AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Make GPU mapping of memory as uncached if HSA_DISABLE_CACHE is set

Browse Source 
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 <Oak.Zeng@amd.com>


[ROCm/ROCR-Runtime commit: f132fb2cd0]
This commit is contained in:
Oak Zeng
2021-01-05 13:13:24 -06:00
committed by Kent Russell
parent 2e82a94d97
commit c189479766
1 changed files with 9 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/rocr-runtime/src/fmm.c
+9 -8
View File
@@ -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)