-
Notifications
You must be signed in to change notification settings - Fork 25
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Add a check_doubles rule and documentation to run the double-precision requiring tests separately if target has doubles #135
Comments
All the known failures have been documented on the wiki as part of the 0.9
release. If you have additional tests that fail for you then we should
investigate.
…On Thu, Sep 8, 2022, 14:22 Pekka Jääskeläinen ***@***.***> wrote:
Are these known? Happens with my Iris Xe mobile GPU.
Shall we remove them as known-failing tests or have another check rule for
known-passing OpenCL tests?
The following tests FAILED:
108 - Unit_hipHostGetFlags_Basic - int (Failed)
109 - Unit_hipHostGetFlags_Basic - float (Failed)
110 - Unit_hipHostGetFlags_Basic - double (Failed)
111 - Unit_hipMallocManaged_MultiChunkSingleDevice (Failed)
112 - Unit_hipMallocManaged_MultiChunkMultiDevice (Failed)
115 - Unit_hipMallocManaged_TwoPointers - int (Failed)
116 - Unit_hipMallocManaged_TwoPointers - float (Failed)
117 - Unit_hipMallocManaged_TwoPointers - double (Failed)
118 - Unit_hipMallocManaged_DeviceContextChange - unsigned char (Failed)
119 - Unit_hipMallocManaged_DeviceContextChange - int (Failed)
120 - Unit_hipMallocManaged_DeviceContextChange - float (Failed)
121 - Unit_hipMallocManaged_DeviceContextChange - double (Failed)
187 - Unit_hipMemcpy_KernelLaunch - int (Failed)
188 - Unit_hipMemcpy_KernelLaunch - float (Failed)
189 - Unit_hipMemcpy_KernelLaunch - double (Failed)
193 - Unit_hipMemcpy_MultiThreadWithSerialization (Subprocess aborted)
197 - Unit_hipMemcpyAsync_KernelLaunch - int (Failed)
198 - Unit_hipMemcpyAsync_KernelLaunch - float (Failed)
199 - Unit_hipMemcpyAsync_KernelLaunch - double (Failed)
204 - Unit_hipMemcpyAsync_hipMultiMemcpyMultiThread - int (Subprocess aborted)
205 - Unit_hipMemcpyAsync_hipMultiMemcpyMultiThread - float (SEGFAULT)
206 - Unit_hipMemcpyAsync_hipMultiMemcpyMultiThread - double (Subprocess aborted)
215 - Unit_ldg (Failed)
450 - Unit_deviceFunctions_CompileTest_modf_double (Failed)
454 - Unit_deviceFunctions_CompileTest_norm_double (Failed)
463 - Unit_deviceFunctions_CompileTest_rhypot_double (Failed)
465 - Unit_deviceFunctions_CompileTest_rnorm_double (Failed)
466 - Unit_deviceFunctions_CompileTest_rnorm3d_double (Failed)
467 - Unit_deviceFunctions_CompileTest_rnorm4d_double (Failed)
474 - Unit_deviceFunctions_CompileTest_sincos_double (Failed)
475 - Unit_deviceFunctions_CompileTest_sincospi_double (Failed)
528 - Unit_hipGetDeviceProperties_ArchPropertiesTst (Failed)
540 - Unit_hipStreamPerThread_MultiThread (Subprocess aborted)
541 - Unit_hipStreamPerThread_DeviceReset_1 (Subprocess aborted)
626 - cuda-reduction (Failed)
Errors while running CTest
—
Reply to this email directly, view it on GitHub
<#135>, or unsubscribe
<https://github.com/notifications/unsubscribe-auth/ACCJBQNYVIOMJ2TKNEF54KLV5HD7JANCNFSM6AAAAAAQHULNIE>
.
You are receiving this because you are subscribed to this thread.Message
ID: ***@***.***>
|
I haven't seen this error before. Are you using an outdated driver? Trying to understand what could cause a failure to emulate.
I thought we enable double support by using
If we fail to compile we throw an error and print the compilation log:
besides changing the error from |
This is an extension, meaning it's not mandatory to support it. I do not know if we can make it more graceful, as long as the user sees it's because of missing capabilities and not just crash, I think there's not much else we can do. |
Ok so sounds like we don't have an issue in that regard then since we already print the exact error.
Sounds like good idea and something we can consider for the next release. I haven't come across a device that doesn't support double precision (so no way of testing) so it doesn't seem like a high-priority issue at this time. |
All of these cases except the one reported in Issue #134 seem to be caused by the double precision issue. I'll change this ticket to fix it for 0.9 via the double precision test tag and another test suite for tests that are not supposed to work with non-double capable HW, and let's leave Issue #137 open for later consideration. |
Not sure if we really need this one as there's a workaround for the only known target where the issue appears. |
All testing was done with llvm 14
…On Thu, Sep 8, 2022, 14:34 Pekka Jääskeläinen ***@***.***> wrote:
Have you tested with LLVM 15 or 14?
—
Reply to this email directly, view it on GitHub
<#135 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/ACCJBQLJ6JJHC7RRMI4WQDTV5HFMBANCNFSM6AAAAAAQHULNIE>
.
You are receiving this because you commented.Message ID:
***@***.***>
|
Are these known? Happens with my Iris Xe mobile GPU with LLVM 15.
OpenCL:
Level0:
Some errors from logs (LZ):
The "Double type is not supported on this platform." is a HW/driver issue. OpenCL/SPIR-V doesn't require double support or its SW emulation. https://registry.khronos.org/OpenCL/sdk/1.0/docs/man/xhtml/cl_khr_fp64.html Can we fail more gracefully at runtime? Should we add a separate test suite for the double tests since they are not supposed to work on OpenCL/LZ devices which do not provide double support?
The text was updated successfully, but these errors were encountered: