AI/rocm-systems
2
0
Fork 0
Koodi Ongelmat Pull-pyynnöt Actions Paketit Projektit Julkaisut Wiki Toiminta

Remove the use of IS_DGPU()

Selaa lähdekoodia 
The information can be obtained directly from node id. Also improve the
whole logic for future compatibility.

Change-Id: I130733be4e7930d5953d5e81409905e60c2ec35e
Signed-off-by: Yong Zhao <Yong.Zhao@amd.com>
This commit is contained in:
Yong Zhao
2018-08-02 13:25:44 -04:00
committed by Yong Zhao
vanhempi c21927f425
commit 4bb90d048c
4 muutettua tiedostoa jossa 44 lisäystä ja 37 poistoa
Download Patch File Download Diff File Expand all files Collapse all files
src/libhsakmt.h
+3 -3
Näytä tiedosto
@@ -100,13 +100,13 @@ enum asic_family_type {
CHIP_RAVEN,
CHIP_VEGA20
};
#define IS_DGPU(chip) ((chip) != CHIP_KAVERI && (chip) != CHIP_CARRIZO && \
(chip) != CHIP_RAVEN)
#define IS_SOC15(chip) ((chip) >= CHIP_VEGA10)
HSAKMT_STATUS validate_nodeid(uint32_t nodeid, uint32_t *gpu_id);
HSAKMT_STATUS gpuid_to_nodeid(uint32_t gpu_id, uint32_t* node_id);
uint16_t get_device_id_by_node(HSAuint32 node_id);
bool prefer_ats(HSAuint32 node_id);
uint16_t get_device_id_by_node_id(HSAuint32 node_id);
uint16_t get_device_id_by_gpu_id(HSAuint32 gpu_id);
int get_drm_render_fd_by_gpu_id(HSAuint32 gpu_id);
HSAKMT_STATUS validate_nodeid_array(uint32_t **gpu_id_array,
src/pmc_table.c
+1 -1
Näytä tiedosto
@@ -1712,7 +1712,7 @@ HSAKMT_STATUS get_block_properties(uint32_t node_id,
enum perf_block_id block_id,
struct perf_counter_block *block)
{
uint16_t dev_id = get_device_id_by_node(node_id);
uint16_t dev_id = get_device_id_by_node_id(node_id);
enum asic_family_type asic;
if (block_id > PERFCOUNTER_BLOCKID__MAX ||
src/queues.c
+32 -32
Näytä tiedosto
@@ -50,68 +50,68 @@ struct device_info {
uint32_t doorbell_size;
};
struct device_info kaveri_device_info = {
const struct device_info kaveri_device_info = {
.asic_family = CHIP_KAVERI,
.eop_buffer_size = 0,
.doorbell_size = DOORBELL_SIZE_GFX7,
};
struct device_info hawaii_device_info = {
const struct device_info hawaii_device_info = {
.asic_family = CHIP_HAWAII,
.eop_buffer_size = 0,
.doorbell_size = DOORBELL_SIZE_GFX7,
};
struct device_info carrizo_device_info = {
const struct device_info carrizo_device_info = {
.asic_family = CHIP_CARRIZO,
.eop_buffer_size = 4096,
.doorbell_size = DOORBELL_SIZE_GFX8,
};
struct device_info tonga_device_info = {
const struct device_info tonga_device_info = {
.asic_family = CHIP_TONGA,
.eop_buffer_size = TONGA_PAGE_SIZE,
.doorbell_size = DOORBELL_SIZE_GFX8,
};
struct device_info fiji_device_info = {
const struct device_info fiji_device_info = {
.asic_family = CHIP_FIJI,
.eop_buffer_size = TONGA_PAGE_SIZE,
.doorbell_size = DOORBELL_SIZE_GFX8,
};
struct device_info polaris10_device_info = {
const struct device_info polaris10_device_info = {
.asic_family = CHIP_POLARIS10,
.eop_buffer_size = TONGA_PAGE_SIZE,
.doorbell_size = DOORBELL_SIZE_GFX8,
};
struct device_info polaris11_device_info = {
const struct device_info polaris11_device_info = {
.asic_family = CHIP_POLARIS11,
.eop_buffer_size = TONGA_PAGE_SIZE,
.doorbell_size = DOORBELL_SIZE_GFX8,
};
struct device_info vega10_device_info = {
const struct device_info vega10_device_info = {
.asic_family = CHIP_VEGA10,
.eop_buffer_size = 4096,
.doorbell_size = DOORBELL_SIZE_GFX9,
};
struct device_info raven_device_info = {
const struct device_info raven_device_info = {
.asic_family = CHIP_RAVEN,
.eop_buffer_size = 4096,
.doorbell_size = DOORBELL_SIZE_GFX9,
};
struct device_info vega20_device_info = {
const struct device_info vega20_device_info = {
.asic_family = CHIP_VEGA20,
.eop_buffer_size = 4096,
.doorbell_size = DOORBELL_SIZE_GFX9,
};
static struct device_info *dev_lookup_table[] = {
static const struct device_info *dev_lookup_table[] = {
[CHIP_KAVERI] = &kaveri_device_info,
[CHIP_HAWAII] = &hawaii_device_info,
[CHIP_CARRIZO] = &carrizo_device_info,
@@ -124,11 +124,6 @@ static struct device_info *dev_lookup_table[] = {
[CHIP_RAVEN] = &raven_device_info
};
struct device_id {
uint16_t dev_id;
struct device_info *dev_info;
};
struct queue {
uint32_t queue_id;
uint64_t wptr;
@@ -138,6 +133,7 @@ struct queue {
uint32_t ctx_save_restore_size;
uint32_t ctl_stack_size;
const struct device_info *dev_info;
bool use_ats;
/* This queue structure is allocated from GPU with page aligned size
* but only small bytes are used. We use the extra space in the end for
* cu_mask bits array.
@@ -180,7 +176,7 @@ HSAKMT_STATUS init_process_doorbells(unsigned int NumNodes)
return ret;
}
static struct device_info *get_device_info_by_dev_id(uint16_t dev_id)
static const struct device_info *get_device_info_by_dev_id(uint16_t dev_id)
{
enum asic_family_type asic;
@@ -193,7 +189,7 @@ static struct device_info *get_device_info_by_dev_id(uint16_t dev_id)
static void get_doorbell_map_info(uint16_t dev_id,
struct process_doorbells *doorbell)
{
struct device_info *dev_info;
const struct device_info *dev_info;
dev_info = get_device_info_by_dev_id(dev_id);
@@ -301,7 +297,7 @@ static HSAKMT_STATUS map_doorbell(HSAuint32 NodeId, HSAuint32 gpu_id,
return HSAKMT_STATUS_SUCCESS;
}
get_doorbell_map_info(get_device_id_by_node(NodeId),
get_doorbell_map_info(get_device_id_by_node_id(NodeId),
&doorbells[NodeId]);
if (doorbells[NodeId].use_gpuvm) {
@@ -414,20 +410,20 @@ void free_exec_aligned_memory_gpu(void *addr, uint32_t size, uint32_t align)
* Allocates memory aligned to sysconf(_SC_PAGESIZE)
*/
static void *allocate_exec_aligned_memory(uint32_t size,
enum asic_family_type type,
bool use_ats,
uint32_t NodeId,
bool DeviceLocal)
{
if (IS_DGPU(type))
if (!use_ats)
return allocate_exec_aligned_memory_gpu(size, PAGE_SIZE, NodeId,
DeviceLocal, DeviceLocal);
return allocate_exec_aligned_memory_cpu(size);
}
static void free_exec_aligned_memory(void *addr, uint32_t size, uint32_t align,
enum asic_family_type type)
bool use_ats)
{
if (IS_DGPU(type))
if (!use_ats)
free_exec_aligned_memory_gpu(addr, size, align);
else
munmap(addr, size);
@@ -438,11 +434,11 @@ static void free_queue(struct queue *q)
if (q->eop_buffer)
free_exec_aligned_memory(q->eop_buffer,
q->dev_info->eop_buffer_size,
PAGE_SIZE, q->dev_info->asic_family);
PAGE_SIZE, q->use_ats);
if (q->ctx_save_restore)
free_exec_aligned_memory(q->ctx_save_restore,
q->ctx_save_restore_size,
PAGE_SIZE, q->dev_info->asic_family);
PAGE_SIZE, q->use_ats);
free_exec_aligned_memory((void *)q, sizeof(*q), PAGE_SIZE, q->dev_info->asic_family);
}
@@ -460,7 +456,7 @@ static int handle_concrete_asic(struct queue *q,
if (dev_info->eop_buffer_size > 0) {
q->eop_buffer =
allocate_exec_aligned_memory(q->dev_info->eop_buffer_size,
dev_info->asic_family,
q->use_ats,
NodeId, true);
if (!q->eop_buffer)
return HSAKMT_STATUS_NO_MEMORY;
@@ -476,7 +472,7 @@ static int handle_concrete_asic(struct queue *q,
args->ctl_stack_size = q->ctl_stack_size;
q->ctx_save_restore =
allocate_exec_aligned_memory(q->ctx_save_restore_size,
dev_info->asic_family,
q->use_ats,
NodeId, false);
if (!q->ctx_save_restore)
return HSAKMT_STATUS_NO_MEMORY;
@@ -507,10 +503,11 @@ HSAKMT_STATUS HSAKMTAPI hsaKmtCreateQueue(HSAuint32 NodeId,
uint16_t dev_id;
uint64_t doorbell_mmap_offset;
unsigned int doorbell_offset;
struct device_info *dev_info;
const struct device_info *dev_info;
int err;
HsaNodeProperties props;
uint32_t cu_num, i;
bool use_ats;
CHECK_KFD_OPEN();
@@ -522,17 +519,22 @@ HSAKMT_STATUS HSAKMTAPI hsaKmtCreateQueue(HSAuint32 NodeId,
if (result != HSAKMT_STATUS_SUCCESS)
return result;
dev_id = get_device_id_by_node(NodeId);
use_ats = prefer_ats(NodeId);
dev_id = get_device_id_by_node_id(NodeId);
dev_info = get_device_info_by_dev_id(dev_id);
struct queue *q = allocate_exec_aligned_memory(sizeof(*q),
dev_info->asic_family,
use_ats,
NodeId, false);
if (!q)
return HSAKMT_STATUS_NO_MEMORY;
memset(q, 0, sizeof(*q));
q->use_ats = use_ats;
q->dev_info = dev_info;
/* By default, CUs are all turned on. Initialize cu_mask to '1
* for all CU bits.
*/
@@ -550,8 +552,6 @@ HSAKMT_STATUS HSAKMTAPI hsaKmtCreateQueue(HSAuint32 NodeId,
args.gpu_id = gpu_id;
q->dev_info = dev_info;
switch (Type) {
case HSA_QUEUE_COMPUTE:
args.queue_type = KFD_IOC_QUEUE_TYPE_COMPUTE;
src/topology.c
+8 -1
Näytä tiedosto
@@ -1939,7 +1939,7 @@ out:
return err;
}
uint16_t get_device_id_by_node(HSAuint32 node_id)
uint16_t get_device_id_by_node_id(HSAuint32 node_id)
{
if (!g_props || !_system || _system->NumNodes <= node_id)
return 0;
@@ -1947,6 +1947,13 @@ uint16_t get_device_id_by_node(HSAuint32 node_id)
return g_props[node_id].node.DeviceId;
}
bool prefer_ats(HSAuint32 node_id)
{
return g_props[node_id].node.Capability.ui32.HSAMMUPresent
&& g_props[node_id].node.NumCPUCores
&& g_props[node_id].node.NumFComputeCores;
}
uint16_t get_device_id_by_gpu_id(HSAuint32 gpu_id)
{
unsigned int i;