The lifetime of the kernel descriptor handle is unknown, so we currently uses the record id to identify the kernel name
Change-Id: I8802ce245bb021bcb43dd97cd6334bc88de0faf4
If rocprofiler_set_api_trace_sync_callback is not called by the API client then it will be automatically saved in the buffer and it will be in async to the user to see the API trace data, sample for async api tracing is added to samples/tracer
Change-Id: I01266a12b0eec172fdcffd7f04c89c2fe96174bb
Records right now will report basic raw data, no pre-processing in API anymore, and it will be left to the tool to process data from the output of synchronous callbacks of the API tracing records output
Change-Id: I8e5845e49e1228cb2fb60ca188ee748745b6a1af
This is an attempt to support basic and derived counters for navi21. This code will not work correctly unless we add navi counters to metrics.xml and gfx_metrics.xml
Change-Id: Ied06a81345a6fbb02fa0fde1889d94bbe64e9a03
Use hsa header files from /opt/rocm-ver/include rather than using wrapper files from /opt/rocm-ver/hsa/include/hsa
Change-Id: Id7a9bde19447cd2a0fd6e03b11c08471f09c2a46
Fixed exception thrown when ROCP_HSA_INTERCEPT not set or set to 0;
Fixed ROCM hsa_init() failed with error 4096 when trying to read hardware performance counters;
Fixed LD_LIBRARY_PATH to include necessary library;
Change-Id: Idcb7ff807a79f4267374c34041d3bca33d85f532
Changed derived metrics to double from int64.
Fixed standalone test due to int64 to float change
Fixed intercept test due to int64 to float change.
Change-Id: I49631c187406ae9dd94a869b3bb13772012e8cdf
L2 flush is triggered by explicit cache flush PM4 packet in aqlprofile
packets to GPU. This cache flush is used to sync up CPU and GPU to make
sure perfomance counters copied to profile output buffer is visible to
CPU. To get rid of this cache flush the followings are done:
1. This explicit cache flush packet is removed from aqlprofile code
(another commit to aqlprofile code).
2. This commit which changed profile output buffer to use kernarg
memory since it is uncached for GPU.
After these changes profile counter values when copied by GPU to output
buffer they are guaranteed to be visible to CPU.
Change-Id: Ie953949c85fbee2f4369f1de966bcfb33daec084
Concurrent profiling relies on the aqlprofile read_api
and tracker. This patch set those options to enable
the concurrent profiling.
Change-Id: Ib97d4d8facfbc11f2684d83109397cd13f117d5e
This patch adds barrier packets, together with extra signals,
to enforce the completion order of read packets w.r.t dispatch.
And, PmcStopper is added to stop the profiling finally.
Change-Id: I8e8d3a41d86e42be1d9e5afd44c247be876cf1a5
The profiling was only enabled in serial mode, i.e., kernels
are serialized in execution, and counters are reset at each
kernel start and read at kernel completion. This patch adds
the concurrent mode, by issuing the process-level start
packet to reset counters, and then reading twice at kernel
start and end time to obtain the counter value difference.
The new concurrent profiling usage needs the integration
with the corresponding augment at aqlprofile side.
Change-Id: I94b4442eadc8c64b8fba51b1e4916fc8b895ad21
gobhardw
Giovanni LB
Ammar ELWazir
Sriraksha Nagaraj
Saurabh Verma
Kiumars Sabeti
Ranjith Ramakrishnan
Mark Laws
Chun Yang
AMD
Rachida Kebichi
Evgeny
Xianwei Zhang