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

Allocate and map doorbells in SVM for discrete GPUs

Browse Source 
Allocate doorbells for dGPUs in the SVM aperture and map them for
GPU access. This is necessary to allow GPU-initiated submissions to
user mode queues.

Depends on new doorbell BO allocation flag in KFD.

Change-Id: I0737bef4a4764bb4a66c43846707ead2108f6601


[ROCm/ROCR-Runtime commit: 2e0a6eb371]
This commit is contained in:
Felix Kuehling
2016-09-16 15:56:57 -04:00
parent 6b33ada07b
commit fee7a91fb9
4 changed files with 143 additions and 21 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/rocr-runtime/src/fmm.h
+1
View File
@@ -52,6 +52,7 @@ void fmm_destroy_process_apertures(void);
*/
void* fmm_allocate_scratch(uint32_t gpu_id, uint64_t MemorySizeInBytes);
void* fmm_allocate_device(uint32_t gpu_id, uint64_t MemorySizeInBytes, HsaMemFlags flags);
void* fmm_allocate_doorbell(uint32_t gpu_id, uint64_t MemorySizeInBytes, uint64_t doorbell_offset);
void* fmm_allocate_host(uint32_t node_id, uint64_t MemorySizeInBytes,
HsaMemFlags flags);
void* fmm_open_graphic_handle(uint32_t gpu_id,