diff options
Diffstat (limited to 'lib/Target/AMDGPU')
40 files changed, 1050 insertions, 264 deletions
diff --git a/lib/Target/AMDGPU/AMDGPU.td b/lib/Target/AMDGPU/AMDGPU.td index 2e7e39a54d33..569ad3844b25 100644 --- a/lib/Target/AMDGPU/AMDGPU.td +++ b/lib/Target/AMDGPU/AMDGPU.td @@ -141,6 +141,19 @@ class SubtargetFeatureLDSBankCount <int Value> : SubtargetFeature < def FeatureLDSBankCount16 : SubtargetFeatureLDSBankCount<16>; def FeatureLDSBankCount32 : SubtargetFeatureLDSBankCount<32>; +class SubtargetFeatureISAVersion <int Major, int Minor, int Stepping> + : SubtargetFeature < + "isaver"#Major#"."#Minor#"."#Stepping, + "IsaVersion", + "ISAVersion"#Major#"_"#Minor#"_"#Stepping, + "Instruction set version number" +>; + +def FeatureISAVersion7_0_0 : SubtargetFeatureISAVersion <7,0,0>; +def FeatureISAVersion7_0_1 : SubtargetFeatureISAVersion <7,0,1>; +def FeatureISAVersion8_0_0 : SubtargetFeatureISAVersion <8,0,0>; +def FeatureISAVersion8_0_1 : SubtargetFeatureISAVersion <8,0,1>; + class SubtargetFeatureLocalMemorySize <int Value> : SubtargetFeature< "localmemorysize"#Value, "LocalMemorySize", diff --git a/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp b/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp index afc6bcb52bb8..709d7531d38b 100644 --- a/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp +++ b/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp @@ -17,7 +17,9 @@ // #include "AMDGPUAsmPrinter.h" +#include "MCTargetDesc/AMDGPUTargetStreamer.h" #include "InstPrinter/AMDGPUInstPrinter.h" +#include "Utils/AMDGPUBaseInfo.h" #include "AMDGPU.h" #include "AMDKernelCodeT.h" #include "AMDGPUSubtarget.h" @@ -89,6 +91,15 @@ AMDGPUAsmPrinter::AMDGPUAsmPrinter(TargetMachine &TM, std::unique_ptr<MCStreamer> Streamer) : AsmPrinter(TM, std::move(Streamer)) {} +void AMDGPUAsmPrinter::EmitFunctionBodyStart() { + const AMDGPUSubtarget &STM = MF->getSubtarget<AMDGPUSubtarget>(); + SIProgramInfo KernelInfo; + if (STM.isAmdHsaOS()) { + getSIProgramInfo(KernelInfo, *MF); + EmitAmdKernelCodeT(*MF, KernelInfo); + } +} + void AMDGPUAsmPrinter::EmitEndOfAsmFile(Module &M) { // This label is used to mark the end of the .text section. @@ -113,13 +124,18 @@ bool AMDGPUAsmPrinter::runOnMachineFunction(MachineFunction &MF) { const AMDGPUSubtarget &STM = MF.getSubtarget<AMDGPUSubtarget>(); SIProgramInfo KernelInfo; - if (STM.isAmdHsaOS()) { - getSIProgramInfo(KernelInfo, MF); - EmitAmdKernelCodeT(MF, KernelInfo); - OutStreamer->EmitCodeAlignment(2 << (MF.getAlignment() - 1)); - } else if (STM.getGeneration() >= AMDGPUSubtarget::SOUTHERN_ISLANDS) { - getSIProgramInfo(KernelInfo, MF); - EmitProgramInfoSI(MF, KernelInfo); + if (STM.getGeneration() >= AMDGPUSubtarget::SOUTHERN_ISLANDS) { + if (!STM.isAmdHsaOS()) { + getSIProgramInfo(KernelInfo, MF); + EmitProgramInfoSI(MF, KernelInfo); + } + // Emit directives + AMDGPUTargetStreamer *TS = + static_cast<AMDGPUTargetStreamer *>(OutStreamer->getTargetStreamer()); + TS->EmitDirectiveHSACodeObjectVersion(1, 0); + AMDGPU::IsaVersion ISA = STM.getIsaVersion(); + TS->EmitDirectiveHSACodeObjectISA(ISA.Major, ISA.Minor, ISA.Stepping, + "AMD", "AMDGPU"); } else { EmitProgramInfoR600(MF); } @@ -459,125 +475,28 @@ void AMDGPUAsmPrinter::EmitProgramInfoSI(const MachineFunction &MF, } void AMDGPUAsmPrinter::EmitAmdKernelCodeT(const MachineFunction &MF, - const SIProgramInfo &KernelInfo) const { + const SIProgramInfo &KernelInfo) const { const SIMachineFunctionInfo *MFI = MF.getInfo<SIMachineFunctionInfo>(); const AMDGPUSubtarget &STM = MF.getSubtarget<AMDGPUSubtarget>(); amd_kernel_code_t header; - memset(&header, 0, sizeof(header)); - - header.amd_code_version_major = AMD_CODE_VERSION_MAJOR; - header.amd_code_version_minor = AMD_CODE_VERSION_MINOR; - - header.struct_byte_size = sizeof(amd_kernel_code_t); - - header.target_chip = STM.getAmdKernelCodeChipID(); - - header.kernel_code_entry_byte_offset = (1ULL << MF.getAlignment()); + AMDGPU::initDefaultAMDKernelCodeT(header, STM.getFeatureBits()); header.compute_pgm_resource_registers = KernelInfo.ComputePGMRSrc1 | (KernelInfo.ComputePGMRSrc2 << 32); + header.code_properties = + AMD_CODE_PROPERTY_ENABLE_SGPR_KERNARG_SEGMENT_PTR | + AMD_CODE_PROPERTY_IS_PTR64; - // Code Properties: - header.code_properties = AMD_CODE_PROPERTY_ENABLE_SGPR_KERNARG_SEGMENT_PTR | - AMD_CODE_PROPERTY_IS_PTR64; - - if (KernelInfo.FlatUsed) - header.code_properties |= AMD_CODE_PROPERTY_ENABLE_SGPR_FLAT_SCRATCH_INIT; - - if (KernelInfo.ScratchBlocks) - header.code_properties |= AMD_CODE_PROPERTY_ENABLE_SGPR_PRIVATE_SEGMENT_SIZE; - - header.workitem_private_segment_byte_size = KernelInfo.ScratchSize; - header.workgroup_group_segment_byte_size = KernelInfo.LDSSize; - - // MFI->ABIArgOffset is the number of bytes for the kernel arguments - // plus 36. 36 is the number of bytes reserved at the begining of the - // input buffer to store work-group size information. - // FIXME: We should be adding the size of the implicit arguments - // to this value. header.kernarg_segment_byte_size = MFI->ABIArgOffset; - header.wavefront_sgpr_count = KernelInfo.NumSGPR; header.workitem_vgpr_count = KernelInfo.NumVGPR; - // FIXME: What values do I put for these alignments - header.kernarg_segment_alignment = 0; - header.group_segment_alignment = 0; - header.private_segment_alignment = 0; - - header.code_type = 1; // HSA_EXT_CODE_KERNEL - - header.wavefront_size = STM.getWavefrontSize(); - - MCSectionELF *VersionSection = - OutContext.getELFSection(".hsa.version", ELF::SHT_PROGBITS, 0); - OutStreamer->SwitchSection(VersionSection); - OutStreamer->EmitBytes(Twine("HSA Code Unit:" + - Twine(header.hsail_version_major) + "." + - Twine(header.hsail_version_minor) + ":" + - "AMD:" + - Twine(header.amd_code_version_major) + "." + - Twine(header.amd_code_version_minor) + ":" + - "GFX8.1:0").str()); - - OutStreamer->SwitchSection(getObjFileLowering().getTextSection()); - - if (isVerbose()) { - OutStreamer->emitRawComment("amd_code_version_major = " + - Twine(header.amd_code_version_major), false); - OutStreamer->emitRawComment("amd_code_version_minor = " + - Twine(header.amd_code_version_minor), false); - OutStreamer->emitRawComment("struct_byte_size = " + - Twine(header.struct_byte_size), false); - OutStreamer->emitRawComment("target_chip = " + - Twine(header.target_chip), false); - OutStreamer->emitRawComment(" compute_pgm_rsrc1: " + - Twine::utohexstr(KernelInfo.ComputePGMRSrc1), - false); - OutStreamer->emitRawComment(" compute_pgm_rsrc2: " + - Twine::utohexstr(KernelInfo.ComputePGMRSrc2), - false); - OutStreamer->emitRawComment("enable_sgpr_private_segment_buffer = " + - Twine((bool)(header.code_properties & - AMD_CODE_PROPERTY_ENABLE_SGPR_PRIVATE_SEGMENT_SIZE)), false); - OutStreamer->emitRawComment("enable_sgpr_kernarg_segment_ptr = " + - Twine((bool)(header.code_properties & - AMD_CODE_PROPERTY_ENABLE_SGPR_KERNARG_SEGMENT_PTR)), false); - OutStreamer->emitRawComment("private_element_size = 2 ", false); - OutStreamer->emitRawComment("is_ptr64 = " + - Twine((bool)(header.code_properties & AMD_CODE_PROPERTY_IS_PTR64)), false); - OutStreamer->emitRawComment("workitem_private_segment_byte_size = " + - Twine(header.workitem_private_segment_byte_size), - false); - OutStreamer->emitRawComment("workgroup_group_segment_byte_size = " + - Twine(header.workgroup_group_segment_byte_size), - false); - OutStreamer->emitRawComment("gds_segment_byte_size = " + - Twine(header.gds_segment_byte_size), false); - OutStreamer->emitRawComment("kernarg_segment_byte_size = " + - Twine(header.kernarg_segment_byte_size), false); - OutStreamer->emitRawComment("wavefront_sgpr_count = " + - Twine(header.wavefront_sgpr_count), false); - OutStreamer->emitRawComment("workitem_vgpr_count = " + - Twine(header.workitem_vgpr_count), false); - OutStreamer->emitRawComment("code_type = " + Twine(header.code_type), false); - OutStreamer->emitRawComment("wavefront_size = " + - Twine((int)header.wavefront_size), false); - OutStreamer->emitRawComment("optimization_level = " + - Twine(header.optimization_level), false); - OutStreamer->emitRawComment("hsail_profile = " + - Twine(header.hsail_profile), false); - OutStreamer->emitRawComment("hsail_machine_model = " + - Twine(header.hsail_machine_model), false); - OutStreamer->emitRawComment("hsail_version_major = " + - Twine(header.hsail_version_major), false); - OutStreamer->emitRawComment("hsail_version_minor = " + - Twine(header.hsail_version_minor), false); - } - OutStreamer->EmitBytes(StringRef((char*)&header, sizeof(header))); + AMDGPUTargetStreamer *TS = + static_cast<AMDGPUTargetStreamer *>(OutStreamer->getTargetStreamer()); + TS->EmitAMDKernelCodeT(header); } bool AMDGPUAsmPrinter::PrintAsmOperand(const MachineInstr *MI, unsigned OpNo, diff --git a/lib/Target/AMDGPU/AMDGPUAsmPrinter.h b/lib/Target/AMDGPU/AMDGPUAsmPrinter.h index 92072512e6b5..345af9b85e15 100644 --- a/lib/Target/AMDGPU/AMDGPUAsmPrinter.h +++ b/lib/Target/AMDGPU/AMDGPUAsmPrinter.h @@ -97,6 +97,8 @@ public: /// Implemented in AMDGPUMCInstLower.cpp void EmitInstruction(const MachineInstr *MI) override; + void EmitFunctionBodyStart() override; + void EmitEndOfAsmFile(Module &M) override; bool PrintAsmOperand(const MachineInstr *MI, unsigned OpNo, @@ -108,6 +110,6 @@ protected: size_t DisasmLineMaxLen; }; -} // namespace llvm +} // End anonymous llvm #endif diff --git a/lib/Target/AMDGPU/AMDGPUISelLowering.cpp b/lib/Target/AMDGPU/AMDGPUISelLowering.cpp index 570473d85585..d56838ec2019 100644 --- a/lib/Target/AMDGPU/AMDGPUISelLowering.cpp +++ b/lib/Target/AMDGPU/AMDGPUISelLowering.cpp @@ -68,7 +68,7 @@ public: }; int DiagnosticInfoUnsupported::KindID = 0; -} // namespace +} static bool allocateStack(unsigned ValNo, MVT ValVT, MVT LocVT, diff --git a/lib/Target/AMDGPU/AMDGPUInstrInfo.h b/lib/Target/AMDGPU/AMDGPUInstrInfo.h index 31ae9a3c7760..86d3962b3856 100644 --- a/lib/Target/AMDGPU/AMDGPUInstrInfo.h +++ b/lib/Target/AMDGPU/AMDGPUInstrInfo.h @@ -198,7 +198,7 @@ namespace AMDGPU { int16_t getNamedOperandIdx(uint16_t Opcode, uint16_t NamedIndex); } // End namespace AMDGPU -} // namespace llvm +} // End llvm namespace #define AMDGPU_FLAG_REGISTER_LOAD (UINT64_C(1) << 63) #define AMDGPU_FLAG_REGISTER_STORE (UINT64_C(1) << 62) diff --git a/lib/Target/AMDGPU/AMDGPUMachineFunction.h b/lib/Target/AMDGPU/AMDGPUMachineFunction.h index e17b41ad5f21..f5e4694e76f6 100644 --- a/lib/Target/AMDGPU/AMDGPUMachineFunction.h +++ b/lib/Target/AMDGPU/AMDGPUMachineFunction.h @@ -41,5 +41,5 @@ public: bool IsKernel; }; -} // namespace llvm +} #endif diff --git a/lib/Target/AMDGPU/AMDGPUSubtarget.cpp b/lib/Target/AMDGPU/AMDGPUSubtarget.cpp index 605ccd0e1361..0779d1d786b2 100644 --- a/lib/Target/AMDGPU/AMDGPUSubtarget.cpp +++ b/lib/Target/AMDGPU/AMDGPUSubtarget.cpp @@ -72,6 +72,7 @@ AMDGPUSubtarget::AMDGPUSubtarget(const Triple &TT, StringRef GPU, StringRef FS, WavefrontSize(0), CFALUBug(false), LocalMemorySize(0), EnableVGPRSpilling(false), SGPRInitBug(false), IsGCN(false), GCN1Encoding(false), GCN3Encoding(false), CIInsts(false), LDSBankCount(0), + IsaVersion(ISAVersion0_0_0), FrameLowering(TargetFrameLowering::StackGrowsUp, 64 * 16, // Maximum stack alignment (long16) 0), @@ -109,6 +110,10 @@ unsigned AMDGPUSubtarget::getAmdKernelCodeChipID() const { } } +AMDGPU::IsaVersion AMDGPUSubtarget::getIsaVersion() const { + return AMDGPU::getIsaVersion(getFeatureBits()); +} + bool AMDGPUSubtarget::isVGPRSpillingEnabled( const SIMachineFunctionInfo *MFI) const { return MFI->getShaderType() == ShaderType::COMPUTE || EnableVGPRSpilling; @@ -131,3 +136,4 @@ void AMDGPUSubtarget::overrideSchedPolicy(MachineSchedPolicy &Policy, Policy.OnlyBottomUp = false; } } + diff --git a/lib/Target/AMDGPU/AMDGPUSubtarget.h b/lib/Target/AMDGPU/AMDGPUSubtarget.h index 0d40d14f8203..30f50eb1d2f3 100644 --- a/lib/Target/AMDGPU/AMDGPUSubtarget.h +++ b/lib/Target/AMDGPU/AMDGPUSubtarget.h @@ -20,6 +20,8 @@ #include "AMDGPUIntrinsicInfo.h" #include "AMDGPUSubtarget.h" #include "R600ISelLowering.h" +#include "AMDKernelCodeT.h" +#include "Utils/AMDGPUBaseInfo.h" #include "llvm/ADT/StringExtras.h" #include "llvm/ADT/StringRef.h" #include "llvm/Target/TargetSubtargetInfo.h" @@ -48,6 +50,14 @@ public: FIXED_SGPR_COUNT_FOR_INIT_BUG = 80 }; + enum { + ISAVersion0_0_0, + ISAVersion7_0_0, + ISAVersion7_0_1, + ISAVersion8_0_0, + ISAVersion8_0_1 + }; + private: std::string DevName; bool Is64bit; @@ -77,6 +87,7 @@ private: bool CIInsts; bool FeatureDisable; int LDSBankCount; + unsigned IsaVersion; AMDGPUFrameLowering FrameLowering; std::unique_ptr<AMDGPUTargetLowering> TLInfo; @@ -236,6 +247,8 @@ public: unsigned getAmdKernelCodeChipID() const; + AMDGPU::IsaVersion getIsaVersion() const; + bool enableMachineScheduler() const override { return true; } @@ -275,6 +288,13 @@ public: bool enableSubRegLiveness() const override { return true; } + + /// \brief Returns the offset in bytes from the start of the input buffer + /// of the first explicit kernel argument. + unsigned getExplicitKernelArgOffset() const { + return isAmdHsaOS() ? 0 : 36; + } + }; } // End namespace llvm diff --git a/lib/Target/AMDGPU/AMDKernelCodeT.h b/lib/Target/AMDGPU/AMDKernelCodeT.h index eaffb854793c..a9ba60c8cbad 100644 --- a/lib/Target/AMDGPU/AMDKernelCodeT.h +++ b/lib/Target/AMDGPU/AMDKernelCodeT.h @@ -12,9 +12,12 @@ #ifndef AMDKERNELCODET_H #define AMDKERNELCODET_H +#include "llvm/MC/SubtargetFeature.h" + #include <cstddef> #include <cstdint> +#include "llvm/Support/Debug.h" //---------------------------------------------------------------------------// // AMD Kernel Code, and its dependencies // //---------------------------------------------------------------------------// @@ -142,7 +145,7 @@ enum amd_code_property_mask_t { /// the GPU flat scratch (SH_STATIC_MEM_CONFIG.ELEMENT_SIZE). This /// is generally DWORD. /// - /// Use values from the amd_element_byte_size_t enum. + /// uSE VALUES FROM THE AMD_ELEMENT_BYTE_SIZE_T ENUM. AMD_CODE_PROPERTY_PRIVATE_ELEMENT_SIZE_SHIFT = 11, AMD_CODE_PROPERTY_PRIVATE_ELEMENT_SIZE_WIDTH = 2, AMD_CODE_PROPERTY_PRIVATE_ELEMENT_SIZE = ((1 << AMD_CODE_PROPERTY_PRIVATE_ELEMENT_SIZE_WIDTH) - 1) << AMD_CODE_PROPERTY_PRIVATE_ELEMENT_SIZE_SHIFT, @@ -171,7 +174,11 @@ enum amd_code_property_mask_t { /// Indicate if code generated has support for debugging. AMD_CODE_PROPERTY_IS_DEBUG_SUPPORTED_SHIFT = 15, AMD_CODE_PROPERTY_IS_DEBUG_SUPPORTED_WIDTH = 1, - AMD_CODE_PROPERTY_IS_DEBUG_SUPPORTED = ((1 << AMD_CODE_PROPERTY_IS_DEBUG_SUPPORTED_WIDTH) - 1) << AMD_CODE_PROPERTY_IS_DEBUG_SUPPORTED_SHIFT + AMD_CODE_PROPERTY_IS_DEBUG_SUPPORTED = ((1 << AMD_CODE_PROPERTY_IS_DEBUG_SUPPORTED_WIDTH) - 1) << AMD_CODE_PROPERTY_IS_DEBUG_SUPPORTED_SHIFT, + + AMD_CODE_PROPERTY_IS_XNACK_SUPPORTED_SHIFT = 15, + AMD_CODE_PROPERTY_IS_XNACK_SUPPORTED_WIDTH = 1, + AMD_CODE_PROPERTY_IS_XNACK_SUPPORTED = ((1 << AMD_CODE_PROPERTY_IS_XNACK_SUPPORTED_WIDTH) - 1) << AMD_CODE_PROPERTY_IS_XNACK_SUPPORTED_SHIFT }; /// @brief The hsa_ext_control_directives_t specifies the values for the HSAIL @@ -369,7 +376,7 @@ typedef struct hsa_ext_control_directives_s { /// Scratch Wave Offset must be added by the kernel code and moved to /// SGPRn-4 for use as the FLAT SCRATCH BASE in flat memory instructions. /// -/// The second SGPR is 32 bit byte size of a single work-item’s scratch +/// The second SGPR is 32 bit byte size of a single work-item's scratch /// memory usage. This is directly loaded from the dispatch packet Private /// Segment Byte Size and rounded up to a multiple of DWORD. /// @@ -385,7 +392,7 @@ typedef struct hsa_ext_control_directives_s { /// /// Private Segment Size (enable_sgpr_private_segment_size): /// Number of User SGPR registers: 1. The 32 bit byte size of a single -/// work-item’s scratch memory allocation. This is the value from the dispatch +/// work-item's scratch memory allocation. This is the value from the dispatch /// packet. Private Segment Byte Size rounded up by CP to a multiple of DWORD. /// /// \todo [Does CP need to round this to >4 byte alignment?] @@ -433,7 +440,7 @@ typedef struct hsa_ext_control_directives_s { /// present /// /// Work-Group Info (enable_sgpr_workgroup_info): -/// Number of System SGPR registers: 1. {first_wave, 14’b0000, +/// Number of System SGPR registers: 1. {first_wave, 14'b0000, /// ordered_append_term[10:0], threadgroup_size_in_waves[5:0]} /// /// Private Segment Wave Byte Offset @@ -499,25 +506,14 @@ typedef struct hsa_ext_control_directives_s { /// Alternatively scalar loads can be used if the kernarg offset is uniform, as /// the kernarg segment is constant for the duration of the kernel execution. /// -typedef struct amd_kernel_code_s { - /// The AMD major version of the Code Object. Must be the value - /// AMD_CODE_VERSION_MAJOR. - amd_code_version32_t amd_code_version_major; - /// The AMD minor version of the Code Object. Minor versions must be - /// backward compatible. Must be the value - /// AMD_CODE_VERSION_MINOR. - amd_code_version32_t amd_code_version_minor; - - /// The byte size of this struct. Must be set to - /// sizeof(amd_kernel_code_t). Used for backward - /// compatibility. - uint32_t struct_byte_size; - - /// The target chip instruction set for which code has been - /// generated. Values are from the E_SC_INSTRUCTION_SET enumeration - /// in sc/Interface/SCCommon.h. - uint32_t target_chip; +typedef struct amd_kernel_code_s { + uint32_t amd_kernel_code_version_major; + uint32_t amd_kernel_code_version_minor; + uint16_t amd_machine_kind; + uint16_t amd_machine_version_major; + uint16_t amd_machine_version_minor; + uint16_t amd_machine_version_stepping; /// Byte offset (possibly negative) from start of amd_kernel_code_t /// object to kernel's entry point instruction. The actual code for @@ -535,10 +531,6 @@ typedef struct amd_kernel_code_s { /// and size. The offset is from the start (possibly negative) of /// amd_kernel_code_t object. Set both to 0 if no prefetch /// information is available. - /// - /// \todo ttye 11/15/2013 Is the prefetch definition we want? Did - /// not make the size a uint64_t as prefetching more than 4GiB seems - /// excessive. int64_t kernel_code_prefetch_byte_offset; uint64_t kernel_code_prefetch_byte_size; @@ -553,11 +545,11 @@ typedef struct amd_kernel_code_s { /// Shader program settings for CS. Contains COMPUTE_PGM_RSRC1 and /// COMPUTE_PGM_RSRC2 registers. - amd_compute_pgm_resource_register64_t compute_pgm_resource_registers; + uint64_t compute_pgm_resource_registers; /// Code properties. See amd_code_property_mask_t for a full list of /// properties. - amd_code_property32_t code_properties; + uint32_t code_properties; /// The amount of memory required for the combined private, spill /// and arg segments for a work-item in bytes. If @@ -629,76 +621,21 @@ typedef struct amd_kernel_code_s { /// The maximum byte alignment of variables used by the kernel in /// the specified memory segment. Expressed as a power of two. Must /// be at least HSA_POWERTWO_16. - hsa_powertwo8_t kernarg_segment_alignment; - hsa_powertwo8_t group_segment_alignment; - hsa_powertwo8_t private_segment_alignment; - - uint8_t reserved3; - - /// Type of code object. - hsa_ext_code_kind32_t code_type; - - /// Reserved for code properties if any are defined in the future. - /// There are currently no code properties so this field must be 0. - uint32_t reserved4; + uint8_t kernarg_segment_alignment; + uint8_t group_segment_alignment; + uint8_t private_segment_alignment; /// Wavefront size expressed as a power of two. Must be a power of 2 /// in range 1..64 inclusive. Used to support runtime query that /// obtains wavefront size, which may be used by application to /// allocated dynamic group memory and set the dispatch work-group /// size. - hsa_powertwo8_t wavefront_size; - - /// The optimization level specified when the kernel was - /// finalized. - uint8_t optimization_level; - - /// The HSAIL profile defines which features are used. This - /// information is from the HSAIL version directive. If this - /// amd_kernel_code_t is not generated from an HSAIL compilation - /// unit then must be 0. - hsa_ext_brig_profile8_t hsail_profile; - - /// The HSAIL machine model gives the address sizes used by the - /// code. This information is from the HSAIL version directive. If - /// not generated from an HSAIL compilation unit then must still - /// indicate for what machine mode the code is generated. - hsa_ext_brig_machine_model8_t hsail_machine_model; - - /// The HSAIL major version. This information is from the HSAIL - /// version directive. If this amd_kernel_code_t is not - /// generated from an HSAIL compilation unit then must be 0. - uint32_t hsail_version_major; - - /// The HSAIL minor version. This information is from the HSAIL - /// version directive. If this amd_kernel_code_t is not - /// generated from an HSAIL compilation unit then must be 0. - uint32_t hsail_version_minor; - - /// Reserved for HSAIL target options if any are defined in the - /// future. There are currently no target options so this field - /// must be 0. - uint16_t reserved5; - - /// Reserved. Must be 0. - uint16_t reserved6; - - /// The values should be the actually values used by the finalizer - /// in generating the code. This may be the union of values - /// specified as finalizer arguments and explicit HSAIL control - /// directives. If the finalizer chooses to ignore a control - /// directive, and not generate constrained code, then the control - /// directive should not be marked as enabled even though it was - /// present in the HSAIL or finalizer argument. The values are - /// intended to reflect the constraints that the code actually - /// requires to correctly execute, not the values that were - /// actually specified at finalize time. - hsa_ext_control_directives_t control_directive; - - /// The code can immediately follow the amd_kernel_code_t, or can - /// come after subsequent amd_kernel_code_t structs when there are - /// multiple kernels in the compilation unit. + uint8_t wavefront_size; + int32_t call_convention; + uint8_t reserved3[12]; + uint64_t runtime_loader_kernel_symbol; + uint64_t control_directives[16]; } amd_kernel_code_t; #endif // AMDKERNELCODET_H diff --git a/lib/Target/AMDGPU/AsmParser/AMDGPUAsmParser.cpp b/lib/Target/AMDGPU/AsmParser/AMDGPUAsmParser.cpp index 80081d40d089..2018983bc306 100644 --- a/lib/Target/AMDGPU/AsmParser/AMDGPUAsmParser.cpp +++ b/lib/Target/AMDGPU/AsmParser/AMDGPUAsmParser.cpp @@ -8,6 +8,9 @@ //===----------------------------------------------------------------------===// #include "MCTargetDesc/AMDGPUMCTargetDesc.h" +#include "MCTargetDesc/AMDGPUTargetStreamer.h" +#include "Utils/AMDGPUBaseInfo.h" +#include "AMDKernelCodeT.h" #include "SIDefines.h" #include "llvm/ADT/APFloat.h" #include "llvm/ADT/SmallString.h" @@ -314,6 +317,13 @@ class AMDGPUAsmParser : public MCTargetAsmParser { /// } +private: + bool ParseDirectiveMajorMinor(uint32_t &Major, uint32_t &Minor); + bool ParseDirectiveHSACodeObjectVersion(); + bool ParseDirectiveHSACodeObjectISA(); + bool ParseAMDKernelCodeTValue(StringRef ID, amd_kernel_code_t &Header); + bool ParseDirectiveAMDKernelCodeT(); + public: AMDGPUAsmParser(MCSubtargetInfo &STI, MCAsmParser &_Parser, const MCInstrInfo &MII, @@ -329,6 +339,11 @@ public: setAvailableFeatures(ComputeAvailableFeatures(STI.getFeatureBits())); } + AMDGPUTargetStreamer &getTargetStreamer() { + MCTargetStreamer &TS = *getParser().getStreamer().getTargetStreamer(); + return static_cast<AMDGPUTargetStreamer &>(TS); + } + unsigned getForcedEncodingSize() const { return ForcedEncodingSize; } @@ -403,7 +418,7 @@ struct OptionalOperand { bool (*ConvertResult)(int64_t&); }; -} // namespace +} static unsigned getRegClass(bool IsVgpr, unsigned RegWidth) { if (IsVgpr) { @@ -581,7 +596,304 @@ bool AMDGPUAsmParser::MatchAndEmitInstruction(SMLoc IDLoc, unsigned &Opcode, llvm_unreachable("Implement any new match types added!"); } +bool AMDGPUAsmParser::ParseDirectiveMajorMinor(uint32_t &Major, + uint32_t &Minor) { + if (getLexer().isNot(AsmToken::Integer)) + return TokError("invalid major version"); + + Major = getLexer().getTok().getIntVal(); + Lex(); + + if (getLexer().isNot(AsmToken::Comma)) + return TokError("minor version number required, comma expected"); + Lex(); + + if (getLexer().isNot(AsmToken::Integer)) + return TokError("invalid minor version"); + + Minor = getLexer().getTok().getIntVal(); + Lex(); + + return false; +} + +bool AMDGPUAsmParser::ParseDirectiveHSACodeObjectVersion() { + + uint32_t Major; + uint32_t Minor; + + if (ParseDirectiveMajorMinor(Major, Minor)) + return true; + + getTargetStreamer().EmitDirectiveHSACodeObjectVersion(Major, Minor); + return false; +} + +bool AMDGPUAsmParser::ParseDirectiveHSACodeObjectISA() { + + uint32_t Major; + uint32_t Minor; + uint32_t Stepping; + StringRef VendorName; + StringRef ArchName; + + // If this directive has no arguments, then use the ISA version for the + // targeted GPU. + if (getLexer().is(AsmToken::EndOfStatement)) { + AMDGPU::IsaVersion Isa = AMDGPU::getIsaVersion(STI.getFeatureBits()); + getTargetStreamer().EmitDirectiveHSACodeObjectISA(Isa.Major, Isa.Minor, + Isa.Stepping, + "AMD", "AMDGPU"); + return false; + } + + + if (ParseDirectiveMajorMinor(Major, Minor)) + return true; + + if (getLexer().isNot(AsmToken::Comma)) + return TokError("stepping version number required, comma expected"); + Lex(); + + if (getLexer().isNot(AsmToken::Integer)) + return TokError("invalid stepping version"); + + Stepping = getLexer().getTok().getIntVal(); + Lex(); + + if (getLexer().isNot(AsmToken::Comma)) + return TokError("vendor name required, comma expected"); + Lex(); + + if (getLexer().isNot(AsmToken::String)) + return TokError("invalid vendor name"); + + VendorName = getLexer().getTok().getStringContents(); + Lex(); + + if (getLexer().isNot(AsmToken::Comma)) + return TokError("arch name required, comma expected"); + Lex(); + + if (getLexer().isNot(AsmToken::String)) + return TokError("invalid arch name"); + + ArchName = getLexer().getTok().getStringContents(); + Lex(); + + getTargetStreamer().EmitDirectiveHSACodeObjectISA(Major, Minor, Stepping, + VendorName, ArchName); + return false; +} + +bool AMDGPUAsmParser::ParseAMDKernelCodeTValue(StringRef ID, + amd_kernel_code_t &Header) { + + if (getLexer().isNot(AsmToken::Equal)) + return TokError("expected '='"); + Lex(); + + if (getLexer().isNot(AsmToken::Integer)) + return TokError("amd_kernel_code_t values must be integers"); + + uint64_t Value = getLexer().getTok().getIntVal(); + Lex(); + + if (ID == "kernel_code_version_major") + Header.amd_kernel_code_version_major = Value; + else if (ID == "kernel_code_version_minor") + Header.amd_kernel_code_version_minor = Value; + else if (ID == "machine_kind") + Header.amd_machine_kind = Value; + else if (ID == "machine_version_major") + Header.amd_machine_version_major = Value; + else if (ID == "machine_version_minor") + Header.amd_machine_version_minor = Value; + else if (ID == "machine_version_stepping") + Header.amd_machine_version_stepping = Value; + else if (ID == "kernel_code_entry_byte_offset") + Header.kernel_code_entry_byte_offset = Value; + else if (ID == "kernel_code_prefetch_byte_size") + Header.kernel_code_prefetch_byte_size = Value; + else if (ID == "max_scratch_backing_memory_byte_size") + Header.max_scratch_backing_memory_byte_size = Value; + else if (ID == "compute_pgm_rsrc1_vgprs") + Header.compute_pgm_resource_registers |= S_00B848_VGPRS(Value); + else if (ID == "compute_pgm_rsrc1_sgprs") + Header.compute_pgm_resource_registers |= S_00B848_SGPRS(Value); + else if (ID == "compute_pgm_rsrc1_priority") + Header.compute_pgm_resource_registers |= S_00B848_PRIORITY(Value); + else if (ID == "compute_pgm_rsrc1_float_mode") + Header.compute_pgm_resource_registers |= S_00B848_FLOAT_MODE(Value); + else if (ID == "compute_pgm_rsrc1_priv") + Header.compute_pgm_resource_registers |= S_00B848_PRIV(Value); + else if (ID == "compute_pgm_rsrc1_dx10_clamp") + Header.compute_pgm_resource_registers |= S_00B848_DX10_CLAMP(Value); + else if (ID == "compute_pgm_rsrc1_debug_mode") + Header.compute_pgm_resource_registers |= S_00B848_DEBUG_MODE(Value); + else if (ID == "compute_pgm_rsrc1_ieee_mode") + Header.compute_pgm_resource_registers |= S_00B848_IEEE_MODE(Value); + else if (ID == "compute_pgm_rsrc2_scratch_en") + Header.compute_pgm_resource_registers |= (S_00B84C_SCRATCH_EN(Value) << 32); + else if (ID == "compute_pgm_rsrc2_user_sgpr") + Header.compute_pgm_resource_registers |= (S_00B84C_USER_SGPR(Value) << 32); + else if (ID == "compute_pgm_rsrc2_tgid_x_en") + Header.compute_pgm_resource_registers |= (S_00B84C_TGID_X_EN(Value) << 32); + else if (ID == "compute_pgm_rsrc2_tgid_y_en") + Header.compute_pgm_resource_registers |= (S_00B84C_TGID_Y_EN(Value) << 32); + else if (ID == "compute_pgm_rsrc2_tgid_z_en") + Header.compute_pgm_resource_registers |= (S_00B84C_TGID_Z_EN(Value) << 32); + else if (ID == "compute_pgm_rsrc2_tg_size_en") + Header.compute_pgm_resource_registers |= (S_00B84C_TG_SIZE_EN(Value) << 32); + else if (ID == "compute_pgm_rsrc2_tidig_comp_cnt") + Header.compute_pgm_resource_registers |= + (S_00B84C_TIDIG_COMP_CNT(Value) << 32); + else if (ID == "compute_pgm_rsrc2_excp_en_msb") + Header.compute_pgm_resource_registers |= + (S_00B84C_EXCP_EN_MSB(Value) << 32); + else if (ID == "compute_pgm_rsrc2_lds_size") + Header.compute_pgm_resource_registers |= (S_00B84C_LDS_SIZE(Value) << 32); + else if (ID == "compute_pgm_rsrc2_excp_en") + Header.compute_pgm_resource_registers |= (S_00B84C_EXCP_EN(Value) << 32); + else if (ID == "compute_pgm_resource_registers") + Header.compute_pgm_resource_registers = Value; + else if (ID == "enable_sgpr_private_segment_buffer") + Header.code_properties |= + (Value << AMD_CODE_PROPERTY_ENABLE_SGPR_PRIVATE_SEGMENT_BUFFER_SHIFT); + else if (ID == "enable_sgpr_dispatch_ptr") + Header.code_properties |= + (Value << AMD_CODE_PROPERTY_ENABLE_SGPR_DISPATCH_PTR_SHIFT); + else if (ID == "enable_sgpr_queue_ptr") + Header.code_properties |= + (Value << AMD_CODE_PROPERTY_ENABLE_SGPR_QUEUE_PTR_SHIFT); + else if (ID == "enable_sgpr_kernarg_segment_ptr") + Header.code_properties |= + (Value << AMD_CODE_PROPERTY_ENABLE_SGPR_KERNARG_SEGMENT_PTR_SHIFT); + else if (ID == "enable_sgpr_dispatch_id") + Header.code_properties |= + (Value << AMD_CODE_PROPERTY_ENABLE_SGPR_DISPATCH_ID_SHIFT); + else if (ID == "enable_sgpr_flat_scratch_init") + Header.code_properties |= + (Value << AMD_CODE_PROPERTY_ENABLE_SGPR_FLAT_SCRATCH_INIT_SHIFT); + else if (ID == "enable_sgpr_private_segment_size") + Header.code_properties |= + (Value << AMD_CODE_PROPERTY_ENABLE_SGPR_PRIVATE_SEGMENT_SIZE_SHIFT); + else if (ID == "enable_sgpr_grid_workgroup_count_x") + Header.code_properties |= + (Value << AMD_CODE_PROPERTY_ENABLE_SGPR_GRID_WORKGROUP_COUNT_X_SHIFT); + else if (ID == "enable_sgpr_grid_workgroup_count_y") + Header.code_properties |= + (Value << AMD_CODE_PROPERTY_ENABLE_SGPR_GRID_WORKGROUP_COUNT_Y_SHIFT); + else if (ID == "enable_sgpr_grid_workgroup_count_z") + Header.code_properties |= + (Value << AMD_CODE_PROPERTY_ENABLE_SGPR_GRID_WORKGROUP_COUNT_Z_SHIFT); + else if (ID == "enable_ordered_append_gds") + Header.code_properties |= + (Value << AMD_CODE_PROPERTY_ENABLE_ORDERED_APPEND_GDS_SHIFT); + else if (ID == "private_element_size") + Header.code_properties |= + (Value << AMD_CODE_PROPERTY_PRIVATE_ELEMENT_SIZE_SHIFT); + else if (ID == "is_ptr64") + Header.code_properties |= + (Value << AMD_CODE_PROPERTY_IS_PTR64_SHIFT); + else if (ID == "is_dynamic_callstack") + Header.code_properties |= + (Value << AMD_CODE_PROPERTY_IS_DYNAMIC_CALLSTACK_SHIFT); + else if (ID == "is_debug_enabled") + Header.code_properties |= + (Value << AMD_CODE_PROPERTY_IS_DEBUG_SUPPORTED_SHIFT); + else if (ID == "is_xnack_enabled") + Header.code_properties |= + (Value << AMD_CODE_PROPERTY_IS_XNACK_SUPPORTED_SHIFT); + else if (ID == "workitem_private_segment_byte_size") + Header.workitem_private_segment_byte_size = Value; + else if (ID == "workgroup_group_segment_byte_size") + Header.workgroup_group_segment_byte_size = Value; + else if (ID == "gds_segment_byte_size") + Header.gds_segment_byte_size = Value; + else if (ID == "kernarg_segment_byte_size") + Header.kernarg_segment_byte_size = Value; + else if (ID == "workgroup_fbarrier_count") + Header.workgroup_fbarrier_count = Value; + else if (ID == "wavefront_sgpr_count") + Header.wavefront_sgpr_count = Value; + else if (ID == "workitem_vgpr_count") + Header.workitem_vgpr_count = Value; + else if (ID == "reserved_vgpr_first") + Header.reserved_vgpr_first = Value; + else if (ID == "reserved_vgpr_count") + Header.reserved_vgpr_count = Value; + else if (ID == "reserved_sgpr_first") + Header.reserved_sgpr_first = Value; + else if (ID == "reserved_sgpr_count") + Header.reserved_sgpr_count = Value; + else if (ID == "debug_wavefront_private_segment_offset_sgpr") + Header.debug_wavefront_private_segment_offset_sgpr = Value; + else if (ID == "debug_private_segment_buffer_sgpr") + Header.debug_private_segment_buffer_sgpr = Value; + else if (ID == "kernarg_segment_alignment") + Header.kernarg_segment_alignment = Value; + else if (ID == "group_segment_alignment") + Header.group_segment_alignment = Value; + else if (ID == "private_segment_alignment") + Header.private_segment_alignment = Value; + else if (ID == "wavefront_size") + Header.wavefront_size = Value; + else if (ID == "call_convention") + Header.call_convention = Value; + else if (ID == "runtime_loader_kernel_symbol") + Header.runtime_loader_kernel_symbol = Value; + else + return TokError("amd_kernel_code_t value not recognized."); + + return false; +} + +bool AMDGPUAsmParser::ParseDirectiveAMDKernelCodeT() { + + amd_kernel_code_t Header; + AMDGPU::initDefaultAMDKernelCodeT(Header, STI.getFeatureBits()); + + while (true) { + + if (getLexer().isNot(AsmToken::EndOfStatement)) + return TokError("amd_kernel_code_t values must begin on a new line"); + + // Lex EndOfStatement. This is in a while loop, because lexing a comment + // will set the current token to EndOfStatement. + while(getLexer().is(AsmToken::EndOfStatement)) + Lex(); + + if (getLexer().isNot(AsmToken::Identifier)) + return TokError("expected value identifier or .end_amd_kernel_code_t"); + + StringRef ID = getLexer().getTok().getIdentifier(); + Lex(); + + if (ID == ".end_amd_kernel_code_t") + break; + + if (ParseAMDKernelCodeTValue(ID, Header)) + return true; + } + + getTargetStreamer().EmitAMDKernelCodeT(Header); + + return false; +} + bool AMDGPUAsmParser::ParseDirective(AsmToken DirectiveID) { + StringRef IDVal = DirectiveID.getString(); + + if (IDVal == ".hsa_code_object_version") + return ParseDirectiveHSACodeObjectVersion(); + + if (IDVal == ".hsa_code_object_isa") + return ParseDirectiveHSACodeObjectISA(); + + if (IDVal == ".amd_kernel_code_t") + return ParseDirectiveAMDKernelCodeT(); + return true; } diff --git a/lib/Target/AMDGPU/AsmParser/LLVMBuild.txt b/lib/Target/AMDGPU/AsmParser/LLVMBuild.txt index 63d44d1e06f1..dab0c6f585af 100644 --- a/lib/Target/AMDGPU/AsmParser/LLVMBuild.txt +++ b/lib/Target/AMDGPU/AsmParser/LLVMBuild.txt @@ -19,5 +19,5 @@ type = Library name = AMDGPUAsmParser parent = AMDGPU -required_libraries = MC MCParser AMDGPUDesc AMDGPUInfo Support +required_libraries = MC MCParser AMDGPUDesc AMDGPUInfo AMDGPUUtils Support add_to_library_groups = AMDGPU diff --git a/lib/Target/AMDGPU/CMakeLists.txt b/lib/Target/AMDGPU/CMakeLists.txt index 3e5ff1f3c6d4..9460bf6b9338 100644 --- a/lib/Target/AMDGPU/CMakeLists.txt +++ b/lib/Target/AMDGPU/CMakeLists.txt @@ -62,3 +62,4 @@ add_subdirectory(AsmParser) add_subdirectory(InstPrinter) add_subdirectory(TargetInfo) add_subdirectory(MCTargetDesc) +add_subdirectory(Utils) diff --git a/lib/Target/AMDGPU/LLVMBuild.txt b/lib/Target/AMDGPU/LLVMBuild.txt index c6861df91ed6..38c5489586f1 100644 --- a/lib/Target/AMDGPU/LLVMBuild.txt +++ b/lib/Target/AMDGPU/LLVMBuild.txt @@ -16,7 +16,7 @@ ;===------------------------------------------------------------------------===; [common] -subdirectories = AsmParser InstPrinter MCTargetDesc TargetInfo +subdirectories = AsmParser InstPrinter MCTargetDesc TargetInfo Utils [component_0] type = TargetGroup @@ -29,5 +29,5 @@ has_asmprinter = 1 type = Library name = AMDGPUCodeGen parent = AMDGPU -required_libraries = Analysis AsmPrinter CodeGen Core IPO MC AMDGPUAsmParser AMDGPUAsmPrinter AMDGPUDesc AMDGPUInfo Scalar SelectionDAG Support Target TransformUtils +required_libraries = Analysis AsmPrinter CodeGen Core IPO MC AMDGPUAsmParser AMDGPUAsmPrinter AMDGPUDesc AMDGPUInfo AMDGPUUtils Scalar SelectionDAG Support Target TransformUtils add_to_library_groups = AMDGPU diff --git a/lib/Target/AMDGPU/MCTargetDesc/AMDGPUAsmBackend.cpp b/lib/Target/AMDGPU/MCTargetDesc/AMDGPUAsmBackend.cpp index 8bed2deef4cd..468563c44982 100644 --- a/lib/Target/AMDGPU/MCTargetDesc/AMDGPUAsmBackend.cpp +++ b/lib/Target/AMDGPU/MCTargetDesc/AMDGPUAsmBackend.cpp @@ -127,11 +127,14 @@ bool AMDGPUAsmBackend::writeNopData(uint64_t Count, MCObjectWriter *OW) const { namespace { class ELFAMDGPUAsmBackend : public AMDGPUAsmBackend { + bool Is64Bit; + public: - ELFAMDGPUAsmBackend(const Target &T) : AMDGPUAsmBackend(T) { } + ELFAMDGPUAsmBackend(const Target &T, bool Is64Bit) : + AMDGPUAsmBackend(T), Is64Bit(Is64Bit) { } MCObjectWriter *createObjectWriter(raw_pwrite_stream &OS) const override { - return createAMDGPUELFObjectWriter(OS); + return createAMDGPUELFObjectWriter(Is64Bit, OS); } }; @@ -140,5 +143,8 @@ public: MCAsmBackend *llvm::createAMDGPUAsmBackend(const Target &T, const MCRegisterInfo &MRI, const Triple &TT, StringRef CPU) { - return new ELFAMDGPUAsmBackend(T); + Triple TargetTriple(TT); + + // Use 64-bit ELF for amdgcn + return new ELFAMDGPUAsmBackend(T, TargetTriple.getArch() == Triple::amdgcn); } diff --git a/lib/Target/AMDGPU/MCTargetDesc/AMDGPUELFObjectWriter.cpp b/lib/Target/AMDGPU/MCTargetDesc/AMDGPUELFObjectWriter.cpp index 59f45ff02d88..820f17df8960 100644 --- a/lib/Target/AMDGPU/MCTargetDesc/AMDGPUELFObjectWriter.cpp +++ b/lib/Target/AMDGPU/MCTargetDesc/AMDGPUELFObjectWriter.cpp @@ -18,7 +18,7 @@ namespace { class AMDGPUELFObjectWriter : public MCELFObjectTargetWriter { public: - AMDGPUELFObjectWriter(); + AMDGPUELFObjectWriter(bool Is64Bit); protected: unsigned GetRelocType(const MCValue &Target, const MCFixup &Fixup, bool IsPCRel) const override { @@ -30,10 +30,11 @@ protected: } // End anonymous namespace -AMDGPUELFObjectWriter::AMDGPUELFObjectWriter() - : MCELFObjectTargetWriter(false, 0, 0, false) { } +AMDGPUELFObjectWriter::AMDGPUELFObjectWriter(bool Is64Bit) + : MCELFObjectTargetWriter(Is64Bit, ELF::ELFOSABI_AMDGPU_HSA, + ELF::EM_AMDGPU, false) { } -MCObjectWriter *llvm::createAMDGPUELFObjectWriter(raw_pwrite_stream &OS) { - MCELFObjectTargetWriter *MOTW = new AMDGPUELFObjectWriter(); +MCObjectWriter *llvm::createAMDGPUELFObjectWriter(bool Is64Bit, raw_pwrite_stream &OS) { + MCELFObjectTargetWriter *MOTW = new AMDGPUELFObjectWriter(Is64Bit); return createELFObjectWriter(MOTW, OS, true); } diff --git a/lib/Target/AMDGPU/MCTargetDesc/AMDGPUFixupKinds.h b/lib/Target/AMDGPU/MCTargetDesc/AMDGPUFixupKinds.h index fa3b3c3d9489..01021d67ffd9 100644 --- a/lib/Target/AMDGPU/MCTargetDesc/AMDGPUFixupKinds.h +++ b/lib/Target/AMDGPU/MCTargetDesc/AMDGPUFixupKinds.h @@ -28,7 +28,7 @@ enum Fixups { LastTargetFixupKind, NumTargetFixupKinds = LastTargetFixupKind - FirstTargetFixupKind }; -} // namespace AMDGPU -} // namespace llvm +} +} #endif diff --git a/lib/Target/AMDGPU/MCTargetDesc/AMDGPUMCTargetDesc.cpp b/lib/Target/AMDGPU/MCTargetDesc/AMDGPUMCTargetDesc.cpp index a7d3dd1345f9..7172e4bb9335 100644 --- a/lib/Target/AMDGPU/MCTargetDesc/AMDGPUMCTargetDesc.cpp +++ b/lib/Target/AMDGPU/MCTargetDesc/AMDGPUMCTargetDesc.cpp @@ -14,6 +14,7 @@ #include "AMDGPUMCTargetDesc.h" #include "AMDGPUMCAsmInfo.h" +#include "AMDGPUTargetStreamer.h" #include "InstPrinter/AMDGPUInstPrinter.h" #include "SIDefines.h" #include "llvm/MC/MCCodeGenInfo.h" @@ -72,6 +73,19 @@ static MCInstPrinter *createAMDGPUMCInstPrinter(const Triple &T, return new AMDGPUInstPrinter(MAI, MII, MRI); } +static MCTargetStreamer *createAMDGPUAsmTargetStreamer(MCStreamer &S, + formatted_raw_ostream &OS, + MCInstPrinter *InstPrint, + bool isVerboseAsm) { + return new AMDGPUTargetAsmStreamer(S, OS); +} + +static MCTargetStreamer * createAMDGPUObjectTargetStreamer( + MCStreamer &S, + const MCSubtargetInfo &STI) { + return new AMDGPUTargetELFStreamer(S); +} + extern "C" void LLVMInitializeAMDGPUTargetMC() { for (Target *T : {&TheAMDGPUTarget, &TheGCNTarget}) { RegisterMCAsmInfo<AMDGPUMCAsmInfo> X(*T); @@ -84,7 +98,15 @@ extern "C" void LLVMInitializeAMDGPUTargetMC() { TargetRegistry::RegisterMCAsmBackend(*T, createAMDGPUAsmBackend); } + // R600 specific registration TargetRegistry::RegisterMCCodeEmitter(TheAMDGPUTarget, createR600MCCodeEmitter); + + // GCN specific registration TargetRegistry::RegisterMCCodeEmitter(TheGCNTarget, createSIMCCodeEmitter); + + TargetRegistry::RegisterAsmTargetStreamer(TheGCNTarget, + createAMDGPUAsmTargetStreamer); + TargetRegistry::RegisterObjectTargetStreamer(TheGCNTarget, + createAMDGPUObjectTargetStreamer); } diff --git a/lib/Target/AMDGPU/MCTargetDesc/AMDGPUMCTargetDesc.h b/lib/Target/AMDGPU/MCTargetDesc/AMDGPUMCTargetDesc.h index ac611b862a1a..5d1b86b8c0c2 100644 --- a/lib/Target/AMDGPU/MCTargetDesc/AMDGPUMCTargetDesc.h +++ b/lib/Target/AMDGPU/MCTargetDesc/AMDGPUMCTargetDesc.h @@ -46,8 +46,9 @@ MCCodeEmitter *createSIMCCodeEmitter(const MCInstrInfo &MCII, MCAsmBackend *createAMDGPUAsmBackend(const Target &T, const MCRegisterInfo &MRI, const Triple &TT, StringRef CPU); -MCObjectWriter *createAMDGPUELFObjectWriter(raw_pwrite_stream &OS); -} // namespace llvm +MCObjectWriter *createAMDGPUELFObjectWriter(bool Is64Bit, + raw_pwrite_stream &OS); +} // End llvm namespace #define GET_REGINFO_ENUM #include "AMDGPUGenRegisterInfo.inc" diff --git a/lib/Target/AMDGPU/MCTargetDesc/AMDGPUTargetStreamer.cpp b/lib/Target/AMDGPU/MCTargetDesc/AMDGPUTargetStreamer.cpp new file mode 100644 index 000000000000..09e6cb1f1ffc --- /dev/null +++ b/lib/Target/AMDGPU/MCTargetDesc/AMDGPUTargetStreamer.cpp @@ -0,0 +1,297 @@ +//===-- AMDGPUTargetStreamer.cpp - Mips Target Streamer Methods -----------===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +// +// This file provides AMDGPU specific target streamer methods. +// +//===----------------------------------------------------------------------===// + +#include "AMDGPUTargetStreamer.h" +#include "SIDefines.h" +#include "llvm/ADT/Twine.h" +#include "llvm/MC/MCContext.h" +#include "llvm/MC/MCELFStreamer.h" +#include "llvm/MC/MCObjectFileInfo.h" +#include "llvm/MC/MCSectionELF.h" +#include "llvm/Support/ELF.h" +#include "llvm/Support/FormattedStream.h" + +using namespace llvm; + +AMDGPUTargetStreamer::AMDGPUTargetStreamer(MCStreamer &S) + : MCTargetStreamer(S) { } + +//===----------------------------------------------------------------------===// +// AMDGPUTargetAsmStreamer +//===----------------------------------------------------------------------===// + +AMDGPUTargetAsmStreamer::AMDGPUTargetAsmStreamer(MCStreamer &S, + formatted_raw_ostream &OS) + : AMDGPUTargetStreamer(S), OS(OS) { } + +void +AMDGPUTargetAsmStreamer::EmitDirectiveHSACodeObjectVersion(uint32_t Major, + uint32_t Minor) { + OS << "\t.hsa_code_object_version " << + Twine(Major) << "," << Twine(Minor) << '\n'; +} + +void +AMDGPUTargetAsmStreamer::EmitDirectiveHSACodeObjectISA(uint32_t Major, + uint32_t Minor, + uint32_t Stepping, + StringRef VendorName, + StringRef ArchName) { + OS << "\t.hsa_code_object_isa " << + Twine(Major) << "," << Twine(Minor) << "," << Twine(Stepping) << + ",\"" << VendorName << "\",\"" << ArchName << "\"\n"; + +} + +void +AMDGPUTargetAsmStreamer::EmitAMDKernelCodeT(const amd_kernel_code_t &Header) { + uint64_t ComputePgmRsrc2 = (Header.compute_pgm_resource_registers >> 32); + bool EnableSGPRPrivateSegmentBuffer = (Header.code_properties & + AMD_CODE_PROPERTY_ENABLE_SGPR_PRIVATE_SEGMENT_BUFFER); + bool EnableSGPRDispatchPtr = (Header.code_properties & + AMD_CODE_PROPERTY_ENABLE_SGPR_DISPATCH_PTR); + bool EnableSGPRQueuePtr = (Header.code_properties & + AMD_CODE_PROPERTY_ENABLE_SGPR_QUEUE_PTR); + bool EnableSGPRKernargSegmentPtr = (Header.code_properties & + AMD_CODE_PROPERTY_ENABLE_SGPR_KERNARG_SEGMENT_PTR); + bool EnableSGPRDispatchID = (Header.code_properties & + AMD_CODE_PROPERTY_ENABLE_SGPR_DISPATCH_ID); + bool EnableSGPRFlatScratchInit = (Header.code_properties & + AMD_CODE_PROPERTY_ENABLE_SGPR_FLAT_SCRATCH_INIT); + bool EnableSGPRPrivateSegmentSize = (Header.code_properties & + AMD_CODE_PROPERTY_ENABLE_SGPR_PRIVATE_SEGMENT_SIZE); + bool EnableSGPRGridWorkgroupCountX = (Header.code_properties & + AMD_CODE_PROPERTY_ENABLE_SGPR_GRID_WORKGROUP_COUNT_X); + bool EnableSGPRGridWorkgroupCountY = (Header.code_properties & + AMD_CODE_PROPERTY_ENABLE_SGPR_GRID_WORKGROUP_COUNT_Y); + bool EnableSGPRGridWorkgroupCountZ = (Header.code_properties & + AMD_CODE_PROPERTY_ENABLE_SGPR_GRID_WORKGROUP_COUNT_Z); + bool EnableOrderedAppendGDS = (Header.code_properties & + AMD_CODE_PROPERTY_ENABLE_ORDERED_APPEND_GDS); + uint32_t PrivateElementSize = (Header.code_properties & + AMD_CODE_PROPERTY_PRIVATE_ELEMENT_SIZE) >> + AMD_CODE_PROPERTY_PRIVATE_ELEMENT_SIZE_SHIFT; + bool IsPtr64 = (Header.code_properties & AMD_CODE_PROPERTY_IS_PTR64); + bool IsDynamicCallstack = (Header.code_properties & + AMD_CODE_PROPERTY_IS_DYNAMIC_CALLSTACK); + bool IsDebugEnabled = (Header.code_properties & + AMD_CODE_PROPERTY_IS_DEBUG_SUPPORTED); + bool IsXNackEnabled = (Header.code_properties & + AMD_CODE_PROPERTY_IS_XNACK_SUPPORTED); + + OS << "\t.amd_kernel_code_t\n" << + "\t\tkernel_code_version_major = " << + Header.amd_kernel_code_version_major << '\n' << + "\t\tkernel_code_version_minor = " << + Header.amd_kernel_code_version_minor << '\n' << + "\t\tmachine_kind = " << + Header.amd_machine_kind << '\n' << + "\t\tmachine_version_major = " << + Header.amd_machine_version_major << '\n' << + "\t\tmachine_version_minor = " << + Header.amd_machine_version_minor << '\n' << + "\t\tmachine_version_stepping = " << + Header.amd_machine_version_stepping << '\n' << + "\t\tkernel_code_entry_byte_offset = " << + Header.kernel_code_entry_byte_offset << '\n' << + "\t\tkernel_code_prefetch_byte_size = " << + Header.kernel_code_prefetch_byte_size << '\n' << + "\t\tmax_scratch_backing_memory_byte_size = " << + Header.max_scratch_backing_memory_byte_size << '\n' << + "\t\tcompute_pgm_rsrc1_vgprs = " << + G_00B848_VGPRS(Header.compute_pgm_resource_registers) << '\n' << + "\t\tcompute_pgm_rsrc1_sgprs = " << + G_00B848_SGPRS(Header.compute_pgm_resource_registers) << '\n' << + "\t\tcompute_pgm_rsrc1_priority = " << + G_00B848_PRIORITY(Header.compute_pgm_resource_registers) << '\n' << + "\t\tcompute_pgm_rsrc1_float_mode = " << + G_00B848_FLOAT_MODE(Header.compute_pgm_resource_registers) << '\n' << + "\t\tcompute_pgm_rsrc1_priv = " << + G_00B848_PRIV(Header.compute_pgm_resource_registers) << '\n' << + "\t\tcompute_pgm_rsrc1_dx10_clamp = " << + G_00B848_DX10_CLAMP(Header.compute_pgm_resource_registers) << '\n' << + "\t\tcompute_pgm_rsrc1_debug_mode = " << + G_00B848_DEBUG_MODE(Header.compute_pgm_resource_registers) << '\n' << + "\t\tcompute_pgm_rsrc1_ieee_mode = " << + G_00B848_IEEE_MODE(Header.compute_pgm_resource_registers) << '\n' << + "\t\tcompute_pgm_rsrc2_scratch_en = " << + G_00B84C_SCRATCH_EN(ComputePgmRsrc2) << '\n' << + "\t\tcompute_pgm_rsrc2_user_sgpr = " << + G_00B84C_USER_SGPR(ComputePgmRsrc2) << '\n' << + "\t\tcompute_pgm_rsrc2_tgid_x_en = " << + G_00B84C_TGID_X_EN(ComputePgmRsrc2) << '\n' << + "\t\tcompute_pgm_rsrc2_tgid_y_en = " << + G_00B84C_TGID_Y_EN(ComputePgmRsrc2) << '\n' << + "\t\tcompute_pgm_rsrc2_tgid_z_en = " << + G_00B84C_TGID_Z_EN(ComputePgmRsrc2) << '\n' << + "\t\tcompute_pgm_rsrc2_tg_size_en = " << + G_00B84C_TG_SIZE_EN(ComputePgmRsrc2) << '\n' << + "\t\tcompute_pgm_rsrc2_tidig_comp_cnt = " << + G_00B84C_TIDIG_COMP_CNT(ComputePgmRsrc2) << '\n' << + "\t\tcompute_pgm_rsrc2_excp_en_msb = " << + G_00B84C_EXCP_EN_MSB(ComputePgmRsrc2) << '\n' << + "\t\tcompute_pgm_rsrc2_lds_size = " << + G_00B84C_LDS_SIZE(ComputePgmRsrc2) << '\n' << + "\t\tcompute_pgm_rsrc2_excp_en = " << + G_00B84C_EXCP_EN(ComputePgmRsrc2) << '\n' << + + "\t\tenable_sgpr_private_segment_buffer = " << + EnableSGPRPrivateSegmentBuffer << '\n' << + "\t\tenable_sgpr_dispatch_ptr = " << + EnableSGPRDispatchPtr << '\n' << + "\t\tenable_sgpr_queue_ptr = " << + EnableSGPRQueuePtr << '\n' << + "\t\tenable_sgpr_kernarg_segment_ptr = " << + EnableSGPRKernargSegmentPtr << '\n' << + "\t\tenable_sgpr_dispatch_id = " << + EnableSGPRDispatchID << '\n' << + "\t\tenable_sgpr_flat_scratch_init = " << + EnableSGPRFlatScratchInit << '\n' << + "\t\tenable_sgpr_private_segment_size = " << + EnableSGPRPrivateSegmentSize << '\n' << + "\t\tenable_sgpr_grid_workgroup_count_x = " << + EnableSGPRGridWorkgroupCountX << '\n' << + "\t\tenable_sgpr_grid_workgroup_count_y = " << + EnableSGPRGridWorkgroupCountY << '\n' << + "\t\tenable_sgpr_grid_workgroup_count_z = " << + EnableSGPRGridWorkgroupCountZ << '\n' << + "\t\tenable_ordered_append_gds = " << + EnableOrderedAppendGDS << '\n' << + "\t\tprivate_element_size = " << + PrivateElementSize << '\n' << + "\t\tis_ptr64 = " << + IsPtr64 << '\n' << + "\t\tis_dynamic_callstack = " << + IsDynamicCallstack << '\n' << + "\t\tis_debug_enabled = " << + IsDebugEnabled << '\n' << + "\t\tis_xnack_enabled = " << + IsXNackEnabled << '\n' << + "\t\tworkitem_private_segment_byte_size = " << + Header.workitem_private_segment_byte_size << '\n' << + "\t\tworkgroup_group_segment_byte_size = " << + Header.workgroup_group_segment_byte_size << '\n' << + "\t\tgds_segment_byte_size = " << + Header.gds_segment_byte_size << '\n' << + "\t\tkernarg_segment_byte_size = " << + Header.kernarg_segment_byte_size << '\n' << + "\t\tworkgroup_fbarrier_count = " << + Header.workgroup_fbarrier_count << '\n' << + "\t\twavefront_sgpr_count = " << + Header.wavefront_sgpr_count << '\n' << + "\t\tworkitem_vgpr_count = " << + Header.workitem_vgpr_count << '\n' << + "\t\treserved_vgpr_first = " << + Header.reserved_vgpr_first << '\n' << + "\t\treserved_vgpr_count = " << + Header.reserved_vgpr_count << '\n' << + "\t\treserved_sgpr_first = " << + Header.reserved_sgpr_first << '\n' << + "\t\treserved_sgpr_count = " << + Header.reserved_sgpr_count << '\n' << + "\t\tdebug_wavefront_private_segment_offset_sgpr = " << + Header.debug_wavefront_private_segment_offset_sgpr << '\n' << + "\t\tdebug_private_segment_buffer_sgpr = " << + Header.debug_private_segment_buffer_sgpr << '\n' << + "\t\tkernarg_segment_alignment = " << + (uint32_t)Header.kernarg_segment_alignment << '\n' << + "\t\tgroup_segment_alignment = " << + (uint32_t)Header.group_segment_alignment << '\n' << + "\t\tprivate_segment_alignment = " << + (uint32_t)Header.private_segment_alignment << '\n' << + "\t\twavefront_size = " << + (uint32_t)Header.wavefront_size << '\n' << + "\t\tcall_convention = " << + Header.call_convention << '\n' << + "\t\truntime_loader_kernel_symbol = " << + Header.runtime_loader_kernel_symbol << '\n' << + // TODO: control_directives + "\t.end_amd_kernel_code_t\n"; + +} + +//===----------------------------------------------------------------------===// +// AMDGPUTargetELFStreamer +//===----------------------------------------------------------------------===// + +AMDGPUTargetELFStreamer::AMDGPUTargetELFStreamer(MCStreamer &S) + : AMDGPUTargetStreamer(S), Streamer(S) { } + +MCELFStreamer &AMDGPUTargetELFStreamer::getStreamer() { + return static_cast<MCELFStreamer &>(Streamer); +} + +void +AMDGPUTargetELFStreamer::EmitDirectiveHSACodeObjectVersion(uint32_t Major, + uint32_t Minor) { + MCStreamer &OS = getStreamer(); + MCSectionELF *Note = OS.getContext().getELFSection(".note", ELF::SHT_NOTE, 0); + + unsigned NameSZ = 4; + + OS.PushSection(); + OS.SwitchSection(Note); + OS.EmitIntValue(NameSZ, 4); // namesz + OS.EmitIntValue(8, 4); // descz + OS.EmitIntValue(NT_AMDGPU_HSA_CODE_OBJECT_VERSION, 4); // type + OS.EmitBytes(StringRef("AMD", NameSZ)); // name + OS.EmitIntValue(Major, 4); // desc + OS.EmitIntValue(Minor, 4); + OS.EmitValueToAlignment(4); + OS.PopSection(); +} + +void +AMDGPUTargetELFStreamer::EmitDirectiveHSACodeObjectISA(uint32_t Major, + uint32_t Minor, + uint32_t Stepping, + StringRef VendorName, + StringRef ArchName) { + MCStreamer &OS = getStreamer(); + MCSectionELF *Note = OS.getContext().getELFSection(".note", ELF::SHT_NOTE, 0); + + unsigned NameSZ = 4; + uint16_t VendorNameSize = VendorName.size() + 1; + uint16_t ArchNameSize = ArchName.size() + 1; + unsigned DescSZ = sizeof(VendorNameSize) + sizeof(ArchNameSize) + + sizeof(Major) + sizeof(Minor) + sizeof(Stepping) + + VendorNameSize + ArchNameSize; + + OS.PushSection(); + OS.SwitchSection(Note); + OS.EmitIntValue(NameSZ, 4); // namesz + OS.EmitIntValue(DescSZ, 4); // descsz + OS.EmitIntValue(NT_AMDGPU_HSA_ISA, 4); // type + OS.EmitBytes(StringRef("AMD", 4)); // name + OS.EmitIntValue(VendorNameSize, 2); // desc + OS.EmitIntValue(ArchNameSize, 2); + OS.EmitIntValue(Major, 4); + OS.EmitIntValue(Minor, 4); + OS.EmitIntValue(Stepping, 4); + OS.EmitBytes(VendorName); + OS.EmitIntValue(0, 1); // NULL terminate VendorName + OS.EmitBytes(ArchName); + OS.EmitIntValue(0, 1); // NULL terminte ArchName + OS.EmitValueToAlignment(4); + OS.PopSection(); +} + +void +AMDGPUTargetELFStreamer::EmitAMDKernelCodeT(const amd_kernel_code_t &Header) { + + MCStreamer &OS = getStreamer(); + OS.PushSection(); + OS.SwitchSection(OS.getContext().getObjectFileInfo()->getTextSection()); + OS.EmitBytes(StringRef((const char*)&Header, sizeof(Header))); + OS.PopSection(); +} diff --git a/lib/Target/AMDGPU/MCTargetDesc/AMDGPUTargetStreamer.h b/lib/Target/AMDGPU/MCTargetDesc/AMDGPUTargetStreamer.h new file mode 100644 index 000000000000..d37677c6b863 --- /dev/null +++ b/lib/Target/AMDGPU/MCTargetDesc/AMDGPUTargetStreamer.h @@ -0,0 +1,77 @@ +//===-- AMDGPUTargetStreamer.h - AMDGPU Target Streamer --------*- C++ -*--===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// + +#include "AMDKernelCodeT.h" +#include "llvm/MC/MCStreamer.h" +#include "llvm/MC/MCSymbol.h" +#include "llvm/Support/Debug.h" +namespace llvm { + +class MCELFStreamer; + +class AMDGPUTargetStreamer : public MCTargetStreamer { +public: + AMDGPUTargetStreamer(MCStreamer &S); + virtual void EmitDirectiveHSACodeObjectVersion(uint32_t Major, + uint32_t Minor) = 0; + + virtual void EmitDirectiveHSACodeObjectISA(uint32_t Major, uint32_t Minor, + uint32_t Stepping, + StringRef VendorName, + StringRef ArchName) = 0; + + virtual void EmitAMDKernelCodeT(const amd_kernel_code_t &Header) = 0; +}; + +class AMDGPUTargetAsmStreamer : public AMDGPUTargetStreamer { + formatted_raw_ostream &OS; +public: + AMDGPUTargetAsmStreamer(MCStreamer &S, formatted_raw_ostream &OS); + void EmitDirectiveHSACodeObjectVersion(uint32_t Major, + uint32_t Minor) override; + + void EmitDirectiveHSACodeObjectISA(uint32_t Major, uint32_t Minor, + uint32_t Stepping, StringRef VendorName, + StringRef ArchName) override; + + void EmitAMDKernelCodeT(const amd_kernel_code_t &Header) override; +}; + +class AMDGPUTargetELFStreamer : public AMDGPUTargetStreamer { + + enum NoteType { + NT_AMDGPU_HSA_CODE_OBJECT_VERSION = 1, + NT_AMDGPU_HSA_HSAIL = 2, + NT_AMDGPU_HSA_ISA = 3, + NT_AMDGPU_HSA_PRODUCER = 4, + NT_AMDGPU_HSA_PRODUCER_OPTIONS = 5, + NT_AMDGPU_HSA_EXTENSION = 6, + NT_AMDGPU_HSA_HLDEBUG_DEBUG = 101, + NT_AMDGPU_HSA_HLDEBUG_TARGET = 102 + }; + + MCStreamer &Streamer; + +public: + AMDGPUTargetELFStreamer(MCStreamer &S); + + MCELFStreamer &getStreamer(); + + void EmitDirectiveHSACodeObjectVersion(uint32_t Major, + uint32_t Minor) override; + + void EmitDirectiveHSACodeObjectISA(uint32_t Major, uint32_t Minor, + uint32_t Stepping, StringRef VendorName, + StringRef ArchName) override; + + void EmitAMDKernelCodeT(const amd_kernel_code_t &Header) override; + +}; + +} diff --git a/lib/Target/AMDGPU/MCTargetDesc/CMakeLists.txt b/lib/Target/AMDGPU/MCTargetDesc/CMakeLists.txt index 151d0d5f83de..8306a051ff98 100644 --- a/lib/Target/AMDGPU/MCTargetDesc/CMakeLists.txt +++ b/lib/Target/AMDGPU/MCTargetDesc/CMakeLists.txt @@ -5,6 +5,7 @@ add_llvm_library(LLVMAMDGPUDesc AMDGPUMCCodeEmitter.cpp AMDGPUMCTargetDesc.cpp AMDGPUMCAsmInfo.cpp + AMDGPUTargetStreamer.cpp R600MCCodeEmitter.cpp SIMCCodeEmitter.cpp ) diff --git a/lib/Target/AMDGPU/Makefile b/lib/Target/AMDGPU/Makefile index 2e2de5020867..219f34daa24f 100644 --- a/lib/Target/AMDGPU/Makefile +++ b/lib/Target/AMDGPU/Makefile @@ -18,6 +18,6 @@ BUILT_SOURCES = AMDGPUGenRegisterInfo.inc AMDGPUGenInstrInfo.inc \ AMDGPUGenIntrinsics.inc AMDGPUGenDFAPacketizer.inc \ AMDGPUGenAsmWriter.inc AMDGPUGenAsmMatcher.inc -DIRS = AsmParser InstPrinter TargetInfo MCTargetDesc +DIRS = AsmParser InstPrinter TargetInfo MCTargetDesc Utils include $(LEVEL)/Makefile.common diff --git a/lib/Target/AMDGPU/Processors.td b/lib/Target/AMDGPU/Processors.td index c0ffede51999..69efb8b8bc43 100644 --- a/lib/Target/AMDGPU/Processors.td +++ b/lib/Target/AMDGPU/Processors.td @@ -104,7 +104,7 @@ def : ProcessorModel<"hainan", SIQuarterSpeedModel, [FeatureSouthernIslands]>; //===----------------------------------------------------------------------===// def : ProcessorModel<"bonaire", SIQuarterSpeedModel, - [FeatureSeaIslands, FeatureLDSBankCount32] + [FeatureSeaIslands, FeatureLDSBankCount32, FeatureISAVersion7_0_0] >; def : ProcessorModel<"kabini", SIQuarterSpeedModel, @@ -112,11 +112,12 @@ def : ProcessorModel<"kabini", SIQuarterSpeedModel, >; def : ProcessorModel<"kaveri", SIQuarterSpeedModel, - [FeatureSeaIslands, FeatureLDSBankCount32] + [FeatureSeaIslands, FeatureLDSBankCount32, FeatureISAVersion7_0_0] >; def : ProcessorModel<"hawaii", SIFullSpeedModel, - [FeatureSeaIslands, FeatureFastFMAF32, FeatureLDSBankCount32] + [FeatureSeaIslands, FeatureFastFMAF32, FeatureLDSBankCount32, + FeatureISAVersion7_0_1] >; def : ProcessorModel<"mullins", SIQuarterSpeedModel, @@ -127,11 +128,13 @@ def : ProcessorModel<"mullins", SIQuarterSpeedModel, //===----------------------------------------------------------------------===// def : ProcessorModel<"tonga", SIQuarterSpeedModel, - [FeatureVolcanicIslands, FeatureSGPRInitBug] + [FeatureVolcanicIslands, FeatureSGPRInitBug, FeatureISAVersion8_0_0] >; def : ProcessorModel<"iceland", SIQuarterSpeedModel, - [FeatureVolcanicIslands, FeatureSGPRInitBug] + [FeatureVolcanicIslands, FeatureSGPRInitBug, FeatureISAVersion8_0_0] >; -def : ProcessorModel<"carrizo", SIQuarterSpeedModel, [FeatureVolcanicIslands]>; +def : ProcessorModel<"carrizo", SIQuarterSpeedModel, + [FeatureVolcanicIslands, FeatureISAVersion8_0_1] +>; diff --git a/lib/Target/AMDGPU/R600Defines.h b/lib/Target/AMDGPU/R600Defines.h index 6ff0a2204cfa..51d87eda31d1 100644 --- a/lib/Target/AMDGPU/R600Defines.h +++ b/lib/Target/AMDGPU/R600Defines.h @@ -48,7 +48,7 @@ namespace R600_InstFlag { IS_EXPORT = (1 << 17), LDS_1A2D = (1 << 18) }; -} // namespace R600_InstFlag +} #define HAS_NATIVE_OPERANDS(Flags) ((Flags) & R600_InstFlag::NATIVE_OPERANDS) @@ -138,7 +138,7 @@ namespace OpName { VEC_COUNT }; -} // namespace OpName +} //===----------------------------------------------------------------------===// // Config register definitions diff --git a/lib/Target/AMDGPU/R600ISelLowering.h b/lib/Target/AMDGPU/R600ISelLowering.h index c25287806988..c06d3c4fd309 100644 --- a/lib/Target/AMDGPU/R600ISelLowering.h +++ b/lib/Target/AMDGPU/R600ISelLowering.h @@ -75,6 +75,6 @@ private: SDNode *PostISelFolding(MachineSDNode *N, SelectionDAG &DAG) const override; }; -} // namespace llvm +} // End namespace llvm; #endif diff --git a/lib/Target/AMDGPU/R600InstrInfo.cpp b/lib/Target/AMDGPU/R600InstrInfo.cpp index 5ef883cbcadd..855fa9fe45b2 100644 --- a/lib/Target/AMDGPU/R600InstrInfo.cpp +++ b/lib/Target/AMDGPU/R600InstrInfo.cpp @@ -697,15 +697,10 @@ R600InstrInfo::AnalyzeBranch(MachineBasicBlock &MBB, // Most of the following comes from the ARM implementation of AnalyzeBranch // If the block has no terminators, it just falls into the block after it. - MachineBasicBlock::iterator I = MBB.end(); - if (I == MBB.begin()) + MachineBasicBlock::iterator I = MBB.getLastNonDebugInstr(); + if (I == MBB.end()) return false; - --I; - while (I->isDebugValue()) { - if (I == MBB.begin()) - return false; - --I; - } + // AMDGPU::BRANCH* instructions are only available after isel and are not // handled if (isBranch(I->getOpcode())) diff --git a/lib/Target/AMDGPU/R600InstrInfo.h b/lib/Target/AMDGPU/R600InstrInfo.h index 9c5f76c882f1..dee4c2b9ae31 100644 --- a/lib/Target/AMDGPU/R600InstrInfo.h +++ b/lib/Target/AMDGPU/R600InstrInfo.h @@ -298,6 +298,6 @@ int getLDSNoRetOp(uint16_t Opcode); } //End namespace AMDGPU -} // namespace llvm +} // End llvm namespace #endif diff --git a/lib/Target/AMDGPU/R600MachineFunctionInfo.h b/lib/Target/AMDGPU/R600MachineFunctionInfo.h index f5556c1e81fc..263561edd30d 100644 --- a/lib/Target/AMDGPU/R600MachineFunctionInfo.h +++ b/lib/Target/AMDGPU/R600MachineFunctionInfo.h @@ -29,6 +29,6 @@ public: unsigned StackSize; }; -} // namespace llvm +} // End llvm namespace #endif diff --git a/lib/Target/AMDGPU/R600OptimizeVectorRegisters.cpp b/lib/Target/AMDGPU/R600OptimizeVectorRegisters.cpp index a1a1b4043429..0c06ccc736d0 100644 --- a/lib/Target/AMDGPU/R600OptimizeVectorRegisters.cpp +++ b/lib/Target/AMDGPU/R600OptimizeVectorRegisters.cpp @@ -375,7 +375,7 @@ bool R600VectorRegMerger::runOnMachineFunction(MachineFunction &Fn) { return false; } -} // namespace +} llvm::FunctionPass *llvm::createR600VectorRegMerger(TargetMachine &tm) { return new R600VectorRegMerger(tm); diff --git a/lib/Target/AMDGPU/R600TextureIntrinsicsReplacer.cpp b/lib/Target/AMDGPU/R600TextureIntrinsicsReplacer.cpp index 93bcf680a022..2fc7b02f673f 100644 --- a/lib/Target/AMDGPU/R600TextureIntrinsicsReplacer.cpp +++ b/lib/Target/AMDGPU/R600TextureIntrinsicsReplacer.cpp @@ -296,7 +296,7 @@ public: char R600TextureIntrinsicsReplacer::ID = 0; -} // namespace +} FunctionPass *llvm::createR600TextureIntrinsicsReplacer() { return new R600TextureIntrinsicsReplacer(); diff --git a/lib/Target/AMDGPU/SIDefines.h b/lib/Target/AMDGPU/SIDefines.h index f1b4ba1ac07d..4c3263911c40 100644 --- a/lib/Target/AMDGPU/SIDefines.h +++ b/lib/Target/AMDGPU/SIDefines.h @@ -39,7 +39,7 @@ enum { WQM = 1 << 20, VGPRSpill = 1 << 21 }; -} // namespace SIInstrFlags +} namespace llvm { namespace AMDGPU { @@ -74,7 +74,7 @@ namespace SIInstrFlags { P_NORMAL = 1 << 8, // Positive normal P_INFINITY = 1 << 9 // Positive infinity }; -} // namespace SIInstrFlags +} namespace SISrcMods { enum { @@ -100,16 +100,41 @@ namespace SIOutMods { #define R_00B848_COMPUTE_PGM_RSRC1 0x00B848 #define S_00B028_VGPRS(x) (((x) & 0x3F) << 0) #define S_00B028_SGPRS(x) (((x) & 0x0F) << 6) + #define R_00B84C_COMPUTE_PGM_RSRC2 0x00B84C #define S_00B84C_SCRATCH_EN(x) (((x) & 0x1) << 0) +#define G_00B84C_SCRATCH_EN(x) (((x) >> 0) & 0x1) +#define C_00B84C_SCRATCH_EN 0xFFFFFFFE #define S_00B84C_USER_SGPR(x) (((x) & 0x1F) << 1) +#define G_00B84C_USER_SGPR(x) (((x) >> 1) & 0x1F) +#define C_00B84C_USER_SGPR 0xFFFFFFC1 #define S_00B84C_TGID_X_EN(x) (((x) & 0x1) << 7) +#define G_00B84C_TGID_X_EN(x) (((x) >> 7) & 0x1) +#define C_00B84C_TGID_X_EN 0xFFFFFF7F #define S_00B84C_TGID_Y_EN(x) (((x) & 0x1) << 8) +#define G_00B84C_TGID_Y_EN(x) (((x) >> 8) & 0x1) +#define C_00B84C_TGID_Y_EN 0xFFFFFEFF #define S_00B84C_TGID_Z_EN(x) (((x) & 0x1) << 9) +#define G_00B84C_TGID_Z_EN(x) (((x) >> 9) & 0x1) +#define C_00B84C_TGID_Z_EN 0xFFFFFDFF #define S_00B84C_TG_SIZE_EN(x) (((x) & 0x1) << 10) +#define G_00B84C_TG_SIZE_EN(x) (((x) >> 10) & 0x1) +#define C_00B84C_TG_SIZE_EN 0xFFFFFBFF #define S_00B84C_TIDIG_COMP_CNT(x) (((x) & 0x03) << 11) - +#define G_00B84C_TIDIG_COMP_CNT(x) (((x) >> 11) & 0x03) +#define C_00B84C_TIDIG_COMP_CNT 0xFFFFE7FF +/* CIK */ +#define S_00B84C_EXCP_EN_MSB(x) (((x) & 0x03) << 13) +#define G_00B84C_EXCP_EN_MSB(x) (((x) >> 13) & 0x03) +#define C_00B84C_EXCP_EN_MSB 0xFFFF9FFF +/* */ #define S_00B84C_LDS_SIZE(x) (((x) & 0x1FF) << 15) +#define G_00B84C_LDS_SIZE(x) (((x) >> 15) & 0x1FF) +#define C_00B84C_LDS_SIZE 0xFF007FFF +#define S_00B84C_EXCP_EN(x) (((x) & 0x7F) << 24) +#define G_00B84C_EXCP_EN(x) (((x) >> 24) & 0x7F) +#define C_00B84C_EXCP_EN + #define R_0286CC_SPI_PS_INPUT_ENA 0x0286CC diff --git a/lib/Target/AMDGPU/SIISelLowering.cpp b/lib/Target/AMDGPU/SIISelLowering.cpp index 12d08cf4c7f5..ead1a3743473 100644 --- a/lib/Target/AMDGPU/SIISelLowering.cpp +++ b/lib/Target/AMDGPU/SIISelLowering.cpp @@ -583,7 +583,8 @@ SDValue SITargetLowering::LowerFormalArguments( if (VA.isMemLoc()) { VT = Ins[i].VT; EVT MemVT = Splits[i].VT; - const unsigned Offset = 36 + VA.getLocMemOffset(); + const unsigned Offset = Subtarget->getExplicitKernelArgOffset() + + VA.getLocMemOffset(); // The first 36 bytes of the input buffer contains information about // thread group and global sizes. SDValue Arg = LowerParameter(DAG, VT, MemVT, DL, DAG.getRoot(), @@ -2211,8 +2212,9 @@ SDValue SITargetLowering::CreateLiveInRegister(SelectionDAG &DAG, std::pair<unsigned, const TargetRegisterClass *> SITargetLowering::getRegForInlineAsmConstraint(const TargetRegisterInfo *TRI, - const std::string &Constraint, + const std::string &Constraint_, MVT VT) const { + StringRef Constraint(Constraint_); if (Constraint == "r") { switch(VT.SimpleTy) { default: llvm_unreachable("Unhandled type for 'r' inline asm constraint"); @@ -2232,8 +2234,9 @@ SITargetLowering::getRegForInlineAsmConstraint(const TargetRegisterInfo *TRI, } if (RC) { - unsigned Idx = std::atoi(Constraint.substr(2).c_str()); - if (Idx < RC->getNumRegs()) + uint32_t Idx; + bool Failed = Constraint.substr(2).getAsInteger(10, Idx); + if (!Failed && Idx < RC->getNumRegs()) return std::make_pair(RC->getRegister(Idx), RC); } } diff --git a/lib/Target/AMDGPU/SIInstrInfo.cpp b/lib/Target/AMDGPU/SIInstrInfo.cpp index 47bc17823b3f..eb96bd0227b2 100644 --- a/lib/Target/AMDGPU/SIInstrInfo.cpp +++ b/lib/Target/AMDGPU/SIInstrInfo.cpp @@ -440,22 +440,22 @@ SIInstrInfo::copyPhysReg(MachineBasicBlock &MBB, } } -unsigned SIInstrInfo::commuteOpcode(const MachineInstr &MI) const { +int SIInstrInfo::commuteOpcode(const MachineInstr &MI) const { const unsigned Opcode = MI.getOpcode(); int NewOpc; // Try to map original to commuted opcode NewOpc = AMDGPU::getCommuteRev(Opcode); - // Check if the commuted (REV) opcode exists on the target. - if (NewOpc != -1 && pseudoToMCOpcode(NewOpc) != -1) - return NewOpc; + if (NewOpc != -1) + // Check if the commuted (REV) opcode exists on the target. + return pseudoToMCOpcode(NewOpc) != -1 ? NewOpc : -1; // Try to map commuted to original opcode NewOpc = AMDGPU::getCommuteOrig(Opcode); - // Check if the original (non-REV) opcode exists on the target. - if (NewOpc != -1 && pseudoToMCOpcode(NewOpc) != -1) - return NewOpc; + if (NewOpc != -1) + // Check if the original (non-REV) opcode exists on the target. + return pseudoToMCOpcode(NewOpc) != -1 ? NewOpc : -1; return Opcode; } @@ -771,6 +771,10 @@ MachineInstr *SIInstrInfo::commuteInstruction(MachineInstr *MI, if (MI->getNumOperands() < 3) return nullptr; + int CommutedOpcode = commuteOpcode(*MI); + if (CommutedOpcode == -1) + return nullptr; + int Src0Idx = AMDGPU::getNamedOperandIdx(MI->getOpcode(), AMDGPU::OpName::src0); assert(Src0Idx != -1 && "Should always have src0 operand"); @@ -833,7 +837,7 @@ MachineInstr *SIInstrInfo::commuteInstruction(MachineInstr *MI, } if (MI) - MI->setDesc(get(commuteOpcode(*MI))); + MI->setDesc(get(CommutedOpcode)); return MI; } @@ -2716,8 +2720,13 @@ MachineOperand *SIInstrInfo::getNamedOperand(MachineInstr &MI, uint64_t SIInstrInfo::getDefaultRsrcDataFormat() const { uint64_t RsrcDataFormat = AMDGPU::RSRC_DATA_FORMAT; - if (ST.isAmdHsaOS()) + if (ST.isAmdHsaOS()) { RsrcDataFormat |= (1ULL << 56); + if (ST.getGeneration() >= AMDGPUSubtarget::VOLCANIC_ISLANDS) + // Set MTYPE = 2 + RsrcDataFormat |= (2ULL << 59); + } + return RsrcDataFormat; } diff --git a/lib/Target/AMDGPU/SIInstrInfo.h b/lib/Target/AMDGPU/SIInstrInfo.h index 6fafb945c993..0382272068d2 100644 --- a/lib/Target/AMDGPU/SIInstrInfo.h +++ b/lib/Target/AMDGPU/SIInstrInfo.h @@ -117,7 +117,7 @@ public: // register. If there is no hardware instruction that can store to \p // DstRC, then AMDGPU::COPY is returned. unsigned getMovOpcode(const TargetRegisterClass *DstRC) const; - unsigned commuteOpcode(const MachineInstr &MI) const; + int commuteOpcode(const MachineInstr &MI) const; MachineInstr *commuteInstruction(MachineInstr *MI, bool NewMI = false) const override; diff --git a/lib/Target/AMDGPU/SIInstrInfo.td b/lib/Target/AMDGPU/SIInstrInfo.td index 93e4ca74ec38..fcb58d5da3b0 100644 --- a/lib/Target/AMDGPU/SIInstrInfo.td +++ b/lib/Target/AMDGPU/SIInstrInfo.td @@ -1740,7 +1740,7 @@ multiclass VOP3_VCC_Inst <vop3 op, string opName, InputModsNoDefault:$src2_modifiers, P.Src2RC64:$src2, ClampMod:$clamp, omod:$omod), - " $dst, $src0_modifiers, $src1_modifiers, $src2_modifiers"#"$clamp"#"$omod", + "$dst, $src0_modifiers, $src1_modifiers, $src2_modifiers"#"$clamp"#"$omod", [(set P.DstVT:$dst, (node (P.Src0VT (VOP3Mods0 P.Src0VT:$src0, i32:$src0_modifiers, i1:$clamp, i32:$omod)), diff --git a/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.cpp b/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.cpp new file mode 100644 index 000000000000..b76b4007003f --- /dev/null +++ b/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.cpp @@ -0,0 +1,60 @@ +//===-- AMDGPUBaseInfo.cpp - AMDGPU Base encoding information--------------===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// +#include "AMDGPUBaseInfo.h" +#include "llvm/MC/SubtargetFeature.h" + +#define GET_SUBTARGETINFO_ENUM +#include "AMDGPUGenSubtargetInfo.inc" +#undef GET_SUBTARGETINFO_ENUM + +namespace llvm { +namespace AMDGPU { + +IsaVersion getIsaVersion(const FeatureBitset &Features) { + + if (Features.test(FeatureISAVersion7_0_0)) + return {7, 0, 0}; + + if (Features.test(FeatureISAVersion7_0_1)) + return {7, 0, 1}; + + if (Features.test(FeatureISAVersion8_0_0)) + return {8, 0, 0}; + + if (Features.test(FeatureISAVersion8_0_1)) + return {8, 0, 1}; + + return {0, 0, 0}; +} + +void initDefaultAMDKernelCodeT(amd_kernel_code_t &Header, + const FeatureBitset &Features) { + + IsaVersion ISA = getIsaVersion(Features); + + memset(&Header, 0, sizeof(Header)); + + Header.amd_kernel_code_version_major = 1; + Header.amd_kernel_code_version_minor = 0; + Header.amd_machine_kind = 1; // AMD_MACHINE_KIND_AMDGPU + Header.amd_machine_version_major = ISA.Major; + Header.amd_machine_version_minor = ISA.Minor; + Header.amd_machine_version_stepping = ISA.Stepping; + Header.kernel_code_entry_byte_offset = sizeof(Header); + // wavefront_size is specified as a power of 2: 2^6 = 64 threads. + Header.wavefront_size = 6; + // These alignment values are specified in powers of two, so alignment = + // 2^n. The minimum alignment is 2^4 = 16. + Header.kernarg_segment_alignment = 4; + Header.group_segment_alignment = 4; + Header.private_segment_alignment = 4; +} + +} // End namespace AMDGPU +} // End namespace llvm diff --git a/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.h b/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.h new file mode 100644 index 000000000000..f57028cc5bfd --- /dev/null +++ b/lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.h @@ -0,0 +1,34 @@ +//===-- AMDGPUBaseInfo.h - Top level definitions for AMDGPU -----*- C++ -*-===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// + +#ifndef LLVM_LIB_TARGET_AMDGPU_UTILS_AMDGPUBASEINFO_H +#define LLVM_LIB_TARGET_AMDGPU_UTILS_AMDGPUBASEINFO_H + +#include "AMDKernelCodeT.h" + +namespace llvm { + +class FeatureBitset; + +namespace AMDGPU { + +struct IsaVersion { + unsigned Major; + unsigned Minor; + unsigned Stepping; +}; + +IsaVersion getIsaVersion(const FeatureBitset &Features); +void initDefaultAMDKernelCodeT(amd_kernel_code_t &Header, + const FeatureBitset &Features); + +} // end namespace AMDGPU +} // end namespace llvm + +#endif diff --git a/lib/Target/AMDGPU/Utils/CMakeLists.txt b/lib/Target/AMDGPU/Utils/CMakeLists.txt new file mode 100644 index 000000000000..2c07aeab7dd3 --- /dev/null +++ b/lib/Target/AMDGPU/Utils/CMakeLists.txt @@ -0,0 +1,3 @@ +add_llvm_library(LLVMAMDGPUUtils + AMDGPUBaseInfo.cpp + ) diff --git a/lib/Target/AMDGPU/Utils/LLVMBuild.txt b/lib/Target/AMDGPU/Utils/LLVMBuild.txt new file mode 100644 index 000000000000..dec5360e3bc7 --- /dev/null +++ b/lib/Target/AMDGPU/Utils/LLVMBuild.txt @@ -0,0 +1,23 @@ +;===- ./lib/Target/AMDGPU/Utils/LLVMBuild.txt ------------------*- Conf -*--===; +; +; The LLVM Compiler Infrastructure +; +; This file is distributed under the University of Illinois Open Source +; License. See LICENSE.TXT for details. +; +;===------------------------------------------------------------------------===; +; +; This is an LLVMBuild description file for the components in this subdirectory. +; +; For more information on the LLVMBuild system, please see: +; +; http://llvm.org/docs/LLVMBuild.html +; +;===------------------------------------------------------------------------===; + +[component_0] +type = Library +name = AMDGPUUtils +parent = AMDGPU +required_libraries = Support +add_to_library_groups = AMDGPU diff --git a/lib/Target/AMDGPU/Utils/Makefile b/lib/Target/AMDGPU/Utils/Makefile new file mode 100644 index 000000000000..1019e726d50e --- /dev/null +++ b/lib/Target/AMDGPU/Utils/Makefile @@ -0,0 +1,16 @@ +##===- lib/Target/AMDGPU/Utils/Makefile --------------------*- Makefile -*-===## +# +# The LLVM Compiler Infrastructure +# +# This file is distributed under the University of Illinois Open Source +# License. See LICENSE.TXT for details. +# +##===----------------------------------------------------------------------===## +LEVEL = ../../../.. +LIBRARYNAME = LLVMAMDGPUUtils + +# Hack: we need to include 'main' AMDGPU target directory to grab private +# headers +CPP.Flags += -I$(PROJ_OBJ_DIR)/.. -I$(PROJ_SRC_DIR)/.. + +include $(LEVEL)/Makefile.common |