Disable loop unrolling in clang

szkarpinski · szkarpinski · commit 58ea73d6199a · 2021-12-14T15:28:58.000+01:00
Signed-off-by: Szymon Karpiński &lt;hugo@staszic.waw.pl&gt;
diff --git a/dali/kernels/slice/slice_gpu.cuh b/dali/kernels/slice/slice_gpu.cuh
@@ -139,7 +139,9 @@ __device__ void SliceFunc(OutputType *__restrict__ out, const InputType *__restr
   }
 
   for (; offset < block_end; offset += blockDim.x * coalesced_values<OutputType>) {
+    #ifndef __clang__
     #pragma unroll
+    #endif
     for (uint64_t i = 0; i < coalesced_values<OutputType>; i++) {
       uint64_t idx = offset + i;
       if (idx >= block_end) break;

Original file line number	Diff line number	Diff line change
`@@ -139,7 +139,9 @@ __device__ void SliceFunc(OutputType __restrict__ out, const InputType __restr`
`139`	`139`	`}`
`140`	`140`
`141`	`141`	`for (; offset < block_end; offset += blockDim.x * coalesced_values<OutputType>) {`
	`142`	`+ #ifndef __clang__`
`142`	`143`	`#pragma unroll`
	`144`	`+ #endif`
`143`	`145`	`for (uint64_t i = 0; i < coalesced_values<OutputType>; i++) {`
`144`	`146`	`uint64_t idx = offset + i;`
`145`	`147`	`if (idx >= block_end) break;`