[SYCL] defer kernel diagnostics #181
Conversation
| return S.DeviceKnownEmittedFns.count(FD) > 0; | ||
| } | ||
|
|
||
| Sema::DeviceDiagBuilder Sema::SYCLDiagIfDeviceCode(SourceLocation Loc, |
There was a problem hiding this comment.
Hi Premanand, I need something like this function to avoid emitting an error for a function that has varargs and which is not part of DEVICE code.
In my case I have "include <stdio.h>" in the beginning of the source file and no uses of any functions like printf(f, ...) in DEVICE code, but clang still reports an error on Windows for such code.
Q1: Would your fix help with this error: "variadic function cannot use spir_function calling convention" emitted from SemaType.cpp:7014 ?
If NO, can this patch be extended to handle that case?
Q2: This MergeRequest was created almost 2 weeks ago. Is there some ETA for this fix approval/merge?
bader
added
the
upstream
premanandrao
force-pushed
the
remote_defer_diag
branch
2 times, most recently
from
241f1c6 to
30d5d1a
Compare
Signed-off-by: Premanand M Rao <premanand.m.rao@intel.com>
premanandrao
force-pushed
the
remote_defer_diag
branch
from
30d5d1a to
b2aec3b
Compare
| @@ -1494,7 +1494,8 @@ static QualType ConvertDeclSpecToType(TypeProcessingState &state) { | |||
| Result = Context.DoubleTy; | |||
| break; | |||
| case DeclSpec::TST_float128: | |||
| if (!S.Context.getTargetInfo().hasFloat128Type() && | |||
| if (!S.Context.getTargetInfo().hasFloat128Type() && | |||
| !S.getLangOpts().SYCLIsDevice && | |||
There was a problem hiding this comment.
Because of this change we silent usage of __float128 in device code.
Interesting observation: __float128 type variable in device code turns into fp128 type variable in LLVM IR and after conversion to SPIRV and back, I see half (!) type variable.
@bader , @AlexeySachkov FYI.
But, of course, It helps do not emit error: __float128 is not supported on this target from host code when we compile for device.
There was a problem hiding this comment.
Because of this change we silent usage of
__float128in device code.
It sounds like a bug. We should not allow __float128 in device code, so I guess the right fix is use differed diagnostic instead of skipping it.
* [SYCL] Moved tests introduced in intel#3255 * [SYCL] Fix issues found in CI - fix target tripple used by default to spir64-unknown-unknown-sycldevice; - remove obsolet LIT parameters (target_triple, host_triple); - enable tests for CPU and ACC devices.
Signed-off-by: Premanand M Rao premanand.m.rao@intel.com