Index: lib/Target/Hexagon/HexagonTargetObjectFile.h =================================================================== --- lib/Target/Hexagon/HexagonTargetObjectFile.h +++ lib/Target/Hexagon/HexagonTargetObjectFile.h @@ -33,6 +33,8 @@ unsigned getSmallDataSize() const; + const Function *getLutUsedFunction(const GlobalObject *GO) const; + private: MCSectionELF *SmallDataSection; MCSectionELF *SmallBSSSection; @@ -43,6 +45,10 @@ MCSection *selectSmallSectionForGlobal(const GlobalObject *GO, SectionKind Kind, const TargetMachine &TM) const; + + MCSection *selectSectionForLookupTable(const GlobalObject *GO, + const TargetMachine &TM, + const Function *Fn) const; }; } // namespace llvm Index: lib/Target/Hexagon/HexagonTargetObjectFile.cpp =================================================================== --- lib/Target/Hexagon/HexagonTargetObjectFile.cpp +++ lib/Target/Hexagon/HexagonTargetObjectFile.cpp @@ -49,6 +49,10 @@ cl::Hidden, cl::init(false), cl::desc("Trace global value placement")); +static cl::opt + EmitLutInText("hexagon-emit-lut-text", cl::Hidden, cl::init(false), + cl::desc("Emit hexagon lookup tables in function section")); + // TraceGVPlacement controls messages for all builds. For builds with assertions // (debug or release), messages are also controlled by the usual debug flags // (e.g. -debug and -debug-only=globallayout) @@ -132,6 +136,15 @@ << (Kind.isBSS() ? "kind_bss " : "" ) << (Kind.isBSSLocal() ? "kind_bss_local " : "" )); + // If the lookup table is used by more than one function, do not place + // it in text section. + if (EmitLutInText && GO->getName().startswith("switch.table")) { + const Function *Fn = getLutUsedFunction(GO); + if (Fn) { + return selectSectionForLookupTable(GO, TM, Fn); + } + } + if (isGlobalInSmallSection(GO, TM)) return selectSmallSectionForGlobal(GO, Kind, TM); @@ -393,3 +406,39 @@ // Otherwise, we work the same as ELF. return TargetLoweringObjectFileELF::SelectSectionForGlobal(GO, Kind, TM); } + +// Return the function that uses the lookup table. If there are more +// than one live function that uses this look table, bail out and place +// the lookup table in default section. +const Function * +HexagonTargetObjectFile::getLutUsedFunction(const GlobalObject *GO) const { + const Function *ReturnFn = nullptr; + for (auto U : GO->users()) { + // validate each instance of user to be a live function. + auto *I = dyn_cast(U); + if (I) { + auto *Bb = I->getParent(); + if (Bb) { + auto *UserFn = Bb->getParent(); + if (!ReturnFn) { + ReturnFn = UserFn; + } else if (ReturnFn != UserFn) + return nullptr; + } + } + } + return ReturnFn; +} + +MCSection *HexagonTargetObjectFile::selectSectionForLookupTable( + const GlobalObject *GO, const TargetMachine &TM, const Function *Fn) const { + + SectionKind Kind = SectionKind::getText(); + // If the function has explicit section, place the lookup table in this + // explicit section. + if (Fn->hasSection()) + return getExplicitSectionGlobal(Fn, Kind, TM); + + const auto *FuncObj = dyn_cast(Fn); + return SelectSectionForGlobal(FuncObj, Kind, TM); +} Index: test/CodeGen/Hexagon/switch-lut-explicit-section.ll =================================================================== --- /dev/null +++ test/CodeGen/Hexagon/switch-lut-explicit-section.ll @@ -0,0 +1,32 @@ +;RUN: llc -O2 -hexagon-emit-lut-text=true < %s | FileCheck --check-prefix=FUNCTEXT %s +;RUN: llc -O2 -hexagon-emit-lut-text=true -function-sections < %s | FileCheck --check-prefix=FUNCTEXT %s + +;This test checks the placement of lookup table in explicit section from the attribute set. +target datalayout = "e-m:e-p:32:32:32-a:0-n16:32-i64:64:64-i32:32:32-i16:16:16-i1:8:8-f32:32:32-f64:64:64-v32:32:32-v64:64:64-v512:512:512-v1024:1024:1024-v2048:2048:2048" +target triple = "hexagon-unknown--elf" + +;FUNCTEXT: .text +;FUNCTEXT: .section{{.*}}tcm.hexagon, +;FUNCTEXT-NOT: .section{{.*}}.rodata +;FUNCTEXT-NOT: .text +;FUNCTEXT: .Lswitch.table: +;FUNCTEXT-NEXT: .word + +@switch.table = private unnamed_addr constant [9 x i32] [i32 9, i32 20, i32 14, i32 22, i32 12, i32 5, i32 98, i32 8, i32 11] #0 + +; Function Attrs: norecurse nounwind readnone +define i32 @foo(i32 %x) local_unnamed_addr #0 section "tcm.hexagon" { +entry: + %0 = icmp ult i32 %x, 9 + br i1 %0, label %switch.lookup, label %return + +switch.lookup: ; preds = %entry + %switch.gep = getelementptr inbounds [9 x i32], [9 x i32]* @switch.table, i32 0, i32 %x + %switch.load = load i32, i32* %switch.gep, align 4 + ret i32 %switch.load + +return: ; preds = %entry + ret i32 19 +} + +attributes #0 = { norecurse nounwind readnone "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="hexagonv60" "target-features"="-hvx-double,-long-calls" "unsafe-fp-math"="false" "use-soft-float"="false" } Index: test/CodeGen/Hexagon/switch-lut-function-section.ll =================================================================== --- /dev/null +++ test/CodeGen/Hexagon/switch-lut-function-section.ll @@ -0,0 +1,30 @@ +;RUN: llc -O2 -hexagon-emit-lut-text=true -function-sections < %s | FileCheck --check-prefix=FUNCTEXT %s + +;This test checks the placement of lookup table in function's text section. +target datalayout = "e-m:e-p:32:32:32-a:0-n16:32-i64:64:64-i32:32:32-i16:16:16-i1:8:8-f32:32:32-f64:64:64-v32:32:32-v64:64:64-v512:512:512-v1024:1024:1024-v2048:2048:2048" +target triple = "hexagon-unknown--elf" + +;FUNCTEXT: .text +;FUNCTEXT: .section{{.*}}text.foo, +;FUNCTEXT-NOT: .section{{.*}}.rodata +;FUNCTEXT: .Lswitch.table: +;FUNCTEXT-NEXT: .word + +@switch.table = private unnamed_addr constant [9 x i32] [i32 9, i32 20, i32 14, i32 22, i32 12, i32 5, i32 98, i32 8, i32 11] #0 + +; Function Attrs: norecurse nounwind readnone +define i32 @foo(i32 %x) local_unnamed_addr #0 { +entry: + %0 = icmp ult i32 %x, 9 + br i1 %0, label %switch.lookup, label %return + +switch.lookup: ; preds = %entry + %switch.gep = getelementptr inbounds [9 x i32], [9 x i32]* @switch.table, i32 0, i32 %x + %switch.load = load i32, i32* %switch.gep, align 4 + ret i32 %switch.load + +return: ; preds = %entry + ret i32 19 +} + +attributes #0 = { norecurse nounwind readnone "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="hexagonv60" "target-features"="-hvx-double,-long-calls" "unsafe-fp-math"="false" "use-soft-float"="false" } Index: test/CodeGen/Hexagon/switch-lut-multiple-functions.ll =================================================================== --- /dev/null +++ test/CodeGen/Hexagon/switch-lut-multiple-functions.ll @@ -0,0 +1,42 @@ +;RUN: llc -O2 -hexagon-emit-lut-text=true < %s | FileCheck --check-prefix=TEXT %s +;If the look up table is used by more than one function, we should ignore the +;flag and place it the rodata. +target datalayout = "e-m:e-p:32:32:32-a:0-n16:32-i64:64:64-i32:32:32-i16:16:16-i1:8:8-f32:32:32-f64:64:64-v32:32:32-v64:64:64-v512:512:512-v1024:1024:1024-v2048:2048:2048" +target triple = "hexagon-unknown--elf" + +;TEXT: .text +;TEXT: .section{{.*}}.rodata +;TEXT: .Lswitch.table: +;TEXT-NEXT: .word +@switch.table = private unnamed_addr constant [9 x i32] [i32 9, i32 20, i32 14, i32 22, i32 12, i32 5, i32 98, i32 8, i32 11] + +; Function Attrs: norecurse nounwind readnone +define i32 @foo(i32 %x) local_unnamed_addr #0 { +entry: + %0 = icmp ult i32 %x, 9 + br i1 %0, label %switch.lookup, label %return + +switch.lookup: ; preds = %entry + %switch.gep = getelementptr inbounds [9 x i32], [9 x i32]* @switch.table, i32 0, i32 %x + %switch.load = load i32, i32* %switch.gep, align 4 + ret i32 %switch.load + +return: ; preds = %entry + ret i32 19 +} + +define i32 @goo(i32 %x) local_unnamed_addr #0 { +entry: + %0 = icmp ult i32 %x, 9 + br i1 %0, label %switch.lookup, label %return + +switch.lookup: ; preds = %entry + %switch.gep = getelementptr inbounds [9 x i32], [9 x i32]* @switch.table, i32 0, i32 %x + %switch.load = load i32, i32* %switch.gep, align 4 + ret i32 %switch.load + +return: ; preds = %entry + ret i32 19 +} + +attributes #0 = { norecurse nounwind readnone "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="hexagonv60" "target-features"="-hvx-double,-long-calls" "unsafe-fp-math"="false" "use-soft-float"="false" } Index: test/CodeGen/Hexagon/switch-lut-text-section.ll =================================================================== --- /dev/null +++ test/CodeGen/Hexagon/switch-lut-text-section.ll @@ -0,0 +1,27 @@ +;RUN: llc -O2 -hexagon-emit-lut-text=true < %s | FileCheck --check-prefix=TEXT %s +;This test checks the placement of lookup table in text section. +target datalayout = "e-m:e-p:32:32:32-a:0-n16:32-i64:64:64-i32:32:32-i16:16:16-i1:8:8-f32:32:32-f64:64:64-v32:32:32-v64:64:64-v512:512:512-v1024:1024:1024-v2048:2048:2048" +target triple = "hexagon-unknown--elf" + +;TEXT: .text +;TEXT-NOT: .section{{.*}}.rodata +;TEXT: .Lswitch.table: +;TEXT-NEXT: .word +@switch.table = private unnamed_addr constant [9 x i32] [i32 9, i32 20, i32 14, i32 22, i32 12, i32 5, i32 98, i32 8, i32 11] + +; Function Attrs: norecurse nounwind readnone +define i32 @foo(i32 %x) local_unnamed_addr #0 { +entry: + %0 = icmp ult i32 %x, 9 + br i1 %0, label %switch.lookup, label %return + +switch.lookup: ; preds = %entry + %switch.gep = getelementptr inbounds [9 x i32], [9 x i32]* @switch.table, i32 0, i32 %x + %switch.load = load i32, i32* %switch.gep, align 4 + ret i32 %switch.load + +return: ; preds = %entry + ret i32 19 +} + +attributes #0 = { norecurse nounwind readnone "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="hexagonv60" "target-features"="-hvx-double,-long-calls" "unsafe-fp-math"="false" "use-soft-float"="false" }