Jonathan Kim 7df0167821 Enable D2D SDMA Ganging over xGMI ...

Use all available SDMA engines capped by xGMI bandwith for
all D2D copies within a hive.

By default, set the latency boundary copy size as 4KB and below.
Any copy size in within this boundary will not gang.

Avoid oversubscribing engines by not ganging on engines with
pending non-ganged work.

An enviroment variable HSA_ENABLE_SDMA_GANG has been provided
to override default ganging behaviour.

Change-Id: Iccde76aa1af1d47ea2a151789432c9db4f0ffa8d

2023-07-27 08:58:26 -04:00

..

lnx

Fix non-x86 builds

2023-06-30 01:04:04 -04:00

win

Add function for parse CPUID information

2023-02-22 16:55:42 +00:00

atomic_helpers.h

Drop some unnecessary definitions

2022-04-19 12:22:42 -04:00

flag.cpp

Correct documentation typo.

2021-12-10 16:19:30 -06:00

flag.h

Enable D2D SDMA Ganging over xGMI

2023-07-27 08:58:26 -04:00

lazy_ptr.h

Add rocr namespace to core header and impl files

2020-06-19 22:34:21 -04:00

locks.h

Rework memory locks to allow device parallelism in alloc/free.

2021-11-24 19:22:05 -06:00

os.h

Add function for parse CPUID information

2023-02-22 16:55:42 +00:00

simple_heap.h

Add API for available GPU memory

2022-06-07 10:33:18 -04:00

small_heap.cpp

Add rocr namespace to core header and impl files

2020-06-19 22:34:21 -04:00

small_heap.h

Add rocr namespace to core header and impl files

2020-06-19 22:34:21 -04:00

timer.cpp

Correct inf loop defect in fast clock init.

2022-04-01 16:15:37 -04:00

timer.h

Drop some unnecessary definitions

2022-04-19 12:22:42 -04:00

utils.h

Correct scratch allocation logic to account for asymmetric harvest.

2022-03-02 19:59:30 -06:00