-#! /usr/bin/env python
#
# Copyright (C) 2015 Intel Corporation
#
def __iter__(self):
return self
- def next(self):
+ def __next__(self):
if self.i < len(self.source_types):
i = self.i
self.i += 1
else:
raise StopIteration()
+ next = __next__
+
uint_type = type("unsigned", "u", "GLSL_TYPE_UINT")
int_type = type("int", "i", "GLSL_TYPE_INT")
+uint64_type = type("uint64_t", "u64", "GLSL_TYPE_UINT64")
+int64_type = type("int64_t", "i64", "GLSL_TYPE_INT64")
float_type = type("float", "f", "GLSL_TYPE_FLOAT")
double_type = type("double", "d", "GLSL_TYPE_DOUBLE")
bool_type = type("bool", "b", "GLSL_TYPE_BOOL")
-all_types = (uint_type, int_type, float_type, double_type, bool_type)
-numeric_types = (uint_type, int_type, float_type, double_type)
-signed_numeric_types = (int_type, float_type, double_type)
-integer_types = (uint_type, int_type)
+all_types = (uint_type, int_type, float_type, double_type, uint64_type, int64_type, bool_type)
+numeric_types = (uint_type, int_type, float_type, double_type, uint64_type, int64_type)
+signed_numeric_types = (int_type, float_type, double_type, int64_type)
+integer_types = (uint_type, int_type, uint64_type, int64_type)
real_types = (float_type, double_type)
-# This template is for unary and binary operations that can only have operands
-# of a single type. ir_unop_logic_not is an example.
-constant_template0 = mako.template.Template("""\
- case ${op.get_enum_name()}:
- assert(op[0]->type->base_type == ${op.source_types[0].glsl_type});
- for (unsigned c = 0; c < op[0]->type->components(); c++)
- data.${op.source_types[0].union_field}[c] = ${op.get_c_expression(op.source_types)};
- break;""")
-
-# This template is for unary operations that can have operands of a several
-# different types. ir_unop_bit_not is an example.
-constant_template1 = mako.template.Template("""\
- case ${op.get_enum_name()}:
- switch (op[0]->type->base_type) {
- % for dst_type, src_types in op.signatures():
- case ${src_types[0].glsl_type}:
- for (unsigned c = 0; c < op[0]->type->components(); c++)
- data.${dst_type.union_field}[c] = ${op.get_c_expression(src_types)};
- break;
- % endfor
- default:
- assert(0);
- }
- break;""")
-
-# This template is for unary operations that can have operands of a several
-# different types, and each type has a different C expression. ir_unop_neg is
-# an example.
-constant_template3 = mako.template.Template("""\
+# This template is for operations that can have operands of a several
+# different types, and each type may or may not has a different C expression.
+# This is used by most operations.
+constant_template_common = mako.template.Template("""\
case ${op.get_enum_name()}:
for (unsigned c = 0; c < op[0]->type->components(); c++) {
- switch (this->type->base_type) {
- % for dst_type, src_types in op.signatures():
- case ${src_types[0].glsl_type}:
- data.${dst_type.union_field}[c] = ${op.get_c_expression(src_types)};
- break;
- % endfor
- default:
- assert(0);
- }
- }
- break;""")
-
-# This template is for unary operations that map an operand of one type to an
-# operand of another type. ir_unop_f2b is an example.
-constant_template2 = mako.template.Template("""\
- case ${op.get_enum_name()}:
- assert(op[0]->type->base_type == ${op.source_types[0].glsl_type});
- for (unsigned c = 0; c < op[0]->type->components(); c++)
- data.${op.dest_type.union_field}[c] = ${op.get_c_expression(op.source_types)};
- break;""")
-
-# This template is for operations with an output type that doesn't match the
-# input types.
-constant_template5 = mako.template.Template("""\
- case ${op.get_enum_name()}:
- for (unsigned c = 0; c < components; c++) {
switch (op[0]->type->base_type) {
% for dst_type, src_types in op.signatures():
case ${src_types[0].glsl_type}:
break;
% endfor
default:
- assert(0);
+ unreachable("invalid type");
}
}
break;""")
constant_template_vector_scalar = mako.template.Template("""\
case ${op.get_enum_name()}:
% if "mixed" in op.flags:
- % for i in xrange(op.num_operands):
+ % for i in range(op.num_operands):
assert(op[${i}]->type->base_type == ${op.source_types[0].glsl_type} ||
% for src_type in op.source_types[1:-1]:
op[${i}]->type->base_type == ${src_type.glsl_type} ||
switch (op[0]->type->base_type) {
% for dst_type, src_types in op.signatures():
case ${src_types[0].glsl_type}:
- data.${dst_type.union_field}[c] = ${op.get_c_expression(src_types, ("c0", "c1"))};
+ data.${dst_type.union_field}[c] = ${op.get_c_expression(src_types, ("c0", "c1", "c2"))};
break;
% endfor
default:
- assert(0);
+ unreachable("invalid type");
+ }
+ }
+ break;""")
+
+# This template is for multiplication. It is unique because it has to support
+# matrix * vector and matrix * matrix operations, and those are just different.
+constant_template_mul = mako.template.Template("""\
+ case ${op.get_enum_name()}:
+ /* Check for equal types, or unequal types involving scalars */
+ if ((op[0]->type == op[1]->type && !op[0]->type->is_matrix())
+ || op0_scalar || op1_scalar) {
+ for (unsigned c = 0, c0 = 0, c1 = 0;
+ c < components;
+ c0 += c0_inc, c1 += c1_inc, c++) {
+
+ switch (op[0]->type->base_type) {
+ % for dst_type, src_types in op.signatures():
+ case ${src_types[0].glsl_type}:
+ data.${dst_type.union_field}[c] = ${op.get_c_expression(src_types, ("c0", "c1", "c2"))};
+ break;
+ % endfor
+ default:
+ unreachable("invalid type");
+ }
+ }
+ } else {
+ assert(op[0]->type->is_matrix() || op[1]->type->is_matrix());
+
+ /* Multiply an N-by-M matrix with an M-by-P matrix. Since either
+ * matrix can be a GLSL vector, either N or P can be 1.
+ *
+ * For vec*mat, the vector is treated as a row vector. This
+ * means the vector is a 1-row x M-column matrix.
+ *
+ * For mat*vec, the vector is treated as a column vector. Since
+ * matrix_columns is 1 for vectors, this just works.
+ */
+ const unsigned n = op[0]->type->is_vector()
+ ? 1 : op[0]->type->vector_elements;
+ const unsigned m = op[1]->type->vector_elements;
+ const unsigned p = op[1]->type->matrix_columns;
+ for (unsigned j = 0; j < p; j++) {
+ for (unsigned i = 0; i < n; i++) {
+ for (unsigned k = 0; k < m; k++) {
+ if (op[0]->type->is_double())
+ data.d[i+n*j] += op[0]->value.d[i+n*k]*op[1]->value.d[k+m*j];
+ else
+ data.f[i+n*j] += op[0]->value.f[i+n*k]*op[1]->value.f[k+m*j];
+ }
+ }
}
}
break;""")
${op.c_expression['default']};
break;""")
+# This template is for binary operations that are horizontal. That is, the
+# operation consumes a vector and produces a scalar.
+constant_template_horizontal = mako.template.Template("""\
+ case ${op.get_enum_name()}:
+ switch (op[0]->type->base_type) {
+ % for dst_type, src_types in op.signatures():
+ case ${src_types[0].glsl_type}:
+ data.${dst_type.union_field}[0] = ${op.get_c_expression(src_types)};
+ break;
+ % endfor
+ default:
+ unreachable("invalid type");
+ }
+ break;""")
+
+# This template is for ir_binop_vector_extract.
+constant_template_vector_extract = mako.template.Template("""\
+ case ${op.get_enum_name()}: {
+ const int c = CLAMP(op[1]->value.i[0], 0,
+ (int) op[0]->type->vector_elements - 1);
+
+ switch (op[0]->type->base_type) {
+ % for dst_type, src_types in op.signatures():
+ case ${src_types[0].glsl_type}:
+ data.${dst_type.union_field}[0] = op[0]->value.${src_types[0].union_field}[c];
+ break;
+ % endfor
+ default:
+ unreachable("invalid type");
+ }
+ break;
+ }""")
+
+# This template is for ir_triop_vector_insert.
+constant_template_vector_insert = mako.template.Template("""\
+ case ${op.get_enum_name()}: {
+ const unsigned idx = op[2]->value.u[0];
+
+ memcpy(&data, &op[0]->value, sizeof(data));
+
+ switch (this->type->base_type) {
+ % for dst_type, src_types in op.signatures():
+ case ${src_types[0].glsl_type}:
+ data.${dst_type.union_field}[idx] = op[1]->value.${src_types[0].union_field}[0];
+ break;
+ % endfor
+ default:
+ unreachable("invalid type");
+ }
+ break;
+ }""")
+
+# This template is for ir_quadop_vector.
+constant_template_vector = mako.template.Template("""\
+ case ${op.get_enum_name()}:
+ for (unsigned c = 0; c < this->type->vector_elements; c++) {
+ switch (this->type->base_type) {
+ % for dst_type, src_types in op.signatures():
+ case ${src_types[0].glsl_type}:
+ data.${dst_type.union_field}[c] = op[c]->value.${src_types[0].union_field}[0];
+ break;
+ % endfor
+ default:
+ unreachable("invalid type");
+ }
+ }
+ break;""")
+
+# This template is for ir_triop_lrp.
+constant_template_lrp = mako.template.Template("""\
+ case ${op.get_enum_name()}: {
+ assert(op[0]->type->is_float() || op[0]->type->is_double());
+ assert(op[1]->type->is_float() || op[1]->type->is_double());
+ assert(op[2]->type->is_float() || op[2]->type->is_double());
+
+ unsigned c2_inc = op[2]->type->is_scalar() ? 0 : 1;
+ for (unsigned c = 0, c2 = 0; c < components; c2 += c2_inc, c++) {
+ switch (this->type->base_type) {
+ % for dst_type, src_types in op.signatures():
+ case ${src_types[0].glsl_type}:
+ data.${dst_type.union_field}[c] = ${op.get_c_expression(src_types, ("c", "c", "c2"))};
+ break;
+ % endfor
+ default:
+ unreachable("invalid type");
+ }
+ }
+ break;
+ }""")
+
+# This template is for ir_triop_csel. This expression is really unique
+# because not all of the operands are the same type, and the second operand
+# determines the type of the expression (instead of the first).
+constant_template_csel = mako.template.Template("""\
+ case ${op.get_enum_name()}:
+ for (unsigned c = 0; c < components; c++) {
+ switch (this->type->base_type) {
+ % for dst_type, src_types in op.signatures():
+ case ${src_types[1].glsl_type}:
+ data.${dst_type.union_field}[c] = ${op.get_c_expression(src_types)};
+ break;
+ % endfor
+ default:
+ unreachable("invalid type");
+ }
+ }
+ break;""")
+
vector_scalar_operation = "vector-scalar"
horizontal_operation = "horizontal"
def get_enum_name(self):
- return "ir_{}op_{}".format(("un", "bin", "tri", "quad")[self.num_operands-1], self.name)
+ return "ir_{0}op_{1}".format(("un", "bin", "tri", "quad")[self.num_operands-1], self.name)
def get_template(self):
if self.c_expression is None:
return None
- if self.num_operands == 1:
- if horizontal_operation in self.flags and non_assign_operation in self.flags:
+ if horizontal_operation in self.flags:
+ if non_assign_operation in self.flags:
return constant_template_horizontal_nonassignment.render(op=self)
- elif horizontal_operation in self.flags:
+ elif types_identical_operation in self.flags:
return constant_template_horizontal_single_implementation.render(op=self)
- elif self.dest_type is not None and len(self.source_types) == 1:
- return constant_template2.render(op=self)
- elif self.dest_type is not None:
- return constant_template5.render(op=self)
- elif len(self.source_types) == 1:
- return constant_template0.render(op=self)
- elif len(self.c_expression) == 1 and 'default' in self.c_expression:
- return constant_template1.render(op=self)
else:
- return constant_template3.render(op=self)
- elif self.num_operands == 2:
- if vector_scalar_operation in self.flags:
+ return constant_template_horizontal.render(op=self)
+
+ if self.num_operands == 2:
+ if self.name == "mul":
+ return constant_template_mul.render(op=self)
+ elif self.name == "vector_extract":
+ return constant_template_vector_extract.render(op=self)
+ elif vector_scalar_operation in self.flags:
return constant_template_vector_scalar.render(op=self)
- elif horizontal_operation in self.flags and types_identical_operation in self.flags:
- return constant_template_horizontal_single_implementation.render(op=self)
- elif len(self.source_types) == 1:
- return constant_template0.render(op=self)
- elif self.dest_type is not None:
- return constant_template5.render(op=self)
- else:
- return constant_template3.render(op=self)
-
- return None
-
-
- def get_c_expression(self, types, indices=("c", "c")):
- src0 = "op[0]->value.{}[{}]".format(types[0].union_field, indices[0])
- src1 = "op[1]->value.{}[{}]".format(types[1].union_field, indices[1]) if len(types) >= 2 else "ERROR"
+ elif self.num_operands == 3:
+ if self.name == "vector_insert":
+ return constant_template_vector_insert.render(op=self)
+ elif self.name == "lrp":
+ return constant_template_lrp.render(op=self)
+ elif self.name == "csel":
+ return constant_template_csel.render(op=self)
+ elif self.num_operands == 4:
+ if self.name == "vector":
+ return constant_template_vector.render(op=self)
+
+ return constant_template_common.render(op=self)
+
+
+ def get_c_expression(self, types, indices=("c", "c", "c")):
+ src0 = "op[0]->value.{0}[{1}]".format(types[0].union_field, indices[0])
+ src1 = "op[1]->value.{0}[{1}]".format(types[1].union_field, indices[1]) if len(types) >= 2 else "ERROR"
+ src2 = "op[2]->value.{0}[{1}]".format(types[2].union_field, indices[2]) if len(types) >= 3 else "ERROR"
+ src3 = "op[3]->value.{0}[c]".format(types[3].union_field) if len(types) >= 4 else "ERROR"
expr = self.c_expression[types[0].union_field] if types[0].union_field in self.c_expression else self.c_expression['default']
return expr.format(src0=src0,
- src1=src1)
+ src1=src1,
+ src2=src2,
+ src3=src3)
def signatures(self):
operation("bit_not", 1, printable_name="~", source_types=integer_types, c_expression="~ {src0}"),
operation("logic_not", 1, printable_name="!", source_types=(bool_type,), c_expression="!{src0}"),
operation("neg", 1, source_types=numeric_types, c_expression={'u': "-((int) {src0})", 'default': "-{src0}"}),
- operation("abs", 1, source_types=signed_numeric_types, c_expression={'i': "{src0} < 0 ? -{src0} : {src0}", 'f': "fabsf({src0})", 'd': "fabs({src0})"}),
- operation("sign", 1, source_types=signed_numeric_types, c_expression={'i': "({src0} > 0) - ({src0} < 0)", 'f': "float(({src0} > 0.0F) - ({src0} < 0.0F))", 'd': "double(({src0} > 0.0) - ({src0} < 0.0))"}),
- operation("rcp", 1, source_types=real_types, c_expression={'f': "{src0} != 0.0F ? 1.0F / {src0} : 0.0F", 'd': "{src0} != 0.0 ? 1.0 / {src0} : 0.0"}),
+ operation("abs", 1, source_types=signed_numeric_types, c_expression={'i': "{src0} < 0 ? -{src0} : {src0}", 'f': "fabsf({src0})", 'd': "fabs({src0})", 'i64': "{src0} < 0 ? -{src0} : {src0}"}),
+ operation("sign", 1, source_types=signed_numeric_types, c_expression={'i': "({src0} > 0) - ({src0} < 0)", 'f': "float(({src0} > 0.0F) - ({src0} < 0.0F))", 'd': "double(({src0} > 0.0) - ({src0} < 0.0))", 'i64': "({src0} > 0) - ({src0} < 0)"}),
+ operation("rcp", 1, source_types=real_types, c_expression={'f': "1.0F / {src0}", 'd': "1.0 / {src0}"}),
operation("rsq", 1, source_types=real_types, c_expression={'f': "1.0F / sqrtf({src0})", 'd': "1.0 / sqrt({src0})"}),
operation("sqrt", 1, source_types=real_types, c_expression={'f': "sqrtf({src0})", 'd': "sqrt({src0})"}),
operation("exp", 1, source_types=(float_type,), c_expression="expf({src0})"), # Log base e on gentype
# Boolean-to-float conversion
operation("b2f", 1, source_types=(bool_type,), dest_type=float_type, c_expression="{src0} ? 1.0F : 0.0F"),
# int-to-boolean conversion
- operation("i2b", 1, source_types=integer_types, dest_type=bool_type, c_expression="{src0} ? true : false"),
+ operation("i2b", 1, source_types=(uint_type, int_type), dest_type=bool_type, c_expression="{src0} ? true : false"),
# Boolean-to-int conversion
operation("b2i", 1, source_types=(bool_type,), dest_type=int_type, c_expression="{src0} ? 1 : 0"),
# Unsigned-to-float conversion.
operation("bitcast_u2f", 1, source_types=(uint_type,), dest_type=float_type, c_expression="bitcast_u2f({src0})"),
# 'Bit-identical float-to-uint "conversion"
operation("bitcast_f2u", 1, source_types=(float_type,), dest_type=uint_type, c_expression="bitcast_f2u({src0})"),
+ # Bit-identical u64-to-double "conversion"
+ operation("bitcast_u642d", 1, source_types=(uint64_type,), dest_type=double_type, c_expression="bitcast_u642d({src0})"),
+ # Bit-identical i64-to-double "conversion"
+ operation("bitcast_i642d", 1, source_types=(int64_type,), dest_type=double_type, c_expression="bitcast_i642d({src0})"),
+ # Bit-identical double-to_u64 "conversion"
+ operation("bitcast_d2u64", 1, source_types=(double_type,), dest_type=uint64_type, c_expression="bitcast_d2u64({src0})"),
+ # Bit-identical double-to-i64 "conversion"
+ operation("bitcast_d2i64", 1, source_types=(double_type,), dest_type=int64_type, c_expression="bitcast_d2i64({src0})"),
+ # i64-to-i32 conversion
+ operation("i642i", 1, source_types=(int64_type,), dest_type=int_type, c_expression="{src0}"),
+ # ui64-to-i32 conversion
+ operation("u642i", 1, source_types=(uint64_type,), dest_type=int_type, c_expression="{src0}"),
+ operation("i642u", 1, source_types=(int64_type,), dest_type=uint_type, c_expression="{src0}"),
+ operation("u642u", 1, source_types=(uint64_type,), dest_type=uint_type, c_expression="{src0}"),
+ operation("i642b", 1, source_types=(int64_type,), dest_type=bool_type, c_expression="{src0} != 0"),
+ operation("i642f", 1, source_types=(int64_type,), dest_type=float_type, c_expression="{src0}"),
+ operation("u642f", 1, source_types=(uint64_type,), dest_type=float_type, c_expression="{src0}"),
+ operation("i642d", 1, source_types=(int64_type,), dest_type=double_type, c_expression="{src0}"),
+ operation("u642d", 1, source_types=(uint64_type,), dest_type=double_type, c_expression="{src0}"),
+ operation("i2i64", 1, source_types=(int_type,), dest_type=int64_type, c_expression="{src0}"),
+ operation("u2i64", 1, source_types=(uint_type,), dest_type=int64_type, c_expression="{src0}"),
+ operation("b2i64", 1, source_types=(bool_type,), dest_type=int64_type, c_expression="{src0}"),
+ operation("f2i64", 1, source_types=(float_type,), dest_type=int64_type, c_expression="{src0}"),
+ operation("d2i64", 1, source_types=(double_type,), dest_type=int64_type, c_expression="{src0}"),
+ operation("i2u64", 1, source_types=(int_type,), dest_type=uint64_type, c_expression="{src0}"),
+ operation("u2u64", 1, source_types=(uint_type,), dest_type=uint64_type, c_expression="{src0}"),
+ operation("f2u64", 1, source_types=(float_type,), dest_type=uint64_type, c_expression="{src0}"),
+ operation("d2u64", 1, source_types=(double_type,), dest_type=uint64_type, c_expression="{src0}"),
+ operation("u642i64", 1, source_types=(uint64_type,), dest_type=int64_type, c_expression="{src0}"),
+ operation("i642u64", 1, source_types=(int64_type,), dest_type=uint64_type, c_expression="{src0}"),
+
# Unary floating-point rounding operations.
operation("trunc", 1, source_types=real_types, c_expression={'f': "truncf({src0})", 'd': "trunc({src0})"}),
operation("unpack_half_2x16", 1, printable_name="unpackHalf2x16", source_types=(uint_type,), dest_type=float_type, c_expression="unpack_2x16(unpack_half_1x16, op[0]->value.u[0], &data.f[0], &data.f[1])", flags=frozenset((horizontal_operation, non_assign_operation))),
# Bit operations, part of ARB_gpu_shader5.
- operation("bitfield_reverse", 1, source_types=integer_types, c_expression="bitfield_reverse({src0})"),
- operation("bit_count", 1, source_types=integer_types, dest_type=int_type, c_expression="_mesa_bitcount({src0})"),
- operation("find_msb", 1, source_types=integer_types, dest_type=int_type, c_expression={'u': "find_msb_uint({src0})", 'i': "find_msb_int({src0})"}),
- operation("find_lsb", 1, source_types=integer_types, dest_type=int_type, c_expression="find_msb_uint({src0} & -{src0})"),
+ operation("bitfield_reverse", 1, source_types=(uint_type, int_type), c_expression="bitfield_reverse({src0})"),
+ operation("bit_count", 1, source_types=(uint_type, int_type), dest_type=int_type, c_expression="util_bitcount({src0})"),
+ operation("find_msb", 1, source_types=(uint_type, int_type), dest_type=int_type, c_expression={'u': "find_msb_uint({src0})", 'i': "find_msb_int({src0})"}),
+ operation("find_lsb", 1, source_types=(uint_type, int_type), dest_type=int_type, c_expression="find_msb_uint({src0} & -{src0})"),
operation("saturate", 1, printable_name="sat", source_types=(float_type,), c_expression="CLAMP({src0}, 0.0f, 1.0f)"),
operation("pack_double_2x32", 1, printable_name="packDouble2x32", source_types=(uint_type,), dest_type=double_type, c_expression="memcpy(&data.d[0], &op[0]->value.u[0], sizeof(double))", flags=frozenset((horizontal_operation, non_assign_operation))),
operation("unpack_double_2x32", 1, printable_name="unpackDouble2x32", source_types=(double_type,), dest_type=uint_type, c_expression="memcpy(&data.u[0], &op[0]->value.d[0], sizeof(double))", flags=frozenset((horizontal_operation, non_assign_operation))),
+ # Sampler/Image packing, part of ARB_bindless_texture.
+ operation("pack_sampler_2x32", 1, printable_name="packSampler2x32", source_types=(uint_type,), dest_type=uint64_type, c_expression="memcpy(&data.u64[0], &op[0]->value.u[0], sizeof(uint64_t))", flags=frozenset((horizontal_operation, non_assign_operation))),
+ operation("pack_image_2x32", 1, printable_name="packImage2x32", source_types=(uint_type,), dest_type=uint64_type, c_expression="memcpy(&data.u64[0], &op[0]->value.u[0], sizeof(uint64_t))", flags=frozenset((horizontal_operation, non_assign_operation))),
+ operation("unpack_sampler_2x32", 1, printable_name="unpackSampler2x32", source_types=(uint64_type,), dest_type=uint_type, c_expression="memcpy(&data.u[0], &op[0]->value.u64[0], sizeof(uint64_t))", flags=frozenset((horizontal_operation, non_assign_operation))),
+ operation("unpack_image_2x32", 1, printable_name="unpackImage2x32", source_types=(uint64_type,), dest_type=uint_type, c_expression="memcpy(&data.u[0], &op[0]->value.u64[0], sizeof(uint64_t))", flags=frozenset((horizontal_operation, non_assign_operation))),
+
operation("frexp_sig", 1),
operation("frexp_exp", 1),
# of its length.
operation("ssbo_unsized_array_length", 1),
- # Vote among threads on the value of the boolean argument.
- operation("vote_any", 1),
- operation("vote_all", 1),
- operation("vote_eq", 1),
+ # 64-bit integer packing ops.
+ operation("pack_int_2x32", 1, printable_name="packInt2x32", source_types=(int_type,), dest_type=int64_type, c_expression="memcpy(&data.i64[0], &op[0]->value.i[0], sizeof(int64_t))", flags=frozenset((horizontal_operation, non_assign_operation))),
+ operation("pack_uint_2x32", 1, printable_name="packUint2x32", source_types=(uint_type,), dest_type=uint64_type, c_expression="memcpy(&data.u64[0], &op[0]->value.u[0], sizeof(uint64_t))", flags=frozenset((horizontal_operation, non_assign_operation))),
+ operation("unpack_int_2x32", 1, printable_name="unpackInt2x32", source_types=(int64_type,), dest_type=int_type, c_expression="memcpy(&data.i[0], &op[0]->value.i64[0], sizeof(int64_t))", flags=frozenset((horizontal_operation, non_assign_operation))),
+ operation("unpack_uint_2x32", 1, printable_name="unpackUint2x32", source_types=(uint64_type,), dest_type=uint_type, c_expression="memcpy(&data.u[0], &op[0]->value.u64[0], sizeof(uint64_t))", flags=frozenset((horizontal_operation, non_assign_operation))),
operation("add", 2, printable_name="+", source_types=numeric_types, c_expression="{src0} + {src1}", flags=vector_scalar_operation),
operation("sub", 2, printable_name="-", source_types=numeric_types, c_expression="{src0} - {src1}", flags=vector_scalar_operation),
# "Floating-point or low 32-bit integer multiply."
- operation("mul", 2, printable_name="*"),
+ operation("mul", 2, printable_name="*", source_types=numeric_types, c_expression="{src0} * {src1}"),
operation("imul_high", 2), # Calculates the high 32-bits of a 64-bit multiply.
- operation("div", 2, printable_name="/", source_types=numeric_types, c_expression={'u': "{src1} == 0 ? 0 : {src0} / {src1}", 'i': "{src1} == 0 ? 0 : {src0} / {src1}", 'default': "{src0} / {src1}"}, flags=vector_scalar_operation),
+ operation("div", 2, printable_name="/", source_types=numeric_types, c_expression={'u': "{src1} == 0 ? 0 : {src0} / {src1}", 'i': "{src1} == 0 ? 0 : {src0} / {src1}", 'u64': "{src1} == 0 ? 0 : {src0} / {src1}", 'i64': "{src1} == 0 ? 0 : {src0} / {src1}", 'default': "{src0} / {src1}"}, flags=vector_scalar_operation),
# Returns the carry resulting from the addition of the two arguments.
operation("carry", 2),
#
# We don't use fmod because it rounds toward zero; GLSL specifies the use
# of floor.
- operation("mod", 2, printable_name="%", source_types=numeric_types, c_expression={'u': "{src1} == 0 ? 0 : {src0} % {src1}", 'i': "{src1} == 0 ? 0 : {src0} % {src1}", 'f': "{src0} - {src1} * floorf({src0} / {src1})", 'd': "{src0} - {src1} * floor({src0} / {src1})"}, flags=vector_scalar_operation),
+ operation("mod", 2, printable_name="%", source_types=numeric_types, c_expression={'u': "{src1} == 0 ? 0 : {src0} % {src1}", 'i': "{src1} == 0 ? 0 : {src0} % {src1}", 'f': "{src0} - {src1} * floorf({src0} / {src1})", 'd': "{src0} - {src1} * floor({src0} / {src1})", 'u64': "{src1} == 0 ? 0 : {src0} % {src1}", 'i64': "{src1} == 0 ? 0 : {src0} % {src1}"}, flags=vector_scalar_operation),
# Binary comparison operators which return a boolean vector.
# The type of both operands must be equal.
operation("less", 2, printable_name="<", source_types=numeric_types, dest_type=bool_type, c_expression="{src0} < {src1}"),
- operation("greater", 2, printable_name=">", source_types=numeric_types, dest_type=bool_type, c_expression="{src0} > {src1}"),
- operation("lequal", 2, printable_name="<=", source_types=numeric_types, dest_type=bool_type, c_expression="{src0} <= {src1}"),
operation("gequal", 2, printable_name=">=", source_types=numeric_types, dest_type=bool_type, c_expression="{src0} >= {src1}"),
operation("equal", 2, printable_name="==", source_types=all_types, dest_type=bool_type, c_expression="{src0} == {src1}"),
operation("nequal", 2, printable_name="!=", source_types=all_types, dest_type=bool_type, c_expression="{src0} != {src1}"),
operation("logic_xor", 2, printable_name="^^", source_types=(bool_type,), c_expression="{src0} != {src1}"),
operation("logic_or", 2, printable_name="||", source_types=(bool_type,), c_expression="{src0} || {src1}"),
- operation("dot", 2),
+ operation("dot", 2, source_types=real_types, c_expression={'f': "dot_f(op[0], op[1])", 'd': "dot_d(op[0], op[1])"}, flags=horizontal_operation),
operation("min", 2, source_types=numeric_types, c_expression="MIN2({src0}, {src1})", flags=vector_scalar_operation),
operation("max", 2, source_types=numeric_types, c_expression="MAX2({src0}, {src1})", flags=vector_scalar_operation),
#
# operand0 is the vector
# operand1 is the index of the field to read from operand0
- operation("vector_extract", 2),
+ operation("vector_extract", 2, source_types=all_types, c_expression="anything-except-None"),
# Interpolate fs input at offset
#
operation("interpolate_at_sample", 2),
# Fused floating-point multiply-add, part of ARB_gpu_shader5.
- operation("fma", 3),
+ operation("fma", 3, source_types=real_types, c_expression="{src0} * {src1} + {src2}"),
- operation("lrp", 3),
+ operation("lrp", 3, source_types=real_types, c_expression={'f': "{src0} * (1.0f - {src2}) + ({src1} * {src2})", 'd': "{src0} * (1.0 - {src2}) + ({src1} * {src2})"}),
# Conditional Select
#
# component on vectors).
#
# See also lower_instructions_visitor::ldexp_to_arith
- operation("csel", 3),
+ operation("csel", 3,
+ all_signatures=zip(all_types, zip(len(all_types) * (bool_type,), all_types, all_types)),
+ c_expression="{src0} ? {src1} : {src2}"),
- operation("bitfield_extract", 3),
+ operation("bitfield_extract", 3,
+ all_signatures=((int_type, (uint_type, int_type, int_type)),
+ (int_type, (int_type, int_type, int_type))),
+ c_expression={'u': "bitfield_extract_uint({src0}, {src1}, {src2})",
+ 'i': "bitfield_extract_int({src0}, {src1}, {src2})"}),
# Generate a value with one field of a vector changed
#
# operand0 is the vector
# operand1 is the value to write into the vector result
# operand2 is the index in operand0 to be modified
- operation("vector_insert", 3),
+ operation("vector_insert", 3, source_types=all_types, c_expression="anything-except-None"),
- operation("bitfield_insert", 4),
+ operation("bitfield_insert", 4,
+ all_signatures=((uint_type, (uint_type, uint_type, int_type, int_type)),
+ (int_type, (int_type, int_type, int_type, int_type))),
+ c_expression="bitfield_insert({src0}, {src1}, {src2}, {src3})"),
- operation("vector", 4),
+ operation("vector", 4, source_types=all_types, c_expression="anything-except-None"),
]
% for item in values:
"${item.printable_name}",
% endfor
+};
+
+const char *const ir_expression_operation_enum_strings[] = {
+% for item in values:
+ "${item.name}",
+% endfor
};""")
constant_template = mako.template.Template("""\