#include "linker.h"
#include "main/mtypes.h"
+#include "main/shaderapi.h"
#include "main/shaderobj.h"
#include "main/uniforms.h"
-#include "program/hash_table.h"
-extern "C" {
-#include "main/shaderapi.h"
+#include "program/hash_table.h"
#include "program/prog_instruction.h"
#include "program/prog_optimize.h"
#include "program/prog_print.h"
#include "program/program.h"
#include "program/prog_parameter.h"
#include "program/sampler.h"
-}
+
static int swizzle_for_size(int size);
virtual void visit(ir_if *);
virtual void visit(ir_emit_vertex *);
virtual void visit(ir_end_primitive *);
+ virtual void visit(ir_barrier *);
/*@}*/
src_reg result;
void emit_scalar(ir_instruction *ir, enum prog_opcode op,
dst_reg dst, src_reg src0, src_reg src1);
- void emit_scs(ir_instruction *ir, enum prog_opcode op,
- dst_reg dst, const src_reg &src);
-
bool try_emit_mad(ir_expression *ir,
int mul_operand);
bool try_emit_mad_for_and_not(ir_expression *ir,
int mul_operand);
- bool try_emit_sat(ir_expression *ir);
void emit_swz(ir_expression *ir);
dst_reg dst, src_reg src0, src_reg src1,
unsigned elements)
{
- static const gl_inst_opcode dot_opcodes[] = {
+ static const enum prog_opcode dot_opcodes[] = {
OPCODE_DP2, OPCODE_DP3, OPCODE_DP4
};
emit_scalar(ir, op, dst, src0, undef);
}
-/**
- * Emit an OPCODE_SCS instruction
- *
- * The \c SCS opcode functions a bit differently than the other Mesa (or
- * ARB_fragment_program) opcodes. Instead of splatting its result across all
- * four components of the destination, it writes one value to the \c x
- * component and another value to the \c y component.
- *
- * \param ir IR instruction being processed
- * \param op Either \c OPCODE_SIN or \c OPCODE_COS depending on which
- * value is desired.
- * \param dst Destination register
- * \param src Source register
- */
-void
-ir_to_mesa_visitor::emit_scs(ir_instruction *ir, enum prog_opcode op,
- dst_reg dst,
- const src_reg &src)
-{
- /* Vertex programs cannot use the SCS opcode.
- */
- if (this->prog->Target == GL_VERTEX_PROGRAM_ARB) {
- emit_scalar(ir, op, dst, src);
- return;
- }
-
- const unsigned component = (op == OPCODE_SIN) ? 0 : 1;
- const unsigned scs_mask = (1U << component);
- int done_mask = ~dst.writemask;
- src_reg tmp;
-
- assert(op == OPCODE_SIN || op == OPCODE_COS);
-
- /* If there are compnents in the destination that differ from the component
- * that will be written by the SCS instrution, we'll need a temporary.
- */
- if (scs_mask != unsigned(dst.writemask)) {
- tmp = get_temp(glsl_type::vec4_type);
- }
-
- for (unsigned i = 0; i < 4; i++) {
- unsigned this_mask = (1U << i);
- src_reg src0 = src;
-
- if ((done_mask & this_mask) != 0)
- continue;
-
- /* The source swizzle specified which component of the source generates
- * sine / cosine for the current component in the destination. The SCS
- * instruction requires that this value be swizzle to the X component.
- * Replace the current swizzle with a swizzle that puts the source in
- * the X component.
- */
- unsigned src0_swiz = GET_SWZ(src.swizzle, i);
-
- src0.swizzle = MAKE_SWIZZLE4(src0_swiz, src0_swiz,
- src0_swiz, src0_swiz);
- for (unsigned j = i + 1; j < 4; j++) {
- /* If there is another enabled component in the destination that is
- * derived from the same inputs, generate its value on this pass as
- * well.
- */
- if (!(done_mask & (1 << j)) &&
- GET_SWZ(src0.swizzle, j) == src0_swiz) {
- this_mask |= (1 << j);
- }
- }
-
- if (this_mask != scs_mask) {
- ir_to_mesa_instruction *inst;
- dst_reg tmp_dst = dst_reg(tmp);
-
- /* Emit the SCS instruction.
- */
- inst = emit(ir, OPCODE_SCS, tmp_dst, src0);
- inst->dst.writemask = scs_mask;
-
- /* Move the result of the SCS instruction to the desired location in
- * the destination.
- */
- tmp.swizzle = MAKE_SWIZZLE4(component, component,
- component, component);
- inst = emit(ir, OPCODE_SCS, dst, tmp);
- inst->dst.writemask = this_mask;
- } else {
- /* Emit the SCS instruction to write directly to the destination.
- */
- ir_to_mesa_instruction *inst = emit(ir, OPCODE_SCS, dst, src0);
- inst->dst.writemask = scs_mask;
- }
-
- done_mask |= this_mask;
- }
-}
-
src_reg
ir_to_mesa_visitor::src_reg_for_float(float val)
{
*/
return 1;
}
+ break;
+ case GLSL_TYPE_DOUBLE:
+ if (type->is_matrix()) {
+ if (type->vector_elements > 2)
+ return type->matrix_columns * 2;
+ else
+ return type->matrix_columns;
+ } else {
+ if (type->vector_elements > 2)
+ return 2;
+ else
+ return 1;
+ }
+ break;
case GLSL_TYPE_ARRAY:
assert(type->length > 0);
return type_size(type->fields.array) * type->length;
return size;
case GLSL_TYPE_SAMPLER:
case GLSL_TYPE_IMAGE:
+ case GLSL_TYPE_SUBROUTINE:
/* Samplers take up one slot in UNIFORMS[], but they're baked in
* at link time.
*/
if (ir->data.mode == ir_var_uniform && strncmp(ir->name, "gl_", 3) == 0) {
unsigned int i;
- const ir_state_slot *const slots = ir->state_slots;
- assert(ir->state_slots != NULL);
+ const ir_state_slot *const slots = ir->get_state_slots();
+ assert(slots != NULL);
/* Check if this statevar's setup in the STATE file exactly
* matches how we'll want to reference it as a
* temporary storage and hope that it'll get copy-propagated
* out.
*/
- for (i = 0; i < ir->num_state_slots; i++) {
+ for (i = 0; i < ir->get_num_state_slots(); i++) {
if (slots[i].swizzle != SWIZZLE_XYZW) {
break;
}
variable_storage *storage;
dst_reg dst;
- if (i == ir->num_state_slots) {
+ if (i == ir->get_num_state_slots()) {
/* We'll set the index later. */
storage = new(mem_ctx) variable_storage(ir, PROGRAM_STATE_VAR, -1);
this->variables.push_tail(storage);
* of the type. However, this had better match the number of state
* elements that we're going to copy into the new temporary.
*/
- assert((int) ir->num_state_slots == type_size(ir->type));
+ assert((int) ir->get_num_state_slots() == type_size(ir->type));
storage = new(mem_ctx) variable_storage(ir, PROGRAM_TEMPORARY,
this->next_temp);
}
- for (unsigned int i = 0; i < ir->num_state_slots; i++) {
+ for (unsigned int i = 0; i < ir->get_num_state_slots(); i++) {
int index = _mesa_add_state_reference(this->prog->Parameters,
(gl_state_index *)slots[i].tokens);
}
if (storage->file == PROGRAM_TEMPORARY &&
- dst.index != storage->index + (int) ir->num_state_slots) {
+ dst.index != storage->index + (int) ir->get_num_state_slots()) {
linker_error(this->shader_program,
"failed to load builtin uniform `%s' "
"(%d/%d regs loaded)\n",
return true;
}
-bool
-ir_to_mesa_visitor::try_emit_sat(ir_expression *ir)
-{
- /* Saturates were only introduced to vertex programs in
- * NV_vertex_program3, so don't give them to drivers in the VP.
- */
- if (this->prog->Target == GL_VERTEX_PROGRAM_ARB)
- return false;
-
- ir_rvalue *sat_src = ir->as_rvalue_to_saturate();
- if (!sat_src)
- return false;
-
- sat_src->accept(this);
- src_reg src = this->result;
-
- /* If we generated an expression instruction into a temporary in
- * processing the saturate's operand, apply the saturate to that
- * instruction. Otherwise, generate a MOV to do the saturate.
- *
- * Note that we have to be careful to only do this optimization if
- * the instruction in question was what generated src->result. For
- * example, ir_dereference_array might generate a MUL instruction
- * to create the reladdr, and return us a src reg using that
- * reladdr. That MUL result is not the value we're trying to
- * saturate.
- */
- ir_expression *sat_src_expr = sat_src->as_expression();
- ir_to_mesa_instruction *new_inst;
- new_inst = (ir_to_mesa_instruction *)this->instructions.get_tail();
- if (sat_src_expr && (sat_src_expr->operation == ir_binop_mul ||
- sat_src_expr->operation == ir_binop_add ||
- sat_src_expr->operation == ir_binop_dot)) {
- new_inst->saturate = true;
- } else {
- this->result = get_temp(ir->type);
- ir_to_mesa_instruction *inst;
- inst = emit(ir, OPCODE_MOV, dst_reg(this->result), src);
- inst->saturate = true;
- }
-
- return true;
-}
-
void
ir_to_mesa_visitor::reladdr_to_temp(ir_instruction *ir,
src_reg *reg, int *num_reladdr)
ir_to_mesa_visitor::visit(ir_expression *ir)
{
unsigned int operand;
- src_reg op[Elements(ir->operands)];
+ src_reg op[ARRAY_SIZE(ir->operands)];
src_reg result_src;
dst_reg result_dst;
return;
}
- if (try_emit_sat(ir))
- return;
-
if (ir->operation == ir_quadop_vector) {
this->emit_swz(ir);
return;
case ir_unop_cos:
emit_scalar(ir, OPCODE_COS, result_dst, op[0]);
break;
- case ir_unop_sin_reduced:
- emit_scs(ir, OPCODE_SIN, result_dst, op[0]);
- break;
- case ir_unop_cos_reduced:
- emit_scs(ir, OPCODE_COS, result_dst, op[0]);
- break;
case ir_unop_dFdx:
emit(ir, OPCODE_DDX, result_dst, op[0]);
assert(!"not reached: should be handled by ir_div_to_mul_rcp");
break;
case ir_binop_mod:
- /* Floating point should be lowered by MOD_TO_FRACT in the compiler. */
+ /* Floating point should be lowered by MOD_TO_FLOOR in the compiler. */
assert(ir->type->is_integer());
emit(ir, OPCODE_MUL, result_dst, op[0], op[1]);
break;
case ir_unop_pack_unorm_2x16:
case ir_unop_pack_unorm_4x8:
case ir_unop_pack_half_2x16:
+ case ir_unop_pack_double_2x32:
case ir_unop_unpack_snorm_2x16:
case ir_unop_unpack_snorm_4x8:
case ir_unop_unpack_unorm_2x16:
case ir_unop_unpack_half_2x16:
case ir_unop_unpack_half_2x16_split_x:
case ir_unop_unpack_half_2x16_split_y:
+ case ir_unop_unpack_double_2x32:
case ir_binop_pack_half_2x16_split:
case ir_unop_bitfield_reverse:
case ir_unop_bit_count:
case ir_unop_find_msb:
case ir_unop_find_lsb:
+ case ir_unop_d2f:
+ case ir_unop_f2d:
+ case ir_unop_d2i:
+ case ir_unop_i2d:
+ case ir_unop_d2u:
+ case ir_unop_u2d:
+ case ir_unop_d2b:
+ case ir_unop_frexp_sig:
+ case ir_unop_frexp_exp:
assert(!"not supported");
break;
case ir_binop_min:
case ir_unop_dFdx_fine:
case ir_unop_dFdy_coarse:
case ir_unop_dFdy_fine:
+ case ir_unop_subroutine_to_int:
assert(!"not supported");
break;
ir->val->accept(this);
src = this->result;
assert(src.file != PROGRAM_UNDEFINED);
+ assert(ir->type->vector_elements > 0);
for (i = 0; i < 4; i++) {
if (i < ir->type->vector_elements) {
assert(!"Geometry shaders not supported.");
}
+void
+ir_to_mesa_visitor::visit(ir_barrier *)
+{
+ unreachable("GLSL barrier() not supported.");
+}
+
ir_to_mesa_visitor::ir_to_mesa_visitor()
{
result.file = PROGRAM_UNDEFINED;
if (type->is_vector() || type->is_scalar()) {
size = type->vector_elements;
+ if (type->is_double())
+ size *= 2;
} else {
size = type_size(type) * 4;
}
ir_variable *var = node->as_variable();
if ((var == NULL) || (var->data.mode != ir_var_uniform)
- || var->is_in_uniform_block() || (strncmp(var->name, "gl_", 3) == 0))
+ || var->is_in_buffer_block() || (strncmp(var->name, "gl_", 3) == 0))
continue;
add.process(var);
if (!found)
continue;
+ struct gl_uniform_storage *storage =
+ &shader_program->UniformStorage[location];
+
+ /* Do not associate any uniform storage to built-in uniforms */
+ if (storage->builtin)
+ continue;
+
if (location != last_location) {
- struct gl_uniform_storage *storage =
- &shader_program->UniformStorage[location];
enum gl_uniform_driver_format format = uniform_native;
unsigned columns = 0;
+ int dmul = 4 * sizeof(float);
switch (storage->type->base_type) {
case GLSL_TYPE_UINT:
assert(ctx->Const.NativeIntegers);
(ctx->Const.NativeIntegers) ? uniform_native : uniform_int_float;
columns = 1;
break;
+
+ case GLSL_TYPE_DOUBLE:
+ if (storage->type->vector_elements > 2)
+ dmul *= 2;
+ /* fallthrough */
case GLSL_TYPE_FLOAT:
format = uniform_native;
columns = storage->type->matrix_columns;
break;
case GLSL_TYPE_BOOL:
- if (ctx->Const.NativeIntegers) {
- format = (ctx->Const.UniformBooleanTrue == 1)
- ? uniform_bool_int_0_1 : uniform_bool_int_0_not0;
- } else {
- format = uniform_bool_float;
- }
+ format = uniform_native;
columns = 1;
break;
case GLSL_TYPE_SAMPLER:
case GLSL_TYPE_IMAGE:
+ case GLSL_TYPE_SUBROUTINE:
format = uniform_native;
columns = 1;
break;
}
_mesa_uniform_attach_driver_storage(storage,
- 4 * sizeof(float) * columns,
- 4 * sizeof(float),
+ dmul * columns,
+ dmul,
format,
¶ms->ParameterValues[i]);
mesa_inst->Opcode = inst->op;
mesa_inst->CondUpdate = inst->cond_update;
if (inst->saturate)
- mesa_inst->SaturateMode = SATURATE_ZERO_ONE;
+ mesa_inst->Saturate = GL_TRUE;
mesa_inst->DstReg.File = inst->dst.file;
mesa_inst->DstReg.Index = inst->dst.index;
mesa_inst->DstReg.CondMask = inst->dst.cond_mask;
/* Lowering */
do_mat_op_to_vec(ir);
- GLenum target = _mesa_shader_stage_to_program(prog->_LinkedShaders[i]->Stage);
- lower_instructions(ir, (MOD_TO_FRACT | DIV_TO_MUL_RCP | EXP_TO_EXP2
+ lower_instructions(ir, (MOD_TO_FLOOR | DIV_TO_MUL_RCP | EXP_TO_EXP2
| LOG_TO_LOG2 | INT_DIV_TO_MUL_RCP
- | ((options->EmitNoPow) ? POW_TO_EXP2 : 0)
- | ((target == GL_VERTEX_PROGRAM_ARB) ? SAT_TO_CLAMP
- : 0)));
+ | ((options->EmitNoPow) ? POW_TO_EXP2 : 0)));
progress = do_lower_jumps(ir, true, true, options->EmitNoMainReturn, options->EmitNoCont, options->EmitNoLoops) || progress;
if (options->EmitNoIndirectInput || options->EmitNoIndirectOutput
|| options->EmitNoIndirectTemp || options->EmitNoIndirectUniform)
progress =
- lower_variable_index_to_cond_assign(ir,
+ lower_variable_index_to_cond_assign(prog->_LinkedShaders[i]->Stage, ir,
options->EmitNoIndirectInput,
options->EmitNoIndirectOutput,
options->EmitNoIndirectTemp,
{
unsigned int i;
- _mesa_clear_shader_program_data(ctx, prog);
+ _mesa_clear_shader_program_data(prog);
prog->LinkStatus = GL_TRUE;
if (prog->LinkStatus) {
if (!ctx->Driver.LinkShader(ctx, prog)) {
prog->LinkStatus = GL_FALSE;
+ } else {
+ build_program_resource_list(ctx, prog);
}
}
projects / mesa.git / blobdiff