[SYCL][Driver] Add option to specify the location of ocloc

clang/docs/ClangLinkerWrapper.md

@@ -35,6 +35,7 @@ OPTIONS:
   --linker-path=<path>   The linker executable to invoke
   -L <dir>               Add <dir> to the library search path
   -l <libname>           Search for library <libname>
+  --ocloc-path=<dir>     Path to the ocloc tool used for Intel GPU AOT compilation
   --opt-level=<O0, O1, O2, or O3>
                          Optimization level for LTO
   --override-image=<kind=file>

clang/docs/ClangSYCLLinker.md

@@ -50,6 +50,7 @@ OPTIONS:
   -u <symbol>                   Force undefined symbol during linking
   --module-split-mode=<mode>    Module split mode: 'translation_unit' (default), 'kernel', or 'link_unit'
   --ocloc-options=<value>       Options passed to ocloc for Intel GPU AOT compilation
+  --ocloc-path=<dir>            Path to the ocloc tool used for Intel GPU AOT compilation
   --opencl-aot-options=<value>  Options passed to opencl-aot for Intel CPU AOT compilation
   -o <path>                     Path to file to write output
   --save-temps                  Save intermediate results

clang/include/clang/Options/Options.td

@@ -8002,6 +8002,11 @@ def fsycl_help_EQ
 def fsycl_help : Flag<["-"], "fsycl-help">, Alias<fsycl_help_EQ>,
   Flags<[NoXarchOption]>, AliasArgs<["all"]>,
   HelpText<"Emit help information from all of the offline compilation tools">;
+def ocloc_path_EQ : Joined<["--"], "ocloc-path=">,
+  Visibility<[ClangOption, CLOption]>,
+  Flags<[NoXarchOption, NoArgumentUnused]>, MetaVarName<"<dir>">,
+  HelpText<"Path to the ocloc tool, which is used for ahead of time "
+           "compilation targeting Intel GPUs">;
 def fsycl_libspirv_path_EQ : Joined<["-"], "fsycl-libspirv-path=">,
   HelpText<"Path to libspirv library">;
 def fno_sycl_libspirv : Flag<["-"], "fno-sycl-libspirv">,

clang/lib/Driver/Driver.cpp

@@ -336,6 +336,16 @@ InputArgList Driver::ParseArgStrings(ArrayRef<const char *> ArgStrings,
                            diag::warn_drv_empty_joined_argument,
                            SourceLocation()) > DiagnosticsEngine::Warning;
     }
+
+    // An empty --ocloc-path= is rejected here for consistent usage for areas
+    // that consume it.
+    if (A->getOption().matches(options::OPT_ocloc_path_EQ) &&
+        A->containsValue("")) {
+      Diag(diag::err_drv_invalid_value) << A->getSpelling() << A->getValue();
+      ContainsError |= Diags.getDiagnosticLevel(diag::err_drv_invalid_value,
+                                                SourceLocation()) >
+                       DiagnosticsEngine::Warning;
+    }
   }
 
   for (const Arg *A : Args.filtered(options::OPT_UNKNOWN)) {
@@ -2817,6 +2827,10 @@ void Driver::PrintHelp(bool ShowHidden) const {
 // Print the help from any of the given tools which are used for AOT
 // compilation for SYCL
 void Driver::PrintSYCLToolHelp(const Compilation &C) const {
+  // Do not run any external tools if the command line was already rejected.
+  if (C.containsError())
+    return;
+
   SmallVector<std::tuple<llvm::Triple, StringRef, StringRef, StringRef>, 4>
       HelpArgs;
   // Populate the vector with the tools and help options
@@ -2841,15 +2855,22 @@ void Driver::PrintSYCLToolHelp(const Compilation &C) const {
 
   // Go through the args and emit the help information for each.
   for (auto &HA : HelpArgs) {
-    llvm::outs() << "Emitting help information for " << std::get<1>(HA) << '\n'
-        << "Use triple of '" << std::get<0>(HA).normalize() <<
-        "' to enable ahead of time compilation\n";
+    StringRef ToolName = std::get<1>(HA);
+    llvm::outs() << "Emitting help information for " << ToolName << '\n'
+                 << "Use triple of '" << std::get<0>(HA).normalize()
+                 << "' to enable ahead of time compilation\n";
     // Flush out the buffer before calling the external tool.
     llvm::outs().flush();
-    std::vector<StringRef> ToolArgs = {std::get<1>(HA), std::get<2>(HA),
+    std::vector<StringRef> ToolArgs = {ToolName, std::get<2>(HA),
                                        std::get<3>(HA)};
-    SmallString<128> ExecPath(
-        C.getDefaultToolChain().GetProgramPath(std::get<1>(HA).data()));
+    SmallString<128> ExecPath;
+    // The lookup for ocloc is shared with the AOT compilation step, which
+    // honors any user provided --ocloc-path=.
+    if (ToolName == "ocloc")
+      ExecPath = tools::SYCL::gen::getOclocPath(C, C.getDefaultToolChain(),
+                                                C.getArgs());
+    else
+      ExecPath = C.getDefaultToolChain().GetProgramPath(ToolName.data());
     // do not run the tools with -###.
     if (C.getArgs().hasArg(options::OPT__HASH_HASH_HASH)) {
       llvm::errs() << "\"" << ExecPath << "\" \"" << ToolArgs[1] << "\"";
@@ -2859,12 +2880,13 @@ void Driver::PrintSYCLToolHelp(const Compilation &C) const {
       continue;
     }
     auto ToolBinary = llvm::sys::findProgramByName(ExecPath);
-    if (ToolBinary.getError()) {
+    if (ToolBinary.getError() || !llvm::sys::fs::can_execute(*ToolBinary)) {
       C.getDriver().Diag(diag::err_drv_command_failure) << ExecPath;
       continue;
     }
     // Run the Tool.
-    llvm::sys::ExecuteAndWait(ToolBinary.get(), ToolArgs);
+    if (llvm::sys::ExecuteAndWait(ToolBinary.get(), ToolArgs))
+      C.getDriver().Diag(diag::err_drv_command_failure) << ExecPath;
   }
 }
 

clang/lib/Driver/ToolChains/Clang.cpp

@@ -12146,6 +12146,11 @@ void LinkerWrapper::ConstructJob(Compilation &C, const JobAction &JA,
   // Add any SYCL offloading specific options to the clang-linker-wrapper
   if (C.hasOffloadToolChain<Action::OFK_SYCL>()) {
 
+    // Forward the user provided location for ocloc.
+    if (Arg *A = Args.getLastArg(options::OPT_ocloc_path_EQ))
+      CmdArgs.push_back(
+          Args.MakeArgString(Twine("--ocloc-path=") + A->getValue()));
+
     if (Args.hasArg(options::OPT_fsycl_link_EQ))
       CmdArgs.push_back(Args.MakeArgString("--sycl-device-link"));
 

clang/lib/Driver/ToolChains/SPIRV.cpp

@@ -186,6 +186,10 @@ void SPIRV::Linker::ConstructJob(Compilation &C, const JobAction &JA,
     Linker = ToolChain.GetProgramPath("clang-sycl-linker");
     if (Args.hasArg(options::OPT_v))
       CmdArgs.push_back("-v");
+    // Forward the user provided location to ocloc.
+    if (Arg *A = Args.getLastArg(options::OPT_ocloc_path_EQ))
+      CmdArgs.push_back(
+          Args.MakeArgString(Twine("--ocloc-path=") + A->getValue()));
   } else if (!llvm::sys::fs::can_execute(Linker) &&
              !C.getArgs().hasArg(clang::options::OPT__HASH_HASH_HASH)) {
     C.getDriver().Diag(clang::diag::err_drv_no_spv_tools) << getShortName();

clang/lib/Driver/ToolChains/SYCL.cpp

@@ -1044,14 +1044,27 @@ void SYCL::Linker::ConstructJob(Compilation &C, const JobAction &JA,
                            SpirvInputs);
 }
 
-static const char *makeExeName(Compilation &C, StringRef Name) {
+static const char *makeExeName(const Compilation &C, StringRef Name) {
   llvm::SmallString<8> ExeName(Name);
   const ToolChain *HostTC = C.getSingleOffloadToolChain<Action::OFK_Host>();
   if (HostTC->getTriple().isWindowsMSVCEnvironment())
     ExeName.append(".exe");
   return C.getArgs().MakeArgString(ExeName);
 }
 
+const char *SYCL::gen::getOclocPath(const Compilation &C, const ToolChain &TC,
+                                    const llvm::opt::ArgList &Args) {
+  const char *ExeName = makeExeName(C, "ocloc");
+  // A user provided --ocloc-path= takes precedence over any ocloc that is
+  // found via the program paths or the PATH environment variable.
+  if (Arg *A = Args.getLastArg(options::OPT_ocloc_path_EQ)) {
+    SmallString<128> OclocPath(A->getValue());
+    llvm::sys::path::append(OclocPath, ExeName);
+    return C.getArgs().MakeArgString(OclocPath);
+  }
+  return C.getArgs().MakeArgString(TC.GetProgramPath(ExeName));
+}
+
 // Determine if any of the given arguments contain any PVC based values for
 // the -device option.
 static bool hasPVCDevice(const ArgStringList &CmdArgs, std::string &DevArg) {
@@ -1115,9 +1128,7 @@ void SYCL::gen::BackendCompiler::ConstructJob(Compilation &C,
                                 Device);
   TC.TranslateLinkerTargetArgs(getToolChain().getTriple(), Args, CmdArgs,
                                Device);
-  SmallString<128> ExecPath(
-      getToolChain().GetProgramPath(makeExeName(C, "ocloc")));
-  const char *Exec = C.getArgs().MakeArgString(ExecPath);
+  const char *Exec = SYCL::gen::getOclocPath(C, getToolChain(), Args);
   auto Cmd = std::make_unique<Command>(JA, *this, ResponseFileSupport::None(),
                                        Exec, CmdArgs, ArrayRef<InputInfo>{});
   if (!ForeachInputs.empty()) {

clang/lib/Driver/ToolChains/SYCL.h

@@ -82,6 +82,13 @@ StringRef resolveGenDevice(StringRef DeviceName);
 SmallString<64> getGenDeviceMacro(StringRef DeviceName);
 StringRef getGenGRFFlag(StringRef GRFMode);
 
+// Returns the full path of the ocloc tool to be used for AOT compilation and
+// for emitting the ocloc help information.  A user provided --ocloc-path= is
+// honored above all other lookup locations.  If not found, the tool (ocloc) is
+// returned with no directory.
+const char *getOclocPath(const Compilation &C, const ToolChain &TC,
+                         const llvm::opt::ArgList &Args);
+
 // Prefix for GPU specific targets used for -fsycl-targets
 constexpr char IntelGPU[] = "intel_gpu_";
 constexpr char NvidiaGPU[] = "nvidia_gpu_";

clang/test/Driver/clang-linker-wrapper.cpp

@@ -138,6 +138,21 @@
 // CHK-NO-CMDS-AOT-GEN-LINKERARG: sycl-post-link{{.*}} -o {{[^,]*}}.table {{.*}}.bc
 // CHK-NO-CMDS-AOT-GEN-LINKERARG: ocloc{{.*}} -device pvc -output
 
+// Check that --ocloc-path= provides the location of the ocloc tool.
+// RUN: clang-linker-wrapper --ocloc-path=/my/ocloc/dir --linker-path=/usr/bin/ld -o /dev/null %t1.o --dry-run 2>&1 | FileCheck -check-prefix=CHK-OCLOC-PATH %s
+// CHK-OCLOC-PATH: "/my/ocloc/dir{{[/\\]+}}ocloc" -output_no_suffix
+// Check that --ocloc-path= is not forwarded on to the host linker.
+// CHK-OCLOC-PATH-NOT: ld{{.*}} --ocloc-path=
+
+// Check the diagnostic emitted when ocloc cannot be found in the given
+// directory.
+// RUN: not clang-linker-wrapper --ocloc-path=%t.no-ocloc-here --linker-path=/usr/bin/ld -o /dev/null %t1.o 2>&1 | FileCheck -check-prefix=CHK-OCLOC-PATH-ERR %s
+// CHK-OCLOC-PATH-ERR: unable to find 'ocloc' in '{{.*}}no-ocloc-here'
+
+// Check that an empty --ocloc-path= is rejected.
+// RUN: not clang-linker-wrapper --ocloc-path= --linker-path=/usr/bin/ld -o /dev/null %t1.o --dry-run 2>&1 | FileCheck -check-prefix=CHK-OCLOC-PATH-NOARG %s
+// CHK-OCLOC-PATH-NOARG: no directory given for '--ocloc-path='
+
 /// Check for list of commands for standalone clang-linker-wrapper run for sycl (AOT for Intel CPU)
 // -------
 // Generate .o file as linker wrapper input.

clang/test/Driver/sycl-ocloc-path.cpp

@@ -0,0 +1,115 @@
+///
+/// Tests for --ocloc-path=, which provides the location of the externally
+/// acquired ocloc tool used for Intel GPU AOT compilation.
+///
+
+// REQUIRES: x86-registered-target
+
+/// Check that --ocloc-path= is used for the old offloading model.
+// RUN:   %clang -### -fsycl --no-offload-new-driver -fsycl-targets=spir64_gen \
+// RUN:     --ocloc-path=/my/ocloc/dir %s 2>&1 \
+// RUN:   | FileCheck -check-prefix=CHK-OCLOC-PATH-OLD %s
+// RUN:   %clang -### -fsycl --no-offload-new-driver \
+// RUN:     -fsycl-targets=intel_gpu_pvc --ocloc-path=/my/ocloc/dir %s 2>&1 \
+// RUN:   | FileCheck -check-prefix=CHK-OCLOC-PATH-OLD %s
+// CHK-OCLOC-PATH-OLD: "/my/ocloc/dir{{[/\\]+}}ocloc{{(\.exe)?}}" "-output"
+
+/// Check that the user provided location wins over an ocloc that is visible
+/// via the PATH.  The fake ocloc must be findable, which means it needs the
+/// execute bit set on linux and the executable extension on windows.
+// RUN:   rm -rf %t.dir && mkdir -p %t.dir
+// RUN:   %if system-windows %{ touch %t.dir/ocloc.exe %} \
+// RUN:   %else %{ touch %t.dir/ocloc && chmod +x %t.dir/ocloc %}
+// RUN:   env "PATH=%t.dir%{pathsep}%PATH%" %clang -### -fsycl \
+// RUN:     --no-offload-new-driver -fsycl-targets=spir64_gen \
+// RUN:     --ocloc-path=/my/ocloc/dir %s 2>&1 \
+// RUN:   | FileCheck -check-prefix=CHK-OCLOC-PATH-OLD %s
+
+/// Check that the 'exe' name is used for windows.
+// RUN:   %clang_cl -### -fsycl --no-offload-new-driver \
+// RUN:     -fsycl-targets=spir64_gen --ocloc-path=/my/ocloc/dir -- %s 2>&1 \
+// RUN:   | FileCheck -check-prefix=CHK-OCLOC-PATH-OLD-WIN %s
+// RUN:   %clang -### -target x86_64-pc-windows-msvc -fsycl \
+// RUN:     --no-offload-new-driver -fsycl-targets=spir64_gen \
+// RUN:     --ocloc-path=/my/ocloc/dir %s 2>&1 \
+// RUN:   | FileCheck -check-prefix=CHK-OCLOC-PATH-OLD-WIN %s
+// CHK-OCLOC-PATH-OLD-WIN: "/my/ocloc/dir{{[/\\]+}}ocloc.exe" "-output"
+
+/// Check that --ocloc-path= is forwarded to the clang-linker-wrapper for the
+/// new offloading model.
+// RUN:   %clang -### -fsycl --offload-new-driver -fsycl-targets=spir64_gen \
+// RUN:     --sysroot=%S/Inputs/SYCL --ocloc-path=/my/ocloc/dir %s 2>&1 \
+// RUN:   | FileCheck -check-prefix=CHK-OCLOC-PATH-NEW %s
+// RUN:   %clang -### -fsycl --offload-new-driver \
+// RUN:     -fsycl-targets=intel_gpu_pvc --sysroot=%S/Inputs/SYCL \
+// RUN:     --ocloc-path=/my/ocloc/dir %s 2>&1 \
+// RUN:   | FileCheck -check-prefix=CHK-OCLOC-PATH-NEW %s
+// CHK-OCLOC-PATH-NEW: clang-linker-wrapper{{.*}} "--ocloc-path=/my/ocloc/dir"
+
+/// Check that --ocloc-path= is forwarded to the clang-sycl-linker.
+// RUN:   touch %t.bc
+// RUN:   %clangxx -### --target=spirv64 --sycl-link \
+// RUN:     --ocloc-path=/my/ocloc/dir %t.bc 2>&1 \
+// RUN:   | FileCheck -check-prefix=CHK-OCLOC-PATH-SYCL-LINK %s
+// CHK-OCLOC-PATH-SYCL-LINK: clang-sycl-linker{{.*}} "--ocloc-path=/my/ocloc/dir"
+
+/// Check that --ocloc-path= is used when emitting the ocloc help information.
+// RUN:   %clang -### -fsycl -fsycl-help=gen --ocloc-path=/my/ocloc/dir %s 2>&1 \
+// RUN:   | FileCheck -check-prefix=CHK-OCLOC-PATH-HELP %s
+// CHK-OCLOC-PATH-HELP: Emitting help information for ocloc
+// CHK-OCLOC-PATH-HELP: "/my/ocloc/dir{{[/\\]+}}ocloc{{(\.exe)?}}" "--help"
+
+/// Check that the 'exe' name is used for windows when emitting the ocloc help
+/// information.
+// RUN:   %clang -### -target x86_64-pc-windows-msvc -fsycl -fsycl-help=gen \
+// RUN:     --ocloc-path=/my/ocloc/dir %s 2>&1 \
+// RUN:   | FileCheck -check-prefix=CHK-OCLOC-PATH-HELP-WIN %s
+// CHK-OCLOC-PATH-HELP-WIN: "/my/ocloc/dir{{[/\\]+}}ocloc.exe" "--help"
+
+/// Check the diagnostic emitted when the given directory does not contain a
+/// usable ocloc.  Without -### the tool is actually launched, so the composed
+/// path is expected to be diagnosed instead of being silently ignored.
+// RUN:   rm -rf %t.empty.dir && mkdir -p %t.empty.dir
+// RUN:   not %clang -fsycl -fsycl-help=gen --ocloc-path=%t.empty.dir %s 2>&1 \
+// RUN:   | FileCheck -check-prefix=CHK-OCLOC-PATH-HELP-ERR %s
+// CHK-OCLOC-PATH-HELP-ERR: error: unable to execute command: {{.*}}ocloc
+
+/// Check that an empty --ocloc-path= is rejected instead of falling back to
+/// another ocloc.
+// RUN:   not %clang -### -fsycl --no-offload-new-driver \
+// RUN:     -fsycl-targets=spir64_gen --ocloc-path= %s 2>&1 \
+// RUN:   | FileCheck -check-prefix=CHK-OCLOC-PATH-EMPTY %s
+// RUN:   not %clang -fsycl -fsycl-help=gen --ocloc-path= %s 2>&1 \
+// RUN:   | FileCheck -check-prefix=CHK-OCLOC-PATH-EMPTY %s
+// CHK-OCLOC-PATH-EMPTY: error: invalid value '' in '--ocloc-path='
+// CHK-OCLOC-PATH-EMPTY-NOT: Emitting help information
+
+/// Check that --ocloc-path= does not warn as unused when no AOT compilation
+/// for Intel GPU is being performed.
+// RUN:   %clang -### -fsycl -fsycl-targets=spir64 --ocloc-path=/my/ocloc/dir \
+// RUN:     --sysroot=%S/Inputs/SYCL %s 2>&1 \
+// RUN:   | FileCheck -check-prefix=CHK-OCLOC-PATH-UNUSED %s
+// CHK-OCLOC-PATH-UNUSED-NOT: warning: argument unused during compilation
+
+/// Check that a --ocloc-path= directory whose name contains spaces is
+/// composed and quoted correctly.
+// RUN:   %clang -### -fsycl --no-offload-new-driver -fsycl-targets=spir64_gen \
+// RUN:     --ocloc-path="/my/ocloc dir/with spaces" %s 2>&1 \
+// RUN:   | FileCheck -check-prefix=CHK-OCLOC-PATH-SPACES-OLD %s
+// CHK-OCLOC-PATH-SPACES-OLD: "/my/ocloc dir/with spaces{{[/\\]+}}ocloc{{(\.exe)?}}" "-output"
+// RUN:   %clang -### -fsycl --offload-new-driver -fsycl-targets=spir64_gen \
+// RUN:     --sysroot=%S/Inputs/SYCL --ocloc-path="/my/ocloc dir/with spaces" %s 2>&1 \
+// RUN:   | FileCheck -check-prefix=CHK-OCLOC-PATH-SPACES-NEW %s
+// CHK-OCLOC-PATH-SPACES-NEW: clang-linker-wrapper{{.*}} "--ocloc-path=/my/ocloc dir/with spaces"
+// RUN:   %clangxx -### --target=spirv64 --sycl-link \
+// RUN:     --ocloc-path="/my/ocloc dir/with spaces" %t.bc 2>&1 \
+// RUN:   | FileCheck -check-prefix=CHK-OCLOC-PATH-SPACES-SYCL-LINK %s
+// CHK-OCLOC-PATH-SPACES-SYCL-LINK: clang-sycl-linker{{.*}} "--ocloc-path=/my/ocloc dir/with spaces"
+
+/// Check that a real --ocloc-path= directory containing spaces is actually
+/// located and used to launch the tool.
+// RUN:   rm -rf "%t.dir with spaces" && mkdir -p "%t.dir with spaces"
+// RUN:   not %clang -fsycl -fsycl-help=gen \
+// RUN:     --ocloc-path="%t.dir with spaces" %s 2>&1 \
+// RUN:   | FileCheck -check-prefix=CHK-OCLOC-PATH-SPACES-ERR %s
+// CHK-OCLOC-PATH-SPACES-ERR: error: unable to execute command: {{.*}}dir with spaces{{[/\\]+}}ocloc

clang/test/OffloadTools/clang-sycl-linker/basic.ll

@@ -118,6 +118,22 @@
 ; AOT-INTEL-GPU-NEXT: sycl-bundle: image kind: o, triple: spirv64, arch: bmg_g21
 ; AOT-INTEL-GPU-NOT:  {{.+}}
 ;
+; Test that --ocloc-path= provides the location of the ocloc tool.
+; RUN: clang-sycl-linker --dry-run -v --module-split-mode=link_unit -arch=bmg_g21 %t/input1.bc %t/input2.bc -o %t/aot-gpu.out --ocloc-path=/my/ocloc/dir 2>&1 \
+; RUN:   | FileCheck %s --check-prefix=AOT-OCLOC-PATH
+; AOT-OCLOC-PATH: "/my/ocloc/dir{{[/\\]+}}ocloc" {{.*}}-device bmg_g21
+;
+; Test the diagnostic emitted when ocloc cannot be found in the given
+; directory.
+; RUN: not clang-sycl-linker -v --module-split-mode=link_unit -arch=bmg_g21 %t/input1.bc %t/input2.bc -o %t/aot-gpu.out --ocloc-path=%t/no-ocloc-here 2>&1 \
+; RUN:   | FileCheck %s --check-prefix=AOT-OCLOC-PATH-ERR
+; AOT-OCLOC-PATH-ERR: unable to find 'ocloc' in '{{.*}}no-ocloc-here'
+;
+; Test that an empty --ocloc-path= is rejected.
+; RUN: not clang-sycl-linker --dry-run -v -arch=bmg_g21 %t/input1.bc --ocloc-path= -o %t/aot-gpu.out 2>&1 \
+; RUN:   | FileCheck %s --check-prefix=AOT-OCLOC-PATH-NOARG
+; AOT-OCLOC-PATH-NOARG: no directory given for '--ocloc-path='
+
 ; Test that all --ocloc-options are passed to ocloc, even if they contain spaces or quotes.
 ; RUN: clang-sycl-linker --dry-run -v --module-split-mode=link_unit -arch=bmg_g21 %t/input1.bc -o %t/aot-gpu.out 2>&1 \
 ; RUN:     --ocloc-options="-a -b" --ocloc-options=-c --ocloc-options="d" --ocloc-options='"-e -f"' \

clang/tools/clang-linker-wrapper/ClangLinkerWrapper.cpp

@@ -334,10 +334,32 @@ Expected<std::string> findProgram(StringRef Name, ArrayRef<StringRef> Paths) {
     return Name.str();
   if (!Path)
     return createStringError(Path.getError(),
-                             "Unable to find '" + Name + "' in path");
+                             "unable to find '" + Name + "' in path");
   return *Path;
 }
 
+/// Locate the 'ocloc' tool used for Intel GPU AOT compilation.
+Expected<std::string> findOcloc(const ArgList &Args) {
+  if (Arg *A = Args.getLastArg(OPT_ocloc_path_EQ)) {
+    StringRef Dir = A->getValue();
+    if (Dir.empty())
+      return createStringError("no directory given for '" + A->getSpelling() +
+                               "'");
+    if (DryRun) {
+      SmallString<128> OclocPath(Dir);
+      sys::path::append(OclocPath, "ocloc");
+      return std::string(OclocPath);
+    }
+    // Only look in the given directory.  The tool name is resolved by
+    // findProgramByName, which takes care of any platform specific executable
+    // extension.
+    if (ErrorOr<std::string> Path = sys::findProgramByName("ocloc", {Dir}))
+      return *Path;
+    return createStringError("unable to find 'ocloc' in '" + Dir + "'");
+  }
+  return findProgram("ocloc", {getExecutableDir("ocloc")});
+}
+
 bool linkerSupportsLTO(const ArgList &Args) {
   llvm::Triple Triple(Args.getLastArgValue(OPT_triple_EQ));
   return Triple.isNVPTX() || Triple.isAMDGPU() ||
@@ -1082,8 +1104,7 @@ runAOTCompileIntelGPU(StringRef InputFile, const ArgList &Args,
   const llvm::Triple Triple(Args.getLastArgValue(OPT_triple_EQ));
   StringRef Arch(Args.getLastArgValue(OPT_arch_EQ));
   SmallVector<StringRef, 8> CmdArgs;
-  Expected<std::string> OclocPath =
-      findProgram("ocloc", {getExecutableDir("ocloc")});
+  Expected<std::string> OclocPath = findOcloc(Args);
   if (!OclocPath)
     return OclocPath.takeError();