[HIP] Pass -fno-hip-fp32-correctly-rounded-divide-sqrt to clang -cc1

[yaxunl]

Summary

-fno-hip-fp32-correctly-rounded-divide-sqrt affects clang codegen
and should be passed to clang -cc1 by clang driver.

Fixes: https://github.com/llvm/llvm-project/issues/63653

clang/lib/Driver/ToolChains/Clang.cpp

StringRef InlineThresh =
         Args.getLastArgValue(options::OPT_fgpu_inline_threshold_EQ);
     if (!InlineThresh.empty()) {
       std::string ArgStr =
           std::string("-inline-threshold=") + InlineThresh.str();
       CmdArgs.append({"-mllvm", Args.MakeArgStringRef(ArgStr)});
     }
   }
 
+  if (IsHIPDevice)
+    Args.addOptOutFlag(CmdArgs,
+                       options::OPT_fhip_fp32_correctly_rounded_divide_sqrt,
+                       options::OPT_fno_hip_fp32_correctly_rounded_divide_sqrt);
+
   // OpenMP offloading device jobs take the argument -fopenmp-host-ir-file-path
   // to specify the result of the compile phase on the host, so the meaningful
   // device declarations can be identified. Also, -fopenmp-is-device is passed
   // along to tell the frontend that it is generating code for a device, so that
   // only the relevant declarations are emitted.
   if (IsOpenMPDevice) {
     CmdArgs.push_back("-fopenmp-is-device");
     if (OpenMPDeviceInput) {

clang/test/Driver/hip-options.hip

 
 // RUN: %clang -### -nogpuinc -nogpulib -mamdgpu-ieee -mno-amdgpu-ieee -ffast-math \
 // RUN:   --cuda-gpu-arch=gfx906  %s 2>&1 | FileCheck -check-prefixes=IEEE-OFF %s
 // IEEE-OFF: clang{{.*}} "-triple" "amdgcn-amd-amdhsa" {{.*}} "-mno-amdgpu-ieee"
 
 // RUN: %clang -### -nogpuinc -nogpulib -mamdgpu-ieee -mno-amdgpu-ieee -ffast-math \
 // RUN:   --cuda-gpu-arch=gfx906  %s 2>&1 | FileCheck -check-prefixes=IEEE-OFF-NEG %s
 // IEEE-OFF-NEG-NOT: clang{{.*}} "-triple" "amdgcn-amd-amdhsa" {{.*}} "-mamdgpu-ieee"
+
+// Check -fno-hip-fp32-correctly-rounded-divide-sqrt is passed to -cc1 but
+// (default) -fhip-fp32-correctly-rounded-divide-sqrt is not.
+
+// RUN: %clang -### -nogpuinc -nogpulib -fno-hip-fp32-correctly-rounded-divide-sqrt \
+// RUN:   --cuda-gpu-arch=gfx906  %s 2>&1 | FileCheck -check-prefixes=NOCRDS %s
+// NOCRDS: clang{{.*}} "-triple" "amdgcn-amd-amdhsa" {{.*}} "-fno-hip-fp32-correctly-rounded-divide-sqrt"
+
+// RUN: %clang -### -nogpuinc -nogpulib \
+// RUN:   --cuda-gpu-arch=gfx906  %s 2>&1 | FileCheck -check-prefixes=CRDS %s
+// CRDS-NOT: "-f{{(no-)?}}hip-fp32-correctly-rounded-divide-sqrt"
+
+// RUN: %clang -### -nogpuinc -nogpulib -fhip-fp32-correctly-rounded-divide-sqrt \
+// RUN:   --cuda-gpu-arch=gfx906  %s 2>&1 | FileCheck -check-prefixes=CRDS %s
+// CRDS-NOT: "-f{{(no-)?}}hip-fp32-correctly-rounded-divide-sqrt"