#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 "ir_visitor.h"
#include "glsl_types.h"
#include "program/hash_table.h"
-#if defined(_MSC_VER) && (_MSC_VER < 1800)
-static int isnormal(double x)
-{
- return _fpclass(x) == _FPCLASS_NN || _fpclass(x) == _FPCLASS_PN;
-}
-#elif defined(__SUNPRO_CC)
-#include <ieeefp.h>
-static int isnormal(double x)
-{
- return fpclass(x) == FP_NORMAL;
-}
-#endif
-
-#if defined(_MSC_VER)
-static double copysign(double x, double y)
-{
- return _copysign(x, y);
-}
-#endif
-
static float
-dot(ir_constant *op0, ir_constant *op1)
+dot_f(ir_constant *op0, ir_constant *op1)
{
assert(op0->type->is_float() && op1->type->is_float());
return result;
}
+static double
+dot_d(ir_constant *op0, ir_constant *op1)
+{
+ assert(op0->type->is_double() && op1->type->is_double());
+
+ double result = 0;
+ for (unsigned c = 0; c < op0->type->components(); c++)
+ result += op0->value.d[c] * op1->value.d[c];
+
+ return result;
+}
+
/* This method is the only one supported by gcc. Unions in particular
* are iffy, and read-through-converted-pointer is killed by strict
* aliasing. OTOH, the compiler sees through the memcpy, so the
* follows:
*
* packSnorm4x8: round(clamp(c, -1, +1) * 127.0)
- *
- * We must first cast the float to an int, because casting a negative
- * float to a uint is undefined.
*/
- return (uint8_t) (int8_t)
- _mesa_round_to_even(CLAMP(x, -1.0f, +1.0f) * 127.0f);
+ return (uint8_t)
+ _mesa_lroundevenf(CLAMP(x, -1.0f, +1.0f) * 127.0f);
}
/**
* follows:
*
* packSnorm2x16: round(clamp(c, -1, +1) * 32767.0)
- *
- * We must first cast the float to an int, because casting a negative
- * float to a uint is undefined.
*/
- return (uint16_t) (int16_t)
- _mesa_round_to_even(CLAMP(x, -1.0f, +1.0f) * 32767.0f);
+ return (uint16_t)
+ _mesa_lroundevenf(CLAMP(x, -1.0f, +1.0f) * 32767.0f);
}
/**
*
* packUnorm4x8: round(clamp(c, 0, +1) * 255.0)
*/
- return (uint8_t) _mesa_round_to_even(CLAMP(x, 0.0f, 1.0f) * 255.0f);
+ return (uint8_t) (int) _mesa_roundevenf(CLAMP(x, 0.0f, 1.0f) * 255.0f);
}
/**
*
* packUnorm2x16: round(clamp(c, 0, +1) * 65535.0)
*/
- return (uint16_t) _mesa_round_to_even(CLAMP(x, 0.0f, 1.0f) * 65535.0f);
+ return (uint16_t) (int)
+ _mesa_roundevenf(CLAMP(x, 0.0f, 1.0f) * 65535.0f);
}
/**
ir_constant *
-ir_rvalue::constant_expression_value(struct hash_table *variable_context)
+ir_rvalue::constant_expression_value(struct hash_table *)
{
assert(this->type->is_error());
return NULL;
if (this->type->is_error())
return NULL;
- ir_constant *op[Elements(this->operands)] = { NULL, };
+ ir_constant *op[ARRAY_SIZE(this->operands)] = { NULL, };
ir_constant_data data;
memset(&data, 0, sizeof(data));
case ir_binop_lshift:
case ir_binop_rshift:
case ir_binop_ldexp:
+ case ir_binop_interpolate_at_offset:
+ case ir_binop_interpolate_at_sample:
case ir_binop_vector_extract:
case ir_triop_csel:
case ir_triop_bitfield_extract:
data.b[0] = true;
}
break;
-
- case ir_unop_trunc:
+ 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.f[c] = truncf(op[0]->value.f[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:
- assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);
for (unsigned c = 0; c < op[0]->type->components(); c++) {
- data.f[c] = _mesa_round_to_even(op[0]->value.f[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:
- assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);
for (unsigned c = 0; c < op[0]->type->components(); c++) {
- data.f[c] = ceilf(op[0]->value.f[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:
- assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);
for (unsigned c = 0; c < op[0]->type->components(); c++) {
- data.f[c] = floorf(op[0]->value.f[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 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:
- case ir_unop_sin_reduced:
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:
- case ir_unop_cos_reduced:
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]);
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);
}
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);
}
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:
- assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);
for (unsigned c = 0; c < op[0]->type->components(); c++) {
switch (this->type->base_type) {
case GLSL_TYPE_UINT:
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:
- assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);
for (unsigned c = 0; c < op[0]->type->components(); c++) {
- data.f[c] = 1.0F / sqrtf(op[0]->value.f[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:
- assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);
for (unsigned c = 0; c < op[0]->type->components(); c++) {
- data.f[c] = sqrtf(op[0]->value.f[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;
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_binop_dot:
- data.f[0] = dot(op[0], op[1]);
+ 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:
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);
}
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);
}
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);
}
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);
}
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);
}
for (unsigned j = 0; j < p; j++) {
for (unsigned i = 0; i < n; i++) {
for (unsigned k = 0; k < m; k++) {
- data.f[i+n*j] += op[0]->value.f[i+n*k]*op[1]->value.f[k+m*j];
+ 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];
}
}
}
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);
}
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);
}
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);
}
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);
}
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);
}
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);
}
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);
}
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);
}
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;
}
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 */
+ uint64_t temp;
+ temp = (uint64_t)op[0]->value.u[0] | ((uint64_t)op[0]->value.u[1] << 32);
+ data.d[0] = *(double *)&temp;
+
+ break;
+ }
+ case ir_unop_unpack_double_2x32:
+ /* XXX needs to be checked on big-endian */
+ data.u[0] = *(uint32_t *)&op[0]->value.d[0];
+ data.u[1] = *((uint32_t *)&op[0]->value.d[0] + 1);
+ break;
+
case ir_triop_bitfield_extract: {
int offset = op[1]->value.i[0];
int bits = op[2]->value.i[0];
case ir_binop_ldexp:
for (unsigned c = 0; c < components; c++) {
- data.f[c] = ldexp(op[0]->value.f[c], op[1]->value.i[c]);
- /* Flush subnormal values to zero. */
- if (!isnormal(data.f[c]))
- data.f[c] = copysign(0.0f, op[0]->value.f[c]);
+ if (op[0]->type->base_type == GLSL_TYPE_DOUBLE) {
+ data.d[c] = ldexp(op[0]->value.d[c], op[1]->value.i[c]);
+ /* Flush subnormal values to zero. */
+ if (!isnormal(data.d[c]))
+ data.d[c] = copysign(0.0, op[0]->value.d[c]);
+ } else {
+ data.f[c] = ldexpf(op[0]->value.f[c], op[1]->value.i[c]);
+ /* Flush subnormal values to zero. */
+ if (!isnormal(data.f[c]))
+ data.f[c] = copysignf(0.0f, op[0]->value.f[c]);
+ }
}
break;
case ir_triop_fma:
- assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);
- assert(op[1]->type->base_type == GLSL_TYPE_FLOAT);
- assert(op[2]->type->base_type == GLSL_TYPE_FLOAT);
+ 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++) {
- data.f[c] = op[0]->value.f[c] * op[1]->value.f[c]
- + op[2]->value.f[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;
case ir_triop_lrp: {
- assert(op[0]->type->base_type == GLSL_TYPE_FLOAT);
- assert(op[1]->type->base_type == GLSL_TYPE_FLOAT);
- assert(op[2]->type->base_type == GLSL_TYPE_FLOAT);
+ 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++) {
- 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]);
+ 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;
}
case ir_triop_csel:
for (unsigned c = 0; c < components; c++) {
- data.u[c] = op[0]->value.b[c] ? op[1]->value.u[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 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;
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;
default:
assert(0);
}
ir_constant *
-ir_texture::constant_expression_value(struct hash_table *variable_context)
+ir_texture::constant_expression_value(struct hash_table *)
{
/* texture lookups aren't constant expressions */
return NULL;
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_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;
default: assert(!"Should not get here."); break;
}
}
break;
+ case GLSL_TYPE_DOUBLE:
+ for (unsigned i = 0; i < column_type->vector_elements; i++)
+ data.d[i] = array->value.d[mat_idx + i];
+
+ break;
+
default:
assert(!"Should not get here.");
break;
ir_constant *
-ir_dereference_record::constant_expression_value(struct hash_table *variable_context)
+ir_dereference_record::constant_expression_value(struct hash_table *)
{
ir_constant *v = this->record->constant_expression_value();
ir_constant *
-ir_assignment::constant_expression_value(struct hash_table *variable_context)
+ir_assignment::constant_expression_value(struct hash_table *)
{
/* FINISHME: Handle CEs involving assignment (return RHS) */
return NULL;
ir_constant *
-ir_constant::constant_expression_value(struct hash_table *variable_context)
+ir_constant::constant_expression_value(struct hash_table *)
{
return this;
}
struct hash_table *variable_context,
ir_constant **result)
{
- foreach_list(n, &body) {
- ir_instruction *inst = (ir_instruction *)n;
+ foreach_in_list(ir_instruction, inst, &body) {
switch(inst->ir_type) {
/* (declare () type symbol) */
*/
const exec_node *parameter_info = origin ? origin->parameters.head : parameters.head;
- foreach_list(n, actual_parameters) {
- ir_constant *constant = ((ir_rvalue *) n)->constant_expression_value(variable_context);
+ foreach_in_list(ir_rvalue, n, actual_parameters) {
+ ir_constant *constant = n->constant_expression_value(variable_context);
if (constant == NULL) {
hash_table_dtor(deref_hash);
return NULL;
projects / mesa.git / blobdiff