<div dir="ltr">Is this actually all you have to do to use the new ELF linker instead of the old one for AMDGPU? I'm asking because this is a rather small patch. Are you planning to do something in addition to this?</div><div class="gmail_extra"><br><div class="gmail_quote">On Wed, Jan 6, 2016 at 7:59 PM, Tom Stellard via llvm-commits <span dir="ltr"><<a href="mailto:llvm-commits@lists.llvm.org" target="_blank">llvm-commits@lists.llvm.org</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">Author: tstellar<br>
Date: Wed Jan 6 21:59:08 2016<br>
New Revision: 257023<br>
<br>
URL: <a href="http://llvm.org/viewvc/llvm-project?rev=257023&view=rev" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project?rev=257023&view=rev</a><br>
Log:<br>
[ELF] Add AMDGPU support<br>
<br>
Summary: This will allow us to remove the AMDGPU support from old ELF.<br>
<br>
Reviewers: rafael, ruiu<br>
<br>
Differential Revision: <a href="http://reviews.llvm.org/D15895" rel="noreferrer" target="_blank">http://reviews.llvm.org/D15895</a><br>
<br>
Added:<br>
lld/trunk/test/ELF/amdgpu-entry.s<br>
lld/trunk/test/ELF/amdgpu-globals.s<br>
lld/trunk/test/ELF/amdgpu-kernels.s<br>
Modified:<br>
lld/trunk/ELF/Driver.cpp<br>
lld/trunk/ELF/OutputSections.cpp<br>
lld/trunk/ELF/Target.cpp<br>
lld/trunk/ELF/Writer.cpp<br>
<br>
Modified: lld/trunk/ELF/Driver.cpp<br>
URL: <a href="http://llvm.org/viewvc/llvm-project/lld/trunk/ELF/Driver.cpp?rev=257023&r1=257022&r2=257023&view=diff" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/lld/trunk/ELF/Driver.cpp?rev=257023&r1=257022&r2=257023&view=diff</a><br>
==============================================================================<br>
--- lld/trunk/ELF/Driver.cpp (original)<br>
+++ lld/trunk/ELF/Driver.cpp Wed Jan 6 21:59:08 2016<br>
@@ -253,6 +253,9 @@ void LinkerDriver::createFiles(opt::Inpu<br>
<br>
if (Config->GnuHash && Config->EMachine == EM_MIPS)<br>
error("The .gnu.hash section is not compatible with the MIPS target.");<br>
+<br>
+ if (!Config->Entry.empty() && Config->EMachine == EM_AMDGPU)<br>
+ error("-e option is not valid for AMDGPU.");<br>
}<br>
<br>
template <class ELFT> void LinkerDriver::link(opt::InputArgList &Args) {<br>
@@ -261,7 +264,10 @@ template <class ELFT> void LinkerDriver:<br>
<br>
if (!Config->Shared) {<br>
// Add entry symbol.<br>
- if (Config->Entry.empty())<br>
+ //<br>
+ // There is no entry symbol for AMDGPU binaries, so skip adding one to avoid<br>
+ // having and undefined symbol.<br>
+ if (Config->Entry.empty() && Config->EMachine != EM_AMDGPU)<br>
Config->Entry = (Config->EMachine == EM_MIPS) ? "__start" : "_start";<br>
<br>
// In the assembly for 32 bit x86 the _GLOBAL_OFFSET_TABLE_ symbol<br>
<br>
Modified: lld/trunk/ELF/OutputSections.cpp<br>
URL: <a href="http://llvm.org/viewvc/llvm-project/lld/trunk/ELF/OutputSections.cpp?rev=257023&r1=257022&r2=257023&view=diff" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/lld/trunk/ELF/OutputSections.cpp?rev=257023&r1=257022&r2=257023&view=diff</a><br>
==============================================================================<br>
--- lld/trunk/ELF/OutputSections.cpp (original)<br>
+++ lld/trunk/ELF/OutputSections.cpp Wed Jan 6 21:59:08 2016<br>
@@ -788,6 +788,11 @@ typename ELFFile<ELFT>::uintX_t elf2::ge<br>
InputSectionBase<ELFT> *SC = DR.Section;<br>
if (!SC)<br>
return DR.Sym.st_value;<br>
+<br>
+ // Symbol offsets for AMDGPU need to be the offset in bytes of the symbol<br>
+ // from the beginning of the section.<br>
+ if (Config->EMachine == EM_AMDGPU)<br>
+ return SC->getOffset(DR.Sym);<br>
if (DR.Sym.getType() == STT_TLS)<br>
return SC->OutSec->getVA() + SC->getOffset(DR.Sym) -<br>
Out<ELFT>::TlsPhdr->p_vaddr;<br>
@@ -1325,7 +1330,11 @@ void SymbolTableSection<ELFT>::writeLoca<br>
continue;<br>
const OutputSectionBase<ELFT> *OutSec = Section->OutSec;<br>
ESym->st_shndx = OutSec->SectionIndex;<br>
- VA += OutSec->getVA() + Section->getOffset(Sym);<br>
+ VA = Section->getOffset(Sym);<br>
+ // Symbol offsets for AMDGPU need to be the offset in bytes of the<br>
+ // symbol from the beginning of the section.<br>
+ if (Config->EMachine != EM_AMDGPU)<br>
+ VA += OutSec->getVA();<br>
}<br>
ESym->st_name = StrTabSec.addString(SymName);<br>
ESym->st_size = Sym.st_size;<br>
<br>
Modified: lld/trunk/ELF/Target.cpp<br>
URL: <a href="http://llvm.org/viewvc/llvm-project/lld/trunk/ELF/Target.cpp?rev=257023&r1=257022&r2=257023&view=diff" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/lld/trunk/ELF/Target.cpp?rev=257023&r1=257022&r2=257023&view=diff</a><br>
==============================================================================<br>
--- lld/trunk/ELF/Target.cpp (original)<br>
+++ lld/trunk/ELF/Target.cpp Wed Jan 6 21:59:08 2016<br>
@@ -191,6 +191,22 @@ public:<br>
uint8_t *PairedLoc = nullptr) const override;<br>
};<br>
<br>
+class AMDGPUTargetInfo final : public TargetInfo {<br>
+public:<br>
+ AMDGPUTargetInfo();<br>
+ void writeGotPltEntry(uint8_t *Buf, uint64_t Plt) const override;<br>
+ void writePltZeroEntry(uint8_t *Buf, uint64_t GotEntryAddr,<br>
+ uint64_t PltEntryAddr) const override;<br>
+ void writePltEntry(uint8_t *Buf, uint64_t GotAddr, uint64_t GotEntryAddr,<br>
+ uint64_t PltEntryAddr, int32_t Index,<br>
+ unsigned RelOff) const override;<br>
+ bool relocNeedsGot(uint32_t Type, const SymbolBody &S) const override;<br>
+ bool relocNeedsPlt(uint32_t Type, const SymbolBody &S) const override;<br>
+ void relocateOne(uint8_t *Loc, uint8_t *BufEnd, uint32_t Type, uint64_t P,<br>
+ uint64_t SA, uint64_t ZA = 0,<br>
+ uint8_t *PairedLoc = nullptr) const override;<br>
+};<br>
+<br>
template <class ELFT> class MipsTargetInfo final : public TargetInfo {<br>
public:<br>
MipsTargetInfo();<br>
@@ -216,6 +232,8 @@ TargetInfo *createTarget() {<br>
return new X86TargetInfo();<br>
case EM_AARCH64:<br>
return new AArch64TargetInfo();<br>
+ case EM_AMDGPU:<br>
+ return new AMDGPUTargetInfo();<br>
case EM_MIPS:<br>
switch (Config->EKind) {<br>
case ELF32LEKind:<br>
@@ -1315,6 +1333,38 @@ void AArch64TargetInfo::relocateOne(uint<br>
}<br>
}<br>
<br>
+AMDGPUTargetInfo::AMDGPUTargetInfo() {}<br>
+<br>
+void AMDGPUTargetInfo::writeGotPltEntry(uint8_t *Buf, uint64_t Plt) const {<br>
+ llvm_unreachable("not implemented");<br>
+}<br>
+<br>
+void AMDGPUTargetInfo::writePltZeroEntry(uint8_t *Buf, uint64_t GotEntryAddr,<br>
+ uint64_t PltEntryAddr) const {<br>
+ llvm_unreachable("not implemented");<br>
+}<br>
+<br>
+void AMDGPUTargetInfo::writePltEntry(uint8_t *Buf, uint64_t GotAddr,<br>
+ uint64_t GotEntryAddr,<br>
+ uint64_t PltEntryAddr, int32_t Index,<br>
+ unsigned RelOff) const {<br>
+ llvm_unreachable("not implemented");<br>
+}<br>
+<br>
+bool AMDGPUTargetInfo::relocNeedsGot(uint32_t Type, const SymbolBody &S) const {<br>
+ return false;<br>
+}<br>
+<br>
+bool AMDGPUTargetInfo::relocNeedsPlt(uint32_t Type, const SymbolBody &S) const {<br>
+ return false;<br>
+}<br>
+<br>
+void AMDGPUTargetInfo::relocateOne(uint8_t *Loc, uint8_t *BufEnd, uint32_t Type,<br>
+ uint64_t P, uint64_t SA, uint64_t ZA,<br>
+ uint8_t *PairedLoc) const {<br>
+ llvm_unreachable("not implemented");<br>
+}<br>
+<br>
template <class ELFT> MipsTargetInfo<ELFT>::MipsTargetInfo() {<br>
PageSize = 65536;<br>
GotHeaderEntriesNum = 2;<br>
<br>
Modified: lld/trunk/ELF/Writer.cpp<br>
URL: <a href="http://llvm.org/viewvc/llvm-project/lld/trunk/ELF/Writer.cpp?rev=257023&r1=257022&r2=257023&view=diff" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/lld/trunk/ELF/Writer.cpp?rev=257023&r1=257022&r2=257023&view=diff</a><br>
==============================================================================<br>
--- lld/trunk/ELF/Writer.cpp (original)<br>
+++ lld/trunk/ELF/Writer.cpp Wed Jan 6 21:59:08 2016<br>
@@ -975,6 +975,18 @@ static uint32_t toPhdrFlags(uint64_t Fla<br>
return Ret;<br>
}<br>
<br>
+/// For AMDGPU we need to use custom segment kinds in order to specify which<br>
+/// address space data should be loaded into.<br>
+template <class ELFT><br>
+static uint32_t getAmdgpuPhdr(OutputSectionBase<ELFT> *Sec) {<br>
+ uint32_t Flags = Sec->getFlags();<br>
+ if (Flags & SHF_AMDGPU_HSA_CODE)<br>
+ return PT_AMDGPU_HSA_LOAD_CODE_AGENT;<br>
+ if ((Flags & SHF_AMDGPU_HSA_GLOBAL) && !(Flags & SHF_AMDGPU_HSA_AGENT))<br>
+ return PT_AMDGPU_HSA_LOAD_GLOBAL_PROGRAM;<br>
+ return PT_LOAD;<br>
+}<br>
+<br>
template <class ELFT><br>
void Writer<ELFT>::updateRelro(Elf_Phdr *Cur, Elf_Phdr *GnuRelroPhdr,<br>
uintX_t VA) {<br>
@@ -1032,7 +1044,9 @@ template <class ELFT> void Writer<ELFT>:<br>
if (PH->p_flags != Flags) {<br>
// Flags changed. Create a new PT_LOAD.<br>
PH = &Phdrs[++PhdrIdx];<br>
- setPhdr(PH, PT_LOAD, Flags, FileOff, VA, 0, Target->getPageSize());<br>
+ uint32_t PTType = (Config->EMachine != EM_AMDGPU) ? (uint32_t)PT_LOAD<br>
+ : getAmdgpuPhdr(Sec);<br>
+ setPhdr(PH, PTType, Flags, FileOff, VA, 0, Target->getPageSize());<br>
}<br>
<br>
if (Sec->getFlags() & SHF_TLS) {<br>
<br>
Added: lld/trunk/test/ELF/amdgpu-entry.s<br>
URL: <a href="http://llvm.org/viewvc/llvm-project/lld/trunk/test/ELF/amdgpu-entry.s?rev=257023&view=auto" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/lld/trunk/test/ELF/amdgpu-entry.s?rev=257023&view=auto</a><br>
==============================================================================<br>
--- lld/trunk/test/ELF/amdgpu-entry.s (added)<br>
+++ lld/trunk/test/ELF/amdgpu-entry.s Wed Jan 6 21:59:08 2016<br>
@@ -0,0 +1,14 @@<br>
+# RUN: llvm-mc -filetype=obj -triple=amdgcpu--amdgcn -mcpu=kaveri %s -o %t.o<br>
+# RUN: not lld -e kernel0 -flavor gnu %t.o -o %t<br>
+<br>
+.hsa_code_object_version 1,0<br>
+.hsa_code_object_isa 7,0,0,"AMD","AMDGPU"<br>
+<br>
+.hsatext<br>
+.globl kernel0<br>
+.align 256<br>
+.amdgpu_hsa_kernel kernel0<br>
+kernel0:<br>
+ s_endpgm<br>
+.Lfunc_end0:<br>
+ .size kernel0, .Lfunc_end0-kernel0<br>
<br>
Added: lld/trunk/test/ELF/amdgpu-globals.s<br>
URL: <a href="http://llvm.org/viewvc/llvm-project/lld/trunk/test/ELF/amdgpu-globals.s?rev=257023&view=auto" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/lld/trunk/test/ELF/amdgpu-globals.s?rev=257023&view=auto</a><br>
==============================================================================<br>
--- lld/trunk/test/ELF/amdgpu-globals.s (added)<br>
+++ lld/trunk/test/ELF/amdgpu-globals.s Wed Jan 6 21:59:08 2016<br>
@@ -0,0 +1,138 @@<br>
+# RUN: llvm-mc -filetype=obj -triple=amdgcpu--amdgcn -mcpu=kaveri %s -o %t.o<br>
+# RUN: lld -flavor gnu %t.o -o %t<br>
+# RUN: llvm-readobj -sections -symbols -program-headers %t | FileCheck %s<br>
+<br>
+# Requires: amdgpu<br>
+<br>
+ .amdgpu_hsa_module_global module_global_program<br>
+ .size module_global_program, 4<br>
+ .hsadata_global_program<br>
+module_global_program:<br>
+ .long 0 ; 0x0<br>
+<br>
+ .amdgpu_hsa_program_global program_global_program<br>
+ .size program_global_program, 4<br>
+ .hsadata_global_program<br>
+program_global_program:<br>
+ .long 0 ; 0x0<br>
+<br>
+ .amdgpu_hsa_module_global module_global_agent<br>
+ .size module_global_agent, 4<br>
+ .hsadata_global_agent<br>
+module_global_agent:<br>
+ .long 0 ; 0x0<br>
+<br>
+ .amdgpu_hsa_program_global program_global_agent<br>
+ .size program_global_agent, 4<br>
+ .hsadata_global_agent<br>
+program_global_agent:<br>
+ .long 0 ; 0x0<br>
+<br>
+ .amdgpu_hsa_module_global module_global_readonly<br>
+ .size module_global_readonly, 4<br>
+ .hsatext<br>
+module_global_readonly:<br>
+ .long 0 ; 0x0<br>
+<br>
+ .amdgpu_hsa_program_global program_global_readonly<br>
+ .size program_global_readonly, 4<br>
+ .hsatext<br>
+program_global_readonly:<br>
+ .long 0 ; 0x0<br>
+<br>
+# CHECK: Section {<br>
+# CHECK: Name: .hsadata_global_program<br>
+# CHECK: Type: SHT_PROGBITS (0x1)<br>
+# CHECK: Flags [ (0x100003)<br>
+# CHECK: SHF_ALLOC (0x2)<br>
+# CHECK: SHF_AMDGPU_HSA_GLOBAL (0x100000)<br>
+# CHECK: SHF_WRITE (0x1)<br>
+# CHECK: ]<br>
+# CHECK: Address: [[HSADATA_GLOBAL_PROGRAM_ADDR:[0-9xa-f]+]]<br>
+# CHECK: }<br>
+<br>
+# CHECK: Section {<br>
+# CHECK: Name: .hsadata_global_agent<br>
+# CHECK: Type: SHT_PROGBITS (0x1)<br>
+# CHECK: Flags [ (0x900003)<br>
+# CHECK: SHF_ALLOC (0x2)<br>
+# CHECK: SHF_AMDGPU_HSA_AGENT (0x800000)<br>
+# CHECK: SHF_AMDGPU_HSA_GLOBAL (0x100000)<br>
+# CHECK: SHF_WRITE (0x1)<br>
+# CHECK: ]<br>
+# CHECK: }<br>
+<br>
+# CHECK: Section {<br>
+# CHECK: Name: .hsatext<br>
+# CHECK: Type: SHT_PROGBITS<br>
+# CHECK: Flags [ (0xC00007)<br>
+# CHECK: SHF_ALLOC (0x2)<br>
+# CHECK: SHF_AMDGPU_HSA_AGENT (0x800000)<br>
+# CHECK: SHF_AMDGPU_HSA_CODE (0x400000)<br>
+# CHECK: SHF_EXECINSTR (0x4)<br>
+# CHECK: SHF_WRITE (0x1)<br>
+# CHECK: ]<br>
+# CHECK: Address: [[HSATEXT_ADDR:[0-9xa-f]+]]<br>
+# CHECK: }<br>
+<br>
+# CHECK: Symbol {<br>
+# CHECK: Name: module_global_agent<br>
+# CHECK: Value: 0x0<br>
+# CHECK: Size: 4<br>
+# CHECK: Binding: Local<br>
+# CHECK: Section: .hsadata_global_agent<br>
+# CHECK: }<br>
+<br>
+# CHECK: Symbol {<br>
+# CHECK: Name: module_global_program<br>
+# CHECK: Value: 0x0<br>
+# CHECK: Size: 4<br>
+# CHECK: Binding: Local<br>
+# CHECK: Section: .hsadata_global_program<br>
+# CHECK: }<br>
+<br>
+# CHECK: Symbol {<br>
+# CHECK: Name: module_global_readonly<br>
+# CHECK: Value: 0x0<br>
+# CHECK: Size: 4<br>
+# CHECK: Binding: Local<br>
+# CHECK: Type: Object<br>
+# CHECK: Section: .hsatext<br>
+# CHECK: }<br>
+<br>
+# CHECK: Symbol {<br>
+# CHECK: Name: program_global_agent<br>
+# CHECK: Value: 0x4<br>
+# CHECK: Size: 4<br>
+# CHECK: Binding: Global<br>
+# CHECK: Type: Object<br>
+# CHECK: Section: .hsadata_global_agent<br>
+# CHECK: }<br>
+<br>
+# CHECK: Symbol {<br>
+# CHECK: Name: program_global_program<br>
+# CHECK: Value: 0x4<br>
+# CHECK: Size: 4<br>
+# CHECK: Binding: Global<br>
+# CHECK: Type: Object<br>
+# CHECK: Section: .hsadata_global_program<br>
+# CHECK: }<br>
+<br>
+# CHECK: Symbol {<br>
+# CHECK: Name: program_global_readonly<br>
+# CHECK: Value: 0x4<br>
+# CHECK: Size: 4<br>
+# CHECK: Binding: Global<br>
+# CHECK: Type: Object<br>
+# CHECK: Section: .hsatext<br>
+# CHECK: }<br>
+<br>
+# CHECK: ProgramHeader {<br>
+# CHECK: Type: PT_AMDGPU_HSA_LOAD_GLOBAL_PROGRAM<br>
+# CHECK: VirtualAddress: [[HSADATA_GLOBAL_PROGRAM_ADDR]]<br>
+# CHECK: }<br>
+<br>
+# CHECK: ProgramHeader {<br>
+# CHECK: Type: PT_AMDGPU_HSA_LOAD_CODE_AGENT<br>
+# CHECK: VirtualAddress: [[HSATEXT_ADDR]]<br>
+# CHECK: }<br>
<br>
Added: lld/trunk/test/ELF/amdgpu-kernels.s<br>
URL: <a href="http://llvm.org/viewvc/llvm-project/lld/trunk/test/ELF/amdgpu-kernels.s?rev=257023&view=auto" rel="noreferrer" target="_blank">http://llvm.org/viewvc/llvm-project/lld/trunk/test/ELF/amdgpu-kernels.s?rev=257023&view=auto</a><br>
==============================================================================<br>
--- lld/trunk/test/ELF/amdgpu-kernels.s (added)<br>
+++ lld/trunk/test/ELF/amdgpu-kernels.s Wed Jan 6 21:59:08 2016<br>
@@ -0,0 +1,61 @@<br>
+# RUN: llvm-mc -filetype=obj -triple=amdgcpu--amdgcn -mcpu=kaveri %s -o %t.o<br>
+# RUN: lld -flavor gnu %t.o -o %t<br>
+# RUN: llvm-readobj -sections -symbols -program-headers %t | FileCheck %s<br>
+<br>
+.hsa_code_object_version 1,0<br>
+.hsa_code_object_isa 7,0,0,"AMD","AMDGPU"<br>
+<br>
+.hsatext<br>
+.globl kernel0<br>
+.align 256<br>
+.amdgpu_hsa_kernel kernel0<br>
+kernel0:<br>
+ s_endpgm<br>
+.Lfunc_end0:<br>
+ .size kernel0, .Lfunc_end0-kernel0<br>
+<br>
+.globl kernel1<br>
+.align 256<br>
+.amdgpu_hsa_kernel kernel1<br>
+kernel1:<br>
+ s_endpgm<br>
+ s_endpgm<br>
+.Lfunc_end1:<br>
+ .size kernel1, .Lfunc_end1-kernel1<br>
+<br>
+<br>
+# CHECK: Section {<br>
+# CHECK: Name: .hsatext<br>
+# CHECK: Type: SHT_PROGBITS<br>
+# CHECK: Flags [ (0xC00007)<br>
+# CHECK: SHF_ALLOC (0x2)<br>
+# CHECK: SHF_AMDGPU_HSA_AGENT (0x800000)<br>
+# CHECK: SHF_AMDGPU_HSA_CODE (0x400000)<br>
+# CHECK: SHF_EXECINSTR (0x4)<br>
+# CHECK: SHF_WRITE (0x1)<br>
+# CHECK: ]<br>
+# CHECK: Address: [[HSATEXT_ADDR:[0-9xa-f]+]]<br>
+# CHECK: }<br>
+<br>
+# CHECK: Symbol {<br>
+# CHECK: Name: kernel0<br>
+# CHECK: Value: 0x0<br>
+# CHECK: Size: 4<br>
+# CHECK: Binding: Global<br>
+# CHECK: Type: AMDGPU_HSA_KERNEL<br>
+# CHECK: Section: .hsatext<br>
+# CHECK: }<br>
+<br>
+# CHECK: Symbol {<br>
+# CHECK: Name: kernel1<br>
+# CHECK: Value: 0x100<br>
+# CHECK: Size: 8<br>
+# CHECK: Binding: Global<br>
+# CHECK: Type: AMDGPU_HSA_KERNEL<br>
+# CHECK: Section: .hsatext<br>
+# CHECK: }<br>
+<br>
+# CHECK: ProgramHeader {<br>
+# CHECK: Type: PT_AMDGPU_HSA_LOAD_CODE_AGENT<br>
+# CHECK: VirtualAddress: [[HSATEXT_ADDR]]<br>
+# CHECK: }<br>
<br>
<br>
_______________________________________________<br>
llvm-commits mailing list<br>
<a href="mailto:llvm-commits@lists.llvm.org">llvm-commits@lists.llvm.org</a><br>
<a href="http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits" rel="noreferrer" target="_blank">http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits</a><br>
</blockquote></div><br></div>