/**
* \file ir_to_mesa.cpp
*
- * Translates the IR to ARB_fragment_program text if possible,
- * printing the result
+ * Translate GLSL IR to Mesa's gl_program representation.
*/
#include <stdio.h>
#include "program/program.h"
#include "program/prog_uniform.h"
#include "program/prog_parameter.h"
+#include "program/sampler.h"
}
static int swizzle_for_size(int size);
typedef struct ir_to_mesa_src_reg {
ir_to_mesa_src_reg(int file, int index, const glsl_type *type)
{
- this->file = file;
+ this->file = (gl_register_file) file;
this->index = index;
if (type && (type->is_scalar() || type->is_vector() || type->is_matrix()))
this->swizzle = swizzle_for_size(type->vector_elements);
this->reladdr = NULL;
}
- int file; /**< PROGRAM_* from Mesa */
+ gl_register_file file; /**< PROGRAM_* from Mesa */
int index; /**< temporary index, VERT_ATTRIB_*, FRAG_ATTRIB_*, etc. */
GLuint swizzle; /**< SWIZZLE_XYZWONEZERO swizzles from Mesa. */
int negate; /**< NEGATE_XYZW mask from mesa */
class ir_to_mesa_instruction : public exec_node {
public:
+ /* Callers of this ralloc-based new need not call delete. It's
+ * easier to just ralloc_free 'ctx' (or any of its ancestors). */
+ static void* operator new(size_t size, void *ctx)
+ {
+ void *node;
+
+ node = rzalloc_size(ctx, size);
+ assert(node != NULL);
+
+ return node;
+ }
+
enum prog_opcode op;
ir_to_mesa_dst_reg dst_reg;
ir_to_mesa_src_reg src_reg[3];
/** Pointer to the ir source this tree came from for debugging */
ir_instruction *ir;
GLboolean cond_update;
+ bool saturate;
int sampler; /**< sampler index */
int tex_target; /**< One of TEXTURE_*_INDEX */
GLboolean tex_shadow;
class variable_storage : public exec_node {
public:
- variable_storage(ir_variable *var, int file, int index)
+ variable_storage(ir_variable *var, gl_register_file file, int index)
: file(file), index(index), var(var)
{
/* empty */
}
- int file;
+ gl_register_file file;
int index;
ir_variable *var; /* variable that maps to this, if any */
};
function_entry *current_function;
- GLcontext *ctx;
+ struct gl_context *ctx;
struct gl_program *prog;
+ struct gl_shader_program *shader_program;
+ struct gl_shader_compiler_options *options;
int next_temp;
ir_to_mesa_src_reg src1,
ir_to_mesa_src_reg src2);
+ /**
+ * Emit the correct dot-product instruction for the type of arguments
+ *
+ * \sa ir_to_mesa_emit_op2
+ */
+ void ir_to_mesa_emit_dp(ir_instruction *ir,
+ ir_to_mesa_dst_reg dst,
+ ir_to_mesa_src_reg src0,
+ ir_to_mesa_src_reg src1,
+ unsigned elements);
+
void ir_to_mesa_emit_scalar_op1(ir_instruction *ir,
enum prog_opcode op,
ir_to_mesa_dst_reg dst,
ir_to_mesa_src_reg src0,
ir_to_mesa_src_reg src1);
+ void emit_scs(ir_instruction *ir, enum prog_opcode op,
+ ir_to_mesa_dst_reg dst,
+ const ir_to_mesa_src_reg &src);
+
GLboolean try_emit_mad(ir_expression *ir,
int mul_operand);
+ GLboolean try_emit_sat(ir_expression *ir);
- int add_uniform(const char *name,
- const glsl_type *type,
- ir_constant *constant);
- void add_aggregate_uniform(ir_instruction *ir,
- const char *name,
- const struct glsl_type *type,
- ir_constant *constant,
- struct ir_to_mesa_dst_reg temp);
+ void emit_swz(ir_expression *ir);
- struct hash_table *sampler_map;
+ bool process_move_condition(ir_rvalue *ir);
- void set_sampler_location(ir_variable *sampler, int location);
- int get_sampler_location(ir_variable *sampler);
+ void copy_propagate(void);
void *mem_ctx;
};
PROGRAM_ADDRESS, 0, WRITEMASK_X, COND_TR, NULL
};
-static int swizzle_for_size(int size)
+static void
+fail_link(struct gl_shader_program *prog, const char *fmt, ...) PRINTFLIKE(2, 3);
+
+static void
+fail_link(struct gl_shader_program *prog, const char *fmt, ...)
+{
+ va_list args;
+ va_start(args, fmt);
+ ralloc_vasprintf_append(&prog->InfoLog, fmt, args);
+ va_end(args);
+
+ prog->LinkStatus = GL_FALSE;
+}
+
+static int
+swizzle_for_size(int size)
{
int size_swizzles[4] = {
MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_X, SWIZZLE_X, SWIZZLE_X),
MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_W),
};
+ assert((size >= 1) && (size <= 4));
return size_swizzles[size - 1];
}
}
void
-ir_to_mesa_visitor::set_sampler_location(ir_variable *sampler, int location)
-{
- if (this->sampler_map == NULL) {
- this->sampler_map = hash_table_ctor(0, hash_table_pointer_hash,
- hash_table_pointer_compare);
- }
-
- hash_table_insert(this->sampler_map, (void *)(uintptr_t)location, sampler);
-}
-
-int
-ir_to_mesa_visitor::get_sampler_location(ir_variable *sampler)
+ir_to_mesa_visitor::ir_to_mesa_emit_dp(ir_instruction *ir,
+ ir_to_mesa_dst_reg dst,
+ ir_to_mesa_src_reg src0,
+ ir_to_mesa_src_reg src1,
+ unsigned elements)
{
- void *result = hash_table_find(this->sampler_map, sampler);
+ static const gl_inst_opcode dot_opcodes[] = {
+ OPCODE_DP2, OPCODE_DP3, OPCODE_DP4
+ };
- return (int)(uintptr_t)result;
+ ir_to_mesa_emit_op3(ir, dot_opcodes[elements - 2],
+ dst, src0, src1, ir_to_mesa_undef);
}
inline ir_to_mesa_dst_reg
GLuint src0_swiz = GET_SWZ(src0.swizzle, i);
GLuint src1_swiz = GET_SWZ(src1.swizzle, i);
for (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 &&
GET_SWZ(src1.swizzle, j) == src1_swiz) {
ir_to_mesa_emit_scalar_op2(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,
+ ir_to_mesa_dst_reg dst,
+ const ir_to_mesa_src_reg &src)
+{
+ /* Vertex programs cannot use the SCS opcode.
+ */
+ if (this->prog->Target == GL_VERTEX_PROGRAM_ARB) {
+ ir_to_mesa_emit_scalar_op1(ir, op, dst, src);
+ return;
+ }
+
+ const unsigned component = (op == OPCODE_SIN) ? 0 : 1;
+ const unsigned scs_mask = (1U << component);
+ int done_mask = ~dst.writemask;
+ ir_to_mesa_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);
+ ir_to_mesa_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;
+ ir_to_mesa_dst_reg tmp_dst = ir_to_mesa_dst_reg_from_src(tmp);
+
+ /* Emit the SCS instruction.
+ */
+ inst = ir_to_mesa_emit_op1(ir, OPCODE_SCS, tmp_dst, src0);
+ inst->dst_reg.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 = ir_to_mesa_emit_op1(ir, OPCODE_SCS, dst, tmp);
+ inst->dst_reg.writemask = this_mask;
+ } else {
+ /* Emit the SCS instruction to write directly to the destination.
+ */
+ ir_to_mesa_instruction *inst =
+ ir_to_mesa_emit_op1(ir, OPCODE_SCS, dst, src0);
+ inst->dst_reg.writemask = scs_mask;
+ }
+
+ done_mask |= this_mask;
+ }
+}
+
struct ir_to_mesa_src_reg
ir_to_mesa_visitor::src_reg_for_float(float val)
{
return 1;
}
case GLSL_TYPE_ARRAY:
+ assert(type->length > 0);
return type_size(type->fields.array) * type->length;
case GLSL_TYPE_STRUCT:
size = 0;
}
return size;
case GLSL_TYPE_SAMPLER:
- /* Samplers take up no register space, since they're baked in at
- * link time.
+ /* Samplers take up one slot in UNIFORMS[], but they're baked in
+ * at link time.
*/
- return 0;
+ return 1;
default:
assert(0);
return 0;
fp->OriginUpperLeft = ir->origin_upper_left;
fp->PixelCenterInteger = ir->pixel_center_integer;
+
+ } else if (strcmp(ir->name, "gl_FragDepth") == 0) {
+ struct gl_fragment_program *fp = (struct gl_fragment_program *)this->prog;
+ switch (ir->depth_layout) {
+ case ir_depth_layout_none:
+ fp->FragDepthLayout = FRAG_DEPTH_LAYOUT_NONE;
+ break;
+ case ir_depth_layout_any:
+ fp->FragDepthLayout = FRAG_DEPTH_LAYOUT_ANY;
+ break;
+ case ir_depth_layout_greater:
+ fp->FragDepthLayout = FRAG_DEPTH_LAYOUT_GREATER;
+ break;
+ case ir_depth_layout_less:
+ fp->FragDepthLayout = FRAG_DEPTH_LAYOUT_LESS;
+ break;
+ case ir_depth_layout_unchanged:
+ fp->FragDepthLayout = FRAG_DEPTH_LAYOUT_UNCHANGED;
+ break;
+ default:
+ assert(0);
+ break;
+ }
+ }
+
+ if (ir->mode == ir_var_uniform && strncmp(ir->name, "gl_", 3) == 0) {
+ unsigned int i;
+ const struct gl_builtin_uniform_desc *statevar;
+
+ for (i = 0; _mesa_builtin_uniform_desc[i].name; i++) {
+ if (strcmp(ir->name, _mesa_builtin_uniform_desc[i].name) == 0)
+ break;
+ }
+
+ if (!_mesa_builtin_uniform_desc[i].name) {
+ fail_link(this->shader_program,
+ "Failed to find builtin uniform `%s'\n", ir->name);
+ return;
+ }
+
+ statevar = &_mesa_builtin_uniform_desc[i];
+
+ int array_count;
+ if (ir->type->is_array()) {
+ array_count = ir->type->length;
+ } else {
+ array_count = 1;
+ }
+
+ /* Check if this statevar's setup in the STATE file exactly
+ * matches how we'll want to reference it as a
+ * struct/array/whatever. If not, then we need to move it into
+ * temporary storage and hope that it'll get copy-propagated
+ * out.
+ */
+ for (i = 0; i < statevar->num_elements; i++) {
+ if (statevar->elements[i].swizzle != SWIZZLE_XYZW) {
+ break;
+ }
+ }
+
+ struct variable_storage *storage;
+ ir_to_mesa_dst_reg dst;
+ if (i == statevar->num_elements) {
+ /* We'll set the index later. */
+ storage = new(mem_ctx) variable_storage(ir, PROGRAM_STATE_VAR, -1);
+ this->variables.push_tail(storage);
+
+ dst = ir_to_mesa_undef_dst;
+ } else {
+ storage = new(mem_ctx) variable_storage(ir, PROGRAM_TEMPORARY,
+ this->next_temp);
+ this->variables.push_tail(storage);
+ this->next_temp += type_size(ir->type);
+
+ dst = ir_to_mesa_dst_reg_from_src(ir_to_mesa_src_reg(PROGRAM_TEMPORARY,
+ storage->index,
+ NULL));
+ }
+
+
+ for (int a = 0; a < array_count; a++) {
+ for (unsigned int i = 0; i < statevar->num_elements; i++) {
+ struct gl_builtin_uniform_element *element = &statevar->elements[i];
+ int tokens[STATE_LENGTH];
+
+ memcpy(tokens, element->tokens, sizeof(element->tokens));
+ if (ir->type->is_array()) {
+ tokens[1] = a;
+ }
+
+ int index = _mesa_add_state_reference(this->prog->Parameters,
+ (gl_state_index *)tokens);
+
+ if (storage->file == PROGRAM_STATE_VAR) {
+ if (storage->index == -1) {
+ storage->index = index;
+ } else {
+ assert(index ==
+ (int)(storage->index + a * statevar->num_elements + i));
+ }
+ } else {
+ ir_to_mesa_src_reg src(PROGRAM_STATE_VAR, index, NULL);
+ src.swizzle = element->swizzle;
+ ir_to_mesa_emit_op1(ir, OPCODE_MOV, dst, src);
+ /* even a float takes up a whole vec4 reg in a struct/array. */
+ dst.index++;
+ }
+ }
+ }
+ if (storage->file == PROGRAM_TEMPORARY &&
+ dst.index != storage->index + type_size(ir->type)) {
+ fail_link(this->shader_program,
+ "failed to load builtin uniform `%s' (%d/%d regs loaded)\n",
+ ir->name, dst.index - storage->index,
+ type_size(ir->type));
+ }
}
}
void
ir_to_mesa_visitor::visit(ir_loop *ir)
{
- assert(!ir->from);
- assert(!ir->to);
- assert(!ir->increment);
- assert(!ir->counter);
+ ir_dereference_variable *counter = NULL;
+
+ if (ir->counter != NULL)
+ counter = new(ir) ir_dereference_variable(ir->counter);
+
+ if (ir->from != NULL) {
+ assert(ir->counter != NULL);
+
+ ir_assignment *a = new(ir) ir_assignment(counter, ir->from, NULL);
+
+ a->accept(this);
+ delete a;
+ }
ir_to_mesa_emit_op0(NULL, OPCODE_BGNLOOP);
+
+ if (ir->to) {
+ ir_expression *e =
+ new(ir) ir_expression(ir->cmp, glsl_type::bool_type,
+ counter, ir->to);
+ ir_if *if_stmt = new(ir) ir_if(e);
+
+ ir_loop_jump *brk = new(ir) ir_loop_jump(ir_loop_jump::jump_break);
+
+ if_stmt->then_instructions.push_tail(brk);
+
+ if_stmt->accept(this);
+
+ delete if_stmt;
+ delete e;
+ delete brk;
+ }
+
visit_exec_list(&ir->body_instructions, this);
+
+ if (ir->increment) {
+ ir_expression *e =
+ new(ir) ir_expression(ir_binop_add, counter->type,
+ counter, ir->increment);
+
+ ir_assignment *a = new(ir) ir_assignment(counter, e, NULL);
+
+ a->accept(this);
+ delete a;
+ delete e;
+ }
+
ir_to_mesa_emit_op0(NULL, OPCODE_ENDLOOP);
}
return true;
}
+GLboolean
+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);
+ ir_to_mesa_src_reg src = this->result;
+
+ this->result = get_temp(ir->type);
+ ir_to_mesa_instruction *inst;
+ inst = ir_to_mesa_emit_op1(ir, OPCODE_MOV,
+ ir_to_mesa_dst_reg_from_src(this->result),
+ src);
+ inst->saturate = true;
+
+ return true;
+}
+
void
ir_to_mesa_visitor::reladdr_to_temp(ir_instruction *ir,
ir_to_mesa_src_reg *reg, int *num_reladdr)
(*num_reladdr)--;
}
+void
+ir_to_mesa_visitor::emit_swz(ir_expression *ir)
+{
+ /* Assume that the vector operator is in a form compatible with OPCODE_SWZ.
+ * This means that each of the operands is either an immediate value of -1,
+ * 0, or 1, or is a component from one source register (possibly with
+ * negation).
+ */
+ uint8_t components[4] = { 0 };
+ bool negate[4] = { false };
+ ir_variable *var = NULL;
+
+ for (unsigned i = 0; i < ir->type->vector_elements; i++) {
+ ir_rvalue *op = ir->operands[i];
+
+ assert(op->type->is_scalar());
+
+ while (op != NULL) {
+ switch (op->ir_type) {
+ case ir_type_constant: {
+
+ assert(op->type->is_scalar());
+
+ const ir_constant *const c = op->as_constant();
+ if (c->is_one()) {
+ components[i] = SWIZZLE_ONE;
+ } else if (c->is_zero()) {
+ components[i] = SWIZZLE_ZERO;
+ } else if (c->is_negative_one()) {
+ components[i] = SWIZZLE_ONE;
+ negate[i] = true;
+ } else {
+ assert(!"SWZ constant must be 0.0 or 1.0.");
+ }
+
+ op = NULL;
+ break;
+ }
+
+ case ir_type_dereference_variable: {
+ ir_dereference_variable *const deref =
+ (ir_dereference_variable *) op;
+
+ assert((var == NULL) || (deref->var == var));
+ components[i] = SWIZZLE_X;
+ var = deref->var;
+ op = NULL;
+ break;
+ }
+
+ case ir_type_expression: {
+ ir_expression *const expr = (ir_expression *) op;
+
+ assert(expr->operation == ir_unop_neg);
+ negate[i] = true;
+
+ op = expr->operands[0];
+ break;
+ }
+
+ case ir_type_swizzle: {
+ ir_swizzle *const swiz = (ir_swizzle *) op;
+
+ components[i] = swiz->mask.x;
+ op = swiz->val;
+ break;
+ }
+
+ default:
+ assert(!"Should not get here.");
+ return;
+ }
+ }
+ }
+
+ assert(var != NULL);
+
+ ir_dereference_variable *const deref =
+ new(mem_ctx) ir_dereference_variable(var);
+
+ this->result.file = PROGRAM_UNDEFINED;
+ deref->accept(this);
+ if (this->result.file == PROGRAM_UNDEFINED) {
+ ir_print_visitor v;
+ printf("Failed to get tree for expression operand:\n");
+ deref->accept(&v);
+ exit(1);
+ }
+
+ ir_to_mesa_src_reg src;
+
+ src = this->result;
+ src.swizzle = MAKE_SWIZZLE4(components[0],
+ components[1],
+ components[2],
+ components[3]);
+ src.negate = ((unsigned(negate[0]) << 0)
+ | (unsigned(negate[1]) << 1)
+ | (unsigned(negate[2]) << 2)
+ | (unsigned(negate[3]) << 3));
+
+ /* Storage for our result. Ideally for an assignment we'd be using the
+ * actual storage for the result here, instead.
+ */
+ const ir_to_mesa_src_reg result_src = get_temp(ir->type);
+ ir_to_mesa_dst_reg result_dst = ir_to_mesa_dst_reg_from_src(result_src);
+
+ /* Limit writes to the channels that will be used by result_src later.
+ * This does limit this temp's use as a temporary for multi-instruction
+ * sequences.
+ */
+ result_dst.writemask = (1 << ir->type->vector_elements) - 1;
+
+ ir_to_mesa_emit_op1(ir, OPCODE_SWZ, result_dst, src);
+ this->result = result_src;
+}
+
void
ir_to_mesa_visitor::visit(ir_expression *ir)
{
unsigned int operand;
- struct ir_to_mesa_src_reg op[2];
+ struct ir_to_mesa_src_reg op[Elements(ir->operands)];
struct ir_to_mesa_src_reg result_src;
struct ir_to_mesa_dst_reg result_dst;
- const glsl_type *vec4_type = glsl_type::get_instance(GLSL_TYPE_FLOAT, 4, 1);
- const glsl_type *vec3_type = glsl_type::get_instance(GLSL_TYPE_FLOAT, 3, 1);
- const glsl_type *vec2_type = glsl_type::get_instance(GLSL_TYPE_FLOAT, 2, 1);
/* Quick peephole: Emit OPCODE_MAD(a, b, c) instead of ADD(MUL(a, b), c)
*/
if (try_emit_mad(ir, 0))
return;
}
+ if (try_emit_sat(ir))
+ return;
+
+ if (ir->operation == ir_quadop_vector) {
+ this->emit_swz(ir);
+ return;
+ }
for (operand = 0; operand < ir->get_num_operands(); operand++) {
this->result.file = PROGRAM_UNDEFINED;
assert(!ir->operands[operand]->type->is_matrix());
}
+ int vector_elements = ir->operands[0]->type->vector_elements;
+ if (ir->operands[1]) {
+ vector_elements = MAX2(vector_elements,
+ ir->operands[1]->type->vector_elements);
+ }
+
this->result.file = PROGRAM_UNDEFINED;
/* Storage for our result. Ideally for an assignment we'd be using
case ir_unop_cos:
ir_to_mesa_emit_scalar_op1(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:
ir_to_mesa_emit_op1(ir, OPCODE_DDX, result_dst, op[0]);
ir_to_mesa_emit_op1(ir, OPCODE_DDY, result_dst, op[0]);
break;
+ case ir_unop_noise: {
+ const enum prog_opcode opcode =
+ prog_opcode(OPCODE_NOISE1
+ + (ir->operands[0]->type->vector_elements) - 1);
+ assert((opcode >= OPCODE_NOISE1) && (opcode <= OPCODE_NOISE4));
+
+ ir_to_mesa_emit_op1(ir, opcode, result_dst, op[0]);
+ break;
+ }
+
case ir_binop_add:
ir_to_mesa_emit_op2(ir, OPCODE_ADD, result_dst, op[0], op[1]);
break;
ir_to_mesa_emit_op2(ir, OPCODE_SGE, result_dst, op[0], op[1]);
break;
case ir_binop_equal:
+ ir_to_mesa_emit_op2(ir, OPCODE_SEQ, result_dst, op[0], op[1]);
+ break;
+ case ir_binop_nequal:
+ ir_to_mesa_emit_op2(ir, OPCODE_SNE, result_dst, op[0], op[1]);
+ break;
+ case ir_binop_all_equal:
/* "==" operator producing a scalar boolean. */
if (ir->operands[0]->type->is_vector() ||
ir->operands[1]->type->is_vector()) {
ir_to_mesa_src_reg temp = get_temp(glsl_type::vec4_type);
ir_to_mesa_emit_op2(ir, OPCODE_SNE,
ir_to_mesa_dst_reg_from_src(temp), op[0], op[1]);
- ir_to_mesa_emit_op2(ir, OPCODE_DP4, result_dst, temp, temp);
+ ir_to_mesa_emit_dp(ir, result_dst, temp, temp, vector_elements);
ir_to_mesa_emit_op2(ir, OPCODE_SEQ,
result_dst, result_src, src_reg_for_float(0.0));
} else {
ir_to_mesa_emit_op2(ir, OPCODE_SEQ, result_dst, op[0], op[1]);
}
break;
- case ir_binop_nequal:
+ case ir_binop_any_nequal:
/* "!=" operator producing a scalar boolean. */
if (ir->operands[0]->type->is_vector() ||
ir->operands[1]->type->is_vector()) {
ir_to_mesa_src_reg temp = get_temp(glsl_type::vec4_type);
ir_to_mesa_emit_op2(ir, OPCODE_SNE,
ir_to_mesa_dst_reg_from_src(temp), op[0], op[1]);
- ir_to_mesa_emit_op2(ir, OPCODE_DP4, result_dst, temp, temp);
+ ir_to_mesa_emit_dp(ir, result_dst, temp, temp, vector_elements);
ir_to_mesa_emit_op2(ir, OPCODE_SNE,
result_dst, result_src, src_reg_for_float(0.0));
} else {
ir_to_mesa_emit_op2(ir, OPCODE_SNE, result_dst, op[0], op[1]);
}
break;
+
+ case ir_unop_any:
+ assert(ir->operands[0]->type->is_vector());
+ ir_to_mesa_emit_dp(ir, result_dst, op[0], op[0],
+ ir->operands[0]->type->vector_elements);
+ ir_to_mesa_emit_op2(ir, OPCODE_SNE,
+ result_dst, result_src, src_reg_for_float(0.0));
+ break;
+
case ir_binop_logic_xor:
ir_to_mesa_emit_op2(ir, OPCODE_SNE, result_dst, op[0], op[1]);
break;
break;
case ir_binop_dot:
- if (ir->operands[0]->type == vec4_type) {
- assert(ir->operands[1]->type == vec4_type);
- ir_to_mesa_emit_op2(ir, OPCODE_DP4,
- result_dst,
- op[0], op[1]);
- } else if (ir->operands[0]->type == vec3_type) {
- assert(ir->operands[1]->type == vec3_type);
- ir_to_mesa_emit_op2(ir, OPCODE_DP3,
- result_dst,
- op[0], op[1]);
- } else if (ir->operands[0]->type == vec2_type) {
- assert(ir->operands[1]->type == vec2_type);
- ir_to_mesa_emit_op2(ir, OPCODE_DP2,
- result_dst,
- op[0], op[1]);
- }
- break;
-
- case ir_binop_cross:
- ir_to_mesa_emit_op2(ir, OPCODE_XPD, result_dst, op[0], op[1]);
+ assert(ir->operands[0]->type->is_vector());
+ assert(ir->operands[0]->type == ir->operands[1]->type);
+ ir_to_mesa_emit_dp(ir, result_dst, op[0], op[1],
+ ir->operands[0]->type->vector_elements);
break;
case ir_unop_sqrt:
+ /* sqrt(x) = x * rsq(x). */
ir_to_mesa_emit_scalar_op1(ir, OPCODE_RSQ, result_dst, op[0]);
- ir_to_mesa_emit_scalar_op1(ir, OPCODE_RCP, result_dst, result_src);
- /* For incoming channels < 0, set the result to 0. */
+ ir_to_mesa_emit_op2(ir, OPCODE_MUL, result_dst, result_src, op[0]);
+ /* For incoming channels <= 0, set the result to 0. */
+ op[0].negate = ~op[0].negate;
ir_to_mesa_emit_op3(ir, OPCODE_CMP, result_dst,
- op[0], src_reg_for_float(0.0), result_src);
+ op[0], result_src, src_reg_for_float(0.0));
break;
case ir_unop_rsq:
ir_to_mesa_emit_scalar_op1(ir, OPCODE_RSQ, result_dst, op[0]);
case ir_unop_f2b:
case ir_unop_i2b:
ir_to_mesa_emit_op2(ir, OPCODE_SNE, result_dst,
- result_src, src_reg_for_float(0.0));
+ op[0], src_reg_for_float(0.0));
break;
case ir_unop_trunc:
ir_to_mesa_emit_op1(ir, OPCODE_TRUNC, result_dst, op[0]);
case ir_binop_bit_and:
case ir_binop_bit_xor:
case ir_binop_bit_or:
+ case ir_unop_round_even:
assert(!"GLSL 1.30 features unsupported");
break;
+
+ case ir_quadop_vector:
+ /* This operation should have already been handled.
+ */
+ assert(!"Should not get here.");
+ break;
}
this->result = result_src;
this->result = src_reg;
}
-static const struct {
- const char *name;
- const char *field;
- int tokens[STATE_LENGTH];
- int swizzle;
- bool array_indexed;
-} statevars[] = {
- {"gl_DepthRange", "near",
- {STATE_DEPTH_RANGE, 0, 0}, SWIZZLE_XXXX, false},
- {"gl_DepthRange", "far",
- {STATE_DEPTH_RANGE, 0, 0}, SWIZZLE_YYYY, false},
- {"gl_DepthRange", "diff",
- {STATE_DEPTH_RANGE, 0, 0}, SWIZZLE_ZZZZ, false},
-
- {"gl_ClipPlane", NULL,
- {STATE_CLIPPLANE, 0, 0}, SWIZZLE_XYZW, true}
-,
- {"gl_Point", "size",
- {STATE_POINT_SIZE}, SWIZZLE_XXXX, false},
- {"gl_Point", "sizeMin",
- {STATE_POINT_SIZE}, SWIZZLE_YYYY, false},
- {"gl_Point", "sizeMax",
- {STATE_POINT_SIZE}, SWIZZLE_ZZZZ, false},
- {"gl_Point", "fadeThresholdSize",
- {STATE_POINT_SIZE}, SWIZZLE_WWWW, false},
- {"gl_Point", "distanceConstantAttenuation",
- {STATE_POINT_ATTENUATION}, SWIZZLE_XXXX, false},
- {"gl_Point", "distanceLinearAttenuation",
- {STATE_POINT_ATTENUATION}, SWIZZLE_YYYY, false},
- {"gl_Point", "distanceQuadraticAttenuation",
- {STATE_POINT_ATTENUATION}, SWIZZLE_ZZZZ, false},
-
- {"gl_FrontMaterial", "emission",
- {STATE_MATERIAL, 0, STATE_EMISSION}, SWIZZLE_XYZW, false},
- {"gl_FrontMaterial", "ambient",
- {STATE_MATERIAL, 0, STATE_AMBIENT}, SWIZZLE_XYZW, false},
- {"gl_FrontMaterial", "diffuse",
- {STATE_MATERIAL, 0, STATE_DIFFUSE}, SWIZZLE_XYZW, false},
- {"gl_FrontMaterial", "specular",
- {STATE_MATERIAL, 0, STATE_SPECULAR}, SWIZZLE_XYZW, false},
- {"gl_FrontMaterial", "shininess",
- {STATE_MATERIAL, 0, STATE_SHININESS}, SWIZZLE_XXXX, false},
-
- {"gl_BackMaterial", "emission",
- {STATE_MATERIAL, 1, STATE_EMISSION}, SWIZZLE_XYZW, false},
- {"gl_BackMaterial", "ambient",
- {STATE_MATERIAL, 1, STATE_AMBIENT}, SWIZZLE_XYZW, false},
- {"gl_BackMaterial", "diffuse",
- {STATE_MATERIAL, 1, STATE_DIFFUSE}, SWIZZLE_XYZW, false},
- {"gl_BackMaterial", "specular",
- {STATE_MATERIAL, 1, STATE_SPECULAR}, SWIZZLE_XYZW, false},
- {"gl_BackMaterial", "shininess",
- {STATE_MATERIAL, 1, STATE_SHININESS}, SWIZZLE_XXXX, false},
-
- {"gl_LightSource", "ambient",
- {STATE_LIGHT, 0, STATE_AMBIENT}, SWIZZLE_XYZW, true},
- {"gl_LightSource", "diffuse",
- {STATE_LIGHT, 0, STATE_DIFFUSE}, SWIZZLE_XYZW, true},
- {"gl_LightSource", "specular",
- {STATE_LIGHT, 0, STATE_SPECULAR}, SWIZZLE_XYZW, true},
- {"gl_LightSource", "position",
- {STATE_LIGHT, 0, STATE_POSITION}, SWIZZLE_XYZW, true},
- {"gl_LightSource", "halfVector",
- {STATE_LIGHT, 0, STATE_HALF_VECTOR}, SWIZZLE_XYZW, true},
- {"gl_LightSource", "spotDirection",
- {STATE_LIGHT, 0, STATE_SPOT_DIRECTION}, SWIZZLE_XYZW, true},
- {"gl_LightSource", "spotCosCutoff",
- {STATE_LIGHT, 0, STATE_SPOT_DIRECTION}, SWIZZLE_WWWW, true},
- {"gl_LightSource", "spotCutoff",
- {STATE_LIGHT, 0, STATE_SPOT_CUTOFF}, SWIZZLE_XXXX, true},
- {"gl_LightSource", "spotExponent",
- {STATE_LIGHT, 0, STATE_ATTENUATION}, SWIZZLE_WWWW, true},
- {"gl_LightSource", "constantAttenuation",
- {STATE_LIGHT, 0, STATE_ATTENUATION}, SWIZZLE_XXXX, true},
- {"gl_LightSource", "linearAttenuation",
- {STATE_LIGHT, 0, STATE_ATTENUATION}, SWIZZLE_YYYY, true},
- {"gl_LightSource", "quadraticAttenuation",
- {STATE_LIGHT, 0, STATE_ATTENUATION}, SWIZZLE_ZZZZ, true},
-
- {"gl_LightModel", "ambient",
- {STATE_LIGHTMODEL_AMBIENT, 0}, SWIZZLE_XYZW, false},
-
- {"gl_FrontLightModelProduct", "sceneColor",
- {STATE_LIGHTMODEL_SCENECOLOR, 0}, SWIZZLE_XYZW, false},
- {"gl_BackLightModelProduct", "sceneColor",
- {STATE_LIGHTMODEL_SCENECOLOR, 1}, SWIZZLE_XYZW, false},
-
- {"gl_FrontLightProduct", "ambient",
- {STATE_LIGHTPROD, 0, 0, STATE_AMBIENT}, SWIZZLE_XYZW, true},
- {"gl_FrontLightProduct", "diffuse",
- {STATE_LIGHTPROD, 0, 0, STATE_DIFFUSE}, SWIZZLE_XYZW, true},
- {"gl_FrontLightProduct", "specular",
- {STATE_LIGHTPROD, 0, 0, STATE_SPECULAR}, SWIZZLE_XYZW, true},
-
- {"gl_BackLightProduct", "ambient",
- {STATE_LIGHTPROD, 0, 1, STATE_AMBIENT}, SWIZZLE_XYZW, true},
- {"gl_BackLightProduct", "diffuse",
- {STATE_LIGHTPROD, 0, 1, STATE_DIFFUSE}, SWIZZLE_XYZW, true},
- {"gl_BackLightProduct", "specular",
- {STATE_LIGHTPROD, 0, 1, STATE_SPECULAR}, SWIZZLE_XYZW, true},
-
- {"gl_TextureEnvColor", NULL,
- {STATE_TEXENV_COLOR, 0}, SWIZZLE_XYZW, true},
-
- {"gl_EyePlaneS", NULL,
- {STATE_TEXGEN, 0, STATE_TEXGEN_EYE_S}, SWIZZLE_XYZW, true},
- {"gl_EyePlaneT", NULL,
- {STATE_TEXGEN, 0, STATE_TEXGEN_EYE_T}, SWIZZLE_XYZW, true},
- {"gl_EyePlaneR", NULL,
- {STATE_TEXGEN, 0, STATE_TEXGEN_EYE_R}, SWIZZLE_XYZW, true},
- {"gl_EyePlaneQ", NULL,
- {STATE_TEXGEN, 0, STATE_TEXGEN_EYE_Q}, SWIZZLE_XYZW, true},
-
- {"gl_ObjectPlaneS", NULL,
- {STATE_TEXGEN, 0, STATE_TEXGEN_OBJECT_S}, SWIZZLE_XYZW, true},
- {"gl_ObjectPlaneT", NULL,
- {STATE_TEXGEN, 0, STATE_TEXGEN_OBJECT_T}, SWIZZLE_XYZW, true},
- {"gl_ObjectPlaneR", NULL,
- {STATE_TEXGEN, 0, STATE_TEXGEN_OBJECT_R}, SWIZZLE_XYZW, true},
- {"gl_ObjectPlaneQ", NULL,
- {STATE_TEXGEN, 0, STATE_TEXGEN_OBJECT_Q}, SWIZZLE_XYZW, true},
-
- {"gl_Fog", "color",
- {STATE_FOG_COLOR}, SWIZZLE_XYZW, false},
- {"gl_Fog", "density",
- {STATE_FOG_PARAMS}, SWIZZLE_XXXX, false},
- {"gl_Fog", "start",
- {STATE_FOG_PARAMS}, SWIZZLE_YYYY, false},
- {"gl_Fog", "end",
- {STATE_FOG_PARAMS}, SWIZZLE_ZZZZ, false},
- {"gl_Fog", "scale",
- {STATE_FOG_PARAMS}, SWIZZLE_WWWW, false},
-};
-
-static ir_to_mesa_src_reg
-get_builtin_uniform_reg(struct gl_program *prog,
- const char *name, int array_index, const char *field)
-{
- unsigned int i;
- ir_to_mesa_src_reg src_reg;
- int tokens[STATE_LENGTH];
-
- for (i = 0; i < Elements(statevars); i++) {
- if (strcmp(statevars[i].name, name) != 0)
- continue;
- if (!field && statevars[i].field) {
- assert(!"FINISHME: whole-structure state var dereference");
- }
- if (field && (!statevars[i].field || strcmp(statevars[i].field, field) != 0))
- continue;
- break;
- }
-
- if (i == Elements(statevars)) {
- printf("builtin uniform %s%s%s not found\n",
- name,
- field ? "." : "",
- field ? field : "");
- abort();
- }
-
- memcpy(&tokens, statevars[i].tokens, sizeof(tokens));
- if (statevars[i].array_indexed)
- tokens[1] = array_index;
-
- src_reg.file = PROGRAM_STATE_VAR;
- src_reg.index = _mesa_add_state_reference(prog->Parameters,
- (gl_state_index *)tokens);
- src_reg.swizzle = statevars[i].swizzle;
- src_reg.negate = 0;
- src_reg.reladdr = false;
-
- return src_reg;
-}
-
-static int
-add_matrix_ref(struct gl_program *prog, int *tokens)
-{
- int base_pos = -1;
- int i;
-
- /* Add a ref for each column. It looks like the reason we do
- * it this way is that _mesa_add_state_reference doesn't work
- * for things that aren't vec4s, so the tokens[2]/tokens[3]
- * range has to be equal.
- */
- for (i = 0; i < 4; i++) {
- tokens[2] = i;
- tokens[3] = i;
- int pos = _mesa_add_state_reference(prog->Parameters,
- (gl_state_index *)tokens);
- if (base_pos == -1)
- base_pos = pos;
- else
- assert(base_pos + i == pos);
- }
-
- return base_pos;
-}
-
-static variable_storage *
-get_builtin_matrix_ref(void *mem_ctx, struct gl_program *prog, ir_variable *var,
- ir_rvalue *array_index)
-{
- /*
- * NOTE: The ARB_vertex_program extension specified that matrices get
- * loaded in registers in row-major order. With GLSL, we want column-
- * major order. So, we need to transpose all matrices here...
- */
- static const struct {
- const char *name;
- int matrix;
- int modifier;
- } matrices[] = {
- { "gl_ModelViewMatrix", STATE_MODELVIEW_MATRIX, STATE_MATRIX_TRANSPOSE },
- { "gl_ModelViewMatrixInverse", STATE_MODELVIEW_MATRIX, STATE_MATRIX_INVTRANS },
- { "gl_ModelViewMatrixTranspose", STATE_MODELVIEW_MATRIX, 0 },
- { "gl_ModelViewMatrixInverseTranspose", STATE_MODELVIEW_MATRIX, STATE_MATRIX_INVERSE },
-
- { "gl_ProjectionMatrix", STATE_PROJECTION_MATRIX, STATE_MATRIX_TRANSPOSE },
- { "gl_ProjectionMatrixInverse", STATE_PROJECTION_MATRIX, STATE_MATRIX_INVTRANS },
- { "gl_ProjectionMatrixTranspose", STATE_PROJECTION_MATRIX, 0 },
- { "gl_ProjectionMatrixInverseTranspose", STATE_PROJECTION_MATRIX, STATE_MATRIX_INVERSE },
-
- { "gl_ModelViewProjectionMatrix", STATE_MVP_MATRIX, STATE_MATRIX_TRANSPOSE },
- { "gl_ModelViewProjectionMatrixInverse", STATE_MVP_MATRIX, STATE_MATRIX_INVTRANS },
- { "gl_ModelViewProjectionMatrixTranspose", STATE_MVP_MATRIX, 0 },
- { "gl_ModelViewProjectionMatrixInverseTranspose", STATE_MVP_MATRIX, STATE_MATRIX_INVERSE },
-
- { "gl_TextureMatrix", STATE_TEXTURE_MATRIX, STATE_MATRIX_TRANSPOSE },
- { "gl_TextureMatrixInverse", STATE_TEXTURE_MATRIX, STATE_MATRIX_INVTRANS },
- { "gl_TextureMatrixTranspose", STATE_TEXTURE_MATRIX, 0 },
- { "gl_TextureMatrixInverseTranspose", STATE_TEXTURE_MATRIX, STATE_MATRIX_INVERSE },
-
- { "gl_NormalMatrix", STATE_MODELVIEW_MATRIX, STATE_MATRIX_INVERSE },
-
- };
- unsigned int i;
- variable_storage *entry;
-
- /* C++ gets angry when we try to use an int as a gl_state_index, so we use
- * ints for gl_state_index. Make sure they're compatible.
- */
- assert(sizeof(gl_state_index) == sizeof(int));
-
- for (i = 0; i < Elements(matrices); i++) {
- if (strcmp(var->name, matrices[i].name) == 0) {
- int tokens[STATE_LENGTH];
- int base_pos = -1;
-
- tokens[0] = matrices[i].matrix;
- tokens[4] = matrices[i].modifier;
- if (matrices[i].matrix == STATE_TEXTURE_MATRIX) {
- ir_constant *index = array_index->constant_expression_value();
- if (index) {
- tokens[1] = index->value.i[0];
- base_pos = add_matrix_ref(prog, tokens);
- } else {
- for (i = 0; i < var->type->length; i++) {
- tokens[1] = i;
- int pos = add_matrix_ref(prog, tokens);
- if (base_pos == -1)
- base_pos = pos;
- else
- assert(base_pos + (int)i * 4 == pos);
- }
- }
- } else {
- tokens[1] = 0; /* unused array index */
- base_pos = add_matrix_ref(prog, tokens);
- }
- tokens[4] = matrices[i].modifier;
-
- entry = new(mem_ctx) variable_storage(var,
- PROGRAM_STATE_VAR,
- base_pos);
-
- return entry;
- }
- }
-
- return NULL;
-}
-
-int
-ir_to_mesa_visitor::add_uniform(const char *name,
- const glsl_type *type,
- ir_constant *constant)
-{
- int len;
-
- if (type->is_vector() ||
- type->is_scalar()) {
- len = type->vector_elements;
- } else {
- len = type_size(type) * 4;
- }
-
- float *values = NULL;
- if (constant && type->is_array()) {
- values = (float *)malloc(type->length * 4 * sizeof(float));
-
- assert(type->fields.array->is_scalar() ||
- type->fields.array->is_vector() ||
- !"FINISHME: uniform array initializers for non-vector");
-
- for (unsigned int i = 0; i < type->length; i++) {
- ir_constant *element = constant->array_elements[i];
- unsigned int c;
-
- for (c = 0; c < type->fields.array->vector_elements; c++) {
- switch (type->fields.array->base_type) {
- case GLSL_TYPE_FLOAT:
- values[4 * i + c] = element->value.f[c];
- break;
- case GLSL_TYPE_INT:
- values[4 * i + c] = element->value.i[c];
- break;
- case GLSL_TYPE_UINT:
- values[4 * i + c] = element->value.u[c];
- break;
- case GLSL_TYPE_BOOL:
- values[4 * i + c] = element->value.b[c];
- break;
- default:
- assert(!"not reached");
- }
- }
- }
- } else if (constant) {
- values = (float *)malloc(16 * sizeof(float));
- for (unsigned int i = 0; i < type->components(); i++) {
- switch (type->base_type) {
- case GLSL_TYPE_FLOAT:
- values[i] = constant->value.f[i];
- break;
- case GLSL_TYPE_INT:
- values[i] = constant->value.i[i];
- break;
- case GLSL_TYPE_UINT:
- values[i] = constant->value.u[i];
- break;
- case GLSL_TYPE_BOOL:
- values[i] = constant->value.b[i];
- break;
- default:
- assert(!"not reached");
- }
- }
- }
-
- int loc = _mesa_add_uniform(this->prog->Parameters,
- name,
- len,
- type->gl_type,
- values);
- free(values);
-
- return loc;
-}
-
-/* Recursively add all the members of the aggregate uniform as uniform names
- * to Mesa, moving those uniforms to our structured temporary.
- */
-void
-ir_to_mesa_visitor::add_aggregate_uniform(ir_instruction *ir,
- const char *name,
- const struct glsl_type *type,
- ir_constant *constant,
- struct ir_to_mesa_dst_reg temp)
-{
- int loc;
-
- if (type->is_record()) {
- void *mem_ctx = talloc_new(NULL);
- ir_constant *field_constant = NULL;
-
- if (constant)
- field_constant = (ir_constant *)constant->components.get_head();
-
- for (unsigned int i = 0; i < type->length; i++) {
- const glsl_type *field_type = type->fields.structure[i].type;
-
- add_aggregate_uniform(ir,
- talloc_asprintf(mem_ctx, "%s.%s", name,
- type->fields.structure[i].name),
- field_type, field_constant, temp);
- temp.index += type_size(field_type);
-
- if (constant)
- field_constant = (ir_constant *)field_constant->next;
- }
-
- talloc_free(mem_ctx);
-
- return;
- }
-
- assert(type->is_vector() || type->is_scalar() || !"FINISHME: other types");
-
- loc = add_uniform(name, type, constant);
-
- ir_to_mesa_src_reg uniform(PROGRAM_UNIFORM, loc, type);
-
- for (int i = 0; i < type_size(type); i++) {
- ir_to_mesa_emit_op1(ir, OPCODE_MOV, temp, uniform);
- temp.index++;
- uniform.index++;
- }
-}
-
-
void
ir_to_mesa_visitor::visit(ir_dereference_variable *ir)
{
variable_storage *entry = find_variable_storage(ir->var);
- unsigned int loc;
+ ir_variable *var = ir->var;
if (!entry) {
- switch (ir->var->mode) {
+ switch (var->mode) {
case ir_var_uniform:
- entry = get_builtin_matrix_ref(this->mem_ctx, this->prog, ir->var,
- NULL);
- if (entry)
- break;
-
- /* FINISHME: Fix up uniform name for arrays and things */
- if (ir->var->type->base_type == GLSL_TYPE_SAMPLER ||
- (ir->var->type->base_type == GLSL_TYPE_ARRAY &&
- ir->var->type->fields.array->base_type == GLSL_TYPE_SAMPLER)) {
- int array_length;
-
- if (ir->var->type->base_type == GLSL_TYPE_ARRAY)
- array_length = ir->var->type->length;
- else
- array_length = 1;
- int sampler = _mesa_add_sampler(this->prog->Parameters,
- ir->var->name,
- ir->var->type->gl_type,
- array_length);
- set_sampler_location(ir->var, sampler);
-
- entry = new(mem_ctx) variable_storage(ir->var, PROGRAM_SAMPLER,
- sampler);
- this->variables.push_tail(entry);
- break;
- }
-
- assert(ir->var->type->gl_type != 0 &&
- ir->var->type->gl_type != GL_INVALID_ENUM);
-
- /* Oh, the joy of aggregate types in Mesa. Like constants,
- * we can only really do vec4s. So, make a temp, chop the
- * aggregate up into vec4s, and move those vec4s to the temp.
- */
- if (ir->var->type->is_record()) {
- ir_to_mesa_src_reg temp = get_temp(ir->var->type);
-
- entry = new(mem_ctx) variable_storage(ir->var,
- temp.file,
- temp.index);
- this->variables.push_tail(entry);
-
- add_aggregate_uniform(ir->var, ir->var->name, ir->var->type,
- ir->var->constant_value,
- ir_to_mesa_dst_reg_from_src(temp));
- break;
- }
-
- loc = add_uniform(ir->var->name,
- ir->var->type,
- ir->var->constant_value);
-
- /* Always mark the uniform used at this point. If it isn't
- * used, dead code elimination should have nuked the decl already.
- */
- this->prog->Parameters->Parameters[loc].Used = GL_TRUE;
-
- entry = new(mem_ctx) variable_storage(ir->var, PROGRAM_UNIFORM, loc);
+ entry = new(mem_ctx) variable_storage(var, PROGRAM_UNIFORM,
+ var->location);
this->variables.push_tail(entry);
break;
case ir_var_in:
- case ir_var_out:
case ir_var_inout:
/* The linker assigns locations for varyings and attributes,
* including deprecated builtins (like gl_Color), user-assign
*
* FINISHME: We would hit this path for function arguments. Fix!
*/
- assert(ir->var->location != -1);
- if (ir->var->mode == ir_var_in ||
- ir->var->mode == ir_var_inout) {
- entry = new(mem_ctx) variable_storage(ir->var,
- PROGRAM_INPUT,
- ir->var->location);
-
- if (this->prog->Target == GL_VERTEX_PROGRAM_ARB &&
- ir->var->location >= VERT_ATTRIB_GENERIC0) {
- _mesa_add_attribute(prog->Attributes,
- ir->var->name,
- _mesa_sizeof_glsl_type(ir->var->type->gl_type),
- ir->var->type->gl_type,
- ir->var->location - VERT_ATTRIB_GENERIC0);
- }
- } else {
- entry = new(mem_ctx) variable_storage(ir->var,
- PROGRAM_OUTPUT,
- ir->var->location);
- }
-
+ assert(var->location != -1);
+ entry = new(mem_ctx) variable_storage(var,
+ PROGRAM_INPUT,
+ var->location);
+ if (this->prog->Target == GL_VERTEX_PROGRAM_ARB &&
+ var->location >= VERT_ATTRIB_GENERIC0) {
+ _mesa_add_attribute(this->prog->Attributes,
+ var->name,
+ _mesa_sizeof_glsl_type(var->type->gl_type),
+ var->type->gl_type,
+ var->location - VERT_ATTRIB_GENERIC0);
+ }
+ break;
+ case ir_var_out:
+ assert(var->location != -1);
+ entry = new(mem_ctx) variable_storage(var,
+ PROGRAM_OUTPUT,
+ var->location);
break;
+ case ir_var_system_value:
+ entry = new(mem_ctx) variable_storage(var,
+ PROGRAM_SYSTEM_VALUE,
+ var->location);
+ break;
case ir_var_auto:
case ir_var_temporary:
- entry = new(mem_ctx) variable_storage(ir->var, PROGRAM_TEMPORARY,
+ entry = new(mem_ctx) variable_storage(var, PROGRAM_TEMPORARY,
this->next_temp);
this->variables.push_tail(entry);
- next_temp += type_size(ir->var->type);
+ next_temp += type_size(var->type);
break;
}
if (!entry) {
- printf("Failed to make storage for %s\n", ir->var->name);
+ printf("Failed to make storage for %s\n", var->name);
exit(1);
}
}
- this->result = ir_to_mesa_src_reg(entry->file, entry->index, ir->var->type);
+ this->result = ir_to_mesa_src_reg(entry->file, entry->index, var->type);
}
void
ir_to_mesa_visitor::visit(ir_dereference_array *ir)
{
- ir_variable *var = ir->variable_referenced();
ir_constant *index;
ir_to_mesa_src_reg src_reg;
- ir_dereference_variable *deref_var = ir->array->as_dereference_variable();
int element_size = type_size(ir->type);
index = ir->array_index->constant_expression_value();
- if (deref_var && strncmp(deref_var->var->name,
- "gl_TextureMatrix",
- strlen("gl_TextureMatrix")) == 0) {
- struct variable_storage *entry;
-
- entry = get_builtin_matrix_ref(this->mem_ctx, this->prog, deref_var->var,
- ir->array_index);
- assert(entry);
-
- ir_to_mesa_src_reg src_reg(entry->file, entry->index, ir->type);
-
- if (index) {
- src_reg.reladdr = NULL;
- } else {
- ir_to_mesa_src_reg index_reg = get_temp(glsl_type::float_type);
-
- ir->array_index->accept(this);
- ir_to_mesa_emit_op2(ir, OPCODE_MUL,
- ir_to_mesa_dst_reg_from_src(index_reg),
- this->result, src_reg_for_float(element_size));
-
- src_reg.reladdr = talloc(mem_ctx, ir_to_mesa_src_reg);
- memcpy(src_reg.reladdr, &index_reg, sizeof(index_reg));
- }
-
- this->result = src_reg;
- return;
- }
-
- if (strncmp(var->name, "gl_", 3) == 0 && var->mode == ir_var_uniform &&
- !var->type->is_matrix()) {
- ir_dereference_record *record = NULL;
- if (ir->array->ir_type == ir_type_dereference_record)
- record = (ir_dereference_record *)ir->array;
-
- assert(index || !"FINISHME: variable-indexed builtin uniform access");
-
- this->result = get_builtin_uniform_reg(prog,
- var->name,
- index->value.i[0],
- record ? record->field : NULL);
- }
-
ir->array->accept(this);
src_reg = this->result;
this->result, src_reg_for_float(element_size));
}
- src_reg.reladdr = talloc(mem_ctx, ir_to_mesa_src_reg);
+ src_reg.reladdr = ralloc(mem_ctx, ir_to_mesa_src_reg);
memcpy(src_reg.reladdr, &index_reg, sizeof(index_reg));
}
unsigned int i;
const glsl_type *struct_type = ir->record->type;
int offset = 0;
- ir_variable *var = ir->record->variable_referenced();
-
- if (strncmp(var->name, "gl_", 3) == 0 && var->mode == ir_var_uniform) {
- assert(var);
-
- this->result = get_builtin_uniform_reg(prog,
- var->name,
- 0,
- ir->field);
- return;
- }
ir->record->accept(this);
break;
offset += type_size(struct_type->fields.structure[i].type);
}
- this->result.swizzle = swizzle_for_size(ir->type->vector_elements);
+
+ /* If the type is smaller than a vec4, replicate the last channel out. */
+ if (ir->type->is_scalar() || ir->type->is_vector())
+ this->result.swizzle = swizzle_for_size(ir->type->vector_elements);
+ else
+ this->result.swizzle = SWIZZLE_NOOP;
+
this->result.index += offset;
}
return ir_to_mesa_dst_reg_from_src(v->result);
}
+/**
+ * Process the condition of a conditional assignment
+ *
+ * Examines the condition of a conditional assignment to generate the optimal
+ * first operand of a \c CMP instruction. If the condition is a relational
+ * operator with 0 (e.g., \c ir_binop_less), the value being compared will be
+ * used as the source for the \c CMP instruction. Otherwise the comparison
+ * is processed to a boolean result, and the boolean result is used as the
+ * operand to the CMP instruction.
+ */
+bool
+ir_to_mesa_visitor::process_move_condition(ir_rvalue *ir)
+{
+ ir_rvalue *src_ir = ir;
+ bool negate = true;
+ bool switch_order = false;
+
+ ir_expression *const expr = ir->as_expression();
+ if ((expr != NULL) && (expr->get_num_operands() == 2)) {
+ bool zero_on_left = false;
+
+ if (expr->operands[0]->is_zero()) {
+ src_ir = expr->operands[1];
+ zero_on_left = true;
+ } else if (expr->operands[1]->is_zero()) {
+ src_ir = expr->operands[0];
+ zero_on_left = false;
+ }
+
+ /* a is - 0 + - 0 +
+ * (a < 0) T F F ( a < 0) T F F
+ * (0 < a) F F T (-a < 0) F F T
+ * (a <= 0) T T F (-a < 0) F F T (swap order of other operands)
+ * (0 <= a) F T T ( a < 0) T F F (swap order of other operands)
+ * (a > 0) F F T (-a < 0) F F T
+ * (0 > a) T F F ( a < 0) T F F
+ * (a >= 0) F T T ( a < 0) T F F (swap order of other operands)
+ * (0 >= a) T T F (-a < 0) F F T (swap order of other operands)
+ *
+ * Note that exchanging the order of 0 and 'a' in the comparison simply
+ * means that the value of 'a' should be negated.
+ */
+ if (src_ir != ir) {
+ switch (expr->operation) {
+ case ir_binop_less:
+ switch_order = false;
+ negate = zero_on_left;
+ break;
+
+ case ir_binop_greater:
+ switch_order = false;
+ negate = !zero_on_left;
+ break;
+
+ case ir_binop_lequal:
+ switch_order = true;
+ negate = !zero_on_left;
+ break;
+
+ case ir_binop_gequal:
+ switch_order = true;
+ negate = zero_on_left;
+ break;
+
+ default:
+ /* This isn't the right kind of comparison afterall, so make sure
+ * the whole condition is visited.
+ */
+ src_ir = ir;
+ break;
+ }
+ }
+ }
+
+ src_ir->accept(this);
+
+ /* We use the OPCODE_CMP (a < 0 ? b : c) for conditional moves, and the
+ * condition we produced is 0.0 or 1.0. By flipping the sign, we can
+ * choose which value OPCODE_CMP produces without an extra instruction
+ * computing the condition.
+ */
+ if (negate)
+ this->result.negate = ~this->result.negate;
+
+ return switch_order;
+}
+
void
ir_to_mesa_visitor::visit(ir_assignment *ir)
{
assert(!ir->lhs->type->is_scalar() && !ir->lhs->type->is_vector());
l.writemask = WRITEMASK_XYZW;
} else if (ir->lhs->type->is_scalar()) {
- /* FINISHME: This hack makes writing to gl_FragData, which lives in the
+ /* FINISHME: This hack makes writing to gl_FragDepth, which lives in the
* FINISHME: W component of fragment shader output zero, work correctly.
*/
l.writemask = WRITEMASK_XYZW;
} else {
+ int swizzles[4];
+ int first_enabled_chan = 0;
+ int rhs_chan = 0;
+
assert(ir->lhs->type->is_vector());
l.writemask = ir->write_mask;
+
+ for (int i = 0; i < 4; i++) {
+ if (l.writemask & (1 << i)) {
+ first_enabled_chan = GET_SWZ(r.swizzle, i);
+ break;
+ }
+ }
+
+ /* Swizzle a small RHS vector into the channels being written.
+ *
+ * glsl ir treats write_mask as dictating how many channels are
+ * present on the RHS while Mesa IR treats write_mask as just
+ * showing which channels of the vec4 RHS get written.
+ */
+ for (int i = 0; i < 4; i++) {
+ if (l.writemask & (1 << i))
+ swizzles[i] = GET_SWZ(r.swizzle, rhs_chan++);
+ else
+ swizzles[i] = first_enabled_chan;
+ }
+ r.swizzle = MAKE_SWIZZLE4(swizzles[0], swizzles[1],
+ swizzles[2], swizzles[3]);
}
assert(l.file != PROGRAM_UNDEFINED);
assert(r.file != PROGRAM_UNDEFINED);
if (ir->condition) {
- ir_to_mesa_src_reg condition;
-
- ir->condition->accept(this);
- condition = this->result;
+ const bool switch_order = this->process_move_condition(ir->condition);
+ ir_to_mesa_src_reg condition = this->result;
- /* We use the OPCODE_CMP (a < 0 ? b : c) for conditional moves,
- * and the condition we produced is 0.0 or 1.0. By flipping the
- * sign, we can choose which value OPCODE_CMP produces without
- * an extra computing the condition.
- */
- condition.negate = ~condition.negate;
for (i = 0; i < type_size(ir->lhs->type); i++) {
- ir_to_mesa_emit_op3(ir, OPCODE_CMP, l,
- condition, r, ir_to_mesa_src_reg_from_dst(l));
+ if (switch_order) {
+ ir_to_mesa_emit_op3(ir, OPCODE_CMP, l,
+ condition, ir_to_mesa_src_reg_from_dst(l), r);
+ } else {
+ ir_to_mesa_emit_op3(ir, OPCODE_CMP, l,
+ condition, r, ir_to_mesa_src_reg_from_dst(l));
+ }
+
l.index++;
r.index++;
}
ir_to_mesa_visitor::visit(ir_constant *ir)
{
ir_to_mesa_src_reg src_reg;
- GLfloat stack_vals[4];
+ GLfloat stack_vals[4] = { 0 };
GLfloat *values = stack_vals;
unsigned int i;
return entry;
}
- entry = talloc(mem_ctx, function_entry);
+ entry = ralloc(mem_ctx, function_entry);
entry->sig = sig;
entry->sig_id = this->next_signature_id++;
entry->bgn_inst = NULL;
this->result = entry->return_reg;
}
-
void
ir_to_mesa_visitor::visit(ir_texture *ir)
{
if (ir->shadow_comparitor)
inst->tex_shadow = GL_TRUE;
- ir_variable *sampler = ir->sampler->variable_referenced();
-
- /* generate the mapping, remove when we generate storage at
- * declaration time
- */
- ir->sampler->accept(this);
-
- inst->sampler = get_sampler_location(sampler);
-
- ir_dereference_array *sampler_array = ir->sampler->as_dereference_array();
- if (sampler_array) {
- ir_constant *array_index =
- sampler_array->array_index->constant_expression_value();
-
- /* GLSL 1.10 and 1.20 allowed variable sampler array indices,
- * while GLSL 1.30 requires that the array indices be constant
- * integer expressions. We don't expect any driver to actually
- * work with a really variable array index, and in 1.20 all that
- * would work would be an unrolled loop counter, so assert that
- * we ended up with a constant at least..
- */
- assert(array_index);
- inst->sampler += array_index->value.i[0];
- }
+ inst->sampler = _mesa_get_sampler_uniform_value(ir->sampler,
+ this->shader_program,
+ this->prog);
- const glsl_type *sampler_type = sampler->type;
- while (sampler_type->base_type == GLSL_TYPE_ARRAY)
- sampler_type = sampler_type->fields.array;
+ const glsl_type *sampler_type = ir->sampler->type;
switch (sampler_type->sampler_dimensionality) {
case GLSL_SAMPLER_DIM_1D:
void
ir_to_mesa_visitor::visit(ir_discard *ir)
{
- assert(ir->condition == NULL); /* FINISHME */
+ struct gl_fragment_program *fp = (struct gl_fragment_program *)this->prog;
+
+ if (ir->condition) {
+ ir->condition->accept(this);
+ this->result.negate = ~this->result.negate;
+ ir_to_mesa_emit_op1(ir, OPCODE_KIL, ir_to_mesa_undef_dst, this->result);
+ } else {
+ ir_to_mesa_emit_op0(ir, OPCODE_KIL_NV);
+ }
- ir_to_mesa_emit_op0(ir, OPCODE_KIL_NV);
+ fp->UsesKill = GL_TRUE;
}
void
ir->condition->accept(this);
assert(this->result.file != PROGRAM_UNDEFINED);
- if (ctx->Shader.EmitCondCodes) {
+ if (this->options->EmitCondCodes) {
cond_inst = (ir_to_mesa_instruction *)this->instructions.get_tail();
/* See if we actually generated any instruction for generating
result.file = PROGRAM_UNDEFINED;
next_temp = 1;
next_signature_id = 1;
- sampler_map = NULL;
current_function = NULL;
- mem_ctx = talloc_new(NULL);
+ mem_ctx = ralloc_context(NULL);
}
ir_to_mesa_visitor::~ir_to_mesa_visitor()
{
- talloc_free(mem_ctx);
- if (this->sampler_map)
- hash_table_dtor(this->sampler_map);
+ ralloc_free(mem_ctx);
}
static struct prog_src_register
struct prog_src_register mesa_reg;
mesa_reg.File = reg.file;
- assert(reg.index < (1 << INST_INDEX_BITS) - 1);
+ assert(reg.index < (1 << INST_INDEX_BITS));
mesa_reg.Index = reg.index;
mesa_reg.Swizzle = reg.swizzle;
mesa_reg.RelAddr = reg.reladdr != NULL;
}
}
- if_stack = talloc_zero_array(v->mem_ctx, int, if_count);
- loop_stack = talloc_zero_array(v->mem_ctx, int, loop_count);
+ if_stack = rzalloc_array(v->mem_ctx, int, if_count);
+ loop_stack = rzalloc_array(v->mem_ctx, int, loop_count);
for (i = 0; i < num_instructions; i++) {
switch (mesa_instructions[i].Opcode) {
for (i = 0; i < prog->NumInstructions; i++) {
struct prog_instruction *inst = &prog->Instructions[i];
- /* Instead of just using the uniform's value to map to a
- * sampler, Mesa first allocates a separate number for the
- * sampler (_mesa_add_sampler), then we reindex it down to a
- * small integer (sampler_map[], SamplersUsed), then that gets
- * mapped to the uniform's value, and we get an actual sampler.
- */
if (_mesa_is_tex_instruction(inst->Opcode)) {
prog->SamplerTargets[inst->TexSrcUnit] =
(gl_texture_index)inst->TexSrcTarget;
_mesa_update_shader_textures_used(prog);
}
-/* Each stage has some uniforms in its Parameters list. The Uniforms
- * list for the linked shader program has a pointer to these uniforms
- * in each of the stage's Parameters list, so that their values can be
- * updated when a uniform is set.
+struct uniform_sort {
+ struct gl_uniform *u;
+ int pos;
+};
+
+/* The shader_program->Uniforms list is almost sorted in increasing
+ * uniform->{Frag,Vert}Pos locations, but not quite when there are
+ * uniforms shared between targets. We need to add parameters in
+ * increasing order for the targets.
+ */
+static int
+sort_uniforms(const void *a, const void *b)
+{
+ struct uniform_sort *u1 = (struct uniform_sort *)a;
+ struct uniform_sort *u2 = (struct uniform_sort *)b;
+
+ return u1->pos - u2->pos;
+}
+
+/* Add the uniforms to the parameters. The linker chose locations
+ * in our parameters lists (which weren't created yet), which the
+ * uniforms code will use to poke values into our parameters list
+ * when uniforms are updated.
*/
static void
-link_uniforms_to_shared_uniform_list(struct gl_uniform_list *uniforms,
- struct gl_program *prog)
+add_uniforms_to_parameters_list(struct gl_shader_program *shader_program,
+ struct gl_shader *shader,
+ struct gl_program *prog)
{
unsigned int i;
+ unsigned int next_sampler = 0, num_uniforms = 0;
+ struct uniform_sort *sorted_uniforms;
+
+ sorted_uniforms = ralloc_array(NULL, struct uniform_sort,
+ shader_program->Uniforms->NumUniforms);
+
+ for (i = 0; i < shader_program->Uniforms->NumUniforms; i++) {
+ struct gl_uniform *uniform = shader_program->Uniforms->Uniforms + i;
+ int parameter_index = -1;
+
+ switch (shader->Type) {
+ case GL_VERTEX_SHADER:
+ parameter_index = uniform->VertPos;
+ break;
+ case GL_FRAGMENT_SHADER:
+ parameter_index = uniform->FragPos;
+ break;
+ case GL_GEOMETRY_SHADER:
+ parameter_index = uniform->GeomPos;
+ break;
+ }
+
+ /* Only add uniforms used in our target. */
+ if (parameter_index != -1) {
+ sorted_uniforms[num_uniforms].pos = parameter_index;
+ sorted_uniforms[num_uniforms].u = uniform;
+ num_uniforms++;
+ }
+ }
+
+ qsort(sorted_uniforms, num_uniforms, sizeof(struct uniform_sort),
+ sort_uniforms);
+
+ for (i = 0; i < num_uniforms; i++) {
+ struct gl_uniform *uniform = sorted_uniforms[i].u;
+ int parameter_index = sorted_uniforms[i].pos;
+ const glsl_type *type = uniform->Type;
+ unsigned int size;
+
+ if (type->is_vector() ||
+ type->is_scalar()) {
+ size = type->vector_elements;
+ } else {
+ size = type_size(type) * 4;
+ }
+
+ gl_register_file file;
+ if (type->is_sampler() ||
+ (type->is_array() && type->fields.array->is_sampler())) {
+ file = PROGRAM_SAMPLER;
+ } else {
+ file = PROGRAM_UNIFORM;
+ }
+
+ GLint index = _mesa_lookup_parameter_index(prog->Parameters, -1,
+ uniform->Name);
+
+ if (index < 0) {
+ index = _mesa_add_parameter(prog->Parameters, file,
+ uniform->Name, size, type->gl_type,
+ NULL, NULL, 0x0);
+
+ /* Sampler uniform values are stored in prog->SamplerUnits,
+ * and the entry in that array is selected by this index we
+ * store in ParameterValues[].
+ */
+ if (file == PROGRAM_SAMPLER) {
+ for (unsigned int j = 0; j < size / 4; j++)
+ prog->Parameters->ParameterValues[index + j][0] = next_sampler++;
+ }
+
+ /* The location chosen in the Parameters list here (returned
+ * from _mesa_add_uniform) has to match what the linker chose.
+ */
+ if (index != parameter_index) {
+ fail_link(shader_program, "Allocation of uniform `%s' to target "
+ "failed (%d vs %d)\n",
+ uniform->Name, index, parameter_index);
+ }
+ }
+ }
+
+ ralloc_free(sorted_uniforms);
+}
+
+static void
+set_uniform_initializer(struct gl_context *ctx, void *mem_ctx,
+ struct gl_shader_program *shader_program,
+ const char *name, const glsl_type *type,
+ ir_constant *val)
+{
+ if (type->is_record()) {
+ ir_constant *field_constant;
+
+ field_constant = (ir_constant *)val->components.get_head();
+
+ for (unsigned int i = 0; i < type->length; i++) {
+ const glsl_type *field_type = type->fields.structure[i].type;
+ const char *field_name = ralloc_asprintf(mem_ctx, "%s.%s", name,
+ type->fields.structure[i].name);
+ set_uniform_initializer(ctx, mem_ctx, shader_program, field_name,
+ field_type, field_constant);
+ field_constant = (ir_constant *)field_constant->next;
+ }
+ return;
+ }
+
+ int loc = _mesa_get_uniform_location(ctx, shader_program, name);
+
+ if (loc == -1) {
+ fail_link(shader_program,
+ "Couldn't find uniform for initializer %s\n", name);
+ return;
+ }
+
+ for (unsigned int i = 0; i < (type->is_array() ? type->length : 1); i++) {
+ ir_constant *element;
+ const glsl_type *element_type;
+ if (type->is_array()) {
+ element = val->array_elements[i];
+ element_type = type->fields.array;
+ } else {
+ element = val;
+ element_type = type;
+ }
+
+ void *values;
+
+ if (element_type->base_type == GLSL_TYPE_BOOL) {
+ int *conv = ralloc_array(mem_ctx, int, element_type->components());
+ for (unsigned int j = 0; j < element_type->components(); j++) {
+ conv[j] = element->value.b[j];
+ }
+ values = (void *)conv;
+ element_type = glsl_type::get_instance(GLSL_TYPE_INT,
+ element_type->vector_elements,
+ 1);
+ } else {
+ values = &element->value;
+ }
+
+ if (element_type->is_matrix()) {
+ _mesa_uniform_matrix(ctx, shader_program,
+ element_type->matrix_columns,
+ element_type->vector_elements,
+ loc, 1, GL_FALSE, (GLfloat *)values);
+ loc += element_type->matrix_columns;
+ } else {
+ _mesa_uniform(ctx, shader_program, loc, element_type->matrix_columns,
+ values, element_type->gl_type);
+ loc += type_size(element_type);
+ }
+ }
+}
+
+static void
+set_uniform_initializers(struct gl_context *ctx,
+ struct gl_shader_program *shader_program)
+{
+ void *mem_ctx = NULL;
+
+ for (unsigned int i = 0; i < MESA_SHADER_TYPES; i++) {
+ struct gl_shader *shader = shader_program->_LinkedShaders[i];
+
+ if (shader == NULL)
+ continue;
+
+ foreach_iter(exec_list_iterator, iter, *shader->ir) {
+ ir_instruction *ir = (ir_instruction *)iter.get();
+ ir_variable *var = ir->as_variable();
+
+ if (!var || var->mode != ir_var_uniform || !var->constant_value)
+ continue;
+
+ if (!mem_ctx)
+ mem_ctx = ralloc_context(NULL);
+
+ set_uniform_initializer(ctx, mem_ctx, shader_program, var->name,
+ var->type, var->constant_value);
+ }
+ }
+
+ ralloc_free(mem_ctx);
+}
+
+/*
+ * On a basic block basis, tracks available PROGRAM_TEMPORARY register
+ * channels for copy propagation and updates following instructions to
+ * use the original versions.
+ *
+ * The ir_to_mesa_visitor lazily produces code assuming that this pass
+ * will occur. As an example, a TXP production before this pass:
+ *
+ * 0: MOV TEMP[1], INPUT[4].xyyy;
+ * 1: MOV TEMP[1].w, INPUT[4].wwww;
+ * 2: TXP TEMP[2], TEMP[1], texture[0], 2D;
+ *
+ * and after:
+ *
+ * 0: MOV TEMP[1], INPUT[4].xyyy;
+ * 1: MOV TEMP[1].w, INPUT[4].wwww;
+ * 2: TXP TEMP[2], INPUT[4].xyyw, texture[0], 2D;
+ *
+ * which allows for dead code elimination on TEMP[1]'s writes.
+ */
+void
+ir_to_mesa_visitor::copy_propagate(void)
+{
+ ir_to_mesa_instruction **acp = rzalloc_array(mem_ctx,
+ ir_to_mesa_instruction *,
+ this->next_temp * 4);
+ int *acp_level = rzalloc_array(mem_ctx, int, this->next_temp * 4);
+ int level = 0;
+
+ foreach_iter(exec_list_iterator, iter, this->instructions) {
+ ir_to_mesa_instruction *inst = (ir_to_mesa_instruction *)iter.get();
+
+ assert(inst->dst_reg.file != PROGRAM_TEMPORARY
+ || inst->dst_reg.index < this->next_temp);
+
+ /* First, do any copy propagation possible into the src regs. */
+ for (int r = 0; r < 3; r++) {
+ ir_to_mesa_instruction *first = NULL;
+ bool good = true;
+ int acp_base = inst->src_reg[r].index * 4;
+
+ if (inst->src_reg[r].file != PROGRAM_TEMPORARY ||
+ inst->src_reg[r].reladdr)
+ continue;
+
+ /* See if we can find entries in the ACP consisting of MOVs
+ * from the same src register for all the swizzled channels
+ * of this src register reference.
+ */
+ for (int i = 0; i < 4; i++) {
+ int src_chan = GET_SWZ(inst->src_reg[r].swizzle, i);
+ ir_to_mesa_instruction *copy_chan = acp[acp_base + src_chan];
+
+ if (!copy_chan) {
+ good = false;
+ break;
+ }
+
+ assert(acp_level[acp_base + src_chan] <= level);
+
+ if (!first) {
+ first = copy_chan;
+ } else {
+ if (first->src_reg[0].file != copy_chan->src_reg[0].file ||
+ first->src_reg[0].index != copy_chan->src_reg[0].index) {
+ good = false;
+ break;
+ }
+ }
+ }
+
+ if (good) {
+ /* We've now validated that we can copy-propagate to
+ * replace this src register reference. Do it.
+ */
+ inst->src_reg[r].file = first->src_reg[0].file;
+ inst->src_reg[r].index = first->src_reg[0].index;
+
+ int swizzle = 0;
+ for (int i = 0; i < 4; i++) {
+ int src_chan = GET_SWZ(inst->src_reg[r].swizzle, i);
+ ir_to_mesa_instruction *copy_inst = acp[acp_base + src_chan];
+ swizzle |= (GET_SWZ(copy_inst->src_reg[0].swizzle, src_chan) <<
+ (3 * i));
+ }
+ inst->src_reg[r].swizzle = swizzle;
+ }
+ }
+
+ switch (inst->op) {
+ case OPCODE_BGNLOOP:
+ case OPCODE_ENDLOOP:
+ /* End of a basic block, clear the ACP entirely. */
+ memset(acp, 0, sizeof(*acp) * this->next_temp * 4);
+ break;
+
+ case OPCODE_IF:
+ ++level;
+ break;
+
+ case OPCODE_ENDIF:
+ case OPCODE_ELSE:
+ /* Clear all channels written inside the block from the ACP, but
+ * leaving those that were not touched.
+ */
+ for (int r = 0; r < this->next_temp; r++) {
+ for (int c = 0; c < 4; c++) {
+ if (!acp[4 * r + c])
+ continue;
+
+ if (acp_level[4 * r + c] >= level)
+ acp[4 * r + c] = NULL;
+ }
+ }
+ if (inst->op == OPCODE_ENDIF)
+ --level;
+ break;
+
+ default:
+ /* Continuing the block, clear any written channels from
+ * the ACP.
+ */
+ if (inst->dst_reg.file == PROGRAM_TEMPORARY && inst->dst_reg.reladdr) {
+ /* Any temporary might be written, so no copy propagation
+ * across this instruction.
+ */
+ memset(acp, 0, sizeof(*acp) * this->next_temp * 4);
+ } else if (inst->dst_reg.file == PROGRAM_OUTPUT &&
+ inst->dst_reg.reladdr) {
+ /* Any output might be written, so no copy propagation
+ * from outputs across this instruction.
+ */
+ for (int r = 0; r < this->next_temp; r++) {
+ for (int c = 0; c < 4; c++) {
+ if (!acp[4 * r + c])
+ continue;
+
+ if (acp[4 * r + c]->src_reg[0].file == PROGRAM_OUTPUT)
+ acp[4 * r + c] = NULL;
+ }
+ }
+ } else if (inst->dst_reg.file == PROGRAM_TEMPORARY ||
+ inst->dst_reg.file == PROGRAM_OUTPUT) {
+ /* Clear where it's used as dst. */
+ if (inst->dst_reg.file == PROGRAM_TEMPORARY) {
+ for (int c = 0; c < 4; c++) {
+ if (inst->dst_reg.writemask & (1 << c)) {
+ acp[4 * inst->dst_reg.index + c] = NULL;
+ }
+ }
+ }
+
+ /* Clear where it's used as src. */
+ for (int r = 0; r < this->next_temp; r++) {
+ for (int c = 0; c < 4; c++) {
+ if (!acp[4 * r + c])
+ continue;
+
+ int src_chan = GET_SWZ(acp[4 * r + c]->src_reg[0].swizzle, c);
- for (i = 0; i < prog->Parameters->NumParameters; i++) {
- const struct gl_program_parameter *p = prog->Parameters->Parameters + i;
+ if (acp[4 * r + c]->src_reg[0].file == inst->dst_reg.file &&
+ acp[4 * r + c]->src_reg[0].index == inst->dst_reg.index &&
+ inst->dst_reg.writemask & (1 << src_chan))
+ {
+ acp[4 * r + c] = NULL;
+ }
+ }
+ }
+ }
+ break;
+ }
- if (p->Type == PROGRAM_UNIFORM || p->Type == PROGRAM_SAMPLER) {
- struct gl_uniform *uniform =
- _mesa_append_uniform(uniforms, p->Name, prog->Target, i);
- if (uniform)
- uniform->Initialized = p->Initialized;
+ /* If this is a copy, add it to the ACP. */
+ if (inst->op == OPCODE_MOV &&
+ inst->dst_reg.file == PROGRAM_TEMPORARY &&
+ !inst->dst_reg.reladdr &&
+ !inst->saturate &&
+ !inst->src_reg[0].reladdr &&
+ !inst->src_reg[0].negate) {
+ for (int i = 0; i < 4; i++) {
+ if (inst->dst_reg.writemask & (1 << i)) {
+ acp[4 * inst->dst_reg.index + i] = inst;
+ acp_level[4 * inst->dst_reg.index + i] = level;
+ }
+ }
}
}
+
+ ralloc_free(acp_level);
+ ralloc_free(acp);
}
-struct gl_program *
-get_mesa_program(GLcontext *ctx, struct gl_shader_program *shader_program,
+
+/**
+ * Convert a shader's GLSL IR into a Mesa gl_program.
+ */
+static struct gl_program *
+get_mesa_program(struct gl_context *ctx,
+ struct gl_shader_program *shader_program,
struct gl_shader *shader)
{
ir_to_mesa_visitor v;
GLenum target;
const char *target_string;
GLboolean progress;
+ struct gl_shader_compiler_options *options =
+ &ctx->ShaderCompilerOptions[_mesa_shader_type_to_index(shader->Type)];
switch (shader->Type) {
case GL_VERTEX_SHADER:
target = GL_FRAGMENT_PROGRAM_ARB;
target_string = "fragment";
break;
+ case GL_GEOMETRY_SHADER:
+ target = GL_GEOMETRY_PROGRAM_NV;
+ target_string = "geometry";
+ break;
default:
assert(!"should not be reached");
- break;
+ return NULL;
}
validate_ir_tree(shader->ir);
prog->Attributes = _mesa_new_parameter_list();
v.ctx = ctx;
v.prog = prog;
+ v.shader_program = shader_program;
+ v.options = options;
+
+ add_uniforms_to_parameters_list(shader_program, shader, prog);
/* Emit Mesa IR for main(). */
visit_exec_list(shader->ir, &v);
mesa_instructions =
(struct prog_instruction *)calloc(num_instructions,
sizeof(*mesa_instructions));
- mesa_instruction_annotation = talloc_array(v.mem_ctx, ir_instruction *,
+ mesa_instruction_annotation = ralloc_array(v.mem_ctx, ir_instruction *,
num_instructions);
+ v.copy_propagate();
+
+ /* Convert ir_mesa_instructions into prog_instructions.
+ */
mesa_inst = mesa_instructions;
i = 0;
foreach_iter(exec_list_iterator, iter, v.instructions) {
- ir_to_mesa_instruction *inst = (ir_to_mesa_instruction *)iter.get();
+ const ir_to_mesa_instruction *inst = (ir_to_mesa_instruction *)iter.get();
mesa_inst->Opcode = inst->op;
mesa_inst->CondUpdate = inst->cond_update;
+ if (inst->saturate)
+ mesa_inst->SaturateMode = SATURATE_ZERO_ONE;
mesa_inst->DstReg.File = inst->dst_reg.file;
mesa_inst->DstReg.Index = inst->dst_reg.index;
mesa_inst->DstReg.CondMask = inst->dst_reg.cond_mask;
mesa_inst->TexShadow = inst->tex_shadow;
mesa_instruction_annotation[i] = inst->ir;
- if (ctx->Shader.EmitNoIfs && mesa_inst->Opcode == OPCODE_IF) {
- shader_program->InfoLog =
- talloc_asprintf_append(shader_program->InfoLog,
- "Couldn't flatten if statement\n");
- shader_program->LinkStatus = false;
+ /* Set IndirectRegisterFiles. */
+ if (mesa_inst->DstReg.RelAddr)
+ prog->IndirectRegisterFiles |= 1 << mesa_inst->DstReg.File;
+
+ /* Update program's bitmask of indirectly accessed register files */
+ for (unsigned src = 0; src < 3; src++)
+ if (mesa_inst->SrcReg[src].RelAddr)
+ prog->IndirectRegisterFiles |= 1 << mesa_inst->SrcReg[src].File;
+
+ if (options->EmitNoIfs && mesa_inst->Opcode == OPCODE_IF) {
+ fail_link(shader_program, "Couldn't flatten if statement\n");
}
switch (mesa_inst->Opcode) {
mesa_inst++;
i++;
+
+ if (!shader_program->LinkStatus)
+ break;
+ }
+
+ if (!shader_program->LinkStatus) {
+ free(mesa_instructions);
+ _mesa_reference_program(ctx, &shader->Program, NULL);
+ return NULL;
}
set_branchtargets(&v, mesa_instructions, num_instructions);
}
extern "C" {
+
+/**
+ * Called via ctx->Driver.CompilerShader().
+ * This is a no-op.
+ * XXX can we remove the ctx->Driver.CompileShader() hook?
+ */
GLboolean
-_mesa_ir_compile_shader(GLcontext *ctx, struct gl_shader *shader)
+_mesa_ir_compile_shader(struct gl_context *ctx, struct gl_shader *shader)
{
assert(shader->CompileStatus);
(void) ctx;
return GL_TRUE;
}
+
+/**
+ * Link a shader.
+ * Called via ctx->Driver.LinkShader()
+ * This actually involves converting GLSL IR into Mesa gl_programs with
+ * code lowering and other optimizations.
+ */
GLboolean
-_mesa_ir_link_shader(GLcontext *ctx, struct gl_shader_program *prog)
+_mesa_ir_link_shader(struct gl_context *ctx, struct gl_shader_program *prog)
{
assert(prog->LinkStatus);
- for (unsigned i = 0; i < prog->_NumLinkedShaders; i++) {
+ for (unsigned i = 0; i < MESA_SHADER_TYPES; i++) {
+ if (prog->_LinkedShaders[i] == NULL)
+ continue;
+
bool progress;
exec_list *ir = prog->_LinkedShaders[i]->ir;
+ const struct gl_shader_compiler_options *options =
+ &ctx->ShaderCompilerOptions[_mesa_shader_type_to_index(prog->_LinkedShaders[i]->Type)];
do {
progress = false;
/* Lowering */
do_mat_op_to_vec(ir);
- do_mod_to_fract(ir);
- do_div_to_mul_rcp(ir);
- do_explog_to_explog2(ir);
+ lower_instructions(ir, (MOD_TO_FRACT | DIV_TO_MUL_RCP | EXP_TO_EXP2
+ | LOG_TO_LOG2
+ | ((options->EmitNoPow) ? POW_TO_EXP2 : 0)));
+
+ progress = do_lower_jumps(ir, true, true, options->EmitNoMainReturn, options->EmitNoCont, options->EmitNoLoops) || progress;
+
+ progress = do_common_optimization(ir, true, options->MaxUnrollIterations) || progress;
+
+ progress = lower_quadop_vector(ir, true) || progress;
+
+ if (options->EmitNoIfs) {
+ progress = lower_discard(ir) || progress;
+ progress = lower_if_to_cond_assign(ir) || progress;
+ }
- progress = do_common_optimization(ir, true) || progress;
+ if (options->EmitNoNoise)
+ progress = lower_noise(ir) || progress;
- if (ctx->Shader.EmitNoIfs)
- progress = do_if_to_cond_assign(ir) || progress;
+ /* If there are forms of indirect addressing that the driver
+ * cannot handle, perform the lowering pass.
+ */
+ if (options->EmitNoIndirectInput || options->EmitNoIndirectOutput
+ || options->EmitNoIndirectTemp || options->EmitNoIndirectUniform)
+ progress =
+ lower_variable_index_to_cond_assign(ir,
+ options->EmitNoIndirectInput,
+ options->EmitNoIndirectOutput,
+ options->EmitNoIndirectTemp,
+ options->EmitNoIndirectUniform)
+ || progress;
progress = do_vec_index_to_cond_assign(ir) || progress;
} while (progress);
validate_ir_tree(ir);
}
- for (unsigned i = 0; i < prog->_NumLinkedShaders; i++) {
+ for (unsigned i = 0; i < MESA_SHADER_TYPES; i++) {
struct gl_program *linked_prog;
- bool ok = true;
- linked_prog = get_mesa_program(ctx, prog, prog->_LinkedShaders[i]);
+ if (prog->_LinkedShaders[i] == NULL)
+ continue;
- link_uniforms_to_shared_uniform_list(prog->Uniforms, linked_prog);
+ linked_prog = get_mesa_program(ctx, prog, prog->_LinkedShaders[i]);
- switch (prog->_LinkedShaders[i]->Type) {
- case GL_VERTEX_SHADER:
- _mesa_reference_vertprog(ctx, &prog->VertexProgram,
- (struct gl_vertex_program *)linked_prog);
- ok = ctx->Driver.ProgramStringNotify(ctx, GL_VERTEX_PROGRAM_ARB,
- linked_prog);
- break;
- case GL_FRAGMENT_SHADER:
- _mesa_reference_fragprog(ctx, &prog->FragmentProgram,
- (struct gl_fragment_program *)linked_prog);
- ok = ctx->Driver.ProgramStringNotify(ctx, GL_FRAGMENT_PROGRAM_ARB,
- linked_prog);
- break;
- }
- if (!ok) {
- return GL_FALSE;
+ if (linked_prog) {
+ bool ok = true;
+
+ switch (prog->_LinkedShaders[i]->Type) {
+ case GL_VERTEX_SHADER:
+ _mesa_reference_vertprog(ctx, &prog->VertexProgram,
+ (struct gl_vertex_program *)linked_prog);
+ ok = ctx->Driver.ProgramStringNotify(ctx, GL_VERTEX_PROGRAM_ARB,
+ linked_prog);
+ break;
+ case GL_FRAGMENT_SHADER:
+ _mesa_reference_fragprog(ctx, &prog->FragmentProgram,
+ (struct gl_fragment_program *)linked_prog);
+ ok = ctx->Driver.ProgramStringNotify(ctx, GL_FRAGMENT_PROGRAM_ARB,
+ linked_prog);
+ break;
+ case GL_GEOMETRY_SHADER:
+ _mesa_reference_geomprog(ctx, &prog->GeometryProgram,
+ (struct gl_geometry_program *)linked_prog);
+ ok = ctx->Driver.ProgramStringNotify(ctx, GL_GEOMETRY_PROGRAM_NV,
+ linked_prog);
+ break;
+ }
+ if (!ok) {
+ return GL_FALSE;
+ }
}
+
_mesa_reference_program(ctx, &linked_prog, NULL);
}
return GL_TRUE;
}
+
+/**
+ * Compile a GLSL shader. Called via glCompileShader().
+ */
void
-_mesa_glsl_compile_shader(GLcontext *ctx, struct gl_shader *shader)
+_mesa_glsl_compile_shader(struct gl_context *ctx, struct gl_shader *shader)
{
struct _mesa_glsl_parse_state *state =
new(shader) _mesa_glsl_parse_state(ctx, shader->Type, shader);
const char *source = shader->Source;
+ /* Check if the user called glCompileShader without first calling
+ * glShaderSource. This should fail to compile, but not raise a GL_ERROR.
+ */
+ if (source == NULL) {
+ shader->CompileStatus = GL_FALSE;
+ return;
+ }
+
state->error = preprocess(state, &source, &state->info_log,
- &ctx->Extensions);
+ &ctx->Extensions, ctx->API);
if (ctx->Shader.Flags & GLSL_DUMP) {
printf("GLSL source for shader %d:\n", shader->Name);
_mesa_glsl_lexer_dtor(state);
}
- talloc_free(shader->ir);
+ ralloc_free(shader->ir);
shader->ir = new(shader) exec_list;
if (!state->error && !state->translation_unit.is_empty())
_mesa_ast_to_hir(shader->ir, state);
/* Do some optimization at compile time to reduce shader IR size
* and reduce later work if the same shader is linked multiple times
*/
- while (do_common_optimization(shader->ir, false))
+ while (do_common_optimization(shader->ir, false, 32))
;
validate_ir_tree(shader->ir);
/* Retain any live IR, but trash the rest. */
reparent_ir(shader->ir, shader->ir);
- talloc_free(state);
+ ralloc_free(state);
if (shader->CompileStatus) {
if (!ctx->Driver.CompileShader(ctx, shader))
}
}
+
+/**
+ * Link a GLSL shader program. Called via glLinkProgram().
+ */
void
-_mesa_glsl_link_shader(GLcontext *ctx, struct gl_shader_program *prog)
+_mesa_glsl_link_shader(struct gl_context *ctx, struct gl_shader_program *prog)
{
unsigned int i;
for (i = 0; i < prog->NumShaders; i++) {
if (!prog->Shaders[i]->CompileStatus) {
- prog->InfoLog =
- talloc_asprintf_append(prog->InfoLog,
- "linking with uncompiled shader");
+ fail_link(prog, "linking with uncompiled shader");
prog->LinkStatus = GL_FALSE;
}
}
prog->Varying = _mesa_new_parameter_list();
_mesa_reference_vertprog(ctx, &prog->VertexProgram, NULL);
_mesa_reference_fragprog(ctx, &prog->FragmentProgram, NULL);
+ _mesa_reference_geomprog(ctx, &prog->GeometryProgram, NULL);
if (prog->LinkStatus) {
link_shaders(ctx, prog);
-
- /* We don't use the linker's uniforms list, and cook up our own at
- * generate time.
- */
- _mesa_free_uniform_list(prog->Uniforms);
- prog->Uniforms = _mesa_new_uniform_list();
}
if (prog->LinkStatus) {
}
}
+ set_uniform_initializers(ctx, prog);
+
if (ctx->Shader.Flags & GLSL_DUMP) {
if (!prog->LinkStatus) {
printf("GLSL shader program %d failed to link\n", prog->Name);
projects / mesa.git / blobdiff