#include "compiler/nir/nir_builtin_builder.h"
#include "compiler/nir/nir_deref.h"
#include "main/errors.h"
-#include "main/imports.h"
#include "main/mtypes.h"
#include "main/shaderobj.h"
#include "util/u_math.h"
}
}
+ if (!glsl_type_is_vector_or_scalar(ir->return_type) &&
+ !ir->return_type->is_void()) {
+ unsupported = true;
+ return visit_stop;
+ }
+
return visit_continue;
}
break;
+ case GLSL_TYPE_UINT16:
+ /* Only float base types can be matrices. */
+ assert(cols == 1);
+
+ for (unsigned r = 0; r < rows; r++)
+ ret->values[r].u16 = ir->value.u16[r];
+ break;
+
case GLSL_TYPE_INT:
/* Only float base types can be matrices. */
assert(cols == 1);
break;
+ case GLSL_TYPE_INT16:
+ /* Only float base types can be matrices. */
+ assert(cols == 1);
+
+ for (unsigned r = 0; r < rows; r++)
+ ret->values[r].i16 = ir->value.i16[r];
+ break;
+
case GLSL_TYPE_FLOAT:
+ case GLSL_TYPE_FLOAT16:
case GLSL_TYPE_DOUBLE:
if (cols > 1) {
ret->elements = ralloc_array(mem_ctx, nir_constant *, cols);
col_const->values[r].f32 = ir->value.f[c * rows + r];
break;
+ case GLSL_TYPE_FLOAT16:
+ for (unsigned r = 0; r < rows; r++)
+ col_const->values[r].u16 = ir->value.f16[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];
ret->values[r].f32 = ir->value.f[r];
break;
+ case GLSL_TYPE_FLOAT16:
+ for (unsigned r = 0; r < rows; r++)
+ ret->values[r].u16 = ir->value.f16[r];
+ break;
+
case GLSL_TYPE_DOUBLE:
for (unsigned r = 0; r < rows; r++)
ret->values[r].f64 = ir->value.d[r];
var->data.precision = ir->data.precision;
var->data.explicit_location = ir->data.explicit_location;
+ var->data.matrix_layout = ir->data.matrix_layout;
var->data.from_named_ifc_block = ir->data.from_named_ifc_block;
var->data.compact = false;
if (field->memory_restrict)
qualifiers |= ACCESS_RESTRICT;
}
-
+
parent_type = cur->type;
}
op = nir_intrinsic_image_deref_samples;
break;
case ir_intrinsic_ssbo_store:
- op = nir_intrinsic_store_ssbo;
- break;
case ir_intrinsic_ssbo_load:
- op = nir_intrinsic_load_ssbo;
- break;
case ir_intrinsic_ssbo_atomic_add:
- op = ir->return_deref->type->is_integer_32_64()
- ? nir_intrinsic_ssbo_atomic_add : nir_intrinsic_ssbo_atomic_fadd;
- break;
case ir_intrinsic_ssbo_atomic_and:
- op = nir_intrinsic_ssbo_atomic_and;
- break;
case ir_intrinsic_ssbo_atomic_or:
- op = nir_intrinsic_ssbo_atomic_or;
- break;
case ir_intrinsic_ssbo_atomic_xor:
- op = nir_intrinsic_ssbo_atomic_xor;
- break;
case ir_intrinsic_ssbo_atomic_min:
- assert(ir->return_deref);
- if (ir->return_deref->type == glsl_type::int_type)
- op = nir_intrinsic_ssbo_atomic_imin;
- else if (ir->return_deref->type == glsl_type::uint_type)
- op = nir_intrinsic_ssbo_atomic_umin;
- else if (ir->return_deref->type == glsl_type::float_type)
- op = nir_intrinsic_ssbo_atomic_fmin;
- else
- unreachable("Invalid type");
- break;
case ir_intrinsic_ssbo_atomic_max:
- assert(ir->return_deref);
- if (ir->return_deref->type == glsl_type::int_type)
- op = nir_intrinsic_ssbo_atomic_imax;
- else if (ir->return_deref->type == glsl_type::uint_type)
- op = nir_intrinsic_ssbo_atomic_umax;
- else if (ir->return_deref->type == glsl_type::float_type)
- op = nir_intrinsic_ssbo_atomic_fmax;
- else
- unreachable("Invalid type");
- break;
case ir_intrinsic_ssbo_atomic_exchange:
- op = nir_intrinsic_ssbo_atomic_exchange;
- break;
case ir_intrinsic_ssbo_atomic_comp_swap:
- op = ir->return_deref->type->is_integer_32_64()
- ? nir_intrinsic_ssbo_atomic_comp_swap
- : nir_intrinsic_ssbo_atomic_fcomp_swap;
- break;
+ /* SSBO store/loads should only have been lowered in GLSL IR for
+ * non-nir drivers, NIR drivers make use of gl_nir_lower_buffers()
+ * instead.
+ */
+ unreachable("Invalid operation nir doesn't want lowered ssbo "
+ "store/loads");
case ir_intrinsic_shader_clock:
op = nir_intrinsic_shader_clock;
break;
break;
case nir_intrinsic_shader_clock:
nir_ssa_dest_init(&instr->instr, &instr->dest, 2, 32, NULL);
- instr->num_components = 2;
+ nir_intrinsic_set_memory_scope(instr, NIR_SCOPE_SUBGROUP);
nir_builder_instr_insert(&b, &instr->instr);
break;
case nir_intrinsic_begin_invocation_interlock:
nir_builder_instr_insert(&b, &instr->instr);
break;
}
- case nir_intrinsic_load_ssbo: {
- exec_node *param = ir->actual_parameters.get_head();
- ir_rvalue *block = ((ir_instruction *)param)->as_rvalue();
-
- param = param->get_next();
- ir_rvalue *offset = ((ir_instruction *)param)->as_rvalue();
-
- instr->src[0] = nir_src_for_ssa(evaluate_rvalue(block));
- instr->src[1] = nir_src_for_ssa(evaluate_rvalue(offset));
-
- const glsl_type *type = ir->return_deref->var->type;
- instr->num_components = type->vector_elements;
- intrinsic_set_std430_align(instr, type);
-
- /* Setup destination register */
- unsigned bit_size = type->is_boolean() ? 32 : glsl_get_bit_size(type);
- nir_ssa_dest_init(&instr->instr, &instr->dest,
- type->vector_elements, bit_size, NULL);
-
- /* Insert the created nir instruction now since in the case of boolean
- * result we will need to emit another instruction after it
- */
- nir_builder_instr_insert(&b, &instr->instr);
-
- /*
- * In SSBO/UBO's, a true boolean value is any non-zero value, but we
- * consider a true boolean to be ~0. Fix this up with a != 0
- * comparison.
- */
- if (type->is_boolean())
- ret = nir_i2b(&b, &instr->dest.ssa);
- break;
- }
- case nir_intrinsic_ssbo_atomic_add:
- case nir_intrinsic_ssbo_atomic_imin:
- case nir_intrinsic_ssbo_atomic_umin:
- case nir_intrinsic_ssbo_atomic_imax:
- case nir_intrinsic_ssbo_atomic_umax:
- case nir_intrinsic_ssbo_atomic_and:
- case nir_intrinsic_ssbo_atomic_or:
- case nir_intrinsic_ssbo_atomic_xor:
- case nir_intrinsic_ssbo_atomic_exchange:
- case nir_intrinsic_ssbo_atomic_comp_swap:
- case nir_intrinsic_ssbo_atomic_fadd:
- case nir_intrinsic_ssbo_atomic_fmin:
- case nir_intrinsic_ssbo_atomic_fmax:
- case nir_intrinsic_ssbo_atomic_fcomp_swap: {
- int param_count = ir->actual_parameters.length();
- assert(param_count == 3 || param_count == 4);
-
- /* Block index */
- exec_node *param = ir->actual_parameters.get_head();
- ir_instruction *inst = (ir_instruction *) param;
- instr->src[0] = nir_src_for_ssa(evaluate_rvalue(inst->as_rvalue()));
-
- /* Offset */
- param = param->get_next();
- inst = (ir_instruction *) param;
- instr->src[1] = nir_src_for_ssa(evaluate_rvalue(inst->as_rvalue()));
-
- /* data1 parameter (this is always present) */
- param = param->get_next();
- inst = (ir_instruction *) param;
- instr->src[2] = nir_src_for_ssa(evaluate_rvalue(inst->as_rvalue()));
-
- /* data2 parameter (only with atomic_comp_swap) */
- if (param_count == 4) {
- assert(op == nir_intrinsic_ssbo_atomic_comp_swap ||
- op == nir_intrinsic_ssbo_atomic_fcomp_swap);
- param = param->get_next();
- inst = (ir_instruction *) param;
- instr->src[3] = nir_src_for_ssa(evaluate_rvalue(inst->as_rvalue()));
- }
-
- /* Atomic result */
- assert(ir->return_deref);
- nir_ssa_dest_init(&instr->instr, &instr->dest,
- ir->return_deref->type->vector_elements, 32, NULL);
- nir_builder_instr_insert(&b, &instr->instr);
- break;
- }
case nir_intrinsic_load_shared: {
exec_node *param = ir->actual_parameters.get_head();
ir_rvalue *offset = ((ir_instruction *)param)->as_rvalue();
/* The value in shared memory is a 32-bit value */
if (type->is_boolean())
- ret = nir_i2b(&b, &instr->dest.ssa);
+ ret = nir_b2b1(&b, &instr->dest.ssa);
break;
}
case nir_intrinsic_store_shared: {
nir_ssa_def *nir_val = evaluate_rvalue(val);
/* The value in shared memory is a 32-bit value */
if (val->type->is_boolean())
- nir_val = nir_b2i32(&b, nir_val);
+ nir_val = nir_b2b32(&b, nir_val);
instr->src[0] = nir_src_for_ssa(nir_val);
instr->num_components = val->type->vector_elements;
nir_builder_instr_insert(&b, &instr->instr);
break;
}
+ case nir_intrinsic_vote_ieq:
+ instr->num_components = 1;
+ /* fall-through */
case nir_intrinsic_vote_any:
- case nir_intrinsic_vote_all:
- case nir_intrinsic_vote_ieq: {
+ case nir_intrinsic_vote_all: {
nir_ssa_dest_init(&instr->instr, &instr->dest, 1, 1, NULL);
- instr->num_components = 1;
ir_rvalue *value = (ir_rvalue *) ir->actual_parameters.get_head();
instr->src[0] = nir_src_for_ssa(evaluate_rvalue(value));
}
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;
}
{
/* Some special cases */
switch (ir->operation) {
- case ir_binop_ubo_load: {
- nir_intrinsic_instr *load =
- nir_intrinsic_instr_create(this->shader, nir_intrinsic_load_ubo);
- unsigned bit_size = ir->type->is_boolean() ? 32 :
- glsl_get_bit_size(ir->type);
- load->num_components = ir->type->vector_elements;
- load->src[0] = nir_src_for_ssa(evaluate_rvalue(ir->operands[0]));
- load->src[1] = nir_src_for_ssa(evaluate_rvalue(ir->operands[1]));
- intrinsic_set_std430_align(load, ir->type);
- add_instr(&load->instr, ir->type->vector_elements, bit_size);
-
- /*
- * In UBO's, a true boolean value is any non-zero value, but we consider
- * a true boolean to be ~0. Fix this up with a != 0 comparison.
- */
-
- if (ir->type->is_boolean())
- this->result = nir_i2b(&b, &load->dest.ssa);
-
- return;
- }
-
case ir_unop_interpolate_at_centroid:
case ir_binop_interpolate_at_offset:
case ir_binop_interpolate_at_sample: {
return;
}
+ case ir_binop_ubo_load:
+ /* UBO loads should only have been lowered in GLSL IR for non-nir drivers,
+ * NIR drivers make use of gl_nir_lower_buffers() instead.
+ */
+ unreachable("Invalid operation nir doesn't want lowered ubo loads");
default:
break;
}
case ir_unop_b2i64:
case ir_unop_d2f:
case ir_unop_f2d:
+ case ir_unop_f162f:
+ case ir_unop_f2f16:
+ case ir_unop_f162b:
+ case ir_unop_b2f16:
+ case ir_unop_i2i:
+ case ir_unop_u2u:
case ir_unop_d2i:
case ir_unop_d2u:
case ir_unop_d2b:
break;
}
+ case ir_unop_f2fmp: {
+ result = nir_build_alu(&b, nir_op_f2fmp, srcs[0], NULL, NULL, NULL);
+ break;
+ }
+
+ case ir_unop_i2imp: {
+ result = nir_build_alu(&b, nir_op_i2imp, srcs[0], NULL, NULL, NULL);
+ break;
+ }
+
+ case ir_unop_u2ump: {
+ result = nir_build_alu(&b, nir_op_u2ump, srcs[0], NULL, NULL, NULL);
+ break;
+ }
+
case ir_unop_bitcast_i2f:
case ir_unop_bitcast_f2i:
case ir_unop_bitcast_u2f:
result = nir_find_lsb(&b, srcs[0]);
break;
- case ir_unop_noise:
- switch (ir->type->vector_elements) {
- case 1:
- switch (ir->operands[0]->type->vector_elements) {
- case 1: result = nir_fnoise1_1(&b, srcs[0]); break;
- case 2: result = nir_fnoise1_2(&b, srcs[0]); break;
- case 3: result = nir_fnoise1_3(&b, srcs[0]); break;
- case 4: result = nir_fnoise1_4(&b, srcs[0]); break;
- default: unreachable("not reached");
- }
- break;
- case 2:
- switch (ir->operands[0]->type->vector_elements) {
- case 1: result = nir_fnoise2_1(&b, srcs[0]); break;
- case 2: result = nir_fnoise2_2(&b, srcs[0]); break;
- case 3: result = nir_fnoise2_3(&b, srcs[0]); break;
- case 4: result = nir_fnoise2_4(&b, srcs[0]); break;
- default: unreachable("not reached");
- }
- break;
- case 3:
- switch (ir->operands[0]->type->vector_elements) {
- case 1: result = nir_fnoise3_1(&b, srcs[0]); break;
- case 2: result = nir_fnoise3_2(&b, srcs[0]); break;
- case 3: result = nir_fnoise3_3(&b, srcs[0]); break;
- case 4: result = nir_fnoise3_4(&b, srcs[0]); break;
- default: unreachable("not reached");
- }
- break;
- case 4:
- switch (ir->operands[0]->type->vector_elements) {
- case 1: result = nir_fnoise4_1(&b, srcs[0]); break;
- case 2: result = nir_fnoise4_2(&b, srcs[0]); break;
- case 3: result = nir_fnoise4_3(&b, srcs[0]); break;
- case 4: result = nir_fnoise4_4(&b, srcs[0]); break;
- default: unreachable("not reached");
- }
- break;
- default:
- unreachable("not reached");
- }
- break;
case ir_unop_get_buffer_size: {
nir_intrinsic_instr *load = nir_intrinsic_instr_create(
this->shader,
result = nir_bcsel(&b, srcs[0], srcs[1], srcs[2]);
break;
case ir_triop_bitfield_extract:
- result = (out_type == GLSL_TYPE_INT) ?
- nir_ibitfield_extract(&b, srcs[0], srcs[1], srcs[2]) :
- nir_ubitfield_extract(&b, srcs[0], srcs[1], srcs[2]);
+ result = ir->type->is_int_16_32() ?
+ nir_ibitfield_extract(&b, nir_i2i32(&b, srcs[0]), nir_i2i32(&b, srcs[1]), nir_i2i32(&b, srcs[2])) :
+ nir_ubitfield_extract(&b, nir_u2u32(&b, srcs[0]), nir_i2i32(&b, srcs[1]), nir_i2i32(&b, srcs[2]));
break;
case ir_quadop_bitfield_insert:
- result = nir_bitfield_insert(&b, srcs[0], srcs[1], srcs[2], srcs[3]);
+ result = nir_bitfield_insert(&b,
+ nir_u2u32(&b, srcs[0]), nir_u2u32(&b, srcs[1]),
+ nir_i2i32(&b, srcs[2]), nir_i2i32(&b, srcs[3]));
break;
case ir_quadop_vector:
result = nir_vec(&b, srcs, ir->type->vector_elements);
case GLSL_TYPE_FLOAT:
instr->dest_type = nir_type_float;
break;
+ case GLSL_TYPE_FLOAT16:
+ instr->dest_type = nir_type_float16;
+ break;
+ case GLSL_TYPE_INT16:
+ instr->dest_type = nir_type_int16;
+ break;
+ case GLSL_TYPE_UINT16:
+ instr->dest_type = nir_type_uint16;
+ break;
case GLSL_TYPE_INT:
instr->dest_type = nir_type_int;
break;
projects / mesa.git / blobdiff