March 2024 - Wine-commits - list.winehq.org

Giovanni Mascellani : vkd3d-shader/ir: Add jumps to the structured programs.
by Alexandre Julliard 14 Mar '24

14 Mar '24

Module: vkd3d Branch: master Commit: 2b3e21caeb142b8c0d6faac7d8426443243e1ad8 URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/2b3e21caeb142b8c0d6faac7d8426… Author: Giovanni Mascellani <gmascellani(a)codeweavers.com> Date: Tue Feb 6 15:08:49 2024 +0100 vkd3d-shader/ir: Add jumps to the structured programs. --- libs/vkd3d-shader/ir.c | 168 ++++++++++++++++++++++++++++++++++++++++++++++++- 1 file changed, 166 insertions(+), 2 deletions(-) diff --git a/libs/vkd3d-shader/ir.c b/libs/vkd3d-shader/ir.c index 034d56f3..8c74f2b0 100644 --- a/libs/vkd3d-shader/ir.c +++ b/libs/vkd3d-shader/ir.c @@ -3157,6 +3157,11 @@ struct vsir_cfg_structure STRUCTURE_TYPE_BLOCK, /* Execute a loop, which is identified by an index. */ STRUCTURE_TYPE_LOOP, + /* Execute a `return' or a (possibly) multilevel `break' or + * `continue', targeting a loop by its index. If `condition' + * is non-NULL, then the jump is conditional (this is + * currently not allowed for `return'). */ + STRUCTURE_TYPE_JUMP, } type; union { @@ -3166,6 +3171,21 @@ struct vsir_cfg_structure struct vsir_cfg_structure_list body; unsigned idx; } loop; + struct + { + enum vsir_cfg_jump_type + { + /* NONE is available as an intermediate value, but it + * is not allowed in valid structured programs. */ + JUMP_NONE, + JUMP_BREAK, + JUMP_CONTINUE, + JUMP_RET, + } type; + unsigned int target; + struct vkd3d_shader_src_param *condition; + bool invert_condition; + } jump; } u; }; @@ -3987,6 +4007,70 @@ static enum vkd3d_result vsir_cfg_generate_synthetic_loop_intervals(struct vsir_ return VKD3D_OK; } +struct vsir_cfg_edge_action +{ + enum vsir_cfg_jump_type jump_type; + unsigned int target; + struct vsir_block *successor; +}; + +static void vsir_cfg_compute_edge_action(struct vsir_cfg *cfg, struct vsir_block *block, + struct vsir_block *successor, struct vsir_cfg_edge_action *action) +{ + unsigned int i; + + action->target = UINT_MAX; + action->successor = successor; + + if (successor->order_pos <= block->order_pos) + { + /* The successor is before the current block, so we have to + * use `continue'. The target loop is the innermost that + * contains the current block and has the successor as + * `continue' target. */ + for (i = 0; i < cfg->loop_interval_count; ++i) + { + struct cfg_loop_interval *interval = &cfg->loop_intervals[i]; + + if (interval->begin == successor->order_pos && block->order_pos < interval->end) + action->target = i; + + if (interval->begin > successor->order_pos) + break; + } + + assert(action->target != UINT_MAX); + action->jump_type = JUMP_CONTINUE; + } + else + { + /* The successor is after the current block, so we have to use + * `break', or possibly just jump to the following block. The + * target loop is the outermost that contains the current + * block and has the successor as `break' target. */ + for (i = 0; i < cfg->loop_interval_count; ++i) + { + struct cfg_loop_interval *interval = &cfg->loop_intervals[i]; + + if (interval->begin <= block->order_pos && interval->end == successor->order_pos) + { + action->target = i; + break; + } + } + + if (action->target == UINT_MAX) + { + assert(successor->order_pos == block->order_pos + 1); + action->jump_type = JUMP_NONE; + } + else + { + action->jump_type = JUMP_BREAK; + } + } +} + static enum vkd3d_result vsir_cfg_build_structured_program(struct vsir_cfg *cfg) { unsigned int i, stack_depth = 1, open_interval_idx = 0; @@ -4001,6 +4085,7 @@ static enum vkd3d_result vsir_cfg_build_structured_program(struct vsir_cfg *cfg) for (i = 0; i < cfg->order.count; ++i) { + struct vsir_block *block = cfg->order.blocks[i]; struct vsir_cfg_structure *structure; assert(stack_depth > 0); @@ -4024,9 +4109,88 @@ static enum vkd3d_result vsir_cfg_build_structured_program(struct vsir_cfg *cfg) /* Execute the block. */ if (!(structure = vsir_cfg_structure_list_append(stack[stack_depth - 1], STRUCTURE_TYPE_BLOCK))) goto fail; - structure->u.block = cfg->order.blocks[i]; + structure->u.block = block; - /* TODO: process the jump after the block. */ + /* Generate between zero and two jump instructions. */ + switch (block->end->handler_idx) + { + case VKD3DSIH_BRANCH: + { + struct vsir_cfg_edge_action action_true, action_false; + bool invert_condition = false; + + if (vsir_register_is_label(&block->end->src[0].reg)) + { + unsigned int target = label_from_src_param(&block->end->src[0]); + struct vsir_block *successor = &cfg->blocks[target - 1]; + + vsir_cfg_compute_edge_action(cfg, block, successor, &action_true); + action_false = action_true; + } + else + { + unsigned int target = label_from_src_param(&block->end->src[1]); + struct vsir_block *successor = &cfg->blocks[target - 1]; + + vsir_cfg_compute_edge_action(cfg, block, successor, &action_true); + + target = label_from_src_param(&block->end->src[2]); + successor = &cfg->blocks[target - 1]; + + vsir_cfg_compute_edge_action(cfg, block, successor, &action_false); + } + + /* This will happen if the branch is unconditional, + * but also if it's conditional with the same target + * in both branches, which can happen in some corner + * cases, e.g. when converting switch instructions to + * selection ladders. */ + if (action_true.successor == action_false.successor) + { + assert(action_true.jump_type == action_false.jump_type); + } + else + { + /* At most one branch can just fall through to the + * next block, in which case we make sure it's the + * false branch. */ + if (action_true.jump_type == JUMP_NONE) + { + struct vsir_cfg_edge_action tmp = action_true; + action_true = action_false; + action_false = tmp; + invert_condition = true; + } + + assert(action_true.jump_type != JUMP_NONE); + + if (!(structure = vsir_cfg_structure_list_append(stack[stack_depth - 1], STRUCTURE_TYPE_JUMP))) + goto fail; + structure->u.jump.type = action_true.jump_type; + structure->u.jump.target = action_true.target; + structure->u.jump.condition = &block->end->src[0]; + structure->u.jump.invert_condition = invert_condition; + } + + if (action_false.jump_type != JUMP_NONE) + { + if (!(structure = vsir_cfg_structure_list_append(stack[stack_depth - 1], STRUCTURE_TYPE_JUMP))) + goto fail; + structure->u.jump.type = action_false.jump_type; + structure->u.jump.target = action_false.target; + } + break; + } + + case VKD3DSIH_RET: + if (!(structure = vsir_cfg_structure_list_append(stack[stack_depth - 1], STRUCTURE_TYPE_JUMP))) + goto fail; + structure->u.jump.type = JUMP_RET; + break; + + default: + vkd3d_unreachable(); + } /* Close loop intervals. */ while (stack_depth > 0)

1 0

Conor McCarthy : vkd3d-shader/spirv: Support zero-initialisation for workgroup memory.
by Alexandre Julliard 14 Mar '24

14 Mar '24

Module: vkd3d Branch: master Commit: 6dd54eeb095a937f5b75020d8fb68251748fdff3 URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/6dd54eeb095a937f5b75020d8fb68… Author: Conor McCarthy <cmccarthy(a)codeweavers.com> Date: Fri Dec 1 12:54:08 2023 +1000 vkd3d-shader/spirv: Support zero-initialisation for workgroup memory. --- libs/vkd3d-shader/spirv.c | 11 ++++++----- libs/vkd3d-shader/tpf.c | 2 ++ libs/vkd3d-shader/vkd3d_shader_private.h | 2 ++ 3 files changed, 10 insertions(+), 5 deletions(-) diff --git a/libs/vkd3d-shader/spirv.c b/libs/vkd3d-shader/spirv.c index bd0bfb8e..50fca5bf 100644 --- a/libs/vkd3d-shader/spirv.c +++ b/libs/vkd3d-shader/spirv.c @@ -6435,9 +6435,9 @@ static void spirv_compiler_emit_resource_declaration(struct spirv_compiler *comp } static void spirv_compiler_emit_workgroup_memory(struct spirv_compiler *compiler, - const struct vkd3d_shader_register *reg, unsigned int size, unsigned int structure_stride) + const struct vkd3d_shader_register *reg, unsigned int size, unsigned int structure_stride, bool zero_init) { - uint32_t type_id, array_type_id, length_id, pointer_type_id, var_id; + uint32_t type_id, array_type_id, length_id, pointer_type_id, var_id, init_id; struct vkd3d_spirv_builder *builder = &compiler->spirv_builder; const SpvStorageClass storage_class = SpvStorageClassWorkgroup; struct vkd3d_symbol reg_symbol; @@ -6447,8 +6447,9 @@ static void spirv_compiler_emit_workgroup_memory(struct spirv_compiler *compiler array_type_id = vkd3d_spirv_get_op_type_array(builder, type_id, length_id); pointer_type_id = vkd3d_spirv_get_op_type_pointer(builder, storage_class, array_type_id); + init_id = zero_init ? vkd3d_spirv_get_op_constant_null(builder, array_type_id) : 0; var_id = vkd3d_spirv_build_op_variable(builder, &builder->global_stream, - pointer_type_id, storage_class, 0); + pointer_type_id, storage_class, init_id); spirv_compiler_emit_register_debug_name(builder, var_id, reg); @@ -6464,7 +6465,7 @@ static void spirv_compiler_emit_dcl_tgsm_raw(struct spirv_compiler *compiler, { const struct vkd3d_shader_tgsm_raw *tgsm_raw = &instruction->declaration.tgsm_raw; spirv_compiler_emit_workgroup_memory(compiler, &tgsm_raw->reg.reg, - tgsm_raw->byte_count / 4, 0); + tgsm_raw->byte_count / 4, 0, tgsm_raw->zero_init); } static void spirv_compiler_emit_dcl_tgsm_structured(struct spirv_compiler *compiler, @@ -6473,7 +6474,7 @@ static void spirv_compiler_emit_dcl_tgsm_structured(struct spirv_compiler *compi const struct vkd3d_shader_tgsm_structured *tgsm_structured = &instruction->declaration.tgsm_structured; unsigned int stride = tgsm_structured->byte_stride / 4; spirv_compiler_emit_workgroup_memory(compiler, &tgsm_structured->reg.reg, - tgsm_structured->structure_count * stride, stride); + tgsm_structured->structure_count * stride, stride, tgsm_structured->zero_init); } static void spirv_compiler_emit_dcl_input(struct spirv_compiler *compiler, diff --git a/libs/vkd3d-shader/tpf.c b/libs/vkd3d-shader/tpf.c index bd558693..5c6c8e42 100644 --- a/libs/vkd3d-shader/tpf.c +++ b/libs/vkd3d-shader/tpf.c @@ -1263,6 +1263,7 @@ static void shader_sm5_read_dcl_tgsm_raw(struct vkd3d_shader_instruction *ins, u ins->declaration.tgsm_raw.byte_count = *tokens; if (ins->declaration.tgsm_raw.byte_count % 4) FIXME("Byte count %u is not multiple of 4.\n", ins->declaration.tgsm_raw.byte_count); + ins->declaration.tgsm_raw.zero_init = false; } static void shader_sm5_read_dcl_tgsm_structured(struct vkd3d_shader_instruction *ins, uint32_t opcode, @@ -1274,6 +1275,7 @@ static void shader_sm5_read_dcl_tgsm_structured(struct vkd3d_shader_instruction ins->declaration.tgsm_structured.structure_count = *tokens; if (ins->declaration.tgsm_structured.byte_stride % 4) FIXME("Byte stride %u is not multiple of 4.\n", ins->declaration.tgsm_structured.byte_stride); + ins->declaration.tgsm_structured.zero_init = false; } static void shader_sm5_read_dcl_resource_structured(struct vkd3d_shader_instruction *ins, uint32_t opcode, diff --git a/libs/vkd3d-shader/vkd3d_shader_private.h b/libs/vkd3d-shader/vkd3d_shader_private.h index ba018070..c11f66e3 100644 --- a/libs/vkd3d-shader/vkd3d_shader_private.h +++ b/libs/vkd3d-shader/vkd3d_shader_private.h @@ -1089,6 +1089,7 @@ struct vkd3d_shader_tgsm_raw { struct vkd3d_shader_dst_param reg; unsigned int byte_count; + bool zero_init; }; struct vkd3d_shader_tgsm_structured @@ -1096,6 +1097,7 @@ struct vkd3d_shader_tgsm_structured struct vkd3d_shader_dst_param reg; unsigned int byte_stride; unsigned int structure_count; + bool zero_init; }; struct vkd3d_shader_thread_group_size

1 0

Conor McCarthy : vkd3d-shader/dxil: Implement raw groupshared address space global variables.
by Alexandre Julliard 14 Mar '24

14 Mar '24

Module: vkd3d Branch: master Commit: 0965a3608a18256048f31883cd42780a775c57f0 URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/0965a3608a18256048f31883cd427… Author: Conor McCarthy <cmccarthy(a)codeweavers.com> Date: Fri Mar 1 15:55:38 2024 +1000 vkd3d-shader/dxil: Implement raw groupshared address space global variables. --- libs/vkd3d-shader/dxil.c | 125 +++++++++++++++++++++++++++---- libs/vkd3d-shader/vkd3d_shader_private.h | 1 + 2 files changed, 111 insertions(+), 15 deletions(-) diff --git a/libs/vkd3d-shader/dxil.c b/libs/vkd3d-shader/dxil.c index de51588b..905c2632 100644 --- a/libs/vkd3d-shader/dxil.c +++ b/libs/vkd3d-shader/dxil.c @@ -755,6 +755,7 @@ struct sm6_parser unsigned int indexable_temp_count; unsigned int icb_count; + unsigned int tgsm_count; struct sm6_value *values; size_t value_count; @@ -2267,6 +2268,12 @@ static void register_init_ssa_scalar(struct vkd3d_shader_register *reg, const st register_init_ssa_vector(reg, sm6_type_get_scalar_type(type, 0), 1, value, sm6); } +static void register_make_constant_uint(struct vkd3d_shader_register *reg, unsigned int value) +{ + vsir_register_init(reg, VKD3DSPR_IMMCONST, VKD3D_DATA_UINT, 0); + reg->u.immconst_u32[0] = value; +} + static void dst_param_init(struct vkd3d_shader_dst_param *param) { param->write_mask = VKD3DSP_WRITEMASK_0; @@ -2330,6 +2337,12 @@ static void src_param_init_vector_from_reg(struct vkd3d_shader_src_param *param, param->reg = *reg; } +static void src_param_make_constant_uint(struct vkd3d_shader_src_param *param, unsigned int value) +{ + src_param_init(param); + register_make_constant_uint(&param->reg, value); +} + static void register_index_address_init(struct vkd3d_shader_register_index *idx, const struct sm6_value *address, struct sm6_parser *sm6) { @@ -3009,6 +3022,30 @@ static void sm6_parser_declare_indexable_temp(struct sm6_parser *sm6, const stru register_init_with_id(&dst->u.reg, VKD3DSPR_IDXTEMP, data_type, ins->declaration.indexable_temp.register_idx); } +static void sm6_parser_declare_tgsm_raw(struct sm6_parser *sm6, const struct sm6_type *elem_type, + unsigned int alignment, unsigned int init, struct sm6_value *dst) +{ + enum vkd3d_data_type data_type = vkd3d_data_type_from_sm6_type(elem_type); + struct vkd3d_shader_instruction *ins; + unsigned int byte_count; + + ins = sm6_parser_add_instruction(sm6, VKD3DSIH_DCL_TGSM_RAW); + dst_param_init(&ins->declaration.tgsm_raw.reg); + register_init_with_id(&ins->declaration.tgsm_raw.reg.reg, VKD3DSPR_GROUPSHAREDMEM, data_type, sm6->tgsm_count++); + dst->u.reg = ins->declaration.tgsm_raw.reg.reg; + ins->declaration.tgsm_raw.alignment = alignment; + byte_count = elem_type->u.width / 8u; + if (byte_count != 4) + { + FIXME("Unsupported byte count %u.\n", byte_count); + vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_INVALID_OPERAND, + "Raw TGSM byte count %u is not supported.", byte_count); + } + ins->declaration.tgsm_raw.byte_count = byte_count; + /* The initialiser value index will be resolved later when forward references can be handled. */ + ins->flags = init; +} + static bool sm6_parser_declare_global(struct sm6_parser *sm6, const struct dxil_record *record) { const struct sm6_type *type, *scalar_type; @@ -3134,10 +3171,22 @@ static bool sm6_parser_declare_global(struct sm6_parser *sm6, const struct dxil_ } else if (address_space == ADDRESS_SPACE_GROUPSHARED) { - FIXME("Unsupported TGSM.\n"); - vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_INVALID_OPERAND, - "TGSM global variables are not supported."); - return false; + if (!sm6_type_is_numeric(scalar_type)) + { + WARN("Unsupported type class %u.\n", scalar_type->class); + vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_INVALID_OPERAND, + "TGSM variables of type class %u are not supported.", scalar_type->class); + return false; + } + if (count != 1) + { + FIXME("Unsupported structured TGSM.\n"); + vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_INVALID_OPERAND, + "Structured TGSM global variables are not supported."); + return false; + } + + sm6_parser_declare_tgsm_raw(sm6, scalar_type, alignment, init, dst); } else { @@ -3173,6 +3222,38 @@ static const struct vkd3d_shader_immediate_constant_buffer *resolve_forward_init return NULL; } +static bool resolve_forward_zero_initialiser(size_t index, struct sm6_parser *sm6) +{ + const struct sm6_value *value; + + if (!index) + return false; + + --index; + if (!(value = sm6_parser_get_value_safe(sm6, index)) + || (!sm6_value_is_icb(value) && !sm6_value_is_constant(value) && !sm6_value_is_undef(value))) + { + WARN("Invalid initialiser index %zu.\n", index); + vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_INVALID_OPERAND, + "TGSM initialiser value index %zu is invalid.", index); + return false; + } + else if ((sm6_value_is_icb(value) && value->u.icb->is_null) || sm6_value_is_constant_zero(value)) + { + return true; + } + else if (sm6_value_is_undef(value)) + { + /* In VSIR, initialisation with undefined values of objects is implied, not explicit. */ + return false; + } + + FIXME("Non-zero initialisers are not supported.\n"); + vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_INVALID_OPERAND, + "Non-zero TGSM initialisers are not supported."); + return false; +} + static enum vkd3d_result sm6_parser_globals_init(struct sm6_parser *sm6) { size_t i, count, base_value_idx = sm6->value_count; @@ -3246,6 +3327,11 @@ static enum vkd3d_result sm6_parser_globals_init(struct sm6_parser *sm6) { ins->declaration.icb = resolve_forward_initialiser((uintptr_t)ins->declaration.icb, sm6); } + else if (ins->handler_idx == VKD3DSIH_DCL_TGSM_RAW) + { + ins->declaration.tgsm_raw.zero_init = resolve_forward_zero_initialiser(ins->flags, sm6); + ins->flags = 0; + } } for (i = base_value_idx; i < sm6->value_count; ++i) { @@ -5549,8 +5635,8 @@ static void sm6_parser_emit_load(struct sm6_parser *sm6, const struct dxil_recor struct vkd3d_shader_instruction *ins, struct sm6_value *dst) { const struct sm6_type *elem_type = NULL, *pointee_type; - struct vkd3d_shader_src_param *src_param; - unsigned int alignment, i = 0; + unsigned int alignment, operand_count, i = 0; + struct vkd3d_shader_src_param *src_params; const struct sm6_value *ptr; uint64_t alignment_code; @@ -5587,12 +5673,15 @@ static void sm6_parser_emit_load(struct sm6_parser *sm6, const struct dxil_recor if (record->operands[i]) WARN("Ignoring volatile modifier.\n"); - vsir_instruction_init(ins, &sm6->p.location, VKD3DSIH_MOV); + operand_count = 1 + (ptr->u.reg.type == VKD3DSPR_GROUPSHAREDMEM); + vsir_instruction_init(ins, &sm6->p.location, (operand_count > 1) ? VKD3DSIH_LD_RAW : VKD3DSIH_MOV); - if (!(src_param = instruction_src_params_alloc(ins, 1, sm6))) + if (!(src_params = instruction_src_params_alloc(ins, operand_count, sm6))) return; - src_param_init_from_value(&src_param[0], ptr); - src_param->reg.alignment = alignment; + if (operand_count > 1) + src_param_make_constant_uint(&src_params[0], 0); + src_param_init_from_value(&src_params[operand_count - 1], ptr); + src_params[operand_count - 1].reg.alignment = alignment; instruction_dst_param_init_ssa_scalar(ins, sm6); } @@ -5710,11 +5799,11 @@ static void sm6_parser_emit_ret(struct sm6_parser *sm6, const struct dxil_record static void sm6_parser_emit_store(struct sm6_parser *sm6, const struct dxil_record *record, struct vkd3d_shader_instruction *ins, struct sm6_value *dst) { - struct vkd3d_shader_src_param *src_param; + unsigned int i = 0, alignment, operand_count; + struct vkd3d_shader_src_param *src_params; struct vkd3d_shader_dst_param *dst_param; const struct sm6_type *pointee_type; const struct sm6_value *ptr, *src; - unsigned int i = 0, alignment; uint64_t alignment_code; if (!(ptr = sm6_parser_get_value_by_ref(sm6, record, NULL, &i)) @@ -5747,16 +5836,22 @@ static void sm6_parser_emit_store(struct sm6_parser *sm6, const struct dxil_reco if (record->operands[i]) WARN("Ignoring volatile modifier.\n"); - vsir_instruction_init(ins, &sm6->p.location, VKD3DSIH_MOV); + operand_count = 1 + (ptr->u.reg.type == VKD3DSPR_GROUPSHAREDMEM); + vsir_instruction_init(ins, &sm6->p.location, (operand_count > 1) ? VKD3DSIH_STORE_RAW : VKD3DSIH_MOV); - if (!(src_param = instruction_src_params_alloc(ins, 1, sm6))) + if (!(src_params = instruction_src_params_alloc(ins, operand_count, sm6))) return; - src_param_init_from_value(&src_param[0], src); + if (operand_count > 1) + src_param_make_constant_uint(&src_params[0], 0); + src_param_init_from_value(&src_params[operand_count - 1], src); dst_param = instruction_dst_params_alloc(ins, 1, sm6); dst_param_init(dst_param); dst_param->reg = ptr->u.reg; dst_param->reg.alignment = alignment; + /* Groupshared stores contain the address in the src params. */ + if (dst_param->reg.type != VKD3DSPR_IDXTEMP) + dst_param->reg.idx_count = 1; } static void sm6_parser_emit_switch(struct sm6_parser *sm6, const struct dxil_record *record, diff --git a/libs/vkd3d-shader/vkd3d_shader_private.h b/libs/vkd3d-shader/vkd3d_shader_private.h index c11f66e3..d9a8b539 100644 --- a/libs/vkd3d-shader/vkd3d_shader_private.h +++ b/libs/vkd3d-shader/vkd3d_shader_private.h @@ -1088,6 +1088,7 @@ struct vkd3d_shader_tgsm struct vkd3d_shader_tgsm_raw { struct vkd3d_shader_dst_param reg; + unsigned int alignment; unsigned int byte_count; bool zero_init; };

1 0

Conor McCarthy : vkd3d-shader/dxil: Implement structured groupshared address space global variables.
by Alexandre Julliard 14 Mar '24

14 Mar '24

Module: vkd3d Branch: master Commit: 5e63164d01f2fb0b3cff72cc2a5eb0c21e6e8ddb URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/5e63164d01f2fb0b3cff72cc2a5eb… Author: Conor McCarthy <cmccarthy(a)codeweavers.com> Date: Fri Mar 1 15:56:31 2024 +1000 vkd3d-shader/dxil: Implement structured groupshared address space global variables. --- libs/vkd3d-shader/dxil.c | 115 ++++++++++++++++++++++++------- libs/vkd3d-shader/vkd3d_shader_private.h | 1 + 2 files changed, 92 insertions(+), 24 deletions(-) diff --git a/libs/vkd3d-shader/dxil.c b/libs/vkd3d-shader/dxil.c index 905c2632..09b0b0d9 100644 --- a/libs/vkd3d-shader/dxil.c +++ b/libs/vkd3d-shader/dxil.c @@ -543,6 +543,7 @@ struct sm6_value { const struct sm6_type *type; enum sm6_value_type value_type; + unsigned int structure_stride; bool is_undefined; union { @@ -3033,6 +3034,7 @@ static void sm6_parser_declare_tgsm_raw(struct sm6_parser *sm6, const struct sm6 dst_param_init(&ins->declaration.tgsm_raw.reg); register_init_with_id(&ins->declaration.tgsm_raw.reg.reg, VKD3DSPR_GROUPSHAREDMEM, data_type, sm6->tgsm_count++); dst->u.reg = ins->declaration.tgsm_raw.reg.reg; + dst->structure_stride = 0; ins->declaration.tgsm_raw.alignment = alignment; byte_count = elem_type->u.width / 8u; if (byte_count != 4) @@ -3046,6 +3048,33 @@ static void sm6_parser_declare_tgsm_raw(struct sm6_parser *sm6, const struct sm6 ins->flags = init; } +static void sm6_parser_declare_tgsm_structured(struct sm6_parser *sm6, const struct sm6_type *elem_type, + unsigned int count, unsigned int alignment, unsigned int init, struct sm6_value *dst) +{ + enum vkd3d_data_type data_type = vkd3d_data_type_from_sm6_type(elem_type); + struct vkd3d_shader_instruction *ins; + unsigned int structure_stride; + + ins = sm6_parser_add_instruction(sm6, VKD3DSIH_DCL_TGSM_STRUCTURED); + dst_param_init(&ins->declaration.tgsm_structured.reg); + register_init_with_id(&ins->declaration.tgsm_structured.reg.reg, VKD3DSPR_GROUPSHAREDMEM, + data_type, sm6->tgsm_count++); + dst->u.reg = ins->declaration.tgsm_structured.reg.reg; + structure_stride = elem_type->u.width / 8u; + if (structure_stride != 4) + { + FIXME("Unsupported structure stride %u.\n", structure_stride); + vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_INVALID_OPERAND, + "Structured TGSM byte stride %u is not supported.", structure_stride); + } + dst->structure_stride = structure_stride; + ins->declaration.tgsm_structured.alignment = alignment; + ins->declaration.tgsm_structured.byte_stride = structure_stride; + ins->declaration.tgsm_structured.structure_count = count; + /* The initialiser value index will be resolved later when forward references can be handled. */ + ins->flags = init; +} + static bool sm6_parser_declare_global(struct sm6_parser *sm6, const struct dxil_record *record) { const struct sm6_type *type, *scalar_type; @@ -3178,15 +3207,10 @@ static bool sm6_parser_declare_global(struct sm6_parser *sm6, const struct dxil_ "TGSM variables of type class %u are not supported.", scalar_type->class); return false; } - if (count != 1) - { - FIXME("Unsupported structured TGSM.\n"); - vkd3d_shader_parser_error(&sm6->p, VKD3D_SHADER_ERROR_DXIL_INVALID_OPERAND, - "Structured TGSM global variables are not supported."); - return false; - } - - sm6_parser_declare_tgsm_raw(sm6, scalar_type, alignment, init, dst); + if (count == 1) + sm6_parser_declare_tgsm_raw(sm6, scalar_type, alignment, init, dst); + else + sm6_parser_declare_tgsm_structured(sm6, scalar_type, count, alignment, init, dst); } else { @@ -3332,6 +3356,11 @@ static enum vkd3d_result sm6_parser_globals_init(struct sm6_parser *sm6) ins->declaration.tgsm_raw.zero_init = resolve_forward_zero_initialiser(ins->flags, sm6); ins->flags = 0; } + else if (ins->handler_idx == VKD3DSIH_DCL_TGSM_STRUCTURED) + { + ins->declaration.tgsm_structured.zero_init = resolve_forward_zero_initialiser(ins->flags, sm6); + ins->flags = 0; + } } for (i = base_value_idx; i < sm6->value_count; ++i) { @@ -5627,6 +5656,7 @@ static void sm6_parser_emit_gep(struct sm6_parser *sm6, const struct dxil_record register_index_address_init(&reg->idx[1], elem_value, sm6); reg->idx[1].is_in_bounds = is_in_bounds; reg->idx_count = 2; + dst->structure_stride = src->structure_stride; ins->handler_idx = VKD3DSIH_NOP; } @@ -5673,15 +5703,34 @@ static void sm6_parser_emit_load(struct sm6_parser *sm6, const struct dxil_recor if (record->operands[i]) WARN("Ignoring volatile modifier.\n"); - operand_count = 1 + (ptr->u.reg.type == VKD3DSPR_GROUPSHAREDMEM); - vsir_instruction_init(ins, &sm6->p.location, (operand_count > 1) ? VKD3DSIH_LD_RAW : VKD3DSIH_MOV); + if (ptr->structure_stride) + { + assert(ptr->u.reg.type == VKD3DSPR_GROUPSHAREDMEM); + vsir_instruction_init(ins, &sm6->p.location, VKD3DSIH_LD_STRUCTURED); + + if (!(src_params = instruction_src_params_alloc(ins, 3, sm6))) + return; + if (ptr->u.reg.idx[1].rel_addr) + src_params[0] = *ptr->u.reg.idx[1].rel_addr; + else + src_param_make_constant_uint(&src_params[0], ptr->u.reg.idx[1].offset); + /* Struct offset is always zero as there is no struct, just an array. */ + src_param_make_constant_uint(&src_params[1], 0); + src_param_init_from_value(&src_params[2], ptr); + src_params[2].reg.alignment = alignment; + } + else + { + operand_count = 1 + (ptr->u.reg.type == VKD3DSPR_GROUPSHAREDMEM); + vsir_instruction_init(ins, &sm6->p.location, (operand_count > 1) ? VKD3DSIH_LD_RAW : VKD3DSIH_MOV); - if (!(src_params = instruction_src_params_alloc(ins, operand_count, sm6))) - return; - if (operand_count > 1) - src_param_make_constant_uint(&src_params[0], 0); - src_param_init_from_value(&src_params[operand_count - 1], ptr); - src_params[operand_count - 1].reg.alignment = alignment; + if (!(src_params = instruction_src_params_alloc(ins, operand_count, sm6))) + return; + if (operand_count > 1) + src_param_make_constant_uint(&src_params[0], 0); + src_param_init_from_value(&src_params[operand_count - 1], ptr); + src_params[operand_count - 1].reg.alignment = alignment; + } instruction_dst_param_init_ssa_scalar(ins, sm6); } @@ -5836,14 +5885,32 @@ static void sm6_parser_emit_store(struct sm6_parser *sm6, const struct dxil_reco if (record->operands[i]) WARN("Ignoring volatile modifier.\n"); - operand_count = 1 + (ptr->u.reg.type == VKD3DSPR_GROUPSHAREDMEM); - vsir_instruction_init(ins, &sm6->p.location, (operand_count > 1) ? VKD3DSIH_STORE_RAW : VKD3DSIH_MOV); + if (ptr->structure_stride) + { + assert(ptr->u.reg.type == VKD3DSPR_GROUPSHAREDMEM); + vsir_instruction_init(ins, &sm6->p.location, VKD3DSIH_STORE_STRUCTURED); - if (!(src_params = instruction_src_params_alloc(ins, operand_count, sm6))) - return; - if (operand_count > 1) - src_param_make_constant_uint(&src_params[0], 0); - src_param_init_from_value(&src_params[operand_count - 1], src); + if (!(src_params = instruction_src_params_alloc(ins, 3, sm6))) + return; + if (ptr->u.reg.idx[1].rel_addr) + src_params[0] = *ptr->u.reg.idx[1].rel_addr; + else + src_param_make_constant_uint(&src_params[0], ptr->u.reg.idx[1].offset); + /* Struct offset is always zero as there is no struct, just an array. */ + src_param_make_constant_uint(&src_params[1], 0); + src_param_init_from_value(&src_params[2], src); + } + else + { + operand_count = 1 + (ptr->u.reg.type == VKD3DSPR_GROUPSHAREDMEM); + vsir_instruction_init(ins, &sm6->p.location, (operand_count > 1) ? VKD3DSIH_STORE_RAW : VKD3DSIH_MOV); + + if (!(src_params = instruction_src_params_alloc(ins, operand_count, sm6))) + return; + if (operand_count > 1) + src_param_make_constant_uint(&src_params[0], 0); + src_param_init_from_value(&src_params[operand_count - 1], src); + } dst_param = instruction_dst_params_alloc(ins, 1, sm6); dst_param_init(dst_param); diff --git a/libs/vkd3d-shader/vkd3d_shader_private.h b/libs/vkd3d-shader/vkd3d_shader_private.h index d9a8b539..c540aee9 100644 --- a/libs/vkd3d-shader/vkd3d_shader_private.h +++ b/libs/vkd3d-shader/vkd3d_shader_private.h @@ -1096,6 +1096,7 @@ struct vkd3d_shader_tgsm_raw struct vkd3d_shader_tgsm_structured { struct vkd3d_shader_dst_param reg; + unsigned int alignment; unsigned int byte_stride; unsigned int structure_count; bool zero_init;

1 0

Conor McCarthy : vkd3d-shader/spirv: Emit a trace message if TGSM alignment is ignored.
by Alexandre Julliard 14 Mar '24

14 Mar '24

Module: vkd3d Branch: master Commit: b22632ff1a9361431bfcce41630cd70be3708a5b URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/b22632ff1a9361431bfcce41630cd… Author: Conor McCarthy <cmccarthy(a)codeweavers.com> Date: Fri Mar 1 15:57:14 2024 +1000 vkd3d-shader/spirv: Emit a trace message if TGSM alignment is ignored. This would cause a lot of warning spam if it was a warning. --- libs/vkd3d-shader/spirv.c | 11 ++++++++--- 1 file changed, 8 insertions(+), 3 deletions(-) diff --git a/libs/vkd3d-shader/spirv.c b/libs/vkd3d-shader/spirv.c index 50fca5bf..5403e544 100644 --- a/libs/vkd3d-shader/spirv.c +++ b/libs/vkd3d-shader/spirv.c @@ -6435,13 +6435,18 @@ static void spirv_compiler_emit_resource_declaration(struct spirv_compiler *comp } static void spirv_compiler_emit_workgroup_memory(struct spirv_compiler *compiler, - const struct vkd3d_shader_register *reg, unsigned int size, unsigned int structure_stride, bool zero_init) + const struct vkd3d_shader_register *reg, unsigned int alignment, unsigned int size, + unsigned int structure_stride, bool zero_init) { uint32_t type_id, array_type_id, length_id, pointer_type_id, var_id, init_id; struct vkd3d_spirv_builder *builder = &compiler->spirv_builder; const SpvStorageClass storage_class = SpvStorageClassWorkgroup; struct vkd3d_symbol reg_symbol; + /* Alignment is supported only in the Kernel execution model. */ + if (alignment) + TRACE("Ignoring alignment %u.\n", alignment); + type_id = vkd3d_spirv_get_type_id(builder, VKD3D_SHADER_COMPONENT_UINT, 1); length_id = spirv_compiler_get_constant_uint(compiler, size); array_type_id = vkd3d_spirv_get_op_type_array(builder, type_id, length_id); @@ -6464,7 +6469,7 @@ static void spirv_compiler_emit_dcl_tgsm_raw(struct spirv_compiler *compiler, const struct vkd3d_shader_instruction *instruction) { const struct vkd3d_shader_tgsm_raw *tgsm_raw = &instruction->declaration.tgsm_raw; - spirv_compiler_emit_workgroup_memory(compiler, &tgsm_raw->reg.reg, + spirv_compiler_emit_workgroup_memory(compiler, &tgsm_raw->reg.reg, tgsm_raw->alignment, tgsm_raw->byte_count / 4, 0, tgsm_raw->zero_init); } @@ -6473,7 +6478,7 @@ static void spirv_compiler_emit_dcl_tgsm_structured(struct spirv_compiler *compi { const struct vkd3d_shader_tgsm_structured *tgsm_structured = &instruction->declaration.tgsm_structured; unsigned int stride = tgsm_structured->byte_stride / 4; - spirv_compiler_emit_workgroup_memory(compiler, &tgsm_structured->reg.reg, + spirv_compiler_emit_workgroup_memory(compiler, &tgsm_structured->reg.reg, tgsm_structured->alignment, tgsm_structured->structure_count * stride, stride, tgsm_structured->zero_init); }

1 0

Conor McCarthy : vkd3d-shader/dxil: Implement DX instruction Barrier.
by Alexandre Julliard 14 Mar '24

14 Mar '24

Module: vkd3d Branch: master Commit: f08976be33a605faecaa198f2c3d6ccacd1d1070 URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/f08976be33a605faecaa198f2c3d6… Author: Conor McCarthy <cmccarthy(a)codeweavers.com> Date: Thu Jun 15 11:43:00 2023 +1000 vkd3d-shader/dxil: Implement DX instruction Barrier. --- libs/vkd3d-shader/dxil.c | 31 +++++++++++++++++++++++++++++++ 1 file changed, 31 insertions(+) diff --git a/libs/vkd3d-shader/dxil.c b/libs/vkd3d-shader/dxil.c index 09b0b0d9..b3e4f053 100644 --- a/libs/vkd3d-shader/dxil.c +++ b/libs/vkd3d-shader/dxil.c @@ -393,6 +393,7 @@ enum dx_intrinsic_opcode DX_TEXTURE_GATHER_CMP = 74, DX_ATOMIC_BINOP = 78, DX_ATOMIC_CMP_XCHG = 79, + DX_BARRIER = 80, DX_DERIV_COARSEX = 83, DX_DERIV_COARSEY = 84, DX_DERIV_FINEX = 85, @@ -465,6 +466,14 @@ enum dxil_atomic_binop_code ATOMIC_BINOP_INVALID, }; +enum dxil_sync_flags +{ + SYNC_THREAD_GROUP = 0x1, + SYNC_GLOBAL_UAV = 0x2, + SYNC_THREAD_GROUP_UAV = 0x4, + SYNC_GROUP_SHARED_MEMORY = 0x8, +}; + struct sm6_pointer_info { const struct sm6_type *type; @@ -4104,6 +4113,27 @@ static void sm6_parser_emit_dx_atomic_binop(struct sm6_parser *sm6, enum dx_intr dst->u.reg = dst_params[0].reg; } +static void sm6_parser_emit_dx_barrier(struct sm6_parser *sm6, enum dx_intrinsic_opcode op, + const struct sm6_value **operands, struct function_emission_state *state) +{ + struct vkd3d_shader_instruction *ins = state->ins; + enum dxil_sync_flags flags; + + vsir_instruction_init(ins, &sm6->p.location, VKD3DSIH_SYNC); + flags = sm6_value_get_constant_uint(operands[0]); + ins->flags = flags & (SYNC_THREAD_GROUP | SYNC_THREAD_GROUP_UAV); + if (flags & SYNC_GLOBAL_UAV) + ins->flags |= VKD3DSSF_GLOBAL_UAV; + if (flags & SYNC_GROUP_SHARED_MEMORY) + ins->flags |= VKD3DSSF_GROUP_SHARED_MEMORY; + if (flags &= ~(SYNC_THREAD_GROUP | SYNC_GLOBAL_UAV | SYNC_THREAD_GROUP_UAV | SYNC_GROUP_SHARED_MEMORY)) + { + FIXME("Unhandled flags %#x.\n", flags); + vkd3d_shader_parser_warning(&sm6->p, VKD3D_SHADER_WARNING_DXIL_IGNORING_OPERANDS, + "Barrier flags %#x are unhandled.", flags); + } +} + static void sm6_parser_emit_dx_cbuffer_load(struct sm6_parser *sm6, enum dx_intrinsic_opcode op, const struct sm6_value **operands, struct function_emission_state *state) { @@ -4933,6 +4963,7 @@ static const struct sm6_dx_opcode_info sm6_dx_op_table[] = [DX_ATAN ] = {"g", "R", sm6_parser_emit_dx_unary}, [DX_ATOMIC_BINOP ] = {"o", "HciiiR", sm6_parser_emit_dx_atomic_binop}, [DX_ATOMIC_CMP_XCHG ] = {"o", "HiiiRR", sm6_parser_emit_dx_atomic_binop}, + [DX_BARRIER ] = {"v", "c", sm6_parser_emit_dx_barrier}, [DX_BFREV ] = {"m", "R", sm6_parser_emit_dx_unary}, [DX_BUFFER_LOAD ] = {"o", "Hii", sm6_parser_emit_dx_buffer_load}, [DX_BUFFER_STORE ] = {"v", "Hiiooooc", sm6_parser_emit_dx_buffer_store},

1 0

Paul Gofman : dsdmo: Force debug info in critical sections.
by Alexandre Julliard 14 Mar '24

14 Mar '24

Module: wine Branch: master Commit: c4ab638083cacca88d2223f0cd357fdfb3115abb URL: https://gitlab.winehq.org/wine/wine/-/commit/c4ab638083cacca88d2223f0cd357f… Author: Paul Gofman <pgofman(a)codeweavers.com> Date: Thu Mar 7 16:26:37 2024 -0600 dsdmo: Force debug info in critical sections. --- dlls/dsdmo/main.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/dlls/dsdmo/main.c b/dlls/dsdmo/main.c index 77345b1b5d1..d6f36b53159 100644 --- a/dlls/dsdmo/main.c +++ b/dlls/dsdmo/main.c @@ -588,7 +588,7 @@ static void effect_init(struct effect *effect, IUnknown *outer, const struct eff effect->IMediaParams_iface.lpVtbl = &effect_media_params_vtbl; effect->IMediaParamInfo_iface.lpVtbl = &effect_media_param_info_vtbl; - InitializeCriticalSection(&effect->cs); + InitializeCriticalSectionEx(&effect->cs, 0, RTL_CRITICAL_SECTION_FLAG_FORCE_DEBUG_INFO); effect->cs.DebugInfo->Spare[0] = (DWORD_PTR)(__FILE__ ": effect.cs"); effect->ops = ops;

1 0

Paul Gofman : qasf: Force debug info in critical sections.
by Alexandre Julliard 14 Mar '24

14 Mar '24

Module: wine Branch: master Commit: 90ee219c44d4a3a78051e27d1a6d4520088acb6b URL: https://gitlab.winehq.org/wine/wine/-/commit/90ee219c44d4a3a78051e27d1a6d45… Author: Paul Gofman <pgofman(a)codeweavers.com> Date: Thu Mar 7 16:27:00 2024 -0600 qasf: Force debug info in critical sections. --- dlls/qasf/asfreader.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/dlls/qasf/asfreader.c b/dlls/qasf/asfreader.c index 7a832e9a718..91fdc6fbb34 100644 --- a/dlls/qasf/asfreader.c +++ b/dlls/qasf/asfreader.c @@ -1038,7 +1038,7 @@ HRESULT asf_reader_create(IUnknown *outer, IUnknown **out) strmbase_filter_init(&object->filter, outer, &CLSID_WMAsfReader, &filter_ops); object->IFileSourceFilter_iface.lpVtbl = &file_source_vtbl; - InitializeCriticalSection(&object->status_cs); + InitializeCriticalSectionEx(&object->status_cs, 0, RTL_CRITICAL_SECTION_FLAG_FORCE_DEBUG_INFO); object->status_cs.DebugInfo->Spare[0] = (DWORD_PTR)(__FILE__ ": status_cs"); TRACE("Created WM ASF reader %p.\n", object);

1 0

Paul Gofman : services: Force debug info in critical sections.
by Alexandre Julliard 14 Mar '24

14 Mar '24

Module: wine Branch: master Commit: 95b0e65b079bee3dd30248e127d0d62841e24266 URL: https://gitlab.winehq.org/wine/wine/-/commit/95b0e65b079bee3dd30248e127d0d6… Author: Paul Gofman <pgofman(a)codeweavers.com> Date: Thu Mar 7 16:31:23 2024 -0600 services: Force debug info in critical sections. --- programs/services/services.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/programs/services/services.c b/programs/services/services.c index b45022eaec8..273a146c784 100644 --- a/programs/services/services.c +++ b/programs/services/services.c @@ -667,7 +667,7 @@ static DWORD scmdatabase_create(struct scmdatabase **db) list_init(&(*db)->processes); list_init(&(*db)->services); - InitializeCriticalSection(&(*db)->cs); + InitializeCriticalSectionEx(&(*db)->cs, 0, RTL_CRITICAL_SECTION_FLAG_FORCE_DEBUG_INFO); (*db)->cs.DebugInfo->Spare[0] = (DWORD_PTR)(__FILE__ ": scmdatabase"); err = RegCreateKeyExW(HKEY_LOCAL_MACHINE, SZ_SERVICES_KEY, 0, NULL,

1 0

Paul Gofman : explorer: Force debug info in critical sections.
by Alexandre Julliard 14 Mar '24

14 Mar '24

Module: wine Branch: master Commit: 9115dc0abadfab6c8b16a3ba310d82a138cbfb3a URL: https://gitlab.winehq.org/wine/wine/-/commit/9115dc0abadfab6c8b16a3ba310d82… Author: Paul Gofman <pgofman(a)codeweavers.com> Date: Thu Mar 7 16:32:28 2024 -0600 explorer: Force debug info in critical sections. --- programs/explorer/desktop.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/programs/explorer/desktop.c b/programs/explorer/desktop.c index 4e8f9a84c07..25bba7cb7cb 100644 --- a/programs/explorer/desktop.c +++ b/programs/explorer/desktop.c @@ -2402,7 +2402,7 @@ static void shellwindows_init(void) CoInitialize(NULL); shellwindows.IShellWindows_iface.lpVtbl = &shellwindowsvtbl; - InitializeCriticalSection(&shellwindows.cs); + InitializeCriticalSectionEx(&shellwindows.cs, 0, RTL_CRITICAL_SECTION_FLAG_FORCE_DEBUG_INFO); shellwindows.cs.DebugInfo->Spare[0] = (DWORD_PTR)(__FILE__ ": shellwindows.cs"); hr = CoRegisterClassObject(&CLSID_ShellWindows,

1 0