raphiaura/KytyPS5
1
0
Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-28 13:56:59 +01:00
Code Issues Packages Projects Releases Wiki Activity

graphics: draw patched-offset indirect draws from GPU-read records

Browse Source
This commit is contained in:
nmzik
2026-10-27 09:31:50 +01:00
parent 09753bf8ff
commit a5b3673a39
14 changed files with 118 additions and 21 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graphics/guest_gpu/graphicsRun.cpp
+26 -7
View File
@@ -814,6 +814,27 @@ static void PatchIndirectDrawOffsets(CommandProcessor& cp, IndirectDrawRegisters
write(registers.draw_index, draw_index);
}
// Writes the GS user-data slots the packet patches; false when the shaders cannot read the
// offsets from GPU records: a first-index or subdraw-index destination, or another stage's
// user data.
static bool IndirectDrawOffsets(IndirectDrawRegisters registers, ShaderIndirectOffsets& offsets) {
if (registers.index_offset != Pm4::SH_NOP || registers.draw_index != Pm4::SH_NOP) {
return false;
}
for (const auto [location, slot]: {std::pair {registers.vertex_offset, &offsets.vertex},
std::pair {registers.instance_offset, &offsets.instance}}) {
if (location == Pm4::SH_NOP) {
continue;
}
const uint32_t index = location - Pm4::SPI_SHADER_USER_DATA_GS_0;
if (index >= HW::UserSgprInfo::SGPRS_MAX) {
return false;
}
*slot = static_cast<uint8_t>(index);
}
return true;
}
void CommandProcessor::DrawIndirect(uint32_t data_offset, IndirectDrawRegisters registers,
uint32_t draw_initiator, bool indexed) {
const auto args_size =
@@ -845,13 +866,11 @@ void CommandProcessor::DrawIndirectMulti(uint32_t data_offset, uint32_t max_coun
// NUM_INSTANCES keeps the last draw's instance count.
m_num_instances_addr = args_addr + static_cast<uint64_t>(draw_count - 1) * stride_in_bytes +
offsetof(vk::DrawIndirectCommand, instanceCount);
// With no SGPRs to patch, the GPU reads the records if the shaders allow. Shaders apply
// the records' offsets only through those SGPRs; Vulkan applies them, which matches
// while they are zero.
const DrawIndirectArgs records {args_addr, draw_count, stride_in_bytes};
if (registers.vertex_offset == Pm4::SH_NOP && registers.instance_offset == Pm4::SH_NOP &&
registers.index_offset == Pm4::SH_NOP && registers.draw_index == Pm4::SH_NOP &&
(!indexed || m_index_buffer_size != 0) &&
// The GPU reads the records if the shaders allow; the VS prolog loads the patched SGPRs from
// Vulkan's bases. With no SGPR patched, Vulkan applies the offsets, which matches while they
// are zero.
DrawIndirectArgs records {args_addr, draw_count, stride_in_bytes};
if (IndirectDrawOffsets(registers, records.offsets) && (!indexed || m_index_buffer_size != 0) &&
(indexed ? executor.DrawIndex(m_submit_id, CurrentBuffer(),
{.index_count = m_index_buffer_size,
.index_addr = IndexAddress(0),