mirror of
https://gitlab.freedesktop.org/mesa/mesa.git
synced 2025-12-30 18:40:13 +01:00
77f5221233
The main motivation for this is that threaded compilation can fall
over if we were to allocate IR inside constant_expression_value()
when calling it on a builtin. This is because builtins are shared
across the whole OpenGL context.
f81ede4699 worked around the problem by cloning the entire
builtin before constant_expression_value() could be called on
it. However cloning the whole function each time we referenced
it lead to a significant reduction in the GLSL IR compiler
performance. This change along with the following patch
helps fix that performance regression.
Other advantages are that we reduce the number of calls to
ralloc_parent(), and for loop unrolling we free constants after
they are used rather than leaving them hanging around.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
497 lines
20 KiB
C++
497 lines
20 KiB
C++
/*
|
|
* Copyright (c) 2015 Intel Corporation
|
|
*
|
|
* Permission is hereby granted, free of charge, to any person obtaining a
|
|
* copy of this software and associated documentation files (the "Software"),
|
|
* to deal in the Software without restriction, including without limitation
|
|
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
|
|
* and/or sell copies of the Software, and to permit persons to whom the
|
|
* Software is furnished to do so, subject to the following conditions:
|
|
*
|
|
* The above copyright notice and this permission notice (including the next
|
|
* paragraph) shall be included in all copies or substantial portions of the
|
|
* Software.
|
|
*
|
|
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
|
|
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
|
|
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
|
|
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
|
|
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
|
|
* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
|
|
* DEALINGS IN THE SOFTWARE.
|
|
*/
|
|
|
|
/**
|
|
* \file lower_buffer_access.cpp
|
|
*
|
|
* Helper for IR lowering pass to replace dereferences of buffer object based
|
|
* shader variables with intrinsic function calls.
|
|
*
|
|
* This helper is used by lowering passes for UBOs, SSBOs and compute shader
|
|
* shared variables.
|
|
*/
|
|
|
|
#include "lower_buffer_access.h"
|
|
#include "ir_builder.h"
|
|
#include "main/macros.h"
|
|
#include "util/list.h"
|
|
#include "glsl_parser_extras.h"
|
|
|
|
using namespace ir_builder;
|
|
|
|
namespace lower_buffer_access {
|
|
|
|
static inline int
|
|
writemask_for_size(unsigned n)
|
|
{
|
|
return ((1 << n) - 1);
|
|
}
|
|
|
|
/**
|
|
* Takes a deref and recursively calls itself to break the deref down to the
|
|
* point that the reads or writes generated are contiguous scalars or vectors.
|
|
*/
|
|
void
|
|
lower_buffer_access::emit_access(void *mem_ctx,
|
|
bool is_write,
|
|
ir_dereference *deref,
|
|
ir_variable *base_offset,
|
|
unsigned int deref_offset,
|
|
bool row_major,
|
|
int matrix_columns,
|
|
unsigned int packing,
|
|
unsigned int write_mask)
|
|
{
|
|
if (deref->type->is_record()) {
|
|
unsigned int field_offset = 0;
|
|
|
|
for (unsigned i = 0; i < deref->type->length; i++) {
|
|
const struct glsl_struct_field *field =
|
|
&deref->type->fields.structure[i];
|
|
ir_dereference *field_deref =
|
|
new(mem_ctx) ir_dereference_record(deref->clone(mem_ctx, NULL),
|
|
field->name);
|
|
|
|
field_offset =
|
|
glsl_align(field_offset,
|
|
field->type->std140_base_alignment(row_major));
|
|
|
|
emit_access(mem_ctx, is_write, field_deref, base_offset,
|
|
deref_offset + field_offset,
|
|
row_major, 1, packing,
|
|
writemask_for_size(field_deref->type->vector_elements));
|
|
|
|
field_offset += field->type->std140_size(row_major);
|
|
}
|
|
return;
|
|
}
|
|
|
|
if (deref->type->is_array()) {
|
|
unsigned array_stride = packing == GLSL_INTERFACE_PACKING_STD430 ?
|
|
deref->type->fields.array->std430_array_stride(row_major) :
|
|
glsl_align(deref->type->fields.array->std140_size(row_major), 16);
|
|
|
|
for (unsigned i = 0; i < deref->type->length; i++) {
|
|
ir_constant *element = new(mem_ctx) ir_constant(i);
|
|
ir_dereference *element_deref =
|
|
new(mem_ctx) ir_dereference_array(deref->clone(mem_ctx, NULL),
|
|
element);
|
|
emit_access(mem_ctx, is_write, element_deref, base_offset,
|
|
deref_offset + i * array_stride,
|
|
row_major, 1, packing,
|
|
writemask_for_size(element_deref->type->vector_elements));
|
|
}
|
|
return;
|
|
}
|
|
|
|
if (deref->type->is_matrix()) {
|
|
for (unsigned i = 0; i < deref->type->matrix_columns; i++) {
|
|
ir_constant *col = new(mem_ctx) ir_constant(i);
|
|
ir_dereference *col_deref =
|
|
new(mem_ctx) ir_dereference_array(deref->clone(mem_ctx, NULL), col);
|
|
|
|
if (row_major) {
|
|
/* For a row-major matrix, the next column starts at the next
|
|
* element.
|
|
*/
|
|
int size_mul = deref->type->is_64bit() ? 8 : 4;
|
|
emit_access(mem_ctx, is_write, col_deref, base_offset,
|
|
deref_offset + i * size_mul,
|
|
row_major, deref->type->matrix_columns, packing,
|
|
writemask_for_size(col_deref->type->vector_elements));
|
|
} else {
|
|
int size_mul;
|
|
|
|
/* std430 doesn't round up vec2 size to a vec4 size */
|
|
if (packing == GLSL_INTERFACE_PACKING_STD430 &&
|
|
deref->type->vector_elements == 2 &&
|
|
!deref->type->is_64bit()) {
|
|
size_mul = 8;
|
|
} else {
|
|
/* std140 always rounds the stride of arrays (and matrices) to a
|
|
* vec4, so matrices are always 16 between columns/rows. With
|
|
* doubles, they will be 32 apart when there are more than 2 rows.
|
|
*
|
|
* For both std140 and std430, if the member is a
|
|
* three-'component vector with components consuming N basic
|
|
* machine units, the base alignment is 4N. For vec4, base
|
|
* alignment is 4N.
|
|
*/
|
|
size_mul = (deref->type->is_64bit() &&
|
|
deref->type->vector_elements > 2) ? 32 : 16;
|
|
}
|
|
|
|
emit_access(mem_ctx, is_write, col_deref, base_offset,
|
|
deref_offset + i * size_mul,
|
|
row_major, deref->type->matrix_columns, packing,
|
|
writemask_for_size(col_deref->type->vector_elements));
|
|
}
|
|
}
|
|
return;
|
|
}
|
|
|
|
assert(deref->type->is_scalar() || deref->type->is_vector());
|
|
|
|
if (!row_major) {
|
|
ir_rvalue *offset =
|
|
add(base_offset, new(mem_ctx) ir_constant(deref_offset));
|
|
unsigned mask =
|
|
is_write ? write_mask : (1 << deref->type->vector_elements) - 1;
|
|
insert_buffer_access(mem_ctx, deref, deref->type, offset, mask, -1);
|
|
} else {
|
|
unsigned N = deref->type->is_64bit() ? 8 : 4;
|
|
|
|
/* We're dereffing a column out of a row-major matrix, so we
|
|
* gather the vector from each stored row.
|
|
*/
|
|
assert(deref->type->is_float() || deref->type->is_double());
|
|
|
|
/* Matrices, row_major or not, are stored as if they were
|
|
* arrays of vectors of the appropriate size in std140.
|
|
* Arrays have their strides rounded up to a vec4, so the
|
|
* matrix stride is always 16. However a double matrix may either be 16
|
|
* or 32 depending on the number of columns.
|
|
*/
|
|
assert(matrix_columns <= 4);
|
|
unsigned matrix_stride = 0;
|
|
/* Matrix stride for std430 mat2xY matrices are not rounded up to
|
|
* vec4 size. From OpenGL 4.3 spec, section 7.6.2.2 "Standard Uniform
|
|
* Block Layout":
|
|
*
|
|
* "2. If the member is a two- or four-component vector with components
|
|
* consuming N basic machine units, the base alignment is 2N or 4N,
|
|
* respectively." [...]
|
|
* "4. If the member is an array of scalars or vectors, the base alignment
|
|
* and array stride are set to match the base alignment of a single array
|
|
* element, according to rules (1), (2), and (3), and rounded up to the
|
|
* base alignment of a vec4." [...]
|
|
* "7. If the member is a row-major matrix with C columns and R rows, the
|
|
* matrix is stored identically to an array of R row vectors with C
|
|
* components each, according to rule (4)." [...]
|
|
* "When using the std430 storage layout, shader storage blocks will be
|
|
* laid out in buffer storage identically to uniform and shader storage
|
|
* blocks using the std140 layout, except that the base alignment and
|
|
* stride of arrays of scalars and vectors in rule 4 and of structures in
|
|
* rule 9 are not rounded up a multiple of the base alignment of a vec4."
|
|
*/
|
|
if (packing == GLSL_INTERFACE_PACKING_STD430 && matrix_columns == 2)
|
|
matrix_stride = 2 * N;
|
|
else
|
|
matrix_stride = glsl_align(matrix_columns * N, 16);
|
|
|
|
const glsl_type *deref_type = deref->type->is_float() ?
|
|
glsl_type::float_type : glsl_type::double_type;
|
|
|
|
for (unsigned i = 0; i < deref->type->vector_elements; i++) {
|
|
ir_rvalue *chan_offset =
|
|
add(base_offset,
|
|
new(mem_ctx) ir_constant(deref_offset + i * matrix_stride));
|
|
if (!is_write || ((1U << i) & write_mask))
|
|
insert_buffer_access(mem_ctx, deref, deref_type, chan_offset,
|
|
(1U << i), i);
|
|
}
|
|
}
|
|
}
|
|
|
|
/**
|
|
* Determine if a thing being dereferenced is row-major
|
|
*
|
|
* There is some trickery here.
|
|
*
|
|
* If the thing being dereferenced is a member of uniform block \b without an
|
|
* instance name, then the name of the \c ir_variable is the field name of an
|
|
* interface type. If this field is row-major, then the thing referenced is
|
|
* row-major.
|
|
*
|
|
* If the thing being dereferenced is a member of uniform block \b with an
|
|
* instance name, then the last dereference in the tree will be an
|
|
* \c ir_dereference_record. If that record field is row-major, then the
|
|
* thing referenced is row-major.
|
|
*/
|
|
bool
|
|
lower_buffer_access::is_dereferenced_thing_row_major(const ir_rvalue *deref)
|
|
{
|
|
bool matrix = false;
|
|
const ir_rvalue *ir = deref;
|
|
|
|
while (true) {
|
|
matrix = matrix || ir->type->without_array()->is_matrix();
|
|
|
|
switch (ir->ir_type) {
|
|
case ir_type_dereference_array: {
|
|
const ir_dereference_array *const array_deref =
|
|
(const ir_dereference_array *) ir;
|
|
|
|
ir = array_deref->array;
|
|
break;
|
|
}
|
|
|
|
case ir_type_dereference_record: {
|
|
const ir_dereference_record *const record_deref =
|
|
(const ir_dereference_record *) ir;
|
|
|
|
ir = record_deref->record;
|
|
|
|
const int idx = record_deref->field_idx;
|
|
assert(idx >= 0);
|
|
|
|
const enum glsl_matrix_layout matrix_layout =
|
|
glsl_matrix_layout(ir->type->fields.structure[idx].matrix_layout);
|
|
|
|
switch (matrix_layout) {
|
|
case GLSL_MATRIX_LAYOUT_INHERITED:
|
|
break;
|
|
case GLSL_MATRIX_LAYOUT_COLUMN_MAJOR:
|
|
return false;
|
|
case GLSL_MATRIX_LAYOUT_ROW_MAJOR:
|
|
return matrix || deref->type->without_array()->is_record();
|
|
}
|
|
|
|
break;
|
|
}
|
|
|
|
case ir_type_dereference_variable: {
|
|
const ir_dereference_variable *const var_deref =
|
|
(const ir_dereference_variable *) ir;
|
|
|
|
const enum glsl_matrix_layout matrix_layout =
|
|
glsl_matrix_layout(var_deref->var->data.matrix_layout);
|
|
|
|
switch (matrix_layout) {
|
|
case GLSL_MATRIX_LAYOUT_INHERITED: {
|
|
/* For interface block matrix variables we handle inherited
|
|
* layouts at HIR generation time, but we don't do that for shared
|
|
* variables, which are always column-major
|
|
*/
|
|
MAYBE_UNUSED ir_variable *var = deref->variable_referenced();
|
|
assert((var->is_in_buffer_block() && !matrix) ||
|
|
var->data.mode == ir_var_shader_shared);
|
|
return false;
|
|
}
|
|
case GLSL_MATRIX_LAYOUT_COLUMN_MAJOR:
|
|
return false;
|
|
case GLSL_MATRIX_LAYOUT_ROW_MAJOR:
|
|
return matrix || deref->type->without_array()->is_record();
|
|
}
|
|
|
|
unreachable("invalid matrix layout");
|
|
break;
|
|
}
|
|
|
|
default:
|
|
return false;
|
|
}
|
|
}
|
|
|
|
/* The tree must have ended with a dereference that wasn't an
|
|
* ir_dereference_variable. That is invalid, and it should be impossible.
|
|
*/
|
|
unreachable("invalid dereference tree");
|
|
return false;
|
|
}
|
|
|
|
/**
|
|
* This function initializes various values that will be used later by
|
|
* emit_access when actually emitting loads or stores.
|
|
*
|
|
* Note: const_offset is an input as well as an output, clients must
|
|
* initialize it to the offset of the variable in the underlying block, and
|
|
* this function will adjust it by adding the constant offset of the member
|
|
* being accessed into that variable.
|
|
*/
|
|
void
|
|
lower_buffer_access::setup_buffer_access(void *mem_ctx,
|
|
ir_rvalue *deref,
|
|
ir_rvalue **offset,
|
|
unsigned *const_offset,
|
|
bool *row_major,
|
|
int *matrix_columns,
|
|
const glsl_struct_field **struct_field,
|
|
enum glsl_interface_packing packing)
|
|
{
|
|
*offset = new(mem_ctx) ir_constant(0u);
|
|
*row_major = is_dereferenced_thing_row_major(deref);
|
|
*matrix_columns = 1;
|
|
|
|
/* Calculate the offset to the start of the region of the UBO
|
|
* dereferenced by *rvalue. This may be a variable offset if an
|
|
* array dereference has a variable index.
|
|
*/
|
|
while (deref) {
|
|
switch (deref->ir_type) {
|
|
case ir_type_dereference_variable: {
|
|
deref = NULL;
|
|
break;
|
|
}
|
|
|
|
case ir_type_dereference_array: {
|
|
ir_dereference_array *deref_array = (ir_dereference_array *) deref;
|
|
unsigned array_stride;
|
|
if (deref_array->array->type->is_vector()) {
|
|
/* We get this when storing or loading a component out of a vector
|
|
* with a non-constant index. This happens for v[i] = f where v is
|
|
* a vector (or m[i][j] = f where m is a matrix). If we don't
|
|
* lower that here, it gets turned into v = vector_insert(v, i,
|
|
* f), which loads the entire vector, modifies one component and
|
|
* then write the entire thing back. That breaks if another
|
|
* thread or SIMD channel is modifying the same vector.
|
|
*/
|
|
array_stride = 4;
|
|
if (deref_array->array->type->is_64bit())
|
|
array_stride *= 2;
|
|
} else if (deref_array->array->type->is_matrix() && *row_major) {
|
|
/* When loading a vector out of a row major matrix, the
|
|
* step between the columns (vectors) is the size of a
|
|
* float, while the step between the rows (elements of a
|
|
* vector) is handled below in emit_ubo_loads.
|
|
*/
|
|
array_stride = 4;
|
|
if (deref_array->array->type->is_64bit())
|
|
array_stride *= 2;
|
|
*matrix_columns = deref_array->array->type->matrix_columns;
|
|
} else if (deref_array->type->without_array()->is_interface()) {
|
|
/* We're processing an array dereference of an interface instance
|
|
* array. The thing being dereferenced *must* be a variable
|
|
* dereference because interfaces cannot be embedded in other
|
|
* types. In terms of calculating the offsets for the lowering
|
|
* pass, we don't care about the array index. All elements of an
|
|
* interface instance array will have the same offsets relative to
|
|
* the base of the block that backs them.
|
|
*/
|
|
deref = deref_array->array->as_dereference();
|
|
break;
|
|
} else {
|
|
/* Whether or not the field is row-major (because it might be a
|
|
* bvec2 or something) does not affect the array itself. We need
|
|
* to know whether an array element in its entirety is row-major.
|
|
*/
|
|
const bool array_row_major =
|
|
is_dereferenced_thing_row_major(deref_array);
|
|
|
|
/* The array type will give the correct interface packing
|
|
* information
|
|
*/
|
|
if (packing == GLSL_INTERFACE_PACKING_STD430) {
|
|
array_stride = deref_array->type->std430_array_stride(array_row_major);
|
|
} else {
|
|
array_stride = deref_array->type->std140_size(array_row_major);
|
|
array_stride = glsl_align(array_stride, 16);
|
|
}
|
|
}
|
|
|
|
ir_rvalue *array_index = deref_array->array_index;
|
|
if (array_index->type->base_type == GLSL_TYPE_INT)
|
|
array_index = i2u(array_index);
|
|
|
|
ir_constant *const_index =
|
|
array_index->constant_expression_value(mem_ctx, NULL);
|
|
if (const_index) {
|
|
*const_offset += array_stride * const_index->value.u[0];
|
|
} else {
|
|
*offset = add(*offset,
|
|
mul(array_index,
|
|
new(mem_ctx) ir_constant(array_stride)));
|
|
}
|
|
deref = deref_array->array->as_dereference();
|
|
break;
|
|
}
|
|
|
|
case ir_type_dereference_record: {
|
|
ir_dereference_record *deref_record = (ir_dereference_record *) deref;
|
|
const glsl_type *struct_type = deref_record->record->type;
|
|
unsigned intra_struct_offset = 0;
|
|
|
|
for (unsigned int i = 0; i < struct_type->length; i++) {
|
|
const glsl_type *type = struct_type->fields.structure[i].type;
|
|
|
|
ir_dereference_record *field_deref = new(mem_ctx)
|
|
ir_dereference_record(deref_record->record,
|
|
struct_type->fields.structure[i].name);
|
|
const bool field_row_major =
|
|
is_dereferenced_thing_row_major(field_deref);
|
|
|
|
ralloc_free(field_deref);
|
|
|
|
unsigned field_align = 0;
|
|
|
|
if (packing == GLSL_INTERFACE_PACKING_STD430)
|
|
field_align = type->std430_base_alignment(field_row_major);
|
|
else
|
|
field_align = type->std140_base_alignment(field_row_major);
|
|
|
|
if (struct_type->fields.structure[i].offset != -1) {
|
|
intra_struct_offset = struct_type->fields.structure[i].offset;
|
|
}
|
|
|
|
intra_struct_offset = glsl_align(intra_struct_offset, field_align);
|
|
|
|
assert(deref_record->field_idx >= 0);
|
|
if (i == (unsigned) deref_record->field_idx) {
|
|
if (struct_field)
|
|
*struct_field = &struct_type->fields.structure[i];
|
|
break;
|
|
}
|
|
|
|
if (packing == GLSL_INTERFACE_PACKING_STD430)
|
|
intra_struct_offset += type->std430_size(field_row_major);
|
|
else
|
|
intra_struct_offset += type->std140_size(field_row_major);
|
|
|
|
/* If the field just examined was itself a structure, apply rule
|
|
* #9:
|
|
*
|
|
* "The structure may have padding at the end; the base offset
|
|
* of the member following the sub-structure is rounded up to
|
|
* the next multiple of the base alignment of the structure."
|
|
*/
|
|
if (type->without_array()->is_record()) {
|
|
intra_struct_offset = glsl_align(intra_struct_offset,
|
|
field_align);
|
|
|
|
}
|
|
}
|
|
|
|
*const_offset += intra_struct_offset;
|
|
deref = deref_record->record->as_dereference();
|
|
break;
|
|
}
|
|
|
|
case ir_type_swizzle: {
|
|
ir_swizzle *deref_swizzle = (ir_swizzle *) deref;
|
|
|
|
assert(deref_swizzle->mask.num_components == 1);
|
|
|
|
*const_offset += deref_swizzle->mask.x * sizeof(int);
|
|
deref = deref_swizzle->val->as_dereference();
|
|
break;
|
|
}
|
|
|
|
default:
|
|
assert(!"not reached");
|
|
deref = NULL;
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
|
|
} /* namespace lower_buffer_access */
|