#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"
-/* Using C99 rounding functions for roundToEven() implementation is
- * difficult, because round(), rint, and nearbyint() are affected by
- * fesetenv(), which the application may have done for its own
- * purposes. Mesa's IROUND macro is close to what we want, but it
- * rounds away from 0 on n + 0.5.
- */
-static int
-round_to_even(float val)
+static float
+dot_f(ir_constant *op0, ir_constant *op1)
{
- int rounded = IROUND(val);
+ assert(op0->type->is_float() && op1->type->is_float());
- if (val - floor(val) == 0.5) {
- if (rounded % 2 != 0)
- rounded += val > 0 ? -1 : 1;
- }
+ float result = 0;
+ for (unsigned c = 0; c < op0->type->components(); c++)
+ result += op0->value.f[c] * op1->value.f[c];
- return rounded;
+ return result;
}
-static float
-dot(ir_constant *op0, ir_constant *op1)
+static double
+dot_d(ir_constant *op0, ir_constant *op1)
{
- assert(op0->type->is_float() && op1->type->is_float());
+ assert(op0->type->is_double() && op1->type->is_double());
- float result = 0;
+ double result = 0;
for (unsigned c = 0; c < op0->type->components(); c++)
- result += op0->value.f[c] * op1->value.f[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
+ * resulting asm is reasonable.
+ */
+static float
+bitcast_u2f(unsigned int u)
+{
+ assert(sizeof(float) == sizeof(unsigned int));
+ float f;
+ memcpy(&f, &u, sizeof(f));
+ return f;
+}
+
+static unsigned int
+bitcast_f2u(float f)
+{
+ assert(sizeof(float) == sizeof(unsigned int));
+ unsigned int u;
+ memcpy(&u, &f, sizeof(f));
+ return u;
+}
+
+/**
+ * Evaluate one component of a floating-point 4x8 unpacking function.
+ */
+typedef uint8_t
+(*pack_1x8_func_t)(float);
+
+/**
+ * Evaluate one component of a floating-point 2x16 unpacking function.
+ */
+typedef uint16_t
+(*pack_1x16_func_t)(float);
+
+/**
+ * Evaluate one component of a floating-point 4x8 unpacking function.
+ */
+typedef float
+(*unpack_1x8_func_t)(uint8_t);
+
+/**
+ * Evaluate one component of a floating-point 2x16 unpacking function.
+ */
+typedef float
+(*unpack_1x16_func_t)(uint16_t);
+
+/**
+ * Evaluate a 2x16 floating-point packing function.
+ */
+static uint32_t
+pack_2x16(pack_1x16_func_t pack_1x16,
+ float x, float y)
+{
+ /* From section 8.4 of the GLSL ES 3.00 spec:
+ *
+ * packSnorm2x16
+ * -------------
+ * The first component of the vector will be written to the least
+ * significant bits of the output; the last component will be written to
+ * the most significant bits.
+ *
+ * The specifications for the other packing functions contain similar
+ * language.
+ */
+ uint32_t u = 0;
+ u |= ((uint32_t) pack_1x16(x) << 0);
+ u |= ((uint32_t) pack_1x16(y) << 16);
+ return u;
+}
+
+/**
+ * Evaluate a 4x8 floating-point packing function.
+ */
+static uint32_t
+pack_4x8(pack_1x8_func_t pack_1x8,
+ float x, float y, float z, float w)
+{
+ /* From section 8.4 of the GLSL 4.30 spec:
+ *
+ * packSnorm4x8
+ * ------------
+ * The first component of the vector will be written to the least
+ * significant bits of the output; the last component will be written to
+ * the most significant bits.
+ *
+ * The specifications for the other packing functions contain similar
+ * language.
+ */
+ uint32_t u = 0;
+ u |= ((uint32_t) pack_1x8(x) << 0);
+ u |= ((uint32_t) pack_1x8(y) << 8);
+ u |= ((uint32_t) pack_1x8(z) << 16);
+ u |= ((uint32_t) pack_1x8(w) << 24);
+ return u;
+}
+
+/**
+ * Evaluate a 2x16 floating-point unpacking function.
+ */
+static void
+unpack_2x16(unpack_1x16_func_t unpack_1x16,
+ uint32_t u,
+ float *x, float *y)
+{
+ /* From section 8.4 of the GLSL ES 3.00 spec:
+ *
+ * unpackSnorm2x16
+ * ---------------
+ * The first component of the returned vector will be extracted from
+ * the least significant bits of the input; the last component will be
+ * extracted from the most significant bits.
+ *
+ * The specifications for the other unpacking functions contain similar
+ * language.
+ */
+ *x = unpack_1x16((uint16_t) (u & 0xffff));
+ *y = unpack_1x16((uint16_t) (u >> 16));
+}
+
+/**
+ * Evaluate a 4x8 floating-point unpacking function.
+ */
+static void
+unpack_4x8(unpack_1x8_func_t unpack_1x8, uint32_t u,
+ float *x, float *y, float *z, float *w)
+{
+ /* From section 8.4 of the GLSL 4.30 spec:
+ *
+ * unpackSnorm4x8
+ * --------------
+ * The first component of the returned vector will be extracted from
+ * the least significant bits of the input; the last component will be
+ * extracted from the most significant bits.
+ *
+ * The specifications for the other unpacking functions contain similar
+ * language.
+ */
+ *x = unpack_1x8((uint8_t) (u & 0xff));
+ *y = unpack_1x8((uint8_t) (u >> 8));
+ *z = unpack_1x8((uint8_t) (u >> 16));
+ *w = unpack_1x8((uint8_t) (u >> 24));
+}
+
+/**
+ * Evaluate one component of packSnorm4x8.
+ */
+static uint8_t
+pack_snorm_1x8(float x)
+{
+ /* From section 8.4 of the GLSL 4.30 spec:
+ *
+ * packSnorm4x8
+ * ------------
+ * The conversion for component c of v to fixed point is done as
+ * follows:
+ *
+ * packSnorm4x8: round(clamp(c, -1, +1) * 127.0)
+ */
+ return (uint8_t)
+ _mesa_lroundevenf(CLAMP(x, -1.0f, +1.0f) * 127.0f);
+}
+
+/**
+ * Evaluate one component of packSnorm2x16.
+ */
+static uint16_t
+pack_snorm_1x16(float x)
+{
+ /* From section 8.4 of the GLSL ES 3.00 spec:
+ *
+ * packSnorm2x16
+ * -------------
+ * The conversion for component c of v to fixed point is done as
+ * follows:
+ *
+ * packSnorm2x16: round(clamp(c, -1, +1) * 32767.0)
+ */
+ return (uint16_t)
+ _mesa_lroundevenf(CLAMP(x, -1.0f, +1.0f) * 32767.0f);
+}
+
+/**
+ * Evaluate one component of unpackSnorm4x8.
+ */
+static float
+unpack_snorm_1x8(uint8_t u)
+{
+ /* From section 8.4 of the GLSL 4.30 spec:
+ *
+ * unpackSnorm4x8
+ * --------------
+ * The conversion for unpacked fixed-point value f to floating point is
+ * done as follows:
+ *
+ * unpackSnorm4x8: clamp(f / 127.0, -1, +1)
+ */
+ return CLAMP((int8_t) u / 127.0f, -1.0f, +1.0f);
+}
+
+/**
+ * Evaluate one component of unpackSnorm2x16.
+ */
+static float
+unpack_snorm_1x16(uint16_t u)
+{
+ /* From section 8.4 of the GLSL ES 3.00 spec:
+ *
+ * unpackSnorm2x16
+ * ---------------
+ * The conversion for unpacked fixed-point value f to floating point is
+ * done as follows:
+ *
+ * unpackSnorm2x16: clamp(f / 32767.0, -1, +1)
+ */
+ return CLAMP((int16_t) u / 32767.0f, -1.0f, +1.0f);
+}
+
+/**
+ * Evaluate one component packUnorm4x8.
+ */
+static uint8_t
+pack_unorm_1x8(float x)
+{
+ /* From section 8.4 of the GLSL 4.30 spec:
+ *
+ * packUnorm4x8
+ * ------------
+ * The conversion for component c of v to fixed point is done as
+ * follows:
+ *
+ * packUnorm4x8: round(clamp(c, 0, +1) * 255.0)
+ */
+ return (uint8_t) (int) _mesa_roundevenf(CLAMP(x, 0.0f, 1.0f) * 255.0f);
+}
+
+/**
+ * Evaluate one component packUnorm2x16.
+ */
+static uint16_t
+pack_unorm_1x16(float x)
+{
+ /* From section 8.4 of the GLSL ES 3.00 spec:
+ *
+ * packUnorm2x16
+ * -------------
+ * The conversion for component c of v to fixed point is done as
+ * follows:
+ *
+ * packUnorm2x16: round(clamp(c, 0, +1) * 65535.0)
+ */
+ return (uint16_t) (int)
+ _mesa_roundevenf(CLAMP(x, 0.0f, 1.0f) * 65535.0f);
+}
+
+/**
+ * Evaluate one component of unpackUnorm4x8.
+ */
+static float
+unpack_unorm_1x8(uint8_t u)
+{
+ /* From section 8.4 of the GLSL 4.30 spec:
+ *
+ * unpackUnorm4x8
+ * --------------
+ * The conversion for unpacked fixed-point value f to floating point is
+ * done as follows:
+ *
+ * unpackUnorm4x8: f / 255.0
+ */
+ return (float) u / 255.0f;
+}
+
+/**
+ * Evaluate one component of unpackUnorm2x16.
+ */
+static float
+unpack_unorm_1x16(uint16_t u)
+{
+ /* From section 8.4 of the GLSL ES 3.00 spec:
+ *
+ * unpackUnorm2x16
+ * ---------------
+ * The conversion for unpacked fixed-point value f to floating point is
+ * done as follows:
+ *
+ * unpackUnorm2x16: f / 65535.0
+ */
+ return (float) u / 65535.0f;
+}
+
+/**
+ * Evaluate one component of packHalf2x16.
+ */
+static uint16_t
+pack_half_1x16(float x)
+{
+ return _mesa_float_to_half(x);
+}
+
+/**
+ * Evaluate one component of unpackHalf2x16.
+ */
+static float
+unpack_half_1x16(uint16_t u)
+{
+ return _mesa_half_to_float(u);
+}
+
+/**
+ * Get the constant that is ultimately referenced by an r-value, in a constant
+ * expression evaluation context.
+ *
+ * The offset is used when the reference is to a specific column of a matrix.
+ */
+static bool
+constant_referenced(const ir_dereference *deref,
+ struct hash_table *variable_context,
+ ir_constant *&store, int &offset)
+{
+ store = NULL;
+ offset = 0;
+
+ if (variable_context == NULL)
+ return false;
+
+ switch (deref->ir_type) {
+ case ir_type_dereference_array: {
+ const ir_dereference_array *const da =
+ (const ir_dereference_array *) deref;
+
+ 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())
+ break;
+
+ const int index = index_c->type->base_type == GLSL_TYPE_INT ?
+ index_c->get_int_component(0) :
+ index_c->get_uint_component(0);
+
+ ir_constant *substore;
+ int suboffset;
+
+ const ir_dereference *const deref = da->array->as_dereference();
+ if (!deref)
+ break;
+
+ if (!constant_referenced(deref, variable_context, substore, suboffset))
+ break;
+
+ const glsl_type *const vt = da->array->type;
+ if (vt->is_array()) {
+ store = substore->get_array_element(index);
+ offset = 0;
+ } else if (vt->is_matrix()) {
+ store = substore;
+ offset = index * vt->vector_elements;
+ } else if (vt->is_vector()) {
+ store = substore;
+ offset = suboffset + index;
+ }
+
+ break;
+ }
+
+ case ir_type_dereference_record: {
+ const ir_dereference_record *const dr =
+ (const ir_dereference_record *) deref;
+
+ const ir_dereference *const deref = dr->record->as_dereference();
+ if (!deref)
+ break;
+
+ ir_constant *substore;
+ int suboffset;
+
+ if (!constant_referenced(deref, variable_context, substore, suboffset))
+ break;
+
+ /* Since we're dropping it on the floor...
+ */
+ assert(suboffset == 0);
+
+ store = substore->get_record_field(dr->field);
+ break;
+ }
+
+ case ir_type_dereference_variable: {
+ const ir_dereference_variable *const dv =
+ (const ir_dereference_variable *) deref;
+
+ store = (ir_constant *) hash_table_find(variable_context, dv->var);
+ break;
+ }
+
+ default:
+ assert(!"Should not get here.");
+ break;
+ }
+
+ return store != NULL;
+}
+
+
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));
}
if (op[1] != NULL)
- assert(op[0]->type->base_type == op[1]->type->base_type ||
- this->operation == ir_binop_lshift ||
- this->operation == ir_binop_rshift);
+ switch (this->operation) {
+ 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:
+ break;
+
+ default:
+ assert(op[0]->type->base_type == op[1]->type->base_type);
+ break;
+ }
bool op0_scalar = op[0]->type->is_scalar();
bool op1_scalar = op[1] != NULL && op[1]->type->is_scalar();
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.i[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.u[c] = op[0]->value.i[c];
}
break;
- case ir_unop_any:
- assert(op[0]->type->is_boolean());
- data.b[0] = false;
+ case ir_unop_bitcast_i2f:
+ assert(op[0]->type->base_type == GLSL_TYPE_INT);
for (unsigned c = 0; c < op[0]->type->components(); c++) {
- if (op[0]->value.b[c])
- data.b[0] = true;
+ data.f[c] = bitcast_u2f(op[0]->value.i[c]);
}
break;
-
- case ir_unop_trunc:
+ 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.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] = 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_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++) {
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);
}
}
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;
}
break;
+ case ir_unop_bitfield_reverse:
+ /* http://graphics.stanford.edu/~seander/bithacks.html#BitReverseObvious */
+ for (unsigned c = 0; c < components; c++) {
+ unsigned int v = op[0]->value.u[c]; // input bits to be reversed
+ unsigned int r = v; // r will be reversed bits of v; first get LSB of v
+ int s = sizeof(v) * CHAR_BIT - 1; // extra shift needed at end
+
+ for (v >>= 1; v; v >>= 1) {
+ r <<= 1;
+ r |= v & 1;
+ s--;
+ }
+ r <<= s; // shift when v's highest bits are zero
+
+ data.u[c] = r;
+ }
+ break;
+
+ case ir_unop_bit_count:
+ for (unsigned c = 0; c < components; c++) {
+ unsigned count = 0;
+ unsigned v = op[0]->value.u[c];
+
+ for (; v; count++) {
+ v &= v - 1;
+ }
+ data.u[c] = count;
+ }
+ break;
+
+ case ir_unop_find_msb:
+ for (unsigned c = 0; c < components; c++) {
+ int v = op[0]->value.i[c];
+
+ if (v == 0 || (op[0]->type->base_type == GLSL_TYPE_INT && v == -1))
+ data.i[c] = -1;
+ else {
+ int count = 0;
+ int top_bit = op[0]->type->base_type == GLSL_TYPE_UINT
+ ? 0 : v & (1 << 31);
+
+ while (((v & (1 << 31)) == top_bit) && count != 32) {
+ count++;
+ v <<= 1;
+ }
+
+ data.i[c] = 31 - count;
+ }
+ }
+ break;
+
+ case ir_unop_find_lsb:
+ for (unsigned c = 0; c < components; c++) {
+ if (op[0]->value.i[c] == 0)
+ data.i[c] = -1;
+ else {
+ unsigned pos = 0;
+ unsigned v = op[0]->value.u[c];
+
+ for (; !(v & 1); v >>= 1) {
+ pos++;
+ }
+ data.u[c] = pos;
+ }
+ }
+ 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];
+
+ for (unsigned c = 0; c < components; 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_bfm: {
+ int bits = op[0]->value.i[0];
+ int offset = op[1]->value.i[0];
+
+ for (unsigned c = 0; c < components; c++) {
+ if (bits == 0)
+ data.u[c] = op[0]->value.u[c];
+ else if (offset < 0 || bits < 0)
+ data.u[c] = 0; /* Undefined for bitfieldInsert, per spec. */
+ else if (offset + bits > 32)
+ data.u[c] = 0; /* Undefined for bitfieldInsert, per spec. */
+ else
+ data.u[c] = ((1 << bits) - 1) << offset;
+ }
+ break;
+ }
+
+ case ir_binop_ldexp:
+ for (unsigned c = 0; c < components; 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 ||
+ 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;
+
+ 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;
+ }
+
+ 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;
+ }
+
+ case ir_quadop_bitfield_insert: {
+ int offset = op[2]->value.i[0];
+ int bits = op[3]->value.i[0];
+
+ for (unsigned c = 0; c < components; 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 = ((1 << 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;
+ }
+
case ir_quadop_vector:
for (unsigned c = 0; c < this->type->vector_elements; c++) {
switch (this->type->base_type) {
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;
}
}
}
-void
-ir_dereference_variable::constant_referenced(struct hash_table *variable_context,
- ir_constant *&store, int &offset) const
-{
- if (variable_context) {
- store = (ir_constant *)hash_table_find(variable_context, var);
- offset = 0;
- } else {
- store = NULL;
- offset = 0;
- }
-}
-
ir_constant *
ir_dereference_variable::constant_expression_value(struct hash_table *variable_context)
{
/* The constant_value of a uniform variable is its initializer,
* not the lifetime constant value of the uniform.
*/
- if (var->mode == ir_var_uniform)
+ if (var->data.mode == ir_var_uniform)
return NULL;
if (!var->constant_value)
}
-void
-ir_dereference_array::constant_referenced(struct hash_table *variable_context,
- ir_constant *&store, int &offset) const
-{
- ir_constant *index_c = array_index->constant_expression_value(variable_context);
-
- if (!index_c || !index_c->type->is_scalar() || !index_c->type->is_integer()) {
- store = 0;
- offset = 0;
- return;
- }
-
- int index = index_c->type->base_type == GLSL_TYPE_INT ?
- index_c->get_int_component(0) :
- index_c->get_uint_component(0);
-
- ir_constant *substore;
- int suboffset;
- const ir_dereference *deref = array->as_dereference();
- if (!deref) {
- store = 0;
- offset = 0;
- return;
- }
-
- deref->constant_referenced(variable_context, substore, suboffset);
-
- if (!substore) {
- store = 0;
- offset = 0;
- return;
- }
-
- const glsl_type *vt = substore->type;
- if (vt->is_array()) {
- store = substore->get_array_element(index);
- offset = 0;
- return;
- }
- if (vt->is_matrix()) {
- store = substore;
- offset = index * vt->vector_elements;
- return;
- }
- if (vt->is_vector()) {
- store = substore;
- offset = suboffset + index;
- return;
- }
-
- store = 0;
- offset = 0;
-}
-
ir_constant *
ir_dereference_array::constant_expression_value(struct hash_table *variable_context)
{
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;
}
-void
-ir_dereference_record::constant_referenced(struct hash_table *variable_context,
- ir_constant *&store, int &offset) const
-{
- ir_constant *substore;
- int suboffset;
- const ir_dereference *deref = record->as_dereference();
- if (!deref) {
- store = 0;
- offset = 0;
- return;
- }
-
- deref->constant_referenced(variable_context, substore, suboffset);
-
- if (!substore) {
- store = 0;
- offset = 0;
- return;
- }
-
- store = substore->get_record_field(field);
- offset = 0;
-}
-
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) */
ir_constant *store = NULL;
int offset = 0;
- asg->lhs->constant_referenced(variable_context, store, offset);
- if (!store)
+ if (!constant_referenced(asg->lhs, variable_context, store, offset))
return false;
ir_constant *value = asg->rhs->constant_expression_value(variable_context);
ir_constant *store = NULL;
int offset = 0;
- call->return_deref->constant_referenced(variable_context, store, offset);
- if (!store)
+ if (!constant_referenced(call->return_deref, variable_context,
+ store, offset))
return false;
ir_constant *value = call->constant_expression_value(variable_context);
* "Function calls to user-defined functions (non-built-in functions)
* cannot be used to form constant expressions."
*/
- if (!this->is_builtin)
+ if (!this->is_builtin())
return NULL;
/*
*/
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);
- if (constant == NULL)
- return NULL;
+ 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;
+ }
+
ir_variable *var = (ir_variable *)parameter_info;
hash_table_insert(deref_hash, constant, var);