From d17d371c40209d9db42d3c8041d5e7d8679166d7 Mon Sep 17 00:00:00 2001 From: Andrea Di Biagio Date: Tue, 28 Aug 2018 14:27:01 +0000 Subject: [PATCH] [llvm-mca][TimelineView] Force the same number of executions for every entry in the 'wait-times' table. This patch also uses colors to highlight problematic wait-time entries. A problematic entry is an entry with an high wait time that tends to match (or exceed) the size of the scheduler's buffer. Color RED is used if an instruction had to wait an average number of cycles which is bigger than (or equal to) the size of the underlying scheduler's buffer. Color YELLOW is used if the time (in cycles) spend waiting for the operands or pipeline resources is bigger than half the size of the underlying scheduler's buffer. Color MAGENTA is used if an instruction does not consume buffer resources according to the scheduling model. llvm-svn: 340825 --- .../llvm-mca/X86/BtVer2/clear-super-register-2.s | 28 ++-- llvm/test/tools/llvm-mca/X86/BtVer2/pipes-fpu.s | 6 +- llvm/test/tools/llvm-mca/X86/BtVer2/pr37790.s | 4 +- llvm/tools/llvm-mca/Views/TimelineView.cpp | 161 +++++++++++++-------- llvm/tools/llvm-mca/Views/TimelineView.h | 15 +- 5 files changed, 125 insertions(+), 89 deletions(-) diff --git a/llvm/test/tools/llvm-mca/X86/BtVer2/clear-super-register-2.s b/llvm/test/tools/llvm-mca/X86/BtVer2/clear-super-register-2.s index 28cd8af..6d8d8a9 100644 --- a/llvm/test/tools/llvm-mca/X86/BtVer2/clear-super-register-2.s +++ b/llvm/test/tools/llvm-mca/X86/BtVer2/clear-super-register-2.s @@ -104,17 +104,17 @@ vandps %xmm4, %xmm1, %xmm0 # CHECK-NEXT: 1. 2 9.5 0.5 35.5 vaddps %xmm0, %xmm1, %xmm3 # CHECK-NEXT: 2. 2 11.5 0.0 33.5 vaddps %ymm3, %ymm1, %ymm4 # CHECK-NEXT: 3. 2 12.5 2.0 31.5 vaddps %ymm3, %ymm1, %ymm4 -# CHECK-NEXT: 4. 1 5.0 4.0 29.0 vaddps %ymm3, %ymm1, %ymm4 -# CHECK-NEXT: 5. 1 6.0 6.0 27.0 vaddps %ymm3, %ymm1, %ymm4 -# CHECK-NEXT: 6. 1 7.0 7.0 26.0 vaddps %ymm3, %ymm1, %ymm4 -# CHECK-NEXT: 7. 1 8.0 8.0 24.0 vaddps %ymm3, %ymm1, %ymm4 -# CHECK-NEXT: 8. 1 9.0 9.0 23.0 vaddps %ymm3, %ymm1, %ymm4 -# CHECK-NEXT: 9. 1 10.0 10.0 21.0 vaddps %ymm3, %ymm1, %ymm4 -# CHECK-NEXT: 10. 1 11.0 11.0 20.0 vaddps %ymm3, %ymm1, %ymm4 -# CHECK-NEXT: 11. 1 12.0 12.0 18.0 vaddps %ymm3, %ymm1, %ymm4 -# CHECK-NEXT: 12. 1 13.0 13.0 17.0 vaddps %ymm3, %ymm1, %ymm4 -# CHECK-NEXT: 13. 1 14.0 14.0 15.0 vaddps %ymm3, %ymm1, %ymm4 -# CHECK-NEXT: 14. 1 15.0 15.0 14.0 vaddps %ymm3, %ymm1, %ymm4 -# CHECK-NEXT: 15. 1 16.0 16.0 12.0 vaddps %ymm3, %ymm1, %ymm4 -# CHECK-NEXT: 16. 1 17.0 17.0 11.0 vaddps %ymm3, %ymm1, %ymm4 -# CHECK-NEXT: 17. 1 19.0 0.0 10.0 vandps %xmm4, %xmm1, %xmm0 +# CHECK-NEXT: 4. 2 13.5 4.0 30.5 vaddps %ymm3, %ymm1, %ymm4 +# CHECK-NEXT: 5. 2 14.5 6.0 28.5 vaddps %ymm3, %ymm1, %ymm4 +# CHECK-NEXT: 6. 2 15.5 7.5 27.5 vaddps %ymm3, %ymm1, %ymm4 +# CHECK-NEXT: 7. 2 16.5 9.0 25.5 vaddps %ymm3, %ymm1, %ymm4 +# CHECK-NEXT: 8. 2 17.5 10.5 24.5 vaddps %ymm3, %ymm1, %ymm4 +# CHECK-NEXT: 9. 2 18.5 12.0 22.5 vaddps %ymm3, %ymm1, %ymm4 +# CHECK-NEXT: 10. 2 19.5 13.5 21.5 vaddps %ymm3, %ymm1, %ymm4 +# CHECK-NEXT: 11. 2 20.5 15.0 19.5 vaddps %ymm3, %ymm1, %ymm4 +# CHECK-NEXT: 12. 2 21.5 16.5 18.5 vaddps %ymm3, %ymm1, %ymm4 +# CHECK-NEXT: 13. 2 22.5 18.0 16.5 vaddps %ymm3, %ymm1, %ymm4 +# CHECK-NEXT: 14. 2 23.5 19.5 15.5 vaddps %ymm3, %ymm1, %ymm4 +# CHECK-NEXT: 15. 2 21.0 21.0 13.5 vaddps %ymm3, %ymm1, %ymm4 +# CHECK-NEXT: 16. 2 22.0 22.0 12.5 vaddps %ymm3, %ymm1, %ymm4 +# CHECK-NEXT: 17. 2 24.0 0.0 11.5 vandps %xmm4, %xmm1, %xmm0 diff --git a/llvm/test/tools/llvm-mca/X86/BtVer2/pipes-fpu.s b/llvm/test/tools/llvm-mca/X86/BtVer2/pipes-fpu.s index fc942d8..54b8d13 100644 --- a/llvm/test/tools/llvm-mca/X86/BtVer2/pipes-fpu.s +++ b/llvm/test/tools/llvm-mca/X86/BtVer2/pipes-fpu.s @@ -103,6 +103,6 @@ vsqrtps %ymm0, %ymm2 # CHECK-NEXT: 2. 2 1.0 1.0 28.0 vcvttps2dq %xmm0, %xmm2 # CHECK-NEXT: 3. 2 1.0 1.0 29.5 vpclmulqdq $0, %xmm0, %xmm1, %xmm2 # CHECK-NEXT: 4. 2 1.0 1.0 28.0 vaddps %xmm0, %xmm1, %xmm2 -# CHECK-NEXT: 5. 1 1.0 1.0 0.0 vsqrtps %xmm0, %xmm2 -# CHECK-NEXT: 6. 1 1.0 1.0 17.0 vaddps %ymm0, %ymm1, %ymm2 -# CHECK-NEXT: 7. 1 20.0 20.0 0.0 vsqrtps %ymm0, %ymm2 +# CHECK-NEXT: 5. 2 29.5 29.5 0.0 vsqrtps %xmm0, %xmm2 +# CHECK-NEXT: 6. 2 1.0 1.0 45.5 vaddps %ymm0, %ymm1, %ymm2 +# CHECK-NEXT: 7. 2 48.5 48.5 0.0 vsqrtps %ymm0, %ymm2 diff --git a/llvm/test/tools/llvm-mca/X86/BtVer2/pr37790.s b/llvm/test/tools/llvm-mca/X86/BtVer2/pr37790.s index 19597f8..5ac7901 100644 --- a/llvm/test/tools/llvm-mca/X86/BtVer2/pr37790.s +++ b/llvm/test/tools/llvm-mca/X86/BtVer2/pr37790.s @@ -37,5 +37,5 @@ stmxcsr (%rsp) # CHECK-NEXT: [3]: Average time elapsed from WB until retire stage # CHECK: [0] [1] [2] [3] -# CHECK-NEXT: 0. 1 1.0 1.0 0.0 int3 -# CHECK-NEXT: 1. 1 101.0 0.0 0.0 stmxcsr (%rsp) +# CHECK-NEXT: 0. 2 1.0 0.5 0.0 int3 +# CHECK-NEXT: 1. 2 100.5 0.0 0.0 stmxcsr (%rsp) diff --git a/llvm/tools/llvm-mca/Views/TimelineView.cpp b/llvm/tools/llvm-mca/Views/TimelineView.cpp index 79dfa3a..90a3eeb 100644 --- a/llvm/tools/llvm-mca/Views/TimelineView.cpp +++ b/llvm/tools/llvm-mca/Views/TimelineView.cpp @@ -18,41 +18,63 @@ using namespace llvm; namespace mca { -void TimelineView::initialize(unsigned MaxIterations) { - unsigned NumInstructions = - AsmSequence.getNumIterations() * AsmSequence.size(); +TimelineView::TimelineView(const MCSubtargetInfo &sti, MCInstPrinter &Printer, + const SourceMgr &S, unsigned MaxIterations, + unsigned Cycles) + : STI(sti), MCIP(Printer), AsmSequence(S), CurrentCycle(0), + MaxCycle(Cycles == 0 ? 80 : Cycles), LastCycle(0), WaitTime(S.size()), + UsedBuffer(S.size()) { + unsigned NumInstructions = AsmSequence.size(); if (!MaxIterations) MaxIterations = DEFAULT_ITERATIONS; - unsigned NumEntries = - std::min(NumInstructions, MaxIterations * AsmSequence.size()); - Timeline.resize(NumEntries); - TimelineViewEntry NullTVEntry = {0, 0, 0, 0, 0}; - std::fill(Timeline.begin(), Timeline.end(), NullTVEntry); - - WaitTime.resize(AsmSequence.size()); - WaitTimeEntry NullWTEntry = {0, 0, 0, 0}; + NumInstructions *= std::min(MaxIterations, AsmSequence.getNumIterations()); + Timeline.resize(NumInstructions); + + WaitTimeEntry NullWTEntry = {0, 0, 0}; std::fill(WaitTime.begin(), WaitTime.end(), NullWTEntry); } +void TimelineView::onReservedBuffers(const InstRef &IR, + ArrayRef Buffers) { + if (IR.getSourceIndex() >= AsmSequence.size()) + return; + + const MCSchedModel &SM = STI.getSchedModel(); + std::pair BufferInfo = {0, 0}; + for (const unsigned Buffer : Buffers) { + const MCProcResourceDesc &MCDesc = *SM.getProcResource(Buffer); + if (MCDesc.BufferSize <= 0) + continue; + unsigned OtherSize = static_cast(MCDesc.BufferSize); + if (!BufferInfo.first || BufferInfo.second > OtherSize) { + BufferInfo.first = Buffer; + BufferInfo.second = OtherSize; + } + } + + UsedBuffer[IR.getSourceIndex()] = BufferInfo; +} + void TimelineView::onEvent(const HWInstructionEvent &Event) { const unsigned Index = Event.IR.getSourceIndex(); - if (CurrentCycle >= MaxCycle || Index >= Timeline.size()) + if (Index >= Timeline.size()) return; + switch (Event.Type) { case HWInstructionEvent::Retired: { TimelineViewEntry &TVEntry = Timeline[Index]; - TVEntry.CycleRetired = CurrentCycle; + if (CurrentCycle < MaxCycle) + TVEntry.CycleRetired = CurrentCycle; // Update the WaitTime entry which corresponds to this Index. WaitTimeEntry &WTEntry = WaitTime[Index % AsmSequence.size()]; - WTEntry.Executions++; WTEntry.CyclesSpentInSchedulerQueue += TVEntry.CycleIssued - TVEntry.CycleDispatched; assert(TVEntry.CycleDispatched <= TVEntry.CycleReady); WTEntry.CyclesSpentInSQWhileReady += TVEntry.CycleIssued - TVEntry.CycleReady; WTEntry.CyclesSpentAfterWBAndBeforeRetire += - (TVEntry.CycleRetired - 1) - TVEntry.CycleExecuted; + (CurrentCycle - 1) - TVEntry.CycleExecuted; break; } case HWInstructionEvent::Ready: @@ -70,57 +92,83 @@ void TimelineView::onEvent(const HWInstructionEvent &Event) { default: return; } - LastCycle = std::max(LastCycle, CurrentCycle); + if (CurrentCycle < MaxCycle) + LastCycle = std::max(LastCycle, CurrentCycle); +} + +static raw_ostream::Colors chooseColor(unsigned CumulativeCycles, + unsigned Executions, + unsigned BufferSize) { + if (CumulativeCycles && BufferSize == 0) + return raw_ostream::MAGENTA; + if (CumulativeCycles >= (BufferSize * Executions)) + return raw_ostream::RED; + if ((CumulativeCycles * 2) >= (BufferSize * Executions)) + return raw_ostream::YELLOW; + return raw_ostream::SAVEDCOLOR; +} + +static void tryChangeColor(raw_ostream &OS, unsigned Cycles, + unsigned Executions, unsigned BufferSize) { + if (!OS.has_colors()) + return; + + raw_ostream::Colors Color = chooseColor(Cycles, Executions, BufferSize); + if (Color == raw_ostream::SAVEDCOLOR) { + OS.resetColor(); + return; + } + OS.changeColor(Color, /* bold */ true, /* BG */ false); } void TimelineView::printWaitTimeEntry(formatted_raw_ostream &OS, const WaitTimeEntry &Entry, - unsigned SourceIndex) const { + unsigned SourceIndex, + unsigned Executions) const { OS << SourceIndex << '.'; OS.PadToColumn(7); - if (Entry.Executions == 0) { - OS << "- - - - "; - } else { - double AverageTime1, AverageTime2, AverageTime3; - unsigned Executions = Entry.Executions; - AverageTime1 = (double)Entry.CyclesSpentInSchedulerQueue / Executions; - AverageTime2 = (double)Entry.CyclesSpentInSQWhileReady / Executions; - AverageTime3 = (double)Entry.CyclesSpentAfterWBAndBeforeRetire / Executions; - - OS << Executions; - OS.PadToColumn(13); - - OS << format("%.1f", floor((AverageTime1 * 10) + 0.5) / 10); - OS.PadToColumn(20); - OS << format("%.1f", floor((AverageTime2 * 10) + 0.5) / 10); - OS.PadToColumn(27); - OS << format("%.1f", floor((AverageTime3 * 10) + 0.5) / 10); - OS.PadToColumn(34); - } -} + double AverageTime1, AverageTime2, AverageTime3; + AverageTime1 = (double)Entry.CyclesSpentInSchedulerQueue / Executions; + AverageTime2 = (double)Entry.CyclesSpentInSQWhileReady / Executions; + AverageTime3 = (double)Entry.CyclesSpentAfterWBAndBeforeRetire / Executions; -void TimelineView::printAverageWaitTimes(raw_ostream &OS) const { - if (WaitTime.empty()) - return; + OS << Executions; + OS.PadToColumn(13); + unsigned BufferSize = UsedBuffer[SourceIndex].second; + tryChangeColor(OS, Entry.CyclesSpentInSchedulerQueue, Executions, BufferSize); + OS << format("%.1f", floor((AverageTime1 * 10) + 0.5) / 10); + OS.PadToColumn(20); + tryChangeColor(OS, Entry.CyclesSpentInSQWhileReady, Executions, BufferSize); + OS << format("%.1f", floor((AverageTime2 * 10) + 0.5) / 10); + OS.PadToColumn(27); + tryChangeColor(OS, Entry.CyclesSpentAfterWBAndBeforeRetire, Executions, + STI.getSchedModel().MicroOpBufferSize); + OS << format("%.1f", floor((AverageTime3 * 10) + 0.5) / 10); - std::string Buffer; - raw_string_ostream TempStream(Buffer); - formatted_raw_ostream FOS(TempStream); + if (OS.has_colors()) + OS.resetColor(); + OS.PadToColumn(34); +} - FOS << "\n\nAverage Wait times (based on the timeline view):\n" - << "[0]: Executions\n" - << "[1]: Average time spent waiting in a scheduler's queue\n" - << "[2]: Average time spent waiting in a scheduler's queue while ready\n" - << "[3]: Average time elapsed from WB until retire stage\n\n"; - FOS << " [0] [1] [2] [3]\n"; +void TimelineView::printAverageWaitTimes(raw_ostream &OS) const { + std::string Header = + "\n\nAverage Wait times (based on the timeline view):\n" + "[0]: Executions\n" + "[1]: Average time spent waiting in a scheduler's queue\n" + "[2]: Average time spent waiting in a scheduler's queue while ready\n" + "[3]: Average time elapsed from WB until retire stage\n\n" + " [0] [1] [2] [3]\n"; + OS << Header; - // Use a different string stream for the instruction. + // Use a different string stream for printing instructions. std::string Instruction; raw_string_ostream InstrStream(Instruction); + formatted_raw_ostream FOS(OS); + unsigned Executions = Timeline.size() / AsmSequence.size(); for (unsigned I = 0, E = WaitTime.size(); I < E; ++I) { - printWaitTimeEntry(FOS, WaitTime[I], I); + printWaitTimeEntry(FOS, WaitTime[I], I, Executions); // Append the instruction info at the end of the line. const MCInst &Inst = AsmSequence.getMCInstFromIndex(I); @@ -133,9 +181,6 @@ void TimelineView::printAverageWaitTimes(raw_ostream &OS) const { FOS << " " << Str << '\n'; FOS.flush(); Instruction = ""; - - OS << Buffer; - Buffer = ""; } } @@ -202,20 +247,15 @@ static void printTimelineHeader(formatted_raw_ostream &OS, unsigned Cycles) { } void TimelineView::printTimeline(raw_ostream &OS) const { - std::string Buffer; - raw_string_ostream StringStream(Buffer); - formatted_raw_ostream FOS(StringStream); - + formatted_raw_ostream FOS(OS); printTimelineHeader(FOS, LastCycle); FOS.flush(); - OS << Buffer; // Use a different string stream for the instruction. std::string Instruction; raw_string_ostream InstrStream(Instruction); for (unsigned I = 0, E = Timeline.size(); I < E; ++I) { - Buffer = ""; const TimelineViewEntry &Entry = Timeline[I]; if (Entry.CycleRetired == 0) return; @@ -234,7 +274,6 @@ void TimelineView::printTimeline(raw_ostream &OS) const { FOS << " " << Str << '\n'; FOS.flush(); Instruction = ""; - OS << Buffer; } } } // namespace mca diff --git a/llvm/tools/llvm-mca/Views/TimelineView.h b/llvm/tools/llvm-mca/Views/TimelineView.h index a5d7765..3dc3bfe 100644 --- a/llvm/tools/llvm-mca/Views/TimelineView.h +++ b/llvm/tools/llvm-mca/Views/TimelineView.h @@ -135,23 +135,22 @@ class TimelineView : public View { std::vector Timeline; struct WaitTimeEntry { - unsigned Executions; unsigned CyclesSpentInSchedulerQueue; unsigned CyclesSpentInSQWhileReady; unsigned CyclesSpentAfterWBAndBeforeRetire; }; std::vector WaitTime; + std::vector> UsedBuffer; void printTimelineViewEntry(llvm::formatted_raw_ostream &OS, const TimelineViewEntry &E, unsigned Iteration, unsigned SourceIndex) const; void printWaitTimeEntry(llvm::formatted_raw_ostream &OS, - const WaitTimeEntry &E, unsigned Index) const; + const WaitTimeEntry &E, unsigned Index, + unsigned Executions) const; const unsigned DEFAULT_ITERATIONS = 10; - void initialize(unsigned MaxIterations); - // Display characters for the TimelineView report output. struct DisplayChar { static const char Dispatched = 'D'; @@ -165,15 +164,13 @@ class TimelineView : public View { public: TimelineView(const llvm::MCSubtargetInfo &sti, llvm::MCInstPrinter &Printer, const SourceMgr &Sequence, unsigned MaxIterations, - unsigned Cycles) - : STI(sti), MCIP(Printer), AsmSequence(Sequence), CurrentCycle(0), - MaxCycle(Cycles == 0 ? 80 : Cycles), LastCycle(0) { - initialize(MaxIterations); - } + unsigned Cycles); // Event handlers. void onCycleEnd() override { ++CurrentCycle; } void onEvent(const HWInstructionEvent &Event) override; + void onReservedBuffers(const InstRef &IR, + llvm::ArrayRef Buffers) override; // print functionalities. void printTimeline(llvm::raw_ostream &OS) const; -- 2.7.4