[PATCH v7 0/8] MR181: vkd3d-shader: Normalise signatures and input/output registers to the Shader Model 6 pattern.
-- v7: vkd3d-shader/ir: Normalise signatures and input/output registers to the Shader Model 6 pattern. vkd3d-shader/tpf: Emit an error if an index range is declared for default control point phase input/output. https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/181
From: Conor McCarthy <cmccarthy(a)codeweavers.com> --- libs/vkd3d-shader/d3dbc.c | 3 ++ libs/vkd3d-shader/ir.c | 31 ++++++------- libs/vkd3d-shader/tpf.c | 55 ++++++++++++++++++++++++ libs/vkd3d-shader/vkd3d_shader_private.h | 2 + 4 files changed, 73 insertions(+), 18 deletions(-) diff --git a/libs/vkd3d-shader/d3dbc.c b/libs/vkd3d-shader/d3dbc.c index 823df6f6..d4f3ef66 100644 --- a/libs/vkd3d-shader/d3dbc.c +++ b/libs/vkd3d-shader/d3dbc.c @@ -465,6 +465,7 @@ static void shader_sm1_parse_src_param(uint32_t param, const struct vkd3d_shader src->reg.idx[1].rel_addr = NULL; src->reg.idx[2].offset = ~0u; src->reg.idx[2].rel_addr = NULL; + src->reg.idx_count = 1; src->swizzle = swizzle_from_sm1((param & VKD3D_SM1_SWIZZLE_MASK) >> VKD3D_SM1_SWIZZLE_SHIFT); src->modifiers = (param & VKD3D_SM1_SRC_MODIFIER_MASK) >> VKD3D_SM1_SRC_MODIFIER_SHIFT; } @@ -483,6 +484,7 @@ static void shader_sm1_parse_dst_param(uint32_t param, const struct vkd3d_shader dst->reg.idx[1].rel_addr = NULL; dst->reg.idx[2].offset = ~0u; dst->reg.idx[2].rel_addr = NULL; + dst->reg.idx_count = 1; dst->write_mask = (param & VKD3D_SM1_WRITEMASK_MASK) >> VKD3D_SM1_WRITEMASK_SHIFT; dst->modifiers = (param & VKD3D_SM1_DST_MODIFIER_MASK) >> VKD3D_SM1_DST_MODIFIER_SHIFT; dst->shift = (param & VKD3D_SM1_DSTSHIFT_MASK) >> VKD3D_SM1_DSTSHIFT_SHIFT; @@ -664,6 +666,7 @@ static void shader_sm1_read_immconst(struct vkd3d_shader_sm1_parser *sm1, const src_param->reg.idx[1].rel_addr = NULL; src_param->reg.idx[2].offset = ~0u; src_param->reg.idx[2].rel_addr = NULL; + src_param->reg.idx_count = 0; src_param->reg.immconst_type = type; memcpy(src_param->reg.u.immconst_uint, *ptr, count * sizeof(uint32_t)); src_param->swizzle = VKD3D_SHADER_NO_SWIZZLE; diff --git a/libs/vkd3d-shader/ir.c b/libs/vkd3d-shader/ir.c index a3ab03f2..cce23afd 100644 --- a/libs/vkd3d-shader/ir.c +++ b/libs/vkd3d-shader/ir.c @@ -43,7 +43,7 @@ static void shader_register_eliminate_phase_addressing(struct vkd3d_shader_regis { unsigned int i; - for (i = 0; i < ARRAY_SIZE(reg->idx) && reg->idx[i].offset != ~0u; ++i) + for (i = 0; i < reg->idx_count; ++i) { if (reg->idx[i].rel_addr && shader_register_is_phase_instance_id(®->idx[i].rel_addr->reg)) { @@ -73,6 +73,7 @@ static void shader_instruction_eliminate_phase_instance_id(struct vkd3d_shader_i reg->idx[1].rel_addr = NULL; reg->idx[2].offset = ~0u; reg->idx[2].rel_addr = NULL; + reg->idx_count = 0; reg->immconst_type = VKD3D_IMMCONST_SCALAR; reg->u.immconst_uint[0] = instance_id; continue; @@ -222,8 +223,8 @@ static enum vkd3d_result shader_normaliser_flatten_phases(struct vkd3d_shader_no return VKD3D_OK; } -static void shader_register_init(struct vkd3d_shader_register *reg, - enum vkd3d_shader_register_type reg_type, enum vkd3d_data_type data_type) +static void shader_register_init(struct vkd3d_shader_register *reg, enum vkd3d_shader_register_type reg_type, + enum vkd3d_data_type data_type, unsigned int idx_count) { reg->type = reg_type; reg->precision = VKD3D_SHADER_REGISTER_PRECISION_DEFAULT; @@ -235,6 +236,7 @@ static void shader_register_init(struct vkd3d_shader_register *reg, reg->idx[1].rel_addr = NULL; reg->idx[2].offset = ~0u; reg->idx[2].rel_addr = NULL; + reg->idx_count = idx_count; reg->immconst_type = VKD3D_IMMCONST_SCALAR; } @@ -286,7 +288,7 @@ static struct vkd3d_shader_src_param *shader_normaliser_create_outpointid_param( if (!(rel_addr = shader_src_param_allocator_get(&normaliser->instructions.src_params, 1))) return NULL; - shader_register_init(&rel_addr->reg, VKD3DSPR_OUTPOINTID, VKD3D_DATA_UINT); + shader_register_init(&rel_addr->reg, VKD3DSPR_OUTPOINTID, VKD3D_DATA_UINT, 0); rel_addr->swizzle = 0; rel_addr->modifiers = 0; @@ -300,32 +302,25 @@ static bool shader_dst_param_normalise_outpointid(struct vkd3d_shader_dst_param if (normaliser_is_in_control_point_phase(normaliser) && reg->type == VKD3DSPR_OUTPUT) { - if (reg->idx[2].offset != ~0u) - { - FIXME("Cannot insert phase id.\n"); - return false; - } - if (reg->idx[1].offset != ~0u) - { - WARN("Unexpected address at index 1.\n"); - reg->idx[2] = reg->idx[1]; - } + /* The TPF reader validates idx_count. */ + assert(reg->idx_count == 1); reg->idx[1] = reg->idx[0]; /* The control point id param is implicit here. Avoid later complications by inserting it. */ reg->idx[0].offset = 0; reg->idx[0].rel_addr = normaliser->outpointid_param; + ++reg->idx_count; } return true; } -static void shader_dst_param_io_init(struct vkd3d_shader_dst_param *param, - const struct signature_element *e, enum vkd3d_shader_register_type reg_type) +static void shader_dst_param_io_init(struct vkd3d_shader_dst_param *param, const struct signature_element *e, + enum vkd3d_shader_register_type reg_type, unsigned int idx_count) { param->write_mask = e->mask; param->modifiers = 0; param->shift = 0; - shader_register_init(¶m->reg, reg_type, vkd3d_data_type_from_component_type(e->component_type)); + shader_register_init(¶m->reg, reg_type, vkd3d_data_type_from_component_type(e->component_type), idx_count); } static enum vkd3d_result shader_normaliser_emit_hs_input(struct vkd3d_shader_normaliser *normaliser, @@ -369,7 +364,7 @@ static enum vkd3d_result shader_normaliser_emit_hs_input(struct vkd3d_shader_nor param = &ins->declaration.dst; } - shader_dst_param_io_init(param, e, VKD3DSPR_INPUT); + shader_dst_param_io_init(param, e, VKD3DSPR_INPUT, 2); param->reg.idx[0].offset = input_control_point_count; param->reg.idx[1].offset = i; diff --git a/libs/vkd3d-shader/tpf.c b/libs/vkd3d-shader/tpf.c index fcd7cf93..b77942ff 100644 --- a/libs/vkd3d-shader/tpf.c +++ b/libs/vkd3d-shader/tpf.c @@ -564,6 +564,8 @@ struct vkd3d_shader_sm4_parser unsigned int output_map[MAX_REG_OUTPUT]; + enum vkd3d_shader_opcode phase; + struct vkd3d_shader_parser p; }; @@ -1669,6 +1671,8 @@ static bool shader_sm4_read_param(struct vkd3d_shader_sm4_parser *priv, const ui return false; } + param->idx_count = order; + if (register_type == VKD3D_SM4_RT_IMMCONST || register_type == VKD3D_SM4_RT_IMMCONST64) { enum vkd3d_sm4_dimension dimension = (token & VKD3D_SM4_DIMENSION_MASK) >> VKD3D_SM4_DIMENSION_SHIFT; @@ -1710,6 +1714,7 @@ static bool shader_sm4_read_param(struct vkd3d_shader_sm4_parser *priv, const ui * other values up one slot. Normalize to SM5.1. */ param->idx[2] = param->idx[1]; param->idx[1] = param->idx[0]; + ++param->idx_count; } map_register(priv, param); @@ -1742,6 +1747,47 @@ static uint32_t swizzle_from_sm4(uint32_t s) return vkd3d_shader_create_swizzle(s & 0x3, (s >> 2) & 0x3, (s >> 4) & 0x3, (s >> 6) & 0x3); } +static bool register_is_input_output(const struct vkd3d_shader_register *reg) +{ + switch (reg->type) + { + case VKD3DSPR_INPUT: + case VKD3DSPR_OUTPUT: + case VKD3DSPR_COLOROUT: + case VKD3DSPR_INCONTROLPOINT: + case VKD3DSPR_OUTCONTROLPOINT: + case VKD3DSPR_PATCHCONST: + return true; + + default: + return false; + } +} + +static bool register_is_control_point_input(const struct vkd3d_shader_register *reg, + const struct vkd3d_shader_sm4_parser *priv) +{ + return reg->type == VKD3DSPR_INCONTROLPOINT || reg->type == VKD3DSPR_OUTCONTROLPOINT + || (reg->type == VKD3DSPR_INPUT && (priv->phase == VKD3DSIH_HS_CONTROL_POINT_PHASE + || priv->p.shader_version.type == VKD3D_SHADER_TYPE_GEOMETRY)); +} + +static bool shader_sm4_validate_input_output_register(struct vkd3d_shader_sm4_parser *priv, + const struct vkd3d_shader_register *reg) +{ + unsigned int idx_count = 1 + register_is_control_point_input(reg, priv); + + if (reg->idx_count != idx_count) + { + vkd3d_shader_parser_error(&priv->p, VKD3D_SHADER_ERROR_TPF_INVALID_REGISTER_INDEX_COUNT, + "Invalid index count %u for register type %#x; expected count %u.", + reg->idx_count, reg->type, idx_count); + return false; + } + + return true; +} + static bool shader_sm4_read_src_param(struct vkd3d_shader_sm4_parser *priv, const uint32_t **ptr, const uint32_t *end, enum vkd3d_data_type data_type, struct vkd3d_shader_src_param *src_param) { @@ -1793,6 +1839,9 @@ static bool shader_sm4_read_src_param(struct vkd3d_shader_sm4_parser *priv, cons } } + if (register_is_input_output(&src_param->reg) && !shader_sm4_validate_input_output_register(priv, &src_param->reg)) + return false; + return true; } @@ -1830,6 +1879,9 @@ static bool shader_sm4_read_dst_param(struct vkd3d_shader_sm4_parser *priv, cons dst_param->modifiers = 0; dst_param->shift = 0; + if (register_is_input_output(&dst_param->reg) && !shader_sm4_validate_input_output_register(priv, &dst_param->reg)) + return false; + return true; } @@ -1967,6 +2019,9 @@ static void shader_sm4_read_instruction(struct vkd3d_shader_sm4_parser *sm4, str } ins->handler_idx = opcode_info->handler_idx; + if (ins->handler_idx == VKD3DSIH_HS_CONTROL_POINT_PHASE || ins->handler_idx == VKD3DSIH_HS_FORK_PHASE + || ins->handler_idx == VKD3DSIH_HS_JOIN_PHASE) + sm4->phase = ins->handler_idx; ins->flags = 0; ins->coissue = false; ins->raw = false; diff --git a/libs/vkd3d-shader/vkd3d_shader_private.h b/libs/vkd3d-shader/vkd3d_shader_private.h index f1641c1e..2546c0d2 100644 --- a/libs/vkd3d-shader/vkd3d_shader_private.h +++ b/libs/vkd3d-shader/vkd3d_shader_private.h @@ -74,6 +74,7 @@ enum vkd3d_shader_error VKD3D_SHADER_ERROR_TPF_MISMATCHED_CF = 1000, VKD3D_SHADER_ERROR_TPF_INVALID_REGISTER_RANGE = 1001, VKD3D_SHADER_ERROR_TPF_OUT_OF_MEMORY = 1002, + VKD3D_SHADER_ERROR_TPF_INVALID_REGISTER_INDEX_COUNT = 1003, VKD3D_SHADER_ERROR_SPV_DESCRIPTOR_BINDING_NOT_FOUND = 2000, VKD3D_SHADER_ERROR_SPV_INVALID_REGISTER_TYPE = 2001, @@ -676,6 +677,7 @@ struct vkd3d_shader_register bool non_uniform; enum vkd3d_data_type data_type; struct vkd3d_shader_register_index idx[3]; + unsigned int idx_count; enum vkd3d_immconst_type immconst_type; union { -- GitLab https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/181
From: Conor McCarthy <cmccarthy(a)codeweavers.com> --- libs/vkd3d-shader/tpf.c | 31 ++++++++++++++++++++++++ libs/vkd3d-shader/vkd3d_shader_private.h | 1 + 2 files changed, 32 insertions(+) diff --git a/libs/vkd3d-shader/tpf.c b/libs/vkd3d-shader/tpf.c index b77942ff..3dd2b446 100644 --- a/libs/vkd3d-shader/tpf.c +++ b/libs/vkd3d-shader/tpf.c @@ -2200,6 +2200,29 @@ static bool shader_sm4_init(struct vkd3d_shader_sm4_parser *sm4, const uint32_t return true; } +static bool shader_sm4_parser_validate_signature(struct vkd3d_shader_sm4_parser *sm4, + const struct shader_signature *signature, const char *name) +{ + unsigned int i, register_idx, register_count; + + for (i = 0; i < signature->element_count; ++i) + { + register_idx = signature->elements[i].register_index; + register_count = signature->elements[i].register_count; + if (register_idx != ~0u && (register_idx >= MAX_REG_OUTPUT || MAX_REG_OUTPUT - register_idx < register_count)) + { + WARN("%s signature element %u unhandled register index %u, count %u.\n", + name, i, register_idx, register_count); + vkd3d_shader_parser_error(&sm4->p, VKD3D_SHADER_ERROR_TPF_TOO_MANY_REGISTERS, + "%s signature element %u register index %u, count %u exceeds maximum index of %u.", name, + i, register_idx, register_count, MAX_REG_OUTPUT - 1); + return false; + } + } + + return true; +} + int vkd3d_shader_sm4_parser_create(const struct vkd3d_shader_compile_info *compile_info, struct vkd3d_shader_message_context *message_context, struct vkd3d_shader_parser **parser) { @@ -2233,6 +2256,14 @@ int vkd3d_shader_sm4_parser_create(const struct vkd3d_shader_compile_info *compi return VKD3D_ERROR_INVALID_ARGUMENT; } + if (!shader_sm4_parser_validate_signature(sm4, &shader_desc->input_signature, "Input") + || !shader_sm4_parser_validate_signature(sm4, &shader_desc->output_signature, "Output") + || !shader_sm4_parser_validate_signature(sm4, &shader_desc->patch_constant_signature, "Patch constant")) + { + shader_sm4_destroy(&sm4->p); + return VKD3D_ERROR_INVALID_SHADER; + } + instructions = &sm4->p.instructions; while (sm4->ptr != sm4->end) { diff --git a/libs/vkd3d-shader/vkd3d_shader_private.h b/libs/vkd3d-shader/vkd3d_shader_private.h index 2546c0d2..3b421813 100644 --- a/libs/vkd3d-shader/vkd3d_shader_private.h +++ b/libs/vkd3d-shader/vkd3d_shader_private.h @@ -75,6 +75,7 @@ enum vkd3d_shader_error VKD3D_SHADER_ERROR_TPF_INVALID_REGISTER_RANGE = 1001, VKD3D_SHADER_ERROR_TPF_OUT_OF_MEMORY = 1002, VKD3D_SHADER_ERROR_TPF_INVALID_REGISTER_INDEX_COUNT = 1003, + VKD3D_SHADER_ERROR_TPF_TOO_MANY_REGISTERS = 1004, VKD3D_SHADER_ERROR_SPV_DESCRIPTOR_BINDING_NOT_FOUND = 2000, VKD3D_SHADER_ERROR_SPV_INVALID_REGISTER_TYPE = 2001, -- GitLab https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/181
From: Conor McCarthy <cmccarthy(a)codeweavers.com> --- libs/vkd3d-shader/tpf.c | 9 ++++++++- libs/vkd3d-shader/vkd3d_shader_private.h | 2 ++ 2 files changed, 10 insertions(+), 1 deletion(-) diff --git a/libs/vkd3d-shader/tpf.c b/libs/vkd3d-shader/tpf.c index 3dd2b446..0daadb9c 100644 --- a/libs/vkd3d-shader/tpf.c +++ b/libs/vkd3d-shader/tpf.c @@ -2203,7 +2203,7 @@ static bool shader_sm4_init(struct vkd3d_shader_sm4_parser *sm4, const uint32_t static bool shader_sm4_parser_validate_signature(struct vkd3d_shader_sm4_parser *sm4, const struct shader_signature *signature, const char *name) { - unsigned int i, register_idx, register_count; + unsigned int i, register_idx, register_count, mask; for (i = 0; i < signature->element_count; ++i) { @@ -2218,6 +2218,13 @@ static bool shader_sm4_parser_validate_signature(struct vkd3d_shader_sm4_parser i, register_idx, register_count, MAX_REG_OUTPUT - 1); return false; } + + if (!vkd3d_bitmask_is_contiguous(mask = signature->elements[i].mask)) + { + WARN("%s signature element %u mask %#x is not contiguous.\n", name, i, mask); + vkd3d_shader_parser_warning(&sm4->p, VKD3D_SHADER_WARNING_TPF_MASK_NOT_CONTIGUOUS, + "%s signature element %u mask %#x is not contiguous.", name, i, mask); + } } return true; diff --git a/libs/vkd3d-shader/vkd3d_shader_private.h b/libs/vkd3d-shader/vkd3d_shader_private.h index 3b421813..d2ea97e8 100644 --- a/libs/vkd3d-shader/vkd3d_shader_private.h +++ b/libs/vkd3d-shader/vkd3d_shader_private.h @@ -77,6 +77,8 @@ enum vkd3d_shader_error VKD3D_SHADER_ERROR_TPF_INVALID_REGISTER_INDEX_COUNT = 1003, VKD3D_SHADER_ERROR_TPF_TOO_MANY_REGISTERS = 1004, + VKD3D_SHADER_WARNING_TPF_MASK_NOT_CONTIGUOUS = 1300, + VKD3D_SHADER_ERROR_SPV_DESCRIPTOR_BINDING_NOT_FOUND = 2000, VKD3D_SHADER_ERROR_SPV_INVALID_REGISTER_TYPE = 2001, VKD3D_SHADER_ERROR_SPV_INVALID_DESCRIPTOR_BINDING = 2002, -- GitLab https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/181
From: Conor McCarthy <cmccarthy(a)codeweavers.com> --- libs/vkd3d-shader/tpf.c | 72 +++++++++++++++++++++--- libs/vkd3d-shader/vkd3d_shader_private.h | 1 + 2 files changed, 65 insertions(+), 8 deletions(-) diff --git a/libs/vkd3d-shader/tpf.c b/libs/vkd3d-shader/tpf.c index 0daadb9c..8b796dc1 100644 --- a/libs/vkd3d-shader/tpf.c +++ b/libs/vkd3d-shader/tpf.c @@ -565,6 +565,9 @@ struct vkd3d_shader_sm4_parser unsigned int output_map[MAX_REG_OUTPUT]; enum vkd3d_shader_opcode phase; + unsigned int input_register_masks[MAX_REG_OUTPUT]; + unsigned int output_register_masks[MAX_REG_OUTPUT]; + unsigned int patch_constant_register_masks[MAX_REG_OUTPUT]; struct vkd3d_shader_parser p; }; @@ -1427,7 +1430,7 @@ static void map_register(const struct vkd3d_shader_sm4_parser *sm4, struct vkd3d if (reg_idx >= ARRAY_SIZE(sm4->output_map)) { - ERR("Invalid output index %u.\n", reg_idx); + /* Validated later */ break; } @@ -1772,10 +1775,20 @@ static bool register_is_control_point_input(const struct vkd3d_shader_register * || priv->p.shader_version.type == VKD3D_SHADER_TYPE_GEOMETRY)); } +static unsigned int mask_from_swizzle(unsigned int swizzle) +{ + return (1u << vkd3d_swizzle_get_component(swizzle, 0)) + | (1u << vkd3d_swizzle_get_component(swizzle, 1)) + | (1u << vkd3d_swizzle_get_component(swizzle, 2)) + | (1u << vkd3d_swizzle_get_component(swizzle, 3)); +} + static bool shader_sm4_validate_input_output_register(struct vkd3d_shader_sm4_parser *priv, - const struct vkd3d_shader_register *reg) + const struct vkd3d_shader_register *reg, unsigned int mask) { unsigned int idx_count = 1 + register_is_control_point_input(reg, priv); + const unsigned int *masks; + unsigned int register_idx; if (reg->idx_count != idx_count) { @@ -1785,6 +1798,41 @@ static bool shader_sm4_validate_input_output_register(struct vkd3d_shader_sm4_pa return false; } + switch (reg->type) + { + case VKD3DSPR_INPUT: + case VKD3DSPR_INCONTROLPOINT: + masks = priv->input_register_masks; + break; + case VKD3DSPR_OUTPUT: + masks = (priv->phase == VKD3DSIH_HS_FORK_PHASE || priv->phase == VKD3DSIH_HS_JOIN_PHASE) + ? priv->patch_constant_register_masks + : priv->output_register_masks; + break; + case VKD3DSPR_COLOROUT: + case VKD3DSPR_OUTCONTROLPOINT: + masks = priv->output_register_masks; + break; + case VKD3DSPR_PATCHCONST: + masks = priv->patch_constant_register_masks; + break; + + default: + vkd3d_unreachable(); + } + + register_idx = reg->idx[reg->idx_count - 1].offset; + /* The signature element registers have already been checked against MAX_REG_OUTPUT. */ + if (register_idx >= MAX_REG_OUTPUT || (masks[register_idx] & mask) != mask) + { + WARN("Failed to find signature element for register type %#x, index %u and mask %#x.\n", + reg->type, register_idx, mask); + vkd3d_shader_parser_error(&priv->p, VKD3D_SHADER_ERROR_TPF_INVALID_IO_REGISTER, + "Could not find signature element matching register type %#x, index %u and mask %#x.", + reg->type, register_idx, mask); + return false; + } + return true; } @@ -1839,7 +1887,8 @@ static bool shader_sm4_read_src_param(struct vkd3d_shader_sm4_parser *priv, cons } } - if (register_is_input_output(&src_param->reg) && !shader_sm4_validate_input_output_register(priv, &src_param->reg)) + if (register_is_input_output(&src_param->reg) && !shader_sm4_validate_input_output_register(priv, + &src_param->reg, mask_from_swizzle(src_param->swizzle))) return false; return true; @@ -1879,7 +1928,8 @@ static bool shader_sm4_read_dst_param(struct vkd3d_shader_sm4_parser *priv, cons dst_param->modifiers = 0; dst_param->shift = 0; - if (register_is_input_output(&dst_param->reg) && !shader_sm4_validate_input_output_register(priv, &dst_param->reg)) + if (register_is_input_output(&dst_param->reg) && !shader_sm4_validate_input_output_register(priv, + &dst_param->reg, dst_param->write_mask)) return false; return true; @@ -2201,7 +2251,7 @@ static bool shader_sm4_init(struct vkd3d_shader_sm4_parser *sm4, const uint32_t } static bool shader_sm4_parser_validate_signature(struct vkd3d_shader_sm4_parser *sm4, - const struct shader_signature *signature, const char *name) + const struct shader_signature *signature, unsigned int *masks, const char *name) { unsigned int i, register_idx, register_count, mask; @@ -2225,6 +2275,9 @@ static bool shader_sm4_parser_validate_signature(struct vkd3d_shader_sm4_parser vkd3d_shader_parser_warning(&sm4->p, VKD3D_SHADER_WARNING_TPF_MASK_NOT_CONTIGUOUS, "%s signature element %u mask %#x is not contiguous.", name, i, mask); } + + if (register_idx != ~0u) + masks[register_idx] |= mask; } return true; @@ -2263,9 +2316,12 @@ int vkd3d_shader_sm4_parser_create(const struct vkd3d_shader_compile_info *compi return VKD3D_ERROR_INVALID_ARGUMENT; } - if (!shader_sm4_parser_validate_signature(sm4, &shader_desc->input_signature, "Input") - || !shader_sm4_parser_validate_signature(sm4, &shader_desc->output_signature, "Output") - || !shader_sm4_parser_validate_signature(sm4, &shader_desc->patch_constant_signature, "Patch constant")) + if (!shader_sm4_parser_validate_signature(sm4, &shader_desc->input_signature, + sm4->input_register_masks, "Input") + || !shader_sm4_parser_validate_signature(sm4, &shader_desc->output_signature, + sm4->output_register_masks, "Output") + || !shader_sm4_parser_validate_signature(sm4, &shader_desc->patch_constant_signature, + sm4->patch_constant_register_masks, "Patch constant")) { shader_sm4_destroy(&sm4->p); return VKD3D_ERROR_INVALID_SHADER; diff --git a/libs/vkd3d-shader/vkd3d_shader_private.h b/libs/vkd3d-shader/vkd3d_shader_private.h index d2ea97e8..55c25f64 100644 --- a/libs/vkd3d-shader/vkd3d_shader_private.h +++ b/libs/vkd3d-shader/vkd3d_shader_private.h @@ -76,6 +76,7 @@ enum vkd3d_shader_error VKD3D_SHADER_ERROR_TPF_OUT_OF_MEMORY = 1002, VKD3D_SHADER_ERROR_TPF_INVALID_REGISTER_INDEX_COUNT = 1003, VKD3D_SHADER_ERROR_TPF_TOO_MANY_REGISTERS = 1004, + VKD3D_SHADER_ERROR_TPF_INVALID_IO_REGISTER = 1005, VKD3D_SHADER_WARNING_TPF_MASK_NOT_CONTIGUOUS = 1300, -- GitLab https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/181
From: Conor McCarthy <cmccarthy(a)codeweavers.com> --- libs/vkd3d-shader/spirv.c | 23 ------ libs/vkd3d-shader/tpf.c | 92 ++++++++++++++++++++++-- libs/vkd3d-shader/vkd3d_shader_private.h | 2 + 3 files changed, 90 insertions(+), 27 deletions(-) diff --git a/libs/vkd3d-shader/spirv.c b/libs/vkd3d-shader/spirv.c index e13ce0a2..fa4d056e 100644 --- a/libs/vkd3d-shader/spirv.c +++ b/libs/vkd3d-shader/spirv.c @@ -6195,8 +6195,6 @@ static bool spirv_compiler_check_index_range(struct spirv_compiler *compiler, struct vkd3d_shader_register_info reg_info; struct vkd3d_shader_register current_reg; struct vkd3d_symbol reg_symbol; - unsigned int i; - uint32_t id; current_reg = *reg; vkd3d_symbol_make_register(®_symbol, ¤t_reg); @@ -6212,27 +6210,6 @@ static bool spirv_compiler_check_index_range(struct spirv_compiler *compiler, FIXME("Unhandled register %#x.\n", reg->type); return false; } - id = reg_info.id; - - for (i = reg->idx[0].offset; i < reg->idx[0].offset + range->register_count; ++i) - { - current_reg.idx[0].offset = i; - vkd3d_symbol_make_register(®_symbol, ¤t_reg); - - if (range->dst.write_mask != reg_info.write_mask - || vkd3d_write_mask_component_count(reg_info.write_mask) != 1) - { - FIXME("Unhandled index range write mask %#x (%#x).\n", - range->dst.write_mask, reg_info.write_mask); - return false; - } - - if (reg_info.id != id) - { - FIXME("Unhandled index range %#x, %u.\n", reg->type, i); - return false; - } - } return true; } diff --git a/libs/vkd3d-shader/tpf.c b/libs/vkd3d-shader/tpf.c index 8b796dc1..1a347a5f 100644 --- a/libs/vkd3d-shader/tpf.c +++ b/libs/vkd3d-shader/tpf.c @@ -569,6 +569,10 @@ struct vkd3d_shader_sm4_parser unsigned int output_register_masks[MAX_REG_OUTPUT]; unsigned int patch_constant_register_masks[MAX_REG_OUTPUT]; + unsigned int input_index_range_masks[MAX_REG_OUTPUT]; + unsigned int output_index_range_masks[MAX_REG_OUTPUT]; + unsigned int patch_constant_index_range_masks[MAX_REG_OUTPUT]; + struct vkd3d_shader_parser p; }; @@ -823,12 +827,93 @@ static void shader_sm4_read_dcl_sampler(struct vkd3d_shader_instruction *ins, ui shader_sm4_read_register_space(priv, &tokens, end, &ins->declaration.sampler.range.space); } +static bool sm4_parser_is_in_fork_or_join_phase(const struct vkd3d_shader_sm4_parser *sm4) +{ + return sm4->phase == VKD3DSIH_HS_FORK_PHASE || sm4->phase == VKD3DSIH_HS_JOIN_PHASE; +} + static void shader_sm4_read_dcl_index_range(struct vkd3d_shader_instruction *ins, uint32_t opcode, uint32_t opcode_token, const uint32_t *tokens, unsigned int token_count, struct vkd3d_shader_sm4_parser *priv) { + struct vkd3d_shader_index_range *index_range = &ins->declaration.index_range; + unsigned int i, register_idx, register_count, write_mask; + unsigned int *io_masks, *range_masks; + enum vkd3d_shader_register_type type; + shader_sm4_read_dst_param(priv, &tokens, &tokens[token_count], VKD3D_DATA_OPAQUE, - &ins->declaration.index_range.dst); - ins->declaration.index_range.register_count = *tokens; + &index_range->dst); + index_range->register_count = *tokens; + + register_idx = index_range->dst.reg.idx[index_range->dst.reg.idx_count - 1].offset; + register_count = index_range->register_count; + write_mask = index_range->dst.write_mask; + + if (vkd3d_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; + range_masks = priv->input_index_range_masks; + break; + case VKD3DSPR_OUTPUT: + if (sm4_parser_is_in_fork_or_join_phase(priv)) + { + io_masks = priv->patch_constant_register_masks; + range_masks = priv->patch_constant_index_range_masks; + } + else + { + io_masks = priv->output_register_masks; + range_masks = priv->output_index_range_masks; + } + break; + case VKD3DSPR_COLOROUT: + case VKD3DSPR_OUTCONTROLPOINT: + io_masks = priv->output_register_masks; + range_masks = priv->output_index_range_masks; + break; + case VKD3DSPR_PATCHCONST: + io_masks = priv->patch_constant_register_masks; + range_masks = priv->patch_constant_index_range_masks; + break; + + default: + WARN("Unhandled register type %#x.\n", type); + vkd3d_shader_parser_error(&priv->p, VKD3D_SHADER_ERROR_TPF_INVALID_INDEX_RANGE_DCL, + "Invalid register type %#x for index range base %u, count %u, mask %#x.", + type, register_idx, register_count, write_mask); + return; + } + + for (i = 0; i < register_count; ++i) + { + if (range_masks[register_idx + i] & write_mask) + { + WARN("Detected index range collision for base %u, count %u, mask %#x.\n", + register_idx, register_count, write_mask); + vkd3d_shader_parser_error(&priv->p, VKD3D_SHADER_ERROR_TPF_INVALID_INDEX_RANGE_DCL, + "Register index range base %u, count %u, mask %#x collides with a previous declaration.", + register_idx, register_count, write_mask); + 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", + register_idx, register_count, write_mask); + vkd3d_shader_parser_error(&priv->p, VKD3D_SHADER_ERROR_TPF_INVALID_INDEX_RANGE_DCL, + "Input/output registers matching index range base %u, count %u, mask %#x were not declared.", + register_idx, register_count, write_mask); + return; + } + range_masks[register_idx + i] |= write_mask; + } } static void shader_sm4_read_dcl_output_topology(struct vkd3d_shader_instruction *ins, uint32_t opcode, @@ -1805,8 +1890,7 @@ static bool shader_sm4_validate_input_output_register(struct vkd3d_shader_sm4_pa masks = priv->input_register_masks; break; case VKD3DSPR_OUTPUT: - masks = (priv->phase == VKD3DSIH_HS_FORK_PHASE || priv->phase == VKD3DSIH_HS_JOIN_PHASE) - ? priv->patch_constant_register_masks + masks = sm4_parser_is_in_fork_or_join_phase(priv) ? priv->patch_constant_register_masks : priv->output_register_masks; break; case VKD3DSPR_COLOROUT: diff --git a/libs/vkd3d-shader/vkd3d_shader_private.h b/libs/vkd3d-shader/vkd3d_shader_private.h index 55c25f64..f085deb8 100644 --- a/libs/vkd3d-shader/vkd3d_shader_private.h +++ b/libs/vkd3d-shader/vkd3d_shader_private.h @@ -77,8 +77,10 @@ enum vkd3d_shader_error VKD3D_SHADER_ERROR_TPF_INVALID_REGISTER_INDEX_COUNT = 1003, VKD3D_SHADER_ERROR_TPF_TOO_MANY_REGISTERS = 1004, VKD3D_SHADER_ERROR_TPF_INVALID_IO_REGISTER = 1005, + VKD3D_SHADER_ERROR_TPF_INVALID_INDEX_RANGE_DCL = 1006, VKD3D_SHADER_WARNING_TPF_MASK_NOT_CONTIGUOUS = 1300, + VKD3D_SHADER_WARNING_TPF_UNHANDLED_INDEX_RANGE_MASK = 1301, VKD3D_SHADER_ERROR_SPV_DESCRIPTOR_BINDING_NOT_FOUND = 2000, VKD3D_SHADER_ERROR_SPV_INVALID_REGISTER_TYPE = 2001, -- GitLab https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/181
From: Conor McCarthy <cmccarthy(a)codeweavers.com> --- libs/vkd3d-shader/tpf.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/libs/vkd3d-shader/tpf.c b/libs/vkd3d-shader/tpf.c index 1a347a5f..e90fb64d 100644 --- a/libs/vkd3d-shader/tpf.c +++ b/libs/vkd3d-shader/tpf.c @@ -729,7 +729,7 @@ static void shader_sm4_set_descriptor_register_range(struct vkd3d_shader_sm4_par { FIXME("Invalid register range [%u:%u].\n", range->first, range->last); vkd3d_shader_parser_error(&sm4->p, VKD3D_SHADER_ERROR_TPF_INVALID_REGISTER_RANGE, - "Last register %u must not be less than first register %u in range.\n", range->last, range->first); + "Last register %u must not be less than first register %u in range.", range->last, range->first); } } -- GitLab https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/181
From: Conor McCarthy <cmccarthy(a)codeweavers.com> In the fork/join phases these registers have two-dimensional addressing. --- libs/vkd3d-shader/tpf.c | 13 +++++++++++++ 1 file changed, 13 insertions(+) diff --git a/libs/vkd3d-shader/tpf.c b/libs/vkd3d-shader/tpf.c index e90fb64d..077956ae 100644 --- a/libs/vkd3d-shader/tpf.c +++ b/libs/vkd3d-shader/tpf.c @@ -565,6 +565,7 @@ struct vkd3d_shader_sm4_parser unsigned int output_map[MAX_REG_OUTPUT]; enum vkd3d_shader_opcode phase; + bool has_control_point_phase; unsigned int input_register_masks[MAX_REG_OUTPUT]; unsigned int output_register_masks[MAX_REG_OUTPUT]; unsigned int patch_constant_register_masks[MAX_REG_OUTPUT]; @@ -892,6 +893,17 @@ static void shader_sm4_read_dcl_index_range(struct vkd3d_shader_instruction *ins return; } + if (priv->p.shader_version.type == VKD3D_SHADER_TYPE_HULL && !priv->has_control_point_phase + && (range_masks == priv->input_index_range_masks || range_masks == priv->output_index_range_masks)) + { + WARN("Index range base %u, count %u, mask %#x applies to a default phase.\n", + register_idx, register_count, write_mask); + vkd3d_shader_parser_error(&priv->p, VKD3D_SHADER_ERROR_TPF_INVALID_INDEX_RANGE_DCL, + "Register index range base %u, count %u, mask %#x applies to a default control point phase.", + register_idx, register_count, write_mask); + return; + } + for (i = 0; i < register_count; ++i) { if (range_masks[register_idx + i] & write_mask) @@ -2156,6 +2168,7 @@ static void shader_sm4_read_instruction(struct vkd3d_shader_sm4_parser *sm4, str if (ins->handler_idx == VKD3DSIH_HS_CONTROL_POINT_PHASE || ins->handler_idx == VKD3DSIH_HS_FORK_PHASE || ins->handler_idx == VKD3DSIH_HS_JOIN_PHASE) sm4->phase = ins->handler_idx; + sm4->has_control_point_phase |= ins->handler_idx == VKD3DSIH_HS_CONTROL_POINT_PHASE; ins->flags = 0; ins->coissue = false; ins->raw = false; -- GitLab https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/181
From: Conor McCarthy <cmccarthy(a)codeweavers.com> In Shader Model 6 each signature element can span a range of register indices, or 'rows', and system values do not share a register index with non-system values. Inputs and outputs are referenced by element index instead of register index. This patch merges multiple signature elements into a single element under the following conditions: - The register index in a load or store is specified dynamically by including a relative address parameter with a base register index. The dcl_index_range instruction is used to identify these. - A register declaration is split across multiple elements which declare different components of the register. - A patch constant function writes tessellation factors. These are an array in SPIR-V, but in SM 5.x each factor is declared as a separate register, and these are dynamically indexed by the fork/join instance id. Elimination of multiple fork/join phases converts the indices to constants, but merging the signature elements into a single arrayed element matches the SPIR-V output. All references to input/output register indices are converted to element indices. If a relative address is present, the element index is moved up a slot so it cannot be confused with a constant offset. Existing code only handles register index relative addressing for tessellation factors. This patch adds generic support for it. --- libs/vkd3d-shader/dxbc.c | 2 + libs/vkd3d-shader/ir.c | 587 ++++++++++++++++++- libs/vkd3d-shader/spirv.c | 696 ++++++++--------------- libs/vkd3d-shader/vkd3d_shader_private.h | 26 +- tests/d3d12.c | 1 - 5 files changed, 847 insertions(+), 465 deletions(-) diff --git a/libs/vkd3d-shader/dxbc.c b/libs/vkd3d-shader/dxbc.c index 743ba369..3e3f06fa 100644 --- a/libs/vkd3d-shader/dxbc.c +++ b/libs/vkd3d-shader/dxbc.c @@ -373,6 +373,8 @@ static int shader_parse_signature(const struct vkd3d_shader_dxbc_section_desc *s { uint32_t name_offset, mask; + e[i].sort_index = i; + if (has_stream_index) read_dword(&ptr, &e[i].stream_index); else diff --git a/libs/vkd3d-shader/ir.c b/libs/vkd3d-shader/ir.c index cce23afd..87c416d4 100644 --- a/libs/vkd3d-shader/ir.c +++ b/libs/vkd3d-shader/ir.c @@ -247,11 +247,12 @@ static void shader_instruction_init(struct vkd3d_shader_instruction *ins, enum v } void shader_normaliser_init(struct vkd3d_shader_normaliser *normaliser, - struct vkd3d_shader_instruction_array *instructions) + struct vkd3d_shader_instruction_array *instructions, struct vkd3d_shader_message_context *message_context) { memset(normaliser, 0, sizeof(*normaliser)); normaliser->phase = VKD3DSIH_INVALID; normaliser->instructions = *instructions; + normaliser->message_context = message_context; memset(instructions, 0, sizeof(*instructions)); } @@ -438,6 +439,590 @@ enum vkd3d_result shader_normaliser_normalise_hull_shader_control_point_io(struc return VKD3D_OK; } +static unsigned int shader_signature_find_element_for_reg(const struct shader_signature *signature, + unsigned int reg_idx, unsigned int write_mask, struct vkd3d_shader_normaliser *normaliser) +{ + unsigned int i; + + for (i = 0; i < signature->element_count; ++i) + { + struct signature_element *e = &signature->elements[i]; + if (e->register_index <= reg_idx && e->register_index + e->register_count > reg_idx + && (e->mask & write_mask) == write_mask) + { + return i; + } + } + + /* Validated in the TPF reader. */ + vkd3d_unreachable(); +} + +static unsigned int range_map_get_register_count(uint8_t range_map[][VKD3D_VEC4_SIZE], + unsigned int register_idx, unsigned int write_mask, struct vkd3d_shader_normaliser *normaliser) +{ + if (register_idx >= MAX_REG_OUTPUT) + { + /* An error was already emitted if this occurred while setting the counts, + * and returning 0 is safe otherwise. */ + return 0; + } + return range_map[register_idx][vkd3d_write_mask_get_component_idx(write_mask)]; +} + +static void range_map_set_register_range(uint8_t range_map[][VKD3D_VEC4_SIZE], unsigned int register_idx, + unsigned int register_count, unsigned int write_mask, bool is_dcl_indexrange, + struct vkd3d_shader_normaliser *normaliser) +{ + unsigned int i, j, r, c, component_idx, component_count; + + assert(write_mask <= VKD3DSP_WRITEMASK_ALL); + component_idx = vkd3d_write_mask_get_component_idx(write_mask); + component_count = vkd3d_write_mask_component_count(write_mask); + + assert(register_idx < MAX_REG_OUTPUT && MAX_REG_OUTPUT - register_idx >= register_count); + + if (range_map[register_idx][component_idx] > register_count && is_dcl_indexrange) + { + /* Validated in the TPF reader. */ + assert(range_map[register_idx][component_idx] != UINT8_MAX); + return; + } + if (range_map[register_idx][component_idx] == register_count) + { + /* Already done. This happens when fxc splits a register declaration by + * component(s). The dcl_indexrange instructions are split too. */ + return; + } + range_map[register_idx][component_idx] = register_count; + + for (i = 0; i < register_count; ++i) + { + r = register_idx + i; + for (j = !i; j < component_count; ++j) + { + c = component_idx + j; + /* A synthetic patch constant range which overlaps an existing range can start upstream of it + * for fork/join phase instancing, but ranges declared by dcl_indexrange should not overlap. + * The latter is validated in the TPF reader. */ + assert(!range_map[r][c] || !is_dcl_indexrange); + range_map[r][c] = UINT8_MAX; + } + } +} + +static void shader_normaliser_add_index_range(struct vkd3d_shader_normaliser *normaliser, + const struct vkd3d_shader_instruction *ins) +{ + const struct vkd3d_shader_index_range *range = &ins->declaration.index_range; + const struct vkd3d_shader_register *reg = &range->dst.reg; + unsigned int reg_idx, write_mask, element_idx; + const struct shader_signature *signature; + uint8_t (*range_map)[VKD3D_VEC4_SIZE]; + + switch (reg->type) + { + case VKD3DSPR_INPUT: + case VKD3DSPR_INCONTROLPOINT: + range_map = normaliser->input_range_map; + signature = normaliser->input_signature; + break; + case VKD3DSPR_OUTCONTROLPOINT: + range_map = normaliser->output_range_map; + signature = normaliser->output_signature; + break; + case VKD3DSPR_OUTPUT: + if (!normaliser_is_in_fork_or_join_phase(normaliser)) + { + range_map = normaliser->output_range_map; + signature = normaliser->output_signature; + break; + } + /* fall through */ + case VKD3DSPR_PATCHCONST: + range_map = normaliser->pc_range_map; + signature = normaliser->patch_constant_signature; + break; + default: + /* Validated in the TPF reader. */ + vkd3d_unreachable(); + } + + reg_idx = reg->idx[reg->idx_count - 1].offset; + write_mask = range->dst.write_mask; + element_idx = shader_signature_find_element_for_reg(signature, reg_idx, write_mask, normaliser); + range_map_set_register_range(range_map, reg_idx, range->register_count, + signature->elements[element_idx].mask, true, normaliser); +} + +static int signature_element_mask_compare(const void *a, const void *b) +{ + const struct signature_element *e = a, *f = b; + int ret; + + return (ret = vkd3d_u32_compare(e->mask, f->mask)) ? ret : vkd3d_u32_compare(e->register_index, f->register_index); +} + +static bool sysval_semantics_should_merge(const struct signature_element *e, const struct signature_element *f) +{ + if (e->sysval_semantic < VKD3D_SHADER_SV_TESS_FACTOR_QUADEDGE + || e->sysval_semantic > VKD3D_SHADER_SV_TESS_FACTOR_LINEDEN) + return false; + + return e->sysval_semantic == f->sysval_semantic + /* Line detail and density must be merged together to match the SPIR-V array. + * This deletes one of the two sysvals, but these are not used. */ + || (e->sysval_semantic == VKD3D_SHADER_SV_TESS_FACTOR_LINEDET + && f->sysval_semantic == VKD3D_SHADER_SV_TESS_FACTOR_LINEDEN) + || (e->sysval_semantic == VKD3D_SHADER_SV_TESS_FACTOR_LINEDEN + && f->sysval_semantic == VKD3D_SHADER_SV_TESS_FACTOR_LINEDET); +} + +/* Merge tess factor sysvals because they are an array in SPIR-V. */ +static void shader_signature_map_patch_constant_index_ranges(struct shader_signature *s, + uint8_t range_map[][VKD3D_VEC4_SIZE], struct vkd3d_shader_normaliser *normaliser) +{ + struct signature_element *e, *f; + unsigned int i, j, register_count; + + qsort(s->elements, s->element_count, sizeof(s->elements[0]), signature_element_mask_compare); + + for (i = 0; i < s->element_count; i += register_count) + { + e = &s->elements[i]; + register_count = 1; + + if (!e->sysval_semantic) + continue; + + for (j = i + 1; j < s->element_count; ++j, ++register_count) + { + f = &s->elements[j]; + if (f->register_index != e->register_index + register_count || !sysval_semantics_should_merge(e, f)) + break; + } + if (register_count < 2) + continue; + + range_map_set_register_range(range_map, e->register_index, register_count, e->mask, false, normaliser); + } +} + +static int signature_element_register_compare(const void *a, const void *b) +{ + const struct signature_element *e = a, *f = b; + + return vkd3d_u32_compare(e->register_index, f->register_index); +} + +static int signature_element_index_compare(const void *a, const void *b) +{ + const struct signature_element *e = a, *f = b; + + return vkd3d_u32_compare(e->sort_index, f->sort_index); +} + +static bool shader_signature_merge(struct shader_signature *s, uint8_t range_map[][VKD3D_VEC4_SIZE], + bool is_patch_constant, struct vkd3d_shader_normaliser *normaliser) +{ + unsigned int i, j, element_count, new_count, register_count; + struct signature_element *elements; + struct signature_element *e, *f; + + element_count = s->element_count; + if (!(elements = vkd3d_malloc(element_count * sizeof(*elements)))) + return false; + memcpy(elements, s->elements, element_count * sizeof(*elements)); + + qsort(elements, element_count, sizeof(elements[0]), signature_element_register_compare); + + for (i = 0, new_count = 0; i < element_count; i = j, elements[new_count++] = *e) + { + e = &elements[i]; + j = i + 1; + + if (e->register_index == ~0u) + continue; + + /* Do not merge if the register index will be relative-addressed. */ + if (range_map_get_register_count(range_map, e->register_index, e->mask, normaliser) > 1) + continue; + + for (; j < element_count; ++j) + { + f = &elements[j]; + + /* Merge different components of the same register unless sysvals are different, + * or it will be relative-addressed. */ + if (f->register_index != e->register_index || f->sysval_semantic != e->sysval_semantic + || range_map_get_register_count(range_map, f->register_index, f->mask, normaliser) > 1) + break; + + TRACE("Merging %s, reg %u, mask %#x, sysval %#x with %s, mask %#x, sysval %#x.\n", e->semantic_name, + e->register_index, e->mask, e->sysval_semantic, f->semantic_name, f->mask, f->sysval_semantic); + assert(!(e->mask & f->mask)); + + e->mask |= f->mask; + e->used_mask |= f->used_mask; + e->semantic_index = min(e->semantic_index, f->semantic_index); + } + } + element_count = new_count; + /* Signature 's' is a copy of the original signature struct, so we can replace + * the 'elements' pointer without freeing it. */ + s->elements = elements; + s->element_count = element_count; + + if (is_patch_constant) + shader_signature_map_patch_constant_index_ranges(s, range_map, normaliser); + + for (i = 0, new_count = 0; i < element_count; i += register_count, elements[new_count++] = *e) + { + e = &elements[i]; + register_count = 1; + + if (e->register_index >= MAX_REG_OUTPUT) + continue; + + register_count = range_map_get_register_count(range_map, e->register_index, e->mask, normaliser); + assert(register_count != UINT8_MAX); + register_count += !register_count; + + if (register_count > 1) + { + TRACE("Merging %s, base reg %u, count %u.\n", e->semantic_name, e->register_index, register_count); + e->register_count = register_count; + } + } + element_count = new_count; + + /* Restoring the original order is required for sensible trace output. */ + qsort(elements, element_count, sizeof(elements[0]), signature_element_index_compare); + + s->element_count = element_count; + + return true; +} + +static bool sysval_semantic_is_tess_factor(enum vkd3d_shader_sysval_semantic sysval_semantic) +{ + return sysval_semantic >= VKD3D_SHADER_SV_TESS_FACTOR_QUADEDGE + && sysval_semantic <= VKD3D_SHADER_SV_TESS_FACTOR_LINEDEN; +} + +static unsigned int shader_register_normalise_arrayed_addressing(struct vkd3d_shader_register *reg, + unsigned int id_idx, unsigned int register_index) +{ + assert(id_idx < ARRAY_SIZE(reg->idx) - 1); + + /* For a relative-addressed register index, move the id up a slot to separate it from the address, + * because rel_addr can be replaced with a constant offset in some cases. */ + if (reg->idx[id_idx].rel_addr) + { + 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; + ++id_idx; + } + /* Otherwise we have no address for the arrayed register, so insert one. This happens e.g. where + * tessellation level registers are merged into an array because they're an array in SPIR-V. */ + else + { + ++id_idx; + memmove(®->idx[1], ®->idx[0], id_idx * sizeof(reg->idx[0])); + reg->idx[0].rel_addr = NULL; + reg->idx[0].offset = reg->idx[id_idx].offset - register_index; + } + + return id_idx; +} + +static bool shader_dst_param_io_normalise(struct vkd3d_shader_dst_param *dst_param, bool is_io_dcl, + struct vkd3d_shader_normaliser *normaliser) + { + unsigned int id_idx, reg_idx, write_mask, element_idx; + struct vkd3d_shader_register *reg = &dst_param->reg; + struct vkd3d_shader_dst_param **dcl_params; + const struct shader_signature *signature; + const struct signature_element *e; + + if ((reg->type == VKD3DSPR_OUTPUT && normaliser_is_in_fork_or_join_phase(normaliser)) + || reg->type == VKD3DSPR_PATCHCONST) + { + signature = normaliser->patch_constant_signature; + /* Convert patch constant outputs to the patch constant register type to avoid the need + * to convert compiler symbols when accessed as inputs in a later stage. */ + reg->type = VKD3DSPR_PATCHCONST; + dcl_params = normaliser->pc_dcl_params; + } + else if (reg->type == VKD3DSPR_OUTPUT || dst_param->reg.type == VKD3DSPR_COLOROUT) + { + signature = normaliser->output_signature; + dcl_params = normaliser->output_dcl_params; + } + else if (dst_param->reg.type == VKD3DSPR_INCONTROLPOINT || dst_param->reg.type == VKD3DSPR_INPUT) + { + signature = normaliser->input_signature; + dcl_params = normaliser->input_dcl_params; + } + else + { + return true; + } + + id_idx = reg->idx_count - 1; + reg_idx = reg->idx[id_idx].offset; + write_mask = dst_param->write_mask; + element_idx = shader_signature_find_element_for_reg(signature, reg_idx, write_mask, normaliser); + e = &signature->elements[element_idx]; + + dst_param->write_mask >>= vkd3d_write_mask_get_component_idx(e->mask); + if (is_io_dcl) + { + /* Validated in the TPF reader. */ + assert(element_idx < ARRAY_SIZE(normaliser->input_dcl_params)); + + if (dcl_params[element_idx]) + { + /* Merge split declarations into a single one. */ + dcl_params[element_idx]->write_mask |= dst_param->write_mask; + /* Turn this into a nop. */ + return false; + } + else + { + dcl_params[element_idx] = dst_param; + } + } + + if (normaliser_is_in_control_point_phase(normaliser) && reg->type == VKD3DSPR_OUTPUT) + { + if (is_io_dcl) + { + /* Emit an array size for the control points for consistency with inputs. */ + reg->idx[0].offset = normaliser->output_control_point_count; + } + else + { + /* The control point id param. */ + assert(reg->idx[0].rel_addr); + } + id_idx = 1; + } + + if ((e->register_count > 1 || sysval_semantic_is_tess_factor(e->sysval_semantic))) + { + if (is_io_dcl) + { + /* For control point I/O, idx 0 contains the control point count. + * Ensure it is moved up to the next slot. */ + reg->idx[id_idx].offset = reg->idx[0].offset; + reg->idx[0].offset = e->register_count; + ++id_idx; + } + else + { + id_idx = shader_register_normalise_arrayed_addressing(reg, id_idx, e->register_index); + } + } + + /* Replace the register index with the signature element index */ + reg->idx[id_idx].offset = element_idx; + reg->idx_count = id_idx + 1; + + return true; +} + +static void shader_src_param_io_normalise(struct vkd3d_shader_src_param *src_param, + struct vkd3d_shader_normaliser *normaliser) +{ + unsigned int i, id_idx, reg_idx, write_mask, element_idx, component_idx; + struct vkd3d_shader_register *reg = &src_param->reg; + const struct shader_signature *signature; + const struct signature_element *e; + + /* Input/output registers from one phase can be used as inputs in + * subsequent phases. Specifically: + * + * - Control phase inputs are available as "vicp" in fork and join + * phases. + * - Control phase outputs are available as "vocp" in fork and join + * phases. + * - Fork phase patch constants are available as "vpc" in join + * phases. + * + * We handle "vicp" here by converting INCONTROLPOINT src registers to + * type INPUT so they match the control phase declarations. We handle + * "vocp" by converting OUTCONTROLPOINT registers to type OUTPUT. + * Merging fork and join phases handles "vpc". */ + + switch (reg->type) + { + case VKD3DSPR_PATCHCONST: + signature = normaliser->patch_constant_signature; + break; + case VKD3DSPR_INCONTROLPOINT: + if (normaliser->shader_type == VKD3D_SHADER_TYPE_HULL) + reg->type = VKD3DSPR_INPUT; + /* fall through */ + case VKD3DSPR_INPUT: + signature = normaliser->input_signature; + break; + case VKD3DSPR_OUTCONTROLPOINT: + if (normaliser->shader_type == VKD3D_SHADER_TYPE_HULL) + reg->type = VKD3DSPR_OUTPUT; + /* fall through */ + case VKD3DSPR_OUTPUT: + signature = normaliser->output_signature; + break; + default: + return; + } + + id_idx = reg->idx_count - 1; + reg_idx = reg->idx[id_idx].offset; + write_mask = VKD3DSP_WRITEMASK_0 << vkd3d_swizzle_get_component(src_param->swizzle, 0); + element_idx = shader_signature_find_element_for_reg(signature, reg_idx, write_mask, normaliser); + + e = &signature->elements[element_idx]; + if ((e->register_count > 1 || sysval_semantic_is_tess_factor(e->sysval_semantic))) + id_idx = shader_register_normalise_arrayed_addressing(reg, id_idx, e->register_index); + reg->idx[id_idx].offset = element_idx; + reg->idx_count = id_idx + 1; + + if ((component_idx = vkd3d_write_mask_get_component_idx(e->mask))) + { + for (i = 0; i < VKD3D_VEC4_SIZE; ++i) + if (vkd3d_swizzle_get_component(src_param->swizzle, i)) + src_param->swizzle -= component_idx << VKD3D_SHADER_SWIZZLE_SHIFT(i); + } +} + +static void shader_instruction_normalise_io_params(struct vkd3d_shader_instruction *ins, + struct vkd3d_shader_normaliser *normaliser) +{ + struct vkd3d_shader_register *reg; + bool keep = true; + unsigned int i; + + switch (ins->handler_idx) + { + case VKD3DSIH_DCL_INPUT: + if (normaliser->shader_type == VKD3D_SHADER_TYPE_HULL) + { + reg = &ins->declaration.dst.reg; + /* We don't need to keep OUTCONTROLPOINT or PATCHCONST input declarations since their + * equivalents were declared earlier, but INCONTROLPOINT may be the first occurrence. */ + if (reg->type == VKD3DSPR_OUTCONTROLPOINT || reg->type == VKD3DSPR_PATCHCONST) + vkd3d_shader_instruction_make_nop(ins); + else if (reg->type == VKD3DSPR_INCONTROLPOINT) + reg->type = VKD3DSPR_INPUT; + } + /* fall through */ + case VKD3DSIH_DCL_INPUT_PS: + case VKD3DSIH_DCL_OUTPUT: + keep = shader_dst_param_io_normalise(&ins->declaration.dst, true, normaliser); + break; + case VKD3DSIH_DCL_INPUT_SGV: + case VKD3DSIH_DCL_INPUT_SIV: + case VKD3DSIH_DCL_INPUT_PS_SGV: + case VKD3DSIH_DCL_INPUT_PS_SIV: + case VKD3DSIH_DCL_OUTPUT_SIV: + keep = shader_dst_param_io_normalise(&ins->declaration.register_semantic.reg, true, + normaliser); + break; + case VKD3DSIH_HS_CONTROL_POINT_PHASE: + case VKD3DSIH_HS_FORK_PHASE: + case VKD3DSIH_HS_JOIN_PHASE: + normaliser->phase = ins->handler_idx; + memset(normaliser->input_dcl_params, 0, sizeof(normaliser->input_dcl_params)); + memset(normaliser->output_dcl_params, 0, sizeof(normaliser->output_dcl_params)); + memset(normaliser->pc_dcl_params, 0, sizeof(normaliser->pc_dcl_params)); + break; + default: + if (shader_instruction_is_dcl(ins)) + break; + for (i = 0; i < ins->dst_count; ++i) + shader_dst_param_io_normalise((struct vkd3d_shader_dst_param *)&ins->dst[i], false, normaliser); + for (i = 0; i < ins->src_count; ++i) + shader_src_param_io_normalise((struct vkd3d_shader_src_param *)&ins->src[i], normaliser); + break; + } + + if (!keep) + shader_instruction_init(ins, VKD3DSIH_NOP); +} + +enum vkd3d_result shader_normaliser_normalise_io_registers(struct vkd3d_shader_normaliser *normaliser, + enum vkd3d_shader_type shader_type, struct shader_signature *input_signature, + struct shader_signature *output_signature, struct shader_signature *patch_constant_signature) +{ + struct vkd3d_shader_instruction_array *instructions = &normaliser->instructions; + struct vkd3d_shader_instruction *ins; + bool has_control_point_phase; + unsigned int i, j; + + normaliser->phase = VKD3DSIH_INVALID; + normaliser->shader_type = shader_type; + normaliser->input_signature = input_signature; + normaliser->output_signature = output_signature; + normaliser->patch_constant_signature = patch_constant_signature; + + for (i = 0, has_control_point_phase = false; i < instructions->count; ++i) + { + ins = &instructions->elements[i]; + + switch (ins->handler_idx) + { + case VKD3DSIH_DCL_OUTPUT_CONTROL_POINT_COUNT: + normaliser->output_control_point_count = ins->declaration.count; + break; + case VKD3DSIH_DCL_INDEX_RANGE: + shader_normaliser_add_index_range(normaliser, ins); + vkd3d_shader_instruction_make_nop(ins); + break; + case VKD3DSIH_HS_CONTROL_POINT_PHASE: + has_control_point_phase = true; + /* fall through */ + case VKD3DSIH_HS_FORK_PHASE: + case VKD3DSIH_HS_JOIN_PHASE: + normaliser->phase = ins->handler_idx; + break; + default: + break; + } + } + + if (normaliser->shader_type == VKD3D_SHADER_TYPE_HULL && !has_control_point_phase) + { + /* Inputs and outputs must match for the default phase, so merge ranges must match too. */ + for (i = 0; i < MAX_REG_OUTPUT; ++i) + { + for (j = 0; j < VKD3D_VEC4_SIZE; ++j) + { + if (!normaliser->input_range_map[i][j] && normaliser->output_range_map[i][j]) + normaliser->input_range_map[i][j] = normaliser->output_range_map[i][j]; + else if (normaliser->input_range_map[i][j] && !normaliser->output_range_map[i][j]) + normaliser->output_range_map[i][j] = normaliser->input_range_map[i][j]; + else assert(normaliser->input_range_map[i][j] == normaliser->output_range_map[i][j]); + } + } + } + + if (!shader_signature_merge(input_signature, normaliser->input_range_map, false, normaliser) + || !shader_signature_merge(output_signature, normaliser->output_range_map, false, normaliser) + || !shader_signature_merge(patch_constant_signature, normaliser->pc_range_map, true, normaliser)) + { + return VKD3D_ERROR_OUT_OF_MEMORY; + } + + normaliser->phase = VKD3DSIH_INVALID; + for (i = 0; i < normaliser->instructions.count; ++i) + shader_instruction_normalise_io_params(&normaliser->instructions.elements[i], normaliser); + + return VKD3D_OK; +} + void shader_normaliser_destroy(struct vkd3d_shader_normaliser *normaliser) { shader_instruction_array_destroy(&normaliser->instructions); diff --git a/libs/vkd3d-shader/spirv.c b/libs/vkd3d-shader/spirv.c index fa4d056e..2c7c033b 100644 --- a/libs/vkd3d-shader/spirv.c +++ b/libs/vkd3d-shader/spirv.c @@ -1962,7 +1962,6 @@ struct vkd3d_symbol_register_data uint32_t member_idx; enum vkd3d_shader_component_type component_type; unsigned int write_mask; - uint32_t dcl_mask; unsigned int structure_stride; unsigned int binding_base_idx; bool is_aggregate; /* An aggregate, i.e. a structure or an array. */ @@ -2058,8 +2057,15 @@ static void vkd3d_symbol_make_register(struct vkd3d_symbol *symbol, symbol->type = VKD3D_SYMBOL_REGISTER; memset(&symbol->key, 0, sizeof(symbol->key)); symbol->key.reg.type = reg->type; - if ((vkd3d_shader_register_is_input(reg) || vkd3d_shader_register_is_output(reg)) && reg->idx[1].offset != ~0u) - symbol->key.reg.idx = reg->idx[1].offset; + if (vkd3d_shader_register_is_input(reg) || vkd3d_shader_register_is_output(reg) + || vkd3d_shader_register_is_patch_constant(reg)) + { + unsigned int i; + for (i = 2; i > 0; --i) + if (reg->idx[i].offset != ~0u) + break; + symbol->key.reg.idx = reg->idx[i].offset; + } else if (reg->type != VKD3DSPR_IMMCONSTBUFFER) symbol->key.reg.idx = reg->idx[0].offset; } @@ -2074,7 +2080,6 @@ static void vkd3d_symbol_set_register_info(struct vkd3d_symbol *symbol, symbol->info.reg.member_idx = 0; symbol->info.reg.component_type = component_type; symbol->info.reg.write_mask = write_mask; - symbol->info.reg.dcl_mask = 0; symbol->info.reg.structure_stride = 0; symbol->info.reg.binding_base_idx = 0; symbol->info.reg.is_aggregate = false; @@ -2243,9 +2248,9 @@ struct spirv_compiler const struct vkd3d_shader_spirv_target_info *spirv_target_info; bool after_declarations_section; - const struct shader_signature *input_signature; - const struct shader_signature *output_signature; - const struct shader_signature *patch_constant_signature; + struct shader_signature input_signature; + struct shader_signature output_signature; + struct shader_signature patch_constant_signature; const struct vkd3d_shader_transform_feedback_info *xfb_info; struct vkd3d_shader_output_info { @@ -2422,9 +2427,9 @@ static struct spirv_compiler *spirv_compiler_create(const struct vkd3d_shader_ve compiler->shader_type = shader_version->type; - compiler->input_signature = &shader_desc->input_signature; - compiler->output_signature = &shader_desc->output_signature; - compiler->patch_constant_signature = &shader_desc->patch_constant_signature; + compiler->input_signature = shader_desc->input_signature; + compiler->output_signature = shader_desc->output_signature; + compiler->patch_constant_signature = shader_desc->patch_constant_signature; if ((shader_interface = vkd3d_find_struct(compile_info->next, INTERFACE_INFO))) { @@ -2981,18 +2986,27 @@ static uint32_t spirv_compiler_emit_variable(struct spirv_compiler *compiler, static uint32_t spirv_compiler_emit_array_variable(struct spirv_compiler *compiler, struct vkd3d_spirv_stream *stream, SpvStorageClass storage_class, - enum vkd3d_shader_component_type component_type, unsigned int component_count, unsigned int array_length) + enum vkd3d_shader_component_type component_type, unsigned int component_count, unsigned int array_length, + unsigned int outer_array_length) { struct vkd3d_spirv_builder *builder = &compiler->spirv_builder; uint32_t type_id, length_id, ptr_type_id; - if (!array_length) + if (!array_length && !outer_array_length) return spirv_compiler_emit_variable(compiler, stream, storage_class, component_type, component_count); type_id = vkd3d_spirv_get_type_id(builder, component_type, component_count); - length_id = spirv_compiler_get_constant_uint(compiler, array_length); - type_id = vkd3d_spirv_get_op_type_array(builder, type_id, length_id); + if (array_length) + { + length_id = spirv_compiler_get_constant_uint(compiler, array_length); + type_id = vkd3d_spirv_get_op_type_array(builder, type_id, length_id); + } + if (outer_array_length) + { + length_id = spirv_compiler_get_constant_uint(compiler, outer_array_length); + type_id = vkd3d_spirv_get_op_type_array(builder, type_id, length_id); + } ptr_type_id = vkd3d_spirv_get_op_type_pointer(builder, storage_class, type_id); return vkd3d_spirv_build_op_variable(builder, stream, ptr_type_id, storage_class, 0); } @@ -3357,24 +3371,14 @@ static void spirv_compiler_emit_dereference_register(struct spirv_compiler *comp } else if (register_info->is_aggregate) { - if (reg->type == VKD3DSPR_INPUT || reg->type == VKD3DSPR_INCONTROLPOINT) - { - /* Indices for these are swapped compared to the generated SPIR-V. */ - if (reg->idx[1].offset != ~0u) - indexes[index_count++] = spirv_compiler_emit_register_addressing(compiler, ®->idx[1]); - if (reg->idx[0].offset != ~0u) - indexes[index_count++] = spirv_compiler_emit_register_addressing(compiler, ®->idx[0]); - } - else - { - struct vkd3d_shader_register_index reg_idx = reg->idx[0]; - - if (reg->idx[1].rel_addr) - FIXME("Relative addressing not implemented.\n"); - - reg_idx.offset = register_info->member_idx; - indexes[index_count++] = spirv_compiler_emit_register_addressing(compiler, ®_idx); - } + /* Indices for these are swapped compared to the generated SPIR-V. */ + if (reg->idx[2].offset != ~0u) + indexes[index_count++] = spirv_compiler_emit_register_addressing(compiler, ®->idx[1]); + if (reg->idx[1].offset != ~0u) + indexes[index_count++] = spirv_compiler_emit_register_addressing(compiler, ®->idx[0]); + if (!index_count) + /* A register sysval which is an array in SPIR-V, e.g. SAMPLEMASK. */ + indexes[index_count++] = spirv_compiler_get_constant_uint(compiler, 0); } else { @@ -4253,35 +4257,12 @@ static const struct vkd3d_spirv_builtin *vkd3d_get_spirv_builtin(const struct sp if ((builtin = get_spirv_builtin_for_register(reg_type))) return builtin; - if (sysval != VKD3D_SIV_NONE || (reg_type != VKD3DSPR_OUTPUT && reg_type != VKD3DSPR_COLOROUT)) + if (sysval != VKD3D_SIV_NONE || (reg_type != VKD3DSPR_OUTPUT && reg_type != VKD3DSPR_COLOROUT + && reg_type != VKD3DSPR_PATCHCONST)) FIXME("Unhandled builtin (register type %#x, sysval %#x).\n", reg_type, sysval); return NULL; } -static const struct signature_element *vkd3d_find_signature_element_for_reg( - const struct shader_signature *signature, unsigned int *signature_element_index, - unsigned int reg_idx, DWORD write_mask) -{ - unsigned int signature_idx; - - for (signature_idx = 0; signature_idx < signature->element_count; ++signature_idx) - { - if (signature->elements[signature_idx].register_index == reg_idx - && (signature->elements[signature_idx].mask & write_mask) == write_mask) - { - if (signature_element_index) - *signature_element_index = signature_idx; - return &signature->elements[signature_idx]; - } - } - - FIXME("Could not find shader signature element (register %u, write mask %#x).\n", - reg_idx, write_mask); - if (signature_element_index) - *signature_element_index = ~0u; - return NULL; -} - static uint32_t spirv_compiler_get_invocation_id(struct spirv_compiler *compiler) { struct vkd3d_shader_register r; @@ -4402,7 +4383,8 @@ static void spirv_compiler_decorate_xfb_output(struct spirv_compiler *compiler, } static uint32_t spirv_compiler_emit_builtin_variable(struct spirv_compiler *compiler, - const struct vkd3d_spirv_builtin *builtin, SpvStorageClass storage_class, unsigned int array_size) + const struct vkd3d_spirv_builtin *builtin, SpvStorageClass storage_class, unsigned int array_size, + unsigned int outer_array_length) { struct vkd3d_spirv_builder *builder = &compiler->spirv_builder; uint32_t id; @@ -4411,7 +4393,7 @@ static uint32_t spirv_compiler_emit_builtin_variable(struct spirv_compiler *comp id = spirv_compiler_emit_array_variable(compiler, &builder->global_stream, storage_class, - builtin->component_type, builtin->component_count, array_size); + builtin->component_type, builtin->component_count, array_size, outer_array_length); vkd3d_spirv_add_iface_variable(builder, id); spirv_compiler_decorate_builtin(compiler, id, builtin->spirv_builtin); @@ -4423,54 +4405,41 @@ static uint32_t spirv_compiler_emit_builtin_variable(struct spirv_compiler *comp return id; } -static bool needs_private_io_variable(const struct shader_signature *signature, - unsigned int reg_idx, const struct vkd3d_spirv_builtin *builtin, - unsigned int *component_count, unsigned int *out_write_mask) +static bool needs_private_io_variable(const struct vkd3d_spirv_builtin *builtin) { - unsigned int write_mask = 0; - bool have_sysval = false; - unsigned int i, count; - - /* Always use private variables for arrayed builtins. These are generally - * scalars on the D3D side, so would need extra array indices when - * accessing them. It may be feasible to insert those indices at the point - * where the builtins are used, but it's not clear it's worth the effort. */ - if (builtin && (builtin->spirv_array_size || builtin->fixup_pfn)) - return true; - - if (*component_count == VKD3D_VEC4_SIZE) - return false; - - for (i = 0, count = 0; i < signature->element_count; ++i) - { - const struct signature_element *current = &signature->elements[i]; - - if (current->register_index != reg_idx) - continue; + return builtin && builtin->fixup_pfn; +} - write_mask |= current->mask; - ++count; +static unsigned int shader_signature_next_location(const struct shader_signature *signature) +{ + unsigned int i, max_row; - if (current->sysval_semantic) - have_sysval = true; - } + if (!signature) + return 0; - if (count == 1) - return false; + for (i = 0, max_row = 0; i < signature->element_count; ++i) + max_row = max(max_row, signature->elements[i].register_index + signature->elements[i].register_count); + return max_row; +} - if (builtin || have_sysval) - return true; +static unsigned int shader_register_get_io_indices(const struct vkd3d_shader_register *reg, + unsigned int *array_size, unsigned int *outer_array_size) +{ + unsigned int i, element_idx; - if (!vkd3d_bitmask_is_contiguous(write_mask)) + *outer_array_size = 0; + *array_size = 0; + element_idx = reg->idx[0].offset; + for (i = 1; i < 3; ++i) { - FIXME("Write mask %#x is non-contiguous.\n", write_mask); - return true; + if (reg->idx[i].offset == ~0u) + break; + *outer_array_size = *array_size; + *array_size = element_idx; + element_idx = reg->idx[i].offset; } - assert(vkd3d_write_mask_component_count(write_mask) >= *component_count); - *component_count = vkd3d_write_mask_component_count(write_mask); - *out_write_mask = write_mask; - return false; + return element_idx; } static uint32_t spirv_compiler_emit_input(struct spirv_compiler *compiler, @@ -4478,6 +4447,7 @@ static uint32_t spirv_compiler_emit_input(struct spirv_compiler *compiler, enum vkd3d_shader_interpolation_mode interpolation_mode) { struct vkd3d_spirv_builder *builder = &compiler->spirv_builder; + unsigned int array_size, outer_array_size, element_idx; const struct vkd3d_shader_register *reg = &dst->reg; unsigned int component_idx, input_component_count; const struct signature_element *signature_element; @@ -4485,41 +4455,23 @@ static uint32_t spirv_compiler_emit_input(struct spirv_compiler *compiler, enum vkd3d_shader_component_type component_type; uint32_t type_id, ptr_type_id, float_type_id; const struct vkd3d_spirv_builtin *builtin; + unsigned int write_mask, reg_write_mask; struct vkd3d_symbol *symbol = NULL; uint32_t val_id, input_id, var_id; struct vkd3d_symbol reg_symbol; - struct vkd3d_symbol tmp_symbol; SpvStorageClass storage_class; struct rb_entry *entry = NULL; bool use_private_var = false; - unsigned int write_mask; - unsigned int array_size; - unsigned int reg_idx; uint32_t i, index; assert(!reg->idx[0].rel_addr); assert(!reg->idx[1].rel_addr); - if (reg->idx[1].offset != ~0u) - { - array_size = reg->idx[0].offset; - reg_idx = reg->idx[1].offset; - } - else - { - array_size = 0; - reg_idx = reg->idx[0].offset; - } - shader_signature = reg->type == VKD3DSPR_PATCHCONST - ? compiler->patch_constant_signature : compiler->input_signature; + ? &compiler->patch_constant_signature : &compiler->input_signature; - if (!(signature_element = vkd3d_find_signature_element_for_reg(shader_signature, - NULL, reg_idx, dst->write_mask))) - { - FIXME("No signature element for shader input, ignoring shader input.\n"); - return 0; - } + element_idx = shader_register_get_io_indices(reg, &array_size, &outer_array_size); + signature_element = &shader_signature->elements[element_idx]; if (compiler->shader_type == VKD3D_SHADER_TYPE_HULL && !sysval && signature_element->sysval_semantic) sysval = vkd3d_siv_from_sysval(signature_element->sysval_semantic); @@ -4541,12 +4493,16 @@ static uint32_t spirv_compiler_emit_input(struct spirv_compiler *compiler, component_idx = vkd3d_write_mask_get_component_idx(signature_element->mask); } - if (needs_private_io_variable(shader_signature, reg_idx, builtin, &input_component_count, &write_mask) - && (compiler->shader_type != VKD3D_SHADER_TYPE_HULL - || (reg->type != VKD3DSPR_INCONTROLPOINT && reg->type != VKD3DSPR_PATCHCONST))) + if (needs_private_io_variable(builtin)) + { use_private_var = true; + reg_write_mask = write_mask; + } else + { component_idx = vkd3d_write_mask_get_component_idx(write_mask); + reg_write_mask = write_mask >> component_idx; + } storage_class = SpvStorageClassInput; @@ -4554,101 +4510,59 @@ static uint32_t spirv_compiler_emit_input(struct spirv_compiler *compiler, if ((entry = rb_get(&compiler->symbol_table, ®_symbol))) { + /* Except for vicp there should be one declaration per signature element. Sources of + * duplicate declarations are: a single register split into multiple declarations having + * different components, which should have been merged, and declarations in one phase + * being repeated in another (i.e. vcp/vocp), which should have been deleted. */ + if (reg->type != VKD3DSPR_INPUT || !is_in_fork_or_join_phase(compiler)) + FIXME("Duplicate input definition found.\n"); symbol = RB_ENTRY_VALUE(entry, struct vkd3d_symbol, entry); - input_id = symbol->id; - } - else if (compiler->shader_type == VKD3D_SHADER_TYPE_HULL - && (reg->type == VKD3DSPR_INCONTROLPOINT || reg->type == VKD3DSPR_PATCHCONST)) - { - /* Input/output registers from one phase can be used as inputs in - * subsequent phases. Specifically: - * - * - Control phase inputs are available as "vicp" in fork and join - * phases. - * - Control phase outputs are available as "vocp" in fork and join - * phases. - * - Fork phase patch constants are available as "vpc" in join - * phases. - * - * We handle "vicp" and "vpc" here by creating aliases to the shader's - * global inputs and outputs. We handle "vocp" in - * spirv_compiler_leave_shader_phase(). */ - - tmp_symbol = reg_symbol; - if (reg->type == VKD3DSPR_PATCHCONST) - tmp_symbol.key.reg.type = VKD3DSPR_OUTPUT; - else - tmp_symbol.key.reg.type = VKD3DSPR_INPUT; - - if ((entry = rb_get(&compiler->symbol_table, &tmp_symbol))) - { - symbol = RB_ENTRY_VALUE(entry, struct vkd3d_symbol, entry); - tmp_symbol = *symbol; - tmp_symbol.key.reg.type = reg->type; - spirv_compiler_put_symbol(compiler, &tmp_symbol); - - input_id = symbol->id; - } - else - { - if (reg->type == VKD3DSPR_PATCHCONST) - ERR("Patch constant register %u was not declared in a previous phase.\n", reg_idx); - else - ERR("Input control point register %u was not declared in a previous phase.\n", reg_idx); - } + return symbol->id; } - if (!symbol || ~symbol->info.reg.dcl_mask & write_mask) + if (builtin) { - if (builtin) - { - input_id = spirv_compiler_emit_builtin_variable(compiler, builtin, storage_class, array_size); - if (reg->type == VKD3DSPR_PATCHCONST) - vkd3d_spirv_build_op_decorate(builder, input_id, SpvDecorationPatch, NULL, 0); - } - else - { - unsigned int location = reg_idx; - - input_id = spirv_compiler_emit_array_variable(compiler, &builder->global_stream, - storage_class, component_type, input_component_count, array_size); - vkd3d_spirv_add_iface_variable(builder, input_id); - if (reg->type == VKD3DSPR_PATCHCONST) - { - vkd3d_spirv_build_op_decorate(builder, input_id, SpvDecorationPatch, NULL, 0); - location += compiler->input_signature->element_count; - } - vkd3d_spirv_build_op_decorate1(builder, input_id, SpvDecorationLocation, location); - if (component_idx) - vkd3d_spirv_build_op_decorate1(builder, input_id, SpvDecorationComponent, component_idx); - - spirv_compiler_emit_interpolation_decorations(compiler, input_id, interpolation_mode); - } + input_id = spirv_compiler_emit_builtin_variable(compiler, builtin, storage_class, array_size, + outer_array_size); + if (reg->type == VKD3DSPR_PATCHCONST) + vkd3d_spirv_build_op_decorate(builder, input_id, SpvDecorationPatch, NULL, 0); } - - if (!symbol) + else { - var_id = input_id; - if (use_private_var) + unsigned int location = signature_element->register_index; + + input_id = spirv_compiler_emit_array_variable(compiler, &builder->global_stream, + storage_class, component_type, input_component_count, array_size, outer_array_size); + vkd3d_spirv_add_iface_variable(builder, input_id); + if (reg->type == VKD3DSPR_PATCHCONST) { - storage_class = SpvStorageClassPrivate; - var_id = spirv_compiler_emit_array_variable(compiler, &builder->global_stream, - storage_class, VKD3D_SHADER_COMPONENT_FLOAT, VKD3D_VEC4_SIZE, array_size); + vkd3d_spirv_build_op_decorate(builder, input_id, SpvDecorationPatch, NULL, 0); + location += shader_signature_next_location(&compiler->input_signature); } + vkd3d_spirv_build_op_decorate1(builder, input_id, SpvDecorationLocation, location); + if (component_idx) + vkd3d_spirv_build_op_decorate1(builder, input_id, SpvDecorationComponent, component_idx); - vkd3d_symbol_set_register_info(®_symbol, var_id, storage_class, - use_private_var ? VKD3D_SHADER_COMPONENT_FLOAT : component_type, - use_private_var ? VKD3DSP_WRITEMASK_ALL : write_mask); - reg_symbol.info.reg.dcl_mask |= write_mask; - spirv_compiler_put_symbol(compiler, ®_symbol); - - spirv_compiler_emit_register_debug_name(builder, var_id, reg); + spirv_compiler_emit_interpolation_decorations(compiler, input_id, interpolation_mode); } - else + + var_id = input_id; + if (use_private_var) { - symbol->info.reg.dcl_mask |= write_mask; + storage_class = SpvStorageClassPrivate; + var_id = spirv_compiler_emit_array_variable(compiler, &builder->global_stream, + storage_class, VKD3D_SHADER_COMPONENT_FLOAT, VKD3D_VEC4_SIZE, array_size, outer_array_size); } + vkd3d_symbol_set_register_info(®_symbol, var_id, storage_class, + use_private_var ? VKD3D_SHADER_COMPONENT_FLOAT : component_type, + use_private_var ? VKD3DSP_WRITEMASK_ALL : reg_write_mask); + reg_symbol.info.reg.is_aggregate = array_size || outer_array_size; + assert(!builtin || !builtin->spirv_array_size || use_private_var || array_size || outer_array_size); + spirv_compiler_put_symbol(compiler, ®_symbol); + + spirv_compiler_emit_register_debug_name(builder, var_id, reg); + if (use_private_var) { type_id = vkd3d_spirv_get_type_id(builder, component_type, input_component_count); @@ -4673,7 +4587,7 @@ static uint32_t spirv_compiler_emit_input(struct spirv_compiler *compiler, ptr_type_id = vkd3d_spirv_get_op_type_pointer(builder, SpvStorageClassInput, type_id); index = spirv_compiler_get_constant_uint(compiler, builtin->member_idx); val_id = vkd3d_spirv_build_op_in_bounds_access_chain1(builder, ptr_type_id, input_id, index); - dst_reg.idx[0].offset = reg_idx + i; + dst_reg.idx[0].offset = element_idx + i; } val_id = vkd3d_spirv_build_op_load(builder, type_id, val_id, SpvMemoryAccessMaskNone); @@ -4723,12 +4637,11 @@ static void spirv_compiler_emit_input_register(struct spirv_compiler *compiler, if ((entry = rb_get(&compiler->symbol_table, ®_symbol))) return; - input_id = spirv_compiler_emit_builtin_variable(compiler, builtin, SpvStorageClassInput, 0); + input_id = spirv_compiler_emit_builtin_variable(compiler, builtin, SpvStorageClassInput, 0, 0); write_mask = vkd3d_write_mask_from_component_count(builtin->component_count); vkd3d_symbol_set_register_info(®_symbol, input_id, SpvStorageClassInput, builtin->component_type, write_mask); - reg_symbol.info.reg.dcl_mask = write_mask; reg_symbol.info.reg.is_aggregate = builtin->spirv_array_size; spirv_compiler_put_symbol(compiler, ®_symbol); spirv_compiler_emit_register_debug_name(builder, input_id, reg); @@ -4797,38 +4710,10 @@ static void calculate_clip_or_cull_distance_mask(const struct signature_element *mask |= (e->mask & VKD3DSP_WRITEMASK_ALL) << (VKD3D_VEC4_SIZE * e->semantic_index); } -static uint32_t calculate_sysval_array_mask(struct spirv_compiler *compiler, - const struct shader_signature *signature, enum vkd3d_shader_input_sysval_semantic sysval) -{ - const struct vkd3d_spirv_builtin *sig_builtin; - const struct vkd3d_spirv_builtin *builtin; - const struct signature_element *e; - uint32_t signature_idx, mask = 0; - - if (!(builtin = get_spirv_builtin_for_sysval(compiler, sysval))) - { - FIXME("Unhandled sysval %#x.\n", sysval); - return 0; - } - - for (signature_idx = 0; signature_idx < signature->element_count; ++signature_idx) - { - e = &signature->elements[signature_idx]; - - sig_builtin = get_spirv_builtin_for_sysval(compiler, - vkd3d_siv_from_sysval_indexed(e->sysval_semantic, e->semantic_index)); - - if (sig_builtin && sig_builtin->spirv_builtin == builtin->spirv_builtin) - mask |= (e->mask & VKD3DSP_WRITEMASK_ALL) << (VKD3D_VEC4_SIZE * sig_builtin->member_idx); - } - - return mask; -} - /* Emits arrayed SPIR-V built-in variables. */ static void spirv_compiler_emit_shader_signature_outputs(struct spirv_compiler *compiler) { - const struct shader_signature *output_signature = compiler->output_signature; + const struct shader_signature *output_signature = &compiler->output_signature; uint32_t clip_distance_mask = 0, clip_distance_id = 0; uint32_t cull_distance_mask = 0, cull_distance_id = 0; const struct vkd3d_spirv_builtin *builtin; @@ -4858,7 +4743,7 @@ static void spirv_compiler_emit_shader_signature_outputs(struct spirv_compiler * count = vkd3d_popcount(clip_distance_mask); builtin = get_spirv_builtin_for_sysval(compiler, VKD3D_SIV_CLIP_DISTANCE); clip_distance_id = spirv_compiler_emit_builtin_variable(compiler, - builtin, SpvStorageClassOutput, count); + builtin, SpvStorageClassOutput, count, 0); } if (cull_distance_mask) @@ -4866,7 +4751,7 @@ static void spirv_compiler_emit_shader_signature_outputs(struct spirv_compiler * count = vkd3d_popcount(cull_distance_mask); builtin = get_spirv_builtin_for_sysval(compiler, VKD3D_SIV_CULL_DISTANCE); cull_distance_id = spirv_compiler_emit_builtin_variable(compiler, - builtin, SpvStorageClassOutput, count); + builtin, SpvStorageClassOutput, count, 0); } for (i = 0; i < output_signature->element_count; ++i) @@ -4913,13 +4798,12 @@ static void spirv_compiler_emit_output_register(struct spirv_compiler *compiler, return; } - output_id = spirv_compiler_emit_builtin_variable(compiler, builtin, SpvStorageClassOutput, 0); + output_id = spirv_compiler_emit_builtin_variable(compiler, builtin, SpvStorageClassOutput, 0, 0); vkd3d_symbol_make_register(®_symbol, reg); write_mask = vkd3d_write_mask_from_component_count(builtin->component_count); vkd3d_symbol_set_register_info(®_symbol, output_id, SpvStorageClassOutput, builtin->component_type, write_mask); - reg_symbol.info.reg.dcl_mask = write_mask; reg_symbol.info.reg.is_aggregate = builtin->spirv_array_size; spirv_compiler_put_symbol(compiler, ®_symbol); spirv_compiler_emit_register_execution_mode(compiler, reg); @@ -4942,7 +4826,7 @@ static uint32_t spirv_compiler_emit_shader_phase_builtin_variable(struct spirv_c if (variable_id && *variable_id) return *variable_id; - id = spirv_compiler_emit_builtin_variable(compiler, builtin, SpvStorageClassOutput, 0); + id = spirv_compiler_emit_builtin_variable(compiler, builtin, SpvStorageClassOutput, 0, 0); if (is_in_fork_or_join_phase(compiler)) vkd3d_spirv_build_op_decorate(builder, id, SpvDecorationPatch, NULL, 0); @@ -4961,36 +4845,28 @@ static void spirv_compiler_emit_output(struct spirv_compiler *compiler, enum vkd3d_shader_component_type component_type; const struct shader_signature *shader_signature; const struct vkd3d_spirv_builtin *builtin; - struct vkd3d_symbol *symbol = NULL; + unsigned int array_size, outer_array_size; + unsigned int write_mask, reg_write_mask; bool use_private_variable = false; struct vkd3d_symbol reg_symbol; SpvStorageClass storage_class; - struct rb_entry *entry = NULL; - unsigned int signature_idx; - unsigned int write_mask; - unsigned int array_size; + unsigned int element_idx; bool is_patch_constant; uint32_t id, var_id; is_patch_constant = is_in_fork_or_join_phase(compiler); - shader_signature = is_patch_constant ? compiler->patch_constant_signature : compiler->output_signature; - - array_size = is_in_control_point_phase(compiler) ? compiler->output_control_point_count : 0; + shader_signature = is_patch_constant ? &compiler->patch_constant_signature : &compiler->output_signature; - if (!(signature_element = vkd3d_find_signature_element_for_reg(shader_signature, - &signature_idx, reg->idx[0].offset, dst->write_mask))) - { - FIXME("No signature element for shader output, ignoring shader output.\n"); - return; - } + element_idx = shader_register_get_io_indices(reg, &array_size, &outer_array_size); + signature_element = &shader_signature->elements[element_idx]; builtin = vkd3d_get_spirv_builtin(compiler, dst->reg.type, sysval); write_mask = signature_element->mask; - component_idx = vkd3d_write_mask_get_component_idx(dst->write_mask); - output_component_count = vkd3d_write_mask_component_count(signature_element->mask); + component_idx = vkd3d_write_mask_get_component_idx(write_mask); + output_component_count = vkd3d_write_mask_component_count(write_mask); if (builtin) { component_type = builtin->component_type; @@ -5006,116 +4882,97 @@ static void spirv_compiler_emit_output(struct spirv_compiler *compiler, storage_class = SpvStorageClassOutput; if (get_shader_output_swizzle(compiler, signature_element->register_index) != VKD3D_SHADER_NO_SWIZZLE - || needs_private_io_variable(shader_signature, signature_element->register_index, - builtin, &output_component_count, &write_mask) - || is_patch_constant) + || (compiler->output_info[element_idx].id && compiler->output_info[element_idx].array_element_mask) + || needs_private_io_variable(builtin)) + { use_private_variable = true; + reg_write_mask = write_mask; + } else + { component_idx = vkd3d_write_mask_get_component_idx(write_mask); + reg_write_mask = write_mask >> component_idx; + } vkd3d_symbol_make_register(®_symbol, reg); - if ((entry = rb_get(&compiler->symbol_table, ®_symbol))) + if (rb_get(&compiler->symbol_table, ®_symbol)) { - symbol = RB_ENTRY_VALUE(entry, struct vkd3d_symbol, entry); - id = symbol->id; + /* See spirv_compiler_emit_input() for possible causes. */ + FIXME("Duplicate output definition found.\n"); + return; } - if (!symbol || ~symbol->info.reg.dcl_mask & write_mask) + if (compiler->output_info[element_idx].id) { - if (compiler->output_info[signature_idx].id) - { - id = compiler->output_info[signature_idx].id; - if (compiler->output_info[signature_idx].array_element_mask) - use_private_variable = true; - } - else if (builtin) - { - if (spirv_compiler_get_current_shader_phase(compiler)) - id = spirv_compiler_emit_shader_phase_builtin_variable(compiler, builtin); - else - id = spirv_compiler_emit_builtin_variable(compiler, builtin, storage_class, array_size); - - if (builtin->spirv_array_size) - compiler->output_info[signature_idx].array_element_mask = - calculate_sysval_array_mask(compiler, shader_signature, sysval); - - spirv_compiler_emit_register_execution_mode(compiler, &dst->reg); - } + id = compiler->output_info[element_idx].id; + } + else if (builtin) + { + if (spirv_compiler_get_current_shader_phase(compiler)) + id = spirv_compiler_emit_shader_phase_builtin_variable(compiler, builtin); else - { - unsigned int location = reg->idx[0].offset; + id = spirv_compiler_emit_builtin_variable(compiler, builtin, storage_class, array_size, outer_array_size); - if (is_patch_constant) - location += compiler->output_signature->element_count; + spirv_compiler_emit_register_execution_mode(compiler, &dst->reg); + } + else + { + unsigned int location = signature_element->register_index; - id = spirv_compiler_emit_array_variable(compiler, &builder->global_stream, - storage_class, component_type, output_component_count, array_size); - vkd3d_spirv_add_iface_variable(builder, id); + if (is_patch_constant) + location += shader_signature_next_location(&compiler->output_signature); - if (is_dual_source_blending(compiler) && reg->idx[0].offset < 2) - { - vkd3d_spirv_build_op_decorate1(builder, id, SpvDecorationLocation, 0); - vkd3d_spirv_build_op_decorate1(builder, id, SpvDecorationIndex, reg->idx[0].offset); - } - else - { - vkd3d_spirv_build_op_decorate1(builder, id, SpvDecorationLocation, location); - } + id = spirv_compiler_emit_array_variable(compiler, &builder->global_stream, + storage_class, component_type, output_component_count, array_size, outer_array_size); + vkd3d_spirv_add_iface_variable(builder, id); - if (component_idx) - vkd3d_spirv_build_op_decorate1(builder, id, SpvDecorationComponent, component_idx); + if (is_dual_source_blending(compiler) && signature_element->register_index < 2) + { + vkd3d_spirv_build_op_decorate1(builder, id, SpvDecorationLocation, 0); + vkd3d_spirv_build_op_decorate1(builder, id, SpvDecorationIndex, signature_element->register_index); + } + else + { + vkd3d_spirv_build_op_decorate1(builder, id, SpvDecorationLocation, location); } - if (is_patch_constant) - vkd3d_spirv_build_op_decorate(builder, id, SpvDecorationPatch, NULL, 0); - - spirv_compiler_decorate_xfb_output(compiler, id, output_component_count, signature_element); - - compiler->output_info[signature_idx].id = id; - compiler->output_info[signature_idx].component_type = component_type; + if (component_idx) + vkd3d_spirv_build_op_decorate1(builder, id, SpvDecorationComponent, component_idx); } - if (!symbol) - { - var_id = id; - if (use_private_variable) - storage_class = SpvStorageClassPrivate; - if (is_patch_constant) - var_id = compiler->hs.patch_constants_id; - else if (use_private_variable) - var_id = spirv_compiler_emit_variable(compiler, &builder->global_stream, - storage_class, VKD3D_SHADER_COMPONENT_FLOAT, VKD3D_VEC4_SIZE); + if (is_patch_constant) + vkd3d_spirv_build_op_decorate(builder, id, SpvDecorationPatch, NULL, 0); - vkd3d_symbol_set_register_info(®_symbol, var_id, storage_class, - use_private_variable ? VKD3D_SHADER_COMPONENT_FLOAT : component_type, - use_private_variable ? VKD3DSP_WRITEMASK_ALL : write_mask); - reg_symbol.info.reg.is_aggregate = use_private_variable ? is_patch_constant : array_size; - if (is_patch_constant) - { - reg_symbol.info.reg.member_idx = reg->idx[0].offset; - } - reg_symbol.info.reg.dcl_mask = write_mask; + spirv_compiler_decorate_xfb_output(compiler, id, output_component_count, signature_element); - spirv_compiler_put_symbol(compiler, ®_symbol); + compiler->output_info[element_idx].id = id; + compiler->output_info[element_idx].component_type = component_type; - if (!is_patch_constant) - spirv_compiler_emit_register_debug_name(builder, var_id, reg); - } - else + var_id = id; + if (use_private_variable) { - symbol->info.reg.dcl_mask |= write_mask; - var_id = symbol->id; + storage_class = SpvStorageClassPrivate; + var_id = spirv_compiler_emit_variable(compiler, &builder->global_stream, + storage_class, VKD3D_SHADER_COMPONENT_FLOAT, VKD3D_VEC4_SIZE); } + vkd3d_symbol_set_register_info(®_symbol, var_id, storage_class, + use_private_variable ? VKD3D_SHADER_COMPONENT_FLOAT : component_type, + use_private_variable ? VKD3DSP_WRITEMASK_ALL : reg_write_mask); + reg_symbol.info.reg.is_aggregate = array_size || outer_array_size; + assert(!builtin || !builtin->spirv_array_size || use_private_variable || array_size || outer_array_size); + + spirv_compiler_put_symbol(compiler, ®_symbol); + + if (!is_patch_constant) + spirv_compiler_emit_register_debug_name(builder, var_id, reg); + if (use_private_variable) { - unsigned int idx = spirv_compiler_get_output_variable_index(compiler, reg->idx[0].offset); + unsigned int idx = spirv_compiler_get_output_variable_index(compiler, element_idx); compiler->private_output_variable[idx] = var_id; compiler->private_output_variable_write_mask[idx] |= dst->write_mask; - if (is_patch_constant) - compiler->private_output_variable_array_idx[idx] = spirv_compiler_get_constant_uint( - compiler, reg->idx[0].offset); if (!compiler->epilogue_function_id) compiler->epilogue_function_id = vkd3d_spirv_alloc_id(builder); } @@ -5167,6 +5024,9 @@ static void spirv_compiler_emit_store_shader_output(struct spirv_compiler *compi use_mask |= element->used_mask; } } + index = vkd3d_write_mask_get_component_idx(output->mask); + dst_write_mask >>= index; + use_mask >>= index; write_mask &= dst_write_mask; if (!write_mask) @@ -5250,7 +5110,7 @@ static void spirv_compiler_emit_shader_epilogue_function(struct spirv_compiler * is_patch_constant = is_in_fork_or_join_phase(compiler); - signature = is_patch_constant ? compiler->patch_constant_signature : compiler->output_signature; + signature = is_patch_constant ? &compiler->patch_constant_signature : &compiler->output_signature; function_id = compiler->epilogue_function_id; @@ -5289,8 +5149,7 @@ static void spirv_compiler_emit_shader_epilogue_function(struct spirv_compiler * if (!compiler->output_info[i].id) continue; - variable_idx = spirv_compiler_get_output_variable_index(compiler, - signature->elements[i].register_index); + variable_idx = spirv_compiler_get_output_variable_index(compiler, i); if (!param_id[variable_idx]) continue; @@ -5320,24 +5179,6 @@ static void spirv_compiler_emit_hull_shader_builtins(struct spirv_compiler *comp spirv_compiler_emit_input_register(compiler, &dst); } -static void spirv_compiler_emit_hull_shader_patch_constants(struct spirv_compiler *compiler) -{ - const struct shader_signature *signature = compiler->patch_constant_signature; - struct vkd3d_spirv_builder *builder = &compiler->spirv_builder; - uint32_t register_count = 0; - unsigned int signature_idx; - - for (signature_idx = 0; signature_idx < signature->element_count; ++signature_idx) - register_count = max(register_count, signature->elements[signature_idx].register_index + 1); - - if (!register_count) - return; - - compiler->hs.patch_constants_id = spirv_compiler_emit_array_variable(compiler, &builder->global_stream, - SpvStorageClassPrivate, VKD3D_SHADER_COMPONENT_FLOAT, VKD3D_VEC4_SIZE, register_count); - vkd3d_spirv_build_op_name(builder, compiler->hs.patch_constants_id, "opc"); -} - static void spirv_compiler_emit_initial_declarations(struct spirv_compiler *compiler) { const struct vkd3d_shader_transform_feedback_info *xfb_info = compiler->xfb_info; @@ -5351,7 +5192,6 @@ static void spirv_compiler_emit_initial_declarations(struct spirv_compiler *comp case VKD3D_SHADER_TYPE_HULL: vkd3d_spirv_set_execution_model(builder, SpvExecutionModelTessellationControl); spirv_compiler_emit_hull_shader_builtins(compiler); - spirv_compiler_emit_hull_shader_patch_constants(compiler); break; case VKD3D_SHADER_TYPE_DOMAIN: vkd3d_spirv_set_execution_model(builder, SpvExecutionModelTessellationEvaluation); @@ -5380,8 +5220,6 @@ 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); - - spirv_compiler_emit_shader_signature_outputs(compiler); } } @@ -5468,7 +5306,7 @@ static void spirv_compiler_emit_dcl_indexable_temp(struct spirv_compiler *compil vkd3d_spirv_begin_function_stream_insertion(builder, function_location); id = spirv_compiler_emit_array_variable(compiler, &builder->function_stream, - SpvStorageClassFunction, VKD3D_SHADER_COMPONENT_FLOAT, VKD3D_VEC4_SIZE, temp->register_size); + SpvStorageClassFunction, VKD3D_SHADER_COMPONENT_FLOAT, VKD3D_VEC4_SIZE, temp->register_size, 0); spirv_compiler_emit_register_debug_name(builder, id, ®); @@ -6170,7 +6008,8 @@ static void spirv_compiler_emit_dcl_output(struct spirv_compiler *compiler, { const struct vkd3d_shader_dst_param *dst = &instruction->declaration.dst; - if (vkd3d_shader_register_is_output(&dst->reg)) + if (vkd3d_shader_register_is_output(&dst->reg) + || (is_in_fork_or_join_phase(compiler) && vkd3d_shader_register_is_patch_constant(&dst->reg))) spirv_compiler_emit_output(compiler, dst, VKD3D_SIV_NONE); else spirv_compiler_emit_output_register(compiler, dst); @@ -6188,41 +6027,6 @@ static void spirv_compiler_emit_dcl_output_siv(struct spirv_compiler *compiler, spirv_compiler_emit_output(compiler, dst, sysval); } -static bool spirv_compiler_check_index_range(struct spirv_compiler *compiler, - const struct vkd3d_shader_index_range *range) -{ - const struct vkd3d_shader_register *reg = &range->dst.reg; - struct vkd3d_shader_register_info reg_info; - struct vkd3d_shader_register current_reg; - struct vkd3d_symbol reg_symbol; - - current_reg = *reg; - vkd3d_symbol_make_register(®_symbol, ¤t_reg); - if (!spirv_compiler_get_register_info(compiler, ¤t_reg, ®_info)) - { - ERR("Failed to get register info.\n"); - return false; - } - - /* FIXME: We should check if it's an array. */ - if (!reg_info.is_aggregate) - { - FIXME("Unhandled register %#x.\n", reg->type); - return false; - } - - return true; -} - -static void spirv_compiler_emit_dcl_index_range(struct spirv_compiler *compiler, - const struct vkd3d_shader_instruction *instruction) -{ - const struct vkd3d_shader_index_range *range = &instruction->declaration.index_range; - - if (!spirv_compiler_check_index_range(compiler, range)) - FIXME("Ignoring dcl_index_range %#x %u.\n", range->dst.reg.type, range->register_count); -} - static void spirv_compiler_emit_dcl_stream(struct spirv_compiler *compiler, const struct vkd3d_shader_instruction *instruction) { @@ -6286,7 +6090,7 @@ static void spirv_compiler_emit_point_size(struct spirv_compiler *compiler) || compiler->write_tess_geom_point_size) { vkd3d_spirv_build_op_store(&compiler->spirv_builder, - spirv_compiler_emit_builtin_variable(compiler, &point_size, SpvStorageClassOutput, 0), + spirv_compiler_emit_builtin_variable(compiler, &point_size, SpvStorageClassOutput, 0, 0), spirv_compiler_get_constant_float(compiler, 1.0f), SpvMemoryAccessMaskNone); } } @@ -6422,12 +6226,8 @@ static void spirv_compiler_emit_default_control_point_phase(struct spirv_compile static void spirv_compiler_leave_shader_phase(struct spirv_compiler *compiler) { - const struct shader_signature *signature = compiler->output_signature; + const struct shader_signature *signature = &compiler->output_signature; struct vkd3d_spirv_builder *builder = &compiler->spirv_builder; - struct vkd3d_symbol reg_symbol, *symbol; - struct vkd3d_shader_register reg; - struct rb_entry *entry; - unsigned int i; if (is_in_control_point_phase(compiler) && compiler->emit_default_control_point_phase) spirv_compiler_emit_default_control_point_phase(compiler); @@ -6437,11 +6237,6 @@ static void spirv_compiler_leave_shader_phase(struct spirv_compiler *compiler) compiler->temp_id = 0; compiler->temp_count = 0; - /* - * vocp inputs in fork and join shader phases are outputs of the control - * point phase. Reinsert symbols for vocp registers while leaving the - * control point phase. - */ if (is_in_control_point_phase(compiler)) { if (compiler->epilogue_function_id) @@ -6450,42 +6245,12 @@ static void spirv_compiler_leave_shader_phase(struct spirv_compiler *compiler) spirv_compiler_emit_shader_epilogue_function(compiler); } - memset(®, 0, sizeof(reg)); - /* Fork and join phases share output registers (patch constants). * Control point phase has separate output registers. */ memset(compiler->output_info, 0, signature->element_count * sizeof(*compiler->output_info)); memset(compiler->private_output_variable, 0, sizeof(compiler->private_output_variable)); memset(compiler->private_output_variable_array_idx, 0, sizeof(compiler->private_output_variable_array_idx)); memset(compiler->private_output_variable_write_mask, 0, sizeof(compiler->private_output_variable_write_mask)); - - for (i = 0; i < signature->element_count; ++i) - { - const struct signature_element *e = &signature->elements[i]; - - reg.type = VKD3DSPR_OUTPUT; - reg.idx[0].offset = e->register_index; - reg.idx[1].offset = ~0u; - vkd3d_symbol_make_register(®_symbol, ®); - if ((entry = rb_get(&compiler->symbol_table, ®_symbol))) - { - rb_remove(&compiler->symbol_table, entry); - - symbol = RB_ENTRY_VALUE(entry, struct vkd3d_symbol, entry); - - reg.type = VKD3DSPR_OUTCONTROLPOINT; - reg.idx[1].offset = reg.idx[0].offset; - reg.idx[0].offset = compiler->output_control_point_count; - vkd3d_symbol_make_register(symbol, ®); - symbol->info.reg.is_aggregate = false; - - if (rb_put(&compiler->symbol_table, symbol, entry) == -1) - { - ERR("Failed to insert vocp symbol entry (%s).\n", debug_vkd3d_symbol(symbol)); - vkd3d_symbol_free(entry, NULL); - } - } - } } } @@ -6524,15 +6289,15 @@ static void spirv_compiler_enter_shader_phase(struct spirv_compiler *compiler, static void spirv_compiler_emit_default_control_point_phase(struct spirv_compiler *compiler) { - const struct shader_signature *output_signature = compiler->output_signature; - const struct shader_signature *input_signature = compiler->input_signature; + const struct shader_signature *output_signature = &compiler->output_signature; + const struct shader_signature *input_signature = &compiler->input_signature; struct vkd3d_spirv_builder *builder = &compiler->spirv_builder; enum vkd3d_shader_component_type component_type; struct vkd3d_shader_src_param invocation; + unsigned int component_count, array_size; struct vkd3d_shader_register input_reg; uint32_t type_id, output_ptr_type_id; uint32_t input_id, output_id, dst_id; - unsigned int component_count; uint32_t invocation_id; unsigned int i; @@ -6562,18 +6327,22 @@ static void spirv_compiler_emit_default_control_point_phase(struct spirv_compile assert(input->mask == output->mask); assert(input->component_type == output->component_type); - input_reg.idx[1].offset = input->register_index; + input_reg.idx[1].offset = i; input_id = spirv_compiler_get_register_id(compiler, &input_reg); component_type = output->component_type; component_count = vkd3d_write_mask_component_count(output->mask); - output_id = spirv_compiler_emit_array_variable(compiler, &builder->global_stream, - SpvStorageClassOutput, component_type, component_count, compiler->output_control_point_count); + type_id = vkd3d_spirv_get_type_id(builder, component_type, component_count); + if ((array_size = (input->register_count > 1) ? input->register_count : 0)) + type_id = vkd3d_spirv_get_op_type_array(builder, type_id, spirv_compiler_get_constant_uint(compiler, + array_size)); + + output_id = spirv_compiler_emit_array_variable(compiler, &builder->global_stream, SpvStorageClassOutput, + component_type, component_count, array_size, compiler->output_control_point_count); vkd3d_spirv_add_iface_variable(builder, output_id); vkd3d_spirv_build_op_decorate1(builder, output_id, SpvDecorationLocation, output->register_index); vkd3d_spirv_build_op_name(builder, output_id, "vocp%u", output->register_index); - type_id = vkd3d_spirv_get_type_id(builder, component_type, component_count); output_ptr_type_id = vkd3d_spirv_get_op_type_pointer(builder, SpvStorageClassOutput, type_id); dst_id = vkd3d_spirv_build_op_access_chain1(builder, output_ptr_type_id, output_id, invocation_id); @@ -9379,9 +9148,6 @@ static int spirv_compiler_handle_instruction(struct spirv_compiler *compiler, case VKD3DSIH_DCL_OUTPUT_SIV: spirv_compiler_emit_dcl_output_siv(compiler, instruction); break; - case VKD3DSIH_DCL_INDEX_RANGE: - spirv_compiler_emit_dcl_index_range(compiler, instruction); - break; case VKD3DSIH_DCL_STREAM: spirv_compiler_emit_dcl_stream(compiler, instruction); break; @@ -9678,17 +9444,24 @@ static int spirv_compiler_generate_spirv(struct spirv_compiler *compiler, compiler->location.column = 0; compiler->location.line = 1; - if (compiler->shader_type == VKD3D_SHADER_TYPE_HULL) + shader_normaliser_init(&normaliser, instructions, compiler->message_context); + if (compiler->shader_type == VKD3D_SHADER_TYPE_HULL + && (result = shader_normaliser_flatten_hull_shader_phases(&normaliser)) >= 0) { - shader_normaliser_init(&normaliser, instructions); - if ((result = shader_normaliser_flatten_hull_shader_phases(&normaliser)) >= 0) - result = shader_normaliser_normalise_hull_shader_control_point_io(&normaliser, - &parser->shader_desc.input_signature); - instructions = &normaliser.instructions; - - if (result >= 0 && TRACE_ON()) - vkd3d_shader_trace(instructions, &parser->shader_version); + result = shader_normaliser_normalise_hull_shader_control_point_io(&normaliser, + &parser->shader_desc.input_signature); } + if (result >= 0) + result = shader_normaliser_normalise_io_registers(&normaliser, parser->shader_version.type, + &compiler->input_signature, &compiler->output_signature, &compiler->patch_constant_signature); + + instructions = &normaliser.instructions; + + if (result >= 0 && TRACE_ON()) + vkd3d_shader_trace(instructions, &parser->shader_version); + + if (compiler->shader_type != VKD3D_SHADER_TYPE_HULL) + spirv_compiler_emit_shader_signature_outputs(compiler); for (i = 0; i < instructions->count && result >= 0; ++i) { @@ -9696,8 +9469,7 @@ static int spirv_compiler_generate_spirv(struct spirv_compiler *compiler, result = spirv_compiler_handle_instruction(compiler, &instructions->elements[i]); } - if (instructions == &normaliser.instructions) - shader_normaliser_destroy(&normaliser); + shader_normaliser_destroy(&normaliser); if (result < 0) return result; diff --git a/libs/vkd3d-shader/vkd3d_shader_private.h b/libs/vkd3d-shader/vkd3d_shader_private.h index f085deb8..4faf9551 100644 --- a/libs/vkd3d-shader/vkd3d_shader_private.h +++ b/libs/vkd3d-shader/vkd3d_shader_private.h @@ -785,6 +785,7 @@ enum vkd3d_shader_input_sysval_semantic struct signature_element { + unsigned int sort_index; const char *semantic_name; unsigned int semantic_index; unsigned int stream_index; @@ -958,6 +959,11 @@ static inline bool vkd3d_shader_register_is_output(const struct vkd3d_shader_reg return reg->type == VKD3DSPR_OUTPUT || reg->type == VKD3DSPR_COLOROUT; } +static inline bool vkd3d_shader_register_is_patch_constant(const struct vkd3d_shader_register *reg) +{ + return reg->type == VKD3DSPR_PATCHCONST; +} + struct vkd3d_shader_location { const char *source_name; @@ -1364,7 +1370,14 @@ int dxbc_writer_write(struct dxbc_writer *dxbc, struct vkd3d_shader_code *code); struct vkd3d_shader_normaliser { + struct vkd3d_shader_message_context *message_context; + bool failed; + struct vkd3d_shader_instruction_array instructions; + enum vkd3d_shader_type shader_type; + struct shader_signature *input_signature; + struct shader_signature *output_signature; + struct shader_signature *patch_constant_signature; unsigned int max_temp_count; unsigned int temp_dcl_idx; @@ -1372,15 +1385,26 @@ struct vkd3d_shader_normaliser unsigned int instance_count; unsigned int phase_body_idx; enum vkd3d_shader_opcode phase; + unsigned int output_control_point_count; struct vkd3d_shader_src_param *outpointid_param; + + struct vkd3d_shader_dst_param *input_dcl_params[MAX_REG_OUTPUT]; + struct vkd3d_shader_dst_param *output_dcl_params[MAX_REG_OUTPUT]; + struct vkd3d_shader_dst_param *pc_dcl_params[MAX_REG_OUTPUT]; + uint8_t input_range_map[MAX_REG_OUTPUT][VKD3D_VEC4_SIZE]; + uint8_t output_range_map[MAX_REG_OUTPUT][VKD3D_VEC4_SIZE]; + uint8_t pc_range_map[MAX_REG_OUTPUT][VKD3D_VEC4_SIZE]; }; void shader_normaliser_init(struct vkd3d_shader_normaliser *normaliser, - struct vkd3d_shader_instruction_array *instructions); + struct vkd3d_shader_instruction_array *instructions, struct vkd3d_shader_message_context *message_context); enum vkd3d_result shader_normaliser_flatten_hull_shader_phases(struct vkd3d_shader_normaliser *normaliser); enum vkd3d_result shader_normaliser_normalise_hull_shader_control_point_io(struct vkd3d_shader_normaliser *normaliser, const struct shader_signature *input_signature); +enum vkd3d_result shader_normaliser_normalise_io_registers(struct vkd3d_shader_normaliser *normaliser, + enum vkd3d_shader_type shader_type, struct shader_signature *input_signature, + struct shader_signature *output_signature, struct shader_signature *patch_constant_signature); void shader_normaliser_destroy(struct vkd3d_shader_normaliser *normaliser); #endif /* __VKD3D_SHADER_PRIVATE_H */ diff --git a/tests/d3d12.c b/tests/d3d12.c index 23e1e97c..1eb17a96 100644 --- a/tests/d3d12.c +++ b/tests/d3d12.c @@ -36457,7 +36457,6 @@ static void test_vs_ps_relative_addressing(void) transition_resource_state(command_list, context.render_target, D3D12_RESOURCE_STATE_RENDER_TARGET, D3D12_RESOURCE_STATE_COPY_SOURCE); - todo check_sub_resource_uint(context.render_target, 0, queue, command_list, 0xff00ff00, 0); ID3D12Resource_Release(vb); -- GitLab https://gitlab.winehq.org/wine/vkd3d/-/merge_requests/181
participants (2)
-
Conor McCarthy -
Conor McCarthy (＠cmccarthy)