vkd3d-shader/hlsl: Lower separated samplers for SM1.

The combined sampler is created as a SAMPLER instead of a TEXTURE
because that fits all our current infrastructure. The only problem is
that in the CTAB it must appear as a Texture, so the new field
hlsl_type.is_combined_sampler is added.

Co-authored-by: Elizabeth Figura <zfigura@codeweavers.com>

libs/vkd3d-shader/d3dbc.c

@@ -1570,9 +1570,14 @@ D3DXPARAMETER_CLASS hlsl_sm1_class(const struct hlsl_type *type)
     vkd3d_unreachable();
 }
 
-D3DXPARAMETER_TYPE hlsl_sm1_base_type(const struct hlsl_type *type)
+D3DXPARAMETER_TYPE hlsl_sm1_base_type(const struct hlsl_type *type, bool is_combined_sampler)
 {
-    switch (type->class)
+    enum hlsl_type_class class = type->class;
+
+    if (is_combined_sampler)
+        class = HLSL_CLASS_TEXTURE;
+
+    switch (class)
     {
         case HLSL_CLASS_SCALAR:
         case HLSL_CLASS_VECTOR:
@@ -1639,7 +1644,7 @@ D3DXPARAMETER_TYPE hlsl_sm1_base_type(const struct hlsl_type *type)
             break;
 
         case HLSL_CLASS_ARRAY:
-            return hlsl_sm1_base_type(type->e.array.type);
+            return hlsl_sm1_base_type(type->e.array.type, is_combined_sampler);
 
         case HLSL_CLASS_STRUCT:
             return D3DXPT_VOID;
@@ -1677,7 +1682,8 @@ D3DXPARAMETER_TYPE hlsl_sm1_base_type(const struct hlsl_type *type)
     vkd3d_unreachable();
 }
 
-static void write_sm1_type(struct vkd3d_bytecode_buffer *buffer, struct hlsl_type *type, unsigned int ctab_start)
+static void write_sm1_type(struct vkd3d_bytecode_buffer *buffer,
+        struct hlsl_type *type, bool is_combined_sampler, unsigned int ctab_start)
 {
     const struct hlsl_type *array_type = hlsl_get_multiarray_element_type(type);
     unsigned int array_size = hlsl_get_multiarray_size(type);
@@ -1697,7 +1703,7 @@ static void write_sm1_type(struct vkd3d_bytecode_buffer *buffer, struct hlsl_typ
             struct hlsl_struct_field *field = &array_type->e.record.fields[i];
 
             field->name_bytecode_offset = put_string(buffer, field->name);
-            write_sm1_type(buffer, field->type, ctab_start);
+            write_sm1_type(buffer, field->type, false, ctab_start);
         }
 
         fields_offset = bytecode_align(buffer) - ctab_start;
@@ -1711,7 +1717,8 @@ static void write_sm1_type(struct vkd3d_bytecode_buffer *buffer, struct hlsl_typ
         }
     }
 
-    type->bytecode_offset = put_u32(buffer, vkd3d_make_u32(hlsl_sm1_class(type), hlsl_sm1_base_type(array_type)));
+    type->bytecode_offset = put_u32(buffer,
+            vkd3d_make_u32(hlsl_sm1_class(type), hlsl_sm1_base_type(array_type, is_combined_sampler)));
     put_u32(buffer, vkd3d_make_u32(type->dimy, type->dimx));
     put_u32(buffer, vkd3d_make_u32(array_size, field_count));
     put_u32(buffer, fields_offset);
@@ -1836,7 +1843,7 @@ void write_sm1_uniforms(struct hlsl_ctx *ctx, struct vkd3d_bytecode_buffer *buff
             name_offset = put_string(buffer, var->name);
             set_u32(buffer, var_offset, name_offset - ctab_start);
 
-            write_sm1_type(buffer, var->data_type, ctab_start);
+            write_sm1_type(buffer, var->data_type, var->is_combined_sampler, ctab_start);
             set_u32(buffer, var_offset + 3 * sizeof(uint32_t), var->data_type->bytecode_offset - ctab_start);
 
             if (var->default_values)

libs/vkd3d-shader/fx.c

@@ -1021,8 +1021,8 @@ static uint32_t get_fx_2_type_class(const struct hlsl_type *type)
     return hlsl_sm1_class(type);
 }
 
-static uint32_t write_fx_2_parameter(const struct hlsl_type *type, const char *name, const struct hlsl_semantic *semantic,
+static uint32_t write_fx_2_parameter(const struct hlsl_type *type, const char *name,
-        struct fx_write_context *fx)
+        const struct hlsl_semantic *semantic, bool is_combined_sampler, struct fx_write_context *fx)
 {
     struct vkd3d_bytecode_buffer *buffer = &fx->unstructured;
     uint32_t semantic_offset, offset, elements_count = 0, name_offset;
@@ -1038,7 +1038,7 @@ static uint32_t write_fx_2_parameter(const struct hlsl_type *type, const char *n
     name_offset = write_string(name, fx);
     semantic_offset = semantic->raw_name ? write_string(semantic->raw_name, fx) : 0;
 
-    offset = put_u32(buffer, hlsl_sm1_base_type(type));
+    offset = put_u32(buffer, hlsl_sm1_base_type(type, is_combined_sampler));
     put_u32(buffer, get_fx_2_type_class(type));
     put_u32(buffer, name_offset);
     put_u32(buffer, semantic_offset);
@@ -1074,7 +1074,7 @@ static uint32_t write_fx_2_parameter(const struct hlsl_type *type, const char *n
 
             /* Validated in check_invalid_object_fields(). */
             VKD3D_ASSERT(hlsl_is_numeric_type(field->type));
-            write_fx_2_parameter(field->type, field->name, &field->semantic, fx);
+            write_fx_2_parameter(field->type, field->name, &field->semantic, false, fx);
         }
     }
 
@@ -1335,7 +1335,7 @@ static void write_fx_2_parameters(struct fx_write_context *fx)
         if (!is_type_supported_fx_2(ctx, var->data_type, &var->loc))
             continue;
 
-        desc_offset = write_fx_2_parameter(var->data_type, var->name, &var->semantic, fx);
+        desc_offset = write_fx_2_parameter(var->data_type, var->name, &var->semantic, var->is_combined_sampler, fx);
         value_offset = write_fx_2_initial_value(var, fx);
 
         flags = 0;
@@ -1358,7 +1358,7 @@ static void write_fx_2_annotation(struct hlsl_ir_var *var, struct fx_write_conte
     struct vkd3d_bytecode_buffer *buffer = &fx->structured;
     uint32_t desc_offset, value_offset;
 
-    desc_offset = write_fx_2_parameter(var->data_type, var->name, &var->semantic, fx);
+    desc_offset = write_fx_2_parameter(var->data_type, var->name, &var->semantic, var->is_combined_sampler, fx);
     value_offset = write_fx_2_initial_value(var, fx);
 
     put_u32(buffer, desc_offset);

libs/vkd3d-shader/hlsl.h

@@ -535,6 +535,10 @@ struct hlsl_ir_var
      * element of a struct, and thus needs to be aligned when packed in the signature. */
     bool force_align;
 
+    /* Whether this is a sampler that was created from the combination of a
+     * sampler and a texture for SM<4 backwards compatibility. */
+    bool is_combined_sampler;
+
     uint32_t is_input_semantic : 1;
     uint32_t is_output_semantic : 1;
     uint32_t is_uniform : 1;
@@ -1643,7 +1647,7 @@ bool hlsl_transform_ir(struct hlsl_ctx *ctx, bool (*func)(struct hlsl_ctx *ctx,
         struct hlsl_block *block, void *context);
 
 D3DXPARAMETER_CLASS hlsl_sm1_class(const struct hlsl_type *type);
-D3DXPARAMETER_TYPE hlsl_sm1_base_type(const struct hlsl_type *type);
+D3DXPARAMETER_TYPE hlsl_sm1_base_type(const struct hlsl_type *type, bool is_combined_sampler);
 
 void write_sm1_uniforms(struct hlsl_ctx *ctx, struct vkd3d_bytecode_buffer *buffer);
 int d3dbc_compile(struct vsir_program *program, uint64_t config_flags,

libs/vkd3d-shader/hlsl_codegen.c

@@ -2790,6 +2790,87 @@ static bool lower_nonconstant_array_loads(struct hlsl_ctx *ctx, struct hlsl_ir_n
 
     return true;
 }
+
+/* Lower samples from separate texture and sampler variables to samples from
+ * synthetized combined samplers. That is, translate SM4-style samples in the
+ * source to SM1-style samples in the bytecode. */
+static bool lower_separate_samples(struct hlsl_ctx *ctx, struct hlsl_ir_node *instr, void *context)
+{
+    struct hlsl_ir_var *var, *resource, *sampler;
+    struct hlsl_ir_resource_load *load;
+    struct vkd3d_string_buffer *name;
+    struct hlsl_type *sampler_type;
+    unsigned int sampler_dim;
+
+    if (instr->type != HLSL_IR_RESOURCE_LOAD)
+        return false;
+    load = hlsl_ir_resource_load(instr);
+
+    if (load->load_type != HLSL_RESOURCE_SAMPLE
+            && load->load_type != HLSL_RESOURCE_SAMPLE_LOD
+            && load->load_type != HLSL_RESOURCE_SAMPLE_LOD_BIAS)
+        return false;
+
+    if (!load->sampler.var)
+        return false;
+    resource = load->resource.var;
+    sampler = load->sampler.var;
+
+    VKD3D_ASSERT(hlsl_type_is_resource(resource->data_type));
+    VKD3D_ASSERT(hlsl_type_is_resource(sampler->data_type));
+    if (sampler->data_type->class == HLSL_CLASS_ARRAY)
+    {
+        /* Only supported by d3dcompiler if the sampler is the first component
+         * of the sampler array. */
+        hlsl_error(ctx, &instr->loc, VKD3D_SHADER_ERROR_HLSL_NOT_IMPLEMENTED,
+                "Lower separated samples with sampler arrays.");
+        return false;
+    }
+    if (resource->data_type->class == HLSL_CLASS_ARRAY)
+    {
+        hlsl_fixme(ctx, &instr->loc, "Lower separated samples with resource arrays.");
+        return false;
+    }
+    if (!resource->is_uniform)
+        return false;
+    if(!sampler->is_uniform)
+        return false;
+
+    if (!(name = hlsl_get_string_buffer(ctx)))
+        return false;
+    vkd3d_string_buffer_printf(name, "%s+%s", sampler->name, resource->name);
+    sampler_dim = hlsl_get_multiarray_element_type(resource->data_type)->sampler_dim;
+
+    TRACE("Lowering to combined sampler %s.\n", debugstr_a(name->buffer));
+
+    if (!(var = hlsl_get_var(ctx->globals, name->buffer)))
+    {
+        sampler_type = ctx->builtin_types.sampler[sampler_dim];
+
+        if (!(var = hlsl_new_synthetic_var_named(ctx, name->buffer, sampler_type, &instr->loc, false)))
+        {
+            hlsl_release_string_buffer(ctx, name);
+            return false;
+        }
+        var->storage_modifiers |= HLSL_STORAGE_UNIFORM;
+        var->is_combined_sampler = true;
+        var->is_uniform = 1;
+
+        list_remove(&var->scope_entry);
+        list_add_after(&sampler->scope_entry, &var->scope_entry);
+
+        list_add_after(&sampler->extern_entry, &var->extern_entry);
+    }
+    hlsl_release_string_buffer(ctx, name);
+
+    /* Only change the deref's var, keep the path. */
+    load->resource.var = var;
+    hlsl_cleanup_deref(&load->sampler);
+    load->sampler.var = NULL;
+
+    return true;
+}
+
 /* 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)
@@ -2901,6 +2982,27 @@ static void insert_ensuring_decreasing_bind_count(struct list *list, struct hlsl
     list_add_tail(list, &to_add->extern_entry);
 }
 
+static bool sort_synthetic_combined_samplers_first(struct hlsl_ctx *ctx)
+{
+    struct list separated_resources;
+    struct hlsl_ir_var *var, *next;
+
+    list_init(&separated_resources);
+
+    LIST_FOR_EACH_ENTRY_SAFE(var, next, &ctx->extern_vars, struct hlsl_ir_var, extern_entry)
+    {
+        if (var->is_combined_sampler)
+        {
+            list_remove(&var->extern_entry);
+            insert_ensuring_decreasing_bind_count(&separated_resources, var, HLSL_REGSET_SAMPLERS);
+        }
+    }
+
+    list_move_head(&ctx->extern_vars, &separated_resources);
+
+    return false;
+}
+
 static bool sort_synthetic_separated_samplers_first(struct hlsl_ctx *ctx)
 {
     struct list separated_resources;
@@ -10260,9 +10362,13 @@ static void process_entry_function(struct hlsl_ctx *ctx,
     lower_ir(ctx, lower_casts_to_bool, body);
     lower_ir(ctx, lower_int_dot, body);
 
+    if (hlsl_version_lt(ctx, 4, 0))
+        hlsl_transform_ir(ctx, lower_separate_samples, body, NULL);
+
     hlsl_transform_ir(ctx, validate_dereferences, body, NULL);
     hlsl_transform_ir(ctx, track_object_components_sampler_dim, body, NULL);
-    if (profile->major_version >= 4)
+
+    if (hlsl_version_ge(ctx, 4, 0))
         hlsl_transform_ir(ctx, lower_combined_samples, body, NULL);
 
     do
@@ -10270,7 +10376,10 @@ static void process_entry_function(struct hlsl_ctx *ctx,
     while (hlsl_transform_ir(ctx, dce, body, NULL));
 
     hlsl_transform_ir(ctx, track_components_usage, body, NULL);
-    sort_synthetic_separated_samplers_first(ctx);
+    if (hlsl_version_lt(ctx, 4, 0))
+        sort_synthetic_combined_samplers_first(ctx);
+    else
+        sort_synthetic_separated_samplers_first(ctx);
 
     if (profile->major_version < 4)
     {

tests/hlsl/combined-samplers.shader_test

@@ -95,8 +95,8 @@ float4 main() : sv_target
 
 [test]
 todo(msl) draw quad
-if(sm<4)  todo probe (0, 0) rgba (1, 1, 1, 11)
+if(sm<4)  probe (0, 0) rgba(1, 1, 1, 11)
-if(sm>=4) probe (0, 0) rgba (10, 10, 10, 11)
+if(sm>=4) probe (0, 0) rgba(10, 10, 10, 11)
 
 
 [pixel shader]
@@ -110,8 +110,8 @@ float4 main() : sv_target
 
 [test]
 todo(msl) draw quad
-if(sm<4)  todo probe (0, 0) rgba (1, 1, 1, 11)
+if(sm<4)  probe (0, 0) rgba(1, 1, 1, 11)
-if(sm>=4) probe (0, 0) rgba (10, 10, 10, 11)
+if(sm>=4) probe (0, 0) rgba(10, 10, 10, 11)
 
 
 [pixel shader todo(sm<4)]
@@ -202,7 +202,7 @@ float4 main() : sv_target
 
 [test]
 todo(msl) draw quad
-if(sm<4)  todo probe (0, 0) rgba(453021, 453021, 453021, 111111)
+if(sm<4)  probe (0, 0) rgba(453021, 453021, 453021, 111111)
 if(sm>=4) probe (0, 0) rgba(12333, 12333, 12333, 111111)
 
 

tests/hlsl/conditional.shader_test

@@ -229,7 +229,7 @@ size (2d, 2, 2)
 filter linear linear linear
 address clamp clamp clamp
 
-[pixel shader todo(sm<4)]
+[pixel shader]
 Texture2D tex;
 sampler sam;
 float a;
@@ -243,14 +243,14 @@ float4 main() : sv_target
 
 [test]
 uniform 0 float -2
-todo(sm<4 | msl) draw quad
+todo(msl) draw quad
 probe (0, 0) rgba (0.0, 0.0, 0.0, 4.0)
 uniform 0 float 4
-todo(sm<4 | msl) draw quad
+todo(msl) draw quad
 probe (0, 0) rgba (1.0, 2.0, 3.0, 4.0)
 
 
-[pixel shader todo(sm<4)]
+[pixel shader]
 Texture2D tex;
 sampler sam;
 float a;
@@ -267,11 +267,11 @@ float4 main() : sv_target
 
 [test]
 uniform 0 float 2
-todo(sm<4 | msl) draw quad
+todo(msl) draw quad
 probe (0, 0) rgba (0.0, 0.0, 0.0, 4.0)
 uniform 0 float 1
-todo(sm<4 | msl) draw quad
+todo(msl) draw quad
 probe (0, 0) rgba (1.0, 0.0, 0.0, 4.0)
 uniform 0 float 0
-todo(sm<4 | msl) draw quad
+todo(msl) draw quad
 probe (0, 0) rgba (1.0, 1.0, 0.0, 4.0)

tests/hlsl/sample-bias.shader_test

@@ -18,7 +18,7 @@ void main(float4 pos : position, out float2 tex : texcoord, out float4 out_pos :
     out_pos = pos;
 }
 
-[pixel shader todo(sm<4)]
+[pixel shader]
 sampler s;
 Texture2D t;
 uniform float bias;
@@ -32,16 +32,16 @@ float4 main(float2 coord : texcoord) : sv_target
 
 [test]
 uniform 0 float4 6.5 0.0 0.0 0.0
-todo(sm<4 | msl) draw quad
+todo(msl) draw quad
 probe (0, 0) rgba (10.0, 0.0, 10.0, 0.0)
 
 uniform 0 float4 7.5 0.0 0.0 0.0
-todo(sm<4 | msl) draw quad
+todo(msl) draw quad
-probe (0, 0) rgba (4.0, 0.0, 10.0, 0.0)
+todo(sm<4) probe (0, 0) rgba(4.0, 0.0, 10.0, 0.0)
 
 uniform 0 float4 8.5 0.0 0.0 0.0
-todo(sm<4 | msl) draw quad
+todo(msl) draw quad
-probe (0, 0) rgba (0.0, 0.0, 10.0, 0.0)
+todo(sm<4) probe (0, 0) rgba(0.0, 0.0, 10.0, 0.0)
 
 [require]
 shader model >= 3.0

tests/hlsl/sampler.shader_test

@@ -7,7 +7,7 @@ size (2d, 2, 2)
 0.0 0.0 0.0 0.0     0.0 0.0 0.0 0.0
 0.0 0.0 0.0 0.0     1.0 0.0 1.0 0.0
 
-[pixel shader todo(sm<4)]
+[pixel shader]
 sampler s;
 Texture2D t;
 
@@ -17,10 +17,10 @@ float4 main() : sv_target
 }
 
 [test]
-todo(sm<4 | msl) draw quad
+todo(msl) draw quad
 probe (0, 0) rgba (0.25, 0, 0.25, 0)
 
-[pixel shader todo(sm<4)]
+[pixel shader]
 SamplerState s;
 Texture2D t;
 
@@ -30,7 +30,7 @@ float4 main() : sv_target
 }
 
 [test]
-todo(sm<4 | msl) draw quad
+todo(msl) draw quad
 probe (0, 0) rgba (0.25, 0, 0.25, 0)
 
 [pixel shader fail]

tests/hlsl/static-initializer.shader_test

@@ -136,7 +136,7 @@ float4 main(Texture2D tex2) : sv_target
 }
 
 
-[pixel shader todo(sm<4)]
+[pixel shader]
 Texture2D real_tex;
 static Texture2D tex = real_tex;
 sampler sam;
@@ -147,11 +147,11 @@ float4 main() : sv_target
 }
 
 [test]
-todo(sm<4 | msl) draw quad
+todo(msl) draw quad
 probe (0, 0) rgba (1, 2, 3, 4)
 
 
-[pixel shader todo(sm<4)]
+[pixel shader]
 Texture2D real_tex;
 static Texture2D tex;
 sampler sam;
@@ -163,7 +163,7 @@ float4 main() : sv_target
 }
 
 [test]
-todo(sm<4 | msl) draw quad
+todo(msl) draw quad
 probe (0, 0) rgba (1, 2, 3, 4)