mirrors/shadPS4
1
0
Fork 0
mirror of https://github.com/shadps4-emu/shadPS4.git synced 2025-07-22 18:15:14 +00:00
Code Issues Packages Projects Releases Wiki Activity

liverpool: Use span for acb too

Browse Source
This commit is contained in:
IndecisiveTurtle 2025-07-01 14:56:51 +03:00
parent 53a48bdbcd
commit 0b410a16ad
3 changed files with 18 additions and 20 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
src/core/libraries/gnmdriver/gnmdriver.cpp
View File

@ -2834,7 +2834,7 @@ void RegisterlibSceGnmDriver(Core::Loader::SymbolsResolver* sym) {
} }
if (Config::copyGPUCmdBuffers()) { if (Config::copyGPUCmdBuffers()) {
liverpool->reserveCopyBufferSpace(); liverpool->ReserveCopyBufferSpace();
} }
Platform::IrqC::Instance()->Register(Platform::InterruptId::GpuIdle, ResetSubmissionLock, Platform::IrqC::Instance()->Register(Platform::InterruptId::GpuIdle, ResetSubmissionLock,

32
src/video_core/amdgpu/liverpool.cpp
View File

@ -812,31 +812,32 @@ Liverpool::Task Liverpool::ProcessGraphics(std::span<const u32> dcb, std::span<c
} }
template <bool is_indirect> template <bool is_indirect>
Liverpool::Task Liverpool::ProcessCompute(const u32* acb, u32 acb_dwords, u32 vqid) { Liverpool::Task Liverpool::ProcessCompute(std::span<const u32> acb, u32 vqid) {
FIBER_ENTER(acb_task_name[vqid]); FIBER_ENTER(acb_task_name[vqid]);
auto& queue = asc_queues[{vqid}]; auto& queue = asc_queues[{vqid}];
auto base_addr = reinterpret_cast<VAddr>(acb); auto base_addr = reinterpret_cast<VAddr>(acb.data());
while (acb_dwords > 0) { while (!acb.empty()) {
ProcessCommands(); ProcessCommands();
auto* header = reinterpret_cast<const PM4Header*>(acb); auto* header = reinterpret_cast<const PM4Header*>(acb.data());
u32 next_dw_off = header->type3.NumWords() + 1; u32 next_dw_off = header->type3.NumWords() + 1;
// If we have a buffered packet, use it. // If we have a buffered packet, use it.
if (queue.tmp_dwords > 0) [[unlikely]] { if (queue.tmp_dwords > 0) [[unlikely]] {
header = reinterpret_cast<const PM4Header*>(queue.tmp_packet.data()); header = reinterpret_cast<const PM4Header*>(queue.tmp_packet.data());
next_dw_off = header->type3.NumWords() + 1 - queue.tmp_dwords; next_dw_off = header->type3.NumWords() + 1 - queue.tmp_dwords;
std::memcpy(queue.tmp_packet.data() + queue.tmp_dwords, acb, next_dw_off * sizeof(u32)); std::memcpy(queue.tmp_packet.data() + queue.tmp_dwords, acb.data(),
next_dw_off * sizeof(u32));
queue.tmp_dwords = 0; queue.tmp_dwords = 0;
} }
// If the packet is split across ring boundary, buffer until next submission // If the packet is split across ring boundary, buffer until next submission
if (next_dw_off > acb_dwords) [[unlikely]] { if (next_dw_off > acb.size()) [[unlikely]] {
std::memcpy(queue.tmp_packet.data(), acb, acb_dwords * sizeof(u32)); std::memcpy(queue.tmp_packet.data(), acb.data(), acb.size_bytes());
queue.tmp_dwords = acb_dwords; queue.tmp_dwords = acb.size();
if constexpr (!is_indirect) { if constexpr (!is_indirect) {
*queue.read_addr += acb_dwords; *queue.read_addr += acb.size();
*queue.read_addr %= queue.ring_size_dw; *queue.read_addr %= queue.ring_size_dw;
} }
break; break;
@ -845,9 +846,7 @@ Liverpool::Task Liverpool::ProcessCompute(const u32* acb, u32 acb_dwords, u32 vq
if (header->type == 2) { if (header->type == 2) {
// Type-2 packet are used for padding purposes // Type-2 packet are used for padding purposes
next_dw_off = 1; next_dw_off = 1;
acb += next_dw_off; acb = NextPacket(acb, next_dw_off);
acb_dwords -= next_dw_off;
if constexpr (!is_indirect) { if constexpr (!is_indirect) {
*queue.read_addr += next_dw_off; *queue.read_addr += next_dw_off;
*queue.read_addr %= queue.ring_size_dw; *queue.read_addr %= queue.ring_size_dw;
@ -869,8 +868,8 @@ Liverpool::Task Liverpool::ProcessCompute(const u32* acb, u32 acb_dwords, u32 vq
} }
case PM4ItOpcode::IndirectBuffer: { case PM4ItOpcode::IndirectBuffer: {
const auto* indirect_buffer = reinterpret_cast<const PM4CmdIndirectBuffer*>(header); const auto* indirect_buffer = reinterpret_cast<const PM4CmdIndirectBuffer*>(header);
auto task = ProcessCompute<true>(indirect_buffer->Address<const u32>(), auto task = ProcessCompute<true>(
indirect_buffer->ib_size, vqid); {indirect_buffer->Address<const u32>(), indirect_buffer->ib_size}, vqid);
RESUME_ASC(task, vqid); RESUME_ASC(task, vqid);
while (!task.handle.done()) { while (!task.handle.done()) {
@ -1032,8 +1031,7 @@ Liverpool::Task Liverpool::ProcessCompute(const u32* acb, u32 acb_dwords, u32 vq
static_cast<u32>(opcode), header->type3.NumWords()); static_cast<u32>(opcode), header->type3.NumWords());
} }
acb += next_dw_off; acb = NextPacket(acb, next_dw_off);
acb_dwords -= next_dw_off;
if constexpr (!is_indirect) { if constexpr (!is_indirect) {
*queue.read_addr += next_dw_off; *queue.read_addr += next_dw_off;
@ -1103,7 +1101,7 @@ void Liverpool::SubmitAsc(u32 gnm_vqid, std::span<const u32> acb) {
auto& queue = mapped_queues[gnm_vqid]; auto& queue = mapped_queues[gnm_vqid];
const auto vqid = gnm_vqid - 1; const auto vqid = gnm_vqid - 1;
const auto& task = ProcessCompute(acb.data(), acb.size(), vqid); const auto& task = ProcessCompute(acb, vqid);
{ {
std::scoped_lock lock{queue.m_access}; std::scoped_lock lock{queue.m_access};
queue.submits.emplace(task.handle); queue.submits.emplace(task.handle);

4
src/video_core/amdgpu/liverpool.h
View File

@ -1519,7 +1519,7 @@ public:
submit_cv.notify_one(); submit_cv.notify_one();
} }
void reserveCopyBufferSpace() { void ReserveCopyBufferSpace() {
GpuQueue& gfx_queue = mapped_queues[GfxQueueId]; GpuQueue& gfx_queue = mapped_queues[GfxQueueId];
std::scoped_lock<std::mutex> lk(gfx_queue.m_access); std::scoped_lock<std::mutex> lk(gfx_queue.m_access);
@ -1582,7 +1582,7 @@ private:
Task ProcessGraphics(std::span<const u32> dcb, std::span<const u32> ccb); Task ProcessGraphics(std::span<const u32> dcb, std::span<const u32> ccb);
Task ProcessCeUpdate(std::span<const u32> ccb); Task ProcessCeUpdate(std::span<const u32> ccb);
template <bool is_indirect = false> template <bool is_indirect = false>
Task ProcessCompute(const u32* acb, u32 acb_dwords, u32 vqid); Task ProcessCompute(std::span<const u32> acb, u32 vqid);
void ProcessCommands(); void ProcessCommands();
void Process(std::stop_token stoken); void Process(std::stop_token stoken);