forked from suyu/suyu
buffer_base: Add support for cached CPU writes
Some games usually write memory pages currently used by the GPU, causing rendering issues (e.g. flashing geometry and shadows on Link's Awakening). To workaround this issue, Guest CPU writes are delayed until the command buffer finishes processing, but the pages are updated immediately. The overall behavior is: - CPU writes are cached until they are flushed, they update the page state, but don't change the modification state. Cached writes stop pages from being flushed, in case games have meaningful data in it. - Command processing writes (e.g. push constants) update the page state and are marked to the command processor as dirty. They don't remove the state of cached writes.
This commit is contained in:
parent
c86d770af9
commit
9d8ca6cc4a
1 changed files with 145 additions and 61 deletions
|
@ -19,6 +19,7 @@ namespace VideoCommon {
|
||||||
|
|
||||||
enum class BufferFlagBits {
|
enum class BufferFlagBits {
|
||||||
Picked = 1 << 0,
|
Picked = 1 << 0,
|
||||||
|
CachedWrites = 1 << 1,
|
||||||
};
|
};
|
||||||
DECLARE_ENUM_FLAG_OPERATORS(BufferFlagBits)
|
DECLARE_ENUM_FLAG_OPERATORS(BufferFlagBits)
|
||||||
|
|
||||||
|
@ -40,7 +41,7 @@ class BufferBase {
|
||||||
static constexpr u64 BYTES_PER_WORD = PAGES_PER_WORD * BYTES_PER_PAGE;
|
static constexpr u64 BYTES_PER_WORD = PAGES_PER_WORD * BYTES_PER_PAGE;
|
||||||
|
|
||||||
/// Vector tracking modified pages tightly packed with small vector optimization
|
/// Vector tracking modified pages tightly packed with small vector optimization
|
||||||
union WrittenWords {
|
union WordsArray {
|
||||||
/// Returns the pointer to the words state
|
/// Returns the pointer to the words state
|
||||||
[[nodiscard]] const u64* Pointer(bool is_short) const noexcept {
|
[[nodiscard]] const u64* Pointer(bool is_short) const noexcept {
|
||||||
return is_short ? &stack : heap;
|
return is_short ? &stack : heap;
|
||||||
|
@ -55,49 +56,59 @@ class BufferBase {
|
||||||
u64* heap; ///< Not-small buffers pointer to the storage
|
u64* heap; ///< Not-small buffers pointer to the storage
|
||||||
};
|
};
|
||||||
|
|
||||||
struct GpuCpuWords {
|
struct Words {
|
||||||
explicit GpuCpuWords() = default;
|
explicit Words() = default;
|
||||||
explicit GpuCpuWords(u64 size_bytes_) : size_bytes{size_bytes_} {
|
explicit Words(u64 size_bytes_) : size_bytes{size_bytes_} {
|
||||||
if (IsShort()) {
|
if (IsShort()) {
|
||||||
cpu.stack = ~u64{0};
|
cpu.stack = ~u64{0};
|
||||||
gpu.stack = 0;
|
gpu.stack = 0;
|
||||||
|
cached_cpu.stack = 0;
|
||||||
|
untracked.stack = ~u64{0};
|
||||||
} else {
|
} else {
|
||||||
// Share allocation between CPU and GPU pages and set their default values
|
// Share allocation between CPU and GPU pages and set their default values
|
||||||
const size_t num_words = NumWords();
|
const size_t num_words = NumWords();
|
||||||
u64* const alloc = new u64[num_words * 2];
|
u64* const alloc = new u64[num_words * 4];
|
||||||
cpu.heap = alloc;
|
cpu.heap = alloc;
|
||||||
gpu.heap = alloc + num_words;
|
gpu.heap = alloc + num_words;
|
||||||
|
cached_cpu.heap = alloc + num_words * 2;
|
||||||
|
untracked.heap = alloc + num_words * 3;
|
||||||
std::fill_n(cpu.heap, num_words, ~u64{0});
|
std::fill_n(cpu.heap, num_words, ~u64{0});
|
||||||
std::fill_n(gpu.heap, num_words, 0);
|
std::fill_n(gpu.heap, num_words, 0);
|
||||||
|
std::fill_n(cached_cpu.heap, num_words, 0);
|
||||||
|
std::fill_n(untracked.heap, num_words, ~u64{0});
|
||||||
}
|
}
|
||||||
// Clean up tailing bits
|
// Clean up tailing bits
|
||||||
const u64 last_local_page =
|
const u64 last_word_size = size_bytes % BYTES_PER_WORD;
|
||||||
Common::DivCeil(size_bytes % BYTES_PER_WORD, BYTES_PER_PAGE);
|
const u64 last_local_page = Common::DivCeil(last_word_size, BYTES_PER_PAGE);
|
||||||
const u64 shift = (PAGES_PER_WORD - last_local_page) % PAGES_PER_WORD;
|
const u64 shift = (PAGES_PER_WORD - last_local_page) % PAGES_PER_WORD;
|
||||||
u64& last_word = cpu.Pointer(IsShort())[NumWords() - 1];
|
const u64 last_word = (~u64{0} << shift) >> shift;
|
||||||
last_word = (last_word << shift) >> shift;
|
cpu.Pointer(IsShort())[NumWords() - 1] = last_word;
|
||||||
|
untracked.Pointer(IsShort())[NumWords() - 1] = last_word;
|
||||||
}
|
}
|
||||||
|
|
||||||
~GpuCpuWords() {
|
~Words() {
|
||||||
Release();
|
Release();
|
||||||
}
|
}
|
||||||
|
|
||||||
GpuCpuWords& operator=(GpuCpuWords&& rhs) noexcept {
|
Words& operator=(Words&& rhs) noexcept {
|
||||||
Release();
|
Release();
|
||||||
size_bytes = rhs.size_bytes;
|
size_bytes = rhs.size_bytes;
|
||||||
cpu = rhs.cpu;
|
cpu = rhs.cpu;
|
||||||
gpu = rhs.gpu;
|
gpu = rhs.gpu;
|
||||||
|
cached_cpu = rhs.cached_cpu;
|
||||||
|
untracked = rhs.untracked;
|
||||||
rhs.cpu.heap = nullptr;
|
rhs.cpu.heap = nullptr;
|
||||||
return *this;
|
return *this;
|
||||||
}
|
}
|
||||||
|
|
||||||
GpuCpuWords(GpuCpuWords&& rhs) noexcept
|
Words(Words&& rhs) noexcept
|
||||||
: size_bytes{rhs.size_bytes}, cpu{rhs.cpu}, gpu{rhs.gpu} {
|
: size_bytes{rhs.size_bytes}, cpu{rhs.cpu}, gpu{rhs.gpu},
|
||||||
|
cached_cpu{rhs.cached_cpu}, untracked{rhs.untracked} {
|
||||||
rhs.cpu.heap = nullptr;
|
rhs.cpu.heap = nullptr;
|
||||||
}
|
}
|
||||||
|
|
||||||
GpuCpuWords& operator=(const GpuCpuWords&) = delete;
|
Words& operator=(const Words&) = delete;
|
||||||
GpuCpuWords(const GpuCpuWords&) = delete;
|
Words(const Words&) = delete;
|
||||||
|
|
||||||
/// Returns true when the buffer fits in the small vector optimization
|
/// Returns true when the buffer fits in the small vector optimization
|
||||||
[[nodiscard]] bool IsShort() const noexcept {
|
[[nodiscard]] bool IsShort() const noexcept {
|
||||||
|
@ -118,8 +129,17 @@ class BufferBase {
|
||||||
}
|
}
|
||||||
|
|
||||||
u64 size_bytes = 0;
|
u64 size_bytes = 0;
|
||||||
WrittenWords cpu;
|
WordsArray cpu;
|
||||||
WrittenWords gpu;
|
WordsArray gpu;
|
||||||
|
WordsArray cached_cpu;
|
||||||
|
WordsArray untracked;
|
||||||
|
};
|
||||||
|
|
||||||
|
enum class Type {
|
||||||
|
CPU,
|
||||||
|
GPU,
|
||||||
|
CachedCPU,
|
||||||
|
Untracked,
|
||||||
};
|
};
|
||||||
|
|
||||||
public:
|
public:
|
||||||
|
@ -132,68 +152,93 @@ public:
|
||||||
BufferBase& operator=(const BufferBase&) = delete;
|
BufferBase& operator=(const BufferBase&) = delete;
|
||||||
BufferBase(const BufferBase&) = delete;
|
BufferBase(const BufferBase&) = delete;
|
||||||
|
|
||||||
|
BufferBase& operator=(BufferBase&&) = default;
|
||||||
|
BufferBase(BufferBase&&) = default;
|
||||||
|
|
||||||
/// Returns the inclusive CPU modified range in a begin end pair
|
/// Returns the inclusive CPU modified range in a begin end pair
|
||||||
[[nodiscard]] std::pair<u64, u64> ModifiedCpuRegion(VAddr query_cpu_addr,
|
[[nodiscard]] std::pair<u64, u64> ModifiedCpuRegion(VAddr query_cpu_addr,
|
||||||
u64 query_size) const noexcept {
|
u64 query_size) const noexcept {
|
||||||
const u64 offset = query_cpu_addr - cpu_addr;
|
const u64 offset = query_cpu_addr - cpu_addr;
|
||||||
return ModifiedRegion<false>(offset, query_size);
|
return ModifiedRegion<Type::CPU>(offset, query_size);
|
||||||
}
|
}
|
||||||
|
|
||||||
/// Returns the inclusive GPU modified range in a begin end pair
|
/// Returns the inclusive GPU modified range in a begin end pair
|
||||||
[[nodiscard]] std::pair<u64, u64> ModifiedGpuRegion(VAddr query_cpu_addr,
|
[[nodiscard]] std::pair<u64, u64> ModifiedGpuRegion(VAddr query_cpu_addr,
|
||||||
u64 query_size) const noexcept {
|
u64 query_size) const noexcept {
|
||||||
const u64 offset = query_cpu_addr - cpu_addr;
|
const u64 offset = query_cpu_addr - cpu_addr;
|
||||||
return ModifiedRegion<true>(offset, query_size);
|
return ModifiedRegion<Type::GPU>(offset, query_size);
|
||||||
}
|
}
|
||||||
|
|
||||||
/// Returns true if a region has been modified from the CPU
|
/// Returns true if a region has been modified from the CPU
|
||||||
[[nodiscard]] bool IsRegionCpuModified(VAddr query_cpu_addr, u64 query_size) const noexcept {
|
[[nodiscard]] bool IsRegionCpuModified(VAddr query_cpu_addr, u64 query_size) const noexcept {
|
||||||
const u64 offset = query_cpu_addr - cpu_addr;
|
const u64 offset = query_cpu_addr - cpu_addr;
|
||||||
return IsRegionModified<false>(offset, query_size);
|
return IsRegionModified<Type::CPU>(offset, query_size);
|
||||||
}
|
}
|
||||||
|
|
||||||
/// Returns true if a region has been modified from the GPU
|
/// Returns true if a region has been modified from the GPU
|
||||||
[[nodiscard]] bool IsRegionGpuModified(VAddr query_cpu_addr, u64 query_size) const noexcept {
|
[[nodiscard]] bool IsRegionGpuModified(VAddr query_cpu_addr, u64 query_size) const noexcept {
|
||||||
const u64 offset = query_cpu_addr - cpu_addr;
|
const u64 offset = query_cpu_addr - cpu_addr;
|
||||||
return IsRegionModified<true>(offset, query_size);
|
return IsRegionModified<Type::GPU>(offset, query_size);
|
||||||
}
|
}
|
||||||
|
|
||||||
/// Mark region as CPU modified, notifying the rasterizer about this change
|
/// Mark region as CPU modified, notifying the rasterizer about this change
|
||||||
void MarkRegionAsCpuModified(VAddr dirty_cpu_addr, u64 size) {
|
void MarkRegionAsCpuModified(VAddr dirty_cpu_addr, u64 size) {
|
||||||
ChangeRegionState<true, true>(words.cpu, dirty_cpu_addr, size);
|
ChangeRegionState<Type::CPU, true>(dirty_cpu_addr, size);
|
||||||
}
|
}
|
||||||
|
|
||||||
/// Unmark region as CPU modified, notifying the rasterizer about this change
|
/// Unmark region as CPU modified, notifying the rasterizer about this change
|
||||||
void UnmarkRegionAsCpuModified(VAddr dirty_cpu_addr, u64 size) {
|
void UnmarkRegionAsCpuModified(VAddr dirty_cpu_addr, u64 size) {
|
||||||
ChangeRegionState<false, true>(words.cpu, dirty_cpu_addr, size);
|
ChangeRegionState<Type::CPU, false>(dirty_cpu_addr, size);
|
||||||
}
|
}
|
||||||
|
|
||||||
/// Mark region as modified from the host GPU
|
/// Mark region as modified from the host GPU
|
||||||
void MarkRegionAsGpuModified(VAddr dirty_cpu_addr, u64 size) noexcept {
|
void MarkRegionAsGpuModified(VAddr dirty_cpu_addr, u64 size) noexcept {
|
||||||
ChangeRegionState<true, false>(words.gpu, dirty_cpu_addr, size);
|
ChangeRegionState<Type::GPU, true>(dirty_cpu_addr, size);
|
||||||
}
|
}
|
||||||
|
|
||||||
/// Unmark region as modified from the host GPU
|
/// Unmark region as modified from the host GPU
|
||||||
void UnmarkRegionAsGpuModified(VAddr dirty_cpu_addr, u64 size) noexcept {
|
void UnmarkRegionAsGpuModified(VAddr dirty_cpu_addr, u64 size) noexcept {
|
||||||
ChangeRegionState<false, false>(words.gpu, dirty_cpu_addr, size);
|
ChangeRegionState<Type::GPU, false>(dirty_cpu_addr, size);
|
||||||
|
}
|
||||||
|
|
||||||
|
/// Mark region as modified from the CPU
|
||||||
|
/// but don't mark it as modified until FlusHCachedWrites is called.
|
||||||
|
void CachedCpuWrite(VAddr dirty_cpu_addr, u64 size) {
|
||||||
|
flags |= BufferFlagBits::CachedWrites;
|
||||||
|
ChangeRegionState<Type::CachedCPU, true>(dirty_cpu_addr, size);
|
||||||
|
}
|
||||||
|
|
||||||
|
/// Flushes cached CPU writes, and notify the rasterizer about the deltas
|
||||||
|
void FlushCachedWrites() noexcept {
|
||||||
|
flags &= ~BufferFlagBits::CachedWrites;
|
||||||
|
const u64 num_words = NumWords();
|
||||||
|
const u64* const cached_words = Array<Type::CachedCPU>();
|
||||||
|
u64* const untracked_words = Array<Type::Untracked>();
|
||||||
|
u64* const cpu_words = Array<Type::CPU>();
|
||||||
|
for (u64 word_index = 0; word_index < num_words; ++word_index) {
|
||||||
|
const u64 cached_bits = cached_words[word_index];
|
||||||
|
NotifyRasterizer<false>(word_index, untracked_words[word_index], cached_bits);
|
||||||
|
untracked_words[word_index] |= cached_bits;
|
||||||
|
cpu_words[word_index] |= cached_bits;
|
||||||
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
/// Call 'func' for each CPU modified range and unmark those pages as CPU modified
|
/// Call 'func' for each CPU modified range and unmark those pages as CPU modified
|
||||||
template <typename Func>
|
template <typename Func>
|
||||||
void ForEachUploadRange(VAddr query_cpu_range, u64 size, Func&& func) {
|
void ForEachUploadRange(VAddr query_cpu_range, u64 size, Func&& func) {
|
||||||
ForEachModifiedRange<false, true>(query_cpu_range, size, func);
|
ForEachModifiedRange<Type::CPU>(query_cpu_range, size, func);
|
||||||
}
|
}
|
||||||
|
|
||||||
/// Call 'func' for each GPU modified range and unmark those pages as GPU modified
|
/// Call 'func' for each GPU modified range and unmark those pages as GPU modified
|
||||||
template <typename Func>
|
template <typename Func>
|
||||||
void ForEachDownloadRange(VAddr query_cpu_range, u64 size, Func&& func) {
|
void ForEachDownloadRange(VAddr query_cpu_range, u64 size, Func&& func) {
|
||||||
ForEachModifiedRange<true, false>(query_cpu_range, size, func);
|
ForEachModifiedRange<Type::GPU>(query_cpu_range, size, func);
|
||||||
}
|
}
|
||||||
|
|
||||||
/// Call 'func' for each GPU modified range and unmark those pages as GPU modified
|
/// Call 'func' for each GPU modified range and unmark those pages as GPU modified
|
||||||
template <typename Func>
|
template <typename Func>
|
||||||
void ForEachDownloadRange(Func&& func) {
|
void ForEachDownloadRange(Func&& func) {
|
||||||
ForEachModifiedRange<true, false>(cpu_addr, SizeBytes(), func);
|
ForEachModifiedRange<Type::GPU>(cpu_addr, SizeBytes(), func);
|
||||||
}
|
}
|
||||||
|
|
||||||
/// Mark buffer as picked
|
/// Mark buffer as picked
|
||||||
|
@ -216,6 +261,11 @@ public:
|
||||||
return True(flags & BufferFlagBits::Picked);
|
return True(flags & BufferFlagBits::Picked);
|
||||||
}
|
}
|
||||||
|
|
||||||
|
/// Returns true when the buffer has pending cached writes
|
||||||
|
[[nodiscard]] bool HasCachedWrites() const noexcept {
|
||||||
|
return True(flags & BufferFlagBits::CachedWrites);
|
||||||
|
}
|
||||||
|
|
||||||
/// Returns the base CPU address of the buffer
|
/// Returns the base CPU address of the buffer
|
||||||
[[nodiscard]] VAddr CpuAddr() const noexcept {
|
[[nodiscard]] VAddr CpuAddr() const noexcept {
|
||||||
return cpu_addr;
|
return cpu_addr;
|
||||||
|
@ -233,26 +283,48 @@ public:
|
||||||
}
|
}
|
||||||
|
|
||||||
private:
|
private:
|
||||||
|
template <Type type>
|
||||||
|
u64* Array() noexcept {
|
||||||
|
if constexpr (type == Type::CPU) {
|
||||||
|
return words.cpu.Pointer(IsShort());
|
||||||
|
} else if constexpr (type == Type::GPU) {
|
||||||
|
return words.gpu.Pointer(IsShort());
|
||||||
|
} else if constexpr (type == Type::CachedCPU) {
|
||||||
|
return words.cached_cpu.Pointer(IsShort());
|
||||||
|
} else if constexpr (type == Type::Untracked) {
|
||||||
|
return words.untracked.Pointer(IsShort());
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|
||||||
|
template <Type type>
|
||||||
|
const u64* Array() const noexcept {
|
||||||
|
if constexpr (type == Type::CPU) {
|
||||||
|
return words.cpu.Pointer(IsShort());
|
||||||
|
} else if constexpr (type == Type::GPU) {
|
||||||
|
return words.gpu.Pointer(IsShort());
|
||||||
|
} else if constexpr (type == Type::CachedCPU) {
|
||||||
|
return words.cached_cpu.Pointer(IsShort());
|
||||||
|
} else if constexpr (type == Type::Untracked) {
|
||||||
|
return words.untracked.Pointer(IsShort());
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|
||||||
/**
|
/**
|
||||||
* Change the state of a range of pages
|
* Change the state of a range of pages
|
||||||
*
|
*
|
||||||
* @param written_words Pages to be marked or unmarked as modified
|
|
||||||
* @param dirty_addr Base address to mark or unmark as modified
|
* @param dirty_addr Base address to mark or unmark as modified
|
||||||
* @param size Size in bytes to mark or unmark as modified
|
* @param size Size in bytes to mark or unmark as modified
|
||||||
*
|
|
||||||
* @tparam enable True when the bits will be set to one, false for zero
|
|
||||||
* @tparam notify_rasterizer True when the rasterizer has to be notified about the changes
|
|
||||||
*/
|
*/
|
||||||
template <bool enable, bool notify_rasterizer>
|
template <Type type, bool enable>
|
||||||
void ChangeRegionState(WrittenWords& written_words, u64 dirty_addr,
|
void ChangeRegionState(u64 dirty_addr, s64 size) noexcept(type == Type::GPU) {
|
||||||
s64 size) noexcept(!notify_rasterizer) {
|
|
||||||
const s64 difference = dirty_addr - cpu_addr;
|
const s64 difference = dirty_addr - cpu_addr;
|
||||||
const u64 offset = std::max<s64>(difference, 0);
|
const u64 offset = std::max<s64>(difference, 0);
|
||||||
size += std::min<s64>(difference, 0);
|
size += std::min<s64>(difference, 0);
|
||||||
if (offset >= SizeBytes() || size < 0) {
|
if (offset >= SizeBytes() || size < 0) {
|
||||||
return;
|
return;
|
||||||
}
|
}
|
||||||
u64* const state_words = written_words.Pointer(IsShort());
|
u64* const untracked_words = Array<Type::Untracked>();
|
||||||
|
u64* const state_words = Array<type>();
|
||||||
const u64 offset_end = std::min(offset + size, SizeBytes());
|
const u64 offset_end = std::min(offset + size, SizeBytes());
|
||||||
const u64 begin_page_index = offset / BYTES_PER_PAGE;
|
const u64 begin_page_index = offset / BYTES_PER_PAGE;
|
||||||
const u64 begin_word_index = begin_page_index / PAGES_PER_WORD;
|
const u64 begin_word_index = begin_page_index / PAGES_PER_WORD;
|
||||||
|
@ -268,13 +340,19 @@ private:
|
||||||
u64 bits = ~u64{0};
|
u64 bits = ~u64{0};
|
||||||
bits = (bits >> right_offset) << right_offset;
|
bits = (bits >> right_offset) << right_offset;
|
||||||
bits = (bits << left_offset) >> left_offset;
|
bits = (bits << left_offset) >> left_offset;
|
||||||
if constexpr (notify_rasterizer) {
|
if constexpr (type == Type::CPU || type == Type::CachedCPU) {
|
||||||
NotifyRasterizer<!enable>(word_index, state_words[word_index], bits);
|
NotifyRasterizer<!enable>(word_index, untracked_words[word_index], bits);
|
||||||
}
|
}
|
||||||
if constexpr (enable) {
|
if constexpr (enable) {
|
||||||
state_words[word_index] |= bits;
|
state_words[word_index] |= bits;
|
||||||
|
if constexpr (type == Type::CPU || type == Type::CachedCPU) {
|
||||||
|
untracked_words[word_index] |= bits;
|
||||||
|
}
|
||||||
} else {
|
} else {
|
||||||
state_words[word_index] &= ~bits;
|
state_words[word_index] &= ~bits;
|
||||||
|
if constexpr (type == Type::CPU || type == Type::CachedCPU) {
|
||||||
|
untracked_words[word_index] &= ~bits;
|
||||||
|
}
|
||||||
}
|
}
|
||||||
page_index = 0;
|
page_index = 0;
|
||||||
++word_index;
|
++word_index;
|
||||||
|
@ -291,7 +369,7 @@ private:
|
||||||
* @tparam add_to_rasterizer True when the rasterizer should start tracking the new pages
|
* @tparam add_to_rasterizer True when the rasterizer should start tracking the new pages
|
||||||
*/
|
*/
|
||||||
template <bool add_to_rasterizer>
|
template <bool add_to_rasterizer>
|
||||||
void NotifyRasterizer(u64 word_index, u64 current_bits, u64 new_bits) {
|
void NotifyRasterizer(u64 word_index, u64 current_bits, u64 new_bits) const {
|
||||||
u64 changed_bits = (add_to_rasterizer ? current_bits : ~current_bits) & new_bits;
|
u64 changed_bits = (add_to_rasterizer ? current_bits : ~current_bits) & new_bits;
|
||||||
VAddr addr = cpu_addr + word_index * BYTES_PER_WORD;
|
VAddr addr = cpu_addr + word_index * BYTES_PER_WORD;
|
||||||
while (changed_bits != 0) {
|
while (changed_bits != 0) {
|
||||||
|
@ -315,21 +393,20 @@ private:
|
||||||
* @param query_cpu_range Base CPU address to loop over
|
* @param query_cpu_range Base CPU address to loop over
|
||||||
* @param size Size in bytes of the CPU range to loop over
|
* @param size Size in bytes of the CPU range to loop over
|
||||||
* @param func Function to call for each turned off region
|
* @param func Function to call for each turned off region
|
||||||
*
|
|
||||||
* @tparam gpu True for host GPU pages, false for CPU pages
|
|
||||||
* @tparam notify_rasterizer True when the rasterizer should be notified about state changes
|
|
||||||
*/
|
*/
|
||||||
template <bool gpu, bool notify_rasterizer, typename Func>
|
template <Type type, typename Func>
|
||||||
void ForEachModifiedRange(VAddr query_cpu_range, s64 size, Func&& func) {
|
void ForEachModifiedRange(VAddr query_cpu_range, s64 size, Func&& func) {
|
||||||
|
static_assert(type != Type::Untracked);
|
||||||
|
|
||||||
const s64 difference = query_cpu_range - cpu_addr;
|
const s64 difference = query_cpu_range - cpu_addr;
|
||||||
const u64 query_begin = std::max<s64>(difference, 0);
|
const u64 query_begin = std::max<s64>(difference, 0);
|
||||||
size += std::min<s64>(difference, 0);
|
size += std::min<s64>(difference, 0);
|
||||||
if (query_begin >= SizeBytes() || size < 0) {
|
if (query_begin >= SizeBytes() || size < 0) {
|
||||||
return;
|
return;
|
||||||
}
|
}
|
||||||
const u64* const cpu_words = words.cpu.Pointer(IsShort());
|
u64* const untracked_words = Array<Type::Untracked>();
|
||||||
|
u64* const state_words = Array<type>();
|
||||||
const u64 query_end = query_begin + std::min(static_cast<u64>(size), SizeBytes());
|
const u64 query_end = query_begin + std::min(static_cast<u64>(size), SizeBytes());
|
||||||
u64* const state_words = (gpu ? words.gpu : words.cpu).Pointer(IsShort());
|
|
||||||
u64* const words_begin = state_words + query_begin / BYTES_PER_WORD;
|
u64* const words_begin = state_words + query_begin / BYTES_PER_WORD;
|
||||||
u64* const words_end = state_words + Common::DivCeil(query_end, BYTES_PER_WORD);
|
u64* const words_end = state_words + Common::DivCeil(query_end, BYTES_PER_WORD);
|
||||||
|
|
||||||
|
@ -345,7 +422,8 @@ private:
|
||||||
const u64 word_index_end = std::distance(state_words, last_modified_word);
|
const u64 word_index_end = std::distance(state_words, last_modified_word);
|
||||||
|
|
||||||
const unsigned local_page_begin = std::countr_zero(*first_modified_word);
|
const unsigned local_page_begin = std::countr_zero(*first_modified_word);
|
||||||
const unsigned local_page_end = PAGES_PER_WORD - std::countl_zero(last_modified_word[-1]);
|
const unsigned local_page_end =
|
||||||
|
static_cast<unsigned>(PAGES_PER_WORD) - std::countl_zero(last_modified_word[-1]);
|
||||||
const u64 word_page_begin = word_index_begin * PAGES_PER_WORD;
|
const u64 word_page_begin = word_index_begin * PAGES_PER_WORD;
|
||||||
const u64 word_page_end = (word_index_end - 1) * PAGES_PER_WORD;
|
const u64 word_page_end = (word_index_end - 1) * PAGES_PER_WORD;
|
||||||
const u64 query_page_begin = query_begin / BYTES_PER_PAGE;
|
const u64 query_page_begin = query_begin / BYTES_PER_PAGE;
|
||||||
|
@ -371,11 +449,13 @@ private:
|
||||||
const u64 current_word = state_words[word_index] & bits;
|
const u64 current_word = state_words[word_index] & bits;
|
||||||
state_words[word_index] &= ~bits;
|
state_words[word_index] &= ~bits;
|
||||||
|
|
||||||
// Exclude CPU modified pages when visiting GPU pages
|
if constexpr (type == Type::CPU) {
|
||||||
const u64 word = current_word & ~(gpu ? cpu_words[word_index] : 0);
|
const u64 current_bits = untracked_words[word_index] & bits;
|
||||||
if constexpr (notify_rasterizer) {
|
untracked_words[word_index] &= ~bits;
|
||||||
NotifyRasterizer<true>(word_index, word, ~u64{0});
|
NotifyRasterizer<true>(word_index, current_bits, ~u64{0});
|
||||||
}
|
}
|
||||||
|
// Exclude CPU modified pages when visiting GPU pages
|
||||||
|
const u64 word = current_word & ~(type == Type::GPU ? untracked_words[word_index] : 0);
|
||||||
u64 page = page_begin;
|
u64 page = page_begin;
|
||||||
page_begin = 0;
|
page_begin = 0;
|
||||||
|
|
||||||
|
@ -416,17 +496,20 @@ private:
|
||||||
* @param offset Offset in bytes from the start of the buffer
|
* @param offset Offset in bytes from the start of the buffer
|
||||||
* @param size Size in bytes of the region to query for modifications
|
* @param size Size in bytes of the region to query for modifications
|
||||||
*/
|
*/
|
||||||
template <bool gpu>
|
template <Type type>
|
||||||
[[nodiscard]] bool IsRegionModified(u64 offset, u64 size) const noexcept {
|
[[nodiscard]] bool IsRegionModified(u64 offset, u64 size) const noexcept {
|
||||||
const u64* const cpu_words = words.cpu.Pointer(IsShort());
|
static_assert(type != Type::Untracked);
|
||||||
const u64* const state_words = (gpu ? words.gpu : words.cpu).Pointer(IsShort());
|
|
||||||
|
const u64* const untracked_words = Array<Type::Untracked>();
|
||||||
|
const u64* const state_words = Array<type>();
|
||||||
const u64 num_query_words = size / BYTES_PER_WORD + 1;
|
const u64 num_query_words = size / BYTES_PER_WORD + 1;
|
||||||
const u64 word_begin = offset / BYTES_PER_WORD;
|
const u64 word_begin = offset / BYTES_PER_WORD;
|
||||||
const u64 word_end = std::min(word_begin + num_query_words, NumWords());
|
const u64 word_end = std::min(word_begin + num_query_words, NumWords());
|
||||||
const u64 page_limit = Common::DivCeil(offset + size, BYTES_PER_PAGE);
|
const u64 page_limit = Common::DivCeil(offset + size, BYTES_PER_PAGE);
|
||||||
u64 page_index = (offset / BYTES_PER_PAGE) % PAGES_PER_WORD;
|
u64 page_index = (offset / BYTES_PER_PAGE) % PAGES_PER_WORD;
|
||||||
for (u64 word_index = word_begin; word_index < word_end; ++word_index, page_index = 0) {
|
for (u64 word_index = word_begin; word_index < word_end; ++word_index, page_index = 0) {
|
||||||
const u64 word = state_words[word_index] & ~(gpu ? cpu_words[word_index] : 0);
|
const u64 off_word = type == Type::GPU ? untracked_words[word_index] : 0;
|
||||||
|
const u64 word = state_words[word_index] & ~off_word;
|
||||||
if (word == 0) {
|
if (word == 0) {
|
||||||
continue;
|
continue;
|
||||||
}
|
}
|
||||||
|
@ -445,13 +528,13 @@ private:
|
||||||
*
|
*
|
||||||
* @param offset Offset in bytes from the start of the buffer
|
* @param offset Offset in bytes from the start of the buffer
|
||||||
* @param size Size in bytes of the region to query for modifications
|
* @param size Size in bytes of the region to query for modifications
|
||||||
*
|
|
||||||
* @tparam gpu True to query GPU modified pages, false for CPU pages
|
|
||||||
*/
|
*/
|
||||||
template <bool gpu>
|
template <Type type>
|
||||||
[[nodiscard]] std::pair<u64, u64> ModifiedRegion(u64 offset, u64 size) const noexcept {
|
[[nodiscard]] std::pair<u64, u64> ModifiedRegion(u64 offset, u64 size) const noexcept {
|
||||||
const u64* const cpu_words = words.cpu.Pointer(IsShort());
|
static_assert(type != Type::Untracked);
|
||||||
const u64* const state_words = (gpu ? words.gpu : words.cpu).Pointer(IsShort());
|
|
||||||
|
const u64* const untracked_words = Array<Type::Untracked>();
|
||||||
|
const u64* const state_words = Array<type>();
|
||||||
const u64 num_query_words = size / BYTES_PER_WORD + 1;
|
const u64 num_query_words = size / BYTES_PER_WORD + 1;
|
||||||
const u64 word_begin = offset / BYTES_PER_WORD;
|
const u64 word_begin = offset / BYTES_PER_WORD;
|
||||||
const u64 word_end = std::min(word_begin + num_query_words, NumWords());
|
const u64 word_end = std::min(word_begin + num_query_words, NumWords());
|
||||||
|
@ -460,7 +543,8 @@ private:
|
||||||
u64 begin = std::numeric_limits<u64>::max();
|
u64 begin = std::numeric_limits<u64>::max();
|
||||||
u64 end = 0;
|
u64 end = 0;
|
||||||
for (u64 word_index = word_begin; word_index < word_end; ++word_index) {
|
for (u64 word_index = word_begin; word_index < word_end; ++word_index) {
|
||||||
const u64 word = state_words[word_index] & ~(gpu ? cpu_words[word_index] : 0);
|
const u64 off_word = type == Type::GPU ? untracked_words[word_index] : 0;
|
||||||
|
const u64 word = state_words[word_index] & ~off_word;
|
||||||
if (word == 0) {
|
if (word == 0) {
|
||||||
continue;
|
continue;
|
||||||
}
|
}
|
||||||
|
@ -488,7 +572,7 @@ private:
|
||||||
|
|
||||||
RasterizerInterface* rasterizer = nullptr;
|
RasterizerInterface* rasterizer = nullptr;
|
||||||
VAddr cpu_addr = 0;
|
VAddr cpu_addr = 0;
|
||||||
GpuCpuWords words;
|
Words words;
|
||||||
BufferFlagBits flags{};
|
BufferFlagBits flags{};
|
||||||
};
|
};
|
||||||
|
|
||||||
|
|
Loading…
Reference in a new issue