swung0x48/MobileGL
1
0
Fork 0
mirror of https://github.com/MobileGL-Dev/MobileGL synced 2026-11-07 01:53:24 +09:00
Code Issues Packages Projects Releases Wiki Activity
Files
393002a46e0432bb04b2b1667869a9d6aec79d8f
MobileGL/MobileGL/MG_Pipe/PipeApply.cpp
T

3971 lines
249 KiB
C++
Raw Blame History

// MobileGL - MobileGL/MG_Pipe/PipeApply.cpp
// Copyright (c) 2025-2026 MobileGL-Dev
// Licensed under the GNU Lesser General Public License v3.0:
// https://www.gnu.org/licenses/gpl-3.0.txt
// https://www.gnu.org/licenses/lgpl-3.0.txt
// SPDX-License-Identifier: LGPL-3.0-only
// End of Source File Header
// The in-process applier (PipeApply.h). Compiled only under MOBILEGL_PIPE_PUSH.
//
// This is the one .cpp under MG_Pipe/ that reaches UP to MG_Backend/MGPipe/PipeInputs.h,
// and that is the point: under split it becomes the server, and the server is where the
// working state lives. Nothing in MG_Pipe's HEADERS reaches it, so purity gate A
// (MGPipeValueTypes.h's include closure) is untouched.
#include <MG_Pipe/PipeApply.h>
#include <MG_Backend/MGPipe/PipeInputs.h>
#if MOBILEGL_BUILD_DISAGGREGATED
// MGPipeUnmigratedEmulation's split arm reads MG_Config::Transport, which is what tells an
// emulation site whether it is running inside a server or in the monolith it was written for.
#include <Config.h>
#endif
#if MOBILEGL_PIPE_VERIFY
// THE ONE PLACE THE PROGRAM ARCHIVE'S CODEC IS CALLED, and it is compiled into the VERIFY
// build only. In monolith the archive does not travel - MGPProgramDesc's seven blob refs are
// declared with Size 0 and the two structs ride beside the record through the entry point's
// companion pointers - so a push build pays nothing for it and the header keeps its forward
// declarations. The verify build serialises, deserialises and compares before storing, which
// makes the codec LIVE CODE WITH A GATE rather than dead code with a unit test.
#include <MG_State/GLState/ProgramState/ProgramArtifactsCodec.h>
#endif
#if MOBILEGL_BUILD_DISAGGREGATED
// P7 wave 0: the seam the Magma wire funnels in MG_Backend die through. Declared in MG_Pipe and
// DEFINED below, so MG_Backend names no MG_Remote symbol of its own to reach Session::Fail.
#include <MG_Pipe/PipeSessionFail.h>
// R-6's tier gate. One spelling, asked at the one place the decline is decided. Outside the
// MOBILEGL_PIPE_VERIFY block above on purpose: the tier is a property of the BUILD, not of the
// comparator, and a split build without the comparator still declines every acquisition.
#include <MG_Remote/Client/PersistentMapTracker.h>
// P5c (ct), CONTRACT-P5C.md §6 layer 2: MGPipeApplierReset's role guard asks
// ServerLoop::OnApplyThread() - the one predicate that tells the GL thread from the thread
// that owns g_applier under an active transport - and ClientSession::Active(), which is what
// tells a live wire from a configured-but-wireless one (the bring-up window and the
// server-role-only fixture: there the direct call is the only reset that exists).
#include <MG_Remote/Client/ClientSession.h>
#include <MG_Remote/Server/ServerLoop.h>
#include <MG_Remote/Server/ServerSession.h>
#include <MG_Remote/Server/StagedTextureStore.h>
#include <MG_Remote/CapsCodec.h>
#endif
#include <algorithm>
#include <atomic>
#include <chrono>
#include <cmath>
#include <cstdarg>
#include <cstdio>
#include <cstdlib>
#include <cstring>
#include <utility>
#if defined(__ANDROID__)
#include <sys/system_properties.h>
#endif
// THE VERDICT OF EVERY TRIP WIRE IN THIS FILE, IN ONE PLACE.
//
// MOBILEGL_ASSERT is inert at INFO (Defines.h), which is the level every P2 gate builds at,
// so nothing below is left to an assertion. A poison or verify build stops the process; a
// shipped push build logs at error level and carries on from a DEFINED state, and the
// applier counts the divergence so a unit case can see the wire fire there too. Only the
// poison/verify arm writes the "Fatal{...}" marker G4 greps the retrace logs for.
#if MOBILEGL_PIPE_POISON || MOBILEGL_PIPE_VERIFY
#define MGP_TRIP_WIRE_TAG(name) "Fatal{" name "}"
#define MGP_TRIP_WIRE_REPORT(...) \
do { \
MGLOG_F(__VA_ARGS__); \
std::abort(); \
} while (0)
#else
#define MGP_TRIP_WIRE_TAG(name) name
#define MGP_TRIP_WIRE_REPORT(...) MGLOG_E(__VA_ARGS__)
#endif
// The 25 capabilities whose storage is a plain `<Name>Enabled` bool. Written ONCE and used
// twice - once for the switch arms of DeriveCapability and once for the chunk set that guards
// the capability walk - so the two cannot drift apart. The three P2 gave storage to
// (DepthClamp, FramebufferSrgb, TextureCubeMapSeamless) are in the list like any other; the
// three that are NOT are Blend (BlendStates[i].Enabled), ScissorTest (a 16-bit mask) and the
// eight ClipDistances (an 8-bit mask), each handled by name below.
#define MGP_PLAIN_CAPABILITY_LIST(X) \
X(ColorLogicOp) \
X(DebugOutput) \
X(DebugOutputSynchronous) \
X(DepthClamp) \
X(DepthTest) \
X(CullFace) \
X(Dither) \
X(FramebufferSrgb) \
X(LineSmooth) \
X(Multisample) \
X(PolygonOffsetFill) \
X(PolygonOffsetLine) \
X(PolygonOffsetPoint) \
X(PolygonSmooth) \
X(PrimitiveRestart) \
X(PrimitiveRestartFixedIndex) \
X(RasterizerDiscard) \
X(SampleAlphaToCoverage) \
X(SampleAlphaToOne) \
X(SampleCoverage) \
X(SampleMask) \
X(SampleShading) \
X(StencilTest) \
X(TextureCubeMapSeamless) \
X(ProgramPointSize)
namespace MobileGL::MG_Pipe {
#if MOBILEGL_BUILD_DISAGGREGATED
// ----------------------------------------------------------------------------
// THE BACKEND-SIDE SESSION-FAIL SEAM (P7 wave 0). PipeSessionFail.h holds the argument;
// this is the whole implementation.
//
// It lives in this file for the reason MGP_TRIP_WIRE_REPORT does, forty lines up: this is
// where MG_Pipe already states "the verdict of every trip wire, in one place", and a seam
// whose default is an abort belongs next to the other one rather than in a file of its own
// that a reader has to be told about. Split-only, so a pull build gains nothing (G1) and a
// push-monolith build - where the three Magma wire funnels are not compiled either - gains
// nothing to call.
namespace {
MGPipeSessionFailHook g_sessionFailHook = nullptr;
} // namespace
void MGPipeInstallSessionFailHook(MGPipeSessionFailHook hook) { g_sessionFailHook = hook; }
MGPipeSessionFailHook MGPipeSessionFailHookInstalled() { return g_sessionFailHook; }
void MGPipeSessionFail(MGPipeFatalFamily family, const char* fmt, ...) {
// The same 512-byte buffer and the same fallback SessionFail and WireLogFatal use, so a
// death that travels this seam reads identically to one that did not.
char line[512];
va_list args;
va_start(args, fmt);
const int written = std::vsnprintf(line, sizeof(line), fmt, args);
va_end(args);
if (written < 0) {
std::snprintf(line, sizeof(line),
"MGPipe: unformattable Fatal diagnostic (format=%s)", fmt);
}
if (g_sessionFailHook != nullptr) g_sessionFailHook(family, line);
// NO HOOK: exactly what the three funnels did before this seam existed - the line, then
// the abort. Reached in a server image whose role init has not run (a unit case that
// drives the applier headless) and in the client process, where there is no session to
// fail. The abort carries no `Fatal{` of its own because `line` already does;
// fatal_census.py's FUNNEL_SITES names this function for that reason.
MGLOG_F("%s", line);
std::abort();
}
namespace {
MGPipeSessionLatchHook g_sessionLatchHook = nullptr;
MGPipeSessionLatchArmedHook g_sessionLatchArmedHook = nullptr;
} // namespace
void MGPipeInstallSessionLatchHook(MGPipeSessionLatchHook hook, MGPipeSessionLatchArmedHook armed) {
g_sessionLatchHook = hook;
g_sessionLatchArmedHook = armed;
}
MGPipeSessionLatchHook MGPipeSessionLatchHookInstalled() { return g_sessionLatchHook; }
bool MGPipeSessionLatchArmed() { return g_sessionLatchArmedHook != nullptr && g_sessionLatchArmedHook(); }
bool MGPipeDebugDeviceLossDue() {
long at = 0;
if (const char* value = std::getenv("MOBILEGL_DEBUG_INJECT_DEVICE_LOST_AT")) {
at = std::strtol(value, nullptr, 10);
}
#if defined(__ANDROID__)
else {
char property[PROP_VALUE_MAX] = {};
if (__system_property_get("debug.mobilegl.inject_device_lost_at", property) > 0) {
at = std::strtol(property, nullptr, 10);
}
}
#endif
static std::atomic<long> armedAt{0};
static std::atomic<long> seen{0};
static std::atomic<bool> fired{false};
if (at <= 0) {
// Unset (or 0) disarms: the same value set again counts afresh.
if (armedAt.load(std::memory_order_relaxed) != 0) armedAt.store(0, std::memory_order_relaxed);
return false;
}
// A NEW VALUE RE-ARMS. One server process outlives many sessions, so a second loss is
// injected by setting the knob again (another value, or 0 and then a value): the count
// restarts from the checks made after it was seen.
if (long previous = armedAt.load(std::memory_order_acquire);
previous != at && armedAt.compare_exchange_strong(previous, at, std::memory_order_acq_rel)) {
seen.store(0, std::memory_order_release);
fired.store(false, std::memory_order_release);
}
if (fired.load(std::memory_order_acquire)) return false;
if (seen.fetch_add(1, std::memory_order_acq_rel) + 1 < at) return false;
if (fired.exchange(true, std::memory_order_acq_rel)) return false;
MGLOG_E("MGPipe: debug knob inject_device_lost_at=%ld - this device check reports the device LOST", at);
return true;
}
bool MGPipeDebugSessionLossDue(unsigned clientPid) {
if (clientPid == 0) return false;
// The first call on a thread only starts the clock, so a session is first asked 100 ms
// after it started rendering - never in the middle of its first make-current.
thread_local auto lastAsked = std::chrono::steady_clock::time_point{};
const auto now = std::chrono::steady_clock::now();
if (lastAsked == std::chrono::steady_clock::time_point{}) {
lastAsked = now;
return false;
}
if (now - lastAsked < std::chrono::milliseconds(100)) return false;
lastAsked = now;
long target = 0;
if (const char* value = std::getenv("MOBILEGL_DEBUG_INJECT_DEVICE_LOST_PID")) {
target = std::strtol(value, nullptr, 10);
}
#if defined(__ANDROID__)
else {
char property[PROP_VALUE_MAX] = {};
if (__system_property_get("debug.mobilegl.inject_device_lost_pid", property) > 0) {
target = std::strtol(property, nullptr, 10);
}
}
#endif
// Once per value: the session it hit is over, and a later session of the same process (the
// recovery's) must not be lost again by the same setting. Another pid - or unset/0 and
// then the same one - re-arms it.
static std::atomic<long> firedFor{0};
if (target <= 0) {
if (firedFor.load(std::memory_order_relaxed) != 0) firedFor.store(0, std::memory_order_relaxed);
return false;
}
if (static_cast<unsigned long>(target) != clientPid) return false;
long previous = firedFor.load(std::memory_order_acquire);
if (previous == target || !firedFor.compare_exchange_strong(previous, target, std::memory_order_acq_rel)) {
return false;
}
MGLOG_E("MGPipe: debug knob inject_device_lost_pid=%ld - this session's device is reported LOST", target);
return true;
}
bool MGPipeSessionLatch(MGPipeFatalFamily family, const char* fmt, ...) {
char line[512];
va_list args;
va_start(args, fmt);
const int written = std::vsnprintf(line, sizeof(line), fmt, args);
va_end(args);
if (written < 0) {
std::snprintf(line, sizeof(line),
"MGPipe: unformattable Fatal diagnostic (format=%s)", fmt);
}
if (g_sessionLatchHook != nullptr) return g_sessionLatchHook(family, line);
// NO HOOK: no session to latch, so the death the fail seam would have given.
MGPipeSessionFail(family, "%s", line);
}
#endif
namespace {
// ----------------------------------------------------------------------------
// WHICH CHUNKS EACH DERIVATION READS.
//
// The derivation is called after every scatter, and a scatter usually moves ONE
// chunk: a per-frame glViewport sends dynamic chunk D0 and nothing else (D8). So the
// three wide loops and the 35-arm capability switch are guarded by the chunks whose
// bytes they read, and a scatter that did not touch those bytes does not pay for them.
//
// Nothing here is hand-mapped. Every constant is
// MGPipeRenderStateChunkBitsCovering(offsetof(member), sizeof(member)) over the
// members the guarded block actually reads, so the ONLY claim a reader has to check
// is "does this block read anything else?" - and a boundary move re-computes the
// guards rather than invalidating them.
// ----------------------------------------------------------------------------
using RSP = RenderStateParameters;
#define MGP_CHUNKS_OF(member) MGPipeRenderStateChunkBitsCovering(offsetof(RSP, member), sizeof(RSP::member))
// The per-draw-buffer loop reads BlendStates (equations, factors, Enabled) and
// ColorMasks, and nothing else.
constexpr Uint32 kChunksBlendLoop = MGP_CHUNKS_OF(BlendStates) | MGP_CHUNKS_OF(ColorMasks);
// m_viewportIndexed[16] and the rounded m_viewport both read Viewports, and nothing else.
constexpr Uint32 kChunksViewportLoop = MGP_CHUNKS_OF(Viewports);
constexpr Uint32 kChunksDepthRangeLoop = MGP_CHUNKS_OF(DepthRanges);
constexpr Uint32 kChunksScissorEnableLoop = MGP_CHUNKS_OF(ScissorTestEnabledMask);
// DeriveCapability's sources: the 25 plain bools, plus the three masks/arrays the
// three special arms read.
#define MGP_CAPABILITY_CHUNKS(capability) | MGP_CHUNKS_OF(capability##Enabled)
constexpr Uint32 kChunksCapabilityWalk = MGP_CHUNKS_OF(BlendStates) |
MGP_CHUNKS_OF(ScissorTestEnabledMask) |
MGP_CHUNKS_OF(ClipDistanceEnabledMask)
MGP_PLAIN_CAPABILITY_LIST(MGP_CAPABILITY_CHUNKS);
#undef MGP_CAPABILITY_CHUNKS
// The patch trio's chunk, which is what arms the set_patch_state trip wire: the
// question that wire asks is whether the chunk-P0 bytes in the working block are the
// APPLIER'S, and only a scatter puts them there.
constexpr Uint32 kChunksPatchTrio = MGP_CHUNKS_OF(PatchVertices) |
MGP_CHUNKS_OF(PatchDefaultOuterLevel) |
MGP_CHUNKS_OF(PatchDefaultInnerLevel);
// Which chunks ONE capability's answer is read out of - the same sources
// DeriveCapability reads, written from the same list so the two cannot drift. This is
// what arms the residual trip wire PER CAPABILITY: a bind alone owns the pipeline
// half, and the eight ClipDistances are answered from ClipDistanceEnabledMask in
// DYNAMIC chunk D7, so between a bind and the first set_dynamic_state exactly those
// eight are unanswerable and the other 27 are not.
constexpr Uint32 CapabilitySourceChunks(CapabilityInput cap) {
#define MGP_CAPABILITY_SOURCE(capability) \
case CapabilityInput::capability: \
return MGP_CHUNKS_OF(capability##Enabled);
switch (cap) {
MGP_PLAIN_CAPABILITY_LIST(MGP_CAPABILITY_SOURCE)
case CapabilityInput::Blend:
return MGP_CHUNKS_OF(BlendStates);
case CapabilityInput::ScissorTest:
return MGP_CHUNKS_OF(ScissorTestEnabledMask);
case CapabilityInput::ClipDistance0:
case CapabilityInput::ClipDistance1:
case CapabilityInput::ClipDistance2:
case CapabilityInput::ClipDistance3:
case CapabilityInput::ClipDistance4:
case CapabilityInput::ClipDistance5:
case CapabilityInput::ClipDistance6:
case CapabilityInput::ClipDistance7:
return MGP_CHUNKS_OF(ClipDistanceEnabledMask);
// A capability with no storage cannot be answered from any byte, and
// DeriveCapability says so with a compile-time false. Demanding the whole table
// keeps such a value out of the comparison until every chunk is owned, which is
// the conservative direction: a wire that cannot be answered must not fire.
default:
return kMGPipeAllGlobalChunks;
}
#undef MGP_CAPABILITY_SOURCE
}
// The scalar copies are left unguarded on purpose: they are ~20 stores and two
// 28-byte struct copies, so guarding each would cost more branches than it saves
// stores - and an unguarded copy cannot go stale, which keeps the risk of the scoping
// confined to the four guards above.
#undef MGP_CHUNKS_OF
} // namespace
// The applier's door into PipeInputs' storage, the write-side twin of PipeFill.cpp's
// MGPipeFillAccess. It does NOT stamp the poison generations: a stamp says "the filler
// published this field for THIS verb", and that statement belongs to the walk that
// called the applier, not to the applier - MG_Impl/Pipe/PipeFill.cpp stamps what it
// emitted, exactly as it stamps what it copied.
struct MGPipeApplyAccess {
static RenderStateParameters& RenderState(PipeInputs& inputs) { return inputs.m_renderState; }
static PixelStoreParameters& PackState(PipeInputs& inputs) { return inputs.m_pixelStore[0]; }
static Bool* Capabilities(PipeInputs& inputs) { return inputs.m_capability; }
static PipeInputs::CurrentVertexAttributeValue* VertexAttribDefaults(PipeInputs& inputs) {
return inputs.m_currentVertexAttribute;
}
static void SetRenderStateVersions(PipeInputs& inputs, Uint parameters, Uint pipeline) {
inputs.m_renderStateParametersVersion = parameters;
inputs.m_pipelineStateVersion = pipeline;
}
static void SetRenderStateParametersVersion(PipeInputs& inputs, Uint parameters) {
inputs.m_renderStateParametersVersion = parameters;
}
static void SetPatchState(PipeInputs& inputs, Uint vertices, const FloatVec4& outer,
const FloatVec2& inner) {
inputs.m_patchVertices = vertices;
inputs.m_patchDefaultOuterLevel = outer;
inputs.m_patchDefaultInnerLevel = inner;
}
// P5c (rv, CONTRACT-P5C.md §5.3): set_context_values' write, one field per payload
// member, on the server side of the wire. Like PackState it is a plain store - the
// stamp is the filler/verb boundary's statement, not the applier's (see the struct's
// header comment).
static void SetContextValues(PipeInputs& inputs, const MGPContextValues& values) {
inputs.m_activeTextureUnit = static_cast<Int>(values.ActiveTextureUnit);
inputs.m_maxTouchedTextureUnit = static_cast<Int>(values.MaxTouchedTextureUnit);
for (SizeT i = 0; i < PipeInputs::kBufferTargetCount; ++i) {
inputs.m_touchedBindingPointCount[i] =
static_cast<SizeT>(values.TouchedBufferBindingPointCount[i]);
}
inputs.m_transformFeedbackActive = values.IsTransformFeedbackActive != 0;
inputs.m_transformFeedbackPaused = values.IsTransformFeedbackPaused != 0;
inputs.m_transformFeedbackGeneration = values.TransformFeedbackGeneration;
inputs.m_boundTransformFeedbackLifetimeId = values.BoundTransformFeedbackLifetimeId;
inputs.m_transformFeedbackCapturedVertices = values.TransformFeedbackCapturedVertices;
}
// ----------------------------------------------------------------------------
// D5: the 29 PipeInputs fields that are PURE FUNCTIONS of RenderStateParameters.
//
// Once bind_render_state / set_dynamic_state have assembled the working block,
// copying these out of GLContext a second time would be exactly the per-verb pull
// P2 exists to remove - so the applier DERIVES them instead. Each line below is a
// transcription of the RenderState getter of the same name (RenderState.cpp);
// GLContext's accessors are one-line forwards to those, so this block and the pull
// path answer the same question from the same bytes.
//
// This departs from P1 brief D4's "no derivation logic is re-implemented in
// PipeInputs", deliberately and with a guard: MOBILEGL_PIPE_VERIFY's compare-at-read
// re-reads every one of these from the live context AT EVERY BACKEND READ and
// compares field-wise, so a transcription error is caught on the first draw that
// reads it. RenderStateSpansTest.DerivationMatchesTheFrontendGetters walks every
// setter and checks all 29 against GLContext on top of that.
// ----------------------------------------------------------------------------
// RenderState::IsCapabilityEnabled, transcribed against the assembled block. One of
// the two derivations that is not a field copy, and the reason D3's three storage
// holes had to close FIRST: before P2, DepthClamp, FramebufferSrgb and
// TextureCubeMapSeamless fell to `default: return false` and this could not have
// been written at all.
static Bool DeriveCapability(const RenderStateParameters& p, CapabilityInput cap) {
#define MGP_DERIVE_CAPABILITY(capability) \
case CapabilityInput::capability: \
return p.capability##Enabled;
switch (cap) {
MGP_PLAIN_CAPABILITY_LIST(MGP_DERIVE_CAPABILITY)
// The non-indexed query of an INDEXED capability answers for index 0
// (GL 4.6 core 22.1) - RenderState.cpp says it in the same words.
case CapabilityInput::Blend:
return p.BlendStates[0].Enabled;
case CapabilityInput::ScissorTest:
return (p.ScissorTestEnabledMask & 1u) != 0;
// CapabilityInput lists ClipDistance0..7 contiguously, so the subtraction below
// is in range for exactly the eight values that reach here - RenderState.cpp's
// file-local ClipDistanceBit is the same expression.
case CapabilityInput::ClipDistance0:
case CapabilityInput::ClipDistance1:
case CapabilityInput::ClipDistance2:
case CapabilityInput::ClipDistance3:
case CapabilityInput::ClipDistance4:
case CapabilityInput::ClipDistance5:
case CapabilityInput::ClipDistance6:
case CapabilityInput::ClipDistance7:
return (p.ClipDistanceEnabledMask &
(1u << (static_cast<Uint>(cap) - static_cast<Uint>(CapabilityInput::ClipDistance0)))) != 0;
default:
return false;
}
#undef MGP_DERIVE_CAPABILITY
}
// `chunkBits` names the GLOBAL chunks the scatter that called this actually moved;
// kMGPipeAllGlobalChunks is the whole-block form. See the guard constants at the top
// of this file for why the four wide walks are scoped and the scalars are not.
static void DeriveRenderStateFields(PipeInputs& inputs, Uint32 chunkBits) {
const RenderStateParameters& p = inputs.m_renderState;
// Per draw buffer: GetBlendEquationIndexed, GetBlendFuncIndexed,
// GetColorMaskIndexed and IsCapabilityEnabledIndexed(Blend).
if ((chunkBits & kChunksBlendLoop) != 0) {
for (Uint i = 0; i < kMGMaxDrawBuffers; ++i) {
const PerBufferBlendState& blend = p.BlendStates[i];
inputs.m_blendEquation[i][0] = blend.ColorEquation;
inputs.m_blendEquation[i][1] = blend.AlphaEquation;
inputs.m_blendFunc[i][0] = blend.SrcFactorRGB;
inputs.m_blendFunc[i][1] = blend.DstFactorRGB;
inputs.m_blendFunc[i][2] = blend.SrcFactorAlpha;
inputs.m_blendFunc[i][3] = blend.DstFactorAlpha;
inputs.m_colorMask[i] = p.ColorMasks[i];
inputs.m_capabilityIndexed.Blend[i] = blend.Enabled;
}
}
// GetViewportIndexed, and GetViewport: viewport 0 ROUNDED - the other derivation
// that is not a field copy. glGetIntegerv on floating-point state rounds to
// nearest (GL 4.6 core 22.2), and truncating a 63.5-wide viewport would also hand
// the backends a rectangle one pixel short of what was asked for. std::lround,
// exactly as RenderState::GetViewport does it.
if ((chunkBits & kChunksViewportLoop) != 0) {
for (Uint i = 0; i < PipeInputs::kMaxViewports; ++i) {
inputs.m_viewportIndexed[i] = p.Viewports[i];
}
const FloatVec4& viewport = p.Viewports[0];
inputs.m_viewport =
IntVec4(static_cast<Int>(std::lround(viewport.x())), static_cast<Int>(std::lround(viewport.y())),
static_cast<Int>(std::lround(viewport.z())), static_cast<Int>(std::lround(viewport.w())));
}
// GetDepthRangeIndexed. Its own chunk (D2) - a glClearColor moves that chunk and
// a glViewport does not, so it cannot ride with the viewports.
if ((chunkBits & kChunksDepthRangeLoop) != 0) {
for (Uint i = 0; i < PipeInputs::kMaxViewports; ++i) {
inputs.m_depthRange[i] = p.DepthRanges[i];
}
}
// IsCapabilityEnabledIndexed(ScissorTest): 16 bits of one pipeline word.
if ((chunkBits & kChunksScissorEnableLoop) != 0) {
for (Uint i = 0; i < PipeInputs::kMaxViewports; ++i) {
inputs.m_capabilityIndexed.ScissorTest[i] = (p.ScissorTestEnabledMask & (1u << i)) != 0;
}
}
// The scalar copies, in the order MGP_COVERAGE_EMITTED_LIST names them.
inputs.m_blendColor = p.BlendColor;
inputs.m_clampReadColor = p.ClampReadColor;
inputs.m_clearColor = p.ClearColor;
inputs.m_clearDepth = p.ClearDepth;
inputs.m_clearStencil = p.ClearStencil;
inputs.m_cullFaceMode = p.CullFaceModeSetting;
inputs.m_depthFunc = p.DepthFunc;
inputs.m_depthMask = p.DepthMask;
inputs.m_lineWidth = p.LineWidth;
inputs.m_logicOp = p.LogicOp;
inputs.m_minSampleShadingValue = p.MinSampleShadingValue;
inputs.m_patchDefaultInnerLevel = p.PatchDefaultInnerLevel;
inputs.m_patchDefaultOuterLevel = p.PatchDefaultOuterLevel;
inputs.m_patchVertices = p.PatchVertices;
inputs.m_polygonModeFront = p.PolygonModeFront;
inputs.m_polygonOffsetFactor = p.PolygonOffsetFactor;
inputs.m_polygonOffsetUnits = p.PolygonOffsetUnits;
inputs.m_primitiveRestartIndex = p.PrimitiveRestartIndex;
inputs.m_provokingVertexMode = p.ProvokingVertexModeSetting;
// GetScissorBox answers for rectangle 0, like GetViewport - but WITHOUT any
// rounding, because the scissor rectangle is integer state to begin with.
inputs.m_scissorBox = p.ScissorBoxes[0];
// GetStencilState: Front is index 0 and Back is index 1 on both sides
// (RenderState.cpp's GetStencilFaceIndex and PipeInputs::GetStencilState agree),
// so the two faces copy straight across.
for (SizeT face = 0; face < PipeInputs::kStencilFaceCount; ++face) {
inputs.m_stencil[face] = p.StencilStates[face];
}
// The 35-arm switch, dispatched 35 times. The widest single thing the derivation
// does, and the one a per-frame glViewport most obviously must not pay for.
if ((chunkBits & kChunksCapabilityWalk) != 0) {
for (SizeT i = 0; i < PipeInputs::kCapabilityCount; ++i) {
inputs.m_capability[i] = DeriveCapability(p, static_cast<CapabilityInput>(i));
}
}
}
};
namespace {
// CapabilityInput in enum order, so the residual block's bit i and this name agree by
// construction. The static_assert below is what makes a capability added to the enum
// without a name here a build break rather than an "<unknown>" in a Fatal line.
constexpr const char* kCapabilityNames[] = {
"Blend",
"ClipDistance0",
"ClipDistance1",
"ClipDistance2",
"ClipDistance3",
"ClipDistance4",
"ClipDistance5",
"ClipDistance6",
"ClipDistance7",
"ColorLogicOp",
"CullFace",
"DebugOutput",
"DebugOutputSynchronous",
"DepthClamp",
"DepthTest",
"Dither",
"FramebufferSrgb",
"LineSmooth",
"Multisample",
"PolygonOffsetFill",
"PolygonOffsetLine",
"PolygonOffsetPoint",
"PolygonSmooth",
"PrimitiveRestart",
"PrimitiveRestartFixedIndex",
"RasterizerDiscard",
"SampleAlphaToCoverage",
"SampleAlphaToOne",
"SampleCoverage",
"SampleShading",
"SampleMask",
"ScissorTest",
"StencilTest",
"TextureCubeMapSeamless",
"ProgramPointSize",
};
constexpr SizeT kCapabilityCount = static_cast<SizeT>(CapabilityInput::CapabilityInputCount);
static_assert(sizeof(kCapabilityNames) / sizeof(kCapabilityNames[0]) == kCapabilityCount,
"CapabilityInput gained a value; name it here or the residual trip wire "
"cannot say which capability diverged");
static_assert(kCapabilityCount <= 64,
"ResidualValueBlock::CapabilityBits is a Uint64; 35 bits fit, 65 would not");
// ============================================================================
// P14 S5: THE APPLIER REGISTRY.
// ============================================================================
//
// A REFERENCE TO A NEVER-DESTROYED BLOCK, for MGPipeSlots()' reason
// (MG_Impl/Pipe/SlotAllocator.cpp): resource_destroy and delete_vertex_elements are
// raised from ~BufferObject / ~VertexArrayObject, and those objects are released by
// exit handlers that run after this translation unit's own globals are gone.
//
// THE DEFAULT IS THE OLD SINGLETON, unrenamed in what it means: the applier of a thread
// that belongs to nobody - the client process, the monolith, the inproc one-process
// shape, every unit case. Its object records are the group it owns alone.
MGPipeObjectRecords& g_defaultObjects = *new MGPipeObjectRecords{};
MGPipeApplierState& g_defaultApplier = *new MGPipeApplierState(g_defaultObjects);
// Installed once by the session registry (MG_Remote/Server/SessionRuntime.cpp), never
// replaced. Null in every process that serves no session, which is what keeps the
// monolith's `MGPipeApplier()` a load and a branch.
MGPipeApplierKeyResolver g_applierKeyResolver = nullptr;
// One served session's appliers. Groups own object records; Contexts own working state
// and point at their group. Both are held through UniquePtr so a table's ADDRESS is
// stable across a rehash - MGPipeApplierState binds its group by reference - and the
// maps are std::unordered_map rather than the house UnorderedMap (a flat hash map) for
// that same reason at the map level: a reference into it may not be invalidated by an
// insert somewhere else in the same call.
struct ApplierSession {
std::unordered_map<Uint64, UniquePtr<MGPipeObjectRecords>> Groups;
std::unordered_map<Uint64, UniquePtr<MGPipeApplierState>> Contexts;
std::unordered_map<Uint64, Uint64> ContextGroup;
std::unordered_map<Uint64, Uint32> GroupContexts;
};
std::unordered_map<Uint64, ApplierSession> g_applierSessions;
std::mutex g_applierSessionsMutex;
// Bumped by EVERY registry mutation. A thread's cached applier pointer is only trusted
// while it was bound at this generation, which is what makes "a context was destroyed on
// another thread" unable to leave a dangling cache behind.
std::atomic<Uint64> g_applierRegistryGeneration{1};
// A context that started its own share group is keyed by its own token, in a namespace
// of its own: the frontend mints share-group tokens and context tokens from separate
// counters, so without this a group 7 and a context 7 would be one group.
constexpr Uint64 kMGPipeOwnShareGroupBit = Uint64{1} << 63;
// `shareGroupToken` 0 means "this context started its own group" - the sentinel S1's
// ContextRuntime documents and the one a caller that frames a context by hand (a test,
// the in-process server helpers) leaves behind.
Uint64 ApplierShareGroupKey(Uint64 contextToken, Uint64 shareGroupToken) {
return shareGroupToken != 0 ? shareGroupToken : (contextToken | kMGPipeOwnShareGroupBit);
}
// The three tables of the applier this thread's records are being applied through.
// `generation` is the registry's, read here so a caller can bind a pointer under one
// generation and publish it under the same one.
MGPipeApplierState* FindApplierForKey(const MGPipeApplierKey& key) {
std::lock_guard<std::mutex> guard(g_applierSessionsMutex);
ApplierSession& session = g_applierSessions[key.SessionKey];
const auto found = session.Contexts.find(key.ContextToken);
if (found != session.Contexts.end() && found->second) return found->second.get();
// A context the session never registered - a token that predates the first
// CreateContext, or the 0 of a session that has not bound one yet. It gets the same
// shape as any other: a working state in a group of its own.
const Uint64 groupKey = ApplierShareGroupKey(key.ContextToken, 0);
auto group = session.Groups.find(groupKey);
if (group == session.Groups.end()) {
group = session.Groups.emplace(groupKey, MakeUnique<MGPipeObjectRecords>()).first;
}
auto applier = MakeUnique<MGPipeApplierState>(*group->second);
MGPipeApplierState* const raw = applier.get();
session.Contexts.emplace(key.ContextToken, std::move(applier));
session.ContextGroup.emplace(key.ContextToken, groupKey);
++session.GroupContexts[groupKey];
return raw;
}
thread_local MGPipeApplierState* t_applierCached = nullptr;
thread_local Uint64 t_applierCachedSession = 0;
thread_local Uint64 t_applierCachedContext = 0;
thread_local Uint64 t_applierCachedGeneration = 0;
} // namespace
Uint64 MGPipeNextApplierSerialBase() {
// Constant-initialized, so the default applier's static construction may already use it.
static std::atomic<Uint64> appliers{0};
return (appliers.fetch_add(1, std::memory_order_relaxed) + 1) << 40;
}
MGPipeApplierState& MGPipeApplier() {
const MGPipeApplierKeyResolver resolve = g_applierKeyResolver;
if (resolve == nullptr) return g_defaultApplier;
MGPipeApplierKey key;
if (!resolve(&key)) return g_defaultApplier;
// THE FAST PATH, and the reason this is a cache rather than a lookup: this function is
// called from the backend per state read. A hit is three TLS loads, one atomic load and
// two compares; a miss is the registry, which only moves when a context is created,
// destroyed or made current.
const Uint64 generation = g_applierRegistryGeneration.load(std::memory_order_acquire);
if (t_applierCached != nullptr && t_applierCachedGeneration == generation &&
t_applierCachedSession == key.SessionKey && t_applierCachedContext == key.ContextToken) {
return *t_applierCached;
}
MGPipeApplierState* const applier = FindApplierForKey(key);
t_applierCached = applier;
t_applierCachedSession = key.SessionKey;
t_applierCachedContext = key.ContextToken;
t_applierCachedGeneration = generation;
return *applier;
}
void MGPipeSetApplierKeyResolver(MGPipeApplierKeyResolver resolver) { g_applierKeyResolver = resolver; }
void MGPipeApplierRegisterContext(Uint64 sessionKey, Uint64 contextToken, Uint64 shareGroupToken) {
std::lock_guard<std::mutex> guard(g_applierSessionsMutex);
ApplierSession& session = g_applierSessions[sessionKey];
const Uint64 groupKey = ApplierShareGroupKey(contextToken, shareGroupToken);
auto group = session.Groups.find(groupKey);
if (group == session.Groups.end()) {
group = session.Groups.emplace(groupKey, MakeUnique<MGPipeObjectRecords>()).first;
}
const auto existing = session.Contexts.find(contextToken);
if (existing != session.Contexts.end() && existing->second) {
// Idempotent: the same token registered twice keeps the state it already has. The
// session refuses a duplicate CreateContext before reaching here, so this arm is a
// caller that re-registers deliberately.
g_applierRegistryGeneration.fetch_add(1, std::memory_order_acq_rel);
return;
}
session.Contexts.emplace(contextToken, MakeUnique<MGPipeApplierState>(*group->second));
session.ContextGroup[contextToken] = groupKey;
++session.GroupContexts[groupKey];
g_applierRegistryGeneration.fetch_add(1, std::memory_order_acq_rel);
}
void MGPipeApplierUnregisterContext(Uint64 sessionKey, Uint64 contextToken) {
std::lock_guard<std::mutex> guard(g_applierSessionsMutex);
const auto sessionIt = g_applierSessions.find(sessionKey);
if (sessionIt == g_applierSessions.end()) return;
ApplierSession& session = sessionIt->second;
const auto context = session.Contexts.find(contextToken);
if (context == session.Contexts.end()) return;
const auto groupIt = session.ContextGroup.find(contextToken);
const Uint64 groupKey = groupIt != session.ContextGroup.end() ? groupIt->second : 0;
// The OBJECT RECORDS GO WITH THE LAST CONTEXT OF THE GROUP, which is the per-group
// spelling of MGPipeApplierReleaseObjectRecords' scope. A surviving context of the same
// group keeps every record - that is what sharing means.
const auto refs = session.GroupContexts.find(groupKey);
const Bool lastOfGroup = refs != session.GroupContexts.end() && refs->second <= 1;
session.Contexts.erase(context);
if (groupIt != session.ContextGroup.end()) session.ContextGroup.erase(groupIt);
if (lastOfGroup) {
if (refs != session.GroupContexts.end()) session.GroupContexts.erase(refs);
const auto group = session.Groups.find(groupKey);
if (group != session.Groups.end()) session.Groups.erase(group);
} else if (refs != session.GroupContexts.end()) {
--refs->second;
}
if (session.Contexts.empty()) g_applierSessions.erase(sessionIt);
g_applierRegistryGeneration.fetch_add(1, std::memory_order_acq_rel);
}
void MGPipeApplierReleaseSession(Uint64 sessionKey) {
std::lock_guard<std::mutex> guard(g_applierSessionsMutex);
if (g_applierSessions.erase(sessionKey) == 0) return;
g_applierRegistryGeneration.fetch_add(1, std::memory_order_acq_rel);
}
Uint64 MGPipeApplierShareGroupCountForTesting(Uint64 sessionKey) {
std::lock_guard<std::mutex> guard(g_applierSessionsMutex);
const auto sessionIt = g_applierSessions.find(sessionKey);
return sessionIt == g_applierSessions.end() ? 0 : static_cast<Uint64>(sessionIt->second.Groups.size());
}
Uint64 MGPipeApplierShareGroupKeyFor(Uint64 sessionKey, Uint64 contextToken) {
std::lock_guard<std::mutex> guard(g_applierSessionsMutex);
const auto sessionIt = g_applierSessions.find(sessionKey);
if (sessionIt != g_applierSessions.end()) {
const auto groupIt = sessionIt->second.ContextGroup.find(contextToken);
if (groupIt != sessionIt->second.ContextGroup.end()) return groupIt->second;
}
// Not registered - the same fallback FindApplierForKey takes, spelled by the same one
// function so the two cannot drift.
return ApplierShareGroupKey(contextToken, 0);
}
namespace {
// The installed handle-shaped resource table. Null until a backend registers one,
// which is what makes the client half landable on its own: with nothing here every
// frontend dispatch falls through to the op table this one replaces, and the tree
// behaves exactly as it did.
const MGPipeResourceOps* g_resourceOps = nullptr;
MGPipeRenderStateCsoRecord* FindCso(MGPipeHandle handle) {
if (handle.Slot >= MGPipeApplier().RenderStateCsos.size()) return nullptr;
MGPipeRenderStateCsoRecord& record = MGPipeApplier().RenderStateCsos[handle.Slot];
if (!record.Live || record.Gen != handle.Gen) return nullptr;
return &record;
}
constexpr Uint32 kAllPipelineChunks =
static_cast<Uint32>((Uint64{1} << kMGPipePipelineChunkCount) - 1);
// ----------------------------------------------------------------------------
// P3a: resolving a handle, growing a slot table, and the bounds gate.
// ----------------------------------------------------------------------------
// Grows a slot-indexed record table so `slot` is in it, or returns NULL when the slot
// is outside the table's bound. Slot spaces are DENSE per kind - the allocator is a
// free list plus a high-water mark - which is exactly why the server's object table is
// an array a handle indexes rather than a map, and why this grows only when a new
// high-water mark arrives.
//
// AND WHY IT IS BOUNDED. `slot` is a client-supplied Uint32 that arrives in a payload,
// and this is the one number in the family that reaches an ALLOCATOR: unbounded, a
// corrupt 0xFFFFFFFE asks for a four-billion-entry vector from inside the same commit
// that polices Blob.Size, the destination range, Level, RegionCount and Start + Count.
// The bounds are kMGPipeMax{Resource,VertexElements}Slots (PipeApply.h) and a slot at
// or above one is the callers' Fatal{ProtocolCorruption}, with the identity in the
// line like its siblings - never a resize.
template <class Record>
Record* RecordAt(Vector<Record>& records, Uint32 slot, Uint32 slotLimit) {
if (slot >= slotLimit) return nullptr;
if (slot >= records.size()) records.resize(static_cast<SizeT>(slot) + 1);
return &records[slot];
}
// Null means "this applier does not have that resource": an out-of-range slot, a slot
// that is not live, or a handle whose generation has moved on because the slot was
// recycled under it. FindCso above is the same three questions for the CSO store.
MGPipeResourceRecord* FindResource(MGPipeHandle res) {
if (res.Slot >= MGPipeApplier().Resources.size()) return nullptr;
MGPipeResourceRecord& record = MGPipeApplier().Resources[res.Slot];
if (!record.Live || record.Gen != res.Gen) return nullptr;
return &record;
}
MGPipeVertexElementsRecord* FindVertexElements(MGPipeHandle cso) {
if (cso.Slot >= MGPipeApplier().VertexElementsCsos.size()) return nullptr;
MGPipeVertexElementsRecord& record = MGPipeApplier().VertexElementsCsos[cso.Slot];
if (!record.Live || record.Gen != cso.Gen) return nullptr;
return &record;
}
// ----------------------------------------------------------------------------
// P4a: THE SAME THREE QUESTIONS, ASKED OF A TABLE RATHER THAN OF THE ONE TABLE.
//
// The slot spaces of kinds Buffer, Texture and Renderbuffer are INDEPENDENT - the
// allocator is per kind - so three different live objects can hold slot 7 at once and
// one slot-indexed table would alias all three onto one record. The record TYPE is
// shared, because the descriptor is one discriminated descriptor; only the table is
// per kind, and choosing it is what the two selectors below do.
// ----------------------------------------------------------------------------
template <class Record>
Record* FindIn(Vector<Record>& records, MGPipeHandle handle) {
if (handle.Slot >= records.size()) return nullptr;
Record& record = records[handle.Slot];
if (!record.Live || record.Gen != handle.Gen) return nullptr;
return &record;
}
// Null means "this is not a resource target this catalogue names", which is a corrupt
// descriptor rather than an unknown object: acting on the wrong table would create,
// respecify or destroy an unrelated LIVE object that happens to hold the same slot in
// another kind's space, and that is the "act outside its own storage" class.
Vector<MGPipeResourceRecord>* ResourceTableForTarget(Uint8 target) {
if (target == kMGPipeResourceTargetBuffer) return &MGPipeApplier().Resources;
if (target == static_cast<Uint8>(MGPipeResourceTarget::Renderbuffer)) {
return &MGPipeApplier().RenderbufferResources;
}
if (target < static_cast<Uint8>(MGPipeResourceTarget::Count)) {
// Every remaining enumerator is a texture target, and they share one table
// because they share one kind: MGPipeKind::Texture. Tex1D..TexCubeArray,
// Tex2DMS/MSArray, TexBuffer and TexRect are all one slot space.
return &MGPipeApplier().TextureResources;
}
return nullptr;
}
// resource_destroy carries no descriptor, so its discriminator is the handle's KIND.
Vector<MGPipeResourceRecord>* ResourceTableForKind(Uint32 kind) {
switch (static_cast<MGPipeKind>(kind)) {
case MGPipeKind::Buffer:
return &MGPipeApplier().Resources;
case MGPipeKind::Texture:
return &MGPipeApplier().TextureResources;
case MGPipeKind::Renderbuffer:
return &MGPipeApplier().RenderbufferResources;
default:
return nullptr;
}
}
// A sub-data record's own discriminator, and it is DELIBERATELY NOT the table selector
// above: MGPSubData::Target carries the UPLOAD target as well - a cube face is one, and
// those are not MGPipeResourceTarget enumerators - so the buffer question is asked of
// the whole field, which is the one question that has an answer for every value.
// kMGPipeResourceTargetBuffer is 0, a buffer has no upload target to pack beside it, and
// no texture emission leaves the whole field 0: that is the contract the emitter is
// held to.
Bool SubDataNamesABuffer(const MGPSubData& record) {
return record.Target == kMGPipeResourceTargetBuffer;
}
// THE VALUE SPACE OF MGPSubData::Target IS THE CONTRACT'S, AND THIS FILE NO LONGER
// RESTATES IT. c0c (ID-12, the DV-3 seam) moved the encoding into MGPipeTypes.h as
// MGPipePackSubDataTarget / MGPipeSubDataResourceTargetOf / MGPipeSubDataUploadTargetOf,
// so wire v2's local `SubDataResourceTargetOf` - written because those helpers were not
// on that round's base - is gone and every half-read below goes through the contract's
// accessor. Nothing here open-codes `& 0xFF`.
//
// The applier still matches the WHOLE field for the buffer question above, exactly as
// ID-12 says it does, and c0c's own static_assert is what keeps the two readings in
// step: a buffer record's Target is exactly kMGPipeResourceTargetBuffer.
// ID-18 M4: DOES THIS RESPECIFY REDEFINE ANY STORAGE, OR IS IT CARRYING METADATA?
//
// The question exists because a sticky BindMask / ImageBindableHint bit has exactly one
// way onto the wire - a respecify - and an IMMUTABLE texture has no further respecify to
// ride on. So B re-publishes the descriptor when the mask moves, and the applier has to
// tell that call apart from a real redefinition: the first must NOT drop the pending
// uploads standing against the storage (nothing replaced them), the second must.
//
// THE COMPARISON IS FIELD BY FIELD AND NOT A memcmp. MGPResourceDesc carries Pad0/Pad1
// and neither side of this comparison is guaranteed to have written them - the client
// fills a stack descriptor field by field - so a byte comparison of 88 bytes is a coin
// flip on padding, which is the defect clientsp's SamplerParameters cache already paid
// for once.
//
// WHAT IS NOT COMPARED IS THE SHORT LIST AND IT IS THE DEFINITION: BindMask and
// ImageBindableHint (the two this call exists to carry), GlNameForDiag (diagnostics, and
// never an identity - section 4.2.1) and the two padding words. EVERYTHING ELSE IS
// COMPARED, which is deliberately WIDER than "the fields that define an allocation":
// being wide can only mis-classify a metadata update as a redefinition, whose cost is
// one level's pending upload dropped by a call B does not emit; being narrow would
// mis-classify a REDEFINITION as metadata and keep boxes in a coordinate system that no
// longer exists, which is an upload past the end of the new level. The asymmetry decides
// the direction.
//
// A BUFFER IS NEVER METADATA-ONLY, and that is a rule rather than a consequence.
// glBufferData at an unchanged size is the canonical ORPHANING idiom - a real
// reallocation whose whole purpose is that the old store is gone - and glBufferStorage
// is the one entry point in the catalogue allowed a synchronous acknowledgement
// (MGPipeResourceRespecifyNeedsAck). Excluding the target here is what makes ID-18 M4's
// "no reallocation ack" true by construction instead of by a second suppression rule,
// and it costs nothing: the mask that M4 is about is a TEXTURE's (and a renderbuffer's),
// and a buffer has no pending upload for the arm to protect anyway.
Bool RespecifyRedefinesNoStorage(const MGPResourceDesc& stored, const MGPResourceDesc& next) {
if (next.Target == kMGPipeResourceTargetBuffer) return false;
if (stored.Target != next.Target) return false;
if (stored.StorageKind != next.StorageKind) return false;
if (stored.InternalFormat != next.InternalFormat) return false;
if (stored.Width != next.Width || stored.Height != next.Height ||
stored.Depth != next.Depth) {
return false;
}
if (stored.ArrayLayers != next.ArrayLayers || stored.Levels != next.Levels ||
stored.Samples != next.Samples) {
return false;
}
if (stored.FixedSampleLocations != next.FixedSampleLocations) return false;
if (stored.Immutable != next.Immutable) return false;
if (stored.HasDefinedContent != next.HasDefinedContent) return false;
if (stored.Usage != next.Usage || stored.StorageFlags != next.StorageFlags) return false;
// The view and buffer-texture fields: a texture view's storage OWNER and a buffer
// texture's backing range are its storage exactly as an extent is one, and
// re-pointing either is a redefinition however unchanged the extent looks.
if (!(stored.ViewOf == next.ViewOf)) return false;
if (!(stored.BufferForTexBuffer == next.BufferForTexBuffer)) return false;
if (stored.BufOffset != next.BufOffset || stored.BufSize != next.BufSize) return false;
return true;
}
// WHY A DEAD HANDLE IS NOT A TRIP WIRE HERE, and the bounds faults below are - and
// why it is nonetheless COUNTED rather than silently dropped.
//
// A make-current no longer takes the records with it (MGPipeApplierReset), so the
// shared-store case that used to arrive here every context switch does not arrive at
// all: a buffer that outlives a switch keeps its record and its writes keep landing.
// What is left is (a) a genuine protocol error - an unknown slot, a stale generation -
// and (b) ONE legal sequence, which is why this is still a no-op and not a wire:
//
// the served context is torn down
// -> MGPipeApplierReleaseObjectRecords() (the applier goes away with it)
// -> ~BufferObject / ~VertexArrayObject for every object the context still owns
// -> resource_destroy / delete_vertex_elements, each naming a record that the
// line above has already dropped.
//
// Every one of those death notices is legal, unavoidable and arrives after the
// records are gone, and a wire here would abort a verify lane on the ordinary shutdown
// of a context. So the refusal stays a DEFINED no-op - nothing stored, nothing
// dispatched, no serial moved, and the debug assertion names it, which is the shape
// bind_render_state already uses for a dead CSO.
//
// But MOBILEGL_ASSERT compiles out at INFO, which is what all three gate builds and
// every shipped build are, so a no-op alone would make case (a) - a dropped
// glBufferSubData - invisible in every build that matters. Both refusal paths
// therefore go through ResolveResource / ResolveVertexElements below, which COUNT into
// MGPipeApplierState::Refused{Resource,VertexInput}Calls. That is the observable: a
// legal sequence leaves it at 0 and a refused call moves it, in every build.
//
// The faults below are the other class entirely: a record that does not describe its
// own bytes would have the BACKEND read or write outside a store, which is memory
// corruption rather than a dropped call, so those get the tag ARCHITECTURE.md reserves
// for exactly this - Fatal{ProtocolCorruption} - and the record's identity in the line.
constexpr const char* kResourceRefusalNote =
"the record is not this applier's; the call is dropped, not applied";
// The two resolvers every entry point below uses. One place resolves, asserts and
// counts, so a call that forgets one of the three cannot exist.
MGPipeResourceRecord* ResolveResource(const char* call, MGPipeHandle res) {
MGPipeResourceRecord* record = FindResource(res);
MOBILEGL_ASSERT(record != nullptr, "%s named {slot=%u, gen=%u}: %s", call, res.Slot, res.Gen,
kResourceRefusalNote);
if (record == nullptr) ++MGPipeApplier().RefusedResourceCalls;
return record;
}
// P4a: the same resolver over a chosen table. A texture's and a renderbuffer's
// resource_* calls are RESOURCE calls and count into the resource counter, exactly as a
// buffer's do; the object counter beside it is for the five families that have no
// resource call at all (framebuffer, sampler, sampler view, program, texture params).
MGPipeResourceRecord* ResolveResourceIn(Vector<MGPipeResourceRecord>& table, const char* call,
MGPipeHandle res) {
MGPipeResourceRecord* record = FindIn(table, res);
MOBILEGL_ASSERT(record != nullptr, "%s named {slot=%u, gen=%u}: %s", call, res.Slot, res.Gen,
kResourceRefusalNote);
if (record == nullptr) ++MGPipeApplier().RefusedResourceCalls;
return record;
}
// P4a's object families. ONE counter for the five, because they share one legal refusal
// sequence - the teardown order beside kResourceRefusalNote - and because what an
// operator reading a log needs to know is that an object call was dropped; the line
// itself names which call and which handle.
template <class Record>
Record* ResolveObject(Vector<Record>& table, const char* call, MGPipeHandle handle) {
Record* record = FindIn(table, handle);
MOBILEGL_ASSERT(record != nullptr, "%s named {slot=%u, gen=%u}: %s", call, handle.Slot,
handle.Gen, kResourceRefusalNote);
if (record == nullptr) ++MGPipeApplier().RefusedObjectCalls;
return record;
}
MGPipeVertexElementsRecord* ResolveVertexElements(const char* call, MGPipeHandle cso) {
MGPipeVertexElementsRecord* record = FindVertexElements(cso);
MOBILEGL_ASSERT(record != nullptr, "%s named {slot=%u, gen=%u}: %s", call, cso.Slot, cso.Gen,
kResourceRefusalNote);
if (record == nullptr) ++MGPipeApplier().RefusedVertexInputCalls;
return record;
}
// Returns the fault, or null when [offset, offset+size) lies inside `width` bytes.
// Written so nothing can overflow: `offset > width` is answered before the subtraction
// that the second question needs.
const char* BufferRangeFault(Uint64 offset, Uint64 size, Uint64 width) {
if (offset > width) return "the offset starts past the resource's declared storage";
if (size > width - offset) return "the range runs past the resource's declared storage";
return nullptr;
}
// The buffer half of MGPSubData is a CONVENTION over a texture record's box
// (MGPipeTypes.h): offset in UnionBox.X, size in UnionBox.W, no level and no regions.
// MGPipeSetSubDataBufferRange is its only encoder, so every field it writes is a field
// the applier can hold the record to - which is what makes a hand-rolled or corrupted
// record visible instead of being read as a plausible range.
//
// P4a SPLIT IT IN TWO RATHER THAN WIDENING IT. Every statement below is a statement
// about the BUFFER convention - "no level", "no sub-regions", "the blob is the box's
// own byte size" - and every one of them is false for a texture, which carries a real
// level, a real region list and a blob whose length no other field describes. A single
// predicate that tried to hold both would have to be right about which record it was
// looking at anyway, so the branch is at the call and each half states only what it
// can actually check. SubDataTextureFault is the other half.
const char* SubDataBoxFault(const MGPSubData& record) {
// The box's first coordinate is a signed Int32 on the wire and the encoder never
// writes a negative one; read back as unsigned (which is what the decoder does,
// deliberately, rather than sign-extending) a corrupt one lands above the
// encodable bound and is refused here.
if (MGPipeSubDataBufferOffset(record) > 0x7FFFFFFFull) {
return "the destination offset is above the bound one record can encode";
}
if (record.Level != 0) return "the buffer half carries a mip level";
if (record.RegionCount != 0) return "the buffer half carries sub-regions";
// THE BLOB RULE, and it is the SAME rule create_vertex_elements is held to
// (MGPipeTypes.h states it on both records): a declared blob length must be
// exactly the byte length the record's other fields describe, and a length of 0
// means "this record does not declare its blob" - which is what a monolith
// emission is, because the bytes travel beside the record through the entry
// point's companion pointer. So the gate is inert while the client leaves the
// field zero and becomes a real one on the first record a transport truncates.
if (record.Blob.Size != 0 && record.Blob.Size != MGPipeSubDataBufferSize(record)) {
return "the declared blob length is not the record's own byte size";
}
return nullptr;
}
// ----------------------------------------------------------------------------
// P4a: the texture half of the sub-data validator, and the pending-upload set.
// ----------------------------------------------------------------------------
// A box is EMPTY when any extent is zero, and an empty box unions to nothing. Written
// once because the accumulation below needs it in three places.
Bool BoxIsEmpty(const MGPBox& box) { return box.W == 0 || box.H == 0 || box.D == 0; }
// A box stays inside the range one record can encode: the origin is a signed Int32 and
// the extent an unsigned Uint32, so an origin plus an extent that leaves the positive
// Int32 range is a record no encoder writes and is refused before any arithmetic below
// has to survive it. It is the same bound, and the same reason, as the buffer half's
// "above the bound one record can encode".
Bool BoxIsEncodable(Int32 origin, Uint32 extent) {
return origin >= 0 && static_cast<Int64>(origin) + static_cast<Int64>(extent) <= 0x7FFFFFFF;
}
Bool BoxIsEncodable(const MGPBox& box) {
return BoxIsEncodable(box.X, box.W) && BoxIsEncodable(box.Y, box.H) &&
BoxIsEncodable(box.Z, box.D);
}
Bool AxisContains(Int32 outerOrigin, Uint32 outerExtent, Int32 innerOrigin, Uint32 innerExtent) {
const Int64 outerEnd = static_cast<Int64>(outerOrigin) + static_cast<Int64>(outerExtent);
const Int64 innerEnd = static_cast<Int64>(innerOrigin) + static_cast<Int64>(innerExtent);
return innerOrigin >= outerOrigin && innerEnd <= outerEnd;
}
// THE ONE INVARIANT THE TEXTURE HALF CAN ACTUALLY CHECK, and it is the one that matters:
// the union box IS the union of the regions (MipmapStorage maintains both through one
// MarkDirtyRegion, and "0 regions" means "the box is the whole story"). The SERVER picks
// the upload shape from the pair - one box job, or N rect jobs, and Mali prices that
// choice at ~6 ms/frame - so a region outside the box means the two shapes describe
// different texels and whichever the server picks is wrong: the box misses the region's
// texels, and the region writes where the box never said it would.
const char* SubDataTextureFault(const MGPSubData& record, const MGPSubRegion* regions) {
if (record.Level >= kMGPipeMaxTextureLevels) {
return "the level is above the bound any texture's storage can have";
}
if (!BoxIsEncodable(record.UnionBox)) {
return "the union box has a negative origin or runs past the bound one record can encode";
}
if (record.RegionCount > kMGPipeMaxPendingUploadRegions) {
return "the record declares more sub-regions than one upload may carry";
}
if (record.RegionCount != 0 && regions == nullptr) {
return "the record declares sub-regions and carries none";
}
for (Uint32 i = 0; i < record.RegionCount; ++i) {
const MGPSubRegion& region = regions[i];
if (!BoxIsEncodable(region.X, region.W) || !BoxIsEncodable(region.Y, region.H) ||
!BoxIsEncodable(region.Z, region.D)) {
return "a sub-region has a negative origin or runs past the bound one record can encode";
}
if (!AxisContains(record.UnionBox.X, record.UnionBox.W, region.X, region.W) ||
!AxisContains(record.UnionBox.Y, record.UnionBox.H, region.Y, region.H) ||
!AxisContains(record.UnionBox.Z, record.UnionBox.D, region.Z, region.D)) {
return "a sub-region is not inside the union box the record declares";
}
}
// WHAT IS DELIBERATELY NOT CHECKED, so that a later reader does not add it back as
// an oversight. (a) The level against MGPResourceDesc::Levels: a MUTABLE texture
// defines its levels one glTexImage2D at a time, so the descriptor's level count is
// not an upper bound at every instant and a gate on it would refuse a legal upload
// to a level the next respecify is about to declare. (b) The box against the
// descriptor's extents: the record addresses the LEVEL's coordinate system, and a
// view remaps that space, so the arithmetic is the storage owner's and not this
// applier's. (c) The blob: no field of a texture record describes its own byte
// length - the strides are per region and the level shadow's size is not carried -
// so the one Blob rule has nothing to cross-check here and stays inert by
// construction rather than by omission.
return nullptr;
}
MGPBox UnionOfBoxes(const MGPBox& a, const MGPBox& b) {
if (BoxIsEmpty(a)) return b;
if (BoxIsEmpty(b)) return a;
const Int32 x = a.X < b.X ? a.X : b.X;
const Int32 y = a.Y < b.Y ? a.Y : b.Y;
const Int32 z = a.Z < b.Z ? a.Z : b.Z;
const Int64 xEnd = std::max(static_cast<Int64>(a.X) + a.W, static_cast<Int64>(b.X) + b.W);
const Int64 yEnd = std::max(static_cast<Int64>(a.Y) + a.H, static_cast<Int64>(b.Y) + b.H);
const Int64 zEnd = std::max(static_cast<Int64>(a.Z) + a.D, static_cast<Int64>(b.Z) + b.D);
return MGPBox{x, y, z, static_cast<Uint32>(xEnd - x), static_cast<Uint32>(yEnd - y),
static_cast<Uint32>(zEnd - z)};
}
// D-D5's SAFETY NET, and the whole reason it is server-side state. The client clears
// its own dirty flags AT EMISSION, for the levels whose record this applier accepted;
// Espryt's upload loop has bail arms - an incomplete texture returns early, a
// multisample target refreshes and skips - that today leave the frontend flag set, so a
// naive move of the clear to the client would lose exactly those texels. The emitted
// shape accumulates here instead, it survives any number of bails, and Espryt consumes
// and clears an entry only where it actually uploads.
//
// ACCUMULATION IS THE CLIENT'S OWN MODEL, ONE LEVEL UP. MipmapStorage keeps a union box
// and, behind it, a bounded disjoint rect list, and answers "0 rects" for everything it
// cannot describe that way - which means "upload the box instead" and covers every
// reason at once. So: boxes union; rect lists concatenate; and the moment either side
// says "box only", or the list would outgrow its bound, the entry becomes box only.
// Never a dropped region - the box still covers every texel the dropped list named.
// Returns false when the record names more distinct (target, level) keys than any
// texture can have, which the caller reports as a corrupt record.
Bool AccumulatePendingUpload(MGPipeResourceRecord& stored, const MGPSubData& record,
const MGPSubRegion* regions) {
for (MGPipeResourceRecord::PendingUpload& entry : stored.PendingUploads) {
if (entry.UploadTarget != record.Target || entry.Level != record.Level) continue;
entry.UnionBox = UnionOfBoxes(entry.UnionBox, record.UnionBox);
if (record.RegionCount == 0 || entry.Regions.empty()) {
// A whole-level contribution (the emitter's box-only spelling since P11 M2)
// owns every texel of the level, so the entry is that box, exactly. The one
// other box-only record is the emitter's oversize fallback (a footprint past
// kMGPipeMaxPendingUploadRegions goes out as the storage's cover, which it
// withholds as 0 at 3/4 of the box), a partial box spanning gaps - already
// inexact, and logged once where it is emitted (TextureEmit.h).
entry.Regions.clear();
return true;
}
if (static_cast<Uint64>(entry.Regions.size()) + record.RegionCount > kMGPipeMaxPendingUploadRegions) {
// The one lossy answer left: the box covers every texel the list named, and
// also the gaps between them (see kMGPipeMaxPendingUploadRegions).
MGLOG_W_ONCE("MGPipe: a texture level's pending upload outgrew %u regions and becomes its union "
"box, which spans the gaps between the writes",
kMGPipeMaxPendingUploadRegions);
entry.Regions.clear();
return true;
}
entry.Regions.insert(entry.Regions.end(), regions, regions + record.RegionCount);
return true;
}
if (stored.PendingUploads.size() >= kMGPipeMaxPendingUploads) return false;
MGPipeResourceRecord::PendingUpload entry;
entry.UploadTarget = record.Target;
entry.Level = record.Level;
entry.UnionBox = record.UnionBox;
// A first contribution with no regions makes the entry BOX ONLY from the start, and
// that is why an empty list above means box only rather than "not filled in yet":
// an entry is never created without its first contribution.
if (record.RegionCount != 0) entry.Regions.assign(regions, regions + record.RegionCount);
stored.PendingUploads.push_back(std::move(entry));
return true;
}
#if MOBILEGL_PIPE_VERIFY && !MOBILEGL_BUILD_DISAGGREGATED
// D-A4's pin, AND P5 (b1) IS THE PHASE IT WAS WAITING FOR. It said "HasLiveHostWrites
// is always false in this phase and is written by nobody: it exists so the phase that
// pushes persistent-mapped host writes can set it with no new record kind", and a
// verify build refused to let such a producer arrive unannounced.
//
// The producer is announced: MGPSubData::HasLiveHostWrites, set by
// MGPipeEmitResourceSubData from BufferObject::HasLiveHostWritesForWire and read by
// ApplyBufferWrite below. So the pin is LIFTED FOR A SPLIT BUILD ONLY, and left
// standing everywhere else - in a monolith build nothing sets the bit and the wire is
// still the thing that would say so if something started to. That is the whole value
// of the pin and it survives the phase it was written for.
void PinNoLiveHostWrites(const MGPipeResourceRecord& record, MGPipeHandle res, const char* call) {
if (!record.HasLiveHostWrites) return;
MGP_TRIP_WIRE_REPORT("MGPipe: " MGP_TRIP_WIRE_TAG("PipeLiveHostWrites")
" %s {slot=%u, gen=%u}: the resource record says host writes are live, and "
"no path in this phase may set that",
call, res.Slot, res.Gen);
}
#else
void PinNoLiveHostWrites(const MGPipeResourceRecord&, MGPipeHandle, const char*) {}
#endif
#if MOBILEGL_PIPE_VERIFY
// P5's pin, RELAXED BY P7 WAVE 3 (CONTRACT-P7 §6) - and relaxed rather than deleted,
// because the thing it was guarding is now reachable instead of hypothetical.
//
// WHAT IT USED TO SAY. "The respecify SCOPE has a wire carrier (MGPResourceDesc's
// HasRespecifiedLevel + the pair) and NO PRODUCER: every descriptor P5 builds is
// whole-resource, and the per-level scope still arrives the old way, as the trailing
// MGPRespecifiedLevel* this function does not look at." So any descriptor arriving with
// the carrier set was, by definition, a producer nobody had announced, and a verify
// build refused it.
//
// WHY THAT IS NOW WRONG, AND EXACTLY HOW FAR. The producer landed:
// Wire_Escape_ResourceRespecify (Client/WireTables.cpp:425-433) writes the carrier from
// the pointer through MGPipeSetRespecifiedLevel, and the server codec
// (Wire/PipeWireCodec.cpp:1677-1683) rebuilds the pointer from the carrier. So under a
// split transport EVERY per-level glTexImage*D arrives here with the carrier set, and
// the old pin made verify x split unmeasurable - it fired on the server's apply thread
// during bring-up, before a single verify case could arm:
//
// Fatal{PipeRespecifyScope} resource_respecify {slot=12, gen=0}: the descriptor
// carries a per-level respecify scope (target=258, level=0), and no path in this
// phase may set one
//
// WHAT SURVIVES, AND IT IS THE WHOLE INVARIANT. The pin never cared that the carrier
// was set; it cared that THE TWO HALVES OF THE SCOPE COULD DISAGREE - "the first thing
// that can go wrong is exactly one of the two moving". That hazard did not go away when
// the producer landed, it became LIVE, and it is silent: the applier below drops pending
// uploads by the POINTER, so a descriptor that declares (target, level) while this call
// hands over a null pointer takes the WHOLE-RESOURCE arm and eats every other level's
// texels, and one that hands over a different pair erases the wrong key and keeps the
// one the client stopped owing. Neither shows up as anything but missing pixels, a
// frame later, on whichever level the wire dropped.
//
// So the rule becomes CONTRACT-P7 §6's: A PRODUCER OF A PER-LEVEL SCOPE MUST COVER THE
// RANGE IT DECLARES. Carrier set => the call passes a pointer, and that pointer names
// exactly the declared (uploadTarget, level).
//
// ONE-DIRECTIONAL ON PURPOSE. A whole-resource descriptor says NOTHING about the
// pointer, and must not: that is the monolith shape and it is still the majority of the
// calls in the tree. MG_Impl/Pipe/TextureEmit.h builds a value-initialized descriptor
// (carrier clear, by MGPipeClearRespecifiedLevel's own "the safe default") and states
// the scope in the trailing pointer alone, which is legal, load-bearing and untouched
// here. Checking that direction too would red every monolith glTexImage*D in the tree
// and would be asserting a convention, not an invariant - the pointer is the caller's
// statement, and a caller that has not been converted to the carrier has not lied.
void PinRespecifyScopeCoversItsDeclaration(const MGPResourceDesc& desc,
const MGPRespecifiedLevel* level, MGPipeHandle res,
const char* call) {
if (MGPipeRespecifyIsWholeResource(desc)) return;
const Uint16 declaredTarget = MGPipeRespecifiedUploadTargetOf(desc);
const Uint16 declaredLevel = MGPipeRespecifiedLevelOf(desc);
if (level != nullptr && level->UploadTarget == declaredTarget &&
level->Level == declaredLevel) {
return;
}
MGP_TRIP_WIRE_REPORT("MGPipe: " MGP_TRIP_WIRE_TAG("PipeRespecifyScope")
" %s {slot=%u, gen=%u}: the descriptor declares a per-level respecify "
"scope (target=%u, level=%u) and this call covers %s(target=%u, "
"level=%u), so a per-level producer is not covering the range it "
"declares",
call, res.Slot, res.Gen, static_cast<unsigned>(declaredTarget),
static_cast<unsigned>(declaredLevel),
level == nullptr ? "NOTHING " : "",
level == nullptr ? 0u : static_cast<unsigned>(level->UploadTarget),
level == nullptr ? 0u : static_cast<unsigned>(level->Level));
}
#else
void PinRespecifyScopeCoversItsDeclaration(const MGPResourceDesc&, const MGPRespecifiedLevel*,
MGPipeHandle, const char*) {}
#endif
#if MOBILEGL_PIPE_VERIFY
// P5 b1's wire, and the one statement about HasLiveHostWrites that IS still always
// true once the field has a producer: the bit is the BUFFER family's. It answers "does
// this resource have a live host writer right now", which only a buffer can have - a
// texture's host writes are the unpack path's and are announced by the upload itself -
// and the two emitters that set it (MGPipeEmitResourceSubData,
// MGPipeEmitBufferSubDataResident) are both buffer-only. A texture sub-data carrying
// it is therefore an encoder that copied a field across payload halves, which is
// precisely the class of defect a shared record type invites.
//
// It replaces the always-false pin in a split build. That pin could not survive the
// producer it existed to announce; this one can, and unlike that one it is compiled in
// EVERY verify build, including build-verify-split - the matrix cell where the field
// can actually be non-zero.
void PinLiveHostWritesNamesABuffer(const MGPSubData& record, const char* call) {
if (record.HasLiveHostWrites == 0) return;
MGP_TRIP_WIRE_REPORT("MGPipe: " MGP_TRIP_WIRE_TAG("PipeLiveHostWritesTarget")
" %s {slot=%u, gen=%u}: a record whose resource target is not a "
"buffer (target=%u) declares live host writes, and only the buffer "
"family has any",
call, record.Res.Slot, record.Res.Gen, record.Target);
}
#else
void PinLiveHostWritesNamesABuffer(const MGPSubData&, const char*) {}
#endif
// The one gate every content-carrying buffer write goes through. resource_subdata and
// buffer_subdata_resident differ only in which backend hook takes the bytes and in the
// fact that one of them is allowed to be absent, so a second copy of this arithmetic
// would be a second place to get it wrong.
// RETURNS THE ACCEPTANCE, for D-D5 step 1's reason: the emitter clears its own dirty
// state on the strength of this answer and neither refusal path is otherwise visible
// from the call site (a dead handle is a counted no-op, a corrupt record is a Fatal
// that deliberately moves no counter).
Bool ApplyBufferWrite(const char* call, const MGPSubData& record, const void* bytes, Bool resident) {
MGPipeResourceRecord* stored = ResolveResource(call, record.Res);
if (stored == nullptr) return false;
const Uint64 offset = MGPipeSubDataBufferOffset(record);
const Uint64 size = MGPipeSubDataBufferSize(record);
const char* fault = SubDataBoxFault(record);
if (fault == nullptr) fault = BufferRangeFault(offset, size, stored->Desc.Width);
if (fault == nullptr && size != 0 && bytes == nullptr) {
fault = "a non-empty write carries no bytes";
}
if (fault != nullptr) {
MGP_TRIP_WIRE_REPORT("MGPipe: " MGP_TRIP_WIRE_TAG("ProtocolCorruption")
" %s {slot=%u, gen=%u, glName=%u}: %s (offset=%llu, size=%llu, "
"storage=%u bytes)",
call, record.Res.Slot, record.Res.Gen, stored->Desc.GlNameForDiag,
fault, static_cast<unsigned long long>(offset),
static_cast<unsigned long long>(size), stored->Desc.Width);
return false;
}
PinNoLiveHostWrites(*stored, record.Res, call);
#if MOBILEGL_BUILD_DISAGGREGATED
// P5 (b1): THE PRODUCER. The record's own statement about the resource, applied
// before the serial moves so that a probe re-entered from inside the backend hook
// below already sees it. It is an assignment and not an OR: the bit is a STATE,
// and a content record emitted while nothing maps the buffer is exactly how the
// state goes back to false - which is why the falling edge pushes one block
// (BufferObject::NotePersistentMapStateChanged) rather than relying on the next
// ordinary write to arrive.
stored->HasLiveHostWrites = record.HasLiveHostWrites != 0;
#endif
// THE SERIAL MOVES BEFORE THE BACKEND IS TOLD, and that order is load-bearing:
// the backend stamps its own synced serial from this record inside the hook, so a
// bump afterwards would leave the twin stamped one mutation behind and the next
// draw would re-upload what it had just landed.
++stored->Serial;
// THE RANGE IS LANDED FROM HERE ON, so every path below returns true: whether a
// backend table is registered, and whether it implements the optional resident
// hook, is a property of the BUILD and not of the record. An emitter that read
// "not accepted" off an unregistered table would keep re-sending a write the
// applier has already taken responsibility for.
if (g_resourceOps == nullptr) return true;
if (resident) {
// kOptional, and the frontend already checks the same way for the table this
// one replaces: a backend that does not implement the resident path leaves the
// member null and the write is landed by its ordinary sub-data route instead.
if (g_resourceOps->SubDataResident != nullptr) {
g_resourceOps->SubDataResident(record.Res, record, bytes);
}
return true;
}
if (g_resourceOps->SubData != nullptr) g_resourceOps->SubData(record.Res, record, bytes);
return true;
}
#if MOBILEGL_BUILD_DISAGGREGATED
// Texture records can have a server consumer without the buffer ops table (Magma).
// Their staged bytes still belong to the server before SEG_STAGE retires. Keep the
// existing backend hooks authoritative when present, and use this fallback otherwise.
void AdoptTextureWithoutBackendHook(const MGPipeResourceRecord& stored, const MGPSubData& upload,
const void* bytes, const MGPSubRegion* regions) {
if (MG_Config::Transport == MG_Config::TransportMode::Monolith || !bytes || upload.Blob.Size == 0) return;
auto& store = MG_Remote::Server::ServerStagedTexture();
const auto& desc = stored.Desc;
const Uint64 key = MG_Remote::Server::StagedTextureStore::KeyForHandle(upload.Res);
const Uint16 target = MGPipeSubDataUploadTargetOf(upload.Target);
const IntVec3 extent = MG_Remote::Server::StagedTextureUploadExtent(desc, upload);
// RegionCount == 0 IS the whole-level spelling - "the run is the level shadow" - so it
// is adopted with the spelling that replaces the level; every other record names a RUN
// of the level by its own box (fix A2: a level too large to stage whole crosses as
// slabs, one record each, and the first of them is at the level's first byte).
if (upload.RegionCount == 0) {
store.Adopt(key, target, upload.Level, extent, bytes,
static_cast<SizeT>(upload.Blob.Size));
return;
}
store.AdoptRun(key, target, upload.Level, extent,
MG_Remote::Server::StagedTextureRunImageOffset(upload, regions), bytes,
static_cast<SizeT>(upload.Blob.Size));
}
void DefineTextureWithoutBackendHook(const MGPResourceDesc& desc, const MGPRespecifiedLevel* level) {
if (MG_Config::Transport == MG_Config::TransportMode::Monolith) return;
auto& store = MG_Remote::Server::ServerStagedTexture();
const Uint64 key = MG_Remote::Server::StagedTextureStore::KeyForHandle(desc.Resource);
const auto define = [&](Uint16 target, Uint16 mip, const IntVec3& extent) {
store.NoteLevelDefined(key, target, mip, extent, desc.Target, desc.InternalFormat);
};
if (level) {
define(MGPipeSubDataUploadTargetOf(level->UploadTarget), level->Level,
IntVec3{static_cast<Int>(level->Width), static_cast<Int>(level->Height),
static_cast<Int>(level->Depth)});
return;
}
store.ResetLevels(key);
if (!desc.Immutable || !desc.Levels) return;
const auto defineChain = [&](TextureUploadTarget target) {
for (Uint32 mip = 0; mip < desc.Levels; ++mip)
define(static_cast<Uint16>(target), static_cast<Uint16>(mip),
MG_Remote::Server::StagedTextureMipExtent(desc.Target, desc.Width,
desc.Height, desc.Depth, mip));
};
switch (static_cast<MGPipeResourceTarget>(desc.Target)) {
case MGPipeResourceTarget::Tex1D: defineChain(TextureUploadTarget::Texture1D); break;
case MGPipeResourceTarget::Tex2D: defineChain(TextureUploadTarget::Texture2D); break;
case MGPipeResourceTarget::Tex3D: defineChain(TextureUploadTarget::Texture3D); break;
case MGPipeResourceTarget::Tex1DArray: defineChain(TextureUploadTarget::Texture1DArray); break;
case MGPipeResourceTarget::Tex2DArray: defineChain(TextureUploadTarget::Texture2DArray); break;
case MGPipeResourceTarget::TexCube:
for (Uint32 face = static_cast<Uint32>(TextureUploadTarget::CubeMapPositiveX);
face <= static_cast<Uint32>(TextureUploadTarget::CubeMapNegativeZ); ++face)
defineChain(static_cast<TextureUploadTarget>(face));
break;
case MGPipeResourceTarget::TexCubeArray: defineChain(TextureUploadTarget::CubeMapArray); break;
case MGPipeResourceTarget::Tex2DMS: defineChain(TextureUploadTarget::Texture2DMultisample); break;
case MGPipeResourceTarget::Tex2DMSArray: defineChain(TextureUploadTarget::Texture2DMultisampleArray); break;
case MGPipeResourceTarget::TexRect: defineChain(TextureUploadTarget::TextureRectangle); break;
case MGPipeResourceTarget::TexBuffer: defineChain(TextureUploadTarget::TextureBuffer); break;
default: break;
}
}
#endif
// The texture half of resource_subdata, and it DISPATCHES TO NOBODY at GL-call time:
// a texture write marks a level dirty and Espryt uploads it at its own sync point, out
// of the accumulated set below. So the whole of this function is the gate, the
// accumulation, the serial and - P5c (tx), disaggregated builds only - the adoption
// hook that moves the staged bytes into the server's staged-texture store while they
// are still alive. MGPipeResourceOps grew its three texture members for exactly that
// hook; the monolith shape of everything above them is unchanged.
Bool ApplyTextureUpload(const MGPSubData& record, const void* bytes, const MGPSubRegion* regions) {
MGPipeResourceRecord* stored =
ResolveResourceIn(MGPipeApplier().TextureResources, "resource_subdata", record.Res);
if (stored == nullptr) return false;
const char* fault = SubDataTextureFault(record, regions);
// A whole-level upload declares no regions and a non-empty box; a record that
// declares neither has nothing to upload and nothing to accumulate, which is a
// shape the drain list cannot produce.
if (fault == nullptr && BoxIsEmpty(record.UnionBox) && record.RegionCount == 0) {
fault = "the record describes no texels at all";
}
if (fault == nullptr && bytes == nullptr) {
fault = "a texture upload carries no bytes";
}
if (fault != nullptr) {
MGP_TRIP_WIRE_REPORT("MGPipe: " MGP_TRIP_WIRE_TAG("ProtocolCorruption")
" resource_subdata {slot=%u, gen=%u, glName=%u}: %s (target=%u, "
"level=%u, box=%d,%d,%d %ux%ux%u, regions=%u)",
record.Res.Slot, record.Res.Gen, stored->Desc.GlNameForDiag, fault,
record.Target, record.Level, record.UnionBox.X, record.UnionBox.Y,
record.UnionBox.Z, record.UnionBox.W, record.UnionBox.H,
record.UnionBox.D, record.RegionCount);
return false;
}
if (!AccumulatePendingUpload(*stored, record, regions)) {
MGP_TRIP_WIRE_REPORT("MGPipe: " MGP_TRIP_WIRE_TAG("ProtocolCorruption")
" resource_subdata {slot=%u, gen=%u, glName=%u}: the resource names "
"more distinct (upload target, level) pairs than any texture can "
"have (%u)",
record.Res.Slot, record.Res.Gen, stored->Desc.GlNameForDiag,
kMGPipeMaxPendingUploads);
return false;
}
// The serial moves for the buffer half's reason: the twin stamps its own synced
// serial from inside the sync that reads this record, so a bump afterwards would
// leave it one mutation behind and the next draw would re-upload what it had just
// landed. The ACCEPTANCE is what the client reads to clear its own dirty flag - the
// record was accumulated, so the texels are the server's now, and that is the true
// this returns.
++stored->Serial;
#if MOBILEGL_BUILD_DISAGGREGATED
// P5c (tx): THE STAGED BYTES ARE ADOPTED HERE, at the last instant `bytes` is known
// alive (rule C: SEG_STAGE retires when this record does, and PendingUpload
// deliberately holds no byte pointer). The hook copies the run into the server's
// staged-texture store keyed by this record's own handle; it is a no-op in
// monolith, so the monolith shape keeps P5's pointer-dropping expression exactly.
// The REGIONS travel with it because a record may carry a RUN of its level rather
// than the whole of it (fix A2): they are what places that run in the level image,
// and the applier's pending set - which already has them - is not the store's.
if (g_resourceOps != nullptr && g_resourceOps->TextureSubData != nullptr) {
g_resourceOps->TextureSubData(record.Res, record, bytes, regions);
} else {
AdoptTextureWithoutBackendHook(*stored, record, bytes, regions);
}
#endif
return true;
}
// ----------------------------------------------------------------------------
// P4a: the three kVarTail unit sets share one body.
//
// They differ in exactly one thing - what an entry IS - and in nothing else: the same
// window rule, the same bound, the same "the entries outside the window are not
// cleared", the same serial discipline. set_vertex_buffers wrote this arithmetic once
// already; a second, third and fourth copy of it would be three more places to get the
// Start + Count overflow wrong.
//
// THE WINDOW IS THE BOUND AND ENTRIES OUTSIDE IT ARE NOT CLEARED. The record is "the
// last set as received": a set that names four units has said nothing about the other
// 188, and clearing them would unbind textures the client never mentioned.
//
// THE ENTRY'S OWN Unit FIELD IS NOT POLICED, deliberately and for MGPVertexBuffer::
// BindingIndex's reason: the DESTINATION is Start + i, which is the window this
// function has already bounded, and the field beside it is the client's own label for
// the entry. Policing it would hand the emitter a contract this applier cannot justify
// - the two agree by construction or the emitter is broken in a way a unit test on the
// emitter's side is the right place to catch.
template <class Entry, class ArrayT>
Bool ApplyUnitWindow(const char* call, Uint32 start, Uint32 count, Uint64 contentHash,
const Entry* tail, ArrayT& destination, Uint32& destinationStart,
Uint32& destinationCount) {
const Uint64 capacity = destination.size();
const Uint64 end = Uint64{start} + Uint64{count};
const char* fault = nullptr;
if (end > capacity) {
fault = "the window runs past the merged texture-unit space";
} else if (count != 0 && tail == nullptr) {
fault = "a non-empty set carries no entries";
}
if (fault != nullptr) {
MGP_TRIP_WIRE_REPORT("MGPipe: " MGP_TRIP_WIRE_TAG("ProtocolCorruption")
" %s {start=%u, count=%u, hash=%llu}: %s (the applier holds %llu "
"entries)",
call, start, count, static_cast<unsigned long long>(contentHash),
fault, static_cast<unsigned long long>(capacity));
return false;
}
for (Uint32 i = 0; i < count; ++i) {
destination[start + i] = tail[i];
}
destinationStart = start;
destinationCount = count;
return true;
}
// ----------------------------------------------------------------------------
// P4a: the ShaderCso band split, mirrored from the allocator's.
//
// The composite band starts at 983040, so ONE program-pipeline composite in a
// slot-indexed vector would grow that vector to ~983k records of ~240 bytes each - a
// 236 MB spike on the first pipeline draw, which would be a defect this commit
// introduced rather than one it inherited. Both spaces are kept dense against their OWN
// high-water mark instead, exactly as MGPipeSlotAllocator keeps the band in a table of
// its own.
//
// THE SERVER STILL NEVER LEARNS A HANDLE IS A COMPOSITE. The split is an indexing detail
// on this side of the wire: create / bind / delete_shader_state and set_global_constants
// name a composite exactly as they name any other program, and none of them branches on
// it outside these two functions.
// ----------------------------------------------------------------------------
MGPipeShaderCsoRecord* ShaderCsoRecordAt(Uint32 slot) {
if (MGPipeIsCompositeShaderSlot(slot)) {
return RecordAt(MGPipeApplier().CompositeShaderCsos, slot - kMGPipeShaderCsoCompositeSlotBase,
kMGPipeShaderCsoSlotLimit - kMGPipeShaderCsoCompositeSlotBase);
}
// The ordinary table is bounded by the BAND'S BASE and not by the slot limit: an
// ordinary program can never be handed a band slot (the allocator refuses it), so a
// slot at or above the base that is not a composite is out of range by definition.
return RecordAt(MGPipeApplier().ShaderCsos, slot, kMGPipeShaderCsoCompositeSlotBase);
}
MGPipeShaderCsoRecord* FindShaderCso(MGPipeHandle cso) {
if (MGPipeIsCompositeShaderSlot(cso.Slot)) {
return FindIn(MGPipeApplier().CompositeShaderCsos,
MGPipeHandle{cso.Slot - kMGPipeShaderCsoCompositeSlotBase, cso.Gen});
}
return FindIn(MGPipeApplier().ShaderCsos, cso);
}
MGPipeShaderCsoRecord* ResolveShaderCso(const char* call, MGPipeHandle cso) {
MGPipeShaderCsoRecord* record = FindShaderCso(cso);
MOBILEGL_ASSERT(record != nullptr, "%s named {slot=%u, gen=%u}: %s", call, cso.Slot, cso.Gen,
kResourceRefusalNote);
if (record == nullptr) ++MGPipeApplier().RefusedObjectCalls;
return record;
}
#if MOBILEGL_PIPE_VERIFY
// THE ARCHIVE'S ROUND TRIP, and it is a VERIFY-ONLY gate on a path that does not
// serialise anything in a shipped build. What it asks is the question a transport will
// ask on the first day it exists and nobody can ask afterwards: does the archive this
// client is publishing survive being turned into bytes and back? A field the codec
// skips in BOTH directions round-trips perfectly, which is why the unit suite reads the
// members back explicitly; what THIS adds is that it runs over every real program the
// verify lane links rather than over one hand-built instance.
void PinProgramArchiveRoundTrip(const MGPProgramDesc& desc,
const MG_State::GLState::LinkArtifacts& link,
const MG_State::GLState::SpirvArtifacts& spirv) {
Vector<Uint8> encoded;
MG_State::GLState::EncodeProgramArtifacts(link, spirv, encoded);
MG_State::GLState::LinkArtifacts decodedLink;
MG_State::GLState::SpirvArtifacts decodedSpirv;
Vector<Uint8> reencoded;
const char* fault = nullptr;
if (encoded.empty()) {
fault = "the encoder produced no bytes for a program that has artefacts";
} else if (!MG_State::GLState::DecodeProgramArtifacts(encoded.data(), encoded.size(),
decodedLink, decodedSpirv)) {
fault = "the archive this client would put on the wire does not decode";
} else {
MG_State::GLState::EncodeProgramArtifacts(decodedLink, decodedSpirv, reencoded);
if (reencoded != encoded) {
fault = "the archive does not survive its own round trip";
} else if (decodedLink.program != nullptr) {
// The live glslang TProgram is the one member the tables deliberately omit;
// a decode that reconstructed one would be carrying the compiler front end
// across a boundary that exists to keep it on this side.
fault = "the decoded archive carries a live program object";
}
}
if (fault == nullptr) return;
MGP_TRIP_WIRE_REPORT("MGPipe: " MGP_TRIP_WIRE_TAG("PipeVerifyDiffer")
" program-archive create_shader_state {slot=%u, gen=%u}: %s (%llu bytes "
"encoded, %llu re-encoded)",
desc.Cso.Slot, desc.Cso.Gen, fault,
static_cast<unsigned long long>(encoded.size()),
static_cast<unsigned long long>(reencoded.size()));
}
#endif
} // namespace
// MGPipeApplier() is defined with the registry at the top of this file: it is the one
// function that has to know whose applier the calling thread is in.
// P5c (rv, CONTRACT-P5C.md §5.3): the three texture shutters' server-side answers. See
// MGPipeApplierState::TextureShutterSerial / ContextSerial for the bump rule.
Uint64 MGPipeApplierTextureShutterSerial() { return MGPipeApplier().TextureShutterSerial; }
Uint64 MGPipeApplierContextSerial() { return MGPipeApplier().ContextSerial; }
void MGPipeApplierNoteTextureStateMoved() { ++MGPipeApplier().TextureShutterSerial; }
void MGPipeSetResourceOps(const MGPipeResourceOps* ops) { g_resourceOps = ops; }
const MGPipeResourceOps* MGPipeGetResourceOps() { return g_resourceOps; }
namespace {
// P4a's BELT (ID-39). See MGPipeApplierState::RefusedNoConsumer for the whole argument;
// in one line: acceptance is a contract with the client, and accepting a record on a
// backend that consumes none of them makes the emitter clear a dirty flag the legacy
// pull path still owed.
//
// IT IS THE SAME SIGNAL THE CLIENT'S GATE READS, deliberately - g_resourceOps is the
// table a backend installs at its own bring-up and uninstalls at teardown, and
// MGPipeApplierReset does NOT touch it (see there), so the gate cannot flap between a
// make-current and the emissions that follow it. The two therefore agree on every path
// except the one this exists for: an emitter called directly, without passing PipeFill.
//
// EVERY CALLER PLACES IT AFTER THE RECORD'S OWN SHAPE CHECKS AND BEFORE ANYTHING MOVES.
// A malformed record is Fatal{ProtocolCorruption} whether or not anything would have
// read it - a trip wire that fires only on some backends is a trip wire nobody can
// trust - so the refusal is the LAST thing asked and the first thing that stops the
// write. The three unit-set entry points are the exception and say so at their call
// site: ApplyUnitWindow validates and writes in one step, so the question has to be
// asked in front of it.
Bool NoP4aConsumer() {
#if MOBILEGL_BUILD_DISAGGREGATED
if (MG_Config::Transport != MG_Config::TransportMode::Monolith) {
const auto* session = MG_Remote::Server::ServerSession::Active();
if (session && session->CallMaskIsSet() && MG_Remote::MGCapsServerConsumes(
session->CallMask(), kMGPipeSubsystemTextureResources)) return false;
}
#endif
if (g_resourceOps != nullptr) return false;
++MGPipeApplier().RefusedNoConsumer;
return true;
}
// P5e (sb, CONTRACT-P5E.md §5.6): the three binding-point windows, cleared in one
// place because BOTH resets clear them and the two used to drift on exactly this kind
// of member (the framebuffer family's four). The SERIAL is deliberately NOT touched
// here - each caller advances it beside its own siblings, where the "advance, never
// zero" rule is written out.
void ClearShaderBufferWindows() {
for (Uint32 cls = 0; cls < kMGPipeShaderBufferClassCount; ++cls) {
MGPipeApplier().BoundShaderBuffers[cls] = {};
MGPipeApplier().ShaderBufferStart[cls] = 0;
MGPipeApplier().ShaderBufferCount[cls] = 0;
for (Uint32 w = 0; w < kMGPipeShaderBufferWritableMaskWords; ++w) {
MGPipeApplier().ShaderBufferWritableMask[cls][w] = 0;
}
}
}
} // namespace
void MGPipeApplierReset() {
#if MOBILEGL_BUILD_DISAGGREGATED
// P5c (ct), CONTRACT-P5C.md §5.1 / §6 layer 2. With an active transport g_applier is
// SERVER-PRIVATE and the GL thread's reset crosses as the applier_reset RECORD
// (PipeFill.cpp's FreshlyPrimed arm emits it; ServerVerbSink::OnApplierReset runs this
// function ON the apply thread). Reaching here from any other thread WITH A LIVE WIRE
// is the direct call the record replaced - a write into server memory with no wire
// shape, and exactly the revert the red-once gate must catch - so it is a named Fatal
// rather than a silent reset of state another role owns. The window the Fatal
// deliberately does NOT cover is a configured-but-wireless transport: before
// ClientSession::Start() there is no wire for a record to cross (§6 layer 2's
// documented bring-up exception), and a ServerLoop fixture with no client at all has
// none either - there this call is the only reset that exists. Monolith keeps the
// direct call, byte for byte (G1); in a pull build none of this is compiled at all.
// D10: THE MIDDLE CONJUNCT WAS ClientSession::Active() != nullptr ALONE, WHICH IS
// PERMANENTLY NULL IN A SERVER PROCESS - so this guard could not arm at all under
// spawn, in the shape where calling the reset off the apply thread is most likely and
// least recoverable. MGPipeSessionLive() is the same question asked in a way both roles
// can answer, and MGPipeServerArm() is the process-scoped half of "am I the server".
//
// BOTH ARE ADDED AS DISJUNCTS, not substituted. The originals are the only facts a
// fixture without a full session has, and swapping them out silently disarms every such
// test - measured.
if (MG_Config::Transport != MG_Config::TransportMode::Monolith &&
(MG_Remote::Client::ClientSession::Active() != nullptr || MG_Pipe::MGPipeSessionLive()) &&
!MG_Remote::Server::ServerLoop::OnApplyThread() && !MG_Pipe::MGPipeServerArm()) {
MGLOG_F("MGPipe: Fatal{RoleViolation, \"g_applier\"} - MGPipeApplierReset() called "
"off the apply thread with an active transport; under split the reset "
"crosses as the applier_reset record (CONTRACT-P5C.md §5.1)");
std::abort();
}
#endif
MGPipeApplier().RenderStateCsos.clear();
MGPipeApplier().BoundRenderStateCso = kMGPipeNullHandle;
MGPipeApplier().Residual = ResidualValueBlock{};
MGPipeApplier().HasResidual = false;
MGPipeApplier().ScatteredChunkBits = 0;
MGPipeApplier().ResidualCapabilitiesCompared = 0;
MGPipeApplier().ResidualDivergences = 0;
MGPipeApplier().PatchCarrierComparisons = 0;
MGPipeApplier().PatchCarrierDivergences = 0;
// P3a. THIS RUNS AT EVERY CHANGE OF THE CURRENT CONTEXT, not once per fresh one:
// MGPipeTracker::Update resets itself whenever the context pointer moves and the
// emitter calls this from the walk that follows, so a make-current BACK to a context
// that is still alive lands here too. Everything cleared below is therefore something
// a returning context may not inherit, and nothing else is cleared.
//
// THE OBJECT RECORDS ARE NOT CLEARED. A GL object lives in a share group, not in a
// context: the buffer a returning context is about to write to is the same buffer with
// the same storage, and its record is where the extent and the mutation serial that
// D-A4 re-keys IsBufferDrawClean onto now live. Dropping them here made every
// glBufferSubData after a context switch resolve to nothing and be dropped, with the
// only trace an assertion that compiles out at INFO. They go at the object's own death
// (resource_destroy, delete_vertex_elements) and at MGPipeApplierReleaseObjectRecords.
//
// The OP TABLE is deliberately not cleared either - it is installed and uninstalled by
// the backend's own bring-up and teardown, not by a state reset.
MGPipeApplier().RefusedResourceCalls = 0;
MGPipeApplier().RefusedVertexInputCalls = 0;
MGPipeApplier().RefusedObjectCalls = 0;
MGPipeApplier().RefusedNoConsumer = 0;
MGPipeApplier().BoundVertexElements = kMGPipeNullHandle;
MGPipeApplier().VertexBuffers = {};
MGPipeApplier().VertexBufferStart = 0;
MGPipeApplier().VertexBufferCount = 0;
MGPipeApplier().VertexFetchBaseInstance = 0;
MGPipeApplier().IndexBuffer = MGPIndexBuffer{};
// m6 / wire n6, written down rather than left to be rediscovered: THIS counter is
// per-applier and is zeroed at every make-current, while MG_Util::PipeStats' `mpr` -
// emitted from the CLIENT at MG_Impl/Pipe/PipeFill.cpp's MGPipeEmitMapPersistent - is
// process-wide and is windowed by EndFrame. The two therefore disagree across a context
// switch, by design and not by accident: this one answers "how many round trips has THIS
// applier been asked for since it was last reset", which is what a unit case driving the
// applier directly wants, and PipeStats' answers "how many did the process take in this
// window", which is what a lane reading a log line wants.
//
// WHICH ONE THE GATES ASSERT ON, because that was the open question: G10
// (StorageBufferRegrow) and G12 read PipeStats' `mpr` out of the lane's own log through
// MG_IntegrationTest/Harness/PipeStatsWindow.h - they cannot link this symbol at all, on
// Android or anywhere else - so a make-current inside a scenario cannot silently reset
// what they measure. Nothing outside MG_Test reads the member below.
MGPipeApplier().MapPersistentRoundtrips = 0;
// THE TWO SERIALS ADVANCE; THEY ARE NOT ZEROED. They are MGGens, and an MGGen that
// walks backwards is not one. There are exactly three things a reset can do to a
// version whose data it has just cleared:
// - carry the count over: the twin's memo still matches state that is now empty, so
// the very next draw reads clean over a cleared window. Wrong immediately;
// - restart at 0: the counter then walks back up through every value it has already
// stamped into a twin, and a VAO twin does NOT die with a make-current
// (OnBackendContextDestroyed runs on destroy) and has no context generation beside
// the serial - D-G4 deletes the identity patch that used to close exactly this
// hole. Wrong later, and reliably, because a context whose per-activation call
// count is stable lands on a stamped value every time;
// - advance: the clearing is itself announced, no stamped value can ever recur, and
// the first compare after the switch is a mismatch, which is the safe direction.
++MGPipeApplier().VertexBuffersSerial;
++MGPipeApplier().IndexBufferSerial;
// P5c (rv): a make-current is a fresh server, and the two shutter serials are what the
// PipeInputs texture-shutter accessors answer with - so both ADVANCE here, for the
// serials' own reason: a counter that restarts walks back through values already
// stamped into a memo that outlived the switch. ContextSerial's move is also
// GetTextureContextId's whole job (the backends key their per-context memos on it).
++MGPipeApplier().TextureShutterSerial;
++MGPipeApplier().ContextSerial;
// ---- P4a's working state, cleared for the same reason and with the same serial rule
// (D-J4). The OBJECT records - texture and renderbuffer resources, sampler CSOs,
// sampler views, shader CSOs and, since ID-19(b), the FRAMEBUFFER RECORDS - are
// deliberately NOT here: a texture lives in a share group exactly as a buffer does, and
// its record is where the extent, the parameters and the pending-upload set the backend
// reads now live.
//
// WHAT A MAKE-CURRENT TAKES OF THE FRAMEBUFFER FAMILY IS THE TWO BOUND HANDLES AND
// NOTHING ELSE, and that is the same observable it always took: DrawFramebuffer() /
// ReadFramebuffer() answer null after this, exactly as the zeroed records used to answer
// a null Fbo. FramebufferRecords stays because a framebuffer that is only ever addressed
// BY NAME - the DSA blit and clear entry points - has no re-emission trigger at all, so
// dropping its record here would lose it for the life of the process.
MGPipeApplier().BoundFramebuffer = {};
MGPipeApplier().BoundSamplerViews = {};
MGPipeApplier().SamplerViewStart = 0;
MGPipeApplier().SamplerViewCount = 0;
MGPipeApplier().BoundSamplerStates = {};
MGPipeApplier().SamplerStateStart = 0;
MGPipeApplier().SamplerStateCount = 0;
MGPipeApplier().BoundShaderImages = {};
MGPipeApplier().ShaderImageStart = 0;
MGPipeApplier().ShaderImageCount = 0;
MGPipeApplier().DrawProgram = kMGPipeNullHandle;
MGPipeApplier().DispatchProgram = kMGPipeNullHandle;
MGPipeApplier().BoundShaderCso = kMGPipeNullHandle;
#if MOBILEGL_BUILD_DISAGGREGATED
MGPipeApplier().BoundStreamOutputLifetimeId = 0;
// Begin is emitted once per span, not on make-current. Its immutable object
// snapshot survives with the resource/program records until End or release;
// set_context_values restores the returning context's bound lifetime id.
#endif
// P5e (sb, CONTRACT-P5E.md §5.6): the three binding-point windows are per-context
// WORKING state and go with the rest of it - a returning context has its own
// glBindBufferBase history and may not inherit the one this applier was left holding.
// The SERIAL advances rather than restarting, for the reason written out above: the
// clearing is itself a change every backend memo has to hear about, and a counter that
// restarts walks back through values already stamped into a memo that outlived the
// switch. The EMITTER's latch resets with it (PipeFill.cpp's FreshlyPrimed arm
// invalidates the three suppressor slots), or the first emission after a make-current
// would be suppressed as unchanged and the server would draw with a cleared window.
ClearShaderBufferWindows();
++MGPipeApplier().FramebufferSerial;
++MGPipeApplier().SamplerViewsSerial;
++MGPipeApplier().SamplerStatesSerial;
++MGPipeApplier().ShaderImagesSerial;
++MGPipeApplier().ProgramBindingSerial;
++MGPipeApplier().ShaderBuffersSerial;
}
void MGPipeApplierReleaseObjectRecords() {
#if MOBILEGL_BUILD_DISAGGREGATED
if (MG_Config::Transport != MG_Config::TransportMode::Monolith &&
(g_resourceOps == nullptr || g_resourceOps->TextureDestroy == nullptr)) {
MG_Remote::Server::ServerStagedTexture().DropAll();
}
#endif
// The served context is going away and this applier with it. Under split that is one
// applier per served context; in the monolith there is one applier behind every
// context, so nothing wires this - see PipeApply.h. The two serials advance here for
// MGPipeApplierReset's reason: state was cleared, and a twin that outlives it must not
// be able to match a value it has already seen.
MGPipeApplier().Resources.clear();
MGPipeApplier().VertexElementsCsos.clear();
MGPipeApplier().BoundVertexElements = kMGPipeNullHandle;
++MGPipeApplier().VertexBuffersSerial;
++MGPipeApplier().IndexBufferSerial;
// P5c (rv): same rule for the shutter serials - the served context is going away, and
// a memo that outlives it must not match a value these have already answered with.
++MGPipeApplier().TextureShutterSerial;
++MGPipeApplier().ContextSerial;
// P4a's five object tables go with them, and the working handles they could name go
// too - a bound shader CSO whose record has just been dropped must not survive as a
// handle the next call resolves against.
MGPipeApplier().TextureResources.clear();
MGPipeApplier().RenderbufferResources.clear();
MGPipeApplier().SamplerCsos.clear();
MGPipeApplier().SamplerViewCsos.clear();
MGPipeApplier().ShaderCsos.clear();
MGPipeApplier().CompositeShaderCsos.clear();
// AND THE FRAMEBUFFER RECORDS, which MGPipeApplierReset above deliberately keeps. Here
// there is nothing for them to outlive: the served context is going away and the
// framebuffers it owned are container objects that go with it. A table left standing
// would answer a lookup from the NEXT context on a slot the allocator has re-minted -
// with the same generation, if the next context happens to reach the same one - and
// that is the aliasing every other table is cleared to prevent.
MGPipeApplier().FramebufferRecords.clear();
MGPipeApplier().DrawProgram = kMGPipeNullHandle;
MGPipeApplier().DispatchProgram = kMGPipeNullHandle;
MGPipeApplier().BoundShaderCso = kMGPipeNullHandle;
#if MOBILEGL_BUILD_DISAGGREGATED
MGPipeApplier().BoundStreamOutputLifetimeId = 0;
MGPipeApplier().StreamOutputSpans.clear();
#endif
// AND "THE WORKING HANDLES THEY COULD NAME" IS ALL OF THEM, NOT JUST THE THREE ABOVE.
// Every framebuffer record holds eleven MGPSurface::Res naming texture and renderbuffer
// records this function has just dropped - which is why the table itself goes above -
// and the three unit windows hold entries naming sampler-view, sampler-CSO and texture
// records it has just dropped. A window left populated here is a set of handles into
// empty tables: the next resolve either refuses and counts (a teardown-time refusal
// storm nothing asked for) or, on a slot the next context re-mints, resolves to somebody
// else's record. So this takes the same working state MGPipeApplierReset does - which is
// the honest reading of the rule as well: Reset keeps the object records BECAUSE they
// outlive a make-current, and here they do not outlive anything.
MGPipeApplier().BoundFramebuffer = {};
MGPipeApplier().BoundSamplerViews = {};
MGPipeApplier().SamplerViewStart = 0;
MGPipeApplier().SamplerViewCount = 0;
MGPipeApplier().BoundSamplerStates = {};
MGPipeApplier().SamplerStateStart = 0;
MGPipeApplier().SamplerStateCount = 0;
MGPipeApplier().BoundShaderImages = {};
MGPipeApplier().ShaderImageStart = 0;
MGPipeApplier().ShaderImageCount = 0;
// P5e (sb): and the three binding-point windows, for the paragraph above's reason in
// its own words - every MGPBufferRange::Res names a resource record this function has
// just dropped, so a window left populated here is a set of handles into an empty
// table, which the next resolve either refuses or answers with somebody else's record
// on a slot the next context re-mints.
ClearShaderBufferWindows();
++MGPipeApplier().FramebufferSerial;
++MGPipeApplier().SamplerViewsSerial;
++MGPipeApplier().SamplerStatesSerial;
++MGPipeApplier().ShaderImagesSerial;
++MGPipeApplier().ProgramBindingSerial;
++MGPipeApplier().ShaderBuffersSerial;
}
void MGPipeApplyCreateRenderState(const MGPRenderStateDesc& desc, const void* chunkBytes) {
MOBILEGL_ASSERT(desc.Cso.Slot >= kMGPipeFirstAllocatableSlot,
"create_render_state named the reserved slot 0");
MGPipeRenderStateCsoRecord* recordAt =
RecordAt(MGPipeApplier().RenderStateCsos, desc.Cso.Slot, kMGPipeMaxRenderStateCsoSlots);
if (recordAt == nullptr) {
#if MOBILEGL_BUILD_DISAGGREGATED
MGPipeSessionFail(MGPipeFatalFamily::ProtocolCorruption,
"MGPipe: Fatal{ProtocolCorruption, \"CreateRenderState.Cso.Slot\"} - "
"create_render_state {slot=%u, gen=%u} is outside the record table bound (%u)",
desc.Cso.Slot, desc.Cso.Gen, kMGPipeMaxRenderStateCsoSlots);
#else
MGP_TRIP_WIRE_REPORT("MGPipe: " MGP_TRIP_WIRE_TAG("ProtocolCorruption")
" create_render_state {slot=%u, gen=%u} is outside the record table bound (%u)",
desc.Cso.Slot, desc.Cso.Gen, kMGPipeMaxRenderStateCsoSlots);
#endif
return;
}
MGPipeRenderStateCsoRecord& record = *recordAt;
// NEITHER BRANCH MAY LEAVE ITS BAD CASE TO MOBILEGL_ASSERT. The slot the client is
// naming may be a RECYCLED one whose record still holds the previous occupant's 396
// bytes; in an INFO build - which is what every gate and every shipped build is - an
// assertion is a no-op, so inheriting nothing onto those bytes and then scattering
// the delta chunks on top would hand out a record that is half one CSO and half
// another, with no gate able to see it. Both arms therefore start from a DEFINED
// base and report through this file's trip-wire verdict.
if (MGPipeHandleIsNull(desc.BaseCso)) {
// A brand-new CSO carries its whole content; there is no earlier record to
// inherit the unnamed chunks from.
record.PipelineBytes = {};
if ((desc.ChunkMask & kAllPipelineChunks) != kAllPipelineChunks) {
MGP_TRIP_WIRE_REPORT("MGPipe: " MGP_TRIP_WIRE_TAG("PipeIncompleteCso")
" create_render_state {slot=%u, gen=%u} with no BaseCso named chunks "
"0x%x, not the whole pipeline half 0x%x; the rest is zeroed",
desc.Cso.Slot, desc.Cso.Gen, desc.ChunkMask,
static_cast<Uint32>(kAllPipelineChunks));
}
} else {
const MGPipeRenderStateCsoRecord* base = FindCso(desc.BaseCso);
record.PipelineBytes =
base != nullptr ? base->PipelineBytes : Array<Uint8, kMGPipePipelineChunkBytes>{};
if (base == nullptr) {
MGP_TRIP_WIRE_REPORT("MGPipe: " MGP_TRIP_WIRE_TAG("PipeDeadBaseCso")
" create_render_state {slot=%u, gen=%u} named a dead BaseCso "
"{slot=%u, gen=%u}; the delta chunks land on zeroed bytes, not on "
"the recycled slot's previous occupant",
desc.Cso.Slot, desc.Cso.Gen, desc.BaseCso.Slot, desc.BaseCso.Gen);
}
}
// The chunk bytes land in the record's own gathered order, so the record is always a
// complete pipeline half whatever mask minted it.
RenderStateParameters staging{};
MGPipeScatterPipelineBytes(record.PipelineBytes.data(), staging);
MGPipeScatterPipelineChunks(chunkBytes, desc.ChunkMask, staging);
MGPipeGatherPipelineBytes(staging, record.PipelineBytes.data());
record.Gen = desc.Cso.Gen;
record.Live = true;
}
void MGPipeApplyBindRenderState(const MGPBindRenderState& bind) {
const MGPipeRenderStateCsoRecord* record = FindCso(bind.Cso);
MOBILEGL_ASSERT(record != nullptr, "bind_render_state named a dead CSO {slot=%u, gen=%u}",
bind.Cso.Slot, bind.Cso.Gen);
if (record == nullptr) return;
PipeInputs& inputs = gPipeInputs;
MGPipeScatterPipelineBytes(record->PipelineBytes.data(),
MGPipeApplyAccess::RenderState(inputs));
MGPipeApplyAccess::SetRenderStateVersions(inputs, bind.Version, bind.PipelineVersion);
MGPipeApplier().BoundRenderStateCso = bind.Cso;
// A bind scatters the WHOLE pipeline half - the record is always a complete one,
// whatever mask minted it - so the pipeline chunks are all "moved" here, and all
// enter the applier's ledger of the bytes it owns.
const Uint32 moved = MGPipeGlobalChunkBitsOfPipelineMask(kAllPipelineChunks);
MGPipeApplier().ScatteredChunkBits |= moved;
MGPipeDeriveRenderStateFieldsForChunks(inputs, moved);
}
void MGPipeApplyDeleteRenderState(const MGPHandleOnly& handle) {
MOBILEGL_ASSERT(handle.Kind == static_cast<Uint32>(MGPipeKind::RenderStateCso),
"delete_render_state on kind %u", handle.Kind);
MGPipeRenderStateCsoRecord* record = FindCso(handle.Handle);
if (record == nullptr) return;
record->Live = false;
// The Gen stays: it is the CLIENT allocator that bumps it when the slot is handed
// out again (MGPipeHandles.h: "Gen increments only when a SLOT IS REUSED"), and a
// server-side bump here would put the two identities out of step.
if (MGPipeApplier().BoundRenderStateCso == handle.Handle) {
MGPipeApplier().BoundRenderStateCso = kMGPipeNullHandle;
}
}
void MGPipeApplySetDynamicState(const MGPDynamicState& dyn, const void* chunkBytes) {
PipeInputs& inputs = gPipeInputs;
MGPipeScatterDynamicChunks(chunkBytes, dyn.ChunkMask, MGPipeApplyAccess::RenderState(inputs));
MGPipeApplyAccess::SetRenderStateParametersVersion(inputs, dyn.Version);
const Uint32 moved = MGPipeGlobalChunkBitsOfDynamicMask(dyn.ChunkMask);
MGPipeApplier().ScatteredChunkBits |= moved;
MGPipeDeriveRenderStateFieldsForChunks(inputs, moved);
}
void MGPipeApplySetPixelPackState(const MGPPixelPackState& pack) {
MGPipeApplyAccess::PackState(gPipeInputs) = pack.Pack;
}
// set_context_values (P5c rv, CONTRACT-P5C.md §5.3): the residual-value record's server
// half. A plain store of the whole POD - there is no dirty mask and no field-level gate,
// because the record crosses only when one of its source values moved (the client's
// whole-record hash suppression) and a suppressed record means "nothing moved", never
// "field invalid" (§1).
void MGPipeApplySetContextValues(const MGPContextValues& values) {
static_assert(std::extent_v<decltype(MGPContextValues::TouchedBufferBindingPointCount)> ==
PipeInputs::kBufferTargetCount,
"the record's per-target array and PipeInputs' have drifted");
MGPipeApplyAccess::SetContextValues(gPipeInputs, values);
// P5e (CONTRACT-P5E.md §2.1, escalation (i)): mirror the XFB-active flag into the
// applier state as well. The barriered predicate has to be computable on the SERVER
// from applier state alone - gPipeInputs is the block the run-ahead client stops
// filling, so a predicate that read the answer out of THERE would be reading the very
// memory the rule exists to stop it reading. One store, inert until the caps bit is
// published, and it is what makes the two roles' third clause the same clause.
MGPipeApplier().IsTransformFeedbackActive = values.IsTransformFeedbackActive != 0;
#if MOBILEGL_BUILD_DISAGGREGATED
MGPipeApplier().BoundStreamOutputLifetimeId = values.BoundTransformFeedbackLifetimeId;
#endif
}
void MGPipeApplySetPatchState(const MGPPatchState& patch) {
PipeInputs& inputs = gPipeInputs;
RenderStateParameters& working = MGPipeApplyAccess::RenderState(inputs);
const FloatVec4 outer(patch.Outer[0], patch.Outer[1], patch.Outer[2], patch.Outer[3]);
const FloatVec2 inner(patch.Inner[0], patch.Inner[1]);
// THE SECOND TRIP WIRE (D6, D10). The patch trio travels TWICE - once in pipeline
// chunk P0, because it is pipeline state, and once as set_patch_state, because both
// backends bake it into the synthesized control stage from a shader-build path. The
// redundancy is the point: if the two carriers ever part, a stale set_patch_state
// silently clobbers what bind_render_state scattered and the tessellation levels a
// draw uses stop being the ones its CSO was minted for.
//
// Compared BITWISE, because a NaN outer level is a legal glPatchParameterfv value
// (ARCHITECTURE.md 5.2) and must compare equal to itself.
//
// ARMED BY THE APPLIER'S OWN SCATTER LEDGER, not by "some CSO has been bound"
// (PipeApply.h, MGPipeApplierState::ScatteredChunkBits). The question is whether the
// chunk-P0 bytes in the working block are the applier's, and that single condition
// covers both contracts this wire needs: the ORDERING one - a set_patch_state that
// legitimately precedes the first bind of a context has nothing to agree with yet -
// and the VERB-CLASS one - with the render-state subsystem off those bytes are the
// per-verb fill loop's, and FillPoints.def does not publish GetRenderStateParameters
// at kDispatch or kTextureOp, so they go stale there.
//
// It runs in the shipped push build too, because a wire that is compiled out of
// every build a device runs is not a wire. The cost is a 24-byte memcmp on a call
// that is emitted when the tessellation state CHANGES, i.e. about once per program.
if ((MGPipeApplier().ScatteredChunkBits & kChunksPatchTrio) == kChunksPatchTrio) {
++MGPipeApplier().PatchCarrierComparisons;
const Bool agrees = working.PatchVertices == patch.Vertices &&
std::memcmp(&working.PatchDefaultOuterLevel, &outer, sizeof(outer)) == 0 &&
std::memcmp(&working.PatchDefaultInnerLevel, &inner, sizeof(inner)) == 0;
if (!agrees) {
++MGPipeApplier().PatchCarrierDivergences;
MGP_TRIP_WIRE_REPORT("MGPipe: " MGP_TRIP_WIRE_TAG("PipePatchCarriersDiffer")
" set_patch_state says vertices=%u outer=(%g,%g,%g,%g) inner=(%g,%g); "
"chunk P0 delivered vertices=%u outer=(%g,%g,%g,%g) inner=(%g,%g)",
patch.Vertices, static_cast<double>(outer.x()),
static_cast<double>(outer.y()), static_cast<double>(outer.z()),
static_cast<double>(outer.w()), static_cast<double>(inner.x()),
static_cast<double>(inner.y()), working.PatchVertices,
static_cast<double>(working.PatchDefaultOuterLevel.x()),
static_cast<double>(working.PatchDefaultOuterLevel.y()),
static_cast<double>(working.PatchDefaultOuterLevel.z()),
static_cast<double>(working.PatchDefaultOuterLevel.w()),
static_cast<double>(working.PatchDefaultInnerLevel.x()),
static_cast<double>(working.PatchDefaultInnerLevel.y()));
}
}
working.PatchVertices = patch.Vertices;
working.PatchDefaultOuterLevel = outer;
working.PatchDefaultInnerLevel = inner;
MGPipeApplyAccess::SetPatchState(inputs, working.PatchVertices, working.PatchDefaultOuterLevel,
working.PatchDefaultInnerLevel);
}
void MGPipeApplySetVertexAttribDefaults(const MGPVertexAttribDefaults& hdr,
const MGPAttribValue* tail) {
PipeInputs& inputs = gPipeInputs;
PipeInputs::CurrentVertexAttributeValue* slots = MGPipeApplyAccess::VertexAttribDefaults(inputs);
// The loop walks the MASK's 32 bits, not the slot array, and every named bit consumes
// its tail entry even when there is no slot to write it to: a named-but-unstorable
// attribute that did not consume would write every attribute after it from the wrong
// entry. PipeInputs::kMaxVertexAttribs is VertexArrayObject's 32 today, so the
// out-of-range arm is unreachable - the guard is what keeps that true if the two ever
// stop agreeing. None of the three consistency checks may be left to MOBILEGL_ASSERT,
// which is inert at INFO: a malformed tail would then desynchronise the attribute
// writes without a word in exactly the builds that ship.
Uint32 consumed = 0;
const char* fault = nullptr;
for (Uint32 location = 0; location < 32 && fault == nullptr; ++location) {
if ((hdr.Mask & (Uint32{1} << location)) == 0) continue;
if (consumed >= hdr.Count) {
fault = "Mask names more attributes than Count";
break;
}
const MGPAttribValue& value = tail[consumed++];
if (value.Location != location) {
fault = "tail out of ascending location order";
break;
}
if (location >= PipeInputs::kMaxVertexAttribs) {
fault = "Mask names a location the block has no slot for";
continue;
}
PipeInputs::CurrentVertexAttributeValue& slot = slots[location];
// P5c (rv, CONTRACT-P5C.md §5.3): the record carries ALL THREE VIEWS VERBATIM, as
// the frontend computed them, and the applier writes each view from its own array.
// The pre-rv shape - one Data[4] memcpied into all three views - could not
// reproduce the frontend's NUMERIC cross-view conversion (glVertexAttrib4f(loc,
// 1.5f, ...) leaves 1 in intValue and 0x3FC00000 in floatValue), which is exactly
// why the field stayed EMITTED-AND-STILL-PULLED until rv. ValueClass rides along
// for the comparator; the applier no longer needs it to rebuild anything.
std::memcpy(slot.floatValue.data(), value.FloatView, sizeof(slot.floatValue));
std::memcpy(slot.intValue.data(), value.IntView, sizeof(slot.intValue));
std::memcpy(slot.uintValue.data(), value.UintView, sizeof(slot.uintValue));
}
if (fault == nullptr && consumed != hdr.Count) {
fault = "Count does not match the attributes Mask names";
}
if (fault != nullptr) {
MGP_TRIP_WIRE_REPORT("MGPipe: " MGP_TRIP_WIRE_TAG("PipeAttribTailMalformed")
" set_vertex_attrib_defaults: %s (Mask=0x%x, Count=%u, consumed=%u)",
fault, hdr.Mask, hdr.Count, consumed);
}
}
void MGPipeApplySetResidualValueState(const ResidualValueBlock& block) {
MGPipeApplier().Residual = block;
MGPipeApplier().HasResidual = true;
MGPipeApplier().ResidualCapabilitiesCompared = 0;
// THE TRIP WIRE (ARCHITECTURE.md 9.4, P2 brief D9). CapabilityBits is redundant with
// the assembled working block by design: every one of the 35 capabilities is
// answerable from RenderStateParameters now that P2 closed the three storage holes.
// So the day a later call takes a capability over and forgets to carry it, the two
// answers part and this says so on the next draw - which is what a migration carrier
// is for.
//
// THE ORACLE IS THE WORKING BLOCK, AND THE WIRE IS ARMED PER CAPABILITY by the
// applier's own scatter ledger: capability i is compared only once every chunk its
// answer is read out of has been scattered by this applier. That is not a weakening,
// it is the wire's whole precondition:
//
// - with the render-state subsystem off (MOBILEGL_PIPE_PUSH=0x10 is a legal
// configuration - D14's per-subsystem A/B) the ledger is empty and the wire says
// nothing at all, which is right: the working block is then the per-verb fill
// loop's, published per verb CLASS, so at a kDispatch or kTextureOp verb it holds
// the previous draw's bytes and disagreeing with it means nothing;
// - with it on, the applier is the block's only writer and its bytes are current at
// every verb of every class - including the two above, which is the case a
// RenderStateSpansTest case drives on purpose.
//
// NOT PipeInputs::m_capability, which the earlier form compared against and which
// FillPoints.def does publish at seven classes rather than five: where that
// publication is what makes m_capability fresh, the fill loop filled it out of the
// same GLContext the client built CapabilityBits from, so the comparison is a
// tautology. The redundancy this wire exists to check is between the CARRIED bits and
// the ASSEMBLED block.
const RenderStateParameters& working = MGPipeApplyAccess::RenderState(gPipeInputs);
const Uint32 owned = MGPipeApplier().ScatteredChunkBits;
for (SizeT i = 0; i < kCapabilityCount; ++i) {
const CapabilityInput cap = static_cast<CapabilityInput>(i);
const Uint32 sources = CapabilitySourceChunks(cap);
if ((owned & sources) != sources) continue;
++MGPipeApplier().ResidualCapabilitiesCompared;
const Bool carried = ((block.CapabilityBits >> i) & 1ull) != 0;
const Bool assembledBit = MGPipeApplyAccess::DeriveCapability(working, cap);
if (carried == assembledBit) continue;
++MGPipeApplier().ResidualDivergences;
MGP_TRIP_WIRE_REPORT("MGPipe: " MGP_TRIP_WIRE_TAG("PipeResidualDiverged, \"%s\"")
" carried=%d assembled=%d",
kCapabilityNames[i], static_cast<int>(carried),
static_cast<int>(assembledBit));
}
}
// ================================================================================
// P3a: the nine resource entry points (D-A1, D-A2).
//
// WHAT THE APPLIER OWNS HERE IS IDENTITY, EXTENT AND ORDER - NOT CONTENT. A buffer's
// bytes are the backend's; what crosses is a {slot, gen} handle, a flat descriptor and,
// where the call carries content, the client's own shadow base. So each body below does
// three things in this order: resolve the handle against this applier's record, check the
// record against its own declared extent, and only then move the record and hand the call
// to the backend.
//
// NOTHING REGISTERS MGPipeResourceOps IN THIS PACKAGE, so every dispatch below is a null
// check that falls through, and the tree behaves exactly as it did. That is deliberate and
// it is what makes this commit landable on its own: the frontend still dispatches the op
// table these replace, the backend that will register one is a later package, and the
// records these bodies keep are already correct when it does.
//
// THE THREE SERIALS ARE MGGen-CLASS: server-owned, monotone, never crossing the line. No
// MGPipe call may require the client to supply or know one (ARCHITECTURE.md 4.2.2), which
// is why they are incremented here rather than carried in a payload.
// ================================================================================
Bool MGPipeApplyResourceCreate(const MGPResourceDesc& desc) {
#if MOBILEGL_BUILD_DISAGGREGATED
if (!MGPipeRespecifyIsWholeResource(desc)) {
MGP_TRIP_WIRE_REPORT("MGPipe: Fatal{ProtocolCorruption, \"ResourceCreate.RespecifyScope\"} - "
"resource_create cannot carry a per-level respecify extent");
}
if (desc.Target != static_cast<Uint8>(MGPipeResourceTarget::TexBuffer) &&
(desc.BufOffset != 0 || desc.BufSize != 0)) {
MGP_TRIP_WIRE_REPORT("MGPipe: Fatal{ProtocolCorruption, \"ResourceCreate.BufferRange\"} - "
"only a texture-buffer resource may carry BufOffset/BufSize");
}
#endif
MOBILEGL_ASSERT(desc.Resource.Slot >= kMGPipeFirstAllocatableSlot,
"resource_create named the reserved slot 0");
if (desc.Resource.Slot < kMGPipeFirstAllocatableSlot) return false;
// P4a: THE TABLE IS CHOSEN BY THE DESCRIPTOR'S TARGET, and getting that wrong is the one
// way this call can damage an object it was not about - slot 7 is a live Buffer, a live
// Texture and a live Renderbuffer at the same time, in three independent slot spaces.
Vector<MGPipeResourceRecord>* table = ResourceTableForTarget(desc.Target);
if (table == nullptr) {
MGP_TRIP_WIRE_REPORT("MGPipe: " MGP_TRIP_WIRE_TAG("ProtocolCorruption")
" resource_create {slot=%u, gen=%u, glName=%u}: the descriptor names no "
"resource target (%u)",
desc.Resource.Slot, desc.Resource.Gen, desc.GlNameForDiag, desc.Target);
return false;
}
// A CREATE STARTS THE RECORD OVER rather than editing it. The slot it names may be a
// RECYCLED one whose record still describes the previous occupant, and inheriting one
// field of that - a Width, a Serial, an Immutable, a pending upload - is precisely how a
// buffer at a recycled address inherits its predecessor's contents. The generation is
// the client allocator's answer to "is this still the same GL object", so it is taken
// from the handle and nothing else survives.
MGPipeResourceRecord* record = RecordAt(*table, desc.Resource.Slot, kMGPipeMaxResourceSlots);
if (record == nullptr) {
MGP_TRIP_WIRE_REPORT("MGPipe: " MGP_TRIP_WIRE_TAG("ProtocolCorruption")
" resource_create {slot=%u, gen=%u, glName=%u}: the slot is outside the "
"record table's bound (%u)",
desc.Resource.Slot, desc.Resource.Gen, desc.GlNameForDiag,
kMGPipeMaxResourceSlots);
return false;
}
// P4a's belt, and only over the P4a half of this call: the BUFFER row is P3a's and its
// consumer question is the frontend's (MGPipeResourceSubsystemEnabled), which has
// already answered it before the emission - a buffer record can only arrive here on a
// backend that registered the table, so asking again would be a second copy of that
// rule in the one path that runs. A texture or renderbuffer create declined here leaves
// the client's publication latch at false, which is what makes the next respecify's
// self-healing create do the right thing if a table appears later.
if (desc.Target != kMGPipeResourceTargetBuffer && NoP4aConsumer()) return false;
*record = MGPipeResourceRecord{};
record->Gen = desc.Resource.Gen;
record->Live = true;
record->Desc = desc;
// Serial stays 0: a create is not a mutation. The descriptor a create carries defines
// no storage - that is the first respecify's job, and a backend tolerates a resource
// that has none - and a fresh backend twin starts its own synced serial at 0, so the
// two agree from the first instant without either side publishing anything.
//
// EVERY NON-BUFFER TARGET STORES AND RETURNS. Espryt allocates a texture's storage
// lazily inside SyncMipmapsToBackend and a renderbuffer's inside its own SyncToBackend,
// so there is no GL-call-time hook to dispatch to and P4a adds none: the record IS the
// publication, and the backend reads it at the sync point it already has.
if (desc.Target != kMGPipeResourceTargetBuffer) return true;
if (g_resourceOps != nullptr && g_resourceOps->Create != nullptr) {
g_resourceOps->Create(desc.Resource, desc);
}
return true;
}
Bool MGPipeApplyAttachSharedImage(MGPipeHandle texture, Uint64 imageId) {
MGPipeResourceRecord* record = FindIn(MGPipeApplier().TextureResources, texture);
if (record == nullptr) return false;
record->SharedImageId = imageId;
// The storage moved, so every twin's synced serial is stale, and the sampling answer with it.
++record->Serial;
MGPipeApplierNoteTextureStateMoved();
return true;
}
Bool MGPipeApplyResourceRespecify(const MGPResourceDesc& desc, const void* initialBytes,
const MGPRespecifiedLevel* level) {
Vector<MGPipeResourceRecord>* table = ResourceTableForTarget(desc.Target);
if (table == nullptr) {
MGP_TRIP_WIRE_REPORT("MGPipe: " MGP_TRIP_WIRE_TAG("ProtocolCorruption")
" resource_respecify {slot=%u, gen=%u, glName=%u}: the descriptor names "
"no resource target (%u)",
desc.Resource.Slot, desc.Resource.Gen, desc.GlNameForDiag, desc.Target);
return false;
}
// P4a's belt, for MGPipeApplyResourceCreate's reason and in front of the resolution
// rather than after it: with no consumer no create was ever accepted, so resolving
// first would report the absence as RefusedResourceCalls - the counter that means "a
// seam defect" - for a state that is by design.
if (desc.Target != kMGPipeResourceTargetBuffer && NoP4aConsumer()) return false;
MGPipeResourceRecord* record = ResolveResourceIn(*table, "resource_respecify", desc.Resource);
if (record == nullptr) return false;
PinNoLiveHostWrites(*record, desc.Resource, "resource_respecify");
PinRespecifyScopeCoversItsDeclaration(desc, level, desc.Resource, "resource_respecify");
#if MOBILEGL_BUILD_DISAGGREGATED
if (level != nullptr &&
(desc.Target == kMGPipeResourceTargetBuffer ||
desc.Target == static_cast<Uint8>(MGPipeResourceTarget::Renderbuffer) ||
desc.Target == static_cast<Uint8>(MGPipeResourceTarget::TexBuffer))) {
MGP_TRIP_WIRE_REPORT("MGPipe: Fatal{ProtocolCorruption, \"ResourceRespecify.Target\"} - "
"this resource target cannot name a texture mip level");
}
if (!MGPipeRespecifyIsWholeResource(desc) &&
!MGPipeRespecifiedExtentCarrierIsCanonical(desc)) {
MGP_TRIP_WIRE_REPORT("MGPipe: Fatal{ProtocolCorruption, \"ResourceRespecify.ExtentCarrier\"} - "
"a named image level's depth word must fit in 32 bits (no high BufSize bits)");
}
if (level == nullptr && desc.Target != static_cast<Uint8>(MGPipeResourceTarget::TexBuffer) &&
(desc.BufOffset != 0 || desc.BufSize != 0)) {
MGP_TRIP_WIRE_REPORT("MGPipe: Fatal{ProtocolCorruption, \"ResourceRespecify.BufferRange\"} - "
"a whole non-buffer resource may not carry BufOffset/BufSize");
}
#endif
// IS THIS A REDEFINITION AT ALL? Asked BEFORE the descriptor is replaced, because the
// stored one is the only thing there is to compare against (ID-18 M4). See
// RespecifyRedefinesNoStorage: true means every storage-defining field is unchanged and
// the call exists to carry a BindMask / ImageBindableHint that moved after the
// allocation - the immutable-texture case, where there is no later respecify to ride on.
const Bool metadataOnly = RespecifyRedefinesNoStorage(record->Desc, desc);
// The descriptor is replaced WHOLE, because that is what a respecify is: the store's
// extent, usage, storage flags, immutability and defined-content flag are all restated
// by the call that redefines it, and the backend reads them from here instead of
// asking a frontend object for them. A metadata update replaces it too - that is how
// the mask arrives - and by construction only the non-storage fields differ.
record->Desc = desc;
// A shared image is the storage this call replaces - unless the call replaces none.
if ((level != nullptr || !metadataOnly) && record->SharedImageId != 0) {
MGLOG_D("MGPipe: texture %u leaves shared image %llu: a respecify (%s) redefined its storage",
desc.GlNameForDiag, static_cast<unsigned long long>(record->SharedImageId),
level != nullptr ? "one level" : "whole resource");
record->SharedImageId = 0;
}
#if MOBILEGL_BUILD_DISAGGREGATED
if (level != nullptr && desc.Target != static_cast<Uint8>(MGPipeResourceTarget::TexBuffer)) {
// Scope and BufOffset/BufSize are transient respecify carriers, not resource state.
MGPipeClearRespecifiedLevel(record->Desc);
record->Desc.BufOffset = 0;
record->Desc.BufSize = 0;
}
#endif
// THE SERIAL MOVES EITHER WAY, and for a metadata update it is the entire publication:
// the twin re-derives its storage flags from the new mask at its next sync and decides
// for itself whether the backend needs a recreate.
++record->Serial;
// P5c (rv): a respecify can move a texture's shape, which is exactly what the
// sampling-resolution shutter guarded - the server-side answer moves with it.
MGPipeApplierNoteTextureStateMoved();
// A RESPECIFY REDEFINES A STORE, SO THE PENDING UPLOADS AGAINST THE STORE IT REPLACES
// GO WITH IT - AND ONLY THOSE. They are boxes and rects in a level's coordinate system
// and that space has just been replaced, so a box kept across a shrink would have
// Espryt upload past the end of the new level. But WHICH storage a respecify replaces
// is the call's to say and the descriptor's to say nothing about:
//
// - `level == nullptr` is a WHOLE-RESOURCE redefinition - every glBufferData /
// glBufferStorage, every glTexStorage* (which defines every level at once), every
// texture view - and every key goes.
// - a non-null `level` is ONE glTexImage*D on a mutable texture, which redefines that
// (uploadTarget, level) AND NOTHING ELSE. Dropping the other keys here would lose
// texels the client no longer owes: MG_State's AllocateStorage / MarkStorageDirty
// are per (uploadTarget, level) too, so the re-mark this call causes covers ITS
// level only, while every other level's dirty flag was cleared at its own emission
// (D-D5 step 1). The canonical sequence that loses them is glTexImage2D(0, data) ->
// draw (the shape is emitted and accumulated; Espryt bails because the texture is
// not yet mipmap-complete, Managers.cpp's incomplete-texture early return, which is
// the arm this set exists for) -> glTexImage2D(1, data), which under a blanket
// clear destroys level 0's entry before anything ever uploaded it.
//
// - and a METADATA update (ID-18 M4) with a NULL level drops NOTHING. It refines the
// whole-resource arm rather than contradicting it: the rule is "the uploads against
// the storage this call REPLACES go with it", and a call whose storage-defining
// fields all equal the stored descriptor replaces no storage, so no level's
// coordinate system has moved and every pending box is still described in the
// space it was accumulated in. B re-emits the descriptor when a sticky bind bit
// moves - with a null level, deliberately - which can land between a
// glTexSubImage2D and the sync that consumes it; eating those texels there would be
// C1's bug with a different trigger, and just as silent.
//
// - A NAMED LEVEL IS DROPPED WHETHER OR NOT THE base descriptor moved (P4a final
// review C-1). The companion scope now carries that level's exact W/H/D as well as
// its (uploadTarget, level) key, so the texture twin and staged-byte store update the
// same coordinate system. A metadata-only mask republish passes null and therefore
// cannot consume a standing upload on its behalf.
//
// A buffer never has a pending upload at all, so all three arms are inert for P3a's
// half - which is also why a buffer is never classified as metadata-only (below).
if (level != nullptr) {
// The keys are unique by AccumulatePendingUpload's construction - it looks for the
// pair before it appends - so this erases at most one entry and stops.
for (auto it = record->PendingUploads.begin(); it != record->PendingUploads.end(); ++it) {
if (it->UploadTarget != level->UploadTarget || it->Level != level->Level) continue;
record->PendingUploads.erase(it);
break;
}
} else if (metadataOnly) {
// nothing to drop, deliberately.
} else {
record->PendingUploads.clear();
}
#if MOBILEGL_BUILD_DISAGGREGATED
// P5c (tx): the staged-texture store's defined-ness and drop bookkeeping rides THE SAME
// scope rules as the pending-set drops above - a named level redefines that one
// (uploadTarget, level), a whole-resource respecify drops every level, and a metadata
// update replaces no storage and is not delivered. Without this the store could not say
// whether a level exists at all (a null-data glTexImage*D carries no sub-data), and a
// whole-resource respecify would leave levels keyed against a replaced coordinate
// system. Textures only: a buffer's storage is the ops table's own Respecify hook, and
// a renderbuffer has no levels.
if (desc.Target != kMGPipeResourceTargetBuffer &&
desc.Target != static_cast<Uint8>(MGPipeResourceTarget::Renderbuffer) &&
(level != nullptr || !metadataOnly)) {
if (g_resourceOps != nullptr && g_resourceOps->TextureRespecify != nullptr)
g_resourceOps->TextureRespecify(desc.Resource, desc, level);
else
DefineTextureWithoutBackendHook(desc, level);
}
#endif
// resource_respecify is the catalogue's only kNeedsAck call, and the per-record half
// of that flag is MGPipeResourceRespecifyNeedsAck(desc): glBufferStorage is a real
// synchronous allocation and the only entry point allowed a synchronous ack, while
// glBufferData travels through the same call and must not acknowledge one. In monolith
// the acknowledgement IS the return of this function - the applier is one call away -
// so the predicate has nothing to gate here and is deliberately not branched on: a
// branch whose arms were identical would be dead code the transport would then have to
// find and remove. PipeCatalogueTest.ResourceRespecifyAcksOnlyImmutableStorage is what
// keeps the predicate honest until the doorbell reads it.
//
// ID-18 M4 ASKS FOR "NO REALLOCATION ACK" ON A METADATA UPDATE, AND IT IS TRUE HERE BY
// CONSTRUCTION RATHER THAN BY A BRANCH: the predicate is false for every target that is
// not a buffer, and RespecifyRedefinesNoStorage refuses the buffer target outright, so
// MGPipeResourceRespecifyNeedsAck(desc) is false for every record this applier
// classifies as metadata-only. That is why the classification excludes buffers rather
// than suppressing the ack afterwards - a suppression would have been a second rule the
// transport would then have to learn.
if (desc.Target != kMGPipeResourceTargetBuffer) return true;
if (g_resourceOps != nullptr && g_resourceOps->Respecify != nullptr) {
g_resourceOps->Respecify(desc.Resource, desc, initialBytes);
}
return true;
}
Bool MGPipeApplyResourceSubData(const MGPSubData& record, const void* bytes,
const MGPSubRegion* regions) {
// ONE CALL, TWO HALVES, and the branch is one comparison. For a buffer the applier
// stores NOTHING per record - the contents are the backend's, and the range is the
// backend's to land - so its whole job is the gate and the serial. For a texture there
// is no backend hook at all and the whole job is the gate, the accumulated shape and
// the serial.
if (SubDataNamesABuffer(record)) {
MOBILEGL_ASSERT(regions == nullptr,
"resource_subdata: the buffer half declares no sub-regions and carries none");
return ApplyBufferWrite("resource_subdata", record, bytes, /*resident=*/false);
}
// AND THE TEXTURE HALF IS NOT "EVERYTHING ELSE". The table selector's own argument -
// acting on the wrong table would have this applier act outside the storage the record
// names - is exactly the argument for refusing a resource-target half that names no
// texture, and Renderbuffer is the reachable one: it is enumerator 10 and therefore a
// perfectly well-formed MGPSubData::Target, so without this a renderbuffer record would
// accumulate a pending upload onto whatever TEXTURE holds slot N in the texture slot
// space. Renderbuffers have no sub-data path at all, so no correct client can produce
// one and this is a protocol fault rather than a dropped call.
PinLiveHostWritesNamesABuffer(record, "resource_subdata");
const Uint8 resourceTarget = MGPipeSubDataResourceTargetOf(record.Target);
if (resourceTarget == kMGPipeResourceTargetBuffer ||
resourceTarget == static_cast<Uint8>(MGPipeResourceTarget::Renderbuffer) ||
resourceTarget >= static_cast<Uint8>(MGPipeResourceTarget::Count)) {
// BOTH HALVES GO IN THE LINE. The packed field alone leaves a reader doing the
// arithmetic the encoding exists to stop anyone doing by hand, and the upload half
// is what says whether the emitter packed the pair the wrong way round.
MGP_TRIP_WIRE_REPORT("MGPipe: " MGP_TRIP_WIRE_TAG("ProtocolCorruption")
" resource_subdata {slot=%u, gen=%u}: the record's resource target names "
"no texture to upload into (target=%u, resource target=%u, upload "
"target=%u)",
record.Res.Slot, record.Res.Gen, record.Target,
static_cast<Uint32>(resourceTarget),
static_cast<Uint32>(MGPipeSubDataUploadTargetOf(record.Target)));
return false;
}
// P4a's belt, AFTER the target fault above and before the upload is accumulated: this
// is the call whose acceptance clears a level's dirty flags (D-D5 / ID-18 M3), so it is
// the one that turned "no consumer" into lost texels on Magma.
if (NoP4aConsumer()) return false;
return ApplyTextureUpload(record, bytes, regions);
}
void MGPipeApplyBufferSubDataResident(const MGPSubData& record, const void* bytes) {
// `bytes` is the application's staging store and is valid for THE DURATION OF THE CALL
// ONLY, which is why the backend hook copies rather than remembering the pointer.
ApplyBufferWrite("buffer_subdata_resident", record, bytes, /*resident=*/true);
}
// THE FOUR BUFFER-ONLY CALLS, AND WHAT POLICES THAT. resource_flush_range,
// resource_readback and map_persistent resolve through ResolveResource - i.e. against
// MGPipeApplier().Resources and never the texture or renderbuffer table - which is right, because
// a mapped range, a readback and a persistent donation exist for buffers only. They are
// buffer-only BY CATALOGUE and not by check: MGPFlushRange and MGPReadback carry no kind
// at all, and map_persistent's MGPHandleOnly is not asked for one, so there is nothing here
// to compare and a mistyped record can only be caught where a discriminator exists.
// unmap_persistent is that place, and it Fatals - see there. If a later phase gives any of
// these three a kind, the gate belongs beside that field on the same terms.
void MGPipeApplyResourceFlushRange(const MGPFlushRange& record, const void* bytes) {
MGPipeResourceRecord* stored = ResolveResource("resource_flush_range", record.Res);
if (stored == nullptr) return;
const char* fault = BufferRangeFault(record.Offset, record.Size, stored->Desc.Width);
// item-12 / R-13.2 / ID-37: the "a non-empty flush carries no bytes" fault is a MONOLITH
// rule. Under an ACTIVE TRANSPORT resource_flush_range carries no companion pointer AT ALL
// (contract table 1 row 20 / R-13.2): the covering resource_subdata already staged those
// bytes into the server-owned StagedShadowStore (R-11), and Ops_H_FlushRange reads that
// shadow, not the record. So a null `bytes` on a non-empty flush is the NORMAL split case,
// not corruption - and a flush whose bytes were genuinely never staged is caught more
// precisely downstream by R-11's Fatal{StageSnapshotTooNarrow}, which names the missing
// subdata rather than the flush. The check stays exact under monolith
// (MG_Config::Transport is a constexpr Monolith in a non-disaggregated build, so this is
// `&& true` there and the codegen is unchanged). This was the only seam between the joint
// inproc lane's 14/21 and 21/21: all seven PersistentCoherentMapScenario entries aborted
// here (c1-v2.md 6).
if (fault == nullptr && record.Size != 0 && bytes == nullptr &&
MG_Config::Transport == MG_Config::TransportMode::Monolith) {
fault = "a non-empty flush carries no bytes";
}
if (fault != nullptr) {
MGP_TRIP_WIRE_REPORT("MGPipe: " MGP_TRIP_WIRE_TAG("ProtocolCorruption")
" resource_flush_range {slot=%u, gen=%u, glName=%u}: %s "
"(offset=%llu, size=%llu, storage=%u bytes)",
record.Res.Slot, record.Res.Gen, stored->Desc.GlNameForDiag, fault,
static_cast<unsigned long long>(record.Offset),
static_cast<unsigned long long>(record.Size), stored->Desc.Width);
return;
}
PinNoLiveHostWrites(*stored, record.Res, "resource_flush_range");
++stored->Serial;
// record.AccessFlags are the application's REAL mapping flags and this applier does
// not normalise them: the backend reads INVALIDATE_RANGE / INVALIDATE_BUFFER /
// UNSYNCHRONIZED per call to choose its upload shape, and merging them here would take
// that choice away from the side that pays for it.
if (g_resourceOps != nullptr && g_resourceOps->FlushRange != nullptr) {
g_resourceOps->FlushRange(record.Res, record, bytes);
}
}
void MGPipeApplyResourceReadback(const MGPReadback& record) {
MGPipeResourceRecord* stored = ResolveResource("resource_readback", record.Res);
if (stored == nullptr) return;
const char* fault = BufferRangeFault(record.Offset, record.Size, stored->Desc.Width);
if (fault != nullptr) {
MGP_TRIP_WIRE_REPORT("MGPipe: " MGP_TRIP_WIRE_TAG("ProtocolCorruption")
" resource_readback {slot=%u, gen=%u, glName=%u}: %s "
"(offset=%llu, size=%llu, storage=%u bytes)",
record.Res.Slot, record.Res.Gen, stored->Desc.GlNameForDiag, fault,
static_cast<unsigned long long>(record.Offset),
static_cast<unsigned long long>(record.Size), stored->Desc.Width);
return;
}
// The readback READS the store this flag describes, so it is one of the calls whose
// answer would change silently the day a producer sets it (D-A4).
PinNoLiveHostWrites(*stored, record.Res, "resource_readback");
// NO SERIAL MOVES. A readback does not mutate the resource; it produces host bytes out
// of it. Bumping here would tell the backend twin its store had changed and buy a
// re-upload of what it had just been read out of.
//
// THE ANSWER TRAVELS BACK THROUGH THE REVERSE CHANNEL, NOT THROUGH THIS FUNCTION, and
// the applier's contribution to that is the ORDER: the hook is called synchronously
// and this returns only once it has finished, so the writeback into the client's
// shadow and the mutation-epoch bump that must follow it have both happened before the
// caller reads. The epoch bump stays server-side and happens AFTER the writeback,
// never before - a rule the reverse channel needs as much as the forward one, because
// a bump that overtook its writeback would leave the draw-clean memo stale behind it.
//
// With no table registered nothing answers, and nothing asks either: the frontend is
// still on the path this call replaces.
if (g_resourceOps != nullptr && g_resourceOps->Readback != nullptr) {
g_resourceOps->Readback(record.Res, record);
}
}
void MGPipeApplyResourceDestroy(const MGPHandleOnly& handle) {
// THE KIND IS THE DISCRIMINATOR HERE, because a destroy carries no descriptor, and it is
// a Fatal rather than an assertion for the reason resource_create's target is: the three
// slot spaces are independent, so a destroy routed to the wrong table would drop the
// record of a live object of another kind that happens to hold the same slot.
Vector<MGPipeResourceRecord>* table = ResourceTableForKind(handle.Kind);
if (table == nullptr) {
MGP_TRIP_WIRE_REPORT("MGPipe: " MGP_TRIP_WIRE_TAG("ProtocolCorruption")
" resource_destroy {slot=%u, gen=%u}: the handle names no resource kind "
"(%u)",
handle.Handle.Slot, handle.Handle.Gen, handle.Kind);
return;
}
MGPipeResourceRecord* record = ResolveResourceIn(*table, "resource_destroy", handle.Handle);
if (record == nullptr) return;
// The record is dropped WHOLE and the generation is kept. The client allocator owns
// the Gen bump, and it takes it on the next HANDOUT of the slot rather than on the
// free, so a server-side bump here would put the two identities out of step and a
// double free could skip a generation. Everything else goes: a stale read of a
// destroyed slot must find nothing, not the extent of the buffer that used to be there.
const Uint32 gen = record->Gen;
*record = MGPipeResourceRecord{};
record->Gen = gen;
// P5c (rv): a destroyed object may still be BOUND somewhere the texture shutters guard;
// the server's answer for them moves with the death rather than waiting for the next
// unit-set record to say so.
MGPipeApplierNoteTextureStateMoved();
// The applier's own state is consistent before the backend hears the news, so a hook
// that looked back at this applier could not see a resource that is already gone. The
// CLIENT frees the slot after this returns, in that order, because the allocator
// forgets the lifetime id on free and a notice resolved twice finds nothing the second
// time.
//
// AND ONLY A BUFFER IS HANDED ON, for resource_create's reason: the op table is the
// buffer family's, its Destroy takes a handle whose kind that backend registered for,
// and a texture's death is read out of the record at the sync that would have used it.
#if MOBILEGL_BUILD_DISAGGREGATED
// P5c (tx): with ONE exception - the staged-texture store is keyed by the handle, so
// the death must reach it or a recycled slot's stale levels would answer for the
// successor. This does not hand the texture to the buffer family's Destroy.
if (static_cast<MGPipeKind>(handle.Kind) == MGPipeKind::Texture) {
if (g_resourceOps != nullptr && g_resourceOps->TextureDestroy != nullptr)
g_resourceOps->TextureDestroy(handle.Handle);
else if (MG_Config::Transport != MG_Config::TransportMode::Monolith)
MG_Remote::Server::ServerStagedTexture().Drop(
MG_Remote::Server::StagedTextureStore::KeyForHandle(handle.Handle));
}
#endif
if (static_cast<MGPipeKind>(handle.Kind) != MGPipeKind::Buffer) return;
if (g_resourceOps != nullptr && g_resourceOps->Destroy != nullptr) {
g_resourceOps->Destroy(handle.Handle);
}
}
void* MGPipeApplyMapPersistent(const MGPHandleOnly& handle, Uint64 size, const void* seedBytes) {
// COUNTED FIRST AND UNCONDITIONALLY, because the counter is acquisition ATTEMPTS and
// not acquisitions: every one of them - mint or decline - needs an answer from the
// resource owner, and under a transport a decline costs the same round trip as a mint.
// Defined that way the number is identical in both modes, equals "one per storage
// definition", and is non-zero and assertable today; defined as "round trips actually
// taken" it would be 0 by construction in monolith and could never go red.
++MGPipeApplier().MapPersistentRoundtrips;
MGPipeResourceRecord* record = ResolveResource("map_persistent", handle.Handle);
if (record == nullptr) return nullptr;
// THE ONE CALL A LATER PHASE ATTACHES THE PRODUCER TO. D-A4 and ARCHITECTURE.md name
// the persistent-map push as exactly where HasLiveHostWrites gets set, so this is the
// call the pin must sit on: a producer landing under it here is the semantic change
// the flag exists to announce, and the wire is what refuses to let it arrive unnamed.
PinNoLiveHostWrites(*record, handle.Handle, "map_persistent");
#if MOBILEGL_BUILD_DISAGGREGATED
// R-6: A SPLIT BUILD RUNS AT TIER T2 AND DECLINES EVERY ACQUISITION, ALWAYS.
//
// The mint returns a raw void* that the client stores as the store's base
// (BufferObject.cpp:238 / :603 / :658). Across a process boundary that address is
// meaningless, and under `inproc` it is WORSE than meaningless: it happens to work,
// so a lane that kept adoption alive would be green for a reason spawn cannot
// reproduce, and persistent-map-push - an exit-gate counter - would be structurally
// zero (PipeStats.cpp says so in as many words). Forcing T2 here rather than at the
// three client call sites is what keeps `mpr` identical between the arms: the
// roundtrip is COUNTED above, unconditionally, because a decline costs the same round
// trip as a mint.
//
// The frontend already tolerates a decline in all three places, and has since P3a.
if (MG_Config::Transport != MG_Config::TransportMode::Monolith &&
MG_Remote::Client::AdoptTierIsEmulate()) {
return nullptr;
}
#endif
// NO SERIAL MOVES and NO DESCRIPTOR CHANGES: the donation re-mints the backend's own
// driver object, which is a server-local event that the backend's own id generation
// already catches, and the client's view of the store's extent is untouched by it.
//
// A NULL RETURN IS A DECLINE, not a failure - the acquisition is allowed to say no,
// the caller already has that branch, and that is why the call is kOptional as well as
// kReplySlot. An unregistered table declines every acquisition, which is exactly the
// answer a build with no migrated backend should give.
if (g_resourceOps == nullptr || g_resourceOps->MapPersistent == nullptr) return nullptr;
return g_resourceOps->MapPersistent(handle.Handle, size, seedBytes);
}
void MGPipeApplyUnmapPersistent(const MGPHandleOnly& handle) {
// THE SAME VERDICT resource_destroy GIVES, AND FOR THE SAME REASON. This is the only
// one of the four buffer-only calls that carries a discriminator at all, and an
// assertion is not a check: MOBILEGL_ASSERT compiles out at INFO, which is what all
// three gate builds and every shipped build are, so a mistyped record used to walk
// straight into ResolveResource and alias whatever BUFFER holds that slot - the three
// slot spaces being independent - which is exactly the failure resource_destroy's
// ResourceTableForKind Fatal exists to stop. Acting on another kind's storage is the
// corrupt-record verdict, not the dropped-call one, so it moves no refusal counter.
if (static_cast<MGPipeKind>(handle.Kind) != MGPipeKind::Buffer) {
MGP_TRIP_WIRE_REPORT("MGPipe: " MGP_TRIP_WIRE_TAG("ProtocolCorruption")
" unmap_persistent {slot=%u, gen=%u}: the persistent donation is the "
"buffer family's and the handle names another kind (%u)",
handle.Handle.Slot, handle.Handle.Gen, handle.Kind);
return;
}
MGPipeResourceRecord* record = ResolveResource("unmap_persistent", handle.Handle);
if (record == nullptr) return;
// Never emitted by this phase's own paths - the donation is permanent for the store's
// life and the retire happens inside the backend - so this exists to keep the pair
// complete and to give the transport both halves.
if (g_resourceOps != nullptr && g_resourceOps->UnmapPersistent != nullptr) {
g_resourceOps->UnmapPersistent(handle.Handle);
}
}
#if MOBILEGL_BUILD_DISAGGREGATED
Bool MGPipeApplyAdoptExternal(const MGPHandleOnly& handle, void* ahb, Uint64 size) {
// The same kind verdict unmap_persistent gives: a store is the buffer family's.
if (static_cast<MGPipeKind>(handle.Kind) != MGPipeKind::Buffer) {
MGP_TRIP_WIRE_REPORT("MGPipe: " MGP_TRIP_WIRE_TAG("ProtocolCorruption")
" adopt_external {slot=%u, gen=%u}: a T0 store is the buffer family's "
"and the handle names another kind (%u)",
handle.Handle.Slot, handle.Handle.Gen, handle.Kind);
return false;
}
MGPipeResourceRecord* record = ResolveResource("adopt_external", handle.Handle);
if (record == nullptr) return false;
// The store the AHB replaces is the record's whole extent; a handle whose width moved
// (a respecify the client has not caught up with) is not this store any more.
if (size == 0 || size != record->Desc.Width) return false;
// The same pin map_persistent takes: adoption is where a producer of live host writes
// would attach, and T0 attaches none - the client writes the AHB through its lock, as a
// monolith client writes its adopted map, and the applier is told nothing either way.
PinNoLiveHostWrites(*record, handle.Handle, "adopt_external");
if (g_resourceOps == nullptr || g_resourceOps->ImportExternal == nullptr) return false;
return g_resourceOps->ImportExternal(handle.Handle, ahb, size);
}
#endif
// ================================================================================
// P3a: the five vertex-input entry points (D-G, D-H, D-I).
//
// THESE FIVE DISPATCH TO NOBODY, and that is not an omission: MGPipeResourceOps is the
// RESOURCE family's table, and the vertex-input calls have no backend hook because the
// backend does not act on them when they arrive. It reads them at its own draw-time sync,
// out of the state below, which is what the three serials here are for - they are what
// retires the twin's wrapping-Uint16-plus-identity patches, so a compare that used to ask
// "is my Uint16 configuration version still the frontend's?" asks "is my Uint64 serial
// still the server's?" instead.
//
// Nothing emits them in this package either: their dirty bits map to no subsystem, the two
// subsystem bits are not in MG_Impl/Pipe/PipeFill.cpp's kMGPipeWiredSubsystems, and the
// emitters beside them are stubs. The state below is therefore correct and unread until
// the packages that wire both ends land.
// ================================================================================
void MGPipeApplyCreateVertexElements(const MGPVertexElements& desc, const void* blobBytes) {
MOBILEGL_ASSERT(desc.Cso.Slot >= kMGPipeFirstAllocatableSlot,
"create_vertex_elements named the reserved slot 0");
if (desc.Cso.Slot < kMGPipeFirstAllocatableSlot) return;
// THE COUNTS ARE CHECKED BEFORE A BYTE OF THE BLOB IS TOUCHED, and the check is the
// record's own self-description: the blob is MGPVertexAttribWire[AttributeCount]
// immediately followed by MGPVertexBindingPointWire[BindingPointCount]. THIS is the
// reason the second view travels at all - a record that declares a BindingPointCount
// it does not carry would otherwise be a shape this gate had to police forever with
// nothing to police it against.
//
// Both counts are bounded by GL's attribute limit, which is also the size of the two
// arrays they are unpacked into, so the bound and the destination cannot drift apart.
// THE COUNTS ARE WHAT BOUNDS THE READ; the declared blob length is a cross-check.
//
// THE BLOB RULE IS THE SAME ONE resource_subdata IS HELD TO (SubDataBoxFault above,
// and MGPipeTypes.h states it on both records): a non-zero Blob.Size must be exactly
// the length the record's other fields describe, and a zero Blob.Size means "this
// record does not declare its blob" - which is what a monolith emission is, since the
// bytes travel beside the record through blobBytes and no MGPBlobRef is filled. One
// rule, both blob-carrying families: a transport that fills the field gets a real gate
// on the first truncated record, and a client that leaves it zero does not abort a
// verify build over a field it never used.
const Uint64 attributeBytes = Uint64{desc.AttributeCount} * sizeof(MGPVertexAttribWire);
const Uint64 bindingBytes = Uint64{desc.BindingPointCount} * sizeof(MGPVertexBindingPointWire);
const Uint64 declared = attributeBytes + bindingBytes;
const char* fault = nullptr;
if (desc.AttributeCount > kMGPipeMaxVertexAttribs) {
fault = "the declared attribute count is above GL's attribute limit";
} else if (desc.BindingPointCount > kMGPipeMaxVertexAttribs) {
fault = "the declared binding-point count is above GL's attribute limit";
} else if (desc.Blob.Size != 0 && desc.Blob.Size != declared) {
fault = "the declared blob length is not the byte length the two counts describe";
} else if (declared != 0 && blobBytes == nullptr) {
fault = "a non-empty blob carries no bytes";
}
if (fault != nullptr) {
MGP_TRIP_WIRE_REPORT("MGPipe: " MGP_TRIP_WIRE_TAG("ProtocolCorruption")
" create_vertex_elements {slot=%u, gen=%u}: %s (attributes=%u, "
"bindingPoints=%u, that describes %llu bytes, blob declares %llu)",
desc.Cso.Slot, desc.Cso.Gen, fault, desc.AttributeCount,
desc.BindingPointCount, static_cast<unsigned long long>(declared),
static_cast<unsigned long long>(desc.Blob.Size));
return;
}
MGPipeVertexElementsRecord* recordAt =
RecordAt(MGPipeApplier().VertexElementsCsos, desc.Cso.Slot, kMGPipeMaxVertexElementsSlots);
if (recordAt == nullptr) {
MGP_TRIP_WIRE_REPORT("MGPipe: " MGP_TRIP_WIRE_TAG("ProtocolCorruption")
" create_vertex_elements {slot=%u, gen=%u}: the slot is outside the record "
"table's bound (%u)",
desc.Cso.Slot, desc.Cso.Gen, kMGPipeMaxVertexElementsSlots);
return;
}
MGPipeVertexElementsRecord& record = *recordAt;
// A RE-CREATE ON THE SAME HANDLE IS HOW A CONFIGURATION CHANGE TRAVELS - the handle is
// minted per frontend vertex array and a generation moves only when a slot is reused -
// so an existing record of the same identity keeps its serial and counts up from it. A
// record of a DIFFERENT identity is a recycled slot and starts over, or the walks below
// would read the previous occupant's attributes out of the array's tail.
if (!record.Live || record.Gen != desc.Cso.Gen) {
record = MGPipeVertexElementsRecord{};
record.Gen = desc.Cso.Gen;
}
// Zeroed first, so a configuration that shrinks does not leave the entries above its
// new count describing the one before it. memcpy rather than a typed store because a
// blob pointer carries no alignment guarantee.
record.Attributes = {};
record.BindingPoints = {};
const Uint8* blob = static_cast<const Uint8*>(blobBytes);
if (attributeBytes != 0) {
std::memcpy(record.Attributes.data(), blob, static_cast<SizeT>(attributeBytes));
}
if (bindingBytes != 0) {
std::memcpy(record.BindingPoints.data(), blob + attributeBytes, static_cast<SizeT>(bindingBytes));
}
record.AttributeCount = desc.AttributeCount;
record.BindingPointCount = desc.BindingPointCount;
record.Live = true;
// Serial 0 means "never created", so the first create of an identity lands on 1 and a
// backend twin that has synced nothing can never accidentally match a live record.
++record.ContentSerial;
// AND IT DOES NOT REBIND. A create on the bound handle changes what the binding points
// AT, which the serial already says; a create on any other handle must not steal the
// binding.
}
void MGPipeApplyBindVertexElements(const MGPHandleOnly& handle) {
MOBILEGL_ASSERT(handle.Kind == static_cast<Uint32>(MGPipeKind::VertexElementsCso),
"bind_vertex_elements on kind %u", handle.Kind);
// The null handle is legal and means "no vertex array bound" - GL's unbound state is a
// state, not an error, and the backend has a branch for it.
if (MGPipeHandleIsNull(handle.Handle)) {
MGPipeApplier().BoundVertexElements = kMGPipeNullHandle;
return;
}
// A dead handle leaves the PREVIOUS binding untouched, which is bind_render_state's
// precedent for the same question, and is counted like every other refusal.
const MGPipeVertexElementsRecord* record = ResolveVertexElements("bind_vertex_elements", handle.Handle);
if (record == nullptr) return;
MGPipeApplier().BoundVertexElements = handle.Handle;
}
void MGPipeApplyDeleteVertexElements(const MGPHandleOnly& handle) {
MOBILEGL_ASSERT(handle.Kind == static_cast<Uint32>(MGPipeKind::VertexElementsCso),
"delete_vertex_elements on kind %u", handle.Kind);
// A death notice on a record this applier does not have is the SAME refusal every
// other entry point makes, with the same verdict and the same counter: n1's
// inconsistency (a bare `return` with no assertion and no reason) is closed by routing
// it through the resolver rather than by giving it a private answer. It is also the
// one refusal a legal sequence produces - the teardown order beside
// kResourceRefusalNote - which is why it stays a no-op.
MGPipeVertexElementsRecord* record = ResolveVertexElements("delete_vertex_elements", handle.Handle);
if (record == nullptr) return;
// Dropped whole, generation kept, for resource_destroy's reason: the client allocator
// owns the Gen bump and takes it on the next handout of the slot. Emitted from ONE
// place - the frontend object's death notice - so there is no second path to keep in
// step with this one.
const Uint32 gen = record->Gen;
*record = MGPipeVertexElementsRecord{};
record->Gen = gen;
if (MGPipeApplier().BoundVertexElements == handle.Handle) {
MGPipeApplier().BoundVertexElements = kMGPipeNullHandle;
}
}
void MGPipeApplySetVertexBuffers(const MGPVertexBuffers& hdr, const MGPVertexBuffer* tail) {
// The window is the bound: Start + Count entries land in an array of exactly GL's
// attribute limit, and a var-tail header that describes more than its destination can
// hold is the same class of fault as a blob outside its segment.
const Uint64 end = Uint64{hdr.Start} + Uint64{hdr.Count};
const char* fault = nullptr;
if (end > kMGPipeMaxVertexAttribs) {
fault = "the window runs past GL's attribute limit";
} else if (hdr.Count != 0 && tail == nullptr) {
fault = "a non-empty set carries no entries";
}
if (fault != nullptr) {
MGP_TRIP_WIRE_REPORT("MGPipe: " MGP_TRIP_WIRE_TAG("ProtocolCorruption")
" set_vertex_buffers {start=%u, count=%u, hash=%llu}: %s (the applier holds "
"%u entries)",
hdr.Start, hdr.Count, static_cast<unsigned long long>(hdr.ContentHash),
fault, kMGPipeMaxVertexAttribs);
return;
}
// Copied into the window the record declares and nowhere else. The entries outside it
// are not cleared: this record is "the last set as received", and a set that names
// four entries has said nothing about the rest.
for (Uint32 i = 0; i < hdr.Count; ++i) {
MGPipeApplier().VertexBuffers[hdr.Start + i] = tail[i];
}
MGPipeApplier().VertexBufferStart = hdr.Start;
MGPipeApplier().VertexBufferCount = hdr.Count;
// THE RAW VALUE IS STORED, NOT A RESOLVED SHIFT, and the resolution is one step
// further out on purpose. Whether the fetch shift has to be emulated at all is a
// BACKEND CAPABILITY - a device with native base-instance support does it in hardware
// and shifts nothing - and emulation is server-owned, so the answer belongs to the
// backend arm that computes the per-attribute byte shift out of this value and each
// attribute's own stride and divisor. This applier is below MG_Backend and may not ask
// the question; storing the raw value keeps the one answer in the one place that can
// give it.
//
// The client sends it once per SET rather than per entry, and it is a ContentHash
// input: set_vertex_buffers is suppressed on an unchanged hash, so a base instance
// that moved while the buffer set did not would otherwise never arrive and the server
// would keep the previous shift.
MGPipeApplier().VertexFetchBaseInstance = hdr.BaseInstance;
++MGPipeApplier().VertexBuffersSerial;
}
void MGPipeApplySetIndexBuffer(const MGPIndexBuffer& record) {
// Stored verbatim, with no gate over it, and each of the three fields has its own
// reason to be taken as sent:
// - Res may legitimately be the null handle: no element-array buffer is bound, which
// is the state a client-memory index draw is in;
// - Offset and IndexSize are the DRAW's, not the binding's, and are 0 and 0 until a
// draw supplies them - so there is no extent here to check a range against, and
// the draw verb overrides them anyway;
// - it is an INDEPENDENT call and NOT a subset of the vertex-elements configuration
// (D5): a rebind of the index slot must move this serial without touching the
// configuration's, which is exactly what the backend's two separate compares need.
MGPipeApplier().IndexBuffer = record;
++MGPipeApplier().IndexBufferSerial;
}
void MGPipeDeriveRenderStateFields(PipeInputs& inputs) {
// The derivation itself lives in MGPipeApplyAccess above, because that is the one
// struct PipeInputs names as a friend - see D5 there for what it recomputes and which
// getter each line was transcribed from.
MGPipeApplyAccess::DeriveRenderStateFields(inputs, kMGPipeAllGlobalChunks);
}
void MGPipeDeriveRenderStateFieldsForChunks(PipeInputs& inputs, Uint32 globalChunkBits) {
MGPipeApplyAccess::DeriveRenderStateFields(inputs, globalChunkBits);
}
// ================================================================================
// P4a: the fifteen object and working-state entry points
// ================================================================================
//
// NOT ONE OF THEM DISPATCHES TO A BACKEND FUNCTION POINTER, and that is the structural
// decision the whole phase rests on rather than an omission. Nothing in these families
// reaches the backend at GL-call time today - texture storage only marks a level dirty,
// texture parameters run from the draw-time sync, renderbuffer storage is allocated inside
// SyncToBackend, a sampler twin is created lazily from the program pass, and the
// framebuffer, the unit sets and the program are all resolved at PrepareForDraw - so every
// call below is either an OBJECT RECORD the applier stores or WORKING STATE the applier
// stores, and the backend reads the applier at the sync points it already has, keyed on a
// server-owned Serial instead of a frontend version. MGPipeResourceOps is unchanged.
//
// THE SAME THREE STEPS, IN THE SAME ORDER, AS P3a's NINE: resolve the handle against this
// applier's record, check the record against its own declared extent, and only then move
// the record. The two verdicts are the same two, and the difference between them is the
// whole of this file's discipline - a call naming a record this applier does not have is a
// DEFINED NO-OP that is COUNTED (RefusedObjectCalls), because one legal sequence produces
// it; a record that would make the server index or allocate outside its own storage is
// Fatal{ProtocolCorruption} with the record's identity in the line.
//
// NOTHING EMITS ANY OF THEM IN THIS PACKAGE. The four subsystem bits are not in
// kMGPipeWiredSubsystems, every client emitter beside them is still a stub, and no backend
// reads the records yet - so the state below is correct and unread until the packages that
// wire both ends land, and this tree is behaviourally identical to the contract commit's.
// ================================================================================
// w1: the framebuffer record, per bound target
// ================================================================================
void MGPipeApplySetFramebufferState(const MGPFramebufferState& state) {
// GL HAS TWO INDEPENDENT FRAMEBUFFER BINDINGS AND THIS RECORD DESCRIBES ONE FRAMEBUFFER,
// so Target says which binding - if any - the description also BINDS it to. Both is one
// object bound to both and moves both handles from one call; Named
// (MGPipeFramebufferTarget::Named) moves neither, which is how the DSA entry points
// describe a framebuffer they are about to blit into or clear without claiming it is
// bound (ID-19). A value above Named is not a target this server has, and guessing one
// would put a draw's attachments into the read binding or the other way round.
const char* fault = nullptr;
if (state.Target >= static_cast<Uint8>(MGPipeFramebufferTarget::Count)) {
fault = "the record names no framebuffer binding target";
}
// THE RECORD IS ADDRESSED BY THE FRAMEBUFFER HANDLE NOW, so the handle has to be one.
// The null handle is refused rather than stored: {0,0} is what "nothing is bound to this
// binding" reads as on the other side of BoundFramebuffer, and a record installed there
// would be answered to every caller asking about an EMPTY binding. Every emitter has a
// handle for every framebuffer it describes - kMGPipeDefaultFramebuffer {0,1} for the
// default one, a minted {slot, gen} otherwise - so a null here is a malformed record.
else if (MGPipeHandleIsNull(state.Fbo)) {
fault = "the record names the null framebuffer handle";
}
// And the same bound gate the five object tables carry, for the same reason: Slot is a
// client-supplied Uint32 arriving in a payload and this is the one number in the call
// that reaches an allocator.
else if (state.Fbo.Slot >= kMGPipeMaxFramebufferSlots) {
fault = "the framebuffer slot is outside the record table's bound";
}
// THE DRAW-BUFFER ARRAY IS AN INDEX INTO THIS RECORD'S OWN Color[], and -1 is NONE. An
// entry outside that range would have the server read a colour attachment the record
// does not carry: the wire array is 8 wide, the driver's MaxColorAttachments is not
// clamped to it, and a truncated record arriving here as a plausible index is exactly
// what the framebuffer subsystem's cap refusal exists to prevent upstream.
for (Uint32 i = 0; fault == nullptr && i < kMGPipeMaxColorAttachments; ++i) {
if (state.DrawBuffers[i] < -1 ||
state.DrawBuffers[i] >= static_cast<Int8>(kMGPipeMaxColorAttachments)) {
fault = "a draw-buffer entry names a colour attachment outside the record's own array";
}
}
if (fault != nullptr) {
MGP_TRIP_WIRE_REPORT("MGPipe: " MGP_TRIP_WIRE_TAG("ProtocolCorruption")
" set_framebuffer_state {slot=%u, gen=%u, target=%u}: %s (the record "
"carries %u colour attachments)",
state.Fbo.Slot, state.Fbo.Gen, state.Target, fault,
kMGPipeMaxColorAttachments);
return;
}
// NO HANDLE IS RESOLVED HERE AND NONE MAY BE. A framebuffer has a handle but no wire
// lifetime - the catalogue has no framebuffer create and no framebuffer destroy,
// because a framebuffer is state and this call is the only one that names one - so
// there is nothing to look up, this entry point never counts a refusal, and the record
// it writes OVERWRITES whatever the slot held. That last part is the point rather than
// a compromise: the slot's previous occupant is dead by construction (the client frees
// a framebuffer's slot when the object dies and the allocator bumps the generation on
// reuse), and the record arriving now describes whatever holds the slot today.
//
// The surfaces' Res handles are not resolved either: an attachment PINS its texture,
// and in monolith the frontend's own SharedPtr is that keep-alive, so a refusal here
// would be enforcing a lifetime rule monolith cannot need and split has not defined.
// P4a's belt, after the record's own shape checks and before the table is touched. This
// entry point is the ONE the client can reach without passing PipeFill's gate -
// GL_Framebuffer.cpp's PipePublishFramebufferByName calls the emitter directly at the
// fifteen DSA sites - so on a backend with no consumer this is where those records
// stop. RefusedNoConsumer and not RefusedObjectCalls: the framebuffer family's counter
// contract (see the header) is that no framebuffer call ever moves that one.
if (NoP4aConsumer()) return;
MGPipeFramebufferRecord* record =
RecordAt(MGPipeApplier().FramebufferRecords, state.Fbo.Slot, kMGPipeMaxFramebufferSlots);
// Unreachable: the bound was checked above, before anything moved. The null check is
// what makes that an argument rather than an assumption.
if (record == nullptr) return;
record->Gen = state.Fbo.Gen;
record->Live = true;
record->State = state;
// AND ONLY THEN THE BINDINGS. Named moves neither, which is the whole of the fourth
// target: "this record describes the framebuffer it names; no binding changes".
if (state.Target == static_cast<Uint8>(MGPipeFramebufferTarget::Draw) ||
state.Target == static_cast<Uint8>(MGPipeFramebufferTarget::Both)) {
MGPipeApplier().BoundFramebuffer[static_cast<SizeT>(MGPipeFramebufferTarget::Draw)] = state.Fbo;
}
if (state.Target == static_cast<Uint8>(MGPipeFramebufferTarget::Read) ||
state.Target == static_cast<Uint8>(MGPipeFramebufferTarget::Both)) {
MGPipeApplier().BoundFramebuffer[static_cast<SizeT>(MGPipeFramebufferTarget::Read)] = state.Fbo;
}
// ONE SERIAL FOR THE FAMILY, and it moves once per applied record - a Both record is one
// record and a Named record is one record. It is what retires the four g_fboSynced*
// arrays and the twin's own {slot version, object version, backend id generation}
// quadruple: a compare that used to ask "is my memo still the frontend's" asks "is my
// serial still the server's". A Named record moves it for the same reason a bound one
// does - a twin that memoised a framebuffer's attachments has to hear that they moved,
// and whether the framebuffer happens to be bound is not that question.
++MGPipeApplier().FramebufferSerial;
}
// THE READ SIDE OF THE TABLE (ID-19(b)/(d)). Package D asks it by handle at every DSA entry
// point; the two accessors under it ask it through a binding, which is the question package
// E's SyncCurrentFBOByRecord has always asked.
const MGPFramebufferState* MGPipeApplierState::FramebufferRecordFor(MGPipeHandle fbo) const {
// "Nothing is bound" and "no record" are BOTH null and only the second of them is worth
// a word: the null handle is what a make-current leaves in BoundFramebuffer and what an
// unbound target reads as, it arrives here on every draw of a context that has not
// described its framebuffers yet, and logging it would be a storm about the ordinary
// transitional state. It is separated first so the generation test below can be loud.
if (MGPipeHandleIsNull(fbo)) return nullptr;
if (fbo.Slot >= FramebufferRecords.size()) return nullptr;
const MGPipeFramebufferRecord& record = FramebufferRecords[fbo.Slot];
// Nothing has been written at this slot. Also not an error, and also silent: it is what
// every framebuffer looks like before its first set_framebuffer_state, and the caller's
// answer to it is to run its pre-handle arm.
if (!record.Live) return nullptr;
if (record.Gen != fbo.Gen) {
// AND THIS ONE IS. A record exists at the slot and it describes a DIFFERENT object:
// either the caller is holding a handle whose framebuffer died and whose slot has
// been recycled, or an emitter minted the successor without ever describing it. On
// the handle arm the visible consequence is a DSA blit or clear into a driver
// framebuffer that never got its attachments, which is precisely the failure
// ID-19 exists to close - so this is the loud refusal P3a's lesson demands and
// never a quiet fall-back.
++StaleFramebufferRecordLookups;
MGLOG_E_ONCE("MGPipe: the framebuffer record at slot %u is generation %u and the "
"lookup named generation %u; refusing it rather than answering with "
"another framebuffer's attachments",
fbo.Slot, record.Gen, fbo.Gen);
return nullptr;
}
return &record.State;
}
const MGPFramebufferState* MGPipeApplierState::DrawFramebuffer() const {
return FramebufferRecordFor(BoundFramebuffer[static_cast<SizeT>(MGPipeFramebufferTarget::Draw)]);
}
const MGPFramebufferState* MGPipeApplierState::ReadFramebuffer() const {
return FramebufferRecordFor(BoundFramebuffer[static_cast<SizeT>(MGPipeFramebufferTarget::Read)]);
}
// ================================================================================
// w2: sampler CSOs, sampler views and the three unit sets
// ================================================================================
void MGPipeApplyCreateSamplerState(const MGPSamplerDesc& desc, const SamplerParameters* parameters) {
MOBILEGL_ASSERT(desc.Cso.Slot >= kMGPipeFirstAllocatableSlot,
"create_sampler_state named the reserved slot 0");
if (desc.Cso.Slot < kMGPipeFirstAllocatableSlot) return;
// THE ONE BLOB RULE, on the family's own blob: a non-zero Parameters.Size must be
// exactly one SamplerParameters, and a zero means "this record does not declare its
// blob" - which is what a monolith emission is, because the value rides beside the
// record through the companion pointer. Either way the bytes read are bounded by the
// TYPE and not by the declared length, so the length is a cross-check and never the
// safety property.
const char* fault = nullptr;
if (desc.Parameters.Size != 0 && desc.Parameters.Size != sizeof(SamplerParameters)) {
fault = "the declared blob length is not one SamplerParameters";
} else if (parameters == nullptr) {
fault = "the record declares no parameters and carries none";
}
if (fault != nullptr) {
MGP_TRIP_WIRE_REPORT("MGPipe: " MGP_TRIP_WIRE_TAG("ProtocolCorruption")
" create_sampler_state {slot=%u, gen=%u}: %s (one SamplerParameters is "
"%llu bytes, the blob declares %llu)",
desc.Cso.Slot, desc.Cso.Gen, fault,
static_cast<unsigned long long>(sizeof(SamplerParameters)),
static_cast<unsigned long long>(desc.Parameters.Size));
return;
}
MGPipeSamplerCsoRecord* recordAt =
RecordAt(MGPipeApplier().SamplerCsos, desc.Cso.Slot, kMGPipeMaxSamplerCsoSlots);
if (recordAt == nullptr) {
MGP_TRIP_WIRE_REPORT("MGPipe: " MGP_TRIP_WIRE_TAG("ProtocolCorruption")
" create_sampler_state {slot=%u, gen=%u}: the slot is outside the record "
"table's bound (%u)",
desc.Cso.Slot, desc.Cso.Gen, kMGPipeMaxSamplerCsoSlots);
return;
}
// P4a's belt, after the blob rule and the slot bound and before the record moves.
if (NoP4aConsumer()) return;
MGPipeSamplerCsoRecord& record = *recordAt;
// A CREATE STARTS THE RECORD OVER AND LEAVES Serial AT 0; A RE-ISSUE ON A LIVE IDENTITY
// COUNTS UP. The first half is what stops a recycled slot contributing one field of its
// predecessor, and it is why a fresh backend twin starting its own synced serial at 0
// agrees with a fresh record without either side publishing anything. The second is how
// a value change travels on a handle whose Gen moves only on slot reuse - and it is a
// mutation, so it moves the serial.
if (!record.Live || record.Gen != desc.Cso.Gen) {
record = MGPipeSamplerCsoRecord{};
record.Gen = desc.Cso.Gen;
} else {
++record.Serial;
}
record.Live = true;
// BY VALUE, INCLUDING borderColorForm. All three border representations are always
// numerically populated, so the value alone cannot say which driver entry point to use,
// and the backend's redundancy filter compares all four. The three trailing padding
// bytes are why MOBILEGL_PIPE_VERIFY compares this FIELD BY FIELD through
// PipeFields.def's MGP_FIELDS_SamplerParameters rather than as bytes.
record.Params = *parameters;
}
void MGPipeApplyDeleteSamplerState(const MGPHandleOnly& handle) {
MOBILEGL_ASSERT(handle.Kind == static_cast<Uint32>(MGPipeKind::SamplerCso),
"delete_sampler_state on kind %u", handle.Kind);
MGPipeSamplerCsoRecord* record =
ResolveObject(MGPipeApplier().SamplerCsos, "delete_sampler_state", handle.Handle);
if (record == nullptr) return;
// Dropped whole, generation kept, for resource_destroy's reason: the client allocator
// owns the Gen bump and takes it on the next handout of the slot, so a server-side bump
// here would put the two identities out of step.
//
// AND THE UNIT SET IS NOT SWEPT. bind_sampler_states is "the last set as received" and
// the client re-emits the whole window from its own resolved set at the next verb, so
// walking 192 entries here to blank a handle the next set is about to overwrite would
// buy nothing and would break the one rule the window has.
const Uint32 gen = record->Gen;
*record = MGPipeSamplerCsoRecord{};
record->Gen = gen;
}
void MGPipeApplyCreateSamplerView(const MGPSamplerView& view) {
MOBILEGL_ASSERT(view.Cso.Slot >= kMGPipeFirstAllocatableSlot,
"create_sampler_view named the reserved slot 0");
if (view.Cso.Slot < kMGPipeFirstAllocatableSlot) return;
MGPipeSamplerViewRecord* recordAt =
RecordAt(MGPipeApplier().SamplerViewCsos, view.Cso.Slot, kMGPipeMaxSamplerViewSlots);
if (recordAt == nullptr) {
MGP_TRIP_WIRE_REPORT("MGPipe: " MGP_TRIP_WIRE_TAG("ProtocolCorruption")
" create_sampler_view {slot=%u, gen=%u}: the slot is outside the record "
"table's bound (%u)",
view.Cso.Slot, view.Cso.Gen, kMGPipeMaxSamplerViewSlots);
return;
}
// P4a's belt, after the slot bound and before the record moves.
if (NoP4aConsumer()) return;
MGPipeSamplerViewRecord& record = *recordAt;
// RE-ISSUING ON THE SAME HANDLE IS HOW A RESTRICTION CHANGE TRAVELS - a view is
// identity-addressed one per texture object, minted off that object's lifetime id, and
// Gen moves only on slot reuse - so it bumps the serial and starts nothing over. AND IT
// DOES NOT REBIND: a re-issue on a view some unit is holding changes what that unit
// sees, which the serial already says; it must not make some other unit see it.
if (!record.Live || record.Gen != view.Cso.Gen) {
record = MGPipeSamplerViewRecord{};
record.Gen = view.Cso.Gen;
} else {
++record.Serial;
}
record.Live = true;
record.View = view;
// THE TEXTURE'S OWN ViewCso IS WRITTEN HERE, and it is a SILENT lookup rather than a
// resolved one: the texture-resource bit and the sampler bit are independent, so a view
// arriving before - or without - the texture record is an ordering fact and not a
// refusal. When the record is there this is the back-pointer that lets a sync reach a
// texture's view without walking every view the applier holds.
MGPipeResourceRecord* texture = FindIn(MGPipeApplier().TextureResources, view.Texture);
if (texture != nullptr) texture->ViewCso = view.Cso;
}
void MGPipeApplyDeleteSamplerView(const MGPHandleOnly& handle) {
MOBILEGL_ASSERT(handle.Kind == static_cast<Uint32>(MGPipeKind::SamplerViewCso),
"delete_sampler_view on kind %u", handle.Kind);
MGPipeSamplerViewRecord* record =
ResolveObject(MGPipeApplier().SamplerViewCsos, "delete_sampler_view", handle.Handle);
if (record == nullptr) return;
// The back-pointer goes first, while the record that names the texture still exists,
// and only if that texture still names THIS view - a texture whose view has already
// been re-minted must not have the new handle cleared out from under it.
MGPipeResourceRecord* texture = FindIn(MGPipeApplier().TextureResources, record->View.Texture);
if (texture != nullptr && texture->ViewCso == handle.Handle) {
texture->ViewCso = kMGPipeNullHandle;
}
const Uint32 gen = record->Gen;
*record = MGPipeSamplerViewRecord{};
record->Gen = gen;
}
Bool MGPipeApplySetTextureParams(const MGPTextureParams& params) {
// P4a's belt, and FIRST here because this call's first act is a resolution: with no
// consumer no texture create was accepted, so resolving would report the absence as
// RefusedObjectCalls - the counter that means a seam defect - for the designed state.
if (NoP4aConsumer()) return false;
// ADDRESSED BY RESOURCE AND BY NOTHING ELSE, which is the whole point of the call: a
// texture that is only an FBO attachment, only an image-unit binding or only a
// glCopyImageSubData endpoint has no sampler view to hang its parameters on, and the
// READ-attachment case reaches no parameter push at all today. The record exists the
// moment the parameters move, whether or not anything is bound.
MGPipeResourceRecord* record =
ResolveObject(MGPipeApplier().TextureResources, "set_texture_params", params.Res);
if (record == nullptr) return false;
// EVERY ITextureObject OWNS A SamplerObject, so the built-in sampler CSO is not
// optional and a null handle is not "no sampler" - it is a record that would have the
// backend sample a texture with whatever filter and wrap state the unit last left
// behind. GL 4.6 table 23.18 makes filter/wrap/compare/border sampler state and Espryt
// pushes it onto the TEXTURE with glTexParameter*, so this handle is the only thing
// that says which values those are.
if (MGPipeHandleIsNull(params.BuiltinSampler)) {
MGP_TRIP_WIRE_REPORT("MGPipe: " MGP_TRIP_WIRE_TAG("ProtocolCorruption")
" set_texture_params {slot=%u, gen=%u, glName=%u}: the record names no "
"built-in sampler CSO, and every texture object owns one",
params.Res.Slot, params.Res.Gen, record->Desc.GlNameForDiag);
return false;
}
// AND THE CSO IT NAMES IS NOT RESOLVED. The sampler subsystem is its own bit and may be
// clear while the texture bit is set, so a record that names a CSO this applier has not
// been told about is an ORDERING fact rather than a corrupt one; refusing it would make
// one legal A/B arm drop every texture's parameters. What the record carries is the
// identity, and the identity is what the backend resolves at its own sync point.
record->Params = params;
// The serial moves BEFORE anything downstream is told, for ApplyBufferWrite's reason.
// It plus the record's own ForceResync / SamplerResync bytes are what retire the twin's
// m_syncedTextureParamsVersion + m_forceTextureParamsResync pair - and the two resync
// bytes are CARRIED, never cleared here: the server ORs them into its own flags and
// clears its own copy, and the client never clears a server flag.
++record->ParamsSerial;
// P5c (rv): a parameter change is what the sampling-resolution shutter guarded.
MGPipeApplierNoteTextureStateMoved();
return true;
}
// The three of them, and NO STAGE DIMENSION on any of them: MobileGL's texture-unit space
// is one merged array of 192, the same unit may be sampled from two stages, and stage is
// derived server-side from the reflection archive only where the target API needs it.
//
// A NULL HANDLE IN A TAIL ENTRY IS LEGAL EVERYWHERE HERE and is not a refusal: a unit the
// program does not resolve carries a null view, a unit with no sampler object carries a
// null sampler CSO (the texture's built-in sampler applies then, exactly as today), and a
// unit with no texture carries a null resource. None of the three is resolved against a
// record either - a set is WORKING STATE, the records it names are OBJECT state, and the
// backend resolves the pair at its own sync point where both are current.
//
// P4a's belt sits IN FRONT of ApplyUnitWindow on all three, and this is the one place it
// is not last: ApplyUnitWindow validates the window and writes it in the same step, so
// there is no point between the two to stand at. The cost is that on a backend with no
// consumer a malformed window is declined rather than Fatal - which is the right trade the
// one way round it can be made, because that backend would never have been handed the
// window at all (the client's gate stops it) and the shipped configuration that DOES
// consume these still trips the wire.
void MGPipeApplySetSamplerViews(const MGPSamplerViews& hdr, const MGPBoundView* tail) {
if (NoP4aConsumer()) return;
if (!ApplyUnitWindow("set_sampler_views", hdr.Start, hdr.Count, hdr.ContentHash, tail,
MGPipeApplier().BoundSamplerViews, MGPipeApplier().SamplerViewStart,
MGPipeApplier().SamplerViewCount)) {
return;
}
++MGPipeApplier().SamplerViewsSerial;
// P5c (rv): a texture bind is what the bind-generation shutter guarded.
MGPipeApplierNoteTextureStateMoved();
}
void MGPipeApplyBindSamplerStates(const MGPSamplerStates& hdr, const MGPipeHandle* tail) {
if (NoP4aConsumer()) return;
if (!ApplyUnitWindow("bind_sampler_states", hdr.Start, hdr.Count, hdr.ContentHash, tail,
MGPipeApplier().BoundSamplerStates, MGPipeApplier().SamplerStateStart,
MGPipeApplier().SamplerStateCount)) {
return;
}
++MGPipeApplier().SamplerStatesSerial;
// P5c (rv): a sampler bind or parameter change is what the shutters guarded.
MGPipeApplierNoteTextureStateMoved();
}
void MGPipeApplySetShaderImages(const MGPShaderImages& hdr, const MGPImageView* tail) {
// The image set carries InternalFormat and Access per entry and BOTH are live
// glBindImageTexture state the format-less image bake keys on, so they are stored as
// sent and recast on the server - the record carries the application's format, and the
// bind-format recast that turns a GL_RG32F bind into something 19 of 26 non-core
// formats on Adreno will accept is the backend's, not this applier's.
if (NoP4aConsumer()) return;
if (!ApplyUnitWindow("set_shader_images", hdr.Start, hdr.Count, hdr.ContentHash, tail,
MGPipeApplier().BoundShaderImages, MGPipeApplier().ShaderImageStart,
MGPipeApplier().ShaderImageCount)) {
return;
}
++MGPipeApplier().ShaderImagesSerial;
// P5c (rv): an image bind moves the texture bind generation's guarded set too.
MGPipeApplierNoteTextureStateMoved();
}
// ---------------------------------------------------------------------------------
// P5e's two entry points, DECLARED BY c0e (MG_Remote/CONTRACT-P5E.md §1)
// ---------------------------------------------------------------------------------
//
// set_shader_buffers HAS ITS BODY (package sb); set_program_bindings still refuses by name
// until package pg writes its own. A record that reached an unbodied one would be a binding
// set the server accepted and dropped, so the answer there is a named abort rather than a
// quiet return: the picture would be wrong and the lane would be green, which is the one
// failure mode this campaign spends its refusals on.
// set_shader_buffers, ONE RECORD PER CLASS (CONTRACT-P5E.md §5.6, rulings 10/11). The whole
// body is "validate the window, copy the tail into that class's array, take the mask, move
// the serial" - the binding points are WORKING STATE and the objects they name are resolved
// by the backend at its own sync point, from the handle each range carries.
//
// IT DOES NOT GO THROUGH ApplyUnitWindow, and the reason is worth a line rather than a
// diff: that helper's refusal message names "the merged texture-unit space" and the three
// unit sets share one capacity, while this row has THREE windows keyed by Class and two
// refusals of its own that the contract names by string ("SetShaderBuffers.Class" /
// ".Count"). A shared helper that had to grow a class parameter and a message parameter
// would be the same code with more ways to pass the wrong one.
//
// AND THERE IS NO NoP4aConsumer BELT HERE, deliberately. P4a's belt exists because a client
// clears a per-level dirty flag on ACCEPTANCE, so accepting on a backend that consumes
// nothing loses uploads (ID-39). This family's liveness gate is the client's own
// P5eFamilyIsLive (PipeFill.cpp), which asks R-8 for bit 13 in particular - a server that
// does not publish bit 13 never receives one of these records at all - and storing a window
// no backend reads costs nothing and loses nothing, because the emitter's only latch is the
// set-hash suppressor and the record it latched did cross.
void MGPipeApplySetShaderBuffers(const MGPShaderBuffers& hdr, const MGPBufferRange* tail) {
if (hdr.Class >= kMGPipeShaderBufferClassCount) {
MGP_TRIP_WIRE_REPORT("MGPipe: " MGP_TRIP_WIRE_TAG("ProtocolCorruption")
" SetShaderBuffers.Class - class %u names no binding-point "
"array; the three are Uniform=0, ShaderStorage=1, "
"AtomicCounter=2",
hdr.Class);
return;
}
const Uint64 end = Uint64{hdr.Start} + Uint64{hdr.Count};
if (end > kMGPipeMaxBufferBindingPoints || (hdr.Count != 0 && tail == nullptr)) {
MGP_TRIP_WIRE_REPORT("MGPipe: " MGP_TRIP_WIRE_TAG("ProtocolCorruption")
" SetShaderBuffers.Count {class=%u, start=%u, count=%u}: the "
"window runs past the %u binding points the wire carries, or a "
"non-empty set carries no ranges",
hdr.Class, hdr.Start, hdr.Count,
static_cast<unsigned>(kMGPipeMaxBufferBindingPoints));
return;
}
// THE WINDOW IS THE BOUND AND ENTRIES OUTSIDE IT ARE NOT CLEARED, exactly as the three
// unit sets and set_vertex_buffers work: the record is "the last set as received", and
// a binding at or above Count means "nothing bound" to every reader (the frontend
// array's default says the same thing, which is what makes the two pictures agree).
auto& destination = MGPipeApplier().BoundShaderBuffers[hdr.Class];
for (Uint32 i = 0; i < hdr.Count; ++i) destination[hdr.Start + i] = tail[i];
MGPipeApplier().ShaderBufferStart[hdr.Class] = hdr.Start;
MGPipeApplier().ShaderBufferCount[hdr.Class] = hdr.Count;
for (Uint32 w = 0; w < kMGPipeShaderBufferWritableMaskWords; ++w) {
MGPipeApplier().ShaderBufferWritableMask[hdr.Class][w] = hdr.WritableMask[w];
}
// ONE SERIAL FOR ALL THREE CLASSES, and it moves on EVERY applied record - a backend
// memo that asks "have the binding points moved since I last bound them" must hear
// about a uniform set exactly as it hears about a storage set, and splitting the serial
// per class would make each of the four consumers pick which one to read.
++MGPipeApplier().ShaderBuffersSerial;
}
// P5e (pg) WRITES THE SECOND BODY. The three post-link mutable reflection fields -
// glUniformBlockBinding's block-to-point map, glUniform1i's sampler unit per uniform
// LOCATION, and glShaderStorageBlockBinding's name-keyed override set - all live INSIDE
// the frontend's LinkArtifacts and all move after the link that produced it. An archive
// snapshot alone therefore answers the wrong question: a twin built from it would bind the
// uniform blocks the program was LINKED with rather than the ones it is BOUND with.
void MGPipeApplySetProgramBindings(const MGPProgramBindings& hdr, const Int32* blockBindings,
const MGPProgramSamplerUnit* samplerUnits,
const MGPProgramStorageOverride* storageOverrides,
const char* const* storageOverrideNames) {
// THE BOUNDS ARE THE DECODER'S AND ARE RE-ASSERTED HERE, because the monolith adapter
// reaches this entry point without passing the decoder at all: under monolith the only
// validation a record ever gets is the one the applier does itself, which is the rule
// every sibling in this file already keeps.
const char* fault = nullptr;
if (hdr.BlockBindingCount > kMGPipeMaxProgramBlockBindings) fault = "BlockBindingCount";
else if (hdr.SamplerUnitCount > kMGPipeMaxProgramSamplerUnits) fault = "SamplerUnitCount";
else if (hdr.StorageOverrideCount > kMGPipeMaxProgramStorageOverrides) fault = "StorageOverrideCount";
else if (hdr.BlockBindingCount != 0 && blockBindings == nullptr) fault = "BlockBindings";
else if (hdr.SamplerUnitCount != 0 && samplerUnits == nullptr) fault = "SamplerUnits";
else if (hdr.StorageOverrideCount != 0 &&
(storageOverrides == nullptr || storageOverrideNames == nullptr)) {
fault = "StorageOverrides";
}
if (fault != nullptr) {
MGP_TRIP_WIRE_REPORT("MGPipe: " MGP_TRIP_WIRE_TAG("ProtocolCorruption")
" set_program_bindings {slot=%u, gen=%u}: %s is out of range or "
"its tail is absent (%u/%u/%u declared)",
hdr.Cso.Slot, hdr.Cso.Gen, fault, hdr.BlockBindingCount,
hdr.SamplerUnitCount, hdr.StorageOverrideCount);
return;
}
// P4a's belt, for MGPipeApplyBindShaderState's reason: with no consumer this applier
// holds no shader CSO record at all, so the resolution below would report the designed
// state as RefusedObjectCalls.
if (NoP4aConsumer()) return;
MGPipeShaderCsoRecord* record = ResolveShaderCso("set_program_bindings", hdr.Cso);
if (record == nullptr) return;
// WHOLE-SET REPLACEMENT, NOT A MERGE, in all three tails. Each one describes the
// program's complete state for its index space as the client observed it at emit; a
// merge would leave a block binding the application has since reset to its declared
// default sitting on top of the archive's value for ever, and there is no per-entry
// "cleared" token on the wire that could say otherwise.
record->BlockBindings.assign(blockBindings, blockBindings + hdr.BlockBindingCount);
record->SamplerUnits.assign(samplerUnits, samplerUnits + hdr.SamplerUnitCount);
record->StorageOverrides.clear();
record->StorageOverrides.reserve(hdr.StorageOverrideCount);
for (Uint32 i = 0; i < hdr.StorageOverrideCount; ++i) {
MGPipeProgramStorageOverride entry;
// COPIED, NOT POINTED AT (rule C). The name rides SEG_STAGE and the staged run
// retires with the record that named it; the rebuild that reads this may be a
// frame later, and by then those bytes are somebody else's.
entry.Name = storageOverrideNames[i] != nullptr ? String(storageOverrideNames[i]) : String();
entry.Binding = storageOverrides[i].Binding;
record->StorageOverrides.push_back(Move(entry));
}
record->Signature = hdr.Signature;
// ADVANCED, NEVER RETURNED TO A VALUE IT HAS HANDED OUT - VertexBuffersSerial's rule.
// Both texture-unit memos and the program twin's clean condition key on it (ruling 6),
// so a serial that went backwards would make a memo match state it has never seen.
++record->BindingsSerial;
}
// ---------------------------------------------------------------------------------
// P5e: the barriered predicate and the applier's current-record flag (§2.1, §4.4)
// ---------------------------------------------------------------------------------
Bool MGPipeBarriered(MGPWireOp op, const void* payload, const MGPipeApplierState& st) {
// 1. The static column. Every reply row, present, and the rows whose apply still pulls
// a BARRIER_PULLED field or probes the client allocator.
if (MGPipeWaitClassFor(op) != kWaitNone) return true;
// 2. Escalation (i): an open transform-feedback span makes every verb inside it
// barriered. XFB is not migrated by P5e - StartPendingTransformFeedback holds
// frontend SharedPtrs across two verbs, which rule C forbids an applier entry point
// outright - so the span keeps P5C's semantics and the cost on the steady Espryt
// path is one bool test per draw.
if (MGPipeCallClassFor(op) == kCtxVerb && st.IsTransformFeedbackActive) return true;
// 3. Escalation (ii): a draw whose enabled attributes include a client-memory array.
// The server uploads those bytes by dereferencing a raw client pointer, so the draw
// cannot run ahead. Under run-ahead the client refuses it before it is published
// (§5.1), so this clause is what the LOCKSTEP arm - where the draw is legal - uses.
if (op == MGPWireOp::DrawVbo && payload != nullptr) {
const auto& draw = *static_cast<const MGPDrawInfo*>(payload);
if ((draw.Flags & static_cast<Uint8>(kDrawClientArrays)) != 0) return true;
// ---- ESCALATION (iii) WAS HERE AND IS WITHDRAWN (P5e ra2, ID-133 then ID-136) ---
//
// ID-133 escalated a PLAIN multi-draw (`NumDraws > 1 && !kDrawIsIndirect`) so that
// MultiDrawImpl::RunIndirect's read of the client's GL_DRAW_INDIRECT_BUFFER binding
// became a legal barriered pull instead of an unbarriered Fatal. It worked, and it
// cost a rendezvous on EVERY plain glMultiDraw* on the DEFAULT tier: there is one
// draw opcode - all twenty entry points collapse onto draw_vbo - and the tier is
// chosen on the server per batch (MultiDraw.cpp's ResolveTierForBatch), so no
// predicate both roles can compute names the arm that actually needed it.
//
// THE CHECK ON AN ESCALATION IS "WHICH ARM PAYS FOR IT", NOT "WHICH LANE GOES
// GREEN", and that is the rule this pair of rulings exists to record. The read was
// a save/restore of a GL binding NAME around the tier's own scratch buffer, not a
// data dependency, so the answer was to ask the side that did the binding:
// BoundDrawIndirectBufferId now takes the handle arm its neighbour
// ResolveBoundIndexBuffer already had, and the pull is GONE rather than legalised.
//
// Both halves landed in ONE commit on purpose: the pull retired without this clause
// withdrawn is a cost with no reason, and this clause withdrawn without the pull
// retired puts 18 lane entries back on the unbarriered arm. Neither is a state to
// gate or to measure, so neither was ever a head.
}
return false;
}
// Server-private and deliberately NOT a member of MGPipeApplierState: it is not per-context
// state the applier owns, it is "which record is the apply thread inside right now", and a
// reader (the allocator guard) that had to find the right context's applier first would be
// asking a harder question than the one it needs answered.
//
// TRUE IS THE DEFAULT AND THE LOCKSTEP ANSWER. On a server that does not publish
// kCapRunAheadApply every record is barriered, the client is parked in its own wait for
// each of them, and the two named exemption scopes of CONTRACT-P5C §3.1 keep working
// exactly as they do today. Nothing writes this until package id sets it from
// MGPipeBarriered in the sink's ApplyOne.
namespace {
thread_local Bool g_currentRecordBarriered = true;
// P5e (gl), ID-128. FALSE IS THE DEFAULT AND IT IS THE STRICT ANSWER: "this record was
// barriered by escalation" admits a pull that neither the wait class nor the retiring
// phase admits, so a reader that runs before any writer must say no. The stamp above
// defaults the other way because ITS safe answer is "the client is parked".
thread_local Bool g_currentRecordBarrieredByEscalation = false;
} // namespace
Bool MGPipeApplierCurrentRecordIsBarriered() { return g_currentRecordBarriered; }
void MGPipeApplierSetCurrentRecordBarriered(Bool barriered) {
g_currentRecordBarriered = barriered;
}
Bool MGPipeApplierCurrentRecordIsBarrieredByEscalation() {
return g_currentRecordBarrieredByEscalation;
}
void MGPipeApplierSetCurrentRecordBarrieredByEscalation(Bool escalated) {
g_currentRecordBarrieredByEscalation = escalated;
}
// ================================================================================
// w3: the shader CSO, the two program bindings and the default uniform block
//
// AND THE SERVER NEVER RE-LINKS. glslang lives entirely on the client and SPIRV-Cross
// entirely on the server, a file-level cut: what crosses is the per-stage SPIR-V and the
// reflection archive, never source, and there is no server-side compile pool to hand a
// failure back from. Link failure needs no synchronous return either - it is one error log
// plus a bind-program-0 empty draw today, GL_LINK_STATUS is never withdrawn, and the
// synchronous query is answered by the client out of its own ProgramObject.
//
// WHAT THE SERVER STILL SPECIALISES IS NOT THIS RECORD'S BUSINESS. A backend program
// depends on eight more inputs than the artefacts - the draw framebuffer's clamp masks and
// fragColor broadcast count, the storage-block binding signature, the live image formats,
// the patch parameters - and every one of them now arrives through ANOTHER record this
// applier holds. create_shader_state publishes the artefacts; the verb specialises.
// ================================================================================
void MGPipeApplyCreateShaderState(const MGPProgramDesc& desc,
const MG_State::GLState::LinkArtifacts* link,
const MG_State::GLState::SpirvArtifacts* spirv,
SharedPtr<const MG_State::GLState::ProgramArchive> archive) {
MOBILEGL_ASSERT(desc.Cso.Slot >= kMGPipeFirstAllocatableSlot,
"create_shader_state named the reserved slot 0");
if (desc.Cso.Slot < kMGPipeFirstAllocatableSlot) return;
// THE ONE BLOB RULE, over seven blob refs at once. All seven are declared with Size 0 in
// monolith - "this record does not declare its blob" - and the artefacts ride beside the
// record through the two companion pointers, so a record that declares nothing AND
// carries nothing describes a program with no reflection and no SPIR-V, which is not a
// program. The declared lengths themselves cross-check nothing here: no other field of
// MGPProgramDesc describes how long a module or an archive is, so the rule is inert on
// this record by construction rather than by omission, and it becomes a real one when
// the transport that fills those refs in lands with the splitter that owns them.
const char* fault = nullptr;
if (link == nullptr || spirv == nullptr) {
fault = "the record declares no blobs and carries no artefacts";
} else if (desc.GlobalUboSize > kMGPipeMaxGlobalConstantsBytes) {
fault = "the default uniform block is larger than any program may declare";
}
if (fault != nullptr) {
MGP_TRIP_WIRE_REPORT("MGPipe: " MGP_TRIP_WIRE_TAG("ProtocolCorruption")
" create_shader_state {slot=%u, gen=%u}: %s (stages=0x%x, "
"globalUboSize=%u, bound %u)",
desc.Cso.Slot, desc.Cso.Gen, fault, desc.StageMask, desc.GlobalUboSize,
kMGPipeMaxGlobalConstantsBytes);
return;
}
MGPipeShaderCsoRecord* recordAt = ShaderCsoRecordAt(desc.Cso.Slot);
if (recordAt == nullptr) {
MGP_TRIP_WIRE_REPORT("MGPipe: " MGP_TRIP_WIRE_TAG("ProtocolCorruption")
" create_shader_state {slot=%u, gen=%u}: the slot is outside the record "
"table's bound (%u)",
desc.Cso.Slot, desc.Cso.Gen, kMGPipeMaxShaderCsoSlots);
return;
}
#if MOBILEGL_PIPE_VERIFY
// BEFORE STORING, so a lane that aborts here aborts on the record that was wrong rather
// than on the next one that reads it.
PinProgramArchiveRoundTrip(desc, *link, *spirv);
#endif
// P4a's belt, after the blob rule, the slot bound and the verify round trip - all three
// are checks on the RECORD and stay honest on every backend - and before it is stored.
if (NoP4aConsumer()) return;
MGPipeShaderCsoRecord& record = *recordAt;
// A RE-ISSUE ON THE SAME HANDLE IS HOW A RELINK TRAVELS: the handle is minted per
// frontend program and Gen moves only on slot reuse, so an existing record of the same
// identity keeps its serial and counts up from it, and a record of a different identity
// is a recycled slot and starts over.
if (!record.Live || record.Gen != desc.Cso.Gen) {
record = MGPipeShaderCsoRecord{};
record.Gen = desc.Cso.Gen;
} else {
++record.Serial;
// A RELINK REPLACES THE DEFAULT UNIFORM BLOCK'S LAYOUT, so the image and the version
// keyed to the old one go with it - keeping them would leave a block sized to a
// layout that no longer exists, and the sentinel is exactly the value that says
// "nothing has been uploaded for this program". The client re-emits against the
// layout that now exists; the serial announces the clearing so a twin cannot match
// what it uploaded before the relink.
record.GlobalConstants.clear();
record.GlobalConstantsVersion = ~Uint32{0};
++record.GlobalConstantsSerial;
}
record.Live = true;
// The DESCRIPTOR is what is stored, and the artefacts are not: in monolith the backend
// reads the frontend's own archive, and under split the descriptor is what the
// deserialised archive is attached to. Either way the applier owns identity, extent and
// order, and never content.
record.Desc = desc;
// P5e (pg), gap G-A: "under split the descriptor is what the deserialised archive is
// attached to" is now literally true, and this is the attachment. ADOPTED
// UNCONDITIONALLY, INCLUDING A NULL: a monolith create must CLEAR an archive a
// previous split-arm create left behind, because the record outlives a transport
// change (MGPipeApplierReset keeps the CSO table - a program lives in a share group)
// and a stale archive is worse than none.
//
// AFTER Desc, so a reader that sees the new descriptor sees the artefacts it describes;
// the apply thread is the only writer and the only reader, so ordering here is about
// reading this function rather than about memory visibility.
record.Archive = Move(archive);
// THE RE-ISSUE CLEARS THE THREE BINDING TAILS for the reason it clears GlobalConstants
// one branch up: they are indices INTO the archive that has just been replaced, so a
// surviving tail would name block indices and uniform locations of a program that no
// longer exists. Unconditional rather than inside the else-branch above, because a
// recycled slot must not inherit its predecessor's bindings either - and a fresh record
// has empty tails, so the clear costs a live program nothing. The client re-emits
// set_program_bindings immediately after this create (ProgramEmit.h), which is the
// ordering the applier relies on: emitting them BEFORE would put them where this line
// wipes them.
record.BlockBindings.clear();
record.SamplerUnits.clear();
record.StorageOverrides.clear();
record.Signature = 0;
++record.BindingsSerial;
}
void MGPipeApplyBindShaderState(const MGPHandleOnly& handle) {
MOBILEGL_ASSERT(handle.Kind == static_cast<Uint32>(MGPipeKind::ShaderCso),
"bind_shader_state on kind %u", handle.Kind);
// P4a's belt, and first for set_texture_params' reason: with no consumer this applier
// holds no shader CSO record, so the resolution below would report the designed state
// as RefusedObjectCalls. The null-handle unbind is behind it too - a binding this
// applier never made is not one it may clear.
if (NoP4aConsumer()) return;
// The null handle is legal and means "nothing bound", which is a GL state and not an
// error; a DEAD handle leaves the previous binding untouched and is counted, which is
// bind_render_state's precedent for the same question.
if (MGPipeHandleIsNull(handle.Handle)) {
MGPipeApplier().BoundShaderCso = kMGPipeNullHandle;
++MGPipeApplier().ProgramBindingSerial;
return;
}
const MGPipeShaderCsoRecord* record = ResolveShaderCso("bind_shader_state", handle.Handle);
if (record == nullptr) return;
MGPipeApplier().BoundShaderCso = handle.Handle;
++MGPipeApplier().ProgramBindingSerial;
}
void MGPipeApplyDeleteShaderState(const MGPHandleOnly& handle) {
MOBILEGL_ASSERT(handle.Kind == static_cast<Uint32>(MGPipeKind::ShaderCso),
"delete_shader_state on kind %u", handle.Kind);
// A COMPOSITE'S SLOT HAS TWO INDEPENDENT RELEASE PATHS - the pipeline cache's eviction
// and the composite program's own destructor - and both arrive here through one client
// helper. The second is a refusal: the record is already gone, so it is counted and
// dropped, which is the same defined no-op every other death notice gets and is why the
// double free is a proven no-op rather than a race.
MGPipeShaderCsoRecord* record = ResolveShaderCso("delete_shader_state", handle.Handle);
if (record == nullptr) return;
const Uint32 gen = record->Gen;
*record = MGPipeShaderCsoRecord{};
record->Gen = gen;
// UNLIKE THE UNIT SETS, THE THREE PROGRAM BINDINGS ARE CLEARED. They are single handles
// rather than a window of "the last set as received", and each of them is resolved
// against the record that has just been dropped - leaving one behind would mean the
// next verb resolving a binding to a record this applier no longer has, which is the
// refusal counter firing for a state the applier itself created.
Bool cleared = false;
if (MGPipeApplier().BoundShaderCso == handle.Handle) {
MGPipeApplier().BoundShaderCso = kMGPipeNullHandle;
cleared = true;
}
if (MGPipeApplier().DrawProgram == handle.Handle) {
MGPipeApplier().DrawProgram = kMGPipeNullHandle;
cleared = true;
}
if (MGPipeApplier().DispatchProgram == handle.Handle) {
MGPipeApplier().DispatchProgram = kMGPipeNullHandle;
cleared = true;
}
if (cleared) ++MGPipeApplier().ProgramBindingSerial;
}
// TWO CALLS AND NOT ONE, because the frontend has two joins and two PipeInputs slots: a
// draw resolves GetProgramForDraw and a dispatch resolves GetProgramForDispatch, and a
// pipeline object flattened into a composite is the first of those and never the second.
void MGPipeApplySetDrawProgram(const MGPHandleOnly& handle) {
MOBILEGL_ASSERT(handle.Kind == static_cast<Uint32>(MGPipeKind::ShaderCso),
"set_draw_program on kind %u", handle.Kind);
if (NoP4aConsumer()) return; // P4a's belt, for MGPipeApplyBindShaderState's reason.
if (MGPipeHandleIsNull(handle.Handle)) {
MGPipeApplier().DrawProgram = kMGPipeNullHandle;
++MGPipeApplier().ProgramBindingSerial;
return;
}
const MGPipeShaderCsoRecord* record = ResolveShaderCso("set_draw_program", handle.Handle);
if (record == nullptr) return;
MGPipeApplier().DrawProgram = handle.Handle;
++MGPipeApplier().ProgramBindingSerial;
}
void MGPipeApplySetDispatchProgram(const MGPHandleOnly& handle) {
MOBILEGL_ASSERT(handle.Kind == static_cast<Uint32>(MGPipeKind::ShaderCso),
"set_dispatch_program on kind %u", handle.Kind);
if (NoP4aConsumer()) return; // P4a's belt, for MGPipeApplyBindShaderState's reason.
if (MGPipeHandleIsNull(handle.Handle)) {
MGPipeApplier().DispatchProgram = kMGPipeNullHandle;
++MGPipeApplier().ProgramBindingSerial;
return;
}
const MGPipeShaderCsoRecord* record = ResolveShaderCso("set_dispatch_program", handle.Handle);
if (record == nullptr) return;
MGPipeApplier().DispatchProgram = handle.Handle;
++MGPipeApplier().ProgramBindingSerial;
}
void MGPipeApplySetGlobalConstants(const MGPGlobalConstants& record, const void* bytes) {
// ON THE PROGRAM'S RECORD, not in the working state, and that is what makes it survive a
// make-current: the block is (ShaderCso, Version) keyed and belongs to the program, not
// to the context that last uploaded it.
//
// P4a's belt goes in front of the resolution for set_texture_params' reason, and it is
// in front of the fault block too because THIS call's bounds come out of the resolved
// program's own GlobalUboSize - there is nothing to validate against until the record
// is in hand.
if (NoP4aConsumer()) return;
MGPipeShaderCsoRecord* stored = ResolveShaderCso("set_global_constants", record.ShaderCso);
if (stored == nullptr) return;
// THE LENGTH IS THE PROGRAM'S OWN GlobalUboSize, which arrived on the create and was
// bounded there - so this call can only ever allocate what the create already declared,
// and the blob rule has a real field to cross-check against for once.
const Uint32 size = stored->Desc.GlobalUboSize;
const char* fault = nullptr;
if (record.Version == ~Uint32{0}) {
fault = "the version is the backends' never-uploaded sentinel, which no record may carry";
} else if (record.Blob.Size != 0 && record.Blob.Size != size) {
fault = "the declared blob length is not the program's own default uniform block size";
} else if (size > kMGPipeMaxGlobalConstantsBytes) {
fault = "the program's default uniform block is larger than any program may declare";
} else if (size != 0 && bytes == nullptr) {
fault = "a non-empty block carries no bytes";
}
if (fault != nullptr) {
MGP_TRIP_WIRE_REPORT("MGPipe: " MGP_TRIP_WIRE_TAG("ProtocolCorruption")
" set_global_constants {slot=%u, gen=%u}: %s (version=%u, the program "
"declares %u bytes, the blob declares %llu)",
record.ShaderCso.Slot, record.ShaderCso.Gen, fault, record.Version, size,
static_cast<unsigned long long>(record.Blob.Size));
return;
}
if (size == 0) {
stored->GlobalConstants.clear();
} else {
const Uint8* image = static_cast<const Uint8*>(bytes);
stored->GlobalConstants.assign(image, image + size);
}
stored->GlobalConstantsVersion = record.Version;
// Its OWN serial, beside the record's: a block upload is not a relink, and a twin that
// memoises "which block image have I staged" must not be told the program moved.
++stored->GlobalConstantsSerial;
}
// THE MONOLITH BODY IS A NO-OP AND THAT IS THE WHOLE OF IT: the emulation this names still
// runs, exactly as it does today, on the same code path. What the call site buys is that
// the set of emulations a split server cannot serve is NAMED, GREPPABLE and PINNED, so P5
// and P8 give it teeth by editing one function instead of rediscovering five call sites.
//
// It takes a literal and does nothing with it. Not a log line, not a counter: it sits on
// paths a frame can reach many times, and ROADMAP.md forbids committing hot-path
// instrumentation.
// P5 GIVES IT TEETH, and this is the whole of it: ONE function edit arms FIVE call sites
// (Managers.cpp:5334, DirectGLES.cpp:8051, :8702, :8997, :10623), exactly as the header
// and P4a's comment above promised. Every one of them reaches back into the CLIENT's
// address space - a texture shadow, a CPU mipmap fallback, a host-side copy - and a server
// has no client address space to reach into, so degrading silently is the one outcome that
// must not happen.
//
// THE ARM IS THE TRANSPORT, NOT THE BUILD. build-split runs MOBILEGL_TRANSPORT=monolith in
// every unit and integration-gpu lane, and several of the five are on ordinary monolith
// paths that those lanes exercise (glGenerateMipmap reaches two of them); arming on the
// build would turn 1117 integration cases red for running code that is correct in the role
// they run it in. None of the five is on P5's reduced path, so under a real transport this
// costs nothing and catches a lot.
void MGPipeUnmigratedEmulation(const char* name) {
#if MOBILEGL_BUILD_DISAGGREGATED
if (MG_Config::Transport != MG_Config::TransportMode::Monolith) {
MGLOG_F("MGPipe: Fatal{UnmigratedEmulation, \"%s\"}", name != nullptr ? name : "<null>");
std::abort();
}
#endif
(void)name;
}
} // namespace MobileGL::MG_Pipe
Reference in New Issue View Git Blame Copy Permalink