/**
* This just eliminates instructions with null dst so far.
*/
static void
eliminate_dead_code(struct toy_compiler *tc)
{
struct toy_inst *inst;
tc_head(tc);
while ((inst = tc_next(tc)) != NULL) {
switch (inst->opcode) {
case GEN6_OPCODE_IF:
case GEN6_OPCODE_ELSE:
case GEN6_OPCODE_ENDIF:
case GEN6_OPCODE_WHILE:
case GEN6_OPCODE_BREAK:
case GEN6_OPCODE_CONT:
case GEN6_OPCODE_SEND:
case GEN6_OPCODE_SENDC:
case GEN6_OPCODE_NOP:
/* never eliminated */
break;
default:
if (tdst_is_null(inst->dst) || !inst->dst.writemask) {
/* math is always GEN6_COND_NORMAL */
if ((inst->opcode == GEN6_OPCODE_MATH ||
inst->cond_modifier == GEN6_COND_NORMAL) &&
!inst->acc_wr_ctrl)
tc_discard_inst(tc, inst);
}
break;
}
}
}
/**
* Compile the shader.
*/
static bool
gs_compile(struct gs_compile_context *gcc)
{
struct toy_compiler *tc = &gcc->tc;
struct ilo_shader *sh = gcc->shader;
get_num_prims_static(gcc);
if (gcc->is_static) {
tc_head(tc);
gs_init_vars(gcc);
gs_ff_sync(gcc, tdst_d(gcc->vars.tmp), tsrc_imm_d(gcc->static_data.total_prims));
gs_COPY1(tc, gcc->vars.urb_write_header, 0, tsrc_from(tdst_d(gcc->vars.tmp)), 0);
if (gcc->write_so)
gs_COPY4(tc, gcc->vars.so_index, 0, tsrc_from(tdst_d(gcc->vars.tmp)), 1);
tc_tail(tc);
}
else {
tc_fail(tc, "no control flow support");
return false;
}
if (!gcc->write_vue)
gs_discard(gcc);
gs_lower_virtual_opcodes(gcc);
toy_compiler_legalize_for_ra(tc);
toy_compiler_optimize(tc);
toy_compiler_allocate_registers(tc,
gcc->first_free_grf,
gcc->last_free_grf,
1);
toy_compiler_legalize_for_asm(tc);
if (tc->fail) {
ilo_err("failed to legalize GS instructions: %s\n", tc->reason);
return false;
}
if (ilo_debug & ILO_DEBUG_GS) {
ilo_printf("legalized instructions:\n");
toy_compiler_dump(tc);
ilo_printf("\n");
}
sh->kernel = toy_compiler_assemble(tc, &sh->kernel_size);
if (!sh->kernel)
return false;
if (ilo_debug & ILO_DEBUG_GS) {
ilo_printf("disassembly:\n");
toy_compiler_disassemble(tc->dev, sh->kernel, sh->kernel_size, false);
ilo_printf("\n");
}
return true;
}
/**
* Get the number of (tessellated) primitives generated by this shader.
* Return false if that is unknown until runtime.
*/
static void
get_num_prims_static(struct gs_compile_context *gcc)
{
struct toy_compiler *tc = &gcc->tc;
const struct toy_inst *inst;
int num_vertices_in_prim = 0, if_depth = 0, do_depth = 0;
bool is_static = true;
tc_head(tc);
while ((inst = tc_next_no_skip(tc)) != NULL) {
switch (inst->opcode) {
case GEN6_OPCODE_IF:
if_depth++;
break;
case GEN6_OPCODE_ENDIF:
if_depth--;
break;
case TOY_OPCODE_DO:
do_depth++;
break;
case GEN6_OPCODE_WHILE:
do_depth--;
break;
case TOY_OPCODE_EMIT:
if (if_depth || do_depth) {
is_static = false;
}
else {
gcc->static_data.total_vertices++;
num_vertices_in_prim++;
if (num_vertices_in_prim >= gcc->out_vue_min_count)
gcc->static_data.total_prims++;
}
break;
case TOY_OPCODE_ENDPRIM:
if (if_depth || do_depth) {
is_static = false;
}
else {
const int vertidx = gcc->static_data.total_vertices - 1;
const int idx = vertidx / 32;
const int subidx = vertidx % 32;
gcc->static_data.last_vertex[idx] |= 1 << subidx;
num_vertices_in_prim = 0;
}
break;
default:
break;
}
if (!is_static)
break;
}
gcc->is_static = is_static;
}
/**
* Dump the instructions added to the compiler.
*/
void
toy_compiler_dump(struct toy_compiler *tc)
{
struct toy_inst *inst;
int pc;
pc = 0;
tc_head(tc);
while ((inst = tc_next_no_skip(tc)) != NULL) {
/* we do not generate code for markers */
if (inst->marker)
ilo_printf("marker:");
else
ilo_printf("%6d:", pc++);
tc_dump_inst(tc, inst);
}
}
static void
gs_lower_virtual_opcodes(struct gs_compile_context *gcc)
{
struct toy_compiler *tc = &gcc->tc;
struct toy_inst *inst;
tc_head(tc);
while ((inst = tc_next(tc)) != NULL) {
switch (inst->opcode) {
case TOY_OPCODE_TGSI_IN:
case TOY_OPCODE_TGSI_CONST:
case TOY_OPCODE_TGSI_SV:
case TOY_OPCODE_TGSI_IMM:
gs_lower_opcode_tgsi_direct(gcc, inst);
break;
case TOY_OPCODE_TGSI_INDIRECT_FETCH:
case TOY_OPCODE_TGSI_INDIRECT_STORE:
/* TODO similar to VS */
tc_fail(tc, "no indirection support");
tc_discard_inst(tc, inst);
break;
case TOY_OPCODE_TGSI_TEX:
case TOY_OPCODE_TGSI_TXB:
case TOY_OPCODE_TGSI_TXD:
case TOY_OPCODE_TGSI_TXL:
case TOY_OPCODE_TGSI_TXP:
case TOY_OPCODE_TGSI_TXF:
case TOY_OPCODE_TGSI_TXQ:
case TOY_OPCODE_TGSI_TXQ_LZ:
case TOY_OPCODE_TGSI_TEX2:
case TOY_OPCODE_TGSI_TXB2:
case TOY_OPCODE_TGSI_TXL2:
case TOY_OPCODE_TGSI_SAMPLE:
case TOY_OPCODE_TGSI_SAMPLE_I:
case TOY_OPCODE_TGSI_SAMPLE_I_MS:
case TOY_OPCODE_TGSI_SAMPLE_B:
case TOY_OPCODE_TGSI_SAMPLE_C:
case TOY_OPCODE_TGSI_SAMPLE_C_LZ:
case TOY_OPCODE_TGSI_SAMPLE_D:
case TOY_OPCODE_TGSI_SAMPLE_L:
case TOY_OPCODE_TGSI_GATHER4:
case TOY_OPCODE_TGSI_SVIEWINFO:
case TOY_OPCODE_TGSI_SAMPLE_POS:
case TOY_OPCODE_TGSI_SAMPLE_INFO:
/* TODO similar to VS */
tc_fail(tc, "no sampling support");
tc_discard_inst(tc, inst);
break;
case TOY_OPCODE_EMIT:
gs_lower_opcode_emit(gcc, inst);
tc_discard_inst(tc, inst);
break;
case TOY_OPCODE_ENDPRIM:
gs_lower_opcode_endprim(gcc, inst);
tc_discard_inst(tc, inst);
break;
default:
break;
}
}
tc_head(tc);
while ((inst = tc_next(tc)) != NULL) {
switch (inst->opcode) {
case TOY_OPCODE_INV:
case TOY_OPCODE_LOG:
case TOY_OPCODE_EXP:
case TOY_OPCODE_SQRT:
case TOY_OPCODE_RSQ:
case TOY_OPCODE_SIN:
case TOY_OPCODE_COS:
case TOY_OPCODE_FDIV:
case TOY_OPCODE_POW:
case TOY_OPCODE_INT_DIV_QUOTIENT:
case TOY_OPCODE_INT_DIV_REMAINDER:
toy_compiler_lower_math(tc, inst);
break;
case TOY_OPCODE_URB_WRITE:
toy_compiler_lower_to_send(tc, inst, false, GEN6_SFID_URB);
break;
default:
if (inst->opcode > 127)
tc_fail(tc, "unhandled virtual opcode");
break;
}
}
}
static void
vs_lower_virtual_opcodes(struct vs_compile_context *vcc)
{
struct toy_compiler *tc = &vcc->tc;
struct toy_inst *inst;
tc_head(tc);
while ((inst = tc_next(tc)) != NULL) {
switch (inst->opcode) {
case TOY_OPCODE_TGSI_IN:
case TOY_OPCODE_TGSI_CONST:
case TOY_OPCODE_TGSI_SV:
case TOY_OPCODE_TGSI_IMM:
vs_lower_opcode_tgsi_direct(vcc, inst);
break;
case TOY_OPCODE_TGSI_INDIRECT_FETCH:
case TOY_OPCODE_TGSI_INDIRECT_STORE:
vs_lower_opcode_tgsi_indirect(vcc, inst);
break;
case TOY_OPCODE_TGSI_TEX:
case TOY_OPCODE_TGSI_TXB:
case TOY_OPCODE_TGSI_TXD:
case TOY_OPCODE_TGSI_TXL:
case TOY_OPCODE_TGSI_TXP:
case TOY_OPCODE_TGSI_TXF:
case TOY_OPCODE_TGSI_TXQ:
case TOY_OPCODE_TGSI_TXQ_LZ:
case TOY_OPCODE_TGSI_TEX2:
case TOY_OPCODE_TGSI_TXB2:
case TOY_OPCODE_TGSI_TXL2:
case TOY_OPCODE_TGSI_SAMPLE:
case TOY_OPCODE_TGSI_SAMPLE_I:
case TOY_OPCODE_TGSI_SAMPLE_I_MS:
case TOY_OPCODE_TGSI_SAMPLE_B:
case TOY_OPCODE_TGSI_SAMPLE_C:
case TOY_OPCODE_TGSI_SAMPLE_C_LZ:
case TOY_OPCODE_TGSI_SAMPLE_D:
case TOY_OPCODE_TGSI_SAMPLE_L:
case TOY_OPCODE_TGSI_GATHER4:
case TOY_OPCODE_TGSI_SVIEWINFO:
case TOY_OPCODE_TGSI_SAMPLE_POS:
case TOY_OPCODE_TGSI_SAMPLE_INFO:
vs_lower_opcode_tgsi_sampling(vcc, inst);
break;
case TOY_OPCODE_INV:
case TOY_OPCODE_LOG:
case TOY_OPCODE_EXP:
case TOY_OPCODE_SQRT:
case TOY_OPCODE_RSQ:
case TOY_OPCODE_SIN:
case TOY_OPCODE_COS:
case TOY_OPCODE_FDIV:
case TOY_OPCODE_POW:
case TOY_OPCODE_INT_DIV_QUOTIENT:
case TOY_OPCODE_INT_DIV_REMAINDER:
toy_compiler_lower_math(tc, inst);
break;
case TOY_OPCODE_URB_WRITE:
vs_lower_opcode_urb_write(tc, inst);
break;
default:
if (inst->opcode > 127)
tc_fail(tc, "unhandled virtual opcode");
break;
}
}
}
static void
fs_lower_virtual_opcodes(struct fs_compile_context *fcc)
{
struct toy_compiler *tc = &fcc->tc;
struct toy_inst *inst;
/* lower TGSI's first, as they might be lowered to other virtual opcodes */
tc_head(tc);
while ((inst = tc_next(tc)) != NULL) {
switch (inst->opcode) {
case TOY_OPCODE_TGSI_IN:
case TOY_OPCODE_TGSI_CONST:
case TOY_OPCODE_TGSI_SV:
case TOY_OPCODE_TGSI_IMM:
fs_lower_opcode_tgsi_direct(fcc, inst);
break;
case TOY_OPCODE_TGSI_INDIRECT_FETCH:
case TOY_OPCODE_TGSI_INDIRECT_STORE:
fs_lower_opcode_tgsi_indirect(fcc, inst);
break;
case TOY_OPCODE_TGSI_TEX:
case TOY_OPCODE_TGSI_TXB:
case TOY_OPCODE_TGSI_TXD:
case TOY_OPCODE_TGSI_TXL:
case TOY_OPCODE_TGSI_TXP:
case TOY_OPCODE_TGSI_TXF:
case TOY_OPCODE_TGSI_TXQ:
case TOY_OPCODE_TGSI_TXQ_LZ:
case TOY_OPCODE_TGSI_TEX2:
case TOY_OPCODE_TGSI_TXB2:
case TOY_OPCODE_TGSI_TXL2:
case TOY_OPCODE_TGSI_SAMPLE:
case TOY_OPCODE_TGSI_SAMPLE_I:
case TOY_OPCODE_TGSI_SAMPLE_I_MS:
case TOY_OPCODE_TGSI_SAMPLE_B:
case TOY_OPCODE_TGSI_SAMPLE_C:
case TOY_OPCODE_TGSI_SAMPLE_C_LZ:
case TOY_OPCODE_TGSI_SAMPLE_D:
case TOY_OPCODE_TGSI_SAMPLE_L:
case TOY_OPCODE_TGSI_GATHER4:
case TOY_OPCODE_TGSI_SVIEWINFO:
case TOY_OPCODE_TGSI_SAMPLE_POS:
case TOY_OPCODE_TGSI_SAMPLE_INFO:
fs_lower_opcode_tgsi_sampling(fcc, inst);
break;
}
}
tc_head(tc);
while ((inst = tc_next(tc)) != NULL) {
switch (inst->opcode) {
case TOY_OPCODE_INV:
case TOY_OPCODE_LOG:
case TOY_OPCODE_EXP:
case TOY_OPCODE_SQRT:
case TOY_OPCODE_RSQ:
case TOY_OPCODE_SIN:
case TOY_OPCODE_COS:
case TOY_OPCODE_FDIV:
case TOY_OPCODE_POW:
case TOY_OPCODE_INT_DIV_QUOTIENT:
case TOY_OPCODE_INT_DIV_REMAINDER:
toy_compiler_lower_math(tc, inst);
break;
case TOY_OPCODE_DDX:
case TOY_OPCODE_DDY:
fs_lower_opcode_derivative(tc, inst);
break;
case TOY_OPCODE_FB_WRITE:
fs_lower_opcode_fb_write(tc, inst);
break;
case TOY_OPCODE_KIL:
fs_lower_opcode_kil(tc, inst);
break;
default:
if (inst->opcode > 127)
tc_fail(tc, "unhandled virtual opcode");
break;
}
}
}
