*/
extern "C" {
+
+#include <sys/types.h>
+
#include "main/macros.h"
#include "main/shaderobj.h"
#include "program/prog_parameter.h"
GRF = BRW_GENERAL_REGISTER_FILE,
MRF = BRW_MESSAGE_REGISTER_FILE,
IMM = BRW_IMMEDIATE_VALUE,
- FIXED_HW_REG,
+ FIXED_HW_REG, /* a struct brw_reg */
+ UNIFORM, /* prog_data->params[hw_reg] */
BAD_FILE
};
FS_OPCODE_DDX,
FS_OPCODE_DDY,
FS_OPCODE_LINTERP,
+ FS_OPCODE_TEX,
+ FS_OPCODE_TXB,
+ FS_OPCODE_TXL,
+ FS_OPCODE_DISCARD,
};
static int using_new_fs = -1;
struct brw_shader *shader;
shader = talloc_zero(NULL, struct brw_shader);
- shader->base.Type = type;
- shader->base.Name = name;
if (shader) {
+ shader->base.Type = type;
+ shader->base.Name = name;
_mesa_init_shader(ctx, &shader->base);
}
struct brw_shader_program *prog;
prog = talloc_zero(NULL, struct brw_shader_program);
if (prog) {
+ prog->base.Name = name;
_mesa_init_shader_program(ctx, &prog->base);
}
return &prog->base;
void *mem_ctx = talloc_new(NULL);
bool progress;
+ if (shader->ir)
+ talloc_free(shader->ir);
shader->ir = new(shader) exec_list;
clone_ir_list(mem_ctx, shader->ir, shader->base.ir);
do_mat_op_to_vec(shader->ir);
+ do_mod_to_fract(shader->ir);
do_div_to_mul_rcp(shader->ir);
do_sub_to_add_neg(shader->ir);
do_explog_to_explog2(shader->ir);
progress = do_common_optimization(shader->ir, true) || progress;
} while (progress);
- reparent_ir(shader->ir, shader);
+ validate_ir_tree(shader->ir);
+
+ reparent_ir(shader->ir, shader->ir);
talloc_free(mem_ctx);
}
}
case GLSL_TYPE_INT:
case GLSL_TYPE_FLOAT:
case GLSL_TYPE_BOOL:
- if (type->is_matrix()) {
- /* In case of incoming uniform/varying matrices, match their
- * allocation behavior. FINISHME: We could just use
- * glsl_type->components() for variables and temps within the
- * shader.
- */
- return type->matrix_columns * 4;
- } else {
- return type->vector_elements;
- }
+ return type->components();
case GLSL_TYPE_ARRAY:
/* FINISHME: uniform/varying arrays. */
return type_size(type->fields.array) * type->length;
{
void *node;
- node = talloc_size(ctx, size);
+ node = talloc_zero_size(ctx, size);
assert(node != NULL);
return node;
}
- fs_inst()
+ void init()
{
this->opcode = BRW_OPCODE_NOP;
this->saturate = false;
this->conditional_mod = BRW_CONDITIONAL_NONE;
this->predicated = false;
+ this->sampler = 0;
+ this->shadow_compare = false;
+ }
+
+ fs_inst()
+ {
+ init();
+ }
+
+ fs_inst(int opcode)
+ {
+ init();
+ this->opcode = opcode;
}
fs_inst(int opcode, fs_reg dst, fs_reg src0)
{
+ init();
this->opcode = opcode;
this->dst = dst;
this->src[0] = src0;
- this->saturate = false;
- this->conditional_mod = BRW_CONDITIONAL_NONE;
- this->predicated = false;
}
fs_inst(int opcode, fs_reg dst, fs_reg src0, fs_reg src1)
{
+ init();
this->opcode = opcode;
this->dst = dst;
this->src[0] = src0;
this->src[1] = src1;
- this->saturate = false;
- this->conditional_mod = BRW_CONDITIONAL_NONE;
- this->predicated = false;
}
fs_inst(int opcode, fs_reg dst, fs_reg src0, fs_reg src1, fs_reg src2)
{
+ init();
this->opcode = opcode;
this->dst = dst;
this->src[0] = src0;
this->src[1] = src1;
this->src[2] = src2;
- this->saturate = false;
- this->conditional_mod = BRW_CONDITIONAL_NONE;
- this->predicated = false;
}
int opcode; /* BRW_OPCODE_* or FS_OPCODE_* */
bool saturate;
bool predicated;
int conditional_mod; /**< BRW_CONDITIONAL_* */
+
+ int mlen; /** SEND message length */
+ int sampler;
+ bool shadow_compare;
+
+ /** @{
+ * Annotation for the generated IR. One of the two can be set.
+ */
+ ir_instruction *ir;
+ const char *annotation;
+ /** @} */
};
class fs_visitor : public ir_visitor
this->p = &c->func;
this->brw = p->brw;
this->intel = &brw->intel;
+ this->ctx = &intel->ctx;
this->mem_ctx = talloc_new(NULL);
this->shader = shader;
this->fail = false;
this->frag_data = NULL;
this->frag_depth = NULL;
this->first_non_payload_grf = 0;
+
+ this->current_annotation = NULL;
+ this->annotation_string = NULL;
+ this->annotation_ir = NULL;
}
~fs_visitor()
{
void visit(ir_function_signature *ir);
fs_inst *emit(fs_inst inst);
+ void assign_curb_setup();
void assign_urb_setup();
void assign_regs();
void generate_code();
void generate_fb_write(fs_inst *inst);
void generate_linterp(fs_inst *inst, struct brw_reg dst,
struct brw_reg *src);
+ void generate_tex(fs_inst *inst, struct brw_reg dst, struct brw_reg src);
void generate_math(fs_inst *inst, struct brw_reg dst, struct brw_reg *src);
+ void generate_discard(fs_inst *inst);
void emit_dummy_fs();
void emit_interpolation();
struct brw_context *brw;
struct intel_context *intel;
+ GLcontext *ctx;
struct brw_wm_compile *c;
struct brw_compile *p;
struct brw_shader *shader;
ir_variable *frag_color, *frag_data, *frag_depth;
int first_non_payload_grf;
+ /** @{ debug annotation info */
+ const char *current_annotation;
+ ir_instruction *base_ir;
+ const char **annotation_string;
+ ir_instruction **annotation_ir;
+ /** @} */
+
bool fail;
/* Result of last visit() method. */
{
fs_reg *reg = NULL;
- /* FINISHME */
- assert(ir->mode != ir_var_uniform);
-
if (strcmp(ir->name, "gl_FragColor") == 0) {
this->frag_color = ir;
} else if (strcmp(ir->name, "gl_FragData") == 0) {
this->frag_data = ir;
} else if (strcmp(ir->name, "gl_FragDepth") == 0) {
this->frag_depth = ir;
+ assert(!"FINISHME: this hangs currently.");
}
if (ir->mode == ir_var_in) {
reg = &this->interp_attrs[ir->location];
}
+ if (ir->mode == ir_var_uniform) {
+ const float *vec_values;
+ int param_index = c->prog_data.nr_params;
+
+ /* FINISHME: This is wildly incomplete. */
+ assert(ir->type->is_scalar() || ir->type->is_vector() ||
+ ir->type->is_sampler());
+
+ const struct gl_program *fp = &this->brw->fragment_program->Base;
+ /* Our support for uniforms is piggy-backed on the struct
+ * gl_fragment_program, because that's where the values actually
+ * get stored, rather than in some global gl_shader_program uniform
+ * store.
+ */
+ vec_values = fp->Parameters->ParameterValues[ir->location];
+ for (unsigned int i = 0; i < ir->type->vector_elements; i++) {
+ c->prog_data.param[c->prog_data.nr_params++] = &vec_values[i];
+ }
+
+ reg = new(this->mem_ctx) fs_reg(UNIFORM, param_index);
+ }
+
if (!reg)
reg = new(this->mem_ctx) fs_reg(this, ir->type);
void
fs_visitor::visit(ir_dereference_array *ir)
{
- assert(!"FINISHME");
+ ir_constant *index;
+ int element_size;
+
+ ir->array->accept(this);
+ index = ir->array_index->as_constant();
+
+ if (ir->type->is_matrix()) {
+ element_size = ir->type->vector_elements;
+ } else {
+ element_size = type_size(ir->type);
+ }
+
+ if (index) {
+ assert(this->result.file == UNIFORM ||
+ (this->result.file == GRF &&
+ this->result.reg != 0));
+ this->result.reg_offset += index->value.i[0] * element_size;
+ } else {
+ assert(!"FINISHME: non-constant matrix column");
+ }
}
void
emit(fs_inst(BRW_OPCODE_ADD, this->result, op[0], fs_reg(-1)));
break;
case ir_unop_neg:
- this->result = op[0];
op[0].negate = ~op[0].negate;
+ this->result = op[0];
break;
case ir_unop_abs:
- this->result = op[0];
op[0].abs = true;
+ this->result = op[0];
break;
case ir_unop_sign:
temp = fs_reg(this, ir->type);
- inst = emit(fs_inst(BRW_OPCODE_CMP, this->result, op[0], fs_reg(0.0f)));
+ emit(fs_inst(BRW_OPCODE_MOV, this->result, fs_reg(0.0f)));
+
+ inst = emit(fs_inst(BRW_OPCODE_CMP, reg_null, op[0], fs_reg(0.0f)));
inst->conditional_mod = BRW_CONDITIONAL_G;
+ inst = emit(fs_inst(BRW_OPCODE_MOV, this->result, fs_reg(1.0f)));
+ inst->predicated = true;
- inst = emit(fs_inst(BRW_OPCODE_CMP, temp, op[0], fs_reg(0.0f)));
+ inst = emit(fs_inst(BRW_OPCODE_CMP, reg_null, op[0], fs_reg(0.0f)));
inst->conditional_mod = BRW_CONDITIONAL_L;
-
- temp.negate = true;
- emit(fs_inst(BRW_OPCODE_ADD, this->result, this->result, temp));
+ inst = emit(fs_inst(BRW_OPCODE_MOV, this->result, fs_reg(-1.0f)));
+ inst->predicated = true;
break;
case ir_unop_rcp:
case ir_binop_less:
inst = emit(fs_inst(BRW_OPCODE_CMP, this->result, op[0], op[1]));
inst->conditional_mod = BRW_CONDITIONAL_L;
+ emit(fs_inst(BRW_OPCODE_AND, this->result, this->result, fs_reg(0x1)));
break;
case ir_binop_greater:
inst = emit(fs_inst(BRW_OPCODE_CMP, this->result, op[0], op[1]));
inst->conditional_mod = BRW_CONDITIONAL_G;
+ emit(fs_inst(BRW_OPCODE_AND, this->result, this->result, fs_reg(0x1)));
break;
case ir_binop_lequal:
inst = emit(fs_inst(BRW_OPCODE_CMP, this->result, op[0], op[1]));
inst->conditional_mod = BRW_CONDITIONAL_LE;
+ emit(fs_inst(BRW_OPCODE_AND, this->result, this->result, fs_reg(0x1)));
break;
case ir_binop_gequal:
inst = emit(fs_inst(BRW_OPCODE_CMP, this->result, op[0], op[1]));
inst->conditional_mod = BRW_CONDITIONAL_GE;
+ emit(fs_inst(BRW_OPCODE_AND, this->result, this->result, fs_reg(0x1)));
break;
case ir_binop_equal:
inst = emit(fs_inst(BRW_OPCODE_CMP, this->result, op[0], op[1]));
inst->conditional_mod = BRW_CONDITIONAL_Z;
+ emit(fs_inst(BRW_OPCODE_AND, this->result, this->result, fs_reg(0x1)));
break;
case ir_binop_nequal:
inst = emit(fs_inst(BRW_OPCODE_CMP, this->result, op[0], op[1]));
inst->conditional_mod = BRW_CONDITIONAL_NZ;
+ emit(fs_inst(BRW_OPCODE_AND, this->result, this->result, fs_reg(0x1)));
break;
case ir_binop_logic_xor:
}
for (i = 0; i < type_size(ir->lhs->type); i++) {
- if (i < 4 && !(write_mask & (1 << i)))
- continue;
-
- inst = emit(fs_inst(BRW_OPCODE_MOV, l, r));
- if (ir->condition)
- inst->predicated = true;
+ if (i >= 4 || (write_mask & (1 << i))) {
+ inst = emit(fs_inst(BRW_OPCODE_MOV, l, r));
+ if (ir->condition)
+ inst->predicated = true;
+ }
l.reg_offset++;
r.reg_offset++;
}
void
fs_visitor::visit(ir_texture *ir)
{
- assert(!"FINISHME");
+ int base_mrf = 2;
+ fs_inst *inst = NULL;
+ unsigned int mlen = 0;
+
+ ir->coordinate->accept(this);
+ fs_reg coordinate = this->result;
+
+ if (ir->projector) {
+ fs_reg inv_proj = fs_reg(this, glsl_type::float_type);
+
+ ir->projector->accept(this);
+ emit(fs_inst(FS_OPCODE_RCP, inv_proj, this->result));
+
+ fs_reg proj_coordinate = fs_reg(this, ir->coordinate->type);
+ for (unsigned int i = 0; i < ir->coordinate->type->vector_elements; i++) {
+ emit(fs_inst(BRW_OPCODE_MUL, proj_coordinate, coordinate, inv_proj));
+ coordinate.reg_offset++;
+ proj_coordinate.reg_offset++;
+ }
+ proj_coordinate.reg_offset = 0;
+
+ coordinate = proj_coordinate;
+ }
+
+ for (mlen = 0; mlen < ir->coordinate->type->vector_elements; mlen++) {
+ emit(fs_inst(BRW_OPCODE_MOV, fs_reg(MRF, base_mrf + mlen), coordinate));
+ coordinate.reg_offset++;
+ }
+
+ /* Pre-Ironlake, the 8-wide sampler always took u,v,r. */
+ if (intel->gen < 5)
+ mlen = 3;
+
+ if (ir->shadow_comparitor) {
+ /* For shadow comparisons, we have to supply u,v,r. */
+ mlen = 3;
+
+ ir->shadow_comparitor->accept(this);
+ emit(fs_inst(BRW_OPCODE_MOV, fs_reg(MRF, base_mrf + mlen), this->result));
+ mlen++;
+ }
+
+ /* Do we ever want to handle writemasking on texture samples? Is it
+ * performance relevant?
+ */
+ fs_reg dst = fs_reg(this, glsl_type::vec4_type);
+
+ switch (ir->op) {
+ case ir_tex:
+ inst = emit(fs_inst(FS_OPCODE_TEX, dst, fs_reg(MRF, base_mrf)));
+ break;
+ case ir_txb:
+ ir->lod_info.bias->accept(this);
+ emit(fs_inst(BRW_OPCODE_MOV, fs_reg(MRF, base_mrf + mlen), this->result));
+ mlen++;
+
+ inst = emit(fs_inst(FS_OPCODE_TXB, dst, fs_reg(MRF, base_mrf)));
+ break;
+ case ir_txl:
+ ir->lod_info.lod->accept(this);
+ emit(fs_inst(BRW_OPCODE_MOV, fs_reg(MRF, base_mrf + mlen), this->result));
+ mlen++;
+
+ inst = emit(fs_inst(FS_OPCODE_TXL, dst, fs_reg(MRF, base_mrf)));
+ break;
+ case ir_txd:
+ case ir_txf:
+ assert(!"GLSL 1.30 features unsupported");
+ break;
+ }
+
+ this->result = dst;
+
+ if (ir->shadow_comparitor)
+ inst->shadow_compare = true;
+ inst->mlen = mlen;
}
void
void
fs_visitor::visit(ir_discard *ir)
{
- assert(!"FINISHME");
+ assert(ir->condition == NULL); /* FINISHME */
+
+ emit(fs_inst(FS_OPCODE_DISCARD));
}
void
void
fs_visitor::visit(ir_if *ir)
{
- assert(!"FINISHME");
+ fs_inst *inst;
+
+ /* Don't point the annotation at the if statement, because then it plus
+ * the then and else blocks get printed.
+ */
+ this->base_ir = ir->condition;
+
+ /* Generate the condition into the condition code. */
+ ir->condition->accept(this);
+ inst = emit(fs_inst(BRW_OPCODE_MOV, fs_reg(brw_null_reg()), this->result));
+ inst->conditional_mod = BRW_CONDITIONAL_NZ;
+
+ inst = emit(fs_inst(BRW_OPCODE_IF));
+ inst->predicated = true;
+
+ foreach_iter(exec_list_iterator, iter, ir->then_instructions) {
+ ir_instruction *ir = (ir_instruction *)iter.get();
+ this->base_ir = ir;
+
+ ir->accept(this);
+ }
+
+ if (!ir->else_instructions.is_empty()) {
+ emit(fs_inst(BRW_OPCODE_ELSE));
+
+ foreach_iter(exec_list_iterator, iter, ir->else_instructions) {
+ ir_instruction *ir = (ir_instruction *)iter.get();
+ this->base_ir = ir;
+
+ ir->accept(this);
+ }
+ }
+
+ emit(fs_inst(BRW_OPCODE_ENDIF));
}
void
foreach_iter(exec_list_iterator, iter, sig->body) {
ir_instruction *ir = (ir_instruction *)iter.get();
+ this->base_ir = ir;
ir->accept(this);
}
fs_inst *list_inst = new(mem_ctx) fs_inst;
*list_inst = inst;
+ list_inst->annotation = this->current_annotation;
+ list_inst->ir = this->base_ir;
+
this->instructions.push_tail(list_inst);
return list_inst;
*/
fs_reg src_reg = reg_undef;
- /* Compute the pixel centers. */
+ this->current_annotation = "compute pixel centers";
this->pixel_x = fs_reg(this, glsl_type::uint_type);
this->pixel_y = fs_reg(this, glsl_type::uint_type);
emit(fs_inst(BRW_OPCODE_ADD,
fs_reg(stride(suboffset(g1_uw, 5), 2, 4, 0)),
fs_reg(brw_imm_v(0x11001100))));
- /* Compute the offsets from vertex 0 to the pixel centers */
+ this->current_annotation = "compute pixel deltas from v0";
this->delta_x = fs_reg(this, glsl_type::float_type);
this->delta_y = fs_reg(this, glsl_type::float_type);
emit(fs_inst(BRW_OPCODE_ADD,
this->pixel_y,
fs_reg(brw_vec1_grf(1, 1))));
+ this->current_annotation = "compute pos.w and 1/pos.w";
/* Compute wpos. Unlike many other varying inputs, we usually need it
* to produce 1/w, and the varying variable wouldn't show up.
*/
if (var->location == 0)
continue;
+ this->current_annotation = talloc_asprintf(this->mem_ctx,
+ "interpolate %s "
+ "(FRAG_ATTRIB[%d])",
+ var->name,
+ var->location);
emit_pinterp(var->location);
}
+ this->current_annotation = NULL;
}
void
void
fs_visitor::emit_fb_writes()
{
+ this->current_annotation = "FB write";
+
assert(this->frag_color || !"FINISHME: MRT");
fs_reg color = *(variable_storage(this->frag_color));
emit(fs_inst(FS_OPCODE_FB_WRITE,
fs_reg(0),
fs_reg(0)));
+
+ this->current_annotation = NULL;
}
void
break;
}
- brw_MOV(p, brw_message_reg(2), src[0]);
if (inst->opcode == FS_OPCODE_POW) {
brw_MOV(p, brw_message_reg(3), src[1]);
}
op,
inst->saturate ? BRW_MATH_SATURATE_SATURATE :
BRW_MATH_SATURATE_NONE,
- 2, brw_null_reg(),
+ 2, src[0],
BRW_MATH_DATA_VECTOR,
BRW_MATH_PRECISION_FULL);
}
+void
+fs_visitor::generate_tex(fs_inst *inst, struct brw_reg dst, struct brw_reg src)
+{
+ int msg_type = -1;
+ int rlen = 4;
+
+ if (intel->gen == 5) {
+ switch (inst->opcode) {
+ case FS_OPCODE_TEX:
+ if (inst->shadow_compare) {
+ msg_type = BRW_SAMPLER_MESSAGE_SAMPLE_COMPARE_GEN5;
+ } else {
+ msg_type = BRW_SAMPLER_MESSAGE_SAMPLE_GEN5;
+ }
+ break;
+ case FS_OPCODE_TXB:
+ if (inst->shadow_compare) {
+ msg_type = BRW_SAMPLER_MESSAGE_SAMPLE_BIAS_COMPARE_GEN5;
+ } else {
+ msg_type = BRW_SAMPLER_MESSAGE_SAMPLE_BIAS_GEN5;
+ }
+ break;
+ }
+ } else {
+ switch (inst->opcode) {
+ case FS_OPCODE_TEX:
+ /* Note that G45 and older determines shadow compare and dispatch width
+ * from message length for most messages.
+ */
+ if (inst->shadow_compare) {
+ msg_type = BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE_COMPARE;
+ } else {
+ msg_type = BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE;
+ }
+ case FS_OPCODE_TXB:
+ if (inst->shadow_compare) {
+ assert(!"FINISHME: shadow compare with bias.");
+ msg_type = BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE_BIAS;
+ } else {
+ msg_type = BRW_SAMPLER_MESSAGE_SIMD16_SAMPLE_BIAS;
+ rlen = 8;
+ }
+ break;
+ }
+ }
+ assert(msg_type != -1);
+
+ /* g0 header. */
+ src.nr--;
+
+ brw_SAMPLE(p,
+ retype(dst, BRW_REGISTER_TYPE_UW),
+ src.nr,
+ retype(brw_vec8_grf(0, 0), BRW_REGISTER_TYPE_UW),
+ SURF_INDEX_TEXTURE(inst->sampler),
+ inst->sampler,
+ WRITEMASK_XYZW,
+ msg_type,
+ rlen,
+ inst->mlen + 1,
+ 0,
+ 1,
+ BRW_SAMPLER_SIMD_MODE_SIMD8);
+}
+
+void
+fs_visitor::generate_discard(fs_inst *inst)
+{
+ struct brw_reg g0 = retype(brw_vec1_grf(0, 0), BRW_REGISTER_TYPE_UW);
+ brw_push_insn_state(p);
+ brw_set_mask_control(p, BRW_MASK_DISABLE);
+ brw_NOT(p, c->emit_mask_reg, brw_mask_reg(1)); /* IMASK */
+ brw_AND(p, g0, c->emit_mask_reg, g0);
+ brw_pop_insn_state(p);
+}
+
static void
trivial_assign_reg(int header_size, fs_reg *reg)
{
}
}
+void
+fs_visitor::assign_curb_setup()
+{
+ c->prog_data.first_curbe_grf = c->key.nr_payload_regs;
+ c->prog_data.curb_read_length = ALIGN(c->prog_data.nr_params, 8) / 8;
+
+ /* Map the offsets in the UNIFORM file to fixed HW regs. */
+ foreach_iter(exec_list_iterator, iter, this->instructions) {
+ fs_inst *inst = (fs_inst *)iter.get();
+
+ for (unsigned int i = 0; i < 3; i++) {
+ if (inst->src[i].file == UNIFORM) {
+ int constant_nr = inst->src[i].hw_reg + inst->src[i].reg_offset;
+ struct brw_reg brw_reg = brw_vec1_grf(c->prog_data.first_curbe_grf +
+ constant_nr / 8,
+ constant_nr % 8);
+
+ inst->src[i].file = FIXED_HW_REG;
+ inst->src[i].fixed_hw_reg = brw_reg;
+ }
+ }
+ }
+}
+
void
fs_visitor::assign_urb_setup()
{
- int urb_start = c->key.nr_payload_regs; /* FINISHME: push constants */
+ int urb_start = c->prog_data.first_curbe_grf + c->prog_data.curb_read_length;
int interp_reg_nr[FRAG_ATTRIB_MAX];
c->prog_data.urb_read_length = 0;
last_grf = MAX2(last_grf, inst->src[1].hw_reg);
}
- this->grf_used = last_grf;
+ this->grf_used = last_grf + 1;
}
static struct brw_reg brw_reg_from_fs_reg(fs_reg *reg)
brw_reg = brw_imm_f(reg->imm.f);
break;
case BRW_REGISTER_TYPE_D:
- brw_reg = brw_imm_f(reg->imm.i);
+ brw_reg = brw_imm_d(reg->imm.i);
break;
case BRW_REGISTER_TYPE_UD:
- brw_reg = brw_imm_f(reg->imm.u);
+ brw_reg = brw_imm_ud(reg->imm.u);
break;
default:
assert(!"not reached");
case BAD_FILE:
/* Probably unused. */
brw_reg = brw_null_reg();
+ break;
+ case UNIFORM:
+ assert(!"not reached");
+ brw_reg = brw_null_reg();
+ break;
}
if (reg->abs)
brw_reg = brw_abs(brw_reg);
void
fs_visitor::generate_code()
{
+ unsigned int annotation_len = 0;
+ int last_native_inst = 0;
+ struct brw_instruction *if_stack[16];
+ int if_stack_depth = 0;
+
+ memset(&if_stack, 0, sizeof(if_stack));
foreach_iter(exec_list_iterator, iter, this->instructions) {
fs_inst *inst = (fs_inst *)iter.get();
struct brw_reg src[3], dst;
case BRW_OPCODE_MUL:
brw_MUL(p, dst, src[0], src[1]);
break;
+
+ case BRW_OPCODE_FRC:
+ brw_FRC(p, dst, src[0]);
+ break;
+ case BRW_OPCODE_RNDD:
+ brw_RNDD(p, dst, src[0]);
+ break;
+ case BRW_OPCODE_RNDZ:
+ brw_RNDZ(p, dst, src[0]);
+ break;
+
+ case BRW_OPCODE_AND:
+ brw_AND(p, dst, src[0], src[1]);
+ break;
+ case BRW_OPCODE_OR:
+ brw_OR(p, dst, src[0], src[1]);
+ break;
+ case BRW_OPCODE_XOR:
+ brw_XOR(p, dst, src[0], src[1]);
+ break;
+
+ case BRW_OPCODE_CMP:
+ brw_CMP(p, dst, inst->conditional_mod, src[0], src[1]);
+ break;
+ case BRW_OPCODE_SEL:
+ brw_SEL(p, dst, src[0], src[1]);
+ break;
+
+ case BRW_OPCODE_IF:
+ assert(if_stack_depth < 16);
+ if_stack[if_stack_depth] = brw_IF(p, BRW_EXECUTE_8);
+ if_stack_depth++;
+ break;
+ case BRW_OPCODE_ELSE:
+ if_stack[if_stack_depth - 1] =
+ brw_ELSE(p, if_stack[if_stack_depth - 1]);
+ break;
+ case BRW_OPCODE_ENDIF:
+ if_stack_depth--;
+ brw_ENDIF(p , if_stack[if_stack_depth]);
+ break;
case FS_OPCODE_RCP:
case FS_OPCODE_RSQ:
case FS_OPCODE_SQRT:
case FS_OPCODE_LINTERP:
generate_linterp(inst, dst, src);
break;
+ case FS_OPCODE_TEX:
+ case FS_OPCODE_TXB:
+ case FS_OPCODE_TXL:
+ generate_tex(inst, dst, src[0]);
+ break;
+ case FS_OPCODE_DISCARD:
+ generate_discard(inst);
+ break;
case FS_OPCODE_FB_WRITE:
generate_fb_write(inst);
break;
default:
- assert(!"not reached");
+ if (inst->opcode < (int)ARRAY_SIZE(brw_opcodes)) {
+ _mesa_problem(ctx, "Unsupported opcode `%s' in FS",
+ brw_opcodes[inst->opcode].name);
+ } else {
+ _mesa_problem(ctx, "Unsupported opcode %d in FS", inst->opcode);
+ }
+ this->fail = true;
+ }
+
+ if (annotation_len < p->nr_insn) {
+ annotation_len *= 2;
+ if (annotation_len < 16)
+ annotation_len = 16;
+
+ this->annotation_string = talloc_realloc(this->mem_ctx,
+ annotation_string,
+ const char *,
+ annotation_len);
+ this->annotation_ir = talloc_realloc(this->mem_ctx,
+ annotation_ir,
+ ir_instruction *,
+ annotation_len);
+ }
+
+ for (unsigned int i = last_native_inst; i < p->nr_insn; i++) {
+ this->annotation_string[i] = inst->annotation;
+ this->annotation_ir[i] = inst->ir;
}
+ last_native_inst = p->nr_insn;
}
}
/* Generate FS IR for main(). (the visitor only descends into
* functions called "main").
*/
- visit_exec_list(shader->ir, &v);
+ foreach_iter(exec_list_iterator, iter, *shader->ir) {
+ ir_instruction *ir = (ir_instruction *)iter.get();
+ v.base_ir = ir;
+ ir->accept(&v);
+ }
if (v.fail)
return GL_FALSE;
v.emit_fb_writes();
+ v.assign_curb_setup();
v.assign_urb_setup();
v.assign_regs();
}
v.generate_code();
if (INTEL_DEBUG & DEBUG_WM) {
+ const char *last_annotation_string = NULL;
+ ir_instruction *last_annotation_ir = NULL;
+
printf("Native code for fragment shader %d:\n", prog->Name);
- for (unsigned int i = 0; i < p->nr_insn; i++)
+ for (unsigned int i = 0; i < p->nr_insn; i++) {
+ if (last_annotation_ir != v.annotation_ir[i]) {
+ last_annotation_ir = v.annotation_ir[i];
+ if (last_annotation_ir) {
+ printf(" ");
+ last_annotation_ir->print();
+ printf("\n");
+ }
+ }
+ if (last_annotation_string != v.annotation_string[i]) {
+ last_annotation_string = v.annotation_string[i];
+ if (last_annotation_string)
+ printf(" %s\n", last_annotation_string);
+ }
brw_disasm(stdout, &p->store[i], intel->gen);
+ }
printf("\n");
}
- c->prog_data.nr_params = 0; /* FINISHME */
- c->prog_data.first_curbe_grf = c->key.nr_payload_regs;
- c->prog_data.curb_read_length = 0; /* FINISHME */
c->prog_data.total_grf = v.grf_used;
c->prog_data.total_scratch = 0;
projects / mesa.git / blobdiff