[SYCL][ESIMD] Test for a fix of error when a scalar offset is provided as a parameter to the API #1534
Conversation
|
/verify with intel/llvm#8075 |
|
/verify with intel/llvm#8075 |
|
/verify with intel/llvm#8075 |
|
Test failures: |
| @@ -98,48 +98,59 @@ int main(void) { | |||
| bool Pass = true; | |||
|
|
|||
| Pass &= test<int8_t, 1>(Q); | |||
| #ifndef USE_SCALAR_OFFSET | |||
There was a problem hiding this comment.
Just a note: This situation reveals that this test is written in a bad way: gather and scatter are used in one kernel and gather is input for gather. This makes it impossible to test gather with scalar offset and N>1 and it creates opportunity for double-error (1 in gather, 1 in scatter) giving a pass.
I had some changes in my local ws splitting this test to gather and scatter.
| if (q.get_backend() != sycl::backend::ext_intel_esimd_emulator) | ||
| passed &= test_int_types<8, ImplStore>(q, cfg); | ||
| #ifndef USE_DWORD_ATOMICS | ||
| if (q.get_backend() != sycl::backend::ext_intel_esimd_emulator) | ||
| passed &= test<float, 8, ImplStore>(q, cfg); | ||
| #endif // USE_DWORD_ATOMICS | ||
| #endif |
There was a problem hiding this comment.
nit (here and in few other places):
| #endif | |
| #endif // !USE_SCALAR_OFFSET |
|
/verify with intel/llvm#8075 |
|
Test failures: |
[SYCL][Matrix] fix ATS-M double bug
…as a parameter to the API (intel#1534)
…as a parameter to the API (intel/llvm-test-suite#1534)
Complementary compiler PR: intel/llvm#8075