[AMD] Guard ROCm 7.0 build from using hipMemcpyBatchAsync (#35128)

This commit is contained in:
Alan Kao
2026-08-17 02:22:58 -07:00
committed by GitHub
parent 92bce3d7bb
commit 056808723e
@@ -995,9 +995,9 @@ inline void transfer_kv_page_first_direct_impl(
};
#if defined(USE_ROCM) || !defined(CUDA_VERSION) || CUDA_VERSION < 12080
#if defined(USE_ROCM)
// Opt-in HIP batch copy path (mirrors cudaMemcpyBatchAsync); disabled by
// default, falls back to per-page copy below.
#if defined(USE_ROCM) && defined(HIP_VERSION) && HIP_VERSION >= 70200000
// Opt-in HIP batch copy path (mirrors cudaMemcpyBatchAsync); requires
// ROCm >= 7.2. Disabled by default, falls back to per-page copy below.
constexpr bool kEnableHipBatch = false;
if (kEnableHipBatch) {
std::vector<void*> b_srcs, b_dsts;