SemaAMDGPU.cpp

Go to the documentation of this file.

//===------ SemaAMDGPU.cpp ------- AMDGPU target-specific routines --------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
//
//  This file implements semantic analysis functions specific to AMDGPU.
//
//===----------------------------------------------------------------------===//
 
#include "clang/Sema/SemaAMDGPU.h"
#include "clang/AST/Decl.h"
#include "clang/AST/DynamicRecursiveASTVisitor.h"
#include "clang/AST/Expr.h"
#include "clang/Basic/DiagnosticFrontend.h"
#include "clang/Basic/DiagnosticSema.h"
#include "clang/Basic/TargetBuiltins.h"
#include "clang/Basic/TargetInfo.h"
#include "clang/Sema/Ownership.h"
#include "clang/Sema/Scope.h"
#include "clang/Sema/Sema.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/ADT/StringExtras.h"
#include "llvm/ADT/StringMap.h"
#include "llvm/Support/AMDGPUAddrSpace.h"
#include "llvm/Support/AtomicOrdering.h"
#include "llvm/TargetParser/AMDGPUTargetParser.h"
#include "llvm/TargetParser/AtomicScope.h"
#include <cstdint>
#include <utility>
 
namespace clang {
 
SemaAMDGPU::SemaAMDGPU(Sema &S) : SemaBase(S) {}
 
bool SemaAMDGPU::CheckAMDGCNBuiltinFunctionCall(const TargetInfo &TI,
                                                unsigned BuiltinID,
                                                CallExpr *TheCall) {
  const auto *FD = SemaRef.getCurFunctionDecl(/*AllowLambda=*/true);
  assert(FD && "AMDGPU builtins should not be used outside of a function");
  llvm::StringMap<bool> CallerFeatureMap;
  getASTContext().getFunctionFeatureMap(CallerFeatureMap, FD);
  bool HasGFX950Insts =
      Builtin::evaluateRequiredTargetFeatures("gfx950-insts", CallerFeatureMap);
 
  switch (BuiltinID) {
  case AMDGPU::BI__builtin_amdgcn_raw_ptr_buffer_load_lds:
  case AMDGPU::BI__builtin_amdgcn_raw_ptr_buffer_load_async_lds:
  case AMDGPU::BI__builtin_amdgcn_struct_ptr_buffer_load_lds:
  case AMDGPU::BI__builtin_amdgcn_struct_ptr_buffer_load_async_lds:
  case AMDGPU::BI__builtin_amdgcn_load_to_lds:
  case AMDGPU::BI__builtin_amdgcn_load_async_to_lds:
  case AMDGPU::BI__builtin_amdgcn_global_load_lds:
  case AMDGPU::BI__builtin_amdgcn_global_load_async_lds: {
    constexpr const int SizeIdx = 2;
    llvm::APSInt Size;
    Expr *ArgExpr = TheCall->getArg(SizeIdx);
    // Check for instantiation-dependent expressions (e.g., involving template
    // parameters). These will be checked again during template instantiation.
    if (ArgExpr->isInstantiationDependent())
      return false;
    [[maybe_unused]] ExprResult R =
        SemaRef.VerifyIntegerConstantExpression(ArgExpr, &Size);
    assert(!R.isInvalid());
    switch (Size.getSExtValue()) {
    case 1:
    case 2:
    case 4:
      return false;
    case 12:
    case 16: {
      if (HasGFX950Insts)
        return false;
      [[fallthrough]];
    }
    default:
      SemaRef.targetDiag(ArgExpr->getExprLoc(),
                         diag::err_amdgcn_load_lds_size_invalid_value)
          << ArgExpr->getSourceRange();
      SemaRef.targetDiag(ArgExpr->getExprLoc(),
                         diag::note_amdgcn_load_lds_size_valid_value)
          << HasGFX950Insts << ArgExpr->getSourceRange();
      return true;
    }
  }
  case AMDGPU::BI__builtin_amdgcn_uicmp:
  case AMDGPU::BI__builtin_amdgcn_uicmpl:
  case AMDGPU::BI__builtin_amdgcn_sicmp:
  case AMDGPU::BI__builtin_amdgcn_sicmpl:
  case AMDGPU::BI__builtin_amdgcn_fcmp:
  case AMDGPU::BI__builtin_amdgcn_fcmpf: {
    // These builtins are deprecated in favor of
    // __builtin_amdgcn_ballot_{w32|w64}. Suggest the replacement matching the
    // wavefront size of the calling function.
    bool IsWave32 = Builtin::evaluateRequiredTargetFeatures("wavefrontsize32",
                                                            CallerFeatureMap);
    Diag(TheCall->getBeginLoc(), diag::warn_deprecated_builtin)
        << getASTContext().BuiltinInfo.getQuotedName(BuiltinID)
        << (IsWave32 ? "__builtin_amdgcn_ballot_w32"
                     : "__builtin_amdgcn_ballot_w64");
    return false;
  }
  case AMDGPU::BI__builtin_amdgcn_get_fpenv:
  case AMDGPU::BI__builtin_amdgcn_set_fpenv:
    return false;
  case AMDGPU::BI__builtin_amdgcn_atomic_inc32:
  case AMDGPU::BI__builtin_amdgcn_atomic_inc64:
  case AMDGPU::BI__builtin_amdgcn_atomic_dec32:
  case AMDGPU::BI__builtin_amdgcn_atomic_dec64:
  case AMDGPU::BI__builtin_amdgcn_fence: {
    bool IsFence = BuiltinID == AMDGPU::BI__builtin_amdgcn_fence;
    unsigned OrderIndex = IsFence ? 0 : 2;
    unsigned ScopeIndex = IsFence ? 1 : 3;
    Expr *OrderExpr = TheCall->getArg(OrderIndex);
    Expr *ScopeExpr = TheCall->getArg(ScopeIndex);
 
    // Checks requiring constant evaluation are deferred until instantiation.
    if (OrderExpr->isInstantiationDependent() ||
        ScopeExpr->isInstantiationDependent())
      return false;
 
    Expr::EvalResult OrderResult;
    if (!OrderExpr->EvaluateAsInt(OrderResult, getASTContext()))
      return Diag(OrderExpr->getExprLoc(), diag::err_typecheck_expect_int)
             << OrderExpr->getType();
    uint64_t Ord = OrderResult.Val.getInt().getZExtValue();
 
    // Check validity of memory ordering as per C11 / C++11's memory model.
    // Only fence needs check. Atomic dec/inc allow all memory orders.
    if (!llvm::isValidAtomicOrderingCABI(Ord))
      return Diag(OrderExpr->getBeginLoc(),
                  diag::warn_atomic_op_has_invalid_memory_order)
             << 0 << OrderExpr->getSourceRange();
    switch (static_cast<llvm::AtomicOrderingCABI>(Ord)) {
    case llvm::AtomicOrderingCABI::relaxed:
    case llvm::AtomicOrderingCABI::consume:
      if (IsFence)
        return Diag(OrderExpr->getBeginLoc(),
                    diag::warn_atomic_op_has_invalid_memory_order)
               << 0 << OrderExpr->getSourceRange();
      break;
    case llvm::AtomicOrderingCABI::acquire:
    case llvm::AtomicOrderingCABI::release:
    case llvm::AtomicOrderingCABI::acq_rel:
    case llvm::AtomicOrderingCABI::seq_cst:
      break;
    }
 
    Expr::EvalResult ScopeResult;
    // Check that sync scope is a constant literal
    if (!ScopeExpr->EvaluateAsConstantExpr(ScopeResult, getASTContext()))
      return Diag(ScopeExpr->getExprLoc(), diag::err_expr_not_string_literal)
             << ScopeExpr->getType();
 
    // Reject any string that is not a valid synchronization scope name.
    std::optional<std::string> ScopeName =
        ScopeExpr->tryEvaluateString(getASTContext());
    const llvm::Triple &TT = TI.getTriple();
    if (ScopeName && !llvm::parseAtomicScopeIRString(TT, *ScopeName)) {
      return Diag(ScopeExpr->getExprLoc(), diag::err_invalid_sync_scope)
             << *ScopeName << ScopeExpr->getSourceRange();
    }
 
    return false;
  }
  case AMDGPU::BI__builtin_amdgcn_s_setreg:
    return SemaRef.BuiltinConstantArgRange(TheCall, /*ArgNum=*/0, /*Low=*/0,
                                           /*High=*/UINT16_MAX);
  case AMDGPU::BI__builtin_amdgcn_s_wait_event: {
    llvm::APSInt Result;
    if (SemaRef.BuiltinConstantArg(TheCall, 0, Result))
      return true;
 
    bool IsGFX12Plus = Builtin::evaluateRequiredTargetFeatures(
        "gfx12-insts", CallerFeatureMap);
 
    // gfx11 -> gfx12 changed the interpretation of the bitmask. gfx12 inverted
    // the intepretation for export_ready, but shifted the used bit by 1. Thus
    // waiting for the export_ready event can use a value of 2 universally.
    if (((IsGFX12Plus && !Result[1]) || (!IsGFX12Plus && Result[0])) ||
        Result.getZExtValue() > 2) {
      Expr *ArgExpr = TheCall->getArg(0);
      SemaRef.targetDiag(ArgExpr->getExprLoc(),
                         diag::warn_amdgpu_s_wait_event_mask_no_effect_target)
          << ArgExpr->getSourceRange();
      SemaRef.targetDiag(ArgExpr->getExprLoc(),
                         diag::note_amdgpu_s_wait_event_suggested_value)
          << ArgExpr->getSourceRange();
    }
 
    return false;
  }
  case AMDGPU::BI__builtin_amdgcn_mov_dpp:
    return checkMovDPPFunctionCall(TheCall, 5, 1);
  case AMDGPU::BI__builtin_amdgcn_mov_dpp8:
    return checkMovDPPFunctionCall(TheCall, 2, 1);
  case AMDGPU::BI__builtin_amdgcn_update_dpp:
    return checkMovDPPFunctionCall(TheCall, 6, 2);
  case AMDGPU::BI__builtin_amdgcn_cvt_scale_pk8_f16_fp8:
  case AMDGPU::BI__builtin_amdgcn_cvt_scale_pk8_bf16_fp8:
  case AMDGPU::BI__builtin_amdgcn_cvt_scale_pk8_f16_bf8:
  case AMDGPU::BI__builtin_amdgcn_cvt_scale_pk8_bf16_bf8:
  case AMDGPU::BI__builtin_amdgcn_cvt_scale_pk8_f16_fp4:
  case AMDGPU::BI__builtin_amdgcn_cvt_scale_pk8_bf16_fp4:
  case AMDGPU::BI__builtin_amdgcn_cvt_scale_pk8_f32_fp8:
  case AMDGPU::BI__builtin_amdgcn_cvt_scale_pk8_f32_bf8:
  case AMDGPU::BI__builtin_amdgcn_cvt_scale_pk8_f32_fp4:
  case AMDGPU::BI__builtin_amdgcn_cvt_scale_pk16_f16_fp6:
  case AMDGPU::BI__builtin_amdgcn_cvt_scale_pk16_bf16_fp6:
  case AMDGPU::BI__builtin_amdgcn_cvt_scale_pk16_f16_bf6:
  case AMDGPU::BI__builtin_amdgcn_cvt_scale_pk16_bf16_bf6:
  case AMDGPU::BI__builtin_amdgcn_cvt_scale_pk16_f32_fp6:
  case AMDGPU::BI__builtin_amdgcn_cvt_scale_pk16_f32_bf6:
    return SemaRef.BuiltinConstantArgRange(TheCall, 2, 0, 15);
  case AMDGPU::BI__builtin_amdgcn_av_load_b128:
    return checkAVLoadStore(TheCall, /*IsStore=*/false);
  case AMDGPU::BI__builtin_amdgcn_av_store_b128:
    return checkAVLoadStore(TheCall, /*IsStore=*/true);
  case AMDGPU::BI__builtin_amdgcn_cooperative_atomic_load_32x4B:
  case AMDGPU::BI__builtin_amdgcn_cooperative_atomic_load_16x8B:
  case AMDGPU::BI__builtin_amdgcn_cooperative_atomic_load_8x16B:
    return checkCoopAtomicFunctionCall(TheCall, /*IsStore=*/false);
  case AMDGPU::BI__builtin_amdgcn_cooperative_atomic_store_32x4B:
  case AMDGPU::BI__builtin_amdgcn_cooperative_atomic_store_16x8B:
  case AMDGPU::BI__builtin_amdgcn_cooperative_atomic_store_8x16B:
    return checkCoopAtomicFunctionCall(TheCall, /*IsStore=*/true);
  case AMDGPU::BI__builtin_amdgcn_flat_load_monitor_b32:
  case AMDGPU::BI__builtin_amdgcn_flat_load_monitor_b64:
  case AMDGPU::BI__builtin_amdgcn_flat_load_monitor_b128:
  case AMDGPU::BI__builtin_amdgcn_global_load_monitor_b32:
  case AMDGPU::BI__builtin_amdgcn_global_load_monitor_b64:
  case AMDGPU::BI__builtin_amdgcn_global_load_monitor_b128:
    return checkAtomicMonitorLoad(TheCall);
  case AMDGPU::BI__builtin_amdgcn_image_load_1d_v4f32_i32:
  case AMDGPU::BI__builtin_amdgcn_image_load_1darray_v4f32_i32:
  case AMDGPU::BI__builtin_amdgcn_image_load_1d_v4f16_i32:
  case AMDGPU::BI__builtin_amdgcn_image_load_1darray_v4f16_i32:
  case AMDGPU::BI__builtin_amdgcn_image_load_2d_f32_i32:
  case AMDGPU::BI__builtin_amdgcn_image_load_2d_v4f32_i32:
  case AMDGPU::BI__builtin_amdgcn_image_load_2d_v4f16_i32:
  case AMDGPU::BI__builtin_amdgcn_image_load_2darray_f32_i32:
  case AMDGPU::BI__builtin_amdgcn_image_load_2darray_v4f32_i32:
  case AMDGPU::BI__builtin_amdgcn_image_load_2darray_v4f16_i32:
  case AMDGPU::BI__builtin_amdgcn_image_load_3d_v4f32_i32:
  case AMDGPU::BI__builtin_amdgcn_image_load_3d_v4f16_i32:
  case AMDGPU::BI__builtin_amdgcn_image_load_cube_v4f32_i32:
  case AMDGPU::BI__builtin_amdgcn_image_load_cube_v4f16_i32:
  case AMDGPU::BI__builtin_amdgcn_image_load_mip_1d_v4f32_i32:
  case AMDGPU::BI__builtin_amdgcn_image_load_mip_1d_v4f16_i32:
  case AMDGPU::BI__builtin_amdgcn_image_load_mip_1darray_v4f32_i32:
  case AMDGPU::BI__builtin_amdgcn_image_load_mip_1darray_v4f16_i32:
  case AMDGPU::BI__builtin_amdgcn_image_load_mip_2d_f32_i32:
  case AMDGPU::BI__builtin_amdgcn_image_load_mip_2d_v4f32_i32:
  case AMDGPU::BI__builtin_amdgcn_image_load_mip_2d_v4f16_i32:
  case AMDGPU::BI__builtin_amdgcn_image_load_mip_2darray_f32_i32:
  case AMDGPU::BI__builtin_amdgcn_image_load_mip_2darray_v4f32_i32:
  case AMDGPU::BI__builtin_amdgcn_image_load_mip_2darray_v4f16_i32:
  case AMDGPU::BI__builtin_amdgcn_image_load_mip_3d_v4f32_i32:
  case AMDGPU::BI__builtin_amdgcn_image_load_mip_3d_v4f16_i32:
  case AMDGPU::BI__builtin_amdgcn_image_load_mip_cube_v4f32_i32:
  case AMDGPU::BI__builtin_amdgcn_image_load_mip_cube_v4f16_i32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_1d_v4f32_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_1darray_v4f32_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_1d_v4f16_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_1darray_v4f16_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_2d_f32_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_2d_v4f32_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_2d_v4f16_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_2darray_f32_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_2darray_v4f32_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_2darray_v4f16_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_3d_v4f32_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_3d_v4f16_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_cube_v4f32_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_cube_v4f16_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_lz_1d_v4f32_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_lz_1d_v4f16_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_lz_1darray_v4f32_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_lz_1darray_v4f16_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_lz_2d_f32_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_lz_2d_v4f32_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_lz_2d_v4f16_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_lz_2darray_f32_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_lz_2darray_v4f32_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_lz_2darray_v4f16_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_lz_3d_v4f32_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_lz_3d_v4f16_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_lz_cube_v4f32_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_lz_cube_v4f16_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_l_1d_v4f32_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_l_1d_v4f16_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_l_1darray_v4f32_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_l_1darray_v4f16_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_l_2d_f32_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_l_2d_v4f16_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_l_2d_v4f32_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_l_2darray_f32_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_l_2darray_v4f32_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_l_2darray_v4f16_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_l_3d_v4f32_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_l_3d_v4f16_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_l_cube_v4f32_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_l_cube_v4f16_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_d_1d_v4f32_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_d_1d_v4f16_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_d_1darray_v4f32_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_d_1darray_v4f16_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_d_2d_f32_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_d_2d_v4f32_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_d_2d_v4f16_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_d_2darray_f32_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_d_2darray_v4f32_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_d_2darray_v4f16_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_d_3d_v4f32_f32:
  case AMDGPU::BI__builtin_amdgcn_image_sample_d_3d_v4f16_f32:
  case AMDGPU::BI__builtin_amdgcn_image_gather4_lz_2d_v4f32_f32:
  case AMDGPU::BI__builtin_amdgcn_image_gather4_lz_2d_v4f16_f32: {
    StringRef FeatureList(
        getASTContext().BuiltinInfo.getRequiredFeatures(BuiltinID));
    if (!Builtin::evaluateRequiredTargetFeatures(FeatureList,
                                                 CallerFeatureMap)) {
      Diag(TheCall->getBeginLoc(), diag::err_builtin_needs_feature)
          << FD->getDeclName() << FeatureList;
      return false;
    }
 
    unsigned ArgCount = TheCall->getNumArgs() - 1;
    llvm::APSInt Result;
 
    // Compilain about dmask values which are too huge to fully fit into 4 bits
    // (which is the actual size of the dmask in corresponding HW instructions).
    constexpr unsigned DMaskArgNo = 0;
    constexpr int Low = 0;
    constexpr int High = 15;
    if (SemaRef.BuiltinConstantArg(TheCall, DMaskArgNo, Result) ||
        SemaRef.BuiltinConstantArgRange(TheCall, DMaskArgNo, Low, High,
                                        /* RangeIsError = */ true))
      return true;
 
    // Dmask indicates which elements should be returned and it is not possible
    // to return more values than there are elements in return type.
    int NumElementsInRetTy = 1;
    const Type *RetTy = TheCall->getType().getTypePtr();
    if (auto *VTy = dyn_cast<VectorType>(RetTy))
      NumElementsInRetTy = VTy->getNumElements();
    int NumActiveBitsInDMask =
        llvm::popcount(static_cast<uint8_t>(Result.getExtValue()));
    if (NumActiveBitsInDMask > NumElementsInRetTy) {
      Diag(TheCall->getBeginLoc(),
           diag::err_amdgcn_dmask_has_too_many_bits_set);
      return true;
    }
 
    // For gather, only one bit can be set indicating which exact component to
    // return.
    bool ExtraGatherChecks =
        (BuiltinID ==
             AMDGPU::BI__builtin_amdgcn_image_gather4_lz_2d_v4f32_f32 ||
         BuiltinID ==
             AMDGPU::BI__builtin_amdgcn_image_gather4_lz_2d_v4f16_f32) &&
        SemaRef.BuiltinConstantArgPower2(TheCall, 0);
 
    return ExtraGatherChecks ||
           (SemaRef.BuiltinConstantArg(TheCall, ArgCount, Result)) ||
           (SemaRef.BuiltinConstantArg(TheCall, (ArgCount - 1), Result));
  }
  case AMDGPU::BI__builtin_amdgcn_image_store_1d_v4f32_i32:
  case AMDGPU::BI__builtin_amdgcn_image_store_1darray_v4f32_i32:
  case AMDGPU::BI__builtin_amdgcn_image_store_1d_v4f16_i32:
  case AMDGPU::BI__builtin_amdgcn_image_store_1darray_v4f16_i32:
  case AMDGPU::BI__builtin_amdgcn_image_store_2d_f32_i32:
  case AMDGPU::BI__builtin_amdgcn_image_store_2d_v4f32_i32:
  case AMDGPU::BI__builtin_amdgcn_image_store_2d_v4f16_i32:
  case AMDGPU::BI__builtin_amdgcn_image_store_2darray_f32_i32:
  case AMDGPU::BI__builtin_amdgcn_image_store_2darray_v4f32_i32:
  case AMDGPU::BI__builtin_amdgcn_image_store_2darray_v4f16_i32:
  case AMDGPU::BI__builtin_amdgcn_image_store_3d_v4f32_i32:
  case AMDGPU::BI__builtin_amdgcn_image_store_3d_v4f16_i32:
  case AMDGPU::BI__builtin_amdgcn_image_store_cube_v4f32_i32:
  case AMDGPU::BI__builtin_amdgcn_image_store_cube_v4f16_i32:
  case AMDGPU::BI__builtin_amdgcn_image_store_mip_1d_v4f32_i32:
  case AMDGPU::BI__builtin_amdgcn_image_store_mip_1d_v4f16_i32:
  case AMDGPU::BI__builtin_amdgcn_image_store_mip_1darray_v4f32_i32:
  case AMDGPU::BI__builtin_amdgcn_image_store_mip_1darray_v4f16_i32:
  case AMDGPU::BI__builtin_amdgcn_image_store_mip_2d_f32_i32:
  case AMDGPU::BI__builtin_amdgcn_image_store_mip_2d_v4f32_i32:
  case AMDGPU::BI__builtin_amdgcn_image_store_mip_2d_v4f16_i32:
  case AMDGPU::BI__builtin_amdgcn_image_store_mip_2darray_f32_i32:
  case AMDGPU::BI__builtin_amdgcn_image_store_mip_2darray_v4f32_i32:
  case AMDGPU::BI__builtin_amdgcn_image_store_mip_2darray_v4f16_i32:
  case AMDGPU::BI__builtin_amdgcn_image_store_mip_3d_v4f32_i32:
  case AMDGPU::BI__builtin_amdgcn_image_store_mip_3d_v4f16_i32:
  case AMDGPU::BI__builtin_amdgcn_image_store_mip_cube_v4f32_i32:
  case AMDGPU::BI__builtin_amdgcn_image_store_mip_cube_v4f16_i32: {
    StringRef FeatureList(
        getASTContext().BuiltinInfo.getRequiredFeatures(BuiltinID));
    if (!Builtin::evaluateRequiredTargetFeatures(FeatureList,
                                                 CallerFeatureMap)) {
      Diag(TheCall->getBeginLoc(), diag::err_builtin_needs_feature)
          << FD->getDeclName() << FeatureList;
      return false;
    }
 
    unsigned ArgCount = TheCall->getNumArgs() - 1;
    llvm::APSInt Result;
 
    // Complain about dmask values which are too huge to fully fit into 4 bits
    // (which is the actual size of the dmask in corresponding HW instructions).
    constexpr unsigned DMaskArgNo = 1;
    return SemaRef.BuiltinConstantArgRange(TheCall, DMaskArgNo, /*Low=*/0,
                                           /*High=*/15,
                                           /*RangeIsError=*/true) ||
           SemaRef.BuiltinConstantArg(TheCall, ArgCount, Result) ||
           SemaRef.BuiltinConstantArg(TheCall, (ArgCount - 1), Result);
  }
  case AMDGPU::BI__builtin_amdgcn_wmma_i32_16x16x64_iu8:
  case AMDGPU::BI__builtin_amdgcn_swmmac_i32_16x16x128_iu8: {
    if (BuiltinID == AMDGPU::BI__builtin_amdgcn_wmma_i32_16x16x64_iu8) {
      if (SemaRef.checkArgCountRange(TheCall, 7, 8))
        return true;
      if (TheCall->getNumArgs() == 7)
        return false;
    } else if (BuiltinID ==
               AMDGPU::BI__builtin_amdgcn_swmmac_i32_16x16x128_iu8) {
      if (SemaRef.checkArgCountRange(TheCall, 8, 9))
        return true;
      if (TheCall->getNumArgs() == 8)
        return false;
    }
    // Check if the last argument (clamp operand) is a constant and is
    // convertible to bool.
    Expr *ClampArg = TheCall->getArg(TheCall->getNumArgs() - 1);
    // 1) Ensure clamp argument is a constant expression
    llvm::APSInt ClampValue;
    if (!SemaRef.VerifyIntegerConstantExpression(ClampArg, &ClampValue)
             .isUsable())
      return true;
    // 2) Check if the argument can be converted to bool type
    if (!SemaRef.Context.hasSameType(ClampArg->getType(),
                                     SemaRef.Context.BoolTy)) {
      // Try to convert to bool
      QualType BoolTy = SemaRef.Context.BoolTy;
      ExprResult ClampExpr(ClampArg);
      SemaRef.CheckSingleAssignmentConstraints(BoolTy, ClampExpr);
      if (ClampExpr.isInvalid())
        return true;
    }
    return false;
  }
  case AMDGPU::BI__builtin_amdgcn_wmma_f32_16x16x32_bf16:
  case AMDGPU::BI__builtin_amdgcn_wmma_f32_16x16x4_f32:
  case AMDGPU::BI__builtin_amdgcn_wmma_f32_16x16x32_f16:
  case AMDGPU::BI__builtin_amdgcn_wmma_f16_16x16x32_f16:
  case AMDGPU::BI__builtin_amdgcn_wmma_bf16_16x16x32_bf16:
  case AMDGPU::BI__builtin_amdgcn_wmma_bf16f32_16x16x32_bf16:
    return SemaRef.BuiltinConstantArgRange(TheCall, /*ArgNum=*/0, /*Low=*/0,
                                           /*High=*/0) ||
           SemaRef.BuiltinConstantArgRange(TheCall, /*ArgNum=*/2, /*Low=*/0,
                                           /*High=*/0);
  default:
    return false;
  }
}
 
bool SemaAMDGPU::checkAtomicOrderingCABIArg(Expr *E, bool MayLoad,
                                            bool MayStore) {
  Expr::EvalResult AtomicOrdArgRes;
  if (!E->EvaluateAsInt(AtomicOrdArgRes, getASTContext()))
    llvm_unreachable("Intrinsic requires imm for atomic ordering argument!");
  auto Ord =
      llvm::AtomicOrderingCABI(AtomicOrdArgRes.Val.getInt().getZExtValue());
 
  // Atomic ordering cannot be acq_rel in any case, acquire for stores or
  // release for loads.
  if (!llvm::isValidAtomicOrderingCABI((unsigned)Ord) ||
      (!(MayLoad && MayStore) && (Ord == llvm::AtomicOrderingCABI::acq_rel)) ||
      (!MayLoad && Ord == llvm::AtomicOrderingCABI::acquire) ||
      (!MayStore && Ord == llvm::AtomicOrderingCABI::release)) {
    return Diag(E->getBeginLoc(), diag::warn_atomic_op_has_invalid_memory_order)
           << 0 << E->getSourceRange();
  }
 
  return false;
}
 
// Check that the first argument to TheCall is a global or generic pointer.
static bool checkGlobalOrFlatPointerArg(SemaAMDGPU &S, CallExpr *TheCall) {
  Expr *PtrArg = TheCall->getArg(0);
  QualType PtrTy = PtrArg->getType()->getPointeeType();
  unsigned AS =
      S.getASTContext().getTargetAddressSpace(PtrTy.getAddressSpace());
  if (AS != llvm::AMDGPUAS::FLAT_ADDRESS &&
      AS != llvm::AMDGPUAS::GLOBAL_ADDRESS) {
    return S.Diag(TheCall->getBeginLoc(),
                  diag::err_amdgcn_global_or_flat_pointer_required)
           << PtrArg->getSourceRange();
  }
  return false;
}
 
static bool checkScopeAsInt(SemaAMDGPU &S, Expr *Scope) {
  if (Scope->isValueDependent())
    return false;
  auto ScopeModel = AtomicScopeModel::create(AtomicScopeModelKind::Generic);
  if (std::optional<llvm::APSInt> Result =
          Scope->getIntegerConstantExpr(S.SemaRef.Context)) {
    if (!ScopeModel->isValid(Result->getZExtValue())) {
      return S.Diag(Scope->getBeginLoc(),
                    diag::err_atomic_op_has_invalid_sync_scope)
             << Scope->getSourceRange();
    }
  }
  return false;
}
 
bool SemaAMDGPU::checkAVLoadStore(CallExpr *TheCall, bool IsStore) {
  if (checkGlobalOrFlatPointerArg(*this, TheCall))
    return true;
 
  Expr *Scope = TheCall->getArg(TheCall->getNumArgs() - 1);
  return checkScopeAsInt(*this, Scope);
}
 
bool SemaAMDGPU::checkCoopAtomicFunctionCall(CallExpr *TheCall, bool IsStore) {
  bool Fail = checkGlobalOrFlatPointerArg(*this, TheCall);
 
  Expr *AO = TheCall->getArg(IsStore ? 2 : 1);
  Expr *Scope = TheCall->getArg(TheCall->getNumArgs() - 1);
 
  if (AO->isValueDependent() || Scope->isValueDependent())
    return false;
 
  // Check atomic ordering
  Fail |=
      checkAtomicOrderingCABIArg(TheCall->getArg(IsStore ? 2 : 1),
                                 /*MayLoad=*/!IsStore, /*MayStore=*/IsStore);
 
  // Last argument is the syncscope as a string literal.
  if (!isa<StringLiteral>(Scope->IgnoreParenImpCasts())) {
    Diag(TheCall->getBeginLoc(), diag::err_expr_not_string_literal)
        << Scope->getSourceRange();
    Fail = true;
  }
 
  return Fail;
}
 
bool SemaAMDGPU::checkAtomicMonitorLoad(CallExpr *TheCall) {
  Expr *AO = TheCall->getArg(1);
  Expr *Scope = TheCall->getArg(TheCall->getNumArgs() - 1);
 
  if (AO->isValueDependent() || Scope->isValueDependent())
    return false;
 
  bool Fail = checkAtomicOrderingCABIArg(AO, /*MayLoad=*/true,
                                         /*MayStore=*/false);
  Fail |= checkScopeAsInt(*this, Scope);
  return Fail;
}
 
bool SemaAMDGPU::checkMovDPPFunctionCall(CallExpr *TheCall, unsigned NumArgs,
                                         unsigned NumDataArgs) {
  assert(NumDataArgs <= 2);
  if (SemaRef.checkArgCountRange(TheCall, NumArgs, NumArgs))
    return true;
  Expr *Args[2];
  QualType ArgTys[2];
  for (unsigned I = 0; I != NumDataArgs; ++I) {
    Args[I] = TheCall->getArg(I);
    ArgTys[I] = Args[I]->getType();
    // TODO: Vectors can also be supported.
    if (!ArgTys[I]->isArithmeticType() || ArgTys[I]->isAnyComplexType()) {
      SemaRef.Diag(Args[I]->getBeginLoc(),
                   diag::err_typecheck_cond_expect_int_float)
          << ArgTys[I] << Args[I]->getSourceRange();
      return true;
    }
  }
  if (NumDataArgs < 2)
    return false;
 
  if (getASTContext().hasSameUnqualifiedType(ArgTys[0], ArgTys[1]))
    return false;
 
  if (((ArgTys[0]->isUnsignedIntegerType() &&
        ArgTys[1]->isSignedIntegerType()) ||
       (ArgTys[0]->isSignedIntegerType() &&
        ArgTys[1]->isUnsignedIntegerType())) &&
      getASTContext().getTypeSize(ArgTys[0]) ==
          getASTContext().getTypeSize(ArgTys[1]))
    return false;
 
  SemaRef.Diag(Args[1]->getBeginLoc(),
               diag::err_typecheck_call_different_arg_types)
      << ArgTys[0] << ArgTys[1];
  return true;
}
 
static bool
checkAMDGPUFlatWorkGroupSizeArguments(Sema &S, Expr *MinExpr, Expr *MaxExpr,
                                      const AMDGPUFlatWorkGroupSizeAttr &Attr) {
  // Accept template arguments for now as they depend on something else.
  // We'll get to check them when they eventually get instantiated.
  if (MinExpr->isValueDependent() || MaxExpr->isValueDependent())
    return false;
 
  uint32_t Min = 0;
  if (!S.checkUInt32Argument(Attr, MinExpr, Min, 0))
    return true;
 
  uint32_t Max = 0;
  if (!S.checkUInt32Argument(Attr, MaxExpr, Max, 1))
    return true;
 
  if (Min == 0 && Max != 0) {
    S.Diag(Attr.getLocation(), diag::err_attribute_argument_invalid)
        << &Attr << 0;
    return true;
  }
  if (Min > Max) {
    S.Diag(Attr.getLocation(), diag::err_attribute_argument_invalid)
        << &Attr << 1;
    return true;
  }
 
  return false;
}
 
AMDGPUFlatWorkGroupSizeAttr *
SemaAMDGPU::CreateAMDGPUFlatWorkGroupSizeAttr(const AttributeCommonInfo &CI,
                                              Expr *MinExpr, Expr *MaxExpr) {
  ASTContext &Context = getASTContext();
  AMDGPUFlatWorkGroupSizeAttr TmpAttr(Context, CI, MinExpr, MaxExpr);
 
  if (checkAMDGPUFlatWorkGroupSizeArguments(SemaRef, MinExpr, MaxExpr, TmpAttr))
    return nullptr;
  return ::new (Context)
      AMDGPUFlatWorkGroupSizeAttr(Context, CI, MinExpr, MaxExpr);
}
 
void SemaAMDGPU::addAMDGPUFlatWorkGroupSizeAttr(Decl *D,
                                                const AttributeCommonInfo &CI,
                                                Expr *MinExpr, Expr *MaxExpr) {
  if (auto *Attr = CreateAMDGPUFlatWorkGroupSizeAttr(CI, MinExpr, MaxExpr))
    D->addAttr(Attr);
}
 
void SemaAMDGPU::handleAMDGPUFlatWorkGroupSizeAttr(Decl *D,
                                                   const ParsedAttr &AL) {
  Expr *MinExpr = AL.getArgAsExpr(0);
  Expr *MaxExpr = AL.getArgAsExpr(1);
 
  addAMDGPUFlatWorkGroupSizeAttr(D, AL, MinExpr, MaxExpr);
}
 
static bool checkAMDGPUWavesPerEUArguments(Sema &S, Expr *MinExpr,
                                           Expr *MaxExpr,
                                           const AMDGPUWavesPerEUAttr &Attr) {
  if (S.DiagnoseUnexpandedParameterPack(MinExpr) ||
      (MaxExpr && S.DiagnoseUnexpandedParameterPack(MaxExpr)))
    return true;
 
  // Accept template arguments for now as they depend on something else.
  // We'll get to check them when they eventually get instantiated.
  if (MinExpr->isValueDependent() || (MaxExpr && MaxExpr->isValueDependent()))
    return false;
 
  uint32_t Min = 0;
  if (!S.checkUInt32Argument(Attr, MinExpr, Min, 0))
    return true;
 
  uint32_t Max = 0;
  if (MaxExpr && !S.checkUInt32Argument(Attr, MaxExpr, Max, 1))
    return true;
 
  if (Min == 0 && Max != 0) {
    S.Diag(Attr.getLocation(), diag::err_attribute_argument_invalid)
        << &Attr << 0;
    return true;
  }
  if (Max != 0 && Min > Max) {
    S.Diag(Attr.getLocation(), diag::err_attribute_argument_invalid)
        << &Attr << 1;
    return true;
  }
 
  return false;
}
 
AMDGPUWavesPerEUAttr *
SemaAMDGPU::CreateAMDGPUWavesPerEUAttr(const AttributeCommonInfo &CI,
                                       Expr *MinExpr, Expr *MaxExpr) {
  ASTContext &Context = getASTContext();
  AMDGPUWavesPerEUAttr TmpAttr(Context, CI, MinExpr, MaxExpr);
 
  if (checkAMDGPUWavesPerEUArguments(SemaRef, MinExpr, MaxExpr, TmpAttr))
    return nullptr;
 
  return ::new (Context) AMDGPUWavesPerEUAttr(Context, CI, MinExpr, MaxExpr);
}
 
void SemaAMDGPU::addAMDGPUWavesPerEUAttr(Decl *D, const AttributeCommonInfo &CI,
                                         Expr *MinExpr, Expr *MaxExpr) {
  if (auto *Attr = CreateAMDGPUWavesPerEUAttr(CI, MinExpr, MaxExpr))
    D->addAttr(Attr);
}
 
void SemaAMDGPU::handleAMDGPUWavesPerEUAttr(Decl *D, const ParsedAttr &AL) {
  if (!AL.checkAtLeastNumArgs(SemaRef, 1) || !AL.checkAtMostNumArgs(SemaRef, 2))
    return;
 
  Expr *MinExpr = AL.getArgAsExpr(0);
  Expr *MaxExpr = (AL.getNumArgs() > 1) ? AL.getArgAsExpr(1) : nullptr;
 
  addAMDGPUWavesPerEUAttr(D, AL, MinExpr, MaxExpr);
}
 
void SemaAMDGPU::handleAMDGPUNumSGPRAttr(Decl *D, const ParsedAttr &AL) {
  Diag(AL.getLoc(), diag::warn_amdgpu_num_reg_attr_deprecated) << AL;
 
  uint32_t NumSGPR = 0;
  Expr *NumSGPRExpr = AL.getArgAsExpr(0);
  if (!SemaRef.checkUInt32Argument(AL, NumSGPRExpr, NumSGPR))
    return;
 
  D->addAttr(::new (getASTContext())
                 AMDGPUNumSGPRAttr(getASTContext(), AL, NumSGPR));
}
 
void SemaAMDGPU::handleAMDGPUNumVGPRAttr(Decl *D, const ParsedAttr &AL) {
  Diag(AL.getLoc(), diag::warn_amdgpu_num_reg_attr_deprecated) << AL;
 
  uint32_t NumVGPR = 0;
  Expr *NumVGPRExpr = AL.getArgAsExpr(0);
  if (!SemaRef.checkUInt32Argument(AL, NumVGPRExpr, NumVGPR))
    return;
 
  D->addAttr(::new (getASTContext())
                 AMDGPUNumVGPRAttr(getASTContext(), AL, NumVGPR));
}
 
static bool
checkAMDGPUMaxNumWorkGroupsArguments(Sema &S, Expr *XExpr, Expr *YExpr,
                                     Expr *ZExpr,
                                     const AMDGPUMaxNumWorkGroupsAttr &Attr) {
  if (S.DiagnoseUnexpandedParameterPack(XExpr) ||
      (YExpr && S.DiagnoseUnexpandedParameterPack(YExpr)) ||
      (ZExpr && S.DiagnoseUnexpandedParameterPack(ZExpr)))
    return true;
 
  // Accept template arguments for now as they depend on something else.
  // We'll get to check them when they eventually get instantiated.
  if (XExpr->isValueDependent() || (YExpr && YExpr->isValueDependent()) ||
      (ZExpr && ZExpr->isValueDependent()))
    return false;
 
  uint32_t NumWG = 0;
  Expr *Exprs[3] = {XExpr, YExpr, ZExpr};
  for (int i = 0; i < 3; i++) {
    if (Exprs[i]) {
      if (!S.checkUInt32Argument(Attr, Exprs[i], NumWG, i,
                                 /*StrictlyUnsigned=*/true))
        return true;
      if (NumWG == 0) {
        S.Diag(Attr.getLoc(), diag::err_attribute_argument_is_zero)
            << &Attr << Exprs[i]->getSourceRange();
        return true;
      }
    }
  }
 
  return false;
}
 
AMDGPUMaxNumWorkGroupsAttr *SemaAMDGPU::CreateAMDGPUMaxNumWorkGroupsAttr(
    const AttributeCommonInfo &CI, Expr *XExpr, Expr *YExpr, Expr *ZExpr) {
  ASTContext &Context = getASTContext();
  AMDGPUMaxNumWorkGroupsAttr TmpAttr(Context, CI, XExpr, YExpr, ZExpr);
  assert(!SemaRef.isSFINAEContext() &&
         "Can't produce SFINAE diagnostic pointing to temporary attribute");
 
  if (checkAMDGPUMaxNumWorkGroupsArguments(SemaRef, XExpr, YExpr, ZExpr,
                                           TmpAttr))
    return nullptr;
 
  return ::new (Context)
      AMDGPUMaxNumWorkGroupsAttr(Context, CI, XExpr, YExpr, ZExpr);
}
 
void SemaAMDGPU::addAMDGPUMaxNumWorkGroupsAttr(Decl *D,
                                               const AttributeCommonInfo &CI,
                                               Expr *XExpr, Expr *YExpr,
                                               Expr *ZExpr) {
  if (auto *Attr = CreateAMDGPUMaxNumWorkGroupsAttr(CI, XExpr, YExpr, ZExpr))
    D->addAttr(Attr);
}
 
void SemaAMDGPU::handleAMDGPUMaxNumWorkGroupsAttr(Decl *D,
                                                  const ParsedAttr &AL) {
  Expr *YExpr = (AL.getNumArgs() > 1) ? AL.getArgAsExpr(1) : nullptr;
  Expr *ZExpr = (AL.getNumArgs() > 2) ? AL.getArgAsExpr(2) : nullptr;
  addAMDGPUMaxNumWorkGroupsAttr(D, AL, AL.getArgAsExpr(0), YExpr, ZExpr);
}
 
Expr *SemaAMDGPU::ExpandAMDGPUPredicateBuiltIn(Expr *E) {
  CallExpr *CE = cast<CallExpr>(E->IgnoreParens());
  ASTContext &Ctx = getASTContext();
  QualType BoolTy = Ctx.getLogicalOperationType();
  SourceLocation Loc = CE->getExprLoc();
 
  if (!CE->getBuiltinCallee())
    return *ExpandedPredicates
                .insert(SemaRef.BuildBoolLiteral(Loc, false).get())
                .first;
 
  bool P = false;
  unsigned BI = CE->getBuiltinCallee();
  if (Ctx.BuiltinInfo.isAuxBuiltinID(BI))
    BI = Ctx.BuiltinInfo.getAuxBuiltinID(BI);
 
  if (BI == AMDGPU::BI__builtin_amdgcn_processor_is) {
    auto *GFX = dyn_cast<StringLiteral>(CE->getArg(0)->IgnoreParenCasts());
    if (!GFX) {
      Diag(Loc, diag::err_amdgcn_processor_is_arg_not_literal);
      return nullptr;
    }
 
    StringRef N = GFX->getString();
    const TargetInfo &TI = Ctx.getTargetInfo();
    if (llvm::AMDGPU::parseArchAMDGCN(N) == llvm::AMDGPU::GK_NONE) {
      Diag(Loc, diag::err_amdgcn_processor_is_arg_invalid_value) << N;
      SmallVector<StringRef, 64> ValidList;
      llvm::AMDGPU::fillValidArchListAMDGCN(ValidList);
      if (!ValidList.empty())
        Diag(Loc, diag::note_amdgcn_processor_is_valid_options)
            << llvm::join(ValidList, ", ");
      return nullptr;
    }
    if (TI.getTriple().isSPIRV()) {
      CE->setType(BoolTy);
      return *ExpandedPredicates.insert(CE).first;
    }
 
    P = TI.isProcessorName(N);
  } else {
    Expr *Arg = CE->getArg(0);
    if (!Arg || Arg->getType() != Ctx.BuiltinFnTy) {
      Diag(Loc, diag::err_amdgcn_is_invocable_arg_invalid_value) << Arg;
      return nullptr;
    }
 
    if (Ctx.getTargetInfo().getTriple().isSPIRV()) {
      CE->setType(BoolTy);
      return *ExpandedPredicates.insert(CE).first;
    }
 
    auto *FD = cast<FunctionDecl>(Arg->getReferencedDeclOfCallee());
 
    StringRef RF = Ctx.BuiltinInfo.getRequiredFeatures(FD->getBuiltinID());
    llvm::StringMap<bool> CF;
    Ctx.getFunctionFeatureMap(CF, FD);
 
    P = Builtin::evaluateRequiredTargetFeatures(RF, CF);
  }
 
  return *ExpandedPredicates.insert(SemaRef.BuildBoolLiteral(Loc, P).get())
              .first;
}
 
bool SemaAMDGPU::IsPredicate(Expr *E) const {
  return ExpandedPredicates.contains(E);
}
 
void SemaAMDGPU::AddPotentiallyUnguardedBuiltinUser(FunctionDecl *FD) {
  PotentiallyUnguardedBuiltinUsers.insert(FD);
}
 
bool SemaAMDGPU::HasPotentiallyUnguardedBuiltinUsage(FunctionDecl *FD) const {
  return PotentiallyUnguardedBuiltinUsers.contains(FD);
}
 
namespace {
/// This class implements -Wamdgpu-unguarded-builtin-usage.
///
/// This is done with a traversal of the AST of a function that includes a
/// call to a target specific builtin. Whenever we encounter an \c if of the
/// form: \c if(__builtin_amdgcn_is_invocable), we consider the then statement
/// guarded.
class DiagnoseUnguardedBuiltins : public DynamicRecursiveASTVisitor {
  // TODO: this could eventually be extended to consider what happens when there
  //       are multiple target architectures specified via target("arch=gfxXXX")
  //       target("arch=gfxyyy") etc., as well as feature disabling via "-XXX".
  Sema &SemaRef;
 
  SmallVector<StringRef> TargetFeatures;
  SmallVector<std::pair<SourceLocation, StringRef>> CurrentGFXIP;
  SmallVector<unsigned> GuardedBuiltins;
 
  static Expr *FindPredicate(Expr *Cond) {
    if (auto *CE = dyn_cast<CallExpr>(Cond)) {
      if (CE->getBuiltinCallee() == AMDGPU::BI__builtin_amdgcn_is_invocable ||
          CE->getBuiltinCallee() == AMDGPU::BI__builtin_amdgcn_processor_is)
        return Cond;
    } else if (auto *UO = dyn_cast<UnaryOperator>(Cond)) {
      return FindPredicate(UO->getSubExpr());
    } else if (auto *BO = dyn_cast<BinaryOperator>(Cond)) {
      if ((Cond = FindPredicate(BO->getLHS())))
        return Cond;
      return FindPredicate(BO->getRHS());
    }
    return nullptr;
  }
 
  bool EnterPredicateGuardedContext(CallExpr *P);
  void ExitPredicateGuardedContext(bool WasProcessorCheck);
  bool TraverseGuardedStmt(Stmt *S, CallExpr *P);
 
public:
  DiagnoseUnguardedBuiltins(Sema &SemaRef) : SemaRef(SemaRef) {
    if (auto *TAT = SemaRef.getCurFunctionDecl(true)->getAttr<TargetAttr>()) {
      // We use the somewhat misnamed x86 accessors because they provide exactly
      // what we require.
      TAT->getX86AddedFeatures(TargetFeatures);
      if (auto GFXIP = TAT->getX86Architecture())
        CurrentGFXIP.emplace_back(TAT->getLocation(), *GFXIP);
    }
  }
 
  bool TraverseLambdaExpr(LambdaExpr *LE) override {
    if (SemaRef.AMDGPU().HasPotentiallyUnguardedBuiltinUsage(
            LE->getCallOperator()))
      return true; // We have already handled this.
    return DynamicRecursiveASTVisitor::TraverseLambdaExpr(LE);
  }
 
  bool TraverseStmt(Stmt *S) override {
    if (!S)
      return true;
    return DynamicRecursiveASTVisitor::TraverseStmt(S);
  }
 
  void IssueDiagnostics(Stmt *S) { TraverseStmt(S); }
 
  bool TraverseIfStmt(IfStmt *If) override {
    if (auto *CE = dyn_cast_or_null<CallExpr>(FindPredicate(If->getCond())))
      return TraverseGuardedStmt(If, CE);
    return DynamicRecursiveASTVisitor::TraverseIfStmt(If);
  }
 
  bool TraverseCaseStmt(CaseStmt *CS) override {
    return TraverseStmt(CS->getSubStmt());
  }
 
  bool TraverseConditionalOperator(ConditionalOperator *CO) override {
    if (auto *CE = dyn_cast_or_null<CallExpr>(FindPredicate(CO->getCond())))
      return TraverseGuardedStmt(CO, CE);
    return DynamicRecursiveASTVisitor::TraverseConditionalOperator(CO);
  }
 
  bool VisitAsmStmt(AsmStmt *ASM) override;
  bool VisitCallExpr(CallExpr *CE) override;
};
 
bool DiagnoseUnguardedBuiltins::EnterPredicateGuardedContext(CallExpr *P) {
  bool IsProcessorCheck =
      P->getBuiltinCallee() == AMDGPU::BI__builtin_amdgcn_processor_is;
 
  if (IsProcessorCheck) {
    StringRef G = cast<clang::StringLiteral>(P->getArg(0))->getString();
    // TODO: handle generic ISAs.
    if (!CurrentGFXIP.empty() && G != CurrentGFXIP.back().second) {
      SemaRef.Diag(P->getExprLoc(),
                   diag::err_amdgcn_conflicting_is_processor_options)
          << P;
      SemaRef.Diag(CurrentGFXIP.back().first,
                   diag::note_amdgcn_previous_is_processor_guard);
    }
    CurrentGFXIP.emplace_back(P->getExprLoc(), G);
  } else {
    auto *FD = cast<FunctionDecl>(
        cast<DeclRefExpr>(P->getArg(0))->getReferencedDeclOfCallee());
    GuardedBuiltins.push_back(FD->getBuiltinID());
  }
 
  return IsProcessorCheck;
}
 
void DiagnoseUnguardedBuiltins::ExitPredicateGuardedContext(bool WasProcCheck) {
  if (WasProcCheck)
    CurrentGFXIP.pop_back();
  else
    GuardedBuiltins.pop_back();
}
 
inline std::pair<Stmt *, Stmt *> GetTraversalOrder(Stmt *S) {
  std::pair<Stmt *, Stmt *> Ordered;
  Expr *Condition = nullptr;
 
  if (auto *CO = dyn_cast<ConditionalOperator>(S)) {
    Condition = CO->getCond();
    Ordered = {CO->getTrueExpr(), CO->getFalseExpr()};
  } else if (auto *If = dyn_cast<IfStmt>(S)) {
    Condition = If->getCond();
    Ordered = {If->getThen(), If->getElse()};
  }
 
  if (auto *UO = dyn_cast<UnaryOperator>(Condition))
    if (UO->getOpcode() == UnaryOperatorKind::UO_LNot)
      std::swap(Ordered.first, Ordered.second);
 
  return Ordered;
}
 
bool DiagnoseUnguardedBuiltins::TraverseGuardedStmt(Stmt *S, CallExpr *P) {
  assert(S && "Unexpected missing Statement!");
  assert(P && "Unexpected missing Predicate!");
 
  auto [Guarded, Unguarded] = GetTraversalOrder(S);
 
  bool WasProcessorCheck = EnterPredicateGuardedContext(P);
 
  bool Continue = TraverseStmt(Guarded);
 
  ExitPredicateGuardedContext(WasProcessorCheck);
 
  return Continue && TraverseStmt(Unguarded);
}
 
bool DiagnoseUnguardedBuiltins::VisitAsmStmt(AsmStmt *ASM) {
  // TODO: should we check if the ASM is valid for the target? Can we?
  if (!CurrentGFXIP.empty())
    return true;
 
  std::string S = ASM->generateAsmString(SemaRef.getASTContext());
  SemaRef.Diag(ASM->getAsmLoc(), diag::warn_amdgcn_unguarded_asm_stmt) << S;
  SemaRef.Diag(ASM->getAsmLoc(), diag::note_amdgcn_unguarded_asm_silence) << S;
 
  return true;
}
 
bool DiagnoseUnguardedBuiltins::VisitCallExpr(CallExpr *CE) {
  unsigned ID = CE->getBuiltinCallee();
  Builtin::Context &BInfo = SemaRef.getASTContext().BuiltinInfo;
 
  if (!ID)
    return true;
  if (!BInfo.isTSBuiltin(ID))
    return true;
  if (ID == AMDGPU::BI__builtin_amdgcn_processor_is ||
      ID == AMDGPU::BI__builtin_amdgcn_is_invocable)
    return true;
  if (llvm::find(GuardedBuiltins, ID) != GuardedBuiltins.end())
    return true;
 
  StringRef FL(BInfo.getRequiredFeatures(ID));
  llvm::StringMap<bool> FeatureMap;
  if (CurrentGFXIP.empty()) {
    for (auto &&F : TargetFeatures)
      FeatureMap[F] = true;
    for (auto &&GID : GuardedBuiltins)
      for (auto &&F : llvm::split(BInfo.getRequiredFeatures(GID), ','))
        FeatureMap[F] = true;
  } else {
    static const llvm::Triple AMDGCN(llvm::Triple::amdgpu,
                                     llvm::Triple::NoSubArch, llvm::Triple::AMD,
                                     llvm::Triple::AMDHSA);
    llvm::AMDGPU::fillAMDGPUFeatureMap(CurrentGFXIP.back().second, AMDGCN,
                                       FeatureMap);
  }
 
  FunctionDecl *BI = CE->getDirectCallee();
  SourceLocation BICallLoc = CE->getExprLoc();
  if (Builtin::evaluateRequiredTargetFeatures(FL, FeatureMap)) {
    SemaRef.Diag(BICallLoc, diag::warn_amdgcn_unguarded_builtin) << BI;
    SemaRef.Diag(BICallLoc, diag::note_amdgcn_unguarded_builtin_silence) << BI;
  } else {
    StringRef GFXIP = CurrentGFXIP.empty() ? "" : CurrentGFXIP.back().second;
    SemaRef.Diag(BICallLoc, diag::err_amdgcn_incompatible_builtin)
        << BI << FL << !CurrentGFXIP.empty() << GFXIP;
    if (!CurrentGFXIP.empty())
      SemaRef.Diag(CurrentGFXIP.back().first,
                   diag::note_amdgcn_previous_is_processor_guard);
  }
 
  return true;
}
} // Unnamed namespace
 
void SemaAMDGPU::DiagnoseUnguardedBuiltinUsage(FunctionDecl *FD) {
  DiagnoseUnguardedBuiltins(SemaRef).IssueDiagnostics(FD->getBody());
}
 
bool SemaAMDGPU::checkAMDGPUTypeSupport(QualType Ty, SourceLocation Loc) {
  ASTContext &Ctx = getASTContext();
  llvm::Triple TT = Ctx.getTargetInfo().getTriple();
  const Type *BaseTy = Ty->getPointeeOrArrayElementType();
 
  if (Ctx.getTargetInfo().getTriple().isSPIRV()) {
    // The AMDGPU named barrier type requires special handling in the back-end
    // and is not supported for SPIR-V
    if (BaseTy->isAMDGPUNamedBarrierType()) {
      SemaRef.Diag(Loc, diag::err_amdgpu_target_ext_type_unsupported)
          << Ty << TT.str();
      return false;
    }
  }
 
  return true;
}
 
static FieldDecl *getNamedBarrierField(const RecordDecl *R) {
  for (FieldDecl *FD : R->fields()) {
    QualType FDTy = FD->getType();
    if (FDTy->isAMDGPUNamedBarrierTypeOrWrapper())
      return FD;
  }
 
  return nullptr;
}
 
void SemaAMDGPU::checkNamedBarrierWrapper(RecordDecl *R) {
  ASTContext &Context = getASTContext();
  if (R->isInvalidDecl())
    return;
 
  if (!Context.getTargetInfo().hasAMDGPUTypes() &&
      (!Context.getAuxTargetInfo() ||
       !Context.getAuxTargetInfo()->hasAMDGPUTypes()))
    return;
 
  bool IsWrapper = false;
  std::function<void()> DiagWrapperNote;
 
  // First, check if this is a named barrier wrapper by virtue of the class
  // declaring a named barrier field. This covers both C and C++.
  if (FieldDecl *NamedBarrField = getNamedBarrierField(R)) {
    // If this record contains a named barrier field, it must have only one
    // field.
    if (R->getNumFields() > 1) {
      SemaRef.Diag(NamedBarrField->getLocation(),
                   diag::err_amdgpu_invalid_field_not_a_wrapper)
          << NamedBarrField->getType();
      SemaRef.Diag(
          R->getLocation(),
          diag::note_amdgpu_not_a_named_barrier_wrapper_too_many_fields)
          << R->getDeclName();
      return;
    }
 
    IsWrapper = true;
    DiagWrapperNote = [this, R, NamedBarrField]() {
      SemaRef.Diag(NamedBarrField->getLocation(),
                   diag::note_amdgpu_named_barrier_reason_field)
          << R->getDeclName() << NamedBarrField->getDeclName();
    };
  }
 
  // Then, for C++ classes, check if this is a named barrier wrapper by virtue
  // of inheriting one.
  const auto *CxxR = dyn_cast<CXXRecordDecl>(R);
  if (CxxR && !IsWrapper) {
    for (CXXBaseSpecifier BS : CxxR->bases()) {
      const RecordDecl *Base = BS.getType()->getAsRecordDecl();
      if (!Base || !Base->hasAttr<AMDGPUNamedBarrierWrapperAttr>())
        continue;
 
      IsWrapper = true;
      DiagWrapperNote = [this, BS, R]() {
        // Print using the CXXBaseSpecifier type as it includes the template
        // parameters.
        SemaRef.Diag(BS.getBeginLoc(),
                     diag::note_amdgpu_named_barrier_reason_inherited)
            << R->getDeclName() << BS.getType();
      };
    }
  }
 
  if (!IsWrapper)
    return;
 
  // Set the attribute even if the wrapper may be found to be invalid later.
  R->addAttr(
      AMDGPUNamedBarrierWrapperAttr::CreateImplicit(Context, SourceRange()));
 
  // This is a wrapper CXXRecordDecl, it must have a C++11 standard layout.
  if (CxxR && !CxxR->isCXX11StandardLayout()) {
    SemaRef.Diag(R->getLocation(),
                 diag::err_amdgpu_named_barrier_wrapper_non_standard_layout)
        << R->getDeclName();
    assert(DiagWrapperNote &&
           "IsWrapper is set but no context diagnostic provided");
    DiagWrapperNote();
  }
}
} // namespace clang