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

agc: encode acquire-memory byte ranges in standard packets

Browse Source
This commit is contained in:
nmzik
2026-10-15 23:34:42 +02:00
parent c8caa27936
commit a5ceb48184
6 changed files with 75 additions and 68 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tests/ShaderRecompilerComputeTests.cpp
+60 -24
View File
@@ -45742,29 +45742,6 @@ void CheckVulkan13FeatureRequirements() {
std::printf("[host] %-32s ok\n", "Vulkan13FeatureRequirements");
}
void CheckPm4AcquireMemNoOp(RenderContext &renderer) {
GraphicsInitJmpTables();
CommandProcessor processor(renderer, 0);
const std::array<uint32_t, 6> standard_payload{0x00400000u, 1u, 0u,
0u, 0u, 10u};
const std::array<uint32_t, 7> custom_payload{
0xdeadbeefu, 0xffffffffu, 0x80000000u, 0x13579bdfu,
0x2468ace0u, 0xaaaaaaaau, 0x55555555u};
const auto standard_before = standard_payload;
const auto custom_before = custom_payload;
Require("Pm4AcquireMemNoOp", "recognized packets",
CpOpAcquireMem(processor, 0xC0055800u, standard_payload.data(), 0,
0) == 6 &&
CpOpAcquireMem(processor, 0xc0061050u, custom_payload.data(), 0,
0) == 7 &&
standard_payload == standard_before &&
custom_payload == custom_before,
"ACQUIRE_MEM did not consume both pinned packet forms as "
"side-effect-free "
"no-ops");
std::printf("[host] %-32s ok\n", "Pm4AcquireMemNoOp");
}
void CheckPm4SyntheticOcclusionCounterDump(RenderContext &renderer) {
GraphicsInitJmpTables();
CommandProcessor processor(renderer, 0);
@@ -46531,6 +46508,64 @@ struct CommandBufferLayout {
static_assert(offsetof(CommandBufferLayout, cursor_up) == 0x10);
static_assert(offsetof(CommandBufferLayout, reserved_dw) == 0x30);
void CheckPm4AcquireMemPackets(RenderContext &renderer) {
constexpr const char *name = "Pm4AcquireMemPackets";
GraphicsInitJmpTables();
CommandProcessor processor(renderer, 0);
// Byte ranges cover every touched 256-byte block, including partial first and last blocks.
struct Case {
const char *label;
bool async;
uint8_t engine;
uint32_t cb_db, gcr;
uint64_t base, size;
uint32_t poll;
std::array<uint32_t, 8> expected;
};
const Case cases[]{
{"captured byte range", false, 1, 0, 0x9000, 0x2036000308ull, 0x20, 400,
{0xc0065800, 0x80000000, 1, 0, 0x20360003, 0, 25, 0x9000}},
{"crossing high address", false, 0, 0x7fc0, 0x20, 0xab12345678f0ull, 0x40, 400,
{0xc0065800, 0x7fc0, 2, 0, 0x12345678, 0xab, 25, 0x20}},
{"full range", false, 0, 0x7fc0, 0x20, 0, UINT64_MAX, 400,
{0xc0065800, 0x7fc0, UINT32_MAX, 0xff, 0, 0, 25, 0x20}},
{"wide range and saturated poll", false, 1, 0, 0x380, 0, 0x10000000000ull, UINT32_MAX,
{0xc0065800, 0x80000000, 0, 1, 0, 0, 0xffff, 0x380}},
{"async byte range", true, 0, 0, 0x9000, 0x2036000308ull, 0x20, 400,
{0xc0065800, 0, 1, 0, 0x20360003, 0, 25, 0x9000}},
};
Require(name, "packet sizes", Gen5::AgcDcbAcquireMemGetSize() == 32 &&
Gen5::AgcAcbAcquireMemGetSize() == 32,
"native single ACQUIRE_MEM packet size changed");
for (const auto &test : cases) {
std::array<uint32_t, 8> packet{};
CommandBufferLayout layout{packet.data(), packet.data() + packet.size(),
packet.data(), packet.data() + packet.size(),
nullptr, nullptr, 0};
auto *buffer = reinterpret_cast<Gen5::CommandBuffer *>(&layout);
const auto *base = reinterpret_cast<const volatile void *>(test.base);
auto *emitted = test.async
? Gen5::AgcAcbAcquireMem(buffer, test.gcr, base, test.size, test.poll)
: Gen5::AgcDcbAcquireMem(buffer, test.engine, test.cb_db, test.gcr,
base, test.size, test.poll);
Require(name, test.label, emitted == packet.data() &&
layout.cursor_up == packet.data() + packet.size() && packet == test.expected,
"ACQUIRE_MEM differs from the pinned native packet");
const auto tick = renderer.GetCommandScheduler().CurrentTick();
Pm4Execution execution;
Require(name, test.label,
processor.Process(execution, packet) == Pm4ProcessResult::Complete &&
packet == test.expected && renderer.GetCommandScheduler().CurrentTick() == tick,
"native ACQUIRE_MEM was not consumed without mutation or queue submission");
}
const std::array<uint32_t, 7> ps4_packet{0xc0055800, 0x00400000, 1, 0, 0, 0, 10};
Pm4Execution execution;
Require(name, "native PS4 packet",
processor.Process(execution, ps4_packet) == Pm4ProcessResult::Complete,
"seven-DWORD ACQUIRE_MEM was not consumed");
std::printf("[host] %-32s ok\n", name);
}
void CheckPm4WaitPackets(RenderContext &renderer) {
GraphicsInitJmpTables();
CommandProcessor processor(renderer, 0);
@@ -48240,6 +48275,7 @@ int main(int argc, char **argv) {
}
if (argc == 2 && std::strcmp(argv[1], "--pm4-wait-only") == 0) {
VulkanHarness vulkan;
CheckPm4AcquireMemPackets(vulkan.RuntimeRenderer());
CheckPm4WaitPackets(vulkan.RuntimeRenderer());
CheckPm4WaitResume(vulkan.RuntimeRenderer());
return 0;
@@ -48512,7 +48548,7 @@ int main(int argc, char **argv) {
CheckErrorDialogLifecycle();
CheckSigninDialogLifecycle();
CheckVulkan13FeatureRequirements();
CheckPm4AcquireMemNoOp(vulkan.RuntimeRenderer());
CheckPm4AcquireMemPackets(vulkan.RuntimeRenderer());
CheckPm4SyntheticOcclusionCounterDump(vulkan.RuntimeRenderer());
CheckPm4Predication(vulkan.RuntimeRenderer());
CheckPm4StencilInfoValueLane(vulkan.RuntimeRenderer());