fix(agc): start DB_Z_INFO and DB_STENCIL_INFO at their clear-state value

core/libs/prx/libSceAgcDriver/Execution/include/QueueState.hpp

@@ -221,7 +221,7 @@ inline Registers BuildInitialContextRegisters() {
     Registers result{
         {0x200, 0}, {0x201, 0}, {0x202, 0xcc0010}, {0x203, 0},
         {0x204, 0}, {0x205, 0}, {0x206, 1087}, {0x207, 0},
-        {0x0, 0}, {0x2, 0}, {0x3, 0}, {0x4, 0}, {0x8, 0}, {0x9, 0x3f800000}, {0xa, 0}, {0xb, 0},
+        {0x0, 0}, {0x2, 0}, {0x3, 0}, {0x4, 0}, {0x8, 0}, {0x9, 0x3f800000}, {0xa, 0}, {0xb, 0}, {0x10, 0}, {0x11, 0},
         {0x80, 0}, {0x83, 0xffff}, {0x8c, 0xaa99aaaa}, {0x8d, 0}, {0x8e, 0}, {0x8f, 0},
         {0xc, 0}, {0xd, 0x40004000}, {0x81, 0x80000000}, {0x82, 0x40004000},
         {0x90, 0x80000000}, {0x91, 0x40004000},

core/libs/prx/libSceAgcDriver/tests/Graphics.cpp

@@ -863,6 +863,57 @@ void msaaEnableClearOneSampleTargetTests() {
     expectFailure([&] { DecodeState(queue); }, "a color target's sample count differs from the rasterizer's");
 }
 
+void UnwrittenDepthPlaneTests() {
+    using AgcDriver::Graphics::DecodeState;
+    using AgcDriver::Graphics::DrawRejection;
+    const auto fresh = [](std::uint32_t depthControl) {
+        auto written = makeState();
+        written.context[0x200] = depthControl;
+        written.context[0x1b3] = 2;
+        written.context[0x1b4] = 2;
+        AgcDriver::QueueState queue;
+        queue.shader = written.shader;
+        queue.userConfig = written.userConfig;
+        written.context.ForEach([&](std::uint32_t offset, std::uint32_t value) { queue.context[offset] = value; });
+        return queue;
+    };
+    auto queue = fresh(0x777);
+    auto state = DecodeState(queue);
+    Require(!state.depth && !state.depthTest && !state.depthWrite && !state.stencilTest, "depth and stencil tests on unwritten depth and stencil planes were kept (PPSA24040's DB_DEPTH_CONTROL)");
+    auto rejection = DrawRejection(queue, false);
+    Require(rejection.empty(), "the precheck refused depth and stencil tests on unwritten depth and stencil planes: " + rejection);
+    queue = fresh(0x77f);
+    state = DecodeState(queue);
+    Require(!state.depth && !state.depthBoundsTest, "a depth bounds test on unwritten depth and stencil planes was kept");
+    rejection = DrawRejection(queue, false);
+    Require(rejection.empty(), "the precheck refused a depth bounds test on unwritten depth and stencil planes: " + rejection);
+    queue = fresh(0x007007b3);
+    queue.context[0x010] = 0x80000183;
+    queue.context[0x012] = 0x00001000;
+    queue.context[0x014] = 0x00001000;
+    queue.context[0x007] = 0x0001001f;
+    queue.context[0x00b] = 0x3f800000;
+    state = DecodeState(queue);
+    Require(state.depth && state.depthTest && !state.stencilTest && state.renderExtent.width == 32 && state.renderExtent.height == 2, "a depth plane beside an unwritten DB_STENCIL_INFO did not decode as depth only");
+    rejection = DrawRejection(queue, false);
+    Require(rejection.empty(), "the precheck refused a depth plane beside an unwritten DB_STENCIL_INFO: " + rejection);
+    queue = fresh(0x007007b3);
+    queue.context[0x011] = 0x20000181;
+    queue.context[0x013] = 0x00002000;
+    queue.context[0x015] = 0x00002000;
+    queue.context[0x007] = 0x003f003f;
+    queue.context[0x00b] = 0x3f800000;
+    queue.context[0x10c] = 0x01ffff00;
+    queue.context[0x10d] = 0x01ffff00;
+    state = DecodeState(queue);
+    Require(!state.depthTest && state.stencilTest, "a stencil plane beside an unwritten DB_Z_INFO kept the depth test or lost the stencil test");
+    rejection = DrawRejection(queue, false);
+    Require(rejection.empty(), "the precheck refused a stencil plane beside an unwritten DB_Z_INFO: " + rejection);
+    queue.context[0x010] = 0x80000183;
+    queue.ClearContext();
+    for (const auto offset : {0x010u, 0x011u}) Require(queue.context.contains(offset) && queue.context.at(offset) == 0, "CLEAR_STATE did not reset DB_Z_INFO and DB_STENCIL_INFO to their clear-state value 0");
+}
+
 void MsaaEnableSingleSampleTests() {
     using AgcDriver::Graphics::DecodeState;
     using AgcDriver::Graphics::DrawRejection;
@@ -11030,6 +11081,7 @@ int main(int argc, char** argv) {
         UnwrittenSampleLocationTests();
         sampleLocationGridFallbackTests();
         sampleOrderTests();
+        UnwrittenDepthPlaneTests();
         depthCopyFormatTests();
         quadPixelMaskTests();
         ShaderStageTests();

core/libs/prx/libSceAgcDriver/tests/Pm4.cpp

@@ -192,9 +192,9 @@ void testContextAndBases() {
     check(state.context.contains(0x280) && state.context.at(0x280) == 0x00080008u && state.context.contains(0x281) && state.context.at(0x281) == 0xffff0000u, "initial point context differs from the SDK defaults");
     execute(state, makePacket(0x69, {0x280, 0x00180018u, 0x00400008u}));
     check(state.context.at(0x280) == 0x00180018u && state.context.at(0x281) == 0x00400008u, "explicit point size and limits were not stored");
-    for (const auto offset : {0x10bu, 0x10cu, 0x10du}) execute(state, makePacket(0x69, {offset, 0xffffffffu}));
+    for (const auto offset : {0x010u, 0x011u, 0x10bu, 0x10cu, 0x10du}) execute(state, makePacket(0x69, {offset, 0xffffffffu}));
     execute(state, makePacket(0x12, {0}));
-    for (const auto offset : {0x10bu, 0x10cu, 0x10du}) check(state.context.at(offset) == 0, "CLEAR_STATE did not restore stencil defaults");
+    for (const auto offset : {0x010u, 0x011u, 0x10bu, 0x10cu, 0x10du}) check(state.context.at(offset) == 0, "CLEAR_STATE did not restore depth and stencil defaults");
     check(state.context.at(0x280) == 0x00080008u && state.context.at(0x281) == 0xffff0000u, "CLEAR_STATE did not restore point size and limits");
     execute(state, makePacket(0x69, {0x10, 17}));
     execute(state, makePacket(0x76, {0x20c, 2}));
@@ -1021,8 +1021,9 @@ void testInvalidRegisterListEntries() {
     check(!AgcDriver::Pm4::IndirectRegistersValid(std::array<std::uint32_t, 2>{0xffffffffu, 0}), "the sentinel was taken as a register offset");
     check(AgcDriver::Pm4::IndirectRegistersValid(std::array<std::uint32_t, 4>{0x10, 1, 0x70000011u, 2}), "a valid register list was rejected");
     AgcDriver::QueueState state;
+    const auto initial = state.context;
     expectFailure([&] { AgcDriver::Pm4::ExecuteIndirectRegisters(packet, pairs, state); }, "extended register");
-    check(!state.context.contains(0x10) && !state.context.contains(0x11), "a rejected register list partially changed state");
+    check(state.context == initial, "a rejected register list partially changed state");
     AgcDriver::Pm4::ExecuteIndirectRegisters(packet, pairs, state, true);
     check(state.context.at(0x10) == 41 && state.context.at(0x11) == 42 && !state.context.contains(0x38f), "skipping an invalid entry lost the valid ones or wrote the invalid one");
 }

docs/dev/debt/libSceAgcDriver.md

@@ -67,6 +67,7 @@
 
 - [Initial stencil state](../../../core/libs/prx/libSceAgcDriver/Execution/include/QueueState.hpp) uses zero for `DB_STENCIL_CONTROL`, `DB_STENCILREFMASK` and `DB_STENCILREFMASK_BF` (context `0x10b`-`0x10d`), including after `CLEAR_STATE`. The public V0, V4, V5 and V7-V11 [register-default tables](../../../core/libs/prx/libSceAgc/Command/include/RegisterDefaultsData.hpp) supply these values. Linux's [GFX10 clear-state table](https://github.com/torvalds/linux/blob/master/drivers/gpu/drm/amd/amdgpu/clearstate_gfx10.h) agrees on the control register but uses `0x01000000` for the reference masks; the console reset values have not been measured. Planet Coaster (PPSA01736 v01.017) enables stencil tests without writing the control register.
 - [Multisampled DCC comp-to-single clears](../../../core/libs/prx/libSceAgcDriver/Graphics/src/MultisampleTarget.cpp) (AGC driver) - RADV's multisampled comp-to-single clear dispatch ([`radv_clear_dcc_comp_to_single`](https://github.com/mihawk-99/PS5_Mesa/blob/1ba9874f478657ccf6da72ce55f0260164cc0bd0/src/amd/vulkan/meta/radv_meta_clear.c#L954)) is recognized by its code and not run. Its clear color (user data 5 to 8, which the shader stores as the four dwords of sample 0's texel) is recorded for the surface its image descriptor names, and DCC keys of 0x10 on a multisampled target of that surface, pipe-bank XOR, size and sample count import as a clear to it, for texels of 8 to 128 bits. A record is dropped when an unmap or allocation removal covers its surface's base. The guest's color memory doesn't get the color at the blocks' first pixels, and the recorded color is assumed to be the one the keys refer to: RADV writes 0x10 keys only in a fast clear that runs this dispatch for the same image right after, and initializes DCC to 0x00 or 0xff on an image's first use ([`radv_init_color_image_metadata`](https://github.com/mihawk-99/PS5_Mesa/blob/1ba9874f478657ccf6da72ce55f0260164cc0bd0/src/amd/vulkan/radv_cmd_buffer.c#L16894)). A new image on memory reused within one mapping, with the same address, XOR, size and sample count, whose clear runs a comp-to-single shader other than the recognized one, would still import the old color.
+- [Initial depth and stencil planes](../../../core/libs/prx/libSceAgcDriver/Execution/include/QueueState.hpp) - the console's reset values of `DB_Z_INFO` and `DB_STENCIL_INFO` (context `0x010`-`0x011`) are not measured. Both start, and return after `CLEAR_STATE`, at zero (formats `INVALID`), the value of Linux's [GFX10 clear-state table](https://github.com/torvalds/linux/blob/master/drivers/gpu/drm/amd/amdgpu/clearstate_gfx10.h). The public [register-default tables](../../../core/libs/prx/libSceAgc/Command/include/RegisterDefaultsData.hpp) write `0x80000180` and `0x20000180` instead, also with formats `INVALID`; of their other bits, only `TILE_STENCIL_DISABLE` would change a result: the HTILE layout of a multisampled depth plane whose `DB_STENCIL_INFO` was never written.
 - [Type-0 and type-1 texture descriptors](../../../core/libs/prx/libSceAgcDriver/Graphics/src/GuestTextureResource.cpp) (`TYPE` = 0 or 1 in dword 3) with a nonzero base are bound as null images, like a zero base. TYPE 0 comes from a gfx1036 (RDNA2) measurement in issue #4139 (image sample, load and resinfo return zero, no fault). TYPE 1 comes from a gfx1010 (RDNA1) measurement in #4139 and #5828 (`image_sample_lz`, `image_sample_b`, `image_load` and `image_get_resinfo` return zero, no fault) with format 22 only; TYPE 1 is not measured on RDNA2, with other formats or with depth-compare reads (`image_sample_c*`). Neither type was measured on the console's gfx1013.
 - [Multisampled color targets](../../../core/libs/prx/libSceAgcDriver/Graphics/src/MultisampleTarget.cpp) lose their guest-read guards and cannot be reused after a successful allocation removal or unmap covers their full guarded range. The next target, sampled or storage lookup removes retired cache owners; recorded work and sampled views keep their images alive until their last reference is released. Partial unmaps keep the guard; partial allocation reuse and guest access to live multisampled data remain unsupported.
 - [Stencil reference export replay](../../../core/libs/prx/libSceAgcDriver/Graphics/src/Draw.cpp) on devices without `VK_EXT_shader_stencil_export` replays the low eight bits of the values proved from constant, select and phi exports, with an export proved on every kept path. An export the prover cannot bound replays all 256 eight-bit values, one pass per value per triangle (logged once per process). It records instances, then their triangles, then the reference classes of each triangle. One pipeline filters fragments at shader termination against a dynamic push value before late stencil and depth tests. The [Vulkan stencil-reference replacement rule](https://docs.vulkan.org/refpages/latest/refpages/source/FragStencilRefEXT.html) discards bits above the attachment's stencil width; supported stencil attachments have eight bits. Numeric subgroup observations throw except bounded own-lane mask plumbing and group control with kept-lane and derivative-helper coverage proved. WQM inputs must exclude helper contributions, since [non-quad subgroup operations may treat helpers as inactive](https://docs.vulkan.org/glslext/latest/glslext/khr/GL_KHR_shader_subgroup.html). Readonly guarded SRT reads may report internal host faults; these draws wait and check faults before publishing results, and loops throw. Indirect draws, restart, strips, fans, tessellation, mesh, multisampling, early tests, ordered pixel execution, guest shader memory writes, other fault-producing memory access, clocks, primitive ID inputs, attachment feedback and pipeline face/channel splits throw. An exhausted stage push range throws. Equivalence to console ordering has not been measured on console hardware.