From f132fb2cd0d4368336b990a9a61990fbc9d7cdcd Mon Sep 17 00:00:00 2001 From: Oak Zeng Date: Tue, 5 Jan 2021 13:13:24 -0600 Subject: [PATCH] Make GPU mapping of memory as uncached if HSA_DISABLE_CACHE is set Before gfx90a, coherent memory is uncached. So it was reasonable when environment variable HSA_DISABLE_CACHE is set, memory is mapped as coherent. On gfx90a, coherent memory can be cached, so mapping memory as coherent can't guarantee memory is uncached. When HSA_DISABLE_CACHE is set, we have to map memory as uncached. Change-Id: Ia5ed4cf0ad6aef5644dc8c9e6632b52d606f06f4 Signed-off-by: Oak Zeng --- src/fmm.c | 17 +++++++++-------- 1 file changed, 9 insertions(+), 8 deletions(-) diff --git a/src/fmm.c b/src/fmm.c index a3f4822297..463a3e2fd1 100644 --- a/src/fmm.c +++ b/src/fmm.c @@ -1335,11 +1335,11 @@ void *fmm_allocate_device(uint32_t gpu_id, void *address, uint64_t MemorySizeInB aperture = &gpu_mem[gpu_mem_id].gpuvm_aperture; } - if (!flags.ui32.CoarseGrain || svm.disable_cache) { + if (!flags.ui32.CoarseGrain) ioc_flags |= KFD_IOC_ALLOC_MEM_FLAGS_COHERENT; - if (flags.ui32.Uncached) - ioc_flags |= KFD_IOC_ALLOC_MEM_FLAGS_UNCACHED; - } + + if (flags.ui32.Uncached || svm.disable_cache) + ioc_flags |= KFD_IOC_ALLOC_MEM_FLAGS_UNCACHED; mem = __fmm_allocate_device(gpu_id, address, size, aperture, &mmap_offset, ioc_flags, &vm_obj); @@ -1544,11 +1544,12 @@ static void *fmm_allocate_host_gpu(uint32_t node_id, void *address, else aperture = svm.dgpu_alt_aperture; /* always coherent */ - if (!flags.ui32.CoarseGrain || svm.disable_cache) { + if (!flags.ui32.CoarseGrain) ioc_flags |= KFD_IOC_ALLOC_MEM_FLAGS_COHERENT; - if (flags.ui32.Uncached) - ioc_flags |= KFD_IOC_ALLOC_MEM_FLAGS_UNCACHED; - } + + if (flags.ui32.Uncached || svm.disable_cache) + ioc_flags |= KFD_IOC_ALLOC_MEM_FLAGS_UNCACHED; + ioc_flags |= fmm_translate_hsa_to_ioc_flags(flags); if (flags.ui32.AQLQueueMemory)