Enable fake hot/cold splitting on ARM64

This commit contains fixes for various bugs exposed by enabling fake hot/cold splitting on ARM64: - Branches between hot/cold sections are now always long. - The pseudoinstruction for loading a constant from the cold section did not support loading 16-byte data into vector registers, as it temporarily loaded the constant into an 8-byte integer register. Now, 16-byte constants are loaded directly into vector registers via an `ld1` instruction. - Tests involving loading 16-byte constants exposed the data section is not always aligned to its largest constant. Now, the data section is always aligned to `emitConsDsc.alignment` when calling `eeAllocMem`. - Asserts/NYIs blocking hot/cold splitting on ARM64 have been removed. Fake hot/cold splitting requires we fake unwind info by treating each split function as one hot section. A more architecture-agnostic approach for this has been applied.
dotnet · Jun 14, 2022 · d2bbed8 · d2bbed8
1 parent 7989a93
commit d2bbed8
Show file tree

Hide file tree

Showing 8 changed files with 108 additions and 105 deletions.
diff --git a/src/coreclr/jit/compiler.h b/src/coreclr/jit/compiler.h
@@ -8000,10 +8000,6 @@ XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
     void unwindReserveFuncHelper(FuncInfoDsc* func, bool isHotCode);
     void unwindEmitFuncHelper(FuncInfoDsc* func, void* pHotCode, void* pColdCode, bool isHotCode);
 
-#ifdef DEBUG
-    void fakeUnwindEmitFuncHelper(FuncInfoDsc* func, void* pHotCode);
-#endif // DEBUG
-
 #endif // TARGET_AMD64 || (TARGET_X86 && FEATURE_EH_FUNCLETS)
 
     UNATIVE_OFFSET unwindGetCurrentOffset(FuncInfoDsc* func);

diff --git a/src/coreclr/jit/emit.cpp b/src/coreclr/jit/emit.cpp
@@ -6045,13 +6045,8 @@ unsigned emitter::emitEndCodeGen(Compiler* comp,
     // For arm64/LoongArch64, we want to allocate JIT data always adjacent to code similar to what native compiler does.
     // This way allows us to use a single `ldr` to access such data like float constant/jmp table.
 
-    UNATIVE_OFFSET roDataAlignmentDelta = 0;
-    if (emitConsDsc.dsdOffs && (emitConsDsc.alignment == TARGET_POINTER_SIZE))
-    {
-        UNATIVE_OFFSET roDataAlignment = TARGET_POINTER_SIZE; // 8 Byte align by default.
-        roDataAlignmentDelta = (UNATIVE_OFFSET)ALIGN_UP(emitTotalHotCodeSize, roDataAlignment) - emitTotalHotCodeSize;
-        assert((roDataAlignmentDelta == 0) || (roDataAlignmentDelta == 4));
-    }
+    const UNATIVE_OFFSET roDataAlignmentDelta =
+        (UNATIVE_OFFSET)ALIGN_UP(emitTotalHotCodeSize, emitConsDsc.alignment) - emitTotalHotCodeSize;
 
     args.hotCodeSize  = emitTotalHotCodeSize + roDataAlignmentDelta + emitConsDsc.dsdOffs;
     args.coldCodeSize = emitTotalColdCodeSize;

diff --git a/src/coreclr/jit/emitarm64.cpp b/src/coreclr/jit/emitarm64.cpp
@@ -8422,7 +8422,8 @@ void emitter::emitIns_J(instruction ins, BasicBlock* dst, int instrCount)
     {
         case INS_bl_local:
             idjShort = true;
-        // Fall through.
+            fmt = IF_BI_0A;
+            break;
         case INS_b:
             // Unconditional jump is a single form.
             // Assume is long in case we cross hot/cold sections.
@@ -9825,6 +9826,9 @@ BYTE* emitter::emitOutputLJ(insGroup* ig, BYTE* dst, instrDesc* i)
                 // Special case: emit add + ld1 instructions for loading 16-byte data into vector register.
                 if (isVectorRegister(dstReg) && (opSize == EA_16BYTE))
                 {
+                    // Low 4 bits should be 0 -- 16-byte JIT data should be aligned on 16 bytes.
+                    assert((imm12 & 15) == 0);
+
                     const emitAttr elemSize = EA_1BYTE;
                     const insOpts  opt      = optMakeArrangement(opSize, elemSize);
 

diff --git a/src/coreclr/jit/unwind.cpp b/src/coreclr/jit/unwind.cpp
@@ -69,7 +69,16 @@ void Compiler::unwindGetFuncLocations(FuncInfoDsc*             func,
                 // The hot section only goes up to the cold section
                 assert(fgFirstFuncletBB == nullptr);
 
-                *ppEndLoc = new (this, CMK_UnwindInfo) emitLocation(ehEmitCookie(fgFirstColdBlock));
+#ifdef DEBUG
+                if (JitConfig.JitFakeProcedureSplitting())
+                {
+                    *ppEndLoc = nullptr; // If fake-splitting, "trick" VM by pretending entire function is hot.
+                }
+                else
+#endif // DEBUG
+                {
+                    *ppEndLoc = new (this, CMK_UnwindInfo) emitLocation(ehEmitCookie(fgFirstColdBlock));
+                }
             }
             else
             {
@@ -259,6 +268,13 @@ void Compiler::unwindEmitFuncCFI(FuncInfoDsc* func, void* pHotCode, void* pColdC
     DWORD          unwindCodeBytes = 0;
     BYTE*          pUnwindBlock    = nullptr;
 
+#ifdef DEBUG
+    if (JitConfig.JitFakeProcedureSplitting())
+    {
+        pColdCode = nullptr;
+    }
+#endif // DEBUG
+
     if (func->startLoc == nullptr)
     {
         startOffset = 0;

diff --git a/src/coreclr/jit/unwindamd64.cpp b/src/coreclr/jit/unwindamd64.cpp
@@ -656,18 +656,17 @@ void Compiler::unwindReserve()
 //
 void Compiler::unwindReserveFunc(FuncInfoDsc* func)
 {
-#ifdef DEBUG
-    if (JitConfig.JitFakeProcedureSplitting() && (fgFirstColdBlock != nullptr))
+    unwindReserveFuncHelper(func, true);
+
+    if (fgFirstColdBlock != nullptr)
     {
-        assert(func->funKind == FUNC_ROOT); // No fake-splitting of funclets.
-        unwindReserveFuncHelper(func, true);
-    }
-    else
+#ifdef DEBUG
+        if (JitConfig.JitFakeProcedureSplitting())
+        {
+            assert(func->funKind == FUNC_ROOT); // No splitting of funclets.
+        }
+        else
 #endif // DEBUG
-    {
-        unwindReserveFuncHelper(func, true);
-
-        if (fgFirstColdBlock != nullptr)
         {
             unwindReserveFuncHelper(func, false);
         }
@@ -859,7 +858,17 @@ void Compiler::unwindEmitFuncHelper(FuncInfoDsc* func, void* pHotCode, void* pCo
 
     if (isHotCode)
     {
-        assert(endOffset <= info.compTotalHotCodeSize);
+#ifdef DEBUG
+        if (JitConfig.JitFakeProcedureSplitting() && (fgFirstColdBlock != nullptr))
+        {
+            assert(endOffset <= info.compNativeCodeSize);
+        }
+        else
+#endif // DEBUG
+        {
+            assert(endOffset <= info.compTotalHotCodeSize);
+        }
+
         pColdCode = nullptr;
     }
     else
@@ -890,43 +899,17 @@ void Compiler::unwindEmitFunc(FuncInfoDsc* func, void* pHotCode, void* pColdCode
     static_assert_no_msg(FUNC_HANDLER == (FuncKind)CORJIT_FUNC_HANDLER);
     static_assert_no_msg(FUNC_FILTER == (FuncKind)CORJIT_FUNC_FILTER);
 
-#ifdef DEBUG
-    if (JitConfig.JitFakeProcedureSplitting() && (pColdCode != nullptr))
+    unwindEmitFuncHelper(func, pHotCode, pColdCode, true);
+
+    if (pColdCode != nullptr)
     {
-        fakeUnwindEmitFuncHelper(func, pHotCode);
-    }
-    else
+#ifdef DEBUG
+        if (!JitConfig.JitFakeProcedureSplitting())
 #endif // DEBUG
-    {
-        unwindEmitFuncHelper(func, pHotCode, pColdCode, true);
-
-        if (pColdCode != nullptr)
         {
             unwindEmitFuncHelper(func, pHotCode, pColdCode, false);
         }
     }
 }
 
-#ifdef DEBUG
-void Compiler::fakeUnwindEmitFuncHelper(FuncInfoDsc* func, void* pHotCode)
-{
-    assert(fgFirstColdBlock != nullptr);
-    assert(func->funKind == FUNC_ROOT); // No fake-splitting of funclets.
-
-    const UNATIVE_OFFSET startOffset     = 0;
-    const UNATIVE_OFFSET endOffset       = info.compNativeCodeSize;
-    const DWORD          unwindCodeBytes = sizeof(func->unwindCodes) - func->unwindCodeSlot;
-    BYTE*                pUnwindBlock    = &func->unwindCodes[func->unwindCodeSlot];
-
-    if (opts.dspUnwind)
-    {
-        DumpUnwindInfo(true, startOffset, endOffset, (const UNWIND_INFO* const)pUnwindBlock);
-    }
-
-    // Pass pColdCode = nullptr; VM allocs unwind info for combined hot/cold section
-    eeAllocUnwindInfo((BYTE*)pHotCode, nullptr, startOffset, endOffset, unwindCodeBytes, pUnwindBlock,
-                      (CorJitFuncKind)func->funKind);
-}
-#endif // DEBUG
-
 #endif // TARGET_AMD64
diff --git a/src/coreclr/jit/unwindarm.cpp b/src/coreclr/jit/unwindarm.cpp
@@ -563,13 +563,20 @@ void Compiler::unwindReserve()
 void Compiler::unwindReserveFunc(FuncInfoDsc* func)
 {
     BOOL isFunclet          = (func->funKind == FUNC_ROOT) ? FALSE : TRUE;
-    bool funcHasColdSection = false;
+    bool funcHasColdSection = (fgFirstColdBlock != nullptr);
+
+#ifdef DEBUG
+    if (JitConfig.JitFakeProcedureSplitting() && funcHasColdSection)
+    {
+        funcHasColdSection = false; // "Trick" the VM into thinking we don't have a cold section.
+    }
+#endif // DEBUG
 
 #if defined(FEATURE_CFI_SUPPORT)
     if (generateCFIUnwindCodes())
     {
         DWORD unwindCodeBytes = 0;
-        if (fgFirstColdBlock != nullptr)
+        if (funcHasColdSection)
         {
             eeReserveUnwindInfo(isFunclet, true /*isColdCode*/, unwindCodeBytes);
         }
@@ -584,7 +591,7 @@ void Compiler::unwindReserveFunc(FuncInfoDsc* func)
     // cold section. This needs to be done before we split into fragments, as each
     // of the hot and cold sections can have multiple fragments.
 
-    if (fgFirstColdBlock != NULL)
+    if (funcHasColdSection)
     {
         assert(!isFunclet); // TODO-CQ: support hot/cold splitting with EH
 
@@ -595,8 +602,6 @@ void Compiler::unwindReserveFunc(FuncInfoDsc* func)
         func->uwiCold = new (this, CMK_UnwindInfo) UnwindInfo();
         func->uwiCold->InitUnwindInfo(this, startLoc, endLoc);
         func->uwiCold->HotColdSplitCodes(&func->uwi);
-
-        funcHasColdSection = true;
     }
 
     // First we need to split the function or funclet into fragments that are no larger
@@ -1604,11 +1609,19 @@ void UnwindFragmentInfo::Allocate(
     UNATIVE_OFFSET endOffset;
     UNATIVE_OFFSET codeSize;
 
-    // We don't support hot/cold splitting with EH, so if there is cold code, this
-    // better not be a funclet!
-    // TODO-CQ: support funclets in cold code
-
-    noway_assert(isHotCode || funKind == CORJIT_FUNC_ROOT);
+// We don't support hot/cold splitting with EH, so if there is cold code, this
+// better not be a funclet!
+// TODO-CQ: support funclets in cold code
+#ifdef DEBUG
+    if (JitConfig.JitFakeProcedureSplitting() && (pColdCode != NULL))
+    {
+        noway_assert(isHotCode && (funKind == CORJIT_FUNC_ROOT));
+    }
+    else
+#endif // DEBUG
+    {
+        noway_assert(isHotCode || (funKind == CORJIT_FUNC_ROOT));
+    }
 
     // Compute the final size, and start and end offsets of the fragment
 
@@ -1656,7 +1669,17 @@ void UnwindFragmentInfo::Allocate(
 
     if (isHotCode)
     {
-        assert(endOffset <= uwiComp->info.compTotalHotCodeSize);
+#ifdef DEBUG
+        if (JitConfig.JitFakeProcedureSplitting() && (pColdCode != NULL))
+        {
+            assert(endOffset <= uwiComp->info.compNativeCodeSize);
+        }
+        else
+#endif // DEBUG
+        {
+            assert(endOffset <= uwiComp->info.compTotalHotCodeSize);
+        }
+
         pColdCode = NULL;
     }
     else

diff --git a/src/coreclr/jit/unwindx86.cpp b/src/coreclr/jit/unwindx86.cpp
@@ -113,18 +113,17 @@ void Compiler::unwindEmit(void* pHotCode, void* pColdCode)
 //
 void Compiler::unwindReserveFunc(FuncInfoDsc* func)
 {
-#ifdef DEBUG
-    if (JitConfig.JitFakeProcedureSplitting() && (fgFirstColdBlock != nullptr))
+    unwindReserveFuncHelper(func, true);
+
+    if (fgFirstColdBlock != nullptr)
     {
-        assert(func->funKind == FUNC_ROOT); // No fake-splitting of funclets.
-        unwindReserveFuncHelper(func, true);
-    }
-    else
+#ifdef DEBUG
+        if (JitConfig.JitFakeProcedureSplitting())
+        {
+            assert(func->funKind == FUNC_ROOT); // No splitting of funclets.
+        }
+        else
 #endif // DEBUG
-    {
-        unwindReserveFuncHelper(func, true);
-
-        if (fgFirstColdBlock != nullptr)
         {
             unwindReserveFuncHelper(func, false);
         }
@@ -164,17 +163,13 @@ void Compiler::unwindEmitFunc(FuncInfoDsc* func, void* pHotCode, void* pColdCode
     static_assert_no_msg(FUNC_HANDLER == (FuncKind)CORJIT_FUNC_HANDLER);
     static_assert_no_msg(FUNC_FILTER == (FuncKind)CORJIT_FUNC_FILTER);
 
-#ifdef DEBUG
-    if (JitConfig.JitFakeProcedureSplitting() && (pColdCode != nullptr))
+    unwindEmitFuncHelper(func, pHotCode, pColdCode, true);
+
+    if (pColdCode != nullptr)
     {
-        fakeUnwindEmitFuncHelper(func, pHotCode);
-    }
-    else
+#ifdef DEBUG
+        if (!JitConfig.JitFakeProcedureSplitting())
 #endif // DEBUG
-    {
-        unwindEmitFuncHelper(func, pHotCode, pColdCode, true);
-
-        if (pColdCode != nullptr)
         {
             unwindEmitFuncHelper(func, pHotCode, pColdCode, false);
         }
@@ -258,7 +253,17 @@ void Compiler::unwindEmitFuncHelper(FuncInfoDsc* func, void* pHotCode, void* pCo
 
     if (isHotCode)
     {
-        assert(endOffset <= info.compTotalHotCodeSize);
+#ifdef DEBUG
+        if (JitConfig.JitFakeProcedureSplitting() && (fgFirstColdBlock != nullptr))
+        {
+            assert(endOffset <= info.compNativeCodeSize);
+        }
+        else
+#endif // DEBUG
+        {
+            assert(endOffset <= info.compTotalHotCodeSize);
+        }
+
         pColdCode = nullptr;
     }
     else
@@ -276,22 +281,4 @@ void Compiler::unwindEmitFuncHelper(FuncInfoDsc* func, void* pHotCode, void* pCo
                       (BYTE*)&unwindInfo, (CorJitFuncKind)func->funKind);
 }
 
-#ifdef DEBUG
-void Compiler::fakeUnwindEmitFuncHelper(FuncInfoDsc* func, void* pHotCode)
-{
-    assert(fgFirstColdBlock != nullptr);
-    assert(func->funKind == FUNC_ROOT); // No fake-splitting of funclets.
-
-    const UNATIVE_OFFSET startOffset = 0;
-    const UNATIVE_OFFSET endOffset   = info.compNativeCodeSize;
-
-    UNWIND_INFO unwindInfo;
-    unwindInfo.FunctionLength = (ULONG)(endOffset);
-
-    // Pass pColdCode = nullptr; VM allocs unwind info for combined hot/cold section
-    eeAllocUnwindInfo((BYTE*)pHotCode, nullptr, startOffset, endOffset, sizeof(UNWIND_INFO), (BYTE*)&unwindInfo,
-                      (CorJitFuncKind)func->funKind);
-}
-#endif // DEBUG
-
 #endif // FEATURE_EH_FUNCLETS
diff --git a/src/tests/Common/testenvironment.proj b/src/tests/Common/testenvironment.proj
@@ -36,7 +36,6 @@
       COMPlus_EnableSSE42;
       COMPlus_EnableSSSE3;
       COMPlus_ForceRelocs;
-      COMPlus_GCgen0size;
       COMPlus_GCStress;
       COMPlus_GCName;
       COMPlus_gcServer;