[NFC] Remove LLVM type name altering #3915
Conversation
This is a leftover from early DPC++ prototype
bader
requested review from
AaronBallman,
DenisBakhvalov,
elizabethandrews,
kbobrovs,
premanandrao and
a team
as code owners
|
/summary:run |
| @@ -45,14 +45,14 @@ SYCL_ESIMD_FUNCTION SYCL_EXTERNAL simd<float, 16> foo() { | |||
| simd<ushort, VL> pred; | |||
| v_addr += offsets; | |||
|
|
|||
| __esimd_flat_atomic0<EsimdAtomicOpType::ATOMIC_INC, uint32_t, VL>( | |||
| __esimd_flat_atomic0<EsimdAtomicOpType::inc, uint32_t, VL>( | |||
There was a problem hiding this comment.
Since you are editing this anyway, this needs 1 more step (EsimdAtomicOpType itself is deprecated)
| __esimd_flat_atomic0<EsimdAtomicOpType::inc, uint32_t, VL>( | |
| __esimd_flat_atomic0<atomic_op::inc, uint32_t, VL>( |
There was a problem hiding this comment.
EsimdAtomicOpTypeitself is deprecated
The compiler gives a message only about using ATOMIC_*.
I suggest we update the tests together with adding corresponding diagnostics.
| v_addr.data(), pred.data()); | ||
| // CHECK: %{{[0-9a-zA-Z_.]+}} = call <32 x i32> @llvm.genx.svm.atomic.inc.v32i32.v32i1.v32i64(<32 x i1> %{{[0-9a-zA-Z_.]+}}, <32 x i64> %{{[0-9a-zA-Z_.]+}}, <32 x i32> undef) | ||
|
|
||
| __esimd_flat_atomic1<EsimdAtomicOpType::ATOMIC_ADD, uint32_t, VL>( | ||
| __esimd_flat_atomic1<EsimdAtomicOpType::add, uint32_t, VL>( |
There was a problem hiding this comment.
| __esimd_flat_atomic1<EsimdAtomicOpType::add, uint32_t, VL>( | |
| __esimd_flat_atomic1<atomic_op::add, uint32_t, VL>( |
| v_addr.data(), v1, pred.data()); | ||
| // CHECK: %{{[0-9a-zA-Z_.]+}} = call <32 x i32> @llvm.genx.svm.atomic.add.v32i32.v32i1.v32i64(<32 x i1> %{{[0-9a-zA-Z_.]+}}, <32 x i64> %{{[0-9a-zA-Z_.]+}}, <32 x i32> %{{[0-9a-zA-Z_.]+}}, <32 x i32> undef) | ||
| __esimd_flat_atomic2<EsimdAtomicOpType::ATOMIC_CMPXCHG, uint32_t, VL>( | ||
| __esimd_flat_atomic2<EsimdAtomicOpType::cmpxchg, uint32_t, VL>( |
There was a problem hiding this comment.
| __esimd_flat_atomic2<EsimdAtomicOpType::cmpxchg, uint32_t, VL>( | |
| __esimd_flat_atomic2<atomic_op::cmpxchg, uint32_t, VL>( |
|
I looked at the failures in pre-commit and it seems to be a bug in low-level runtime. I filed a bug report and we probably have to wait for a fix. Meanwhile, we can review the DPC++ changes. |
|
This change unveils a bug in SYCL pipes implementation for Intel FPGA emulator. |
bader
requested review from
kychendev,
smanna12,
sndmitriev and
v-klochkov
as code owners
|
/summary:run |
This is a leftover from early DPC++ prototype