[SYCL] Add -fsycl-fp32-prec-sqrt flag

clang/include/clang/Basic/CodeGenOptions.def

@@ -177,6 +177,7 @@ CODEGENOPT(NoImplicitFloat   , 1, 0) ///< Set when -mno-implicit-float is enable
 CODEGENOPT(NullPointerIsValid , 1, 0) ///< Assume Null pointer deference is defined.
 CODEGENOPT(OpenCLCorrectlyRoundedDivSqrt, 1, 0) ///< -cl-fp32-correctly-rounded-divide-sqrt
 CODEGENOPT(HIPCorrectlyRoundedDivSqrt, 1, 1) ///< -fno-hip-fp32-correctly-rounded-divide-sqrt
+CODEGENOPT(SYCLFp32PrecSqrt, 1, 0) ///< -fsycl-fp32-prec-sqrt
 CODEGENOPT(UniqueInternalLinkageNames, 1, 0) ///< Internal Linkage symbols get unique names.
 CODEGENOPT(SplitMachineFunctions, 1, 0) ///< Split machine functions using profile information.
 

clang/include/clang/Driver/Options.td

@@ -4732,6 +4732,9 @@ def fno_sycl_device_lib_EQ : CommaJoined<["-"], "fno-sycl-device-lib=">, Group<s
   Values<"libc, libm-fp32, libm-fp64, all">, HelpText<"Control exclusion of "
   "device libraries from device binary linkage. Valid arguments "
   "are libc, libm-fp32, libm-fp64, all">;
+def fsycl_fp32_prec_sqrt : Flag<["-"], "fsycl-fp32-prec-sqrt">, Group<sycl_Group>, Flags<[CC1Option]>,
+  HelpText<"SYCL only. Specify that single precision floating-point sqrt is correctly rounded.">,
+  MarshallingInfoFlag<CodeGenOpts<"SYCLFp32PrecSqrt">>;
 
 //===----------------------------------------------------------------------===//
 // FLangOption + CoreOption + NoXarchOption

clang/include/clang/Driver/ToolChain.h

@@ -705,7 +705,8 @@ class ToolChain {
 
   /// Get paths of HIP device libraries.
   virtual llvm::SmallVector<BitCodeLibraryInfo, 12>
-  getHIPDeviceLibs(const llvm::opt::ArgList &Args) const;
+  getHIPDeviceLibs(const llvm::opt::ArgList &Args,
+                   const Action::OffloadKind DeviceOffloadingKind) const;
 
   /// Return sanitizers which are available in this toolchain.
   virtual SanitizerMask getSupportedSanitizers() const;

clang/lib/Driver/ToolChain.cpp

@@ -1097,7 +1097,8 @@ void ToolChain::AddHIPIncludeArgs(const ArgList &DriverArgs,
                                   ArgStringList &CC1Args) const {}
 
 llvm::SmallVector<ToolChain::BitCodeLibraryInfo, 12>
-ToolChain::getHIPDeviceLibs(const ArgList &DriverArgs) const {
+ToolChain::getHIPDeviceLibs(const ArgList &DriverArgs,
+                            const Action::OffloadKind OffloadKind) const {
   return {};
 }
 

clang/lib/Driver/ToolChains/AMDGPU.cpp

@@ -895,9 +895,9 @@ bool AMDGPUToolChain::shouldSkipArgument(const llvm::opt::Arg *A) const {
   return false;
 }
 
-llvm::SmallVector<std::string, 12>
-ROCMToolChain::getCommonDeviceLibNames(const llvm::opt::ArgList &DriverArgs,
-                                       const std::string &GPUArch) const {
+llvm::SmallVector<std::string, 12> ROCMToolChain::getCommonDeviceLibNames(
+    const llvm::opt::ArgList &DriverArgs, const std::string &GPUArch,
+    const Action::OffloadKind DeviceOffloadingKind) const {
   auto Kind = llvm::AMDGPU::parseArchAMDGCN(GPUArch);
   const StringRef CanonArch = llvm::AMDGPU::getArchNameAMDGCN(Kind);
 
@@ -923,6 +923,12 @@ ROCMToolChain::getCommonDeviceLibNames(const llvm::opt::ArgList &DriverArgs,
   bool CorrectSqrt = DriverArgs.hasFlag(
       options::OPT_fhip_fp32_correctly_rounded_divide_sqrt,
       options::OPT_fno_hip_fp32_correctly_rounded_divide_sqrt);
+
+  if (DeviceOffloadingKind == Action::OFK_SYCL) {
+    // When using SYCL, sqrt is only correctly rounded if the flag is specified
+    CorrectSqrt = DriverArgs.hasArg(options::OPT_fsycl_fp32_prec_sqrt);
+  }
+
   bool Wave64 = isWave64(DriverArgs, Kind);
 
   return RocmInstallation.getCommonBitcodeLibs(

clang/lib/Driver/ToolChains/AMDGPU.h

@@ -142,7 +142,8 @@ class LLVM_LIBRARY_VISIBILITY ROCMToolChain : public AMDGPUToolChain {
   // Returns a list of device library names shared by different languages
   llvm::SmallVector<std::string, 12>
   getCommonDeviceLibNames(const llvm::opt::ArgList &DriverArgs,
-                          const std::string &GPUArch) const;
+                          const std::string &GPUArch,
+                          const Action::OffloadKind DeviceOffloadingKind) const;
 };
 
 } // end namespace toolchains

clang/lib/Driver/ToolChains/AMDGPUOpenMP.cpp

@@ -123,7 +123,8 @@ const char *AMDGCN::OpenMPLinker::constructLLVMLinkCommand(
       //  - write an opt pass that sets that on every function it sees and pipe
       //    the device-libs bitcode through that on the way to this llvm-link
       SmallVector<std::string, 12> BCLibs =
-          AMDGPUOpenMPTC.getCommonDeviceLibNames(Args, SubArchName.str());
+          AMDGPUOpenMPTC.getCommonDeviceLibNames(Args, SubArchName.str(),
+                                                 Action::OFK_OpenMP);
       llvm::for_each(BCLibs, [&](StringRef BCFile) {
         CmdArgs.push_back(Args.MakeArgString(BCFile));
       });

clang/lib/Driver/ToolChains/Cuda.cpp

@@ -707,6 +707,10 @@ void CudaToolChain::addClangTargetOptions(
   if (DeviceOffloadingKind == Action::OFK_SYCL) {
     toolchains::SYCLToolChain::AddSYCLIncludeArgs(getDriver(), DriverArgs,
                                                   CC1Args);
+
+    if (DriverArgs.hasArg(options::OPT_fsycl_fp32_prec_sqrt)) {
+      CC1Args.push_back("-fcuda-prec-sqrt");
+    }
   }
 
   auto NoLibSpirv = DriverArgs.hasArg(options::OPT_fno_sycl_libspirv,

clang/lib/Driver/ToolChains/HIPAMD.cpp

@@ -256,11 +256,12 @@ void HIPAMDToolChain::addClangTargetOptions(
     CC1Args.push_back(DriverArgs.MakeArgString(LibSpirvFile));
   }
 
-  llvm::for_each(getHIPDeviceLibs(DriverArgs), [&](auto BCFile) {
-    CC1Args.push_back(BCFile.ShouldInternalize ? "-mlink-builtin-bitcode"
-                                               : "-mlink-bitcode-file");
-    CC1Args.push_back(DriverArgs.MakeArgString(BCFile.Path));
-  });
+  llvm::for_each(
+      getHIPDeviceLibs(DriverArgs, DeviceOffloadingKind), [&](auto BCFile) {
+        CC1Args.push_back(BCFile.ShouldInternalize ? "-mlink-builtin-bitcode"
+                                                   : "-mlink-bitcode-file");
+        CC1Args.push_back(DriverArgs.MakeArgString(BCFile.Path));
+      });
 }
 
 llvm::opt::DerivedArgList *
@@ -355,7 +356,9 @@ VersionTuple HIPAMDToolChain::computeMSVCVersion(const Driver *D,
 }
 
 llvm::SmallVector<ToolChain::BitCodeLibraryInfo, 12>
-HIPAMDToolChain::getHIPDeviceLibs(const llvm::opt::ArgList &DriverArgs) const {
+HIPAMDToolChain::getHIPDeviceLibs(
+    const llvm::opt::ArgList &DriverArgs,
+    const Action::OffloadKind DeviceOffloadingKind) const {
   llvm::SmallVector<BitCodeLibraryInfo, 12> BCLibs;
   if (DriverArgs.hasArg(options::OPT_nogpulib))
     return {};
@@ -412,7 +415,8 @@ HIPAMDToolChain::getHIPDeviceLibs(const llvm::opt::ArgList &DriverArgs) const {
     BCLibs.push_back(RocmInstallation.getHIPPath());
 
     // Add common device libraries like ocml etc.
-    for (auto N : getCommonDeviceLibNames(DriverArgs, GpuArch.str()))
+    for (auto N : getCommonDeviceLibNames(DriverArgs, GpuArch.str(),
+                                          DeviceOffloadingKind))
       BCLibs.push_back(StringRef(N));
 
     // Add instrument lib.

clang/lib/Driver/ToolChains/HIPAMD.h

@@ -86,8 +86,9 @@ class LLVM_LIBRARY_VISIBILITY HIPAMDToolChain final : public ROCMToolChain {
                            llvm::opt::ArgStringList &CC1Args) const override;
   void AddHIPIncludeArgs(const llvm::opt::ArgList &DriverArgs,
                          llvm::opt::ArgStringList &CC1Args) const override;
-  llvm::SmallVector<BitCodeLibraryInfo, 12>
-  getHIPDeviceLibs(const llvm::opt::ArgList &Args) const override;
+  llvm::SmallVector<BitCodeLibraryInfo, 12> getHIPDeviceLibs(
+      const llvm::opt::ArgList &Args,
+      const Action::OffloadKind DeviceOffloadingKind) const override;
 
   SanitizerMask getSupportedSanitizers() const override;
 

clang/lib/Driver/ToolChains/HIPSPV.cpp

@@ -154,7 +154,7 @@ void HIPSPVToolChain::addClangTargetOptions(
     CC1Args.append(
         {"-fvisibility", "hidden", "-fapply-global-visibility-to-externs"});
 
-  llvm::for_each(getHIPDeviceLibs(DriverArgs),
+  llvm::for_each(getHIPDeviceLibs(DriverArgs, DeviceOffloadingKind),
                  [&](const BitCodeLibraryInfo &BCFile) {
                    CC1Args.append({"-mlink-builtin-bitcode",
                                    DriverArgs.MakeArgString(BCFile.Path)});
@@ -206,7 +206,9 @@ void HIPSPVToolChain::AddHIPIncludeArgs(const ArgList &DriverArgs,
 }
 
 llvm::SmallVector<ToolChain::BitCodeLibraryInfo, 12>
-HIPSPVToolChain::getHIPDeviceLibs(const llvm::opt::ArgList &DriverArgs) const {
+HIPSPVToolChain::getHIPDeviceLibs(
+    const llvm::opt::ArgList &DriverArgs,
+    const Action::OffloadKind DeviceOffloadingKind) const {
   llvm::SmallVector<ToolChain::BitCodeLibraryInfo, 12> BCLibs;
   if (DriverArgs.hasArg(options::OPT_nogpulib))
     return {};

clang/lib/Driver/ToolChains/HIPSPV.h

@@ -69,7 +69,8 @@ class LLVM_LIBRARY_VISIBILITY HIPSPVToolChain final : public ToolChain {
   void AddHIPIncludeArgs(const llvm::opt::ArgList &DriverArgs,
                          llvm::opt::ArgStringList &CC1Args) const override;
   llvm::SmallVector<BitCodeLibraryInfo, 12>
-  getHIPDeviceLibs(const llvm::opt::ArgList &Args) const override;
+  getHIPDeviceLibs(const llvm::opt::ArgList &Args,
+                   const Action::OffloadKind DeviceOffloadingKind) const override;
 
   SanitizerMask getSupportedSanitizers() const override;
 

clang/test/Driver/sycl-amdgcn-sqrt.cpp

@@ -0,0 +1,24 @@
+// REQUIRES: clang-driver
+// REQUIRES: amdgpu-registered-target
+// REQUIRES: !system-windows
+
+// RUN: %clang -### \
+// RUN:   -fsycl -fsycl-targets=amdgcn-amd-amdhsa     \
+// RUN:   -Xsycl-target-backend --offload-arch=gfx900 \
+// RUN:   -fsycl-fp32-prec-sqrt \
+// RUN:   --rocm-path=%S/Inputs/rocm \
+// RUN:   %s \
+// RUN: 2>&1 | FileCheck  --check-prefix=CHECK-CORRECT %s
+
+// CHECK-CORRECT: "-mlink-builtin-bitcode" "{{.*}}/amdgcn/bitcode/oclc_correctly_rounded_sqrt_on.bc"
+
+// RUN: %clang -### \
+// RUN:   -fsycl -fsycl-targets=amdgcn-amd-amdhsa     \
+// RUN:   -Xsycl-target-backend --offload-arch=gfx900 \
+// RUN:   --rocm-path=%S/Inputs/rocm \
+// RUN:   %s \
+// RUN: 2>&1 | FileCheck  --check-prefix=CHECK-APPROX %s
+
+// CHECK-APPROX: "-mlink-builtin-bitcode" "{{.*}}/amdgcn/bitcode/oclc_correctly_rounded_sqrt_off.bc"
+
+void func(){};

clang/test/Driver/sycl-no-prec-sqrt.cpp

@@ -0,0 +1,17 @@
+// REQUIRES: clang-driver
+
+// RUN: %clang -### -fsycl \
+// RUN:   -fsycl-fp32-prec-sqrt %s 2>&1 | FileCheck %s
+
+// RUN: %clang -### -fsycl -fsycl-targets=spir64_gen \
+// RUN:   -fsycl-fp32-prec-sqrt %s 2>&1 | FileCheck %s
+//
+// RUN: %clang -### -fsycl -fsycl-targets=spir64_x86_64 \
+// RUN:   -fsycl-fp32-prec-sqrt %s 2>&1 | FileCheck %s
+//
+// RUN: %clang -### -fsycl -fsycl-targets=spir64_fpga \
+// RUN:   -fsycl-fp32-prec-sqrt %s 2>&1 | FileCheck %s
+
+// CHECK: warning: argument unused during compilation: '-fsycl-fp32-prec-sqrt'
+
+void func(){};

clang/test/Driver/sycl-nvptx-sqrt.cpp

@@ -0,0 +1,20 @@
+// REQUIRES: clang-driver
+// REQUIRES: nvptx-registered-target
+// REQUIRES: !system-windows
+
+// RUN: %clang -### \
+// RUN:   -fsycl -fsycl-targets=nvptx64-nvidia-cuda \
+// RUN:   -fsycl-fp32-prec-sqrt \
+// RUN:   %s \
+// RUN: 2>&1 | FileCheck --check-prefix=CHECK-CORRECT %s
+
+// CHECK-CORRECT: "-fcuda-prec-sqrt"
+
+// RUN: %clang -### \
+// RUN:   -fsycl -fsycl-targets=nvptx64-nvidia-cuda \
+// RUN:   %s \
+// RUN: 2>&1 | FileCheck --check-prefix=CHECK-APPROX %s
+
+// CHECK-APPROX-NOT: "-fcuda-prec-sqrt"
+
+void func(){};

sycl/doc/GetStartedGuide.md

@@ -832,7 +832,7 @@ which contains all the symbols required.
   project and may cause compilation issues on some platforms
 * `sycl::sqrt` is not correctly rounded by default as the SYCL specification
   allows lower precision, when porting from CUDA it may be helpful to use
-  `-Xclang -fcuda-prec-sqrt` to use the correctly rounded square root, this is
+  `-fsycl-fp32-prec-sqrt` to use the correctly rounded square root, this is
   significantly slower but matches the default precision used by `nvcc`, and
   this `clang++` flag is equivalent to the `nvcc` `-prec-sqrt` flag, except that
   it defaults to `false`.

sycl/doc/UsersManual.md

@@ -257,6 +257,14 @@ and not recommended to use in production environment.
     options (e.g. -c, -E, -S) may interfere with the expected output set during
     the host compilation.  Doing so is considered undefined behavior.
 
+**`-fsycl-fp32-prec-sqrt`**
+
+    Enable use of correctly rounded `sycl::sqrt` function as defined by IEE754.
+    Without this flag, the default precision requirement for `sycl::sqrt` is 3
+    ULP.
+
+    NOTE: This flag is currently only supported with the CUDA and HIP targets.
+
 # Example: SYCL device code compilation
 
 To invoke SYCL device compiler set `-fsycl-device-only` flag.