mirror of
https://github.com/shadps4-emu/shadPS4.git
synced 2025-07-27 04:25:12 +00:00
resource_tracking_pass: Perform BFS for sharp tracking
* The Witness triggered this
This commit is contained in:
parent
1e842c01bf
commit
1b5a96e307
@ -1029,7 +1029,7 @@ s32 PS4_SYSV_ABI sceGnmInsertPushMarker(u32* cmdbuf, u32 size, const char* marke
|
|||||||
|
|
||||||
if (cmdbuf && marker) {
|
if (cmdbuf && marker) {
|
||||||
const auto len = std::strlen(marker);
|
const auto len = std::strlen(marker);
|
||||||
const u32 packet_size = ((len + 8) >> 2) + ((len + 0xc) >> 3);
|
const u32 packet_size = ((len + 8) >> 2) + ((len + 0xc) >> 3) * 2;
|
||||||
if (packet_size + 2 == size) {
|
if (packet_size + 2 == size) {
|
||||||
auto* nop = reinterpret_cast<PM4CmdNop*>(cmdbuf);
|
auto* nop = reinterpret_cast<PM4CmdNop*>(cmdbuf);
|
||||||
nop->header =
|
nop->header =
|
||||||
|
@ -396,6 +396,7 @@ void Translate(IR::Block* block, u32 block_base, std::span<const GcnInst> inst_l
|
|||||||
case Opcode::IMAGE_SAMPLE_L:
|
case Opcode::IMAGE_SAMPLE_L:
|
||||||
case Opcode::IMAGE_SAMPLE_C_O:
|
case Opcode::IMAGE_SAMPLE_C_O:
|
||||||
case Opcode::IMAGE_SAMPLE_B:
|
case Opcode::IMAGE_SAMPLE_B:
|
||||||
|
case Opcode::IMAGE_SAMPLE_C_LZ_O:
|
||||||
translator.IMAGE_SAMPLE(inst);
|
translator.IMAGE_SAMPLE(inst);
|
||||||
break;
|
break;
|
||||||
case Opcode::IMAGE_ATOMIC_ADD:
|
case Opcode::IMAGE_ATOMIC_ADD:
|
||||||
|
52
src/shader_recompiler/ir/breadth_first_search.h
Normal file
52
src/shader_recompiler/ir/breadth_first_search.h
Normal file
@ -0,0 +1,52 @@
|
|||||||
|
// SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
|
||||||
|
// SPDX-License-Identifier: GPL-2.0-or-later
|
||||||
|
|
||||||
|
#pragma once
|
||||||
|
|
||||||
|
#include <optional>
|
||||||
|
#include <type_traits>
|
||||||
|
#include <boost/container/small_vector.hpp>
|
||||||
|
#include <queue>
|
||||||
|
#include "shader_recompiler/ir/value.h"
|
||||||
|
|
||||||
|
namespace Shader::IR {
|
||||||
|
|
||||||
|
template <typename Pred>
|
||||||
|
auto BreadthFirstSearch(const Value& value, Pred&& pred)
|
||||||
|
-> std::invoke_result_t<Pred, const Inst*> {
|
||||||
|
if (value.IsImmediate()) {
|
||||||
|
// Nothing to do with immediates
|
||||||
|
return std::nullopt;
|
||||||
|
}
|
||||||
|
// Breadth-first search visiting the right most arguments first
|
||||||
|
boost::container::small_vector<const Inst*, 2> visited;
|
||||||
|
std::queue<const Inst*> queue;
|
||||||
|
queue.push(value.InstRecursive());
|
||||||
|
|
||||||
|
while (!queue.empty()) {
|
||||||
|
// Pop one instruction from the queue
|
||||||
|
const Inst* const inst{queue.front()};
|
||||||
|
queue.pop();
|
||||||
|
if (const std::optional result = pred(inst)) {
|
||||||
|
// This is the instruction we were looking for
|
||||||
|
return result;
|
||||||
|
}
|
||||||
|
// Visit the right most arguments first
|
||||||
|
for (size_t arg = inst->NumArgs(); arg--;) {
|
||||||
|
const Value arg_value{inst->Arg(arg)};
|
||||||
|
if (arg_value.IsImmediate()) {
|
||||||
|
continue;
|
||||||
|
}
|
||||||
|
// Queue instruction if it hasn't been visited
|
||||||
|
const Inst* const arg_inst{arg_value.InstRecursive()};
|
||||||
|
if (std::ranges::find(visited, arg_inst) == visited.end()) {
|
||||||
|
visited.push_back(arg_inst);
|
||||||
|
queue.push(arg_inst);
|
||||||
|
}
|
||||||
|
}
|
||||||
|
}
|
||||||
|
// SSA tree has been traversed and the result hasn't been found
|
||||||
|
return std::nullopt;
|
||||||
|
}
|
||||||
|
|
||||||
|
} // namespace Shader::IR
|
@ -4,8 +4,8 @@
|
|||||||
#include <algorithm>
|
#include <algorithm>
|
||||||
#include <deque>
|
#include <deque>
|
||||||
#include <boost/container/small_vector.hpp>
|
#include <boost/container/small_vector.hpp>
|
||||||
|
|
||||||
#include "shader_recompiler/ir/basic_block.h"
|
#include "shader_recompiler/ir/basic_block.h"
|
||||||
|
#include "shader_recompiler/ir/breadth_first_search.h"
|
||||||
#include "shader_recompiler/ir/ir_emitter.h"
|
#include "shader_recompiler/ir/ir_emitter.h"
|
||||||
#include "shader_recompiler/ir/program.h"
|
#include "shader_recompiler/ir/program.h"
|
||||||
#include "shader_recompiler/runtime_info.h"
|
#include "shader_recompiler/runtime_info.h"
|
||||||
@ -244,22 +244,19 @@ SharpLocation TrackSharp(const IR::Inst* inst) {
|
|||||||
const IR::Inst* spgpr_base = inst->Arg(0).InstRecursive();
|
const IR::Inst* spgpr_base = inst->Arg(0).InstRecursive();
|
||||||
|
|
||||||
// Retrieve SGPR pair that holds sbase
|
// Retrieve SGPR pair that holds sbase
|
||||||
const IR::Inst* sbase0 = spgpr_base->Arg(0).InstRecursive();
|
const auto pred = [](const IR::Inst* inst) -> std::optional<IR::ScalarReg> {
|
||||||
const IR::Inst* sbase1 = spgpr_base->Arg(1).InstRecursive();
|
if (inst->GetOpcode() == IR::Opcode::GetUserData) {
|
||||||
while (sbase0->GetOpcode() == IR::Opcode::Phi) {
|
return inst->Arg(0).ScalarReg();
|
||||||
sbase0 = sbase0->Arg(0).TryInstRecursive();
|
|
||||||
}
|
}
|
||||||
while (sbase1->GetOpcode() == IR::Opcode::Phi) {
|
return std::nullopt;
|
||||||
sbase1 = sbase1->Arg(0).TryInstRecursive();
|
};
|
||||||
}
|
const auto base0 = IR::BreadthFirstSearch(spgpr_base->Arg(0), pred);
|
||||||
ASSERT_MSG(sbase0->GetOpcode() == IR::Opcode::GetUserData &&
|
const auto base1 = IR::BreadthFirstSearch(spgpr_base->Arg(1), pred);
|
||||||
sbase1->GetOpcode() == IR::Opcode::GetUserData,
|
ASSERT_MSG(base0 && base1, "Nested resource loads not supported");
|
||||||
"Nested resource loads not supported");
|
|
||||||
const IR::ScalarReg base = sbase0->Arg(0).ScalarReg();
|
|
||||||
|
|
||||||
// Return retrieved location.
|
// Return retrieved location.
|
||||||
return SharpLocation{
|
return SharpLocation{
|
||||||
.sgpr_base = u32(base),
|
.sgpr_base = u32(base0.value()),
|
||||||
.dword_offset = dword_offset,
|
.dword_offset = dword_offset,
|
||||||
};
|
};
|
||||||
}
|
}
|
||||||
|
@ -422,6 +422,13 @@ vk::Format SurfaceFormat(AmdGpu::DataFormat data_format, AmdGpu::NumberFormat nu
|
|||||||
num_format == AmdGpu::NumberFormat::Sint) {
|
num_format == AmdGpu::NumberFormat::Sint) {
|
||||||
return vk::Format::eR16G16Sint;
|
return vk::Format::eR16G16Sint;
|
||||||
}
|
}
|
||||||
|
if (data_format == AmdGpu::DataFormat::Format8_8_8_8 &&
|
||||||
|
num_format == AmdGpu::NumberFormat::Uscaled) {
|
||||||
|
return vk::Format::eR8G8B8A8Uscaled;
|
||||||
|
}
|
||||||
|
if (data_format == AmdGpu::DataFormat::Format16 && num_format == AmdGpu::NumberFormat::Unorm) {
|
||||||
|
return vk::Format::eR16Unorm;
|
||||||
|
}
|
||||||
UNREACHABLE_MSG("Unknown data_format={} and num_format={}", u32(data_format), u32(num_format));
|
UNREACHABLE_MSG("Unknown data_format={} and num_format={}", u32(data_format), u32(num_format));
|
||||||
}
|
}
|
||||||
|
|
||||||
|
Loading…
Reference in New Issue
Block a user