Revision - Addition [Impact SoC: MI200, MI300]
Note: this set of counters are important help understand the
bottleneck.
1. TCC_TAG_STALL
a. Metric: TCC_TAG_STALL/TCC_CYCLE: percentage of time TCC
tag lookup pipeline is stalled
2. TCP_TCR_TCP_STALL_CYCLES
a. Metric: TCP_TCR_TCP_STALL_CYCLES/TCP_GATE_EN1: percentage
of time TCP is stalled by TCR
Revision - Addition [Impact SoC: MI300]
3. TCC_BUBBLE:
a. Definition: Number of 128-byte read requests sent to EA
b. Revised Metric #1, TCC-EA Read BW:
ReadBW = 128 * TCC_BUBBLE
+ 64 * (TCC_EA0_RDREQ - TCC_BUBBLE - TCC_EA0_RDREQ_32B)
+ 32 * TCC_EA0_RDREQ_32B
c. Revised Metric #2: TCC_EA Read Latency
ReadLatency = TCC_EA0_RDREQ_LEVEL / (TCC_BUBBLE + TCC_EA0_RDREQ)
/* [Fineprint] More detailed arithmetic:
* ReadLatency = TCC_EA0_RDREQ_LEVEL / (#32B_req + #64B_req + #128B_req * 2)
*/
Change-Id: I0a2dfc1b64ca97023b1e8ba0f9830330b3034946
[ROCm/rocprofiler commit: 46e02a9866]
The src/core/isa_capture/code_object_track.cpp file includes the
amd-dbgapi header but does not use the library (and is not configured to
use the library).
Remove the unused import.
Change-Id: Icc6c46a1f52355e9e5631395dfad1b44365bdca4
[ROCm/rocprofiler commit: 372fc3910b]
Unittest is failing in ASAN builds. To unblock ASAN builds, disabling the same as a workaround
The test will run in non-asan builds
Change-Id: Ib3c8d7a2c77aa83e9c6260e9b47ccb0564f789e8
[ROCm/rocprofiler commit: 8babd28a50]
Change-Id: I7b8f37867d7aaf6da1301d91028990c343e555f1
- Updated README for kernel name truncation.
- Updated CHANGELOG.
- By default enabled truncation for ATT because csv file names should be smaller
Change-Id: I7b8f37867d7aaf6da1301d91028990c343e555f1
[ROCm/rocprofiler commit: 296c935874]
Change-Id: I74caa4ab89bd765e59686cfbaaf1ce44ab10fe19
- adds counter correctness tests for GRBM, SQWAVES and SQ INSTRUCTIONs
- pandas dependency
- made it compatible with test infra.
Change-Id: I74caa4ab89bd765e59686cfbaaf1ce44ab10fe19
[ROCm/rocprofiler commit: 3f5ddfa0ab]
If hsa_shut_down() is not called from the application then we may still have async calls back
to the rocprofiler to use session parameters, thats why we need to leak the session up till
this is fixed in the ROCR-Runtime
Change-Id: Ia8fcdaa859815d88a272296770d702968132e3e6
[ROCm/rocprofiler commit: c9b5d73ca8]
Change-Id: I050120ba0dd0df84a46f022c0f9b8117057202b1
When we did file plugin changes for omniperf in v2, we removed 2 fields from header:
SIG and OBJ. However from the values we missed removing OBJ. Fix is to remove the OBJ from values.
Change-Id: I050120ba0dd0df84a46f022c0f9b8117057202b1
[ROCm/rocprofiler commit: ab066543b1]
Adding Versioning to rocprof main script
Adding Versioning to file plugins to support legacy v1 output and new v2 output
Fixing headers for v2 output
Change-Id: I352e8c8ed03871466269a792d1bfa572da58bd8a
[ROCm/rocprofiler commit: 4f40fb8a40]
The softlinks librocprofiler64.so and librocprofiler64v2.so was missing in ASAN package
Added the same
Change-Id: I8775b811f08377845dcce6bca88818c1741f3462
[ROCm/rocprofiler commit: a12d47fdb1]
Using `gfxip` when listing counters on the system with multi-GPUs
all belonging to the same family results in showing counters
of the first agent multiple times.
Instead, use the agent's `name` to observe counters of all agents.
Change-Id: I7b46c0670bc7c918e2ba0357fa659df9d23379d4
[ROCm/rocprofiler commit: 03a63791b0]
Mythreya
gobhardw
Ammar ELWazir
Lancelot Six
Parag Bhandari
Ranjith Ramakrishnan
Giovanni LB
Sriraksha Nagaraj
Benjamin Welton
Saurabh Verma
vlaindic