lib/Target/AMDGPU/Utils/AMDKernelCodeTInfo.h

   1 //===--------------------- AMDKernelCodeTInfo.h ---------------------------===//
   2 //
   3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
   4 // See https://llvm.org/LICENSE.txt for license information.
   5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
   6 //
   7 //===----------------------------------------------------------------------===//
   8 //
   9 //===----------------------------------------------------------------------===//
  10 //
  11 /// \file - specifies tables for amd_kernel_code_t structure parsing/printing
  12 //
  13 //===----------------------------------------------------------------------===//
  14
  15 #define QNAME(name) amd_kernel_code_t::name
  16 #define FLD_T(name) decltype(QNAME(name)), &QNAME(name)
  17
  18 #define FIELD2(sname, aname, name) \
  19   RECORD(sname, aname, printField<FLD_T(name)>, parseField<FLD_T(name)>)
  20
  21 #define FIELD(name) FIELD2(name, name, name)
  22
  23
  24 #define PRINTCODEPROP(name) \
  25   printBitField<FLD_T(code_properties),\
  26                 AMD_CODE_PROPERTY_##name##_SHIFT,\
  27                 AMD_CODE_PROPERTY_##name##_WIDTH>
  28
  29 #define PARSECODEPROP(name) \
  30   parseBitField<FLD_T(code_properties),\
  31                 AMD_CODE_PROPERTY_##name##_SHIFT,\
  32                 AMD_CODE_PROPERTY_##name##_WIDTH>
  33
  34 #define CODEPROP(name, shift) \
  35   RECORD(name, name, PRINTCODEPROP(shift), PARSECODEPROP(shift))
  36
  37 // have to define these lambdas because of Set/GetMacro
  38 #define PRINTCOMP(GetMacro, Shift) \
  39 [](StringRef Name, const amd_kernel_code_t &C, raw_ostream &OS) { \
  40    printName(OS, Name) << \
  41      (int)GetMacro(C.compute_pgm_resource_registers >> Shift); \
  42 }
  43 #define PARSECOMP(SetMacro, Shift) \
  44 [](amd_kernel_code_t &C, MCAsmParser &MCParser, raw_ostream &Err) { \
  45    int64_t Value = 0; \
  46    if (!expectAbsExpression(MCParser, Value, Err)) \
  47      return false; \
  48    C.compute_pgm_resource_registers &= ~(SetMacro(0xFFFFFFFFFFFFFFFFULL) << Shift); \
  49    C.compute_pgm_resource_registers |= SetMacro(Value) << Shift; \
  50    return true; \
  51 }
  52
  53 #define COMPPGM(name, aname, GetMacro, SetMacro, Shift) \
  54   RECORD(name, aname, PRINTCOMP(GetMacro, Shift), PARSECOMP(SetMacro, Shift))
  55
  56 #define COMPPGM1(name, aname, AccMacro) \
  57   COMPPGM(name, aname, G_00B848_##AccMacro, S_00B848_##AccMacro, 0)
  58
  59 #define COMPPGM2(name, aname, AccMacro) \
  60   COMPPGM(name, aname, G_00B84C_##AccMacro, S_00B84C_##AccMacro, 32)
  61
  62 ///////////////////////////////////////////////////////////////////////////////
  63 // Begin of the table
  64 // Define RECORD(name, print, parse) in your code to get field definitions
  65 // and include this file
  66
  67 FIELD2(amd_code_version_major,        kernel_code_version_major,  amd_kernel_code_version_major),
  68 FIELD2(amd_code_version_minor,        kernel_code_version_minor,  amd_kernel_code_version_minor),
  69 FIELD2(amd_machine_kind,              machine_kind,               amd_machine_kind),
  70 FIELD2(amd_machine_version_major,     machine_version_major,      amd_machine_version_major),
  71 FIELD2(amd_machine_version_minor,     machine_version_minor,      amd_machine_version_minor),
  72 FIELD2(amd_machine_version_stepping,  machine_version_stepping,   amd_machine_version_stepping),
  73
  74 FIELD(kernel_code_entry_byte_offset),
  75 FIELD(kernel_code_prefetch_byte_size),
  76
  77 COMPPGM1(granulated_workitem_vgpr_count,  compute_pgm_rsrc1_vgprs,          VGPRS),
  78 COMPPGM1(granulated_wavefront_sgpr_count, compute_pgm_rsrc1_sgprs,          SGPRS),
  79 COMPPGM1(priority,                        compute_pgm_rsrc1_priority,       PRIORITY),
  80 COMPPGM1(float_mode,                      compute_pgm_rsrc1_float_mode,     FLOAT_MODE), // TODO: split float_mode
  81 COMPPGM1(priv,                            compute_pgm_rsrc1_priv,           PRIV),
  82 COMPPGM1(enable_dx10_clamp,               compute_pgm_rsrc1_dx10_clamp,     DX10_CLAMP),
  83 COMPPGM1(debug_mode,                      compute_pgm_rsrc1_debug_mode,     DEBUG_MODE),
  84 COMPPGM1(enable_ieee_mode,                compute_pgm_rsrc1_ieee_mode,      IEEE_MODE),
  85 COMPPGM1(enable_wgp_mode,                 compute_pgm_rsrc1_wgp_mode,       WGP_MODE),
  86 COMPPGM1(enable_mem_ordered,              compute_pgm_rsrc1_mem_ordered,    MEM_ORDERED),
  87 COMPPGM1(enable_fwd_progress,             compute_pgm_rsrc1_fwd_progress,   FWD_PROGRESS),
  88 // TODO: bulky
  89 // TODO: cdbg_user
  90 COMPPGM2(enable_sgpr_private_segment_wave_byte_offset, compute_pgm_rsrc2_scratch_en, SCRATCH_EN),
  91 COMPPGM2(user_sgpr_count,                 compute_pgm_rsrc2_user_sgpr,      USER_SGPR),
  92 COMPPGM2(enable_trap_handler,             compute_pgm_rsrc2_trap_handler,   TRAP_HANDLER),
  93 COMPPGM2(enable_sgpr_workgroup_id_x,      compute_pgm_rsrc2_tgid_x_en,      TGID_X_EN),
  94 COMPPGM2(enable_sgpr_workgroup_id_y,      compute_pgm_rsrc2_tgid_y_en,      TGID_Y_EN),
  95 COMPPGM2(enable_sgpr_workgroup_id_z,      compute_pgm_rsrc2_tgid_z_en,      TGID_Z_EN),
  96 COMPPGM2(enable_sgpr_workgroup_info,      compute_pgm_rsrc2_tg_size_en,     TG_SIZE_EN),
  97 COMPPGM2(enable_vgpr_workitem_id,         compute_pgm_rsrc2_tidig_comp_cnt, TIDIG_COMP_CNT),
  98 COMPPGM2(enable_exception_msb,            compute_pgm_rsrc2_excp_en_msb,    EXCP_EN_MSB), // TODO: split enable_exception_msb
  99 COMPPGM2(granulated_lds_size,             compute_pgm_rsrc2_lds_size,       LDS_SIZE),
 100 COMPPGM2(enable_exception,                compute_pgm_rsrc2_excp_en,        EXCP_EN), // TODO: split enable_exception
 101
 102 CODEPROP(enable_sgpr_private_segment_buffer,  ENABLE_SGPR_PRIVATE_SEGMENT_BUFFER),
 103 CODEPROP(enable_sgpr_dispatch_ptr,            ENABLE_SGPR_DISPATCH_PTR),
 104 CODEPROP(enable_sgpr_queue_ptr,               ENABLE_SGPR_QUEUE_PTR),
 105 CODEPROP(enable_sgpr_kernarg_segment_ptr,     ENABLE_SGPR_KERNARG_SEGMENT_PTR),
 106 CODEPROP(enable_sgpr_dispatch_id,             ENABLE_SGPR_DISPATCH_ID),
 107 CODEPROP(enable_sgpr_flat_scratch_init,       ENABLE_SGPR_FLAT_SCRATCH_INIT),
 108 CODEPROP(enable_sgpr_private_segment_size,    ENABLE_SGPR_PRIVATE_SEGMENT_SIZE),
 109 CODEPROP(enable_sgpr_grid_workgroup_count_x,  ENABLE_SGPR_GRID_WORKGROUP_COUNT_X),
 110 CODEPROP(enable_sgpr_grid_workgroup_count_y,  ENABLE_SGPR_GRID_WORKGROUP_COUNT_Y),
 111 CODEPROP(enable_sgpr_grid_workgroup_count_z,  ENABLE_SGPR_GRID_WORKGROUP_COUNT_Z),
 112 CODEPROP(enable_wavefront_size32,             ENABLE_WAVEFRONT_SIZE32),
 113 CODEPROP(enable_ordered_append_gds,           ENABLE_ORDERED_APPEND_GDS),
 114 CODEPROP(private_element_size,                PRIVATE_ELEMENT_SIZE),
 115 CODEPROP(is_ptr64,                            IS_PTR64),
 116 CODEPROP(is_dynamic_callstack,                IS_DYNAMIC_CALLSTACK),
 117 CODEPROP(is_debug_enabled,                    IS_DEBUG_SUPPORTED),
 118 CODEPROP(is_xnack_enabled,                    IS_XNACK_SUPPORTED),
 119
 120 FIELD(workitem_private_segment_byte_size),
 121 FIELD(workgroup_group_segment_byte_size),
 122 FIELD(gds_segment_byte_size),
 123 FIELD(kernarg_segment_byte_size),
 124 FIELD(workgroup_fbarrier_count),
 125 FIELD(wavefront_sgpr_count),
 126 FIELD(workitem_vgpr_count),
 127 FIELD(reserved_vgpr_first),
 128 FIELD(reserved_vgpr_count),
 129 FIELD(reserved_sgpr_first),
 130 FIELD(reserved_sgpr_count),
 131 FIELD(debug_wavefront_private_segment_offset_sgpr),
 132 FIELD(debug_private_segment_buffer_sgpr),
 133 FIELD(kernarg_segment_alignment),
 134 FIELD(group_segment_alignment),
 135 FIELD(private_segment_alignment),
 136 FIELD(wavefront_size),
 137 FIELD(call_convention),
 138 FIELD(runtime_loader_kernel_symbol)
 139 // TODO: control_directive
 140
 141 // end of the table
 142 ///////////////////////////////////////////////////////////////////////////////
 143
 144 #undef QNAME
 145 #undef FLD_T
 146 #undef FIELD2
 147 #undef FIELD
 148 #undef PRINTCODEPROP
 149 #undef PARSECODEPROP
 150 #undef CODEPROP
 151 #undef PRINTCOMP
 152 #undef PAPSECOMP
 153 #undef COMPPGM
 154 #undef COMPPGM1
 155 #undef COMPPGM2