raphiaura/KytyPS5
1
0
Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-20 10:56:58 +02:00
Code Issues Packages Projects Releases Wiki Activity

graphics: store vertex attribute bindings without duplicated arrays

Browse Source
This commit is contained in:
nmzik
2026-09-29 03:25:22 +04:00
parent 8e42e99788
commit b43bf9f7fb
6 changed files with 115 additions and 79 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graphics/host_gpu/renderer/pipeline/pipelineCache.cpp
+9 -13
View File
@@ -761,24 +761,20 @@ PipelineCache::Pipeline& PipelineCache::GetGraphicsPipeline(
vs_input_info.resources_num > ShaderVertexInputInfo::RES_MAX);
key.vertex_input.binding_count = static_cast<uint8_t>(vs_input_info.buffers_num);
key.vertex_input.attribute_count = static_cast<uint8_t>(vs_input_info.resources_num);
uint32_t attributes_num = 0;
for (int binding = 0; binding < vs_input_info.buffers_num; binding++) {
const auto& buffer = vs_input_info.buffers[binding];
EXIT_IF(buffer.attr_num < 0 || buffer.attr_num > ShaderVertexInputBuffer::ATTR_MAX);
attributes_num += static_cast<uint32_t>(buffer.attr_num);
EXIT_IF(attributes_num > static_cast<uint32_t>(vs_input_info.resources_num));
key.vertex_input.bindings[binding] = {.stride = buffer.stride,
.instance = buffer.fetch_index != 0};
for (int attribute = 0; attribute < buffer.attr_num; attribute++) {
const auto index = buffer.attr_indices[attribute];
EXIT_IF(index < 0 || index >= vs_input_info.resources_num);
key.vertex_input.attributes[index] = {
.offset = buffer.attr_offsets[attribute],
.binding = static_cast<uint8_t>(binding),
};
}
}
EXIT_IF(attributes_num != static_cast<uint32_t>(vs_input_info.resources_num));
for (int attribute = 0; attribute < vs_input_info.resources_num; attribute++) {
const auto binding = vs_input_info.resources_dst[attribute].buffer_index;
EXIT_IF(binding < 0 || binding >= vs_input_info.buffers_num);
key.vertex_input.attributes[attribute] = {
.offset = static_cast<uint32_t>(vs_input_info.resources[attribute].Base48() -
vs_input_info.buffers[binding].addr),
.binding = static_cast<uint8_t>(binding),
};
}
}
if (auto iter = m_graphics_pipelines.find(key); iter != m_graphics_pipelines.end()) {
src/graphics/host_gpu/renderer/renderDraw.cpp
+27 -20
View File
@@ -249,8 +249,8 @@ static void LogDrawInputState(const CommandBuffer& buffer, const RenderColorInfo
for (int bi = 0; bi < vs_input_info.buffers_num; bi++) {
const auto& b = vs_input_info.buffers[bi];
LOGF("DrawInputState[%u]: vb[%d] addr=0x%010" PRIx64
" stride=%u records=%u fetch_index=%u attr_num=%d\n",
log_id, bi, b.addr, b.stride, b.num_records, b.fetch_index, b.attr_num);
" stride=%u records=%u fetch_index=%u\n",
log_id, bi, b.addr, b.stride, b.num_records, b.fetch_index);
const auto* bytes = reinterpret_cast<const uint8_t*>(b.addr);
if (bytes != nullptr && b.stride != 0) {
@@ -272,11 +272,13 @@ static void LogDrawInputState(const CommandBuffer& buffer, const RenderColorInfo
raw[5], raw[6], raw[7], raw[8], flt[0], flt[1], flt[2], flt[3], flt[4], flt[5],
flt[6], flt[7], flt[8]);
for (int ai = 0; ai < b.attr_num; ai++) {
const auto res_index = b.attr_indices[ai];
const auto& r = vs_input_info.resources[res_index];
const auto& rd = vs_input_info.resources_dst[res_index];
const auto offset = b.attr_offsets[ai];
for (int ai = 0; ai < vs_input_info.resources_num; ai++) {
const auto& r = vs_input_info.resources[ai];
const auto& rd = vs_input_info.resources_dst[ai];
if (rd.buffer_index != bi) {
continue;
}
const auto offset = static_cast<uint32_t>(r.Base48() - b.addr);
if (offset + 4u <= b.stride &&
r.Format() == Prospero::BufferFormat::k8_8_8_8UNorm) {
uint32_t packed = 0;
@@ -296,14 +298,16 @@ static void LogDrawInputState(const CommandBuffer& buffer, const RenderColorInfo
}
}
for (int ai = 0; ai < b.attr_num; ai++) {
const auto res_index = b.attr_indices[ai];
const auto& r = vs_input_info.resources[res_index];
const auto& rd = vs_input_info.resources_dst[res_index];
LOGF("DrawInputState[%u]: attr[%d] res=%d offset=%u dst=v%d regs=%d fetch_index=%u "
for (int ai = 0; ai < vs_input_info.resources_num; ai++) {
const auto& r = vs_input_info.resources[ai];
const auto& rd = vs_input_info.resources_dst[ai];
if (rd.buffer_index != bi) {
continue;
}
LOGF("DrawInputState[%u]: attr[%d] offset=%u dst=v%d regs=%d fetch_index=%u "
"sharp=%08" PRIx32 " %08" PRIx32 " %08" PRIx32 " %08" PRIx32 "\n",
log_id, ai, res_index, b.attr_offsets[ai], rd.register_start, rd.registers_num,
rd.fetch_index, r.fields[0], r.fields[1], r.fields[2], r.fields[3]);
log_id, ai, static_cast<uint32_t>(r.Base48() - b.addr), rd.register_start,
rd.registers_num, rd.fetch_index, r.fields[0], r.fields[1], r.fields[2], r.fields[3]);
}
}
}
@@ -630,19 +634,22 @@ struct PreparedIndexBuffer {
vk::IndexType type = vk::IndexType::eUint16;
};
static uint64_t VertexBufferDescriptorSize(const ShaderVertexInputBuffer& buffer,
const ShaderVertexInputInfo& info) {
static uint64_t VertexBufferDescriptorSize(int binding, const ShaderVertexInputInfo& info) {
const auto& buffer = info.buffers[binding];
if (buffer.stride != 0 || buffer.num_records == 0) {
return static_cast<uint64_t>(buffer.stride) * buffer.num_records;
}
uint64_t size = 0;
for (int i = 0; i < buffer.attr_num; i++) {
const auto& resource = info.resources[buffer.attr_indices[i]];
for (int i = 0; i < info.resources_num; i++) {
if (info.resources_dst[i].buffer_index != binding) {
continue;
}
const auto& resource = info.resources[i];
// RDNA2 OOB_SELECT=2 only checks NumRecords != 0. A constant attribute still
// fetches its entire format; NumRecords is not a byte count in this mode.
const uint64_t extent = resource.OutOfBounds() == 2
? static_cast<uint64_t>(buffer.attr_offsets[i]) +
? resource.Base48() - buffer.addr +
ShaderRecompiler::Format::GetFormatInfo(resource.Format()).byte_size
: buffer.num_records;
size = std::max(size, extent);
@@ -679,7 +686,7 @@ static PreparedVertexBuffers AcquireVertexBuffers(CommandBuffer& b
uint32_t range_count = 0;
for (int i = 0; i < vs_input_info.buffers_num; i++) {
const auto& vertex = vs_input_info.buffers[i];
const auto size = VertexBufferDescriptorSize(vertex, vs_input_info);
const auto size = VertexBufferDescriptorSize(i, vs_input_info);
sizes[i] = size;
if (size == 0) {
continue;