[hip] Claim builtin type `__float128` supported if the host target supports it.

[yaxunl]

Currently if instructions of float128 get to amdgpu backend, are we going to crash?

[hliao]

In D78513#1993115, @yaxunl wrote:

Currently if instructions of float128 get to amdgpu backend, are we going to crash?

As Float128Format is re-defined as double, there won't be any issue in the backend. But, it won't function as the developer expects. That's quite similar to long double in clang.

[yaxunl]

LGTM. Thanks.

[jdoerfert]

In D78513#1993210, @hliao wrote:

In D78513#1993115, @yaxunl wrote:

Currently if instructions of float128 get to amdgpu backend, are we going to crash?

As Float128Format is re-defined as double, there won't be any issue in the backend. But, it won't function as the developer expects. That's quite similar to long double in clang.

I'd argue that it is quite similarly broken. Silently miscompiling the program is arguably the worst. Are you sure the OpenMP/Sycl way to deal with this is not better? Long story short,
allow the presence of a type that the aux-triple supports but do not allow the use of such a type. The implementation is not perfect yet, e.g., it doesn't allow the above typedef and sometimes a long double is manifested in IR, but for the most part it's functional: clang/test/OpenMP/nvptx_unsupported_type_messages.cpp and clang/test/SemaSYCL/float128.cpp after D74387. TBH, I believe an error is still better than treating long double and fp128 on the device side different and then hope that it somehow works.

FWIW, I landed here after I wrote D95665 to address PR48923. I think the ideas there and here are contrary and we should pick one.

clang/lib/Basic/Targets/AMDGPU.cpp

void AMDGPUTargetInfo::setAuxTarget(const TargetInfo *Aux) {
   // On x86_64 long double is 80-bit extended precision format, which is
   // not supported by AMDGPU. 128-bit floating point format is also not
   // supported by AMDGPU. Therefore keep its own format for these two types.
   auto SaveLongDoubleFormat = LongDoubleFormat;
   auto SaveFloat128Format = Float128Format;
   copyAuxTarget(Aux);
   LongDoubleFormat = SaveLongDoubleFormat;
   Float128Format = SaveFloat128Format;
+  // For certain builtin types support on the host target, claim they are
+  // support to pass the compilation of the host code during the device-side
+  // compilation.
+  // FIXME: As the side effect, we also accept `__float128` uses in the device
+  // code. To rejct these builtin types supported in the host target but not in
+  // the device target, one approach would support `device_builtin` attribute
+  // so that we could tell the device builtin types from the host ones. The
+  // also solves the different representations of the same builtin type, such
+  // as `size_t` in the MSVC environment.
+  if (Aux->hasFloat128Type()) {
+    HasFloat128 = true;
+    Float128Format = DoubleFormat;
+  }
 }

clang/test/SemaCUDA/amdgpu-f128.cu

+// RUN: %clang_cc1 -triple amdgcn-amd-amdhsa -aux-triple x86_64-unknown-linux-gnu -fcuda-is-device -fsyntax-only -verify %s
+
+// expected-no-diagnostics
+typedef __float128 f128_t;