*/
#include <math.h>
-#include "main/core.h" /* for MAX2, MIN2, CLAMP */
#include "util/rounding.h" /* for _mesa_roundeven */
#include "util/half_float.h"
#include "ir.h"
#include "compiler/glsl_types.h"
-#include "program/hash_table.h"
+#include "util/hash_table.h"
+#include "util/u_math.h"
static float
dot_f(ir_constant *op0, ir_constant *op1)
static float
bitcast_u2f(unsigned int u)
{
- assert(sizeof(float) == sizeof(unsigned int));
+ static_assert(sizeof(float) == sizeof(unsigned int),
+ "float and unsigned int size mismatch");
float f;
memcpy(&f, &u, sizeof(f));
return f;
static unsigned int
bitcast_f2u(float f)
{
- assert(sizeof(float) == sizeof(unsigned int));
+ static_assert(sizeof(float) == sizeof(unsigned int),
+ "float and unsigned int size mismatch");
unsigned int u;
memcpy(&u, &f, sizeof(f));
return u;
}
+static double
+bitcast_u642d(uint64_t u)
+{
+ static_assert(sizeof(double) == sizeof(uint64_t),
+ "double and uint64_t size mismatch");
+ double d;
+ memcpy(&d, &u, sizeof(d));
+ return d;
+}
+
+static double
+bitcast_i642d(int64_t i)
+{
+ static_assert(sizeof(double) == sizeof(int64_t),
+ "double and int64_t size mismatch");
+ double d;
+ memcpy(&d, &i, sizeof(d));
+ return d;
+}
+
+static uint64_t
+bitcast_d2u64(double d)
+{
+ static_assert(sizeof(double) == sizeof(uint64_t),
+ "double and uint64_t size mismatch");
+ uint64_t u;
+ memcpy(&u, &d, sizeof(d));
+ return u;
+}
+
+static int64_t
+bitcast_d2i64(double d)
+{
+ static_assert(sizeof(double) == sizeof(int64_t),
+ "double and int64_t size mismatch");
+ int64_t i;
+ memcpy(&i, &d, sizeof(d));
+ return i;
+}
+
/**
* Evaluate one component of a floating-point 4x8 unpacking function.
*/
*
* packUnorm4x8: round(clamp(c, 0, +1) * 255.0)
*/
- return (uint8_t) (int) _mesa_roundevenf(CLAMP(x, 0.0f, 1.0f) * 255.0f);
+ return (uint8_t) (int) _mesa_roundevenf(SATURATE(x) * 255.0f);
}
/**
* packUnorm2x16: round(clamp(c, 0, +1) * 65535.0)
*/
return (uint16_t) (int)
- _mesa_roundevenf(CLAMP(x, 0.0f, 1.0f) * 65535.0f);
+ _mesa_roundevenf(SATURATE(x) * 65535.0f);
}
/**
return _mesa_half_to_float(u);
}
+static int32_t
+iadd_saturate(int32_t a, int32_t b)
+{
+ return CLAMP(int64_t(a) + int64_t(b), INT32_MIN, INT32_MAX);
+}
+
+static int64_t
+iadd64_saturate(int64_t a, int64_t b)
+{
+ if (a < 0 && b < INT64_MIN - a)
+ return INT64_MIN;
+
+ if (a > 0 && b > INT64_MAX - a)
+ return INT64_MAX;
+
+ return a + b;
+}
+
+static int32_t
+isub_saturate(int32_t a, int32_t b)
+{
+ return CLAMP(int64_t(a) - int64_t(b), INT32_MIN, INT32_MAX);
+}
+
+static int64_t
+isub64_saturate(int64_t a, int64_t b)
+{
+ if (b > 0 && a < INT64_MIN + b)
+ return INT64_MIN;
+
+ if (b < 0 && a > INT64_MAX + b)
+ return INT64_MAX;
+
+ return a - b;
+}
+
+static uint64_t
+pack_2x32(uint32_t a, uint32_t b)
+{
+ uint64_t v = a;
+ v |= (uint64_t)b << 32;
+ return v;
+}
+
+static void
+unpack_2x32(uint64_t p, uint32_t *a, uint32_t *b)
+{
+ *a = p & 0xffffffff;
+ *b = (p >> 32);
+}
+
/**
* Get the constant that is ultimately referenced by an r-value, in a constant
* expression evaluation context.
ir_constant *const index_c =
da->array_index->constant_expression_value(variable_context);
- if (!index_c || !index_c->type->is_scalar() || !index_c->type->is_integer())
+ if (!index_c || !index_c->type->is_scalar() ||
+ !index_c->type->is_integer_32())
break;
const int index = index_c->type->base_type == GLSL_TYPE_INT ?
*/
assert(suboffset == 0);
- store = substore->get_record_field(dr->field);
+ store = substore->get_record_field(dr->field_idx);
break;
}
const ir_dereference_variable *const dv =
(const ir_dereference_variable *) deref;
- store = (ir_constant *) hash_table_find(variable_context, dv->var);
+ hash_entry *entry = _mesa_hash_table_search(variable_context, dv->var);
+ if (entry)
+ store = (ir_constant *) entry->data;
break;
}
ir_constant *
-ir_rvalue::constant_expression_value(struct hash_table *)
+ir_rvalue::constant_expression_value(void *, struct hash_table *)
{
assert(this->type->is_error());
return NULL;
return !isnormal(result) ? copysign(0.0, x) : result;
}
+static uint32_t
+bitfield_extract_uint(uint32_t value, int offset, int bits)
+{
+ if (bits == 0)
+ return 0;
+ else if (offset < 0 || bits < 0)
+ return 0; /* Undefined, per spec. */
+ else if (offset + bits > 32)
+ return 0; /* Undefined, per spec. */
+ else {
+ value <<= 32 - bits - offset;
+ value >>= 32 - bits;
+ return value;
+ }
+}
+
+static int32_t
+bitfield_extract_int(int32_t value, int offset, int bits)
+{
+ if (bits == 0)
+ return 0;
+ else if (offset < 0 || bits < 0)
+ return 0; /* Undefined, per spec. */
+ else if (offset + bits > 32)
+ return 0; /* Undefined, per spec. */
+ else {
+ value <<= 32 - bits - offset;
+ value >>= 32 - bits;
+ return value;
+ }
+}
+
+static uint32_t
+bitfield_insert(uint32_t base, uint32_t insert, int offset, int bits)
+{
+ if (bits == 0)
+ return base;
+ else if (offset < 0 || bits < 0)
+ return 0; /* Undefined, per spec. */
+ else if (offset + bits > 32)
+ return 0; /* Undefined, per spec. */
+ else {
+ unsigned insert_mask = ((1ull << bits) - 1) << offset;
+
+ insert <<= offset;
+ insert &= insert_mask;
+ base &= ~insert_mask;
+
+ return base | insert;
+ }
+}
+
ir_constant *
-ir_expression::constant_expression_value(struct hash_table *variable_context)
+ir_expression::constant_expression_value(void *mem_ctx,
+ struct hash_table *variable_context)
{
+ assert(mem_ctx);
+
if (this->type->is_error())
return NULL;
+ const glsl_type *return_type = this->type;
ir_constant *op[ARRAY_SIZE(this->operands)] = { NULL, };
ir_constant_data data;
memset(&data, 0, sizeof(data));
- for (unsigned operand = 0; operand < this->get_num_operands(); operand++) {
- op[operand] = this->operands[operand]->constant_expression_value(variable_context);
+ for (unsigned operand = 0; operand < this->num_operands; operand++) {
+ op[operand] =
+ this->operands[operand]->constant_expression_value(mem_ctx,
+ variable_context);
if (!op[operand])
return NULL;
}
+ for (unsigned operand = 0; operand < this->num_operands; operand++) {
+ switch (op[operand]->type->base_type) {
+ case GLSL_TYPE_FLOAT16: {
+ const struct glsl_type *float_type =
+ glsl_type::get_instance(GLSL_TYPE_FLOAT,
+ op[operand]->type->vector_elements,
+ op[operand]->type->matrix_columns,
+ op[operand]->type->explicit_stride,
+ op[operand]->type->interface_row_major);
+
+ ir_constant_data f;
+ for (unsigned i = 0; i < ARRAY_SIZE(f.f); i++)
+ f.f[i] = _mesa_half_to_float(op[operand]->value.f16[i]);
+
+ op[operand] = new(mem_ctx) ir_constant(float_type, &f);
+ break;
+ }
+ case GLSL_TYPE_INT16: {
+ const struct glsl_type *int_type =
+ glsl_type::get_instance(GLSL_TYPE_INT,
+ op[operand]->type->vector_elements,
+ op[operand]->type->matrix_columns,
+ op[operand]->type->explicit_stride,
+ op[operand]->type->interface_row_major);
+
+ ir_constant_data d;
+ for (unsigned i = 0; i < ARRAY_SIZE(d.i); i++)
+ d.i[i] = op[operand]->value.i16[i];
+
+ op[operand] = new(mem_ctx) ir_constant(int_type, &d);
+ break;
+ }
+ case GLSL_TYPE_UINT16: {
+ const struct glsl_type *uint_type =
+ glsl_type::get_instance(GLSL_TYPE_UINT,
+ op[operand]->type->vector_elements,
+ op[operand]->type->matrix_columns,
+ op[operand]->type->explicit_stride,
+ op[operand]->type->interface_row_major);
+
+ ir_constant_data d;
+ for (unsigned i = 0; i < ARRAY_SIZE(d.u); i++)
+ d.u[i] = op[operand]->value.u16[i];
+
+ op[operand] = new(mem_ctx) ir_constant(uint_type, &d);
+ break;
+ }
+ default:
+ /* nothing to do */
+ break;
+ }
+ }
+
+ switch (return_type->base_type) {
+ case GLSL_TYPE_FLOAT16:
+ return_type = glsl_type::get_instance(GLSL_TYPE_FLOAT,
+ return_type->vector_elements,
+ return_type->matrix_columns,
+ return_type->explicit_stride,
+ return_type->interface_row_major);
+ break;
+ case GLSL_TYPE_INT16:
+ return_type = glsl_type::get_instance(GLSL_TYPE_INT,
+ return_type->vector_elements,
+ return_type->matrix_columns,
+ return_type->explicit_stride,
+ return_type->interface_row_major);
+ break;
+ case GLSL_TYPE_UINT16:
+ return_type = glsl_type::get_instance(GLSL_TYPE_UINT,
+ return_type->vector_elements,
+ return_type->matrix_columns,
+ return_type->explicit_stride,
+ return_type->interface_row_major);
+ break;
+ default:
+ /* nothing to do */
+ break;
+ }
+
if (op[1] != NULL)
switch (this->operation) {
case ir_binop_lshift:
components = op[1]->type->components();
}
- void *ctx = ralloc_parent(this);
-
/* Handle array operations here, rather than below. */
if (op[0]->type->is_array()) {
assert(op[1] != NULL && op[1]->type->is_array());
switch (this->operation) {
case ir_binop_all_equal:
- return new(ctx) ir_constant(op[0]->has_value(op[1]));
+ return new(mem_ctx) ir_constant(op[0]->has_value(op[1]));
case ir_binop_any_nequal:
- return new(ctx) ir_constant(!op[0]->has_value(op[1]));
+ return new(mem_ctx) ir_constant(!op[0]->has_value(op[1]));
default:
break;
}
return NULL;
}
- switch (this->operation) {
- case ir_unop_bit_not:
- switch (op[0]->type->base_type) {
- case GLSL_TYPE_INT:
- for (unsigned c = 0; c < components; c++)
- data.i[c] = ~ op[0]->value.i[c];
- break;
- case GLSL_TYPE_UINT:
- for (unsigned c = 0; c < components; c++)
- data.u[c] = ~ op[0]->value.u[c];
- break;
- default:
- assert(0);
- }
- break;
-
- case ir_unop_logic_not:
- assert(op[0]->type->base_type == GLSL_TYPE_BOOL);
- for (unsigned c = 0; c < op[0]->type->components(); c++)
- data.b[c] = !op[0]->value.b[c];
- break;
-
- case ir_unop_f2i:
- assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);
- for (unsigned c = 0; c < op[0]->type->components(); c++) {
- data.i[c] = (int) op[0]->value.f[c];
- }
- break;
- case ir_unop_f2u:
- assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);
- for (unsigned c = 0; c < op[0]->type->components(); c++) {
- data.u[c] = (unsigned) op[0]->value.f[c];
- }
- break;
- case ir_unop_i2f:
- assert(op[0]->type->base_type == GLSL_TYPE_INT);
- for (unsigned c = 0; c < op[0]->type->components(); c++) {
- data.f[c] = (float) op[0]->value.i[c];
- }
- break;
- case ir_unop_u2f:
- assert(op[0]->type->base_type == GLSL_TYPE_UINT);
- for (unsigned c = 0; c < op[0]->type->components(); c++) {
- data.f[c] = (float) op[0]->value.u[c];
- }
- break;
- case ir_unop_b2f:
- assert(op[0]->type->base_type == GLSL_TYPE_BOOL);
- for (unsigned c = 0; c < op[0]->type->components(); c++) {
- data.f[c] = op[0]->value.b[c] ? 1.0F : 0.0F;
- }
- break;
- case ir_unop_f2b:
- assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);
- for (unsigned c = 0; c < op[0]->type->components(); c++) {
- data.b[c] = op[0]->value.f[c] != 0.0F ? true : false;
- }
- break;
- case ir_unop_b2i:
- assert(op[0]->type->base_type == GLSL_TYPE_BOOL);
- for (unsigned c = 0; c < op[0]->type->components(); c++) {
- data.i[c] = op[0]->value.b[c] ? 1 : 0;
- }
- break;
- case ir_unop_i2b:
- assert(op[0]->type->is_integer());
- for (unsigned c = 0; c < op[0]->type->components(); c++) {
- data.b[c] = op[0]->value.u[c] ? true : false;
- }
- break;
- case ir_unop_u2i:
- assert(op[0]->type->base_type == GLSL_TYPE_UINT);
- for (unsigned c = 0; c < op[0]->type->components(); c++) {
- data.i[c] = op[0]->value.u[c];
- }
- break;
- case ir_unop_i2u:
- assert(op[0]->type->base_type == GLSL_TYPE_INT);
- for (unsigned c = 0; c < op[0]->type->components(); c++) {
- data.u[c] = op[0]->value.i[c];
- }
- break;
- case ir_unop_bitcast_i2f:
- assert(op[0]->type->base_type == GLSL_TYPE_INT);
- for (unsigned c = 0; c < op[0]->type->components(); c++) {
- data.f[c] = bitcast_u2f(op[0]->value.i[c]);
- }
- break;
- case ir_unop_bitcast_f2i:
- assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);
- for (unsigned c = 0; c < op[0]->type->components(); c++) {
- data.i[c] = bitcast_f2u(op[0]->value.f[c]);
- }
- break;
- case ir_unop_bitcast_u2f:
- assert(op[0]->type->base_type == GLSL_TYPE_UINT);
- for (unsigned c = 0; c < op[0]->type->components(); c++) {
- data.f[c] = bitcast_u2f(op[0]->value.u[c]);
- }
- break;
- case ir_unop_bitcast_f2u:
- assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);
- for (unsigned c = 0; c < op[0]->type->components(); c++) {
- data.u[c] = bitcast_f2u(op[0]->value.f[c]);
- }
- break;
- case ir_unop_d2f:
- assert(op[0]->type->base_type == GLSL_TYPE_DOUBLE);
- for (unsigned c = 0; c < op[0]->type->components(); c++) {
- data.f[c] = op[0]->value.d[c];
- }
- break;
- case ir_unop_f2d:
- assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);
- for (unsigned c = 0; c < op[0]->type->components(); c++) {
- data.d[c] = op[0]->value.f[c];
- }
- break;
- case ir_unop_d2i:
- assert(op[0]->type->base_type == GLSL_TYPE_DOUBLE);
- for (unsigned c = 0; c < op[0]->type->components(); c++) {
- data.i[c] = op[0]->value.d[c];
- }
- break;
- case ir_unop_i2d:
- assert(op[0]->type->base_type == GLSL_TYPE_INT);
- for (unsigned c = 0; c < op[0]->type->components(); c++) {
- data.d[c] = op[0]->value.i[c];
- }
- break;
- case ir_unop_d2u:
- assert(op[0]->type->base_type == GLSL_TYPE_DOUBLE);
- for (unsigned c = 0; c < op[0]->type->components(); c++) {
- data.u[c] = op[0]->value.d[c];
- }
- break;
- case ir_unop_u2d:
- assert(op[0]->type->base_type == GLSL_TYPE_UINT);
- for (unsigned c = 0; c < op[0]->type->components(); c++) {
- data.d[c] = op[0]->value.u[c];
- }
- break;
- case ir_unop_d2b:
- assert(op[0]->type->base_type == GLSL_TYPE_DOUBLE);
- for (unsigned c = 0; c < op[0]->type->components(); c++) {
- data.b[c] = op[0]->value.d[c] != 0.0;
- }
- break;
- case ir_unop_trunc:
- for (unsigned c = 0; c < op[0]->type->components(); c++) {
- if (op[0]->type->base_type == GLSL_TYPE_DOUBLE)
- data.d[c] = trunc(op[0]->value.d[c]);
- else
- data.f[c] = truncf(op[0]->value.f[c]);
- }
- break;
-
- case ir_unop_round_even:
- for (unsigned c = 0; c < op[0]->type->components(); c++) {
- if (op[0]->type->base_type == GLSL_TYPE_DOUBLE)
- data.d[c] = _mesa_roundeven(op[0]->value.d[c]);
- else
- data.f[c] = _mesa_roundevenf(op[0]->value.f[c]);
- }
- break;
-
- case ir_unop_ceil:
- for (unsigned c = 0; c < op[0]->type->components(); c++) {
- if (op[0]->type->base_type == GLSL_TYPE_DOUBLE)
- data.d[c] = ceil(op[0]->value.d[c]);
- else
- data.f[c] = ceilf(op[0]->value.f[c]);
- }
- break;
-
- case ir_unop_floor:
- for (unsigned c = 0; c < op[0]->type->components(); c++) {
- if (op[0]->type->base_type == GLSL_TYPE_DOUBLE)
- data.d[c] = floor(op[0]->value.d[c]);
- else
- data.f[c] = floorf(op[0]->value.f[c]);
- }
- break;
-
- case ir_unop_fract:
- for (unsigned c = 0; c < op[0]->type->components(); c++) {
- switch (this->type->base_type) {
- case GLSL_TYPE_FLOAT:
- data.f[c] = op[0]->value.f[c] - floor(op[0]->value.f[c]);
- break;
- case GLSL_TYPE_DOUBLE:
- data.d[c] = op[0]->value.d[c] - floor(op[0]->value.d[c]);
- break;
- default:
- assert(0);
- }
- }
- break;
-
- case ir_unop_sin:
- assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);
- for (unsigned c = 0; c < op[0]->type->components(); c++) {
- data.f[c] = sinf(op[0]->value.f[c]);
- }
- break;
-
- case ir_unop_cos:
- assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);
- for (unsigned c = 0; c < op[0]->type->components(); c++) {
- data.f[c] = cosf(op[0]->value.f[c]);
- }
- break;
-
- case ir_unop_neg:
- for (unsigned c = 0; c < op[0]->type->components(); c++) {
- switch (this->type->base_type) {
- case GLSL_TYPE_UINT:
- data.u[c] = -((int) op[0]->value.u[c]);
- break;
- case GLSL_TYPE_INT:
- data.i[c] = -op[0]->value.i[c];
- break;
- case GLSL_TYPE_FLOAT:
- data.f[c] = -op[0]->value.f[c];
- break;
- case GLSL_TYPE_DOUBLE:
- data.d[c] = -op[0]->value.d[c];
- break;
- default:
- assert(0);
- }
- }
- break;
-
- case ir_unop_abs:
- for (unsigned c = 0; c < op[0]->type->components(); c++) {
- switch (this->type->base_type) {
- case GLSL_TYPE_INT:
- data.i[c] = op[0]->value.i[c];
- if (data.i[c] < 0)
- data.i[c] = -data.i[c];
- break;
- case GLSL_TYPE_FLOAT:
- data.f[c] = fabs(op[0]->value.f[c]);
- break;
- case GLSL_TYPE_DOUBLE:
- data.d[c] = fabs(op[0]->value.d[c]);
- break;
- default:
- assert(0);
- }
- }
- break;
-
- case ir_unop_sign:
- for (unsigned c = 0; c < op[0]->type->components(); c++) {
- switch (this->type->base_type) {
- case GLSL_TYPE_INT:
- data.i[c] = (op[0]->value.i[c] > 0) - (op[0]->value.i[c] < 0);
- break;
- case GLSL_TYPE_FLOAT:
- data.f[c] = float((op[0]->value.f[c] > 0)-(op[0]->value.f[c] < 0));
- break;
- case GLSL_TYPE_DOUBLE:
- data.d[c] = double((op[0]->value.d[c] > 0)-(op[0]->value.d[c] < 0));
- break;
- default:
- assert(0);
- }
- }
- break;
-
- case ir_unop_rcp:
- for (unsigned c = 0; c < op[0]->type->components(); c++) {
- switch (this->type->base_type) {
- case GLSL_TYPE_FLOAT:
- if (op[0]->value.f[c] != 0.0)
- data.f[c] = 1.0F / op[0]->value.f[c];
- break;
- case GLSL_TYPE_DOUBLE:
- if (op[0]->value.d[c] != 0.0)
- data.d[c] = 1.0 / op[0]->value.d[c];
- break;
- default:
- assert(0);
- }
- }
- break;
-
- case ir_unop_rsq:
- for (unsigned c = 0; c < op[0]->type->components(); c++) {
- if (op[0]->type->base_type == GLSL_TYPE_DOUBLE)
- data.d[c] = 1.0 / sqrt(op[0]->value.d[c]);
- else
- data.f[c] = 1.0F / sqrtf(op[0]->value.f[c]);
- }
- break;
-
- case ir_unop_sqrt:
- for (unsigned c = 0; c < op[0]->type->components(); c++) {
- if (op[0]->type->base_type == GLSL_TYPE_DOUBLE)
- data.d[c] = sqrt(op[0]->value.d[c]);
- else
- data.f[c] = sqrtf(op[0]->value.f[c]);
- }
- break;
-
- case ir_unop_exp:
- assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);
- for (unsigned c = 0; c < op[0]->type->components(); c++) {
- data.f[c] = expf(op[0]->value.f[c]);
- }
- break;
-
- case ir_unop_exp2:
- assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);
- for (unsigned c = 0; c < op[0]->type->components(); c++) {
- data.f[c] = exp2f(op[0]->value.f[c]);
- }
- break;
-
- case ir_unop_log:
- assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);
- for (unsigned c = 0; c < op[0]->type->components(); c++) {
- data.f[c] = logf(op[0]->value.f[c]);
- }
- break;
-
- case ir_unop_log2:
- assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);
- for (unsigned c = 0; c < op[0]->type->components(); c++) {
- data.f[c] = log2f(op[0]->value.f[c]);
- }
- break;
-
- case ir_unop_dFdx:
- case ir_unop_dFdx_coarse:
- case ir_unop_dFdx_fine:
- case ir_unop_dFdy:
- case ir_unop_dFdy_coarse:
- case ir_unop_dFdy_fine:
- assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);
- for (unsigned c = 0; c < op[0]->type->components(); c++) {
- data.f[c] = 0.0;
- }
- break;
-
- case ir_unop_pack_snorm_2x16:
- assert(op[0]->type == glsl_type::vec2_type);
- data.u[0] = pack_2x16(pack_snorm_1x16,
- op[0]->value.f[0],
- op[0]->value.f[1]);
- break;
- case ir_unop_pack_snorm_4x8:
- assert(op[0]->type == glsl_type::vec4_type);
- data.u[0] = pack_4x8(pack_snorm_1x8,
- op[0]->value.f[0],
- op[0]->value.f[1],
- op[0]->value.f[2],
- op[0]->value.f[3]);
- break;
- case ir_unop_unpack_snorm_2x16:
- assert(op[0]->type == glsl_type::uint_type);
- unpack_2x16(unpack_snorm_1x16,
- op[0]->value.u[0],
- &data.f[0], &data.f[1]);
- break;
- case ir_unop_unpack_snorm_4x8:
- assert(op[0]->type == glsl_type::uint_type);
- unpack_4x8(unpack_snorm_1x8,
- op[0]->value.u[0],
- &data.f[0], &data.f[1], &data.f[2], &data.f[3]);
- break;
- case ir_unop_pack_unorm_2x16:
- assert(op[0]->type == glsl_type::vec2_type);
- data.u[0] = pack_2x16(pack_unorm_1x16,
- op[0]->value.f[0],
- op[0]->value.f[1]);
- break;
- case ir_unop_pack_unorm_4x8:
- assert(op[0]->type == glsl_type::vec4_type);
- data.u[0] = pack_4x8(pack_unorm_1x8,
- op[0]->value.f[0],
- op[0]->value.f[1],
- op[0]->value.f[2],
- op[0]->value.f[3]);
- break;
- case ir_unop_unpack_unorm_2x16:
- assert(op[0]->type == glsl_type::uint_type);
- unpack_2x16(unpack_unorm_1x16,
- op[0]->value.u[0],
- &data.f[0], &data.f[1]);
- break;
- case ir_unop_unpack_unorm_4x8:
- assert(op[0]->type == glsl_type::uint_type);
- unpack_4x8(unpack_unorm_1x8,
- op[0]->value.u[0],
- &data.f[0], &data.f[1], &data.f[2], &data.f[3]);
- break;
- case ir_unop_pack_half_2x16:
- assert(op[0]->type == glsl_type::vec2_type);
- data.u[0] = pack_2x16(pack_half_1x16,
- op[0]->value.f[0],
- op[0]->value.f[1]);
- break;
- case ir_unop_unpack_half_2x16:
- assert(op[0]->type == glsl_type::uint_type);
- unpack_2x16(unpack_half_1x16,
- op[0]->value.u[0],
- &data.f[0], &data.f[1]);
- break;
- case ir_binop_pow:
- assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);
- for (unsigned c = 0; c < op[0]->type->components(); c++) {
- data.f[c] = powf(op[0]->value.f[c], op[1]->value.f[c]);
- }
- break;
-
- case ir_binop_dot:
- if (op[0]->type->base_type == GLSL_TYPE_DOUBLE)
- data.d[0] = dot_d(op[0], op[1]);
- else
- data.f[0] = dot_f(op[0], op[1]);
- break;
-
- case ir_binop_min:
- assert(op[0]->type == op[1]->type || 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) {
- case GLSL_TYPE_UINT:
- data.u[c] = MIN2(op[0]->value.u[c0], op[1]->value.u[c1]);
- break;
- case GLSL_TYPE_INT:
- data.i[c] = MIN2(op[0]->value.i[c0], op[1]->value.i[c1]);
- break;
- case GLSL_TYPE_FLOAT:
- data.f[c] = MIN2(op[0]->value.f[c0], op[1]->value.f[c1]);
- break;
- case GLSL_TYPE_DOUBLE:
- data.d[c] = MIN2(op[0]->value.d[c0], op[1]->value.d[c1]);
- break;
- default:
- assert(0);
- }
- }
-
- break;
- case ir_binop_max:
- assert(op[0]->type == op[1]->type || 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) {
- case GLSL_TYPE_UINT:
- data.u[c] = MAX2(op[0]->value.u[c0], op[1]->value.u[c1]);
- break;
- case GLSL_TYPE_INT:
- data.i[c] = MAX2(op[0]->value.i[c0], op[1]->value.i[c1]);
- break;
- case GLSL_TYPE_FLOAT:
- data.f[c] = MAX2(op[0]->value.f[c0], op[1]->value.f[c1]);
- break;
- case GLSL_TYPE_DOUBLE:
- data.d[c] = MAX2(op[0]->value.d[c0], op[1]->value.d[c1]);
- break;
- default:
- assert(0);
- }
- }
- break;
-
- case ir_binop_add:
- assert(op[0]->type == op[1]->type || 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) {
- case GLSL_TYPE_UINT:
- data.u[c] = op[0]->value.u[c0] + op[1]->value.u[c1];
- break;
- case GLSL_TYPE_INT:
- data.i[c] = op[0]->value.i[c0] + op[1]->value.i[c1];
- break;
- case GLSL_TYPE_FLOAT:
- data.f[c] = op[0]->value.f[c0] + op[1]->value.f[c1];
- break;
- case GLSL_TYPE_DOUBLE:
- data.d[c] = op[0]->value.d[c0] + op[1]->value.d[c1];
- break;
- default:
- assert(0);
- }
- }
-
- break;
- case ir_binop_sub:
- assert(op[0]->type == op[1]->type || 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) {
- case GLSL_TYPE_UINT:
- data.u[c] = op[0]->value.u[c0] - op[1]->value.u[c1];
- break;
- case GLSL_TYPE_INT:
- data.i[c] = op[0]->value.i[c0] - op[1]->value.i[c1];
- break;
- case GLSL_TYPE_FLOAT:
- data.f[c] = op[0]->value.f[c0] - op[1]->value.f[c1];
- break;
- case GLSL_TYPE_DOUBLE:
- data.d[c] = op[0]->value.d[c0] - op[1]->value.d[c1];
- break;
- default:
- assert(0);
- }
- }
-
- break;
- case ir_binop_mul:
- /* 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) {
- case GLSL_TYPE_UINT:
- data.u[c] = op[0]->value.u[c0] * op[1]->value.u[c1];
- break;
- case GLSL_TYPE_INT:
- data.i[c] = op[0]->value.i[c0] * op[1]->value.i[c1];
- break;
- case GLSL_TYPE_FLOAT:
- data.f[c] = op[0]->value.f[c0] * op[1]->value.f[c1];
- break;
- case GLSL_TYPE_DOUBLE:
- data.d[c] = op[0]->value.d[c0] * op[1]->value.d[c1];
- break;
- default:
- assert(0);
- }
- }
- } 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->base_type == GLSL_TYPE_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;
- case ir_binop_div:
- /* FINISHME: Emit warning when division-by-zero is detected. */
- assert(op[0]->type == op[1]->type || 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) {
- case GLSL_TYPE_UINT:
- if (op[1]->value.u[c1] == 0) {
- data.u[c] = 0;
- } else {
- data.u[c] = op[0]->value.u[c0] / op[1]->value.u[c1];
- }
- break;
- case GLSL_TYPE_INT:
- if (op[1]->value.i[c1] == 0) {
- data.i[c] = 0;
- } else {
- data.i[c] = op[0]->value.i[c0] / op[1]->value.i[c1];
- }
- break;
- case GLSL_TYPE_FLOAT:
- data.f[c] = op[0]->value.f[c0] / op[1]->value.f[c1];
- break;
- case GLSL_TYPE_DOUBLE:
- data.d[c] = op[0]->value.d[c0] / op[1]->value.d[c1];
- break;
- default:
- assert(0);
- }
- }
-
- break;
- case ir_binop_mod:
- /* FINISHME: Emit warning when division-by-zero is detected. */
- assert(op[0]->type == op[1]->type || 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) {
- case GLSL_TYPE_UINT:
- if (op[1]->value.u[c1] == 0) {
- data.u[c] = 0;
- } else {
- data.u[c] = op[0]->value.u[c0] % op[1]->value.u[c1];
- }
- break;
- case GLSL_TYPE_INT:
- if (op[1]->value.i[c1] == 0) {
- data.i[c] = 0;
- } else {
- data.i[c] = op[0]->value.i[c0] % op[1]->value.i[c1];
- }
- break;
- case GLSL_TYPE_FLOAT:
- /* We don't use fmod because it rounds toward zero; GLSL specifies
- * the use of floor.
- */
- data.f[c] = op[0]->value.f[c0] - op[1]->value.f[c1]
- * floorf(op[0]->value.f[c0] / op[1]->value.f[c1]);
- break;
- case GLSL_TYPE_DOUBLE:
- /* We don't use fmod because it rounds toward zero; GLSL specifies
- * the use of floor.
- */
- data.d[c] = op[0]->value.d[c0] - op[1]->value.d[c1]
- * floor(op[0]->value.d[c0] / op[1]->value.d[c1]);
- break;
- default:
- assert(0);
- }
- }
-
- break;
-
- case ir_binop_logic_and:
- assert(op[0]->type->base_type == GLSL_TYPE_BOOL);
- for (unsigned c = 0; c < op[0]->type->components(); c++)
- data.b[c] = op[0]->value.b[c] && op[1]->value.b[c];
- break;
- case ir_binop_logic_xor:
- assert(op[0]->type->base_type == GLSL_TYPE_BOOL);
- for (unsigned c = 0; c < op[0]->type->components(); c++)
- data.b[c] = op[0]->value.b[c] ^ op[1]->value.b[c];
- break;
- case ir_binop_logic_or:
- assert(op[0]->type->base_type == GLSL_TYPE_BOOL);
- for (unsigned c = 0; c < op[0]->type->components(); c++)
- data.b[c] = op[0]->value.b[c] || op[1]->value.b[c];
- break;
-
- case ir_binop_less:
- assert(op[0]->type == op[1]->type);
- for (unsigned c = 0; c < op[0]->type->components(); c++) {
- switch (op[0]->type->base_type) {
- case GLSL_TYPE_UINT:
- data.b[c] = op[0]->value.u[c] < op[1]->value.u[c];
- break;
- case GLSL_TYPE_INT:
- data.b[c] = op[0]->value.i[c] < op[1]->value.i[c];
- break;
- case GLSL_TYPE_FLOAT:
- data.b[c] = op[0]->value.f[c] < op[1]->value.f[c];
- break;
- case GLSL_TYPE_DOUBLE:
- data.b[c] = op[0]->value.d[c] < op[1]->value.d[c];
- break;
- default:
- assert(0);
- }
- }
- break;
- case ir_binop_greater:
- assert(op[0]->type == op[1]->type);
- for (unsigned c = 0; c < op[0]->type->components(); c++) {
- switch (op[0]->type->base_type) {
- case GLSL_TYPE_UINT:
- data.b[c] = op[0]->value.u[c] > op[1]->value.u[c];
- break;
- case GLSL_TYPE_INT:
- data.b[c] = op[0]->value.i[c] > op[1]->value.i[c];
- break;
- case GLSL_TYPE_FLOAT:
- data.b[c] = op[0]->value.f[c] > op[1]->value.f[c];
- break;
- case GLSL_TYPE_DOUBLE:
- data.b[c] = op[0]->value.d[c] > op[1]->value.d[c];
- break;
- default:
- assert(0);
- }
- }
- break;
- case ir_binop_lequal:
- assert(op[0]->type == op[1]->type);
- for (unsigned c = 0; c < op[0]->type->components(); c++) {
- switch (op[0]->type->base_type) {
- case GLSL_TYPE_UINT:
- data.b[c] = op[0]->value.u[c] <= op[1]->value.u[c];
- break;
- case GLSL_TYPE_INT:
- data.b[c] = op[0]->value.i[c] <= op[1]->value.i[c];
- break;
- case GLSL_TYPE_FLOAT:
- data.b[c] = op[0]->value.f[c] <= op[1]->value.f[c];
- break;
- case GLSL_TYPE_DOUBLE:
- data.b[c] = op[0]->value.d[c] <= op[1]->value.d[c];
- break;
- default:
- assert(0);
- }
- }
- break;
- case ir_binop_gequal:
- assert(op[0]->type == op[1]->type);
- for (unsigned c = 0; c < op[0]->type->components(); c++) {
- switch (op[0]->type->base_type) {
- case GLSL_TYPE_UINT:
- data.b[c] = op[0]->value.u[c] >= op[1]->value.u[c];
- break;
- case GLSL_TYPE_INT:
- data.b[c] = op[0]->value.i[c] >= op[1]->value.i[c];
- break;
- case GLSL_TYPE_FLOAT:
- data.b[c] = op[0]->value.f[c] >= op[1]->value.f[c];
- break;
- case GLSL_TYPE_DOUBLE:
- data.b[c] = op[0]->value.d[c] >= op[1]->value.d[c];
- break;
- default:
- assert(0);
- }
- }
- break;
- case ir_binop_equal:
- assert(op[0]->type == op[1]->type);
- for (unsigned c = 0; c < components; c++) {
- switch (op[0]->type->base_type) {
- case GLSL_TYPE_UINT:
- data.b[c] = op[0]->value.u[c] == op[1]->value.u[c];
- break;
- case GLSL_TYPE_INT:
- data.b[c] = op[0]->value.i[c] == op[1]->value.i[c];
- break;
- case GLSL_TYPE_FLOAT:
- data.b[c] = op[0]->value.f[c] == op[1]->value.f[c];
- break;
- case GLSL_TYPE_BOOL:
- data.b[c] = op[0]->value.b[c] == op[1]->value.b[c];
- break;
- case GLSL_TYPE_DOUBLE:
- data.b[c] = op[0]->value.d[c] == op[1]->value.d[c];
- break;
- default:
- assert(0);
- }
- }
- break;
- case ir_binop_nequal:
- assert(op[0]->type == op[1]->type);
- for (unsigned c = 0; c < components; c++) {
- switch (op[0]->type->base_type) {
- case GLSL_TYPE_UINT:
- data.b[c] = op[0]->value.u[c] != op[1]->value.u[c];
- break;
- case GLSL_TYPE_INT:
- data.b[c] = op[0]->value.i[c] != op[1]->value.i[c];
- break;
- case GLSL_TYPE_FLOAT:
- data.b[c] = op[0]->value.f[c] != op[1]->value.f[c];
- break;
- case GLSL_TYPE_BOOL:
- data.b[c] = op[0]->value.b[c] != op[1]->value.b[c];
- break;
- case GLSL_TYPE_DOUBLE:
- data.b[c] = op[0]->value.d[c] != op[1]->value.d[c];
- break;
- default:
- assert(0);
- }
- }
- break;
- case ir_binop_all_equal:
- data.b[0] = op[0]->has_value(op[1]);
- break;
- case ir_binop_any_nequal:
- data.b[0] = !op[0]->has_value(op[1]);
- break;
-
- case ir_binop_lshift:
- for (unsigned c = 0, c0 = 0, c1 = 0;
- c < components;
- c0 += c0_inc, c1 += c1_inc, c++) {
-
- if (op[0]->type->base_type == GLSL_TYPE_INT &&
- op[1]->type->base_type == GLSL_TYPE_INT) {
- data.i[c] = op[0]->value.i[c0] << op[1]->value.i[c1];
-
- } else if (op[0]->type->base_type == GLSL_TYPE_INT &&
- op[1]->type->base_type == GLSL_TYPE_UINT) {
- data.i[c] = op[0]->value.i[c0] << op[1]->value.u[c1];
-
- } else if (op[0]->type->base_type == GLSL_TYPE_UINT &&
- op[1]->type->base_type == GLSL_TYPE_INT) {
- data.u[c] = op[0]->value.u[c0] << op[1]->value.i[c1];
-
- } else if (op[0]->type->base_type == GLSL_TYPE_UINT &&
- op[1]->type->base_type == GLSL_TYPE_UINT) {
- data.u[c] = op[0]->value.u[c0] << op[1]->value.u[c1];
- }
- }
- break;
-
- case ir_binop_rshift:
- for (unsigned c = 0, c0 = 0, c1 = 0;
- c < components;
- c0 += c0_inc, c1 += c1_inc, c++) {
-
- if (op[0]->type->base_type == GLSL_TYPE_INT &&
- op[1]->type->base_type == GLSL_TYPE_INT) {
- data.i[c] = op[0]->value.i[c0] >> op[1]->value.i[c1];
-
- } else if (op[0]->type->base_type == GLSL_TYPE_INT &&
- op[1]->type->base_type == GLSL_TYPE_UINT) {
- data.i[c] = op[0]->value.i[c0] >> op[1]->value.u[c1];
-
- } else if (op[0]->type->base_type == GLSL_TYPE_UINT &&
- op[1]->type->base_type == GLSL_TYPE_INT) {
- data.u[c] = op[0]->value.u[c0] >> op[1]->value.i[c1];
-
- } else if (op[0]->type->base_type == GLSL_TYPE_UINT &&
- op[1]->type->base_type == GLSL_TYPE_UINT) {
- data.u[c] = op[0]->value.u[c0] >> op[1]->value.u[c1];
- }
- }
- break;
-
- case ir_binop_bit_and:
- for (unsigned c = 0, c0 = 0, c1 = 0;
- c < components;
- c0 += c0_inc, c1 += c1_inc, c++) {
-
- switch (op[0]->type->base_type) {
- case GLSL_TYPE_INT:
- data.i[c] = op[0]->value.i[c0] & op[1]->value.i[c1];
- break;
- case GLSL_TYPE_UINT:
- data.u[c] = op[0]->value.u[c0] & op[1]->value.u[c1];
- break;
- default:
- assert(0);
- }
- }
- break;
-
- case ir_binop_bit_or:
- for (unsigned c = 0, c0 = 0, c1 = 0;
- c < components;
- c0 += c0_inc, c1 += c1_inc, c++) {
-
- switch (op[0]->type->base_type) {
- case GLSL_TYPE_INT:
- data.i[c] = op[0]->value.i[c0] | op[1]->value.i[c1];
- break;
- case GLSL_TYPE_UINT:
- data.u[c] = op[0]->value.u[c0] | op[1]->value.u[c1];
- break;
- default:
- assert(0);
- }
- }
- break;
-
- case ir_binop_vector_extract: {
- const int c = CLAMP(op[1]->value.i[0], 0,
- (int) op[0]->type->vector_elements - 1);
-
- switch (op[0]->type->base_type) {
- case GLSL_TYPE_UINT:
- data.u[0] = op[0]->value.u[c];
- break;
- case GLSL_TYPE_INT:
- data.i[0] = op[0]->value.i[c];
- break;
- case GLSL_TYPE_FLOAT:
- data.f[0] = op[0]->value.f[c];
- break;
- case GLSL_TYPE_DOUBLE:
- data.d[0] = op[0]->value.d[c];
- break;
- case GLSL_TYPE_BOOL:
- data.b[0] = op[0]->value.b[c];
- break;
- default:
- assert(0);
- }
- break;
- }
-
- case ir_binop_bit_xor:
- for (unsigned c = 0, c0 = 0, c1 = 0;
- c < components;
- c0 += c0_inc, c1 += c1_inc, c++) {
-
- switch (op[0]->type->base_type) {
- case GLSL_TYPE_INT:
- data.i[c] = op[0]->value.i[c0] ^ op[1]->value.i[c1];
- break;
- case GLSL_TYPE_UINT:
- data.u[c] = op[0]->value.u[c0] ^ op[1]->value.u[c1];
- break;
- default:
- assert(0);
- }
- }
- break;
-
- case ir_unop_bitfield_reverse:
- for (unsigned c = 0; c < op[0]->type->components(); c++) {
- switch (this->type->base_type) {
- case GLSL_TYPE_UINT:
- data.u[c] = bitfield_reverse(op[0]->value.u[c]);
- break;
- case GLSL_TYPE_INT:
- data.i[c] = bitfield_reverse(op[0]->value.i[c]);
- break;
- default:
- assert(0);
- }
- }
- break;
-
- case ir_unop_bit_count:
- for (unsigned c = 0; c < components; c++)
- data.i[c] = _mesa_bitcount(op[0]->value.u[c]);
- break;
-
- case ir_unop_find_msb:
- for (unsigned c = 0; c < components; c++) {
- switch (op[0]->type->base_type) {
- case GLSL_TYPE_UINT:
- data.i[c] = find_msb_uint(op[0]->value.u[c]);
- break;
- case GLSL_TYPE_INT:
- data.i[c] = find_msb_int(op[0]->value.i[c]);
- break;
- default:
- assert(0);
- }
- }
- break;
-
- case ir_unop_find_lsb:
- for (unsigned c = 0; c < components; c++) {
- switch (op[0]->type->base_type) {
- case GLSL_TYPE_UINT:
- data.i[c] = find_msb_uint(op[0]->value.u[c] & -op[0]->value.u[c]);
- break;
- case GLSL_TYPE_INT:
- data.i[c] = find_msb_uint(op[0]->value.i[c] & -op[0]->value.i[c]);
- break;
- default:
- assert(0);
- }
- }
- break;
-
- case ir_unop_saturate:
- for (unsigned c = 0; c < components; c++) {
- data.f[c] = CLAMP(op[0]->value.f[c], 0.0f, 1.0f);
- }
- break;
- case ir_unop_pack_double_2x32:
- /* XXX needs to be checked on big-endian */
- memcpy(&data.d[0], &op[0]->value.u[0], sizeof(double));
- break;
- case ir_unop_unpack_double_2x32:
- /* XXX needs to be checked on big-endian */
- memcpy(&data.u[0], &op[0]->value.d[0], sizeof(double));
- break;
-
- case ir_triop_bitfield_extract: {
- for (unsigned c = 0; c < components; c++) {
- int offset = op[1]->value.i[c];
- int bits = op[2]->value.i[c];
-
- if (bits == 0)
- data.u[c] = 0;
- else if (offset < 0 || bits < 0)
- data.u[c] = 0; /* Undefined, per spec. */
- else if (offset + bits > 32)
- data.u[c] = 0; /* Undefined, per spec. */
- else {
- if (op[0]->type->base_type == GLSL_TYPE_INT) {
- /* int so that the right shift will sign-extend. */
- int value = op[0]->value.i[c];
- value <<= 32 - bits - offset;
- value >>= 32 - bits;
- data.i[c] = value;
- } else {
- unsigned value = op[0]->value.u[c];
- value <<= 32 - bits - offset;
- value >>= 32 - bits;
- data.u[c] = value;
- }
- }
- }
- break;
- }
-
- case ir_binop_ldexp:
- for (unsigned c = 0; c < op[0]->type->components(); c++) {
- switch (this->type->base_type) {
- case GLSL_TYPE_FLOAT:
- data.f[c] = ldexpf_flush_subnormal(op[0]->value.f[c], op[1]->value.i[c]);
- break;
- case GLSL_TYPE_DOUBLE:
- data.d[c] = ldexp_flush_subnormal(op[0]->value.d[c], op[1]->value.i[c]);
- break;
- default:
- assert(0);
- }
- }
- break;
+#include "ir_expression_operation_constant.h"
- case ir_triop_fma:
- assert(op[0]->type->base_type == GLSL_TYPE_FLOAT ||
- op[0]->type->base_type == GLSL_TYPE_DOUBLE);
- assert(op[1]->type->base_type == GLSL_TYPE_FLOAT ||
- op[1]->type->base_type == GLSL_TYPE_DOUBLE);
- assert(op[2]->type->base_type == GLSL_TYPE_FLOAT ||
- op[2]->type->base_type == GLSL_TYPE_DOUBLE);
-
- for (unsigned c = 0; c < components; c++) {
- if (op[0]->type->base_type == GLSL_TYPE_DOUBLE)
- data.d[c] = op[0]->value.d[c] * op[1]->value.d[c]
- + op[2]->value.d[c];
- else
- data.f[c] = op[0]->value.f[c] * op[1]->value.f[c]
- + op[2]->value.f[c];
- }
- break;
+ switch (type->base_type) {
+ case GLSL_TYPE_FLOAT16: {
+ ir_constant_data f;
+ for (unsigned i = 0; i < ARRAY_SIZE(f.f16); i++)
+ f.f16[i] = _mesa_float_to_half(data.f[i]);
- case ir_triop_lrp: {
- assert(op[0]->type->base_type == GLSL_TYPE_FLOAT ||
- op[0]->type->base_type == GLSL_TYPE_DOUBLE);
- assert(op[1]->type->base_type == GLSL_TYPE_FLOAT ||
- op[1]->type->base_type == GLSL_TYPE_DOUBLE);
- assert(op[2]->type->base_type == GLSL_TYPE_FLOAT ||
- op[2]->type->base_type == GLSL_TYPE_DOUBLE);
-
- unsigned c2_inc = op[2]->type->is_scalar() ? 0 : 1;
- for (unsigned c = 0, c2 = 0; c < components; c2 += c2_inc, c++) {
- if (op[0]->type->base_type == GLSL_TYPE_DOUBLE)
- data.d[c] = op[0]->value.d[c] * (1.0 - op[2]->value.d[c2]) +
- (op[1]->value.d[c] * op[2]->value.d[c2]);
- else
- data.f[c] = op[0]->value.f[c] * (1.0f - op[2]->value.f[c2]) +
- (op[1]->value.f[c] * op[2]->value.f[c2]);
- }
- break;
+ return new(mem_ctx) ir_constant(this->type, &f);
}
+ case GLSL_TYPE_INT16: {
+ ir_constant_data d;
+ for (unsigned i = 0; i < ARRAY_SIZE(d.i16); i++)
+ d.i16[i] = data.i[i];
- case ir_triop_csel:
- for (unsigned c = 0; c < components; c++) {
- if (op[1]->type->base_type == GLSL_TYPE_DOUBLE)
- data.d[c] = op[0]->value.b[c] ? op[1]->value.d[c]
- : op[2]->value.d[c];
- else
- data.u[c] = op[0]->value.b[c] ? op[1]->value.u[c]
- : op[2]->value.u[c];
- }
- break;
-
- case ir_triop_vector_insert: {
- const unsigned idx = op[2]->value.u[0];
-
- memcpy(&data, &op[0]->value, sizeof(data));
-
- switch (this->type->base_type) {
- case GLSL_TYPE_INT:
- data.i[idx] = op[1]->value.i[0];
- break;
- case GLSL_TYPE_UINT:
- data.u[idx] = op[1]->value.u[0];
- break;
- case GLSL_TYPE_FLOAT:
- data.f[idx] = op[1]->value.f[0];
- break;
- case GLSL_TYPE_BOOL:
- data.b[idx] = op[1]->value.b[0];
- break;
- case GLSL_TYPE_DOUBLE:
- data.d[idx] = op[1]->value.d[0];
- break;
- default:
- assert(!"Should not get here.");
- break;
- }
- break;
+ return new(mem_ctx) ir_constant(this->type, &d);
}
+ case GLSL_TYPE_UINT16: {
+ ir_constant_data d;
+ for (unsigned i = 0; i < ARRAY_SIZE(d.u16); i++)
+ d.u16[i] = data.u[i];
- case ir_quadop_bitfield_insert: {
- for (unsigned c = 0; c < components; c++) {
- int offset = op[2]->value.i[c];
- int bits = op[3]->value.i[c];
-
- if (bits == 0)
- data.u[c] = op[0]->value.u[c];
- else if (offset < 0 || bits < 0)
- data.u[c] = 0; /* Undefined, per spec. */
- else if (offset + bits > 32)
- data.u[c] = 0; /* Undefined, per spec. */
- else {
- unsigned insert_mask = ((1ull << bits) - 1) << offset;
-
- unsigned insert = op[1]->value.u[c];
- insert <<= offset;
- insert &= insert_mask;
-
- unsigned base = op[0]->value.u[c];
- base &= ~insert_mask;
-
- data.u[c] = base | insert;
- }
- }
- break;
+ return new(mem_ctx) ir_constant(this->type, &d);
}
-
- case ir_quadop_vector:
- for (unsigned c = 0; c < this->type->vector_elements; c++) {
- switch (this->type->base_type) {
- case GLSL_TYPE_INT:
- data.i[c] = op[c]->value.i[0];
- break;
- case GLSL_TYPE_UINT:
- data.u[c] = op[c]->value.u[0];
- break;
- case GLSL_TYPE_FLOAT:
- data.f[c] = op[c]->value.f[0];
- break;
- case GLSL_TYPE_DOUBLE:
- data.d[c] = op[c]->value.d[0];
- break;
- case GLSL_TYPE_BOOL:
- data.b[c] = op[c]->value.b[0];
- break;
- default:
- assert(0);
- }
- }
- break;
-
default:
- /* FINISHME: Should handle all expression types. */
- return NULL;
+ return new(mem_ctx) ir_constant(this->type, &data);
}
-
- return new(ctx) ir_constant(this->type, &data);
}
ir_constant *
-ir_texture::constant_expression_value(struct hash_table *)
+ir_texture::constant_expression_value(void *, struct hash_table *)
{
/* texture lookups aren't constant expressions */
return NULL;
ir_constant *
-ir_swizzle::constant_expression_value(struct hash_table *variable_context)
+ir_swizzle::constant_expression_value(void *mem_ctx,
+ struct hash_table *variable_context)
{
- ir_constant *v = this->val->constant_expression_value(variable_context);
+ assert(mem_ctx);
+
+ ir_constant *v = this->val->constant_expression_value(mem_ctx,
+ variable_context);
if (v != NULL) {
ir_constant_data data = { { 0 } };
for (unsigned i = 0; i < this->mask.num_components; i++) {
switch (v->type->base_type) {
+ case GLSL_TYPE_UINT16:
+ case GLSL_TYPE_INT16: data.u16[i] = v->value.u16[swiz_idx[i]]; break;
case GLSL_TYPE_UINT:
case GLSL_TYPE_INT: data.u[i] = v->value.u[swiz_idx[i]]; break;
case GLSL_TYPE_FLOAT: data.f[i] = v->value.f[swiz_idx[i]]; break;
+ case GLSL_TYPE_FLOAT16: data.f16[i] = v->value.f16[swiz_idx[i]]; break;
case GLSL_TYPE_BOOL: data.b[i] = v->value.b[swiz_idx[i]]; break;
case GLSL_TYPE_DOUBLE:data.d[i] = v->value.d[swiz_idx[i]]; break;
+ case GLSL_TYPE_UINT64:data.u64[i] = v->value.u64[swiz_idx[i]]; break;
+ case GLSL_TYPE_INT64: data.i64[i] = v->value.i64[swiz_idx[i]]; break;
default: assert(!"Should not get here."); break;
}
}
- void *ctx = ralloc_parent(this);
- return new(ctx) ir_constant(this->type, &data);
+ return new(mem_ctx) ir_constant(this->type, &data);
}
return NULL;
}
ir_constant *
-ir_dereference_variable::constant_expression_value(struct hash_table *variable_context)
+ir_dereference_variable::constant_expression_value(void *mem_ctx,
+ struct hash_table *variable_context)
{
assert(var);
+ assert(mem_ctx);
/* Give priority to the context hashtable, if it exists */
if (variable_context) {
- ir_constant *value = (ir_constant *)hash_table_find(variable_context, var);
- if(value)
- return value;
+ hash_entry *entry = _mesa_hash_table_search(variable_context, var);
+
+ if(entry)
+ return (ir_constant *) entry->data;
}
/* The constant_value of a uniform variable is its initializer,
if (!var->constant_value)
return NULL;
- return var->constant_value->clone(ralloc_parent(var), NULL);
+ return var->constant_value->clone(mem_ctx, NULL);
}
ir_constant *
-ir_dereference_array::constant_expression_value(struct hash_table *variable_context)
+ir_dereference_array::constant_expression_value(void *mem_ctx,
+ struct hash_table *variable_context)
{
- ir_constant *array = this->array->constant_expression_value(variable_context);
- ir_constant *idx = this->array_index->constant_expression_value(variable_context);
+ assert(mem_ctx);
+
+ ir_constant *array = this->array->constant_expression_value(mem_ctx, variable_context);
+ ir_constant *idx = this->array_index->constant_expression_value(mem_ctx, variable_context);
if ((array != NULL) && (idx != NULL)) {
- void *ctx = ralloc_parent(this);
if (array->type->is_matrix()) {
/* Array access of a matrix results in a vector.
*/
ir_constant_data data = { { 0 } };
switch (column_type->base_type) {
- case GLSL_TYPE_UINT:
- case GLSL_TYPE_INT:
+ case GLSL_TYPE_FLOAT16:
for (unsigned i = 0; i < column_type->vector_elements; i++)
- data.u[i] = array->value.u[mat_idx + i];
+ data.f16[i] = array->value.f16[mat_idx + i];
break;
break;
default:
- assert(!"Should not get here.");
- break;
+ unreachable("Matrix types are either float or double.");
}
- return new(ctx) ir_constant(column_type, &data);
+ return new(mem_ctx) ir_constant(column_type, &data);
} else if (array->type->is_vector()) {
const unsigned component = idx->value.u[0];
- return new(ctx) ir_constant(array, component);
- } else {
+ return new(mem_ctx) ir_constant(array, component);
+ } else if (array->type->is_array()) {
const unsigned index = idx->value.u[0];
- return array->get_array_element(index)->clone(ctx, NULL);
+ return array->get_array_element(index)->clone(mem_ctx, NULL);
}
}
return NULL;
ir_constant *
-ir_dereference_record::constant_expression_value(struct hash_table *)
+ir_dereference_record::constant_expression_value(void *mem_ctx,
+ struct hash_table *)
{
- ir_constant *v = this->record->constant_expression_value();
+ assert(mem_ctx);
- return (v != NULL) ? v->get_record_field(this->field) : NULL;
+ ir_constant *v = this->record->constant_expression_value(mem_ctx);
+
+ return (v != NULL) ? v->get_record_field(this->field_idx) : NULL;
}
ir_constant *
-ir_assignment::constant_expression_value(struct hash_table *)
+ir_assignment::constant_expression_value(void *, struct hash_table *)
{
/* FINISHME: Handle CEs involving assignment (return RHS) */
return NULL;
ir_constant *
-ir_constant::constant_expression_value(struct hash_table *)
+ir_constant::constant_expression_value(void *, struct hash_table *)
{
return this;
}
ir_constant *
-ir_call::constant_expression_value(struct hash_table *variable_context)
+ir_call::constant_expression_value(void *mem_ctx, struct hash_table *variable_context)
{
- return this->callee->constant_expression_value(&this->actual_parameters, variable_context);
+ assert(mem_ctx);
+
+ return this->callee->constant_expression_value(mem_ctx,
+ &this->actual_parameters,
+ variable_context);
}
-bool ir_function_signature::constant_expression_evaluate_expression_list(const struct exec_list &body,
+bool ir_function_signature::constant_expression_evaluate_expression_list(void *mem_ctx,
+ const struct exec_list &body,
struct hash_table *variable_context,
ir_constant **result)
{
+ assert(mem_ctx);
+
foreach_in_list(ir_instruction, inst, &body) {
switch(inst->ir_type) {
/* (declare () type symbol) */
case ir_type_variable: {
ir_variable *var = inst->as_variable();
- hash_table_insert(variable_context, ir_constant::zero(this, var->type), var);
+ _mesa_hash_table_insert(variable_context, var, ir_constant::zero(this, var->type));
break;
}
case ir_type_assignment: {
ir_assignment *asg = inst->as_assignment();
if (asg->condition) {
- ir_constant *cond = asg->condition->constant_expression_value(variable_context);
+ ir_constant *cond =
+ asg->condition->constant_expression_value(mem_ctx,
+ variable_context);
if (!cond)
return false;
if (!cond->get_bool_component(0))
if (!constant_referenced(asg->lhs, variable_context, store, offset))
return false;
- ir_constant *value = asg->rhs->constant_expression_value(variable_context);
+ ir_constant *value =
+ asg->rhs->constant_expression_value(mem_ctx, variable_context);
if (!value)
return false;
/* (return (expression)) */
case ir_type_return:
assert (result);
- *result = inst->as_return()->value->constant_expression_value(variable_context);
+ *result =
+ inst->as_return()->value->constant_expression_value(mem_ctx,
+ variable_context);
return *result != NULL;
/* (call name (ref) (params))*/
store, offset))
return false;
- ir_constant *value = call->constant_expression_value(variable_context);
+ ir_constant *value =
+ call->constant_expression_value(mem_ctx, variable_context);
if(!value)
return false;
case ir_type_if: {
ir_if *iif = inst->as_if();
- ir_constant *cond = iif->condition->constant_expression_value(variable_context);
+ ir_constant *cond =
+ iif->condition->constant_expression_value(mem_ctx,
+ variable_context);
if (!cond || !cond->type->is_boolean())
return false;
exec_list &branch = cond->get_bool_component(0) ? iif->then_instructions : iif->else_instructions;
*result = NULL;
- if (!constant_expression_evaluate_expression_list(branch, variable_context, result))
+ if (!constant_expression_evaluate_expression_list(mem_ctx, branch,
+ variable_context,
+ result))
return false;
/* If there was a return in the branch chosen, drop out now. */
}
ir_constant *
-ir_function_signature::constant_expression_value(exec_list *actual_parameters, struct hash_table *variable_context)
+ir_function_signature::constant_expression_value(void *mem_ctx,
+ exec_list *actual_parameters,
+ struct hash_table *variable_context)
{
+ assert(mem_ctx);
+
const glsl_type *type = this->return_type;
if (type == glsl_type::void_type)
return NULL;
/*
* Of the builtin functions, only the texture lookups and the noise
- * ones must not be used in constant expressions. They all include
- * specific opcodes so they don't need to be special-cased at this
- * point.
+ * ones must not be used in constant expressions. Texture instructions
+ * include special ir_texture opcodes which can't be constant-folded (see
+ * ir_texture::constant_expression_value). Noise functions, however, we
+ * have to special case here.
*/
+ if (strcmp(this->function_name(), "noise1") == 0 ||
+ strcmp(this->function_name(), "noise2") == 0 ||
+ strcmp(this->function_name(), "noise3") == 0 ||
+ strcmp(this->function_name(), "noise4") == 0)
+ return NULL;
/* Initialize the table of dereferencable names with the function
* parameters. Verify their const-ness on the way.
* We expect the correctness of the number of parameters to have
* been checked earlier.
*/
- hash_table *deref_hash = hash_table_ctor(8, hash_table_pointer_hash,
- hash_table_pointer_compare);
+ hash_table *deref_hash = _mesa_pointer_hash_table_create(NULL);
/* If "origin" is non-NULL, then the function body is there. So we
* have to use the variable objects from the object with the body,
const exec_node *parameter_info = origin ? origin->parameters.get_head_raw() : parameters.get_head_raw();
foreach_in_list(ir_rvalue, n, actual_parameters) {
- ir_constant *constant = n->constant_expression_value(variable_context);
+ ir_constant *constant =
+ n->constant_expression_value(mem_ctx, variable_context);
if (constant == NULL) {
- hash_table_dtor(deref_hash);
+ _mesa_hash_table_destroy(deref_hash, NULL);
return NULL;
}
ir_variable *var = (ir_variable *)parameter_info;
- hash_table_insert(deref_hash, constant, var);
+ _mesa_hash_table_insert(deref_hash, var, constant);
parameter_info = parameter_info->next;
}
/* Now run the builtin function until something non-constant
* happens or we get the result.
*/
- if (constant_expression_evaluate_expression_list(origin ? origin->body : body, deref_hash, &result) && result)
- result = result->clone(ralloc_parent(this), NULL);
+ if (constant_expression_evaluate_expression_list(mem_ctx, origin ? origin->body : body, deref_hash, &result) &&
+ result)
+ result = result->clone(mem_ctx, NULL);
- hash_table_dtor(deref_hash);
+ _mesa_hash_table_destroy(deref_hash, NULL);
return result;
}