}
static inline bool
-is_vec_negative_one(ir_constant *ir)
+is_vec_four(ir_constant *ir)
{
- return (ir == NULL) ? false : ir->is_negative_one();
+ return (ir == NULL) ? false : ir->is_value(4.0, 4);
}
static inline bool
-is_vec_basis(ir_constant *ir)
+is_vec_negative_one(ir_constant *ir)
{
- return (ir == NULL) ? false : ir->is_basis();
+ return (ir == NULL) ? false : ir->is_negative_one();
}
static inline bool
static inline bool
is_less_than_one(ir_constant *ir)
{
+ assert(ir->type->base_type == GLSL_TYPE_FLOAT);
+
if (!is_valid_vec_const(ir))
return false;
static inline bool
is_greater_than_zero(ir_constant *ir)
{
+ assert(ir->type->base_type == GLSL_TYPE_FLOAT);
+
if (!is_valid_vec_const(ir))
return false;
ir_expression *op_expr[4] = {NULL, NULL, NULL, NULL};
unsigned int i;
+ if (ir->operation == ir_binop_mul &&
+ ir->operands[0]->type->is_matrix() &&
+ ir->operands[1]->type->is_vector()) {
+ ir_expression *matrix_mul = ir->operands[0]->as_expression();
+
+ if (matrix_mul && matrix_mul->operation == ir_binop_mul &&
+ matrix_mul->operands[0]->type->is_matrix() &&
+ matrix_mul->operands[1]->type->is_matrix()) {
+
+ return mul(matrix_mul->operands[0],
+ mul(matrix_mul->operands[1], ir->operands[1]));
+ }
+ }
+
assert(ir->get_num_operands() <= 4);
for (i = 0; i < ir->get_num_operands(); i++) {
if (ir->operands[i]->type->is_matrix())
if (op_expr[0]->operation == ir_unop_log2) {
return op_expr[0]->operands[0];
}
+
+ if (!options->EmitNoPow && op_expr[0]->operation == ir_binop_mul) {
+ for (int log2_pos = 0; log2_pos < 2; log2_pos++) {
+ ir_expression *log2_expr =
+ op_expr[0]->operands[log2_pos]->as_expression();
+
+ if (log2_expr && log2_expr->operation == ir_unop_log2) {
+ return new(mem_ctx) ir_expression(ir_binop_pow,
+ ir->type,
+ log2_expr->operands[0],
+ op_expr[0]->operands[1 - log2_pos]);
+ }
+ }
+ }
break;
case ir_unop_log2:
}
break;
+ case ir_unop_f2i:
+ case ir_unop_f2u:
+ if (op_expr[0] && op_expr[0]->operation == ir_unop_trunc) {
+ return new(mem_ctx) ir_expression(ir->operation,
+ ir->type,
+ op_expr[0]->operands[0]);
+ }
+ break;
+
case ir_unop_logic_not: {
enum ir_expression_operation new_op = ir_unop_logic_not;
break;
}
+ case ir_unop_saturate:
+ if (op_expr[0] && op_expr[0]->operation == ir_binop_add) {
+ ir_expression *b2f_0 = op_expr[0]->operands[0]->as_expression();
+ ir_expression *b2f_1 = op_expr[0]->operands[1]->as_expression();
+
+ if (b2f_0 && b2f_0->operation == ir_unop_b2f &&
+ b2f_1 && b2f_1->operation == ir_unop_b2f) {
+ return b2f(logic_or(b2f_0->operands[0], b2f_1->operands[0]));
+ }
+ }
+ break;
+
case ir_binop_add:
if (is_vec_zero(op_const[0]))
return ir->operands[1];
if (is_vec_negative_one(op_const[1]))
return neg(ir->operands[0]);
+ if (op_expr[0] && op_expr[0]->operation == ir_unop_b2f &&
+ op_expr[1] && op_expr[1]->operation == ir_unop_b2f) {
+ return b2f(logic_and(op_expr[0]->operands[0], op_expr[1]->operands[0]));
+ }
/* Reassociate multiplication of constants so that we can do
* constant folding.
if (op_const[1] && !op_const[0])
reassociate_constant(ir, 1, op_const[1], op_expr[0]);
+ /* Optimizes
+ *
+ * (mul (floor (add (abs x) 0.5) (sign x)))
+ *
+ * into
+ *
+ * (trunc (add x (mul (sign x) 0.5)))
+ */
+ for (int i = 0; i < 2; i++) {
+ ir_expression *sign_expr = ir->operands[i]->as_expression();
+ ir_expression *floor_expr = ir->operands[1 - i]->as_expression();
+
+ if (!sign_expr || sign_expr->operation != ir_unop_sign ||
+ !floor_expr || floor_expr->operation != ir_unop_floor)
+ continue;
+
+ ir_expression *add_expr = floor_expr->operands[0]->as_expression();
+ if (!add_expr)
+ continue;
+
+ for (int j = 0; j < 2; j++) {
+ ir_expression *abs_expr = add_expr->operands[j]->as_expression();
+ if (!abs_expr || abs_expr->operation != ir_unop_abs)
+ continue;
+
+ ir_constant *point_five = add_expr->operands[1 - j]->as_constant();
+ if (!point_five->is_value(0.5, 0))
+ continue;
+
+ if (abs_expr->operands[0]->equals(sign_expr->operands[0])) {
+ return trunc(add(abs_expr->operands[0],
+ mul(sign_expr, point_five)));
+ }
+ }
+ }
break;
case ir_binop_div:
- if (is_vec_one(op_const[0]) && ir->type->base_type == GLSL_TYPE_FLOAT) {
+ if (is_vec_one(op_const[0]) && (
+ ir->type->base_type == GLSL_TYPE_FLOAT ||
+ ir->type->base_type == GLSL_TYPE_DOUBLE)) {
return new(mem_ctx) ir_expression(ir_unop_rcp,
ir->operands[1]->type,
ir->operands[1],
if (is_vec_zero(op_const[0]) || is_vec_zero(op_const[1]))
return ir_constant::zero(mem_ctx, ir->type);
- if (is_vec_basis(op_const[0])) {
- unsigned component = 0;
- for (unsigned c = 0; c < op_const[0]->type->vector_elements; c++) {
- if (op_const[0]->value.f[c] == 1.0)
- component = c;
- }
- return new(mem_ctx) ir_swizzle(ir->operands[1], component, 0, 0, 0, 1);
- }
- if (is_vec_basis(op_const[1])) {
- unsigned component = 0;
- for (unsigned c = 0; c < op_const[1]->type->vector_elements; c++) {
- if (op_const[1]->value.f[c] == 1.0)
- component = c;
- }
- return new(mem_ctx) ir_swizzle(ir->operands[0], component, 0, 0, 0, 1);
+ for (int i = 0; i < 2; i++) {
+ if (!op_const[i])
+ continue;
+
+ unsigned components[4] = { 0 }, count = 0;
+
+ for (unsigned c = 0; c < op_const[i]->type->vector_elements; c++) {
+ if (op_const[i]->is_zero())
+ continue;
+
+ components[count] = c;
+ count++;
+ }
+
+ /* No channels had zero values; bail. */
+ if (count >= op_const[i]->type->vector_elements)
+ break;
+
+ ir_expression_operation op = count == 1 ?
+ ir_binop_mul : ir_binop_dot;
+
+ /* Swizzle both operands to remove the channels that were zero. */
+ return new(mem_ctx)
+ ir_expression(op, ir->type,
+ new(mem_ctx) ir_swizzle(ir->operands[0],
+ components, count),
+ new(mem_ctx) ir_swizzle(ir->operands[1],
+ components, count));
}
break;
if (!is_vec_zero(zero))
continue;
- return new(mem_ctx) ir_expression(ir->operation,
- add->operands[0],
- neg(add->operands[1]));
+ /* Depending of the zero position we want to optimize
+ * (0 cmp x+y) into (-x cmp y) or (x+y cmp 0) into (x cmp -y)
+ */
+ if (add_pos == 1) {
+ return new(mem_ctx) ir_expression(ir->operation,
+ neg(add->operands[0]),
+ add->operands[1]);
+ } else {
+ return new(mem_ctx) ir_expression(ir->operation,
+ add->operands[0],
+ neg(add->operands[1]));
+ }
}
break;
+ case ir_binop_all_equal:
+ case ir_binop_any_nequal:
+ if (ir->operands[0]->type->is_scalar() &&
+ ir->operands[1]->type->is_scalar())
+ return new(mem_ctx) ir_expression(ir->operation == ir_binop_all_equal
+ ? ir_binop_equal : ir_binop_nequal,
+ ir->operands[0],
+ ir->operands[1]);
+ break;
+
case ir_binop_rshift:
case ir_binop_lshift:
/* 0 >> x == 0 */
return mul(x, x);
}
+ if (is_vec_four(op_const[1])) {
+ ir_variable *x = new(ir) ir_variable(ir->operands[1]->type, "x",
+ ir_var_temporary);
+ base_ir->insert_before(x);
+ base_ir->insert_before(assign(x, ir->operands[0]));
+
+ ir_variable *squared = new(ir) ir_variable(ir->operands[1]->type,
+ "squared",
+ ir_var_temporary);
+ base_ir->insert_before(squared);
+ base_ir->insert_before(assign(squared, mul(x, x)));
+ return mul(squared, squared);
+ }
+
break;
case ir_binop_min:
case ir_binop_max:
- if (ir->type->base_type != GLSL_TYPE_FLOAT)
+ if (ir->type->base_type != GLSL_TYPE_FLOAT || options->EmitNoSat)
break;
/* Replace min(max) operations and its commutative combinations with
* a saturate operation
*/
for (int op = 0; op < 2; op++) {
- ir_expression *minmax = op_expr[op];
+ ir_expression *inner_expr = op_expr[op];
ir_constant *outer_const = op_const[1 - op];
ir_expression_operation op_cond = (ir->operation == ir_binop_max) ?
ir_binop_min : ir_binop_max;
- if (!minmax || !outer_const || (minmax->operation != op_cond))
+ if (!inner_expr || !outer_const || (inner_expr->operation != op_cond))
continue;
+ /* One of these has to be a constant */
+ if (!inner_expr->operands[0]->as_constant() &&
+ !inner_expr->operands[1]->as_constant())
+ break;
+
/* Found a min(max) combination. Now try to see if its operands
* meet our conditions that we can do just a single saturate operation
*/
for (int minmax_op = 0; minmax_op < 2; minmax_op++) {
- ir_rvalue *inner_val_a = minmax->operands[minmax_op];
- ir_rvalue *inner_val_b = minmax->operands[1 - minmax_op];
-
- if (!inner_val_a || !inner_val_b)
- continue;
+ ir_rvalue *x = inner_expr->operands[minmax_op];
+ ir_rvalue *y = inner_expr->operands[1 - minmax_op];
- /* Found a {min|max} ({max|min} (x, 0.0), 1.0) operation and its variations */
- if ((outer_const->is_one() && inner_val_a->is_zero()) ||
- (inner_val_a->is_one() && outer_const->is_zero()))
- return saturate(inner_val_b);
-
- /* Found a {min|max} ({max|min} (x, 0.0), b) where b < 1.0
- * and its variations
- */
- if (is_less_than_one(outer_const) && inner_val_b->is_zero())
- return expr(ir_binop_min, saturate(inner_val_a), outer_const);
-
- if (!inner_val_b->as_constant())
+ ir_constant *inner_const = y->as_constant();
+ if (!inner_const)
continue;
- if (is_less_than_one(inner_val_b->as_constant()) && outer_const->is_zero())
- return expr(ir_binop_min, saturate(inner_val_a), inner_val_b);
-
- /* Found a {min|max} ({max|min} (x, b), 1.0), where b > 0.0
- * and its variations
- */
- if (outer_const->is_one() && is_greater_than_zero(inner_val_b->as_constant()))
- return expr(ir_binop_max, saturate(inner_val_a), inner_val_b);
- if (inner_val_b->as_constant()->is_one() && is_greater_than_zero(outer_const))
- return expr(ir_binop_max, saturate(inner_val_a), outer_const);
+ /* min(max(x, 0.0), 1.0) is sat(x) */
+ if (ir->operation == ir_binop_min &&
+ inner_const->is_zero() &&
+ outer_const->is_one())
+ return saturate(x);
+
+ /* max(min(x, 1.0), 0.0) is sat(x) */
+ if (ir->operation == ir_binop_max &&
+ inner_const->is_one() &&
+ outer_const->is_zero())
+ return saturate(x);
+
+ /* min(max(x, 0.0), b) where b < 1.0 is sat(min(x, b)) */
+ if (ir->operation == ir_binop_min &&
+ inner_const->is_zero() &&
+ is_less_than_one(outer_const))
+ return saturate(expr(ir_binop_min, x, outer_const));
+
+ /* max(min(x, b), 0.0) where b < 1.0 is sat(min(x, b)) */
+ if (ir->operation == ir_binop_max &&
+ is_less_than_one(inner_const) &&
+ outer_const->is_zero())
+ return saturate(expr(ir_binop_min, x, inner_const));
+
+ /* max(min(x, 1.0), b) where b > 0.0 is sat(max(x, b)) */
+ if (ir->operation == ir_binop_max &&
+ inner_const->is_one() &&
+ is_greater_than_zero(outer_const))
+ return saturate(expr(ir_binop_max, x, outer_const));
+
+ /* min(max(x, b), 1.0) where b > 0.0 is sat(max(x, b)) */
+ if (ir->operation == ir_binop_min &&
+ is_greater_than_zero(inner_const) &&
+ outer_const->is_one())
+ return saturate(expr(ir_binop_max, x, inner_const));
}
}
if (op_expr[0] && op_expr[0]->operation == ir_unop_rcp)
return op_expr[0]->operands[0];
+ if (op_expr[0] && (op_expr[0]->operation == ir_unop_exp2 ||
+ op_expr[0]->operation == ir_unop_exp)) {
+ return new(mem_ctx) ir_expression(op_expr[0]->operation, ir->type,
+ neg(op_expr[0]->operands[0]));
+ }
+
/* While ir_to_mesa.cpp will lower sqrt(x) to rcp(rsq(x)), it does so at
* its IR level, so we can always apply this transformation.
*/
return mul(ir->operands[1], ir->operands[2]);
} else if (is_vec_zero(op_const[1])) {
unsigned op2_components = ir->operands[2]->type->vector_elements;
- ir_constant *one = new(mem_ctx) ir_constant(1.0f, op2_components);
+ ir_constant *one;
+
+ switch (ir->type->base_type) {
+ case GLSL_TYPE_FLOAT:
+ one = new(mem_ctx) ir_constant(1.0f, op2_components);
+ break;
+ case GLSL_TYPE_DOUBLE:
+ one = new(mem_ctx) ir_constant(1.0, op2_components);
+ break;
+ default:
+ one = NULL;
+ unreachable("unexpected type");
+ }
+
return mul(ir->operands[0], add(one, neg(ir->operands[2])));
}
break;