Michal Babej

HeCBench: merkle-hip+chipStar gives different results than merkle-sycl

3

needs investigating

HIP/tests/catch/unit/deviceLib/{Single,Double}PrecisionMathFunctions tests are not very useful

3

``` grep GENERATE HIP/tests/catch/unit/deviceLib/SinglePrecisionMathFunctions/* acosf.cc:GENERATE_KERNEL_FLOAT(acosf, acosf(1.0f)); acoshf.cc:GENERATE_KERNEL_FLOAT(acoshf, acoshf(1.0f)); asinf.cc:GENERATE_KERNEL_FLOAT(asinf, asinf(1.0f)); asinhf.cc:GENERATE_KERNEL_FLOAT(asinhf, asinhf(1.0f)); atan2f.cc:GENERATE_KERNEL_FLOAT(atan2f, atan2f(1.0f, 1.0f)); atanf.cc:GENERATE_KERNEL_FLOAT(atanf, atanf(1.0f)); atanhf.cc:GENERATE_KERNEL_FLOAT(atanhf, atanhf(1.0f)); .... ``` Almost all of the tests call a function with...

myocyte benchmark from HeCBench significantly slower with chipStar than with SYCL [LevelZero backend]

10

Without immediate queues, chipStar is ~100x slower, with immediate queues it is ~10x slower. My initial examination seems to point to many (possibly unnecessary) barrier commands, but anyway this needs...

At least with my laptop (intel i7-1185G7, 23.09 driver) a bunch of `Unit_hipTexture*` tests fail. Needs investigation, possibly a driver issue.

the new device setup should better support devices which don't have all features of GPU (some of the zeGet...Properties return error)

fix generic_address_space, command_buffer_event_sync, test_compiler and images/test_1D_buffer

2

fixes several issues: * `generic_address_space` test: add check for program scope variables. The test uses both generic AS and program-scope variables, however it only checked the generic-AS presence in `clGetDeviceInfo`....