From: Francisco Casas fcasas@codeweavers.com
--- libs/vkd3d-shader/hlsl.c | 70 +++++++++++++++++++ libs/vkd3d-shader/hlsl.h | 4 ++ libs/vkd3d-shader/tpf.c | 143 ++++++++++++++++++++++++-------------- tests/cbuffer.shader_test | 6 +- 4 files changed, 167 insertions(+), 56 deletions(-)
diff --git a/libs/vkd3d-shader/hlsl.c b/libs/vkd3d-shader/hlsl.c index cd0143209..6ae6a8f77 100644 --- a/libs/vkd3d-shader/hlsl.c +++ b/libs/vkd3d-shader/hlsl.c @@ -430,6 +430,51 @@ struct hlsl_type *hlsl_type_get_component_type(struct hlsl_ctx *ctx, struct hlsl return type; }
+unsigned int hlsl_type_get_component_offset(struct hlsl_ctx *ctx, struct hlsl_type *type, + enum hlsl_regset regset, unsigned int index) +{ + struct hlsl_type *next_type; + unsigned int offset = 0; + unsigned int idx; + + while (!type_is_single_component(type)) + { + next_type = type; + idx = traverse_path_from_component_index(ctx, &next_type, &index); + + switch (type->class) + { + case HLSL_CLASS_SCALAR: + case HLSL_CLASS_VECTOR: + case HLSL_CLASS_MATRIX: + if (regset == HLSL_REGSET_NUMERIC) + offset += idx; + break; + + case HLSL_CLASS_STRUCT: + offset += type->e.record.fields[idx].reg_offset[regset]; + break; + + case HLSL_CLASS_ARRAY: + if (regset == HLSL_REGSET_NUMERIC) + offset += idx * align(type->e.array.type->reg_size[regset], 4); + else + offset += idx * type->e.array.type->reg_size[regset]; + break; + + case HLSL_CLASS_OBJECT: + assert(idx == 0); + break; + + default: + vkd3d_unreachable(); + } + type = next_type; + } + + return offset; +} + static bool init_deref(struct hlsl_ctx *ctx, struct hlsl_deref *deref, struct hlsl_ir_var *var, unsigned int path_len) { @@ -2050,6 +2095,31 @@ struct vkd3d_string_buffer *hlsl_type_to_string(struct hlsl_ctx *ctx, const stru } }
+struct vkd3d_string_buffer *hlsl_component_to_string(struct hlsl_ctx *ctx, const struct hlsl_ir_var *var, + unsigned int index) +{ + struct hlsl_type *type = var->data_type, *current_type; + struct vkd3d_string_buffer *buffer; + unsigned int element_index; + + if (!(buffer = hlsl_get_string_buffer(ctx))) + return NULL; + + vkd3d_string_buffer_printf(buffer, "%s", var->name); + + while (!type_is_single_component(type)) + { + current_type = type; + element_index = traverse_path_from_component_index(ctx, &type, &index); + if (current_type->class == HLSL_CLASS_STRUCT) + vkd3d_string_buffer_printf(buffer, ".%s", current_type->e.record.fields[element_index].name); + else + vkd3d_string_buffer_printf(buffer, "[%u]", element_index); + } + + return buffer; +} + const char *debug_hlsl_type(struct hlsl_ctx *ctx, const struct hlsl_type *type) { struct vkd3d_string_buffer *string; diff --git a/libs/vkd3d-shader/hlsl.h b/libs/vkd3d-shader/hlsl.h index 41a2b3f71..3162bca3b 100644 --- a/libs/vkd3d-shader/hlsl.h +++ b/libs/vkd3d-shader/hlsl.h @@ -1049,6 +1049,8 @@ const char *debug_hlsl_writemask(unsigned int writemask); const char *debug_hlsl_swizzle(unsigned int swizzle, unsigned int count);
struct vkd3d_string_buffer *hlsl_type_to_string(struct hlsl_ctx *ctx, const struct hlsl_type *type); +struct vkd3d_string_buffer *hlsl_component_to_string(struct hlsl_ctx *ctx, const struct hlsl_ir_var *var, + unsigned int index); struct vkd3d_string_buffer *hlsl_modifiers_to_string(struct hlsl_ctx *ctx, unsigned int modifiers); const char *hlsl_node_type_to_string(enum hlsl_ir_node_type type);
@@ -1179,6 +1181,8 @@ unsigned int hlsl_type_component_count(const struct hlsl_type *type); unsigned int hlsl_type_get_array_element_reg_size(const struct hlsl_type *type, enum hlsl_regset regset); struct hlsl_type *hlsl_type_get_component_type(struct hlsl_ctx *ctx, struct hlsl_type *type, unsigned int index); +unsigned int hlsl_type_get_component_offset(struct hlsl_ctx *ctx, struct hlsl_type *type, + enum hlsl_regset regset, unsigned int index); bool hlsl_type_is_row_major(const struct hlsl_type *type); unsigned int hlsl_type_minor_size(const struct hlsl_type *type); unsigned int hlsl_type_major_size(const struct hlsl_type *type); diff --git a/libs/vkd3d-shader/tpf.c b/libs/vkd3d-shader/tpf.c index e8e0b9a31..21270b832 100644 --- a/libs/vkd3d-shader/tpf.c +++ b/libs/vkd3d-shader/tpf.c @@ -3000,6 +3000,7 @@ static void sm4_free_extern_resources(struct extern_resource *extern_resources,
static struct extern_resource *sm4_get_extern_resources(struct hlsl_ctx *ctx, unsigned int *count) { + bool separate_components = ctx->profile->major_version == 5 && ctx->profile->minor_version == 0; struct extern_resource *extern_resources = NULL; const struct hlsl_ir_var *var; enum hlsl_regset regset; @@ -3010,38 +3011,100 @@ static struct extern_resource *sm4_get_extern_resources(struct hlsl_ctx *ctx, un
LIST_FOR_EACH_ENTRY(var, &ctx->extern_vars, struct hlsl_ir_var, extern_entry) { - if (!hlsl_type_is_resource(var->data_type)) - continue; - regset = hlsl_type_get_regset(var->data_type); - if (!var->regs[regset].allocated) - continue; - - if (!(hlsl_array_reserve(ctx, (void **)&extern_resources, &capacity, *count + 1, - sizeof(*extern_resources)))) + if (separate_components) { - sm4_free_extern_resources(extern_resources, *count); - *count = 0; - return NULL; - } + unsigned int component_count = hlsl_type_component_count(var->data_type); + unsigned int k, regset_offset;
- if (!(name = hlsl_strdup(ctx, var->name))) - { - sm4_free_extern_resources(extern_resources, *count); - *count = 0; - return NULL; + for (k = 0; k < component_count; ++k) + { + struct hlsl_type *component_type = hlsl_type_get_component_type(ctx, var->data_type, k); + struct vkd3d_string_buffer *name_buffer; + + if (!hlsl_type_is_resource(component_type)) + continue; + + regset = hlsl_type_get_regset(component_type); + regset_offset = hlsl_type_get_component_offset(ctx, var->data_type, regset, k); + + if (regset_offset > var->regs[regset].bind_count) + continue; + + if (var->objects_usage[regset][regset_offset].used) + { + if (!(hlsl_array_reserve(ctx, (void **)&extern_resources, &capacity, *count + 1, + sizeof(*extern_resources)))) + { + sm4_free_extern_resources(extern_resources, *count); + *count = 0; + return NULL; + } + + if (!(name_buffer = hlsl_component_to_string(ctx, var, k))) + { + sm4_free_extern_resources(extern_resources, *count); + *count = 0; + return NULL; + } + if (!(name = hlsl_strdup(ctx, name_buffer->buffer))) + { + sm4_free_extern_resources(extern_resources, *count); + *count = 0; + hlsl_release_string_buffer(ctx, name_buffer); + return NULL; + } + hlsl_release_string_buffer(ctx, name_buffer); + + + extern_resources[*count].var = NULL; + extern_resources[*count].name = name; + extern_resources[*count].data_type = component_type; + + extern_resources[*count].regset = regset; + extern_resources[*count].id = var->regs[regset].id + regset_offset; + extern_resources[*count].bind_count = 1; + + extern_resources[*count].user_packed = false; + + ++*count; + } + } } + else + { + if (!hlsl_type_is_resource(var->data_type)) + continue; + regset = hlsl_type_get_regset(var->data_type); + if (!var->regs[regset].allocated) + continue; + + if (!(hlsl_array_reserve(ctx, (void **)&extern_resources, &capacity, *count + 1, + sizeof(*extern_resources)))) + { + sm4_free_extern_resources(extern_resources, *count); + *count = 0; + return NULL; + } + + if (!(name = hlsl_strdup(ctx, var->name))) + { + sm4_free_extern_resources(extern_resources, *count); + *count = 0; + return NULL; + }
- extern_resources[*count].var = var; - extern_resources[*count].name = name; - extern_resources[*count].data_type = var->data_type; + extern_resources[*count].var = var; + extern_resources[*count].name = name; + extern_resources[*count].data_type = var->data_type;
- extern_resources[*count].regset = regset; - extern_resources[*count].id = var->regs[regset].id; - extern_resources[*count].bind_count = var->regs[regset].bind_count; + extern_resources[*count].regset = regset; + extern_resources[*count].id = var->regs[regset].id; + extern_resources[*count].bind_count = var->regs[regset].bind_count;
- extern_resources[*count].user_packed = !!var->reg_reservation.reg_type; + extern_resources[*count].user_packed = !!var->reg_reservation.reg_type;
- ++*count; + ++*count; + } }
qsort(extern_resources, *count, sizeof(*extern_resources), sm4_compare_extern_resources); @@ -4871,33 +4934,15 @@ static void write_sm4_gather(struct hlsl_ctx *ctx, struct vkd3d_bytecode_buffer static void write_sm4_resource_load(struct hlsl_ctx *ctx, struct vkd3d_bytecode_buffer *buffer, const struct hlsl_ir_resource_load *load) { - const struct hlsl_type *resource_type = load->resource.var->data_type; const struct hlsl_ir_node *texel_offset = load->texel_offset.node; const struct hlsl_ir_node *coords = load->coords.node;
- if (!hlsl_type_is_resource(resource_type)) + if (load->sampler.var && !load->sampler.var->is_uniform) { - hlsl_fixme(ctx, &load->node.loc, "Separate object fields as new variables."); + hlsl_fixme(ctx, &load->node.loc, "Sample using non-uniform sampler variable."); return; }
- if (load->sampler.var) - { - const struct hlsl_type *sampler_type = load->sampler.var->data_type; - - if (!hlsl_type_is_resource(sampler_type)) - { - hlsl_fixme(ctx, &load->node.loc, "Separate object fields as new variables."); - return; - } - - if (!load->sampler.var->is_uniform) - { - hlsl_fixme(ctx, &load->node.loc, "Sample using non-uniform sampler variable."); - return; - } - } - if (!load->resource.var->is_uniform) { hlsl_fixme(ctx, &load->node.loc, "Load from non-uniform resource variable."); @@ -4950,14 +4995,6 @@ static void write_sm4_resource_load(struct hlsl_ctx *ctx, static void write_sm4_resource_store(struct hlsl_ctx *ctx, struct vkd3d_bytecode_buffer *buffer, const struct hlsl_ir_resource_store *store) { - const struct hlsl_type *resource_type = store->resource.var->data_type; - - if (!hlsl_type_is_resource(resource_type)) - { - hlsl_fixme(ctx, &store->node.loc, "Separate object fields as new variables."); - return; - } - if (!store->resource.var->is_uniform) { hlsl_fixme(ctx, &store->node.loc, "Store to non-uniform resource variable."); diff --git a/tests/cbuffer.shader_test b/tests/cbuffer.shader_test index 7e2a91dca..83397c189 100644 --- a/tests/cbuffer.shader_test +++ b/tests/cbuffer.shader_test @@ -694,7 +694,7 @@ shader model >= 5.0 size (1, 1) 0.0 0.0 0.0 0.5
-[pixel shader todo] +[pixel shader] struct apple { float2 a; @@ -718,5 +718,5 @@ uniform 0 float4 0.0 1.0 2.0 3.0 uniform 4 float4 4.0 5.0 6.0 7.0 uniform 8 float4 8.0 9.0 10.0 11.0 uniform 12 float4 12.0 13.0 14.0 15.0 -todo draw quad -todo probe all rgba (124.0, 135.0, 146.0, 150.5) +draw quad +probe all rgba (124.0, 135.0, 146.0, 150.5)