#include "main/mtypes.h"
#include "glsl_symbol_table.h"
+#include "glsl_parser_extras.h"
#include "ir_optimization.h"
#include "linker.h"
#include "link_varyings.h"
#include "main/macros.h"
+#include "program/hash_table.h"
+#include "program.h"
/**
gl_shader *producer, gl_shader *consumer)
{
glsl_symbol_table parameters;
- /* FINISHME: Figure these out dynamically. */
- const char *const producer_stage = "vertex";
- const char *const consumer_stage = "fragment";
+ const char *const producer_stage =
+ _mesa_glsl_shader_target_name(producer->Type);
+ const char *const consumer_stage =
+ _mesa_glsl_shader_target_name(consumer->Type);
/* Find all shader outputs in the "producer" stage.
*/
foreach_list(node, producer->ir) {
ir_variable *const var = ((ir_instruction *) node)->as_variable();
- /* FINISHME: For geometry shaders, this should also look for inout
- * FINISHME: variables.
- */
- if ((var == NULL) || (var->mode != ir_var_out))
+ if ((var == NULL) || (var->mode != ir_var_shader_out))
continue;
parameters.add_variable(var);
foreach_list(node, consumer->ir) {
ir_variable *const input = ((ir_instruction *) node)->as_variable();
- /* FINISHME: For geometry shaders, this should also look for inout
- * FINISHME: variables.
- */
- if ((input == NULL) || (input->mode != ir_var_in))
+ if ((input == NULL) || (input->mode != ir_var_shader_in))
continue;
ir_variable *const output = parameters.get_variable(input->name);
/**
* Initialize this object based on a string that was passed to
- * glTransformFeedbackVaryings. If there is a parse error, the error is
- * reported using linker_error(), and false is returned.
+ * glTransformFeedbackVaryings.
+ *
+ * If the input is mal-formed, this call still succeeds, but it sets
+ * this->var_name to a mal-formed input, so tfeedback_decl::find_output_var()
+ * will fail to find any matching variable.
*/
-bool
+void
tfeedback_decl::init(struct gl_context *ctx, struct gl_shader_program *prog,
const void *mem_ctx, const char *input)
{
this->is_clip_distance_mesa = false;
this->skip_components = 0;
this->next_buffer_separator = false;
+ this->matched_candidate = NULL;
if (ctx->Extensions.ARB_transform_feedback3) {
/* Parse gl_NextBuffer. */
if (strcmp(input, "gl_NextBuffer") == 0) {
this->next_buffer_separator = true;
- return true;
+ return;
}
/* Parse gl_SkipComponents. */
this->skip_components = 4;
if (this->skip_components)
- return true;
+ return;
}
/* Parse a declaration. */
- const char *bracket = strrchr(input, '[');
-
- if (bracket) {
- this->var_name = ralloc_strndup(mem_ctx, input, bracket - input);
- if (sscanf(bracket, "[%u]", &this->array_subscript) != 1) {
- linker_error(prog, "Cannot parse transform feedback varying %s", input);
- return false;
- }
+ const char *base_name_end;
+ long subscript = parse_program_resource_name(input, &base_name_end);
+ this->var_name = ralloc_strndup(mem_ctx, input, base_name_end - input);
+ if (subscript >= 0) {
+ this->array_subscript = subscript;
this->is_subscripted = true;
} else {
- this->var_name = ralloc_strdup(mem_ctx, input);
this->is_subscripted = false;
}
strcmp(this->var_name, "gl_ClipDistance") == 0) {
this->is_clip_distance_mesa = true;
}
-
- return true;
}
/**
- * Assign a location for this tfeedback_decl object based on the location
- * assignment in output_var.
+ * Assign a location for this tfeedback_decl object based on the transform
+ * feedback candidate found by find_candidate.
*
* If an error occurs, the error is reported through linker_error() and false
* is returned.
*/
bool
tfeedback_decl::assign_location(struct gl_context *ctx,
- struct gl_shader_program *prog,
- ir_variable *output_var)
+ struct gl_shader_program *prog)
{
assert(this->is_varying());
- if (output_var->type->is_array()) {
+ unsigned fine_location
+ = this->matched_candidate->toplevel_var->location * 4
+ + this->matched_candidate->toplevel_var->location_frac
+ + this->matched_candidate->offset;
+
+ if (this->matched_candidate->type->is_array()) {
/* Array variable */
const unsigned matrix_cols =
- output_var->type->fields.array->matrix_columns;
+ this->matched_candidate->type->fields.array->matrix_columns;
const unsigned vector_elements =
- output_var->type->fields.array->vector_elements;
+ this->matched_candidate->type->fields.array->vector_elements;
unsigned actual_array_size = this->is_clip_distance_mesa ?
- prog->Vert.ClipDistanceArraySize : output_var->type->array_size();
+ prog->Vert.ClipDistanceArraySize :
+ this->matched_candidate->type->array_size();
if (this->is_subscripted) {
/* Check array bounds. */
actual_array_size);
return false;
}
- if (this->is_clip_distance_mesa) {
- this->location =
- output_var->location + this->array_subscript / 4;
- this->location_frac = this->array_subscript % 4;
- } else {
- unsigned fine_location
- = output_var->location * 4 + output_var->location_frac;
- unsigned array_elem_size = vector_elements * matrix_cols;
- fine_location += array_elem_size * this->array_subscript;
- this->location = fine_location / 4;
- this->location_frac = fine_location % 4;
- }
+ unsigned array_elem_size = this->is_clip_distance_mesa ?
+ 1 : vector_elements * matrix_cols;
+ fine_location += array_elem_size * this->array_subscript;
this->size = 1;
} else {
- this->location = output_var->location;
- this->location_frac = output_var->location_frac;
this->size = actual_array_size;
}
this->vector_elements = vector_elements;
if (this->is_clip_distance_mesa)
this->type = GL_FLOAT;
else
- this->type = output_var->type->fields.array->gl_type;
+ this->type = this->matched_candidate->type->fields.array->gl_type;
} else {
/* Regular variable (scalar, vector, or matrix) */
if (this->is_subscripted) {
this->orig_name, this->var_name);
return false;
}
- this->location = output_var->location;
- this->location_frac = output_var->location_frac;
this->size = 1;
- this->vector_elements = output_var->type->vector_elements;
- this->matrix_columns = output_var->type->matrix_columns;
- this->type = output_var->type->gl_type;
+ this->vector_elements = this->matched_candidate->type->vector_elements;
+ this->matrix_columns = this->matched_candidate->type->matrix_columns;
+ this->type = this->matched_candidate->type->gl_type;
}
+ this->location = fine_location / 4;
+ this->location_frac = fine_location % 4;
/* From GL_EXT_transform_feedback:
* A program will fail to link if:
}
-ir_variable *
-tfeedback_decl::find_output_var(gl_shader_program *prog,
- gl_shader *producer) const
+const tfeedback_candidate *
+tfeedback_decl::find_candidate(gl_shader_program *prog,
+ hash_table *tfeedback_candidates)
{
const char *name = this->is_clip_distance_mesa
? "gl_ClipDistanceMESA" : this->var_name;
- ir_variable *var = producer->symbols->get_variable(name);
- if (var && var->mode == ir_var_out)
- return var;
-
- /* From GL_EXT_transform_feedback:
- * A program will fail to link if:
- *
- * * any variable name specified in the <varyings> array is not
- * declared as an output in the geometry shader (if present) or
- * the vertex shader (if no geometry shader is present);
- */
- linker_error(prog, "Transform feedback varying %s undeclared.",
- this->orig_name);
- return NULL;
+ this->matched_candidate = (const tfeedback_candidate *)
+ hash_table_find(tfeedback_candidates, name);
+ if (!this->matched_candidate) {
+ /* From GL_EXT_transform_feedback:
+ * A program will fail to link if:
+ *
+ * * any variable name specified in the <varyings> array is not
+ * declared as an output in the geometry shader (if present) or
+ * the vertex shader (if no geometry shader is present);
+ */
+ linker_error(prog, "Transform feedback varying %s undeclared.",
+ this->orig_name);
+ }
+ return this->matched_candidate;
}
char **varying_names, tfeedback_decl *decls)
{
for (unsigned i = 0; i < num_names; ++i) {
- if (!decls[i].init(ctx, prog, mem_ctx, varying_names[i]))
- return false;
+ decls[i].init(ctx, prog, mem_ctx, varying_names[i]);
if (!decls[i].is_varying())
continue;
class varying_matches
{
public:
- varying_matches(bool disable_varying_packing);
+ varying_matches(bool disable_varying_packing, bool consumer_is_fs);
~varying_matches();
void record(ir_variable *producer_var, ir_variable *consumer_var);
unsigned assign_locations();
/**
* The location which has been assigned for this varying. This is
* expressed in multiples of a float, with the first generic varying
- * (i.e. the one referred to by VERT_RESULT_VAR0 or FRAG_ATTRIB_VAR0)
- * represented by the value 0.
+ * (i.e. the one referred to by VARYING_SLOT_VAR0) represented by the
+ * value 0.
*/
unsigned generic_location;
} *matches;
* it was allocated.
*/
unsigned matches_capacity;
+
+ const bool consumer_is_fs;
};
-varying_matches::varying_matches(bool disable_varying_packing)
- : disable_varying_packing(disable_varying_packing)
+varying_matches::varying_matches(bool disable_varying_packing,
+ bool consumer_is_fs)
+ : disable_varying_packing(disable_varying_packing),
+ consumer_is_fs(consumer_is_fs)
{
/* Note: this initial capacity is rather arbitrarily chosen to be large
* enough for many cases without wasting an unreasonable amount of space.
* If \c producer_var has already been paired up with a consumer_var, or
* producer_var is part of fixed pipeline functionality (and hence already has
* a location assigned), this function has no effect.
+ *
+ * Note: as a side effect this function may change the interpolation type of
+ * \c producer_var, but only when the change couldn't possibly affect
+ * rendering.
*/
void
varying_matches::record(ir_variable *producer_var, ir_variable *consumer_var)
return;
}
+ if ((consumer_var == NULL && producer_var->type->contains_integer()) ||
+ !consumer_is_fs) {
+ /* Since this varying is not being consumed by the fragment shader, its
+ * interpolation type varying cannot possibly affect rendering. Also,
+ * this variable is non-flat and is (or contains) an integer.
+ *
+ * lower_packed_varyings requires all integer varyings to flat,
+ * regardless of where they appear. We can trivially satisfy that
+ * requirement by changing the interpolation type to flat here.
+ */
+ producer_var->centroid = false;
+ producer_var->interpolation = INTERP_QUALIFIER_FLAT;
+
+ if (consumer_var) {
+ consumer_var->centroid = false;
+ consumer_var->interpolation = INTERP_QUALIFIER_FLAT;
+ }
+ }
+
if (this->num_matches == this->matches_capacity) {
this->matches_capacity *= 2;
this->matches = (match *)
unsigned
varying_matches::compute_packing_class(ir_variable *var)
{
- /* In this initial implementation we conservatively assume that variables
- * can only be packed if their base type (float/int/uint/bool) matches and
- * their interpolation and centroid qualifiers match.
+ /* Without help from the back-end, there is no way to pack together
+ * variables with different interpolation types, because
+ * lower_packed_varyings must choose exactly one interpolation type for
+ * each packed varying it creates.
+ *
+ * However, we can safely pack together floats, ints, and uints, because:
*
- * TODO: relax these restrictions when the driver back-end permits.
+ * - varyings of base type "int" and "uint" must use the "flat"
+ * interpolation type, which can only occur in GLSL 1.30 and above.
+ *
+ * - On platforms that support GLSL 1.30 and above, lower_packed_varyings
+ * can store flat floats as ints without losing any information (using
+ * the ir_unop_bitcast_* opcodes).
+ *
+ * Therefore, the packing class depends only on the interpolation type.
*/
unsigned packing_class = var->centroid ? 1 : 0;
packing_class *= 4;
packing_class += var->interpolation;
- packing_class *= GLSL_TYPE_ERROR;
- packing_class += var->type->get_scalar_type()->base_type;
return packing_class;
}
{
const glsl_type *element_type = var->type;
- /* FINISHME: Support for "varying" records in GLSL 1.50. */
while (element_type->base_type == GLSL_TYPE_ARRAY) {
element_type = element_type->fields.array;
}
- switch (element_type->vector_elements) {
+ switch (element_type->component_slots() % 4) {
case 1: return PACKING_ORDER_SCALAR;
case 2: return PACKING_ORDER_VEC2;
case 3: return PACKING_ORDER_VEC3;
- case 4: return PACKING_ORDER_VEC4;
+ case 0: return PACKING_ORDER_VEC4;
default:
assert(!"Unexpected value of vector_elements");
return PACKING_ORDER_VEC4;
{
/* Only fragment shaders will take a varying variable as an input */
if (shaderType == GL_FRAGMENT_SHADER &&
- var->mode == ir_var_in) {
+ var->mode == ir_var_shader_in) {
switch (var->location) {
- case FRAG_ATTRIB_WPOS:
- case FRAG_ATTRIB_FACE:
- case FRAG_ATTRIB_PNTC:
+ case VARYING_SLOT_POS:
+ case VARYING_SLOT_FACE:
+ case VARYING_SLOT_PNTC:
return false;
default:
return true;
}
+/**
+ * Visitor class that generates tfeedback_candidate structs describing all
+ * possible targets of transform feedback.
+ *
+ * tfeedback_candidate structs are stored in the hash table
+ * tfeedback_candidates, which is passed to the constructor. This hash table
+ * maps varying names to instances of the tfeedback_candidate struct.
+ */
+class tfeedback_candidate_generator : public program_resource_visitor
+{
+public:
+ tfeedback_candidate_generator(void *mem_ctx,
+ hash_table *tfeedback_candidates)
+ : mem_ctx(mem_ctx),
+ tfeedback_candidates(tfeedback_candidates),
+ toplevel_var(NULL),
+ varying_floats(0)
+ {
+ }
+
+ void process(ir_variable *var)
+ {
+ this->toplevel_var = var;
+ this->varying_floats = 0;
+ if (var->is_interface_instance())
+ program_resource_visitor::process(var->interface_type,
+ var->interface_type->name);
+ else
+ program_resource_visitor::process(var);
+ }
+
+private:
+ virtual void visit_field(const glsl_type *type, const char *name,
+ bool row_major)
+ {
+ assert(!type->is_record());
+ assert(!(type->is_array() && type->fields.array->is_record()));
+ assert(!type->is_interface());
+ assert(!(type->is_array() && type->fields.array->is_interface()));
+
+ (void) row_major;
+
+ tfeedback_candidate *candidate
+ = rzalloc(this->mem_ctx, tfeedback_candidate);
+ candidate->toplevel_var = this->toplevel_var;
+ candidate->type = type;
+ candidate->offset = this->varying_floats;
+ hash_table_insert(this->tfeedback_candidates, candidate,
+ ralloc_strdup(this->mem_ctx, name));
+ this->varying_floats += type->component_slots();
+ }
+
+ /**
+ * Memory context used to allocate hash table keys and values.
+ */
+ void * const mem_ctx;
+
+ /**
+ * Hash table in which tfeedback_candidate objects should be stored.
+ */
+ hash_table * const tfeedback_candidates;
+
+ /**
+ * Pointer to the toplevel variable that is being traversed.
+ */
+ ir_variable *toplevel_var;
+
+ /**
+ * Total number of varying floats that have been visited so far. This is
+ * used to determine the offset to each varying within the toplevel
+ * variable.
+ */
+ unsigned varying_floats;
+};
+
+
/**
* Assign locations for all variables that are produced in one pipeline stage
* (the "producer") and consumed in the next stage (the "consumer").
unsigned num_tfeedback_decls,
tfeedback_decl *tfeedback_decls)
{
- /* FINISHME: Set dynamically when geometry shader support is added. */
- const unsigned producer_base = VERT_RESULT_VAR0;
- const unsigned consumer_base = FRAG_ATTRIB_VAR0;
- varying_matches matches(ctx->Const.DisableVaryingPacking);
+ const unsigned producer_base = VARYING_SLOT_VAR0;
+ const unsigned consumer_base = VARYING_SLOT_VAR0;
+ varying_matches matches(ctx->Const.DisableVaryingPacking,
+ consumer && consumer->Type == GL_FRAGMENT_SHADER);
+ hash_table *tfeedback_candidates
+ = hash_table_ctor(0, hash_table_string_hash, hash_table_string_compare);
+ hash_table *consumer_inputs
+ = hash_table_ctor(0, hash_table_string_hash, hash_table_string_compare);
+ hash_table *consumer_interface_inputs
+ = hash_table_ctor(0, hash_table_string_hash, hash_table_string_compare);
/* Operate in a total of three passes.
*
* not being inputs. This lets the optimizer eliminate them.
*/
+ if (consumer) {
+ foreach_list(node, consumer->ir) {
+ ir_variable *const input_var =
+ ((ir_instruction *) node)->as_variable();
+
+ if ((input_var != NULL) && (input_var->mode == ir_var_shader_in)) {
+ if (input_var->interface_type != NULL) {
+ char *const iface_field_name =
+ ralloc_asprintf(mem_ctx, "%s.%s",
+ input_var->interface_type->name,
+ input_var->name);
+ hash_table_insert(consumer_interface_inputs, input_var,
+ iface_field_name);
+ } else {
+ hash_table_insert(consumer_inputs, input_var,
+ ralloc_strdup(mem_ctx, input_var->name));
+ }
+ }
+ }
+ }
+
foreach_list(node, producer->ir) {
ir_variable *const output_var = ((ir_instruction *) node)->as_variable();
- if ((output_var == NULL) || (output_var->mode != ir_var_out))
+ if ((output_var == NULL) || (output_var->mode != ir_var_shader_out))
continue;
- ir_variable *input_var =
- consumer ? consumer->symbols->get_variable(output_var->name) : NULL;
+ tfeedback_candidate_generator g(mem_ctx, tfeedback_candidates);
+ g.process(output_var);
+
+ ir_variable *input_var;
+ if (output_var->interface_type != NULL) {
+ char *const iface_field_name =
+ ralloc_asprintf(mem_ctx, "%s.%s",
+ output_var->interface_type->name,
+ output_var->name);
+ input_var =
+ (ir_variable *) hash_table_find(consumer_interface_inputs,
+ iface_field_name);
+ } else {
+ input_var =
+ (ir_variable *) hash_table_find(consumer_inputs, output_var->name);
+ }
- if (input_var && input_var->mode != ir_var_in)
+ if (input_var && input_var->mode != ir_var_shader_in)
input_var = NULL;
if (input_var) {
if (!tfeedback_decls[i].is_varying())
continue;
- ir_variable *output_var
- = tfeedback_decls[i].find_output_var(prog, producer);
+ const tfeedback_candidate *matched_candidate
+ = tfeedback_decls[i].find_candidate(prog, tfeedback_candidates);
- if (output_var == NULL)
+ if (matched_candidate == NULL) {
+ hash_table_dtor(tfeedback_candidates);
+ hash_table_dtor(consumer_inputs);
+ hash_table_dtor(consumer_interface_inputs);
return false;
-
- if (output_var->is_unmatched_generic_inout) {
- matches.record(output_var, NULL);
}
+
+ if (matched_candidate->toplevel_var->is_unmatched_generic_inout)
+ matches.record(matched_candidate->toplevel_var, NULL);
}
const unsigned slots_used = matches.assign_locations();
if (!tfeedback_decls[i].is_varying())
continue;
- ir_variable *output_var
- = tfeedback_decls[i].find_output_var(prog, producer);
-
- if (!tfeedback_decls[i].assign_location(ctx, prog, output_var))
+ if (!tfeedback_decls[i].assign_location(ctx, prog)) {
+ hash_table_dtor(tfeedback_candidates);
+ hash_table_dtor(consumer_inputs);
+ hash_table_dtor(consumer_interface_inputs);
return false;
+ }
}
+ hash_table_dtor(tfeedback_candidates);
+ hash_table_dtor(consumer_inputs);
+ hash_table_dtor(consumer_interface_inputs);
+
if (ctx->Const.DisableVaryingPacking) {
/* Transform feedback code assumes varyings are packed, so if the driver
* has disabled varying packing, make sure it does not support transform
*/
assert(!ctx->Extensions.EXT_transform_feedback);
} else {
- lower_packed_varyings(mem_ctx, producer_base, slots_used, ir_var_out,
- producer);
+ lower_packed_varyings(mem_ctx, producer_base, slots_used,
+ ir_var_shader_out, producer);
if (consumer) {
- lower_packed_varyings(mem_ctx, consumer_base, slots_used, ir_var_in,
- consumer);
+ lower_packed_varyings(mem_ctx, consumer_base, slots_used,
+ ir_var_shader_in, consumer);
}
}
- unsigned varying_vectors = 0;
-
if (consumer) {
foreach_list(node, consumer->ir) {
ir_variable *const var = ((ir_instruction *) node)->as_variable();
- if ((var == NULL) || (var->mode != ir_var_in))
- continue;
-
- if (var->is_unmatched_generic_inout) {
+ if (var && var->mode == ir_var_shader_in &&
+ var->is_unmatched_generic_inout) {
if (prog->Version <= 120) {
/* On page 25 (page 31 of the PDF) of the GLSL 1.20 spec:
*
* "glsl1-varying read but not written" in piglit.
*/
- linker_error(prog, "fragment shader varying %s not written "
- "by vertex shader\n.", var->name);
+ linker_error(prog, "%s shader varying %s not written "
+ "by %s shader\n.",
+ _mesa_glsl_shader_target_name(consumer->Type),
+ var->name,
+ _mesa_glsl_shader_target_name(producer->Type));
}
/* An 'in' variable is only really a shader input if its
* value is written by the previous stage.
*/
var->mode = ir_var_auto;
- } else if (is_varying_var(consumer->Type, var)) {
- /* The packing rules are used for vertex shader inputs are also
- * used for fragment shader inputs.
- */
- varying_vectors += count_attribute_slots(var->type);
}
}
}
+ return true;
+}
+
+bool
+check_against_varying_limit(struct gl_context *ctx,
+ struct gl_shader_program *prog,
+ gl_shader *consumer)
+{
+ unsigned varying_vectors = 0;
+
+ foreach_list(node, consumer->ir) {
+ ir_variable *const var = ((ir_instruction *) node)->as_variable();
+
+ if (var && var->mode == ir_var_shader_in &&
+ is_varying_var(consumer->Type, var)) {
+ /* The packing rules used for vertex shader inputs are also
+ * used for fragment shader inputs.
+ */
+ varying_vectors += count_attribute_slots(var->type);
+ }
+ }
+
if (ctx->API == API_OPENGLES2 || prog->IsES) {
if (varying_vectors > ctx->Const.MaxVarying) {
- if (ctx->Const.GLSLSkipStrictMaxVaryingLimitCheck) {
- linker_warning(prog, "shader uses too many varying vectors "
- "(%u > %u), but the driver will try to optimize "
- "them out; this is non-portable out-of-spec "
- "behavior\n",
- varying_vectors, ctx->Const.MaxVarying);
- } else {
- linker_error(prog, "shader uses too many varying vectors "
- "(%u > %u)\n",
- varying_vectors, ctx->Const.MaxVarying);
- return false;
- }
+ linker_error(prog, "shader uses too many varying vectors "
+ "(%u > %u)\n",
+ varying_vectors, ctx->Const.MaxVarying);
+ return false;
}
} else {
const unsigned float_components = varying_vectors * 4;
if (float_components > ctx->Const.MaxVarying * 4) {
- if (ctx->Const.GLSLSkipStrictMaxVaryingLimitCheck) {
- linker_warning(prog, "shader uses too many varying components "
- "(%u > %u), but the driver will try to optimize "
- "them out; this is non-portable out-of-spec "
- "behavior\n",
- float_components, ctx->Const.MaxVarying * 4);
- } else {
- linker_error(prog, "shader uses too many varying components "
- "(%u > %u)\n",
- float_components, ctx->Const.MaxVarying * 4);
- return false;
- }
+ linker_error(prog, "shader uses too many varying components "
+ "(%u > %u)\n",
+ float_components, ctx->Const.MaxVarying * 4);
+ return false;
}
}
projects / mesa.git / blobdiff