[vk] Clamp dynamic descriptors based on device limits (#4115)

Should fix this specific crash
```
* thread #82, name = 'GPU', stop reason = Exception 0xc0000005 encountered at address 0x7ff7e5193e89: Access violation reading location 0x00000098
  * frame #0: 0x00007ff7e5193e89 eden.exe`std::unique_ptr<Vulkan::Scheduler::CommandChunk,std::default_delete<Vulkan::Scheduler::CommandChunk> >::operator->(this=<unavailable>) at memory:3453 [inlined]
    frame #1: 0x00007ff7e5193e81 eden.exe`void Vulkan::Scheduler::DispatchWork(this=0x0000000000000000) at vk_scheduler.cpp:146
    frame #2: 0x00007ff7e5193d8f eden.exe`void Vulkan::Scheduler::WaitWorker(this=<unavailable>) at vk_scheduler.cpp:133
    frame #3: 0x00007ff7e54f472e eden.exe`void Vulkan::UpdateDescriptorQueue::Acquire(this=0x0000026c77cfb4b8) at vk_update_descriptor.cpp:41
    frame #4: 0x00007ff7e5537a43 eden.exe`void Vulkan::ASTCDecoderPass::Assemble(this=0x0000026c794fba10, image=0x0000026cfb794850, map=0x00000025e28fcb30, swizzles=<unavailable>) at vk_compute_pass.cpp:603
    frame #5: 0x00007ff7e55043e1 eden.exe`void Vulkan::TextureCacheRuntime::AccelerateImageUpload(this=<unavailable>, image=0x0000026cfb794850, map=0x00000025e28fcb30, swizzles=size=10, z_start=0, z_count=0) at vk_texture_cache.cpp:2482
    frame #6: 0x00007ff7e5516592 eden.exe`void VideoCommon::TextureCache<struct Vulkan::TextureCacheParams>::UploadImageContents<struct Vulkan::StagingBufferRef>(this=<unavailable>, image=0x0000026cfb794850, staging=0x00000025e28fcb30) at texture_cache.h:1147
    frame #7: 0x00007ff7e551523d eden.exe`void VideoCommon::TextureCache<struct Vulkan::TextureCacheParams>::RefreshContents(this=0x0000026c794fd800, image=0x0000026cfb794850, image_id=(index = 3801072224)) at texture_cache.h:1133
    frame #8: 0x00007ff7e5517b50 eden.exe`struct Common::SlotId VideoCommon::TextureCache<struct Vulkan::TextureCacheParams>::JoinImages(this=0x0000026c794fd800, info=<unavailable>, gpu_addr=25374426112, cpu_addr=2254181376) at texture_cache.h:1644
    frame #9: 0x00007ff7e5516ff6 eden.exe`struct Common::SlotId VideoCommon::TextureCache<struct Vulkan::TextureCacheParams>::InsertImage(this=0x0000026c794fd800, info=0x00000025e28fd270, gpu_addr=25374426112, options=0x0) at texture_cache.h:1513
    frame #10: 0x00007ff7e5516a60 eden.exe`struct Common::SlotId VideoCommon::TextureCache<struct Vulkan::TextureCacheParams>::FindOrInsertImage(this=0x0000026c794fd800, info=0x00000025e28fd270, gpu_addr=25374426112, options=0x0) at texture_cache.h:1194
    frame #11: 0x00007ff7e5515a13 eden.exe`struct Common::SlotId VideoCommon::TextureCache<struct Vulkan::TextureCacheParams>::CreateImageView(this=0x0000026c794fd800, config=0x00000025e28fd370) at texture_cache.h:1173
    frame #12: 0x00007ff7e550cd64 eden.exe`struct Common::SlotId VideoCommon::TextureCache<struct Vulkan::TextureCacheParams>::VisitImageView(this=0x0000026c794fd800, index=4586, compute=<unavailable>) at texture_cache.h:554
    frame #13: 0x00007ff7e550d181 eden.exe`void VideoCommon::TextureCache<struct Vulkan::TextureCacheParams>::FillImageViews(this=0x0000026c794fd800, views=size=5, compute=<unavailable>, blacklist=<unavailable>) at texture_cache.h:227
    frame #14: 0x00007ff7e58896e4 eden.exe`Vulkan::GraphicsPipeline::ConfigureImpl<Vulkan::(anonymous namespace)::SimpleStorageSpec>(this=0x0000026f865c1a60, is_indexed=<unavailable>) at vk_graphics_pipeline.cpp:415
    frame #15: 0x00007ff7e5889479 eden.exe`<lambda_1>::operator(pl=<unavailable>, is_indexed=<unavailable>) at vk_graphics_pipeline.h:123 [inlined]
    frame #16: 0x00007ff7e5889474 eden.exe`<lambda_1>::__invoke(pl=<unavailable>, is_indexed=<unavailable>) at vk_graphics_pipeline.h:123
    frame #17: 0x00007ff7e519d2a8 eden.exe`Vulkan::GraphicsPipeline::Configure(this=0x0000026f865c1a60, is_indexed=<unavailable>) at vk_graphics_pipeline.h:105 [inlined]
    frame #18: 0x00007ff7e519d29d eden.exe`Vulkan::RasterizerVulkan::PrepareDraw<`lambda at D:\a\g\g\eden-source\src\video_core\renderer_vulkan\vk_rasterizer.cpp:256:29'>(this=0x0000026c764f9368, is_indexed=<unavailable>, draw_func=0x00000025e28fdb80) at vk_rasterizer.cpp:244
    frame #19: 0x00007ff7e519d1db eden.exe`void Vulkan::RasterizerVulkan::Draw(this=<unavailable>, is_indexed=<unavailable>, instance_count=<unavailable>) at vk_rasterizer.cpp:256
    frame #20: 0x00007ff7e50d9025 eden.exe`void Tegra::HLE_DrawIndexedIndirect::Fallback(this=0x0000026f2af85118, maxwell3d=0x0000026ccbf9eb00, parameters=size=6) at macro.cpp:178
    frame #21: 0x00007ff7e50d8f21 eden.exe`void Tegra::HLE_DrawIndexedIndirect::Execute(this=0x0000026f2af85118, maxwell3d=0x0000026ccbf9eb00, parameters=size=6) at macro.cpp:129
    frame #22: 0x00007ff7e50db607 eden.exe`Tegra::MacroEngine::Execute::<lambda_0>::operator(this=0x00000025e28fde70, acm= Active Type = Tegra::HLE_DrawIndexedIndirect ) at macro.cpp:1362
    frame #23: 0x00007ff7e50dada0 eden.exe`void Tegra::MacroEngine::Execute(this=0x0000026ccbfaa020, maxwell3d=0x0000026ccbf9eb00, method=418, parameters=<unavailable>) at macro.cpp:1392
    frame #24: 0x00007ff7e4ef062b eden.exe`void Tegra::Engines::Maxwell3D::CallMacroMethod(this=0x0000026ccbf9eb00, method=<unavailable>, parameters=<unavailable>) at maxwell_3d.cpp:390
    frame #25: 0x00007ff7e4ef04be eden.exe`void Tegra::Engines::Maxwell3D::ProcessMacro(this=0x0000026ccbf9eb00, method=<unavailable>, base_start=<unavailable>, amount=5, is_last_call=<unavailable>) at maxwell_3d.cpp:223
    frame #26: 0x00007ff7e4ef1673 eden.exe`void Tegra::Engines::Maxwell3D::CallMultiMethod(this=<unavailable>, method=<unavailable>, base_start=0x000001ea398e22f4, amount=<unavailable>, methods_pending=5) at maxwell_3d.cpp:419
    frame #27: 0x00007ff7e4ef535a eden.exe`void Tegra::DmaPusher::CallMultiMethod(this=<unavailable>, base_start=0x000001ea398e22f4, num_methods=<unavailable>) const at dma_pusher.cpp:201
    frame #28: 0x00007ff7e4ef5023 eden.exe`void Tegra::DmaPusher::ProcessCommands(this=0x0000026ccbfba698, commands=size=36) at dma_pusher.cpp:120
    frame #29: 0x00007ff7e4ef4ced eden.exe`bool Tegra::DmaPusher::Step(this=0x0000026ccbfba698) at dma_pusher.cpp:88
    frame #30: 0x00007ff7e4ef49f8 eden.exe`void Tegra::DmaPusher::DispatchCalls(this=0x0000026ccbfba698) at dma_pusher.cpp:40
    frame #31: 0x00007ff7e4ede797 eden.exe`void Tegra::Control::Scheduler::Push(this=<unavailable>, channel=-493895072, entries=0x00000025e28fe2e0) at scheduler.cpp:31
    frame #32: 0x00007ff7e4eedc7b eden.exe`VideoCommon::GPUThread::ThreadManager::StartThread::<lambda_0>::operator(this=<unavailable>, stop_token=stop_token @ 0x00000025e28ffb40) at gpu_thread.cpp:42 [inlined]
    frame #33: 0x00007ff7e4eedae8 eden.exe`std::invoke(_Obj=<unavailable>, _Arg1=0x0000026ccb3edef0) at type_traits:1680 [inlined]
    frame #34: 0x00007ff7e4eedac1 eden.exe`std::thread::_Invoke<std::tuple<`lambda at D:\a\g\g\eden-source\src\video_core\gpu_thread.cpp:29:27',std::stop_token>,0,1>(_RawVals=0x0000026ccb3edef0) at thread:60
    frame #35: 0x00007ff8e87a37b0 ucrtbase.dll`wcsrchr + 336
    ```

Signed-off-by: lizzie <lizzie@eden-emu.dev>

Co-authored-by: CamilleLaVey <camillelavey99@gmail.com>
Reviewed-on: https://git.eden-emu.dev/eden-emu/eden/pulls/4115
Reviewed-by: MaranBr <maranbr@eden-emu.dev>
Reviewed-by: CamilleLaVey <camillelavey99@gmail.com>
This commit is contained in:
lizzie
2026-06-30 04:33:30 +02:00
committed by crueter
parent 1b482fa99b
commit 2068b5d452
21 changed files with 223 additions and 161 deletions
+65 -71
View File
@@ -32,71 +32,62 @@ struct TextureInst {
using TextureInstVector = boost::container::small_vector<TextureInst, 24>;
constexpr u32 DESCRIPTOR_SIZE = 8;
constexpr u32 DESCRIPTOR_SIZE_SHIFT = static_cast<u32>(std::countr_zero(DESCRIPTOR_SIZE));
constexpr u32 DYNAMIC_DESCRIPTOR_CBUF_BYTES = 16 * 1024;
constexpr u32 MAX_DYNAMIC_DESCRIPTOR_COUNT = 1024;
constexpr u32 DESCRIPTOR_SIZE_SHIFT = u32(std::countr_zero(DESCRIPTOR_SIZE));
constexpr u32 DESCRIPTOR_MAX_COUNT = 1024;
u32 DynamicDescriptorSizeShift(const IR::U32& dynamic_offset) {
const IR::Inst* const inst{dynamic_offset.InstRecursive()};
if (!inst || inst->GetOpcode() != IR::Opcode::ShiftLeftLogical32) {
const IR::Inst* const inst = dynamic_offset.InstRecursive();
if (!inst || inst->GetOpcode() != IR::Opcode::ShiftLeftLogical32)
return DESCRIPTOR_SIZE_SHIFT;
}
const IR::Value shift{inst->Arg(1)};
if (!shift.IsImmediate()) {
const IR::Value shift = inst->Arg(1);
if (!shift.IsImmediate())
return DESCRIPTOR_SIZE_SHIFT;
}
const u32 size_shift{shift.U32()};
return size_shift >= DESCRIPTOR_SIZE_SHIFT && size_shift < 31 ? size_shift
: DESCRIPTOR_SIZE_SHIFT;
const u32 size_shift = shift.U32();
return size_shift >= DESCRIPTOR_SIZE_SHIFT && size_shift < 31 ? size_shift : DESCRIPTOR_SIZE_SHIFT;
}
u32 DynamicDescriptorCount(u32 base_offset, u32 size_shift) {
if (size_shift >= 31 || base_offset >= DYNAMIC_DESCRIPTOR_CBUF_BYTES) {
u32 DynamicDescriptorCount(u32 base_offset, u32 size_shift, u32 max_descriptors) {
auto const descriptor_limit = (std::max)(1U, max_descriptors);
auto const max_cbuf_bytes = 16 * descriptor_limit;
if (size_shift >= 31 || base_offset >= max_cbuf_bytes)
return 1;
}
const u32 stride{1U << size_shift};
const u32 available{DYNAMIC_DESCRIPTOR_CBUF_BYTES - base_offset};
if (available < DESCRIPTOR_SIZE) {
auto const stride = 1U << size_shift;
auto const available = max_cbuf_bytes - base_offset;
if (available < DESCRIPTOR_SIZE)
return 1;
}
const u32 available_count{1U + (available - DESCRIPTOR_SIZE) / stride};
return std::min(MAX_DYNAMIC_DESCRIPTOR_COUNT, available_count);
auto const available_count = 1U + (available - DESCRIPTOR_SIZE) / stride;
return std::min(descriptor_limit, available_count);
}
u32 SaturatingSub(u32 lhs, u32 rhs) {
return lhs > rhs ? lhs - rhs : 0;
}
template <typename Descriptors>
u32 StaticDescriptorCount(const Descriptors& descriptors) {
u32 count{};
for (const auto& desc : descriptors) {
if (desc.count <= 1) {
count += desc.count;
}
}
return count;
template <typename T>
[[nodiscard]] u32 StaticDescriptorCount(T const& descriptors) noexcept {
return std::accumulate(descriptors.cbegin(), descriptors.cend(), 0U, [](auto const& acc, auto const& e) {
return acc + (e.count <= 1 ? e.count : 0);
});
}
u32 DynamicSampledTextureCap(const Info& info, const HostTranslateInfo& host_info,
u32 dynamic_arrays) {
if (dynamic_arrays == 0) {
return MAX_DYNAMIC_DESCRIPTOR_COUNT;
u32 DynamicSampledTextureCap(const Info& info, const HostTranslateInfo& host_info, u32 dynamic_arrays) {
auto const sampled_limit = (std::max)(1U, std::min(host_info.max_per_stage_descriptor_sampled_images,
host_info.max_descriptor_set_sampled_images));
auto const resource_limit = (std::max)(1U, host_info.max_per_stage_resources);
if (dynamic_arrays > 0) {
auto const sampled_static_count = StaticDescriptorCount(info.texture_buffer_descriptors) + StaticDescriptorCount(info.texture_descriptors);
auto const resource_static_count =
NumDescriptors(info.constant_buffer_descriptors)
+ NumDescriptors(info.storage_buffers_descriptors)
+ sampled_static_count + NumDescriptors(info.image_buffer_descriptors)
+ NumDescriptors(info.image_descriptors);
auto const sampled_budget = SaturatingSub(sampled_limit, sampled_static_count);
auto const resource_budget = SaturatingSub(resource_limit, resource_static_count);
auto const sampled_cap = sampled_budget / dynamic_arrays;
auto const resource_cap = resource_budget / dynamic_arrays;
return (std::max)(1U, (std::min)(sampled_cap, resource_cap));
}
const u32 sampled_static_count{StaticDescriptorCount(info.texture_buffer_descriptors) +
StaticDescriptorCount(info.texture_descriptors)};
const u32 resource_static_count{
NumDescriptors(info.constant_buffer_descriptors) +
NumDescriptors(info.storage_buffers_descriptors) + sampled_static_count +
NumDescriptors(info.image_buffer_descriptors) + NumDescriptors(info.image_descriptors)};
const u32 sampled_limit{std::min(host_info.max_per_stage_descriptor_sampled_images,
host_info.max_descriptor_set_sampled_images)};
const u32 sampled_budget{SaturatingSub(sampled_limit, sampled_static_count)};
const u32 resource_budget{SaturatingSub(host_info.max_per_stage_resources,
resource_static_count)};
const u32 sampled_cap{sampled_budget / dynamic_arrays};
const u32 resource_cap{resource_budget / dynamic_arrays};
return std::max(1U, std::min({MAX_DYNAMIC_DESCRIPTOR_COUNT, sampled_cap, resource_cap}));
return (std::min)({DESCRIPTOR_MAX_COUNT, sampled_limit, resource_limit});
}
IR::Opcode IndexedInstruction(const IR::Inst& inst) {
@@ -304,21 +295,23 @@ static inline bool IsTexturePixelFormatIntegerCached(Environment& env,
}
std::optional<ConstBufferAddr> Track(const IR::Value& value, Environment& env);
static inline std::optional<ConstBufferAddr> TrackCached(const IR::Value& v, Environment& env) {
std::optional<ConstBufferAddr> Track(const IR::Value& value, Environment& env, const HostTranslateInfo& host_info);
static inline std::optional<ConstBufferAddr> TrackCached(const IR::Value& v, Environment& env, const HostTranslateInfo& host_info) {
if (const IR::Inst* key = v.InstRecursive()) {
if (auto it = env.track_cache.find(key); it != env.track_cache.end()) return it->second;
auto found = Track(v, env);
auto found = Track(v, env, host_info);
if (found) env.track_cache.emplace(key, *found);
return found;
}
return Track(v, env);
return Track(v, env, host_info);
}
std::optional<ConstBufferAddr> TryGetConstBuffer(const IR::Inst* inst, Environment& env);
std::optional<ConstBufferAddr> TryGetConstBuffer(const IR::Inst* inst, Environment& env, const HostTranslateInfo& host_info);
std::optional<ConstBufferAddr> Track(const IR::Value& value, Environment& env) {
return IR::BreadthFirstSearch(value, [&env](const IR::Inst* inst) { return TryGetConstBuffer(inst, env); });
std::optional<ConstBufferAddr> Track(const IR::Value& value, Environment& env, const HostTranslateInfo& host_info) {
return IR::BreadthFirstSearch(value, [&env, &host_info](const IR::Inst* inst) {
return TryGetConstBuffer(inst, env, host_info);
});
}
std::optional<u32> TryGetConstant(IR::Value& value, Environment& env) {
@@ -342,13 +335,13 @@ std::optional<u32> TryGetConstant(IR::Value& value, Environment& env) {
return ReadCbufCached(env, index_number, offset_number);
}
std::optional<ConstBufferAddr> TryGetConstBuffer(const IR::Inst* inst, Environment& env) {
std::optional<ConstBufferAddr> TryGetConstBuffer(const IR::Inst* inst, Environment& env, const HostTranslateInfo& host_info) {
switch (inst->GetOpcode()) {
default:
return std::nullopt;
case IR::Opcode::BitwiseOr32: {
std::optional lhs{TrackCached(inst->Arg(0), env)};
std::optional rhs{TrackCached(inst->Arg(1), env)};
std::optional lhs{TrackCached(inst->Arg(0), env, host_info)};
std::optional rhs{TrackCached(inst->Arg(1), env, host_info)};
if (!lhs || !rhs) {
return std::nullopt;
}
@@ -378,12 +371,11 @@ std::optional<ConstBufferAddr> TryGetConstBuffer(const IR::Inst* inst, Environme
if (!shift.IsImmediate()) {
return std::nullopt;
}
std::optional lhs{TrackCached(inst->Arg(0), env)};
std::optional lhs{TrackCached(inst->Arg(0), env, host_info)};
if (lhs) {
lhs->shift_left = shift.U32();
}
return lhs;
break;
}
case IR::Opcode::BitwiseAnd32: {
IR::Value op1{inst->Arg(0)};
@@ -407,7 +399,7 @@ std::optional<ConstBufferAddr> TryGetConstBuffer(const IR::Inst* inst, Environme
return std::nullopt;
} while (false);
}
std::optional lhs{TrackCached(op1, env)};
std::optional lhs{TrackCached(op1, env, host_info)};
if (lhs) {
lhs->shift_left = static_cast<u32>(std::countr_zero(op2.U32()));
}
@@ -453,7 +445,10 @@ std::optional<ConstBufferAddr> TryGetConstBuffer(const IR::Inst* inst, Environme
} else {
return std::nullopt;
}
const u32 size_shift{DynamicDescriptorSizeShift(dynamic_offset)};
auto const size_shift = DynamicDescriptorSizeShift(dynamic_offset);
auto const sampled_limit = (std::max)(1U, (std::min)(host_info.max_per_stage_descriptor_sampled_images,
host_info.max_descriptor_set_sampled_images));
auto const resource_limit = (std::max)(1U, host_info.max_per_stage_resources);
return ConstBufferAddr{
.index = index.U32(),
.offset = base_offset,
@@ -462,15 +457,15 @@ std::optional<ConstBufferAddr> TryGetConstBuffer(const IR::Inst* inst, Environme
.secondary_offset = 0,
.secondary_shift_left = 0,
.dynamic_offset = dynamic_offset,
.count = DynamicDescriptorCount(base_offset, size_shift),
.count = DynamicDescriptorCount(base_offset, size_shift, (std::min)({DESCRIPTOR_MAX_COUNT, sampled_limit, resource_limit})),
.has_secondary = false,
};
}
TextureInst MakeInst(Environment& env, IR::Block* block, IR::Inst& inst) {
TextureInst MakeInst(Environment& env, IR::Block* block, IR::Inst& inst, const HostTranslateInfo& host_info) {
ConstBufferAddr addr;
if (IsBindless(inst)) {
const std::optional<ConstBufferAddr> track_addr{TrackCached(inst.Arg(0), env)};
const std::optional<ConstBufferAddr> track_addr{TrackCached(inst.Arg(0), env, host_info)};
if (!track_addr) {
throw NotImplementedException("Failed to track bindless texture constant buffer");
@@ -506,15 +501,15 @@ u32 GetTextureHandle(Environment& env, const ConstBufferAddr& cbuf) {
return lhs_raw | rhs_raw;
}
[[maybe_unused]]TextureType ReadTextureType(Environment& env, const ConstBufferAddr& cbuf) {
[[maybe_unused]] TextureType ReadTextureType(Environment& env, const ConstBufferAddr& cbuf) {
return env.ReadTextureType(GetTextureHandle(env, cbuf));
}
[[maybe_unused]]TexturePixelFormat ReadTexturePixelFormat(Environment& env, const ConstBufferAddr& cbuf) {
[[maybe_unused]] TexturePixelFormat ReadTexturePixelFormat(Environment& env, const ConstBufferAddr& cbuf) {
return env.ReadTexturePixelFormat(GetTextureHandle(env, cbuf));
}
[[maybe_unused]]bool IsTexturePixelFormatInteger(Environment& env, const ConstBufferAddr& cbuf) {
[[maybe_unused]] bool IsTexturePixelFormatInteger(Environment& env, const ConstBufferAddr& cbuf) {
return env.IsTexturePixelFormatInteger(GetTextureHandle(env, cbuf));
}
@@ -675,7 +670,7 @@ void TexturePass(Environment& env, IR::Program& program, const HostTranslateInfo
if (!IsTextureInstruction(inst)) {
continue;
}
to_replace.push_back(MakeInst(env, block, inst));
to_replace.push_back(MakeInst(env, block, inst, host_info));
}
}
// Sort instructions to visit textures by constant buffer index, then by offset
@@ -689,8 +684,7 @@ void TexturePass(Environment& env, IR::Program& program, const HostTranslateInfo
program.info.texture_descriptors,
program.info.image_descriptors,
};
const u32 sampled_dynamic_cap{
DynamicSampledTextureCap(program.info, host_info, DynamicSampledTextureArrayCount(to_replace))};
const u32 sampled_dynamic_cap = DynamicSampledTextureCap(program.info, host_info, DynamicSampledTextureArrayCount(to_replace));
for (TextureInst& texture_inst : to_replace) {
// TODO: Handle arrays
IR::Inst* const inst{texture_inst.inst};