Files
Xenia-Canary/src/xenia/gpu/spirv_shader_translator.cc
2020-10-25 20:24:48 +03:00

862 lines
36 KiB
C++

/**
******************************************************************************
* Xenia : Xbox 360 Emulator Research Project *
******************************************************************************
* Copyright 2020 Ben Vanik. All rights reserved. *
* Released under the BSD license - see LICENSE in the root for more details. *
******************************************************************************
*/
#include "xenia/gpu/spirv_shader_translator.h"
#include <memory>
#include <utility>
#include <vector>
#include "third_party/glslang/SPIRV/GLSL.std.450.h"
#include "xenia/base/assert.h"
namespace xe {
namespace gpu {
SpirvShaderTranslator::SpirvShaderTranslator(bool supports_clip_distance,
bool supports_cull_distance)
: supports_clip_distance_(supports_clip_distance),
supports_cull_distance_(supports_cull_distance) {}
void SpirvShaderTranslator::Reset() {
ShaderTranslator::Reset();
builder_.reset();
main_switch_op_.reset();
main_switch_next_pc_phi_operands_.clear();
cf_exec_conditional_merge_ = nullptr;
cf_instruction_predicate_merge_ = nullptr;
}
void SpirvShaderTranslator::StartTranslation() {
// Tool ID 26 "Xenia Emulator Microcode Translator".
// https://github.com/KhronosGroup/SPIRV-Headers/blob/c43a43c7cc3af55910b9bec2a71e3e8a622443cf/include/spirv/spir-v.xml#L79
// TODO(Triang3l): Logger.
builder_ = std::make_unique<spv::Builder>(1 << 16, (26 << 16) | 1, nullptr);
builder_->addCapability(IsSpirvTessEvalShader() ? spv::CapabilityTessellation
: spv::CapabilityShader);
ext_inst_glsl_std_450_ = builder_->import("GLSL.std.450");
builder_->setMemoryModel(spv::AddressingModelLogical,
spv::MemoryModelGLSL450);
builder_->setSource(spv::SourceLanguageUnknown, 0);
type_void_ = builder_->makeVoidType();
type_bool_ = builder_->makeBoolType();
type_int_ = builder_->makeIntType(32);
type_int4_ = builder_->makeVectorType(type_int_, 4);
type_uint_ = builder_->makeUintType(32);
type_uint3_ = builder_->makeVectorType(type_uint_, 3);
type_uint4_ = builder_->makeVectorType(type_uint_, 4);
type_float_ = builder_->makeFloatType(32);
type_float2_ = builder_->makeVectorType(type_float_, 2);
type_float3_ = builder_->makeVectorType(type_float_, 3);
type_float4_ = builder_->makeVectorType(type_float_, 4);
const_int_0_ = builder_->makeIntConstant(0);
id_vector_temp_.clear();
id_vector_temp_.reserve(4);
for (uint32_t i = 0; i < 4; ++i) {
id_vector_temp_.push_back(const_int_0_);
}
const_int4_0_ = builder_->makeCompositeConstant(type_int4_, id_vector_temp_);
const_uint_0_ = builder_->makeUintConstant(0);
id_vector_temp_.clear();
id_vector_temp_.reserve(4);
for (uint32_t i = 0; i < 4; ++i) {
id_vector_temp_.push_back(const_uint_0_);
}
const_uint4_0_ =
builder_->makeCompositeConstant(type_uint4_, id_vector_temp_);
const_float_0_ = builder_->makeFloatConstant(0.0f);
id_vector_temp_.clear();
id_vector_temp_.reserve(4);
for (uint32_t i = 0; i < 4; ++i) {
id_vector_temp_.push_back(const_float_0_);
}
const_float4_0_ =
builder_->makeCompositeConstant(type_float4_, id_vector_temp_);
// Common uniform buffer - bool and loop constants.
id_vector_temp_.clear();
id_vector_temp_.reserve(2);
// 256 bool constants.
id_vector_temp_.push_back(builder_->makeArrayType(
type_uint4_, builder_->makeUintConstant(2), sizeof(uint32_t) * 4));
// Currently (as of October 24, 2020) makeArrayType only uses the stride to
// check if deduplication can be done - the array stride decoration needs to
// be applied explicitly.
builder_->addDecoration(id_vector_temp_.back(), spv::DecorationArrayStride,
sizeof(uint32_t) * 4);
// 32 loop constants.
id_vector_temp_.push_back(builder_->makeArrayType(
type_uint4_, builder_->makeUintConstant(8), sizeof(uint32_t) * 4));
builder_->addDecoration(id_vector_temp_.back(), spv::DecorationArrayStride,
sizeof(uint32_t) * 4);
spv::Id type_bool_loop_constants =
builder_->makeStructType(id_vector_temp_, "XeBoolLoopConstants");
builder_->addMemberName(type_bool_loop_constants, 0, "bool_constants");
builder_->addMemberDecoration(type_bool_loop_constants, 0,
spv::DecorationOffset, 0);
builder_->addMemberName(type_bool_loop_constants, 1, "loop_constants");
builder_->addMemberDecoration(type_bool_loop_constants, 1,
spv::DecorationOffset, sizeof(uint32_t) * 8);
builder_->addDecoration(type_bool_loop_constants, spv::DecorationBlock);
uniform_bool_loop_constants_ = builder_->createVariable(
spv::NoPrecision, spv::StorageClassUniform, type_bool_loop_constants,
"xe_uniform_bool_loop_constants");
builder_->addDecoration(uniform_bool_loop_constants_,
spv::DecorationDescriptorSet,
int(kDescriptorSetBoolLoopConstants));
builder_->addDecoration(uniform_bool_loop_constants_, spv::DecorationBinding,
0);
if (IsSpirvVertexOrTessEvalShader()) {
StartVertexOrTessEvalShaderBeforeMain();
}
// Begin the main function.
std::vector<spv::Id> main_param_types;
std::vector<std::vector<spv::Decoration>> main_precisions;
spv::Block* function_main_entry;
function_main_ = builder_->makeFunctionEntry(
spv::NoPrecision, type_void_, "main", main_param_types, main_precisions,
&function_main_entry);
// Begin ucode translation. Initialize everything, even without defined
// defaults, for safety.
var_main_predicate_ = builder_->createVariable(
spv::NoPrecision, spv::StorageClassFunction, type_bool_,
"xe_var_predicate", builder_->makeBoolConstant(false));
var_main_loop_count_ = builder_->createVariable(
spv::NoPrecision, spv::StorageClassFunction, type_uint4_,
"xe_var_loop_count", const_uint4_0_);
var_main_address_absolute_ = builder_->createVariable(
spv::NoPrecision, spv::StorageClassFunction, type_int_,
"xe_var_address_absolute", const_int_0_);
var_main_address_relative_ = builder_->createVariable(
spv::NoPrecision, spv::StorageClassFunction, type_int4_,
"xe_var_address_relative", const_int4_0_);
uint32_t register_array_size = register_count();
if (register_array_size) {
id_vector_temp_.clear();
id_vector_temp_.reserve(register_array_size);
// TODO(Triang3l): In PS, only initialize starting from the interpolators,
// probably manually. But not very important.
for (uint32_t i = 0; i < register_array_size; ++i) {
id_vector_temp_.push_back(const_float4_0_);
}
spv::Id type_register_array = builder_->makeArrayType(
type_float4_, builder_->makeUintConstant(register_array_size), 0);
var_main_registers_ = builder_->createVariable(
spv::NoPrecision, spv::StorageClassFunction, type_register_array,
"xe_var_registers",
builder_->makeCompositeConstant(type_register_array, id_vector_temp_));
}
// Write the execution model-specific prologue with access to variables in the
// main function.
if (IsSpirvVertexOrTessEvalShader()) {
StartVertexOrTessEvalShaderInMain();
}
// Open the main loop.
spv::Block& main_loop_pre_header = *builder_->getBuildPoint();
main_loop_header_ = &builder_->makeNewBlock();
spv::Block& main_loop_body = builder_->makeNewBlock();
// Added later because the body has nested control flow, but according to the
// specification:
// "The order of blocks in a function must satisfy the rule that blocks appear
// before all blocks they dominate."
main_loop_continue_ =
new spv::Block(builder_->getUniqueId(), *function_main_);
main_loop_merge_ = new spv::Block(builder_->getUniqueId(), *function_main_);
builder_->createBranch(main_loop_header_);
// If no jumps, don't create a switch, but still create a loop so exece can
// break.
bool has_main_switch = !label_addresses().empty();
// Main loop header - based on whether it's the first iteration (entered from
// the function or from the continuation), choose the program counter.
builder_->setBuildPoint(main_loop_header_);
spv::Id main_loop_pc_current = 0;
if (has_main_switch) {
// OpPhi must be the first in the block.
id_vector_temp_.clear();
id_vector_temp_.reserve(4);
id_vector_temp_.push_back(const_int_0_);
id_vector_temp_.push_back(main_loop_pre_header.getId());
main_loop_pc_next_ = builder_->getUniqueId();
id_vector_temp_.push_back(main_loop_pc_next_);
id_vector_temp_.push_back(main_loop_continue_->getId());
main_loop_pc_current =
builder_->createOp(spv::OpPhi, type_int_, id_vector_temp_);
}
uint_vector_temp_.clear();
builder_->createLoopMerge(main_loop_merge_, main_loop_continue_,
spv::LoopControlDontUnrollMask, uint_vector_temp_);
builder_->createBranch(&main_loop_body);
// Main loop body.
builder_->setBuildPoint(&main_loop_body);
if (has_main_switch) {
// Create the program counter switch with cases for every label and for
// label 0.
main_switch_header_ = builder_->getBuildPoint();
main_switch_merge_ =
new spv::Block(builder_->getUniqueId(), *function_main_);
{
std::unique_ptr<spv::Instruction> main_switch_selection_merge_op =
std::make_unique<spv::Instruction>(spv::OpSelectionMerge);
main_switch_selection_merge_op->addIdOperand(main_switch_merge_->getId());
main_switch_selection_merge_op->addImmediateOperand(
spv::SelectionControlDontFlattenMask);
builder_->getBuildPoint()->addInstruction(
std::move(main_switch_selection_merge_op));
}
main_switch_op_ = std::make_unique<spv::Instruction>(spv::OpSwitch);
main_switch_op_->addIdOperand(main_loop_pc_current);
main_switch_op_->addIdOperand(main_switch_merge_->getId());
// The default case (the merge here) must have the header as a predecessor.
main_switch_merge_->addPredecessor(main_switch_header_);
// The instruction will be inserted later, when all cases are filled.
// Insert and enter case 0.
spv::Block* main_switch_case_0_block =
new spv::Block(builder_->getUniqueId(), *function_main_);
main_switch_op_->addImmediateOperand(0);
main_switch_op_->addIdOperand(main_switch_case_0_block->getId());
// Every switch case must have the OpSelectionMerge/OpSwitch block as a
// predecessor.
main_switch_case_0_block->addPredecessor(main_switch_header_);
function_main_->addBlock(main_switch_case_0_block);
builder_->setBuildPoint(main_switch_case_0_block);
}
}
std::vector<uint8_t> SpirvShaderTranslator::CompleteTranslation() {
// Close flow control within the last switch case.
CloseExecConditionals();
bool has_main_switch = !label_addresses().empty();
// After the final exec (if it happened to be not exece, which would already
// have a break branch), break from the switch if it exists, or from the
// loop it doesn't.
if (!builder_->getBuildPoint()->isTerminated()) {
builder_->createBranch(has_main_switch ? main_switch_merge_
: main_loop_merge_);
}
if (has_main_switch) {
// Insert the switch instruction with all cases added as operands.
builder_->setBuildPoint(main_switch_header_);
builder_->getBuildPoint()->addInstruction(std::move(main_switch_op_));
// Build the main switch merge, breaking out of the loop after falling
// through the end or breaking from exece (only continuing if a jump - from
// a guest loop or from jmp/call - was made).
function_main_->addBlock(main_switch_merge_);
builder_->setBuildPoint(main_switch_merge_);
builder_->createBranch(main_loop_merge_);
}
// Main loop continuation - choose the program counter based on the path
// taken (-1 if not from a jump as a safe fallback, which would result in not
// hitting any switch case and reaching the final break in the body).
function_main_->addBlock(main_loop_continue_);
builder_->setBuildPoint(main_loop_continue_);
if (has_main_switch) {
// OpPhi, if added, must be the first in the block.
// If labels were added, but not jumps (for example, due to the call
// instruction not being implemented as of October 18, 2020), send an
// impossible program counter value (-1) to the OpPhi at the next iteration.
if (main_switch_next_pc_phi_operands_.empty()) {
main_switch_next_pc_phi_operands_.push_back(
builder_->makeIntConstant(-1));
}
std::unique_ptr<spv::Instruction> main_loop_pc_next_op =
std::make_unique<spv::Instruction>(
main_loop_pc_next_, type_int_,
main_switch_next_pc_phi_operands_.size() >= 2 ? spv::OpPhi
: spv::OpCopyObject);
for (spv::Id operand : main_switch_next_pc_phi_operands_) {
main_loop_pc_next_op->addIdOperand(operand);
}
builder_->getBuildPoint()->addInstruction(std::move(main_loop_pc_next_op));
}
builder_->createBranch(main_loop_header_);
// Add the main loop merge block and go back to the function.
function_main_->addBlock(main_loop_merge_);
builder_->setBuildPoint(main_loop_merge_);
if (IsSpirvVertexOrTessEvalShader()) {
CompleteVertexOrTessEvalShaderInMain();
}
// End the main function.
builder_->leaveFunction();
// Make the main function the entry point.
spv::ExecutionModel execution_model;
if (IsSpirvFragmentShader()) {
execution_model = spv::ExecutionModelFragment;
builder_->addExecutionMode(function_main_,
spv::ExecutionModeOriginUpperLeft);
} else {
assert_true(IsSpirvVertexOrTessEvalShader());
execution_model = IsSpirvTessEvalShader()
? spv::ExecutionModelTessellationEvaluation
: spv::ExecutionModelVertex;
}
spv::Instruction* entry_point =
builder_->addEntryPoint(execution_model, function_main_, "main");
if (IsSpirvVertexOrTessEvalShader()) {
CompleteVertexOrTessEvalShaderAfterMain(entry_point);
}
// TODO(Triang3l): Avoid copy?
std::vector<unsigned int> module_uints;
builder_->dump(module_uints);
std::vector<uint8_t> module_bytes;
module_bytes.reserve(sizeof(unsigned int) * module_uints.size());
module_bytes.insert(module_bytes.cend(),
reinterpret_cast<const uint8_t*>(module_uints.data()),
reinterpret_cast<const uint8_t*>(module_uints.data()) +
sizeof(unsigned int) * module_uints.size());
return module_bytes;
}
void SpirvShaderTranslator::ProcessLabel(uint32_t cf_index) {
if (cf_index == 0) {
// 0 already added in the beginning.
return;
}
assert_false(label_addresses().empty());
// Close flow control within the previous switch case.
CloseExecConditionals();
spv::Function& function = builder_->getBuildPoint()->getParent();
// Create the next switch case and fallthrough to it.
spv::Block* new_case = new spv::Block(builder_->getUniqueId(), function);
main_switch_op_->addImmediateOperand(cf_index);
main_switch_op_->addIdOperand(new_case->getId());
// Every switch case must have the OpSelectionMerge/OpSwitch block as a
// predecessor.
new_case->addPredecessor(main_switch_header_);
// The previous block may have already been terminated if was exece.
if (!builder_->getBuildPoint()->isTerminated()) {
builder_->createBranch(new_case);
}
function.addBlock(new_case);
builder_->setBuildPoint(new_case);
}
void SpirvShaderTranslator::ProcessExecInstructionBegin(
const ParsedExecInstruction& instr) {
UpdateExecConditionals(instr.type, instr.bool_constant_index,
instr.condition);
}
void SpirvShaderTranslator::ProcessExecInstructionEnd(
const ParsedExecInstruction& instr) {
if (instr.is_end) {
// Break out of the main switch (if exists) and the main loop.
CloseInstructionPredication();
if (!builder_->getBuildPoint()->isTerminated()) {
builder_->createBranch(label_addresses().empty() ? main_loop_merge_
: main_switch_merge_);
}
}
UpdateExecConditionals(instr.type, instr.bool_constant_index,
instr.condition);
}
void SpirvShaderTranslator::ProcessLoopStartInstruction(
const ParsedLoopStartInstruction& instr) {
// loop il<idx>, L<idx> - loop with loop data il<idx>, end @ L<idx>
// Loop control is outside execs - actually close the last exec.
CloseExecConditionals();
EnsureBuildPointAvailable();
id_vector_temp_.clear();
id_vector_temp_.reserve(3);
// Loop constants (member 1).
id_vector_temp_.push_back(builder_->makeIntConstant(1));
// 4-component vector.
id_vector_temp_.push_back(
builder_->makeIntConstant(int(instr.loop_constant_index >> 2)));
// Scalar within the vector.
id_vector_temp_.push_back(
builder_->makeIntConstant(int(instr.loop_constant_index & 3)));
// Count (unsigned) in bits 0:7 of the loop constant (struct member 1),
// initial aL (unsigned) in 8:15.
spv::Id loop_constant =
builder_->createLoad(builder_->createAccessChain(
spv::StorageClassUniform,
uniform_bool_loop_constants_, id_vector_temp_),
spv::NoPrecision);
spv::Id const_int_8 = builder_->makeIntConstant(8);
// Push the count to the loop count stack - move XYZ to YZW and set X to the
// new iteration count (swizzling the way glslang does it for similar GLSL).
spv::Id loop_count_stack_old =
builder_->createLoad(var_main_loop_count_, spv::NoPrecision);
spv::Id loop_count_new =
builder_->createTriOp(spv::OpBitFieldUExtract, type_uint_, loop_constant,
const_int_0_, const_int_8);
id_vector_temp_.clear();
id_vector_temp_.reserve(4);
id_vector_temp_.push_back(loop_count_new);
for (unsigned int i = 0; i < 3; ++i) {
id_vector_temp_.push_back(
builder_->createCompositeExtract(loop_count_stack_old, type_uint_, i));
}
builder_->createStore(
builder_->createCompositeConstruct(type_uint4_, id_vector_temp_),
var_main_loop_count_);
// Push aL - keep the same value as in the previous loop if repeating, or the
// new one otherwise.
spv::Id address_relative_stack_old =
builder_->createLoad(var_main_address_relative_, spv::NoPrecision);
id_vector_temp_.clear();
id_vector_temp_.reserve(4);
if (instr.is_repeat) {
id_vector_temp_.emplace_back();
} else {
id_vector_temp_.push_back(builder_->createUnaryOp(
spv::OpBitcast, type_int_,
builder_->createTriOp(spv::OpBitFieldUExtract, type_uint_,
loop_constant, const_int_8, const_int_8)));
}
for (unsigned int i = 0; i < 3; ++i) {
id_vector_temp_.push_back(builder_->createCompositeExtract(
address_relative_stack_old, type_int_, i));
}
if (instr.is_repeat) {
id_vector_temp_[0] = id_vector_temp_[1];
}
builder_->createStore(
builder_->createCompositeConstruct(type_int4_, id_vector_temp_),
var_main_address_relative_);
// Break (jump to the skip label) if the loop counter is 0 (since the
// condition is checked in the end).
spv::Block& head_block = *builder_->getBuildPoint();
spv::Id loop_count_zero = builder_->createBinOp(
spv::OpIEqual, type_bool_, loop_count_new, const_uint_0_);
spv::Block& skip_block = builder_->makeNewBlock();
spv::Block& body_block = builder_->makeNewBlock();
{
std::unique_ptr<spv::Instruction> selection_merge_op =
std::make_unique<spv::Instruction>(spv::OpSelectionMerge);
selection_merge_op->addIdOperand(body_block.getId());
selection_merge_op->addImmediateOperand(spv::SelectionControlMaskNone);
head_block.addInstruction(std::move(selection_merge_op));
}
{
std::unique_ptr<spv::Instruction> branch_conditional_op =
std::make_unique<spv::Instruction>(spv::OpBranchConditional);
branch_conditional_op->addIdOperand(loop_count_zero);
branch_conditional_op->addIdOperand(skip_block.getId());
branch_conditional_op->addIdOperand(body_block.getId());
// More likely to enter than to skip.
branch_conditional_op->addImmediateOperand(1);
branch_conditional_op->addImmediateOperand(2);
head_block.addInstruction(std::move(branch_conditional_op));
}
skip_block.addPredecessor(&head_block);
body_block.addPredecessor(&head_block);
builder_->setBuildPoint(&skip_block);
main_switch_next_pc_phi_operands_.push_back(
builder_->makeIntConstant(int(instr.loop_skip_address)));
main_switch_next_pc_phi_operands_.push_back(
builder_->getBuildPoint()->getId());
builder_->createBranch(main_loop_continue_);
builder_->setBuildPoint(&body_block);
}
void SpirvShaderTranslator::ProcessLoopEndInstruction(
const ParsedLoopEndInstruction& instr) {
// endloop il<idx>, L<idx> - end loop w/ data il<idx>, head @ L<idx>
// Loop control is outside execs - actually close the last exec.
CloseExecConditionals();
EnsureBuildPointAvailable();
// Subtract 1 from the loop counter (will store later).
spv::Id loop_count_stack_old =
builder_->createLoad(var_main_loop_count_, spv::NoPrecision);
spv::Id loop_count = builder_->createBinOp(
spv::OpISub, type_uint_,
builder_->createCompositeExtract(loop_count_stack_old, type_uint_, 0),
builder_->makeUintConstant(1));
spv::Id address_relative_stack_old =
builder_->createLoad(var_main_address_relative_, spv::NoPrecision);
// Predicated break works like break if (loop_count == 0 || [!]p0).
// Three options, due to logical operations usage (so OpLogicalNot is not
// required):
// - Continue if (loop_count != 0).
// - Continue if (loop_count != 0 && p0), if breaking if !p0.
// - Break if (loop_count == 0 || p0), if breaking if p0.
bool break_is_true = instr.is_predicated_break && instr.predicate_condition;
spv::Id condition =
builder_->createBinOp(break_is_true ? spv::OpIEqual : spv::OpINotEqual,
type_bool_, loop_count, const_uint_0_);
if (instr.is_predicated_break) {
condition = builder_->createBinOp(
instr.predicate_condition ? spv::OpLogicalOr : spv::OpLogicalAnd,
type_bool_, condition,
builder_->createLoad(var_main_predicate_, spv::NoPrecision));
}
spv::Block& body_block = *builder_->getBuildPoint();
spv::Block& continue_block = builder_->makeNewBlock();
spv::Block& break_block = builder_->makeNewBlock();
{
std::unique_ptr<spv::Instruction> selection_merge_op =
std::make_unique<spv::Instruction>(spv::OpSelectionMerge);
selection_merge_op->addIdOperand(break_block.getId());
selection_merge_op->addImmediateOperand(spv::SelectionControlMaskNone);
body_block.addInstruction(std::move(selection_merge_op));
}
{
std::unique_ptr<spv::Instruction> branch_conditional_op =
std::make_unique<spv::Instruction>(spv::OpBranchConditional);
branch_conditional_op->addIdOperand(condition);
// More likely to continue than to break.
if (break_is_true) {
branch_conditional_op->addIdOperand(break_block.getId());
branch_conditional_op->addIdOperand(continue_block.getId());
branch_conditional_op->addImmediateOperand(1);
branch_conditional_op->addImmediateOperand(2);
} else {
branch_conditional_op->addIdOperand(continue_block.getId());
branch_conditional_op->addIdOperand(break_block.getId());
branch_conditional_op->addImmediateOperand(2);
branch_conditional_op->addImmediateOperand(1);
}
body_block.addInstruction(std::move(branch_conditional_op));
}
continue_block.addPredecessor(&body_block);
break_block.addPredecessor(&body_block);
// Continue case.
builder_->setBuildPoint(&continue_block);
// Store the loop count with 1 subtracted.
builder_->createStore(builder_->createCompositeInsert(
loop_count, loop_count_stack_old, type_uint4_, 0),
var_main_loop_count_);
// Extract the value to add to aL (signed, in bits 16:23 of the loop
// constant).
id_vector_temp_.clear();
id_vector_temp_.reserve(3);
// Loop constants (member 1).
id_vector_temp_.push_back(builder_->makeIntConstant(1));
// 4-component vector.
id_vector_temp_.push_back(
builder_->makeIntConstant(int(instr.loop_constant_index >> 2)));
// Scalar within the vector.
id_vector_temp_.push_back(
builder_->makeIntConstant(int(instr.loop_constant_index & 3)));
spv::Id loop_constant =
builder_->createLoad(builder_->createAccessChain(
spv::StorageClassUniform,
uniform_bool_loop_constants_, id_vector_temp_),
spv::NoPrecision);
spv::Id address_relative_old = builder_->createCompositeExtract(
address_relative_stack_old, type_int_, 0);
builder_->createStore(
builder_->createCompositeInsert(
builder_->createBinOp(
spv::OpIAdd, type_int_, address_relative_old,
builder_->createTriOp(
spv::OpBitFieldSExtract, type_int_,
builder_->createUnaryOp(spv::OpBitcast, type_int_,
loop_constant),
builder_->makeIntConstant(16), builder_->makeIntConstant(8))),
address_relative_stack_old, type_int4_, 0),
var_main_address_relative_);
// Jump back to the beginning of the loop body.
main_switch_next_pc_phi_operands_.push_back(
builder_->makeIntConstant(int(instr.loop_body_address)));
main_switch_next_pc_phi_operands_.push_back(
builder_->getBuildPoint()->getId());
builder_->createBranch(main_loop_continue_);
// Break case.
builder_->setBuildPoint(&break_block);
// Pop the current loop off the loop counter and the relative address stacks -
// move YZW to XYZ and set W to 0.
id_vector_temp_.clear();
id_vector_temp_.reserve(4);
for (unsigned int i = 1; i < 4; ++i) {
id_vector_temp_.push_back(
builder_->createCompositeExtract(loop_count_stack_old, type_uint_, i));
}
id_vector_temp_.push_back(const_uint_0_);
builder_->createStore(
builder_->createCompositeConstruct(type_uint4_, id_vector_temp_),
var_main_loop_count_);
id_vector_temp_.clear();
id_vector_temp_.reserve(4);
for (unsigned int i = 1; i < 4; ++i) {
id_vector_temp_.push_back(builder_->createCompositeExtract(
address_relative_stack_old, type_int_, i));
}
id_vector_temp_.push_back(const_int_0_);
builder_->createStore(
builder_->createCompositeConstruct(type_int4_, id_vector_temp_),
var_main_address_relative_);
id_vector_temp_.clear();
id_vector_temp_.reserve(4);
// Now going to fall through to the next control flow instruction.
}
void SpirvShaderTranslator::ProcessJumpInstruction(
const ParsedJumpInstruction& instr) {
// Treat like exec, merge with execs if possible, since it's an if too.
ParsedExecInstruction::Type type;
if (instr.type == ParsedJumpInstruction::Type::kConditional) {
type = ParsedExecInstruction::Type::kConditional;
} else if (instr.type == ParsedJumpInstruction::Type::kPredicated) {
type = ParsedExecInstruction::Type::kPredicated;
} else {
type = ParsedExecInstruction::Type::kUnconditional;
}
UpdateExecConditionals(type, instr.bool_constant_index, instr.condition);
// UpdateExecConditionals may not necessarily close the instruction-level
// predicate check (it's not necessary if the execs are merged), but here the
// instruction itself is on the control flow level, so the predicate check is
// on the control flow level too.
CloseInstructionPredication();
if (builder_->getBuildPoint()->isTerminated()) {
// Unreachable for some reason.
return;
}
main_switch_next_pc_phi_operands_.push_back(
builder_->makeIntConstant(int(instr.target_address)));
main_switch_next_pc_phi_operands_.push_back(
builder_->getBuildPoint()->getId());
builder_->createBranch(main_loop_continue_);
}
void SpirvShaderTranslator::EnsureBuildPointAvailable() {
if (!builder_->getBuildPoint()->isTerminated()) {
return;
}
spv::Block& new_block = builder_->makeNewBlock();
new_block.setUnreachable();
builder_->setBuildPoint(&new_block);
}
void SpirvShaderTranslator::StartVertexOrTessEvalShaderBeforeMain() {
// Create the inputs.
if (IsSpirvTessEvalShader()) {
input_primitive_id_ = builder_->createVariable(
spv::NoPrecision, spv::StorageClassInput, type_int_, "gl_PrimitiveID");
builder_->addDecoration(input_primitive_id_, spv::DecorationBuiltIn,
spv::BuiltInPrimitiveId);
} else {
input_vertex_index_ = builder_->createVariable(
spv::NoPrecision, spv::StorageClassInput, type_int_, "gl_VertexIndex");
builder_->addDecoration(input_vertex_index_, spv::DecorationBuiltIn,
spv::BuiltInVertexIndex);
}
// Create the entire GLSL 4.50 gl_PerVertex output similar to what glslang
// does. Members (like gl_PointSize) don't need to be used, and also
// ClipDistance and CullDistance may exist even if the device doesn't support
// them, as long as the capabilities aren't enabled, and nothing is stored to
// them.
if (supports_clip_distance_) {
builder_->addCapability(spv::CapabilityClipDistance);
}
if (supports_cull_distance_) {
builder_->addCapability(spv::CapabilityCullDistance);
}
std::vector<spv::Id> struct_per_vertex_members;
struct_per_vertex_members.reserve(kOutputPerVertexMemberCount);
struct_per_vertex_members.push_back(type_float4_);
struct_per_vertex_members.push_back(type_float_);
// TODO(Triang3l): Specialization constant for ucp_cull_only_ena, for 6 + 1
// or 1 + 7 array sizes.
struct_per_vertex_members.push_back(builder_->makeArrayType(
type_float_, builder_->makeUintConstant(supports_clip_distance_ ? 6 : 1),
0));
struct_per_vertex_members.push_back(
builder_->makeArrayType(type_float_, builder_->makeUintConstant(1), 0));
spv::Id type_struct_per_vertex =
builder_->makeStructType(struct_per_vertex_members, "gl_PerVertex");
builder_->addMemberDecoration(type_struct_per_vertex,
kOutputPerVertexMemberPosition,
spv::DecorationBuiltIn, spv::BuiltInPosition);
builder_->addMemberDecoration(type_struct_per_vertex,
kOutputPerVertexMemberPointSize,
spv::DecorationBuiltIn, spv::BuiltInPointSize);
builder_->addMemberDecoration(
type_struct_per_vertex, kOutputPerVertexMemberClipDistance,
spv::DecorationBuiltIn, spv::BuiltInClipDistance);
builder_->addMemberDecoration(
type_struct_per_vertex, kOutputPerVertexMemberCullDistance,
spv::DecorationBuiltIn, spv::BuiltInCullDistance);
builder_->addDecoration(type_struct_per_vertex, spv::DecorationBlock);
output_per_vertex_ =
builder_->createVariable(spv::NoPrecision, spv::StorageClassOutput,
type_struct_per_vertex, "xe_out_gl_PerVertex");
}
void SpirvShaderTranslator::StartVertexOrTessEvalShaderInMain() {
var_main_point_size_edge_flag_kill_vertex_ = builder_->createVariable(
spv::NoPrecision, spv::StorageClassFunction, type_float3_,
"xe_var_point_size_edge_flag_kill_vertex");
}
void SpirvShaderTranslator::CompleteVertexOrTessEvalShaderInMain() {}
void SpirvShaderTranslator::CompleteVertexOrTessEvalShaderAfterMain(
spv::Instruction* entry_point) {
if (IsSpirvTessEvalShader()) {
entry_point->addIdOperand(input_primitive_id_);
} else {
entry_point->addIdOperand(input_vertex_index_);
}
entry_point->addIdOperand(output_per_vertex_);
}
void SpirvShaderTranslator::UpdateExecConditionals(
ParsedExecInstruction::Type type, uint32_t bool_constant_index,
bool condition) {
// Check if we can merge the new exec with the previous one, or the jump with
// the previous exec. The instruction-level predicate check is also merged in
// this case.
if (type == ParsedExecInstruction::Type::kConditional) {
// Can merge conditional with conditional, as long as the bool constant and
// the expected values are the same.
if (cf_exec_conditional_merge_ &&
cf_exec_bool_constant_or_predicate_ == bool_constant_index &&
cf_exec_condition_ == condition) {
return;
}
} else if (type == ParsedExecInstruction::Type::kPredicated) {
// Can merge predicated with predicated if the conditions are the same and
// the previous exec hasn't modified the predicate register.
if (!cf_exec_predicate_written_ && cf_exec_conditional_merge_ &&
cf_exec_bool_constant_or_predicate_ == kCfExecBoolConstantPredicate &&
cf_exec_condition_ == condition) {
return;
}
} else {
// Can merge unconditional with unconditional.
assert_true(type == ParsedExecInstruction::Type::kUnconditional);
if (!cf_exec_conditional_merge_) {
return;
}
}
CloseExecConditionals();
if (type == ParsedExecInstruction::Type::kUnconditional) {
return;
}
EnsureBuildPointAvailable();
spv::Id condition_id;
if (type == ParsedExecInstruction::Type::kConditional) {
id_vector_temp_.clear();
id_vector_temp_.reserve(3);
// Bool constants (member 0).
id_vector_temp_.push_back(const_int_0_);
// 128-bit vector.
id_vector_temp_.push_back(
builder_->makeIntConstant(int(bool_constant_index >> 7)));
// 32-bit scalar of a 128-bit vector.
id_vector_temp_.push_back(
builder_->makeIntConstant(int((bool_constant_index >> 5) & 3)));
spv::Id bool_constant_scalar =
builder_->createLoad(builder_->createAccessChain(
spv::StorageClassUniform,
uniform_bool_loop_constants_, id_vector_temp_),
spv::NoPrecision);
condition_id = builder_->createBinOp(
spv::OpINotEqual, type_bool_,
builder_->createBinOp(
spv::OpBitwiseAnd, type_uint_, bool_constant_scalar,
builder_->makeUintConstant(uint32_t(1)
<< (bool_constant_index & 31))),
const_uint_0_);
cf_exec_bool_constant_or_predicate_ = bool_constant_index;
} else if (type == ParsedExecInstruction::Type::kPredicated) {
condition_id = builder_->createLoad(var_main_predicate_, spv::NoPrecision);
cf_exec_bool_constant_or_predicate_ = kCfExecBoolConstantPredicate;
} else {
assert_unhandled_case(type);
return;
}
cf_exec_condition_ = condition;
spv::Function& function = builder_->getBuildPoint()->getParent();
cf_exec_conditional_merge_ =
new spv::Block(builder_->getUniqueId(), function);
{
std::unique_ptr<spv::Instruction> selection_merge_op =
std::make_unique<spv::Instruction>(spv::OpSelectionMerge);
selection_merge_op->addIdOperand(cf_exec_conditional_merge_->getId());
selection_merge_op->addImmediateOperand(spv::SelectionControlMaskNone);
builder_->getBuildPoint()->addInstruction(std::move(selection_merge_op));
}
spv::Block& inner_block = builder_->makeNewBlock();
builder_->createConditionalBranch(
condition_id, condition ? &inner_block : cf_exec_conditional_merge_,
condition ? cf_exec_conditional_merge_ : &inner_block);
builder_->setBuildPoint(&inner_block);
}
void SpirvShaderTranslator::CloseInstructionPredication() {
if (!cf_instruction_predicate_merge_) {
return;
}
spv::Block& inner_block = *builder_->getBuildPoint();
if (!inner_block.isTerminated()) {
builder_->createBranch(cf_instruction_predicate_merge_);
}
inner_block.getParent().addBlock(cf_instruction_predicate_merge_);
builder_->setBuildPoint(cf_instruction_predicate_merge_);
cf_instruction_predicate_merge_ = nullptr;
}
void SpirvShaderTranslator::CloseExecConditionals() {
// Within the exec - instruction-level predicate check.
CloseInstructionPredication();
// Exec level.
if (cf_exec_conditional_merge_) {
spv::Block& inner_block = *builder_->getBuildPoint();
if (!inner_block.isTerminated()) {
builder_->createBranch(cf_exec_conditional_merge_);
}
inner_block.getParent().addBlock(cf_exec_conditional_merge_);
builder_->setBuildPoint(cf_exec_conditional_merge_);
cf_exec_conditional_merge_ = nullptr;
}
// Nothing relies on the predicate value being unchanged now.
cf_exec_predicate_written_ = false;
}
} // namespace gpu
} // namespace xe