From: Zebediah Figura zfigura@codeweavers.com
Co-authored-by: Francisco Casas fcasas@codeweavers.com --- libs/vkd3d-shader/hlsl.c | 1 + libs/vkd3d-shader/hlsl.h | 1 + libs/vkd3d-shader/hlsl_codegen.c | 107 +++++++++++++++++++++-- libs/vkd3d-shader/tpf.c | 18 ++-- tests/hlsl/combined-samplers.shader_test | 26 +++--- 5 files changed, 126 insertions(+), 27 deletions(-)
diff --git a/libs/vkd3d-shader/hlsl.c b/libs/vkd3d-shader/hlsl.c index 8f2bc520..ab508502 100644 --- a/libs/vkd3d-shader/hlsl.c +++ b/libs/vkd3d-shader/hlsl.c @@ -673,6 +673,7 @@ struct hlsl_type *hlsl_new_array_type(struct hlsl_ctx *ctx, struct hlsl_type *ba type->e.array.type = basic_type; type->dimx = basic_type->dimx; type->dimy = basic_type->dimy; + type->sampler_dim = basic_type->sampler_dim; hlsl_type_calculate_reg_size(ctx, type);
list_add_tail(&ctx->types, &type->entry); diff --git a/libs/vkd3d-shader/hlsl.h b/libs/vkd3d-shader/hlsl.h index 80072d5b..ab91974a 100644 --- a/libs/vkd3d-shader/hlsl.h +++ b/libs/vkd3d-shader/hlsl.h @@ -422,6 +422,7 @@ struct hlsl_ir_var uint32_t is_output_semantic : 1; uint32_t is_uniform : 1; uint32_t is_param : 1; + uint32_t is_separated_resource : 1; };
/* Sized array of variables representing a function's parameters. */ diff --git a/libs/vkd3d-shader/hlsl_codegen.c b/libs/vkd3d-shader/hlsl_codegen.c index 11e321ee..2fa2a3f6 100644 --- a/libs/vkd3d-shader/hlsl_codegen.c +++ b/libs/vkd3d-shader/hlsl_codegen.c @@ -2067,6 +2067,97 @@ static bool lower_nonconstant_vector_derefs(struct hlsl_ctx *ctx, struct hlsl_ir return false; }
+/* Lower combined samples and sampler variables to synthesized separated textures and samplers. + * That is, translate SM1-style samples in the source to SM4-style samples in the bytecode. */ +static bool lower_combined_samples(struct hlsl_ctx *ctx, struct hlsl_ir_node *instr, void *context) +{ + struct hlsl_ir_resource_load *load; + struct vkd3d_string_buffer *name; + struct hlsl_ir_var *var; + unsigned int i; + + if (instr->type != HLSL_IR_RESOURCE_LOAD) + return false; + load = hlsl_ir_resource_load(instr); + + switch (load->load_type) + { + case HLSL_RESOURCE_LOAD: + case HLSL_RESOURCE_GATHER_RED: + case HLSL_RESOURCE_GATHER_GREEN: + case HLSL_RESOURCE_GATHER_BLUE: + case HLSL_RESOURCE_GATHER_ALPHA: + case HLSL_RESOURCE_SAMPLE_CMP: + case HLSL_RESOURCE_SAMPLE_CMP_LZ: + case HLSL_RESOURCE_SAMPLE_GRAD: + return false; + + case HLSL_RESOURCE_SAMPLE: + case HLSL_RESOURCE_SAMPLE_LOD: + case HLSL_RESOURCE_SAMPLE_LOD_BIAS: + break; + } + if (load->sampler.var) + return false; + + if (!hlsl_type_is_resource(load->resource.var->data_type)) + { + hlsl_fixme(ctx, &instr->loc, "Lower combined samplers within structs."); + return false; + } + + assert(hlsl_type_get_regset(load->resource.var->data_type) == HLSL_REGSET_SAMPLERS); + + if (!(name = hlsl_get_string_buffer(ctx))) + return false; + vkd3d_string_buffer_printf(name, "<resource>%s", load->resource.var->name); + + TRACE("Lowering to separate resource %s.\n", debugstr_a(name->buffer)); + + if (!(var = hlsl_get_var(ctx->globals, name->buffer))) + { + struct hlsl_type *texture_array_type = hlsl_new_texture_type(ctx, load->sampling_dim, + hlsl_get_vector_type(ctx, HLSL_TYPE_FLOAT, 4), 0); + + /* Create (possibly multi-dimensional) texture array type with the same dims as the sampler array. */ + struct hlsl_type *arr_type = load->resource.var->data_type; + for (i = 0; i < load->resource.path_len; ++i) + { + assert(arr_type->class == HLSL_CLASS_ARRAY); + texture_array_type = hlsl_new_array_type(ctx, texture_array_type, arr_type->e.array.elements_count); + arr_type = arr_type->e.array.type; + } + + if (!(var = hlsl_new_synthetic_var_named(ctx, name->buffer, texture_array_type, &instr->loc, false))) + { + hlsl_release_string_buffer(ctx, name); + return false; + } + var->is_uniform = 1; + var->is_separated_resource = true; + + list_add_tail(&ctx->extern_vars, &var->extern_entry); + } + hlsl_release_string_buffer(ctx, name); + + if (load->sampling_dim != var->data_type->sampler_dim) + { + hlsl_error(ctx, &load->node.loc, VKD3D_SHADER_ERROR_HLSL_INCONSISTENT_SAMPLER, + "Cannot split combined samplers from "%s" if they have different usage dimensions.", + load->resource.var->name); + hlsl_note(ctx, &var->loc, VKD3D_SHADER_LOG_ERROR, "First use as combined sampler is here."); + return false; + + } + + hlsl_copy_deref(ctx, &load->sampler, &load->resource); + load->resource.var = var; + assert(hlsl_deref_get_type(ctx, &load->resource)->base_type == HLSL_TYPE_TEXTURE); + assert(hlsl_deref_get_type(ctx, &load->sampler)->base_type == HLSL_TYPE_SAMPLER); + + return true; +} + /* Lower DIV to RCP + MUL. */ static bool lower_division(struct hlsl_ctx *ctx, struct hlsl_ir_node *instr, void *context) { @@ -3582,7 +3673,7 @@ static void validate_buffer_offsets(struct hlsl_ctx *ctx)
LIST_FOR_EACH_ENTRY(var1, &ctx->extern_vars, struct hlsl_ir_var, extern_entry) { - if (!var1->is_uniform || var1->data_type->class == HLSL_CLASS_OBJECT) + if (!var1->is_uniform || hlsl_type_is_resource(var1->data_type)) continue;
buffer = var1->buffer; @@ -3593,7 +3684,7 @@ static void validate_buffer_offsets(struct hlsl_ctx *ctx) { unsigned int var1_reg_size, var2_reg_size;
- if (!var2->is_uniform || var2->data_type->class == HLSL_CLASS_OBJECT) + if (!var2->is_uniform || hlsl_type_is_resource(var2->data_type)) continue;
if (var1 == var2 || var1->buffer != var2->buffer) @@ -3643,7 +3734,7 @@ static void allocate_buffers(struct hlsl_ctx *ctx)
LIST_FOR_EACH_ENTRY(var, &ctx->extern_vars, struct hlsl_ir_var, extern_entry) { - if (var->is_uniform && var->data_type->class != HLSL_CLASS_OBJECT) + if (var->is_uniform && !hlsl_type_is_resource(var->data_type)) { if (var->is_param) var->buffer = ctx->params_buffer; @@ -4186,6 +4277,12 @@ int hlsl_emit_bytecode(struct hlsl_ctx *ctx, struct hlsl_ir_function_decl *entry hlsl_transform_ir(ctx, lower_casts_to_bool, body, NULL); hlsl_transform_ir(ctx, lower_int_dot, body, NULL);
+ hlsl_transform_ir(ctx, validate_static_object_references, body, NULL); + hlsl_transform_ir(ctx, track_object_components_sampler_dim, body, NULL); + if (profile->major_version >= 4) + hlsl_transform_ir(ctx, lower_combined_samples, body, NULL); + hlsl_transform_ir(ctx, track_object_components_usage, body, NULL); + if (profile->major_version < 4) { hlsl_transform_ir(ctx, lower_division, body, NULL); @@ -4199,10 +4296,6 @@ int hlsl_emit_bytecode(struct hlsl_ctx *ctx, struct hlsl_ir_function_decl *entry hlsl_transform_ir(ctx, lower_abs, body, NULL); }
- hlsl_transform_ir(ctx, validate_static_object_references, body, NULL); - hlsl_transform_ir(ctx, track_object_components_sampler_dim, body, NULL); - hlsl_transform_ir(ctx, track_object_components_usage, body, NULL); - /* TODO: move forward, remove when no longer needed */ transform_derefs(ctx, replace_deref_path_with_offset, body); while (hlsl_transform_ir(ctx, hlsl_fold_constant_exprs, body, NULL)); diff --git a/libs/vkd3d-shader/tpf.c b/libs/vkd3d-shader/tpf.c index cc42bdba..801c688a 100644 --- a/libs/vkd3d-shader/tpf.c +++ b/libs/vkd3d-shader/tpf.c @@ -3005,6 +3005,13 @@ static void sm4_free_extern_resources(struct extern_resource *extern_resources, vkd3d_free(extern_resources); }
+static const char *string_skip_tag(const char *string) +{ + if (!strncmp(string, "<resource>", strlen("<resource>"))) + return string + strlen("<resource>"); + return string; +} + 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; @@ -3053,7 +3060,7 @@ static struct extern_resource *sm4_get_extern_resources(struct hlsl_ctx *ctx, un *count = 0; return NULL; } - if (!(name = hlsl_strdup(ctx, name_buffer->buffer))) + if (!(name = hlsl_strdup(ctx, string_skip_tag(name_buffer->buffer)))) { sm4_free_extern_resources(extern_resources, *count); *count = 0; @@ -3092,7 +3099,7 @@ static struct extern_resource *sm4_get_extern_resources(struct hlsl_ctx *ctx, un return NULL; }
- if (!(name = hlsl_strdup(ctx, var->name))) + if (!(name = hlsl_strdup(ctx, string_skip_tag(var->name)))) { sm4_free_extern_resources(extern_resources, *count); *count = 0; @@ -5070,11 +5077,8 @@ static void write_sm4_resource_load(struct hlsl_ctx *ctx, case HLSL_RESOURCE_SAMPLE_LOD: case HLSL_RESOURCE_SAMPLE_LOD_BIAS: case HLSL_RESOURCE_SAMPLE_GRAD: - if (!load->sampler.var) - { - hlsl_fixme(ctx, &load->node.loc, "SM4 combined sample expression."); - return; - } + /* Combined sample expressions were lowered. */ + assert(load->sampler.var); write_sm4_sample(ctx, buffer, load); break;
diff --git a/tests/hlsl/combined-samplers.shader_test b/tests/hlsl/combined-samplers.shader_test index c14de4bc..16b5438e 100644 --- a/tests/hlsl/combined-samplers.shader_test +++ b/tests/hlsl/combined-samplers.shader_test @@ -35,7 +35,7 @@ size (1, 1) 4.0 4.0 4.0 1.0
-[pixel shader todo] +[pixel shader] sampler sam;
float4 main() : sv_target @@ -44,12 +44,12 @@ float4 main() : sv_target }
[test] -todo draw quad -todo probe all rgba (0, 0, 0, 1) +draw quad +probe all rgba (0, 0, 0, 1)
% Textures for new separated samplers are allocated before regular textures. -[pixel shader todo] +[pixel shader] Texture2D tex; sampler sam;
@@ -59,11 +59,11 @@ float4 main() : sv_target }
[test] -todo draw quad +draw quad todo probe all rgba (10, 10, 10, 11)
-[pixel shader todo] +[pixel shader] Texture2D tex; sampler sam[2];
@@ -73,11 +73,11 @@ float4 main() : sv_target }
[test] -todo draw quad +draw quad todo probe all rgba (21, 21, 21, 11)
-[pixel shader todo] +[pixel shader] sampler sam0; sampler sam1; sampler sam2; @@ -89,11 +89,11 @@ float4 main() : sv_target }
[test] -todo draw quad -todo probe all rgba (12, 12, 12, 111) +draw quad +probe all rgba (12, 12, 12, 111)
-[pixel shader todo] +[pixel shader] Texture2D tex[2][2]; sampler sam;
@@ -104,13 +104,13 @@ float4 main() : sv_target }
[test] -todo draw quad +draw quad todo probe all rgba (104, 104, 104, 111)
% Sampler arrays with components that have different usage dimensions are only forbidden in SM4 upwards. % However, tex2D and tex1D are considered the same dimension for these purposes. -[pixel shader fail todo] +[pixel shader fail] sampler sam[2];
float4 main() : sv_target