#include "program.h"
#include "compiler/nir/nir_control_flow.h"
#include "compiler/nir/nir_builder.h"
+#include "compiler/nir/nir_builtin_builder.h"
#include "compiler/nir/nir_deref.h"
#include "main/errors.h"
#include "main/imports.h"
virtual void visit(ir_loop *);
virtual void visit(ir_if *);
virtual void visit(ir_discard *);
+ virtual void visit(ir_demote *);
virtual void visit(ir_loop_jump *);
virtual void visit(ir_return *);
virtual void visit(ir_call *);
}
}
- /* Remap the locations to slots so those requiring two slots will occupy
- * two locations. For instance, if we have in the IR code a dvec3 attr0 in
- * location 0 and vec4 attr1 in location 1, in NIR attr0 will use
- * locations/slots 0 and 1, and attr1 will use location/slot 2 */
- if (shader->info.stage == MESA_SHADER_VERTEX)
- nir_remap_dual_slot_attributes(shader, &sh->Program->DualSlotInputs);
-
shader->info.name = ralloc_asprintf(shader, "GLSL%d", shader_prog->Name);
if (shader_prog->Label)
shader->info.label = ralloc_strdup(shader, shader_prog->Label);
assert(cols == 1);
for (unsigned r = 0; r < rows; r++)
- ret->values[0][r].u32 = ir->value.u[r];
+ ret->values[r].u32 = ir->value.u[r];
break;
assert(cols == 1);
for (unsigned r = 0; r < rows; r++)
- ret->values[0][r].i32 = ir->value.i[r];
+ ret->values[r].i32 = ir->value.i[r];
break;
case GLSL_TYPE_FLOAT:
- for (unsigned c = 0; c < cols; c++) {
- for (unsigned r = 0; r < rows; r++)
- ret->values[c][r].f32 = ir->value.f[c * rows + r];
- }
- break;
-
case GLSL_TYPE_DOUBLE:
- for (unsigned c = 0; c < cols; c++) {
- for (unsigned r = 0; r < rows; r++)
- ret->values[c][r].f64 = ir->value.d[c * rows + r];
+ if (cols > 1) {
+ ret->elements = ralloc_array(mem_ctx, nir_constant *, cols);
+ ret->num_elements = cols;
+ for (unsigned c = 0; c < cols; c++) {
+ nir_constant *col_const = rzalloc(mem_ctx, nir_constant);
+ col_const->num_elements = 0;
+ switch (ir->type->base_type) {
+ case GLSL_TYPE_FLOAT:
+ for (unsigned r = 0; r < rows; r++)
+ col_const->values[r].f32 = ir->value.f[c * rows + r];
+ break;
+
+ case GLSL_TYPE_DOUBLE:
+ for (unsigned r = 0; r < rows; r++)
+ col_const->values[r].f64 = ir->value.d[c * rows + r];
+ break;
+
+ default:
+ unreachable("Cannot get here from the first level switch");
+ }
+ ret->elements[c] = col_const;
+ }
+ } else {
+ switch (ir->type->base_type) {
+ case GLSL_TYPE_FLOAT:
+ for (unsigned r = 0; r < rows; r++)
+ ret->values[r].f32 = ir->value.f[r];
+ break;
+
+ case GLSL_TYPE_DOUBLE:
+ for (unsigned r = 0; r < rows; r++)
+ ret->values[r].f64 = ir->value.d[r];
+ break;
+
+ default:
+ unreachable("Cannot get here from the first level switch");
+ }
}
break;
assert(cols == 1);
for (unsigned r = 0; r < rows; r++)
- ret->values[0][r].u64 = ir->value.u64[r];
+ ret->values[r].u64 = ir->value.u64[r];
break;
case GLSL_TYPE_INT64:
assert(cols == 1);
for (unsigned r = 0; r < rows; r++)
- ret->values[0][r].i64 = ir->value.i64[r];
+ ret->values[r].i64 = ir->value.i64[r];
break;
case GLSL_TYPE_BOOL:
assert(cols == 1);
for (unsigned r = 0; r < rows; r++)
- ret->values[0][r].b = ir->value.b[r];
+ ret->values[r].b = ir->value.b[r];
break;
return glsl_type::get_array_instance(elem_type, array_type->length);
}
+static unsigned
+get_nir_how_declared(unsigned how_declared)
+{
+ if (how_declared == ir_var_hidden)
+ return nir_var_hidden;
+
+ return nir_var_declared_normally;
+}
+
void
nir_visitor::visit(ir_variable *ir)
{
var->data.centroid = ir->data.centroid;
var->data.sample = ir->data.sample;
var->data.patch = ir->data.patch;
+ var->data.how_declared = get_nir_how_declared(ir->data.how_declared);
var->data.invariant = ir->data.invariant;
var->data.location = ir->data.location;
var->data.stream = ir->data.stream;
+ if (ir->data.stream & (1u << 31))
+ var->data.stream |= NIR_STREAM_PACKED;
+
+ var->data.precision = ir->data.precision;
+ var->data.explicit_location = ir->data.explicit_location;
+ var->data.from_named_ifc_block = ir->data.from_named_ifc_block;
var->data.compact = false;
switch(ir->data.mode) {
break;
case ir_var_shader_in:
- if (shader->info.stage == MESA_SHADER_FRAGMENT &&
- ir->data.location == VARYING_SLOT_FACE) {
- /* For whatever reason, GLSL IR makes gl_FrontFacing an input */
- var->data.location = SYSTEM_VALUE_FRONT_FACE;
- var->data.mode = nir_var_system_value;
- } else if (shader->info.stage == MESA_SHADER_GEOMETRY &&
- ir->data.location == VARYING_SLOT_PRIMITIVE_ID) {
+ if (shader->info.stage == MESA_SHADER_GEOMETRY &&
+ ir->data.location == VARYING_SLOT_PRIMITIVE_ID) {
/* For whatever reason, GLSL IR makes gl_PrimitiveIDIn an input */
var->data.location = SYSTEM_VALUE_PRIMITIVE_ID;
var->data.mode = nir_var_system_value;
unreachable("not reached");
}
- unsigned image_access = 0;
+ unsigned mem_access = 0;
if (ir->data.memory_read_only)
- image_access |= ACCESS_NON_WRITEABLE;
+ mem_access |= ACCESS_NON_WRITEABLE;
if (ir->data.memory_write_only)
- image_access |= ACCESS_NON_READABLE;
+ mem_access |= ACCESS_NON_READABLE;
if (ir->data.memory_coherent)
- image_access |= ACCESS_COHERENT;
+ mem_access |= ACCESS_COHERENT;
if (ir->data.memory_volatile)
- image_access |= ACCESS_VOLATILE;
+ mem_access |= ACCESS_VOLATILE;
if (ir->data.memory_restrict)
- image_access |= ACCESS_RESTRICT;
+ mem_access |= ACCESS_RESTRICT;
/* For UBO and SSBO variables, we need explicit types */
if (var->data.mode & (nir_var_mem_ubo | nir_var_mem_ssbo)) {
var->type = field->type;
if (field->memory_read_only)
- image_access |= ACCESS_NON_WRITEABLE;
+ mem_access |= ACCESS_NON_WRITEABLE;
if (field->memory_write_only)
- image_access |= ACCESS_NON_READABLE;
+ mem_access |= ACCESS_NON_READABLE;
if (field->memory_coherent)
- image_access |= ACCESS_COHERENT;
+ mem_access |= ACCESS_COHERENT;
if (field->memory_volatile)
- image_access |= ACCESS_VOLATILE;
+ mem_access |= ACCESS_VOLATILE;
if (field->memory_restrict)
- image_access |= ACCESS_RESTRICT;
+ mem_access |= ACCESS_RESTRICT;
found = true;
break;
var->data.explicit_binding = ir->data.explicit_binding;
var->data.bindless = ir->data.bindless;
var->data.offset = ir->data.offset;
+ var->data.access = (gl_access_qualifier)mem_access;
- var->data.image.access = (gl_access_qualifier)image_access;
- var->data.image.format = ir->data.image_format;
+ if (var->type->without_array()->is_image()) {
+ var->data.image.format = ir->data.image_format;
+ } else if (var->data.mode == nir_var_shader_out) {
+ var->data.xfb.buffer = ir->data.xfb_buffer;
+ var->data.xfb.stride = ir->data.xfb_stride;
+ }
var->data.fb_fetch_output = ir->data.fb_fetch_output;
var->data.explicit_xfb_buffer = ir->data.explicit_xfb_buffer;
var->data.explicit_xfb_stride = ir->data.explicit_xfb_stride;
- var->data.xfb_buffer = ir->data.xfb_buffer;
- var->data.xfb_stride = ir->data.xfb_stride;
var->num_state_slots = ir->get_num_state_slots();
if (var->num_state_slots > 0) {
nir_builder_instr_insert(&b, &discard->instr);
}
+void
+nir_visitor::visit(ir_demote *ir)
+{
+ nir_intrinsic_instr *demote =
+ nir_intrinsic_instr_create(this->shader, nir_intrinsic_demote);
+
+ nir_builder_instr_insert(&b, &demote->instr);
+}
+
void
nir_visitor::visit(ir_emit_vertex *ir)
{
nir_deref_path path;
nir_deref_path_init(&path, deref, NULL);
- unsigned qualifiers = path.path[0]->var->data.image.access;
+ unsigned qualifiers = path.path[0]->var->data.access;
const glsl_type *parent_type = path.path[0]->type;
for (nir_deref_instr **cur_ptr = &path.path[1]; *cur_ptr; cur_ptr++) {
: nir_intrinsic_image_deref_atomic_fadd;
break;
case ir_intrinsic_image_atomic_min:
- op = nir_intrinsic_image_deref_atomic_min;
+ if (ir->return_deref->type == glsl_type::int_type)
+ op = nir_intrinsic_image_deref_atomic_imin;
+ else if (ir->return_deref->type == glsl_type::uint_type)
+ op = nir_intrinsic_image_deref_atomic_umin;
+ else
+ unreachable("Invalid type");
break;
case ir_intrinsic_image_atomic_max:
- op = nir_intrinsic_image_deref_atomic_max;
+ if (ir->return_deref->type == glsl_type::int_type)
+ op = nir_intrinsic_image_deref_atomic_imax;
+ else if (ir->return_deref->type == glsl_type::uint_type)
+ op = nir_intrinsic_image_deref_atomic_umax;
+ else
+ unreachable("Invalid type");
break;
case ir_intrinsic_image_atomic_and:
op = nir_intrinsic_image_deref_atomic_and;
case ir_intrinsic_image_atomic_comp_swap:
op = nir_intrinsic_image_deref_atomic_comp_swap;
break;
+ case ir_intrinsic_image_atomic_inc_wrap:
+ op = nir_intrinsic_image_deref_atomic_inc_wrap;
+ break;
+ case ir_intrinsic_image_atomic_dec_wrap:
+ op = nir_intrinsic_image_deref_atomic_dec_wrap;
+ break;
case ir_intrinsic_memory_barrier:
op = nir_intrinsic_memory_barrier;
break;
case ir_intrinsic_read_first_invocation:
op = nir_intrinsic_read_first_invocation;
break;
+ case ir_intrinsic_helper_invocation:
+ op = nir_intrinsic_is_helper_invocation;
+ break;
default:
unreachable("not reached");
}
case nir_intrinsic_image_deref_load:
case nir_intrinsic_image_deref_store:
case nir_intrinsic_image_deref_atomic_add:
- case nir_intrinsic_image_deref_atomic_min:
- case nir_intrinsic_image_deref_atomic_max:
+ case nir_intrinsic_image_deref_atomic_imin:
+ case nir_intrinsic_image_deref_atomic_umin:
+ case nir_intrinsic_image_deref_atomic_imax:
+ case nir_intrinsic_image_deref_atomic_umax:
case nir_intrinsic_image_deref_atomic_and:
case nir_intrinsic_image_deref_atomic_or:
case nir_intrinsic_image_deref_atomic_xor:
case nir_intrinsic_image_deref_atomic_comp_swap:
case nir_intrinsic_image_deref_atomic_fadd:
case nir_intrinsic_image_deref_samples:
- case nir_intrinsic_image_deref_size: {
+ case nir_intrinsic_image_deref_size:
+ case nir_intrinsic_image_deref_atomic_inc_wrap:
+ case nir_intrinsic_image_deref_atomic_dec_wrap: {
nir_ssa_undef_instr *instr_undef =
nir_ssa_undef_instr_create(shader, 1, 32);
nir_builder_instr_insert(&b, &instr_undef->instr);
instr->src[3] =
nir_src_for_ssa(evaluate_rvalue((ir_dereference *)param));
param = param->get_next();
+ } else if (op == nir_intrinsic_image_deref_load) {
+ instr->src[3] = nir_src_for_ssa(nir_imm_int(&b, 0)); /* LOD */
}
if (!param->is_tail_sentinel()) {
instr->src[4] =
nir_src_for_ssa(evaluate_rvalue((ir_dereference *)param));
param = param->get_next();
+ } else if (op == nir_intrinsic_image_deref_store) {
+ instr->src[4] = nir_src_for_ssa(nir_imm_int(&b, 0)); /* LOD */
}
+
nir_builder_instr_insert(&b, &instr->instr);
break;
}
nir_builder_instr_insert(&b, &instr->instr);
break;
}
+ case nir_intrinsic_is_helper_invocation: {
+ nir_ssa_dest_init(&instr->instr, &instr->dest, 1, 1, NULL);
+ instr->num_components = 1;
+ nir_builder_instr_insert(&b, &instr->instr);
+ break;
+ }
default:
unreachable("not reached");
}
return;
}
+ case ir_unop_atan:
+ result = nir_atan(&b, srcs[0]);
+ break;
+
case ir_binop_add:
result = type_is_float(out_type) ? nir_fadd(&b, srcs[0], srcs[1])
: nir_iadd(&b, srcs[0], srcs[1]);
break;
}
+ case ir_binop_atan2:
+ result = nir_atan2(&b, srcs[0], srcs[1]);
+ break;
+
case ir_binop_ldexp: result = nir_ldexp(&b, srcs[0], srcs[1]); break;
case ir_triop_fma:
result = nir_ffma(&b, srcs[0], srcs[1], srcs[2]);
void
nir_visitor::visit(ir_barrier *)
{
+ if (shader->info.stage == MESA_SHADER_COMPUTE) {
+ nir_intrinsic_instr *shared_barrier =
+ nir_intrinsic_instr_create(this->shader,
+ nir_intrinsic_memory_barrier_shared);
+ nir_builder_instr_insert(&b, &shared_barrier->instr);
+ } else if (shader->info.stage == MESA_SHADER_TESS_CTRL) {
+ nir_intrinsic_instr *patch_barrier =
+ nir_intrinsic_instr_create(this->shader,
+ nir_intrinsic_memory_barrier_tcs_patch);
+ nir_builder_instr_insert(&b, &patch_barrier->instr);
+ }
+
nir_intrinsic_instr *instr =
- nir_intrinsic_instr_create(this->shader, nir_intrinsic_barrier);
+ nir_intrinsic_instr_create(this->shader, nir_intrinsic_control_barrier);
nir_builder_instr_insert(&b, &instr->instr);
}
NIR_PASS_V(nir, nir_inline_functions);
NIR_PASS_V(nir, nir_opt_deref);
+ /* Do some optimizations to clean up the shader now. By optimizing the
+ * functions in the library, we avoid having to re-do that work every
+ * time we inline a copy of a function. Reducing basic blocks also helps
+ * with compile times.
+ */
+ NIR_PASS_V(nir, nir_lower_vars_to_ssa);
+ NIR_PASS_V(nir, nir_copy_prop);
+ NIR_PASS_V(nir, nir_opt_dce);
+ NIR_PASS_V(nir, nir_opt_cse);
+ NIR_PASS_V(nir, nir_opt_gcm, true);
+ NIR_PASS_V(nir, nir_opt_peephole_select, 1, false, false);
+ NIR_PASS_V(nir, nir_opt_dce);
+
return nir;
}