perf vendor events intel: update
Subsystems (2)
cover perf vendor events intel: update
Sync the Linux kernel perf PMU events with the latest upstream Intel perfmon repository.
This patchset includes event and metric updates for Alder Lake, Alder Lake N, Arrow Lake, Clearwater Forest, Emerald Rapids (including IIO bandwidth metrics), Grand Ridge, Granite Rapids, Lunar Lake, Meteor Lake, Panther Lake, and Sierra Forest, as well as adding initial event support for Nova Lake v1.00.
Chun-Tse Shao (12):
perf vendor events intel: Update alderlake events from 1.39 to 1.40 perf vendor events intel: Update alderlaken events from 1.39 to 1.40 perf vendor events intel: Update arrowlake events from 1.19 to 1.20 perf vendor events intel: Update clearwaterforest events from 1.02 to
1.04
perf vendor events intel: Update grandridge events from 1.12 to 1.13
perf vendor events intel: Update graniterapids events from 1.19 to
1.20 perf vendor events intel: Update lunarlake events from 1.25 to 1.26 perf vendor events intel: Update meteorlake events from 1.21 to 1.22 perf vendor events intel: Add novalake v1.00 events
perf vendor events intel: Update pantherlake events from 1.06 to 1.07 perf vendor events intel: Update sierraforest events from 1.17 to 1.18 perf vendor events intel: Update emeraldrapids metrics
.../arch/x86/alderlake/adl-metrics.json | 80 +- .../pmu-events/arch/x86/alderlake/cache.json | 36 + .../arch/x86/alderlake/metricgroups.json | 2 +- .../arch/x86/alderlake/pipeline.json | 6 +- .../pmu-events/arch/x86/alderlaken/cache.json | 32 + .../arch/x86/arrowlake/arl-metrics.json | 152 +- .../arch/x86/arrowlake/metricgroups.json | 2 +- .../arch/x86/arrowlake/pipeline.json | 2 +- .../arch/x86/arrowlake/uncore-memory.json | 84 +- .../arch/x86/clearwaterforest/memory.json | 33 + .../arch/x86/clearwaterforest/pipeline.json | 19 + .../clearwaterforest/uncore-interconnect.json | 10 + .../x86/clearwaterforest/uncore-memory.json | 2 +- .../arch/x86/emeraldrapids/emr-metrics.json | 135 +- .../arch/x86/emeraldrapids/metricgroups.json | 2 +- .../pmu-events/arch/x86/grandridge/cache.json | 25 + .../arch/x86/grandridge/counter.json | 42 - .../arch/x86/grandridge/floating-point.json | 56 + .../arch/x86/grandridge/memory.json | 57 + .../arch/x86/grandridge/pipeline.json | 140 ++ .../arch/x86/grandridge/uncore-memory.json | 2 +- .../arch/x86/grandridge/virtual-memory.json | 8 + .../arch/x86/graniterapids/gnr-metrics.json | 102 +- .../arch/x86/graniterapids/metricgroups.json | 2 +- .../arch/x86/graniterapids/pipeline.json | 2 +- .../arch/x86/lunarlake/lnl-metrics.json | 142 +- .../arch/x86/lunarlake/metricgroups.json | 2 +- .../arch/x86/lunarlake/pipeline.json | 2 +- tools/perf/pmu-events/arch/x86/mapfile.csv | 21 +- .../pmu-events/arch/x86/meteorlake/cache.json | 11 +- .../arch/x86/meteorlake/metricgroups.json | 2 +- .../arch/x86/meteorlake/mtl-metrics.json | 88 +- .../arch/x86/meteorlake/pipeline.json | 4 +- .../arch/x86/meteorlake/uncore-memory.json | 84 +- .../pmu-events/arch/x86/novalake/cache.json | 749 +++++++ .../pmu-events/arch/x86/novalake/counter.json | 17 + .../arch/x86/novalake/floating-point.json | 452 ++++ .../arch/x86/novalake/frontend.json | 600 +++++ .../pmu-events/arch/x86/novalake/memory.json | 299 +++ .../pmu-events/arch/x86/novalake/other.json | 144 ++ .../arch/x86/novalake/pipeline.json | 1976 +++++++++++++++++ .../arch/x86/novalake/uncore-memory.json | 42 + .../arch/x86/novalake/virtual-memory.json | 338 +++ .../arch/x86/pantherlake/pipeline.json | 2 +- .../arch/x86/sierraforest/counter.json | 77 - .../x86/sierraforest/uncore-interconnect.json | 10 + .../arch/x86/sierraforest/uncore-memory.json | 2 +- 47 files changed, 5615 insertions(+), 482 deletions(-) delete mode 100644 tools/perf/pmu-events/arch/x86/grandridge/counter.json create mode 100644 tools/perf/pmu-events/arch/x86/novalake/cache.json create mode 100644 tools/perf/pmu-events/arch/x86/novalake/counter.json create mode 100644 tools/perf/pmu-events/arch/x86/novalake/floating-point.json create mode 100644 tools/perf/pmu-events/arch/x86/novalake/frontend.json create mode 100644 tools/perf/pmu-events/arch/x86/novalake/memory.json create mode 100644 tools/perf/pmu-events/arch/x86/novalake/other.json create mode 100644 tools/perf/pmu-events/arch/x86/novalake/pipeline.json create mode 100644 tools/perf/pmu-events/arch/x86/novalake/uncore-memory.json create mode 100644 tools/perf/pmu-events/arch/x86/novalake/virtual-memory.json delete mode 100644 tools/perf/pmu-events/arch/x86/sierraforest/counter.json
Signature · 1 line
2.55.0.795.g602f6c329a-goog
PATCH 1/12 perf vendor events intel: Update alderlake events from 1.39 to 1.40
The updated events were published in:
https://github.com/intel/perfmon/commit/7a14cc8feaf86772deb6708e96c8e9fee6d5b1ca
--- .../arch/x86/alderlake/adl-metrics.json | 80 ++++++++++--------- .../pmu-events/arch/x86/alderlake/cache.json | 36 +++++++++ .../arch/x86/alderlake/metricgroups.json | 2 +- .../arch/x86/alderlake/pipeline.json | 6 +- tools/perf/pmu-events/arch/x86/mapfile.csv | 2 +- 5 files changed, 82 insertions(+), 44 deletions(-)
tools/perf/pmu-events/arch/x86/alderlake/adl-metrics.json +41 -39
@@ -800,16 +800,16 @@ }, { "BriefDescription": "Total pipeline cost of external Memory- or Cache-Bandwidth related bottlenecks", - "MetricExpr": "100 * (tma_memory_bound * (tma_dram_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_mem_bandwidth / (tma_mem_bandwidth + tma_mem_latency)) + tma_memory_bound * (tma_l3_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_sq_full / (tma_contested_accesses + tma_data_sharing + tma_l3_hit_latency + tma_sq_full)) + tma_memory_bound * (tma_l1_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_fb_full / (tma_dtlb_load + tma_fb_full + tma_l1_latency_dependency + tma_lock_latency + tma_split_loads + tma_store_fwd_blk)))", + "MetricExpr": "100 * (tma_memory_bound * (tma_l3_miss_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_mem_bandwidth / (tma_mem_bandwidth + tma_mem_latency)) + tma_memory_bound * (tma_l3_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_sq_full / (tma_contested_accesses + tma_data_sharing + tma_l3_hit_latency + tma_sq_full)) + tma_memory_bound * (tma_l1_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_fb_full / (tma_dtlb_load + tma_fb_full + tma_l1_latency_dependency + tma_lock_latency + tma_split_loads + tma_store_fwd_blk)))", "MetricGroup": "BvMB;Mem;MemoryBW;Offcore;tma_issueBW", "MetricName": "tma_bottleneck_data_cache_memory_bandwidth", "MetricThreshold": "tma_bottleneck_data_cache_memory_bandwidth > 20", - "PublicDescription": "Total pipeline cost of external Memory- or Cache-Bandwidth related bottlenecks. Related metrics: tma_fb_full, tma_info_system_dram_bw_use, tma_mem_bandwidth, tma_sq_full", + "PublicDescription": "Total pipeline cost of external Memory- or Cache-Bandwidth related bottlenecks. Related metrics: tma_fb_full, tma_info_system_dram_bw_use, tma_mem_bandwidth, tma_sq_full, tma_uc_bound", "Unit": "cpu_core" }, { "BriefDescription": "Total pipeline cost of external Memory- or Cache-Latency related bottlenecks", - "MetricExpr": "100 * (tma_memory_bound * (tma_dram_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_mem_latency / (tma_mem_bandwidth + tma_mem_latency)) + tma_memory_bound * (tma_l3_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_l3_hit_latency / (tma_contested_accesses + tma_data_sharing + tma_l3_hit_latency + tma_sq_full)) + tma_memory_bound * tma_l2_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound) + tma_memory_bound * (tma_l1_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_l1_latency_dependency / (tma_dtlb_load + tma_fb_full + tma_l1_latency_dependency + tma_lock_latency + tma_split_loads + tma_store_fwd_blk)) + tma_memory_bound * (tma_l1_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_lock_latency / (tma_dtlb_load + tma_fb_full + tma_l1_latency_dependency + tma_lock_latency + tma_split_loads + tma_store_fwd_blk)) + tma_memory_bound * (tma_l1_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_split_loads / (tma_dtlb_load + tma_fb_full + tma_l1_latency_dependency + tma_lock_latency + tma_split_loads + tma_store_fwd_blk)) + tma_memory_bound * (tma_store_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_split_stores / (tma_dtlb_store + tma_false_sharing + tma_split_stores + tma_store_latency + tma_streaming_stores)) + tma_memory_bound * (tma_store_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_store_latency / (tma_dtlb_store + tma_false_sharing + tma_split_stores + tma_store_latency + tma_streaming_stores)))", + "MetricExpr": "100 * (tma_memory_bound * (tma_l3_miss_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_mem_latency / (tma_mem_bandwidth + tma_mem_latency)) + tma_memory_bound * (tma_l3_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_l3_hit_latency / (tma_contested_accesses + tma_data_sharing + tma_l3_hit_latency + tma_sq_full)) + tma_memory_bound * tma_l2_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound) + tma_memory_bound * (tma_l1_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_l1_latency_dependency / (tma_dtlb_load + tma_fb_full + tma_l1_latency_dependency + tma_lock_latency + tma_split_loads + tma_store_fwd_blk)) + tma_memory_bound * (tma_l1_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_lock_latency / (tma_dtlb_load + tma_fb_full + tma_l1_latency_dependency + tma_lock_latency + tma_split_loads + tma_store_fwd_blk)) + tma_memory_bound * (tma_l1_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_split_loads / (tma_dtlb_load + tma_fb_full + tma_l1_latency_dependency + tma_lock_latency + tma_split_loads + tma_store_fwd_blk)) + tma_memory_bound * (tma_store_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_split_stores / (tma_dtlb_store + tma_false_sharing + tma_split_stores + tma_store_latency + tma_streaming_stores)) + tma_memory_bound * (tma_store_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_store_latency / (tma_dtlb_store + tma_false_sharing + tma_split_stores + tma_store_latency + tma_streaming_stores)))", "MetricGroup": "BvML;Mem;MemoryLat;Offcore;tma_issueLat", "MetricName": "tma_bottleneck_data_cache_memory_latency", "MetricThreshold": "tma_bottleneck_data_cache_memory_latency > 20",
@@ -835,7 +835,7 @@ }, { "BriefDescription": "Total pipeline cost of Memory Address Translation related bottlenecks (data-side TLBs)", - "MetricExpr": "100 * (tma_memory_bound * (tma_l1_bound / max(tma_memory_bound, tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_dtlb_load / max(tma_l1_bound, tma_dtlb_load + tma_fb_full + tma_l1_latency_dependency + tma_lock_latency + tma_split_loads + tma_store_fwd_blk)) + tma_memory_bound * (tma_store_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_dtlb_store / (tma_dtlb_store + tma_false_sharing + tma_split_stores + tma_store_latency + tma_streaming_stores)))", + "MetricExpr": "100 * (tma_memory_bound * (tma_l1_bound / max(tma_memory_bound, tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_dtlb_load / max(tma_l1_bound, tma_dtlb_load + tma_fb_full + tma_l1_latency_dependency + tma_lock_latency + tma_split_loads + tma_store_fwd_blk)) + tma_memory_bound * (tma_store_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_dtlb_store / (tma_dtlb_store + tma_false_sharing + tma_split_stores + tma_store_latency + tma_streaming_stores)))", "MetricGroup": "BvMT;Mem;MemoryTLB;Offcore;tma_issueTLB", "MetricName": "tma_bottleneck_memory_data_tlbs", "MetricThreshold": "tma_bottleneck_memory_data_tlbs > 20",
@@ -844,7 +844,7 @@ }, { "BriefDescription": "Total pipeline cost of Memory Synchronization related bottlenecks (data transfers and coherency updates across processors)", - "MetricExpr": "100 * (tma_memory_bound * (tma_l3_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound) * (tma_contested_accesses + tma_data_sharing) / (tma_contested_accesses + tma_data_sharing + tma_l3_hit_latency + tma_sq_full) + tma_store_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound) * tma_false_sharing / (tma_dtlb_store + tma_false_sharing + tma_split_stores + tma_store_latency + tma_streaming_stores - tma_store_latency)) + tma_machine_clears * (1 - tma_other_nukes / tma_other_nukes))", + "MetricExpr": "100 * (tma_memory_bound * (tma_l3_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound) * (tma_contested_accesses + tma_data_sharing) / (tma_contested_accesses + tma_data_sharing + tma_l3_hit_latency + tma_sq_full) + tma_store_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound) * tma_false_sharing / (tma_dtlb_store + tma_false_sharing + tma_split_stores + tma_store_latency + tma_streaming_stores - tma_store_latency)) + tma_machine_clears * (1 - tma_other_nukes / tma_other_nukes))", "MetricGroup": "BvMS;LockCont;Mem;Offcore;tma_issueSyncxn", "MetricName": "tma_bottleneck_memory_synchronization", "MetricThreshold": "tma_bottleneck_memory_synchronization > 10",
@@ -1012,13 +1012,13 @@ "Unit": "cpu_core" }, { - "BriefDescription": "This metric estimates fraction of cycles while the memory subsystem was handling synchronizations due to data-sharing accesses", + "BriefDescription": "This metric estimates fraction of cycles while the memory subsystem was handling synchronizations due to L3 data-sharing accesses", "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "24 * tma_info_system_core_frequency * (cpu_core@MEM_LOAD_L3_HIT_RETIRED.XSNP_NO_FWD@ + cpu_core@MEM_LOAD_L3_HIT_RETIRED.XSNP_FWD@ * (1 - [email protected]_DATA_RD.L3_HIT.SNOOP_HITM@ / ([email protected]_DATA_RD.L3_HIT.SNOOP_HITM@ + [email protected]_DATA_RD.L3_HIT.SNOOP_HIT_WITH_FWD@))) * (1 + cpu_core@MEM_LOAD_RETIRED.FB_HIT@ / cpu_core@MEM_LOAD_RETIRED.L1_MISS@ / 2) / tma_info_thread_clks", "MetricGroup": "BvMS;Offcore;Snoop;TopdownL4;tma_L4_group;tma_issueSyncxn;tma_l3_bound_group", "MetricName": "tma_data_sharing", "MetricThreshold": "tma_data_sharing > 0.05 & (tma_l3_bound > 0.05 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))", - "PublicDescription": "This metric estimates fraction of cycles while the memory subsystem was handling synchronizations due to data-sharing accesses. Data shared by multiple Logical Processors (even just read shared) may cause increased access latency due to cache coherency. Excessive data sharing can drastically harm multithreaded performance. Sample with: MEM_LOAD_L3_HIT_RETIRED.XSNP_NO_FWD. Related metrics: tma_bottleneck_memory_synchronization, tma_contested_accesses, tma_false_sharing, tma_machine_clears, tma_remote_cache", + "PublicDescription": "This metric estimates fraction of cycles while the memory subsystem was handling synchronizations due to L3 data-sharing accesses. Data shared by multiple Logical Processors (even just read shared) may cause increased access latency due to cache coherency. Excessive data sharing can drastically harm multithreaded performance. Sample with: MEM_LOAD_L3_HIT_RETIRED.XSNP_NO_FWD. Related metrics: tma_bottleneck_memory_synchronization, tma_contested_accesses, tma_false_sharing, tma_machine_clears, tma_remote_cache", "ScaleUnit": "100%", "Unit": "cpu_core" },
@@ -1042,16 +1042,6 @@ "ScaleUnit": "100%", "Unit": "cpu_core" }, - { - "BriefDescription": "This metric estimates how often the CPU was stalled on accesses to external memory (DRAM) by loads", - "MetricExpr": "cpu_core@MEMORY_ACTIVITY.STALLS_L3_MISS@ / tma_info_thread_clks", - "MetricGroup": "MemoryBound;TmaL3mem;TopdownL3;tma_L3_group;tma_memory_bound_group", - "MetricName": "tma_dram_bound", - "MetricThreshold": "tma_dram_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2)", - "PublicDescription": "This metric estimates how often the CPU was stalled on accesses to external memory (DRAM) by loads. Better caching can improve the latency and increase performance. Sample with: MEM_LOAD_RETIRED.L3_MISS", - "ScaleUnit": "100%", - "Unit": "cpu_core" - }, { "BriefDescription": "This metric represents Core fraction of cycles in which CPU was likely limited due to DSB (decoded uop cache) fetch pipeline", "MetricExpr": "([email protected]_CYCLES_ANY@ - [email protected]_CYCLES_OK@) / tma_info_core_core_clks / 2",
@@ -1108,7 +1098,7 @@ "MetricGroup": "BvMB;MemoryBW;TopdownL4;tma_L4_group;tma_issueBW;tma_issueSL;tma_issueSmSt;tma_l1_bound_group", "MetricName": "tma_fb_full", "MetricThreshold": "tma_fb_full > 0.3", - "PublicDescription": "This metric does a *rough estimation* of how often L1D Fill Buffer unavailability limited additional L1D miss memory access requests to proceed. The higher the metric value; the deeper the memory hierarchy level the misses are satisfied from (metric values >1 are valid). Often it hints on approaching bandwidth limits (to L2 cache; L3 cache or external memory). Related metrics: tma_bottleneck_data_cache_memory_bandwidth, tma_info_system_dram_bw_use, tma_mem_bandwidth, tma_sq_full, tma_store_latency, tma_streaming_stores", + "PublicDescription": "This metric does a *rough estimation* of how often L1D Fill Buffer unavailability limited additional L1D miss memory access requests to proceed. The higher the metric value; the deeper the memory hierarchy level the misses are satisfied from (metric values >1 are valid). Often it hints on approaching bandwidth limits (to L2 cache; L3 cache or external memory). Related metrics: tma_bottleneck_data_cache_memory_bandwidth, tma_info_system_dram_bw_use, tma_mem_bandwidth, tma_sq_full, tma_store_latency, tma_streaming_stores, tma_uc_bound", "ScaleUnit": "100%", "Unit": "cpu_core" },
@@ -1426,7 +1416,7 @@ }, { "BriefDescription": "Fraction of Uops delivered by the DSB (aka Decoded ICache; or Uop Cache)", - "MetricExpr": "[email protected]_UOPS@ / cpu_core@UOPS_ISSUED.ANY@", + "MetricExpr": "[email protected]_UOPS@ / ([email protected]_UOPS@ + [email protected]@ + [email protected]_UOPS@ + [email protected]_UOPS@)", "MetricGroup": "DSB;Fed;FetchBW;tma_issueFB", "MetricName": "tma_info_frontend_dsb_coverage", "MetricThreshold": "tma_info_frontend_dsb_coverage < 0.7 & tma_info_thread_ipc / 6 > 0.35",
@@ -1485,7 +1475,7 @@ }, { "BriefDescription": "Fraction of Uops delivered by the LSD (Loop Stream Detector; aka Loop Cache)", - "MetricExpr": "[email protected]@ / cpu_core@UOPS_ISSUED.ANY@", + "MetricExpr": "[email protected]@ / ([email protected]_UOPS@ + [email protected]@ + [email protected]_UOPS@ + [email protected]_UOPS@)", "MetricGroup": "Fed;LSD", "MetricName": "tma_info_frontend_lsd_coverage", "Unit": "cpu_core"
@@ -1629,6 +1619,15 @@ "PublicDescription": "Instructions per taken branch. Related metrics: tma_dsb_switches, tma_fetch_bandwidth, tma_info_botlnk_l2_dsb_bandwidth, tma_info_botlnk_l2_dsb_misses, tma_info_frontend_dsb_coverage, tma_lcp", "Unit": "cpu_core" }, + { + "BriefDescription": "AVX preserve/restore assists per kilo instruction", + "MetricExpr": "1e3 * [email protected]_AVX_MIX@ / cpu_core@INST_RETIRED.ANY@", + "MetricGroup": "tma_issueMV", + "MetricName": "tma_info_inst_mix_vectormixpki", + "MetricThreshold": "tma_info_inst_mix_vectormixpki > 0.05", + "PublicDescription": "AVX preserve/restore assists per kilo instruction. Related metrics: tma_mixing_vectors, tma_ms_switches", + "Unit": "cpu_core" + }, { "BriefDescription": "Average per-core data fill bandwidth to the L1 data cache [GB / sec]", "MetricExpr": "tma_info_memory_l1d_cache_fill_bw",
@@ -1850,7 +1849,7 @@ "Unit": "cpu_core" }, { - "BriefDescription": "Mem;Backend;CacheHits", + "BriefDescription": "Instruction-Level-Parallelism (average number of uops executed when there is execution) per physical core", "MetricExpr": "cpu_core@UOPS_EXECUTED.THREAD@ / (cpu_core@UOPS_EXECUTED.CORE_CYCLES_GE_1@ / 2 if #SMT_on else cpu_core@UOPS_EXECUTED.THREAD\\,cmask\\=1@)", "MetricGroup": "Cor;Pipeline;PortsUtil;SMT", "MetricName": "tma_info_pipeline_execute",
@@ -1935,7 +1934,7 @@ "MetricExpr": "64 * (UNC_ARB_TRK_REQUESTS.ALL + UNC_ARB_COH_TRK_REQUESTS.ALL) / 1e6 / tma_info_system_time / 1e3", "MetricGroup": "HPC;MemOffcore;MemoryBW;SoC;tma_issueBW", "MetricName": "tma_info_system_dram_bw_use", - "PublicDescription": "Average external Memory Bandwidth Use for reads and writes [GB / sec]. Related metrics: tma_bottleneck_data_cache_memory_bandwidth, tma_fb_full, tma_mem_bandwidth, tma_sq_full", + "PublicDescription": "Average external Memory Bandwidth Use for reads and writes [GB / sec]. Related metrics: tma_bottleneck_data_cache_memory_bandwidth, tma_fb_full, tma_mem_bandwidth, tma_sq_full, tma_uc_bound", "Unit": "cpu_core" }, {
@@ -2029,13 +2028,6 @@ "MetricName": "tma_info_system_turbo_utilization", "Unit": "cpu_core" }, - { - "BriefDescription": "Measured Average Uncore Frequency for the SoC [GHz]", - "MetricExpr": "tma_info_system_socket_clks / 1e9 / tma_info_system_time", - "MetricGroup": "SoC", - "MetricName": "tma_info_system_uncore_frequency", - "Unit": "cpu_core" - }, { "BriefDescription": "Per-Logical Processor actual clocks when the Logical Processor is active.", "MetricExpr": "cpu_core@CPU_CLK_UNHALTED.THREAD@",
@@ -2204,6 +2196,16 @@ "ScaleUnit": "100%", "Unit": "cpu_core" }, + { + "BriefDescription": "This metric estimates how often the CPU was stalled on accesses to external memory (DRAM) by loads", + "MetricExpr": "cpu_core@MEMORY_ACTIVITY.STALLS_L3_MISS@ / tma_info_thread_clks", + "MetricGroup": "MemoryBound;Offcore;TmaL3mem;TopdownL3;tma_L3_group;tma_memory_bound_group", + "MetricName": "tma_l3_miss_bound", + "MetricThreshold": "tma_l3_miss_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2)", + "PublicDescription": "This metric estimates how often the CPU was stalled on accesses to external memory (DRAM) by loads. Better caching can improve the latency and increase performance. Sample with: MEM_LOAD_RETIRED.L3_MISS", + "ScaleUnit": "100%", + "Unit": "cpu_core" + }, { "BriefDescription": "This metric represents fraction of cycles CPU was stalled due to Length Changing Prefixes (LCPs)", "MetricExpr": "[email protected]@ / tma_info_thread_clks",
@@ -2283,7 +2285,7 @@ { "BriefDescription": "This metric represents fraction of cycles the CPU spent handling cache misses due to lock operations", "MetricConstraint": "NO_GROUP_EVENTS", - "MetricExpr": "(16 * max(0, cpu_core@MEM_INST_RETIRED.LOCK_LOADS@ - cpu_core@L2_RQSTS.ALL_RFO@) + cpu_core@MEM_INST_RETIRED.LOCK_LOADS@ / cpu_core@MEM_INST_RETIRED.ALL_STORES@ * (10 * cpu_core@L2_RQSTS.RFO_HIT@ + min(cpu_core@CPU_CLK_UNHALTED.THREAD@, cpu_core@OFFCORE_REQUESTS_OUTSTANDING.CYCLES_WITH_DEMAND_RFO@))) / tma_info_thread_clks", + "MetricExpr": "cpu_core@LOCK_CYCLES.CACHE_LOCK_DURATION@ / tma_info_thread_clks", "MetricGroup": "LockCont;Offcore;TopdownL4;tma_L4_group;tma_issueRFO;tma_l1_bound_group", "MetricName": "tma_lock_latency", "MetricThreshold": "tma_lock_latency > 0.2 & (tma_l1_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))",
@@ -2315,19 +2317,19 @@ { "BriefDescription": "This metric estimates fraction of cycles where the core's performance was likely hurt due to approaching bandwidth limits of external memory - DRAM ([SPR-HBM] and/or HBM)", "MetricExpr": "min(cpu_core@CPU_CLK_UNHALTED.THREAD@, cpu_core@OFFCORE_REQUESTS_OUTSTANDING.ALL_DATA_RD\\,cmask\\=4@) / tma_info_thread_clks", - "MetricGroup": "BvMB;MemoryBW;Offcore;TopdownL4;tma_L4_group;tma_dram_bound_group;tma_issueBW", + "MetricGroup": "BvMB;MemoryBW;Offcore;TopdownL4;tma_L4_group;tma_issueBW;tma_l3_miss_bound_group", "MetricName": "tma_mem_bandwidth", - "MetricThreshold": "tma_mem_bandwidth > 0.2 & (tma_dram_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))", - "PublicDescription": "This metric estimates fraction of cycles where the core's performance was likely hurt due to approaching bandwidth limits of external memory - DRAM ([SPR-HBM] and/or HBM). The underlying heuristic assumes that a similar off-core traffic is generated by all IA cores. This metric does not aggregate non-data-read requests by this logical processor; requests from other IA Logical Processors/Physical Cores/sockets; or other non-IA devices like GPU; hence the maximum external memory bandwidth limits may or may not be approached when this metric is flagged (see Uncore counters for that). Related metrics: tma_bottleneck_data_cache_memory_bandwidth, tma_fb_full, tma_info_system_dram_bw_use, tma_sq_full", + "MetricThreshold": "tma_mem_bandwidth > 0.2 & (tma_l3_miss_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))", + "PublicDescription": "This metric estimates fraction of cycles where the core's performance was likely hurt due to approaching bandwidth limits of external memory - DRAM ([SPR-HBM] and/or HBM). The underlying heuristic assumes that a similar off-core traffic is generated by all IA cores. This metric does not aggregate non-data-read requests by this logical processor; requests from other IA Logical Processors/Physical Cores/sockets; or other non-IA devices like GPU; hence the maximum external memory bandwidth limits may or may not be approached when this metric is flagged (see Uncore counters for that). Related metrics: tma_bottleneck_data_cache_memory_bandwidth, tma_fb_full, tma_info_system_dram_bw_use, tma_sq_full, tma_uc_bound", "ScaleUnit": "100%", "Unit": "cpu_core" }, { "BriefDescription": "This metric estimates fraction of cycles where the performance was likely hurt due to latency from external memory - DRAM ([SPR-HBM] and/or HBM)", "MetricExpr": "min(cpu_core@CPU_CLK_UNHALTED.THREAD@, cpu_core@OFFCORE_REQUESTS_OUTSTANDING.CYCLES_WITH_DATA_RD@) / tma_info_thread_clks - tma_mem_bandwidth", - "MetricGroup": "BvML;MemoryLat;Offcore;TopdownL4;tma_L4_group;tma_dram_bound_group;tma_issueLat", + "MetricGroup": "BvML;MemoryLat;Offcore;TopdownL4;tma_L4_group;tma_issueLat;tma_l3_miss_bound_group", "MetricName": "tma_mem_latency", - "MetricThreshold": "tma_mem_latency > 0.1 & (tma_dram_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))", + "MetricThreshold": "tma_mem_latency > 0.1 & (tma_l3_miss_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))", "PublicDescription": "This metric estimates fraction of cycles where the performance was likely hurt due to latency from external memory - DRAM ([SPR-HBM] and/or HBM). This metric does not aggregate requests from other Logical Processors/Physical Cores/sockets (see Uncore counters for that). Related metrics: tma_bottleneck_data_cache_memory_latency, tma_l3_hit_latency", "ScaleUnit": "100%", "Unit": "cpu_core"
@@ -2397,7 +2399,7 @@ "MetricGroup": "TopdownL5;tma_L5_group;tma_issueMV;tma_ports_utilized_0_group", "MetricName": "tma_mixing_vectors", "MetricThreshold": "tma_mixing_vectors > 0.05", - "PublicDescription": "This metric estimates penalty in terms of percentage of([SKL+] injected blend uops out of all Uops Issued -- the Count Domain; [ADL+] cycles). Usually a Mixing_Vectors over 5% is worth investigating. Read more in Appendix B1 of the Optimizations Guide for this topic. Related metrics: tma_ms_switches", + "PublicDescription": "This metric estimates penalty in terms of percentage of([SKL+] injected blend uops out of all Uops Issued -- the Count Domain; [ADL+] cycles). Usually a Mixing_Vectors over 5% is worth investigating. Read more in Appendix B1 of the Optimizations Guide for this topic. Related metrics: tma_info_inst_mix_vectormixpki, tma_ms_switches", "ScaleUnit": "100%", "Unit": "cpu_core" },
@@ -2416,7 +2418,7 @@ "MetricGroup": "FetchLat;MicroSeq;TopdownL3;tma_L3_group;tma_fetch_latency_group;tma_issueMC;tma_issueMS;tma_issueMV;tma_issueSO", "MetricName": "tma_ms_switches", "MetricThreshold": "tma_ms_switches > 0.05 & (tma_fetch_latency > 0.1 & tma_frontend_bound > 0.15)", - "PublicDescription": "This metric estimates the fraction of cycles when the CPU was stalled due to switches of uop delivery to the Microcode Sequencer (MS). Commonly used instructions are optimized for delivery by the DSB (decoded i-cache) or MITE (legacy instruction decode) pipelines. Certain operations cannot be handled natively by the execution pipeline; and must be performed by microcode (small programs injected into the execution stream). Switching to the MS too often can negatively impact performance. The MS is designated to deliver long uop flows required by CISC instructions like CPUID; or uncommon conditions like Floating Point Assists when dealing with Denormals. Sample with: FRONTEND_RETIRED.MS_FLOWS. Related metrics: tma_bottleneck_irregular_overhead, tma_clears_resteers, tma_l1_bound, tma_machine_clears, tma_microcode_sequencer, tma_mixing_vectors, tma_serializing_operation", + "PublicDescription": "This metric estimates the fraction of cycles when the CPU was stalled due to switches of uop delivery to the Microcode Sequencer (MS). Commonly used instructions are optimized for delivery by the DSB (decoded i-cache) or MITE (legacy instruction decode) pipelines. Certain operations cannot be handled natively by the execution pipeline; and must be performed by microcode (small programs injected into the execution stream). Switching to the MS too often can negatively impact performance. The MS is designated to deliver long uop flows required by CISC instructions like CPUID; or uncommon conditions like Floating Point Assists when dealing with Denormals. Sample with: FRONTEND_RETIRED.MS_FLOWS. Related metrics: tma_bottleneck_irregular_overhead, tma_clears_resteers, tma_info_inst_mix_vectormixpki, tma_l1_bound, tma_machine_clears, tma_microcode_sequencer, tma_mixing_vectors, tma_serializing_operation", "ScaleUnit": "100%", "Unit": "cpu_core" },
@@ -2606,7 +2608,7 @@ }, { "BriefDescription": "This metric estimates fraction of cycles handling memory load split accesses - load that cross 64-byte cache line boundary", - "MetricExpr": "tma_info_memory_load_miss_real_latency * cpu_core@LD_BLOCKS.NO_SR@ / tma_info_thread_clks", + "MetricExpr": "cpu_core@MEM_INST_RETIRED.SPLIT_LOADS@ * tma_info_memory_load_miss_real_latency / tma_info_thread_clks", "MetricGroup": "TopdownL4;tma_L4_group;tma_l1_bound_group", "MetricName": "tma_split_loads", "MetricThreshold": "tma_split_loads > 0.3",
@@ -2630,7 +2632,7 @@ "MetricGroup": "BvMB;MemoryBW;Offcore;TopdownL4;tma_L4_group;tma_issueBW;tma_l3_bound_group", "MetricName": "tma_sq_full", "MetricThreshold": "tma_sq_full > 0.3 & (tma_l3_bound > 0.05 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))", - "PublicDescription": "This metric measures fraction of cycles where the Super Queue (SQ) was full taking into account all request-types and both hardware SMT threads (Logical Processors). Related metrics: tma_bottleneck_data_cache_memory_bandwidth, tma_fb_full, tma_info_system_dram_bw_use, tma_mem_bandwidth", + "PublicDescription": "This metric measures fraction of cycles where the Super Queue (SQ) was full taking into account all request-types and both hardware SMT threads (Logical Processors). Related metrics: tma_bottleneck_data_cache_memory_bandwidth, tma_fb_full, tma_info_system_dram_bw_use, tma_mem_bandwidth, tma_uc_bound", "ScaleUnit": "100%", "Unit": "cpu_core" },
tools/perf/pmu-events/arch/x86/alderlake/cache.json +36 -0
@@ -1,4 +1,13 @@ [ + { + "BriefDescription": "Counts the number of core requests (demand and L1 prefetchers) rejected by the L2 queue (L2Q) due to a full condition.", + "Counter": "0,1,2,3,4,5", + "EventCode": "0x31", + "EventName": "CORE_REJECT_L2Q.ANY", + "PublicDescription": "Counts the number of (demand and L1 prefetchers) core requests rejected by the L2 queue (L2Q) due to a full or nearly full condition, which likely indicates back pressure from L2Q. It also counts requests that would have gone directly to the External Queue (XQ), but are rejected due to a full or nearly full condition, indicating back pressure from the IDI link. The L2Q may also reject transactions from a core to ensure fairness between cores, or to delay a cores dirty eviction when the address conflicts incoming external snoops. (Note that L2 prefetcher requests that are dropped are not counted by this event). Counts on a per core basis.", + "SampleAfterValue": "200003", + "Unit": "cpu_atom" + }, { "BriefDescription": "L1D.HWPF_MISS", "Counter": "0,1,2,3",
@@ -121,6 +130,15 @@ "UMask": "0x4", "Unit": "cpu_core" }, + { + "BriefDescription": "Counts the number of demand and prefetch transactions that the External Queue (XQ) rejects due to a full or near full condition.", + "Counter": "0,1,2,3,4,5", + "EventCode": "0x30", + "EventName": "L2_REJECT_XQ.ANY", + "PublicDescription": "Counts the number of demand and prefetch transactions that the External Queue (XQ) rejects due to a full or near full condition which likely indicates back pressure from the IDI link. The XQ may reject transactions from the L2Q (non-cacheable requests), BBL (L2 misses) and WOB (L2 write-back victims).", + "SampleAfterValue": "200003", + "Unit": "cpu_atom" + }, { "BriefDescription": "Counts the total number of L2 Cache accesses. Counts on a per core basis.", "Counter": "0,1,2,3,4,5",
@@ -438,6 +456,15 @@ "UMask": "0x8", "Unit": "cpu_atom" }, + { + "BriefDescription": "Counts the number of cycles the core is stalled due to an instruction cache or TLB miss which missed in the L2 cache.", + "Counter": "0,1,2,3,4,5", + "EventCode": "0x34", + "EventName": "MEM_BOUND_STALLS.IFETCH_L2_MISS", + "SampleAfterValue": "1000003", + "UMask": "0x30", + "Unit": "cpu_atom" + }, { "BriefDescription": "Counts the number of cycles the core is stalled due to an instruction cache or TLB miss which hit in the LLC or other core with HITE/F/M.", "Counter": "0,1,2,3,4,5",
@@ -475,6 +502,15 @@ "UMask": "0x1", "Unit": "cpu_atom" }, + { + "BriefDescription": "Counts the number of cycles the core is stalled due to a demand load which missed in the L2 cache.", + "Counter": "0,1,2,3,4,5", + "EventCode": "0x34", + "EventName": "MEM_BOUND_STALLS.LOAD_L2_MISS", + "SampleAfterValue": "1000003", + "UMask": "0x6", + "Unit": "cpu_atom" + }, { "BriefDescription": "Counts the number of cycles the core is stalled due to a demand load which hit in the LLC or other core with HITE/F/M.", "Counter": "0,1,2,3,4,5",
tools/perf/pmu-events/arch/x86/alderlake/metricgroups.json +1 -1
@@ -93,7 +93,6 @@ "tma_code_stlb_miss_group": "Metrics contributing to tma_code_stlb_miss category", "tma_core_bound_group": "Metrics contributing to tma_core_bound category", "tma_divider_group": "Metrics contributing to tma_divider category", - "tma_dram_bound_group": "Metrics contributing to tma_dram_bound category", "tma_dtlb_load_group": "Metrics contributing to tma_dtlb_load category", "tma_dtlb_store_group": "Metrics contributing to tma_dtlb_store category", "tma_fetch_bandwidth_group": "Metrics contributing to tma_fetch_bandwidth category",
@@ -129,6 +128,7 @@ "tma_l1_bound_group": "Metrics contributing to tma_l1_bound category", "tma_l2_bound_group": "Metrics contributing to tma_l2_bound category", "tma_l3_bound_group": "Metrics contributing to tma_l3_bound category", + "tma_l3_miss_bound_group": "Metrics contributing to tma_l3_miss_bound category", "tma_light_operations_group": "Metrics contributing to tma_light_operations category", "tma_load_op_utilization_group": "Metrics contributing to tma_load_op_utilization category", "tma_load_stlb_miss_group": "Metrics contributing to tma_load_stlb_miss category",
tools/perf/pmu-events/arch/x86/alderlake/pipeline.json +3 -3
@@ -550,11 +550,11 @@ "Unit": "cpu_atom" }, { - "BriefDescription": "This event counts the number of mispredicted ret instructions retired. Non PEBS", + "BriefDescription": "This event counts the number of mispredicted ret instructions retired.", "Counter": "0,1,2,3,4,5,6,7", "EventCode": "0xc5", "EventName": "BR_MISP_RETIRED.RET", - "PublicDescription": "This is a non-precise version (that is, does not use PEBS) of the event that counts mispredicted return instructions retired. Available PDIST counters: 0", + "PublicDescription": "This event counts the number of mispredicted ret instructions retired. Available PDIST counters: 0", "SampleAfterValue": "100007", "UMask": "0x8", "Unit": "cpu_core"
@@ -988,7 +988,7 @@ "Counter": "0,1,2,3,4,5,6,7", "EventCode": "0xc0", "EventName": "INST_RETIRED.REP_ITERATION", - "PublicDescription": "Number of iterations of Repeat (REP) string retired instructions such as MOVS, CMPS, and SCAS. Each has a byte, word, and doubleword version and string instructions can be repeated using a repetition prefix, REP, that allows their architectural execution to be repeated a number of times as specified by the RCX register. Note the number of iterations is implementation-dependent.", + "PublicDescription": "Number of iterations of Repeat (REP) string retired instructions such as MOVS, CMPS, and SCAS. Each has a byte, word, and doubleword version and string instructions can be repeated using a repetition prefix, REP, that allows their architectural execution to be repeated a number of times as specified by the RCX register. Note: Since the number of iterations within a REP instruction can be significantly affected by fast strings, this event may vary run to run and not match the architectural number of iterations (specified by RCX)", "SampleAfterValue": "2000003", "UMask": "0x8", "Unit": "cpu_core"
tools/perf/pmu-events/arch/x86/mapfile.csv +1 -1
@@ -1,5 +1,5 @@ Family-model,Version,Filename,EventType -GenuineIntel-6-(97|9A|B7|BA|BF),v1.39,alderlake,core +GenuineIntel-6-(97|9A|B7|BA|BF),v1.40,alderlake,core GenuineIntel-6-BE,v1.39,alderlaken,core GenuineIntel-6-C[56],v1.19,arrowlake,core GenuineIntel-6-(1C|26|27|35|36),v5,bonnell,core
Signature · 1 line
2.55.0.795.g602f6c329a-goog
[email protected]
AI reviewRe: [PATCH v1 01/12] perf vendor events intel: Update alderlake events from 1.39 to 1.40
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider: - [Low] The metric `tma_uc_bound` is referenced as a 'Related metric' in the PublicDescription of several…
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Low] The metric tma_uc_bound is referenced as a 'Related metric' in the PublicDescription of several metrics, but it is never defined for Alderlake. --
--- Patch [1]: [PATCH v1 01/12] perf vendor events intel: Update alderlake events from 1.39 to 1.40 --- commit 681de70c22f5f71f00cbba2dad4cf02fb3e6da9d
Author: Chun-Tse Shao <[email protected]>
perf vendor events intel: Update alderlake events from 1.39 to 1.40
This commit updates the Alderlake performance events from version 1.39 to 1.40, incorporating changes published in the intel/perfmon repository.
Quoted 8 lines
tools/perf/pmu-events/arch/x86/alderlake/adl-metrics.json +0
-0@@ -800,16 +800,16 @@ }, { "BriefDescription": "Total pipeline cost of external Memory- or Cache-Bandwidth related bottlenecks",
[ ... ]
tools/perf/pmu-events/arch/x86/alderlake/adl-metrics.json line 807
Quoted 5 lines
"MetricGroup": "BvMB;Mem;MemoryBW;Offcore;tma_issueBW", "MetricName": "
tma_bottleneck_data_cache_memory_bandwidth", "MetricThreshold": "tma_bottleneck_data_cache_memory_bandwidth> 20",tools/perf/pmu-events/arch/x86/alderlake/adl-metrics.json +1
-1- "PublicDescription": "Total pipeline cost of external Memory- or Cache-Bandwidth related bottlenecks. Related metrics: tma_fb_full, tma_info_system_dram_bw_use, tma_mem_bandwidth, tma_sq_full", + "PublicDescription": "Total pipeline cost of external Memory- or Cache-Bandwidth related bottlenecks. Related metrics: tma_fb_full, tma_info_system_dram_bw_use, tma_mem_bandwidth, tma_sq_full, tma_uc_bound",
[Severity: Low]
Does this reference a missing metric? The metric tma_uc_bound is listed as a related metric here (and in a few other places), but it doesn't appear to be defined in the Alderlake metric files. Could this cause confusion for users viewing the metric descriptions?
Quoted 2 lines
"Unit": "cpu_core"
},Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1
PATCH 2/12 perf vendor events intel: Update alderlaken events from 1.39 to 1.40
In reply to Chun-Tse ShaoThe updated events were published in:
https://github.com/intel/perfmon/commit/7a14cc8feaf86772deb6708e96c8e9fee6d5b1ca
--- .../pmu-events/arch/x86/alderlaken/cache.json | 32 +++++++++++++++++++ tools/perf/pmu-events/arch/x86/mapfile.csv | 2 +- 2 files changed, 33 insertions(+), 1 deletion(-)
tools/perf/pmu-events/arch/x86/alderlaken/cache.json +32 -0
@@ -1,4 +1,20 @@ [ + { + "BriefDescription": "Counts the number of core requests (demand and L1 prefetchers) rejected by the L2 queue (L2Q) due to a full condition.", + "Counter": "0,1,2,3,4,5", + "EventCode": "0x31", + "EventName": "CORE_REJECT_L2Q.ANY", + "PublicDescription": "Counts the number of (demand and L1 prefetchers) core requests rejected by the L2 queue (L2Q) due to a full or nearly full condition, which likely indicates back pressure from L2Q. It also counts requests that would have gone directly to the External Queue (XQ), but are rejected due to a full or nearly full condition, indicating back pressure from the IDI link. The L2Q may also reject transactions from a core to ensure fairness between cores, or to delay a cores dirty eviction when the address conflicts incoming external snoops. (Note that L2 prefetcher requests that are dropped are not counted by this event). Counts on a per core basis.", + "SampleAfterValue": "200003" + }, + { + "BriefDescription": "Counts the number of demand and prefetch transactions that the External Queue (XQ) rejects due to a full or near full condition.", + "Counter": "0,1,2,3,4,5", + "EventCode": "0x30", + "EventName": "L2_REJECT_XQ.ANY", + "PublicDescription": "Counts the number of demand and prefetch transactions that the External Queue (XQ) rejects due to a full or near full condition which likely indicates back pressure from the IDI link. The XQ may reject transactions from the L2Q (non-cacheable requests), BBL (L2 misses) and WOB (L2 write-back victims).", + "SampleAfterValue": "200003" + }, { "BriefDescription": "Counts the total number of L2 Cache accesses. Counts on a per core basis.", "Counter": "0,1,2,3,4,5",
@@ -70,6 +86,14 @@ "SampleAfterValue": "200003", "UMask": "0x8" }, + { + "BriefDescription": "Counts the number of cycles the core is stalled due to an instruction cache or TLB miss which missed in the L2 cache.", + "Counter": "0,1,2,3,4,5", + "EventCode": "0x34", + "EventName": "MEM_BOUND_STALLS.IFETCH_L2_MISS", + "SampleAfterValue": "1000003", + "UMask": "0x30" + }, { "BriefDescription": "Counts the number of cycles the core is stalled due to an instruction cache or TLB miss which hit in the LLC or other core with HITE/F/M.", "Counter": "0,1,2,3,4,5",
@@ -103,6 +127,14 @@ "SampleAfterValue": "200003", "UMask": "0x1" }, + { + "BriefDescription": "Counts the number of cycles the core is stalled due to a demand load which missed in the L2 cache.", + "Counter": "0,1,2,3,4,5", + "EventCode": "0x34", + "EventName": "MEM_BOUND_STALLS.LOAD_L2_MISS", + "SampleAfterValue": "1000003", + "UMask": "0x6" + }, { "BriefDescription": "Counts the number of cycles the core is stalled due to a demand load which hit in the LLC or other core with HITE/F/M.", "Counter": "0,1,2,3,4,5",
tools/perf/pmu-events/arch/x86/mapfile.csv +1 -1
@@ -1,6 +1,6 @@ Family-model,Version,Filename,EventType GenuineIntel-6-(97|9A|B7|BA|BF),v1.40,alderlake,core -GenuineIntel-6-BE,v1.39,alderlaken,core +GenuineIntel-6-BE,v1.40,alderlaken,core GenuineIntel-6-C[56],v1.19,arrowlake,core GenuineIntel-6-(1C|26|27|35|36),v5,bonnell,core GenuineIntel-6-(3D|47),v30,broadwell,core
Signature · 1 line
2.55.0.795.g602f6c329a-goog
PATCH 3/12 perf vendor events intel: Update arrowlake events from 1.19 to 1.20
In reply to Chun-Tse ShaoThe updated events were published in:
https://github.com/intel/perfmon/commit/b23ebe7bc25add0c835565e4bc87e063cb620a02
--- .../arch/x86/arrowlake/arl-metrics.json | 152 ++++++++++-------- .../arch/x86/arrowlake/metricgroups.json | 2 +- .../arch/x86/arrowlake/pipeline.json | 2 +- .../arch/x86/arrowlake/uncore-memory.json | 84 ++++++++-- tools/perf/pmu-events/arch/x86/mapfile.csv | 2 +- 5 files changed, 160 insertions(+), 82 deletions(-)
tools/perf/pmu-events/arch/x86/arrowlake/arl-metrics.json +85 -67
@@ -823,16 +823,16 @@ }, { "BriefDescription": "Total pipeline cost of external Memory- or Cache-Bandwidth related bottlenecks", - "MetricExpr": "100 * (tma_memory_bound * (tma_dram_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_mem_bandwidth / (tma_mem_bandwidth + tma_mem_latency)) + tma_memory_bound * (tma_l3_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_sq_full / (tma_contested_accesses + tma_data_sharing + tma_l3_hit_latency + tma_sq_full)) + tma_memory_bound * (tma_l1_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_fb_full / (tma_dtlb_load + tma_fb_full + tma_l1_latency_capacity + tma_l1_latency_dependency + tma_lock_latency + tma_split_loads + tma_store_early_blk + tma_store_fwd_blk)))", + "MetricExpr": "100 * (tma_memory_bound * (tma_l3_miss_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_mem_bandwidth / (tma_mem_bandwidth + tma_mem_latency + tma_uc_bound)) + tma_memory_bound * (tma_l3_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_sq_full / (tma_contested_accesses + tma_data_sharing + tma_l3_hit_latency + tma_sq_full)) + tma_memory_bound * (tma_l1_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_fb_full / (tma_dtlb_load + tma_fb_full + tma_l1_latency_capacity + tma_lock_latency + tma_split_loads + tma_store_early_blk + tma_store_fwd_blk)))", "MetricGroup": "BvMB;Mem;MemoryBW;Offcore;tma_issueBW", "MetricName": "tma_bottleneck_data_cache_memory_bandwidth", "MetricThreshold": "tma_bottleneck_data_cache_memory_bandwidth > 20", - "PublicDescription": "Total pipeline cost of external Memory- or Cache-Bandwidth related bottlenecks. Related metrics: tma_fb_full, tma_info_system_dram_bw_use, tma_mem_bandwidth, tma_sq_full", + "PublicDescription": "Total pipeline cost of external Memory- or Cache-Bandwidth related bottlenecks. Related metrics: tma_fb_full, tma_info_system_dram_bw_use, tma_mem_bandwidth, tma_sq_full, tma_uc_bound", "Unit": "cpu_core" }, { "BriefDescription": "Total pipeline cost of external Memory- or Cache-Latency related bottlenecks", - "MetricExpr": "100 * (tma_memory_bound * (tma_dram_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_mem_latency / (tma_mem_bandwidth + tma_mem_latency)) + tma_memory_bound * (tma_l3_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_l3_hit_latency / (tma_contested_accesses + tma_data_sharing + tma_l3_hit_latency + tma_sq_full)) + tma_memory_bound * tma_l2_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound) + tma_memory_bound * (tma_l1_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_l1_latency_dependency / (tma_dtlb_load + tma_fb_full + tma_l1_latency_capacity + tma_l1_latency_dependency + tma_lock_latency + tma_split_loads + tma_store_early_blk + tma_store_fwd_blk)) + tma_memory_bound * (tma_l1_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_l1_latency_capacity / (tma_dtlb_load + tma_fb_full + tma_l1_latency_capacity + tma_l1_latency_dependency + tma_lock_latency + tma_split_loads + tma_store_early_blk + tma_store_fwd_blk)) + tma_memory_bound * (tma_l1_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_lock_latency / (tma_dtlb_load + tma_fb_full + tma_l1_latency_capacity + tma_l1_latency_dependency + tma_lock_latency + tma_split_loads + tma_store_early_blk + tma_store_fwd_blk)) + tma_memory_bound * (tma_l1_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_split_loads / (tma_dtlb_load + tma_fb_full + tma_l1_latency_capacity + tma_l1_latency_dependency + tma_lock_latency + tma_split_loads + tma_store_early_blk + tma_store_fwd_blk)) + tma_memory_bound * (tma_store_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_split_stores / (tma_dtlb_store + tma_false_sharing + tma_split_stores + tma_store_latency + tma_streaming_stores)) + tma_memory_bound * (tma_store_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_store_latency / (tma_dtlb_store + tma_false_sharing + tma_split_stores + tma_store_latency + tma_streaming_stores)))", + "MetricExpr": "100 * (tma_memory_bound * (tma_l3_miss_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_mem_latency / (tma_mem_bandwidth + tma_mem_latency + tma_uc_bound)) + tma_memory_bound * (tma_l3_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_l3_hit_latency / (tma_contested_accesses + tma_data_sharing + tma_l3_hit_latency + tma_sq_full)) + tma_memory_bound * tma_l2_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound) + tma_memory_bound * (tma_l1_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_l1_latency_capacity / (tma_dtlb_load + tma_fb_full + tma_l1_latency_capacity + tma_lock_latency + tma_split_loads + tma_store_early_blk + tma_store_fwd_blk)) + tma_memory_bound * (tma_l1_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_lock_latency / (tma_dtlb_load + tma_fb_full + tma_l1_latency_capacity + tma_lock_latency + tma_split_loads + tma_store_early_blk + tma_store_fwd_blk)) + tma_memory_bound * (tma_l1_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_split_loads / (tma_dtlb_load + tma_fb_full + tma_l1_latency_capacity + tma_lock_latency + tma_split_loads + tma_store_early_blk + tma_store_fwd_blk)) + tma_memory_bound * (tma_store_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_split_stores / (tma_dtlb_store + tma_false_sharing + tma_split_stores + tma_store_latency + tma_streaming_stores)) + tma_memory_bound * (tma_store_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_store_latency / (tma_dtlb_store + tma_false_sharing + tma_split_stores + tma_store_latency + tma_streaming_stores)))", "MetricGroup": "BvML;Mem;MemoryLat;Offcore;tma_issueLat", "MetricName": "tma_bottleneck_data_cache_memory_latency", "MetricThreshold": "tma_bottleneck_data_cache_memory_latency > 20",
@@ -858,7 +858,7 @@ }, { "BriefDescription": "Total pipeline cost of Memory Address Translation related bottlenecks (data-side TLBs)", - "MetricExpr": "100 * (tma_memory_bound * (tma_l1_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_dtlb_load / (tma_dtlb_load + tma_fb_full + tma_l1_latency_capacity + tma_l1_latency_dependency + tma_lock_latency + tma_split_loads + tma_store_early_blk + tma_store_fwd_blk)) + tma_memory_bound * (tma_store_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_dtlb_store / (tma_dtlb_store + tma_false_sharing + tma_split_stores + tma_store_latency + tma_streaming_stores)))", + "MetricExpr": "100 * (tma_memory_bound * (tma_l1_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_dtlb_load / (tma_dtlb_load + tma_fb_full + tma_l1_latency_capacity + tma_lock_latency + tma_split_loads + tma_store_early_blk + tma_store_fwd_blk)) + tma_memory_bound * (tma_store_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_dtlb_store / (tma_dtlb_store + tma_false_sharing + tma_split_stores + tma_store_latency + tma_streaming_stores)))", "MetricGroup": "BvMT;Mem;MemoryTLB;Offcore;tma_issueTLB", "MetricName": "tma_bottleneck_memory_data_tlbs", "MetricThreshold": "tma_bottleneck_memory_data_tlbs > 20",
@@ -867,7 +867,7 @@ }, { "BriefDescription": "Total pipeline cost of Memory Synchronization related bottlenecks (data transfers and coherency updates across processors)", - "MetricExpr": "100 * (tma_memory_bound * (tma_l3_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound) * (tma_contested_accesses + tma_data_sharing) / (tma_contested_accesses + tma_data_sharing + tma_l3_hit_latency + tma_sq_full) + tma_store_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound) * tma_false_sharing / (tma_dtlb_store + tma_false_sharing + tma_split_stores + tma_store_latency + tma_streaming_stores - tma_store_latency)) + tma_machine_clears * (1 - tma_other_nukes / tma_other_nukes))", + "MetricExpr": "100 * (tma_memory_bound * (tma_l3_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound) * (tma_contested_accesses + tma_data_sharing) / (tma_contested_accesses + tma_data_sharing + tma_l3_hit_latency + tma_sq_full) + tma_store_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound) * tma_false_sharing / (tma_dtlb_store + tma_false_sharing + tma_split_stores + tma_store_latency + tma_streaming_stores - tma_store_latency)) + tma_machine_clears * (1 - tma_other_nukes / tma_other_nukes))", "MetricGroup": "BvMS;LockCont;Mem;Offcore;tma_issueSyncxn", "MetricName": "tma_bottleneck_memory_synchronization", "MetricThreshold": "tma_bottleneck_memory_synchronization > 10",
@@ -1015,7 +1015,7 @@ }, { "BriefDescription": "This metric represents fraction of cycles the CPU was stalled due to retired misprediction by non-taken conditional branches.", - "MetricExpr": "cpu_core@BR_MISP_RETIRED.COND_NTAKEN_COST@ * cpu_core@BR_MISP_RETIRED.COND_NTAKEN_COST@R / tma_info_thread_clks", + "MetricExpr": "cpu_core@BR_MISP_RETIRED.COND_NTAKEN@ / cpu_core@BR_MISP_RETIRED.ALL_BRANCHES@ * tma_branch_mispredicts", "MetricGroup": "BrMispredicts;TopdownL3;tma_L3_group;tma_branch_mispredicts_group", "MetricName": "tma_cond_nt_mispredicts", "MetricThreshold": "tma_cond_nt_mispredicts > 0.05 & (tma_branch_mispredicts > 0.1 & tma_bad_speculation > 0.15)",
@@ -1024,7 +1024,7 @@ }, { "BriefDescription": "This metric represents fraction of cycles the CPU was stalled due to misprediction by backward-taken conditional branches.", - "MetricExpr": "cpu_core@BR_MISP_RETIRED.COND_TAKEN_BWD_COST@ * cpu_core@BR_MISP_RETIRED.COND_TAKEN_BWD_COST@R / tma_info_thread_clks", + "MetricExpr": "cpu_core@BR_MISP_RETIRED.COND_TAKEN_BWD@ / cpu_core@BR_MISP_RETIRED.ALL_BRANCHES@ * tma_branch_mispredicts", "MetricGroup": "BrMispredicts;TopdownL3;tma_L3_group;tma_branch_mispredicts_group", "MetricName": "tma_cond_tk_bwd_mispredicts", "MetricThreshold": "tma_cond_tk_bwd_mispredicts > 0.05 & (tma_branch_mispredicts > 0.1 & tma_bad_speculation > 0.15)",
@@ -1033,7 +1033,7 @@ }, { "BriefDescription": "This metric represents fraction of cycles the CPU was stalled due to misprediction by forward-taken conditional branches.", - "MetricExpr": "cpu_core@BR_MISP_RETIRED.COND_TAKEN_FWD_COST@ * cpu_core@BR_MISP_RETIRED.COND_TAKEN_FWD_COST@R / tma_info_thread_clks", + "MetricExpr": "cpu_core@BR_MISP_RETIRED.COND_TAKEN_FWD@ / cpu_core@BR_MISP_RETIRED.ALL_BRANCHES@ * tma_branch_mispredicts", "MetricGroup": "BrMispredicts;TopdownL3;tma_L3_group;tma_branch_mispredicts_group", "MetricName": "tma_cond_tk_fwd_mispredicts", "MetricThreshold": "tma_cond_tk_fwd_mispredicts > 0.05 & (tma_branch_mispredicts > 0.1 & tma_bad_speculation > 0.15)",
@@ -1046,7 +1046,7 @@ "MetricGroup": "BvMS;DataSharing;LockCont;Offcore;Snoop;TopdownL4;tma_L4_group;tma_issueSyncxn;tma_l3_bound_group", "MetricName": "tma_contested_accesses", "MetricThreshold": "tma_contested_accesses > 0.05 & (tma_l3_bound > 0.05 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))", - "PublicDescription": "This metric estimates fraction of cycles while the memory subsystem was handling synchronizations due to contested accesses. Contested accesses occur when data written by one Logical Processor are read by another Logical Processor on a different Physical Core. Examples of contested accesses include synchronizations such as locks; true data sharing such as modified locked variables; and false sharing. Sample with: MEM_LOAD_L3_HIT_RETIRED.XSNP_FWD;MEM_LOAD_L3_HIT_RETIRED.XSNP_MISS. Related metrics: tma_bottleneck_memory_synchronization, tma_data_sharing, tma_false_sharing, tma_machine_clears, tma_remote_cache", + "PublicDescription": "This metric estimates fraction of cycles while the memory subsystem was handling synchronizations due to contested accesses. Contested accesses occur when data written by one Logical Processor are read by another Logical Processor on a different Physical Core. Examples of contested accesses include synchronizations such as locks; true data sharing such as modified locked variables; and false sharing. Sample with: MEM_LOAD_L2_MISS_RETIRED.L3.SAME_CBB.HIT_ESF.SNP_MISS;MEM_LOAD_L2_MISS_RETIRED.OTHER_MODULE_L2.SAME_CBB.HIT_M.SNP_FWD. Related metrics: tma_bottleneck_memory_synchronization, tma_data_sharing, tma_false_sharing, tma_machine_clears, tma_remote_cache", "ScaleUnit": "100%", "Unit": "cpu_core" },
@@ -1062,13 +1062,13 @@ "Unit": "cpu_core" }, { - "BriefDescription": "This metric estimates fraction of cycles while the memory subsystem was handling synchronizations due to data-sharing accesses", + "BriefDescription": "This metric estimates fraction of cycles while the memory subsystem was handling synchronizations due to L3 data-sharing accesses", "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "(cpu_core@MEM_LOAD_L3_HIT_RETIRED.XSNP_NO_FWD@ * min(cpu_core@MEM_LOAD_L3_HIT_RETIRED.XSNP_NO_FWD@R, 24 * tma_info_system_core_frequency) + cpu_core@MEM_LOAD_L3_HIT_RETIRED.XSNP_FWD@ * min(cpu_core@MEM_LOAD_L3_HIT_RETIRED.XSNP_FWD@R, 25 * tma_info_system_core_frequency)) * (1 + cpu_core@MEM_LOAD_RETIRED.FB_HIT@ / cpu_core@MEM_LOAD_RETIRED.L1_MISS@ / 2) / tma_info_thread_clks", "MetricGroup": "BvMS;Offcore;Snoop;TopdownL4;tma_L4_group;tma_issueSyncxn;tma_l3_bound_group", "MetricName": "tma_data_sharing", "MetricThreshold": "tma_data_sharing > 0.05 & (tma_l3_bound > 0.05 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))", - "PublicDescription": "This metric estimates fraction of cycles while the memory subsystem was handling synchronizations due to data-sharing accesses. Data shared by multiple Logical Processors (even just read shared) may cause increased access latency due to cache coherency. Excessive data sharing can drastically harm multithreaded performance. Sample with: MEM_LOAD_L3_HIT_RETIRED.XSNP_NO_FWD. Related metrics: tma_bottleneck_memory_synchronization, tma_contested_accesses, tma_false_sharing, tma_machine_clears, tma_remote_cache", + "PublicDescription": "This metric estimates fraction of cycles while the memory subsystem was handling synchronizations due to L3 data-sharing accesses. Data shared by multiple Logical Processors (even just read shared) may cause increased access latency due to cache coherency. Excessive data sharing can drastically harm multithreaded performance. Sample with: MEM_LOAD_L2_MISS_RETIRED.L3.SAME_CBB.HIT_ESF.SNP_HIT_NO_FWD. Related metrics: tma_bottleneck_memory_synchronization, tma_contested_accesses, tma_false_sharing, tma_machine_clears, tma_remote_cache", "ScaleUnit": "100%", "Unit": "cpu_core" },
@@ -1082,19 +1082,9 @@ "ScaleUnit": "100%", "Unit": "cpu_core" }, - { - "BriefDescription": "This metric estimates how often the CPU was stalled on accesses to external memory (DRAM) by loads", - "MetricExpr": "cpu_core@MEMORY_STALLS.MEM@ / tma_info_thread_clks", - "MetricGroup": "MemoryBound;TmaL3mem;TopdownL3;tma_L3_group;tma_memory_bound_group", - "MetricName": "tma_dram_bound", - "MetricThreshold": "tma_dram_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2)", - "PublicDescription": "This metric estimates how often the CPU was stalled on accesses to external memory (DRAM) by loads. Better caching can improve the latency and increase performance. Sample with: MEM_LOAD_RETIRED.L3_MISS", - "ScaleUnit": "100%", - "Unit": "cpu_core" - }, { "BriefDescription": "This metric represents Core fraction of cycles in which CPU was likely limited due to DSB (decoded uop cache) fetch pipeline", - "MetricExpr": "([email protected]_UOPS\\,cmask\\=0x8\\,inv\\=0x1@ / 2 + [email protected]_UOPS@ / ([email protected]_UOPS@ + [email protected]_UOPS@) * (cpu_core@IDQ_BUBBLES.STARVATION_CYCLES@ - cpu_core@IDQ_BUBBLES.FETCH_LATENCY@)) / tma_info_thread_clks", + "MetricExpr": "([email protected]_UOPS\\,cmask\\=0x8\\,inv\\=0x1@ * max(0, 8 - tma_info_pipeline_fetch_dsb) / 8 + [email protected]_UOPS@ / ([email protected]_UOPS@ + [email protected]_UOPS@) * (cpu_core@IDQ_BUBBLES.STARVATION_CYCLES@ - cpu_core@IDQ_BUBBLES.FETCH_LATENCY@)) / tma_info_thread_clks", "MetricGroup": "DSB;FetchBW;TopdownL3;tma_L3_group;tma_fetch_bandwidth_group", "MetricName": "tma_dsb", "MetricThreshold": "tma_dsb > 0.15 & tma_fetch_bandwidth > 0.2",
@@ -1138,7 +1128,7 @@ "MetricGroup": "BvMS;DataSharing;LockCont;Offcore;Snoop;TopdownL4;tma_L4_group;tma_issueSyncxn;tma_store_bound_group", "MetricName": "tma_false_sharing", "MetricThreshold": "tma_false_sharing > 0.05 & (tma_store_bound > 0.2 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))", - "PublicDescription": "This metric roughly estimates how often CPU was handling synchronizations due to False Sharing. False Sharing is a multithreading hiccup; where multiple Logical Processors contend on different data-elements mapped into the same cache line. Sample with: OCR.DEMAND_RFO.L3_HIT.SNOOP_HITM. Related metrics: tma_bottleneck_memory_synchronization, tma_contested_accesses, tma_data_sharing, tma_machine_clears, tma_remote_cache", + "PublicDescription": "This metric roughly estimates how often CPU was handling synchronizations due to False Sharing. False Sharing is a multithreading hiccup; where multiple Logical Processors contend on different data-elements mapped into the same cache line. Sample with: L2_REQUEST.RFO_HIT_XSNP_HIT_FWD. Related metrics: tma_bottleneck_memory_synchronization, tma_contested_accesses, tma_data_sharing, tma_machine_clears, tma_remote_cache", "ScaleUnit": "100%", "Unit": "cpu_core" },
@@ -1148,7 +1138,7 @@ "MetricGroup": "BvMB;MemoryBW;TopdownL4;tma_L4_group;tma_issueBW;tma_issueSL;tma_issueSmSt;tma_l1_bound_group", "MetricName": "tma_fb_full", "MetricThreshold": "tma_fb_full > 0.3", - "PublicDescription": "This metric does a *rough estimation* of how often L1D Fill Buffer unavailability limited additional L1D miss memory access requests to proceed. The higher the metric value; the deeper the memory hierarchy level the misses are satisfied from (metric values >1 are valid). Often it hints on approaching bandwidth limits (to L2 cache; L3 cache or external memory). Related metrics: tma_bottleneck_data_cache_memory_bandwidth, tma_info_system_dram_bw_use, tma_mem_bandwidth, tma_sq_full, tma_store_latency, tma_streaming_stores", + "PublicDescription": "This metric does a *rough estimation* of how often L1D Fill Buffer unavailability limited additional L1D miss memory access requests to proceed. The higher the metric value; the deeper the memory hierarchy level the misses are satisfied from (metric values >1 are valid). Often it hints on approaching bandwidth limits (to L2 cache; L3 cache or external memory). Related metrics: tma_bottleneck_data_cache_memory_bandwidth, tma_info_system_dram_bw_use, tma_mem_bandwidth, tma_sq_full, tma_store_latency, tma_streaming_stores, tma_uc_bound", "ScaleUnit": "100%", "Unit": "cpu_core" },
@@ -1245,7 +1235,7 @@ }, { "BriefDescription": "This metric approximates arithmetic FP vector uops fraction the CPU has retired for 256-bit wide vectors", - "MetricExpr": "cpu_core@FP_ARITH_OPS_RETIRED.VECTOR\\,umask\\=0x30@ / (tma_retiring * tma_info_thread_slots)", + "MetricExpr": "cpu_core@FP_ARITH_OPS_RETIRED.VECTOR_256B@ / (tma_retiring * tma_info_thread_slots)", "MetricGroup": "Compute;Flops;TopdownL5;tma_L5_group;tma_fp_vector_group;tma_issue2P", "MetricName": "tma_fp_vector_256b", "MetricThreshold": "tma_fp_vector_256b > 0.1 & (tma_fp_vector > 0.1 & (tma_fp_arith > 0.2 & tma_light_operations > 0.6))",
@@ -1288,7 +1278,7 @@ }, { "BriefDescription": "This metric represents fraction of cycles the CPU was stalled due to instruction cache misses", - "MetricExpr": "cpu_core@ICACHE_DATA.STALLS@ / tma_info_thread_clks", + "MetricExpr": "cpu_core@ICACHE_DATA.STALLS@ / tma_info_thread_clks + tma_isb_full", "MetricGroup": "BigFootprint;BvBC;FetchLat;IcMiss;TopdownL3;tma_L3_group;tma_fetch_latency_group", "MetricName": "tma_icache_misses", "MetricThreshold": "tma_icache_misses > 0.05 & (tma_fetch_latency > 0.1 & tma_frontend_bound > 0.15)",
@@ -1298,7 +1288,7 @@ }, { "BriefDescription": "This metric represents fraction of cycles the CPU was stalled due to retired misprediction by indirect CALL instructions.", - "MetricExpr": "cpu_core@BR_MISP_RETIRED.INDIRECT_CALL_COST@ * cpu_core@BR_MISP_RETIRED.INDIRECT_CALL_COST@R / tma_info_thread_clks", + "MetricExpr": "cpu_core@BR_MISP_RETIRED.INDIRECT_CALL@ / cpu_core@BR_MISP_RETIRED.ALL_BRANCHES@ * tma_branch_mispredicts", "MetricGroup": "BrMispredicts;TopdownL3;tma_L3_group;tma_branch_mispredicts_group", "MetricName": "tma_ind_call_mispredicts", "MetricThreshold": "tma_ind_call_mispredicts > 0.05 & (tma_branch_mispredicts > 0.1 & tma_bad_speculation > 0.15)",
@@ -1307,7 +1297,7 @@ }, { "BriefDescription": "This metric represents fraction of cycles the CPU was stalled due to retired misprediction by indirect JMP instructions.", - "MetricExpr": "max((cpu_core@BR_MISP_RETIRED.INDIRECT_COST@ * cpu_core@BR_MISP_RETIRED.INDIRECT_COST@R - cpu_core@BR_MISP_RETIRED.INDIRECT_CALL_COST@ * cpu_core@BR_MISP_RETIRED.INDIRECT_CALL_COST@R) / tma_info_thread_clks, 0)", + "MetricExpr": "max((cpu_core@BR_MISP_RETIRED.INDIRECT@ - cpu_core@BR_MISP_RETIRED.INDIRECT_CALL@) / cpu_core@BR_MISP_RETIRED.ALL_BRANCHES@ * tma_branch_mispredicts, 0)", "MetricGroup": "BrMispredicts;TopdownL3;tma_L3_group;tma_branch_mispredicts_group", "MetricName": "tma_ind_jump_mispredicts", "MetricThreshold": "tma_ind_jump_mispredicts > 0.05 & (tma_branch_mispredicts > 0.1 & tma_bad_speculation > 0.15)",
@@ -1477,7 +1467,7 @@ }, { "BriefDescription": "Fraction of Uops delivered by the DSB (aka Decoded ICache; or Uop Cache)", - "MetricExpr": "[email protected]_UOPS@ / cpu_core@UOPS_ISSUED.ANY@", + "MetricExpr": "[email protected]_UOPS@ / ([email protected]_UOPS@ + [email protected]@ + [email protected]_UOPS@ + [email protected]_UOPS@)", "MetricGroup": "DSB;Fed;FetchBW;tma_issueFB", "MetricName": "tma_info_frontend_dsb_coverage", "MetricThreshold": "tma_info_frontend_dsb_coverage < 0.7 & tma_info_thread_ipc / 8 > 0.35",
@@ -1544,7 +1534,7 @@ }, { "BriefDescription": "Fraction of Uops delivered by the LSD (Loop Stream Detector; aka Loop Cache)", - "MetricExpr": "[email protected]@ / cpu_core@UOPS_ISSUED.ANY@", + "MetricExpr": "[email protected]@ / ([email protected]_UOPS@ + [email protected]@ + [email protected]_UOPS@ + [email protected]_UOPS@)", "MetricGroup": "Fed;LSD", "MetricName": "tma_info_frontend_lsd_coverage", "Unit": "cpu_core"
@@ -1703,6 +1693,15 @@ "PublicDescription": "Instructions per taken branch. Related metrics: tma_dsb_switches, tma_fetch_bandwidth, tma_info_botlnk_l2_dsb_bandwidth, tma_info_botlnk_l2_dsb_misses, tma_info_frontend_dsb_coverage, tma_lcp", "Unit": "cpu_core" }, + { + "BriefDescription": "AVX preserve/restore assists per kilo instruction", + "MetricExpr": "1e3 * [email protected]_AVX_MIX@ / cpu_core@INST_RETIRED.ANY@", + "MetricGroup": "tma_issueMV", + "MetricName": "tma_info_inst_mix_vectormixpki", + "MetricThreshold": "tma_info_inst_mix_vectormixpki > 0.05", + "PublicDescription": "AVX preserve/restore assists per kilo instruction. Related metrics: tma_mixing_vectors, tma_ms_switches", + "Unit": "cpu_core" + }, { "BriefDescription": "Fill Buffer (FB) hits per kilo instructions for retired demand loads (L1D misses that merge into ongoing miss-handling entries)", "MetricExpr": "1e3 * cpu_core@MEM_LOAD_RETIRED.FB_HIT@ / cpu_core@INST_RETIRED.ANY@",
@@ -1969,6 +1968,13 @@ "MetricName": "tma_info_pipeline_retire", "Unit": "cpu_core" }, + { + "BriefDescription": "Total percent of cycles without actually retired uops", + "MetricExpr": "cpu_core@UOPS_RETIRED.STALLS@ / tma_info_thread_clks", + "MetricGroup": "Pipeline;Ret", + "MetricName": "tma_info_pipeline_retire_stall", + "Unit": "cpu_core" + }, { "BriefDescription": "Estimated fraction of retirement-cycles dealing with repeat instructions", "MetricExpr": "cpu_core@INST_RETIRED.REP_ITERATION@ / cpu_core@UOPS_RETIRED.SLOTS\\,cmask\\=1@",
@@ -2011,7 +2017,7 @@ "MetricExpr": "32 * UNC_M_TOTAL_DATA / 1e9 / tma_info_system_time", "MetricGroup": "HPC;MemOffcore;MemoryBW;SoC;tma_issueBW", "MetricName": "tma_info_system_dram_bw_use", - "PublicDescription": "Average external Memory Bandwidth Use for reads and writes [GB / sec]. Related metrics: tma_bottleneck_data_cache_memory_bandwidth, tma_fb_full, tma_mem_bandwidth, tma_sq_full", + "PublicDescription": "Average external Memory Bandwidth Use for reads and writes [GB / sec]. Related metrics: tma_bottleneck_data_cache_memory_bandwidth, tma_fb_full, tma_mem_bandwidth, tma_sq_full, tma_uc_bound", "Unit": "cpu_core" }, {
@@ -2082,13 +2088,6 @@ "MetricName": "tma_info_system_turbo_utilization", "Unit": "cpu_core" }, - { - "BriefDescription": "Measured Average Uncore Frequency for the SoC [GHz]", - "MetricExpr": "tma_info_system_socket_clks / 1e9 / tma_info_system_time", - "MetricGroup": "SoC", - "MetricName": "tma_info_system_uncore_frequency", - "Unit": "cpu_core" - }, { "BriefDescription": "Per-Logical Processor actual clocks when the Logical Processor is active.", "MetricExpr": "cpu_core@CPU_CLK_UNHALTED.THREAD@",
@@ -2180,9 +2179,18 @@ "ScaleUnit": "100%", "Unit": "cpu_core" }, + { + "BriefDescription": "This metric represents fraction of cycles Icache tag lookup was stalled due to Instruction Stream Buffer (ISB) being full", + "MetricExpr": "(cpu_core@ICACHE_TAG.STALLS_ISB@ + cpu_core@ICACHE_TAG.STALLS_INUSE@) / tma_info_thread_clks", + "MetricGroup": "TopdownL4;tma_L4_group;tma_icache_misses_group", + "MetricName": "tma_isb_full", + "MetricThreshold": "tma_isb_full > 0.05 & (tma_icache_misses > 0.05 & (tma_fetch_latency > 0.1 & tma_frontend_bound > 0.15))", + "ScaleUnit": "100%", + "Unit": "cpu_core" + }, { "BriefDescription": "This metric represents fraction of cycles the CPU was stalled due to Instruction TLB (ITLB) misses", - "MetricExpr": "cpu_core@ICACHE_TAG.STALLS@ / tma_info_thread_clks", + "MetricExpr": "cpu_core@ICACHE_TAG.STALLS@ / tma_info_thread_clks - tma_isb_full", "MetricGroup": "BigFootprint;BvBC;FetchLat;MemoryTLB;TopdownL3;tma_L3_group;tma_fetch_latency_group", "MetricName": "tma_itlb_misses", "MetricThreshold": "tma_itlb_misses > 0.05 & (tma_fetch_latency > 0.1 & tma_frontend_bound > 0.15)",
@@ -2209,19 +2217,9 @@ "ScaleUnit": "100%", "Unit": "cpu_core" }, - { - "BriefDescription": "This metric ([SKL+] roughly; [LNL]) estimates fraction of cycles with demand load accesses that hit the L1D cache", - "MetricExpr": "4 * cpu_core@DEPENDENT_LOADS.ANY\\,cmask\\=1@ / tma_info_thread_clks", - "MetricGroup": "BvML;MemoryLat;TopdownL4;tma_L4_group;tma_l1_bound_group", - "MetricName": "tma_l1_latency_dependency", - "MetricThreshold": "tma_l1_latency_dependency > 0.1 & (tma_l1_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))", - "PublicDescription": "This metric ([SKL+] roughly; [LNL]) estimates fraction of cycles with demand load accesses that hit the L1D cache. The short latency of the L1D cache may be exposed in pointer-chasing memory access patterns as an example. Sample with: MEM_LOAD_RETIRED.L1_HIT", - "ScaleUnit": "100%", - "Unit": "cpu_core" - }, { "BriefDescription": "This metric estimates how often the CPU was stalled due to L2 cache accesses by loads", - "MetricExpr": "cpu_core@MEMORY_STALLS.L2@ / tma_info_thread_clks", + "MetricExpr": "cpu_core@MEM_LOAD_RETIRED.L2_HIT@ / (cpu_core@MEM_LOAD_RETIRED.L2_HIT@ + cpu_core@MEM_LOAD_MISC_RETIRED.UC@) * cpu_core@MEMORY_STALLS.L2@ / tma_info_thread_clks", "MetricGroup": "BvML;CacheHits;MemoryBound;TmaL3mem;TopdownL3;tma_L3_group;tma_memory_bound_group", "MetricName": "tma_l2_bound", "MetricThreshold": "tma_l2_bound > 0.05 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2)",
@@ -2245,7 +2243,7 @@ "MetricGroup": "CacheHits;MemoryBound;TmaL3mem;TopdownL3;tma_L3_group;tma_memory_bound_group", "MetricName": "tma_l3_bound", "MetricThreshold": "tma_l3_bound > 0.05 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2)", - "PublicDescription": "This metric estimates how often the CPU was stalled due to loads accesses to L3 cache or contended with a sibling Core. Avoiding cache misses (i.e. L2 misses/L3 hits) can improve the latency and increase performance. Sample with: MEM_LOAD_RETIRED.L3_HIT_PS", + "PublicDescription": "This metric estimates how often the CPU was stalled due to loads accesses to L3 cache or contended with a sibling Core. Avoiding cache misses (i.e. L2 misses/L3 hits) can improve the latency and increase performance. Sample with: MEM_LOAD_L2_MISS_RETIRED.L3_HIT_SAME_CBB", "ScaleUnit": "100%", "Unit": "cpu_core" },
@@ -2255,7 +2253,17 @@ "MetricGroup": "BvML;MemoryLat;TopdownL4;tma_L4_group;tma_issueLat;tma_l3_bound_group", "MetricName": "tma_l3_hit_latency", "MetricThreshold": "tma_l3_hit_latency > 0.1 & (tma_l3_bound > 0.05 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))", - "PublicDescription": "This metric estimates fraction of cycles with demand load accesses that hit the L3 cache under unloaded scenarios (possibly L3 latency limited). Avoiding private cache misses (i.e. L2 misses/L3 hits) will improve the latency; reduce contention with sibling physical cores and increase performance. Note the value of this node may overlap with its siblings. Sample with: MEM_LOAD_RETIRED.L3_HIT_PS. Related metrics: tma_bottleneck_data_cache_memory_latency, tma_mem_latency", + "PublicDescription": "This metric estimates fraction of cycles with demand load accesses that hit the L3 cache under unloaded scenarios (possibly L3 latency limited). Avoiding private cache misses (i.e. L2 misses/L3 hits) will improve the latency; reduce contention with sibling physical cores and increase performance. Note the value of this node may overlap with its siblings. Sample with: MEM_LOAD_L2_MISS_RETIRED.L3_HIT_SAME_CBB. Related metrics: tma_bottleneck_data_cache_memory_latency, tma_mem_latency", + "ScaleUnit": "100%", + "Unit": "cpu_core" + }, + { + "BriefDescription": "This metric estimates how often the CPU was stalled on accesses to external memory (DRAM) by loads", + "MetricExpr": "cpu_core@MEMORY_STALLS.MEM@ / tma_info_thread_clks", + "MetricGroup": "MemoryBound;Offcore;TmaL3mem;TopdownL3;tma_L3_group;tma_memory_bound_group", + "MetricName": "tma_l3_miss_bound", + "MetricThreshold": "tma_l3_miss_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2)", + "PublicDescription": "This metric estimates how often the CPU was stalled on accesses to external memory (DRAM) by loads. Better caching can improve the latency and increase performance. Sample with: MEM_LOAD_L2_MISS_RETIRED.L3_MISS", "ScaleUnit": "100%", "Unit": "cpu_core" },
@@ -2348,7 +2356,7 @@ }, { "BriefDescription": "This metric represents Core fraction of cycles in which CPU was likely limited due to LSD (Loop Stream Detector) unit", - "MetricExpr": "[email protected]\\,cmask\\=0x8\\,inv\\=0x1@ / tma_info_thread_clks / 2", + "MetricExpr": "[email protected]\\,cmask\\=0x8\\,inv\\=0x1@ * max(0, 8 - tma_info_pipeline_fetch_lsd) / 8 / tma_info_thread_clks", "MetricGroup": "FetchBW;LSD;TopdownL3;tma_L3_group;tma_fetch_bandwidth_group", "MetricName": "tma_lsd", "MetricThreshold": "tma_lsd > 0.15 & tma_fetch_bandwidth > 0.2",
@@ -2370,19 +2378,19 @@ { "BriefDescription": "This metric estimates fraction of cycles where the core's performance was likely hurt due to approaching bandwidth limits of external memory - DRAM ([SPR-HBM] and/or HBM)", "MetricExpr": "min(cpu_core@CPU_CLK_UNHALTED.THREAD@, cpu_core@OFFCORE_REQUESTS_OUTSTANDING.DATA_RD\\,cmask\\=4@) / tma_info_thread_clks", - "MetricGroup": "BvMB;MemoryBW;Offcore;TopdownL4;tma_L4_group;tma_dram_bound_group;tma_issueBW", + "MetricGroup": "BvMB;MemoryBW;Offcore;TopdownL4;tma_L4_group;tma_issueBW;tma_l3_miss_bound_group", "MetricName": "tma_mem_bandwidth", - "MetricThreshold": "tma_mem_bandwidth > 0.2 & (tma_dram_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))", - "PublicDescription": "This metric estimates fraction of cycles where the core's performance was likely hurt due to approaching bandwidth limits of external memory - DRAM ([SPR-HBM] and/or HBM). The underlying heuristic assumes that a similar off-core traffic is generated by all IA cores. This metric does not aggregate non-data-read requests by this logical processor; requests from other IA Logical Processors/Physical Cores/sockets; or other non-IA devices like GPU; hence the maximum external memory bandwidth limits may or may not be approached when this metric is flagged (see Uncore counters for that). Related metrics: tma_bottleneck_data_cache_memory_bandwidth, tma_fb_full, tma_info_system_dram_bw_use, tma_sq_full", + "MetricThreshold": "tma_mem_bandwidth > 0.2 & (tma_l3_miss_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))", + "PublicDescription": "This metric estimates fraction of cycles where the core's performance was likely hurt due to approaching bandwidth limits of external memory - DRAM ([SPR-HBM] and/or HBM). The underlying heuristic assumes that a similar off-core traffic is generated by all IA cores. This metric does not aggregate non-data-read requests by this logical processor; requests from other IA Logical Processors/Physical Cores/sockets; or other non-IA devices like GPU; hence the maximum external memory bandwidth limits may or may not be approached when this metric is flagged (see Uncore counters for that). Related metrics: tma_bottleneck_data_cache_memory_bandwidth, tma_fb_full, tma_info_system_dram_bw_use, tma_sq_full, tma_uc_bound", "ScaleUnit": "100%", "Unit": "cpu_core" }, { "BriefDescription": "This metric estimates fraction of cycles where the performance was likely hurt due to latency from external memory - DRAM ([SPR-HBM] and/or HBM)", "MetricExpr": "min(cpu_core@CPU_CLK_UNHALTED.THREAD@, cpu_core@OFFCORE_REQUESTS_OUTSTANDING.CYCLES_WITH_DATA_RD@) / tma_info_thread_clks - tma_mem_bandwidth", - "MetricGroup": "BvML;MemoryLat;Offcore;TopdownL4;tma_L4_group;tma_dram_bound_group;tma_issueLat", + "MetricGroup": "BvML;MemoryLat;Offcore;TopdownL4;tma_L4_group;tma_issueLat;tma_l3_miss_bound_group", "MetricName": "tma_mem_latency", - "MetricThreshold": "tma_mem_latency > 0.1 & (tma_dram_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))", + "MetricThreshold": "tma_mem_latency > 0.1 & (tma_l3_miss_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))", "PublicDescription": "This metric estimates fraction of cycles where the performance was likely hurt due to latency from external memory - DRAM ([SPR-HBM] and/or HBM). This metric does not aggregate requests from other Logical Processors/Physical Cores/sockets (see Uncore counters for that). Related metrics: tma_bottleneck_data_cache_memory_latency, tma_l3_hit_latency", "ScaleUnit": "100%", "Unit": "cpu_core"
@@ -2438,7 +2446,7 @@ }, { "BriefDescription": "This metric represents Core fraction of cycles in which CPU was likely limited due to the MITE pipeline (the legacy decode pipeline)", - "MetricExpr": "([email protected]_UOPS\\,cmask\\=0x8\\,inv\\=0x1@ / 2 + [email protected]_UOPS@ / ([email protected]_UOPS@ + [email protected]_UOPS@) * (cpu_core@IDQ_BUBBLES.STARVATION_CYCLES@ - cpu_core@IDQ_BUBBLES.FETCH_LATENCY@)) / tma_info_thread_clks", + "MetricExpr": "([email protected]_UOPS\\,cmask\\=0x8\\,inv\\=0x1@ * max(0, 8 - tma_info_pipeline_fetch_mite) / 8 + [email protected]_UOPS@ / ([email protected]_UOPS@ + [email protected]_UOPS@) * (cpu_core@IDQ_BUBBLES.STARVATION_CYCLES@ - cpu_core@IDQ_BUBBLES.FETCH_LATENCY@)) / tma_info_thread_clks", "MetricGroup": "DSBmiss;FetchBW;TopdownL3;tma_L3_group;tma_fetch_bandwidth_group", "MetricName": "tma_mite", "MetricThreshold": "tma_mite > 0.1 & tma_fetch_bandwidth > 0.2",
@@ -2452,13 +2460,13 @@ "MetricGroup": "TopdownL5;tma_L5_group;tma_issueMV;tma_ports_utilized_0_group", "MetricName": "tma_mixing_vectors", "MetricThreshold": "tma_mixing_vectors > 0.05", - "PublicDescription": "This metric estimates penalty in terms of percentage of([SKL+] injected blend uops out of all Uops Issued -- the Count Domain; [ADL+] cycles). Usually a Mixing_Vectors over 5% is worth investigating. Read more in Appendix B1 of the Optimizations Guide for this topic. Related metrics: tma_ms_switches", + "PublicDescription": "This metric estimates penalty in terms of percentage of([SKL+] injected blend uops out of all Uops Issued -- the Count Domain; [ADL+] cycles). Usually a Mixing_Vectors over 5% is worth investigating. Read more in Appendix B1 of the Optimizations Guide for this topic. Related metrics: tma_info_inst_mix_vectormixpki, tma_ms_switches", "ScaleUnit": "100%", "Unit": "cpu_core" }, { "BriefDescription": "This metric represents Core fraction of cycles in which CPU was likely limited due to the Microcode Sequencer (MS) unit - see Microcode_Sequencer node for details.", - "MetricExpr": "[email protected]_CYCLES_ANY@ / tma_info_thread_clks / 1.8", + "MetricExpr": "[email protected]_CYCLES_ANY@ * (8 - [email protected]_UOPS@ / [email protected]_CYCLES_ANY@) / tma_info_thread_slots", "MetricGroup": "MicroSeq;TopdownL3;tma_L3_group;tma_fetch_bandwidth_group", "MetricName": "tma_ms", "MetricThreshold": "tma_ms > 0.05 & tma_fetch_bandwidth > 0.2",
@@ -2471,7 +2479,7 @@ "MetricGroup": "FetchLat;MicroSeq;TopdownL3;tma_L3_group;tma_fetch_latency_group;tma_issueMC;tma_issueMS;tma_issueMV;tma_issueSO", "MetricName": "tma_ms_switches", "MetricThreshold": "tma_ms_switches > 0.05 & (tma_fetch_latency > 0.1 & tma_frontend_bound > 0.15)", - "PublicDescription": "This metric estimates the fraction of cycles when the CPU was stalled due to switches of uop delivery to the Microcode Sequencer (MS). Commonly used instructions are optimized for delivery by the DSB (decoded i-cache) or MITE (legacy instruction decode) pipelines. Certain operations cannot be handled natively by the execution pipeline; and must be performed by microcode (small programs injected into the execution stream). Switching to the MS too often can negatively impact performance. The MS is designated to deliver long uop flows required by CISC instructions like CPUID; or uncommon conditions like Floating Point Assists when dealing with Denormals. Sample with: IDQ.MS_SWITCHES. Related metrics: tma_bottleneck_irregular_overhead, tma_clears_resteers, tma_l1_bound, tma_machine_clears, tma_microcode_sequencer, tma_mixing_vectors, tma_serializing_operation", + "PublicDescription": "This metric estimates the fraction of cycles when the CPU was stalled due to switches of uop delivery to the Microcode Sequencer (MS). Commonly used instructions are optimized for delivery by the DSB (decoded i-cache) or MITE (legacy instruction decode) pipelines. Certain operations cannot be handled natively by the execution pipeline; and must be performed by microcode (small programs injected into the execution stream). Switching to the MS too often can negatively impact performance. The MS is designated to deliver long uop flows required by CISC instructions like CPUID; or uncommon conditions like Floating Point Assists when dealing with Denormals. Sample with: IDQ.MS_SWITCHES. Related metrics: tma_bottleneck_irregular_overhead, tma_clears_resteers, tma_info_inst_mix_vectormixpki, tma_l1_bound, tma_machine_clears, tma_microcode_sequencer, tma_mixing_vectors, tma_serializing_operation", "ScaleUnit": "100%", "Unit": "cpu_core" },
@@ -2589,7 +2597,7 @@ }, { "BriefDescription": "This metric represents fraction of cycles the CPU was stalled due to retired misprediction by (indirect) RET instructions.", - "MetricExpr": "cpu_core@BR_MISP_RETIRED.RET_COST@ * cpu_core@BR_MISP_RETIRED.RET_COST@R / tma_info_thread_clks", + "MetricExpr": "cpu_core@BR_MISP_RETIRED.RET@ / cpu_core@BR_MISP_RETIRED.ALL_BRANCHES@ * tma_branch_mispredicts", "MetricGroup": "BrMispredicts;TopdownL3;tma_L3_group;tma_branch_mispredicts_group", "MetricName": "tma_ret_mispredicts", "MetricThreshold": "tma_ret_mispredicts > 0.05 & (tma_branch_mispredicts > 0.1 & tma_bad_speculation > 0.15)",
@@ -2664,7 +2672,7 @@ "MetricGroup": "BvMB;MemoryBW;Offcore;TopdownL4;tma_L4_group;tma_issueBW;tma_l3_bound_group", "MetricName": "tma_sq_full", "MetricThreshold": "tma_sq_full > 0.3 & (tma_l3_bound > 0.05 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))", - "PublicDescription": "This metric measures fraction of cycles where the Super Queue (SQ) was full taking into account all request-types and both hardware SMT threads (Logical Processors). Related metrics: tma_bottleneck_data_cache_memory_bandwidth, tma_fb_full, tma_info_system_dram_bw_use, tma_mem_bandwidth", + "PublicDescription": "This metric measures fraction of cycles where the Super Queue (SQ) was full taking into account all request-types and both hardware SMT threads (Logical Processors). Related metrics: tma_bottleneck_data_cache_memory_bandwidth, tma_fb_full, tma_info_system_dram_bw_use, tma_mem_bandwidth, tma_uc_bound", "ScaleUnit": "100%", "Unit": "cpu_core" },
@@ -2679,8 +2687,8 @@ "Unit": "cpu_core" }, { - "BriefDescription": "This metric estimates clocks wasted due to loads blocked due to unknown store address (did not do memory disambiguation) or due to unknown store data", - "MetricExpr": "7 * cpu_core@LD_BLOCKS.STORE_EARLY\\,cmask\\=1@ / tma_info_thread_clks", + "BriefDescription": "This metric roughly estimates clocks wasted due to loads blocked due to unknown store address (did not do memory disambiguation) or due to unknown store data", + "MetricExpr": "cpu_core@LD_BLOCKS.STORE_EARLY\\,cmask\\=1@ / tma_info_thread_clks", "MetricGroup": "TopdownL4;tma_L4_group;tma_l1_bound_group", "MetricName": "tma_store_early_blk", "MetricThreshold": "tma_store_early_blk > 0.2",
@@ -2768,7 +2776,17 @@ "MetricGroup": "MemoryBW;Offcore;TopdownL4;tma_L4_group;tma_issueSmSt;tma_store_bound_group", "MetricName": "tma_streaming_stores", "MetricThreshold": "tma_streaming_stores > 0.2 & (tma_store_bound > 0.2 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))", - "PublicDescription": "This metric estimates how often CPU was stalled due to Streaming store memory accesses; Streaming store optimize out a read request required by RFO stores. Even though store accesses do not typically stall out-of-order CPUs; there are few cases where stores can lead to actual stalls. This metric will be flagged should Streaming stores be a bottleneck. Sample with: OCR.STREAMING_WR.ANY_RESPONSE. Related metrics: tma_fb_full", + "PublicDescription": "This metric estimates how often CPU was stalled due to Streaming store memory accesses; Streaming store optimize out a read request required by RFO stores. Even though store accesses do not typically stall out-of-order CPUs; there are few cases where stores can lead to actual stalls. This metric will be flagged should Streaming stores be a bottleneck. Sample with: OMR.STREAMING_WR.ANY_RESPONSE. Related metrics: tma_fb_full", + "ScaleUnit": "100%", + "Unit": "cpu_core" + }, + { + "BriefDescription": "This metric roughly estimates fraction of cycles where the core's performance was likely hurt due to un-chacheable memory", + "MetricExpr": "cpu_core@MEM_LOAD_MISC_RETIRED.UC@ / (cpu_core@MEM_LOAD_RETIRED.L2_HIT@ + cpu_core@MEM_LOAD_MISC_RETIRED.UC@) * cpu_core@MEMORY_STALLS.L2@ / tma_info_thread_clks", + "MetricGroup": "BvMB;MemoryBW;Offcore;TopdownL4;tma_L4_group;tma_issueBW;tma_l3_miss_bound_group", + "MetricName": "tma_uc_bound", + "MetricThreshold": "tma_uc_bound > 0.2 & (tma_l3_miss_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))", + "PublicDescription": "This metric roughly estimates fraction of cycles where the core's performance was likely hurt due to un-chacheable memory. Related metrics: tma_bottleneck_data_cache_memory_bandwidth, tma_fb_full, tma_info_system_dram_bw_use, tma_mem_bandwidth, tma_sq_full", "ScaleUnit": "100%", "Unit": "cpu_core" },
tools/perf/pmu-events/arch/x86/arrowlake/metricgroups.json +1 -1
@@ -93,7 +93,6 @@ "tma_code_stlb_miss_group": "Metrics contributing to tma_code_stlb_miss category", "tma_core_bound_group": "Metrics contributing to tma_core_bound category", "tma_divider_group": "Metrics contributing to tma_divider category", - "tma_dram_bound_group": "Metrics contributing to tma_dram_bound category", "tma_dtlb_load_group": "Metrics contributing to tma_dtlb_load category", "tma_dtlb_store_group": "Metrics contributing to tma_dtlb_store category", "tma_fetch_bandwidth_group": "Metrics contributing to tma_fetch_bandwidth category",
@@ -129,6 +128,7 @@ "tma_l1_bound_group": "Metrics contributing to tma_l1_bound category", "tma_l2_bound_group": "Metrics contributing to tma_l2_bound category", "tma_l3_bound_group": "Metrics contributing to tma_l3_bound category", + "tma_l3_miss_bound_group": "Metrics contributing to tma_l3_miss_bound category", "tma_light_operations_group": "Metrics contributing to tma_light_operations category", "tma_load_op_utilization_group": "Metrics contributing to tma_load_op_utilization category", "tma_load_stlb_miss_group": "Metrics contributing to tma_load_stlb_miss category",
tools/perf/pmu-events/arch/x86/arrowlake/pipeline.json +1 -1
@@ -1238,7 +1238,7 @@ "Counter": "0,1,2,3,4,5,6,7,8,9", "EventCode": "0xc0", "EventName": "INST_RETIRED.REP_ITERATION", - "PublicDescription": "Number of iterations of Repeat (REP) string retired instructions such as MOVS, CMPS, and SCAS. Each has a byte, word, and doubleword version and string instructions can be repeated using a repetition prefix, REP, that allows their architectural execution to be repeated a number of times as specified by the RCX register. Note the number of iterations is implementation-dependent. Available PDIST counters: 0,1", + "PublicDescription": "Number of iterations of Repeat (REP) string retired instructions such as MOVS, CMPS, and SCAS. Each has a byte, word, and doubleword version and string instructions can be repeated using a repetition prefix, REP, that allows their architectural execution to be repeated a number of times as specified by the RCX register. Note: Since the number of iterations within a REP instruction can be significantly affected by fast strings, this event may vary run to run and not match the architectural number of iterations (specified by RCX) Available PDIST counters: 0,1", "SampleAfterValue": "2000003", "UMask": "0x8", "Unit": "cpu_core"
tools/perf/pmu-events/arch/x86/arrowlake/uncore-memory.json +72 -12
139 lines not on this page · raw message
tools/perf/pmu-events/arch/x86/mapfile.csv +1 -1
@@ -1,7 +1,7 @@ Family-model,Version,Filename,EventType GenuineIntel-6-(97|9A|B7|BA|BF),v1.40,alderlake,core GenuineIntel-6-BE,v1.40,alderlaken,core -GenuineIntel-6-C[56],v1.19,arrowlake,core +GenuineIntel-6-C[56],v1.20,arrowlake,core GenuineIntel-6-(1C|26|27|35|36),v5,bonnell,core GenuineIntel-6-(3D|47),v30,broadwell,core GenuineIntel-6-56,v12,broadwellde,core
Signature · 1 line
2.55.0.795.g602f6c329a-goog
PATCH 4/12 perf vendor events intel: Update clearwaterforest events from 1.02 to 1.04
In reply to Chun-Tse ShaoThe updated events were published in:
https://github.com/intel/perfmon/commit/13983cd535d18b2bfd86a3b9daa374039f78a836
--- .../arch/x86/clearwaterforest/memory.json | 33 +++++++++++++++++++ .../arch/x86/clearwaterforest/pipeline.json | 19 +++++++++++ .../clearwaterforest/uncore-interconnect.json | 10 ++++++ .../x86/clearwaterforest/uncore-memory.json | 2 +- tools/perf/pmu-events/arch/x86/mapfile.csv | 2 +- 5 files changed, 64 insertions(+), 2 deletions(-)
tools/perf/pmu-events/arch/x86/clearwaterforest/memory.json +33 -0
@@ -91,6 +91,17 @@ "SampleAfterValue": "1000003", "UMask": "0x4" }, + { + "BriefDescription": "Counts demand data reads that were supplied by DRAM.", + "Counter": "0,1,2,3,4,5,6,7", + "EventCode": "0xB7", + "EventName": "OCR.DEMAND_DATA_RD.DRAM", + "MSRIndex": "0x1a6,0x1a7", + "MSRValue": "0x7BC000001", + "PublicDescription": "Counts demand data reads that were supplied by DRAM. Available PDIST counters: 0", + "SampleAfterValue": "100003", + "UMask": "0x1" + }, { "BriefDescription": "Counts demand data reads that were not supplied by the L3 cache.", "Counter": "0,1,2,3,4,5,6,7",
@@ -102,6 +113,28 @@ "SampleAfterValue": "100003", "UMask": "0x1" }, + { + "BriefDescription": "Counts demand data reads that were supplied by DRAM attached to this socket.", + "Counter": "0,1,2,3,4,5,6,7", + "EventCode": "0xB7", + "EventName": "OCR.DEMAND_DATA_RD.LOCAL_DRAM", + "MSRIndex": "0x1a6,0x1a7", + "MSRValue": "0x784000001", + "PublicDescription": "Counts demand data reads that were supplied by DRAM attached to this socket. Available PDIST counters: 0", + "SampleAfterValue": "100003", + "UMask": "0x1" + }, + { + "BriefDescription": "Counts demand data reads that were supplied by DRAM attached to another socket.", + "Counter": "0,1,2,3,4,5,6,7", + "EventCode": "0xB7", + "EventName": "OCR.DEMAND_DATA_RD.REMOTE_DRAM", + "MSRIndex": "0x1a6,0x1a7", + "MSRValue": "0x730000001", + "PublicDescription": "Counts demand data reads that were supplied by DRAM attached to another socket. Available PDIST counters: 0", + "SampleAfterValue": "100003", + "UMask": "0x1" + }, { "BriefDescription": "Counts demand read for ownership (RFO) requests and software prefetches for exclusive ownership (PREFETCHW) that were not supplied by the L3 cache.", "Counter": "0,1,2,3,4,5,6,7",
tools/perf/pmu-events/arch/x86/clearwaterforest/pipeline.json +19 -0
@@ -291,6 +291,25 @@ "SampleAfterValue": "1000003", "UMask": "0x50" }, + { + "BriefDescription": "This event is deprecated. [This event is alias to BR_MISP_RETIRED.NEAR_RETURN]", + "Counter": "0,1,2,3,4,5,6,7", + "Deprecated": "1", + "EventCode": "0xc5", + "EventName": "BR_MISP_RETIRED.NEAR_RET", + "PublicDescription": "This event is deprecated. [This event is alias to BR_MISP_RETIRED.NEAR_RETURN] Available PDIST counters: 0,1", + "SampleAfterValue": "1000003", + "UMask": "0x8" + }, + { + "BriefDescription": "Counts the number of mispredicted near RET branch instructions retired. [This event is alias to BR_MISP_RETIRED.NEAR_RET]", + "Counter": "0,1,2,3,4,5,6,7", + "EventCode": "0xc5", + "EventName": "BR_MISP_RETIRED.NEAR_RETURN", + "PublicDescription": "Counts the number of mispredicted near RET branch instructions retired. [This event is alias to BR_MISP_RETIRED.NEAR_RET] Available PDIST counters: 0,1", + "SampleAfterValue": "1000003", + "UMask": "0x8" + }, { "BriefDescription": "Fixed Counter: Counts the number of unhalted core clock cycles. [This event is alias to CPU_CLK_UNHALTED.THREAD]", "Counter": "Fixed counter 1",
tools/perf/pmu-events/arch/x86/clearwaterforest/uncore-interconnect.json +10 -0
@@ -814,6 +814,16 @@ "PerPkg": "1", "Unit": "IRP" }, + { + "BriefDescription": "Counts Timeouts - Set 0 : Cache Inserts of Write Transactions as Secondary", + "Counter": "0,1,2,3", + "EventCode": "0x1E", + "EventName": "UNC_I_MISC0.2ND_WR_INSERT", + "Experimental": "1", + "PerPkg": "1", + "UMask": "0x8", + "Unit": "IRP" + }, { "BriefDescription": "Counts Timeouts - Set 0 : Fastpath Rejects", "Counter": "0,1,2,3",
tools/perf/pmu-events/arch/x86/clearwaterforest/uncore-memory.json +1 -1
@@ -520,7 +520,7 @@ "Unit": "IMC" }, { - "BriefDescription": "DRAM Precharge commands. : Precharge due to (?) : Counts the number of DRAM Precharge commands sent on this channel.", + "BriefDescription": "DRAM Precharge commands. : Precharge due to page table : Counts the number of DRAM Precharge commands sent on this channel.", "Counter": "0,1,2,3", "EventCode": "0x03", "EventName": "UNC_M_PRE_COUNT.PGT",
tools/perf/pmu-events/arch/x86/mapfile.csv +1 -1
@@ -7,7 +7,7 @@ GenuineIntel-6-(3D|47),v30,broadwell,core GenuineIntel-6-56,v12,broadwellde,core GenuineIntel-6-4F,v23,broadwellx,core GenuineIntel-6-55-[56789ABCDEF],v1.25,cascadelakex,core -GenuineIntel-6-DD,v1.02,clearwaterforest,core +GenuineIntel-6-DD,v1.04,clearwaterforest,core GenuineIntel-6-9[6C],v1.05,elkhartlake,core GenuineIntel-6-CF,v1.24,emeraldrapids,core GenuineIntel-6-5[CF],v13,goldmont,core
Signature · 1 line
2.55.0.795.g602f6c329a-goog
PATCH 5/12 perf vendor events intel: Update grandridge events from 1.12 to 1.13
In reply to Chun-Tse ShaoThe updated events were published in:
https://github.com/intel/perfmon/commit/e479bd676826824110a49505b51a92952de91200
--- .../pmu-events/arch/x86/grandridge/cache.json | 25 ++++ .../arch/x86/grandridge/counter.json | 42 ------ .../arch/x86/grandridge/floating-point.json | 56 +++++++ .../arch/x86/grandridge/memory.json | 57 +++++++ .../arch/x86/grandridge/pipeline.json | 140 ++++++++++++++++++ .../arch/x86/grandridge/uncore-memory.json | 2 +- .../arch/x86/grandridge/virtual-memory.json | 8 + tools/perf/pmu-events/arch/x86/mapfile.csv | 2 +- 8 files changed, 288 insertions(+), 44 deletions(-) delete mode 100644 tools/perf/pmu-events/arch/x86/grandridge/counter.json
tools/perf/pmu-events/arch/x86/grandridge/cache.json +25 -0
@@ -1,4 +1,12 @@ [ + { + "BriefDescription": "Counts the number of requests that were not accepted into the L2Q because the L2Q is FULL.", + "Counter": "0,1,2,3,4,5,6,7", + "EventCode": "0x31", + "EventName": "CORE_REJECT_L2Q.ANY", + "PublicDescription": "Counts the number of (demand and L1 prefetchers) core requests rejected by the L2Q due to a full or nearly full w condition which likely indicates back pressure from L2Q. It also counts requests that would have gone directly to the XQ, but are rejected due to a full or nearly full condition, indicating back pressure from the IDI link. The L2Q may also reject transactions from a core to insure fairness between cores, or to delay a cores dirty eviction when the address conflicts incoming external snoops. (Note that L2 prefetcher requests that are dropped are not counted by this event.) Counts on a per core basis.", + "SampleAfterValue": "200003" + }, { "BriefDescription": "Counts the number of L1D cacheline (dirty) evictions caused by load misses, stores, and prefetches.", "Counter": "0,1,2,3,4,5,6,7",
@@ -62,6 +70,14 @@ "SampleAfterValue": "1000003", "UMask": "0x1" }, + { + "BriefDescription": "Counts the number of demand and prefetch transactions that the External Queue (XQ) rejects due to a full or near full condition.", + "Counter": "0,1,2,3,4,5,6,7", + "EventCode": "0x30", + "EventName": "L2_REJECT_XQ.ANY", + "PublicDescription": "Counts the number of demand and prefetch transactions that the External Queue (XQ) rejects due to a full or near full condition which likely indicates back pressure from the IDI link. The XQ may reject transactions from the L2Q (non-cacheable requests), BBL (L2 misses) and WOB (L2 write-back victims).", + "SampleAfterValue": "200003" + }, { "BriefDescription": "Counts the number of L2 Cache Accesses that resulted in a Hit from a front door request only (does not include rejects or recycles), per core event", "Counter": "0,1,2,3,4,5,6,7",
@@ -322,6 +338,15 @@ "SampleAfterValue": "20003", "UMask": "0x1" }, + { + "BriefDescription": "Counts the number of memory uops retired. A single uop that performs both a load AND a store will be counted as 1, not 2 (e.g. ADD [mem], CONST)", + "Counter": "0,1,2,3,4,5,6,7", + "Data_LA": "1", + "EventCode": "0xd0", + "EventName": "MEM_UOPS_RETIRED.ALL", + "SampleAfterValue": "200003", + "UMask": "0x83" + }, { "BriefDescription": "Counts the number of load ops retired.", "Counter": "0,1,2,3,4,5,6,7",
tools/perf/pmu-events/arch/x86/grandridge/counter.json deleted +0 -42
43 lines not on this page · raw message
tools/perf/pmu-events/arch/x86/grandridge/floating-point.json +56 -0
68 lines not on this page · raw message
tools/perf/pmu-events/arch/x86/grandridge/memory.json +57 -0
91 lines not on this page · raw message
tools/perf/pmu-events/arch/x86/grandridge/pipeline.json +140 -0
170 lines not on this page · raw message
tools/perf/pmu-events/arch/x86/grandridge/uncore-memory.json +1 -1
@@ -448,7 +448,7 @@ "Unit": "IMC" }, { - "BriefDescription": "DRAM Precharge commands. : Precharge due to (?) : Counts the number of DRAM Precharge commands sent on this channel.", + "BriefDescription": "DRAM Precharge commands. : Precharge due to page table : Counts the number of DRAM Precharge commands sent on this channel.", "Counter": "0,1,2,3", "EventCode": "0x03", "EventName": "UNC_M_PRE_COUNT.PGT",
tools/perf/pmu-events/arch/x86/grandridge/virtual-memory.json +8 -0
@@ -137,6 +137,14 @@ "SampleAfterValue": "200003", "UMask": "0x10" }, + { + "BriefDescription": "Counts the number of cycles that the head (oldest load) of the load buffer is stalled due to a DTLB miss", + "Counter": "0,1,2,3,4,5,6,7", + "EventCode": "0x05", + "EventName": "LD_HEAD.DTLB_MISS", + "SampleAfterValue": "1000003", + "UMask": "0x10" + }, { "BriefDescription": "Counts the number of cycles that the head (oldest load) of the load buffer and retirement are both stalled due to a DTLB miss.", "Counter": "0,1,2,3,4,5,6,7",
tools/perf/pmu-events/arch/x86/mapfile.csv +1 -1
@@ -12,7 +12,7 @@ GenuineIntel-6-9[6C],v1.05,elkhartlake,core GenuineIntel-6-CF,v1.24,emeraldrapids,core GenuineIntel-6-5[CF],v13,goldmont,core GenuineIntel-6-7A,v1.01,goldmontplus,core -GenuineIntel-6-B6,v1.12,grandridge,core +GenuineIntel-6-B6,v1.13,grandridge,core GenuineIntel-6-A[DE],v1.19,graniterapids,core GenuineIntel-6-(3C|45|46),v36,haswell,core GenuineIntel-6-3F,v29,haswellx,core
Signature · 1 line
2.55.0.795.g602f6c329a-goog
PATCH 6/12 perf vendor events intel: Update graniterapids events from 1.19 to 1.20
In reply to Chun-Tse ShaoThe updated events were published in:
https://github.com/intel/perfmon/commit/084ecb869d75f9e5383354d3fe68a93aa25be112
--- .../arch/x86/graniterapids/gnr-metrics.json | 102 +++++++++--------- .../arch/x86/graniterapids/metricgroups.json | 2 +- .../arch/x86/graniterapids/pipeline.json | 2 +- tools/perf/pmu-events/arch/x86/mapfile.csv | 2 +- 4 files changed, 55 insertions(+), 53 deletions(-)
tools/perf/pmu-events/arch/x86/graniterapids/gnr-metrics.json +52 -50
@@ -452,15 +452,15 @@ }, { "BriefDescription": "Total pipeline cost of external Memory- or Cache-Bandwidth related bottlenecks", - "MetricExpr": "100 * (tma_memory_bound * (tma_dram_bound / (tma_cxl_mem_bound + tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_mem_bandwidth / (tma_mem_bandwidth + tma_mem_latency)) + tma_memory_bound * (tma_l3_bound / (tma_cxl_mem_bound + tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_sq_full / (tma_contested_accesses + tma_data_sharing + tma_l3_hit_latency + tma_sq_full)) + tma_memory_bound * (tma_l1_bound / (tma_cxl_mem_bound + tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_fb_full / (tma_dtlb_load + tma_fb_full + tma_l1_latency_dependency + tma_lock_latency + tma_split_loads + tma_store_fwd_blk)))", + "MetricExpr": "100 * (tma_memory_bound * (tma_l3_miss_bound / (tma_cxl_mem_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_mem_bandwidth / (tma_mem_bandwidth + tma_mem_latency)) + tma_memory_bound * (tma_l3_bound / (tma_cxl_mem_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_sq_full / (tma_contested_accesses + tma_data_sharing + tma_l3_hit_latency + tma_sq_full)) + tma_memory_bound * (tma_l1_bound / (tma_cxl_mem_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_fb_full / (tma_dtlb_load + tma_fb_full + tma_l1_latency_dependency + tma_lock_latency + tma_split_loads + tma_store_fwd_blk)))", "MetricGroup": "BvMB;Mem;MemoryBW;Offcore;tma_issueBW", "MetricName": "tma_bottleneck_data_cache_memory_bandwidth", "MetricThreshold": "tma_bottleneck_data_cache_memory_bandwidth > 20", - "PublicDescription": "Total pipeline cost of external Memory- or Cache-Bandwidth related bottlenecks. Related metrics: tma_fb_full, tma_info_system_dram_bw_use, tma_mem_bandwidth, tma_sq_full" + "PublicDescription": "Total pipeline cost of external Memory- or Cache-Bandwidth related bottlenecks. Related metrics: tma_fb_full, tma_info_system_dram_bw_use, tma_mem_bandwidth, tma_sq_full, tma_uc_bound" }, { "BriefDescription": "Total pipeline cost of external Memory- or Cache-Latency related bottlenecks", - "MetricExpr": "100 * (tma_memory_bound * (tma_dram_bound / (tma_cxl_mem_bound + tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_mem_latency / (tma_mem_bandwidth + tma_mem_latency)) + 0 / (tma_cxl_mem_bound + tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound) * tma_mem_latency / (tma_mem_bandwidth + tma_mem_latency) + tma_memory_bound * (tma_l3_bound / (tma_cxl_mem_bound + tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_l3_hit_latency / (tma_contested_accesses + tma_data_sharing + tma_l3_hit_latency + tma_sq_full)) + tma_memory_bound * tma_l2_bound / (tma_cxl_mem_bound + tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound) + tma_memory_bound * (tma_l1_bound / (tma_cxl_mem_bound + tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_l1_latency_dependency / (tma_dtlb_load + tma_fb_full + tma_l1_latency_dependency + tma_lock_latency + tma_split_loads + tma_store_fwd_blk)) + tma_memory_bound * (tma_l1_bound / (tma_cxl_mem_bound + tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_lock_latency / (tma_dtlb_load + tma_fb_full + tma_l1_latency_dependency + tma_lock_latency + tma_split_loads + tma_store_fwd_blk)) + tma_memory_bound * (tma_l1_bound / (tma_cxl_mem_bound + tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_split_loads / (tma_dtlb_load + tma_fb_full + tma_l1_latency_dependency + tma_lock_latency + tma_split_loads + tma_store_fwd_blk)) + tma_memory_bound * (tma_store_bound / (tma_cxl_mem_bound + tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_split_stores / (tma_dtlb_store + tma_false_sharing + tma_split_stores + tma_store_latency + tma_streaming_stores)) + tma_memory_bound * (tma_store_bound / (tma_cxl_mem_bound + tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_store_latency / (tma_dtlb_store + tma_false_sharing + tma_split_stores + tma_store_latency + tma_streaming_stores)))", + "MetricExpr": "100 * (tma_memory_bound * (tma_l3_miss_bound / (tma_cxl_mem_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_mem_latency / (tma_mem_bandwidth + tma_mem_latency)) + tma_memory_bound * (tma_l3_bound / (tma_cxl_mem_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_l3_hit_latency / (tma_contested_accesses + tma_data_sharing + tma_l3_hit_latency + tma_sq_full)) + tma_memory_bound * tma_l2_bound / (tma_cxl_mem_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound) + tma_memory_bound * (tma_l1_bound / (tma_cxl_mem_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_l1_latency_dependency / (tma_dtlb_load + tma_fb_full + tma_l1_latency_dependency + tma_lock_latency + tma_split_loads + tma_store_fwd_blk)) + tma_memory_bound * (tma_l1_bound / (tma_cxl_mem_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_lock_latency / (tma_dtlb_load + tma_fb_full + tma_l1_latency_dependency + tma_lock_latency + tma_split_loads + tma_store_fwd_blk)) + tma_memory_bound * (tma_l1_bound / (tma_cxl_mem_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_split_loads / (tma_dtlb_load + tma_fb_full + tma_l1_latency_dependency + tma_lock_latency + tma_split_loads + tma_store_fwd_blk)) + tma_memory_bound * (tma_store_bound / (tma_cxl_mem_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_split_stores / (tma_dtlb_store + tma_false_sharing + tma_split_stores + tma_store_latency + tma_streaming_stores)) + tma_memory_bound * (tma_store_bound / (tma_cxl_mem_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_store_latency / (tma_dtlb_store + tma_false_sharing + tma_split_stores + tma_store_latency + tma_streaming_stores)))", "MetricGroup": "BvML;Mem;MemoryLat;Offcore;tma_issueLat", "MetricName": "tma_bottleneck_data_cache_memory_latency", "MetricThreshold": "tma_bottleneck_data_cache_memory_latency > 20",
@@ -483,7 +483,7 @@ }, { "BriefDescription": "Total pipeline cost of Memory Address Translation related bottlenecks (data-side TLBs)", - "MetricExpr": "100 * (tma_memory_bound * (tma_l1_bound / (tma_cxl_mem_bound + tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_dtlb_load / (tma_dtlb_load + tma_fb_full + tma_l1_latency_dependency + tma_lock_latency + tma_split_loads + tma_store_fwd_blk)) + tma_memory_bound * (tma_store_bound / (tma_cxl_mem_bound + tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_dtlb_store / (tma_dtlb_store + tma_false_sharing + tma_split_stores + tma_store_latency + tma_streaming_stores)))", + "MetricExpr": "100 * (tma_memory_bound * (tma_l1_bound / (tma_cxl_mem_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_dtlb_load / (tma_dtlb_load + tma_fb_full + tma_l1_latency_dependency + tma_lock_latency + tma_split_loads + tma_store_fwd_blk)) + tma_memory_bound * (tma_store_bound / (tma_cxl_mem_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_dtlb_store / (tma_dtlb_store + tma_false_sharing + tma_split_stores + tma_store_latency + tma_streaming_stores)))", "MetricGroup": "BvMT;Mem;MemoryTLB;Offcore;tma_issueTLB", "MetricName": "tma_bottleneck_memory_data_tlbs", "MetricThreshold": "tma_bottleneck_memory_data_tlbs > 20",
@@ -491,7 +491,7 @@ }, { "BriefDescription": "Total pipeline cost of Memory Synchronization related bottlenecks (data transfers and coherency updates across processors)", - "MetricExpr": "100 * (tma_memory_bound * (tma_dram_bound / (tma_cxl_mem_bound + tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound) * (tma_mem_latency / (tma_mem_bandwidth + tma_mem_latency)) * tma_remote_cache / (tma_local_mem + tma_remote_cache + tma_remote_mem) + tma_l3_bound / (tma_cxl_mem_bound + tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound) * (tma_contested_accesses + tma_data_sharing) / (tma_contested_accesses + tma_data_sharing + tma_l3_hit_latency + tma_sq_full) + tma_store_bound / (tma_cxl_mem_bound + tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound) * tma_false_sharing / (tma_dtlb_store + tma_false_sharing + tma_split_stores + tma_store_latency + tma_streaming_stores - tma_store_latency)) + tma_machine_clears * (1 - tma_other_nukes / tma_other_nukes))", + "MetricExpr": "100 * (tma_memory_bound * (tma_l3_miss_bound / (tma_cxl_mem_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound) * (tma_mem_latency / (tma_mem_bandwidth + tma_mem_latency)) * tma_remote_cache / (tma_local_mem + tma_remote_cache + tma_remote_mem) + tma_l3_bound / (tma_cxl_mem_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound) * (tma_contested_accesses + tma_data_sharing) / (tma_contested_accesses + tma_data_sharing + tma_l3_hit_latency + tma_sq_full) + tma_store_bound / (tma_cxl_mem_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound) * tma_false_sharing / (tma_dtlb_store + tma_false_sharing + tma_split_stores + tma_store_latency + tma_streaming_stores - tma_store_latency)) + tma_machine_clears * (1 - tma_other_nukes / tma_other_nukes))", "MetricGroup": "BvMS;LockCont;Mem;Offcore;tma_issueSyncxn", "MetricName": "tma_bottleneck_memory_synchronization", "MetricThreshold": "tma_bottleneck_memory_synchronization > 10",
@@ -624,7 +624,7 @@ }, { "BriefDescription": "This metric represents fraction of cycles the CPU was stalled due to retired misprediction by non-taken conditional branches.", - "MetricExpr": "BR_MISP_RETIRED.COND_NTAKEN_COST * BR_MISP_RETIRED.COND_NTAKEN_COST:R / tma_info_thread_clks", + "MetricExpr": "BR_MISP_RETIRED.COND_NTAKEN / BR_MISP_RETIRED.ALL_BRANCHES * tma_branch_mispredicts", "MetricGroup": "BrMispredicts;TopdownL3;tma_L3_group;tma_branch_mispredicts_group", "MetricName": "tma_cond_nt_mispredicts", "MetricThreshold": "tma_cond_nt_mispredicts > 0.05 & (tma_branch_mispredicts > 0.1 & tma_bad_speculation > 0.15)",
@@ -632,7 +632,7 @@ }, { "BriefDescription": "This metric represents fraction of cycles the CPU was stalled due to misprediction by taken conditional branches.", - "MetricExpr": "BR_MISP_RETIRED.COND_TAKEN_COST * BR_MISP_RETIRED.COND_TAKEN_COST:R / tma_info_thread_clks", + "MetricExpr": "BR_MISP_RETIRED.COND_TAKEN / BR_MISP_RETIRED.ALL_BRANCHES * tma_branch_mispredicts", "MetricGroup": "BrMispredicts;TopdownL3;tma_L3_group;tma_branch_mispredicts_group", "MetricName": "tma_cond_tk_mispredicts", "MetricThreshold": "tma_cond_tk_mispredicts > 0.05 & (tma_branch_mispredicts > 0.1 & tma_bad_speculation > 0.15)",
@@ -668,13 +668,13 @@ "ScaleUnit": "100%" }, { - "BriefDescription": "This metric estimates fraction of cycles while the memory subsystem was handling synchronizations due to data-sharing accesses", + "BriefDescription": "This metric estimates fraction of cycles while the memory subsystem was handling synchronizations due to L3 data-sharing accesses", "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "(MEM_LOAD_L3_HIT_RETIRED.XSNP_NO_FWD * min(MEM_LOAD_L3_HIT_RETIRED.XSNP_NO_FWD:R, 74.6 * tma_info_system_core_frequency) + MEM_LOAD_L3_HIT_RETIRED.XSNP_FWD * min(MEM_LOAD_L3_HIT_RETIRED.XSNP_FWD:R, 74.6 * tma_info_system_core_frequency) * (1 - OCR.DEMAND_DATA_RD.L3_HIT.SNOOP_HITM / (OCR.DEMAND_DATA_RD.L3_HIT.SNOOP_HITM + OCR.DEMAND_DATA_RD.L3_HIT.SNOOP_HIT_WITH_FWD))) * (1 + MEM_LOAD_RETIRED.FB_HIT / MEM_LOAD_RETIRED.L1_MISS / 2) / tma_info_thread_clks", "MetricGroup": "BvMS;Offcore;Snoop;TopdownL4;tma_L4_group;tma_issueSyncxn;tma_l3_bound_group", "MetricName": "tma_data_sharing", "MetricThreshold": "tma_data_sharing > 0.05 & (tma_l3_bound > 0.05 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))", - "PublicDescription": "This metric estimates fraction of cycles while the memory subsystem was handling synchronizations due to data-sharing accesses. Data shared by multiple Logical Processors (even just read shared) may cause increased access latency due to cache coherency. Excessive data sharing can drastically harm multithreaded performance. Sample with: MEM_LOAD_L3_HIT_RETIRED.XSNP_NO_FWD. Related metrics: tma_bottleneck_memory_synchronization, tma_contested_accesses, tma_false_sharing, tma_machine_clears, tma_remote_cache", + "PublicDescription": "This metric estimates fraction of cycles while the memory subsystem was handling synchronizations due to L3 data-sharing accesses. Data shared by multiple Logical Processors (even just read shared) may cause increased access latency due to cache coherency. Excessive data sharing can drastically harm multithreaded performance. Sample with: MEM_LOAD_L3_HIT_RETIRED.XSNP_NO_FWD. Related metrics: tma_bottleneck_memory_synchronization, tma_contested_accesses, tma_false_sharing, tma_machine_clears, tma_remote_cache", "ScaleUnit": "100%" }, {
@@ -695,15 +695,6 @@ "PublicDescription": "This metric represents fraction of cycles where the Divider unit was active. Divide and square root instructions are performed by the Divider unit and can take considerably longer latency than integer or Floating Point addition; subtraction; or multiplication. Sample with: ARITH.DIV_ACTIVE", "ScaleUnit": "100%" }, - { - "BriefDescription": "This metric estimates how often the CPU was stalled on accesses to external memory (DRAM) by loads", - "MetricExpr": "(MEMORY_ACTIVITY.STALLS_L3_MISS / tma_info_thread_clks - tma_cxl_mem_bound if #has_pmem > 0 else MEMORY_ACTIVITY.STALLS_L3_MISS / tma_info_thread_clks)", - "MetricGroup": "MemoryBound;TmaL3mem;TopdownL3;tma_L3_group;tma_memory_bound_group", - "MetricName": "tma_dram_bound", - "MetricThreshold": "tma_dram_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2)", - "PublicDescription": "This metric estimates how often the CPU was stalled on accesses to external memory (DRAM) by loads. Better caching can improve the latency and increase performance. Sample with: MEM_LOAD_RETIRED.L3_MISS", - "ScaleUnit": "100%" - }, { "BriefDescription": "This metric represents Core fraction of cycles in which CPU was likely limited due to DSB (decoded uop cache) fetch pipeline", "MetricExpr": "(IDQ.DSB_CYCLES_ANY - IDQ.DSB_CYCLES_OK) / tma_info_core_core_clks / 2",
@@ -755,7 +746,7 @@ "MetricGroup": "BvMB;MemoryBW;TopdownL4;tma_L4_group;tma_issueBW;tma_issueSL;tma_issueSmSt;tma_l1_bound_group", "MetricName": "tma_fb_full", "MetricThreshold": "tma_fb_full > 0.3", - "PublicDescription": "This metric does a *rough estimation* of how often L1D Fill Buffer unavailability limited additional L1D miss memory access requests to proceed. The higher the metric value; the deeper the memory hierarchy level the misses are satisfied from (metric values >1 are valid). Often it hints on approaching bandwidth limits (to L2 cache; L3 cache or external memory). Related metrics: tma_bottleneck_data_cache_memory_bandwidth, tma_info_system_dram_bw_use, tma_mem_bandwidth, tma_sq_full, tma_store_latency, tma_streaming_stores", + "PublicDescription": "This metric does a *rough estimation* of how often L1D Fill Buffer unavailability limited additional L1D miss memory access requests to proceed. The higher the metric value; the deeper the memory hierarchy level the misses are satisfied from (metric values >1 are valid). Often it hints on approaching bandwidth limits (to L2 cache; L3 cache or external memory). Related metrics: tma_bottleneck_data_cache_memory_bandwidth, tma_info_system_dram_bw_use, tma_mem_bandwidth, tma_sq_full, tma_store_latency, tma_streaming_stores, tma_uc_bound", "ScaleUnit": "100%" }, {
@@ -902,7 +893,7 @@ }, { "BriefDescription": "This metric represents fraction of cycles the CPU was stalled due to retired misprediction by indirect CALL instructions.", - "MetricExpr": "BR_MISP_RETIRED.INDIRECT_CALL_COST * BR_MISP_RETIRED.INDIRECT_CALL_COST:R / tma_info_thread_clks", + "MetricExpr": "BR_MISP_RETIRED.INDIRECT_CALL / BR_MISP_RETIRED.ALL_BRANCHES * tma_branch_mispredicts", "MetricGroup": "BrMispredicts;TopdownL3;tma_L3_group;tma_branch_mispredicts_group", "MetricName": "tma_ind_call_mispredicts", "MetricThreshold": "tma_ind_call_mispredicts > 0.05 & (tma_branch_mispredicts > 0.1 & tma_bad_speculation > 0.15)",
@@ -910,7 +901,7 @@ }, { "BriefDescription": "This metric represents fraction of cycles the CPU was stalled due to retired misprediction by indirect JMP instructions.", - "MetricExpr": "max((BR_MISP_RETIRED.INDIRECT_COST * BR_MISP_RETIRED.INDIRECT_COST:R - BR_MISP_RETIRED.INDIRECT_CALL_COST * BR_MISP_RETIRED.INDIRECT_CALL_COST:R) / tma_info_thread_clks, 0)", + "MetricExpr": "max((BR_MISP_RETIRED.INDIRECT - BR_MISP_RETIRED.INDIRECT_CALL) / BR_MISP_RETIRED.ALL_BRANCHES * tma_branch_mispredicts, 0)", "MetricGroup": "BrMispredicts;TopdownL3;tma_L3_group;tma_branch_mispredicts_group", "MetricName": "tma_ind_jump_mispredicts", "MetricThreshold": "tma_ind_jump_mispredicts > 0.05 & (tma_branch_mispredicts > 0.1 & tma_bad_speculation > 0.15)",
@@ -1045,7 +1036,7 @@ }, { "BriefDescription": "Floating Point Operations Per Cycle", - "MetricExpr": "(cpu@FP_ARITH_INST_RETIRED.SCALAR_SINGLE\\,umask\\=0x03@ + 2 * FP_ARITH_INST_RETIRED.128B_PACKED_DOUBLE + 4 * cpu@FP_ARITH_INST_RETIRED.128B_PACKED_SINGLE\\,umask\\=0x18@ + 8 * cpu@FP_ARITH_INST_RETIRED.256B_PACKED_SINGLE\\,umask\\=0x60@ + 16 * FP_ARITH_INST_RETIRED.512B_PACKED_SINGLE) / tma_info_core_core_clks", + "MetricExpr": "(FP_ARITH_INST_RETIRED.SCALAR + 2 * FP_ARITH_INST_RETIRED.128B_PACKED_DOUBLE + 4 * FP_ARITH_INST_RETIRED.4_FLOPS + 8 * FP_ARITH_INST_RETIRED.8_FLOPS + 16 * FP_ARITH_INST_RETIRED.512B_PACKED_SINGLE) / tma_info_core_core_clks", "MetricGroup": "Flops;Ret", "MetricName": "tma_info_core_flopc" },
@@ -1064,7 +1055,7 @@ }, { "BriefDescription": "Fraction of Uops delivered by the DSB (aka Decoded ICache; or Uop Cache)", - "MetricExpr": "IDQ.DSB_UOPS / UOPS_ISSUED.ANY", + "MetricExpr": "IDQ.DSB_UOPS / (IDQ.DSB_UOPS + IDQ.MITE_UOPS + IDQ.MS_UOPS)", "MetricGroup": "DSB;Fed;FetchBW;tma_issueFB", "MetricName": "tma_info_frontend_dsb_coverage", "MetricThreshold": "tma_info_frontend_dsb_coverage < 0.7 & tma_info_thread_ipc / 6 > 0.35",
@@ -1202,7 +1193,7 @@ { "BriefDescription": "Instructions per FP Arithmetic Scalar Half-Precision instruction (lower number means higher occurrence rate)", "MetricExpr": "INST_RETIRED.ANY / FP_ARITH_INST_RETIRED2.SCALAR", - "MetricGroup": "Flops;FpScalar;InsType;Server", + "MetricGroup": "Flops;FpScalar;InsType", "MetricName": "tma_info_inst_mix_iparith_scalar_hp", "MetricThreshold": "tma_info_inst_mix_iparith_scalar_hp < 10", "PublicDescription": "Instructions per FP Arithmetic Scalar Half-Precision instruction (lower number means higher occurrence rate). Values < 1 are possible due to intentional FMA double counting."
@@ -1231,7 +1222,7 @@ }, { "BriefDescription": "Instructions per Floating Point (FP) Operation (lower number means higher occurrence rate)", - "MetricExpr": "INST_RETIRED.ANY / (cpu@FP_ARITH_INST_RETIRED.SCALAR_SINGLE\\,umask\\=0x03@ + 2 * FP_ARITH_INST_RETIRED.128B_PACKED_DOUBLE + 4 * cpu@FP_ARITH_INST_RETIRED.128B_PACKED_SINGLE\\,umask\\=0x18@ + 8 * cpu@FP_ARITH_INST_RETIRED.256B_PACKED_SINGLE\\,umask\\=0x60@ + 16 * FP_ARITH_INST_RETIRED.512B_PACKED_SINGLE)", + "MetricExpr": "INST_RETIRED.ANY / (FP_ARITH_INST_RETIRED.SCALAR + 2 * FP_ARITH_INST_RETIRED.128B_PACKED_DOUBLE + 4 * FP_ARITH_INST_RETIRED.4_FLOPS + 8 * FP_ARITH_INST_RETIRED.8_FLOPS + 16 * FP_ARITH_INST_RETIRED.512B_PACKED_SINGLE)", "MetricGroup": "Flops;InsType", "MetricName": "tma_info_inst_mix_ipflop", "MetricThreshold": "tma_info_inst_mix_ipflop < 10"
@@ -1271,6 +1262,14 @@ "MetricThreshold": "tma_info_inst_mix_iptb < 13", "PublicDescription": "Instructions per taken branch. Related metrics: tma_dsb_switches, tma_fetch_bandwidth, tma_info_botlnk_l2_dsb_bandwidth, tma_info_botlnk_l2_dsb_misses, tma_info_frontend_dsb_coverage, tma_lcp" }, + { + "BriefDescription": "AVX preserve/restore assists per kilo instruction", + "MetricExpr": "1e3 * ASSISTS.SSE_AVX_MIX / INST_RETIRED.ANY", + "MetricGroup": "tma_issueMV", + "MetricName": "tma_info_inst_mix_vectormixpki", + "MetricThreshold": "tma_info_inst_mix_vectormixpki > 0.05", + "PublicDescription": "AVX preserve/restore assists per kilo instruction. Related metrics: tma_mixing_vectors, tma_ms_switches" + }, { "BriefDescription": "Average per-core data fill bandwidth to the L1 data cache [GB / sec]", "MetricExpr": "tma_info_memory_l1d_cache_fill_bw",
@@ -1526,7 +1525,7 @@ "MetricName": "tma_info_memory_tlb_store_stlb_mpki" }, { - "BriefDescription": "Mem;Backend;CacheHits", + "BriefDescription": "Instruction-Level-Parallelism (average number of uops executed when there is execution) per physical core", "MetricExpr": "UOPS_EXECUTED.THREAD / (UOPS_EXECUTED.CORE_CYCLES_GE_1 / 2 if #SMT_on else cpu@UOPS_EXECUTED.THREAD\\,cmask\\=1@)", "MetricGroup": "Cor;Pipeline;PortsUtil;SMT", "MetricName": "tma_info_pipeline_execute"
@@ -1606,11 +1605,11 @@ "MetricExpr": "64 * (UNC_M_CAS_COUNT_SCH0.RD + UNC_M_CAS_COUNT_SCH1.RD + UNC_M_CAS_COUNT_SCH0.WR + UNC_M_CAS_COUNT_SCH1.WR) / 1e9 / tma_info_system_time", "MetricGroup": "HPC;MemOffcore;MemoryBW;SoC;tma_issueBW", "MetricName": "tma_info_system_dram_bw_use", - "PublicDescription": "Average external Memory Bandwidth Use for reads and writes [GB / sec]. Related metrics: tma_bottleneck_data_cache_memory_bandwidth, tma_fb_full, tma_mem_bandwidth, tma_sq_full" + "PublicDescription": "Average external Memory Bandwidth Use for reads and writes [GB / sec]. Related metrics: tma_bottleneck_data_cache_memory_bandwidth, tma_fb_full, tma_mem_bandwidth, tma_sq_full, tma_uc_bound" }, { "BriefDescription": "Giga Floating Point Operations Per Second", - "MetricExpr": "(cpu@FP_ARITH_INST_RETIRED.SCALAR_SINGLE\\,umask\\=0x03@ + 2 * FP_ARITH_INST_RETIRED.128B_PACKED_DOUBLE + 4 * cpu@FP_ARITH_INST_RETIRED.128B_PACKED_SINGLE\\,umask\\=0x18@ + 8 * cpu@FP_ARITH_INST_RETIRED.256B_PACKED_SINGLE\\,umask\\=0x60@ + 16 * FP_ARITH_INST_RETIRED.512B_PACKED_SINGLE) / 1e9 / tma_info_system_time", + "MetricExpr": "(FP_ARITH_INST_RETIRED.SCALAR + 2 * FP_ARITH_INST_RETIRED.128B_PACKED_DOUBLE + 4 * FP_ARITH_INST_RETIRED.4_FLOPS + 8 * FP_ARITH_INST_RETIRED.8_FLOPS + 16 * FP_ARITH_INST_RETIRED.512B_PACKED_SINGLE) / 1e9 / tma_info_system_time", "MetricGroup": "Cor;Flops;HPC", "MetricName": "tma_info_system_gflops", "PublicDescription": "Giga Floating Point Operations Per Second. Aggregate across all supported options of: FP precisions, scalar and vector instructions, vector-width"
@@ -1701,12 +1700,6 @@ "MetricGroup": "Power", "MetricName": "tma_info_system_turbo_utilization" }, - { - "BriefDescription": "Measured Average Uncore Frequency for the SoC [GHz]", - "MetricExpr": "tma_info_system_socket_clks / 1e9 / tma_info_system_time", - "MetricGroup": "SoC", - "MetricName": "tma_info_system_uncore_frequency" - }, { "BriefDescription": "Cross-socket Ultra Path Interconnect (UPI) data transmit bandwidth for data only [MB / sec]", "MetricExpr": "UNC_UPI_TxL_FLITS.ALL_DATA * 64 / 9 / 1e6",
@@ -1862,6 +1855,15 @@ "PublicDescription": "This metric estimates fraction of cycles with demand load accesses that hit the L3 cache under unloaded scenarios (possibly L3 latency limited). Avoiding private cache misses (i.e. L2 misses/L3 hits) will improve the latency; reduce contention with sibling physical cores and increase performance. Note the value of this node may overlap with its siblings. Sample with: MEM_LOAD_RETIRED.L3_HIT_PS. Related metrics: tma_bottleneck_data_cache_memory_latency, tma_mem_latency", "ScaleUnit": "100%" }, + { + "BriefDescription": "This metric estimates how often the CPU was stalled on accesses to external memory (DRAM) by loads", + "MetricExpr": "(MEMORY_ACTIVITY.STALLS_L3_MISS / tma_info_thread_clks - tma_cxl_mem_bound if #has_pmem > 0 else MEMORY_ACTIVITY.STALLS_L3_MISS / tma_info_thread_clks)", + "MetricGroup": "MemoryBound;Offcore;TmaL3mem;TopdownL3;tma_L3_group;tma_memory_bound_group", + "MetricName": "tma_l3_miss_bound", + "MetricThreshold": "tma_l3_miss_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2)", + "PublicDescription": "This metric estimates how often the CPU was stalled on accesses to external memory (DRAM) by loads. Better caching can improve the latency and increase performance. Sample with: MEM_LOAD_RETIRED.L3_MISS", + "ScaleUnit": "100%" + }, { "BriefDescription": "This metric represents fraction of cycles CPU was stalled due to Length Changing Prefixes (LCPs)", "MetricExpr": "DECODE.LCP / tma_info_thread_clks",
@@ -1936,7 +1938,7 @@ "MetricExpr": "MEM_LOAD_L3_MISS_RETIRED.LOCAL_DRAM * MEM_LOAD_L3_MISS_RETIRED.LOCAL_DRAM:R * (1 + MEM_LOAD_RETIRED.FB_HIT / MEM_LOAD_RETIRED.L1_MISS / 2) / tma_info_thread_clks", "MetricGroup": "Server;TopdownL5;tma_L5_group;tma_mem_latency_group", "MetricName": "tma_local_mem", - "MetricThreshold": "tma_local_mem > 0.1 & (tma_mem_latency > 0.1 & (tma_dram_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2)))", + "MetricThreshold": "tma_local_mem > 0.1 & (tma_mem_latency > 0.1 & (tma_l3_miss_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2)))", "PublicDescription": "This metric estimates fraction of cycles while the memory subsystem was handling loads from local memory. Caching will improve the latency and increase performance. Sample with: MEM_LOAD_L3_MISS_RETIRED.LOCAL_DRAM", "ScaleUnit": "100%" },
@@ -1966,24 +1968,24 @@ "MetricExpr": "INT_MISC.MBA_STALLS / tma_info_thread_clks", "MetricGroup": "MemoryBW;Offcore;Server;TopdownL5;tma_L5_group;tma_mem_bandwidth_group", "MetricName": "tma_mba_stalls", - "MetricThreshold": "tma_mba_stalls > 0.1 & (tma_mem_bandwidth > 0.2 & (tma_dram_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2)))", + "MetricThreshold": "tma_mba_stalls > 0.1 & (tma_mem_bandwidth > 0.2 & (tma_l3_miss_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2)))", "ScaleUnit": "100%" }, { "BriefDescription": "This metric estimates fraction of cycles where the core's performance was likely hurt due to approaching bandwidth limits of external memory - DRAM ([SPR-HBM] and/or HBM)", - "MetricExpr": "min(CPU_CLK_UNHALTED.THREAD, cpu@OFFCORE_REQUESTS_OUTSTANDING.DATA_RD\\,cmask\\=4@) / tma_info_thread_clks", - "MetricGroup": "BvMB;MemoryBW;Offcore;TopdownL4;tma_L4_group;tma_dram_bound_group;tma_issueBW", + "MetricExpr": "min(CPU_CLK_UNHALTED.THREAD, cpu@OFFCORE_REQUESTS_OUTSTANDING.DATA_RD\\,cmask\\=12@) / tma_info_thread_clks", + "MetricGroup": "BvMB;MemoryBW;Offcore;TopdownL4;tma_L4_group;tma_issueBW;tma_l3_miss_bound_group", "MetricName": "tma_mem_bandwidth", - "MetricThreshold": "tma_mem_bandwidth > 0.2 & (tma_dram_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))", - "PublicDescription": "This metric estimates fraction of cycles where the core's performance was likely hurt due to approaching bandwidth limits of external memory - DRAM ([SPR-HBM] and/or HBM). The underlying heuristic assumes that a similar off-core traffic is generated by all IA cores. This metric does not aggregate non-data-read requests by this logical processor; requests from other IA Logical Processors/Physical Cores/sockets; or other non-IA devices like GPU; hence the maximum external memory bandwidth limits may or may not be approached when this metric is flagged (see Uncore counters for that). Related metrics: tma_bottleneck_data_cache_memory_bandwidth, tma_fb_full, tma_info_system_dram_bw_use, tma_sq_full", + "MetricThreshold": "tma_mem_bandwidth > 0.2 & (tma_l3_miss_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))", + "PublicDescription": "This metric estimates fraction of cycles where the core's performance was likely hurt due to approaching bandwidth limits of external memory - DRAM ([SPR-HBM] and/or HBM). The underlying heuristic assumes that a similar off-core traffic is generated by all IA cores. This metric does not aggregate non-data-read requests by this logical processor; requests from other IA Logical Processors/Physical Cores/sockets; or other non-IA devices like GPU; hence the maximum external memory bandwidth limits may or may not be approached when this metric is flagged (see Uncore counters for that). Related metrics: tma_bottleneck_data_cache_memory_bandwidth, tma_fb_full, tma_info_system_dram_bw_use, tma_sq_full, tma_uc_bound", "ScaleUnit": "100%" }, { "BriefDescription": "This metric estimates fraction of cycles where the performance was likely hurt due to latency from external memory - DRAM ([SPR-HBM] and/or HBM)", "MetricExpr": "min(CPU_CLK_UNHALTED.THREAD, OFFCORE_REQUESTS_OUTSTANDING.CYCLES_WITH_DATA_RD) / tma_info_thread_clks - tma_mem_bandwidth", - "MetricGroup": "BvML;MemoryLat;Offcore;TopdownL4;tma_L4_group;tma_dram_bound_group;tma_issueLat", + "MetricGroup": "BvML;MemoryLat;Offcore;TopdownL4;tma_L4_group;tma_issueLat;tma_l3_miss_bound_group", "MetricName": "tma_mem_latency", - "MetricThreshold": "tma_mem_latency > 0.1 & (tma_dram_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))", + "MetricThreshold": "tma_mem_latency > 0.1 & (tma_l3_miss_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))", "PublicDescription": "This metric estimates fraction of cycles where the performance was likely hurt due to latency from external memory - DRAM ([SPR-HBM] and/or HBM). This metric does not aggregate requests from other Logical Processors/Physical Cores/sockets (see Uncore counters for that). Related metrics: tma_bottleneck_data_cache_memory_latency, tma_l3_hit_latency", "ScaleUnit": "100%" },
@@ -2047,7 +2049,7 @@ "MetricGroup": "TopdownL5;tma_L5_group;tma_issueMV;tma_ports_utilized_0_group", "MetricName": "tma_mixing_vectors", "MetricThreshold": "tma_mixing_vectors > 0.05", - "PublicDescription": "This metric estimates penalty in terms of percentage of([SKL+] injected blend uops out of all Uops Issued -- the Count Domain; [ADL+] cycles). Usually a Mixing_Vectors over 5% is worth investigating. Read more in Appendix B1 of the Optimizations Guide for this topic. Related metrics: tma_ms_switches", + "PublicDescription": "This metric estimates penalty in terms of percentage of([SKL+] injected blend uops out of all Uops Issued -- the Count Domain; [ADL+] cycles). Usually a Mixing_Vectors over 5% is worth investigating. Read more in Appendix B1 of the Optimizations Guide for this topic. Related metrics: tma_info_inst_mix_vectormixpki, tma_ms_switches", "ScaleUnit": "100%" }, {
@@ -2064,7 +2066,7 @@ "MetricGroup": "FetchLat;MicroSeq;TopdownL3;tma_L3_group;tma_fetch_latency_group;tma_issueMC;tma_issueMS;tma_issueMV;tma_issueSO", "MetricName": "tma_ms_switches", "MetricThreshold": "tma_ms_switches > 0.05 & (tma_fetch_latency > 0.1 & tma_frontend_bound > 0.15)", - "PublicDescription": "This metric estimates the fraction of cycles when the CPU was stalled due to switches of uop delivery to the Microcode Sequencer (MS). Commonly used instructions are optimized for delivery by the DSB (decoded i-cache) or MITE (legacy instruction decode) pipelines. Certain operations cannot be handled natively by the execution pipeline; and must be performed by microcode (small programs injected into the execution stream). Switching to the MS too often can negatively impact performance. The MS is designated to deliver long uop flows required by CISC instructions like CPUID; or uncommon conditions like Floating Point Assists when dealing with Denormals. Sample with: IDQ.MS_SWITCHES. Related metrics: tma_bottleneck_irregular_overhead, tma_clears_resteers, tma_l1_bound, tma_machine_clears, tma_microcode_sequencer, tma_mixing_vectors, tma_serializing_operation", + "PublicDescription": "This metric estimates the fraction of cycles when the CPU was stalled due to switches of uop delivery to the Microcode Sequencer (MS). Commonly used instructions are optimized for delivery by the DSB (decoded i-cache) or MITE (legacy instruction decode) pipelines. Certain operations cannot be handled natively by the execution pipeline; and must be performed by microcode (small programs injected into the execution stream). Switching to the MS too often can negatively impact performance. The MS is designated to deliver long uop flows required by CISC instructions like CPUID; or uncommon conditions like Floating Point Assists when dealing with Denormals. Sample with: IDQ.MS_SWITCHES. Related metrics: tma_bottleneck_irregular_overhead, tma_clears_resteers, tma_info_inst_mix_vectormixpki, tma_l1_bound, tma_machine_clears, tma_microcode_sequencer, tma_mixing_vectors, tma_serializing_operation", "ScaleUnit": "100%" }, {
@@ -2197,25 +2199,25 @@ }, { "BriefDescription": "This metric estimates fraction of cycles while the memory subsystem was handling loads from remote cache in other sockets including synchronizations issues", - "MetricExpr": "(MEM_LOAD_L3_MISS_RETIRED.REMOTE_HITM * PEBS + MEM_LOAD_L3_MISS_RETIRED.REMOTE_FWD * PEBS) * (1 + MEM_LOAD_RETIRED.FB_HIT / MEM_LOAD_RETIRED.L1_MISS / 2) / tma_info_thread_clks", + "MetricExpr": "(MEM_LOAD_L3_MISS_RETIRED.REMOTE_HITM * MEM_LOAD_L3_MISS_RETIRED.REMOTE_HITM:R + MEM_LOAD_L3_MISS_RETIRED.REMOTE_FWD * MEM_LOAD_L3_MISS_RETIRED.REMOTE_FWD:R) * (1 + MEM_LOAD_RETIRED.FB_HIT / MEM_LOAD_RETIRED.L1_MISS / 2) / tma_info_thread_clks", "MetricGroup": "Offcore;Server;Snoop;TopdownL5;tma_L5_group;tma_issueSyncxn;tma_mem_latency_group", "MetricName": "tma_remote_cache", - "MetricThreshold": "tma_remote_cache > 0.05 & (tma_mem_latency > 0.1 & (tma_dram_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2)))", + "MetricThreshold": "tma_remote_cache > 0.05 & (tma_mem_latency > 0.1 & (tma_l3_miss_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2)))", "PublicDescription": "This metric estimates fraction of cycles while the memory subsystem was handling loads from remote cache in other sockets including synchronizations issues. This is caused often due to non-optimal NUMA allocations. #link to NUMA article. Sample with: MEM_LOAD_L3_MISS_RETIRED.REMOTE_HITM_PS;MEM_LOAD_L3_MISS_RETIRED.REMOTE_FWD_PS. Related metrics: tma_bottleneck_memory_synchronization, tma_contested_accesses, tma_data_sharing, tma_false_sharing, tma_machine_clears", "ScaleUnit": "100%" }, { "BriefDescription": "This metric estimates fraction of cycles while the memory subsystem was handling loads from remote memory", - "MetricExpr": "MEM_LOAD_L3_MISS_RETIRED.REMOTE_DRAM * PEBS * (1 + MEM_LOAD_RETIRED.FB_HIT / MEM_LOAD_RETIRED.L1_MISS / 2) / tma_info_thread_clks", + "MetricExpr": "MEM_LOAD_L3_MISS_RETIRED.REMOTE_DRAM * MEM_LOAD_L3_MISS_RETIRED.REMOTE_DRAM:R * (1 + MEM_LOAD_RETIRED.FB_HIT / MEM_LOAD_RETIRED.L1_MISS / 2) / tma_info_thread_clks", "MetricGroup": "Server;Snoop;TopdownL5;tma_L5_group;tma_mem_latency_group", "MetricName": "tma_remote_mem", - "MetricThreshold": "tma_remote_mem > 0.1 & (tma_mem_latency > 0.1 & (tma_dram_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2)))", + "MetricThreshold": "tma_remote_mem > 0.1 & (tma_mem_latency > 0.1 & (tma_l3_miss_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2)))", "PublicDescription": "This metric estimates fraction of cycles while the memory subsystem was handling loads from remote memory. This is caused often due to non-optimal NUMA allocations. #link to NUMA article. Sample with: MEM_LOAD_L3_MISS_RETIRED.REMOTE_DRAM_PS", "ScaleUnit": "100%" }, { "BriefDescription": "This metric represents fraction of cycles the CPU was stalled due to retired misprediction by (indirect) RET instructions.", - "MetricExpr": "BR_MISP_RETIRED.RET_COST * BR_MISP_RETIRED.RET_COST:R / tma_info_thread_clks", + "MetricExpr": "BR_MISP_RETIRED.RET / BR_MISP_RETIRED.ALL_BRANCHES * tma_branch_mispredicts", "MetricGroup": "BrMispredicts;TopdownL3;tma_L3_group;tma_branch_mispredicts_group", "MetricName": "tma_ret_mispredicts", "MetricThreshold": "tma_ret_mispredicts > 0.05 & (tma_branch_mispredicts > 0.1 & tma_bad_speculation > 0.15)",
@@ -2283,7 +2285,7 @@ "MetricGroup": "BvMB;MemoryBW;Offcore;TopdownL4;tma_L4_group;tma_issueBW;tma_l3_bound_group", "MetricName": "tma_sq_full", "MetricThreshold": "tma_sq_full > 0.3 & (tma_l3_bound > 0.05 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))", - "PublicDescription": "This metric measures fraction of cycles where the Super Queue (SQ) was full taking into account all request-types and both hardware SMT threads (Logical Processors). Related metrics: tma_bottleneck_data_cache_memory_bandwidth, tma_fb_full, tma_info_system_dram_bw_use, tma_mem_bandwidth", + "PublicDescription": "This metric measures fraction of cycles where the Super Queue (SQ) was full taking into account all request-types and both hardware SMT threads (Logical Processors). Related metrics: tma_bottleneck_data_cache_memory_bandwidth, tma_fb_full, tma_info_system_dram_bw_use, tma_mem_bandwidth, tma_uc_bound", "ScaleUnit": "100%" }, {
tools/perf/pmu-events/arch/x86/graniterapids/metricgroups.json +1 -1
@@ -90,7 +90,6 @@ "tma_code_stlb_miss_group": "Metrics contributing to tma_code_stlb_miss category", "tma_core_bound_group": "Metrics contributing to tma_core_bound category", "tma_divider_group": "Metrics contributing to tma_divider category", - "tma_dram_bound_group": "Metrics contributing to tma_dram_bound category", "tma_dtlb_load_group": "Metrics contributing to tma_dtlb_load category", "tma_dtlb_store_group": "Metrics contributing to tma_dtlb_store category", "tma_fetch_bandwidth_group": "Metrics contributing to tma_fetch_bandwidth category",
@@ -124,6 +123,7 @@ "tma_l1_bound_group": "Metrics contributing to tma_l1_bound category", "tma_l2_bound_group": "Metrics contributing to tma_l2_bound category", "tma_l3_bound_group": "Metrics contributing to tma_l3_bound category", + "tma_l3_miss_bound_group": "Metrics contributing to tma_l3_miss_bound category", "tma_light_operations_group": "Metrics contributing to tma_light_operations category", "tma_load_op_utilization_group": "Metrics contributing to tma_load_op_utilization category", "tma_load_stlb_miss_group": "Metrics contributing to tma_load_stlb_miss category",
tools/perf/pmu-events/arch/x86/graniterapids/pipeline.json +1 -1
@@ -559,7 +559,7 @@ "Counter": "0,1,2,3,4,5,6,7", "EventCode": "0xc0", "EventName": "INST_RETIRED.REP_ITERATION", - "PublicDescription": "Number of iterations of Repeat (REP) string retired instructions such as MOVS, CMPS, and SCAS. Each has a byte, word, and doubleword version and string instructions can be repeated using a repetition prefix, REP, that allows their architectural execution to be repeated a number of times as specified by the RCX register. Note the number of iterations is implementation-dependent.", + "PublicDescription": "Number of iterations of Repeat (REP) string retired instructions such as MOVS, CMPS, and SCAS. Each has a byte, word, and doubleword version and string instructions can be repeated using a repetition prefix, REP, that allows their architectural execution to be repeated a number of times as specified by the RCX register. Note: Since the number of iterations within a REP instruction can be significantly affected by fast strings, this event may vary run to run and not match the architectural number of iterations (specified by RCX)", "SampleAfterValue": "2000003", "UMask": "0x8" },
tools/perf/pmu-events/arch/x86/mapfile.csv +1 -1
@@ -13,7 +13,7 @@ GenuineIntel-6-CF,v1.24,emeraldrapids,core GenuineIntel-6-5[CF],v13,goldmont,core GenuineIntel-6-7A,v1.01,goldmontplus,core GenuineIntel-6-B6,v1.13,grandridge,core -GenuineIntel-6-A[DE],v1.19,graniterapids,core +GenuineIntel-6-A[DE],v1.20,graniterapids,core GenuineIntel-6-(3C|45|46),v36,haswell,core GenuineIntel-6-3F,v29,haswellx,core GenuineIntel-6-7[DE],v1.24,icelake,core
Signature · 1 line
2.55.0.795.g602f6c329a-goog
PATCH 7/12 perf vendor events intel: Update lunarlake events from 1.25 to 1.26
In reply to Chun-Tse ShaoThe updated events were published in:
https://github.com/intel/perfmon/commit/2ba9dec72a771ed3ce7114d6a16797131871fd61
--- .../arch/x86/lunarlake/lnl-metrics.json | 142 ++++++++++-------- .../arch/x86/lunarlake/metricgroups.json | 2 +- .../arch/x86/lunarlake/pipeline.json | 2 +- tools/perf/pmu-events/arch/x86/mapfile.csv | 2 +- 4 files changed, 83 insertions(+), 65 deletions(-)
tools/perf/pmu-events/arch/x86/lunarlake/lnl-metrics.json +80 -62
@@ -783,16 +783,16 @@ }, { "BriefDescription": "Total pipeline cost of external Memory- or Cache-Bandwidth related bottlenecks", - "MetricExpr": "100 * (tma_memory_bound * (tma_dram_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_mem_bandwidth / (tma_mem_bandwidth + tma_mem_latency)) + tma_memory_bound * (tma_l3_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_sq_full / (tma_contested_accesses + tma_data_sharing + tma_l3_hit_latency + tma_sq_full)) + tma_memory_bound * (tma_l1_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_fb_full / (tma_dtlb_load + tma_fb_full + tma_l1_latency_capacity + tma_l1_latency_dependency + tma_lock_latency + tma_split_loads + tma_store_early_blk + tma_store_fwd_blk)))", + "MetricExpr": "100 * (tma_memory_bound * (tma_l3_miss_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_mem_bandwidth / (tma_mem_bandwidth + tma_mem_latency + tma_uc_bound)) + tma_memory_bound * (tma_l3_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_sq_full / (tma_contested_accesses + tma_data_sharing + tma_l3_hit_latency + tma_sq_full)) + tma_memory_bound * (tma_l1_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_fb_full / (tma_dtlb_load + tma_fb_full + tma_l1_latency_capacity + tma_lock_latency + tma_split_loads + tma_store_early_blk + tma_store_fwd_blk)))", "MetricGroup": "BvMB;Mem;MemoryBW;Offcore;tma_issueBW", "MetricName": "tma_bottleneck_data_cache_memory_bandwidth", "MetricThreshold": "tma_bottleneck_data_cache_memory_bandwidth > 20", - "PublicDescription": "Total pipeline cost of external Memory- or Cache-Bandwidth related bottlenecks. Related metrics: tma_fb_full, tma_info_system_dram_bw_use, tma_mem_bandwidth, tma_sq_full", + "PublicDescription": "Total pipeline cost of external Memory- or Cache-Bandwidth related bottlenecks. Related metrics: tma_fb_full, tma_info_system_dram_bw_use, tma_mem_bandwidth, tma_sq_full, tma_uc_bound", "Unit": "cpu_core" }, { "BriefDescription": "Total pipeline cost of external Memory- or Cache-Latency related bottlenecks", - "MetricExpr": "100 * (tma_memory_bound * (tma_dram_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_mem_latency / (tma_mem_bandwidth + tma_mem_latency)) + tma_memory_bound * (tma_l3_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_l3_hit_latency / (tma_contested_accesses + tma_data_sharing + tma_l3_hit_latency + tma_sq_full)) + tma_memory_bound * tma_l2_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound) + tma_memory_bound * (tma_l1_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_l1_latency_dependency / (tma_dtlb_load + tma_fb_full + tma_l1_latency_capacity + tma_l1_latency_dependency + tma_lock_latency + tma_split_loads + tma_store_early_blk + tma_store_fwd_blk)) + tma_memory_bound * (tma_l1_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_l1_latency_capacity / (tma_dtlb_load + tma_fb_full + tma_l1_latency_capacity + tma_l1_latency_dependency + tma_lock_latency + tma_split_loads + tma_store_early_blk + tma_store_fwd_blk)) + tma_memory_bound * (tma_l1_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_lock_latency / (tma_dtlb_load + tma_fb_full + tma_l1_latency_capacity + tma_l1_latency_dependency + tma_lock_latency + tma_split_loads + tma_store_early_blk + tma_store_fwd_blk)) + tma_memory_bound * (tma_l1_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_split_loads / (tma_dtlb_load + tma_fb_full + tma_l1_latency_capacity + tma_l1_latency_dependency + tma_lock_latency + tma_split_loads + tma_store_early_blk + tma_store_fwd_blk)) + tma_memory_bound * (tma_store_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_split_stores / (tma_dtlb_store + tma_false_sharing + tma_split_stores + tma_store_latency + tma_streaming_stores)) + tma_memory_bound * (tma_store_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_store_latency / (tma_dtlb_store + tma_false_sharing + tma_split_stores + tma_store_latency + tma_streaming_stores)))", + "MetricExpr": "100 * (tma_memory_bound * (tma_l3_miss_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_mem_latency / (tma_mem_bandwidth + tma_mem_latency + tma_uc_bound)) + tma_memory_bound * (tma_l3_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_l3_hit_latency / (tma_contested_accesses + tma_data_sharing + tma_l3_hit_latency + tma_sq_full)) + tma_memory_bound * tma_l2_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound) + tma_memory_bound * (tma_l1_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_l1_latency_capacity / (tma_dtlb_load + tma_fb_full + tma_l1_latency_capacity + tma_lock_latency + tma_split_loads + tma_store_early_blk + tma_store_fwd_blk)) + tma_memory_bound * (tma_l1_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_lock_latency / (tma_dtlb_load + tma_fb_full + tma_l1_latency_capacity + tma_lock_latency + tma_split_loads + tma_store_early_blk + tma_store_fwd_blk)) + tma_memory_bound * (tma_l1_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_split_loads / (tma_dtlb_load + tma_fb_full + tma_l1_latency_capacity + tma_lock_latency + tma_split_loads + tma_store_early_blk + tma_store_fwd_blk)) + tma_memory_bound * (tma_store_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_split_stores / (tma_dtlb_store + tma_false_sharing + tma_split_stores + tma_store_latency + tma_streaming_stores)) + tma_memory_bound * (tma_store_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_store_latency / (tma_dtlb_store + tma_false_sharing + tma_split_stores + tma_store_latency + tma_streaming_stores)))", "MetricGroup": "BvML;Mem;MemoryLat;Offcore;tma_issueLat", "MetricName": "tma_bottleneck_data_cache_memory_latency", "MetricThreshold": "tma_bottleneck_data_cache_memory_latency > 20",
@@ -818,7 +818,7 @@ }, { "BriefDescription": "Total pipeline cost of Memory Address Translation related bottlenecks (data-side TLBs)", - "MetricExpr": "100 * (tma_memory_bound * (tma_l1_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_dtlb_load / (tma_dtlb_load + tma_fb_full + tma_l1_latency_capacity + tma_l1_latency_dependency + tma_lock_latency + tma_split_loads + tma_store_early_blk + tma_store_fwd_blk)) + tma_memory_bound * (tma_store_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_dtlb_store / (tma_dtlb_store + tma_false_sharing + tma_split_stores + tma_store_latency + tma_streaming_stores)))", + "MetricExpr": "100 * (tma_memory_bound * (tma_l1_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_dtlb_load / (tma_dtlb_load + tma_fb_full + tma_l1_latency_capacity + tma_lock_latency + tma_split_loads + tma_store_early_blk + tma_store_fwd_blk)) + tma_memory_bound * (tma_store_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_dtlb_store / (tma_dtlb_store + tma_false_sharing + tma_split_stores + tma_store_latency + tma_streaming_stores)))", "MetricGroup": "BvMT;Mem;MemoryTLB;Offcore;tma_issueTLB", "MetricName": "tma_bottleneck_memory_data_tlbs", "MetricThreshold": "tma_bottleneck_memory_data_tlbs > 20",
@@ -827,7 +827,7 @@ }, { "BriefDescription": "Total pipeline cost of Memory Synchronization related bottlenecks (data transfers and coherency updates across processors)", - "MetricExpr": "100 * (tma_memory_bound * (tma_l3_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound) * (tma_contested_accesses + tma_data_sharing) / (tma_contested_accesses + tma_data_sharing + tma_l3_hit_latency + tma_sq_full) + tma_store_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound) * tma_false_sharing / (tma_dtlb_store + tma_false_sharing + tma_split_stores + tma_store_latency + tma_streaming_stores - tma_store_latency)) + tma_machine_clears * (1 - tma_other_nukes / tma_other_nukes))", + "MetricExpr": "100 * (tma_memory_bound * (tma_l3_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound) * (tma_contested_accesses + tma_data_sharing) / (tma_contested_accesses + tma_data_sharing + tma_l3_hit_latency + tma_sq_full) + tma_store_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound) * tma_false_sharing / (tma_dtlb_store + tma_false_sharing + tma_split_stores + tma_store_latency + tma_streaming_stores - tma_store_latency)) + tma_machine_clears * (1 - tma_other_nukes / tma_other_nukes))", "MetricGroup": "BvMS;LockCont;Mem;Offcore;tma_issueSyncxn", "MetricName": "tma_bottleneck_memory_synchronization", "MetricThreshold": "tma_bottleneck_memory_synchronization > 10",
@@ -975,7 +975,7 @@ }, { "BriefDescription": "This metric represents fraction of cycles the CPU was stalled due to retired misprediction by non-taken conditional branches.", - "MetricExpr": "cpu_core@BR_MISP_RETIRED.COND_NTAKEN_COST@ * cpu_core@BR_MISP_RETIRED.COND_NTAKEN_COST@R / tma_info_thread_clks", + "MetricExpr": "cpu_core@BR_MISP_RETIRED.COND_NTAKEN@ / cpu_core@BR_MISP_RETIRED.ALL_BRANCHES@ * tma_branch_mispredicts", "MetricGroup": "BrMispredicts;TopdownL3;tma_L3_group;tma_branch_mispredicts_group", "MetricName": "tma_cond_nt_mispredicts", "MetricThreshold": "tma_cond_nt_mispredicts > 0.05 & (tma_branch_mispredicts > 0.1 & tma_bad_speculation > 0.15)",
@@ -984,7 +984,7 @@ }, { "BriefDescription": "This metric represents fraction of cycles the CPU was stalled due to misprediction by backward-taken conditional branches.", - "MetricExpr": "cpu_core@BR_MISP_RETIRED.COND_TAKEN_BWD_COST@ * cpu_core@BR_MISP_RETIRED.COND_TAKEN_BWD_COST@R / tma_info_thread_clks", + "MetricExpr": "cpu_core@BR_MISP_RETIRED.COND_TAKEN_BWD@ / cpu_core@BR_MISP_RETIRED.ALL_BRANCHES@ * tma_branch_mispredicts", "MetricGroup": "BrMispredicts;TopdownL3;tma_L3_group;tma_branch_mispredicts_group", "MetricName": "tma_cond_tk_bwd_mispredicts", "MetricThreshold": "tma_cond_tk_bwd_mispredicts > 0.05 & (tma_branch_mispredicts > 0.1 & tma_bad_speculation > 0.15)",
@@ -993,7 +993,7 @@ }, { "BriefDescription": "This metric represents fraction of cycles the CPU was stalled due to misprediction by forward-taken conditional branches.", - "MetricExpr": "cpu_core@BR_MISP_RETIRED.COND_TAKEN_FWD_COST@ * cpu_core@BR_MISP_RETIRED.COND_TAKEN_FWD_COST@R / tma_info_thread_clks", + "MetricExpr": "cpu_core@BR_MISP_RETIRED.COND_TAKEN_FWD@ / cpu_core@BR_MISP_RETIRED.ALL_BRANCHES@ * tma_branch_mispredicts", "MetricGroup": "BrMispredicts;TopdownL3;tma_L3_group;tma_branch_mispredicts_group", "MetricName": "tma_cond_tk_fwd_mispredicts", "MetricThreshold": "tma_cond_tk_fwd_mispredicts > 0.05 & (tma_branch_mispredicts > 0.1 & tma_bad_speculation > 0.15)",
@@ -1022,13 +1022,13 @@ "Unit": "cpu_core" }, { - "BriefDescription": "This metric estimates fraction of cycles while the memory subsystem was handling synchronizations due to data-sharing accesses", + "BriefDescription": "This metric estimates fraction of cycles while the memory subsystem was handling synchronizations due to L3 data-sharing accesses", "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "(cpu_core@MEM_LOAD_L3_HIT_RETIRED.XSNP_NO_FWD@ * min(cpu_core@MEM_LOAD_L3_HIT_RETIRED.XSNP_NO_FWD@R, 24 * tma_info_system_core_frequency) + cpu_core@MEM_LOAD_L3_HIT_RETIRED.XSNP_FWD@ * min(cpu_core@MEM_LOAD_L3_HIT_RETIRED.XSNP_FWD@R, 25 * tma_info_system_core_frequency)) * (1 + cpu_core@MEM_LOAD_RETIRED.FB_HIT@ / cpu_core@MEM_LOAD_RETIRED.L1_MISS@ / 2) / tma_info_thread_clks", "MetricGroup": "BvMS;Offcore;Snoop;TopdownL4;tma_L4_group;tma_issueSyncxn;tma_l3_bound_group", "MetricName": "tma_data_sharing", "MetricThreshold": "tma_data_sharing > 0.05 & (tma_l3_bound > 0.05 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))", - "PublicDescription": "This metric estimates fraction of cycles while the memory subsystem was handling synchronizations due to data-sharing accesses. Data shared by multiple Logical Processors (even just read shared) may cause increased access latency due to cache coherency. Excessive data sharing can drastically harm multithreaded performance. Sample with: MEM_LOAD_L3_HIT_RETIRED.XSNP_NO_FWD. Related metrics: tma_bottleneck_memory_synchronization, tma_contested_accesses, tma_false_sharing, tma_machine_clears, tma_remote_cache", + "PublicDescription": "This metric estimates fraction of cycles while the memory subsystem was handling synchronizations due to L3 data-sharing accesses. Data shared by multiple Logical Processors (even just read shared) may cause increased access latency due to cache coherency. Excessive data sharing can drastically harm multithreaded performance. Sample with: MEM_LOAD_L3_HIT_RETIRED.XSNP_NO_FWD. Related metrics: tma_bottleneck_memory_synchronization, tma_contested_accesses, tma_false_sharing, tma_machine_clears, tma_remote_cache", "ScaleUnit": "100%", "Unit": "cpu_core" },
@@ -1042,19 +1042,9 @@ "ScaleUnit": "100%", "Unit": "cpu_core" }, - { - "BriefDescription": "This metric estimates how often the CPU was stalled on accesses to external memory (DRAM) by loads", - "MetricExpr": "cpu_core@MEMORY_STALLS.MEM@ / tma_info_thread_clks", - "MetricGroup": "MemoryBound;TmaL3mem;TopdownL3;tma_L3_group;tma_memory_bound_group", - "MetricName": "tma_dram_bound", - "MetricThreshold": "tma_dram_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2)", - "PublicDescription": "This metric estimates how often the CPU was stalled on accesses to external memory (DRAM) by loads. Better caching can improve the latency and increase performance. Sample with: MEM_LOAD_RETIRED.L3_MISS", - "ScaleUnit": "100%", - "Unit": "cpu_core" - }, { "BriefDescription": "This metric represents Core fraction of cycles in which CPU was likely limited due to DSB (decoded uop cache) fetch pipeline", - "MetricExpr": "([email protected]_UOPS\\,cmask\\=0x8\\,inv\\=0x1@ / 2 + [email protected]_UOPS@ / ([email protected]_UOPS@ + [email protected]_UOPS@) * (cpu_core@IDQ_BUBBLES.STARVATION_CYCLES@ - cpu_core@IDQ_BUBBLES.FETCH_LATENCY@)) / tma_info_thread_clks", + "MetricExpr": "([email protected]_UOPS\\,cmask\\=0x8\\,inv\\=0x1@ * max(0, 8 - tma_info_pipeline_fetch_dsb) / 8 + [email protected]_UOPS@ / ([email protected]_UOPS@ + [email protected]_UOPS@) * (cpu_core@IDQ_BUBBLES.STARVATION_CYCLES@ - cpu_core@IDQ_BUBBLES.FETCH_LATENCY@)) / tma_info_thread_clks", "MetricGroup": "DSB;FetchBW;TopdownL3;tma_L3_group;tma_fetch_bandwidth_group", "MetricName": "tma_dsb", "MetricThreshold": "tma_dsb > 0.15 & tma_fetch_bandwidth > 0.2",
@@ -1108,7 +1098,7 @@ "MetricGroup": "BvMB;MemoryBW;TopdownL4;tma_L4_group;tma_issueBW;tma_issueSL;tma_issueSmSt;tma_l1_bound_group", "MetricName": "tma_fb_full", "MetricThreshold": "tma_fb_full > 0.3", - "PublicDescription": "This metric does a *rough estimation* of how often L1D Fill Buffer unavailability limited additional L1D miss memory access requests to proceed. The higher the metric value; the deeper the memory hierarchy level the misses are satisfied from (metric values >1 are valid). Often it hints on approaching bandwidth limits (to L2 cache; L3 cache or external memory). Related metrics: tma_bottleneck_data_cache_memory_bandwidth, tma_info_system_dram_bw_use, tma_mem_bandwidth, tma_sq_full, tma_store_latency, tma_streaming_stores", + "PublicDescription": "This metric does a *rough estimation* of how often L1D Fill Buffer unavailability limited additional L1D miss memory access requests to proceed. The higher the metric value; the deeper the memory hierarchy level the misses are satisfied from (metric values >1 are valid). Often it hints on approaching bandwidth limits (to L2 cache; L3 cache or external memory). Related metrics: tma_bottleneck_data_cache_memory_bandwidth, tma_info_system_dram_bw_use, tma_mem_bandwidth, tma_sq_full, tma_store_latency, tma_streaming_stores, tma_uc_bound", "ScaleUnit": "100%", "Unit": "cpu_core" },
@@ -1205,7 +1195,7 @@ }, { "BriefDescription": "This metric approximates arithmetic FP vector uops fraction the CPU has retired for 256-bit wide vectors", - "MetricExpr": "cpu_core@FP_ARITH_OPS_RETIRED.VECTOR\\,umask\\=0x30@ / (tma_retiring * tma_info_thread_slots)", + "MetricExpr": "cpu_core@FP_ARITH_OPS_RETIRED.VECTOR_256B@ / (tma_retiring * tma_info_thread_slots)", "MetricGroup": "Compute;Flops;TopdownL5;tma_L5_group;tma_fp_vector_group;tma_issue2P", "MetricName": "tma_fp_vector_256b", "MetricThreshold": "tma_fp_vector_256b > 0.1 & (tma_fp_vector > 0.1 & (tma_fp_arith > 0.2 & tma_light_operations > 0.6))",
@@ -1248,7 +1238,7 @@ }, { "BriefDescription": "This metric represents fraction of cycles the CPU was stalled due to instruction cache misses", - "MetricExpr": "cpu_core@ICACHE_DATA.STALLS@ / tma_info_thread_clks", + "MetricExpr": "cpu_core@ICACHE_DATA.STALLS@ / tma_info_thread_clks + tma_isb_full", "MetricGroup": "BigFootprint;BvBC;FetchLat;IcMiss;TopdownL3;tma_L3_group;tma_fetch_latency_group", "MetricName": "tma_icache_misses", "MetricThreshold": "tma_icache_misses > 0.05 & (tma_fetch_latency > 0.1 & tma_frontend_bound > 0.15)",
@@ -1258,7 +1248,7 @@ }, { "BriefDescription": "This metric represents fraction of cycles the CPU was stalled due to retired misprediction by indirect CALL instructions.", - "MetricExpr": "cpu_core@BR_MISP_RETIRED.INDIRECT_CALL_COST@ * cpu_core@BR_MISP_RETIRED.INDIRECT_CALL_COST@R / tma_info_thread_clks", + "MetricExpr": "cpu_core@BR_MISP_RETIRED.INDIRECT_CALL@ / cpu_core@BR_MISP_RETIRED.ALL_BRANCHES@ * tma_branch_mispredicts", "MetricGroup": "BrMispredicts;TopdownL3;tma_L3_group;tma_branch_mispredicts_group", "MetricName": "tma_ind_call_mispredicts", "MetricThreshold": "tma_ind_call_mispredicts > 0.05 & (tma_branch_mispredicts > 0.1 & tma_bad_speculation > 0.15)",
@@ -1267,7 +1257,7 @@ }, { "BriefDescription": "This metric represents fraction of cycles the CPU was stalled due to retired misprediction by indirect JMP instructions.", - "MetricExpr": "max((cpu_core@BR_MISP_RETIRED.INDIRECT_COST@ * cpu_core@BR_MISP_RETIRED.INDIRECT_COST@R - cpu_core@BR_MISP_RETIRED.INDIRECT_CALL_COST@ * cpu_core@BR_MISP_RETIRED.INDIRECT_CALL_COST@R) / tma_info_thread_clks, 0)", + "MetricExpr": "max((cpu_core@BR_MISP_RETIRED.INDIRECT@ - cpu_core@BR_MISP_RETIRED.INDIRECT_CALL@) / cpu_core@BR_MISP_RETIRED.ALL_BRANCHES@ * tma_branch_mispredicts, 0)", "MetricGroup": "BrMispredicts;TopdownL3;tma_L3_group;tma_branch_mispredicts_group", "MetricName": "tma_ind_jump_mispredicts", "MetricThreshold": "tma_ind_jump_mispredicts > 0.05 & (tma_branch_mispredicts > 0.1 & tma_bad_speculation > 0.15)",
@@ -1437,7 +1427,7 @@ }, { "BriefDescription": "Fraction of Uops delivered by the DSB (aka Decoded ICache; or Uop Cache)", - "MetricExpr": "[email protected]_UOPS@ / cpu_core@UOPS_ISSUED.ANY@", + "MetricExpr": "[email protected]_UOPS@ / ([email protected]_UOPS@ + [email protected]@ + [email protected]_UOPS@ + [email protected]_UOPS@)", "MetricGroup": "DSB;Fed;FetchBW;tma_issueFB", "MetricName": "tma_info_frontend_dsb_coverage", "MetricThreshold": "tma_info_frontend_dsb_coverage < 0.7 & tma_info_thread_ipc / 8 > 0.35",
@@ -1504,7 +1494,7 @@ }, { "BriefDescription": "Fraction of Uops delivered by the LSD (Loop Stream Detector; aka Loop Cache)", - "MetricExpr": "[email protected]@ / cpu_core@UOPS_ISSUED.ANY@", + "MetricExpr": "[email protected]@ / ([email protected]_UOPS@ + [email protected]@ + [email protected]_UOPS@ + [email protected]_UOPS@)", "MetricGroup": "Fed;LSD", "MetricName": "tma_info_frontend_lsd_coverage", "Unit": "cpu_core"
@@ -1663,6 +1653,15 @@ "PublicDescription": "Instructions per taken branch. Related metrics: tma_dsb_switches, tma_fetch_bandwidth, tma_info_botlnk_l2_dsb_bandwidth, tma_info_botlnk_l2_dsb_misses, tma_info_frontend_dsb_coverage, tma_lcp", "Unit": "cpu_core" }, + { + "BriefDescription": "AVX preserve/restore assists per kilo instruction", + "MetricExpr": "1e3 * [email protected]_AVX_MIX@ / cpu_core@INST_RETIRED.ANY@", + "MetricGroup": "tma_issueMV", + "MetricName": "tma_info_inst_mix_vectormixpki", + "MetricThreshold": "tma_info_inst_mix_vectormixpki > 0.05", + "PublicDescription": "AVX preserve/restore assists per kilo instruction. Related metrics: tma_mixing_vectors, tma_ms_switches", + "Unit": "cpu_core" + }, { "BriefDescription": "Fill Buffer (FB) hits per kilo instructions for retired demand loads (L1D misses that merge into ongoing miss-handling entries)", "MetricExpr": "1e3 * cpu_core@MEM_LOAD_RETIRED.FB_HIT@ / cpu_core@INST_RETIRED.ANY@",
@@ -1929,6 +1928,13 @@ "MetricName": "tma_info_pipeline_retire", "Unit": "cpu_core" }, + { + "BriefDescription": "Total percent of cycles without actually retired uops", + "MetricExpr": "cpu_core@UOPS_RETIRED.STALLS@ / tma_info_thread_clks", + "MetricGroup": "Pipeline;Ret", + "MetricName": "tma_info_pipeline_retire_stall", + "Unit": "cpu_core" + }, { "BriefDescription": "Estimated fraction of retirement-cycles dealing with repeat instructions", "MetricExpr": "cpu_core@INST_RETIRED.REP_ITERATION@ / cpu_core@UOPS_RETIRED.SLOTS\\,cmask\\=1@",
@@ -1971,7 +1977,7 @@ "MetricExpr": "32 * UNC_M_TOTAL_DATA / 1e9 / tma_info_system_time", "MetricGroup": "HPC;MemOffcore;MemoryBW;SoC;tma_issueBW", "MetricName": "tma_info_system_dram_bw_use", - "PublicDescription": "Average external Memory Bandwidth Use for reads and writes [GB / sec]. Related metrics: tma_bottleneck_data_cache_memory_bandwidth, tma_fb_full, tma_mem_bandwidth, tma_sq_full", + "PublicDescription": "Average external Memory Bandwidth Use for reads and writes [GB / sec]. Related metrics: tma_bottleneck_data_cache_memory_bandwidth, tma_fb_full, tma_mem_bandwidth, tma_sq_full, tma_uc_bound", "Unit": "cpu_core" }, {
@@ -2042,13 +2048,6 @@ "MetricName": "tma_info_system_turbo_utilization", "Unit": "cpu_core" }, - { - "BriefDescription": "Measured Average Uncore Frequency for the SoC [GHz]", - "MetricExpr": "tma_info_system_socket_clks / 1e9 / tma_info_system_time", - "MetricGroup": "SoC", - "MetricName": "tma_info_system_uncore_frequency", - "Unit": "cpu_core" - }, { "BriefDescription": "Per-Logical Processor actual clocks when the Logical Processor is active.", "MetricExpr": "cpu_core@CPU_CLK_UNHALTED.THREAD@",
@@ -2140,9 +2139,18 @@ "ScaleUnit": "100%", "Unit": "cpu_core" }, + { + "BriefDescription": "This metric represents fraction of cycles Icache tag lookup was stalled due to Instruction Stream Buffer (ISB) being full", + "MetricExpr": "(cpu_core@ICACHE_TAG.STALLS_ISB@ + cpu_core@ICACHE_TAG.STALLS_INUSE@) / tma_info_thread_clks", + "MetricGroup": "TopdownL4;tma_L4_group;tma_icache_misses_group", + "MetricName": "tma_isb_full", + "MetricThreshold": "tma_isb_full > 0.05 & (tma_icache_misses > 0.05 & (tma_fetch_latency > 0.1 & tma_frontend_bound > 0.15))", + "ScaleUnit": "100%", + "Unit": "cpu_core" + }, { "BriefDescription": "This metric represents fraction of cycles the CPU was stalled due to Instruction TLB (ITLB) misses", - "MetricExpr": "cpu_core@ICACHE_TAG.STALLS@ / tma_info_thread_clks", + "MetricExpr": "cpu_core@ICACHE_TAG.STALLS@ / tma_info_thread_clks - tma_isb_full", "MetricGroup": "BigFootprint;BvBC;FetchLat;MemoryTLB;TopdownL3;tma_L3_group;tma_fetch_latency_group", "MetricName": "tma_itlb_misses", "MetricThreshold": "tma_itlb_misses > 0.05 & (tma_fetch_latency > 0.1 & tma_frontend_bound > 0.15)",
@@ -2169,19 +2177,9 @@ "ScaleUnit": "100%", "Unit": "cpu_core" }, - { - "BriefDescription": "This metric ([SKL+] roughly; [LNL]) estimates fraction of cycles with demand load accesses that hit the L1D cache", - "MetricExpr": "4 * cpu_core@DEPENDENT_LOADS.ANY\\,cmask\\=1@ / tma_info_thread_clks", - "MetricGroup": "BvML;MemoryLat;TopdownL4;tma_L4_group;tma_l1_bound_group", - "MetricName": "tma_l1_latency_dependency", - "MetricThreshold": "tma_l1_latency_dependency > 0.1 & (tma_l1_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))", - "PublicDescription": "This metric ([SKL+] roughly; [LNL]) estimates fraction of cycles with demand load accesses that hit the L1D cache. The short latency of the L1D cache may be exposed in pointer-chasing memory access patterns as an example. Sample with: MEM_LOAD_UOPS_RETIRED.L1_HIT_PS", - "ScaleUnit": "100%", - "Unit": "cpu_core" - }, { "BriefDescription": "This metric estimates how often the CPU was stalled due to L2 cache accesses by loads", - "MetricExpr": "cpu_core@MEMORY_STALLS.L2@ / tma_info_thread_clks", + "MetricExpr": "cpu_core@MEM_LOAD_RETIRED.L2_HIT@ / (cpu_core@MEM_LOAD_RETIRED.L2_HIT@ + cpu_core@MEM_LOAD_MISC_RETIRED.UC@) * cpu_core@MEMORY_STALLS.L2@ / tma_info_thread_clks", "MetricGroup": "BvML;CacheHits;MemoryBound;TmaL3mem;TopdownL3;tma_L3_group;tma_memory_bound_group", "MetricName": "tma_l2_bound", "MetricThreshold": "tma_l2_bound > 0.05 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2)",
@@ -2219,6 +2217,16 @@ "ScaleUnit": "100%", "Unit": "cpu_core" }, + { + "BriefDescription": "This metric estimates how often the CPU was stalled on accesses to external memory (DRAM) by loads", + "MetricExpr": "cpu_core@MEMORY_STALLS.MEM@ / tma_info_thread_clks", + "MetricGroup": "MemoryBound;Offcore;TmaL3mem;TopdownL3;tma_L3_group;tma_memory_bound_group", + "MetricName": "tma_l3_miss_bound", + "MetricThreshold": "tma_l3_miss_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2)", + "PublicDescription": "This metric estimates how often the CPU was stalled on accesses to external memory (DRAM) by loads. Better caching can improve the latency and increase performance. Sample with: MEM_LOAD_RETIRED.L3_MISS", + "ScaleUnit": "100%", + "Unit": "cpu_core" + }, { "BriefDescription": "This metric represents fraction of cycles CPU was stalled due to Length Changing Prefixes (LCPs)", "MetricExpr": "[email protected]@ / tma_info_thread_clks",
@@ -2308,7 +2316,7 @@ }, { "BriefDescription": "This metric represents Core fraction of cycles in which CPU was likely limited due to LSD (Loop Stream Detector) unit", - "MetricExpr": "[email protected]\\,cmask\\=0x8\\,inv\\=0x1@ / tma_info_thread_clks / 2", + "MetricExpr": "[email protected]\\,cmask\\=0x8\\,inv\\=0x1@ * max(0, 8 - tma_info_pipeline_fetch_lsd) / 8 / tma_info_thread_clks", "MetricGroup": "FetchBW;LSD;TopdownL3;tma_L3_group;tma_fetch_bandwidth_group", "MetricName": "tma_lsd", "MetricThreshold": "tma_lsd > 0.15 & tma_fetch_bandwidth > 0.2",
@@ -2330,19 +2338,19 @@ { "BriefDescription": "This metric estimates fraction of cycles where the core's performance was likely hurt due to approaching bandwidth limits of external memory - DRAM ([SPR-HBM] and/or HBM)", "MetricExpr": "min(cpu_core@CPU_CLK_UNHALTED.THREAD@, cpu_core@OFFCORE_REQUESTS_OUTSTANDING.DATA_RD\\,cmask\\=4@) / tma_info_thread_clks", - "MetricGroup": "BvMB;MemoryBW;Offcore;TopdownL4;tma_L4_group;tma_dram_bound_group;tma_issueBW", + "MetricGroup": "BvMB;MemoryBW;Offcore;TopdownL4;tma_L4_group;tma_issueBW;tma_l3_miss_bound_group", "MetricName": "tma_mem_bandwidth", - "MetricThreshold": "tma_mem_bandwidth > 0.2 & (tma_dram_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))", - "PublicDescription": "This metric estimates fraction of cycles where the core's performance was likely hurt due to approaching bandwidth limits of external memory - DRAM ([SPR-HBM] and/or HBM). The underlying heuristic assumes that a similar off-core traffic is generated by all IA cores. This metric does not aggregate non-data-read requests by this logical processor; requests from other IA Logical Processors/Physical Cores/sockets; or other non-IA devices like GPU; hence the maximum external memory bandwidth limits may or may not be approached when this metric is flagged (see Uncore counters for that). Related metrics: tma_bottleneck_data_cache_memory_bandwidth, tma_fb_full, tma_info_system_dram_bw_use, tma_sq_full", + "MetricThreshold": "tma_mem_bandwidth > 0.2 & (tma_l3_miss_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))", + "PublicDescription": "This metric estimates fraction of cycles where the core's performance was likely hurt due to approaching bandwidth limits of external memory - DRAM ([SPR-HBM] and/or HBM). The underlying heuristic assumes that a similar off-core traffic is generated by all IA cores. This metric does not aggregate non-data-read requests by this logical processor; requests from other IA Logical Processors/Physical Cores/sockets; or other non-IA devices like GPU; hence the maximum external memory bandwidth limits may or may not be approached when this metric is flagged (see Uncore counters for that). Related metrics: tma_bottleneck_data_cache_memory_bandwidth, tma_fb_full, tma_info_system_dram_bw_use, tma_sq_full, tma_uc_bound", "ScaleUnit": "100%", "Unit": "cpu_core" }, { "BriefDescription": "This metric estimates fraction of cycles where the performance was likely hurt due to latency from external memory - DRAM ([SPR-HBM] and/or HBM)", "MetricExpr": "min(cpu_core@CPU_CLK_UNHALTED.THREAD@, cpu_core@OFFCORE_REQUESTS_OUTSTANDING.CYCLES_WITH_DATA_RD@) / tma_info_thread_clks - tma_mem_bandwidth", - "MetricGroup": "BvML;MemoryLat;Offcore;TopdownL4;tma_L4_group;tma_dram_bound_group;tma_issueLat", + "MetricGroup": "BvML;MemoryLat;Offcore;TopdownL4;tma_L4_group;tma_issueLat;tma_l3_miss_bound_group", "MetricName": "tma_mem_latency", - "MetricThreshold": "tma_mem_latency > 0.1 & (tma_dram_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))", + "MetricThreshold": "tma_mem_latency > 0.1 & (tma_l3_miss_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))", "PublicDescription": "This metric estimates fraction of cycles where the performance was likely hurt due to latency from external memory - DRAM ([SPR-HBM] and/or HBM). This metric does not aggregate requests from other Logical Processors/Physical Cores/sockets (see Uncore counters for that). Related metrics: tma_bottleneck_data_cache_memory_latency, tma_l3_hit_latency", "ScaleUnit": "100%", "Unit": "cpu_core"
@@ -2398,7 +2406,7 @@ }, { "BriefDescription": "This metric represents Core fraction of cycles in which CPU was likely limited due to the MITE pipeline (the legacy decode pipeline)", - "MetricExpr": "([email protected]_UOPS\\,cmask\\=0x8\\,inv\\=0x1@ / 2 + [email protected]_UOPS@ / ([email protected]_UOPS@ + [email protected]_UOPS@) * (cpu_core@IDQ_BUBBLES.STARVATION_CYCLES@ - cpu_core@IDQ_BUBBLES.FETCH_LATENCY@)) / tma_info_thread_clks", + "MetricExpr": "([email protected]_UOPS\\,cmask\\=0x8\\,inv\\=0x1@ * max(0, 8 - tma_info_pipeline_fetch_mite) / 8 + [email protected]_UOPS@ / ([email protected]_UOPS@ + [email protected]_UOPS@) * (cpu_core@IDQ_BUBBLES.STARVATION_CYCLES@ - cpu_core@IDQ_BUBBLES.FETCH_LATENCY@)) / tma_info_thread_clks", "MetricGroup": "DSBmiss;FetchBW;TopdownL3;tma_L3_group;tma_fetch_bandwidth_group", "MetricName": "tma_mite", "MetricThreshold": "tma_mite > 0.1 & tma_fetch_bandwidth > 0.2",
@@ -2412,13 +2420,13 @@ "MetricGroup": "TopdownL5;tma_L5_group;tma_issueMV;tma_ports_utilized_0_group", "MetricName": "tma_mixing_vectors", "MetricThreshold": "tma_mixing_vectors > 0.05", - "PublicDescription": "This metric estimates penalty in terms of percentage of([SKL+] injected blend uops out of all Uops Issued -- the Count Domain; [ADL+] cycles). Usually a Mixing_Vectors over 5% is worth investigating. Read more in Appendix B1 of the Optimizations Guide for this topic. Related metrics: tma_ms_switches", + "PublicDescription": "This metric estimates penalty in terms of percentage of([SKL+] injected blend uops out of all Uops Issued -- the Count Domain; [ADL+] cycles). Usually a Mixing_Vectors over 5% is worth investigating. Read more in Appendix B1 of the Optimizations Guide for this topic. Related metrics: tma_info_inst_mix_vectormixpki, tma_ms_switches", "ScaleUnit": "100%", "Unit": "cpu_core" }, { "BriefDescription": "This metric represents Core fraction of cycles in which CPU was likely limited due to the Microcode Sequencer (MS) unit - see Microcode_Sequencer node for details.", - "MetricExpr": "[email protected]_CYCLES_ANY@ / tma_info_thread_clks / 1.8", + "MetricExpr": "[email protected]_CYCLES_ANY@ * (8 - [email protected]_UOPS@ / [email protected]_CYCLES_ANY@) / tma_info_thread_slots", "MetricGroup": "MicroSeq;TopdownL3;tma_L3_group;tma_fetch_bandwidth_group", "MetricName": "tma_ms", "MetricThreshold": "tma_ms > 0.05 & tma_fetch_bandwidth > 0.2",
@@ -2431,7 +2439,7 @@ "MetricGroup": "FetchLat;MicroSeq;TopdownL3;tma_L3_group;tma_fetch_latency_group;tma_issueMC;tma_issueMS;tma_issueMV;tma_issueSO", "MetricName": "tma_ms_switches", "MetricThreshold": "tma_ms_switches > 0.05 & (tma_fetch_latency > 0.1 & tma_frontend_bound > 0.15)", - "PublicDescription": "This metric estimates the fraction of cycles when the CPU was stalled due to switches of uop delivery to the Microcode Sequencer (MS). Commonly used instructions are optimized for delivery by the DSB (decoded i-cache) or MITE (legacy instruction decode) pipelines. Certain operations cannot be handled natively by the execution pipeline; and must be performed by microcode (small programs injected into the execution stream). Switching to the MS too often can negatively impact performance. The MS is designated to deliver long uop flows required by CISC instructions like CPUID; or uncommon conditions like Floating Point Assists when dealing with Denormals. Sample with: IDQ.MS_SWITCHES. Related metrics: tma_bottleneck_irregular_overhead, tma_clears_resteers, tma_l1_bound, tma_machine_clears, tma_microcode_sequencer, tma_mixing_vectors, tma_serializing_operation", + "PublicDescription": "This metric estimates the fraction of cycles when the CPU was stalled due to switches of uop delivery to the Microcode Sequencer (MS). Commonly used instructions are optimized for delivery by the DSB (decoded i-cache) or MITE (legacy instruction decode) pipelines. Certain operations cannot be handled natively by the execution pipeline; and must be performed by microcode (small programs injected into the execution stream). Switching to the MS too often can negatively impact performance. The MS is designated to deliver long uop flows required by CISC instructions like CPUID; or uncommon conditions like Floating Point Assists when dealing with Denormals. Sample with: IDQ.MS_SWITCHES. Related metrics: tma_bottleneck_irregular_overhead, tma_clears_resteers, tma_info_inst_mix_vectormixpki, tma_l1_bound, tma_machine_clears, tma_microcode_sequencer, tma_mixing_vectors, tma_serializing_operation", "ScaleUnit": "100%", "Unit": "cpu_core" },
@@ -2549,7 +2557,7 @@ }, { "BriefDescription": "This metric represents fraction of cycles the CPU was stalled due to retired misprediction by (indirect) RET instructions.", - "MetricExpr": "cpu_core@BR_MISP_RETIRED.RET_COST@ * cpu_core@BR_MISP_RETIRED.RET_COST@R / tma_info_thread_clks", + "MetricExpr": "cpu_core@BR_MISP_RETIRED.RET@ / cpu_core@BR_MISP_RETIRED.ALL_BRANCHES@ * tma_branch_mispredicts", "MetricGroup": "BrMispredicts;TopdownL3;tma_L3_group;tma_branch_mispredicts_group", "MetricName": "tma_ret_mispredicts", "MetricThreshold": "tma_ret_mispredicts > 0.05 & (tma_branch_mispredicts > 0.1 & tma_bad_speculation > 0.15)",
@@ -2624,7 +2632,7 @@ "MetricGroup": "BvMB;MemoryBW;Offcore;TopdownL4;tma_L4_group;tma_issueBW;tma_l3_bound_group", "MetricName": "tma_sq_full", "MetricThreshold": "tma_sq_full > 0.3 & (tma_l3_bound > 0.05 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))", - "PublicDescription": "This metric measures fraction of cycles where the Super Queue (SQ) was full taking into account all request-types and both hardware SMT threads (Logical Processors). Related metrics: tma_bottleneck_data_cache_memory_bandwidth, tma_fb_full, tma_info_system_dram_bw_use, tma_mem_bandwidth", + "PublicDescription": "This metric measures fraction of cycles where the Super Queue (SQ) was full taking into account all request-types and both hardware SMT threads (Logical Processors). Related metrics: tma_bottleneck_data_cache_memory_bandwidth, tma_fb_full, tma_info_system_dram_bw_use, tma_mem_bandwidth, tma_uc_bound", "ScaleUnit": "100%", "Unit": "cpu_core" },
@@ -2639,8 +2647,8 @@ "Unit": "cpu_core" }, { - "BriefDescription": "This metric estimates clocks wasted due to loads blocked due to unknown store address (did not do memory disambiguation) or due to unknown store data", - "MetricExpr": "7 * cpu_core@LD_BLOCKS.STORE_EARLY\\,cmask\\=1@ / tma_info_thread_clks", + "BriefDescription": "This metric roughly estimates clocks wasted due to loads blocked due to unknown store address (did not do memory disambiguation) or due to unknown store data", + "MetricExpr": "cpu_core@LD_BLOCKS.STORE_EARLY\\,cmask\\=1@ / tma_info_thread_clks", "MetricGroup": "TopdownL4;tma_L4_group;tma_l1_bound_group", "MetricName": "tma_store_early_blk", "MetricThreshold": "tma_store_early_blk > 0.2",
@@ -2731,6 +2739,16 @@ "ScaleUnit": "100%", "Unit": "cpu_core" }, + { + "BriefDescription": "This metric roughly estimates fraction of cycles where the core's performance was likely hurt due to un-chacheable memory", + "MetricExpr": "cpu_core@MEM_LOAD_MISC_RETIRED.UC@ / (cpu_core@MEM_LOAD_RETIRED.L2_HIT@ + cpu_core@MEM_LOAD_MISC_RETIRED.UC@) * cpu_core@MEMORY_STALLS.L2@ / tma_info_thread_clks", + "MetricGroup": "BvMB;MemoryBW;Offcore;TopdownL4;tma_L4_group;tma_issueBW;tma_l3_miss_bound_group", + "MetricName": "tma_uc_bound", + "MetricThreshold": "tma_uc_bound > 0.2 & (tma_l3_miss_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))", + "PublicDescription": "This metric roughly estimates fraction of cycles where the core's performance was likely hurt due to un-chacheable memory. Related metrics: tma_bottleneck_data_cache_memory_bandwidth, tma_fb_full, tma_info_system_dram_bw_use, tma_mem_bandwidth, tma_sq_full", + "ScaleUnit": "100%", + "Unit": "cpu_core" + }, { "BriefDescription": "This metric represents fraction of cycles the CPU was stalled due to new branch address clears", "MetricExpr": "cpu_core@INT_MISC.UNKNOWN_BRANCH_CYCLES@ / tma_info_thread_clks",
tools/perf/pmu-events/arch/x86/lunarlake/metricgroups.json +1 -1
@@ -93,7 +93,6 @@ "tma_code_stlb_miss_group": "Metrics contributing to tma_code_stlb_miss category", "tma_core_bound_group": "Metrics contributing to tma_core_bound category", "tma_divider_group": "Metrics contributing to tma_divider category", - "tma_dram_bound_group": "Metrics contributing to tma_dram_bound category", "tma_dtlb_load_group": "Metrics contributing to tma_dtlb_load category", "tma_dtlb_store_group": "Metrics contributing to tma_dtlb_store category", "tma_fetch_bandwidth_group": "Metrics contributing to tma_fetch_bandwidth category",
@@ -129,6 +128,7 @@ "tma_l1_bound_group": "Metrics contributing to tma_l1_bound category", "tma_l2_bound_group": "Metrics contributing to tma_l2_bound category", "tma_l3_bound_group": "Metrics contributing to tma_l3_bound category", + "tma_l3_miss_bound_group": "Metrics contributing to tma_l3_miss_bound category", "tma_light_operations_group": "Metrics contributing to tma_light_operations category", "tma_load_op_utilization_group": "Metrics contributing to tma_load_op_utilization category", "tma_load_stlb_miss_group": "Metrics contributing to tma_load_stlb_miss category",
tools/perf/pmu-events/arch/x86/lunarlake/pipeline.json +1 -1
@@ -973,7 +973,7 @@ "Counter": "0,1,2,3,4,5,6,7,8,9", "EventCode": "0xc0", "EventName": "INST_RETIRED.REP_ITERATION", - "PublicDescription": "Number of iterations of Repeat (REP) string retired instructions such as MOVS, CMPS, and SCAS. Each has a byte, word, and doubleword version and string instructions can be repeated using a repetition prefix, REP, that allows their architectural execution to be repeated a number of times as specified by the RCX register. Note the number of iterations is implementation-dependent. Available PDIST counters: 0,1", + "PublicDescription": "Number of iterations of Repeat (REP) string retired instructions such as MOVS, CMPS, and SCAS. Each has a byte, word, and doubleword version and string instructions can be repeated using a repetition prefix, REP, that allows their architectural execution to be repeated a number of times as specified by the RCX register. Note: Since the number of iterations within a REP instruction can be significantly affected by fast strings, this event may vary run to run and not match the architectural number of iterations (specified by RCX) Available PDIST counters: 0,1", "SampleAfterValue": "2000003", "UMask": "0x8", "Unit": "cpu_core"
tools/perf/pmu-events/arch/x86/mapfile.csv +1 -1
@@ -22,7 +22,7 @@ GenuineIntel-6-3A,v24,ivybridge,core GenuineIntel-6-3E,v24,ivytown,core GenuineIntel-6-2D,v24,jaketown,core GenuineIntel-6-(57|85),v16,knightslanding,core -GenuineIntel-6-BD,v1.25,lunarlake,core +GenuineIntel-6-BD,v1.26,lunarlake,core GenuineIntel-6-(AA|AC|B5),v1.21,meteorlake,core GenuineIntel-6-1[AEF],v4,nehalemep,core GenuineIntel-6-2E,v4,nehalemex,core
Signature · 1 line
2.55.0.795.g602f6c329a-goog
PATCH 8/12 perf vendor events intel: Update meteorlake events from 1.21 to 1.22
In reply to Chun-Tse ShaoThe updated events were published in:
https://github.com/intel/perfmon/commit/704ef43e4c0738065a0575622cf7d31867b8d48b
--- tools/perf/pmu-events/arch/x86/mapfile.csv | 2 +- .../pmu-events/arch/x86/meteorlake/cache.json | 11 ++- .../arch/x86/meteorlake/metricgroups.json | 2 +- .../arch/x86/meteorlake/mtl-metrics.json | 88 ++++++++++--------- .../arch/x86/meteorlake/pipeline.json | 4 +- .../arch/x86/meteorlake/uncore-memory.json | 84 +++++++++++++++--- 6 files changed, 131 insertions(+), 60 deletions(-)
tools/perf/pmu-events/arch/x86/mapfile.csv +1 -1
@@ -23,7 +23,7 @@ GenuineIntel-6-3E,v24,ivytown,core GenuineIntel-6-2D,v24,jaketown,core GenuineIntel-6-(57|85),v16,knightslanding,core GenuineIntel-6-BD,v1.26,lunarlake,core -GenuineIntel-6-(AA|AC|B5),v1.21,meteorlake,core +GenuineIntel-6-(AA|AC|B5),v1.22,meteorlake,core GenuineIntel-6-1[AEF],v4,nehalemep,core GenuineIntel-6-2E,v4,nehalemex,core GenuineIntel-6-(CC|D5),v1.06,pantherlake,core
tools/perf/pmu-events/arch/x86/meteorlake/cache.json +10 -1
@@ -1,6 +1,6 @@ [ { - "BriefDescription": "Counts the number of request that were not accepted into the L2Q because the L2Q is FULL.", + "BriefDescription": "Counts the number of requests that were not accepted into the L2Q because the L2Q is FULL.", "Counter": "0,1,2,3,4,5,6,7", "EventCode": "0x31", "EventName": "CORE_REJECT_L2Q.ANY",
@@ -631,6 +631,15 @@ "UMask": "0x50", "Unit": "cpu_atom" }, + { + "BriefDescription": "Counts the number of unhalted cycles when the core is stalled due to a demand load miss which missed all the caches, a snoop was required, and hits in other core or module on the same die. Another core provides the data with a FWD, NO_FWD, or HITM. If the core has access to an L3 cache, an LLC miss refers to an L3 cache miss, otherwise it is an L2 cache miss.", + "Counter": "0,1,2,3,4,5,6,7", + "EventCode": "0x34", + "EventName": "MEM_BOUND_STALLS_LOAD.LLC_MISS_OTHERMOD", + "SampleAfterValue": "1000003", + "UMask": "0x8", + "Unit": "cpu_atom" + }, { "BriefDescription": "Counts the number of unhalted cycles when the core is stalled to a store buffer full condition", "Counter": "0,1,2,3,4,5,6,7",
tools/perf/pmu-events/arch/x86/meteorlake/metricgroups.json +1 -1
@@ -93,7 +93,6 @@ "tma_code_stlb_miss_group": "Metrics contributing to tma_code_stlb_miss category", "tma_core_bound_group": "Metrics contributing to tma_core_bound category", "tma_divider_group": "Metrics contributing to tma_divider category", - "tma_dram_bound_group": "Metrics contributing to tma_dram_bound category", "tma_dtlb_load_group": "Metrics contributing to tma_dtlb_load category", "tma_dtlb_store_group": "Metrics contributing to tma_dtlb_store category", "tma_fetch_bandwidth_group": "Metrics contributing to tma_fetch_bandwidth category",
@@ -129,6 +128,7 @@ "tma_l1_bound_group": "Metrics contributing to tma_l1_bound category", "tma_l2_bound_group": "Metrics contributing to tma_l2_bound category", "tma_l3_bound_group": "Metrics contributing to tma_l3_bound category", + "tma_l3_miss_bound_group": "Metrics contributing to tma_l3_miss_bound category", "tma_light_operations_group": "Metrics contributing to tma_light_operations category", "tma_load_op_utilization_group": "Metrics contributing to tma_load_op_utilization category", "tma_load_stlb_miss_group": "Metrics contributing to tma_load_stlb_miss category",
tools/perf/pmu-events/arch/x86/meteorlake/mtl-metrics.json +45 -43
242 lines not on this page · raw message
tools/perf/pmu-events/arch/x86/meteorlake/pipeline.json +2 -2
@@ -909,7 +909,7 @@ "Counter": "0,1,2,3,4,5,6,7", "EventCode": "0xc0", "EventName": "INST_RETIRED.REP_ITERATION", - "PublicDescription": "Number of iterations of Repeat (REP) string retired instructions such as MOVS, CMPS, and SCAS. Each has a byte, word, and doubleword version and string instructions can be repeated using a repetition prefix, REP, that allows their architectural execution to be repeated a number of times as specified by the RCX register. Note the number of iterations is implementation-dependent.", + "PublicDescription": "Number of iterations of Repeat (REP) string retired instructions such as MOVS, CMPS, and SCAS. Each has a byte, word, and doubleword version and string instructions can be repeated using a repetition prefix, REP, that allows their architectural execution to be repeated a number of times as specified by the RCX register. Note: Since the number of iterations within a REP instruction can be significantly affected by fast strings, this event may vary run to run and not match the architectural number of iterations (specified by RCX)", "SampleAfterValue": "2000003", "UMask": "0x8", "Unit": "cpu_core"
@@ -1443,7 +1443,7 @@ "Unit": "cpu_atom" }, { - "BriefDescription": "Counts the number issue slots not consumed due to a color request for an FCW or MXCSR control register when all 4 colors (copies) are already in use.", + "BriefDescription": "Counts the number of issue slots not consumed due to a color request for an FCW or MXCSR control register when all 4 colors (copies) are already in use.", "Counter": "0,1,2,3,4,5,6,7", "EventCode": "0x75", "EventName": "SERIALIZATION.COLOR_STALLS",
tools/perf/pmu-events/arch/x86/meteorlake/uncore-memory.json +72 -12
139 lines not on this page · raw message
Signature · 1 line
2.55.0.795.g602f6c329a-goog
PATCH 9/12 perf vendor events intel: Add novalake v1.00 events
In reply to Chun-Tse ShaoThe updated events were published in:
https://github.com/intel/perfmon/commit/3aa49b06346e3a3ff40c7beabe63585591200c58
--- tools/perf/pmu-events/arch/x86/mapfile.csv | 1 + .../pmu-events/arch/x86/novalake/cache.json | 749 +++++++ .../pmu-events/arch/x86/novalake/counter.json | 17 + .../arch/x86/novalake/floating-point.json | 452 ++++ .../arch/x86/novalake/frontend.json | 600 +++++ .../pmu-events/arch/x86/novalake/memory.json | 299 +++ .../pmu-events/arch/x86/novalake/other.json | 144 ++ .../arch/x86/novalake/pipeline.json | 1976 +++++++++++++++++ .../arch/x86/novalake/uncore-memory.json | 42 + .../arch/x86/novalake/virtual-memory.json | 338 +++ 10 files changed, 4618 insertions(+) create mode 100644 tools/perf/pmu-events/arch/x86/novalake/cache.json create mode 100644 tools/perf/pmu-events/arch/x86/novalake/counter.json create mode 100644 tools/perf/pmu-events/arch/x86/novalake/floating-point.json create mode 100644 tools/perf/pmu-events/arch/x86/novalake/frontend.json create mode 100644 tools/perf/pmu-events/arch/x86/novalake/memory.json create mode 100644 tools/perf/pmu-events/arch/x86/novalake/other.json create mode 100644 tools/perf/pmu-events/arch/x86/novalake/pipeline.json create mode 100644 tools/perf/pmu-events/arch/x86/novalake/uncore-memory.json create mode 100644 tools/perf/pmu-events/arch/x86/novalake/virtual-memory.json
tools/perf/pmu-events/arch/x86/mapfile.csv +1 -0
@@ -39,6 +39,7 @@ GenuineIntel-6-8[CD],v1.19,tigerlake,core GenuineIntel-6-2C,v5,westmereep-dp,core GenuineIntel-6-25,v4,westmereep-sp,core GenuineIntel-6-2F,v4,westmereex,core +GenuineIntel-18-0[13],v1.00,novalake,core AuthenticAMD-23-([12][0-9A-F]|[0-9A-F]),v2,amdzen1,core AuthenticAMD-23-[[:xdigit:]]+,v1,amdzen2,core AuthenticAMD-25-([245][[:xdigit:]]|[[:xdigit:]]),v1,amdzen3,core
tools/perf/pmu-events/arch/x86/novalake/cache.json new file +749 -0
749 lines not on this page · raw message
tools/perf/pmu-events/arch/x86/novalake/counter.json new file +17 -0
@@ -0,0 +1,17 @@ +[ + { + "Unit": "cpu_atom", + "CountersNumFixed": "7", + "CountersNumGeneric": "8" + }, + { + "Unit": "cpu_core", + "CountersNumFixed": "4", + "CountersNumGeneric": "8" + }, + { + "Unit": "iMC", + "CountersNumFixed": "0", + "CountersNumGeneric": "5" + } +]
tools/perf/pmu-events/arch/x86/novalake/floating-point.json new file +452 -0
452 lines not on this page · raw message
tools/perf/pmu-events/arch/x86/novalake/frontend.json new file +600 -0
600 lines not on this page · raw message
tools/perf/pmu-events/arch/x86/novalake/memory.json new file +299 -0
299 lines not on this page · raw message
tools/perf/pmu-events/arch/x86/novalake/other.json new file +144 -0
144 lines not on this page · raw message
tools/perf/pmu-events/arch/x86/novalake/pipeline.json new file +1976 -0
731 lines not on this page · raw message
tools/perf/pmu-events/arch/x86/novalake/uncore-memory.json new file +42 -0
42 lines not on this page · raw message
tools/perf/pmu-events/arch/x86/novalake/virtual-memory.json new file +338 -0
338 lines not on this page · raw message
The rest of this diff is not shown. View the raw message.
Signature · 1 line
2.55.0.795.g602f6c329a-goog
Part of the diff is not shown here. View the raw message.
[email protected]
AI reviewRe: [PATCH v1 09/12] perf vendor events intel: Add novalake v1.00 events
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider: - [Medium] The CPUID matching regex for novalake expects a zero-padded model ('0[13]'), but perf generates…
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Medium] The CPUID matching regex for novalake expects a zero-padded model ('0[13]'), but perf generates the CPUID string using '%X' without zero-padding, meaning the events will fail to load. --
--- Patch [9]: [PATCH v1 09/12] perf vendor events intel: Add novalake v1.00 events --- commit 5f5b010b3f6aa216e7d15989744fc8646fec9af1
Author: Chun-Tse Shao <[email protected]>
perf vendor events intel: Add novalake v1.00 events
This commit adds performance monitoring events for the Intel Novalake architecture. It introduces new JSON event definition files and updates the x86 mapfile to associate the new models with these events.
tools/perf/pmu-events/arch/x86/mapfile.csv GenuineIntel-6-8[CD],v1.19,tigerlake,core · line 42
Quoted 9 lines
tools/perf/pmu-events/arch/x86/mapfile.csv +1
-0@@ -39,6 +39,7 @@ GenuineIntel-6-8[CD],v1.19,tigerlake,core GenuineIntel-6-2C,v5,westmereep-dp,core GenuineIntel-6-25,v4,westmereep-sp,core GenuineIntel-6-2F,v4,westmereex,core +GenuineIntel-18-0[13],v1.00,novalake,core
[Severity: Medium]
Will this regex fail to match the actual CPUID string generated by perf?
When perf runs on a Novalake processor, get_cpuid_str() formats the CPUID using %s-%u-%X-%X$. For Novalake models 0x01 and 0x03, the %X format specifier does not zero-pad the output. This results in a CPUID string like GenuineIntel-18-1 or GenuineIntel-18-3.
Since this regex explicitly expects a leading zero (0[13]), it looks like it won't match the unpadded string, causing the new events to be unavailable. Should this be GenuineIntel-18-[13] instead?
[ ... ]
Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=9
PATCH 10/12 perf vendor events intel: Update pantherlake events from 1.06 to 1.07
In reply to Chun-Tse ShaoThe updated events were published in:
https://github.com/intel/perfmon/commit/ce70546e9ccca4181d142057171c5bd820c8756d
--- tools/perf/pmu-events/arch/x86/mapfile.csv | 2 +- tools/perf/pmu-events/arch/x86/pantherlake/pipeline.json | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-)
tools/perf/pmu-events/arch/x86/mapfile.csv +1 -1
@@ -26,7 +26,7 @@ GenuineIntel-6-BD,v1.26,lunarlake,core GenuineIntel-6-(AA|AC|B5),v1.22,meteorlake,core GenuineIntel-6-1[AEF],v4,nehalemep,core GenuineIntel-6-2E,v4,nehalemex,core -GenuineIntel-6-(CC|D5),v1.06,pantherlake,core +GenuineIntel-6-(CC|D5|E5),v1.07,pantherlake,core GenuineIntel-6-A7,v1.04,rocketlake,core GenuineIntel-6-2A,v19,sandybridge,core GenuineIntel-6-8F,v1.39,sapphirerapids,core
tools/perf/pmu-events/arch/x86/pantherlake/pipeline.json +1 -1
@@ -1334,7 +1334,7 @@ "Counter": "0,1,2,3,4,5,6,7,8,9", "EventCode": "0xc0", "EventName": "INST_RETIRED.REP_ITERATION", - "PublicDescription": "Number of iterations of Repeat (REP) string retired instructions such as MOVS, CMPS, and SCAS. Each has a byte, word, and doubleword version and string instructions can be repeated using a repetition prefix, REP, that allows their architectural execution to be repeated a number of times as specified by the RCX register. Note the number of iterations is implementation-dependent. Available PDIST counters: 0,1", + "PublicDescription": "Number of iterations of Repeat (REP) string retired instructions such as MOVS, CMPS, and SCAS. Each has a byte, word, and doubleword version and string instructions can be repeated using a repetition prefix, REP, that allows their architectural execution to be repeated a number of times as specified by the RCX register. Note: Since the number of iterations within a REP instruction can be significantly affected by fast strings, this event may vary run to run and not match the architectural number of iterations (specified by RCX) Available PDIST counters: 0,1", "SampleAfterValue": "2000003", "UMask": "0x8", "Unit": "cpu_core"
Signature · 1 line
2.55.0.795.g602f6c329a-goog
PATCH 11/12 perf vendor events intel: Update sierraforest events from 1.17 to 1.18
In reply to Chun-Tse ShaoThe updated events were published in:
https://github.com/intel/perfmon/commit/d1bc6c1e8b32e7a75c70cc939295c11ba9aabc96
--- tools/perf/pmu-events/arch/x86/mapfile.csv | 2 +- .../arch/x86/sierraforest/counter.json | 77 ------------------- .../x86/sierraforest/uncore-interconnect.json | 10 +++ .../arch/x86/sierraforest/uncore-memory.json | 2 +- 4 files changed, 12 insertions(+), 79 deletions(-) delete mode 100644 tools/perf/pmu-events/arch/x86/sierraforest/counter.json
tools/perf/pmu-events/arch/x86/mapfile.csv +1 -1
@@ -30,7 +30,7 @@ GenuineIntel-6-(CC|D5|E5),v1.07,pantherlake,core GenuineIntel-6-A7,v1.04,rocketlake,core GenuineIntel-6-2A,v19,sandybridge,core GenuineIntel-6-8F,v1.39,sapphirerapids,core -GenuineIntel-6-AF,v1.17,sierraforest,core +GenuineIntel-6-AF,v1.18,sierraforest,core GenuineIntel-6-(37|4A|4C|4D|5A),v15,silvermont,core GenuineIntel-6-(4E|5E|8E|9E|A5|A6),v59,skylake,core GenuineIntel-6-55-[01234],v1.37,skylakex,core
tools/perf/pmu-events/arch/x86/sierraforest/counter.json deleted +0 -77
78 lines not on this page · raw message
tools/perf/pmu-events/arch/x86/sierraforest/uncore-interconnect.json +10 -0
@@ -814,6 +814,16 @@ "PerPkg": "1", "Unit": "IRP" }, + { + "BriefDescription": "Counts Timeouts - Set 0 : Cache Inserts of Write Transactions as Secondary", + "Counter": "0,1,2,3", + "EventCode": "0x1E", + "EventName": "UNC_I_MISC0.2ND_WR_INSERT", + "Experimental": "1", + "PerPkg": "1", + "UMask": "0x8", + "Unit": "IRP" + }, { "BriefDescription": "Counts Timeouts - Set 0 : Fastpath Rejects", "Counter": "0,1,2,3",
tools/perf/pmu-events/arch/x86/sierraforest/uncore-memory.json +1 -1
@@ -499,7 +499,7 @@ "Unit": "IMC" }, { - "BriefDescription": "DRAM Precharge commands. : Precharge due to (?) : Counts the number of DRAM Precharge commands sent on this channel.", + "BriefDescription": "DRAM Precharge commands. : Precharge due to page table : Counts the number of DRAM Precharge commands sent on this channel.", "Counter": "0,1,2,3", "EventCode": "0x03", "EventName": "UNC_M_PRE_COUNT.PGT",
Signature · 1 line
2.55.0.795.g602f6c329a-goog
PATCH 12/12 perf vendor events intel: Update emeraldrapids metrics
In reply to Chun-Tse ShaoThe updated events were published in:
https://github.com/intel/perfmon/commit/240735b7d8e0b50fe8f4a64e08399df13cb87ae6
--- .../arch/x86/emeraldrapids/emr-metrics.json | 135 +++++++++++------- .../arch/x86/emeraldrapids/metricgroups.json | 2 +- 2 files changed, 87 insertions(+), 50 deletions(-)
tools/perf/pmu-events/arch/x86/emeraldrapids/emr-metrics.json +86 -49
@@ -84,6 +84,18 @@ "PublicDescription": "Ratio of number of completed page walks (for all page sizes) caused by demand data stores to the total number of completed instructions. This implies it missed in the DTLB and further levels of TLB", "ScaleUnit": "1per_instr" }, + { + "BriefDescription": "Bandwidth observed by the integrated I/O traffic controller (IIO) of IO reads that are initiated by end device controllers that are requesting memory from the CPU", + "MetricExpr": "UNC_IIO_DATA_REQ_OF_CPU.MEM_READ.ALL_PARTS * 4 / 1e6 / duration_time", + "MetricName": "iio_bandwidth_read", + "ScaleUnit": "1MB/s" + }, + { + "BriefDescription": "Bandwidth observed by the integrated I/O traffic controller (IIO) of IO writes that are initiated by end device controllers that are writing memory to the CPU", + "MetricExpr": "UNC_IIO_DATA_REQ_OF_CPU.MEM_WRITE.ALL_PARTS * 4 / 1e6 / duration_time", + "MetricName": "iio_bandwidth_write", + "ScaleUnit": "1MB/s" + }, { "BriefDescription": "Bandwidth of IO reads that are initiated by end device controllers that are requesting memory from the CPU", "MetricExpr": "UNC_CHA_TOR_INSERTS.IO_PCIRDCUR * 64 / 1e6 / duration_time",
@@ -242,6 +254,30 @@ "MetricName": "llc_demand_data_read_miss_to_dram_latency", "ScaleUnit": "1ns" }, + { + "BriefDescription": "Bandwidth (MB/sec) of read requests that miss the last level cache (LLC) and go to local memory", + "MetricExpr": "UNC_CHA_REQUESTS.READS_LOCAL * 64 / 1e6 / duration_time", + "MetricName": "llc_miss_local_memory_bandwidth_read", + "ScaleUnit": "1MB/s" + }, + { + "BriefDescription": "Bandwidth (MB/sec) of write requests that miss the last level cache (LLC) and go to local memory", + "MetricExpr": "UNC_CHA_REQUESTS.WRITES_LOCAL * 64 / 1e6 / duration_time", + "MetricName": "llc_miss_local_memory_bandwidth_write", + "ScaleUnit": "1MB/s" + }, + { + "BriefDescription": "Bandwidth (MB/sec) of read requests that miss the last level cache (LLC) and go to remote memory", + "MetricExpr": "UNC_CHA_REQUESTS.READS_REMOTE * 64 / 1e6 / duration_time", + "MetricName": "llc_miss_remote_memory_bandwidth_read", + "ScaleUnit": "1MB/s" + }, + { + "BriefDescription": "Bandwidth (MB/sec) of write requests that miss the last level cache (LLC) and go to remote memory", + "MetricExpr": "UNC_CHA_REQUESTS.WRITES_REMOTE * 64 / 1e6 / duration_time", + "MetricName": "llc_miss_remote_memory_bandwidth_write", + "ScaleUnit": "1MB/s" + }, { "BriefDescription": "The ratio of number of completed memory load instructions to the total number completed instructions", "MetricExpr": "MEM_INST_RETIRED.ALL_LOADS / INST_RETIRED.ANY",
@@ -403,15 +439,15 @@ }, { "BriefDescription": "Total pipeline cost of external Memory- or Cache-Bandwidth related bottlenecks", - "MetricExpr": "100 * (tma_memory_bound * (tma_dram_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_mem_bandwidth / (tma_mem_bandwidth + tma_mem_latency)) + 0 / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound) * tma_mem_bandwidth / (tma_mem_bandwidth + tma_mem_latency) + tma_memory_bound * (tma_l3_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_sq_full / (tma_contested_accesses + tma_data_sharing + tma_l3_hit_latency + tma_sq_full)) + tma_memory_bound * (tma_l1_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_fb_full / (tma_dtlb_load + tma_fb_full + tma_l1_latency_dependency + tma_lock_latency + tma_split_loads + tma_store_fwd_blk)))", + "MetricExpr": "100 * (tma_memory_bound * (tma_l3_miss_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_mem_bandwidth / (tma_mem_bandwidth + tma_mem_latency)) + 0 / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound) * tma_mem_bandwidth / (tma_mem_bandwidth + tma_mem_latency) + tma_memory_bound * (tma_l3_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_sq_full / (tma_contested_accesses + tma_data_sharing + tma_l3_hit_latency + tma_sq_full)) + tma_memory_bound * (tma_l1_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_fb_full / (tma_dtlb_load + tma_fb_full + tma_l1_latency_dependency + tma_lock_latency + tma_split_loads + tma_store_fwd_blk)))", "MetricGroup": "BvMB;Mem;MemoryBW;Offcore;tma_issueBW", "MetricName": "tma_bottleneck_data_cache_memory_bandwidth", "MetricThreshold": "tma_bottleneck_data_cache_memory_bandwidth > 20", - "PublicDescription": "Total pipeline cost of external Memory- or Cache-Bandwidth related bottlenecks. Related metrics: tma_fb_full, tma_info_system_dram_bw_use, tma_mem_bandwidth, tma_sq_full" + "PublicDescription": "Total pipeline cost of external Memory- or Cache-Bandwidth related bottlenecks. Related metrics: tma_fb_full, tma_info_system_dram_bw_use, tma_mem_bandwidth, tma_sq_full, tma_uc_bound" }, { "BriefDescription": "Total pipeline cost of external Memory- or Cache-Latency related bottlenecks", - "MetricExpr": "100 * (tma_memory_bound * (tma_dram_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_mem_latency / (tma_mem_bandwidth + tma_mem_latency)) + 0 / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound) * tma_mem_latency / (tma_mem_bandwidth + tma_mem_latency) + tma_memory_bound * (tma_l3_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_l3_hit_latency / (tma_contested_accesses + tma_data_sharing + tma_l3_hit_latency + tma_sq_full)) + tma_memory_bound * tma_l2_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound) + tma_memory_bound * (tma_l1_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_l1_latency_dependency / (tma_dtlb_load + tma_fb_full + tma_l1_latency_dependency + tma_lock_latency + tma_split_loads + tma_store_fwd_blk)) + tma_memory_bound * (tma_l1_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_lock_latency / (tma_dtlb_load + tma_fb_full + tma_l1_latency_dependency + tma_lock_latency + tma_split_loads + tma_store_fwd_blk)) + tma_memory_bound * (tma_l1_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_split_loads / (tma_dtlb_load + tma_fb_full + tma_l1_latency_dependency + tma_lock_latency + tma_split_loads + tma_store_fwd_blk)) + tma_memory_bound * (tma_store_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_split_stores / (tma_dtlb_store + tma_false_sharing + tma_split_stores + tma_store_latency + tma_streaming_stores)) + tma_memory_bound * (tma_store_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_store_latency / (tma_dtlb_store + tma_false_sharing + tma_split_stores + tma_store_latency + tma_streaming_stores)))", + "MetricExpr": "100 * (tma_memory_bound * (tma_l3_miss_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_mem_latency / (tma_mem_bandwidth + tma_mem_latency)) + 0 / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound) * tma_mem_latency / (tma_mem_bandwidth + tma_mem_latency) + tma_memory_bound * (tma_l3_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_l3_hit_latency / (tma_contested_accesses + tma_data_sharing + tma_l3_hit_latency + tma_sq_full)) + tma_memory_bound * tma_l2_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound) + tma_memory_bound * (tma_l1_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_l1_latency_dependency / (tma_dtlb_load + tma_fb_full + tma_l1_latency_dependency + tma_lock_latency + tma_split_loads + tma_store_fwd_blk)) + tma_memory_bound * (tma_l1_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_lock_latency / (tma_dtlb_load + tma_fb_full + tma_l1_latency_dependency + tma_lock_latency + tma_split_loads + tma_store_fwd_blk)) + tma_memory_bound * (tma_l1_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_split_loads / (tma_dtlb_load + tma_fb_full + tma_l1_latency_dependency + tma_lock_latency + tma_split_loads + tma_store_fwd_blk)) + tma_memory_bound * (tma_store_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_split_stores / (tma_dtlb_store + tma_false_sharing + tma_split_stores + tma_store_latency + tma_streaming_stores)) + tma_memory_bound * (tma_store_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_store_latency / (tma_dtlb_store + tma_false_sharing + tma_split_stores + tma_store_latency + tma_streaming_stores)))", "MetricGroup": "BvML;Mem;MemoryLat;Offcore;tma_issueLat", "MetricName": "tma_bottleneck_data_cache_memory_latency", "MetricThreshold": "tma_bottleneck_data_cache_memory_latency > 20",
@@ -434,7 +470,7 @@ }, { "BriefDescription": "Total pipeline cost of Memory Address Translation related bottlenecks (data-side TLBs)", - "MetricExpr": "100 * (tma_memory_bound * (tma_l1_bound / max(tma_memory_bound, tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_dtlb_load / max(tma_l1_bound, tma_dtlb_load + tma_fb_full + tma_l1_latency_dependency + tma_lock_latency + tma_split_loads + tma_store_fwd_blk)) + tma_memory_bound * (tma_store_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound)) * (tma_dtlb_store / (tma_dtlb_store + tma_false_sharing + tma_split_stores + tma_store_latency + tma_streaming_stores)))", + "MetricExpr": "100 * (tma_memory_bound * (tma_l1_bound / max(tma_memory_bound, tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_dtlb_load / max(tma_l1_bound, tma_dtlb_load + tma_fb_full + tma_l1_latency_dependency + tma_lock_latency + tma_split_loads + tma_store_fwd_blk)) + tma_memory_bound * (tma_store_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound)) * (tma_dtlb_store / (tma_dtlb_store + tma_false_sharing + tma_split_stores + tma_store_latency + tma_streaming_stores)))", "MetricGroup": "BvMT;Mem;MemoryTLB;Offcore;tma_issueTLB", "MetricName": "tma_bottleneck_memory_data_tlbs", "MetricThreshold": "tma_bottleneck_memory_data_tlbs > 20",
@@ -442,7 +478,7 @@ }, { "BriefDescription": "Total pipeline cost of Memory Synchronization related bottlenecks (data transfers and coherency updates across processors)", - "MetricExpr": "100 * (tma_memory_bound * (tma_dram_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound) * (tma_mem_latency / (tma_mem_bandwidth + tma_mem_latency)) * tma_remote_cache / (tma_local_mem + tma_remote_cache + tma_remote_mem) + tma_l3_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound) * (tma_contested_accesses + tma_data_sharing) / (tma_contested_accesses + tma_data_sharing + tma_l3_hit_latency + tma_sq_full) + tma_store_bound / (tma_dram_bound + tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_store_bound) * tma_false_sharing / (tma_dtlb_store + tma_false_sharing + tma_split_stores + tma_store_latency + tma_streaming_stores - tma_store_latency)) + tma_machine_clears * (1 - tma_other_nukes / tma_other_nukes))", + "MetricExpr": "100 * (tma_memory_bound * (tma_l3_miss_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound) * (tma_mem_latency / (tma_mem_bandwidth + tma_mem_latency)) * tma_remote_cache / (tma_local_mem + tma_remote_cache + tma_remote_mem) + tma_l3_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound) * (tma_contested_accesses + tma_data_sharing) / (tma_contested_accesses + tma_data_sharing + tma_l3_hit_latency + tma_sq_full) + tma_store_bound / (tma_l1_bound + tma_l2_bound + tma_l3_bound + tma_l3_miss_bound + tma_store_bound) * tma_false_sharing / (tma_dtlb_store + tma_false_sharing + tma_split_stores + tma_store_latency + tma_streaming_stores - tma_store_latency)) + tma_machine_clears * (1 - tma_other_nukes / tma_other_nukes))", "MetricGroup": "BvMS;LockCont;Mem;Offcore;tma_issueSyncxn", "MetricName": "tma_bottleneck_memory_synchronization", "MetricThreshold": "tma_bottleneck_memory_synchronization > 10",
@@ -594,13 +630,13 @@ "ScaleUnit": "100%" }, { - "BriefDescription": "This metric estimates fraction of cycles while the memory subsystem was handling synchronizations due to data-sharing accesses", + "BriefDescription": "This metric estimates fraction of cycles while the memory subsystem was handling synchronizations due to L3 data-sharing accesses", "MetricConstraint": "NO_GROUP_EVENTS", "MetricExpr": "74.6 * tma_info_system_core_frequency * (MEM_LOAD_L3_HIT_RETIRED.XSNP_NO_FWD + MEM_LOAD_L3_HIT_RETIRED.XSNP_FWD * (1 - OCR.DEMAND_DATA_RD.L3_HIT.SNOOP_HITM / (OCR.DEMAND_DATA_RD.L3_HIT.SNOOP_HITM + OCR.DEMAND_DATA_RD.L3_HIT.SNOOP_HIT_WITH_FWD))) * (1 + MEM_LOAD_RETIRED.FB_HIT / MEM_LOAD_RETIRED.L1_MISS / 2) / tma_info_thread_clks", "MetricGroup": "BvMS;Offcore;Snoop;TopdownL4;tma_L4_group;tma_issueSyncxn;tma_l3_bound_group", "MetricName": "tma_data_sharing", "MetricThreshold": "tma_data_sharing > 0.05 & (tma_l3_bound > 0.05 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))", - "PublicDescription": "This metric estimates fraction of cycles while the memory subsystem was handling synchronizations due to data-sharing accesses. Data shared by multiple Logical Processors (even just read shared) may cause increased access latency due to cache coherency. Excessive data sharing can drastically harm multithreaded performance. Sample with: MEM_LOAD_L3_HIT_RETIRED.XSNP_NO_FWD. Related metrics: tma_bottleneck_memory_synchronization, tma_contested_accesses, tma_false_sharing, tma_machine_clears, tma_remote_cache", + "PublicDescription": "This metric estimates fraction of cycles while the memory subsystem was handling synchronizations due to L3 data-sharing accesses. Data shared by multiple Logical Processors (even just read shared) may cause increased access latency due to cache coherency. Excessive data sharing can drastically harm multithreaded performance. Sample with: MEM_LOAD_L3_HIT_RETIRED.XSNP_NO_FWD. Related metrics: tma_bottleneck_memory_synchronization, tma_contested_accesses, tma_false_sharing, tma_machine_clears, tma_remote_cache", "ScaleUnit": "100%" }, {
@@ -621,15 +657,6 @@ "PublicDescription": "This metric represents fraction of cycles where the Divider unit was active. Divide and square root instructions are performed by the Divider unit and can take considerably longer latency than integer or Floating Point addition; subtraction; or multiplication. Sample with: ARITH.DIV_ACTIVE", "ScaleUnit": "100%" }, - { - "BriefDescription": "This metric estimates how often the CPU was stalled on accesses to external memory (DRAM) by loads", - "MetricExpr": "MEMORY_ACTIVITY.STALLS_L3_MISS / tma_info_thread_clks", - "MetricGroup": "MemoryBound;TmaL3mem;TopdownL3;tma_L3_group;tma_memory_bound_group", - "MetricName": "tma_dram_bound", - "MetricThreshold": "tma_dram_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2)", - "PublicDescription": "This metric estimates how often the CPU was stalled on accesses to external memory (DRAM) by loads. Better caching can improve the latency and increase performance. Sample with: MEM_LOAD_RETIRED.L3_MISS", - "ScaleUnit": "100%" - }, { "BriefDescription": "This metric represents Core fraction of cycles in which CPU was likely limited due to DSB (decoded uop cache) fetch pipeline", "MetricExpr": "(IDQ.DSB_CYCLES_ANY - IDQ.DSB_CYCLES_OK) / tma_info_core_core_clks / 2",
@@ -681,7 +708,7 @@ "MetricGroup": "BvMB;MemoryBW;TopdownL4;tma_L4_group;tma_issueBW;tma_issueSL;tma_issueSmSt;tma_l1_bound_group", "MetricName": "tma_fb_full", "MetricThreshold": "tma_fb_full > 0.3", - "PublicDescription": "This metric does a *rough estimation* of how often L1D Fill Buffer unavailability limited additional L1D miss memory access requests to proceed. The higher the metric value; the deeper the memory hierarchy level the misses are satisfied from (metric values >1 are valid). Often it hints on approaching bandwidth limits (to L2 cache; L3 cache or external memory). Related metrics: tma_bottleneck_data_cache_memory_bandwidth, tma_info_system_dram_bw_use, tma_mem_bandwidth, tma_sq_full, tma_store_latency, tma_streaming_stores", + "PublicDescription": "This metric does a *rough estimation* of how often L1D Fill Buffer unavailability limited additional L1D miss memory access requests to proceed. The higher the metric value; the deeper the memory hierarchy level the misses are satisfied from (metric values >1 are valid). Often it hints on approaching bandwidth limits (to L2 cache; L3 cache or external memory). Related metrics: tma_bottleneck_data_cache_memory_bandwidth, tma_info_system_dram_bw_use, tma_mem_bandwidth, tma_sq_full, tma_store_latency, tma_streaming_stores, tma_uc_bound", "ScaleUnit": "100%" }, {
@@ -974,7 +1001,7 @@ }, { "BriefDescription": "Fraction of Uops delivered by the DSB (aka Decoded ICache; or Uop Cache)", - "MetricExpr": "IDQ.DSB_UOPS / UOPS_ISSUED.ANY", + "MetricExpr": "IDQ.DSB_UOPS / (IDQ.DSB_UOPS + IDQ.MITE_UOPS + IDQ.MS_UOPS)", "MetricGroup": "DSB;Fed;FetchBW;tma_issueFB", "MetricName": "tma_info_frontend_dsb_coverage", "MetricThreshold": "tma_info_frontend_dsb_coverage < 0.7 & tma_info_thread_ipc / 6 > 0.35",
@@ -1092,7 +1119,7 @@ { "BriefDescription": "Instructions per FP Arithmetic Scalar Half-Precision instruction (lower number means higher occurrence rate)", "MetricExpr": "INST_RETIRED.ANY / FP_ARITH_INST_RETIRED2.SCALAR", - "MetricGroup": "Flops;FpScalar;InsType;Server", + "MetricGroup": "Flops;FpScalar;InsType", "MetricName": "tma_info_inst_mix_iparith_scalar_hp", "MetricThreshold": "tma_info_inst_mix_iparith_scalar_hp < 10", "PublicDescription": "Instructions per FP Arithmetic Scalar Half-Precision instruction (lower number means higher occurrence rate). Values < 1 are possible due to intentional FMA double counting."
@@ -1161,6 +1188,14 @@ "MetricThreshold": "tma_info_inst_mix_iptb < 13", "PublicDescription": "Instructions per taken branch. Related metrics: tma_dsb_switches, tma_fetch_bandwidth, tma_info_botlnk_l2_dsb_bandwidth, tma_info_botlnk_l2_dsb_misses, tma_info_frontend_dsb_coverage, tma_lcp" }, + { + "BriefDescription": "AVX preserve/restore assists per kilo instruction", + "MetricExpr": "1e3 * ASSISTS.SSE_AVX_MIX / INST_RETIRED.ANY", + "MetricGroup": "tma_issueMV", + "MetricName": "tma_info_inst_mix_vectormixpki", + "MetricThreshold": "tma_info_inst_mix_vectormixpki > 0.05", + "PublicDescription": "AVX preserve/restore assists per kilo instruction. Related metrics: tma_mixing_vectors, tma_ms_switches" + }, { "BriefDescription": "Average per-core data fill bandwidth to the L1 data cache [GB / sec]", "MetricExpr": "tma_info_memory_l1d_cache_fill_bw",
@@ -1402,7 +1437,7 @@ "MetricName": "tma_info_memory_tlb_store_stlb_mpki" }, { - "BriefDescription": "Mem;Backend;CacheHits", + "BriefDescription": "Instruction-Level-Parallelism (average number of uops executed when there is execution) per physical core", "MetricExpr": "UOPS_EXECUTED.THREAD / (UOPS_EXECUTED.CORE_CYCLES_GE_1 / 2 if #SMT_on else cpu@UOPS_EXECUTED.THREAD\\,cmask\\=1@)", "MetricGroup": "Cor;Pipeline;PortsUtil;SMT", "MetricName": "tma_info_pipeline_execute"
@@ -1482,7 +1517,7 @@ "MetricExpr": "64 * (UNC_M_CAS_COUNT.RD + UNC_M_CAS_COUNT.WR) / 1e9 / tma_info_system_time", "MetricGroup": "HPC;MemOffcore;MemoryBW;SoC;tma_issueBW", "MetricName": "tma_info_system_dram_bw_use", - "PublicDescription": "Average external Memory Bandwidth Use for reads and writes [GB / sec]. Related metrics: tma_bottleneck_data_cache_memory_bandwidth, tma_fb_full, tma_mem_bandwidth, tma_sq_full" + "PublicDescription": "Average external Memory Bandwidth Use for reads and writes [GB / sec]. Related metrics: tma_bottleneck_data_cache_memory_bandwidth, tma_fb_full, tma_mem_bandwidth, tma_sq_full, tma_uc_bound" }, { "BriefDescription": "Giga Floating Point Operations Per Second",
@@ -1532,13 +1567,6 @@ "MetricName": "tma_info_system_mem_dram_read_latency", "PublicDescription": "Average latency of data read request to external DRAM memory [in nanoseconds]. Accounts for demand loads and L1/L2 data-read prefetches" }, - { - "BriefDescription": "Fraction of Uncore cycles where requests got rejected due to duplicate address already in IRQ ingress queue in the cache homing agent", - "MetricExpr": "UNC_CHA_RxC_IRQ1_REJECT.PA_MATCH / UNC_CHA_CLOCKTICKS", - "MetricGroup": "LockCont;MemOffcore;Server;SoC", - "MetricName": "tma_info_system_mem_irq_duplicate_address", - "MetricThreshold": "tma_info_system_mem_irq_duplicate_address > 0.1" - }, { "BriefDescription": "Average number of parallel data read requests to external memory", "MetricExpr": "UNC_CHA_TOR_OCCUPANCY.IA_MISS_DRD / UNC_CHA_TOR_OCCUPANCY.IA_MISS_DRD@thresh\\=1@",
@@ -1591,12 +1619,6 @@ "MetricGroup": "Power", "MetricName": "tma_info_system_turbo_utilization" }, - { - "BriefDescription": "Measured Average Uncore Frequency for the SoC [GHz]", - "MetricExpr": "tma_info_system_socket_clks / 1e9 / tma_info_system_time", - "MetricGroup": "SoC", - "MetricName": "tma_info_system_uncore_frequency" - }, { "BriefDescription": "Cross-socket Ultra Path Interconnect (UPI) data transmit bandwidth for data only [MB / sec]", "MetricExpr": "UNC_UPI_TxL_FLITS.ALL_DATA * 64 / 9 / 1e6",
@@ -1752,6 +1774,15 @@ "PublicDescription": "This metric estimates fraction of cycles with demand load accesses that hit the L3 cache under unloaded scenarios (possibly L3 latency limited). Avoiding private cache misses (i.e. L2 misses/L3 hits) will improve the latency; reduce contention with sibling physical cores and increase performance. Note the value of this node may overlap with its siblings. Sample with: MEM_LOAD_RETIRED.L3_HIT_PS. Related metrics: tma_bottleneck_data_cache_memory_latency, tma_mem_latency", "ScaleUnit": "100%" }, + { + "BriefDescription": "This metric estimates how often the CPU was stalled on accesses to external memory (DRAM) by loads", + "MetricExpr": "MEMORY_ACTIVITY.STALLS_L3_MISS / tma_info_thread_clks", + "MetricGroup": "MemoryBound;Offcore;TmaL3mem;TopdownL3;tma_L3_group;tma_memory_bound_group", + "MetricName": "tma_l3_miss_bound", + "MetricThreshold": "tma_l3_miss_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2)", + "PublicDescription": "This metric estimates how often the CPU was stalled on accesses to external memory (DRAM) by loads. Better caching can improve the latency and increase performance. Sample with: MEM_LOAD_RETIRED.L3_MISS", + "ScaleUnit": "100%" + }, { "BriefDescription": "This metric represents fraction of cycles CPU was stalled due to Length Changing Prefixes (LCPs)", "MetricExpr": "DECODE.LCP / tma_info_thread_clks",
@@ -1826,14 +1857,14 @@ "MetricExpr": "72 * tma_info_system_core_frequency * MEM_LOAD_L3_MISS_RETIRED.LOCAL_DRAM * (1 + MEM_LOAD_RETIRED.FB_HIT / MEM_LOAD_RETIRED.L1_MISS / 2) / tma_info_thread_clks", "MetricGroup": "Server;TopdownL5;tma_L5_group;tma_mem_latency_group", "MetricName": "tma_local_mem", - "MetricThreshold": "tma_local_mem > 0.1 & (tma_mem_latency > 0.1 & (tma_dram_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2)))", + "MetricThreshold": "tma_local_mem > 0.1 & (tma_mem_latency > 0.1 & (tma_l3_miss_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2)))", "PublicDescription": "This metric estimates fraction of cycles while the memory subsystem was handling loads from local memory. Caching will improve the latency and increase performance. Sample with: MEM_LOAD_L3_MISS_RETIRED.LOCAL_DRAM", "ScaleUnit": "100%" }, { "BriefDescription": "This metric represents fraction of cycles the CPU spent handling cache misses due to lock operations", "MetricConstraint": "NO_GROUP_EVENTS", - "MetricExpr": "(16 * max(0, MEM_INST_RETIRED.LOCK_LOADS - L2_RQSTS.ALL_RFO) + MEM_INST_RETIRED.LOCK_LOADS / MEM_INST_RETIRED.ALL_STORES * (10 * L2_RQSTS.RFO_HIT + min(CPU_CLK_UNHALTED.THREAD, OFFCORE_REQUESTS_OUTSTANDING.CYCLES_WITH_DEMAND_RFO))) / tma_info_thread_clks", + "MetricExpr": "LOCK_CYCLES.CACHE_LOCK_DURATION / tma_info_thread_clks", "MetricGroup": "LockCont;Offcore;TopdownL4;tma_L4_group;tma_issueRFO;tma_l1_bound_group", "MetricName": "tma_lock_latency", "MetricThreshold": "tma_lock_latency > 0.2 & (tma_l1_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))",
@@ -1856,24 +1887,24 @@ "MetricExpr": "INT_MISC.MBA_STALLS / tma_info_thread_clks", "MetricGroup": "MemoryBW;Offcore;Server;TopdownL5;tma_L5_group;tma_mem_bandwidth_group", "MetricName": "tma_mba_stalls", - "MetricThreshold": "tma_mba_stalls > 0.1 & (tma_mem_bandwidth > 0.2 & (tma_dram_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2)))", + "MetricThreshold": "tma_mba_stalls > 0.1 & (tma_mem_bandwidth > 0.2 & (tma_l3_miss_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2)))", "ScaleUnit": "100%" }, { "BriefDescription": "This metric estimates fraction of cycles where the core's performance was likely hurt due to approaching bandwidth limits of external memory - DRAM ([SPR-HBM] and/or HBM)", - "MetricExpr": "min(CPU_CLK_UNHALTED.THREAD, cpu@OFFCORE_REQUESTS_OUTSTANDING.ALL_DATA_RD\\,cmask\\=4@) / tma_info_thread_clks", - "MetricGroup": "BvMB;MemoryBW;Offcore;TopdownL4;tma_L4_group;tma_dram_bound_group;tma_issueBW", + "MetricExpr": "min(CPU_CLK_UNHALTED.THREAD, cpu@OFFCORE_REQUESTS_OUTSTANDING.ALL_DATA_RD\\,cmask\\=12@) / tma_info_thread_clks", + "MetricGroup": "BvMB;MemoryBW;Offcore;TopdownL4;tma_L4_group;tma_issueBW;tma_l3_miss_bound_group", "MetricName": "tma_mem_bandwidth", - "MetricThreshold": "tma_mem_bandwidth > 0.2 & (tma_dram_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))", - "PublicDescription": "This metric estimates fraction of cycles where the core's performance was likely hurt due to approaching bandwidth limits of external memory - DRAM ([SPR-HBM] and/or HBM). The underlying heuristic assumes that a similar off-core traffic is generated by all IA cores. This metric does not aggregate non-data-read requests by this logical processor; requests from other IA Logical Processors/Physical Cores/sockets; or other non-IA devices like GPU; hence the maximum external memory bandwidth limits may or may not be approached when this metric is flagged (see Uncore counters for that). Related metrics: tma_bottleneck_data_cache_memory_bandwidth, tma_fb_full, tma_info_system_dram_bw_use, tma_sq_full", + "MetricThreshold": "tma_mem_bandwidth > 0.2 & (tma_l3_miss_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))", + "PublicDescription": "This metric estimates fraction of cycles where the core's performance was likely hurt due to approaching bandwidth limits of external memory - DRAM ([SPR-HBM] and/or HBM). The underlying heuristic assumes that a similar off-core traffic is generated by all IA cores. This metric does not aggregate non-data-read requests by this logical processor; requests from other IA Logical Processors/Physical Cores/sockets; or other non-IA devices like GPU; hence the maximum external memory bandwidth limits may or may not be approached when this metric is flagged (see Uncore counters for that). Related metrics: tma_bottleneck_data_cache_memory_bandwidth, tma_fb_full, tma_info_system_dram_bw_use, tma_sq_full, tma_uc_bound", "ScaleUnit": "100%" }, { "BriefDescription": "This metric estimates fraction of cycles where the performance was likely hurt due to latency from external memory - DRAM ([SPR-HBM] and/or HBM)", "MetricExpr": "min(CPU_CLK_UNHALTED.THREAD, OFFCORE_REQUESTS_OUTSTANDING.CYCLES_WITH_DATA_RD) / tma_info_thread_clks - tma_mem_bandwidth", - "MetricGroup": "BvML;MemoryLat;Offcore;TopdownL4;tma_L4_group;tma_dram_bound_group;tma_issueLat", + "MetricGroup": "BvML;MemoryLat;Offcore;TopdownL4;tma_L4_group;tma_issueLat;tma_l3_miss_bound_group", "MetricName": "tma_mem_latency", - "MetricThreshold": "tma_mem_latency > 0.1 & (tma_dram_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))", + "MetricThreshold": "tma_mem_latency > 0.1 & (tma_l3_miss_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))", "PublicDescription": "This metric estimates fraction of cycles where the performance was likely hurt due to latency from external memory - DRAM ([SPR-HBM] and/or HBM). This metric does not aggregate requests from other Logical Processors/Physical Cores/sockets (see Uncore counters for that). Related metrics: tma_bottleneck_data_cache_memory_latency, tma_l3_hit_latency", "ScaleUnit": "100%" },
@@ -1937,7 +1968,7 @@ "MetricGroup": "TopdownL5;tma_L5_group;tma_issueMV;tma_ports_utilized_0_group", "MetricName": "tma_mixing_vectors", "MetricThreshold": "tma_mixing_vectors > 0.05", - "PublicDescription": "This metric estimates penalty in terms of percentage of([SKL+] injected blend uops out of all Uops Issued -- the Count Domain; [ADL+] cycles). Usually a Mixing_Vectors over 5% is worth investigating. Read more in Appendix B1 of the Optimizations Guide for this topic. Related metrics: tma_ms_switches", + "PublicDescription": "This metric estimates penalty in terms of percentage of([SKL+] injected blend uops out of all Uops Issued -- the Count Domain; [ADL+] cycles). Usually a Mixing_Vectors over 5% is worth investigating. Read more in Appendix B1 of the Optimizations Guide for this topic. Related metrics: tma_info_inst_mix_vectormixpki, tma_ms_switches", "ScaleUnit": "100%" }, {
@@ -1954,7 +1985,7 @@ "MetricGroup": "FetchLat;MicroSeq;TopdownL3;tma_L3_group;tma_fetch_latency_group;tma_issueMC;tma_issueMS;tma_issueMV;tma_issueSO", "MetricName": "tma_ms_switches", "MetricThreshold": "tma_ms_switches > 0.05 & (tma_fetch_latency > 0.1 & tma_frontend_bound > 0.15)", - "PublicDescription": "This metric estimates the fraction of cycles when the CPU was stalled due to switches of uop delivery to the Microcode Sequencer (MS). Commonly used instructions are optimized for delivery by the DSB (decoded i-cache) or MITE (legacy instruction decode) pipelines. Certain operations cannot be handled natively by the execution pipeline; and must be performed by microcode (small programs injected into the execution stream). Switching to the MS too often can negatively impact performance. The MS is designated to deliver long uop flows required by CISC instructions like CPUID; or uncommon conditions like Floating Point Assists when dealing with Denormals. Sample with: IDQ.MS_SWITCHES. Related metrics: tma_bottleneck_irregular_overhead, tma_clears_resteers, tma_l1_bound, tma_machine_clears, tma_microcode_sequencer, tma_mixing_vectors, tma_serializing_operation", + "PublicDescription": "This metric estimates the fraction of cycles when the CPU was stalled due to switches of uop delivery to the Microcode Sequencer (MS). Commonly used instructions are optimized for delivery by the DSB (decoded i-cache) or MITE (legacy instruction decode) pipelines. Certain operations cannot be handled natively by the execution pipeline; and must be performed by microcode (small programs injected into the execution stream). Switching to the MS too often can negatively impact performance. The MS is designated to deliver long uop flows required by CISC instructions like CPUID; or uncommon conditions like Floating Point Assists when dealing with Denormals. Sample with: IDQ.MS_SWITCHES. Related metrics: tma_bottleneck_irregular_overhead, tma_clears_resteers, tma_info_inst_mix_vectormixpki, tma_l1_bound, tma_machine_clears, tma_microcode_sequencer, tma_mixing_vectors, tma_serializing_operation", "ScaleUnit": "100%" }, {
@@ -2090,7 +2121,7 @@ "MetricExpr": "(133 * tma_info_system_core_frequency * MEM_LOAD_L3_MISS_RETIRED.REMOTE_HITM + 133 * tma_info_system_core_frequency * MEM_LOAD_L3_MISS_RETIRED.REMOTE_FWD) * (1 + MEM_LOAD_RETIRED.FB_HIT / MEM_LOAD_RETIRED.L1_MISS / 2) / tma_info_thread_clks", "MetricGroup": "Offcore;Server;Snoop;TopdownL5;tma_L5_group;tma_issueSyncxn;tma_mem_latency_group", "MetricName": "tma_remote_cache", - "MetricThreshold": "tma_remote_cache > 0.05 & (tma_mem_latency > 0.1 & (tma_dram_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2)))", + "MetricThreshold": "tma_remote_cache > 0.05 & (tma_mem_latency > 0.1 & (tma_l3_miss_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2)))", "PublicDescription": "This metric estimates fraction of cycles while the memory subsystem was handling loads from remote cache in other sockets including synchronizations issues. This is caused often due to non-optimal NUMA allocations. #link to NUMA article. Sample with: MEM_LOAD_L3_MISS_RETIRED.REMOTE_HITM_PS;MEM_LOAD_L3_MISS_RETIRED.REMOTE_FWD_PS. Related metrics: tma_bottleneck_memory_synchronization, tma_contested_accesses, tma_data_sharing, tma_false_sharing, tma_machine_clears", "ScaleUnit": "100%" },
@@ -2099,7 +2130,7 @@ "MetricExpr": "153 * tma_info_system_core_frequency * MEM_LOAD_L3_MISS_RETIRED.REMOTE_DRAM * (1 + MEM_LOAD_RETIRED.FB_HIT / MEM_LOAD_RETIRED.L1_MISS / 2) / tma_info_thread_clks", "MetricGroup": "Server;Snoop;TopdownL5;tma_L5_group;tma_mem_latency_group", "MetricName": "tma_remote_mem", - "MetricThreshold": "tma_remote_mem > 0.1 & (tma_mem_latency > 0.1 & (tma_dram_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2)))", + "MetricThreshold": "tma_remote_mem > 0.1 & (tma_mem_latency > 0.1 & (tma_l3_miss_bound > 0.1 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2)))", "PublicDescription": "This metric estimates fraction of cycles while the memory subsystem was handling loads from remote memory. This is caused often due to non-optimal NUMA allocations. #link to NUMA article. Sample with: MEM_LOAD_L3_MISS_RETIRED.REMOTE_DRAM_PS", "ScaleUnit": "100%" },
@@ -2143,7 +2174,7 @@ }, { "BriefDescription": "This metric estimates fraction of cycles handling memory load split accesses - load that cross 64-byte cache line boundary", - "MetricExpr": "tma_info_memory_load_miss_real_latency * LD_BLOCKS.NO_SR / tma_info_thread_clks", + "MetricExpr": "MEM_INST_RETIRED.SPLIT_LOADS * tma_info_memory_load_miss_real_latency / tma_info_thread_clks", "MetricGroup": "TopdownL4;tma_L4_group;tma_l1_bound_group", "MetricName": "tma_split_loads", "MetricThreshold": "tma_split_loads > 0.3",
@@ -2165,7 +2196,7 @@ "MetricGroup": "BvMB;MemoryBW;Offcore;TopdownL4;tma_L4_group;tma_issueBW;tma_l3_bound_group", "MetricName": "tma_sq_full", "MetricThreshold": "tma_sq_full > 0.3 & (tma_l3_bound > 0.05 & (tma_memory_bound > 0.2 & tma_backend_bound > 0.2))", - "PublicDescription": "This metric measures fraction of cycles where the Super Queue (SQ) was full taking into account all request-types and both hardware SMT threads (Logical Processors). Related metrics: tma_bottleneck_data_cache_memory_bandwidth, tma_fb_full, tma_info_system_dram_bw_use, tma_mem_bandwidth", + "PublicDescription": "This metric measures fraction of cycles where the Super Queue (SQ) was full taking into account all request-types and both hardware SMT threads (Logical Processors). Related metrics: tma_bottleneck_data_cache_memory_bandwidth, tma_fb_full, tma_info_system_dram_bw_use, tma_mem_bandwidth, tma_uc_bound", "ScaleUnit": "100%" }, {
@@ -2277,6 +2308,12 @@ "MetricName": "uncore_frequency", "ScaleUnit": "1GHz" }, + { + "BriefDescription": "Intel(R) Ultra Path Interconnect (UPI) data receive bandwidth (MB/sec)", + "MetricExpr": "UNC_UPI_RxL_FLITS.ALL_DATA * 7.111111111111111 / 1e6 / duration_time", + "MetricName": "upi_data_receive_bw", + "ScaleUnit": "1MB/s" + }, { "BriefDescription": "Intel(R) Ultra Path Interconnect (UPI) data transmit bandwidth (MB/sec)", "MetricExpr": "UNC_UPI_TxL_FLITS.ALL_DATA * 7.111111111111111 / 1e6 / duration_time",
tools/perf/pmu-events/arch/x86/emeraldrapids/metricgroups.json +1 -1
@@ -90,7 +90,6 @@ "tma_code_stlb_miss_group": "Metrics contributing to tma_code_stlb_miss category", "tma_core_bound_group": "Metrics contributing to tma_core_bound category", "tma_divider_group": "Metrics contributing to tma_divider category", - "tma_dram_bound_group": "Metrics contributing to tma_dram_bound category", "tma_dtlb_load_group": "Metrics contributing to tma_dtlb_load category", "tma_dtlb_store_group": "Metrics contributing to tma_dtlb_store category", "tma_fetch_bandwidth_group": "Metrics contributing to tma_fetch_bandwidth category",
@@ -124,6 +123,7 @@ "tma_l1_bound_group": "Metrics contributing to tma_l1_bound category", "tma_l2_bound_group": "Metrics contributing to tma_l2_bound category", "tma_l3_bound_group": "Metrics contributing to tma_l3_bound category", + "tma_l3_miss_bound_group": "Metrics contributing to tma_l3_miss_bound category", "tma_light_operations_group": "Metrics contributing to tma_light_operations category", "tma_load_op_utilization_group": "Metrics contributing to tma_load_op_utilization category", "tma_load_stlb_miss_group": "Metrics contributing to tma_load_stlb_miss category",
--
2.55.0.795.g602f6c329a-goog
The patchset looks good to me. But It depends on this patch
https://lore.kernel.org/all/[email protected]/.
Without this patch, the building on NVL would fail. Thanks.
Quoted 85 lines · Chun-Tse Shao
Sync the Linux kernel perf PMU events with the latest upstream Intel perfmon repository.
This patchset includes event and metric updates for Alder Lake, Alder Lake N, Arrow Lake, Clearwater Forest, Emerald Rapids (including IIO bandwidth metrics), Grand Ridge, Granite Rapids, Lunar Lake, Meteor Lake, Panther Lake, and Sierra Forest, as well as adding initial event support for Nova Lake v1.00.
Chun-Tse Shao (12):
perf vendor events intel: Update alderlake events from 1.39 to 1.40 perf vendor events intel: Update alderlaken events from 1.39 to 1.40 perf vendor events intel: Update arrowlake events from 1.19 to 1.20 perf vendor events intel: Update clearwaterforest events from 1.02 to1.04 perf vendor events intel: Update grandridge events from 1.12 to 1.13 perf vendor events intel: Update graniterapids events from 1.19 to 1.20perf vendor events intel: Update lunarlake events from 1.25 to 1.26 perf vendor events intel: Update meteorlake events from 1.21 to 1.22 perf vendor events intel: Add novalake v1.00 events
perf vendor events intel: Update pantherlake events from 1.06 to 1.07 perf vendor events intel: Update sierraforest events from 1.17 to 1.18 perf vendor events intel: Update emeraldrapids metrics.../arch/x86/alderlake/adl-metrics.json | 80 +- .../pmu-events/arch/x86/alderlake/cache.json | 36 + .../arch/x86/alderlake/metricgroups.json | 2 +- .../arch/x86/alderlake/pipeline.json | 6 +- .../pmu-events/arch/x86/alderlaken/cache.json | 32 + .../arch/x86/arrowlake/arl-metrics.json | 152 +- .../arch/x86/arrowlake/metricgroups.json | 2 +- .../arch/x86/arrowlake/pipeline.json | 2 +- .../arch/x86/arrowlake/uncore-memory.json | 84 +- .../arch/x86/clearwaterforest/memory.json | 33 + .../arch/x86/clearwaterforest/pipeline.json | 19 + .../clearwaterforest/uncore-interconnect.json | 10 + .../x86/clearwaterforest/uncore-memory.json | 2 +- .../arch/x86/emeraldrapids/emr-metrics.json | 135 +- .../arch/x86/emeraldrapids/metricgroups.json | 2 +- .../pmu-events/arch/x86/grandridge/cache.json | 25 + .../arch/x86/grandridge/counter.json | 42 - .../arch/x86/grandridge/floating-point.json | 56 + .../arch/x86/grandridge/memory.json | 57 + .../arch/x86/grandridge/pipeline.json | 140 ++ .../arch/x86/grandridge/uncore-memory.json | 2 +- .../arch/x86/grandridge/virtual-memory.json | 8 + .../arch/x86/graniterapids/gnr-metrics.json | 102 +- .../arch/x86/graniterapids/metricgroups.json | 2 +- .../arch/x86/graniterapids/pipeline.json | 2 +- .../arch/x86/lunarlake/lnl-metrics.json | 142 +- .../arch/x86/lunarlake/metricgroups.json | 2 +- .../arch/x86/lunarlake/pipeline.json | 2 +- tools/perf/pmu-events/arch/x86/mapfile.csv | 21 +- .../pmu-events/arch/x86/meteorlake/cache.json | 11 +- .../arch/x86/meteorlake/metricgroups.json | 2 +- .../arch/x86/meteorlake/mtl-metrics.json | 88 +- .../arch/x86/meteorlake/pipeline.json | 4 +- .../arch/x86/meteorlake/uncore-memory.json | 84 +- .../pmu-events/arch/x86/novalake/cache.json | 749 +++++++ .../pmu-events/arch/x86/novalake/counter.json | 17 + .../arch/x86/novalake/floating-point.json | 452 ++++ .../arch/x86/novalake/frontend.json | 600 +++++ .../pmu-events/arch/x86/novalake/memory.json | 299 +++ .../pmu-events/arch/x86/novalake/other.json | 144 ++ .../arch/x86/novalake/pipeline.json | 1976 +++++++++++++++++ .../arch/x86/novalake/uncore-memory.json | 42 + .../arch/x86/novalake/virtual-memory.json | 338 +++ .../arch/x86/pantherlake/pipeline.json | 2 +- .../arch/x86/sierraforest/counter.json | 77 - .../x86/sierraforest/uncore-interconnect.json | 10 + .../arch/x86/sierraforest/uncore-memory.json | 2 +- 47 files changed, 5615 insertions(+), 482 deletions(-) delete mode 100644 tools/perf/pmu-events/arch/x86/grandridge/counter.json create mode 100644 tools/perf/pmu-events/arch/x86/novalake/cache.json create mode 100644 tools/perf/pmu-events/arch/x86/novalake/counter.json create mode 100644 tools/perf/pmu-events/arch/x86/novalake/floating-point.json create mode 100644 tools/perf/pmu-events/arch/x86/novalake/frontend.json create mode 100644 tools/perf/pmu-events/arch/x86/novalake/memory.json create mode 100644 tools/perf/pmu-events/arch/x86/novalake/other.json create mode 100644 tools/perf/pmu-events/arch/x86/novalake/pipeline.json create mode 100644 tools/perf/pmu-events/arch/x86/novalake/uncore-memory.json create mode 100644 tools/perf/pmu-events/arch/x86/novalake/virtual-memory.json delete mode 100644 tools/perf/pmu-events/arch/x86/sierraforest/counter.json
Namhyung Kim
maintainerHello,
Quoted 3 lines · Dapeng
The patchset looks good to me. But It depends on this patch
https://lore.kernel.org/all/[email protected]/.
Without this patch, the building on NVL would fail. Thanks.
I'll pick that up too.
Quoted 2 lines
Reviewed-by Dapeng Mi <[email protected]>
Thanks for the review!
Namhyung
Quoted 87 lines · Chun-Tse Shao
Quoted 85 lines
Sync the Linux kernel perf PMU events with the latest upstream Intel perfmon repository.
This patchset includes event and metric updates for Alder Lake, Alder Lake N, Arrow Lake, Clearwater Forest, Emerald Rapids (including IIO bandwidth metrics), Grand Ridge, Granite Rapids, Lunar Lake, Meteor Lake, Panther Lake, and Sierra Forest, as well as adding initial event support for Nova Lake v1.00.
Chun-Tse Shao (12):
perf vendor events intel: Update alderlake events from 1.39 to 1.40 perf vendor events intel: Update alderlaken events from 1.39 to 1.40 perf vendor events intel: Update arrowlake events from 1.19 to 1.20 perf vendor events intel: Update clearwaterforest events from 1.02 to1.04 perf vendor events intel: Update grandridge events from 1.12 to 1.13 perf vendor events intel: Update graniterapids events from 1.19 to 1.20perf vendor events intel: Update lunarlake events from 1.25 to 1.26 perf vendor events intel: Update meteorlake events from 1.21 to 1.22 perf vendor events intel: Add novalake v1.00 events
perf vendor events intel: Update pantherlake events from 1.06 to 1.07 perf vendor events intel: Update sierraforest events from 1.17 to 1.18 perf vendor events intel: Update emeraldrapids metrics.../arch/x86/alderlake/adl-metrics.json | 80 +- .../pmu-events/arch/x86/alderlake/cache.json | 36 + .../arch/x86/alderlake/metricgroups.json | 2 +- .../arch/x86/alderlake/pipeline.json | 6 +- .../pmu-events/arch/x86/alderlaken/cache.json | 32 + .../arch/x86/arrowlake/arl-metrics.json | 152 +- .../arch/x86/arrowlake/metricgroups.json | 2 +- .../arch/x86/arrowlake/pipeline.json | 2 +- .../arch/x86/arrowlake/uncore-memory.json | 84 +- .../arch/x86/clearwaterforest/memory.json | 33 + .../arch/x86/clearwaterforest/pipeline.json | 19 + .../clearwaterforest/uncore-interconnect.json | 10 + .../x86/clearwaterforest/uncore-memory.json | 2 +- .../arch/x86/emeraldrapids/emr-metrics.json | 135 +- .../arch/x86/emeraldrapids/metricgroups.json | 2 +- .../pmu-events/arch/x86/grandridge/cache.json | 25 + .../arch/x86/grandridge/counter.json | 42 - .../arch/x86/grandridge/floating-point.json | 56 + .../arch/x86/grandridge/memory.json | 57 + .../arch/x86/grandridge/pipeline.json | 140 ++ .../arch/x86/grandridge/uncore-memory.json | 2 +- .../arch/x86/grandridge/virtual-memory.json | 8 + .../arch/x86/graniterapids/gnr-metrics.json | 102 +- .../arch/x86/graniterapids/metricgroups.json | 2 +- .../arch/x86/graniterapids/pipeline.json | 2 +- .../arch/x86/lunarlake/lnl-metrics.json | 142 +- .../arch/x86/lunarlake/metricgroups.json | 2 +- .../arch/x86/lunarlake/pipeline.json | 2 +- tools/perf/pmu-events/arch/x86/mapfile.csv | 21 +- .../pmu-events/arch/x86/meteorlake/cache.json | 11 +- .../arch/x86/meteorlake/metricgroups.json | 2 +- .../arch/x86/meteorlake/mtl-metrics.json | 88 +- .../arch/x86/meteorlake/pipeline.json | 4 +- .../arch/x86/meteorlake/uncore-memory.json | 84 +- .../pmu-events/arch/x86/novalake/cache.json | 749 +++++++ .../pmu-events/arch/x86/novalake/counter.json | 17 + .../arch/x86/novalake/floating-point.json | 452 ++++ .../arch/x86/novalake/frontend.json | 600 +++++ .../pmu-events/arch/x86/novalake/memory.json | 299 +++ .../pmu-events/arch/x86/novalake/other.json | 144 ++ .../arch/x86/novalake/pipeline.json | 1976 +++++++++++++++++ .../arch/x86/novalake/uncore-memory.json | 42 + .../arch/x86/novalake/virtual-memory.json | 338 +++ .../arch/x86/pantherlake/pipeline.json | 2 +- .../arch/x86/sierraforest/counter.json | 77 - .../x86/sierraforest/uncore-interconnect.json | 10 + .../arch/x86/sierraforest/uncore-memory.json | 2 +- 47 files changed, 5615 insertions(+), 482 deletions(-) delete mode 100644 tools/perf/pmu-events/arch/x86/grandridge/counter.json create mode 100644 tools/perf/pmu-events/arch/x86/novalake/cache.json create mode 100644 tools/perf/pmu-events/arch/x86/novalake/counter.json create mode 100644 tools/perf/pmu-events/arch/x86/novalake/floating-point.json create mode 100644 tools/perf/pmu-events/arch/x86/novalake/frontend.json create mode 100644 tools/perf/pmu-events/arch/x86/novalake/memory.json create mode 100644 tools/perf/pmu-events/arch/x86/novalake/other.json create mode 100644 tools/perf/pmu-events/arch/x86/novalake/pipeline.json create mode 100644 tools/perf/pmu-events/arch/x86/novalake/uncore-memory.json create mode 100644 tools/perf/pmu-events/arch/x86/novalake/virtual-memory.json delete mode 100644 tools/perf/pmu-events/arch/x86/sierraforest/counter.json
Thanks Dapeng and Namhyung, I submitted v2 for fixing the wrong cpu_id for Novalake: lore.kernel.org/[email protected]
-CT
Quoted 104 lines · Namhyung Kim
Hello,
Quoted 3 lines · Dapeng
The patchset looks good to me. But It depends on this patch
https://lore.kernel.org/all/[email protected]/.
Without this patch, the building on NVL would fail. Thanks.I'll pick that up too.
Quoted 2 lines
Reviewed-by Dapeng Mi <[email protected]>Thanks for the review!
Namhyung
Quoted 87 lines · Chun-Tse Shao
Quoted 85 lines
Sync the Linux kernel perf PMU events with the latest upstream Intel perfmon repository.
This patchset includes event and metric updates for Alder Lake, Alder Lake N, Arrow Lake, Clearwater Forest, Emerald Rapids (including IIO bandwidth metrics), Grand Ridge, Granite Rapids, Lunar Lake, Meteor Lake, Panther Lake, and Sierra Forest, as well as adding initial event support for Nova Lake v1.00.
Chun-Tse Shao (12):
perf vendor events intel: Update alderlake events from 1.39 to 1.40 perf vendor events intel: Update alderlaken events from 1.39 to 1.40 perf vendor events intel: Update arrowlake events from 1.19 to 1.20 perf vendor events intel: Update clearwaterforest events from 1.02 to1.04 perf vendor events intel: Update grandridge events from 1.12 to 1.13 perf vendor events intel: Update graniterapids events from 1.19 to 1.20perf vendor events intel: Update lunarlake events from 1.25 to 1.26 perf vendor events intel: Update meteorlake events from 1.21 to 1.22 perf vendor events intel: Add novalake v1.00 events
perf vendor events intel: Update pantherlake events from 1.06 to 1.07 perf vendor events intel: Update sierraforest events from 1.17 to 1.18 perf vendor events intel: Update emeraldrapids metrics.../arch/x86/alderlake/adl-metrics.json | 80 +- .../pmu-events/arch/x86/alderlake/cache.json | 36 + .../arch/x86/alderlake/metricgroups.json | 2 +- .../arch/x86/alderlake/pipeline.json | 6 +- .../pmu-events/arch/x86/alderlaken/cache.json | 32 + .../arch/x86/arrowlake/arl-metrics.json | 152 +- .../arch/x86/arrowlake/metricgroups.json | 2 +- .../arch/x86/arrowlake/pipeline.json | 2 +- .../arch/x86/arrowlake/uncore-memory.json | 84 +- .../arch/x86/clearwaterforest/memory.json | 33 + .../arch/x86/clearwaterforest/pipeline.json | 19 + .../clearwaterforest/uncore-interconnect.json | 10 + .../x86/clearwaterforest/uncore-memory.json | 2 +- .../arch/x86/emeraldrapids/emr-metrics.json | 135 +- .../arch/x86/emeraldrapids/metricgroups.json | 2 +- .../pmu-events/arch/x86/grandridge/cache.json | 25 + .../arch/x86/grandridge/counter.json | 42 - .../arch/x86/grandridge/floating-point.json | 56 + .../arch/x86/grandridge/memory.json | 57 + .../arch/x86/grandridge/pipeline.json | 140 ++ .../arch/x86/grandridge/uncore-memory.json | 2 +- .../arch/x86/grandridge/virtual-memory.json | 8 + .../arch/x86/graniterapids/gnr-metrics.json | 102 +- .../arch/x86/graniterapids/metricgroups.json | 2 +- .../arch/x86/graniterapids/pipeline.json | 2 +- .../arch/x86/lunarlake/lnl-metrics.json | 142 +- .../arch/x86/lunarlake/metricgroups.json | 2 +- .../arch/x86/lunarlake/pipeline.json | 2 +- tools/perf/pmu-events/arch/x86/mapfile.csv | 21 +- .../pmu-events/arch/x86/meteorlake/cache.json | 11 +- .../arch/x86/meteorlake/metricgroups.json | 2 +- .../arch/x86/meteorlake/mtl-metrics.json | 88 +- .../arch/x86/meteorlake/pipeline.json | 4 +- .../arch/x86/meteorlake/uncore-memory.json | 84 +- .../pmu-events/arch/x86/novalake/cache.json | 749 +++++++ .../pmu-events/arch/x86/novalake/counter.json | 17 + .../arch/x86/novalake/floating-point.json | 452 ++++ .../arch/x86/novalake/frontend.json | 600 +++++ .../pmu-events/arch/x86/novalake/memory.json | 299 +++ .../pmu-events/arch/x86/novalake/other.json | 144 ++ .../arch/x86/novalake/pipeline.json | 1976 +++++++++++++++++ .../arch/x86/novalake/uncore-memory.json | 42 + .../arch/x86/novalake/virtual-memory.json | 338 +++ .../arch/x86/pantherlake/pipeline.json | 2 +- .../arch/x86/sierraforest/counter.json | 77 - .../x86/sierraforest/uncore-interconnect.json | 10 + .../arch/x86/sierraforest/uncore-memory.json | 2 +- 47 files changed, 5615 insertions(+), 482 deletions(-) delete mode 100644 tools/perf/pmu-events/arch/x86/grandridge/counter.json create mode 100644 tools/perf/pmu-events/arch/x86/novalake/cache.json create mode 100644 tools/perf/pmu-events/arch/x86/novalake/counter.json create mode 100644 tools/perf/pmu-events/arch/x86/novalake/floating-point.json create mode 100644 tools/perf/pmu-events/arch/x86/novalake/frontend.json create mode 100644 tools/perf/pmu-events/arch/x86/novalake/memory.json create mode 100644 tools/perf/pmu-events/arch/x86/novalake/other.json create mode 100644 tools/perf/pmu-events/arch/x86/novalake/pipeline.json create mode 100644 tools/perf/pmu-events/arch/x86/novalake/uncore-memory.json create mode 100644 tools/perf/pmu-events/arch/x86/novalake/virtual-memory.json delete mode 100644 tools/perf/pmu-events/arch/x86/sierraforest/counter.json
Quoted 10 lines · Chun-Tse Shao
Sync the Linux kernel perf PMU events with the latest upstream Intel perfmon repository.
This patchset includes event and metric updates for Alder Lake, Alder Lake N, Arrow Lake, Clearwater Forest, Emerald Rapids (including IIO bandwidth metrics), Grand Ridge, Granite Rapids, Lunar Lake, Meteor Lake, Panther Lake, and Sierra Forest, as well as adding initial event support for Nova Lake v1.00.
[...]
Applied to perf-tools-next, thanks!
Best regards,
Namhyung