From b6f47bad7c63040a5a216c94e02de769c3d3af9f Mon Sep 17 00:00:00 2001 From: Nusrat Islam Date: Thu, 2 Nov 2023 21:41:39 -0500 Subject: [PATCH] msccl: remove cases from numReduction switch statement [ROCm/rccl commit: f545b94d4b200a50e7d386f205512b640ff6a3fe] --- .../collectives/device/msccl_kernel_impl.h | 20 ++++--------------- 1 file changed, 4 insertions(+), 16 deletions(-) diff --git a/projects/rccl/src/collectives/device/msccl_kernel_impl.h b/projects/rccl/src/collectives/device/msccl_kernel_impl.h index 9699b1d029..90b9051dc9 100644 --- a/projects/rccl/src/collectives/device/msccl_kernel_impl.h +++ b/projects/rccl/src/collectives/device/msccl_kernel_impl.h @@ -352,22 +352,16 @@ __device__ __forceinline__ void mscclRunInterpreter( T o = load(dstIndex); ssize_t srcBaseOffset = gridOffset + (ssize_t)c * sizePerMscclChunk + tid; switch (numReductions) { - case 1: - #pragma unroll - MSCCL_REDUCE_UNROLL_LOOP_A(1); - break; - case 3: - #pragma unroll - MSCCL_REDUCE_UNROLL_LOOP_A(3); - break; case 7: #pragma unroll MSCCL_REDUCE_UNROLL_LOOP_A(7); break; +#if defined(__gfx90a__) case 15: #pragma unroll MSCCL_REDUCE_UNROLL_LOOP_A(15); break; +#endif default: MSCCL_REDUCE_UNROLL_LOOP_A(numReductions); break; @@ -388,22 +382,16 @@ __device__ __forceinline__ void mscclRunInterpreter( T* dst = dstPointer + dstOffset; ssize_t srcBaseOffset = gridOffset + (ssize_t)c * sizePerMscclChunk; switch (numReductions) { - case 1: - #pragma unroll - MSCCL_REDUCE_UNROLL_LOOP_B(1); - break; - case 3: - #pragma unroll - MSCCL_REDUCE_UNROLL_LOOP_B(3); - break; case 7: #pragma unroll MSCCL_REDUCE_UNROLL_LOOP_B(7); break; +#if defined(__gfx90a__) case 15: #pragma unroll MSCCL_REDUCE_UNROLL_LOOP_B(15); break; +#endif default: MSCCL_REDUCE_UNROLL_LOOP_B(numReductions); break;