mirrors/shadPS4
1
0
Fork 0
mirror of https://github.com/shadps4-emu/shadPS4.git synced 2025-08-05 17:02:40 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'shadps4-emu:main' into gcc-ci

Browse Source
This commit is contained in:
tomboylover93 2025-01-02 18:45:51 -03:00 committed by GitHub
commit b4f441e605
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
12 changed files with 255 additions and 121 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
src/core/libraries/kernel/memory.cpp
View File

@ -505,13 +505,13 @@ int PS4_SYSV_ABI posix_munmap(void* addr, size_t len) {
return result;
}
static constexpr int MAX_PTR_APERTURES = 3;
static constexpr int MAX_PRT_APERTURES = 3;
static constexpr VAddr PRT_AREA_START_ADDR = 0x1000000000;
static constexpr size_t PRT_AREA_SIZE = 0xec00000000;
static std::array<std::pair<VAddr, size_t>, MAX_PTR_APERTURES> PrtApertures{};
static std::array<std::pair<VAddr, size_t>, MAX_PRT_APERTURES> PrtApertures{};
int PS4_SYSV_ABI sceKernelSetPrtAperture(int id, VAddr address, size_t size) {
if (id < 0 || id >= MAX_PTR_APERTURES) {
if (id < 0 || id >= MAX_PRT_APERTURES) {
return ORBIS_KERNEL_ERROR_EINVAL;
}
@ -531,12 +531,12 @@ int PS4_SYSV_ABI sceKernelSetPrtAperture(int id, VAddr address, size_t size) {
return ORBIS_OK;
}
int PS4_SYSV_ABI sceKernelGetPrtAperture(int id, VAddr* addres, size_t* size) {
if (id < 0 || id >= MAX_PTR_APERTURES) {
int PS4_SYSV_ABI sceKernelGetPrtAperture(int id, VAddr* address, size_t* size) {
if (id < 0 || id >= MAX_PRT_APERTURES) {
return ORBIS_KERNEL_ERROR_EINVAL;
}
std::tie(*addres, *size) = PrtApertures[id];
std::tie(*address, *size) = PrtApertures[id];
return ORBIS_OK;
}

12
src/core/libraries/np_manager/np_manager.cpp
View File

@ -972,11 +972,8 @@ int PS4_SYSV_ABI sceNpGetGamePresenceStatusA() {
}
int PS4_SYSV_ABI sceNpGetNpId(OrbisUserServiceUserId user_id, OrbisNpId* np_id) {
LOG_INFO(Lib_NpManager, "user_id {}", user_id);
const auto name = Config::getUserName();
std::memset(np_id, 0, sizeof(OrbisNpId));
name.copy(np_id->handle.data, sizeof(np_id->handle.data));
return ORBIS_OK;
LOG_DEBUG(Lib_NpManager, "user_id {}", user_id);
return ORBIS_NP_ERROR_SIGNED_OUT;
}
int PS4_SYSV_ABI sceNpGetNpReachabilityState() {
@ -986,10 +983,7 @@ int PS4_SYSV_ABI sceNpGetNpReachabilityState() {
int PS4_SYSV_ABI sceNpGetOnlineId(s32 user_id, OrbisNpOnlineId* online_id) {
LOG_DEBUG(Lib_NpManager, "user_id {}", user_id);
const auto name = Config::getUserName();
std::memset(online_id, 0, sizeof(OrbisNpOnlineId));
name.copy(online_id->data, sizeof(online_id->data));
return ORBIS_OK;
return ORBIS_NP_ERROR_SIGNED_OUT;
}
int PS4_SYSV_ABI sceNpGetParentalControlInfo() {

4
src/core/libraries/pad/pad.cpp
View File

@ -104,8 +104,8 @@ int PS4_SYSV_ABI scePadGetControllerInformation(s32 handle, OrbisPadControllerIn
pInfo->touchPadInfo.pixelDensity = 1;
pInfo->touchPadInfo.resolution.x = 1920;
pInfo->touchPadInfo.resolution.y = 950;
pInfo->stickInfo.deadZoneLeft = 20;
pInfo->stickInfo.deadZoneRight = 20;
pInfo->stickInfo.deadZoneLeft = 2;
pInfo->stickInfo.deadZoneRight = 2;
pInfo->connectionType = ORBIS_PAD_PORT_TYPE_STANDARD;
pInfo->connectedCount = 1;
pInfo->connected = true;

6
src/shader_recompiler/frontend/translate/data_share.cpp
View File

@ -205,7 +205,6 @@ void Translator::DS_WRITE(int bit_size, bool is_signed, bool is_pair, bool strid
addr, ir.Imm32((u32(inst.control.ds.offset1) << 8u) + u32(inst.control.ds.offset0)));
ir.WriteShared(bit_size, ir.GetVectorReg(data0), addr0);
}
emit_ds_read_barrier = true;
}
void Translator::DS_SWIZZLE_B32(const GcnInst& inst) {
@ -222,11 +221,6 @@ void Translator::DS_SWIZZLE_B32(const GcnInst& inst) {
void Translator::DS_READ(int bit_size, bool is_signed, bool is_pair, bool stride64,
const GcnInst& inst) {
if (emit_ds_read_barrier && profile.needs_lds_barriers) {
ir.Barrier();
emit_ds_read_barrier = false;
}
const IR::U32 addr{ir.GetVectorReg(IR::VectorReg(inst.src[0].code))};
IR::VectorReg dst_reg{inst.dst[0].code};
if (is_pair) {

1
src/shader_recompiler/frontend/translate/translate.h
View File

@ -308,7 +308,6 @@ private:
const RuntimeInfo& runtime_info;
const Profile& profile;
bool opcode_missing = false;
bool emit_ds_read_barrier = false;
};
void Translate(IR::Block* block, u32 block_base, std::span<const GcnInst> inst_list, Info& info,

71
src/shader_recompiler/ir/passes/shared_memory_barrier_pass.cpp
View File

@ -8,6 +8,54 @@
namespace Shader::Optimization {
static void EmitBarrierInBlock(IR::Block* block) {
// This is inteded to insert a barrier when shared memory write and read
// occur in the same basic block. Also checks if branch depth is zero as
// we don't want to insert barrier in potentially divergent code.
bool emit_barrier_on_write = false;
bool emit_barrier_on_read = false;
const auto emit_barrier = [block](bool& emit_cond, IR::Inst& inst) {
if (emit_cond) {
IR::IREmitter ir{*block, IR::Block::InstructionList::s_iterator_to(inst)};
ir.Barrier();
emit_cond = false;
}
};
for (IR::Inst& inst : block->Instructions()) {
if (inst.GetOpcode() == IR::Opcode::LoadSharedU32 ||
inst.GetOpcode() == IR::Opcode::LoadSharedU64) {
emit_barrier(emit_barrier_on_read, inst);
emit_barrier_on_write = true;
}
if (inst.GetOpcode() == IR::Opcode::WriteSharedU32 ||
inst.GetOpcode() == IR::Opcode::WriteSharedU64) {
emit_barrier(emit_barrier_on_write, inst);
emit_barrier_on_read = true;
}
}
}
static void EmitBarrierInMergeBlock(const IR::AbstractSyntaxNode::Data& data) {
// Insert a barrier after divergent conditional blocks.
// This avoids potential softlocks and crashes when some threads
// initialize shared memory and others read from it.
const IR::U1 cond = data.if_node.cond;
const auto insert_barrier =
IR::BreadthFirstSearch(cond, [](IR::Inst* inst) -> std::optional<bool> {
if (inst->GetOpcode() == IR::Opcode::GetAttributeU32 &&
inst->Arg(0).Attribute() == IR::Attribute::LocalInvocationId) {
return true;
}
return std::nullopt;
});
if (insert_barrier) {
IR::Block* const merge = data.if_node.merge;
auto insert_point = std::ranges::find_if_not(merge->Instructions(), IR::IsPhi);
IR::IREmitter ir{*merge, insert_point};
ir.Barrier();
}
}
void SharedMemoryBarrierPass(IR::Program& program, const Profile& profile) {
if (!program.info.uses_shared || !profile.needs_lds_barriers) {
return;
@ -19,27 +67,12 @@ void SharedMemoryBarrierPass(IR::Program& program, const Profile& profile) {
--branch_depth;
continue;
}
if (node.type != Type::If) {
if (node.type == Type::If && branch_depth++ == 0) {
EmitBarrierInMergeBlock(node.data);
continue;
}
u32 curr_depth = branch_depth++;
if (curr_depth != 0) {
continue;
}
const IR::U1 cond = node.data.if_node.cond;
const auto insert_barrier =
IR::BreadthFirstSearch(cond, [](IR::Inst* inst) -> std::optional<bool> {
if (inst->GetOpcode() == IR::Opcode::GetAttributeU32 &&
inst->Arg(0).Attribute() == IR::Attribute::LocalInvocationId) {
return true;
}
return std::nullopt;
});
if (insert_barrier) {
IR::Block* const merge = node.data.if_node.merge;
auto insert_point = std::ranges::find_if_not(merge->Instructions(), IR::IsPhi);
IR::IREmitter ir{*merge, insert_point};
ir.Barrier();
if (node.type == Type::Block && branch_depth == 0) {
EmitBarrierInBlock(node.data.block);
}
}
}

147
src/video_core/buffer_cache/buffer_cache.cpp
View File

@ -259,7 +259,16 @@ void BufferCache::InlineData(VAddr address, const void* value, u32 num_bytes, bo
const BufferId buffer_id = FindBuffer(address, num_bytes);
return &slot_buffers[buffer_id];
}();
const vk::BufferMemoryBarrier2 buf_barrier = {
const vk::BufferMemoryBarrier2 pre_barrier = {
.srcStageMask = vk::PipelineStageFlagBits2::eAllCommands,
.srcAccessMask = vk::AccessFlagBits2::eMemoryRead,
.dstStageMask = vk::PipelineStageFlagBits2::eTransfer,
.dstAccessMask = vk::AccessFlagBits2::eTransferWrite,
.buffer = buffer->Handle(),
.offset = buffer->Offset(address),
.size = num_bytes,
};
const vk::BufferMemoryBarrier2 post_barrier = {
.srcStageMask = vk::PipelineStageFlagBits2::eTransfer,
.srcAccessMask = vk::AccessFlagBits2::eTransferWrite,
.dstStageMask = vk::PipelineStageFlagBits2::eAllCommands,
@ -271,9 +280,14 @@ void BufferCache::InlineData(VAddr address, const void* value, u32 num_bytes, bo
cmdbuf.pipelineBarrier2(vk::DependencyInfo{
.dependencyFlags = vk::DependencyFlagBits::eByRegion,
.bufferMemoryBarrierCount = 1,
.pBufferMemoryBarriers = &buf_barrier,
.pBufferMemoryBarriers = &pre_barrier,
});
cmdbuf.updateBuffer(buffer->Handle(), buffer->Offset(address), num_bytes, value);
cmdbuf.pipelineBarrier2(vk::DependencyInfo{
.dependencyFlags = vk::DependencyFlagBits::eByRegion,
.bufferMemoryBarrierCount = 1,
.pBufferMemoryBarriers = &post_barrier,
});
cmdbuf.updateBuffer(buffer->Handle(), buf_barrier.offset, num_bytes, value);
}
std::pair<Buffer*, u32> BufferCache::ObtainHostUBO(std::span<const u32> data) {
@ -465,21 +479,48 @@ void BufferCache::JoinOverlap(BufferId new_buffer_id, BufferId overlap_id,
};
scheduler.EndRendering();
const auto cmdbuf = scheduler.CommandBuffer();
static constexpr vk::MemoryBarrier READ_BARRIER{
.srcAccessMask = vk::AccessFlagBits::eMemoryWrite,
.dstAccessMask = vk::AccessFlagBits::eTransferRead | vk::AccessFlagBits::eTransferWrite,
const std::array pre_barriers = {
vk::BufferMemoryBarrier2{
.srcStageMask = vk::PipelineStageFlagBits2::eAllCommands,
.srcAccessMask = vk::AccessFlagBits2::eMemoryRead | vk::AccessFlagBits2::eMemoryWrite,
.dstStageMask = vk::PipelineStageFlagBits2::eTransfer,
.dstAccessMask = vk::AccessFlagBits2::eTransferRead,
.buffer = overlap.Handle(),
.offset = 0,
.size = overlap.SizeBytes(),
},
};
static constexpr vk::MemoryBarrier WRITE_BARRIER{
.srcAccessMask = vk::AccessFlagBits::eTransferWrite,
.dstAccessMask = vk::AccessFlagBits::eMemoryRead | vk::AccessFlagBits::eMemoryWrite,
const std::array post_barriers = {
vk::BufferMemoryBarrier2{
.srcStageMask = vk::PipelineStageFlagBits2::eTransfer,
.srcAccessMask = vk::AccessFlagBits2::eTransferRead,
.dstStageMask = vk::PipelineStageFlagBits2::eAllCommands,
.dstAccessMask = vk::AccessFlagBits2::eMemoryWrite,
.buffer = overlap.Handle(),
.offset = 0,
.size = overlap.SizeBytes(),
},
vk::BufferMemoryBarrier2{
.srcStageMask = vk::PipelineStageFlagBits2::eTransfer,
.srcAccessMask = vk::AccessFlagBits2::eTransferWrite,
.dstStageMask = vk::PipelineStageFlagBits2::eAllCommands,
.dstAccessMask = vk::AccessFlagBits2::eMemoryRead | vk::AccessFlagBits2::eMemoryWrite,
.buffer = new_buffer.Handle(),
.offset = dst_base_offset,
.size = overlap.SizeBytes(),
},
};
cmdbuf.pipelineBarrier(vk::PipelineStageFlagBits::eAllCommands,
vk::PipelineStageFlagBits::eTransfer, vk::DependencyFlagBits::eByRegion,
READ_BARRIER, {}, {});
cmdbuf.copyBuffer(overlap.buffer, new_buffer.buffer, copy);
cmdbuf.pipelineBarrier(vk::PipelineStageFlagBits::eTransfer,
vk::PipelineStageFlagBits::eAllCommands,
vk::DependencyFlagBits::eByRegion, WRITE_BARRIER, {}, {});
cmdbuf.pipelineBarrier2(vk::DependencyInfo{
.dependencyFlags = vk::DependencyFlagBits::eByRegion,
.bufferMemoryBarrierCount = 1,
.pBufferMemoryBarriers = pre_barriers.data(),
});
cmdbuf.copyBuffer(overlap.Handle(), new_buffer.Handle(), copy);
cmdbuf.pipelineBarrier2(vk::DependencyInfo{
.dependencyFlags = vk::DependencyFlagBits::eByRegion,
.bufferMemoryBarrierCount = static_cast<u32>(post_barriers.size()),
.pBufferMemoryBarriers = post_barriers.data(),
});
DeleteBuffer(overlap_id);
}
@ -583,21 +624,35 @@ void BufferCache::SynchronizeBuffer(Buffer& buffer, VAddr device_addr, u32 size,
}
scheduler.EndRendering();
const auto cmdbuf = scheduler.CommandBuffer();
static constexpr vk::MemoryBarrier READ_BARRIER{
.srcAccessMask = vk::AccessFlagBits::eMemoryWrite,
.dstAccessMask = vk::AccessFlagBits::eTransferRead | vk::AccessFlagBits::eTransferWrite,
const vk::BufferMemoryBarrier2 pre_barrier = {
.srcStageMask = vk::PipelineStageFlagBits2::eAllCommands,
.srcAccessMask = vk::AccessFlagBits2::eMemoryRead,
.dstStageMask = vk::PipelineStageFlagBits2::eTransfer,
.dstAccessMask = vk::AccessFlagBits2::eTransferWrite,
.buffer = buffer.Handle(),
.offset = 0,
.size = buffer.SizeBytes(),
};
static constexpr vk::MemoryBarrier WRITE_BARRIER{
.srcAccessMask = vk::AccessFlagBits::eTransferWrite,
.dstAccessMask = vk::AccessFlagBits::eMemoryRead | vk::AccessFlagBits::eMemoryWrite,
const vk::BufferMemoryBarrier2 post_barrier = {
.srcStageMask = vk::PipelineStageFlagBits2::eTransfer,
.srcAccessMask = vk::AccessFlagBits2::eTransferWrite,
.dstStageMask = vk::PipelineStageFlagBits2::eAllCommands,
.dstAccessMask = vk::AccessFlagBits2::eMemoryRead | vk::AccessFlagBits2::eMemoryWrite,
.buffer = buffer.Handle(),
.offset = 0,
.size = buffer.SizeBytes(),
};
cmdbuf.pipelineBarrier(vk::PipelineStageFlagBits::eAllCommands,
vk::PipelineStageFlagBits::eTransfer, vk::DependencyFlagBits::eByRegion,
READ_BARRIER, {}, {});
cmdbuf.pipelineBarrier2(vk::DependencyInfo{
.dependencyFlags = vk::DependencyFlagBits::eByRegion,
.bufferMemoryBarrierCount = 1,
.pBufferMemoryBarriers = &pre_barrier,
});
cmdbuf.copyBuffer(src_buffer, buffer.buffer, copies);
cmdbuf.pipelineBarrier(vk::PipelineStageFlagBits::eTransfer,
vk::PipelineStageFlagBits::eAllCommands,
vk::DependencyFlagBits::eByRegion, WRITE_BARRIER, {}, {});
cmdbuf.pipelineBarrier2(vk::DependencyInfo{
.dependencyFlags = vk::DependencyFlagBits::eByRegion,
.bufferMemoryBarrierCount = 1,
.pBufferMemoryBarriers = &post_barrier,
});
}
bool BufferCache::SynchronizeBufferFromImage(Buffer& buffer, VAddr device_addr, u32 size) {
@ -647,10 +702,42 @@ bool BufferCache::SynchronizeBufferFromImage(Buffer& buffer, VAddr device_addr,
}
if (!copies.empty()) {
scheduler.EndRendering();
image.Transit(vk::ImageLayout::eTransferSrcOptimal, vk::AccessFlagBits2::eTransferRead, {});
const vk::BufferMemoryBarrier2 pre_barrier = {
.srcStageMask = vk::PipelineStageFlagBits2::eAllCommands,
.srcAccessMask = vk::AccessFlagBits2::eMemoryRead,
.dstStageMask = vk::PipelineStageFlagBits2::eTransfer,
.dstAccessMask = vk::AccessFlagBits2::eTransferWrite,
.buffer = buffer.Handle(),
.offset = max_offset - size,
.size = size,
};
const vk::BufferMemoryBarrier2 post_barrier = {
.srcStageMask = vk::PipelineStageFlagBits2::eTransfer,
.srcAccessMask = vk::AccessFlagBits2::eTransferWrite,
.dstStageMask = vk::PipelineStageFlagBits2::eAllCommands,
.dstAccessMask = vk::AccessFlagBits2::eMemoryRead,
.buffer = buffer.Handle(),
.offset = max_offset - size,
.size = size,
};
auto barriers = image.GetBarriers(vk::ImageLayout::eTransferSrcOptimal,
vk::AccessFlagBits2::eTransferRead,
vk::PipelineStageFlagBits2::eTransfer, {});
const auto cmdbuf = scheduler.CommandBuffer();
cmdbuf.copyImageToBuffer(image.image, vk::ImageLayout::eTransferSrcOptimal, buffer.buffer,
cmdbuf.pipelineBarrier2(vk::DependencyInfo{
.dependencyFlags = vk::DependencyFlagBits::eByRegion,
.bufferMemoryBarrierCount = 1,
.pBufferMemoryBarriers = &pre_barrier,
.imageMemoryBarrierCount = static_cast<u32>(barriers.size()),
.pImageMemoryBarriers = barriers.data(),
});
cmdbuf.copyImageToBuffer(image.image, vk::ImageLayout::eTransferSrcOptimal, buffer.Handle(),
copies);
cmdbuf.pipelineBarrier2(vk::DependencyInfo{
.dependencyFlags = vk::DependencyFlagBits::eByRegion,
.bufferMemoryBarrierCount = 1,
.pBufferMemoryBarriers = &post_barrier,
});
}
return true;
}

4
src/video_core/renderer_vulkan/liverpool_to_vk.cpp
View File

@ -697,8 +697,8 @@ static const size_t amd_gpu_number_format_bit_size = 4; // All values are under
static size_t GetSurfaceFormatTableIndex(AmdGpu::DataFormat data_format,
AmdGpu::NumberFormat num_format) {
DEBUG_ASSERT(data_format < 1 << amd_gpu_data_format_bit_size);
DEBUG_ASSERT(num_format < 1 << amd_gpu_number_format_bit_size);
DEBUG_ASSERT(u32(data_format) < 1 << amd_gpu_data_format_bit_size);
DEBUG_ASSERT(u32(num_format) < 1 << amd_gpu_number_format_bit_size);
size_t result = static_cast<size_t>(num_format) |
(static_cast<size_t>(data_format) << amd_gpu_number_format_bit_size);
return result;

8
src/video_core/renderer_vulkan/vk_rasterizer.cpp
View File

@ -562,6 +562,12 @@ void Rasterizer::BindBuffers(const Shader::Info& stage, Shader::Backend::Binding
push_data.AddOffset(binding.buffer, adjust);
buffer_infos.emplace_back(vk_buffer->Handle(), offset_aligned,
vsharp.GetSize() + adjust);
if (auto barrier =
vk_buffer->GetBarrier(desc.is_written ? vk::AccessFlagBits2::eShaderWrite
: vk::AccessFlagBits2::eShaderRead,
vk::PipelineStageFlagBits2::eAllCommands)) {
buffer_barriers.emplace_back(*barrier);
}
}
set_writes.push_back({
@ -600,7 +606,7 @@ void Rasterizer::BindBuffers(const Shader::Info& stage, Shader::Backend::Binding
if (auto barrier =
vk_buffer->GetBarrier(desc.is_written ? vk::AccessFlagBits2::eShaderWrite
: vk::AccessFlagBits2::eShaderRead,
vk::PipelineStageFlagBits2::eComputeShader)) {
vk::PipelineStageFlagBits2::eAllCommands)) {
buffer_barriers.emplace_back(*barrier);
}
if (desc.is_written) {

45
src/video_core/texture_cache/texture_cache.cpp
View File

@ -542,31 +542,62 @@ void TextureCache::RefreshImage(Image& image, Vulkan::Scheduler* custom_schedule
sched_ptr->EndRendering();
const auto cmdbuf = sched_ptr->CommandBuffer();
image.Transit(vk::ImageLayout::eTransferDstOptimal, vk::AccessFlagBits2::eTransferWrite, {},
cmdbuf);
const VAddr image_addr = image.info.guest_address;
const size_t image_size = image.info.guest_size_bytes;
const auto [vk_buffer, buf_offset] =
buffer_cache.ObtainViewBuffer(image_addr, image_size, is_gpu_dirty);
// The obtained buffer may be written by a shader so we need to emit a barrier to prevent RAW
// hazard
if (auto barrier = vk_buffer->GetBarrier(vk::AccessFlagBits2::eTransferRead,
vk::PipelineStageFlagBits2::eTransfer)) {
const auto dependencies = vk::DependencyInfo{
cmdbuf.pipelineBarrier2(vk::DependencyInfo{
.dependencyFlags = vk::DependencyFlagBits::eByRegion,
.bufferMemoryBarrierCount = 1,
.pBufferMemoryBarriers = &barrier.value(),
};
cmdbuf.pipelineBarrier2(dependencies);
});
}
const auto [buffer, offset] = tile_manager.TryDetile(vk_buffer->Handle(), buf_offset, image);
const auto [buffer, offset] =
tile_manager.TryDetile(vk_buffer->Handle(), buf_offset, image.info);
for (auto& copy : image_copy) {
copy.bufferOffset += offset;
}
const vk::BufferMemoryBarrier2 pre_barrier{
.srcStageMask = vk::PipelineStageFlagBits2::eAllCommands,
.srcAccessMask = vk::AccessFlagBits2::eMemoryWrite,
.dstStageMask = vk::PipelineStageFlagBits2::eTransfer,
.dstAccessMask = vk::AccessFlagBits2::eTransferRead,
.buffer = buffer,
.offset = offset,
.size = image_size,
};
const vk::BufferMemoryBarrier2 post_barrier{
.srcStageMask = vk::PipelineStageFlagBits2::eTransfer,
.srcAccessMask = vk::AccessFlagBits2::eTransferWrite,
.dstStageMask = vk::PipelineStageFlagBits2::eAllCommands,
.dstAccessMask = vk::AccessFlagBits2::eMemoryRead | vk::AccessFlagBits2::eMemoryWrite,
.buffer = buffer,
.offset = offset,
.size = image_size,
};
const auto image_barriers =
image.GetBarriers(vk::ImageLayout::eTransferDstOptimal, vk::AccessFlagBits2::eTransferWrite,
vk::PipelineStageFlagBits2::eTransfer, {});
cmdbuf.pipelineBarrier2(vk::DependencyInfo{
.dependencyFlags = vk::DependencyFlagBits::eByRegion,
.bufferMemoryBarrierCount = 1,
.pBufferMemoryBarriers = &pre_barrier,
.imageMemoryBarrierCount = static_cast<u32>(image_barriers.size()),
.pImageMemoryBarriers = image_barriers.data(),
});
cmdbuf.copyBufferToImage(buffer, image.image, vk::ImageLayout::eTransferDstOptimal, image_copy);
cmdbuf.pipelineBarrier2(vk::DependencyInfo{
.dependencyFlags = vk::DependencyFlagBits::eByRegion,
.bufferMemoryBarrierCount = 1,
.pBufferMemoryBarriers = &post_barrier,
});
image.flags &= ~ImageFlagBits::Dirty;
}

59
src/video_core/texture_cache/tile_manager.cpp
View File

@ -4,6 +4,7 @@
#include "video_core/renderer_vulkan/vk_instance.h"
#include "video_core/renderer_vulkan/vk_scheduler.h"
#include "video_core/renderer_vulkan/vk_shader_util.h"
#include "video_core/texture_cache/image_info.h"
#include "video_core/texture_cache/image_view.h"
#include "video_core/texture_cache/tile_manager.h"
@ -86,10 +87,10 @@ static vk::Format DemoteImageFormatForDetiling(vk::Format format) {
return format;
}
const DetilerContext* TileManager::GetDetiler(const Image& image) const {
const auto format = DemoteImageFormatForDetiling(image.info.pixel_format);
const DetilerContext* TileManager::GetDetiler(const ImageInfo& info) const {
const auto format = DemoteImageFormatForDetiling(info.pixel_format);
switch (image.info.tiling_mode) {
switch (info.tiling_mode) {
case AmdGpu::TilingMode::Texture_MicroTiled:
switch (format) {
case vk::Format::eR8Uint:
@ -258,23 +259,23 @@ void TileManager::FreeBuffer(ScratchBuffer buffer) {
}
std::pair<vk::Buffer, u32> TileManager::TryDetile(vk::Buffer in_buffer, u32 in_offset,
Image& image) {
if (!image.info.props.is_tiled) {
const ImageInfo& info) {
if (!info.props.is_tiled) {
return {in_buffer, in_offset};
}
const auto* detiler = GetDetiler(image);
const auto* detiler = GetDetiler(info);
if (!detiler) {
if (image.info.tiling_mode != AmdGpu::TilingMode::Texture_MacroTiled &&
image.info.tiling_mode != AmdGpu::TilingMode::Display_MacroTiled &&
image.info.tiling_mode != AmdGpu::TilingMode::Depth_MacroTiled) {
if (info.tiling_mode != AmdGpu::TilingMode::Texture_MacroTiled &&
info.tiling_mode != AmdGpu::TilingMode::Display_MacroTiled &&
info.tiling_mode != AmdGpu::TilingMode::Depth_MacroTiled) {
LOG_ERROR(Render_Vulkan, "Unsupported tiled image: {} ({})",
vk::to_string(image.info.pixel_format), NameOf(image.info.tiling_mode));
vk::to_string(info.pixel_format), NameOf(info.tiling_mode));
}
return {in_buffer, in_offset};
}
const u32 image_size = image.info.guest_size_bytes;
const u32 image_size = info.guest_size_bytes;
// Prepare output buffer
auto out_buffer = AllocBuffer(image_size, true);
@ -317,22 +318,21 @@ std::pair<vk::Buffer, u32> TileManager::TryDetile(vk::Buffer in_buffer, u32 in_o
set_writes);
DetilerParams params;
params.num_levels = image.info.resources.levels;
params.pitch0 = image.info.pitch >> (image.info.props.is_block ? 2u : 0u);
params.height = image.info.size.height;
if (image.info.tiling_mode == AmdGpu::TilingMode::Texture_Volume) {
ASSERT(image.info.resources.levels == 1);
ASSERT(image.info.num_bits >= 32);
const auto tiles_per_row = image.info.pitch / 8u;
const auto tiles_per_slice = tiles_per_row * ((image.info.size.height + 7u) / 8u);
params.num_levels = info.resources.levels;
params.pitch0 = info.pitch >> (info.props.is_block ? 2u : 0u);
params.height = info.size.height;
if (info.tiling_mode == AmdGpu::TilingMode::Texture_Volume) {
ASSERT(info.resources.levels == 1);
ASSERT(info.num_bits >= 32);
const auto tiles_per_row = info.pitch / 8u;
const auto tiles_per_slice = tiles_per_row * ((info.size.height + 7u) / 8u);
params.sizes[0] = tiles_per_row;
params.sizes[1] = tiles_per_slice;
} else {
ASSERT(image.info.resources.levels <= 14);
ASSERT(info.resources.levels <= 14);
std::memset(&params.sizes, 0, sizeof(params.sizes));
for (int m = 0; m < image.info.resources.levels; ++m) {
params.sizes[m] = image.info.mips_layout[m].size * image.info.resources.layers +
for (int m = 0; m < info.resources.levels; ++m) {
params.sizes[m] = info.mips_layout[m].size * info.resources.layers +
(m > 0 ? params.sizes[m - 1] : 0);
}
}
@ -341,20 +341,9 @@ std::pair<vk::Buffer, u32> TileManager::TryDetile(vk::Buffer in_buffer, u32 in_o
&params);
ASSERT((image_size % 64) == 0);
const auto bpp = image.info.num_bits * (image.info.props.is_block ? 16u : 1u);
const auto bpp = info.num_bits * (info.props.is_block ? 16u : 1u);
const auto num_tiles = image_size / (64 * (bpp / 8));
cmdbuf.dispatch(num_tiles, 1, 1);
const vk::BufferMemoryBarrier post_barrier{
.srcAccessMask = vk::AccessFlagBits::eShaderWrite,
.dstAccessMask = vk::AccessFlagBits::eTransferRead,
.buffer = out_buffer.first,
.size = image_size,
};
cmdbuf.pipelineBarrier(vk::PipelineStageFlagBits::eComputeShader,
vk::PipelineStageFlagBits::eTransfer, vk::DependencyFlagBits::eByRegion,
{}, post_barrier, {});
return {out_buffer.first, 0};
}

7
src/video_core/texture_cache/tile_manager.h
View File

@ -5,11 +5,11 @@
#include "common/types.h"
#include "video_core/buffer_cache/buffer.h"
#include "video_core/texture_cache/image.h"
namespace VideoCore {
class TextureCache;
struct ImageInfo;
enum DetilerType : u32 {
Micro8x1,
@ -36,14 +36,15 @@ public:
TileManager(const Vulkan::Instance& instance, Vulkan::Scheduler& scheduler);
~TileManager();
std::pair<vk::Buffer, u32> TryDetile(vk::Buffer in_buffer, u32 in_offset, Image& image);
std::pair<vk::Buffer, u32> TryDetile(vk::Buffer in_buffer, u32 in_offset,
const ImageInfo& info);
ScratchBuffer AllocBuffer(u32 size, bool is_storage = false);
void Upload(ScratchBuffer buffer, const void* data, size_t size);
void FreeBuffer(ScratchBuffer buffer);
private:
const DetilerContext* GetDetiler(const Image& image) const;
const DetilerContext* GetDetiler(const ImageInfo& info) const;
private:
const Vulkan::Instance& instance;