image_coords(ctx, type, &intr->src[1]); bool use_sample = glsl_get_sampler_dim(type) == GLSL_SAMPLER_DIM_MS || glsl_get_sampler_dim(type) == GLSL_SAMPLER_DIM_SUBPASS_MS; SpvId sample = use_sample ? get_src(ctx, &intr->src[2], &atype) : 0; SpvId dest_type = spirv_builder_type_vector(&ctx->builder, base_type, intr->def.num_components); SpvId result = spirv_builder_emit_image_read(&ctx->builder, dest_type, img, coord, 0, sample, 0, sparse); if (sparse) result = extract_sparse_load(ctx, result, dest_type, &intr->def); if (!sparse && mediump) { spirv_builder_emit_decoration(&ctx->builder, result, SpvDecorationRelaxedPrecision); } store_def(ctx, intr->def.index, result, nir_get_nir_type_for_glsl_base_type(glsl_get_sampler_result_type(type))); } static void emit_image_deref_size(struct ntv_context *ctx, nir_intrinsic_instr *intr) { nir_alu_type atype; SpvId img_var = get_src(ctx, &intr->src[0], &atype); nir_deref_instr *deref = nir_src_as_deref(intr->src[0]); nir_variable *var = nir_deref_instr_get_variable(deref); SpvId img_type = find_image_type(ctx, var); const struct glsl_type *type = glsl_without_array(var->type); SpvId img = spirv_builder_emit_load(&ctx->builder, img_type, img_var); unsigned num_components = glsl_get_sampler_coordinate_components(type); /* SPIRV requires 2 components for non-array cube size */ if (glsl_get_sampler_dim(type) == GLSL_SAMPLER_DIM_CUBE && !glsl_sampler_type_is_array(type)) num_components = 2; spirv_builder_emit_cap(&ctx->builder, SpvCapabilityImageQuery); SpvId result = spirv_builder_emit_image_query_size(&ctx->builder, get_uvec_type(ctx, 32, num_components), img, 0); store_def(ctx, intr->def.index, result, nir_type_uint); } static void emit_image_deref_samples(struct ntv_context *ctx, nir_intrinsic_instr *intr) { nir_alu_type atype; SpvId img_var = get_src(ctx, &intr->src[0], &atype); nir_deref_instr *deref = nir_src_as_deref(intr->src[0]); nir_variable *var = nir_deref_instr_get_variable(deref); SpvId img_type = find_image_type(ctx, var); SpvId img = spirv_builder_emit_load(&ctx->builder, img_type, img_var); spirv_builder_emit_cap(&ctx->builder, SpvCapabilityImageQuery); SpvId result = spirv_builder_emit_unop(&ctx->builder, SpvOpImageQuerySamples, get_def_type(ctx, &intr->def, nir_type_uint), img); store_def(ctx, intr->def.index, result, nir_type_uint); } static void emit_image_intrinsic(struct ntv_context *ctx, nir_intrinsic_instr *intr) { nir_alu_type atype, ptype; SpvId param = get_src(ctx, &intr->src[3], &ptype); SpvId img_var = get_src(ctx, &intr->src[0], &atype); nir_deref_instr *deref = nir_src_as_deref(intr->src[0]); nir_variable *var = nir_deref_instr_get_variable(deref); const struct glsl_type *type = glsl_without_array(var->type); bool is_ms; type_to_dim(glsl_get_sampler_dim(type), &is_ms); SpvId sample = is_ms ? get_src(ctx, &intr->src[2], &atype) : emit_uint_const(ctx, 32, 0); SpvId coord = get_image_coords(ctx, type, &intr->src[1]); enum glsl_base_type glsl_result_type = glsl_get_sampler_result_type(type); SpvId base_type = get_glsl_basetype(ctx, glsl_result_type); SpvId texel = spirv_builder_emit_image_texel_pointer(&ctx->builder, base_type, img_var, coord, sample); SpvId param2 = 0; /* The type of Value must be the same as Result Type. * The type of the value pointed to by Pointer must be the same as Result Type. */ nir_alu_type ntype = nir_get_nir_type_for_glsl_base_type(glsl_result_type); if (ptype != ntype) { SpvId cast_type = get_def_type(ctx, &intr->def, ntype); param = emit_bitcast(ctx, cast_type, param); } if (intr->intrinsic == nir_intrinsic_image_deref_atomic_swap) { param2 = get_src(ctx, &intr->src[4], &ptype); if (ptype != ntype) { SpvId cast_type = get_def_type(ctx, &intr->def, ntype); param2 = emit_bitcast(ctx, cast_type, param2); } } handle_atomic_op(ctx, intr, texel, param, param2, ntype); } static void emit_ballot(struct ntv_context *ctx, nir_intrinsic_instr *intr) { spirv_builder_emit_cap(&ctx->builder, SpvCapabilitySubgroupBallotKHR); spirv_builder_emit_extension(&ctx->builder, "SPV_KHR_shader_ballot"); SpvId type = get_def_uvec_type(ctx, &intr->def); nir_alu_type atype; SpvId result = emit_unop(ctx, SpvOpSubgroupBallotKHR, type, get_src(ctx, &intr->src[0], &atype)); store_def(ctx, intr->def.index, result, nir_type_uint); } static void emit_read_first_invocation(struct ntv_context *ctx, nir_intrinsic_instr *intr) { spirv_builder_emit_cap(&ctx->builder, SpvCapabilitySubgroupBallotKHR); spirv_builder_emit_extension(&ctx->builder, "SPV_KHR_shader_ballot"); nir_alu_type atype; SpvId src = get_src(ctx, &intr->src[0], &atype); SpvId type = get_def_type(ctx, &intr->def, atype); SpvId result = emit_unop(ctx, SpvOpSubgroupFirstInvocationKHR, type, src); store_def(ctx, intr->def.index, result, atype); } static void emit_read_invocation(struct ntv_context *ctx, nir_intrinsic_instr *intr) { spirv_builder_emit_cap(&ctx->builder, SpvCapabilitySubgroupBallotKHR); spirv_builder_emit_extension(&ctx->builder, "SPV_KHR_shader_ballot"); nir_alu_type atype, itype; SpvId src = get_src(ctx, &intr->src[0], &atype); SpvId type = get_def_type(ctx, &intr->def, atype); SpvId result = emit_binop(ctx, SpvOpSubgroupReadInvocationKHR, type, src, get_src(ctx, &intr->src[1], &itype)); store_def(ctx, intr->def.index, result, atype); } static void emit_shader_clock(struct ntv_context *ctx, nir_intrinsic_instr *intr) { spirv_builder_emit_cap(&ctx->builder, SpvCapabilityShaderClockKHR); spirv_builder_emit_extension(&ctx->builder, "SPV_KHR_shader_clock"); SpvScope scope = get_scope(nir_intrinsic_memory_scope(intr)); SpvId type = get_def_type(ctx, &intr->def, nir_type_uint); SpvId result = spirv_builder_emit_unop_const(&ctx->builder, SpvOpReadClockKHR, type, scope); store_def(ctx, intr->def.index, result, nir_type_uint); } static void emit_is_sparse_texels_resident(struct ntv_context *ctx, nir_intrinsic_instr *intr) { spirv_builder_emit_cap(&ctx->builder, SpvCapabilitySparseResidency); SpvId type = get_def_type(ctx, &intr->def, nir_type_uint); unsigned index = intr->src[0].ssa->index; assert(index < ctx->num_defs); assert(ctx->resident_defs[index] != 0); SpvId resident = ctx->resident_defs[index]; SpvId result = spirv_builder_emit_unop(&ctx->builder, SpvOpImageSparseTexelsResident, type, resident); store_def(ctx, intr->def.index, result, nir_type_uint); } static void emit_vote(struct ntv_context *ctx, nir_intrinsic_instr *intr) { SpvOp op; switch (intr->intrinsic) { case nir_intrinsic_vote_all: op = SpvOpGroupNonUniformAll; break; case nir_intrinsic_vote_any: op = SpvOpGroupNonUniformAny; break; case nir_intrinsic_vote_ieq: case nir_intrinsic_vote_feq: op = SpvOpGroupNonUniformAllEqual; break; default: unreachable("unknown vote intrinsic"); } spirv_builder_emit_cap(&ctx->builder, SpvCapabilityGroupNonUniformVote); nir_alu_type atype; SpvId result = spirv_builder_emit_vote(&ctx->builder, op, get_src(ctx, &intr->src[0], &atype)); store_def(ctx, intr->def.index, result, nir_type_bool); } static void emit_is_helper_invocation(struct ntv_context *ctx, nir_intrinsic_instr *intr) { spirv_builder_emit_extension(&ctx->builder, "SPV_EXT_demote_to_helper_invocation"); SpvId result = spirv_is_helper_invocation(&ctx->builder); store_def(ctx, intr->def.index, result, nir_type_bool); } static void emit_barrier(struct ntv_context *ctx, nir_intrinsic_instr *intr) { SpvScope scope = get_scope(nir_intrinsic_execution_scope(intr)); SpvScope mem_scope = get_scope(nir_intrinsic_memory_scope(intr)); SpvMemorySemanticsMask semantics = 0; if (nir_intrinsic_memory_scope(intr) != SCOPE_NONE) { nir_variable_mode modes = nir_intrinsic_memory_modes(intr); if (modes & nir_var_image) semantics |= SpvMemorySemanticsImageMemoryMask; if (modes & nir_var_mem_shared) semantics |= SpvMemorySemanticsWorkgroupMemoryMask; if (modes & (nir_var_mem_ssbo | nir_var_mem_global)) semantics |= SpvMemorySemanticsUniformMemoryMask; if (modes & nir_var_mem_global) semantics |= SpvMemorySemanticsCrossWorkgroupMemoryMask; if (modes & (nir_var_shader_out | nir_var_mem_task_payload)) semantics |= SpvMemorySemanticsOutputMemoryMask; if (!modes) semantics = SpvMemorySemanticsWorkgroupMemoryMask | SpvMemorySemanticsUniformMemoryMask | SpvMemorySemanticsImageMemoryMask | SpvMemorySemanticsCrossWorkgroupMemoryMask; semantics |= SpvMemorySemanticsAcquireReleaseMask; } if (nir_intrinsic_execution_scope(intr) != SCOPE_NONE) spirv_builder_emit_control_barrier(&ctx->builder, scope, mem_scope, semantics); else spirv_builder_emit_memory_barrier(&ctx->builder, mem_scope, semantics); } static void emit_derivative(struct ntv_context *ctx, nir_intrinsic_instr *intr) { SpvOp op; switch (intr->intrinsic) { case nir_intrinsic_ddx: op = SpvOpDPdx; break; case nir_intrinsic_ddy: op = SpvOpDPdy; break; case nir_intrinsic_ddx_fine: op = SpvOpDPdxFine; break; case nir_intrinsic_ddy_fine: op = SpvOpDPdyFine; break; case nir_intrinsic_ddx_coarse: op = SpvOpDPdxCoarse; break; case nir_intrinsic_ddy_coarse: op = SpvOpDPdyCoarse; break; default: unreachable("invalid ddx/ddy"); } if (op != SpvOpDPdx && op != SpvOpDPdy) spirv_builder_emit_cap(&ctx->builder, SpvCapabilityDerivativeControl); SpvId type = get_fvec_type(ctx, intr->def.bit_size, intr->def.num_components); nir_alu_type atype; SpvId value = get_src(ctx, &intr->src[0], &atype); if (atype != nir_type_float) value = emit_bitcast(ctx, type, value); SpvId result = emit_unop(ctx, op, type, value); store_def(ctx, intr->def.index, result, nir_type_float); } static void emit_intrinsic(struct ntv_context *ctx, nir_intrinsic_instr *intr) { switch (intr->intrinsic) { case nir_intrinsic_decl_reg: /* Nothing to do */ break; case nir_intrinsic_load_reg: emit_load_reg(ctx, intr); break; case nir_intrinsic_store_reg: emit_store_reg(ctx, intr); break; case nir_intrinsic_terminate: emit_discard(ctx, intr); break; case nir_intrinsic_demote: spirv_builder_emit_cap(&ctx->builder, SpvCapabilityDemoteToHelperInvocation); spirv_builder_emit_demote(&ctx->builder); break; case nir_intrinsic_load_deref: emit_load_deref(ctx, intr); break; case nir_intrinsic_store_deref: emit_store_deref(ctx, intr); break; case nir_intrinsic_load_push_constant_zink: emit_load_push_const(ctx, intr); break; case nir_intrinsic_load_global: case nir_intrinsic_load_global_constant: emit_load_global(ctx, intr); break; case nir_intrinsic_store_global: emit_store_global(ctx, intr); break; case nir_intrinsic_load_front_face: emit_load_front_face(ctx, intr); break; case nir_intrinsic_load_view_index: emit_load_view_index(ctx, intr); break; case nir_intrinsic_load_base_instance: emit_load_uint_input(ctx, intr, &ctx->base_instance_var, "gl_BaseInstance", SpvBuiltInBaseInstance); break; case nir_intrinsic_load_instance_id: emit_load_uint_input(ctx, intr, &ctx->instance_id_var, "gl_InstanceId", SpvBuiltInInstanceIndex); break; case nir_intrinsic_load_base_vertex: emit_load_uint_input(ctx, intr, &ctx->base_vertex_var, "gl_BaseVertex", SpvBuiltInBaseVertex); break; case nir_intrinsic_load_draw_id: emit_load_uint_input(ctx, intr, &ctx->draw_id_var, "gl_DrawID", SpvBuiltInDrawIndex); break; case nir_intrinsic_load_vertex_id: emit_load_uint_input(ctx, intr, &ctx->vertex_id_var, "gl_VertexId", SpvBuiltInVertexIndex); break; case nir_intrinsic_load_primitive_id: emit_load_uint_input(ctx, intr, &ctx->primitive_id_var, "gl_PrimitiveIdIn", SpvBuiltInPrimitiveId); break; case nir_intrinsic_load_invocation_id: emit_load_uint_input(ctx, intr, &ctx->invocation_id_var, "gl_InvocationId", SpvBuiltInInvocationId); break; case nir_intrinsic_load_sample_id: spirv_builder_emit_cap(&ctx->builder, SpvCapabilitySampleRateShading); emit_load_uint_input(ctx, intr, &ctx->sample_id_var, "gl_SampleId", SpvBuiltInSampleId); break; case nir_intrinsic_load_point_coord_maybe_flipped: case nir_intrinsic_load_point_coord: emit_load_vec_input(ctx, intr, &ctx->point_coord_var, "gl_PointCoord", SpvBuiltInPointCoord, nir_type_float); break; case nir_intrinsic_load_sample_pos: emit_load_vec_input(ctx, intr, &ctx->sample_pos_var, "gl_SamplePosition", SpvBuiltInSamplePosition, nir_type_float); break; case nir_intrinsic_load_sample_mask_in: emit_load_uint_input(ctx, intr, &ctx->sample_mask_in_var, "gl_SampleMaskIn", SpvBuiltInSampleMask); break; case nir_intrinsic_emit_vertex: if (ctx->nir->info.gs.vertices_out) //skip vertex emission if !vertices_out spirv_builder_emit_vertex(&ctx->builder, nir_intrinsic_stream_id(intr), ctx->nir->info.stage == MESA_SHADER_GEOMETRY && util_bitcount(ctx->nir->info.gs.active_stream_mask) > 1); break; case nir_intrinsic_end_primitive: spirv_builder_end_primitive(&ctx->builder, nir_intrinsic_stream_id(intr), ctx->nir->info.stage == MESA_SHADER_GEOMETRY && util_bitcount(ctx->nir->info.gs.active_stream_mask) > 1); break; case nir_intrinsic_load_helper_invocation: emit_load_vec_input(ctx, intr, &ctx->helper_invocation_var, "gl_HelperInvocation", SpvBuiltInHelperInvocation, nir_type_bool); break; case nir_intrinsic_load_patch_vertices_in: emit_load_vec_input(ctx, intr, &ctx->tess_patch_vertices_in, "gl_PatchVerticesIn", SpvBuiltInPatchVertices, nir_type_int); break; case nir_intrinsic_load_tess_coord: emit_load_vec_input(ctx, intr, &ctx->tess_coord_var, "gl_TessCoord", SpvBuiltInTessCoord, nir_type_float); break; case nir_intrinsic_barrier: emit_barrier(ctx, intr); break; case nir_intrinsic_interp_deref_at_centroid: case nir_intrinsic_interp_deref_at_sample: case nir_intrinsic_interp_deref_at_offset: emit_interpolate(ctx, intr); break; case nir_intrinsic_deref_atomic: case nir_intrinsic_deref_atomic_swap: emit_deref_atomic_intrinsic(ctx, intr); break; case nir_intrinsic_shared_atomic: case nir_intrinsic_shared_atomic_swap: emit_shared_atomic_intrinsic(ctx, intr); break; case nir_intrinsic_global_atomic: case nir_intrinsic_global_atomic_swap: emit_global_atomic_intrinsic(ctx, intr); break; case nir_intrinsic_begin_invocation_interlock: case nir_intrinsic_end_invocation_interlock: spirv_builder_emit_interlock(&ctx->builder, intr->intrinsic == nir_intrinsic_end_invocation_interlock); break; case nir_intrinsic_get_ssbo_size: emit_get_ssbo_size(ctx, intr); break; case nir_intrinsic_image_deref_store: emit_image_deref_store(ctx, intr); break; case nir_intrinsic_image_deref_sparse_load: case nir_intrinsic_image_deref_load: emit_image_deref_load(ctx, intr); break; case nir_intrinsic_image_deref_size: emit_image_deref_size(ctx, intr); break; case nir_intrinsic_image_deref_samples: emit_image_deref_samples(ctx, intr); break; case nir_intrinsic_image_deref_atomic: case nir_intrinsic_image_deref_atomic_swap: emit_image_intrinsic(ctx, intr); break; case nir_intrinsic_load_workgroup_id: emit_load_vec_input(ctx, intr, &ctx->workgroup_id_var, "gl_WorkGroupID", SpvBuiltInWorkgroupId, nir_type_uint); break; case nir_intrinsic_load_num_workgroups: emit_load_vec_input(ctx, intr, &ctx->num_workgroups_var, "gl_NumWorkGroups", SpvBuiltInNumWorkgroups, nir_type_uint); break; case nir_intrinsic_load_local_invocation_id: emit_load_vec_input(ctx, intr, &ctx->local_invocation_id_var, "gl_LocalInvocationID", SpvBuiltInLocalInvocationId, nir_type_uint); break; case nir_intrinsic_load_global_invocation_id: emit_load_vec_input(ctx, intr, &ctx->global_invocation_id_var, "gl_GlobalInvocationID", SpvBuiltInGlobalInvocationId, nir_type_uint); break; case nir_intrinsic_load_local_invocation_index: emit_load_uint_input(ctx, intr, &ctx->local_invocation_index_var, "gl_LocalInvocationIndex", SpvBuiltInLocalInvocationIndex); break; #define LOAD_SHADER_BALLOT(lowercase, camelcase) \ case nir_intrinsic_load_##lowercase: \ emit_load_uint_input(ctx, intr, &ctx->lowercase##_var, "gl_"#camelcase, SpvBuiltIn##camelcase); \ break LOAD_SHADER_BALLOT(subgroup_id, SubgroupId); LOAD_SHADER_BALLOT(subgroup_eq_mask, SubgroupEqMask); LOAD_SHADER_BALLOT(subgroup_ge_mask, SubgroupGeMask); LOAD_SHADER_BALLOT(subgroup_invocation, SubgroupLocalInvocationId); LOAD_SHADER_BALLOT(subgroup_le_mask, SubgroupLeMask); LOAD_SHADER_BALLOT(subgroup_lt_mask, SubgroupLtMask); LOAD_SHADER_BALLOT(subgroup_size, SubgroupSize); case nir_intrinsic_ballot: emit_ballot(ctx, intr); break; case nir_intrinsic_read_first_invocation: emit_read_first_invocation(ctx, intr); break; case nir_intrinsic_read_invocation: emit_read_invocation(ctx, intr); break; case nir_intrinsic_load_workgroup_size: assert(ctx->local_group_size_var); store_def(ctx, intr->def.index, ctx->local_group_size_var, nir_type_uint); break; case nir_intrinsic_load_shared: emit_load_shared(ctx, intr); break; case nir_intrinsic_store_shared: emit_store_shared(ctx, intr); break; case nir_intrinsic_load_scratch: emit_load_scratch(ctx, intr); break; case nir_intrinsic_store_scratch: emit_store_scratch(ctx, intr); break; case nir_intrinsic_shader_clock: emit_shader_clock(ctx, intr); break; case nir_intrinsic_vote_all: case nir_intrinsic_vote_any: case nir_intrinsic_vote_ieq: case nir_intrinsic_vote_feq: emit_vote(ctx, intr); break; case nir_intrinsic_is_sparse_resident_zink: emit_is_sparse_texels_resident(ctx, intr); break; case nir_intrinsic_is_helper_invocation: emit_is_helper_invocation(ctx, intr); break; case nir_intrinsic_ddx: case nir_intrinsic_ddy: case nir_intrinsic_ddx_fine: case nir_intrinsic_ddy_fine: case nir_intrinsic_ddx_coarse: case nir_intrinsic_ddy_coarse: emit_derivative(ctx, intr); break; default: fprintf(stderr, "emit_intrinsic: not implemented (%s)\n", nir_intrinsic_infos[intr->intrinsic].name); unreachable("unsupported intrinsic"); } } static void emit_undef(struct ntv_context *ctx, nir_undef_instr *undef) { SpvId type = undef->def.bit_size == 1 ? get_bvec_type(ctx, undef->def.num_components) : get_uvec_type(ctx, undef->def.bit_size, undef->def.num_components); store_def(ctx, undef->def.index, spirv_builder_emit_undef(&ctx->builder, type), undef->def.bit_size == 1 ? nir_type_bool : nir_type_uint); } static SpvId get_src_float(struct ntv_context *ctx, nir_src *src) { nir_alu_type atype; SpvId def = get_src(ctx, src, &atype); if (atype == nir_type_float) return def; unsigned num_components = nir_src_num_components(*src); unsigned bit_size = nir_src_bit_size(*src); return bitcast_to_fvec(ctx, def, bit_size, num_components); } static SpvId get_src_int(struct ntv_context *ctx, nir_src *src) { nir_alu_type atype; SpvId def = get_src(ctx, src, &atype); if (atype == nir_type_int) return def; unsigned num_components = nir_src_num_components(*src); unsigned bit_size = nir_src_bit_size(*src); return bitcast_to_ivec(ctx, def, bit_size, num_components); } static inline bool tex_instr_is_lod_allowed(nir_tex_instr *tex) { /* This can only be used with an OpTypeImage that has a Dim operand of 1D, 2D, 3D, or Cube * - SPIR-V: 3.14. Image Operands */ return (tex->sampler_dim == GLSL_SAMPLER_DIM_1D || tex->sampler_dim == GLSL_SAMPLER_DIM_2D || tex->sampler_dim == GLSL_SAMPLER_DIM_3D || tex->sampler_dim == GLSL_SAMPLER_DIM_CUBE || /* External images are interpreted as 2D in type_to_dim, * so LOD is allowed */ tex->sampler_dim == GLSL_SAMPLER_DIM_EXTERNAL || /* RECT will always become 2D, so this is fine */ tex->sampler_dim == GLSL_SAMPLER_DIM_RECT); } static nir_variable * get_tex_srcs(struct ntv_context *ctx, nir_tex_instr *tex, nir_variable **bindless_var, unsigned *coord_components, struct spriv_tex_src *tex_src) { nir_variable *var = NULL; nir_alu_type atype; tex_src->sparse = tex->is_sparse; for (unsigned i = 0; i < tex->num_srcs; i++) { nir_const_value *cv; switch (tex->src[i].src_type) { case nir_tex_src_texture_deref: var = nir_deref_instr_get_variable(nir_instr_as_deref(tex->src[i].src.ssa->parent_instr)); tex_src->src = get_src(ctx, &tex->src[i].src, &atype); break; case nir_tex_src_sampler_deref: tex_src->cl_sampler = get_src(ctx, &tex->src[i].src, &atype); break; case nir_tex_src_coord: if (tex->op == nir_texop_txf || tex->op == nir_texop_txf_ms) tex_src->coord = get_src_int(ctx, &tex->src[i].src); else tex_src->coord = get_src_float(ctx, &tex->src[i].src); *coord_components = nir_src_num_components(tex->src[i].src); break; case nir_tex_src_projector: assert(nir_src_num_components(tex->src[i].src) == 1); tex_src->proj = get_src_float(ctx, &tex->src[i].src); assert(tex_src->proj != 0); break; case nir_tex_src_offset: cv = nir_src_as_const_value(tex->src[i].src); if (cv) { unsigned bit_size = nir_src_bit_size(tex->src[i].src); unsigned num_components = nir_src_num_components(tex->src[i].src); SpvId components[NIR_MAX_VEC_COMPONENTS]; for (int j = 0; j < num_components; ++j) { int64_t tmp = nir_const_value_as_int(cv[j], bit_size); components[j] = emit_int_const(ctx, bit_size, tmp); } if (num_components > 1) { SpvId type = get_ivec_type(ctx, bit_size, num_components); tex_src->const_offset = spirv_builder_const_composite(&ctx->builder, type, components, num_components); } else tex_src->const_offset = components[0]; } else tex_src->offset = get_src_int(ctx, &tex->src[i].src); break; case nir_tex_src_bias: assert(tex->op == nir_texop_txb); tex_src->bias = get_src_float(ctx, &tex->src[i].src); assert(tex_src->bias != 0); break; case nir_tex_src_min_lod: assert(nir_src_num_components(tex->src[i].src) == 1); tex_src->min_lod = get_src_float(ctx, &tex->src[i].src); assert(tex_src->min_lod != 0); break; case nir_tex_src_lod: assert(nir_src_num_components(tex->src[i].src) == 1); if (tex->op == nir_texop_txf || tex->op == nir_texop_txf_ms || tex->op == nir_texop_txs) tex_src->lod = get_src_int(ctx, &tex->src[i].src); else tex_src->lod = get_src_float(ctx, &tex->src[i].src); assert(tex_src->lod != 0); break; case nir_tex_src_ms_index: assert(nir_src_num_components(tex->src[i].src) == 1); tex_src->sample = get_src_int(ctx, &tex->src[i].src); break; case nir_tex_src_comparator: assert(nir_src_num_components(tex->src[i].src) == 1); tex_src->dref = get_src_float(ctx, &tex->src[i].src); assert(tex_src->dref != 0); break; case nir_tex_src_ddx: tex_src->dx = get_src_float(ctx, &tex->src[i].src); assert(tex_src->dx != 0); break; case nir_tex_src_ddy: tex_src->dy = get_src_float(ctx, &tex->src[i].src); assert(tex_src->dy != 0); break; case nir_tex_src_texture_offset: tex_src->tex_offset = get_src_int(ctx, &tex->src[i].src); break; case nir_tex_src_sampler_offset: case nir_tex_src_sampler_handle: /* don't care */ break; case nir_tex_src_texture_handle: tex_src->src = get_src(ctx, &tex->src[i].src, &atype); var = *bindless_var = nir_deref_instr_get_variable(nir_src_as_deref(tex->src[i].src)); break; default: fprintf(stderr, "texture source: %d\n", tex->src[i].src_type); unreachable("unknown texture source"); } } return var; } static SpvId get_texture_load(struct ntv_context *ctx, SpvId sampler_id, nir_tex_instr *tex, SpvId cl_sampler, SpvId image_type, SpvId sampled_type) { if (ctx->stage == MESA_SHADER_KERNEL) { SpvId image_load = spirv_builder_emit_load(&ctx->builder, image_type, sampler_id); if (nir_tex_instr_need_sampler(tex)) { SpvId sampler_load = spirv_builder_emit_load(&ctx->builder, spirv_builder_type_sampler(&ctx->builder), cl_sampler); return spirv_builder_emit_sampled_image(&ctx->builder, sampled_type, image_load, sampler_load); } else { return image_load; } } else { return spirv_builder_emit_load(&ctx->builder, sampled_type, sampler_id); } } static SpvId get_texop_dest_type(struct ntv_context *ctx, const nir_tex_instr *tex) { SpvId actual_dest_type; unsigned num_components = tex->def.num_components; switch (nir_alu_type_get_base_type(tex->dest_type)) { case nir_type_int: actual_dest_type = get_ivec_type(ctx, 32, num_components); break; case nir_type_uint: actual_dest_type = get_uvec_type(ctx, 32, num_components); break; case nir_type_float: actual_dest_type = get_fvec_type(ctx, 32, num_components); break; default: unreachable("unexpected nir_alu_type"); } return actual_dest_type; } static void move_tex_proj_to_coord(struct ntv_context *ctx, unsigned coord_components, struct spriv_tex_src *tex_src) { SpvId constituents[NIR_MAX_VEC_COMPONENTS + 1]; if (coord_components == 1) constituents[0] = tex_src->coord; else { assert(coord_components > 1); SpvId float_type = spirv_builder_type_float(&ctx->builder, 32); for (uint32_t i = 0; i < coord_components; ++i) constituents[i] = spirv_builder_emit_composite_extract(&ctx->builder, float_type, tex_src->coord, &i, 1); } constituents[coord_components++] = tex_src->proj; SpvId vec_type = get_fvec_type(ctx, 32, coord_components); tex_src->coord = spirv_builder_emit_composite_construct(&ctx->builder, vec_type, constituents, coord_components); } static SpvId get_tex_image_to_load( struct ntv_context *ctx, SpvId image_type, bool is_buffer, SpvId load) { return is_buffer || ctx->stage == MESA_SHADER_KERNEL ? load : spirv_builder_emit_image(&ctx->builder, image_type, load); } static SpvId emit_tex_readop(struct ntv_context *ctx, nir_variable *bindless_var, SpvId load, struct spriv_tex_src *tex_src, SpvId dest_type, bool is_buffer, nir_variable *var, SpvId image_type, nir_tex_instr *tex) { SpvId actual_dest_type = get_texop_dest_type(ctx, tex); SpvId result; if (tex_src->offset) spirv_builder_emit_cap(&ctx->builder, SpvCapabilityImageGatherExtended); if (tex_src->min_lod) spirv_builder_emit_cap(&ctx->builder, SpvCapabilityMinLod); if (tex->op == nir_texop_txf || tex->op == nir_texop_txf_ms || tex->op == nir_texop_tg4) { SpvId image = get_tex_image_to_load(ctx, image_type, is_buffer, load); if (tex->op == nir_texop_tg4) { if (tex_src->const_offset) spirv_builder_emit_cap(&ctx->builder, SpvCapabilityImageGatherExtended); result = spirv_builder_emit_image_gather(&ctx->builder, actual_dest_type, load, tex_src, emit_uint_const(ctx, 32, tex->component)); actual_dest_type = dest_type; } else { assert(tex->op == nir_texop_txf_ms || !tex_src->sample); bool is_ms; type_to_dim(glsl_get_sampler_dim(glsl_without_array(var->type)), &is_ms); assert(is_ms || !tex_src->sample); result = spirv_builder_emit_image_fetch(&ctx->builder, actual_dest_type, image, tex_src); } } else { if (tex->op == nir_texop_txl) tex_src->min_lod = 0; result = spirv_builder_emit_image_sample(&ctx->builder, actual_dest_type, load, tex_src); } if (!bindless_var && (var->data.precision == GLSL_PRECISION_MEDIUM || var->data.precision == GLSL_PRECISION_LOW)) { spirv_builder_emit_decoration(&ctx->builder, result, SpvDecorationRelaxedPrecision); } if (tex->is_sparse) result = extract_sparse_load(ctx, result, actual_dest_type, &tex->def); if (tex->def.bit_size != 32) { /* convert FP32 to FP16 */ result = emit_unop(ctx, SpvOpFConvert, dest_type, result); } return result; } static void emit_tex(struct ntv_context *ctx, nir_tex_instr *tex) { assert(tex->op == nir_texop_tex || tex->op == nir_texop_txb || tex->op == nir_texop_txl || tex->op == nir_texop_txd || tex->op == nir_texop_txf || tex->op == nir_texop_txf_ms || tex->op == nir_texop_txs || tex->op == nir_texop_lod || tex->op == nir_texop_tg4 || tex->op == nir_texop_texture_samples || tex->op == nir_texop_query_levels); struct spriv_tex_src tex_src = {0}; unsigned coord_components = 0; nir_variable *bindless_var = NULL; nir_variable *var = get_tex_srcs(ctx, tex, &bindless_var, &coord_components, &tex_src); assert(var); SpvId image_type = find_image_type(ctx, var); assert(image_type); bool is_buffer = glsl_get_sampler_dim(glsl_without_array(var->type)) == GLSL_SAMPLER_DIM_BUF; SpvId sampled_type = is_buffer ? image_type : spirv_builder_type_sampled_image(&ctx->builder, image_type); assert(sampled_type); SpvId sampler_id = tex_src.src; if (tex_src.tex_offset) { SpvId ptr = spirv_builder_type_pointer(&ctx->builder, SpvStorageClassUniformConstant, sampled_type); sampler_id = spirv_builder_emit_access_chain(&ctx->builder, ptr, sampler_id, &tex_src.tex_offset, 1); } SpvId load = get_texture_load(ctx, sampler_id, tex, tex_src.cl_sampler, image_type, sampled_type); if (tex->is_sparse) tex->def.num_components--; SpvId dest_type = get_def_type(ctx, &tex->def, tex->dest_type); if (nir_tex_instr_is_query(tex)) spirv_builder_emit_cap(&ctx->builder, SpvCapabilityImageQuery); if (!tex_instr_is_lod_allowed(tex)) tex_src.lod = 0; else if (ctx->stage != MESA_SHADER_FRAGMENT && tex->op == nir_texop_tex && ctx->explicit_lod && !tex_src.lod) tex_src.lod = emit_float_const(ctx, 32, 0.0); if (tex_src.proj && coord_components > 0) move_tex_proj_to_coord(ctx, coord_components, &tex_src); SpvId result = 0; switch (tex->op) { case nir_texop_txs: { SpvId image = get_tex_image_to_load(ctx, image_type, is_buffer, load); /* Its Dim operand must be one of 1D, 2D, 3D, or Cube * - OpImageQuerySizeLod specification * * Additionally, if its Dim is 1D, 2D, 3D, or Cube, * it must also have either an MS of 1 or a Sampled of 0 or 2. * - OpImageQuerySize specification * * all spirv samplers use these types */ if (!tex_src.lod && tex_instr_is_lod_allowed(tex)) tex_src.lod = emit_uint_const(ctx, 32, 0); result = spirv_builder_emit_image_query_size(&ctx->builder, dest_type, image, tex_src.lod); break; } case nir_texop_query_levels: { SpvId image = get_tex_image_to_load(ctx, image_type, is_buffer, load); result = spirv_builder_emit_image_query_levels(&ctx->builder, dest_type, image); break; } case nir_texop_texture_samples: { SpvId image = get_tex_image_to_load(ctx, image_type, is_buffer, load); result = spirv_builder_emit_unop(&ctx->builder, SpvOpImageQuerySamples, dest_type, image); break; } case nir_texop_lod: { result = spirv_builder_emit_image_query_lod(&ctx->builder, dest_type, load, tex_src.coord); break; } default: result = emit_tex_readop(ctx, bindless_var, load, &tex_src, dest_type, is_buffer, var, image_type, tex); break; } store_def(ctx, tex->def.index, result, tex->dest_type); if (tex->is_sparse) tex->def.num_components++; } static void start_block(struct ntv_context *ctx, SpvId label) { /* terminate previous block if needed */ if (ctx->block_started) spirv_builder_emit_branch(&ctx->builder, label); /* start new block */ spirv_builder_label(&ctx->builder, label); ctx->block_started = true; } static void branch(struct ntv_context *ctx, SpvId label) { assert(ctx->block_started); spirv_builder_emit_branch(&ctx->builder, label); ctx->block_started = false; } static void branch_conditional(struct ntv_context *ctx, SpvId condition, SpvId then_id, SpvId else_id) { assert(ctx->block_started); spirv_builder_emit_branch_conditional(&ctx->builder, condition, then_id, else_id); ctx->block_started = false; } static void emit_jump(struct ntv_context *ctx, nir_jump_instr *jump) { switch (jump->type) { case nir_jump_break: assert(ctx->loop_break); branch(ctx, ctx->loop_break); break; case nir_jump_continue: assert(ctx->loop_cont); branch(ctx, ctx->loop_cont); break; default: unreachable("Unsupported jump type\n"); } } static void emit_deref_var(struct ntv_context *ctx, nir_deref_instr *deref) { assert(deref->deref_type == nir_deref_type_var); struct hash_entry *he = _mesa_hash_table_search(ctx->vars, deref->var); assert(he); SpvId result = (SpvId)(intptr_t)he->data; store_def(ctx, deref->def.index, result, get_nir_alu_type(deref->type)); } static void emit_deref_array(struct ntv_context *ctx, nir_deref_instr *deref) { assert(deref->deref_type == nir_deref_type_array); nir_variable *var = nir_deref_instr_get_variable(deref); if (!nir_src_is_always_uniform(deref->arr.index)) { if (deref->modes & nir_var_mem_ubo) spirv_builder_emit_cap(&ctx->builder, SpvCapabilityUniformBufferArrayDynamicIndexing); if (deref->modes & nir_var_mem_ssbo) spirv_builder_emit_cap(&ctx->builder, SpvCapabilityStorageBufferArrayDynamicIndexing); if (deref->modes & (nir_var_uniform | nir_var_image)) { const struct glsl_type *type = glsl_without_array(var->type); assert(glsl_type_is_sampler(type) || glsl_type_is_image(type)); if (glsl_type_is_sampler(type)) spirv_builder_emit_cap(&ctx->builder, SpvCapabilitySampledImageArrayDynamicIndexing); else spirv_builder_emit_cap(&ctx->builder, SpvCapabilityStorageImageArrayDynamicIndexing); } } SpvStorageClass storage_class = get_storage_class(var); SpvId type; nir_alu_type atype = nir_type_uint; SpvId base = get_src(ctx, &deref->parent, &atype); switch (var->data.mode) { case nir_var_mem_ubo: case nir_var_mem_ssbo: base = get_src(ctx, &deref->parent, &atype); /* this is either the array deref or the array deref */ if (glsl_type_is_struct_or_ifc(deref->type)) { /* array */ type = get_bo_struct_type(ctx, var); break; } /* array */ FALLTHROUGH; case nir_var_function_temp: case nir_var_shader_in: case nir_var_shader_out: base = get_src(ctx, &deref->parent, &atype); type = get_glsl_type(ctx, deref->type); break; case nir_var_uniform: case nir_var_image: { base = get_src(ctx, &deref->parent, &atype); const struct glsl_type *gtype = glsl_without_array(deref->type); type = get_image_type(ctx, var, glsl_type_is_sampler(gtype), glsl_get_sampler_dim(gtype) == GLSL_SAMPLER_DIM_BUF); break; } default: unreachable("Unsupported nir_variable_mode\n"); } nir_alu_type itype; SpvId index = get_src(ctx, &deref->arr.index, &itype); if (itype == nir_type_float) index = emit_bitcast(ctx, get_uvec_type(ctx, 32, 1), index); SpvId ptr_type = spirv_builder_type_pointer(&ctx->builder, storage_class, type); SpvId result = spirv_builder_emit_access_chain(&ctx->builder, ptr_type, base, &index, 1); /* uint is a bit of a lie here, it's really just an opaque type */ store_def(ctx, deref->def.index, result, get_nir_alu_type(deref->type)); } static void emit_deref_struct(struct ntv_context *ctx, nir_deref_instr *deref) { assert(deref->deref_type == nir_deref_type_struct); nir_variable *var = nir_deref_instr_get_variable(deref); SpvStorageClass storage_class = get_storage_class(var); SpvId index = emit_uint_const(ctx, 32, deref->strct.index); SpvId type = (var->data.mode & (nir_var_mem_ubo | nir_var_mem_ssbo)) ? get_bo_array_type(ctx, var) : get_glsl_type(ctx, deref->type); SpvId ptr_type = spirv_builder_type_pointer(&ctx->builder, storage_class, type); nir_alu_type atype; SpvId result = spirv_builder_emit_access_chain(&ctx->builder, ptr_type, get_src(ctx, &deref->parent, &atype), &index, 1); /* uint is a bit of a lie here, it's really just an opaque type */ store_def(ctx, deref->def.index, result, get_nir_alu_type(deref->type)); } static void emit_deref(struct ntv_context *ctx, nir_deref_instr *deref) { switch (deref->deref_type) { case nir_deref_type_var: emit_deref_var(ctx, deref); break; case nir_deref_type_array: emit_deref_array(ctx, deref); break; case nir_deref_type_struct: emit_deref_struct(ctx, deref); break; default: unreachable("unexpected deref_type"); } } static void emit_block(struct ntv_context *ctx, struct nir_block *block) { start_block(ctx, block_label(ctx, block)); nir_foreach_instr(instr, block) { switch (instr->type) { case nir_instr_type_alu: emit_alu(ctx, nir_instr_as_alu(instr)); break; case nir_instr_type_intrinsic: emit_intrinsic(ctx, nir_instr_as_intrinsic(instr)); break; case nir_instr_type_load_const: emit_load_const(ctx, nir_instr_as_load_const(instr)); break; case nir_instr_type_undef: emit_undef(ctx, nir_instr_as_undef(instr)); break; case nir_instr_type_tex: emit_tex(ctx, nir_instr_as_tex(instr)); break; case nir_instr_type_phi: unreachable("nir_instr_type_phi not supported"); break; case nir_instr_type_jump: emit_jump(ctx, nir_instr_as_jump(instr)); break; case nir_instr_type_call: unreachable("nir_instr_type_call not supported"); break; case nir_instr_type_parallel_copy: unreachable("nir_instr_type_parallel_copy not supported"); break; case nir_instr_type_deref: emit_deref(ctx, nir_instr_as_deref(instr)); break; case nir_instr_type_debug_info: unreachable("nir_instr_type_debug_info not supported"); break; } } } static void emit_cf_list(struct ntv_context *ctx, struct exec_list *list); static SpvId get_src_bool(struct ntv_context *ctx, nir_src *src) { assert(nir_src_bit_size(*src) == 1); nir_alu_type atype; return get_src(ctx, src, &atype); } static void emit_if(struct ntv_context *ctx, nir_if *if_stmt) { SpvId condition = get_src_bool(ctx, &if_stmt->condition); SpvId header_id = spirv_builder_new_id(&ctx->builder); SpvId then_id = block_label(ctx, nir_if_first_then_block(if_stmt)); SpvId endif_id = spirv_builder_new_id(&ctx->builder); SpvId else_id = endif_id; bool has_else = !exec_list_is_empty(&if_stmt->else_list); if (has_else) { assert(nir_if_first_else_block(if_stmt)->index < ctx->num_blocks); else_id = block_label(ctx, nir_if_first_else_block(if_stmt)); } /* create a header-block */ start_block(ctx, header_id); spirv_builder_emit_selection_merge(&ctx->builder, endif_id, SpvSelectionControlMaskNone); branch_conditional(ctx, condition, then_id, else_id); emit_cf_list(ctx, &if_stmt->then_list); if (has_else) { if (ctx->block_started) branch(ctx, endif_id); emit_cf_list(ctx, &if_stmt->else_list); } start_block(ctx, endif_id); } static void emit_loop(struct ntv_context *ctx, nir_loop *loop) { assert(!nir_loop_has_continue_construct(loop)); SpvId header_id = spirv_builder_new_id(&ctx->builder); SpvId begin_id = block_label(ctx, nir_loop_first_block(loop)); SpvId break_id = spirv_builder_new_id(&ctx->builder); SpvId cont_id = spirv_builder_new_id(&ctx->builder); /* create a header-block */ start_block(ctx, header_id); spirv_builder_loop_merge(&ctx->builder, break_id, cont_id, SpvLoopControlMaskNone); branch(ctx, begin_id); SpvId save_break = ctx->loop_break; SpvId save_cont = ctx->loop_cont; ctx->loop_break = break_id; ctx->loop_cont = cont_id; emit_cf_list(ctx, &loop->body); ctx->loop_break = save_break; ctx->loop_cont = save_cont; /* loop->body may have already ended our block */ if (ctx->block_started) branch(ctx, cont_id); start_block(ctx, cont_id); branch(ctx, header_id); start_block(ctx, break_id); } static void emit_cf_list(struct ntv_context *ctx, struct exec_list *list) { foreach_list_typed(nir_cf_node, node, node, list) { switch (node->type) { case nir_cf_node_block: emit_block(ctx, nir_cf_node_as_block(node)); 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: unreachable("nir_cf_node_function not supported"); break; } } } static SpvExecutionMode get_input_prim_type_mode(enum mesa_prim type) { switch (type) { case MESA_PRIM_POINTS: return SpvExecutionModeInputPoints; case MESA_PRIM_LINES: case MESA_PRIM_LINE_LOOP: case MESA_PRIM_LINE_STRIP: return SpvExecutionModeInputLines; case MESA_PRIM_TRIANGLE_STRIP: case MESA_PRIM_TRIANGLES: case MESA_PRIM_TRIANGLE_FAN: return SpvExecutionModeTriangles; case MESA_PRIM_QUADS: case MESA_PRIM_QUAD_STRIP: return SpvExecutionModeQuads; break; case MESA_PRIM_POLYGON: unreachable("handle polygons in gs"); break; case MESA_PRIM_LINES_ADJACENCY: case MESA_PRIM_LINE_STRIP_ADJACENCY: return SpvExecutionModeInputLinesAdjacency; case MESA_PRIM_TRIANGLES_ADJACENCY: case MESA_PRIM_TRIANGLE_STRIP_ADJACENCY: return SpvExecutionModeInputTrianglesAdjacency; break; default: debug_printf("unknown geometry shader input mode %u\n", type); unreachable("error!"); break; } return 0; } static SpvExecutionMode get_output_prim_type_mode(enum mesa_prim type) { switch (type) { case MESA_PRIM_POINTS: return SpvExecutionModeOutputPoints; case MESA_PRIM_LINES: case MESA_PRIM_LINE_LOOP: unreachable("MESA_PRIM_LINES/LINE_LOOP passed as gs output"); break; case MESA_PRIM_LINE_STRIP: return SpvExecutionModeOutputLineStrip; case MESA_PRIM_TRIANGLE_STRIP: return SpvExecutionModeOutputTriangleStrip; case MESA_PRIM_TRIANGLES: case MESA_PRIM_TRIANGLE_FAN: //FIXME: not sure if right for output return SpvExecutionModeTriangles; case MESA_PRIM_QUADS: case MESA_PRIM_QUAD_STRIP: return SpvExecutionModeQuads; case MESA_PRIM_POLYGON: unreachable("handle polygons in gs"); break; case MESA_PRIM_LINES_ADJACENCY: case MESA_PRIM_LINE_STRIP_ADJACENCY: unreachable("handle line adjacency in gs"); break; case MESA_PRIM_TRIANGLES_ADJACENCY: case MESA_PRIM_TRIANGLE_STRIP_ADJACENCY: unreachable("handle triangle adjacency in gs"); break; default: debug_printf("unknown geometry shader output mode %u\n", type); unreachable("error!"); break; } return 0; } static SpvExecutionMode get_depth_layout_mode(enum gl_frag_depth_layout depth_layout) { switch (depth_layout) { case FRAG_DEPTH_LAYOUT_NONE: case FRAG_DEPTH_LAYOUT_ANY: return SpvExecutionModeDepthReplacing; case FRAG_DEPTH_LAYOUT_GREATER: return SpvExecutionModeDepthGreater; case FRAG_DEPTH_LAYOUT_LESS: return SpvExecutionModeDepthLess; case FRAG_DEPTH_LAYOUT_UNCHANGED: return SpvExecutionModeDepthUnchanged; default: unreachable("unexpected depth layout"); } } static SpvExecutionMode get_primitive_mode(enum tess_primitive_mode primitive_mode) { switch (primitive_mode) { case TESS_PRIMITIVE_TRIANGLES: return SpvExecutionModeTriangles; case TESS_PRIMITIVE_QUADS: return SpvExecutionModeQuads; case TESS_PRIMITIVE_ISOLINES: return SpvExecutionModeIsolines; default: unreachable("unknown tess prim type!"); } } static SpvExecutionMode get_spacing(enum gl_tess_spacing spacing) { switch (spacing) { case TESS_SPACING_EQUAL: return SpvExecutionModeSpacingEqual; case TESS_SPACING_FRACTIONAL_ODD: return SpvExecutionModeSpacingFractionalOdd; case TESS_SPACING_FRACTIONAL_EVEN: return SpvExecutionModeSpacingFractionalEven; default: unreachable("unknown tess spacing!"); } } struct spirv_shader * nir_to_spirv(struct nir_shader *s, const struct zink_shader_info *sinfo, const struct zink_screen *screen) { const uint32_t spirv_version = screen->spirv_version; struct spirv_shader *ret = NULL; struct ntv_context ctx = {0}; ctx.mem_ctx = ralloc_context(NULL); ctx.nir = s; ctx.builder.mem_ctx = ctx.mem_ctx; assert(spirv_version >= SPIRV_VERSION(1, 0)); ctx.spirv_1_4_interfaces = spirv_version >= SPIRV_VERSION(1, 4); ctx.bindless_set_idx = sinfo->bindless_set_idx; ctx.glsl_types = _mesa_pointer_hash_table_create(ctx.mem_ctx); ctx.bo_array_types = _mesa_pointer_hash_table_create(ctx.mem_ctx); ctx.bo_struct_types = _mesa_pointer_hash_table_create(ctx.mem_ctx); if (!ctx.glsl_types || !ctx.bo_array_types || !ctx.bo_struct_types || !_mesa_hash_table_init(&ctx.image_types, ctx.mem_ctx, _mesa_hash_pointer, _mesa_key_pointer_equal)) goto fail; spirv_builder_emit_cap(&ctx.builder, SpvCapabilityShader); switch (s->info.stage) { case MESA_SHADER_FRAGMENT: if (s->info.fs.uses_sample_shading) spirv_builder_emit_cap(&ctx.builder, SpvCapabilitySampleRateShading); if (s->info.fs.uses_discard && spirv_version < SPIRV_VERSION(1, 6) && screen->info.have_EXT_shader_demote_to_helper_invocation) spirv_builder_emit_extension(&ctx.builder, "SPV_EXT_demote_to_helper_invocation"); break; case MESA_SHADER_VERTEX: if (BITSET_TEST(s->info.system_values_read, SYSTEM_VALUE_INSTANCE_ID) || BITSET_TEST(s->info.system_values_read, SYSTEM_VALUE_DRAW_ID) || BITSET_TEST(s->info.system_values_read, SYSTEM_VALUE_BASE_INSTANCE) || BITSET_TEST(s->info.system_values_read, SYSTEM_VALUE_BASE_VERTEX)) { if (spirv_version < SPIRV_VERSION(1, 3)) spirv_builder_emit_extension(&ctx.builder, "SPV_KHR_shader_draw_parameters"); spirv_builder_emit_cap(&ctx.builder, SpvCapabilityDrawParameters); } break; case MESA_SHADER_TESS_CTRL: case MESA_SHADER_TESS_EVAL: spirv_builder_emit_cap(&ctx.builder, SpvCapabilityTessellation); /* TODO: check features for this */ if (s->info.outputs_written & BITFIELD64_BIT(VARYING_SLOT_PSIZ)) spirv_builder_emit_cap(&ctx.builder, SpvCapabilityTessellationPointSize); break; case MESA_SHADER_GEOMETRY: spirv_builder_emit_cap(&ctx.builder, SpvCapabilityGeometry); if (s->info.gs.active_stream_mask) spirv_builder_emit_cap(&ctx.builder, SpvCapabilityGeometryStreams); if (s->info.outputs_written & BITFIELD64_BIT(VARYING_SLOT_PSIZ)) spirv_builder_emit_cap(&ctx.builder, SpvCapabilityGeometryPointSize); break; default: ; } if (s->info.stage < MESA_SHADER_GEOMETRY) { if (s->info.outputs_written & BITFIELD64_BIT(VARYING_SLOT_LAYER) || s->info.inputs_read & BITFIELD64_BIT(VARYING_SLOT_LAYER)) { if (spirv_version >= SPIRV_VERSION(1, 5)) spirv_builder_emit_cap(&ctx.builder, SpvCapabilityShaderLayer); else { spirv_builder_emit_extension(&ctx.builder, "SPV_EXT_shader_viewport_index_layer"); spirv_builder_emit_cap(&ctx.builder, SpvCapabilityShaderViewportIndexLayerEXT); } } } else if (s->info.stage == MESA_SHADER_FRAGMENT) { /* incredibly, this is legal and intended. * https://github.com/KhronosGroup/SPIRV-Registry/issues/95 */ if (s->info.inputs_read & (BITFIELD64_BIT(VARYING_SLOT_LAYER) | BITFIELD64_BIT(VARYING_SLOT_PRIMITIVE_ID))) spirv_builder_emit_cap(&ctx.builder, SpvCapabilityGeometry); } if (s->info.num_ssbos && spirv_version < SPIRV_VERSION(1, 1)) spirv_builder_emit_extension(&ctx.builder, "SPV_KHR_storage_buffer_storage_class"); if (s->info.stage < MESA_SHADER_FRAGMENT && s->info.outputs_written & BITFIELD64_BIT(VARYING_SLOT_VIEWPORT)) { if (s->info.stage < MESA_SHADER_GEOMETRY) spirv_builder_emit_cap(&ctx.builder, SpvCapabilityShaderViewportIndex); else spirv_builder_emit_cap(&ctx.builder, SpvCapabilityMultiViewport); } if (s->info.stage > MESA_SHADER_VERTEX && s->info.inputs_read & BITFIELD64_BIT(VARYING_SLOT_VIEWPORT)) { if (s->info.stage < MESA_SHADER_GEOMETRY) spirv_builder_emit_cap(&ctx.builder, SpvCapabilityShaderViewportIndex); else spirv_builder_emit_cap(&ctx.builder, SpvCapabilityMultiViewport); } ctx.stage = s->info.stage; ctx.sinfo = sinfo; ctx.GLSL_std_450 = spirv_builder_import(&ctx.builder, "GLSL.std.450"); ctx.explicit_lod = true; spirv_builder_emit_source(&ctx.builder, SpvSourceLanguageUnknown, 0); SpvAddressingModel model = SpvAddressingModelLogical; if (gl_shader_stage_is_compute(s->info.stage)) { if (s->info.cs.ptr_size == 32) model = SpvAddressingModelPhysical32; else if (s->info.cs.ptr_size == 64) { spirv_builder_emit_cap(&ctx.builder, SpvCapabilityPhysicalStorageBufferAddresses); model = SpvAddressingModelPhysicalStorageBuffer64; } else model = SpvAddressingModelLogical; } if (ctx.sinfo->have_vulkan_memory_model) { spirv_builder_emit_cap(&ctx.builder, SpvCapabilityVulkanMemoryModel); spirv_builder_emit_cap(&ctx.builder, SpvCapabilityVulkanMemoryModelDeviceScope); spirv_builder_emit_mem_model(&ctx.builder, model, SpvMemoryModelVulkan); } else { spirv_builder_emit_mem_model(&ctx.builder, model, SpvMemoryModelGLSL450); } if (s->info.stage == MESA_SHADER_FRAGMENT && s->info.outputs_written & BITFIELD64_BIT(FRAG_RESULT_STENCIL)) { spirv_builder_emit_extension(&ctx.builder, "SPV_EXT_shader_stencil_export"); spirv_builder_emit_cap(&ctx.builder, SpvCapabilityStencilExportEXT); } SpvExecutionModel exec_model; switch (s->info.stage) { case MESA_SHADER_VERTEX: exec_model = SpvExecutionModelVertex; break; case MESA_SHADER_TESS_CTRL: exec_model = SpvExecutionModelTessellationControl; break; case MESA_SHADER_TESS_EVAL: exec_model = SpvExecutionModelTessellationEvaluation; break; case MESA_SHADER_GEOMETRY: exec_model = SpvExecutionModelGeometry; break; case MESA_SHADER_FRAGMENT: exec_model = SpvExecutionModelFragment; break; case MESA_SHADER_COMPUTE: case MESA_SHADER_KERNEL: exec_model = SpvExecutionModelGLCompute; break; default: unreachable("invalid stage"); } SpvId type_void = spirv_builder_type_void(&ctx.builder); SpvId type_void_func = spirv_builder_type_function(&ctx.builder, type_void, NULL, 0); SpvId entry_point = spirv_builder_new_id(&ctx.builder); spirv_builder_emit_name(&ctx.builder, entry_point, "main"); ctx.vars = _mesa_hash_table_create(ctx.mem_ctx, _mesa_hash_pointer, _mesa_key_pointer_equal); nir_foreach_variable_with_modes(var, s, nir_var_mem_push_const) input_var_init(&ctx, var); nir_foreach_shader_in_variable(var, s) emit_input(&ctx, var); int max_output = 0; nir_foreach_shader_out_variable(var, s) { /* ignore SPIR-V built-ins, tagged with a sentinel value */ if (var->data.driver_location != UINT_MAX) { assert(var->data.driver_location < INT_MAX); unsigned extent = glsl_count_attribute_slots(var->type, false); max_output = MAX2(max_output, (int)var->data.driver_location + extent); } emit_output(&ctx, var); } uint32_t tcs_vertices_out_word = 0; unsigned ubo_counter[2] = {0}; nir_foreach_variable_with_modes(var, s, nir_var_mem_ubo) ubo_counter[var->data.driver_location != 0]++; nir_foreach_variable_with_modes(var, s, nir_var_mem_ubo) emit_bo(&ctx, var, ubo_counter[var->data.driver_location != 0] > 1); unsigned ssbo_counter = 0; nir_foreach_variable_with_modes(var, s, nir_var_mem_ssbo) ssbo_counter++; nir_foreach_variable_with_modes(var, s, nir_var_mem_ssbo) emit_bo(&ctx, var, ssbo_counter > 1); nir_foreach_variable_with_modes(var, s, nir_var_image) ctx.image_var[var->data.driver_location] = var; nir_foreach_variable_with_modes(var, s, nir_var_uniform) { if (glsl_type_is_sampler(glsl_without_array(var->type))) { if (var->data.descriptor_set == ctx.bindless_set_idx) ctx.bindless_sampler_var[var->data.driver_location] = var; else ctx.sampler_var[var->data.driver_location] = var; ctx.last_sampler = MAX2(ctx.last_sampler, var->data.driver_location); } } nir_foreach_variable_with_modes(var, s, nir_var_image | nir_var_uniform) { const struct glsl_type *type = glsl_without_array(var->type); if (glsl_type_is_bare_sampler(type)) emit_sampler(&ctx, var); else if (glsl_type_is_sampler(type)) emit_image(&ctx, var, get_bare_image_type(&ctx, var, true)); else if (glsl_type_is_image(type)) emit_image(&ctx, var, get_bare_image_type(&ctx, var, false)); } if (sinfo->float_controls.flush_denorms) { unsigned execution_mode = s->info.float_controls_execution_mode; bool flush_16_bit = nir_is_denorm_flush_to_zero(execution_mode, 16); bool flush_32_bit = nir_is_denorm_flush_to_zero(execution_mode, 32); bool flush_64_bit = nir_is_denorm_flush_to_zero(execution_mode, 64); bool preserve_16_bit = nir_is_denorm_preserve(execution_mode, 16); bool preserve_32_bit = nir_is_denorm_preserve(execution_mode, 32); bool preserve_64_bit = nir_is_denorm_preserve(execution_mode, 64); bool emit_cap_flush = false; bool emit_cap_preserve = false; if (!sinfo->float_controls.denorms_all_independence) { bool flush = flush_16_bit && flush_64_bit; bool preserve = preserve_16_bit && preserve_64_bit; if (!sinfo->float_controls.denorms_32_bit_independence) { flush = flush && flush_32_bit; preserve = preserve && preserve_32_bit; flush_32_bit = flush; preserve_32_bit = preserve; } flush_16_bit = flush; flush_64_bit = flush; preserve_16_bit = preserve; preserve_64_bit = preserve; } if (flush_16_bit && sinfo->float_controls.flush_denorms & BITFIELD_BIT(0)) { emit_cap_flush = true; spirv_builder_emit_exec_mode_literal(&ctx.builder, entry_point, SpvExecutionModeDenormFlushToZero, 16); } if (flush_32_bit && sinfo->float_controls.flush_denorms & BITFIELD_BIT(1)) { emit_cap_flush = true; spirv_builder_emit_exec_mode_literal(&ctx.builder, entry_point, SpvExecutionModeDenormFlushToZero, 32); } if (flush_64_bit && sinfo->float_controls.flush_denorms & BITFIELD_BIT(2)) { emit_cap_flush = true; spirv_builder_emit_exec_mode_literal(&ctx.builder, entry_point, SpvExecutionModeDenormFlushToZero, 64); } if (preserve_16_bit && sinfo->float_controls.preserve_denorms & BITFIELD_BIT(0)) { emit_cap_preserve = true; spirv_builder_emit_exec_mode_literal(&ctx.builder, entry_point, SpvExecutionModeDenormPreserve, 16); } if (preserve_32_bit && sinfo->float_controls.preserve_denorms & BITFIELD_BIT(1)) { emit_cap_preserve = true; spirv_builder_emit_exec_mode_literal(&ctx.builder, entry_point, SpvExecutionModeDenormPreserve, 32); } if (preserve_64_bit && sinfo->float_controls.preserve_denorms & BITFIELD_BIT(2)) { emit_cap_preserve = true; spirv_builder_emit_exec_mode_literal(&ctx.builder, entry_point, SpvExecutionModeDenormPreserve, 64); } if (emit_cap_flush) spirv_builder_emit_cap(&ctx.builder, SpvCapabilityDenormFlushToZero); if (emit_cap_preserve) spirv_builder_emit_cap(&ctx.builder, SpvCapabilityDenormPreserve); } switch (s->info.stage) { case MESA_SHADER_FRAGMENT: spirv_builder_emit_exec_mode(&ctx.builder, entry_point, SpvExecutionModeOriginUpperLeft); if (s->info.outputs_written & BITFIELD64_BIT(FRAG_RESULT_DEPTH)) spirv_builder_emit_exec_mode(&ctx.builder, entry_point, get_depth_layout_mode(s->info.fs.depth_layout)); if (s->info.outputs_written & BITFIELD64_BIT(FRAG_RESULT_STENCIL)) spirv_builder_emit_exec_mode(&ctx.builder, entry_point, SpvExecutionModeStencilRefReplacingEXT); if (s->info.fs.early_fragment_tests) spirv_builder_emit_exec_mode(&ctx.builder, entry_point, SpvExecutionModeEarlyFragmentTests); if (s->info.fs.post_depth_coverage) { spirv_builder_emit_extension(&ctx.builder, "SPV_KHR_post_depth_coverage"); spirv_builder_emit_cap(&ctx.builder, SpvCapabilitySampleMaskPostDepthCoverage); spirv_builder_emit_exec_mode(&ctx.builder, entry_point, SpvExecutionModePostDepthCoverage); } if (s->info.fs.pixel_interlock_ordered || s->info.fs.pixel_interlock_unordered || s->info.fs.sample_interlock_ordered || s->info.fs.sample_interlock_unordered) spirv_builder_emit_extension(&ctx.builder, "SPV_EXT_fragment_shader_interlock"); if (s->info.fs.pixel_interlock_ordered || s->info.fs.pixel_interlock_unordered) spirv_builder_emit_cap(&ctx.builder, SpvCapabilityFragmentShaderPixelInterlockEXT); if (s->info.fs.sample_interlock_ordered || s->info.fs.sample_interlock_unordered) spirv_builder_emit_cap(&ctx.builder, SpvCapabilityFragmentShaderSampleInterlockEXT); if (s->info.fs.pixel_interlock_ordered) spirv_builder_emit_exec_mode(&ctx.builder, entry_point, SpvExecutionModePixelInterlockOrderedEXT); if (s->info.fs.pixel_interlock_unordered) spirv_builder_emit_exec_mode(&ctx.builder, entry_point, SpvExecutionModePixelInterlockUnorderedEXT); if (s->info.fs.sample_interlock_ordered) spirv_builder_emit_exec_mode(&ctx.builder, entry_point, SpvExecutionModeSampleInterlockOrderedEXT); if (s->info.fs.sample_interlock_unordered) spirv_builder_emit_exec_mode(&ctx.builder, entry_point, SpvExecutionModeSampleInterlockUnorderedEXT); break; case MESA_SHADER_TESS_CTRL: tcs_vertices_out_word = spirv_builder_emit_exec_mode_literal(&ctx.builder, entry_point, SpvExecutionModeOutputVertices, s->info.tess.tcs_vertices_out); break; case MESA_SHADER_TESS_EVAL: spirv_builder_emit_exec_mode(&ctx.builder, entry_point, get_primitive_mode(s->info.tess._primitive_mode)); spirv_builder_emit_exec_mode(&ctx.builder, entry_point, s->info.tess.ccw ? SpvExecutionModeVertexOrderCcw : SpvExecutionModeVertexOrderCw); spirv_builder_emit_exec_mode(&ctx.builder, entry_point, get_spacing(s->info.tess.spacing)); if (s->info.tess.point_mode) spirv_builder_emit_exec_mode(&ctx.builder, entry_point, SpvExecutionModePointMode); break; case MESA_SHADER_GEOMETRY: spirv_builder_emit_exec_mode(&ctx.builder, entry_point, get_input_prim_type_mode(s->info.gs.input_primitive)); spirv_builder_emit_exec_mode(&ctx.builder, entry_point, get_output_prim_type_mode(s->info.gs.output_primitive)); spirv_builder_emit_exec_mode_literal(&ctx.builder, entry_point, SpvExecutionModeInvocations, s->info.gs.invocations); spirv_builder_emit_exec_mode_literal(&ctx.builder, entry_point, SpvExecutionModeOutputVertices, MAX2(s->info.gs.vertices_out, 1)); break; case MESA_SHADER_KERNEL: case MESA_SHADER_COMPUTE: if (s->info.workgroup_size[0] || s->info.workgroup_size[1] || s->info.workgroup_size[2]) spirv_builder_emit_exec_mode_literal3(&ctx.builder, entry_point, SpvExecutionModeLocalSize, (uint32_t[3]){(uint32_t)s->info.workgroup_size[0], (uint32_t)s->info.workgroup_size[1], (uint32_t)s->info.workgroup_size[2]}); else { SpvId sizes[3]; uint32_t ids[] = {ZINK_WORKGROUP_SIZE_X, ZINK_WORKGROUP_SIZE_Y, ZINK_WORKGROUP_SIZE_Z}; const char *names[] = {"x", "y", "z"}; for (int i = 0; i < 3; i ++) { sizes[i] = spirv_builder_spec_const_uint(&ctx.builder, 32); spirv_builder_emit_specid(&ctx.builder, sizes[i], ids[i]); spirv_builder_emit_name(&ctx.builder, sizes[i], names[i]); } SpvId var_type = get_uvec_type(&ctx, 32, 3); // Even when using LocalSizeId this need to be initialized for nir_intrinsic_load_workgroup_size ctx.local_group_size_var = spirv_builder_spec_const_composite(&ctx.builder, var_type, sizes, 3); spirv_builder_emit_name(&ctx.builder, ctx.local_group_size_var, "gl_LocalGroupSizeARB"); /* WorkgroupSize is deprecated in SPIR-V 1.6 */ if (spirv_version >= SPIRV_VERSION(1, 6)) { spirv_builder_emit_exec_mode_id3(&ctx.builder, entry_point, SpvExecutionModeLocalSizeId, sizes); } else { spirv_builder_emit_builtin(&ctx.builder, ctx.local_group_size_var, SpvBuiltInWorkgroupSize); } } if (s->info.cs.has_variable_shared_mem) { ctx.shared_mem_size = spirv_builder_spec_const_uint(&ctx.builder, 32); spirv_builder_emit_specid(&ctx.builder, ctx.shared_mem_size, ZINK_VARIABLE_SHARED_MEM); spirv_builder_emit_name(&ctx.builder, ctx.shared_mem_size, "variable_shared_mem"); } if (s->info.derivative_group) { SpvCapability caps[] = { 0, SpvCapabilityComputeDerivativeGroupQuadsNV, SpvCapabilityComputeDerivativeGroupLinearNV }; SpvExecutionMode modes[] = { 0, SpvExecutionModeDerivativeGroupQuadsNV, SpvExecutionModeDerivativeGroupLinearNV }; spirv_builder_emit_extension(&ctx.builder, "SPV_NV_compute_shader_derivatives"); spirv_builder_emit_cap(&ctx.builder, caps[s->info.derivative_group]); spirv_builder_emit_exec_mode(&ctx.builder, entry_point, modes[s->info.derivative_group]); ctx.explicit_lod = false; } break; default: break; } if (BITSET_TEST_RANGE(s->info.system_values_read, SYSTEM_VALUE_SUBGROUP_SIZE, SYSTEM_VALUE_SUBGROUP_LT_MASK)) { spirv_builder_emit_cap(&ctx.builder, SpvCapabilitySubgroupBallotKHR); spirv_builder_emit_extension(&ctx.builder, "SPV_KHR_shader_ballot"); } if (s->info.has_transform_feedback_varyings && s->info.stage != MESA_SHADER_FRAGMENT) { spirv_builder_emit_cap(&ctx.builder, SpvCapabilityTransformFeedback); spirv_builder_emit_exec_mode(&ctx.builder, entry_point, SpvExecutionModeXfb); } if (s->info.stage == MESA_SHADER_FRAGMENT && s->info.fs.uses_discard) { ctx.discard_func = spirv_builder_new_id(&ctx.builder); spirv_builder_emit_name(&ctx.builder, ctx.discard_func, "discard"); spirv_builder_function(&ctx.builder, ctx.discard_func, type_void, SpvFunctionControlMaskNone, type_void_func); SpvId label = spirv_builder_new_id(&ctx.builder); spirv_builder_label(&ctx.builder, label); /* kill is deprecated in SPIR-V 1.6, use terminate instead */ if (spirv_version >= SPIRV_VERSION(1, 6)) spirv_builder_emit_terminate(&ctx.builder); else spirv_builder_emit_kill(&ctx.builder); spirv_builder_function_end(&ctx.builder); } spirv_builder_function(&ctx.builder, entry_point, type_void, SpvFunctionControlMaskNone, type_void_func); nir_function_impl *entry = nir_shader_get_entrypoint(s); nir_metadata_require(entry, nir_metadata_block_index); ctx.defs = rzalloc_array_size(ctx.mem_ctx, sizeof(SpvId), entry->ssa_alloc); ctx.def_types = ralloc_array_size(ctx.mem_ctx, sizeof(nir_alu_type), entry->ssa_alloc); if (!ctx.defs || !ctx.def_types) goto fail; if (sinfo->have_sparse) { spirv_builder_emit_cap(&ctx.builder, SpvCapabilitySparseResidency); /* this could be huge, so only alloc if needed since it's extremely unlikely to * ever be used by anything except cts */ ctx.resident_defs = rzalloc_array_size(ctx.mem_ctx, sizeof(SpvId), entry->ssa_alloc); if (!ctx.resident_defs) goto fail; } ctx.num_defs = entry->ssa_alloc; SpvId *block_ids = ralloc_array_size(ctx.mem_ctx, sizeof(SpvId), entry->num_blocks); if (!block_ids) goto fail; for (int i = 0; i < entry->num_blocks; ++i) block_ids[i] = spirv_builder_new_id(&ctx.builder); ctx.block_ids = block_ids; ctx.num_blocks = entry->num_blocks; /* emit a block only for the variable declarations */ start_block(&ctx, spirv_builder_new_id(&ctx.builder)); spirv_builder_begin_local_vars(&ctx.builder); nir_foreach_reg_decl(reg, entry) { if (nir_intrinsic_bit_size(reg) == 1) init_reg(&ctx, reg, nir_type_bool); } nir_foreach_variable_with_modes(var, s, nir_var_shader_temp) emit_shader_temp(&ctx, var); nir_foreach_function_temp_variable(var, entry) emit_temp(&ctx, var); emit_cf_list(&ctx, &entry->body); spirv_builder_return(&ctx.builder); // doesn't belong here, but whatevz spirv_builder_function_end(&ctx.builder); spirv_builder_emit_entry_point(&ctx.builder, exec_model, entry_point, "main", ctx.entry_ifaces, ctx.num_entry_ifaces); size_t num_words = spirv_builder_get_num_words(&ctx.builder); ret = ralloc(NULL, struct spirv_shader); if (!ret) goto fail; ret->words = ralloc_size(ret, sizeof(uint32_t) * num_words); if (!ret->words) goto fail; ret->num_words = spirv_builder_get_words(&ctx.builder, ret->words, num_words, spirv_version, &tcs_vertices_out_word); ret->tcs_vertices_out_word = tcs_vertices_out_word; assert(ret->num_words == num_words); ralloc_free(ctx.mem_ctx); return ret; fail: ralloc_free(ctx.mem_ctx); if (ret) spirv_shader_delete(ret); return NULL; } void spirv_shader_delete(struct spirv_shader *s) { ralloc_free(s); }