From 4126ee2b27b78e0e75526372ab3dff4bec88ca14 Mon Sep 17 00:00:00 2001 From: Yashas Andaluri Date: Tue, 30 Jul 2024 22:41:14 +0530 Subject: [PATCH 1/2] [Hexagon] Do not optimize extended address pointing to another function's block When the constant extender optimization pass encounters an instruction that uses an extended address pointing to another function's block, avoid adding the instruction to the extender list for the current machine function. --- .../Target/Hexagon/HexagonConstExtenders.cpp | 4 + .../CodeGen/Hexagon/cext-opt-block-addr.mir | 173 ++++++++++++++++++ 2 files changed, 177 insertions(+) create mode 100644 llvm/test/CodeGen/Hexagon/cext-opt-block-addr.mir diff --git a/llvm/lib/Target/Hexagon/HexagonConstExtenders.cpp b/llvm/lib/Target/Hexagon/HexagonConstExtenders.cpp index f0933765bbcbda..86ce6b4e05ed27 100644 --- a/llvm/lib/Target/Hexagon/HexagonConstExtenders.cpp +++ b/llvm/lib/Target/Hexagon/HexagonConstExtenders.cpp @@ -1223,6 +1223,10 @@ void HCE::recordExtender(MachineInstr &MI, unsigned OpNum) { if (ER.Kind == MachineOperand::MO_GlobalAddress) if (ER.V.GV->getName().empty()) return; + // Ignore block address that points to block in another function + if (ER.Kind == MachineOperand::MO_BlockAddress) + if (ER.V.BA->getFunction() != &(MI.getMF()->getFunction())) + return; Extenders.push_back(ED); } diff --git a/llvm/test/CodeGen/Hexagon/cext-opt-block-addr.mir b/llvm/test/CodeGen/Hexagon/cext-opt-block-addr.mir new file mode 100644 index 00000000000000..9f140132dcd6c3 --- /dev/null +++ b/llvm/test/CodeGen/Hexagon/cext-opt-block-addr.mir @@ -0,0 +1,173 @@ +# REQUIRES: asserts +# RUN: llc -march=hexagon -run-pass hexagon-cext-opt %s -o - | FileCheck %s + +# Check that the HexagonConstantExtenders pass does not assert when block +# addresses from different functions are used +# CHECK-LABEL: name: wibble +# CHECK: A2_tfrsi blockaddress(@baz +# CHECK: A2_tfrsi blockaddress(@wibble + +--- | + target triple = "hexagon" + + define dso_local void @baz() { + bb: + br label %bb1 + + bb1: ; preds = %bb + %call = tail call fastcc i32 @wibble(i32 poison) + ret void + } + + define internal fastcc i32 @wibble(i32 %arg) { + bb: + %call = tail call i32 @eggs(i32 noundef ptrtoint (ptr blockaddress(@baz, %bb1) to i32)) + br label %bb1 + + bb1: ; preds = %bb + tail call void @baz.1(i32 noundef ptrtoint (ptr blockaddress(@wibble, %bb1) to i32)) + ret i32 %call + } + + declare i32 @eggs(i32 noundef) local_unnamed_addr + + declare void @baz.1(i32 noundef) local_unnamed_addr + +... +--- +name: baz +alignment: 16 +exposesReturnsTwice: false +legalized: false +regBankSelected: false +selected: false +failedISel: false +tracksRegLiveness: true +hasWinCFI: false +callsEHReturn: false +callsUnwindInit: false +hasEHCatchret: false +hasEHScopes: false +hasEHFunclets: false +isOutlined: false +debugInstrRef: false +failsVerification: false +tracksDebugUserValues: false +registers: + - { id: 0, class: intregs, preferred-register: '' } +liveins: [] +frameInfo: + isFrameAddressTaken: false + isReturnAddressTaken: false + hasStackMap: false + hasPatchPoint: false + stackSize: 0 + offsetAdjustment: 0 + maxAlignment: 1 + adjustsStack: false + hasCalls: false + stackProtector: '' + functionContext: '' + maxCallFrameSize: 4294967295 + cvBytesOfCalleeSavedRegisters: 0 + hasOpaqueSPAdjustment: false + hasVAStart: false + hasMustTailInVarArgFunc: false + hasTailCall: true + isCalleeSavedInfoValid: false + localFrameSize: 0 + savePoint: '' + restorePoint: '' +fixedStack: [] +stack: [] +entry_values: [] +callSites: [] +debugValueSubstitutions: [] +constants: [] +machineFunctionInfo: {} +body: | + bb.0.bb: + successors: %bb.1(0x80000000) + + bb.1.bb1 (ir-block-address-taken %ir-block.bb1): + %0:intregs = IMPLICIT_DEF + $r0 = COPY %0 + PS_tailcall_i @wibble, hexagoncsr, implicit $r0 + +... +--- +name: wibble +alignment: 16 +exposesReturnsTwice: false +legalized: false +regBankSelected: false +selected: false +failedISel: false +tracksRegLiveness: true +hasWinCFI: false +callsEHReturn: false +callsUnwindInit: false +hasEHCatchret: false +hasEHScopes: false +hasEHFunclets: false +isOutlined: false +debugInstrRef: false +failsVerification: false +tracksDebugUserValues: false +registers: + - { id: 0, class: intregs, preferred-register: '' } + - { id: 1, class: intregs, preferred-register: '' } + - { id: 2, class: intregs, preferred-register: '' } + - { id: 3, class: intregs, preferred-register: '' } + - { id: 4, class: intregs, preferred-register: '' } +liveins: [] +frameInfo: + isFrameAddressTaken: false + isReturnAddressTaken: false + hasStackMap: false + hasPatchPoint: false + stackSize: 0 + offsetAdjustment: 0 + maxAlignment: 1 + adjustsStack: true + hasCalls: true + stackProtector: '' + functionContext: '' + maxCallFrameSize: 4294967295 + cvBytesOfCalleeSavedRegisters: 0 + hasOpaqueSPAdjustment: false + hasVAStart: false + hasMustTailInVarArgFunc: false + hasTailCall: false + isCalleeSavedInfoValid: false + localFrameSize: 0 + savePoint: '' + restorePoint: '' +fixedStack: [] +stack: [] +entry_values: [] +callSites: [] +debugValueSubstitutions: [] +constants: [] +machineFunctionInfo: {} +body: | + bb.0.bb: + successors: %bb.1(0x80000000) + + %2:intregs = A2_tfrsi blockaddress(@baz, %ir-block.bb1) + ADJCALLSTACKDOWN 0, 0, implicit-def $r29, implicit-def dead $r30, implicit $r31, implicit $r30, implicit $r29 + $r0 = COPY %2 + J2_call @eggs, hexagoncsr, implicit-def dead $pc, implicit-def dead $r31, implicit $r29, implicit $r0, implicit-def $r29, implicit-def $r0 + ADJCALLSTACKUP 0, 0, implicit-def dead $r29, implicit-def dead $r30, implicit-def dead $r31, implicit $r29 + %3:intregs = COPY $r0 + + bb.1.bb1 (ir-block-address-taken %ir-block.bb1): + %4:intregs = A2_tfrsi blockaddress(@wibble, %ir-block.bb1) + ADJCALLSTACKDOWN 0, 0, implicit-def $r29, implicit-def dead $r30, implicit $r31, implicit $r30, implicit $r29 + $r0 = COPY %4 + J2_call @baz.1, hexagoncsr, implicit-def dead $pc, implicit-def dead $r31, implicit $r29, implicit $r0, implicit-def $r29 + ADJCALLSTACKUP 0, 0, implicit-def dead $r29, implicit-def dead $r30, implicit-def dead $r31, implicit $r29 + $r0 = COPY %3 + PS_jmpret $r31, implicit-def dead $pc, implicit $r0 + +... From ed6fedab9938eb95eb5eda73498703e59069e85d Mon Sep 17 00:00:00 2001 From: Yashas Andaluri Date: Wed, 31 Jul 2024 22:28:30 +0530 Subject: [PATCH 2/2] Retrigger checks