LCOV - code coverage report
Current view: top level - lib/Target/AMDGPU/MCTargetDesc - AMDGPUTargetStreamer.cpp (source / functions) Hit Total Coverage
Test: llvm-toolchain.info Lines: 220 259 84.9 %
Date: 2018-10-20 13:21:21 Functions: 24 25 96.0 %
Legend: Lines: hit not hit

          Line data    Source code
       1             : //===-- AMDGPUTargetStreamer.cpp - Mips Target Streamer Methods -----------===//
       2             : //
       3             : //                     The LLVM Compiler Infrastructure
       4             : //
       5             : // This file is distributed under the University of Illinois Open Source
       6             : // License. See LICENSE.TXT for details.
       7             : //
       8             : //===----------------------------------------------------------------------===//
       9             : //
      10             : // This file provides AMDGPU specific target streamer methods.
      11             : //
      12             : //===----------------------------------------------------------------------===//
      13             : 
      14             : #include "AMDGPUTargetStreamer.h"
      15             : #include "AMDGPU.h"
      16             : #include "SIDefines.h"
      17             : #include "Utils/AMDGPUBaseInfo.h"
      18             : #include "Utils/AMDKernelCodeTUtils.h"
      19             : #include "llvm/ADT/Twine.h"
      20             : #include "llvm/BinaryFormat/ELF.h"
      21             : #include "llvm/IR/Constants.h"
      22             : #include "llvm/IR/Function.h"
      23             : #include "llvm/IR/Metadata.h"
      24             : #include "llvm/IR/Module.h"
      25             : #include "llvm/MC/MCContext.h"
      26             : #include "llvm/MC/MCELFStreamer.h"
      27             : #include "llvm/MC/MCObjectFileInfo.h"
      28             : #include "llvm/MC/MCSectionELF.h"
      29             : #include "llvm/Support/FormattedStream.h"
      30             : #include "llvm/Support/TargetParser.h"
      31             : 
      32             : namespace llvm {
      33             : #include "AMDGPUPTNote.h"
      34             : }
      35             : 
      36             : using namespace llvm;
      37             : using namespace llvm::AMDGPU;
      38             : 
      39             : //===----------------------------------------------------------------------===//
      40             : // AMDGPUTargetStreamer
      41             : //===----------------------------------------------------------------------===//
      42             : 
      43          22 : bool AMDGPUTargetStreamer::EmitHSAMetadata(StringRef HSAMetadataString) {
      44          22 :   HSAMD::Metadata HSAMetadata;
      45          44 :   if (HSAMD::fromString(HSAMetadataString, HSAMetadata))
      46             :     return false;
      47             : 
      48          16 :   return EmitHSAMetadata(HSAMetadata);
      49             : }
      50             : 
      51           0 : StringRef AMDGPUTargetStreamer::getArchNameFromElfMach(unsigned ElfMach) {
      52             :   AMDGPU::GPUKind AK;
      53             : 
      54           0 :   switch (ElfMach) {
      55           0 :   case ELF::EF_AMDGPU_MACH_R600_R600:     AK = GK_R600;    break;
      56           0 :   case ELF::EF_AMDGPU_MACH_R600_R630:     AK = GK_R630;    break;
      57           0 :   case ELF::EF_AMDGPU_MACH_R600_RS880:    AK = GK_RS880;   break;
      58           0 :   case ELF::EF_AMDGPU_MACH_R600_RV670:    AK = GK_RV670;   break;
      59           0 :   case ELF::EF_AMDGPU_MACH_R600_RV710:    AK = GK_RV710;   break;
      60           0 :   case ELF::EF_AMDGPU_MACH_R600_RV730:    AK = GK_RV730;   break;
      61           0 :   case ELF::EF_AMDGPU_MACH_R600_RV770:    AK = GK_RV770;   break;
      62           0 :   case ELF::EF_AMDGPU_MACH_R600_CEDAR:    AK = GK_CEDAR;   break;
      63           0 :   case ELF::EF_AMDGPU_MACH_R600_CYPRESS:  AK = GK_CYPRESS; break;
      64           0 :   case ELF::EF_AMDGPU_MACH_R600_JUNIPER:  AK = GK_JUNIPER; break;
      65           0 :   case ELF::EF_AMDGPU_MACH_R600_REDWOOD:  AK = GK_REDWOOD; break;
      66           0 :   case ELF::EF_AMDGPU_MACH_R600_SUMO:     AK = GK_SUMO;    break;
      67           0 :   case ELF::EF_AMDGPU_MACH_R600_BARTS:    AK = GK_BARTS;   break;
      68           0 :   case ELF::EF_AMDGPU_MACH_R600_CAICOS:   AK = GK_CAICOS;  break;
      69           0 :   case ELF::EF_AMDGPU_MACH_R600_CAYMAN:   AK = GK_CAYMAN;  break;
      70           0 :   case ELF::EF_AMDGPU_MACH_R600_TURKS:    AK = GK_TURKS;   break;
      71           0 :   case ELF::EF_AMDGPU_MACH_AMDGCN_GFX600: AK = GK_GFX600;  break;
      72           0 :   case ELF::EF_AMDGPU_MACH_AMDGCN_GFX601: AK = GK_GFX601;  break;
      73           0 :   case ELF::EF_AMDGPU_MACH_AMDGCN_GFX700: AK = GK_GFX700;  break;
      74           0 :   case ELF::EF_AMDGPU_MACH_AMDGCN_GFX701: AK = GK_GFX701;  break;
      75           0 :   case ELF::EF_AMDGPU_MACH_AMDGCN_GFX702: AK = GK_GFX702;  break;
      76           0 :   case ELF::EF_AMDGPU_MACH_AMDGCN_GFX703: AK = GK_GFX703;  break;
      77           0 :   case ELF::EF_AMDGPU_MACH_AMDGCN_GFX704: AK = GK_GFX704;  break;
      78           0 :   case ELF::EF_AMDGPU_MACH_AMDGCN_GFX801: AK = GK_GFX801;  break;
      79           0 :   case ELF::EF_AMDGPU_MACH_AMDGCN_GFX802: AK = GK_GFX802;  break;
      80           0 :   case ELF::EF_AMDGPU_MACH_AMDGCN_GFX803: AK = GK_GFX803;  break;
      81           0 :   case ELF::EF_AMDGPU_MACH_AMDGCN_GFX810: AK = GK_GFX810;  break;
      82           0 :   case ELF::EF_AMDGPU_MACH_AMDGCN_GFX900: AK = GK_GFX900;  break;
      83           0 :   case ELF::EF_AMDGPU_MACH_AMDGCN_GFX902: AK = GK_GFX902;  break;
      84           0 :   case ELF::EF_AMDGPU_MACH_AMDGCN_GFX904: AK = GK_GFX904;  break;
      85           0 :   case ELF::EF_AMDGPU_MACH_AMDGCN_GFX906: AK = GK_GFX906;  break;
      86           0 :   case ELF::EF_AMDGPU_MACH_NONE:          AK = GK_NONE;    break;
      87             :   }
      88             : 
      89           0 :   StringRef GPUName = getArchNameAMDGCN(AK);
      90             :   if (GPUName != "")
      91           0 :     return GPUName;
      92           0 :   return getArchNameR600(AK);
      93             : }
      94             : 
      95         144 : unsigned AMDGPUTargetStreamer::getElfMach(StringRef GPU) {
      96         144 :   AMDGPU::GPUKind AK = parseArchAMDGCN(GPU);
      97         144 :   if (AK == AMDGPU::GPUKind::GK_NONE)
      98          32 :     AK = parseArchR600(GPU);
      99             : 
     100         144 :   switch (AK) {
     101             :   case GK_R600:    return ELF::EF_AMDGPU_MACH_R600_R600;
     102           1 :   case GK_R630:    return ELF::EF_AMDGPU_MACH_R600_R630;
     103           1 :   case GK_RS880:   return ELF::EF_AMDGPU_MACH_R600_RS880;
     104           1 :   case GK_RV670:   return ELF::EF_AMDGPU_MACH_R600_RV670;
     105           1 :   case GK_RV710:   return ELF::EF_AMDGPU_MACH_R600_RV710;
     106           1 :   case GK_RV730:   return ELF::EF_AMDGPU_MACH_R600_RV730;
     107           1 :   case GK_RV770:   return ELF::EF_AMDGPU_MACH_R600_RV770;
     108           1 :   case GK_CEDAR:   return ELF::EF_AMDGPU_MACH_R600_CEDAR;
     109           3 :   case GK_CYPRESS: return ELF::EF_AMDGPU_MACH_R600_CYPRESS;
     110           1 :   case GK_JUNIPER: return ELF::EF_AMDGPU_MACH_R600_JUNIPER;
     111           2 :   case GK_REDWOOD: return ELF::EF_AMDGPU_MACH_R600_REDWOOD;
     112           1 :   case GK_SUMO:    return ELF::EF_AMDGPU_MACH_R600_SUMO;
     113           1 :   case GK_BARTS:   return ELF::EF_AMDGPU_MACH_R600_BARTS;
     114           1 :   case GK_CAICOS:  return ELF::EF_AMDGPU_MACH_R600_CAICOS;
     115           1 :   case GK_CAYMAN:  return ELF::EF_AMDGPU_MACH_R600_CAYMAN;
     116           1 :   case GK_TURKS:   return ELF::EF_AMDGPU_MACH_R600_TURKS;
     117           3 :   case GK_GFX600:  return ELF::EF_AMDGPU_MACH_AMDGCN_GFX600;
     118           6 :   case GK_GFX601:  return ELF::EF_AMDGPU_MACH_AMDGCN_GFX601;
     119          22 :   case GK_GFX700:  return ELF::EF_AMDGPU_MACH_AMDGCN_GFX700;
     120           2 :   case GK_GFX701:  return ELF::EF_AMDGPU_MACH_AMDGCN_GFX701;
     121           1 :   case GK_GFX702:  return ELF::EF_AMDGPU_MACH_AMDGCN_GFX702;
     122           3 :   case GK_GFX703:  return ELF::EF_AMDGPU_MACH_AMDGCN_GFX703;
     123           2 :   case GK_GFX704:  return ELF::EF_AMDGPU_MACH_AMDGCN_GFX704;
     124          16 :   case GK_GFX801:  return ELF::EF_AMDGPU_MACH_AMDGCN_GFX801;
     125          18 :   case GK_GFX802:  return ELF::EF_AMDGPU_MACH_AMDGCN_GFX802;
     126          20 :   case GK_GFX803:  return ELF::EF_AMDGPU_MACH_AMDGCN_GFX803;
     127           3 :   case GK_GFX810:  return ELF::EF_AMDGPU_MACH_AMDGCN_GFX810;
     128          13 :   case GK_GFX900:  return ELF::EF_AMDGPU_MACH_AMDGCN_GFX900;
     129           1 :   case GK_GFX902:  return ELF::EF_AMDGPU_MACH_AMDGCN_GFX902;
     130           1 :   case GK_GFX904:  return ELF::EF_AMDGPU_MACH_AMDGCN_GFX904;
     131           1 :   case GK_GFX906:  return ELF::EF_AMDGPU_MACH_AMDGCN_GFX906;
     132          12 :   case GK_NONE:    return ELF::EF_AMDGPU_MACH_NONE;
     133             :   }
     134             : 
     135           0 :   llvm_unreachable("unknown GPU");
     136             : }
     137             : 
     138             : //===----------------------------------------------------------------------===//
     139             : // AMDGPUTargetAsmStreamer
     140             : //===----------------------------------------------------------------------===//
     141             : 
     142        2236 : AMDGPUTargetAsmStreamer::AMDGPUTargetAsmStreamer(MCStreamer &S,
     143        2236 :                                                  formatted_raw_ostream &OS)
     144        2236 :     : AMDGPUTargetStreamer(S), OS(OS) { }
     145             : 
     146          36 : void AMDGPUTargetAsmStreamer::EmitDirectiveAMDGCNTarget(StringRef Target) {
     147          36 :   OS << "\t.amdgcn_target \"" << Target << "\"\n";
     148          36 : }
     149             : 
     150         288 : void AMDGPUTargetAsmStreamer::EmitDirectiveHSACodeObjectVersion(
     151             :     uint32_t Major, uint32_t Minor) {
     152         288 :   OS << "\t.hsa_code_object_version " <<
     153         576 :         Twine(Major) << "," << Twine(Minor) << '\n';
     154         288 : }
     155             : 
     156             : void
     157         334 : AMDGPUTargetAsmStreamer::EmitDirectiveHSACodeObjectISA(uint32_t Major,
     158             :                                                        uint32_t Minor,
     159             :                                                        uint32_t Stepping,
     160             :                                                        StringRef VendorName,
     161             :                                                        StringRef ArchName) {
     162         334 :   OS << "\t.hsa_code_object_isa " <<
     163        1002 :         Twine(Major) << "," << Twine(Minor) << "," << Twine(Stepping) <<
     164         334 :         ",\"" << VendorName << "\",\"" << ArchName << "\"\n";
     165             : 
     166         334 : }
     167             : 
     168             : void
     169        2256 : AMDGPUTargetAsmStreamer::EmitAMDKernelCodeT(const amd_kernel_code_t &Header) {
     170        2256 :   OS << "\t.amd_kernel_code_t\n";
     171        2256 :   dumpAmdKernelCode(&Header, OS, "\t\t");
     172        2256 :   OS << "\t.end_amd_kernel_code_t\n";
     173        2256 : }
     174             : 
     175        2261 : void AMDGPUTargetAsmStreamer::EmitAMDGPUSymbolType(StringRef SymbolName,
     176             :                                                    unsigned Type) {
     177        2261 :   switch (Type) {
     178           0 :     default: llvm_unreachable("Invalid AMDGPU symbol type");
     179        2261 :     case ELF::STT_AMDGPU_HSA_KERNEL:
     180        2261 :       OS << "\t.amdgpu_hsa_kernel " << SymbolName << '\n' ;
     181             :       break;
     182             :   }
     183        2261 : }
     184             : 
     185        1812 : bool AMDGPUTargetAsmStreamer::EmitISAVersion(StringRef IsaVersionString) {
     186        1812 :   OS << "\t.amd_amdgpu_isa \"" << IsaVersionString << "\"\n";
     187        1812 :   return true;
     188             : }
     189             : 
     190         294 : bool AMDGPUTargetAsmStreamer::EmitHSAMetadata(
     191             :     const AMDGPU::HSAMD::Metadata &HSAMetadata) {
     192             :   std::string HSAMetadataString;
     193         294 :   if (HSAMD::toString(HSAMetadata, HSAMetadataString))
     194             :     return false;
     195             : 
     196         588 :   OS << '\t' << HSAMD::AssemblerDirectiveBegin << '\n';
     197         294 :   OS << HSAMetadataString << '\n';
     198         588 :   OS << '\t' << HSAMD::AssemblerDirectiveEnd << '\n';
     199             :   return true;
     200             : }
     201             : 
     202          42 : bool AMDGPUTargetAsmStreamer::EmitPALMetadata(
     203             :     const PALMD::Metadata &PALMetadata) {
     204             :   std::string PALMetadataString;
     205          42 :   if (PALMD::toString(PALMetadata, PALMetadataString))
     206             :     return false;
     207             : 
     208          84 :   OS << '\t' << PALMD::AssemblerDirective << PALMetadataString << '\n';
     209             :   return true;
     210             : }
     211             : 
     212          39 : void AMDGPUTargetAsmStreamer::EmitAmdhsaKernelDescriptor(
     213             :     const MCSubtargetInfo &STI, StringRef KernelName,
     214             :     const amdhsa::kernel_descriptor_t &KD, uint64_t NextVGPR, uint64_t NextSGPR,
     215             :     bool ReserveVCC, bool ReserveFlatScr, bool ReserveXNACK) {
     216          39 :   IsaVersion IVersion = getIsaVersion(STI.getCPU());
     217             : 
     218          39 :   OS << "\t.amdhsa_kernel " << KernelName << '\n';
     219             : 
     220             : #define PRINT_FIELD(STREAM, DIRECTIVE, KERNEL_DESC, MEMBER_NAME, FIELD_NAME)   \
     221             :   STREAM << "\t\t" << DIRECTIVE << " "                                         \
     222             :          << AMDHSA_BITS_GET(KERNEL_DESC.MEMBER_NAME, FIELD_NAME) << '\n';
     223             : 
     224          39 :   OS << "\t\t.amdhsa_group_segment_fixed_size " << KD.group_segment_fixed_size
     225             :      << '\n';
     226          39 :   OS << "\t\t.amdhsa_private_segment_fixed_size "
     227          39 :      << KD.private_segment_fixed_size << '\n';
     228             : 
     229          39 :   PRINT_FIELD(OS, ".amdhsa_user_sgpr_private_segment_buffer", KD,
     230             :               kernel_code_properties,
     231             :               amdhsa::KERNEL_CODE_PROPERTY_ENABLE_SGPR_PRIVATE_SEGMENT_BUFFER);
     232          39 :   PRINT_FIELD(OS, ".amdhsa_user_sgpr_dispatch_ptr", KD,
     233             :               kernel_code_properties,
     234             :               amdhsa::KERNEL_CODE_PROPERTY_ENABLE_SGPR_DISPATCH_PTR);
     235          39 :   PRINT_FIELD(OS, ".amdhsa_user_sgpr_queue_ptr", KD,
     236             :               kernel_code_properties,
     237             :               amdhsa::KERNEL_CODE_PROPERTY_ENABLE_SGPR_QUEUE_PTR);
     238          39 :   PRINT_FIELD(OS, ".amdhsa_user_sgpr_kernarg_segment_ptr", KD,
     239             :               kernel_code_properties,
     240             :               amdhsa::KERNEL_CODE_PROPERTY_ENABLE_SGPR_KERNARG_SEGMENT_PTR);
     241          39 :   PRINT_FIELD(OS, ".amdhsa_user_sgpr_dispatch_id", KD,
     242             :               kernel_code_properties,
     243             :               amdhsa::KERNEL_CODE_PROPERTY_ENABLE_SGPR_DISPATCH_ID);
     244          39 :   PRINT_FIELD(OS, ".amdhsa_user_sgpr_flat_scratch_init", KD,
     245             :               kernel_code_properties,
     246             :               amdhsa::KERNEL_CODE_PROPERTY_ENABLE_SGPR_FLAT_SCRATCH_INIT);
     247          39 :   PRINT_FIELD(OS, ".amdhsa_user_sgpr_private_segment_size", KD,
     248             :               kernel_code_properties,
     249             :               amdhsa::KERNEL_CODE_PROPERTY_ENABLE_SGPR_PRIVATE_SEGMENT_SIZE);
     250          39 :   PRINT_FIELD(
     251             :       OS, ".amdhsa_system_sgpr_private_segment_wavefront_offset", KD,
     252             :       compute_pgm_rsrc2,
     253             :       amdhsa::COMPUTE_PGM_RSRC2_ENABLE_SGPR_PRIVATE_SEGMENT_WAVEFRONT_OFFSET);
     254          39 :   PRINT_FIELD(OS, ".amdhsa_system_sgpr_workgroup_id_x", KD,
     255             :               compute_pgm_rsrc2,
     256             :               amdhsa::COMPUTE_PGM_RSRC2_ENABLE_SGPR_WORKGROUP_ID_X);
     257          39 :   PRINT_FIELD(OS, ".amdhsa_system_sgpr_workgroup_id_y", KD,
     258             :               compute_pgm_rsrc2,
     259             :               amdhsa::COMPUTE_PGM_RSRC2_ENABLE_SGPR_WORKGROUP_ID_Y);
     260          39 :   PRINT_FIELD(OS, ".amdhsa_system_sgpr_workgroup_id_z", KD,
     261             :               compute_pgm_rsrc2,
     262             :               amdhsa::COMPUTE_PGM_RSRC2_ENABLE_SGPR_WORKGROUP_ID_Z);
     263          39 :   PRINT_FIELD(OS, ".amdhsa_system_sgpr_workgroup_info", KD,
     264             :               compute_pgm_rsrc2,
     265             :               amdhsa::COMPUTE_PGM_RSRC2_ENABLE_SGPR_WORKGROUP_INFO);
     266          39 :   PRINT_FIELD(OS, ".amdhsa_system_vgpr_workitem_id", KD,
     267             :               compute_pgm_rsrc2,
     268             :               amdhsa::COMPUTE_PGM_RSRC2_ENABLE_VGPR_WORKITEM_ID);
     269             : 
     270             :   // These directives are required.
     271          39 :   OS << "\t\t.amdhsa_next_free_vgpr " << NextVGPR << '\n';
     272          39 :   OS << "\t\t.amdhsa_next_free_sgpr " << NextSGPR << '\n';
     273             : 
     274          39 :   if (!ReserveVCC)
     275          38 :     OS << "\t\t.amdhsa_reserve_vcc " << ReserveVCC << '\n';
     276          39 :   if (IVersion.Major >= 7 && !ReserveFlatScr)
     277          30 :     OS << "\t\t.amdhsa_reserve_flat_scratch " << ReserveFlatScr << '\n';
     278          39 :   if (IVersion.Major >= 8 && ReserveXNACK != hasXNACK(STI))
     279           2 :     OS << "\t\t.amdhsa_reserve_xnack_mask " << ReserveXNACK << '\n';
     280             : 
     281          39 :   PRINT_FIELD(OS, ".amdhsa_float_round_mode_32", KD,
     282             :               compute_pgm_rsrc1,
     283             :               amdhsa::COMPUTE_PGM_RSRC1_FLOAT_ROUND_MODE_32);
     284          39 :   PRINT_FIELD(OS, ".amdhsa_float_round_mode_16_64", KD,
     285             :               compute_pgm_rsrc1,
     286             :               amdhsa::COMPUTE_PGM_RSRC1_FLOAT_ROUND_MODE_16_64);
     287          39 :   PRINT_FIELD(OS, ".amdhsa_float_denorm_mode_32", KD,
     288             :               compute_pgm_rsrc1,
     289             :               amdhsa::COMPUTE_PGM_RSRC1_FLOAT_DENORM_MODE_32);
     290          39 :   PRINT_FIELD(OS, ".amdhsa_float_denorm_mode_16_64", KD,
     291             :               compute_pgm_rsrc1,
     292             :               amdhsa::COMPUTE_PGM_RSRC1_FLOAT_DENORM_MODE_16_64);
     293          39 :   PRINT_FIELD(OS, ".amdhsa_dx10_clamp", KD,
     294             :               compute_pgm_rsrc1,
     295             :               amdhsa::COMPUTE_PGM_RSRC1_ENABLE_DX10_CLAMP);
     296          39 :   PRINT_FIELD(OS, ".amdhsa_ieee_mode", KD,
     297             :               compute_pgm_rsrc1,
     298             :               amdhsa::COMPUTE_PGM_RSRC1_ENABLE_IEEE_MODE);
     299          39 :   if (IVersion.Major >= 9)
     300           9 :     PRINT_FIELD(OS, ".amdhsa_fp16_overflow", KD,
     301             :                 compute_pgm_rsrc1,
     302             :                 amdhsa::COMPUTE_PGM_RSRC1_FP16_OVFL);
     303          39 :   PRINT_FIELD(
     304             :       OS, ".amdhsa_exception_fp_ieee_invalid_op", KD,
     305             :       compute_pgm_rsrc2,
     306             :       amdhsa::COMPUTE_PGM_RSRC2_ENABLE_EXCEPTION_IEEE_754_FP_INVALID_OPERATION);
     307          39 :   PRINT_FIELD(OS, ".amdhsa_exception_fp_denorm_src", KD,
     308             :               compute_pgm_rsrc2,
     309             :               amdhsa::COMPUTE_PGM_RSRC2_ENABLE_EXCEPTION_FP_DENORMAL_SOURCE);
     310          39 :   PRINT_FIELD(
     311             :       OS, ".amdhsa_exception_fp_ieee_div_zero", KD,
     312             :       compute_pgm_rsrc2,
     313             :       amdhsa::COMPUTE_PGM_RSRC2_ENABLE_EXCEPTION_IEEE_754_FP_DIVISION_BY_ZERO);
     314          39 :   PRINT_FIELD(OS, ".amdhsa_exception_fp_ieee_overflow", KD,
     315             :               compute_pgm_rsrc2,
     316             :               amdhsa::COMPUTE_PGM_RSRC2_ENABLE_EXCEPTION_IEEE_754_FP_OVERFLOW);
     317          39 :   PRINT_FIELD(OS, ".amdhsa_exception_fp_ieee_underflow", KD,
     318             :               compute_pgm_rsrc2,
     319             :               amdhsa::COMPUTE_PGM_RSRC2_ENABLE_EXCEPTION_IEEE_754_FP_UNDERFLOW);
     320          39 :   PRINT_FIELD(OS, ".amdhsa_exception_fp_ieee_inexact", KD,
     321             :               compute_pgm_rsrc2,
     322             :               amdhsa::COMPUTE_PGM_RSRC2_ENABLE_EXCEPTION_IEEE_754_FP_INEXACT);
     323          39 :   PRINT_FIELD(OS, ".amdhsa_exception_int_div_zero", KD,
     324             :               compute_pgm_rsrc2,
     325             :               amdhsa::COMPUTE_PGM_RSRC2_ENABLE_EXCEPTION_INT_DIVIDE_BY_ZERO);
     326             : #undef PRINT_FIELD
     327             : 
     328          39 :   OS << "\t.end_amdhsa_kernel\n";
     329          39 : }
     330             : 
     331             : //===----------------------------------------------------------------------===//
     332             : // AMDGPUTargetELFStreamer
     333             : //===----------------------------------------------------------------------===//
     334             : 
     335         144 : AMDGPUTargetELFStreamer::AMDGPUTargetELFStreamer(
     336         144 :     MCStreamer &S, const MCSubtargetInfo &STI)
     337         144 :     : AMDGPUTargetStreamer(S), Streamer(S) {
     338         144 :   MCAssembler &MCA = getStreamer().getAssembler();
     339         144 :   unsigned EFlags = MCA.getELFHeaderEFlags();
     340             : 
     341         144 :   EFlags &= ~ELF::EF_AMDGPU_MACH;
     342         144 :   EFlags |= getElfMach(STI.getCPU());
     343             : 
     344         144 :   EFlags &= ~ELF::EF_AMDGPU_XNACK;
     345         144 :   if (AMDGPU::hasXNACK(STI))
     346          21 :     EFlags |= ELF::EF_AMDGPU_XNACK;
     347             : 
     348             :   MCA.setELFHeaderEFlags(EFlags);
     349         144 : }
     350             : 
     351         977 : MCELFStreamer &AMDGPUTargetELFStreamer::getStreamer() {
     352         977 :   return static_cast<MCELFStreamer &>(Streamer);
     353             : }
     354             : 
     355         258 : void AMDGPUTargetELFStreamer::EmitAMDGPUNote(
     356             :     const MCExpr *DescSZ, unsigned NoteType,
     357             :     function_ref<void(MCELFStreamer &)> EmitDesc) {
     358         258 :   auto &S = getStreamer();
     359         258 :   auto &Context = S.getContext();
     360             : 
     361             :   auto NameSZ = sizeof(ElfNote::NoteName);
     362             : 
     363         258 :   S.PushSection();
     364         258 :   S.SwitchSection(Context.getELFSection(
     365         258 :     ElfNote::SectionName, ELF::SHT_NOTE, ELF::SHF_ALLOC));
     366         258 :   S.EmitIntValue(NameSZ, 4);                                  // namesz
     367         258 :   S.EmitValue(DescSZ, 4);                                     // descz
     368         258 :   S.EmitIntValue(NoteType, 4);                                // type
     369         516 :   S.EmitBytes(StringRef(ElfNote::NoteName, NameSZ));          // name
     370         258 :   S.EmitValueToAlignment(4, 0, 1, 0);                         // padding 0
     371         258 :   EmitDesc(S);                                                // desc
     372         258 :   S.EmitValueToAlignment(4, 0, 1, 0);                         // padding 0
     373         258 :   S.PopSection();
     374         258 : }
     375             : 
     376           3 : void AMDGPUTargetELFStreamer::EmitDirectiveAMDGCNTarget(StringRef Target) {}
     377             : 
     378          51 : void AMDGPUTargetELFStreamer::EmitDirectiveHSACodeObjectVersion(
     379             :     uint32_t Major, uint32_t Minor) {
     380             : 
     381          51 :   EmitAMDGPUNote(
     382         102 :     MCConstantExpr::create(8, getContext()),
     383             :     ElfNote::NT_AMDGPU_HSA_CODE_OBJECT_VERSION,
     384             :     [&](MCELFStreamer &OS){
     385          51 :       OS.EmitIntValue(Major, 4);
     386          51 :       OS.EmitIntValue(Minor, 4);
     387             :     }
     388             :   );
     389          51 : }
     390             : 
     391             : void
     392          58 : AMDGPUTargetELFStreamer::EmitDirectiveHSACodeObjectISA(uint32_t Major,
     393             :                                                        uint32_t Minor,
     394             :                                                        uint32_t Stepping,
     395             :                                                        StringRef VendorName,
     396             :                                                        StringRef ArchName) {
     397          58 :   uint16_t VendorNameSize = VendorName.size() + 1;
     398          58 :   uint16_t ArchNameSize = ArchName.size() + 1;
     399             : 
     400          58 :   unsigned DescSZ = sizeof(VendorNameSize) + sizeof(ArchNameSize) +
     401             :     sizeof(Major) + sizeof(Minor) + sizeof(Stepping) +
     402          58 :     VendorNameSize + ArchNameSize;
     403             : 
     404          58 :   EmitAMDGPUNote(
     405         116 :     MCConstantExpr::create(DescSZ, getContext()),
     406             :     ElfNote::NT_AMDGPU_HSA_ISA,
     407             :     [&](MCELFStreamer &OS) {
     408             :       OS.EmitIntValue(VendorNameSize, 2);
     409             :       OS.EmitIntValue(ArchNameSize, 2);
     410             :       OS.EmitIntValue(Major, 4);
     411             :       OS.EmitIntValue(Minor, 4);
     412             :       OS.EmitIntValue(Stepping, 4);
     413             :       OS.EmitBytes(VendorName);
     414             :       OS.EmitIntValue(0, 1); // NULL terminate VendorName
     415             :       OS.EmitBytes(ArchName);
     416             :       OS.EmitIntValue(0, 1); // NULL terminte ArchName
     417             :     }
     418             :   );
     419          58 : }
     420             : 
     421             : void
     422         282 : AMDGPUTargetELFStreamer::EmitAMDKernelCodeT(const amd_kernel_code_t &Header) {
     423             : 
     424         282 :   MCStreamer &OS = getStreamer();
     425         282 :   OS.PushSection();
     426         564 :   OS.EmitBytes(StringRef((const char*)&Header, sizeof(Header)));
     427         282 :   OS.PopSection();
     428         282 : }
     429             : 
     430         287 : void AMDGPUTargetELFStreamer::EmitAMDGPUSymbolType(StringRef SymbolName,
     431             :                                                    unsigned Type) {
     432         287 :   MCSymbolELF *Symbol = cast<MCSymbolELF>(
     433         287 :       getStreamer().getContext().getOrCreateSymbol(SymbolName));
     434         287 :   Symbol->setType(Type);
     435         287 : }
     436             : 
     437          99 : bool AMDGPUTargetELFStreamer::EmitISAVersion(StringRef IsaVersionString) {
     438             :   // Create two labels to mark the beginning and end of the desc field
     439             :   // and a MCExpr to calculate the size of the desc field.
     440          99 :   auto &Context = getContext();
     441          99 :   auto *DescBegin = Context.createTempSymbol();
     442          99 :   auto *DescEnd = Context.createTempSymbol();
     443             :   auto *DescSZ = MCBinaryExpr::createSub(
     444          99 :     MCSymbolRefExpr::create(DescEnd, Context),
     445          99 :     MCSymbolRefExpr::create(DescBegin, Context), Context);
     446             : 
     447         198 :   EmitAMDGPUNote(
     448             :     DescSZ,
     449             :     ELF::NT_AMD_AMDGPU_ISA,
     450             :     [&](MCELFStreamer &OS) {
     451             :       OS.EmitLabel(DescBegin);
     452             :       OS.EmitBytes(IsaVersionString);
     453             :       OS.EmitLabel(DescEnd);
     454             :     }
     455             :   );
     456          99 :   return true;
     457             : }
     458             : 
     459          45 : bool AMDGPUTargetELFStreamer::EmitHSAMetadata(
     460             :     const AMDGPU::HSAMD::Metadata &HSAMetadata) {
     461             :   std::string HSAMetadataString;
     462          45 :   if (HSAMD::toString(HSAMetadata, HSAMetadataString))
     463             :     return false;
     464             : 
     465             :   // Create two labels to mark the beginning and end of the desc field
     466             :   // and a MCExpr to calculate the size of the desc field.
     467          45 :   auto &Context = getContext();
     468          45 :   auto *DescBegin = Context.createTempSymbol();
     469          45 :   auto *DescEnd = Context.createTempSymbol();
     470             :   auto *DescSZ = MCBinaryExpr::createSub(
     471          45 :     MCSymbolRefExpr::create(DescEnd, Context),
     472          45 :     MCSymbolRefExpr::create(DescBegin, Context), Context);
     473             : 
     474          90 :   EmitAMDGPUNote(
     475             :     DescSZ,
     476             :     ELF::NT_AMD_AMDGPU_HSA_METADATA,
     477             :     [&](MCELFStreamer &OS) {
     478             :       OS.EmitLabel(DescBegin);
     479             :       OS.EmitBytes(HSAMetadataString);
     480             :       OS.EmitLabel(DescEnd);
     481             :     }
     482             :   );
     483          45 :   return true;
     484             : }
     485             : 
     486           5 : bool AMDGPUTargetELFStreamer::EmitPALMetadata(
     487             :     const PALMD::Metadata &PALMetadata) {
     488           5 :   EmitAMDGPUNote(
     489          15 :     MCConstantExpr::create(PALMetadata.size() * sizeof(uint32_t), getContext()),
     490             :     ELF::NT_AMD_AMDGPU_PAL_METADATA,
     491             :     [&](MCELFStreamer &OS){
     492          49 :       for (auto I : PALMetadata)
     493          44 :         OS.EmitIntValue(I, sizeof(uint32_t));
     494             :     }
     495             :   );
     496           5 :   return true;
     497             : }
     498             : 
     499           6 : void AMDGPUTargetELFStreamer::EmitAmdhsaKernelDescriptor(
     500             :     const MCSubtargetInfo &STI, StringRef KernelName,
     501             :     const amdhsa::kernel_descriptor_t &KernelDescriptor, uint64_t NextVGPR,
     502             :     uint64_t NextSGPR, bool ReserveVCC, bool ReserveFlatScr,
     503             :     bool ReserveXNACK) {
     504           6 :   auto &Streamer = getStreamer();
     505           6 :   auto &Context = Streamer.getContext();
     506             : 
     507           6 :   MCSymbolELF *KernelDescriptorSymbol = cast<MCSymbolELF>(
     508           6 :       Context.getOrCreateSymbol(Twine(KernelName) + Twine(".kd")));
     509           6 :   KernelDescriptorSymbol->setBinding(ELF::STB_GLOBAL);
     510           6 :   KernelDescriptorSymbol->setType(ELF::STT_OBJECT);
     511             :   KernelDescriptorSymbol->setSize(
     512           6 :       MCConstantExpr::create(sizeof(KernelDescriptor), Context));
     513             : 
     514           6 :   MCSymbolELF *KernelCodeSymbol = cast<MCSymbolELF>(
     515           6 :       Context.getOrCreateSymbol(Twine(KernelName)));
     516           6 :   KernelCodeSymbol->setBinding(ELF::STB_LOCAL);
     517             : 
     518           6 :   Streamer.EmitLabel(KernelDescriptorSymbol);
     519           6 :   Streamer.EmitBytes(StringRef(
     520             :       (const char*)&(KernelDescriptor),
     521           6 :       offsetof(amdhsa::kernel_descriptor_t, kernel_code_entry_byte_offset)));
     522             :   // FIXME: Remove the use of VK_AMDGPU_REL64 in the expression below. The
     523             :   // expression being created is:
     524             :   //   (start of kernel code) - (start of kernel descriptor)
     525             :   // It implies R_AMDGPU_REL64, but ends up being R_AMDGPU_ABS64.
     526           6 :   Streamer.EmitValue(MCBinaryExpr::createSub(
     527          12 :       MCSymbolRefExpr::create(
     528           6 :           KernelCodeSymbol, MCSymbolRefExpr::VK_AMDGPU_REL64, Context),
     529          12 :       MCSymbolRefExpr::create(
     530           6 :           KernelDescriptorSymbol, MCSymbolRefExpr::VK_None, Context),
     531             :       Context),
     532           6 :       sizeof(KernelDescriptor.kernel_code_entry_byte_offset));
     533           6 :   Streamer.EmitBytes(StringRef(
     534             :       (const char*)&(KernelDescriptor) +
     535             :           offsetof(amdhsa::kernel_descriptor_t, kernel_code_entry_byte_offset) +
     536             :           sizeof(KernelDescriptor.kernel_code_entry_byte_offset),
     537             :       sizeof(KernelDescriptor) -
     538             :           offsetof(amdhsa::kernel_descriptor_t, kernel_code_entry_byte_offset) -
     539           6 :           sizeof(KernelDescriptor.kernel_code_entry_byte_offset)));
     540           6 : }

Generated by: LCOV version 1.13