swung0x48/MobileGL
1
0
Fork 0
mirror of https://github.com/MobileGL-Dev/MobileGL synced 2026-09-11 21:58:31 +09:00
Code Issues Packages Projects Releases Wiki Activity

[Test] (MG_Util, MG_State): remove the TEMP stage-timing probes - the async-compile measurement campaign is done

Browse Source
This commit is contained in:
swung0x48
2026-08-10 09:56:05 -04:00
parent c3412ca394
commit d868e1c476
6 changed files with 8 additions and 228 deletions
Download Patch File Download Diff File Expand all files Collapse all files
MobileGL/MG_Util/Debug/TempStageProbe.h
-121
View File
@@ -1,121 +0,0 @@
// MobileGL - MobileGL/MG_Util/Debug/TempStageProbe.h
// 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
//
// TEMP-STAGE-PROBE: TEMPORARY instrumentation. Measures the per-stage wall time of the
// shader frontend (preprocess / glslang parse / glslang link+mapIO / SPIR-V generation /
// spirv-tools optimize / reflection / SPIRV-Cross global-UBO routing) so a shaderpack load
// can be attributed stage by stage. Delete this file together with every call site marked
// TEMP-STAGE-PROBE when the measurement is finished. Every identifier is spelled
// tempStageProbe* / TempStageProbe* / kTempStageProbe* so both spellings are greppable.
#pragma once
#include <MG_Util/Debug/Log.h>
#include <atomic>
#include <chrono>
namespace MobileGL::MG_Util::Debug {
// TEMP-STAGE-PROBE: stage ids. The two *Total entries are SUPERSETS of the stages above
// them (they time the whole task body), kept so the residue inside a task body is
// visible; kTempStageProbeParseReparse is a SUBSET of kTempStageProbeParse (the link's
// consume-once re-parse goes through the same ShaderCompiler::CompileShader).
enum TempStageProbeStageId : int {
kTempStageProbePreprocess = 0,
kTempStageProbeParse,
kTempStageProbeParseReparse,
kTempStageProbeGlslangLink,
kTempStageProbeSpirvGen,
kTempStageProbeSpirvNull,
kTempStageProbeSpirvOpt,
kTempStageProbeReflection,
kTempStageProbeSpvcRouting,
kTempStageProbeCompileTaskTotal,
kTempStageProbeLinkTaskTotal,
kTempStageProbeSpirvTaskTotal,
kTempStageProbeStageCount
};
inline const char* const kTempStageProbeStageNames[kTempStageProbeStageCount] = {
"preprocess",
"parse",
"parse-reparse[subset-of-parse]",
"glslang-link",
"spirv-gen",
"spirv-null[plumbing-only]",
"spirv-opt",
"reflection",
"spvc-routing",
"compiletask-total[superset]",
"linktask-total[superset]",
"spirvtask-total[superset]",
};
inline std::atomic<unsigned long long> tempStageProbeMicros[kTempStageProbeStageCount] = {};
inline std::atomic<unsigned long long> tempStageProbeCalls[kTempStageProbeStageCount] = {};
inline std::atomic<unsigned long long> tempStageProbeLinkCount{0};
inline std::atomic<long long> tempStageProbeLastDumpNanos{0};
// TEMP-STAGE-PROBE: scoped accumulator. Relaxed atomics only - nothing here orders any
// other memory, and the probe must not perturb what it measures.
class TempStageProbeScope {
public:
explicit TempStageProbeScope(int tempStageProbeStageId)
: tempStageProbeStage(tempStageProbeStageId),
tempStageProbeStart(std::chrono::steady_clock::now()) {}
~TempStageProbeScope() {
const auto tempStageProbeElapsedUs = std::chrono::duration_cast<std::chrono::microseconds>(
std::chrono::steady_clock::now() - tempStageProbeStart)
.count();
tempStageProbeMicros[tempStageProbeStage].fetch_add(
static_cast<unsigned long long>(tempStageProbeElapsedUs), std::memory_order_relaxed);
tempStageProbeCalls[tempStageProbeStage].fetch_add(1ull, std::memory_order_relaxed);
}
TempStageProbeScope(const TempStageProbeScope&) = delete;
TempStageProbeScope& operator=(const TempStageProbeScope&) = delete;
private:
int tempStageProbeStage;
std::chrono::steady_clock::time_point tempStageProbeStart;
};
// TEMP-STAGE-PROBE: one MGLOG_I line per stage carrying the running cumulative total.
inline void TempStageProbeDumpTotals(unsigned long long tempStageProbeLinkIndex) {
for (int tempStageProbeIdx = 0; tempStageProbeIdx < kTempStageProbeStageCount; ++tempStageProbeIdx) {
const unsigned long long tempStageProbeUs =
tempStageProbeMicros[tempStageProbeIdx].load(std::memory_order_relaxed);
const unsigned long long tempStageProbeN =
tempStageProbeCalls[tempStageProbeIdx].load(std::memory_order_relaxed);
MGLOG_I("TEMP-STAGE-PROBE: %s total %llu.%03llu ms over %llu calls (after %llu links)",
kTempStageProbeStageNames[tempStageProbeIdx], tempStageProbeUs / 1000ull,
tempStageProbeUs % 1000ull, tempStageProbeN, tempStageProbeLinkIndex);
}
}
// TEMP-STAGE-PROBE: ticks the program-link counter on scope exit (so a link that returns
// early still counts) and dumps every stage's running total every 25 links. The 3-second
// fallback exists so the LAST link of a load also publishes totals - the process is
// force-stopped afterwards, so there is no exit hook to rely on.
class TempStageProbeLinkTick {
public:
~TempStageProbeLinkTick() {
const unsigned long long tempStageProbeIndex =
tempStageProbeLinkCount.fetch_add(1ull, std::memory_order_relaxed) + 1ull;
const long long tempStageProbeNowNanos =
std::chrono::duration_cast<std::chrono::nanoseconds>(
std::chrono::steady_clock::now().time_since_epoch())
.count();
const long long tempStageProbeLastNanos = tempStageProbeLastDumpNanos.load(std::memory_order_relaxed);
const bool tempStageProbeDueByCount = (tempStageProbeIndex % 25ull) == 0ull;
const bool tempStageProbeDueByTime = tempStageProbeNowNanos - tempStageProbeLastNanos > 3000000000LL;
if (!tempStageProbeDueByCount && !tempStageProbeDueByTime) return;
tempStageProbeLastDumpNanos.store(tempStageProbeNowNanos, std::memory_order_relaxed);
TempStageProbeDumpTotals(tempStageProbeIndex);
}
};
} // namespace MobileGL::MG_Util::Debug
MobileGL/MG_Util/ShaderTranspiler/ShaderCompiler.cpp
+2 -26
View File
@@ -32,7 +32,6 @@
#include <MG_Backend/BackendObjects.h>
#include <MG_Util/Converters/GLToStr/GLEnumConverter.h>
#include <MG_Util/Converters/GLToGlslang/ProgramEnumConverter.h>
#include <MG_Util/Debug/TempStageProbe.h> // TEMP-STAGE-PROBE
#include <cstdlib>
namespace MobileGL {
@@ -225,18 +224,8 @@ namespace MobileGL {
return std::unexpected(r);
}
// TEMP-STAGE-PROBE: "parse" - one glslang TShader::parse. Both the first
// attempt and the legacy-#version retry below go through here, and so does
// ProgramLinkTask's consume-once re-parse (separately counted there as
// "parse-reparse") and ShaderCompiler::PrewarmBuiltins' three warm-up parses.
const auto tempStageProbeParseShaderSource = [&](const String& tempStageProbeSource) {
const MobileGL::MG_Util::Debug::TempStageProbeScope tempStageProbeParse(
MobileGL::MG_Util::Debug::kTempStageProbeParse);
return ParseShaderSource(lang, shaderType, tempStageProbeSource, attrib.flags, attrib.env);
};
const String source(attrib.sourceStr);
auto result = tempStageProbeParseShaderSource(source);
auto result = ParseShaderSource(lang, shaderType, source, attrib.flags, attrib.env);
if (result) return result;
// Legacy desktop sources are normalized to "#version 330 core" (with a marker on the
@@ -252,7 +241,7 @@ namespace MobileGL {
return result;
}
auto retryResult = tempStageProbeParseShaderSource(retrySource); // TEMP-STAGE-PROBE
auto retryResult = ParseShaderSource(lang, shaderType, retrySource, attrib.flags, attrib.env);
if (!retryResult) return result;
MGLOG_D("CompileShader: %s only parsed after retargeting its legacy #version to 460",
@@ -385,19 +374,6 @@ namespace MobileGL {
return optimizer.Run(inputBinary.data(), inputBinary.size(), &outputBinary, options);
}
// TEMP-STAGE-PROBE tempStageProbeNullOptimize: SanitizeAndOptimizeBinary minus
// every pass - measures the pure IR plumbing (BuildModule + serialize +
// IRContext teardown) so a device run can separate allocator-bound plumbing
// from transformation work. Remove with the probes.
bool ShaderCompiler::TempProbeNullOptimizeBinary(const Vector<Uint32>& inputBinary,
Vector<uint32_t>& outputBinary) {
using namespace spvtools;
OptimizerOptions options;
options.set_run_validator(false);
Optimizer optimizer(SPV_ENV_VULKAN_1_1);
return optimizer.Run(inputBinary.data(), inputBinary.size(), &outputBinary, options);
}
bool ShaderCompiler::LowerDrawParametersForEssl(const Vector<Uint32>& inputBinary,
Vector<uint32_t>& outputBinary) {
using namespace spvtools;
MobileGL/MG_Util/ShaderTranspiler/ShaderCompiler.h
-4
View File
@@ -22,10 +22,6 @@ namespace MobileGL {
static Result<Vector<Vector<unsigned>>> GetSpirvBinaryFromProgram(const ProgramBinaryAttrib& attrib);
static bool SanitizeAndOptimizeBinary(const Vector<Uint32>& inputBinary,
Vector<uint32_t>& outputBinary);
// TEMP-STAGE-PROBE: SanitizeAndOptimizeBinary with zero passes - pure IR
// plumbing, for the allocator-pathology discriminator. Remove with the probes.
static bool TempProbeNullOptimizeBinary(const Vector<Uint32>& inputBinary,
Vector<uint32_t>& outputBinary);
// Demotes DrawIndex/BaseInstance/BaseVertex builtins to plain Private globals
// (mg_DrawID/mg_BaseInstance/mg_BaseVertex) so SPIRV-Cross can emit ESSL.
// Only for backends without native draw-parameter support (DirectGLES).