[PATCH v7 0/6] MR491: vkd3d-shader/dxil: Handle control flow instructions.
Conflicts with !450 but can probably go upstream before it since backend changes here are minimal. -- v7: vkd3d-shader/dxil: Handle the DXIL SWITCH instruction. vkd3d-shader: Rename shader_instruction_array_add_icb() to shader_instruction_array_add_opaque_param(). vkd3d-shader/dxil: Handle the DXIL PHI instruction. vkd3d-shader/dxil: Handle the DXIL BR instruction conditional variant. vkd3d-shader/dxil: Handle the DXIL BR instruction unconditional variant. https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/491
From: Conor McCarthy <cmccarthy(a)codeweavers.com> --- libs/vkd3d-shader/dxil.c | 31 ++++++++++++++++++++++++++++++- 1 file changed, 30 insertions(+), 1 deletion(-) diff --git a/libs/vkd3d-shader/dxil.c b/libs/vkd3d-shader/dxil.c index 1709212fa..4090293b1 100644 --- a/libs/vkd3d-shader/dxil.c +++ b/libs/vkd3d-shader/dxil.c @@ -448,11 +448,23 @@ struct sm6_symbol const char *name; }; +enum sm6_block_terminator_type +{ + TERMINATOR_RET, +}; + +struct sm6_block_terminator +{ + enum sm6_block_terminator_type type; +}; + struct sm6_block { struct vkd3d_shader_instruction *instructions; size_t instruction_capacity; size_t instruction_count; + + struct sm6_block_terminator terminator; }; struct sm6_function @@ -4061,6 +4073,8 @@ static void sm6_parser_emit_ret(struct sm6_parser *sm6, const struct dxil_record if (record->operand_count) FIXME("Non-void return is not implemented.\n"); + code_block->terminator.type = TERMINATOR_RET; + ins->handler_idx = VKD3DSIH_NOP; } @@ -4299,6 +4313,21 @@ static enum vkd3d_result sm6_parser_function_init(struct sm6_parser *sm6, const return VKD3D_OK; } +static void sm6_block_emit_terminator(const struct sm6_block *block, struct sm6_parser *sm6) +{ + switch (block->terminator.type) + { + case TERMINATOR_RET: + { + sm6_parser_add_instruction(sm6, VKD3DSIH_RET); + break; + } + + default: + vkd3d_unreachable(); + } +} + static bool sm6_block_emit_instructions(struct sm6_block *block, struct sm6_parser *sm6) { struct vkd3d_shader_instruction *ins = sm6_parser_require_space(sm6, block->instruction_count + 1); @@ -4309,7 +4338,7 @@ static bool sm6_block_emit_instructions(struct sm6_block *block, struct sm6_pars memcpy(ins, block->instructions, block->instruction_count * sizeof(*block->instructions)); sm6->p.instructions.count += block->instruction_count; - sm6_parser_add_instruction(sm6, VKD3DSIH_RET); + sm6_block_emit_terminator(block, sm6); return true; } -- GitLab https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/491
From: Conor McCarthy <cmccarthy(a)codeweavers.com> --- libs/vkd3d-shader/d3d_asm.c | 9 ++ libs/vkd3d-shader/dxil.c | 172 ++++++++++++++++++----- libs/vkd3d-shader/ir.c | 20 +++ libs/vkd3d-shader/vkd3d_shader_private.h | 9 ++ 4 files changed, 176 insertions(+), 34 deletions(-) diff --git a/libs/vkd3d-shader/d3d_asm.c b/libs/vkd3d-shader/d3d_asm.c index 82d1d71d9..48c44e89f 100644 --- a/libs/vkd3d-shader/d3d_asm.c +++ b/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", @@ -1646,6 +1647,14 @@ static void shader_dump_instruction(struct vkd3d_d3d_asm_compiler *compiler, switch (ins->handler_idx) { + case VKD3DSIH_BRANCH: + { + const struct vkd3d_shader_branch *branch = &ins->declaration.branch; + vkd3d_string_buffer_printf(buffer, " %sl%u%s", compiler->colours.reg, branch->true_id, + compiler->colours.reset); + break; + } + case VKD3DSIH_DCL: case VKD3DSIH_DCL_UAV_TYPED: vkd3d_string_buffer_printf(buffer, "%s", compiler->colours.opcode); diff --git a/libs/vkd3d-shader/dxil.c b/libs/vkd3d-shader/dxil.c index 4090293b1..b211ca229 100644 --- a/libs/vkd3d-shader/dxil.c +++ b/libs/vkd3d-shader/dxil.c @@ -450,12 +450,14 @@ struct sm6_symbol enum sm6_block_terminator_type { + TERMINATOR_UNCOND_BR, TERMINATOR_RET, }; struct sm6_block_terminator { enum sm6_block_terminator_type type; + const struct sm6_block *true_block; }; struct sm6_block @@ -464,6 +466,9 @@ struct sm6_block size_t instruction_capacity; size_t instruction_count; + /* A nonzero id. */ + unsigned int id; + struct sm6_block_terminator terminator; }; @@ -471,7 +476,8 @@ struct sm6_function { const struct sm6_value *declaration; - struct sm6_block *blocks[1]; + struct sm6_block **blocks; + size_t block_capacity; size_t block_count; size_t value_count; @@ -3182,6 +3188,45 @@ static void sm6_parser_emit_binop(struct sm6_parser *sm6, const struct dxil_reco } } +static const struct sm6_block *sm6_function_get_block(const struct sm6_function *function, uint64_t index, + struct sm6_parser *sm6) +{ + if (index >= function->block_count) + { + WARN("Invalid code block index %#"PRIx64".\n", index); + vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_INVALID_OPERAND_COUNT, + "Invalid code block index %#"PRIx64" for a control flow instruction.", index); + return NULL; + } + return function->blocks[index]; +} + +static void sm6_parser_emit_br(struct sm6_parser *sm6, const struct dxil_record *record, + struct sm6_function *function, struct sm6_block *code_block, struct vkd3d_shader_instruction *ins) +{ + if (record->operand_count != 1 && record->operand_count < 3) + { + WARN("Invalid operand count %u.\n", record->operand_count); + vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_INVALID_OPERAND_COUNT, + "Invalid operand count %u for a branch instruction.", record->operand_count); + return; + } + + if (record->operand_count == 1) + { + code_block->terminator.type = TERMINATOR_UNCOND_BR; + code_block->terminator.true_block = sm6_function_get_block(function, record->operands[0], sm6); + } + else + { + FIXME("Conditional branch is not implemented.\n"); + vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_INVALID_OPERAND_COUNT, + "Conditional branch is not implemented."); + } + + ins->handler_idx = VKD3DSIH_NOP; +} + static void sm6_parser_emit_dx_cbuffer_load(struct sm6_parser *sm6, struct sm6_block *code_block, enum dx_intrinsic_opcode op, const struct sm6_value **operands, struct vkd3d_shader_instruction *ins) { @@ -4166,6 +4211,31 @@ static bool sm6_metadata_get_uint64_value(const struct sm6_parser *sm6, return true; } +static bool sm6_function_blocks_reserve(struct sm6_function *function, unsigned int reserve) +{ + if (!vkd3d_array_reserve((void **)&function->blocks, &function->block_capacity, + reserve, sizeof(*function->blocks))) + { + ERR("Failed to allocate code block array.\n"); + return false; + } + return true; +} + +static struct sm6_block *sm6_function_create_block(struct sm6_function *function) +{ + struct sm6_block *block; + + if (!(block = sm6_block_create())) + return NULL; + + function->blocks[function->block_count++] = block; + /* Set the id to the array index + 1. */ + block->id = function->block_count; + + return block; +} + static enum vkd3d_result sm6_parser_function_init(struct sm6_parser *sm6, const struct dxil_block *block, struct sm6_function *function) { @@ -4205,16 +4275,18 @@ static enum vkd3d_result sm6_parser_function_init(struct sm6_parser *sm6, const WARN("Function contains no blocks.\n"); return VKD3D_ERROR_INVALID_SHADER; } - if (block_count > 1) - { - FIXME("Branched shaders are not supported yet.\n"); - return VKD3D_ERROR_INVALID_SHADER; - } - if (!(function->blocks[0] = sm6_block_create())) - { - ERR("Failed to allocate code block.\n"); + if (!sm6_function_blocks_reserve(function, block_count)) return VKD3D_ERROR_OUT_OF_MEMORY; + + /* Pre-allocate all blocks to simplify instruction parsing. */ + for (i = 0; i < block_count; ++i) + { + if (!sm6_function_create_block(function)) + { + ERR("Failed to allocate code block.\n"); + return VKD3D_ERROR_OUT_OF_MEMORY; + } } function->block_count = block_count; code_block = function->blocks[0]; @@ -4233,10 +4305,9 @@ static enum vkd3d_result sm6_parser_function_init(struct sm6_parser *sm6, const return VKD3D_ERROR_INVALID_SHADER; } - /* block->record_count - 1 is the instruction count, but some instructions - * can emit >1 IR instruction, so extra may be used. */ + /* Some instructions can emit >1 IR instruction, so extra may be used. */ if (!vkd3d_array_reserve((void **)&code_block->instructions, &code_block->instruction_capacity, - max(code_block->instruction_count + 1, block->record_count), sizeof(*code_block->instructions))) + code_block->instruction_count + 1, sizeof(*code_block->instructions))) { ERR("Failed to allocate instructions.\n"); return VKD3D_ERROR_OUT_OF_MEMORY; @@ -4256,6 +4327,10 @@ static enum vkd3d_result sm6_parser_function_init(struct sm6_parser *sm6, const case FUNC_CODE_INST_BINOP: sm6_parser_emit_binop(sm6, record, ins, dst); break; + case FUNC_CODE_INST_BR: + sm6_parser_emit_br(sm6, record, function, code_block, ins); + is_terminator = true; + break; case FUNC_CODE_INST_CALL: sm6_parser_emit_call(sm6, record, code_block, ins, dst); break; @@ -4315,8 +4390,19 @@ static enum vkd3d_result sm6_parser_function_init(struct sm6_parser *sm6, const static void sm6_block_emit_terminator(const struct sm6_block *block, struct sm6_parser *sm6) { + struct vkd3d_shader_instruction *ins; + switch (block->terminator.type) { + case TERMINATOR_UNCOND_BR: + { + if (!block->terminator.true_block) + return; + ins = sm6_parser_add_instruction(sm6, VKD3DSIH_BRANCH); + ins->declaration.branch.true_id = block->terminator.true_block->id; + break; + } + case TERMINATOR_RET: { sm6_parser_add_instruction(sm6, VKD3DSIH_RET); @@ -4328,21 +4414,6 @@ static void sm6_block_emit_terminator(const struct sm6_block *block, struct sm6_ } } -static bool sm6_block_emit_instructions(struct sm6_block *block, struct sm6_parser *sm6) -{ - struct vkd3d_shader_instruction *ins = sm6_parser_require_space(sm6, block->instruction_count + 1); - - if (!ins) - return false; - - memcpy(ins, block->instructions, block->instruction_count * sizeof(*block->instructions)); - sm6->p.instructions.count += block->instruction_count; - - sm6_block_emit_terminator(block, sm6); - - return true; -} - static enum vkd3d_result sm6_parser_module_init(struct sm6_parser *sm6, const struct dxil_block *block, unsigned int level) { @@ -4398,6 +4469,43 @@ 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_instruction *ins; + + ins = sm6_parser_require_space(sm6, 1); + vsir_instruction_init_label(ins, &sm6->p.location, label_id, &sm6->p); + ++sm6->p.instructions.count; +} + +static enum vkd3d_result sm6_function_emit_blocks(const struct sm6_function *function, struct sm6_parser *sm6) +{ + unsigned int i; + + for (i = 0; i < function->block_count; ++i) + { + const struct sm6_block *block = function->blocks[i]; + + /* Space for the label and terminator. */ + if (!sm6_parser_require_space(sm6, block->instruction_count + 2)) + { + vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_OUT_OF_MEMORY, + "Out of memory emitting shader instructions."); + return VKD3D_ERROR_OUT_OF_MEMORY; + } + if (i) + sm6_parser_emit_label(sm6, block->id); + + memcpy(&sm6->p.instructions.elements[sm6->p.instructions.count], block->instructions, + block->instruction_count * sizeof(*block->instructions)); + sm6->p.instructions.count += block->instruction_count; + + sm6_block_emit_terminator(block, sm6); + } + + return VKD3D_OK; +} + static bool sm6_parser_allocate_named_metadata(struct sm6_parser *sm6) { struct dxil_block *block; @@ -5463,6 +5571,7 @@ static void sm6_functions_cleanup(struct sm6_function *functions, size_t count) { for (j = 0; j < functions[i].block_count; ++j) sm6_block_destroy(functions[i].blocks[j]); + vkd3d_free(functions[i].blocks); } vkd3d_free(functions); } @@ -5759,13 +5868,8 @@ static enum vkd3d_result sm6_parser_init(struct sm6_parser *sm6, const uint32_t return VKD3D_ERROR_INVALID_SHADER; } - assert(sm6->function_count == 1); - if (!sm6_block_emit_instructions(fn->blocks[0], sm6)) - { - vkd3d_shader_error(message_context, &location, VKD3D_SHADER_ERROR_DXIL_OUT_OF_MEMORY, - "Out of memory emitting shader instructions."); - return VKD3D_ERROR_OUT_OF_MEMORY; - } + if ((ret = sm6_function_emit_blocks(fn, sm6)) < 0) + return ret; dxil_block_destroy(&sm6->root_block); diff --git a/libs/vkd3d-shader/ir.c b/libs/vkd3d-shader/ir.c index acdb660ea..bf2b0d3d3 100644 --- a/libs/vkd3d-shader/ir.c +++ b/libs/vkd3d-shader/ir.c @@ -320,6 +320,26 @@ void vsir_instruction_init(struct vkd3d_shader_instruction *ins, const struct vk ins->handler_idx = handler_idx; } +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_register_init(&src_param->reg, VKD3DSPR_LABEL, VKD3D_DATA_UINT, 1); + src_param->reg.idx[0].offset = label_id; + src_param->swizzle = 0; + src_param->modifiers = 0; + + 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}; diff --git a/libs/vkd3d-shader/vkd3d_shader_private.h b/libs/vkd3d-shader/vkd3d_shader_private.h index d3989672b..ae4a0ab7b 100644 --- a/libs/vkd3d-shader/vkd3d_shader_private.h +++ b/libs/vkd3d-shader/vkd3d_shader_private.h @@ -235,6 +235,7 @@ enum vkd3d_shader_opcode VKD3DSIH_BEM, VKD3DSIH_BFI, VKD3DSIH_BFREV, + VKD3DSIH_BRANCH, VKD3DSIH_BREAK, VKD3DSIH_BREAKC, VKD3DSIH_BREAKP, @@ -1101,6 +1102,11 @@ struct vkd3d_shader_location unsigned int line, column; }; +struct vkd3d_shader_branch +{ + unsigned int true_id; +}; + struct vkd3d_shader_instruction { struct vkd3d_shader_location location; @@ -1140,6 +1146,7 @@ struct vkd3d_shader_instruction struct vkd3d_shader_index_range index_range; struct vkd3d_shader_indexable_temp indexable_temp; struct vkd3d_shader_function_table_pointer fp; + struct vkd3d_shader_branch branch; } declaration; }; @@ -1155,6 +1162,8 @@ static inline bool vkd3d_shader_ver_le(const struct vkd3d_shader_version *v, uns void vsir_instruction_init(struct vkd3d_shader_instruction *ins, const struct vkd3d_shader_location *location, enum vkd3d_shader_opcode handler_idx); +bool vsir_instruction_init_label(struct vkd3d_shader_instruction *ins, const struct vkd3d_shader_location *location, + unsigned int label_id, void *parser); static inline bool vkd3d_shader_instruction_has_texel_offset(const struct vkd3d_shader_instruction *ins) { -- GitLab https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/491
From: Conor McCarthy <cmccarthy(a)codeweavers.com> --- libs/vkd3d-shader/d3d_asm.c | 15 +++++++-- libs/vkd3d-shader/dxil.c | 39 ++++++++++++++++++++++-- libs/vkd3d-shader/vkd3d_shader_private.h | 1 + 3 files changed, 50 insertions(+), 5 deletions(-) diff --git a/libs/vkd3d-shader/d3d_asm.c b/libs/vkd3d-shader/d3d_asm.c index 48c44e89f..ea0662017 100644 --- a/libs/vkd3d-shader/d3d_asm.c +++ b/libs/vkd3d-shader/d3d_asm.c @@ -1650,8 +1650,19 @@ static void shader_dump_instruction(struct vkd3d_d3d_asm_compiler *compiler, case VKD3DSIH_BRANCH: { const struct vkd3d_shader_branch *branch = &ins->declaration.branch; - vkd3d_string_buffer_printf(buffer, " %sl%u%s", compiler->colours.reg, branch->true_id, - compiler->colours.reset); + if (ins->src_count) + { + shader_dump_instruction_flags(compiler, ins); + vkd3d_string_buffer_printf(buffer, " "); + shader_dump_src_param(compiler, ins->src); + vkd3d_string_buffer_printf(buffer, " ? %sl%u%s : %sl%u%s", compiler->colours.reg, branch->true_id, + compiler->colours.reset, compiler->colours.reg, branch->false_id, compiler->colours.reset); + } + else + { + vkd3d_string_buffer_printf(buffer, " %sl%u%s", compiler->colours.reg, branch->true_id, + compiler->colours.reset); + } break; } diff --git a/libs/vkd3d-shader/dxil.c b/libs/vkd3d-shader/dxil.c index b211ca229..5e15ee2bf 100644 --- a/libs/vkd3d-shader/dxil.c +++ b/libs/vkd3d-shader/dxil.c @@ -451,13 +451,16 @@ struct sm6_symbol enum sm6_block_terminator_type { TERMINATOR_UNCOND_BR, + TERMINATOR_COND_BR, TERMINATOR_RET, }; struct sm6_block_terminator { + struct vkd3d_shader_register conditional_reg; enum sm6_block_terminator_type type; const struct sm6_block *true_block; + const struct sm6_block *false_block; }; struct sm6_block @@ -3204,6 +3207,9 @@ static const struct sm6_block *sm6_function_get_block(const struct sm6_function static void sm6_parser_emit_br(struct sm6_parser *sm6, const struct dxil_record *record, struct sm6_function *function, struct sm6_block *code_block, struct vkd3d_shader_instruction *ins) { + const struct sm6_value *value; + unsigned int i = 2; + if (record->operand_count != 1 && record->operand_count < 3) { WARN("Invalid operand count %u.\n", record->operand_count); @@ -3219,9 +3225,22 @@ static void sm6_parser_emit_br(struct sm6_parser *sm6, const struct dxil_record } else { - FIXME("Conditional branch is not implemented.\n"); - vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_INVALID_OPERAND_COUNT, - "Conditional branch is not implemented."); + if (!sm6->bool_type) + { + WARN("Bool type not found.\n"); + vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_INVALID_MODULE, + "Module does not define a boolean type for conditions."); + return; + } + if (!(value = sm6_parser_get_value_by_ref(sm6, record, sm6->bool_type, &i)) + || !sm6_value_validate_is_bool(value, sm6)) + return; + dxil_record_validate_operand_max_count(record, i, sm6); + + code_block->terminator.type = TERMINATOR_COND_BR; + code_block->terminator.conditional_reg = value->u.reg; + code_block->terminator.true_block = sm6_function_get_block(function, record->operands[0], sm6); + code_block->terminator.false_block = sm6_function_get_block(function, record->operands[1], sm6); } ins->handler_idx = VKD3DSIH_NOP; @@ -4390,6 +4409,7 @@ static enum vkd3d_result sm6_parser_function_init(struct sm6_parser *sm6, const static void sm6_block_emit_terminator(const struct sm6_block *block, struct sm6_parser *sm6) { + struct vkd3d_shader_src_param *src_param; struct vkd3d_shader_instruction *ins; switch (block->terminator.type) @@ -4403,6 +4423,19 @@ static void sm6_block_emit_terminator(const struct sm6_block *block, struct sm6_ break; } + case TERMINATOR_COND_BR: + { + if (!block->terminator.true_block || !block->terminator.false_block) + return; + ins = sm6_parser_add_instruction(sm6, VKD3DSIH_BRANCH); + src_param = instruction_src_params_alloc(ins, 1, sm6); + src_param_init(src_param); + src_param->reg = block->terminator.conditional_reg; + ins->declaration.branch.true_id = block->terminator.true_block->id; + ins->declaration.branch.false_id = block->terminator.false_block->id; + break; + } + case TERMINATOR_RET: { sm6_parser_add_instruction(sm6, VKD3DSIH_RET); diff --git a/libs/vkd3d-shader/vkd3d_shader_private.h b/libs/vkd3d-shader/vkd3d_shader_private.h index ae4a0ab7b..e95d47480 100644 --- a/libs/vkd3d-shader/vkd3d_shader_private.h +++ b/libs/vkd3d-shader/vkd3d_shader_private.h @@ -1105,6 +1105,7 @@ struct vkd3d_shader_location struct vkd3d_shader_branch { unsigned int true_id; + unsigned int false_id; }; struct vkd3d_shader_instruction -- GitLab https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/491
From: Conor McCarthy <cmccarthy(a)codeweavers.com> --- libs/vkd3d-shader/d3d_asm.c | 13 ++ libs/vkd3d-shader/dxil.c | 222 ++++++++++++++++++++++- libs/vkd3d-shader/ir.c | 1 + libs/vkd3d-shader/spirv.c | 7 +- libs/vkd3d-shader/vkd3d_shader_private.h | 8 + 5 files changed, 246 insertions(+), 5 deletions(-) diff --git a/libs/vkd3d-shader/d3d_asm.c b/libs/vkd3d-shader/d3d_asm.c index ea0662017..274c3552f 100644 --- a/libs/vkd3d-shader/d3d_asm.c +++ b/libs/vkd3d-shader/d3d_asm.c @@ -242,6 +242,7 @@ static const char * const shader_opcode_names[] = [VKD3DSIH_NRM ] = "nrm", [VKD3DSIH_OR ] = "or", [VKD3DSIH_PHASE ] = "phase", + [VKD3DSIH_PHI ] = "phi", [VKD3DSIH_POW ] = "pow", [VKD3DSIH_RCP ] = "rcp", [VKD3DSIH_REP ] = "rep", @@ -1876,6 +1877,18 @@ static void shader_dump_instruction(struct vkd3d_d3d_asm_compiler *compiler, shader_print_bool_literal(compiler, " = ", ins->src[0].reg.u.immconst_uint[0], ""); break; + case VKD3DSIH_PHI: + vkd3d_string_buffer_printf(buffer, " "); + shader_dump_dst_param(compiler, ins->dst, false); + for (i = 0; i < ins->src_count; ++i) + { + vkd3d_string_buffer_printf(buffer, ", ["); + shader_dump_src_param(compiler, &ins->src[i]); + vkd3d_string_buffer_printf(buffer, ", %sl%u%s]", compiler->colours.reg, + ins->src[i].reg.block_id, compiler->colours.reset); + } + break; + default: shader_dump_instruction_flags(compiler, ins); diff --git a/libs/vkd3d-shader/dxil.c b/libs/vkd3d-shader/dxil.c index 5e15ee2bf..6aa527356 100644 --- a/libs/vkd3d-shader/dxil.c +++ b/libs/vkd3d-shader/dxil.c @@ -448,6 +448,20 @@ struct sm6_symbol const char *name; }; +struct incoming_value +{ + const struct sm6_block *block; + struct vkd3d_shader_register reg; +}; + +struct sm6_phi +{ + struct vkd3d_shader_register reg; + struct incoming_value *incoming; + size_t incoming_capacity; + size_t incoming_count; +}; + enum sm6_block_terminator_type { TERMINATOR_UNCOND_BR, @@ -472,6 +486,10 @@ struct sm6_block /* A nonzero id. */ unsigned int id; + struct sm6_phi *phi; + size_t phi_capacity; + size_t phi_count; + struct sm6_block_terminator terminator; }; @@ -1926,6 +1944,11 @@ static bool sm6_value_is_icb(const struct sm6_value *value) return value->value_type == VALUE_TYPE_ICB; } +static bool sm6_value_is_ssa(const struct sm6_value *value) +{ + return sm6_value_is_register(value) && register_is_ssa(&value->u.reg); +} + static inline unsigned int sm6_value_get_constant_uint(const struct sm6_value *value) { if (!sm6_value_is_constant(value)) @@ -3012,6 +3035,26 @@ static struct sm6_block *sm6_block_create() return block; } +static struct sm6_phi *sm6_block_phi_require_space(struct sm6_block *block, struct sm6_parser *sm6) +{ + struct sm6_phi *phi; + + if (!vkd3d_array_reserve((void **)&block->phi, &block->phi_capacity, block->phi_count + 1, sizeof(*block->phi))) + { + ERR("Failed to allocate phi array.\n"); + vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_OUT_OF_MEMORY, + "Out of memory allocating a phi instruction."); + return NULL; + } + phi = &block->phi[block->phi_count++]; + + phi->incoming = NULL; + phi->incoming_capacity = 0; + phi->incoming_count = 0; + + return phi; +} + static enum vkd3d_shader_opcode map_binary_op(uint64_t code, const struct sm6_type *type_a, const struct sm6_type *type_b, struct sm6_parser *sm6) { @@ -4128,6 +4171,72 @@ static void sm6_parser_emit_load(struct sm6_parser *sm6, const struct dxil_recor instruction_dst_param_init_ssa_scalar(ins, sm6); } +static void sm6_parser_emit_phi(struct sm6_parser *sm6, const struct dxil_record *record, + struct sm6_function *function, struct sm6_block *code_block, struct vkd3d_shader_instruction *ins, + struct sm6_value *dst) +{ + const struct sm6_type *type; + struct sm6_phi *phi; + unsigned int i, j; + uint64_t src_idx; + + if (!(record->operand_count & 1)) + { + WARN("Invalid operand count %u.\n", record->operand_count); + vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_INVALID_OPERAND_COUNT, + "Invalid operand count %u for phi instruction.", record->operand_count); + return; + } + + if (!(type = sm6_parser_get_type(sm6, record->operands[0]))) + return; + if (!sm6_type_is_numeric(type)) + { + /* dxc doesn't seem to use buffer/resource read return types here. */ + FIXME("Only scalar numeric types are supported.\n"); + vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_INVALID_OPERAND_COUNT, + "Result type class %u of a phi instruction is not scalar numeric.", type->class); + return; + } + + dst->type = type; + register_init_ssa_scalar(&dst->u.reg, type, sm6); + + if (!(phi = sm6_block_phi_require_space(code_block, sm6))) + return; + phi->reg = dst->u.reg; + phi->incoming_count = record->operand_count / 2u; + + if (!vkd3d_array_reserve((void **)&phi->incoming, &phi->incoming_capacity, phi->incoming_count, + sizeof(*phi->incoming))) + { + ERR("Failed to allocate phi incoming array.\n"); + vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_OUT_OF_MEMORY, + "Out of memory allocating a phi incoming array."); + return; + } + + for (i = 1; i < record->operand_count; i += 2) + { + src_idx = sm6->value_count - decode_rotated_signed_value(record->operands[i]); + /* May be a forward reference. */ + if (src_idx >= sm6->cur_max_value) + { + FIXME("Invalid value index %"PRIu64".\n", src_idx); + vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_INVALID_OPERAND, + "Invalid value index %"PRIu64" for a phi incoming value.", src_idx); + return; + } + + j = i / 2u; + /* Store the value index in the register for later resolution. */ + phi->incoming[j].reg.idx[0].offset = src_idx; + phi->incoming[j].block = sm6_function_get_block(function, record->operands[i + 1], sm6); + } + + ins->handler_idx = VKD3DSIH_NOP; +} + static void sm6_parser_emit_ret(struct sm6_parser *sm6, const struct dxil_record *record, struct sm6_block *code_block, struct vkd3d_shader_instruction *ins) { @@ -4255,6 +4364,45 @@ static struct sm6_block *sm6_function_create_block(struct sm6_function *function return block; } +static enum vkd3d_result sm6_function_resolve_phi_incomings(const struct sm6_function *function, + struct sm6_parser *sm6) +{ + const struct sm6_block *block; + size_t i, j, block_idx; + + for (block_idx = 0; block_idx < function->block_count; ++block_idx) + { + block = function->blocks[block_idx]; + + for (i = 0; i < block->phi_count; ++i) + { + struct sm6_phi *phi = &block->phi[i]; + const struct sm6_value *src; + + for (j = 0; j < phi->incoming_count; ++j) + { + src = &sm6->values[phi->incoming[j].reg.idx[0].offset]; + if (!sm6_value_is_constant(src) && !sm6_value_is_undef(src) && !sm6_value_is_ssa(src)) + { + FIXME("PHI incoming value is not a constant or SSA register.\n"); + vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_INVALID_OPERAND, + "A PHI incoming value is not a constant or SSA register."); + return VKD3D_ERROR_INVALID_SHADER; + } + if (src->u.reg.data_type != phi->reg.data_type) + { + WARN("Type mismatch.\n"); + vkd3d_shader_parser_warning(&sm6->p, VKD3D_SHADER_WARNING_DXIL_TYPE_MISMATCH, + "The type of a phi incoming value does not match the result type."); + } + phi->incoming[j].reg = src->u.reg; + } + } + } + + return VKD3D_OK; +} + static enum vkd3d_result sm6_parser_function_init(struct sm6_parser *sm6, const struct dxil_block *block, struct sm6_function *function) { @@ -4368,6 +4516,9 @@ static enum vkd3d_result sm6_parser_function_init(struct sm6_parser *sm6, const case FUNC_CODE_INST_LOAD: sm6_parser_emit_load(sm6, record, ins, dst); break; + case FUNC_CODE_INST_PHI: + sm6_parser_emit_phi(sm6, record, function, code_block, ins, dst); + break; case FUNC_CODE_INST_RET: sm6_parser_emit_ret(sm6, record, code_block, ins); is_terminator = true; @@ -4404,7 +4555,7 @@ static enum vkd3d_result sm6_parser_function_init(struct sm6_parser *sm6, const return VKD3D_ERROR_INVALID_SHADER; } - return VKD3D_OK; + return sm6_function_resolve_phi_incomings(function, sm6); } static void sm6_block_emit_terminator(const struct sm6_block *block, struct sm6_parser *sm6) @@ -4447,6 +4598,45 @@ static void sm6_block_emit_terminator(const struct sm6_block *block, struct sm6_ } } +static void sm6_block_emit_phi(const struct sm6_block *block, struct sm6_parser *sm6) +{ + struct vkd3d_shader_instruction *ins; + unsigned int i, j, incoming_count; + const struct sm6_phi *src_phi; + + sm6_parser_require_space(sm6, block->phi_count); + + for (i = 0; i < block->phi_count; ++i) + { + struct vkd3d_shader_src_param *src_params; + struct vkd3d_shader_dst_param *dst_param; + + src_phi = &block->phi[i]; + incoming_count = src_phi->incoming_count; + + ins = sm6_parser_add_instruction(sm6, VKD3DSIH_PHI); + if (!(src_params = instruction_src_params_alloc(ins, incoming_count, sm6))) + return; + if (!(dst_param = instruction_dst_params_alloc(ins, 1, sm6))) + return; + + for (j = 0; j < incoming_count; ++j) + { + const struct sm6_block *incoming_block = src_phi->incoming[j].block; + + src_param_init(&src_params[j]); + src_params[j].reg = src_phi->incoming[j].reg; + if (incoming_block) + src_params[j].reg.block_id = incoming_block->id; + else + assert(sm6->p.failed); + } + + dst_param_init(dst_param); + dst_param->reg = src_phi->reg; + } +} + static enum vkd3d_result sm6_parser_module_init(struct sm6_parser *sm6, const struct dxil_block *block, unsigned int level) { @@ -4502,6 +4692,14 @@ static enum vkd3d_result sm6_parser_module_init(struct sm6_parser *sm6, const st return VKD3D_OK; } +static void sm6_parser_emit_branch(struct sm6_parser *sm6, unsigned int block_idx) +{ + struct vkd3d_shader_instruction *ins; + + ins = sm6_parser_add_instruction(sm6, VKD3DSIH_BRANCH); + ins->declaration.branch.true_id = block_idx; +} + static void sm6_parser_emit_label(struct sm6_parser *sm6, unsigned int label_id) { struct vkd3d_shader_instruction *ins; @@ -4519,15 +4717,21 @@ static enum vkd3d_result sm6_function_emit_blocks(const struct sm6_function *fun { const struct sm6_block *block = function->blocks[i]; - /* Space for the label and terminator. */ - if (!sm6_parser_require_space(sm6, block->instruction_count + 2)) + /* Space for a branch, label and terminator. */ + if (!sm6_parser_require_space(sm6, block->instruction_count + block->phi_count + 3)) { vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_OUT_OF_MEMORY, "Out of memory emitting shader instructions."); return VKD3D_ERROR_OUT_OF_MEMORY; } - if (i) + if (function->block_count > 1) + { + /* The entry block cannot be a loop header. Emit branch and label in case there is a loop. */ + if (!i) + sm6_parser_emit_branch(sm6, block->id); sm6_parser_emit_label(sm6, block->id); + sm6_block_emit_phi(block, sm6); + } memcpy(&sm6->p.instructions.elements[sm6->p.instructions.count], block->instructions, block->instruction_count * sizeof(*block->instructions)); @@ -5590,9 +5794,19 @@ static void sm6_symtab_cleanup(struct sm6_symbol *symbols, size_t count) vkd3d_free(symbols); } +static void sm6_phi_destroy(struct sm6_phi *phi) +{ + vkd3d_free(phi->incoming); +} + static void sm6_block_destroy(struct sm6_block *block) { + unsigned int i; + vkd3d_free(block->instructions); + for (i = 0; i < block->phi_count; ++i) + sm6_phi_destroy(&block->phi[i]); + vkd3d_free(block->phi); vkd3d_free(block); } diff --git a/libs/vkd3d-shader/ir.c b/libs/vkd3d-shader/ir.c index bf2b0d3d3..80f624a2a 100644 --- a/libs/vkd3d-shader/ir.c +++ b/libs/vkd3d-shader/ir.c @@ -310,6 +310,7 @@ void vsir_register_init(struct vkd3d_shader_register *reg, enum vkd3d_shader_reg reg->idx_count = idx_count; reg->dimension = VSIR_DIMENSION_SCALAR; reg->alignment = 0; + memset(®->u, 0, sizeof(reg->u)); } void vsir_instruction_init(struct vkd3d_shader_instruction *ins, const struct vkd3d_shader_location *location, diff --git a/libs/vkd3d-shader/spirv.c b/libs/vkd3d-shader/spirv.c index be149a0cf..4f59cf3de 100644 --- a/libs/vkd3d-shader/spirv.c +++ b/libs/vkd3d-shader/spirv.c @@ -3785,7 +3785,12 @@ static uint32_t spirv_compiler_emit_load_ssa_reg(struct spirv_compiler *compiler ssa = spirv_compiler_get_ssa_register_info(compiler, reg); val_id = ssa->id; - assert(val_id); + if (!val_id) + { + /* Should only be from a missing instruction implementation. */ + assert(compiler->failed); + return 0; + } assert(vkd3d_swizzle_is_scalar(swizzle)); if (reg->dimension == VSIR_DIMENSION_SCALAR) diff --git a/libs/vkd3d-shader/vkd3d_shader_private.h b/libs/vkd3d-shader/vkd3d_shader_private.h index e95d47480..6d0acc85d 100644 --- a/libs/vkd3d-shader/vkd3d_shader_private.h +++ b/libs/vkd3d-shader/vkd3d_shader_private.h @@ -433,6 +433,7 @@ enum vkd3d_shader_opcode VKD3DSIH_NRM, VKD3DSIH_OR, VKD3DSIH_PHASE, + VKD3DSIH_PHI, VKD3DSIH_POW, VKD3DSIH_RCP, VKD3DSIH_REP, @@ -830,6 +831,8 @@ struct vkd3d_shader_register enum vsir_dimension dimension; /* known address alignment for optimisation, or zero */ unsigned int alignment; + /* for phi incomings: id of a predecessor block dominated by the reg's origin block. */ + unsigned int block_id; union { DWORD immconst_uint[VKD3D_VEC4_SIZE]; @@ -1176,6 +1179,11 @@ static inline bool register_is_constant(const struct vkd3d_shader_register *reg) return (reg->type == VKD3DSPR_IMMCONST || reg->type == VKD3DSPR_IMMCONST64); } +static inline bool register_is_ssa(const struct vkd3d_shader_register *reg) +{ + return reg->type == VKD3DSPR_SSA; +} + struct vkd3d_shader_param_node { struct vkd3d_shader_param_node *next; -- GitLab https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/491
From: Conor McCarthy <cmccarthy(a)codeweavers.com> --- libs/vkd3d-shader/dxil.c | 2 +- libs/vkd3d-shader/tpf.c | 2 +- libs/vkd3d-shader/vkd3d_shader_main.c | 15 +++++++-------- libs/vkd3d-shader/vkd3d_shader_private.h | 9 ++++----- 4 files changed, 13 insertions(+), 15 deletions(-) diff --git a/libs/vkd3d-shader/dxil.c b/libs/vkd3d-shader/dxil.c index 6aa527356..8861385b9 100644 --- a/libs/vkd3d-shader/dxil.c +++ b/libs/vkd3d-shader/dxil.c @@ -2465,7 +2465,7 @@ static enum vkd3d_result value_allocate_constant_array(struct sm6_value *dst, co "Out of memory allocating an immediate constant buffer of count %u.", count); return VKD3D_ERROR_OUT_OF_MEMORY; } - if (!shader_instruction_array_add_icb(&sm6->p.instructions, icb)) + if (!shader_instruction_array_add_opaque_param(&sm6->p.instructions, icb)) { ERR("Failed to store icb object.\n"); vkd3d_free(icb); diff --git a/libs/vkd3d-shader/tpf.c b/libs/vkd3d-shader/tpf.c index 61d14e30d..0652b26fc 100644 --- a/libs/vkd3d-shader/tpf.c +++ b/libs/vkd3d-shader/tpf.c @@ -796,7 +796,7 @@ static void shader_sm4_read_shader_data(struct vkd3d_shader_instruction *ins, ui icb->component_count = VKD3D_VEC4_SIZE; icb->element_count = icb_size / VKD3D_VEC4_SIZE; memcpy(icb->data, tokens, sizeof(*tokens) * icb_size); - shader_instruction_array_add_icb(&priv->p.instructions, icb); + shader_instruction_array_add_opaque_param(&priv->p.instructions, icb); ins->declaration.icb = icb; } diff --git a/libs/vkd3d-shader/vkd3d_shader_main.c b/libs/vkd3d-shader/vkd3d_shader_main.c index 77164c9c9..f2fccec66 100644 --- a/libs/vkd3d-shader/vkd3d_shader_main.c +++ b/libs/vkd3d-shader/vkd3d_shader_main.c @@ -2045,13 +2045,12 @@ bool shader_instruction_array_reserve(struct vkd3d_shader_instruction_array *ins return true; } -bool shader_instruction_array_add_icb(struct vkd3d_shader_instruction_array *instructions, - struct vkd3d_shader_immediate_constant_buffer *icb) +bool shader_instruction_array_add_opaque_param(struct vkd3d_shader_instruction_array *instructions, void *param) { - if (!vkd3d_array_reserve((void **)&instructions->icbs, &instructions->icb_capacity, instructions->icb_count + 1, - sizeof(*instructions->icbs))) + if (!vkd3d_array_reserve((void **)&instructions->opaque_params, &instructions->opaque_param_capacity, + instructions->opaque_param_count + 1, sizeof(*instructions->opaque_params))) return false; - instructions->icbs[instructions->icb_count++] = icb; + instructions->opaque_params[instructions->opaque_param_count++] = param; return true; } @@ -2140,9 +2139,9 @@ void shader_instruction_array_destroy(struct vkd3d_shader_instruction_array *ins vkd3d_free(instructions->elements); shader_param_allocator_destroy(&instructions->dst_params); shader_param_allocator_destroy(&instructions->src_params); - for (i = 0; i < instructions->icb_count; ++i) - vkd3d_free(instructions->icbs[i]); - vkd3d_free(instructions->icbs); + for (i = 0; i < instructions->opaque_param_count; ++i) + vkd3d_free(instructions->opaque_params[i]); + vkd3d_free(instructions->opaque_params); } void vkd3d_shader_build_varying_map(const struct vkd3d_shader_signature *output_signature, diff --git a/libs/vkd3d-shader/vkd3d_shader_private.h b/libs/vkd3d-shader/vkd3d_shader_private.h index 6d0acc85d..39401e0e7 100644 --- a/libs/vkd3d-shader/vkd3d_shader_private.h +++ b/libs/vkd3d-shader/vkd3d_shader_private.h @@ -1223,15 +1223,14 @@ struct vkd3d_shader_instruction_array struct vkd3d_shader_param_allocator src_params; struct vkd3d_shader_param_allocator dst_params; - struct vkd3d_shader_immediate_constant_buffer **icbs; - size_t icb_capacity; - size_t icb_count; + void **opaque_params; + size_t opaque_param_capacity; + size_t opaque_param_count; }; bool shader_instruction_array_init(struct vkd3d_shader_instruction_array *instructions, unsigned int reserve); bool shader_instruction_array_reserve(struct vkd3d_shader_instruction_array *instructions, unsigned int reserve); -bool shader_instruction_array_add_icb(struct vkd3d_shader_instruction_array *instructions, - struct vkd3d_shader_immediate_constant_buffer *icb); +bool shader_instruction_array_add_opaque_param(struct vkd3d_shader_instruction_array *instructions, void *param); bool shader_instruction_array_clone_instruction(struct vkd3d_shader_instruction_array *instructions, unsigned int dst, unsigned int src); void shader_instruction_array_destroy(struct vkd3d_shader_instruction_array *instructions); -- GitLab https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/491
From: Conor McCarthy <cmccarthy(a)codeweavers.com> --- libs/vkd3d-shader/d3d_asm.c | 16 +++ libs/vkd3d-shader/dxil.c | 156 +++++++++++++++++++++++ libs/vkd3d-shader/vkd3d_shader_private.h | 15 +++ 3 files changed, 187 insertions(+) diff --git a/libs/vkd3d-shader/d3d_asm.c b/libs/vkd3d-shader/d3d_asm.c index 274c3552f..385949e12 100644 --- a/libs/vkd3d-shader/d3d_asm.c +++ b/libs/vkd3d-shader/d3d_asm.c @@ -280,6 +280,7 @@ 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", @@ -1889,6 +1890,21 @@ static void shader_dump_instruction(struct vkd3d_d3d_asm_compiler *compiler, } break; + case VKD3DSIH_SWITCH_MONOLITHIC: + { + const struct vkd3d_shader_switch *switch_ = &ins->declaration.switch_; + vkd3d_string_buffer_printf(buffer, " "); + shader_dump_src_param(compiler, ins->src); + for (i = 0; i < switch_->case_count; ++i) + { + vkd3d_string_buffer_printf(buffer, ", %"PRIu64": %sl%u%s", switch_->cases[i].value, + compiler->colours.reg, switch_->cases[i].block_id, compiler->colours.reset); + } + vkd3d_string_buffer_printf(buffer, ", default: %sl%u%s", compiler->colours.reg, switch_->default_id, + compiler->colours.reset); + break; + } + default: shader_dump_instruction_flags(compiler, ins); diff --git a/libs/vkd3d-shader/dxil.c b/libs/vkd3d-shader/dxil.c index 8861385b9..783c355f1 100644 --- a/libs/vkd3d-shader/dxil.c +++ b/libs/vkd3d-shader/dxil.c @@ -466,15 +466,25 @@ enum sm6_block_terminator_type { TERMINATOR_UNCOND_BR, TERMINATOR_COND_BR, + TERMINATOR_SWITCH, TERMINATOR_RET, }; +struct terminator_case +{ + const struct sm6_block *block; + uint64_t value; + bool is_default; +}; + struct sm6_block_terminator { struct vkd3d_shader_register conditional_reg; enum sm6_block_terminator_type type; const struct sm6_block *true_block; const struct sm6_block *false_block; + struct terminator_case *cases; + unsigned int case_count; }; struct sm6_block @@ -1956,6 +1966,13 @@ static inline unsigned int sm6_value_get_constant_uint(const struct sm6_value *v return register_get_uint_value(&value->u.reg); } +static uint64_t sm6_value_get_constant_uint64(const struct sm6_value *value) +{ + if (!sm6_value_is_constant(value)) + return UINT64_MAX; + return register_get_uint64_value(&value->u.reg); +} + static unsigned int sm6_parser_alloc_ssa_id(struct sm6_parser *sm6) { return sm6->ssa_next_id++; @@ -4251,6 +4268,89 @@ static void sm6_parser_emit_ret(struct sm6_parser *sm6, const struct dxil_record ins->handler_idx = VKD3DSIH_NOP; } +static void sm6_parser_emit_switch(struct sm6_parser *sm6, const struct dxil_record *record, + struct sm6_function *function, struct sm6_block *code_block, struct vkd3d_shader_instruction *ins) +{ + struct sm6_block_terminator *terminator = &code_block->terminator; + const struct sm6_type *type; + const struct sm6_value *src; + unsigned int i = 1, j; + + if (record->operand_count < 3 || !(record->operand_count & 1)) + { + WARN("Invalid operand count %u.\n", record->operand_count); + vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_INVALID_OPERAND_COUNT, + "Invalid operand count %u for a switch instruction.", record->operand_count); + return; + } + + if (!(type = sm6_parser_get_type(sm6, record->operands[0]))) + return; + + if (!(src = sm6_parser_get_value_by_ref(sm6, record, type, &i)) + || !sm6_value_validate_is_register(src, sm6)) + return; + assert(i == 2); + + if (src->type != type) + { + WARN("Type mismatch.\n"); + vkd3d_shader_parser_warning(&sm6->p, VKD3D_SHADER_WARNING_DXIL_TYPE_MISMATCH, + "The type of a switch selector value does not match the selector type."); + } + if (!sm6_type_is_integer(type)) + { + WARN("Selector is not scalar integer.\n"); + vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_INVALID_OPERAND, + "Selector type class %u of a switch instruction is not scalar integer.", type->class); + return; + } + + terminator->conditional_reg = src->u.reg; + terminator->type = TERMINATOR_SWITCH; + + terminator->case_count = record->operand_count / 2u; + if (!(terminator->cases = vkd3d_calloc(terminator->case_count, sizeof(*terminator->cases)))) + { + ERR("Failed to allocate case array.\n"); + vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_OUT_OF_MEMORY, + "Out of memory allocating a switch case array."); + return; + } + + /* Executes 'operand_count / 2' times because operand_count is uneven. */ + for (; i < record->operand_count; i += 2) + { + j = i / 2u - 1; + terminator->cases[j].block = sm6_function_get_block(function, record->operands[i], sm6); + /* For structurisation it is convenient to store the default in the case array. */ + terminator->cases[j].is_default = !j; + } + + for (i = 3; i < record->operand_count; i += 2) + { + if (!(src = sm6_parser_get_value_safe(sm6, record->operands[i]))) + return; + + if (src->type != type) + { + WARN("Type mismatch.\n"); + vkd3d_shader_parser_warning(&sm6->p, VKD3D_SHADER_WARNING_DXIL_TYPE_MISMATCH, + "The type of a switch case value does not match the selector type."); + } + if (!sm6_value_is_constant(src)) + { + WARN("Case value is not a constant.\n"); + vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_INVALID_OPERAND, + "A switch case value is not a constant."); + } + + terminator->cases[i / 2u].value = sm6_value_get_constant_uint64(src); + } + + ins->handler_idx = VKD3DSIH_NOP; +} + static void sm6_parser_emit_vselect(struct sm6_parser *sm6, const struct dxil_record *record, struct vkd3d_shader_instruction *ins, struct sm6_value *dst) { @@ -4524,6 +4624,10 @@ static enum vkd3d_result sm6_parser_function_init(struct sm6_parser *sm6, const is_terminator = true; ret_found = true; break; + case FUNC_CODE_INST_SWITCH: + sm6_parser_emit_switch(sm6, record, function, code_block, ins); + is_terminator = true; + break; case FUNC_CODE_INST_VSELECT: sm6_parser_emit_vselect(sm6, record, ins, dst); break; @@ -4562,6 +4666,7 @@ static void sm6_block_emit_terminator(const struct sm6_block *block, struct sm6_ { struct vkd3d_shader_src_param *src_param; struct vkd3d_shader_instruction *ins; + unsigned int i, count; switch (block->terminator.type) { @@ -4587,6 +4692,56 @@ static void sm6_block_emit_terminator(const struct sm6_block *block, struct sm6_ break; } + case TERMINATOR_SWITCH: + { + struct vkd3d_shader_switch_case *cases; + + ins = sm6_parser_add_instruction(sm6, VKD3DSIH_SWITCH_MONOLITHIC); + src_param = instruction_src_params_alloc(ins, 1, sm6); + src_param_init(src_param); + src_param->reg = block->terminator.conditional_reg; + + if (!(cases = vkd3d_calloc(block->terminator.case_count - 1, sizeof(*cases)))) + { + ERR("Failed to allocate case array.\n"); + vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_OUT_OF_MEMORY, + "Out of memory allocating a switch case array."); + return; + } + + for (i = 0, count = 0; i < block->terminator.case_count; ++i) + { + const struct terminator_case *switch_case; + const struct sm6_block *case_block; + + switch_case = &block->terminator.cases[i]; + if (!(case_block = switch_case->block)) + { + assert(sm6->p.failed); + continue; + } + if (switch_case->is_default) + { + ins->declaration.switch_.default_id = case_block->id; + continue; + } + cases[count].value = switch_case->value; + cases[count++].block_id = case_block->id; + } + assert(ins->declaration.switch_.default_id); + + ins->declaration.switch_.cases = cases; + ins->declaration.switch_.case_count = count; + if (!shader_instruction_array_add_opaque_param(&sm6->p.instructions, cases)) + { + ERR("Failed to store case array.\n"); + vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_OUT_OF_MEMORY, + "Out of memory storing a switch case array."); + vkd3d_free(cases); + } + break; + } + case TERMINATOR_RET: { sm6_parser_add_instruction(sm6, VKD3DSIH_RET); @@ -5807,6 +5962,7 @@ static void sm6_block_destroy(struct sm6_block *block) for (i = 0; i < block->phi_count; ++i) sm6_phi_destroy(&block->phi[i]); vkd3d_free(block->phi); + vkd3d_free(block->terminator.cases); vkd3d_free(block); } diff --git a/libs/vkd3d-shader/vkd3d_shader_private.h b/libs/vkd3d-shader/vkd3d_shader_private.h index 39401e0e7..add30331e 100644 --- a/libs/vkd3d-shader/vkd3d_shader_private.h +++ b/libs/vkd3d-shader/vkd3d_shader_private.h @@ -471,6 +471,7 @@ enum vkd3d_shader_opcode VKD3DSIH_SUB, VKD3DSIH_SWAPC, VKD3DSIH_SWITCH, + VKD3DSIH_SWITCH_MONOLITHIC, VKD3DSIH_SYNC, VKD3DSIH_TEX, VKD3DSIH_TEXBEM, @@ -1111,6 +1112,19 @@ struct vkd3d_shader_branch unsigned int false_id; }; +struct vkd3d_shader_switch_case +{ + uint64_t value; + unsigned int block_id; +}; + +struct vkd3d_shader_switch +{ + const struct vkd3d_shader_switch_case *cases; + unsigned int case_count; + unsigned int default_id; +}; + struct vkd3d_shader_instruction { struct vkd3d_shader_location location; @@ -1151,6 +1165,7 @@ struct vkd3d_shader_instruction struct vkd3d_shader_indexable_temp indexable_temp; struct vkd3d_shader_function_table_pointer fp; struct vkd3d_shader_branch branch; + struct vkd3d_shader_switch switch_; } declaration; }; -- GitLab https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/491
I made the above change because using `block->record_count` as a minimum instruction count allocates too much memory if there are multiple code blocks. -- https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/491#note_54529
I just noticed that in some cases FXC generates PHI nodes which mention the same origin block more than once. For example, ```hlsl struct PSInput { float4 color : COLOR; }; float4 main(PSInput input) : SV_TARGET { float y = 0; switch (input.color.x) { case 0: y = 100; break; case 1: y = 101; break; case 2: break; case 3: break; case 4: y = 104; break; case 5: y = 105; break; default: y = 200; break; } input.color.y = y; return input.color; } ``` compiles to ``` define void @main() { %1 = call float @dx.op.loadInput.f32(i32 4, i32 0, i32 0, i8 0, i32 undef) ; LoadInput(inputSigId,rowIndex,colIndex,gsVertexAxis) %2 = call float @dx.op.loadInput.f32(i32 4, i32 0, i32 0, i8 2, i32 undef) ; LoadInput(inputSigId,rowIndex,colIndex,gsVertexAxis) %3 = call float @dx.op.loadInput.f32(i32 4, i32 0, i32 0, i8 3, i32 undef) ; LoadInput(inputSigId,rowIndex,colIndex,gsVertexAxis) %4 = fptosi float %1 to i32 switch i32 %4, label %9 [ i32 0, label %5 i32 1, label %6 i32 2, label %10 i32 3, label %10 i32 4, label %7 i32 5, label %8 ] ; <label>:5 ; preds = %0 br label %10 ; <label>:6 ; preds = %0 br label %10 ; <label>:7 ; preds = %0 br label %10 ; <label>:8 ; preds = %0 br label %10 ; <label>:9 ; preds = %0 br label %10 ; <label>:10 ; preds = %9, %8, %7, %6, %5, %0, %0 %11 = phi float [ 2.000000e+02, %9 ], [ 1.050000e+02, %8 ], [ 1.040000e+02, %7 ], [ 0.000000e+00, %0 ], [ 0.000000e+00, %0 ], [ 1.010000e+02, %6 ], [ 1.000000e+02, %5 ] call void @dx.op.storeOutput.f32(i32 5, i32 0, i32 0, i8 0, float %1) ; StoreOutput(outputSigId,rowIndex,colIndex,value) call void @dx.op.storeOutput.f32(i32 5, i32 0, i32 0, i8 1, float %11) ; StoreOutput(outputSigId,rowIndex,colIndex,value) call void @dx.op.storeOutput.f32(i32 5, i32 0, i32 0, i8 2, float %2) ; StoreOutput(outputSigId,rowIndex,colIndex,value) call void @dx.op.storeOutput.f32(i32 5, i32 0, i32 0, i8 3, float %3) ; StoreOutput(outputSigId,rowIndex,colIndex,value) ret void } ``` As you can see, `%0` is duplicated both in the preds list and in the PHI node. This feels quite broken and it's understandably forbidden by SPIR-V. Could you please add my (or a similar) source code to a shader runner test and implement a deduplication logic, emitting an error if values of duplicated PHI entries do not match? -- https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/491#note_56665
I noticed the same thing early on in development, but haven't dealt with it since I ported the proton structuriser. I'll add deduplication. -- https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/491#note_56673
participants (3)
-
Conor McCarthy -
Conor McCarthy (＠cmccarthy)
-
Giovanni Mascellani (＠giomasce)