From 8b637cc107807cf7a7179fc5c015c7d08c8d4f8f Mon Sep 17 00:00:00 2001 From: Alistair Leslie-Hughes Date: Thu, 18 Jan 2024 11:50:14 +1100 Subject: [PATCH] Updated vkd3d to cd77b2a9be23b9a34d2d6a5cb566ac3873ff29b5. --- libs/vkd3d/include/private/vkd3d_common.h | 20 +- libs/vkd3d/include/private/vkd3d_debug.h | 23 + libs/vkd3d/libs/vkd3d-shader/d3d_asm.c | 2 + libs/vkd3d/libs/vkd3d-shader/dxbc.c | 5 + libs/vkd3d/libs/vkd3d-shader/dxil.c | 23 +- libs/vkd3d/libs/vkd3d-shader/hlsl.c | 1 - libs/vkd3d/libs/vkd3d-shader/hlsl.y | 59 +- libs/vkd3d/libs/vkd3d-shader/hlsl_codegen.c | 6 + libs/vkd3d/libs/vkd3d-shader/ir.c | 758 +++++++++++++++++- libs/vkd3d/libs/vkd3d-shader/spirv.c | 648 +++++---------- libs/vkd3d/libs/vkd3d-shader/tpf.c | 26 +- .../libs/vkd3d-shader/vkd3d_shader_private.h | 14 + libs/vkd3d/libs/vkd3d/resource.c | 4 +- libs/vkd3d/libs/vkd3d/vkd3d_private.h | 2 +- 14 files changed, 1092 insertions(+), 499 deletions(-) diff --git a/libs/vkd3d/include/private/vkd3d_common.h b/libs/vkd3d/include/private/vkd3d_common.h index 4c97fa06e32..63e21c22067 100644 --- a/libs/vkd3d/include/private/vkd3d_common.h +++ b/libs/vkd3d/include/private/vkd3d_common.h @@ -267,16 +267,28 @@ static inline int ascii_strcasecmp(const char *a, const char *b) return c_a - c_b; } +static inline uint64_t vkd3d_atomic_add_fetch_u64(uint64_t volatile *x, uint64_t val) +{ +#if HAVE_SYNC_ADD_AND_FETCH + return __sync_add_and_fetch(x, val); +#elif defined(_WIN32) + return InterlockedAdd64((LONG64 *)x, val); +#else +# error "vkd3d_atomic_add_fetch_u64() not implemented for this platform" +#endif +} + +static inline uint64_t vkd3d_atomic_increment_u64(uint64_t volatile *x) +{ + return vkd3d_atomic_add_fetch_u64(x, 1); +} + #ifndef _WIN32 # if HAVE_SYNC_ADD_AND_FETCH static inline LONG InterlockedIncrement(LONG volatile *x) { return __sync_add_and_fetch(x, 1); } -static inline LONG64 InterlockedIncrement64(LONG64 volatile *x) -{ - return __sync_add_and_fetch(x, 1); -} # else # error "InterlockedIncrement() not implemented for this platform" # endif /* HAVE_SYNC_ADD_AND_FETCH */ diff --git a/libs/vkd3d/include/private/vkd3d_debug.h b/libs/vkd3d/include/private/vkd3d_debug.h index 6708cad344f..663fc311adf 100644 --- a/libs/vkd3d/include/private/vkd3d_debug.h +++ b/libs/vkd3d/include/private/vkd3d_debug.h @@ -104,6 +104,29 @@ static inline const char *debugstr_guid(const GUID *guid) guid->Data4[5], guid->Data4[6], guid->Data4[7]); } +static inline const char *debugstr_hresult(HRESULT hr) +{ + switch (hr) + { +#define TO_STR(u) case u: return #u; + TO_STR(S_OK) + TO_STR(S_FALSE) + TO_STR(E_NOTIMPL) + TO_STR(E_NOINTERFACE) + TO_STR(E_POINTER) + TO_STR(E_ABORT) + TO_STR(E_FAIL) + TO_STR(E_OUTOFMEMORY) + TO_STR(E_INVALIDARG) + TO_STR(DXGI_ERROR_NOT_FOUND) + TO_STR(DXGI_ERROR_MORE_DATA) + TO_STR(DXGI_ERROR_UNSUPPORTED) +#undef TO_STR + default: + return vkd3d_dbg_sprintf("%#x", (int)hr); + } +} + unsigned int vkd3d_env_var_as_uint(const char *name, unsigned int default_value); struct vkd3d_debug_option diff --git a/libs/vkd3d/libs/vkd3d-shader/d3d_asm.c b/libs/vkd3d/libs/vkd3d-shader/d3d_asm.c index 4829956cecf..5685fe4d4a4 100644 --- a/libs/vkd3d/libs/vkd3d-shader/d3d_asm.c +++ b/libs/vkd3d/libs/vkd3d-shader/d3d_asm.c @@ -44,6 +44,7 @@ static const char * const shader_opcode_names[] = [VKD3DSIH_BEM ] = "bem", [VKD3DSIH_BFI ] = "bfi", [VKD3DSIH_BFREV ] = "bfrev", + [VKD3DSIH_BRANCH ] = "branch", [VKD3DSIH_BREAK ] = "break", [VKD3DSIH_BREAKC ] = "breakc", [VKD3DSIH_BREAKP ] = "breakp", @@ -278,6 +279,7 @@ static const char * const shader_opcode_names[] = [VKD3DSIH_SUB ] = "sub", [VKD3DSIH_SWAPC ] = "swapc", [VKD3DSIH_SWITCH ] = "switch", + [VKD3DSIH_SWITCH_MONOLITHIC ] = "switch", [VKD3DSIH_SYNC ] = "sync", [VKD3DSIH_TEX ] = "texld", [VKD3DSIH_TEXBEM ] = "texbem", diff --git a/libs/vkd3d/libs/vkd3d-shader/dxbc.c b/libs/vkd3d/libs/vkd3d-shader/dxbc.c index 37ebc73c099..edb65d2e92f 100644 --- a/libs/vkd3d/libs/vkd3d-shader/dxbc.c +++ b/libs/vkd3d/libs/vkd3d-shader/dxbc.c @@ -552,9 +552,14 @@ static int shdr_handler(const struct vkd3d_shader_dxbc_section_desc *section, void free_shader_desc(struct vkd3d_shader_desc *desc) { + size_t i; + shader_signature_cleanup(&desc->input_signature); shader_signature_cleanup(&desc->output_signature); shader_signature_cleanup(&desc->patch_constant_signature); + for (i = 0; i < desc->block_name_count; ++i) + vkd3d_free((void *)desc->block_names[i]); + vkd3d_free(desc->block_names); } int shader_extract_from_dxbc(const struct vkd3d_shader_code *dxbc, diff --git a/libs/vkd3d/libs/vkd3d-shader/dxil.c b/libs/vkd3d/libs/vkd3d-shader/dxil.c index b39ec204bff..78c1a052539 100644 --- a/libs/vkd3d/libs/vkd3d-shader/dxil.c +++ b/libs/vkd3d/libs/vkd3d-shader/dxil.c @@ -2691,8 +2691,8 @@ static void sm6_parser_declare_icb(struct sm6_parser *sm6, const struct sm6_type } static void sm6_parser_declare_indexable_temp(struct sm6_parser *sm6, const struct sm6_type *elem_type, - unsigned int count, unsigned int alignment, unsigned int init, struct vkd3d_shader_instruction *ins, - struct sm6_value *dst) + unsigned int count, unsigned int alignment, bool has_function_scope, unsigned int init, + struct vkd3d_shader_instruction *ins, struct sm6_value *dst) { enum vkd3d_data_type data_type = vkd3d_data_type_from_sm6_type(elem_type); @@ -2705,6 +2705,7 @@ static void sm6_parser_declare_indexable_temp(struct sm6_parser *sm6, const stru ins->declaration.indexable_temp.alignment = alignment; ins->declaration.indexable_temp.data_type = data_type; ins->declaration.indexable_temp.component_count = 1; + ins->declaration.indexable_temp.has_function_scope = has_function_scope; /* The initialiser value index will be resolved later so forward references can be handled. */ ins->declaration.indexable_temp.initialiser = (void *)(uintptr_t)init; @@ -2832,7 +2833,7 @@ static bool sm6_parser_declare_global(struct sm6_parser *sm6, const struct dxil_ if (is_constant) sm6_parser_declare_icb(sm6, scalar_type, count, alignment, init, dst); else - sm6_parser_declare_indexable_temp(sm6, scalar_type, count, alignment, init, NULL, dst); + sm6_parser_declare_indexable_temp(sm6, scalar_type, count, alignment, false, init, NULL, dst); } else if (address_space == ADDRESS_SPACE_GROUPSHARED) { @@ -3103,7 +3104,7 @@ static void sm6_parser_emit_alloca(struct sm6_parser *sm6, const struct dxil_rec if (packed_operands) WARN("Ignoring flags %#"PRIx64".\n", packed_operands); - sm6_parser_declare_indexable_temp(sm6, elem_type, type[0]->u.array.count, alignment, 0, ins, dst); + sm6_parser_declare_indexable_temp(sm6, elem_type, type[0]->u.array.count, alignment, true, 0, ins, dst); } static enum vkd3d_shader_opcode map_binary_op(uint64_t code, const struct sm6_type *type_a, @@ -4868,6 +4869,18 @@ static enum vkd3d_result sm6_parser_module_init(struct sm6_parser *sm6, const st return VKD3D_OK; } +static void sm6_parser_emit_label(struct sm6_parser *sm6, unsigned int label_id) +{ + struct vkd3d_shader_src_param *src_param; + struct vkd3d_shader_instruction *ins; + + ins = sm6_parser_add_instruction(sm6, VKD3DSIH_LABEL); + + if (!(src_param = instruction_src_params_alloc(ins, 1, sm6))) + return; + vsir_src_param_init_label(src_param, label_id); +} + static bool sm6_parser_allocate_named_metadata(struct sm6_parser *sm6) { struct dxil_block *block; @@ -6216,6 +6229,7 @@ static enum vkd3d_result sm6_parser_init(struct sm6_parser *sm6, const uint32_t } sm6->p.shader_desc.ssa_count = sm6->ssa_next_id; + sm6->p.shader_desc.block_count = 1; if (!(fn = sm6_parser_get_function(sm6, sm6->entry_point))) { @@ -6226,6 +6240,7 @@ static enum vkd3d_result sm6_parser_init(struct sm6_parser *sm6, const uint32_t } assert(sm6->function_count == 1); + sm6_parser_emit_label(sm6, 1); if (!sm6_block_emit_instructions(fn->blocks[0], sm6)) { vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_OUT_OF_MEMORY, diff --git a/libs/vkd3d/libs/vkd3d-shader/hlsl.c b/libs/vkd3d/libs/vkd3d-shader/hlsl.c index 1e247445119..6a5a6d0e3c1 100644 --- a/libs/vkd3d/libs/vkd3d-shader/hlsl.c +++ b/libs/vkd3d/libs/vkd3d-shader/hlsl.c @@ -3363,7 +3363,6 @@ static void declare_predefined_types(struct hlsl_ctx *ctx) effect_types[] = { {"dword", HLSL_CLASS_SCALAR, HLSL_TYPE_UINT, 1, 1}, - {"float", HLSL_CLASS_SCALAR, HLSL_TYPE_FLOAT, 1, 1}, {"vector", HLSL_CLASS_VECTOR, HLSL_TYPE_FLOAT, 4, 1}, {"matrix", HLSL_CLASS_MATRIX, HLSL_TYPE_FLOAT, 4, 4}, {"fxgroup", HLSL_CLASS_OBJECT, HLSL_TYPE_EFFECT_GROUP, 1, 1}, diff --git a/libs/vkd3d/libs/vkd3d-shader/hlsl.y b/libs/vkd3d/libs/vkd3d-shader/hlsl.y index b11cbde26f1..e30b3dc5f55 100644 --- a/libs/vkd3d/libs/vkd3d-shader/hlsl.y +++ b/libs/vkd3d/libs/vkd3d-shader/hlsl.y @@ -4085,6 +4085,49 @@ static struct hlsl_block *add_constructor(struct hlsl_ctx *ctx, struct hlsl_type return params->instrs; } +static bool add_ternary(struct hlsl_ctx *ctx, struct hlsl_block *block, + struct hlsl_ir_node *cond, struct hlsl_ir_node *first, struct hlsl_ir_node *second) +{ + struct hlsl_ir_node *args[HLSL_MAX_OPERANDS] = {0}; + struct hlsl_type *common_type; + + if (first->data_type->class <= HLSL_CLASS_LAST_NUMERIC + && second->data_type->class <= HLSL_CLASS_LAST_NUMERIC) + { + if (!(common_type = get_common_numeric_type(ctx, first, second, &first->loc))) + return false; + + if (!(first = add_implicit_conversion(ctx, block, first, common_type, &first->loc))) + return false; + + if (!(second = add_implicit_conversion(ctx, block, second, common_type, &second->loc))) + return false; + } + else + { + struct vkd3d_string_buffer *first_string, *second_string; + + if (!hlsl_types_are_equal(first->data_type, second->data_type)) + { + first_string = hlsl_type_to_string(ctx, first->data_type); + second_string = hlsl_type_to_string(ctx, second->data_type); + if (first_string && second_string) + hlsl_error(ctx, &first->loc, VKD3D_SHADER_ERROR_HLSL_INVALID_TYPE, + "Ternary argument types '%s' and '%s' do not match.", + first_string->buffer, second_string->buffer); + hlsl_release_string_buffer(ctx, first_string); + hlsl_release_string_buffer(ctx, second_string); + } + + common_type = first->data_type; + } + + args[0] = cond; + args[1] = first; + args[2] = second; + return add_expr(ctx, block, HLSL_OP3_TERNARY, args, common_type, &first->loc); +} + static unsigned int hlsl_offset_dim_count(enum hlsl_sampler_dim dim) { switch (dim) @@ -7202,27 +7245,13 @@ conditional_expr: struct hlsl_ir_node *cond = node_from_block($1); struct hlsl_ir_node *first = node_from_block($3); struct hlsl_ir_node *second = node_from_block($5); - struct hlsl_ir_node *args[HLSL_MAX_OPERANDS] = { 0 }; - struct hlsl_type *common_type; hlsl_block_add_block($1, $3); hlsl_block_add_block($1, $5); destroy_block($3); destroy_block($5); - if (!(common_type = get_common_numeric_type(ctx, first, second, &@3))) - YYABORT; - - if (!(first = add_implicit_conversion(ctx, $1, first, common_type, &@3))) - YYABORT; - - if (!(second = add_implicit_conversion(ctx, $1, second, common_type, &@5))) - YYABORT; - - args[0] = cond; - args[1] = first; - args[2] = second; - if (!add_expr(ctx, $1, HLSL_OP3_TERNARY, args, common_type, &@1)) + if (!add_ternary(ctx, $1, cond, first, second)) YYABORT; $$ = $1; } diff --git a/libs/vkd3d/libs/vkd3d-shader/hlsl_codegen.c b/libs/vkd3d/libs/vkd3d-shader/hlsl_codegen.c index d37bef15cce..1fe141a346a 100644 --- a/libs/vkd3d/libs/vkd3d-shader/hlsl_codegen.c +++ b/libs/vkd3d/libs/vkd3d-shader/hlsl_codegen.c @@ -2890,6 +2890,12 @@ static bool lower_ternary(struct hlsl_ctx *ctx, struct hlsl_ir_node *instr, stru first = expr->operands[1].node; second = expr->operands[2].node; + if (cond->data_type->class > HLSL_CLASS_VECTOR || instr->data_type->class > HLSL_CLASS_VECTOR) + { + hlsl_fixme(ctx, &instr->loc, "Lower ternary of type other than scalar or vector.\n"); + return false; + } + if (ctx->profile->major_version < 4 && ctx->profile->type == VKD3D_SHADER_TYPE_PIXEL) { struct hlsl_ir_node *abs, *neg; diff --git a/libs/vkd3d/libs/vkd3d-shader/ir.c b/libs/vkd3d/libs/vkd3d-shader/ir.c index 28c7d158452..d6978171beb 100644 --- a/libs/vkd3d/libs/vkd3d-shader/ir.c +++ b/libs/vkd3d/libs/vkd3d-shader/ir.c @@ -312,6 +312,21 @@ void vsir_register_init(struct vkd3d_shader_register *reg, enum vkd3d_shader_reg reg->alignment = 0; } +static void vsir_src_param_init(struct vkd3d_shader_src_param *param, enum vkd3d_shader_register_type reg_type, + enum vkd3d_data_type data_type, unsigned int idx_count) +{ + vsir_register_init(¶m->reg, reg_type, data_type, idx_count); + param->swizzle = 0; + param->modifiers = VKD3DSPSM_NONE; +} + +void vsir_src_param_init_label(struct vkd3d_shader_src_param *param, unsigned int label_id) +{ + vsir_src_param_init(param, VKD3DSPR_LABEL, VKD3D_DATA_UINT, 1); + param->reg.dimension = VSIR_DIMENSION_NONE; + param->reg.idx[0].offset = label_id; +} + void vsir_instruction_init(struct vkd3d_shader_instruction *ins, const struct vkd3d_shader_location *location, enum vkd3d_shader_opcode handler_idx) { @@ -320,6 +335,23 @@ void vsir_instruction_init(struct vkd3d_shader_instruction *ins, const struct vk ins->handler_idx = handler_idx; } +static bool vsir_instruction_init_label(struct vkd3d_shader_instruction *ins, const struct vkd3d_shader_location *location, + unsigned int label_id, void *parser) +{ + struct vkd3d_shader_src_param *src_param; + + if (!(src_param = shader_parser_get_src_params(parser, 1))) + return false; + + vsir_src_param_init_label(src_param, label_id); + + vsir_instruction_init(ins, location, VKD3DSIH_LABEL); + ins->src = src_param; + ins->src_count = 1; + + return true; +} + static enum vkd3d_result instruction_array_flatten_hull_shader_phases(struct vkd3d_shader_instruction_array *src_instructions) { struct hull_flattener flattener = {*src_instructions}; @@ -571,7 +603,7 @@ static bool io_normaliser_is_in_control_point_phase(const struct io_normaliser * static unsigned int shader_signature_find_element_for_reg(const struct shader_signature *signature, unsigned int reg_idx, unsigned int write_mask) { - unsigned int i; + unsigned int i, base_write_mask; for (i = 0; i < signature->element_count; ++i) { @@ -583,7 +615,14 @@ static unsigned int shader_signature_find_element_for_reg(const struct shader_si } } - /* Validated in the TPF reader. */ + /* Validated in the TPF reader, but failure in signature_element_range_expand_mask() + * can land us here on an unmatched vector mask. */ + FIXME("Failed to find signature element for register index %u, mask %#x; using scalar mask.\n", + reg_idx, write_mask); + base_write_mask = 1u << vsir_write_mask_get_component_idx(write_mask); + if (base_write_mask != write_mask) + return shader_signature_find_element_for_reg(signature, reg_idx, base_write_mask); + vkd3d_unreachable(); } @@ -750,6 +789,51 @@ static int signature_element_index_compare(const void *a, const void *b) return vkd3d_u32_compare(e->sort_index, f->sort_index); } +static unsigned int signature_element_range_expand_mask(struct signature_element *e, unsigned int register_count, + uint8_t range_map[][VKD3D_VEC4_SIZE]) +{ + unsigned int i, j, component_idx, component_count, merged_write_mask = e->mask; + + /* dcl_indexrange instructions can declare a subset of the full mask, and the masks of + * the elements within the range may differ. TPF's handling of arrayed inputs with + * dcl_indexrange is really just a hack. Here we create a mask which covers all element + * masks, and check for collisions with other ranges. */ + + for (i = 1; i < register_count; ++i) + merged_write_mask |= e[i].mask; + + if (merged_write_mask == e->mask) + return merged_write_mask; + + /* Reaching this point is very rare to begin with, and collisions are even rarer or + * impossible. If the latter shows up, the fallback in shader_signature_find_element_for_reg() + * may be sufficient. */ + + component_idx = vsir_write_mask_get_component_idx(e->mask); + component_count = vsir_write_mask_component_count(e->mask); + + for (i = e->register_index; i < e->register_index + register_count; ++i) + { + for (j = 0; j < component_idx; ++j) + if (range_map[i][j]) + break; + for (j = component_idx + component_count; j < VKD3D_VEC4_SIZE; ++j) + if (range_map[i][j]) + break; + } + + if (i == register_count) + { + WARN("Expanding mask %#x to %#x for %s, base reg %u, count %u.\n", e->mask, merged_write_mask, + e->semantic_name, e->register_index, register_count); + return merged_write_mask; + } + + WARN("Cannot expand mask %#x to %#x for %s, base reg %u, count %u.\n", e->mask, merged_write_mask, + e->semantic_name, e->register_index, register_count); + return e->mask; +} + static bool shader_signature_merge(struct shader_signature *s, uint8_t range_map[][VKD3D_VEC4_SIZE], bool is_patch_constant) { @@ -820,6 +904,7 @@ static bool shader_signature_merge(struct shader_signature *s, uint8_t range_map { TRACE("Merging %s, base reg %u, count %u.\n", e->semantic_name, e->register_index, register_count); e->register_count = register_count; + e->mask = signature_element_range_expand_mask(e, register_count, range_map); } } element_count = new_count; @@ -844,6 +929,13 @@ static unsigned int shader_register_normalise_arrayed_addressing(struct vkd3d_sh reg->idx[id_idx + 1].rel_addr = NULL; reg->idx[id_idx + 1].offset = reg->idx[id_idx].offset; reg->idx[id_idx].offset -= register_index; + if (id_idx) + { + /* idx[id_idx] now contains the array index, which must be moved below the control point id. */ + struct vkd3d_shader_register_index tmp = reg->idx[id_idx]; + reg->idx[id_idx] = reg->idx[id_idx - 1]; + reg->idx[id_idx - 1] = tmp; + } ++id_idx; } /* Otherwise we have no address for the arrayed register, so insert one. This happens e.g. where @@ -1476,6 +1568,665 @@ static enum vkd3d_result normalise_combined_samplers(struct vkd3d_shader_parser return VKD3D_OK; } +struct cf_flattener_if_info +{ + struct vkd3d_shader_src_param *false_param; + unsigned int id; + uint32_t merge_block_id; + unsigned int else_block_id; +}; + +struct cf_flattener_loop_info +{ + unsigned int header_block_id; + unsigned int continue_block_id; + uint32_t merge_block_id; +}; + +struct cf_flattener_switch_case +{ + unsigned int value; + unsigned int block_id; +}; + +struct cf_flattener_switch_info +{ + size_t ins_location; + const struct vkd3d_shader_src_param *condition; + unsigned int id; + unsigned int merge_block_id; + unsigned int default_block_id; + struct cf_flattener_switch_case *cases; + size_t cases_size; + unsigned int cases_count; +}; + +struct cf_flattener_info +{ + union + { + struct cf_flattener_if_info if_; + struct cf_flattener_loop_info loop; + struct cf_flattener_switch_info switch_; + } u; + + enum + { + VKD3D_BLOCK_IF, + VKD3D_BLOCK_LOOP, + VKD3D_BLOCK_SWITCH, + } current_block; + bool inside_block; +}; + +struct cf_flattener +{ + struct vkd3d_shader_parser *parser; + + struct vkd3d_shader_location location; + bool allocation_failed; + + struct vkd3d_shader_instruction *instructions; + size_t instruction_capacity; + size_t instruction_count; + + unsigned int block_id; + const char **block_names; + size_t block_name_capacity; + size_t block_name_count; + + unsigned int branch_id; + unsigned int loop_id; + unsigned int switch_id; + + unsigned int control_flow_depth; + struct cf_flattener_info *control_flow_info; + size_t control_flow_info_size; +}; + +static struct vkd3d_shader_instruction *cf_flattener_require_space(struct cf_flattener *flattener, size_t count) +{ + if (!vkd3d_array_reserve((void **)&flattener->instructions, &flattener->instruction_capacity, + flattener->instruction_count + count, sizeof(*flattener->instructions))) + { + ERR("Failed to allocate instructions.\n"); + flattener->allocation_failed = true; + return NULL; + } + return &flattener->instructions[flattener->instruction_count]; +} + +static bool cf_flattener_copy_instruction(struct cf_flattener *flattener, + const struct vkd3d_shader_instruction *instruction) +{ + struct vkd3d_shader_instruction *dst_ins; + + if (instruction->handler_idx == VKD3DSIH_NOP) + return true; + + if (!(dst_ins = cf_flattener_require_space(flattener, 1))) + return false; + + *dst_ins = *instruction; + ++flattener->instruction_count; + return true; +} + +static unsigned int cf_flattener_alloc_block_id(struct cf_flattener *flattener) +{ + return ++flattener->block_id; +} + +static struct vkd3d_shader_src_param *instruction_src_params_alloc(struct vkd3d_shader_instruction *ins, + unsigned int count, struct cf_flattener *flattener) +{ + struct vkd3d_shader_src_param *params = shader_parser_get_src_params(flattener->parser, count); + if (!params) + { + flattener->allocation_failed = true; + return NULL; + } + ins->src = params; + ins->src_count = count; + return params; +} + +static void cf_flattener_emit_label(struct cf_flattener *flattener, unsigned int label_id) +{ + struct vkd3d_shader_instruction *ins; + + if (!(ins = cf_flattener_require_space(flattener, 1))) + return; + if (vsir_instruction_init_label(ins, &flattener->location, label_id, flattener->parser)) + ++flattener->instruction_count; + else + flattener->allocation_failed = true; +} + +/* For conditional branches, this returns the false target branch parameter. */ +static struct vkd3d_shader_src_param *cf_flattener_emit_branch(struct cf_flattener *flattener, + unsigned int merge_block_id, unsigned int continue_block_id, + const struct vkd3d_shader_src_param *condition, unsigned int true_id, unsigned int false_id, + unsigned int flags) +{ + struct vkd3d_shader_src_param *src_params, *false_branch_param; + struct vkd3d_shader_instruction *ins; + + if (!(ins = cf_flattener_require_space(flattener, 1))) + return NULL; + vsir_instruction_init(ins, &flattener->location, VKD3DSIH_BRANCH); + + if (condition) + { + if (!(src_params = instruction_src_params_alloc(ins, 4 + !!continue_block_id, flattener))) + return NULL; + src_params[0] = *condition; + if (flags == VKD3D_SHADER_CONDITIONAL_OP_Z) + { + vsir_src_param_init_label(&src_params[1], false_id); + vsir_src_param_init_label(&src_params[2], true_id); + false_branch_param = &src_params[1]; + } + else + { + vsir_src_param_init_label(&src_params[1], true_id); + vsir_src_param_init_label(&src_params[2], false_id); + false_branch_param = &src_params[2]; + } + vsir_src_param_init_label(&src_params[3], merge_block_id); + if (continue_block_id) + vsir_src_param_init_label(&src_params[4], continue_block_id); + } + else + { + if (!(src_params = instruction_src_params_alloc(ins, merge_block_id ? 3 : 1, flattener))) + return NULL; + vsir_src_param_init_label(&src_params[0], true_id); + if (merge_block_id) + { + /* An unconditional branch may only have merge information for a loop, which + * must have both a merge block and continue block. */ + vsir_src_param_init_label(&src_params[1], merge_block_id); + vsir_src_param_init_label(&src_params[2], continue_block_id); + } + false_branch_param = NULL; + } + + ++flattener->instruction_count; + + return false_branch_param; +} + +static void cf_flattener_emit_conditional_branch_and_merge(struct cf_flattener *flattener, + const struct vkd3d_shader_src_param *condition, unsigned int true_id, unsigned int flags) +{ + unsigned int merge_block_id; + + merge_block_id = cf_flattener_alloc_block_id(flattener); + cf_flattener_emit_branch(flattener, merge_block_id, 0, condition, true_id, merge_block_id, flags); + cf_flattener_emit_label(flattener, merge_block_id); +} + +static void cf_flattener_emit_unconditional_branch(struct cf_flattener *flattener, unsigned int target_block_id) +{ + cf_flattener_emit_branch(flattener, 0, 0, NULL, target_block_id, 0, 0); +} + +static struct cf_flattener_info *cf_flattener_push_control_flow_level(struct cf_flattener *flattener) +{ + if (!vkd3d_array_reserve((void **)&flattener->control_flow_info, &flattener->control_flow_info_size, + flattener->control_flow_depth + 1, sizeof(*flattener->control_flow_info))) + { + ERR("Failed to allocate control flow info structure.\n"); + flattener->allocation_failed = true; + return NULL; + } + + return &flattener->control_flow_info[flattener->control_flow_depth++]; +} + +static void cf_flattener_pop_control_flow_level(struct cf_flattener *flattener) +{ + struct cf_flattener_info *cf_info; + + cf_info = &flattener->control_flow_info[--flattener->control_flow_depth]; + memset(cf_info, 0, sizeof(*cf_info)); +} + +static struct cf_flattener_info *cf_flattener_find_innermost_loop(struct cf_flattener *flattener) +{ + int depth; + + for (depth = flattener->control_flow_depth - 1; depth >= 0; --depth) + { + if (flattener->control_flow_info[depth].current_block == VKD3D_BLOCK_LOOP) + return &flattener->control_flow_info[depth]; + } + + return NULL; +} + +static struct cf_flattener_info *cf_flattener_find_innermost_breakable_cf_construct(struct cf_flattener *flattener) +{ + int depth; + + for (depth = flattener->control_flow_depth - 1; depth >= 0; --depth) + { + if (flattener->control_flow_info[depth].current_block == VKD3D_BLOCK_LOOP + || flattener->control_flow_info[depth].current_block == VKD3D_BLOCK_SWITCH) + return &flattener->control_flow_info[depth]; + } + + return NULL; +} + +static void VKD3D_PRINTF_FUNC(3, 4) cf_flattener_create_block_name(struct cf_flattener *flattener, + unsigned int block_id, const char *fmt, ...) +{ + struct vkd3d_string_buffer buffer; + size_t block_name_count; + va_list args; + + --block_id; + + block_name_count = max(flattener->block_name_count, block_id + 1); + if (!vkd3d_array_reserve((void **)&flattener->block_names, &flattener->block_name_capacity, + block_name_count, sizeof(*flattener->block_names))) + return; + memset(&flattener->block_names[flattener->block_name_count], 0, + (block_name_count - flattener->block_name_count) * sizeof(*flattener->block_names)); + flattener->block_name_count = block_name_count; + + vkd3d_string_buffer_init(&buffer); + va_start(args, fmt); + vkd3d_string_buffer_vprintf(&buffer, fmt, args); + va_end(args); + + flattener->block_names[block_id] = buffer.buffer; +} + +static bool vsir_instruction_is_dcl(const struct vkd3d_shader_instruction *instruction) +{ + enum vkd3d_shader_opcode handler_idx = instruction->handler_idx; + return (VKD3DSIH_DCL <= handler_idx && handler_idx <= VKD3DSIH_DCL_VERTICES_OUT) + || handler_idx == VKD3DSIH_HS_DECLS; +} + +static enum vkd3d_result cf_flattener_iterate_instruction_array(struct cf_flattener *flattener) +{ + bool main_block_open, is_hull_shader, after_declarations_section; + struct vkd3d_shader_parser *parser = flattener->parser; + struct vkd3d_shader_instruction_array *instructions; + struct vkd3d_shader_instruction *dst_ins; + size_t i; + + instructions = &parser->instructions; + is_hull_shader = parser->shader_version.type == VKD3D_SHADER_TYPE_HULL; + main_block_open = !is_hull_shader; + after_declarations_section = is_hull_shader; + + if (!cf_flattener_require_space(flattener, instructions->count + 1)) + return VKD3D_ERROR_OUT_OF_MEMORY; + + for (i = 0; i < instructions->count; ++i) + { + unsigned int loop_header_block_id, loop_body_block_id, continue_block_id, merge_block_id, true_block_id; + const struct vkd3d_shader_instruction *instruction = &instructions->elements[i]; + const struct vkd3d_shader_src_param *src = instruction->src; + struct cf_flattener_info *cf_info; + + flattener->location = instruction->location; + + /* Declarations should occur before the first code block, which in hull shaders is marked by the first + * phase instruction, and in all other shader types begins with the first label instruction. */ + if (!after_declarations_section && !vsir_instruction_is_dcl(instruction) + && instruction->handler_idx != VKD3DSIH_NOP) + { + after_declarations_section = true; + cf_flattener_emit_label(flattener, cf_flattener_alloc_block_id(flattener)); + } + + cf_info = flattener->control_flow_depth + ? &flattener->control_flow_info[flattener->control_flow_depth - 1] : NULL; + + switch (instruction->handler_idx) + { + case VKD3DSIH_HS_CONTROL_POINT_PHASE: + case VKD3DSIH_HS_FORK_PHASE: + case VKD3DSIH_HS_JOIN_PHASE: + if (!cf_flattener_copy_instruction(flattener, instruction)) + return VKD3D_ERROR_OUT_OF_MEMORY; + if (instruction->handler_idx != VKD3DSIH_HS_CONTROL_POINT_PHASE || !instruction->flags) + after_declarations_section = false; + break; + + case VKD3DSIH_LABEL: + vkd3d_shader_parser_error(parser, VKD3D_SHADER_ERROR_VSIR_NOT_IMPLEMENTED, + "Aborting due to not yet implemented feature: Label instruction."); + return VKD3D_ERROR_NOT_IMPLEMENTED; + + case VKD3DSIH_IF: + if (!(cf_info = cf_flattener_push_control_flow_level(flattener))) + return VKD3D_ERROR_OUT_OF_MEMORY; + + true_block_id = cf_flattener_alloc_block_id(flattener); + merge_block_id = cf_flattener_alloc_block_id(flattener); + cf_info->u.if_.false_param = cf_flattener_emit_branch(flattener, merge_block_id, 0, + src, true_block_id, merge_block_id, instruction->flags); + if (!cf_info->u.if_.false_param) + return VKD3D_ERROR_OUT_OF_MEMORY; + + cf_flattener_emit_label(flattener, true_block_id); + + cf_info->u.if_.id = flattener->branch_id; + cf_info->u.if_.merge_block_id = merge_block_id; + cf_info->u.if_.else_block_id = 0; + cf_info->inside_block = true; + cf_info->current_block = VKD3D_BLOCK_IF; + + cf_flattener_create_block_name(flattener, merge_block_id, "branch%u_merge", flattener->branch_id); + cf_flattener_create_block_name(flattener, true_block_id, "branch%u_true", flattener->branch_id); + ++flattener->branch_id; + break; + + case VKD3DSIH_ELSE: + if (cf_info->inside_block) + cf_flattener_emit_unconditional_branch(flattener, cf_info->u.if_.merge_block_id); + + cf_info->u.if_.else_block_id = cf_flattener_alloc_block_id(flattener); + cf_info->u.if_.false_param->reg.idx[0].offset = cf_info->u.if_.else_block_id; + + cf_flattener_create_block_name(flattener, + cf_info->u.if_.else_block_id, "branch%u_false", cf_info->u.if_.id); + cf_flattener_emit_label(flattener, cf_info->u.if_.else_block_id); + + cf_info->inside_block = true; + break; + + case VKD3DSIH_ENDIF: + if (cf_info->inside_block) + cf_flattener_emit_unconditional_branch(flattener, cf_info->u.if_.merge_block_id); + + cf_flattener_emit_label(flattener, cf_info->u.if_.merge_block_id); + + cf_flattener_pop_control_flow_level(flattener); + break; + + case VKD3DSIH_LOOP: + if (!(cf_info = cf_flattener_push_control_flow_level(flattener))) + return VKD3D_ERROR_OUT_OF_MEMORY; + + loop_header_block_id = cf_flattener_alloc_block_id(flattener); + loop_body_block_id = cf_flattener_alloc_block_id(flattener); + continue_block_id = cf_flattener_alloc_block_id(flattener); + merge_block_id = cf_flattener_alloc_block_id(flattener); + + cf_flattener_emit_unconditional_branch(flattener, loop_header_block_id); + cf_flattener_emit_label(flattener, loop_header_block_id); + cf_flattener_emit_branch(flattener, merge_block_id, continue_block_id, + NULL, loop_body_block_id, 0, 0); + + cf_flattener_emit_label(flattener, loop_body_block_id); + + cf_info->u.loop.header_block_id = loop_header_block_id; + cf_info->u.loop.continue_block_id = continue_block_id; + cf_info->u.loop.merge_block_id = merge_block_id; + cf_info->current_block = VKD3D_BLOCK_LOOP; + cf_info->inside_block = true; + + cf_flattener_create_block_name(flattener, loop_header_block_id, "loop%u_header", flattener->loop_id); + cf_flattener_create_block_name(flattener, loop_body_block_id, "loop%u_body", flattener->loop_id); + cf_flattener_create_block_name(flattener, continue_block_id, "loop%u_continue", flattener->loop_id); + cf_flattener_create_block_name(flattener, merge_block_id, "loop%u_merge", flattener->loop_id); + ++flattener->loop_id; + break; + + case VKD3DSIH_ENDLOOP: + if (cf_info->inside_block) + cf_flattener_emit_unconditional_branch(flattener, cf_info->u.loop.continue_block_id); + + cf_flattener_emit_label(flattener, cf_info->u.loop.continue_block_id); + cf_flattener_emit_unconditional_branch(flattener, cf_info->u.loop.header_block_id); + cf_flattener_emit_label(flattener, cf_info->u.loop.merge_block_id); + + cf_flattener_pop_control_flow_level(flattener); + break; + + case VKD3DSIH_SWITCH: + if (!(cf_info = cf_flattener_push_control_flow_level(flattener))) + return VKD3D_ERROR_OUT_OF_MEMORY; + + merge_block_id = cf_flattener_alloc_block_id(flattener); + + cf_info->u.switch_.ins_location = flattener->instruction_count; + cf_info->u.switch_.condition = src; + + if (!(dst_ins = cf_flattener_require_space(flattener, 1))) + return VKD3D_ERROR_OUT_OF_MEMORY; + vsir_instruction_init(dst_ins, &instruction->location, VKD3DSIH_SWITCH_MONOLITHIC); + ++flattener->instruction_count; + + cf_info->u.switch_.id = flattener->switch_id; + cf_info->u.switch_.merge_block_id = merge_block_id; + cf_info->u.switch_.cases = NULL; + cf_info->u.switch_.cases_size = 0; + cf_info->u.switch_.cases_count = 0; + cf_info->u.switch_.default_block_id = 0; + cf_info->inside_block = false; + cf_info->current_block = VKD3D_BLOCK_SWITCH; + + cf_flattener_create_block_name(flattener, merge_block_id, "switch%u_merge", flattener->switch_id); + ++flattener->switch_id; + + if (!vkd3d_array_reserve((void **)&cf_info->u.switch_.cases, &cf_info->u.switch_.cases_size, + 10, sizeof(*cf_info->u.switch_.cases))) + return VKD3D_ERROR_OUT_OF_MEMORY; + + break; + + case VKD3DSIH_ENDSWITCH: + { + struct vkd3d_shader_src_param *src_params; + unsigned int j; + + if (!cf_info->u.switch_.default_block_id) + cf_info->u.switch_.default_block_id = cf_info->u.switch_.merge_block_id; + + cf_flattener_emit_label(flattener, cf_info->u.switch_.merge_block_id); + + /* The SWITCH instruction is completed when the endswitch + * instruction is processed because we do not know the number + * of case statements or the default block id in advance.*/ + dst_ins = &flattener->instructions[cf_info->u.switch_.ins_location]; + if (!(src_params = instruction_src_params_alloc(dst_ins, cf_info->u.switch_.cases_count * 2 + 3, flattener))) + { + vkd3d_free(cf_info->u.switch_.cases); + return VKD3D_ERROR_OUT_OF_MEMORY; + } + src_params[0] = *cf_info->u.switch_.condition; + vsir_src_param_init_label(&src_params[1], cf_info->u.switch_.default_block_id); + vsir_src_param_init_label(&src_params[2], cf_info->u.switch_.merge_block_id); + for (j = 0; j < cf_info->u.switch_.cases_count; ++j) + { + unsigned int index = j * 2 + 3; + vsir_src_param_init(&src_params[index], VKD3DSPR_IMMCONST, VKD3D_DATA_UINT, 0); + src_params[index].reg.u.immconst_u32[0] = cf_info->u.switch_.cases[j].value; + vsir_src_param_init_label(&src_params[index + 1], cf_info->u.switch_.cases[j].block_id); + } + vkd3d_free(cf_info->u.switch_.cases); + + cf_flattener_pop_control_flow_level(flattener); + break; + } + + case VKD3DSIH_CASE: + { + unsigned int label_id, value; + + if (src->swizzle != VKD3D_SHADER_SWIZZLE(X, X, X, X)) + { + WARN("Unexpected src swizzle %#x.\n", src->swizzle); + vkd3d_shader_parser_error(parser, VKD3D_SHADER_ERROR_VSIR_INVALID_SWIZZLE, + "The swizzle for a switch case value is not scalar X."); + } + value = *src->reg.u.immconst_u32; + + if (!vkd3d_array_reserve((void **)&cf_info->u.switch_.cases, &cf_info->u.switch_.cases_size, + cf_info->u.switch_.cases_count + 1, sizeof(*cf_info->u.switch_.cases))) + return VKD3D_ERROR_OUT_OF_MEMORY; + + label_id = cf_flattener_alloc_block_id(flattener); + if (cf_info->inside_block) /* fall-through */ + cf_flattener_emit_unconditional_branch(flattener, label_id); + + cf_info->u.switch_.cases[cf_info->u.switch_.cases_count].value = value; + cf_info->u.switch_.cases[cf_info->u.switch_.cases_count].block_id = label_id; + ++cf_info->u.switch_.cases_count; + + cf_flattener_emit_label(flattener, label_id); + cf_flattener_create_block_name(flattener, label_id, "switch%u_case%u", cf_info->u.switch_.id, value); + cf_info->inside_block = true; + break; + } + + case VKD3DSIH_DEFAULT: + cf_info->u.switch_.default_block_id = cf_flattener_alloc_block_id(flattener); + if (cf_info->inside_block) /* fall-through */ + cf_flattener_emit_unconditional_branch(flattener, cf_info->u.switch_.default_block_id); + + cf_flattener_emit_label(flattener, cf_info->u.switch_.default_block_id); + + cf_flattener_create_block_name(flattener, cf_info->u.switch_.default_block_id, + "switch%u_default", cf_info->u.switch_.id); + cf_info->inside_block = true; + break; + + case VKD3DSIH_BREAK: + { + struct cf_flattener_info *breakable_cf_info; + + if (!(breakable_cf_info = cf_flattener_find_innermost_breakable_cf_construct(flattener))) + { + FIXME("Unhandled break instruction.\n"); + return VKD3D_ERROR_INVALID_SHADER; + } + + if (breakable_cf_info->current_block == VKD3D_BLOCK_LOOP) + { + cf_flattener_emit_unconditional_branch(flattener, breakable_cf_info->u.loop.merge_block_id); + } + else if (breakable_cf_info->current_block == VKD3D_BLOCK_SWITCH) + { + cf_flattener_emit_unconditional_branch(flattener, breakable_cf_info->u.switch_.merge_block_id); + } + + cf_info->inside_block = false; + break; + } + + case VKD3DSIH_BREAKP: + { + struct cf_flattener_info *loop_cf_info; + + if (!(loop_cf_info = cf_flattener_find_innermost_loop(flattener))) + { + ERR("Invalid 'breakc' instruction outside loop.\n"); + return VKD3D_ERROR_INVALID_SHADER; + } + + cf_flattener_emit_conditional_branch_and_merge(flattener, + src, loop_cf_info->u.loop.merge_block_id, instruction->flags); + break; + } + + case VKD3DSIH_CONTINUE: + { + struct cf_flattener_info *loop_cf_info; + + if (!(loop_cf_info = cf_flattener_find_innermost_loop(flattener))) + { + ERR("Invalid 'continue' instruction outside loop.\n"); + return VKD3D_ERROR_INVALID_SHADER; + } + + cf_flattener_emit_unconditional_branch(flattener, loop_cf_info->u.loop.continue_block_id); + + cf_info->inside_block = false; + break; + } + + case VKD3DSIH_CONTINUEP: + { + struct cf_flattener_info *loop_cf_info; + + if (!(loop_cf_info = cf_flattener_find_innermost_loop(flattener))) + { + ERR("Invalid 'continuec' instruction outside loop.\n"); + return VKD3D_ERROR_INVALID_SHADER; + } + + cf_flattener_emit_conditional_branch_and_merge(flattener, + src, loop_cf_info->u.loop.continue_block_id, instruction->flags); + break; + } + + case VKD3DSIH_RET: + if (!cf_flattener_copy_instruction(flattener, instruction)) + return VKD3D_ERROR_OUT_OF_MEMORY; + + if (cf_info) + cf_info->inside_block = false; + else + main_block_open = false; + break; + + default: + if (!cf_flattener_copy_instruction(flattener, instruction)) + return VKD3D_ERROR_OUT_OF_MEMORY; + break; + } + } + + if (main_block_open) + { + if (!(dst_ins = cf_flattener_require_space(flattener, 1))) + return VKD3D_ERROR_OUT_OF_MEMORY; + vsir_instruction_init(dst_ins, &flattener->location, VKD3DSIH_RET); + ++flattener->instruction_count; + } + + return flattener->allocation_failed ? VKD3D_ERROR_OUT_OF_MEMORY : VKD3D_OK; +} + +static enum vkd3d_result flatten_control_flow_constructs(struct vkd3d_shader_parser *parser) +{ + struct cf_flattener flattener = {0}; + enum vkd3d_result result; + + flattener.parser = parser; + result = cf_flattener_iterate_instruction_array(&flattener); + + if (result >= 0) + { + vkd3d_free(parser->instructions.elements); + parser->instructions.elements = flattener.instructions; + parser->instructions.capacity = flattener.instruction_capacity; + parser->instructions.count = flattener.instruction_count; + parser->shader_desc.block_count = flattener.block_id; + } + else + { + vkd3d_free(flattener.instructions); + } + + vkd3d_free(flattener.control_flow_info); + /* Simpler to always free these in free_shader_desc(). */ + parser->shader_desc.block_names = flattener.block_names; + parser->shader_desc.block_name_count = flattener.block_name_count; + + return result; +} + enum vkd3d_result vkd3d_shader_normalise(struct vkd3d_shader_parser *parser, const struct vkd3d_shader_compile_info *compile_info) { @@ -1504,6 +2255,9 @@ enum vkd3d_result vkd3d_shader_normalise(struct vkd3d_shader_parser *parser, if (result >= 0) remove_dead_code(parser); + if (result >= 0) + result = flatten_control_flow_constructs(parser); + if (result >= 0) result = normalise_combined_samplers(parser); diff --git a/libs/vkd3d/libs/vkd3d-shader/spirv.c b/libs/vkd3d/libs/vkd3d-shader/spirv.c index 7743319bed5..0eeb04bfe59 100644 --- a/libs/vkd3d/libs/vkd3d-shader/spirv.c +++ b/libs/vkd3d/libs/vkd3d-shader/spirv.c @@ -855,20 +855,6 @@ static void vkd3d_spirv_end_function_stream_insertion(struct vkd3d_spirv_builder builder->insertion_location = ~(size_t)0; } -struct vkd3d_spirv_op_branch_conditional -{ - uint32_t opcode; - uint32_t condition_id; - uint32_t true_label; - uint32_t false_label; -}; - -static struct vkd3d_spirv_op_branch_conditional *vkd3d_spirv_as_op_branch_conditional( - struct vkd3d_spirv_stream *stream, size_t location) -{ - return (struct vkd3d_spirv_op_branch_conditional *)&stream->words[location]; -} - static void vkd3d_spirv_build_op_capability(struct vkd3d_spirv_stream *stream, SpvCapability cap) { @@ -1885,8 +1871,6 @@ static void vkd3d_spirv_builder_begin_main_function(struct vkd3d_spirv_builder * vkd3d_spirv_build_op_function(builder, void_id, builder->main_function_id, SpvFunctionControlMaskNone, function_type_id); - vkd3d_spirv_build_op_label(builder, vkd3d_spirv_alloc_id(builder)); - builder->main_function_location = vkd3d_spirv_stream_current_location(&builder->function_stream); } static void vkd3d_spirv_builder_free(struct vkd3d_spirv_builder *builder) @@ -2267,51 +2251,6 @@ static const char *debug_vkd3d_symbol(const struct vkd3d_symbol *symbol) } } -struct vkd3d_if_cf_info -{ - size_t stream_location; - unsigned int id; - uint32_t merge_block_id; - uint32_t else_block_id; -}; - -struct vkd3d_loop_cf_info -{ - uint32_t header_block_id; - uint32_t continue_block_id; - uint32_t merge_block_id; -}; - -struct vkd3d_switch_cf_info -{ - size_t stream_location; - unsigned int id; - uint32_t selector_id; - uint32_t merge_block_id; - uint32_t default_block_id; - uint32_t *case_blocks; - size_t case_blocks_size; - unsigned int case_block_count; -}; - -struct vkd3d_control_flow_info -{ - union - { - struct vkd3d_if_cf_info if_; - struct vkd3d_loop_cf_info loop; - struct vkd3d_switch_cf_info switch_; - } u; - - enum - { - VKD3D_BLOCK_IF, - VKD3D_BLOCK_LOOP, - VKD3D_BLOCK_SWITCH, - } current_block; - bool inside_block; -}; - struct vkd3d_push_constant_buffer_binding { struct vkd3d_shader_register reg; @@ -2365,13 +2304,6 @@ struct spirv_compiler enum vkd3d_shader_type shader_type; - unsigned int branch_id; - unsigned int loop_id; - unsigned int switch_id; - unsigned int control_flow_depth; - struct vkd3d_control_flow_info *control_flow_info; - size_t control_flow_info_size; - struct vkd3d_shader_interface_info shader_interface; struct vkd3d_shader_descriptor_offset_info offset_info; uint32_t descriptor_offsets_member_id; @@ -2380,8 +2312,7 @@ struct spirv_compiler struct vkd3d_push_constant_buffer_binding *push_constants; const struct vkd3d_shader_spirv_target_info *spirv_target_info; - bool main_block_open; - bool after_declarations_section; + bool prolog_emitted; struct shader_signature input_signature; struct shader_signature output_signature; struct shader_signature patch_constant_signature; @@ -2401,7 +2332,9 @@ struct spirv_compiler const struct vkd3d_shader_scan_descriptor_info1 *scan_descriptor_info; unsigned int input_control_point_count; unsigned int output_control_point_count; + bool use_vocp; + bool emit_point_size; enum vkd3d_shader_opcode phase; bool emit_default_control_point_phase; @@ -2422,6 +2355,11 @@ struct spirv_compiler unsigned int ssa_register_count; uint64_t config_flags; + + uint32_t *block_label_ids; + unsigned int block_count; + const char **block_names; + size_t block_name_count; }; static bool is_in_default_phase(const struct spirv_compiler *compiler) @@ -2440,6 +2378,9 @@ static bool is_in_fork_or_join_phase(const struct spirv_compiler *compiler) } static void spirv_compiler_emit_initial_declarations(struct spirv_compiler *compiler); +static size_t spirv_compiler_get_current_function_location(struct spirv_compiler *compiler); +static void spirv_compiler_emit_main_prolog(struct spirv_compiler *compiler); +static void spirv_compiler_emit_io_declarations(struct spirv_compiler *compiler); static const char *spirv_compiler_get_entry_point_name(const struct spirv_compiler *compiler) { @@ -2450,8 +2391,6 @@ static const char *spirv_compiler_get_entry_point_name(const struct spirv_compil static void spirv_compiler_destroy(struct spirv_compiler *compiler) { - vkd3d_free(compiler->control_flow_info); - vkd3d_free(compiler->output_info); vkd3d_free(compiler->push_constants); @@ -2470,6 +2409,7 @@ static void spirv_compiler_destroy(struct spirv_compiler *compiler) shader_signature_cleanup(&compiler->patch_constant_signature); vkd3d_free(compiler->ssa_register_info); + vkd3d_free(compiler->block_label_ids); vkd3d_free(compiler); } @@ -2592,6 +2532,8 @@ static struct spirv_compiler *spirv_compiler_create(const struct vkd3d_shader_ve if ((shader_interface = vkd3d_find_struct(compile_info->next, INTERFACE_INFO))) { compiler->xfb_info = vkd3d_find_struct(compile_info->next, TRANSFORM_FEEDBACK_INFO); + compiler->emit_point_size = compiler->xfb_info && compiler->xfb_info->element_count + && compiler->shader_type != VKD3D_SHADER_TYPE_GEOMETRY; compiler->shader_interface = *shader_interface; if (shader_interface->push_constant_buffer_count) @@ -2805,6 +2747,14 @@ static struct vkd3d_string_buffer *vkd3d_shader_register_range_string(struct spi return buffer; } +static uint32_t spirv_compiler_get_label_id(struct spirv_compiler *compiler, unsigned int block_id) +{ + --block_id; + if (!compiler->block_label_ids[block_id]) + compiler->block_label_ids[block_id] = vkd3d_spirv_alloc_id(&compiler->spirv_builder); + return compiler->block_label_ids[block_id]; +} + static struct vkd3d_shader_descriptor_binding spirv_compiler_get_descriptor_binding( struct spirv_compiler *compiler, const struct vkd3d_shader_register *reg, const struct vkd3d_shader_register_range *range, enum vkd3d_shader_resource_type resource_type, @@ -3520,11 +3470,13 @@ static uint32_t spirv_compiler_get_descriptor_index(struct spirv_compiler *compi index_ids[0] = compiler->descriptor_offsets_member_id; index_ids[1] = spirv_compiler_get_constant_uint(compiler, push_constant_index); ptr_type_id = vkd3d_spirv_get_op_type_pointer(builder, SpvStorageClassPushConstant, type_id); + vkd3d_spirv_begin_function_stream_insertion(builder, + spirv_compiler_get_current_function_location(compiler)); ptr_id = vkd3d_spirv_build_op_in_bounds_access_chain(builder, ptr_type_id, compiler->push_constants_var_id, index_ids, 2); offset_id = vkd3d_spirv_build_op_load(builder, type_id, ptr_id, SpvMemoryAccessMaskNone); - if (!compiler->control_flow_depth) - compiler->descriptor_offset_ids[push_constant_index] = offset_id; + vkd3d_spirv_end_function_stream_insertion(builder); + compiler->descriptor_offset_ids[push_constant_index] = offset_id; } index_id = vkd3d_spirv_build_op_iadd(builder, type_id, index_id, offset_id); } @@ -4770,7 +4722,7 @@ static uint32_t spirv_compiler_emit_builtin_variable_v(struct spirv_compiler *co assert(size_count <= ARRAY_SIZE(sizes)); memcpy(sizes, array_sizes, size_count * sizeof(sizes[0])); array_sizes = sizes; - sizes[size_count - 1] = max(sizes[size_count - 1], builtin->spirv_array_size); + sizes[0] = max(sizes[0], builtin->spirv_array_size); id = spirv_compiler_emit_array_variable(compiler, &builder->global_stream, storage_class, builtin->component_type, builtin->component_count, array_sizes, size_count); @@ -4841,12 +4793,12 @@ static uint32_t spirv_compiler_emit_input(struct spirv_compiler *compiler, builtin = get_spirv_builtin_for_sysval(compiler, sysval); - array_sizes[0] = (reg_type == VKD3DSPR_PATCHCONST ? 0 : compiler->input_control_point_count); - array_sizes[1] = signature_element->register_count; - if (array_sizes[1] == 1 && !vsir_sysval_semantic_is_tess_factor(signature_element->sysval_semantic) - && (!vsir_sysval_semantic_is_clip_cull(signature_element->sysval_semantic) || array_sizes[0])) + array_sizes[0] = signature_element->register_count; + array_sizes[1] = (reg_type == VKD3DSPR_PATCHCONST ? 0 : compiler->input_control_point_count); + if (array_sizes[0] == 1 && !vsir_sysval_semantic_is_tess_factor(signature_element->sysval_semantic) + && (!vsir_sysval_semantic_is_clip_cull(signature_element->sysval_semantic) || array_sizes[1])) { - array_sizes[1] = 0; + array_sizes[0] = 0; } write_mask = signature_element->mask; @@ -5183,10 +5135,10 @@ static void spirv_compiler_emit_output(struct spirv_compiler *compiler, /* Don't use builtins for TCS -> TES varyings. See spirv_compiler_emit_input(). */ if (compiler->shader_type == VKD3D_SHADER_TYPE_HULL && !is_patch_constant) sysval = VKD3D_SHADER_SV_NONE; - array_sizes[0] = (reg_type == VKD3DSPR_PATCHCONST ? 0 : compiler->output_control_point_count); - array_sizes[1] = signature_element->register_count; - if (array_sizes[1] == 1 && !vsir_sysval_semantic_is_tess_factor(signature_element->sysval_semantic)) - array_sizes[1] = 0; + array_sizes[0] = signature_element->register_count; + array_sizes[1] = (reg_type == VKD3DSPR_PATCHCONST ? 0 : compiler->output_control_point_count); + if (array_sizes[0] == 1 && !vsir_sysval_semantic_is_tess_factor(signature_element->sysval_semantic)) + array_sizes[0] = 0; builtin = vkd3d_get_spirv_builtin(compiler, reg_type, sysval); @@ -5547,7 +5499,6 @@ static void spirv_compiler_emit_initial_declarations(struct spirv_compiler *comp if (compiler->shader_type != VKD3D_SHADER_TYPE_HULL) { vkd3d_spirv_builder_begin_main_function(builder); - compiler->main_block_open = true; } } @@ -5650,7 +5601,10 @@ static void spirv_compiler_emit_dcl_indexable_temp(struct spirv_compiler *compil SpvStorageClass storage_class; size_t function_location; - storage_class = SpvStorageClassFunction; + /* Indexable temps may be used by more than one function in hull shaders, and + * declarations generally should not occur within VSIR code blocks unless function + * scope is specified, e.g. DXIL alloca. */ + storage_class = temp->has_function_scope ? SpvStorageClassFunction : SpvStorageClassPrivate; vsir_register_init(®, VKD3DSPR_IDXTEMP, VKD3D_DATA_FLOAT, 1); reg.idx[0].offset = temp->register_idx; @@ -6382,7 +6336,7 @@ static void spirv_compiler_emit_dcl_output_topology(struct spirv_compiler *compi { case VKD3D_PT_POINTLIST: mode = SpvExecutionModeOutputPoints; - spirv_compiler_emit_point_size(compiler); + compiler->emit_point_size = true; break; case VKD3D_PT_LINESTRIP: mode = SpvExecutionModeOutputLineStrip; @@ -6544,20 +6498,45 @@ static void spirv_compiler_enter_shader_phase(struct spirv_compiler *compiler, vkd3d_spirv_build_op_function(builder, void_id, function_id, SpvFunctionControlMaskNone, function_type_id); - vkd3d_spirv_build_op_label(builder, vkd3d_spirv_alloc_id(builder)); - compiler->phase = instruction->handler_idx; spirv_compiler_emit_shader_phase_name(compiler, function_id, NULL); phase = (instruction->handler_idx == VKD3DSIH_HS_CONTROL_POINT_PHASE) ? &compiler->control_point_phase : &compiler->patch_constant_phase; phase->function_id = function_id; - phase->function_location = vkd3d_spirv_stream_current_location(&builder->function_stream); + /* The insertion location must be set after the label is emitted. */ + phase->function_location = 0; if (instruction->handler_idx == VKD3DSIH_HS_CONTROL_POINT_PHASE) compiler->emit_default_control_point_phase = instruction->flags; } +static void spirv_compiler_initialise_block(struct spirv_compiler *compiler) +{ + struct vkd3d_spirv_builder *builder = &compiler->spirv_builder; + + /* Insertion locations must point immediately after the function's initial label. */ + if (compiler->shader_type == VKD3D_SHADER_TYPE_HULL) + { + struct vkd3d_shader_phase *phase = (compiler->phase == VKD3DSIH_HS_CONTROL_POINT_PHASE) + ? &compiler->control_point_phase : &compiler->patch_constant_phase; + if (!phase->function_location) + phase->function_location = vkd3d_spirv_stream_current_location(&builder->function_stream); + } + else if (!builder->main_function_location) + { + builder->main_function_location = vkd3d_spirv_stream_current_location(&builder->function_stream); + } + + /* I/O declarations can result in emission of fixups, which must occur after the initial label. */ + if (!compiler->prolog_emitted) + { + spirv_compiler_emit_main_prolog(compiler); + spirv_compiler_emit_io_declarations(compiler); + compiler->prolog_emitted = true; + } +} + static void spirv_compiler_emit_default_control_point_phase(struct spirv_compiler *compiler) { const struct shader_signature *output_signature = &compiler->output_signature; @@ -6570,6 +6549,8 @@ static void spirv_compiler_emit_default_control_point_phase(struct spirv_compile unsigned int component_count; unsigned int i; + vkd3d_spirv_build_op_label(builder, vkd3d_spirv_alloc_id(builder)); + spirv_compiler_initialise_block(compiler); invocation_id = spirv_compiler_emit_load_invocation_id(compiler); memset(&invocation, 0, sizeof(invocation)); @@ -6664,7 +6645,11 @@ static void spirv_compiler_emit_hull_shader_main(struct spirv_compiler *compiler struct vkd3d_spirv_builder *builder = &compiler->spirv_builder; uint32_t void_id; + /* If a patch constant function used descriptor indexing the offsets must be reloaded. */ + memset(compiler->descriptor_offset_ids, 0, compiler->offset_info.descriptor_table_count + * sizeof(*compiler->descriptor_offset_ids)); vkd3d_spirv_builder_begin_main_function(builder); + vkd3d_spirv_build_op_label(builder, vkd3d_spirv_alloc_id(builder)); void_id = vkd3d_spirv_get_op_type_void(builder); @@ -7608,379 +7593,129 @@ static void spirv_compiler_emit_kill(struct spirv_compiler *compiler, vkd3d_spirv_build_op_label(builder, merge_block_id); } -static struct vkd3d_control_flow_info *spirv_compiler_push_control_flow_level( - struct spirv_compiler *compiler) +static bool spirv_compiler_init_blocks(struct spirv_compiler *compiler, unsigned int block_count) { - if (!vkd3d_array_reserve((void **)&compiler->control_flow_info, &compiler->control_flow_info_size, - compiler->control_flow_depth + 1, sizeof(*compiler->control_flow_info))) - { - ERR("Failed to allocate control flow info structure.\n"); - return NULL; - } + compiler->block_count = block_count; + + if (!(compiler->block_label_ids = vkd3d_calloc(block_count, sizeof(*compiler->block_label_ids)))) + return false; - return &compiler->control_flow_info[compiler->control_flow_depth++]; + return true; } -static void spirv_compiler_pop_control_flow_level(struct spirv_compiler *compiler) +static void spirv_compiler_emit_label(struct spirv_compiler *compiler, + const struct vkd3d_shader_instruction *instruction) { - struct vkd3d_control_flow_info *cf_info; + struct vkd3d_spirv_builder *builder = &compiler->spirv_builder; + const struct vkd3d_shader_src_param *src = instruction->src; + unsigned int block_id = src->reg.idx[0].offset; + uint32_t label_id; + + label_id = spirv_compiler_get_label_id(compiler, block_id); + vkd3d_spirv_build_op_label(builder, label_id); - assert(compiler->control_flow_depth); + --block_id; + if (block_id < compiler->block_name_count && compiler->block_names[block_id]) + vkd3d_spirv_build_op_name(builder, label_id, compiler->block_names[block_id]); - cf_info = &compiler->control_flow_info[--compiler->control_flow_depth]; - memset(cf_info, 0, sizeof(*cf_info)); + spirv_compiler_initialise_block(compiler); } -static struct vkd3d_control_flow_info *spirv_compiler_find_innermost_loop( - struct spirv_compiler *compiler) +static void spirv_compiler_emit_merge(struct spirv_compiler *compiler, + uint32_t merge_block_id, uint32_t continue_block_id) { - int depth; + struct vkd3d_spirv_builder *builder = &compiler->spirv_builder; - for (depth = compiler->control_flow_depth - 1; depth >= 0; --depth) + if (!merge_block_id) + return; + + merge_block_id = spirv_compiler_get_label_id(compiler, merge_block_id); + if (!continue_block_id) { - if (compiler->control_flow_info[depth].current_block == VKD3D_BLOCK_LOOP) - return &compiler->control_flow_info[depth]; + vkd3d_spirv_build_op_selection_merge(builder, merge_block_id, SpvSelectionControlMaskNone); } - - return NULL; -} - -static struct vkd3d_control_flow_info *spirv_compiler_find_innermost_breakable_cf_construct( - struct spirv_compiler *compiler) -{ - int depth; - - for (depth = compiler->control_flow_depth - 1; depth >= 0; --depth) + else { - if (compiler->control_flow_info[depth].current_block == VKD3D_BLOCK_LOOP - || compiler->control_flow_info[depth].current_block == VKD3D_BLOCK_SWITCH) - return &compiler->control_flow_info[depth]; + continue_block_id = spirv_compiler_get_label_id(compiler, continue_block_id); + vkd3d_spirv_build_op_loop_merge(builder, merge_block_id, continue_block_id, SpvLoopControlMaskNone); } - - return NULL; } -static int spirv_compiler_emit_control_flow_instruction(struct spirv_compiler *compiler, +static void spirv_compiler_emit_branch(struct spirv_compiler *compiler, const struct vkd3d_shader_instruction *instruction) { - uint32_t loop_header_block_id, loop_body_block_id, continue_block_id; struct vkd3d_spirv_builder *builder = &compiler->spirv_builder; const struct vkd3d_shader_src_param *src = instruction->src; - uint32_t merge_block_id, val_id, condition_id, true_label; - struct vkd3d_control_flow_info *cf_info; - - cf_info = compiler->control_flow_depth - ? &compiler->control_flow_info[compiler->control_flow_depth - 1] : NULL; + uint32_t condition_id; - switch (instruction->handler_idx) + if (vsir_register_is_label(&src[0].reg)) { - case VKD3DSIH_IF: - if (!(cf_info = spirv_compiler_push_control_flow_level(compiler))) - return VKD3D_ERROR_OUT_OF_MEMORY; - - val_id = spirv_compiler_emit_load_src(compiler, src, VKD3DSP_WRITEMASK_0); - condition_id = spirv_compiler_emit_int_to_bool(compiler, instruction->flags, src->reg.data_type, 1, val_id); - - true_label = vkd3d_spirv_alloc_id(builder); - merge_block_id = vkd3d_spirv_alloc_id(builder); - vkd3d_spirv_build_op_selection_merge(builder, merge_block_id, SpvSelectionControlMaskNone); - cf_info->u.if_.stream_location = vkd3d_spirv_stream_current_location(&builder->function_stream); - vkd3d_spirv_build_op_branch_conditional(builder, condition_id, true_label, merge_block_id); - - vkd3d_spirv_build_op_label(builder, true_label); - - cf_info->u.if_.id = compiler->branch_id; - cf_info->u.if_.merge_block_id = merge_block_id; - cf_info->u.if_.else_block_id = 0; - cf_info->inside_block = true; - cf_info->current_block = VKD3D_BLOCK_IF; - - vkd3d_spirv_build_op_name(builder, merge_block_id, "branch%u_merge", compiler->branch_id); - vkd3d_spirv_build_op_name(builder, true_label, "branch%u_true", compiler->branch_id); - ++compiler->branch_id; - break; - - case VKD3DSIH_ELSE: - assert(compiler->control_flow_depth); - assert(cf_info->current_block == VKD3D_BLOCK_IF); - - if (cf_info->inside_block) - vkd3d_spirv_build_op_branch(builder, cf_info->u.if_.merge_block_id); - - cf_info->u.if_.else_block_id = vkd3d_spirv_alloc_id(builder); - vkd3d_spirv_as_op_branch_conditional(&builder->function_stream, - cf_info->u.if_.stream_location)->false_label = cf_info->u.if_.else_block_id; - vkd3d_spirv_build_op_name(builder, - cf_info->u.if_.else_block_id, "branch%u_false", cf_info->u.if_.id); - vkd3d_spirv_build_op_label(builder, cf_info->u.if_.else_block_id); - cf_info->inside_block = true; - break; - - case VKD3DSIH_ENDIF: - assert(compiler->control_flow_depth); - assert(cf_info->current_block == VKD3D_BLOCK_IF); - - if (cf_info->inside_block) - vkd3d_spirv_build_op_branch(builder, cf_info->u.if_.merge_block_id); - - vkd3d_spirv_build_op_label(builder, cf_info->u.if_.merge_block_id); - - spirv_compiler_pop_control_flow_level(compiler); - break; - - case VKD3DSIH_LOOP: - if (!(cf_info = spirv_compiler_push_control_flow_level(compiler))) - return VKD3D_ERROR_OUT_OF_MEMORY; - - loop_header_block_id = vkd3d_spirv_alloc_id(builder); - loop_body_block_id = vkd3d_spirv_alloc_id(builder); - continue_block_id = vkd3d_spirv_alloc_id(builder); - merge_block_id = vkd3d_spirv_alloc_id(builder); - - vkd3d_spirv_build_op_branch(builder, loop_header_block_id); - vkd3d_spirv_build_op_label(builder, loop_header_block_id); - vkd3d_spirv_build_op_loop_merge(builder, merge_block_id, continue_block_id, SpvLoopControlMaskNone); - vkd3d_spirv_build_op_branch(builder, loop_body_block_id); - - vkd3d_spirv_build_op_label(builder, loop_body_block_id); - - cf_info->u.loop.header_block_id = loop_header_block_id; - cf_info->u.loop.continue_block_id = continue_block_id; - cf_info->u.loop.merge_block_id = merge_block_id; - cf_info->current_block = VKD3D_BLOCK_LOOP; - cf_info->inside_block = true; - - vkd3d_spirv_build_op_name(builder, loop_header_block_id, "loop%u_header", compiler->loop_id); - vkd3d_spirv_build_op_name(builder, loop_body_block_id, "loop%u_body", compiler->loop_id); - vkd3d_spirv_build_op_name(builder, continue_block_id, "loop%u_continue", compiler->loop_id); - vkd3d_spirv_build_op_name(builder, merge_block_id, "loop%u_merge", compiler->loop_id); - ++compiler->loop_id; - break; - - case VKD3DSIH_ENDLOOP: - assert(compiler->control_flow_depth); - assert(cf_info->current_block == VKD3D_BLOCK_LOOP); - - /* The loop block may have already been ended by an unconditional - * break instruction right before the end of the loop. */ - if (cf_info->inside_block) - vkd3d_spirv_build_op_branch(builder, cf_info->u.loop.continue_block_id); - - vkd3d_spirv_build_op_label(builder, cf_info->u.loop.continue_block_id); - vkd3d_spirv_build_op_branch(builder, cf_info->u.loop.header_block_id); - vkd3d_spirv_build_op_label(builder, cf_info->u.loop.merge_block_id); - - spirv_compiler_pop_control_flow_level(compiler); - break; - - case VKD3DSIH_SWITCH: - if (!(cf_info = spirv_compiler_push_control_flow_level(compiler))) - return VKD3D_ERROR_OUT_OF_MEMORY; - - merge_block_id = vkd3d_spirv_alloc_id(builder); - - assert(src->reg.data_type == VKD3D_DATA_INT); - val_id = spirv_compiler_emit_load_src(compiler, src, VKD3DSP_WRITEMASK_0); - - vkd3d_spirv_build_op_selection_merge(builder, merge_block_id, SpvSelectionControlMaskNone); - - cf_info->u.switch_.id = compiler->switch_id; - cf_info->u.switch_.merge_block_id = merge_block_id; - cf_info->u.switch_.stream_location = vkd3d_spirv_stream_current_location(&builder->function_stream); - cf_info->u.switch_.selector_id = val_id; - cf_info->u.switch_.case_blocks = NULL; - cf_info->u.switch_.case_blocks_size = 0; - cf_info->u.switch_.case_block_count = 0; - cf_info->u.switch_.default_block_id = 0; - cf_info->inside_block = false; - cf_info->current_block = VKD3D_BLOCK_SWITCH; - - vkd3d_spirv_build_op_name(builder, merge_block_id, "switch%u_merge", compiler->switch_id); - - ++compiler->switch_id; - - if (!vkd3d_array_reserve((void **)&cf_info->u.switch_.case_blocks, &cf_info->u.switch_.case_blocks_size, - 10, sizeof(*cf_info->u.switch_.case_blocks))) - return VKD3D_ERROR_OUT_OF_MEMORY; - - break; - - case VKD3DSIH_ENDSWITCH: - assert(compiler->control_flow_depth); - assert(cf_info->current_block == VKD3D_BLOCK_SWITCH); - assert(!cf_info->inside_block); - - if (!cf_info->u.switch_.default_block_id) - cf_info->u.switch_.default_block_id = cf_info->u.switch_.merge_block_id; - - vkd3d_spirv_build_op_label(builder, cf_info->u.switch_.merge_block_id); - - /* The OpSwitch instruction is inserted when the endswitch - * instruction is processed because we do not know the number - * of case statements in advance.*/ - vkd3d_spirv_begin_function_stream_insertion(builder, cf_info->u.switch_.stream_location); - vkd3d_spirv_build_op_switch(builder, cf_info->u.switch_.selector_id, - cf_info->u.switch_.default_block_id, cf_info->u.switch_.case_blocks, - cf_info->u.switch_.case_block_count); - vkd3d_spirv_end_function_stream_insertion(builder); - - vkd3d_free(cf_info->u.switch_.case_blocks); - spirv_compiler_pop_control_flow_level(compiler); - break; - - case VKD3DSIH_CASE: - { - uint32_t label_id, value; - - assert(compiler->control_flow_depth); - assert(cf_info->current_block == VKD3D_BLOCK_SWITCH); - - if (src->swizzle != VKD3D_SHADER_SWIZZLE(X, X, X, X)) - { - WARN("Unexpected src swizzle %#x.\n", src->swizzle); - spirv_compiler_warning(compiler, VKD3D_SHADER_WARNING_SPV_INVALID_SWIZZLE, - "The swizzle for a switch case value is not scalar."); - } - assert(src->reg.type == VKD3DSPR_IMMCONST); - value = *src->reg.u.immconst_u32; - - if (!vkd3d_array_reserve((void **)&cf_info->u.switch_.case_blocks, &cf_info->u.switch_.case_blocks_size, - 2 * (cf_info->u.switch_.case_block_count + 1), sizeof(*cf_info->u.switch_.case_blocks))) - return VKD3D_ERROR_OUT_OF_MEMORY; - - label_id = vkd3d_spirv_alloc_id(builder); - if (cf_info->inside_block) /* fall-through */ - vkd3d_spirv_build_op_branch(builder, label_id); - - cf_info->u.switch_.case_blocks[2 * cf_info->u.switch_.case_block_count + 0] = value; - cf_info->u.switch_.case_blocks[2 * cf_info->u.switch_.case_block_count + 1] = label_id; - ++cf_info->u.switch_.case_block_count; - - vkd3d_spirv_build_op_label(builder, label_id); - cf_info->inside_block = true; - vkd3d_spirv_build_op_name(builder, label_id, "switch%u_case%u", cf_info->u.switch_.id, value); - break; - } - - case VKD3DSIH_DEFAULT: - assert(compiler->control_flow_depth); - assert(cf_info->current_block == VKD3D_BLOCK_SWITCH); - assert(!cf_info->u.switch_.default_block_id); - - cf_info->u.switch_.default_block_id = vkd3d_spirv_alloc_id(builder); - if (cf_info->inside_block) /* fall-through */ - vkd3d_spirv_build_op_branch(builder, cf_info->u.switch_.default_block_id); - - vkd3d_spirv_build_op_label(builder, cf_info->u.switch_.default_block_id); - vkd3d_spirv_build_op_name(builder, cf_info->u.switch_.default_block_id, - "switch%u_default", cf_info->u.switch_.id); - cf_info->inside_block = true; - break; - - case VKD3DSIH_BREAK: - { - struct vkd3d_control_flow_info *breakable_cf_info; - - assert(compiler->control_flow_depth); - - if (!(breakable_cf_info = spirv_compiler_find_innermost_breakable_cf_construct(compiler))) - { - FIXME("Unhandled break instruction.\n"); - return VKD3D_ERROR_INVALID_SHADER; - } - - if (breakable_cf_info->current_block == VKD3D_BLOCK_LOOP) - { - vkd3d_spirv_build_op_branch(builder, breakable_cf_info->u.loop.merge_block_id); - } - else if (breakable_cf_info->current_block == VKD3D_BLOCK_SWITCH) - { - /* The current case block may have already been ended by an - * unconditional continue instruction. */ - if (breakable_cf_info->inside_block) - vkd3d_spirv_build_op_branch(builder, breakable_cf_info->u.switch_.merge_block_id); - } - - cf_info->inside_block = false; - break; - } - - case VKD3DSIH_BREAKP: + if (instruction->src_count > 1) { - struct vkd3d_control_flow_info *loop_cf_info; - - assert(compiler->control_flow_depth); - - if (!(loop_cf_info = spirv_compiler_find_innermost_loop(compiler))) - { - ERR("Invalid 'breakc' instruction outside loop.\n"); - return VKD3D_ERROR_INVALID_SHADER; - } - - merge_block_id = spirv_compiler_emit_conditional_branch(compiler, - instruction, loop_cf_info->u.loop.merge_block_id); - vkd3d_spirv_build_op_label(builder, merge_block_id); - break; - } - - case VKD3DSIH_CONTINUE: - { - struct vkd3d_control_flow_info *loop_cf_info; - - assert(compiler->control_flow_depth); - - if (!(loop_cf_info = spirv_compiler_find_innermost_loop(compiler))) - { - ERR("Invalid 'continue' instruction outside loop.\n"); - return VKD3D_ERROR_INVALID_SHADER; - } - - vkd3d_spirv_build_op_branch(builder, loop_cf_info->u.loop.continue_block_id); - - cf_info->inside_block = false; - break; + /* Loop merge only. Must have a merge block and a continue block. */ + spirv_compiler_emit_merge(compiler, src[1].reg.idx[0].offset, src[2].reg.idx[0].offset); } + vkd3d_spirv_build_op_branch(builder, spirv_compiler_get_label_id(compiler, src[0].reg.idx[0].offset)); + return; + } - case VKD3DSIH_CONTINUEP: - { - struct vkd3d_control_flow_info *loop_cf_info; - - if (!(loop_cf_info = spirv_compiler_find_innermost_loop(compiler))) - { - ERR("Invalid 'continuec' instruction outside loop.\n"); - return VKD3D_ERROR_INVALID_SHADER; - } + if (!vkd3d_swizzle_is_scalar(src->swizzle)) + { + WARN("Unexpected src swizzle %#x.\n", src->swizzle); + spirv_compiler_warning(compiler, VKD3D_SHADER_WARNING_SPV_INVALID_SWIZZLE, + "The swizzle for a branch condition value is not scalar."); + } - merge_block_id = spirv_compiler_emit_conditional_branch(compiler, - instruction, loop_cf_info->u.loop.continue_block_id); - vkd3d_spirv_build_op_label(builder, merge_block_id); - break; - } + condition_id = spirv_compiler_emit_load_src(compiler, &src[0], VKD3DSP_WRITEMASK_0); + condition_id = spirv_compiler_emit_int_to_bool(compiler, + VKD3D_SHADER_CONDITIONAL_OP_NZ, src[0].reg.data_type, 1, condition_id); + /* Emit the merge immediately before the branch instruction. */ + spirv_compiler_emit_merge(compiler, src[3].reg.idx[0].offset, + (instruction->src_count > 4) ? src[4].reg.idx[0].offset : 0); + vkd3d_spirv_build_op_branch_conditional(builder, condition_id, + spirv_compiler_get_label_id(compiler, src[1].reg.idx[0].offset), + spirv_compiler_get_label_id(compiler, src[2].reg.idx[0].offset)); +} - case VKD3DSIH_RET: - spirv_compiler_emit_return(compiler, instruction); +static void spirv_compiler_emit_switch(struct spirv_compiler *compiler, + const struct vkd3d_shader_instruction *instruction) +{ + struct vkd3d_spirv_builder *builder = &compiler->spirv_builder; + const struct vkd3d_shader_src_param *src = instruction->src; + uint32_t val_id, default_id; + unsigned int i, word_count; + uint32_t *cases; - if (cf_info) - cf_info->inside_block = false; - else - compiler->main_block_open = false; - break; + if (!vkd3d_swizzle_is_scalar(src[0].swizzle)) + { + WARN("Unexpected src swizzle %#x.\n", src[0].swizzle); + spirv_compiler_warning(compiler, VKD3D_SHADER_WARNING_SPV_INVALID_SWIZZLE, + "The swizzle for a switch value is not scalar."); + } - case VKD3DSIH_RETP: - spirv_compiler_emit_retc(compiler, instruction); - break; + word_count = instruction->src_count - 3; + if (!(cases = vkd3d_calloc(word_count, sizeof(*cases)))) + { + spirv_compiler_error(compiler, VKD3D_SHADER_ERROR_SPV_OUT_OF_MEMORY, + "Failed to allocate %u words for switch cases.", word_count); + return; + } - case VKD3DSIH_DISCARD: - case VKD3DSIH_TEXKILL: - spirv_compiler_emit_kill(compiler, instruction); - break; + val_id = spirv_compiler_emit_load_src(compiler, &src[0], VKD3DSP_WRITEMASK_0); + default_id = spirv_compiler_get_label_id(compiler, src[1].reg.idx[0].offset); + /* No instructions may occur between the merge and the switch. */ + spirv_compiler_emit_merge(compiler, src[2].reg.idx[0].offset, 0); - default: - ERR("Unexpected instruction %#x.\n", instruction->handler_idx); - break; + src = &src[3]; + for (i = 0; i < word_count; i += 2) + { + cases[i] = src[i].reg.u.immconst_u32[0]; + cases[i + 1] = spirv_compiler_get_label_id(compiler, src[i + 1].reg.idx[0].offset); } - return VKD3D_OK; + vkd3d_spirv_build_op_switch(builder, val_id, default_id, cases, word_count / 2u); + + vkd3d_free(cases); } static void spirv_compiler_emit_deriv_instruction(struct spirv_compiler *compiler, @@ -9519,28 +9254,15 @@ static void spirv_compiler_emit_main_prolog(struct spirv_compiler *compiler) { spirv_compiler_emit_push_constant_buffers(compiler); - if (compiler->xfb_info && compiler->xfb_info->element_count - && compiler->shader_type != VKD3D_SHADER_TYPE_GEOMETRY) + if (compiler->emit_point_size) spirv_compiler_emit_point_size(compiler); } -static bool is_dcl_instruction(enum vkd3d_shader_opcode handler_idx) -{ - return (VKD3DSIH_DCL <= handler_idx && handler_idx <= VKD3DSIH_DCL_VERTICES_OUT) - || handler_idx == VKD3DSIH_HS_DECLS; -} - static int spirv_compiler_handle_instruction(struct spirv_compiler *compiler, const struct vkd3d_shader_instruction *instruction) { int ret = VKD3D_OK; - if (!is_dcl_instruction(instruction->handler_idx) && !compiler->after_declarations_section) - { - compiler->after_declarations_section = true; - spirv_compiler_emit_main_prolog(compiler); - } - switch (instruction->handler_idx) { case VKD3DSIH_DCL_GLOBAL_FLAGS: @@ -9727,24 +9449,24 @@ static int spirv_compiler_handle_instruction(struct spirv_compiler *compiler, case VKD3DSIH_F32TOF16: spirv_compiler_emit_f32tof16(compiler, instruction); break; - case VKD3DSIH_BREAK: - case VKD3DSIH_BREAKP: - case VKD3DSIH_CASE: - case VKD3DSIH_CONTINUE: - case VKD3DSIH_CONTINUEP: - case VKD3DSIH_DEFAULT: - case VKD3DSIH_DISCARD: - case VKD3DSIH_ELSE: - case VKD3DSIH_ENDIF: - case VKD3DSIH_ENDLOOP: - case VKD3DSIH_ENDSWITCH: - case VKD3DSIH_IF: - case VKD3DSIH_LOOP: case VKD3DSIH_RET: + spirv_compiler_emit_return(compiler, instruction); + break; case VKD3DSIH_RETP: - case VKD3DSIH_SWITCH: + spirv_compiler_emit_retc(compiler, instruction); + break; + case VKD3DSIH_DISCARD: case VKD3DSIH_TEXKILL: - ret = spirv_compiler_emit_control_flow_instruction(compiler, instruction); + spirv_compiler_emit_kill(compiler, instruction); + break; + case VKD3DSIH_LABEL: + spirv_compiler_emit_label(compiler, instruction); + break; + case VKD3DSIH_BRANCH: + spirv_compiler_emit_branch(compiler, instruction); + break; + case VKD3DSIH_SWITCH_MONOLITHIC: + spirv_compiler_emit_switch(compiler, instruction); break; case VKD3DSIH_DSX: case VKD3DSIH_DSX_COARSE: @@ -9966,6 +9688,9 @@ static int spirv_compiler_generate_spirv(struct spirv_compiler *compiler, if ((result = vkd3d_shader_normalise(parser, compile_info)) < 0) return result; + if (parser->shader_desc.block_count && !spirv_compiler_init_blocks(compiler, parser->shader_desc.block_count)) + return VKD3D_ERROR_OUT_OF_MEMORY; + instructions = parser->instructions; memset(&parser->instructions, 0, sizeof(parser->instructions)); @@ -9976,6 +9701,8 @@ static int spirv_compiler_generate_spirv(struct spirv_compiler *compiler, memset(&shader_desc->output_signature, 0, sizeof(shader_desc->output_signature)); memset(&shader_desc->patch_constant_signature, 0, sizeof(shader_desc->patch_constant_signature)); compiler->use_vocp = parser->shader_desc.use_vocp; + compiler->block_names = parser->shader_desc.block_names; + compiler->block_name_count = parser->shader_desc.block_name_count; compiler->input_control_point_count = shader_desc->input_control_point_count; compiler->output_control_point_count = shader_desc->output_control_point_count; @@ -9983,8 +9710,6 @@ static int spirv_compiler_generate_spirv(struct spirv_compiler *compiler, if (compiler->shader_type != VKD3D_SHADER_TYPE_HULL) spirv_compiler_emit_shader_signature_outputs(compiler); - spirv_compiler_emit_io_declarations(compiler); - for (i = 0; i < instructions.count && result >= 0; ++i) { compiler->location.line = i + 1; @@ -9996,9 +9721,6 @@ static int spirv_compiler_generate_spirv(struct spirv_compiler *compiler, if (result < 0) return result; - if (compiler->main_block_open) - vkd3d_spirv_build_op_return(builder); - if (!is_in_default_phase(compiler)) spirv_compiler_leave_shader_phase(compiler); else diff --git a/libs/vkd3d/libs/vkd3d-shader/tpf.c b/libs/vkd3d/libs/vkd3d-shader/tpf.c index 1f1944bc0da..e4dfb5235ec 100644 --- a/libs/vkd3d/libs/vkd3d-shader/tpf.c +++ b/libs/vkd3d/libs/vkd3d-shader/tpf.c @@ -919,6 +919,7 @@ static void shader_sm4_read_dcl_index_range(struct vkd3d_shader_instruction *ins { struct vkd3d_shader_index_range *index_range = &ins->declaration.index_range; unsigned int i, register_idx, register_count; + const struct shader_signature *signature; enum vkd3d_shader_register_type type; struct sm4_index_range_array *ranges; unsigned int *io_masks; @@ -932,40 +933,38 @@ static void shader_sm4_read_dcl_index_range(struct vkd3d_shader_instruction *ins register_count = index_range->register_count; write_mask = index_range->dst.write_mask; - if (vsir_write_mask_component_count(write_mask) != 1) - { - WARN("Unhandled write mask %#x.\n", write_mask); - vkd3d_shader_parser_warning(&priv->p, VKD3D_SHADER_WARNING_TPF_UNHANDLED_INDEX_RANGE_MASK, - "Index range mask %#x is not scalar.", write_mask); - } - switch ((type = index_range->dst.reg.type)) { case VKD3DSPR_INPUT: case VKD3DSPR_INCONTROLPOINT: io_masks = priv->input_register_masks; ranges = &priv->input_index_ranges; + signature = &priv->p.shader_desc.input_signature; break; case VKD3DSPR_OUTPUT: if (sm4_parser_is_in_fork_or_join_phase(priv)) { io_masks = priv->patch_constant_register_masks; ranges = &priv->patch_constant_index_ranges; + signature = &priv->p.shader_desc.patch_constant_signature; } else { io_masks = priv->output_register_masks; ranges = &priv->output_index_ranges; + signature = &priv->p.shader_desc.output_signature; } break; case VKD3DSPR_COLOROUT: case VKD3DSPR_OUTCONTROLPOINT: io_masks = priv->output_register_masks; ranges = &priv->output_index_ranges; + signature = &priv->p.shader_desc.output_signature; break; case VKD3DSPR_PATCHCONST: io_masks = priv->patch_constant_register_masks; ranges = &priv->patch_constant_index_ranges; + signature = &priv->p.shader_desc.patch_constant_signature; break; default: @@ -1003,6 +1002,18 @@ static void shader_sm4_read_dcl_index_range(struct vkd3d_shader_instruction *ins for (i = 0; i < register_count; ++i) { + const struct signature_element *e = vsir_signature_find_element_for_reg(signature, register_idx + i, write_mask); + /* Index ranges should not contain non-arrayed sysvals. FXC tries to forbid this but it is buggy, + * and can emit a range containing a sysval if the sysval is not actually accessed. */ + if (e && e->sysval_semantic && register_count > 1 && !vsir_sysval_semantic_is_tess_factor(e->sysval_semantic) + && !vsir_sysval_semantic_is_clip_cull(e->sysval_semantic)) + { + WARN("Sysval %u included in an index range declaration.\n", e->sysval_semantic); + vkd3d_shader_parser_error(&priv->p, VKD3D_SHADER_ERROR_TPF_INVALID_INDEX_RANGE_DCL, + "Index range base %u, count %u, mask %#x contains sysval %u.", + register_idx, register_count, write_mask, e->sysval_semantic); + return; + } if ((io_masks[register_idx + i] & write_mask) != write_mask) { WARN("No matching declaration for index range base %u, count %u, mask %#x.\n", @@ -1117,6 +1128,7 @@ static void shader_sm4_read_dcl_indexable_temp(struct vkd3d_shader_instruction * ins->declaration.indexable_temp.alignment = 0; ins->declaration.indexable_temp.data_type = VKD3D_DATA_FLOAT; ins->declaration.indexable_temp.component_count = *tokens; + ins->declaration.indexable_temp.has_function_scope = false; } static void shader_sm4_read_dcl_global_flags(struct vkd3d_shader_instruction *ins, uint32_t opcode, diff --git a/libs/vkd3d/libs/vkd3d-shader/vkd3d_shader_private.h b/libs/vkd3d/libs/vkd3d-shader/vkd3d_shader_private.h index 0af7ea0d266..ace58161e6b 100644 --- a/libs/vkd3d/libs/vkd3d-shader/vkd3d_shader_private.h +++ b/libs/vkd3d/libs/vkd3d-shader/vkd3d_shader_private.h @@ -238,6 +238,7 @@ enum vkd3d_shader_opcode VKD3DSIH_BEM, VKD3DSIH_BFI, VKD3DSIH_BFREV, + VKD3DSIH_BRANCH, VKD3DSIH_BREAK, VKD3DSIH_BREAKC, VKD3DSIH_BREAKP, @@ -472,6 +473,7 @@ enum vkd3d_shader_opcode VKD3DSIH_SUB, VKD3DSIH_SWAPC, VKD3DSIH_SWITCH, + VKD3DSIH_SWITCH_MONOLITHIC, VKD3DSIH_SYNC, VKD3DSIH_TEX, VKD3DSIH_TEXBEM, @@ -818,6 +820,7 @@ struct vkd3d_shader_indexable_temp unsigned int alignment; enum vkd3d_data_type data_type; unsigned int component_count; + bool has_function_scope; const struct vkd3d_shader_immediate_constant_buffer *initialiser; }; @@ -883,6 +886,8 @@ struct vkd3d_shader_src_param enum vkd3d_shader_src_modifier modifiers; }; +void vsir_src_param_init_label(struct vkd3d_shader_src_param *param, unsigned int label_id); + struct vkd3d_shader_index_range { struct vkd3d_shader_dst_param dst; @@ -1016,6 +1021,7 @@ struct vkd3d_shader_desc uint32_t temp_count; unsigned int ssa_count; + unsigned int block_count; struct { @@ -1023,6 +1029,9 @@ struct vkd3d_shader_desc } flat_constant_count[3]; bool use_vocp; + + const char **block_names; + size_t block_name_count; }; struct vkd3d_shader_register_semantic @@ -1185,6 +1194,11 @@ static inline bool register_is_constant(const struct vkd3d_shader_register *reg) return (reg->type == VKD3DSPR_IMMCONST || reg->type == VKD3DSPR_IMMCONST64); } +static inline bool vsir_register_is_label(const struct vkd3d_shader_register *reg) +{ + return reg->type == VKD3DSPR_LABEL; +} + struct vkd3d_shader_param_node { struct vkd3d_shader_param_node *next; diff --git a/libs/vkd3d/libs/vkd3d/resource.c b/libs/vkd3d/libs/vkd3d/resource.c index 163dd6ce210..f9e50335fa8 100644 --- a/libs/vkd3d/libs/vkd3d/resource.c +++ b/libs/vkd3d/libs/vkd3d/resource.c @@ -22,7 +22,7 @@ #define VKD3D_NULL_BUFFER_SIZE 16 #define VKD3D_NULL_VIEW_FORMAT DXGI_FORMAT_R8G8B8A8_UNORM -LONG64 object_global_serial_id; +uint64_t object_global_serial_id; static inline bool is_cpu_accessible_heap(const D3D12_HEAP_PROPERTIES *properties) { @@ -4314,7 +4314,7 @@ static HRESULT d3d12_descriptor_heap_init(struct d3d12_descriptor_heap *descript descriptor_heap->ID3D12DescriptorHeap_iface.lpVtbl = &d3d12_descriptor_heap_vtbl; descriptor_heap->refcount = 1; - descriptor_heap->serial_id = InterlockedIncrement64(&object_global_serial_id); + descriptor_heap->serial_id = vkd3d_atomic_increment_u64(&object_global_serial_id); descriptor_heap->desc = *desc; diff --git a/libs/vkd3d/libs/vkd3d/vkd3d_private.h b/libs/vkd3d/libs/vkd3d/vkd3d_private.h index 9bf4575d46a..4abe7df3a95 100644 --- a/libs/vkd3d/libs/vkd3d/vkd3d_private.h +++ b/libs/vkd3d/libs/vkd3d/vkd3d_private.h @@ -67,7 +67,7 @@ * this number to prevent excessive pool memory use. */ #define VKD3D_MAX_VIRTUAL_HEAP_DESCRIPTORS_PER_TYPE (16 * 1024u) -extern LONG64 object_global_serial_id; +extern uint64_t object_global_serial_id; struct d3d12_command_list; struct d3d12_device; -- 2.43.0