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

Unify fmm_get_aperture_xxx functions

Browse Source 
Unify fmm_get_aperture_base and fmm_get_aperture_limit into one
function. Make the return value to HSAKMT_STATUS.

Change-Id: I0b3f563ffb268947ab891f4935f61788d0af0e01
This commit is contained in:
Harish Kasiviswanathan
2015-10-22 21:32:41 -04:00
parent 5131ab4e64
commit cb53548c89
3 changed files with 37 additions and 54 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/fmm.c
+20 -40
View File
@@ -1035,8 +1035,10 @@ HSAKMT_STATUS fmm_init_process_apertures(void)
return ret;
}
HSAuint64 fmm_get_aperture_limit(aperture_type_e aperture_type, HSAuint32 gpu_id)
HSAKMT_STATUS fmm_get_aperture_base_and_limit(aperture_type_e aperture_type, HSAuint32 gpu_id,
HSAuint64 *aperture_base, HSAuint64 *aperture_limit)
{
HSAKMT_STATUS err = HSAKMT_STATUS_SUCCESS;
int32_t slot = gpu_mem_find_by_gpu_id(gpu_id);
if (slot < 0)
@@ -1044,56 +1046,34 @@ HSAuint64 fmm_get_aperture_limit(aperture_type_e aperture_type, HSAuint32 gpu_id
switch (aperture_type) {
case FMM_GPUVM:
return aperture_is_valid(gpu_mem[slot].gpuvm_aperture.base,
gpu_mem[slot].gpuvm_aperture.limit) ?
PORT_VPTR_TO_UINT64(gpu_mem[slot].gpuvm_aperture.limit) : 0;
if (aperture_is_valid(gpu_mem[slot].gpuvm_aperture.base,
gpu_mem[slot].gpuvm_aperture.limit)) {
*aperture_base = PORT_VPTR_TO_UINT64(gpu_mem[slot].gpuvm_aperture.base);
*aperture_limit = PORT_VPTR_TO_UINT64(gpu_mem[slot].gpuvm_aperture.limit);
}
break;
case FMM_SCRATCH:
return aperture_is_valid(gpu_mem[slot].scratch_aperture.base,
gpu_mem[slot].scratch_aperture.limit) ?
PORT_VPTR_TO_UINT64(gpu_mem[slot].scratch_aperture.limit) : 0;
if (aperture_is_valid(gpu_mem[slot].scratch_aperture.base,
gpu_mem[slot].scratch_aperture.limit)) {
*aperture_base = PORT_VPTR_TO_UINT64(gpu_mem[slot].scratch_aperture.base);
*aperture_limit = PORT_VPTR_TO_UINT64(gpu_mem[slot].scratch_aperture.limit);
}
break;
case FMM_LDS:
return aperture_is_valid(gpu_mem[slot].lds_aperture.base,
gpu_mem[slot].lds_aperture.limit) ?
PORT_VPTR_TO_UINT64(gpu_mem[slot].lds_aperture.limit) : 0;
if (aperture_is_valid(gpu_mem[slot].lds_aperture.base,
gpu_mem[slot].lds_aperture.limit)) {
*aperture_base = PORT_VPTR_TO_UINT64(gpu_mem[slot].lds_aperture.base);
*aperture_limit = PORT_VPTR_TO_UINT64(gpu_mem[slot].lds_aperture.limit);
}
break;
default:
return 0;
err = HSAKMT_STATUS_ERROR;
}
}
HSAuint64 fmm_get_aperture_base(aperture_type_e aperture_type, HSAuint32 gpu_id)
{
int32_t slot = gpu_mem_find_by_gpu_id(gpu_id);
if (slot < 0)
return HSAKMT_STATUS_INVALID_PARAMETER;
switch (aperture_type) {
case FMM_GPUVM:
return aperture_is_valid(gpu_mem[slot].gpuvm_aperture.base,
gpu_mem[slot].gpuvm_aperture.limit) ?
PORT_VPTR_TO_UINT64(gpu_mem[slot].gpuvm_aperture.base) : 0;
break;
case FMM_SCRATCH:
return aperture_is_valid(gpu_mem[slot].scratch_aperture.base,
gpu_mem[slot].scratch_aperture.limit) ?
PORT_VPTR_TO_UINT64(gpu_mem[slot].scratch_aperture.base) : 0;
break;
case FMM_LDS:
return aperture_is_valid(gpu_mem[slot].lds_aperture.base,
gpu_mem[slot].lds_aperture.limit) ?
PORT_VPTR_TO_UINT64(gpu_mem[slot].lds_aperture.base) : 0;
break;
default:
return 0;
}
return err;
}
static int _fmm_map_to_gpu_gtt(uint32_t gpu_id, manageble_aperture_t *aperture,
src/fmm.h
+2 -2
View File
@@ -65,6 +65,6 @@ bool fmm_get_handle(void *address, uint64_t *handle);
/* Topology interface*/
HSAKMT_STATUS fmm_node_added(HSAuint32 gpu_id);
HSAKMT_STATUS fmm_node_removed(HSAuint32 gpu_id);
HSAuint64 fmm_get_aperture_base(aperture_type_e aperture_type, HSAuint32 gpu_id);
HSAuint64 fmm_get_aperture_limit(aperture_type_e aperture_type, HSAuint32 gpu_id);
HSAKMT_STATUS fmm_get_aperture_base_and_limit(aperture_type_e aperture_type, HSAuint32 gpu_id,
HSAuint64 *aperture_base, HSAuint64 *aperture_limit);
#endif /* FMM_H_ */
src/topology.c
+15 -12
View File
@@ -894,8 +894,9 @@ hsaKmtGetNodeMemoryProperties(
HsaMemoryProperties* MemoryProperties //OUT
)
{
HSAKMT_STATUS err;
HSAKMT_STATUS err = HSAKMT_STATUS_SUCCESS;
uint32_t i, gpu_id;
HSAuint64 aperture_limit;
if (!MemoryProperties)
return HSAKMT_STATUS_INVALID_PARAMETER;
@@ -918,7 +919,7 @@ hsaKmtGetNodeMemoryProperties(
err = validate_nodeid(NodeId, &gpu_id);
if (err != HSAKMT_STATUS_SUCCESS)
return err;
goto out;
for (i = 0; i < MIN(node[NodeId].node.NumMemoryBanks, NumBanks); i++) {
assert(node[NodeId].mem);
@@ -926,31 +927,33 @@ hsaKmtGetNodeMemoryProperties(
}
/*Add LDS*/
if (i < NumBanks){
if (i < NumBanks &&
fmm_get_aperture_base_and_limit(FMM_LDS, gpu_id,
&MemoryProperties[i].VirtualBaseAddress, &aperture_limit) == HSAKMT_STATUS_SUCCESS) {
MemoryProperties[i].HeapType = HSA_HEAPTYPE_GPU_LDS;
MemoryProperties[i].SizeInBytes = node[NodeId].node.LDSSizeInKB * 1024;
MemoryProperties[i].VirtualBaseAddress = fmm_get_aperture_base(FMM_LDS, gpu_id);
i++;
}
/*Add Local memory - HSA_HEAPTYPE_FRAME_BUFFER_PRIVATE*/
if ((i < NumBanks) && (node[NodeId].node.LocalMemSize > 0)) {
/* Add Local memory - HSA_HEAPTYPE_FRAME_BUFFER_PRIVATE*/
if (i < NumBanks &&
node[NodeId].node.LocalMemSize > 0 &&
fmm_get_aperture_base_and_limit(FMM_GPUVM, gpu_id,
&MemoryProperties[i].VirtualBaseAddress, &aperture_limit) == HSAKMT_STATUS_SUCCESS) {
MemoryProperties[i].HeapType = HSA_HEAPTYPE_FRAME_BUFFER_PRIVATE;
MemoryProperties[i].SizeInBytes = node[NodeId].node.LocalMemSize;
MemoryProperties[i].VirtualBaseAddress = fmm_get_aperture_base(FMM_GPUVM, gpu_id);
i++;
}
/*Add SCRATCH*/
if (i < NumBanks){
if (i < NumBanks &&
fmm_get_aperture_base_and_limit(FMM_SCRATCH, gpu_id,
&MemoryProperties[i].VirtualBaseAddress, &aperture_limit) == HSAKMT_STATUS_SUCCESS) {
MemoryProperties[i].HeapType = HSA_HEAPTYPE_GPU_SCRATCH;
MemoryProperties[i].VirtualBaseAddress = fmm_get_aperture_base(FMM_SCRATCH, gpu_id);
MemoryProperties[i].SizeInBytes = fmm_get_aperture_limit(FMM_SCRATCH, gpu_id) - MemoryProperties[i].VirtualBaseAddress;
MemoryProperties[i].SizeInBytes = (aperture_limit - MemoryProperties[i].VirtualBaseAddress) + 1;
i++;
}
err = HSAKMT_STATUS_SUCCESS;
out:
pthread_mutex_unlock(&hsakmt_mutex);
return err;