diff --git a/CHANGELOG.md b/CHANGELOG.md index 558a605c7e..98c1824514 100644 --- a/CHANGELOG.md +++ b/CHANGELOG.md @@ -16,24 +16,25 @@ Full documentation for ROCm Compute Profiler is available at [https://rocm.docs. ### Changed * Change normal_unit default to per_kernel -* change dependency from rocm-smi to amd-smi +* Change dependency from rocm-smi to amd-smi +* Decrease profiling time by not collecting counters not used in post analysis ### Resolved issues * Fixed option specs-correction * Fixed kernel name and kernel dispatch filtering when using rocprof v3 +* Fixed not collecting TCC channel counters in rocprof v3 ### Known issues -* gpu id filtering is not supported when using rocprof v3 +* GPU id filtering is not supported when using rocprof v3 -## (Unreleased) ROCm Compute Profiler 3.1.0 for ROCm 6.4.0 +## ROCm Compute Profiler 3.1.0 for ROCm 6.4.0 ### Added * Roofline support for Ubuntu 24.04 * Experimental support rocprofv3 (not enabled as default) -* Experimental feature: Spatial multiplexing ### Resolved issues diff --git a/src/rocprof_compute_base.py b/src/rocprof_compute_base.py index e81615db1d..696e34be18 100644 --- a/src/rocprof_compute_base.py +++ b/src/rocprof_compute_base.py @@ -42,10 +42,7 @@ from utils.logger import ( setup_file_handler, setup_logging_priority, ) -from utils.mi_gpu_spec import ( - get_gpu_series_dict, - parse_mi_gpu_spec, -) +from utils.mi_gpu_spec import get_gpu_series_dict, parse_mi_gpu_spec from utils.specs import MachineSpecs, generate_machine_specs from utils.utils import ( console_debug, diff --git a/src/rocprof_compute_profile/profiler_base.py b/src/rocprof_compute_profile/profiler_base.py index 23ac4580b7..f545ac7932 100644 --- a/src/rocprof_compute_profile/profiler_base.py +++ b/src/rocprof_compute_profile/profiler_base.py @@ -55,11 +55,6 @@ class RocProfCompute_Base: self.__profiler = profiler_mode self.__supported_archs = supported_archs self._soc = soc # OmniSoC obj - self.__perfmon_dir = str( - Path(str(config.rocprof_compute_home)).joinpath( - "rocprof_compute_soc", "profile_configs" - ) - ) self.__filter_hardware_blocks = [ name for name, type in args.filter_blocks.items() if type == "hardware_block" ] diff --git a/src/rocprof_compute_profile/profiler_rocprof_v1.py b/src/rocprof_compute_profile/profiler_rocprof_v1.py index 53e54b70fc..3b875b322e 100644 --- a/src/rocprof_compute_profile/profiler_rocprof_v1.py +++ b/src/rocprof_compute_profile/profiler_rocprof_v1.py @@ -25,6 +25,7 @@ import os from pathlib import Path +import config from rocprof_compute_profile.profiler_base import RocProfCompute_Base from utils.utils import console_log, demarcate, replace_timestamps, store_app_cmd @@ -43,9 +44,14 @@ class rocprof_v1_profiler(RocProfCompute_Base): app_cmd = self.get_args().remaining args = [] - # can be removed in the future. It supports gfx908 + v1 + # rocprof v1 does not support some counters on gfx 908 architecture if soc.get_arch() == "gfx908": - args += ["-m", soc.get_workload_perfmon_dir() + "/" + "metrics.xml"] + metrics_path = str( + Path(str(config.rocprof_compute_home)).joinpath( + "rocprof_compute_soc", "profile_configs", "metrics.xml" + ) + ) + args += ["-m", metrics_path] args += [ # v1 requires request for timestamps diff --git a/src/rocprof_compute_profile/profiler_rocprof_v2.py b/src/rocprof_compute_profile/profiler_rocprof_v2.py index 4b04f92e64..12e649d879 100644 --- a/src/rocprof_compute_profile/profiler_rocprof_v2.py +++ b/src/rocprof_compute_profile/profiler_rocprof_v2.py @@ -26,6 +26,7 @@ import os import shlex from pathlib import Path +import config from rocprof_compute_profile.profiler_base import RocProfCompute_Base from utils.utils import console_log, demarcate, replace_timestamps, store_app_cmd @@ -44,9 +45,14 @@ class rocprof_v2_profiler(RocProfCompute_Base): app_cmd = shlex.split(self.get_args().remaining) args = [] - # can be removed in the future. It supports gfx908 + v2 + # rocprof v2 does not support some counters on gfx 908 architecture if soc.get_arch() == "gfx908": - args += ["-m", soc.get_workload_perfmon_dir() + "/" + "metrics.xml"] + metrics_path = str( + Path(str(config.rocprof_compute_home)).joinpath( + "rocprof_compute_soc", "profile_configs", "metrics.xml" + ) + ) + args += ["-m", metrics_path] args += [ # v2 requires output directory argument diff --git a/src/rocprof_compute_soc/profile_configs/gfx900/pmc_cpc_perf.txt b/src/rocprof_compute_soc/profile_configs/gfx900/pmc_cpc_perf.txt deleted file mode 100644 index bd80137304..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx900/pmc_cpc_perf.txt +++ /dev/null @@ -1,13 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_CYCLES SQ_BUSY_CYCLES SQ_WAVES - -pmc: CPC_CPC_STAT_BUSY CPC_CPC_STAT_IDLE -pmc: CPC_CPC_TCIU_BUSY CPC_CPC_TCIU_IDLE -pmc: CPC_CPC_STAT_STALL CPC_UTCL1_STALL_ON_TRANSLATION -pmc: CPC_CPC_UTCL2IU_BUSY CPC_CPC_UTCL2IU_IDLE -pmc: CPC_CPC_UTCL2IU_STALL CPC_ME1_BUSY_FOR_PACKET_DECODE -pmc: CPC_ME1_DC0_SPI_BUSY - -range: -gpu: - -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx900/pmc_cpf_perf.txt b/src/rocprof_compute_soc/profile_configs/gfx900/pmc_cpf_perf.txt deleted file mode 100644 index e8119fff56..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx900/pmc_cpf_perf.txt +++ /dev/null @@ -1,11 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_CYCLES SQ_BUSY_CYCLES SQ_WAVES - -pmc: CPF_CPF_STAT_BUSY CPF_CPF_STAT_STALL -pmc: CPF_CPF_TCIU_BUSY CPF_CPF_TCIU_STALL -pmc: CPF_CPF_STAT_IDLE CPF_CPF_TCIU_IDLE -pmc: CPF_CMP_UTCL1_STALL_ON_TRANSLATION - -range: -gpu: - -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx900/pmc_spi_perf.txt b/src/rocprof_compute_soc/profile_configs/gfx900/pmc_spi_perf.txt deleted file mode 100644 index 4eb0d22656..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx900/pmc_spi_perf.txt +++ /dev/null @@ -1,10 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_CYCLES SQ_BUSY_CYCLES SQ_WAVES GRBM_SPI_BUSY - -pmc: SPI_CSN_WINDOW_VALID SPI_CSN_BUSY SPI_CSN_NUM_THREADGROUPS SPI_CSN_WAVE SPI_RA_REQ_NO_ALLOC SPI_RA_REQ_NO_ALLOC_CSN -pmc: SPI_RA_RES_STALL_CSN SPI_RA_TMP_STALL_CSN SPI_RA_WAVE_SIMD_FULL_CSN SPI_RA_VGPR_SIMD_FULL_CSN SPI_RA_SGPR_SIMD_FULL_CSN SPI_RA_LDS_CU_FULL_CSN -pmc: SPI_RA_BAR_CU_FULL_CSN SPI_RA_TGLIM_CU_FULL_CSN SPI_RA_WVLIM_STALL_CSN SPI_SWC_CSC_WR SPI_VWC_CSC_WR SPI_RA_BULKY_CU_FULL_CSN - -range: -gpu: - -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx900/pmc_sq_perf1.txt b/src/rocprof_compute_soc/profile_configs/gfx900/pmc_sq_perf1.txt deleted file mode 100644 index 72afa13fb4..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx900/pmc_sq_perf1.txt +++ /dev/null @@ -1,27 +0,0 @@ -#SQ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_INSTS_VMEM_WR SQ_INSTS_VMEM_RD SQ_INSTS_VMEM SQ_INSTS_SALU SQ_INSTS_VSKIPPED -pmc: SQ_INSTS_SMEM SQ_INSTS_FLAT SQ_INSTS_LDS SQ_INSTS_GDS SQ_INSTS_EXP_GDS SQ_INSTS_BRANCH SQ_INSTS_SENDMSG SQ_INSTS -pmc: SQ_WAVE_CYCLES SQ_WAIT_ANY SQ_WAIT_INST_ANY SQ_ACTIVE_INST_ANY SQ_CYCLES SQ_BUSY_CYCLES SQ_BUSY_CU_CYCLES SQ_INSTS_VALU -pmc: SQ_ACTIVE_INST_VMEM SQ_ACTIVE_INST_LDS SQ_ACTIVE_INST_VALU SQ_ACTIVE_INST_SCA SQ_ACTIVE_INST_EXP_GDS SQ_ACTIVE_INST_MISC SQ_ACTIVE_INST_FLAT SQ_INST_CYCLES_VMEM_WR -pmc: SQ_INST_CYCLES_VMEM_RD SQ_INST_CYCLES_SMEM SQ_INST_CYCLES_SALU SQ_THREAD_CYCLES_VALU SQ_IFETCH SQ_LDS_BANK_CONFLICT SQ_LDS_ADDR_CONFLICT SQ_LDS_UNALIGNED_STALL -pmc: SQ_WAVES SQ_WAVES_EQ_64 SQ_WAVES_LT_64 SQ_WAVES_LT_48 SQ_WAVES_LT_32 SQ_WAVES_LT_16 SQ_ITEMS SQ_INSTS_VSKIPPED -pmc: SQ_LDS_MEM_VIOLATIONS SQ_LDS_ATOMIC_RETURN SQ_LDS_IDX_ACTIVE SQ_WAVES_RESTORED SQ_WAVES_SAVED SQ_INSTS_SMEM_NORM - - -#SQ:MI200 -#pmc: SQ_INSTS_MFMA SQ_INSTS_VALU_MFMA_I8 SQ_INSTS_VALU_MFMA_F16 SQ_INSTS_VALU_MFMA_BF16 SQ_INSTS_VALU_MFMA_F32 SQ_INSTS_VALU_MFMA_F64 SQ_VALU_MFMA_BUSY_CYCLES -#pmc: SQ_INSTS_FLAT_LDS_ONLY SQ_INSTS_VALU_MFMA_MOPS_I8 SQ_INSTS_VALU_MFMA_MOPS_F16 SQ_INSTS_VALU_MFMA_MOPS_BF16 SQ_INSTS_VALU_MFMA_MOPS_F32 SQ_INSTS_VALU_MFMA_MOPS_F64 - -#SQC -pmc: SQC_TC_INST_REQ SQC_TC_DATA_READ_REQ SQC_TC_DATA_WRITE_REQ SQC_TC_DATA_ATOMIC_REQ SQC_TC_STALL SQC_TC_REQ SQC_DCACHE_REQ_READ_16 -pmc: SQC_ICACHE_REQ SQC_ICACHE_HITS SQC_ICACHE_MISSES SQC_ICACHE_MISSES_DUPLICATE SQC_DCACHE_INPUT_VALID_READYB SQC_DCACHE_ATOMIC SQC_DCACHE_REQ_READ_8 -pmc: SQC_DCACHE_REQ SQC_DCACHE_HITS SQC_DCACHE_MISSES SQC_DCACHE_MISSES_DUPLICATE SQC_DCACHE_REQ_READ_1 SQC_DCACHE_REQ_READ_2 SQC_DCACHE_REQ_READ_4 - - - -######################################## -# Filtering -######################################## -range: -gpu: -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx900/pmc_sq_perf2.txt b/src/rocprof_compute_soc/profile_configs/gfx900/pmc_sq_perf2.txt deleted file mode 100644 index 7175c0326c..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx900/pmc_sq_perf2.txt +++ /dev/null @@ -1,10 +0,0 @@ -################################################# -# VMEM latency -################################################# -pmc: SQ_INSTS_VMEM SQ_INST_LEVEL_VMEM SQ_ACCUM_PREV_HIRES - - -range: -gpu: - -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx900/pmc_sq_perf3.txt b/src/rocprof_compute_soc/profile_configs/gfx900/pmc_sq_perf3.txt deleted file mode 100644 index 8d36311af0..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx900/pmc_sq_perf3.txt +++ /dev/null @@ -1,11 +0,0 @@ -################################################# -# SMEM latency -################################################# -pmc: SQ_INSTS_SMEM SQ_INST_LEVEL_SMEM SQ_ACCUM_PREV_HIRES - - -range: -gpu: - - -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx900/pmc_sq_perf4.txt b/src/rocprof_compute_soc/profile_configs/gfx900/pmc_sq_perf4.txt deleted file mode 100644 index f8f29b4f54..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx900/pmc_sq_perf4.txt +++ /dev/null @@ -1,8 +0,0 @@ -################################################# -# ifetch latency -################################################# -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_WAVES SQ_IFETCH SQ_IFETCH_LEVEL SQ_ACCUM_PREV_HIRES - -range: -gpu: -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx900/pmc_sq_perf6.txt b/src/rocprof_compute_soc/profile_configs/gfx900/pmc_sq_perf6.txt deleted file mode 100644 index f9144576eb..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx900/pmc_sq_perf6.txt +++ /dev/null @@ -1,9 +0,0 @@ -################################################# -# LDS latency -################################################# -pmc: SQ_INSTS_LDS SQ_INST_LEVEL_LDS SQ_ACCUM_PREV_HIRES - -range: -gpu: - -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx900/pmc_sq_perf8.txt b/src/rocprof_compute_soc/profile_configs/gfx900/pmc_sq_perf8.txt deleted file mode 100644 index cb29ad2ce1..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx900/pmc_sq_perf8.txt +++ /dev/null @@ -1,6 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE CPC_ME1_BUSY_FOR_PACKET_DECODE SQ_CYCLES SQ_WAVES SQ_WAVE_CYCLES SQ_BUSY_CYCLES SQ_LEVEL_WAVES SQ_ACCUM_PREV_HIRES - -gpu: -range: - -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx900/pmc_sqc_perf1.txt b/src/rocprof_compute_soc/profile_configs/gfx900/pmc_sqc_perf1.txt deleted file mode 100644 index e602d35c1f..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx900/pmc_sqc_perf1.txt +++ /dev/null @@ -1,11 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_CYCLES SQ_BUSY_CYCLES SQ_BUSY_CU_CYCLES SQ_WAVES SQ_WAVE_CYCLES - - -pmc: SQC_TC_INST_REQ SQC_TC_DATA_READ_REQ SQC_TC_DATA_WRITE_REQ SQC_TC_DATA_ATOMIC_REQ SQC_TC_STALL SQC_TC_REQ SQC_DCACHE_REQ_READ_16 -pmc: SQC_ICACHE_REQ SQC_ICACHE_HITS SQC_ICACHE_MISSES SQC_ICACHE_MISSES_DUPLICATE SQC_DCACHE_INPUT_VALID_READYB SQC_DCACHE_ATOMIC SQC_DCACHE_REQ_READ_8 -pmc: SQC_DCACHE_REQ SQC_DCACHE_HITS SQC_DCACHE_MISSES SQC_DCACHE_MISSES_DUPLICATE SQC_DCACHE_REQ_READ_1 SQC_DCACHE_REQ_READ_2 SQC_DCACHE_REQ_READ_4 - -range: -gpu: - -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx900/pmc_ta_perf.txt b/src/rocprof_compute_soc/profile_configs/gfx900/pmc_ta_perf.txt deleted file mode 100644 index aaaed2b06b..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx900/pmc_ta_perf.txt +++ /dev/null @@ -1,37 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_CYCLES SQ_BUSY_CYCLES SQ_BUSY_CU_CYCLES SQ_WAVES SQ_WAVE_CYCLES - -pmc: TA_TA_BUSY_sum TA_SH_FIFO_BUSY_sum -pmc: TA_SH_FIFO_CMD_BUSY_sum TA_SH_FIFO_ADDR_BUSY_sum -pmc: TA_SH_FIFO_DATA_BUSY_sum TA_SH_FIFO_DATA_SFIFO_BUSY_sum -pmc: TA_SH_FIFO_DATA_TFIFO_BUSY_sum TA_SQ_TA_CMD_CYCLES_sum -pmc: TA_SP_TA_ADDR_CYCLES_sum TA_SP_TA_DATA_CYCLES_sum - - -# Starvation -pmc: TA_SH_FIFO_ADDR_STARVED_WHILE_BUSY_CYCLES_sum TA_SH_FIFO_CMD_STARVED_WHILE_BUSY_CYCLES_sum -pmc: TA_SH_FIFO_DATA_STARVED_WHILE_BUSY_CYCLES_sum TA_TA_SH_FIFO_STARVED_sum - - - -# buffer access -pmc: TA_BUFFER_WAVEFRONTS_sum TA_BUFFER_READ_WAVEFRONTS_sum -pmc: TA_BUFFER_WRITE_WAVEFRONTS_sum TA_BUFFER_ATOMIC_WAVEFRONTS_sum -pmc: TA_BUFFER_TOTAL_CYCLES_sum TA_BUFFER_COALESCABLE_WAVEFRONTS_sum -pmc: TA_BUFFER_COALESCED_READ_CYCLES_sum TA_BUFFER_COALESCED_WRITE_CYCLES_sum - - -# stalls -pmc: TA_ADDR_STALLED_BY_TC_CYCLES_sum TA_TOTAL_WAVEFRONTS_sum -pmc: TA_ADDR_STALLED_BY_TD_CYCLES_sum TA_DATA_STALLED_BY_TC_CYCLES_sum - -# flat accesses -pmc: TA_FLAT_WAVEFRONTS_sum TA_FLAT_READ_WAVEFRONTS_sum -pmc: TA_FLAT_WRITE_WAVEFRONTS_sum TA_FLAT_ATOMIC_WAVEFRONTS_sum -pmc: TA_FLAT_COALESCEABLE_WAVEFRONTS_sum - - -range: - -gpu: - -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx900/pmc_tcc2_perf.txt b/src/rocprof_compute_soc/profile_configs/gfx900/pmc_tcc2_perf.txt deleted file mode 100644 index 1130c72a9c..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx900/pmc_tcc2_perf.txt +++ /dev/null @@ -1,25 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_CYCLES SQ_BUSY_CYCLES SQ_WAVES - -# MI50: no TCC_RW_REQ -pmc: TCC_CYCLE[0] TCC_HIT[0] TCC_MISS[0] TCC_CYCLE[1] TCC_HIT[1] TCC_MISS[1] TCC_CYCLE[2] TCC_HIT[2] TCC_MISS[2] TCC_CYCLE[3] TCC_HIT[3] TCC_MISS[3] TCC_CYCLE[4] TCC_HIT[4] TCC_MISS[4] TCC_CYCLE[5] TCC_HIT[5] TCC_MISS[5] TCC_CYCLE[6] TCC_HIT[6] TCC_MISS[6] TCC_CYCLE[7] TCC_HIT[7] TCC_MISS[7] TCC_CYCLE[8] TCC_HIT[8] TCC_MISS[8] TCC_CYCLE[9] TCC_HIT[9] TCC_MISS[9] TCC_CYCLE[10] TCC_HIT[10] TCC_MISS[10] TCC_CYCLE[11] TCC_HIT[11] TCC_MISS[11] TCC_CYCLE[12] TCC_HIT[12] TCC_MISS[12] TCC_CYCLE[13] TCC_HIT[13] TCC_MISS[13] TCC_CYCLE[14] TCC_HIT[14] TCC_MISS[14] TCC_CYCLE[15] TCC_HIT[15] TCC_MISS[15] - - - -pmc: TCC_REQ[0] TCC_READ[0] TCC_WRITE[0] TCC_ATOMIC[0] TCC_REQ[1] TCC_READ[1] TCC_WRITE[1] TCC_ATOMIC[1] TCC_REQ[2] TCC_READ[2] TCC_WRITE[2] TCC_ATOMIC[2] TCC_REQ[3] TCC_READ[3] TCC_WRITE[3] TCC_ATOMIC[3] TCC_REQ[4] TCC_READ[4] TCC_WRITE[4] TCC_ATOMIC[4] TCC_REQ[5] TCC_READ[5] TCC_WRITE[5] TCC_ATOMIC[5] TCC_REQ[6] TCC_READ[6] TCC_WRITE[6] TCC_ATOMIC[6] TCC_REQ[7] TCC_READ[7] TCC_WRITE[7] TCC_ATOMIC[7] TCC_REQ[8] TCC_READ[8] TCC_WRITE[8] TCC_ATOMIC[8] TCC_REQ[9] TCC_READ[9] TCC_WRITE[9] TCC_ATOMIC[9] TCC_REQ[10] TCC_READ[10] TCC_WRITE[10] TCC_ATOMIC[10] TCC_REQ[11] TCC_READ[11] TCC_WRITE[11] TCC_ATOMIC[11] TCC_REQ[12] TCC_READ[12] TCC_WRITE[12] TCC_ATOMIC[12] TCC_REQ[13] TCC_READ[13] TCC_WRITE[13] TCC_ATOMIC[13] TCC_REQ[14] TCC_READ[14] TCC_WRITE[14] TCC_ATOMIC[14] TCC_REQ[15] TCC_READ[15] TCC_WRITE[15] TCC_ATOMIC[15] - - - -pmc: TCC_EA_RDREQ[0] TCC_EA_RDREQ_32B[0] TCC_EA_WRREQ[0] TCC_EA_WRREQ_64B[0] TCC_EA_RDREQ[1] TCC_EA_RDREQ_32B[1] TCC_EA_WRREQ[1] TCC_EA_WRREQ_64B[1] TCC_EA_RDREQ[2] TCC_EA_RDREQ_32B[2] TCC_EA_WRREQ[2] TCC_EA_WRREQ_64B[2] TCC_EA_RDREQ[3] TCC_EA_RDREQ_32B[3] TCC_EA_WRREQ[3] TCC_EA_WRREQ_64B[3] TCC_EA_RDREQ[4] TCC_EA_RDREQ_32B[4] TCC_EA_WRREQ[4] TCC_EA_WRREQ_64B[4] TCC_EA_RDREQ[5] TCC_EA_RDREQ_32B[5] TCC_EA_WRREQ[5] TCC_EA_WRREQ_64B[5] TCC_EA_RDREQ[6] TCC_EA_RDREQ_32B[6] TCC_EA_WRREQ[6] TCC_EA_WRREQ_64B[6] TCC_EA_RDREQ[7] TCC_EA_RDREQ_32B[7] TCC_EA_WRREQ[7] TCC_EA_WRREQ_64B[7] TCC_EA_RDREQ[8] TCC_EA_RDREQ_32B[8] TCC_EA_WRREQ[8] TCC_EA_WRREQ_64B[8] TCC_EA_RDREQ[9] TCC_EA_RDREQ_32B[9] TCC_EA_WRREQ[9] TCC_EA_WRREQ_64B[9] TCC_EA_RDREQ[10] TCC_EA_RDREQ_32B[10] TCC_EA_WRREQ[10] TCC_EA_WRREQ_64B[10] TCC_EA_RDREQ[11] TCC_EA_RDREQ_32B[11] TCC_EA_WRREQ[11] TCC_EA_WRREQ_64B[11] TCC_EA_RDREQ[12] TCC_EA_RDREQ_32B[12] TCC_EA_WRREQ[12] TCC_EA_WRREQ_64B[12] TCC_EA_RDREQ[13] TCC_EA_RDREQ_32B[13] TCC_EA_WRREQ[13] TCC_EA_WRREQ_64B[13] TCC_EA_RDREQ[14] TCC_EA_RDREQ_32B[14] TCC_EA_WRREQ[14] TCC_EA_WRREQ_64B[14] TCC_EA_RDREQ[15] TCC_EA_RDREQ_32B[15] TCC_EA_WRREQ[15] TCC_EA_WRREQ_64B[15] - - -pmc: TCC_EA_ATOMIC[0] TCC_EA_RDREQ_LEVEL[0] TCC_EA_WRREQ_LEVEL[0] TCC_EA_ATOMIC_LEVEL[0] TCC_EA_ATOMIC[1] TCC_EA_RDREQ_LEVEL[1] TCC_EA_WRREQ_LEVEL[1] TCC_EA_ATOMIC_LEVEL[1] TCC_EA_ATOMIC[2] TCC_EA_RDREQ_LEVEL[2] TCC_EA_WRREQ_LEVEL[2] TCC_EA_ATOMIC_LEVEL[2] TCC_EA_ATOMIC[3] TCC_EA_RDREQ_LEVEL[3] TCC_EA_WRREQ_LEVEL[3] TCC_EA_ATOMIC_LEVEL[3] TCC_EA_ATOMIC[4] TCC_EA_RDREQ_LEVEL[4] TCC_EA_WRREQ_LEVEL[4] TCC_EA_ATOMIC_LEVEL[4] TCC_EA_ATOMIC[5] TCC_EA_RDREQ_LEVEL[5] TCC_EA_WRREQ_LEVEL[5] TCC_EA_ATOMIC_LEVEL[5] TCC_EA_ATOMIC[6] TCC_EA_RDREQ_LEVEL[6] TCC_EA_WRREQ_LEVEL[6] TCC_EA_ATOMIC_LEVEL[6] TCC_EA_ATOMIC[7] TCC_EA_RDREQ_LEVEL[7] TCC_EA_WRREQ_LEVEL[7] TCC_EA_ATOMIC_LEVEL[7] TCC_EA_ATOMIC[8] TCC_EA_RDREQ_LEVEL[8] TCC_EA_WRREQ_LEVEL[8] TCC_EA_ATOMIC_LEVEL[8] TCC_EA_ATOMIC[9] TCC_EA_RDREQ_LEVEL[9] TCC_EA_WRREQ_LEVEL[9] TCC_EA_ATOMIC_LEVEL[9] TCC_EA_ATOMIC[10] TCC_EA_RDREQ_LEVEL[10] TCC_EA_WRREQ_LEVEL[10] TCC_EA_ATOMIC_LEVEL[10] TCC_EA_ATOMIC[11] TCC_EA_RDREQ_LEVEL[11] TCC_EA_WRREQ_LEVEL[11] TCC_EA_ATOMIC_LEVEL[11] TCC_EA_ATOMIC[12] TCC_EA_RDREQ_LEVEL[12] TCC_EA_WRREQ_LEVEL[12] TCC_EA_ATOMIC_LEVEL[12] TCC_EA_ATOMIC[13] TCC_EA_RDREQ_LEVEL[13] TCC_EA_WRREQ_LEVEL[13] TCC_EA_ATOMIC_LEVEL[13] TCC_EA_ATOMIC[14] TCC_EA_RDREQ_LEVEL[14] TCC_EA_WRREQ_LEVEL[14] TCC_EA_ATOMIC_LEVEL[14] TCC_EA_ATOMIC[15] TCC_EA_RDREQ_LEVEL[15] TCC_EA_WRREQ_LEVEL[15] TCC_EA_ATOMIC_LEVEL[15] - - -pmc: TCC_EA_RDREQ_IO_CREDIT_STALL[0] TCC_EA_RDREQ_GMI_CREDIT_STALL[0] TCC_EA_RDREQ_DRAM_CREDIT_STALL[0] TCC_EA_RDREQ_IO_CREDIT_STALL[1] TCC_EA_RDREQ_GMI_CREDIT_STALL[1] TCC_EA_RDREQ_DRAM_CREDIT_STALL[1] TCC_EA_RDREQ_IO_CREDIT_STALL[2] TCC_EA_RDREQ_GMI_CREDIT_STALL[2] TCC_EA_RDREQ_DRAM_CREDIT_STALL[2] TCC_EA_RDREQ_IO_CREDIT_STALL[3] TCC_EA_RDREQ_GMI_CREDIT_STALL[3] TCC_EA_RDREQ_DRAM_CREDIT_STALL[3] TCC_EA_RDREQ_IO_CREDIT_STALL[4] TCC_EA_RDREQ_GMI_CREDIT_STALL[4] TCC_EA_RDREQ_DRAM_CREDIT_STALL[4] TCC_EA_RDREQ_IO_CREDIT_STALL[5] TCC_EA_RDREQ_GMI_CREDIT_STALL[5] TCC_EA_RDREQ_DRAM_CREDIT_STALL[5] TCC_EA_RDREQ_IO_CREDIT_STALL[6] TCC_EA_RDREQ_GMI_CREDIT_STALL[6] TCC_EA_RDREQ_DRAM_CREDIT_STALL[6] TCC_EA_RDREQ_IO_CREDIT_STALL[7] TCC_EA_RDREQ_GMI_CREDIT_STALL[7] TCC_EA_RDREQ_DRAM_CREDIT_STALL[7] TCC_EA_RDREQ_IO_CREDIT_STALL[8] TCC_EA_RDREQ_GMI_CREDIT_STALL[8] TCC_EA_RDREQ_DRAM_CREDIT_STALL[8] TCC_EA_RDREQ_IO_CREDIT_STALL[9] TCC_EA_RDREQ_GMI_CREDIT_STALL[9] TCC_EA_RDREQ_DRAM_CREDIT_STALL[9] TCC_EA_RDREQ_IO_CREDIT_STALL[10] TCC_EA_RDREQ_GMI_CREDIT_STALL[10] TCC_EA_RDREQ_DRAM_CREDIT_STALL[10] TCC_EA_RDREQ_IO_CREDIT_STALL[11] TCC_EA_RDREQ_GMI_CREDIT_STALL[11] TCC_EA_RDREQ_DRAM_CREDIT_STALL[11] TCC_EA_RDREQ_IO_CREDIT_STALL[12] TCC_EA_RDREQ_GMI_CREDIT_STALL[12] TCC_EA_RDREQ_DRAM_CREDIT_STALL[12] TCC_EA_RDREQ_IO_CREDIT_STALL[13] TCC_EA_RDREQ_GMI_CREDIT_STALL[13] TCC_EA_RDREQ_DRAM_CREDIT_STALL[13] TCC_EA_RDREQ_IO_CREDIT_STALL[14] TCC_EA_RDREQ_GMI_CREDIT_STALL[14] TCC_EA_RDREQ_DRAM_CREDIT_STALL[14] TCC_EA_RDREQ_IO_CREDIT_STALL[15] TCC_EA_RDREQ_GMI_CREDIT_STALL[15] TCC_EA_RDREQ_DRAM_CREDIT_STALL[15] - - -pmc: TCC_EA_WRREQ_IO_CREDIT_STALL[0] TCC_EA_WRREQ_GMI_CREDIT_STALL[0] TCC_EA_WRREQ_DRAM_CREDIT_STALL[0] TCC_TOO_MANY_EA_WRREQS_STALL[0] TCC_EA_WRREQ_IO_CREDIT_STALL[1] TCC_EA_WRREQ_GMI_CREDIT_STALL[1] TCC_EA_WRREQ_DRAM_CREDIT_STALL[1] TCC_TOO_MANY_EA_WRREQS_STALL[1] TCC_EA_WRREQ_IO_CREDIT_STALL[2] TCC_EA_WRREQ_GMI_CREDIT_STALL[2] TCC_EA_WRREQ_DRAM_CREDIT_STALL[2] TCC_TOO_MANY_EA_WRREQS_STALL[2] TCC_EA_WRREQ_IO_CREDIT_STALL[3] TCC_EA_WRREQ_GMI_CREDIT_STALL[3] TCC_EA_WRREQ_DRAM_CREDIT_STALL[3] TCC_TOO_MANY_EA_WRREQS_STALL[3] TCC_EA_WRREQ_IO_CREDIT_STALL[4] TCC_EA_WRREQ_GMI_CREDIT_STALL[4] TCC_EA_WRREQ_DRAM_CREDIT_STALL[4] TCC_TOO_MANY_EA_WRREQS_STALL[4] TCC_EA_WRREQ_IO_CREDIT_STALL[5] TCC_EA_WRREQ_GMI_CREDIT_STALL[5] TCC_EA_WRREQ_DRAM_CREDIT_STALL[5] TCC_TOO_MANY_EA_WRREQS_STALL[5] TCC_EA_WRREQ_IO_CREDIT_STALL[6] TCC_EA_WRREQ_GMI_CREDIT_STALL[6] TCC_EA_WRREQ_DRAM_CREDIT_STALL[6] TCC_TOO_MANY_EA_WRREQS_STALL[6] TCC_EA_WRREQ_IO_CREDIT_STALL[7] TCC_EA_WRREQ_GMI_CREDIT_STALL[7] TCC_EA_WRREQ_DRAM_CREDIT_STALL[7] TCC_TOO_MANY_EA_WRREQS_STALL[7] TCC_EA_WRREQ_IO_CREDIT_STALL[8] TCC_EA_WRREQ_GMI_CREDIT_STALL[8] TCC_EA_WRREQ_DRAM_CREDIT_STALL[8] TCC_TOO_MANY_EA_WRREQS_STALL[8] TCC_EA_WRREQ_IO_CREDIT_STALL[9] TCC_EA_WRREQ_GMI_CREDIT_STALL[9] TCC_EA_WRREQ_DRAM_CREDIT_STALL[9] TCC_TOO_MANY_EA_WRREQS_STALL[9] TCC_EA_WRREQ_IO_CREDIT_STALL[10] TCC_EA_WRREQ_GMI_CREDIT_STALL[10] TCC_EA_WRREQ_DRAM_CREDIT_STALL[10] TCC_TOO_MANY_EA_WRREQS_STALL[10] TCC_EA_WRREQ_IO_CREDIT_STALL[11] TCC_EA_WRREQ_GMI_CREDIT_STALL[11] TCC_EA_WRREQ_DRAM_CREDIT_STALL[11] TCC_TOO_MANY_EA_WRREQS_STALL[11] TCC_EA_WRREQ_IO_CREDIT_STALL[12] TCC_EA_WRREQ_GMI_CREDIT_STALL[12] TCC_EA_WRREQ_DRAM_CREDIT_STALL[12] TCC_TOO_MANY_EA_WRREQS_STALL[12] TCC_EA_WRREQ_IO_CREDIT_STALL[13] TCC_EA_WRREQ_GMI_CREDIT_STALL[13] TCC_EA_WRREQ_DRAM_CREDIT_STALL[13] TCC_TOO_MANY_EA_WRREQS_STALL[13] TCC_EA_WRREQ_IO_CREDIT_STALL[14] TCC_EA_WRREQ_GMI_CREDIT_STALL[14] TCC_EA_WRREQ_DRAM_CREDIT_STALL[14] TCC_TOO_MANY_EA_WRREQS_STALL[14] TCC_EA_WRREQ_IO_CREDIT_STALL[15] TCC_EA_WRREQ_GMI_CREDIT_STALL[15] TCC_EA_WRREQ_DRAM_CREDIT_STALL[15] TCC_TOO_MANY_EA_WRREQS_STALL[15] - -gpu: -kernel: -range: diff --git a/src/rocprof_compute_soc/profile_configs/gfx900/pmc_tcc_perf.txt b/src/rocprof_compute_soc/profile_configs/gfx900/pmc_tcc_perf.txt deleted file mode 100644 index e24afec96f..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx900/pmc_tcc_perf.txt +++ /dev/null @@ -1,18 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_CYCLES SQ_BUSY_CYCLES SQ_WAVES - -pmc: TCC_CYCLE_sum TCC_BUSY_sum TCC_PROBE_sum TCC_PROBE_ALL_sum -pmc: TCC_NC_REQ_sum TCC_UC_REQ_sum TCC_CC_REQ_sum -pmc: TCC_REQ_sum TCC_STREAMING_REQ_sum TCC_HIT_sum TCC_MISS_sum -pmc: TCC_READ_sum TCC_WRITE_sum TCC_ATOMIC_sum TCC_WRITEBACK_sum -pmc: TCC_EA_WRREQ_sum TCC_EA_WRREQ_64B_sum TCC_EA_WR_UNCACHED_32B_sum -pmc: TCC_EA_WRREQ_STALL_sum TCC_EA_WRREQ_IO_CREDIT_STALL_sum TCC_EA_WRREQ_GMI_CREDIT_STALL_sum TCC_EA_WRREQ_DRAM_CREDIT_STALL_sum -pmc: TCC_TOO_MANY_EA_WRREQS_STALL_sum TCC_EA_ATOMIC_sum TCC_EA_RDREQ_sum TCC_EA_RDREQ_32B_sum -pmc: TCC_EA_RD_UNCACHED_32B_sum TCC_EA_RDREQ_IO_CREDIT_STALL_sum TCC_EA_RDREQ_GMI_CREDIT_STALL_sum TCC_EA_RDREQ_DRAM_CREDIT_STALL_sum -pmc: TCC_NORMAL_WRITEBACK_sum TCC_ALL_TC_OP_WB_WRITEBACK_sum TCC_NORMAL_EVICT_sum -pmc: TCC_ALL_TC_OP_INV_EVICT_sum TCC_EA_RDREQ_DRAM_sum TCC_EA_WRREQ_DRAM_sum -pmc: TCC_EA_RDREQ_LEVEL_sum TCC_EA_WRREQ_LEVEL_sum TCC_EA_ATOMIC_LEVEL_sum - -gpu: -kernel: - -range: diff --git a/src/rocprof_compute_soc/profile_configs/gfx900/pmc_tcp_perf.txt b/src/rocprof_compute_soc/profile_configs/gfx900/pmc_tcp_perf.txt deleted file mode 100644 index 83ee6402fd..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx900/pmc_tcp_perf.txt +++ /dev/null @@ -1,18 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_CYCLES SQ_BUSY_CYCLES SQ_BUSY_CU_CYCLES SQ_WAVES SQ_WAVE_CYCLES - -pmc: TCP_GATE_EN1_sum TCP_GATE_EN2_sum TCP_TD_TCP_STALL_CYCLES_sum TCP_TCR_TCP_STALL_CYCLES_sum -pmc: TCP_READ_TAGCONFLICT_STALL_CYCLES_sum TCP_WRITE_TAGCONFLICT_STALL_CYCLES_sum TCP_ATOMIC_TAGCONFLICT_STALL_CYCLES_sum TCP_TA_TCP_STATE_READ_sum -pmc: TCP_VOLATILE_sum TCP_TOTAL_ACCESSES_sum TCP_TOTAL_READ_sum TCP_TOTAL_WRITE_sum -pmc: TCP_TOTAL_ATOMIC_WITH_RET_sum TCP_TOTAL_ATOMIC_WITHOUT_RET_sum TCP_TOTAL_WRITEBACK_INVALIDATES_sum TCP_TOTAL_CACHE_ACCESSES_sum -pmc: TCP_UTCL1_TRANSLATION_MISS_sum TCP_UTCL1_TRANSLATION_HIT_sum TCP_UTCL1_PERMISSION_MISS_sum TCP_UTCL1_REQUEST_sum -pmc: TCP_TCP_LATENCY_sum TCP_TCC_READ_REQ_LATENCY_sum TCP_TCC_WRITE_REQ_LATENCY_sum TCP_TCC_READ_REQ_sum -pmc: TCP_TCC_WRITE_REQ_sum TCP_TCC_ATOMIC_WITH_RET_REQ_sum TCP_TCC_ATOMIC_WITHOUT_RET_REQ_sum TCP_TCC_NC_READ_REQ_sum -pmc: TCP_TCC_NC_WRITE_REQ_sum TCP_TCC_NC_ATOMIC_REQ_sum TCP_TCC_UC_READ_REQ_sum TCP_TCC_UC_WRITE_REQ_sum -pmc: TCP_TCC_UC_ATOMIC_REQ_sum TCP_TCC_CC_READ_REQ_sum TCP_TCC_CC_WRITE_REQ_sum TCP_TCC_CC_ATOMIC_REQ_sum -pmc: TCP_PENDING_STALL_CYCLES_sum - -#pmc: TCA_CYCLE_sum TCA_BUSY_sum - -gpu: -kernel: -range: diff --git a/src/rocprof_compute_soc/profile_configs/gfx900/pmc_td_perf.txt b/src/rocprof_compute_soc/profile_configs/gfx900/pmc_td_perf.txt deleted file mode 100644 index 91d20dfce0..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx900/pmc_td_perf.txt +++ /dev/null @@ -1,11 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_CYCLES SQ_BUSY_CYCLES SQ_BUSY_CU_CYCLES SQ_WAVES SQ_WAVE_CYCLES - -pmc: TD_TD_BUSY_sum TD_TC_STALL_sum -pmc: TD_COALESCABLE_WAVEFRONT_sum TD_LOAD_WAVEFRONT_sum -pmc: TD_ATOMIC_WAVEFRONT_sum TD_STORE_WAVEFRONT_sum - - - -gpu: -range: -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx906/pmc_cpc_perf.txt b/src/rocprof_compute_soc/profile_configs/gfx906/pmc_cpc_perf.txt deleted file mode 100644 index bd80137304..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx906/pmc_cpc_perf.txt +++ /dev/null @@ -1,13 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_CYCLES SQ_BUSY_CYCLES SQ_WAVES - -pmc: CPC_CPC_STAT_BUSY CPC_CPC_STAT_IDLE -pmc: CPC_CPC_TCIU_BUSY CPC_CPC_TCIU_IDLE -pmc: CPC_CPC_STAT_STALL CPC_UTCL1_STALL_ON_TRANSLATION -pmc: CPC_CPC_UTCL2IU_BUSY CPC_CPC_UTCL2IU_IDLE -pmc: CPC_CPC_UTCL2IU_STALL CPC_ME1_BUSY_FOR_PACKET_DECODE -pmc: CPC_ME1_DC0_SPI_BUSY - -range: -gpu: - -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx906/pmc_cpf_perf.txt b/src/rocprof_compute_soc/profile_configs/gfx906/pmc_cpf_perf.txt deleted file mode 100644 index e8119fff56..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx906/pmc_cpf_perf.txt +++ /dev/null @@ -1,11 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_CYCLES SQ_BUSY_CYCLES SQ_WAVES - -pmc: CPF_CPF_STAT_BUSY CPF_CPF_STAT_STALL -pmc: CPF_CPF_TCIU_BUSY CPF_CPF_TCIU_STALL -pmc: CPF_CPF_STAT_IDLE CPF_CPF_TCIU_IDLE -pmc: CPF_CMP_UTCL1_STALL_ON_TRANSLATION - -range: -gpu: - -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx906/pmc_spi_perf.txt b/src/rocprof_compute_soc/profile_configs/gfx906/pmc_spi_perf.txt deleted file mode 100644 index 4eb0d22656..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx906/pmc_spi_perf.txt +++ /dev/null @@ -1,10 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_CYCLES SQ_BUSY_CYCLES SQ_WAVES GRBM_SPI_BUSY - -pmc: SPI_CSN_WINDOW_VALID SPI_CSN_BUSY SPI_CSN_NUM_THREADGROUPS SPI_CSN_WAVE SPI_RA_REQ_NO_ALLOC SPI_RA_REQ_NO_ALLOC_CSN -pmc: SPI_RA_RES_STALL_CSN SPI_RA_TMP_STALL_CSN SPI_RA_WAVE_SIMD_FULL_CSN SPI_RA_VGPR_SIMD_FULL_CSN SPI_RA_SGPR_SIMD_FULL_CSN SPI_RA_LDS_CU_FULL_CSN -pmc: SPI_RA_BAR_CU_FULL_CSN SPI_RA_TGLIM_CU_FULL_CSN SPI_RA_WVLIM_STALL_CSN SPI_SWC_CSC_WR SPI_VWC_CSC_WR SPI_RA_BULKY_CU_FULL_CSN - -range: -gpu: - -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx906/pmc_sq_perf1.txt b/src/rocprof_compute_soc/profile_configs/gfx906/pmc_sq_perf1.txt deleted file mode 100644 index 72afa13fb4..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx906/pmc_sq_perf1.txt +++ /dev/null @@ -1,27 +0,0 @@ -#SQ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_INSTS_VMEM_WR SQ_INSTS_VMEM_RD SQ_INSTS_VMEM SQ_INSTS_SALU SQ_INSTS_VSKIPPED -pmc: SQ_INSTS_SMEM SQ_INSTS_FLAT SQ_INSTS_LDS SQ_INSTS_GDS SQ_INSTS_EXP_GDS SQ_INSTS_BRANCH SQ_INSTS_SENDMSG SQ_INSTS -pmc: SQ_WAVE_CYCLES SQ_WAIT_ANY SQ_WAIT_INST_ANY SQ_ACTIVE_INST_ANY SQ_CYCLES SQ_BUSY_CYCLES SQ_BUSY_CU_CYCLES SQ_INSTS_VALU -pmc: SQ_ACTIVE_INST_VMEM SQ_ACTIVE_INST_LDS SQ_ACTIVE_INST_VALU SQ_ACTIVE_INST_SCA SQ_ACTIVE_INST_EXP_GDS SQ_ACTIVE_INST_MISC SQ_ACTIVE_INST_FLAT SQ_INST_CYCLES_VMEM_WR -pmc: SQ_INST_CYCLES_VMEM_RD SQ_INST_CYCLES_SMEM SQ_INST_CYCLES_SALU SQ_THREAD_CYCLES_VALU SQ_IFETCH SQ_LDS_BANK_CONFLICT SQ_LDS_ADDR_CONFLICT SQ_LDS_UNALIGNED_STALL -pmc: SQ_WAVES SQ_WAVES_EQ_64 SQ_WAVES_LT_64 SQ_WAVES_LT_48 SQ_WAVES_LT_32 SQ_WAVES_LT_16 SQ_ITEMS SQ_INSTS_VSKIPPED -pmc: SQ_LDS_MEM_VIOLATIONS SQ_LDS_ATOMIC_RETURN SQ_LDS_IDX_ACTIVE SQ_WAVES_RESTORED SQ_WAVES_SAVED SQ_INSTS_SMEM_NORM - - -#SQ:MI200 -#pmc: SQ_INSTS_MFMA SQ_INSTS_VALU_MFMA_I8 SQ_INSTS_VALU_MFMA_F16 SQ_INSTS_VALU_MFMA_BF16 SQ_INSTS_VALU_MFMA_F32 SQ_INSTS_VALU_MFMA_F64 SQ_VALU_MFMA_BUSY_CYCLES -#pmc: SQ_INSTS_FLAT_LDS_ONLY SQ_INSTS_VALU_MFMA_MOPS_I8 SQ_INSTS_VALU_MFMA_MOPS_F16 SQ_INSTS_VALU_MFMA_MOPS_BF16 SQ_INSTS_VALU_MFMA_MOPS_F32 SQ_INSTS_VALU_MFMA_MOPS_F64 - -#SQC -pmc: SQC_TC_INST_REQ SQC_TC_DATA_READ_REQ SQC_TC_DATA_WRITE_REQ SQC_TC_DATA_ATOMIC_REQ SQC_TC_STALL SQC_TC_REQ SQC_DCACHE_REQ_READ_16 -pmc: SQC_ICACHE_REQ SQC_ICACHE_HITS SQC_ICACHE_MISSES SQC_ICACHE_MISSES_DUPLICATE SQC_DCACHE_INPUT_VALID_READYB SQC_DCACHE_ATOMIC SQC_DCACHE_REQ_READ_8 -pmc: SQC_DCACHE_REQ SQC_DCACHE_HITS SQC_DCACHE_MISSES SQC_DCACHE_MISSES_DUPLICATE SQC_DCACHE_REQ_READ_1 SQC_DCACHE_REQ_READ_2 SQC_DCACHE_REQ_READ_4 - - - -######################################## -# Filtering -######################################## -range: -gpu: -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx906/pmc_sq_perf2.txt b/src/rocprof_compute_soc/profile_configs/gfx906/pmc_sq_perf2.txt deleted file mode 100644 index 7175c0326c..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx906/pmc_sq_perf2.txt +++ /dev/null @@ -1,10 +0,0 @@ -################################################# -# VMEM latency -################################################# -pmc: SQ_INSTS_VMEM SQ_INST_LEVEL_VMEM SQ_ACCUM_PREV_HIRES - - -range: -gpu: - -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx906/pmc_sq_perf3.txt b/src/rocprof_compute_soc/profile_configs/gfx906/pmc_sq_perf3.txt deleted file mode 100644 index 8d36311af0..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx906/pmc_sq_perf3.txt +++ /dev/null @@ -1,11 +0,0 @@ -################################################# -# SMEM latency -################################################# -pmc: SQ_INSTS_SMEM SQ_INST_LEVEL_SMEM SQ_ACCUM_PREV_HIRES - - -range: -gpu: - - -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx906/pmc_sq_perf4.txt b/src/rocprof_compute_soc/profile_configs/gfx906/pmc_sq_perf4.txt deleted file mode 100644 index f8f29b4f54..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx906/pmc_sq_perf4.txt +++ /dev/null @@ -1,8 +0,0 @@ -################################################# -# ifetch latency -################################################# -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_WAVES SQ_IFETCH SQ_IFETCH_LEVEL SQ_ACCUM_PREV_HIRES - -range: -gpu: -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx906/pmc_sq_perf6.txt b/src/rocprof_compute_soc/profile_configs/gfx906/pmc_sq_perf6.txt deleted file mode 100644 index f9144576eb..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx906/pmc_sq_perf6.txt +++ /dev/null @@ -1,9 +0,0 @@ -################################################# -# LDS latency -################################################# -pmc: SQ_INSTS_LDS SQ_INST_LEVEL_LDS SQ_ACCUM_PREV_HIRES - -range: -gpu: - -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx906/pmc_sq_perf8.txt b/src/rocprof_compute_soc/profile_configs/gfx906/pmc_sq_perf8.txt deleted file mode 100644 index cb29ad2ce1..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx906/pmc_sq_perf8.txt +++ /dev/null @@ -1,6 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE CPC_ME1_BUSY_FOR_PACKET_DECODE SQ_CYCLES SQ_WAVES SQ_WAVE_CYCLES SQ_BUSY_CYCLES SQ_LEVEL_WAVES SQ_ACCUM_PREV_HIRES - -gpu: -range: - -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx906/pmc_sqc_perf1.txt b/src/rocprof_compute_soc/profile_configs/gfx906/pmc_sqc_perf1.txt deleted file mode 100644 index e602d35c1f..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx906/pmc_sqc_perf1.txt +++ /dev/null @@ -1,11 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_CYCLES SQ_BUSY_CYCLES SQ_BUSY_CU_CYCLES SQ_WAVES SQ_WAVE_CYCLES - - -pmc: SQC_TC_INST_REQ SQC_TC_DATA_READ_REQ SQC_TC_DATA_WRITE_REQ SQC_TC_DATA_ATOMIC_REQ SQC_TC_STALL SQC_TC_REQ SQC_DCACHE_REQ_READ_16 -pmc: SQC_ICACHE_REQ SQC_ICACHE_HITS SQC_ICACHE_MISSES SQC_ICACHE_MISSES_DUPLICATE SQC_DCACHE_INPUT_VALID_READYB SQC_DCACHE_ATOMIC SQC_DCACHE_REQ_READ_8 -pmc: SQC_DCACHE_REQ SQC_DCACHE_HITS SQC_DCACHE_MISSES SQC_DCACHE_MISSES_DUPLICATE SQC_DCACHE_REQ_READ_1 SQC_DCACHE_REQ_READ_2 SQC_DCACHE_REQ_READ_4 - -range: -gpu: - -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx906/pmc_ta_perf.txt b/src/rocprof_compute_soc/profile_configs/gfx906/pmc_ta_perf.txt deleted file mode 100644 index be544c627b..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx906/pmc_ta_perf.txt +++ /dev/null @@ -1,25 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_CYCLES SQ_BUSY_CYCLES SQ_BUSY_CU_CYCLES SQ_WAVES SQ_WAVE_CYCLES - -pmc: TA_TA_BUSY_sum - -# buffer access -pmc: TA_BUFFER_WAVEFRONTS_sum TA_BUFFER_READ_WAVEFRONTS_sum -pmc: TA_BUFFER_WRITE_WAVEFRONTS_sum TA_BUFFER_ATOMIC_WAVEFRONTS_sum -pmc: TA_BUFFER_TOTAL_CYCLES_sum -pmc: TA_BUFFER_COALESCED_READ_CYCLES_sum TA_BUFFER_COALESCED_WRITE_CYCLES_sum - - -# stalls -pmc: TA_ADDR_STALLED_BY_TC_CYCLES_sum TA_TOTAL_WAVEFRONTS_sum -pmc: TA_ADDR_STALLED_BY_TD_CYCLES_sum TA_DATA_STALLED_BY_TC_CYCLES_sum - -# flat accesses -pmc: TA_FLAT_WAVEFRONTS_sum TA_FLAT_READ_WAVEFRONTS_sum -pmc: TA_FLAT_WRITE_WAVEFRONTS_sum TA_FLAT_ATOMIC_WAVEFRONTS_sum - - -range: - -gpu: - -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx906/pmc_tcc2_perf.txt b/src/rocprof_compute_soc/profile_configs/gfx906/pmc_tcc2_perf.txt deleted file mode 100644 index 1130c72a9c..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx906/pmc_tcc2_perf.txt +++ /dev/null @@ -1,25 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_CYCLES SQ_BUSY_CYCLES SQ_WAVES - -# MI50: no TCC_RW_REQ -pmc: TCC_CYCLE[0] TCC_HIT[0] TCC_MISS[0] TCC_CYCLE[1] TCC_HIT[1] TCC_MISS[1] TCC_CYCLE[2] TCC_HIT[2] TCC_MISS[2] TCC_CYCLE[3] TCC_HIT[3] TCC_MISS[3] TCC_CYCLE[4] TCC_HIT[4] TCC_MISS[4] TCC_CYCLE[5] TCC_HIT[5] TCC_MISS[5] TCC_CYCLE[6] TCC_HIT[6] TCC_MISS[6] TCC_CYCLE[7] TCC_HIT[7] TCC_MISS[7] TCC_CYCLE[8] TCC_HIT[8] TCC_MISS[8] TCC_CYCLE[9] TCC_HIT[9] TCC_MISS[9] TCC_CYCLE[10] TCC_HIT[10] TCC_MISS[10] TCC_CYCLE[11] TCC_HIT[11] TCC_MISS[11] TCC_CYCLE[12] TCC_HIT[12] TCC_MISS[12] TCC_CYCLE[13] TCC_HIT[13] TCC_MISS[13] TCC_CYCLE[14] TCC_HIT[14] TCC_MISS[14] TCC_CYCLE[15] TCC_HIT[15] TCC_MISS[15] - - - -pmc: TCC_REQ[0] TCC_READ[0] TCC_WRITE[0] TCC_ATOMIC[0] TCC_REQ[1] TCC_READ[1] TCC_WRITE[1] TCC_ATOMIC[1] TCC_REQ[2] TCC_READ[2] TCC_WRITE[2] TCC_ATOMIC[2] TCC_REQ[3] TCC_READ[3] TCC_WRITE[3] TCC_ATOMIC[3] TCC_REQ[4] TCC_READ[4] TCC_WRITE[4] TCC_ATOMIC[4] TCC_REQ[5] TCC_READ[5] TCC_WRITE[5] TCC_ATOMIC[5] TCC_REQ[6] TCC_READ[6] TCC_WRITE[6] TCC_ATOMIC[6] TCC_REQ[7] TCC_READ[7] TCC_WRITE[7] TCC_ATOMIC[7] TCC_REQ[8] TCC_READ[8] TCC_WRITE[8] TCC_ATOMIC[8] TCC_REQ[9] TCC_READ[9] TCC_WRITE[9] TCC_ATOMIC[9] TCC_REQ[10] TCC_READ[10] TCC_WRITE[10] TCC_ATOMIC[10] TCC_REQ[11] TCC_READ[11] TCC_WRITE[11] TCC_ATOMIC[11] TCC_REQ[12] TCC_READ[12] TCC_WRITE[12] TCC_ATOMIC[12] TCC_REQ[13] TCC_READ[13] TCC_WRITE[13] TCC_ATOMIC[13] TCC_REQ[14] TCC_READ[14] TCC_WRITE[14] TCC_ATOMIC[14] TCC_REQ[15] TCC_READ[15] TCC_WRITE[15] TCC_ATOMIC[15] - - - -pmc: TCC_EA_RDREQ[0] TCC_EA_RDREQ_32B[0] TCC_EA_WRREQ[0] TCC_EA_WRREQ_64B[0] TCC_EA_RDREQ[1] TCC_EA_RDREQ_32B[1] TCC_EA_WRREQ[1] TCC_EA_WRREQ_64B[1] TCC_EA_RDREQ[2] TCC_EA_RDREQ_32B[2] TCC_EA_WRREQ[2] TCC_EA_WRREQ_64B[2] TCC_EA_RDREQ[3] TCC_EA_RDREQ_32B[3] TCC_EA_WRREQ[3] TCC_EA_WRREQ_64B[3] TCC_EA_RDREQ[4] TCC_EA_RDREQ_32B[4] TCC_EA_WRREQ[4] TCC_EA_WRREQ_64B[4] TCC_EA_RDREQ[5] TCC_EA_RDREQ_32B[5] TCC_EA_WRREQ[5] TCC_EA_WRREQ_64B[5] TCC_EA_RDREQ[6] TCC_EA_RDREQ_32B[6] TCC_EA_WRREQ[6] TCC_EA_WRREQ_64B[6] TCC_EA_RDREQ[7] TCC_EA_RDREQ_32B[7] TCC_EA_WRREQ[7] TCC_EA_WRREQ_64B[7] TCC_EA_RDREQ[8] TCC_EA_RDREQ_32B[8] TCC_EA_WRREQ[8] TCC_EA_WRREQ_64B[8] TCC_EA_RDREQ[9] TCC_EA_RDREQ_32B[9] TCC_EA_WRREQ[9] TCC_EA_WRREQ_64B[9] TCC_EA_RDREQ[10] TCC_EA_RDREQ_32B[10] TCC_EA_WRREQ[10] TCC_EA_WRREQ_64B[10] TCC_EA_RDREQ[11] TCC_EA_RDREQ_32B[11] TCC_EA_WRREQ[11] TCC_EA_WRREQ_64B[11] TCC_EA_RDREQ[12] TCC_EA_RDREQ_32B[12] TCC_EA_WRREQ[12] TCC_EA_WRREQ_64B[12] TCC_EA_RDREQ[13] TCC_EA_RDREQ_32B[13] TCC_EA_WRREQ[13] TCC_EA_WRREQ_64B[13] TCC_EA_RDREQ[14] TCC_EA_RDREQ_32B[14] TCC_EA_WRREQ[14] TCC_EA_WRREQ_64B[14] TCC_EA_RDREQ[15] TCC_EA_RDREQ_32B[15] TCC_EA_WRREQ[15] TCC_EA_WRREQ_64B[15] - - -pmc: TCC_EA_ATOMIC[0] TCC_EA_RDREQ_LEVEL[0] TCC_EA_WRREQ_LEVEL[0] TCC_EA_ATOMIC_LEVEL[0] TCC_EA_ATOMIC[1] TCC_EA_RDREQ_LEVEL[1] TCC_EA_WRREQ_LEVEL[1] TCC_EA_ATOMIC_LEVEL[1] TCC_EA_ATOMIC[2] TCC_EA_RDREQ_LEVEL[2] TCC_EA_WRREQ_LEVEL[2] TCC_EA_ATOMIC_LEVEL[2] TCC_EA_ATOMIC[3] TCC_EA_RDREQ_LEVEL[3] TCC_EA_WRREQ_LEVEL[3] TCC_EA_ATOMIC_LEVEL[3] TCC_EA_ATOMIC[4] TCC_EA_RDREQ_LEVEL[4] TCC_EA_WRREQ_LEVEL[4] TCC_EA_ATOMIC_LEVEL[4] TCC_EA_ATOMIC[5] TCC_EA_RDREQ_LEVEL[5] TCC_EA_WRREQ_LEVEL[5] TCC_EA_ATOMIC_LEVEL[5] TCC_EA_ATOMIC[6] TCC_EA_RDREQ_LEVEL[6] TCC_EA_WRREQ_LEVEL[6] TCC_EA_ATOMIC_LEVEL[6] TCC_EA_ATOMIC[7] TCC_EA_RDREQ_LEVEL[7] TCC_EA_WRREQ_LEVEL[7] TCC_EA_ATOMIC_LEVEL[7] TCC_EA_ATOMIC[8] TCC_EA_RDREQ_LEVEL[8] TCC_EA_WRREQ_LEVEL[8] TCC_EA_ATOMIC_LEVEL[8] TCC_EA_ATOMIC[9] TCC_EA_RDREQ_LEVEL[9] TCC_EA_WRREQ_LEVEL[9] TCC_EA_ATOMIC_LEVEL[9] TCC_EA_ATOMIC[10] TCC_EA_RDREQ_LEVEL[10] TCC_EA_WRREQ_LEVEL[10] TCC_EA_ATOMIC_LEVEL[10] TCC_EA_ATOMIC[11] TCC_EA_RDREQ_LEVEL[11] TCC_EA_WRREQ_LEVEL[11] TCC_EA_ATOMIC_LEVEL[11] TCC_EA_ATOMIC[12] TCC_EA_RDREQ_LEVEL[12] TCC_EA_WRREQ_LEVEL[12] TCC_EA_ATOMIC_LEVEL[12] TCC_EA_ATOMIC[13] TCC_EA_RDREQ_LEVEL[13] TCC_EA_WRREQ_LEVEL[13] TCC_EA_ATOMIC_LEVEL[13] TCC_EA_ATOMIC[14] TCC_EA_RDREQ_LEVEL[14] TCC_EA_WRREQ_LEVEL[14] TCC_EA_ATOMIC_LEVEL[14] TCC_EA_ATOMIC[15] TCC_EA_RDREQ_LEVEL[15] TCC_EA_WRREQ_LEVEL[15] TCC_EA_ATOMIC_LEVEL[15] - - -pmc: TCC_EA_RDREQ_IO_CREDIT_STALL[0] TCC_EA_RDREQ_GMI_CREDIT_STALL[0] TCC_EA_RDREQ_DRAM_CREDIT_STALL[0] TCC_EA_RDREQ_IO_CREDIT_STALL[1] TCC_EA_RDREQ_GMI_CREDIT_STALL[1] TCC_EA_RDREQ_DRAM_CREDIT_STALL[1] TCC_EA_RDREQ_IO_CREDIT_STALL[2] TCC_EA_RDREQ_GMI_CREDIT_STALL[2] TCC_EA_RDREQ_DRAM_CREDIT_STALL[2] TCC_EA_RDREQ_IO_CREDIT_STALL[3] TCC_EA_RDREQ_GMI_CREDIT_STALL[3] TCC_EA_RDREQ_DRAM_CREDIT_STALL[3] TCC_EA_RDREQ_IO_CREDIT_STALL[4] TCC_EA_RDREQ_GMI_CREDIT_STALL[4] TCC_EA_RDREQ_DRAM_CREDIT_STALL[4] TCC_EA_RDREQ_IO_CREDIT_STALL[5] TCC_EA_RDREQ_GMI_CREDIT_STALL[5] TCC_EA_RDREQ_DRAM_CREDIT_STALL[5] TCC_EA_RDREQ_IO_CREDIT_STALL[6] TCC_EA_RDREQ_GMI_CREDIT_STALL[6] TCC_EA_RDREQ_DRAM_CREDIT_STALL[6] TCC_EA_RDREQ_IO_CREDIT_STALL[7] TCC_EA_RDREQ_GMI_CREDIT_STALL[7] TCC_EA_RDREQ_DRAM_CREDIT_STALL[7] TCC_EA_RDREQ_IO_CREDIT_STALL[8] TCC_EA_RDREQ_GMI_CREDIT_STALL[8] TCC_EA_RDREQ_DRAM_CREDIT_STALL[8] TCC_EA_RDREQ_IO_CREDIT_STALL[9] TCC_EA_RDREQ_GMI_CREDIT_STALL[9] TCC_EA_RDREQ_DRAM_CREDIT_STALL[9] TCC_EA_RDREQ_IO_CREDIT_STALL[10] TCC_EA_RDREQ_GMI_CREDIT_STALL[10] TCC_EA_RDREQ_DRAM_CREDIT_STALL[10] TCC_EA_RDREQ_IO_CREDIT_STALL[11] TCC_EA_RDREQ_GMI_CREDIT_STALL[11] TCC_EA_RDREQ_DRAM_CREDIT_STALL[11] TCC_EA_RDREQ_IO_CREDIT_STALL[12] TCC_EA_RDREQ_GMI_CREDIT_STALL[12] TCC_EA_RDREQ_DRAM_CREDIT_STALL[12] TCC_EA_RDREQ_IO_CREDIT_STALL[13] TCC_EA_RDREQ_GMI_CREDIT_STALL[13] TCC_EA_RDREQ_DRAM_CREDIT_STALL[13] TCC_EA_RDREQ_IO_CREDIT_STALL[14] TCC_EA_RDREQ_GMI_CREDIT_STALL[14] TCC_EA_RDREQ_DRAM_CREDIT_STALL[14] TCC_EA_RDREQ_IO_CREDIT_STALL[15] TCC_EA_RDREQ_GMI_CREDIT_STALL[15] TCC_EA_RDREQ_DRAM_CREDIT_STALL[15] - - -pmc: TCC_EA_WRREQ_IO_CREDIT_STALL[0] TCC_EA_WRREQ_GMI_CREDIT_STALL[0] TCC_EA_WRREQ_DRAM_CREDIT_STALL[0] TCC_TOO_MANY_EA_WRREQS_STALL[0] TCC_EA_WRREQ_IO_CREDIT_STALL[1] TCC_EA_WRREQ_GMI_CREDIT_STALL[1] TCC_EA_WRREQ_DRAM_CREDIT_STALL[1] TCC_TOO_MANY_EA_WRREQS_STALL[1] TCC_EA_WRREQ_IO_CREDIT_STALL[2] TCC_EA_WRREQ_GMI_CREDIT_STALL[2] TCC_EA_WRREQ_DRAM_CREDIT_STALL[2] TCC_TOO_MANY_EA_WRREQS_STALL[2] TCC_EA_WRREQ_IO_CREDIT_STALL[3] TCC_EA_WRREQ_GMI_CREDIT_STALL[3] TCC_EA_WRREQ_DRAM_CREDIT_STALL[3] TCC_TOO_MANY_EA_WRREQS_STALL[3] TCC_EA_WRREQ_IO_CREDIT_STALL[4] TCC_EA_WRREQ_GMI_CREDIT_STALL[4] TCC_EA_WRREQ_DRAM_CREDIT_STALL[4] TCC_TOO_MANY_EA_WRREQS_STALL[4] TCC_EA_WRREQ_IO_CREDIT_STALL[5] TCC_EA_WRREQ_GMI_CREDIT_STALL[5] TCC_EA_WRREQ_DRAM_CREDIT_STALL[5] TCC_TOO_MANY_EA_WRREQS_STALL[5] TCC_EA_WRREQ_IO_CREDIT_STALL[6] TCC_EA_WRREQ_GMI_CREDIT_STALL[6] TCC_EA_WRREQ_DRAM_CREDIT_STALL[6] TCC_TOO_MANY_EA_WRREQS_STALL[6] TCC_EA_WRREQ_IO_CREDIT_STALL[7] TCC_EA_WRREQ_GMI_CREDIT_STALL[7] TCC_EA_WRREQ_DRAM_CREDIT_STALL[7] TCC_TOO_MANY_EA_WRREQS_STALL[7] TCC_EA_WRREQ_IO_CREDIT_STALL[8] TCC_EA_WRREQ_GMI_CREDIT_STALL[8] TCC_EA_WRREQ_DRAM_CREDIT_STALL[8] TCC_TOO_MANY_EA_WRREQS_STALL[8] TCC_EA_WRREQ_IO_CREDIT_STALL[9] TCC_EA_WRREQ_GMI_CREDIT_STALL[9] TCC_EA_WRREQ_DRAM_CREDIT_STALL[9] TCC_TOO_MANY_EA_WRREQS_STALL[9] TCC_EA_WRREQ_IO_CREDIT_STALL[10] TCC_EA_WRREQ_GMI_CREDIT_STALL[10] TCC_EA_WRREQ_DRAM_CREDIT_STALL[10] TCC_TOO_MANY_EA_WRREQS_STALL[10] TCC_EA_WRREQ_IO_CREDIT_STALL[11] TCC_EA_WRREQ_GMI_CREDIT_STALL[11] TCC_EA_WRREQ_DRAM_CREDIT_STALL[11] TCC_TOO_MANY_EA_WRREQS_STALL[11] TCC_EA_WRREQ_IO_CREDIT_STALL[12] TCC_EA_WRREQ_GMI_CREDIT_STALL[12] TCC_EA_WRREQ_DRAM_CREDIT_STALL[12] TCC_TOO_MANY_EA_WRREQS_STALL[12] TCC_EA_WRREQ_IO_CREDIT_STALL[13] TCC_EA_WRREQ_GMI_CREDIT_STALL[13] TCC_EA_WRREQ_DRAM_CREDIT_STALL[13] TCC_TOO_MANY_EA_WRREQS_STALL[13] TCC_EA_WRREQ_IO_CREDIT_STALL[14] TCC_EA_WRREQ_GMI_CREDIT_STALL[14] TCC_EA_WRREQ_DRAM_CREDIT_STALL[14] TCC_TOO_MANY_EA_WRREQS_STALL[14] TCC_EA_WRREQ_IO_CREDIT_STALL[15] TCC_EA_WRREQ_GMI_CREDIT_STALL[15] TCC_EA_WRREQ_DRAM_CREDIT_STALL[15] TCC_TOO_MANY_EA_WRREQS_STALL[15] - -gpu: -kernel: -range: diff --git a/src/rocprof_compute_soc/profile_configs/gfx906/pmc_tcc_perf.txt b/src/rocprof_compute_soc/profile_configs/gfx906/pmc_tcc_perf.txt deleted file mode 100644 index ff5b85ea40..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx906/pmc_tcc_perf.txt +++ /dev/null @@ -1,18 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_CYCLES SQ_BUSY_CYCLES SQ_WAVES - -pmc: TCC_CYCLE_sum TCC_BUSY_sum TCC_PROBE_sum TCC_PROBE_ALL_sum -pmc: TCC_NC_REQ_sum TCC_UC_REQ_sum TCC_CC_REQ_sum -pmc: TCC_REQ_sum TCC_STREAMING_REQ_sum TCC_HIT_sum TCC_MISS_sum -pmc: TCC_READ_sum TCC_WRITE_sum TCC_ATOMIC_sum TCC_WRITEBACK_sum -pmc: TCC_EA_WRREQ_sum TCC_EA_WRREQ_64B_sum TCC_EA_WR_UNCACHED_32B_sum TCC_EA_WRREQ_DRAM_sum -pmc: TCC_EA_WRREQ_STALL_sum TCC_EA_WRREQ_IO_CREDIT_STALL_sum TCC_EA_WRREQ_GMI_CREDIT_STALL_sum TCC_EA_WRREQ_DRAM_CREDIT_STALL_sum -pmc: TCC_EA_RDREQ_sum TCC_EA_RDREQ_32B_sum TCC_EA_RD_UNCACHED_32B_sum TCC_EA_RDREQ_DRAM_sum -pmc: TCC_EA_RDREQ_IO_CREDIT_STALL_sum TCC_EA_RDREQ_GMI_CREDIT_STALL_sum TCC_EA_RDREQ_DRAM_CREDIT_STALL_sum -pmc: TCC_NORMAL_WRITEBACK_sum TCC_ALL_TC_OP_WB_WRITEBACK_sum TCC_NORMAL_EVICT_sum -pmc: TCC_ALL_TC_OP_INV_EVICT_sum TCC_TOO_MANY_EA_WRREQS_STALL_sum TCC_EA_ATOMIC_sum -pmc: TCC_EA_RDREQ_LEVEL_sum TCC_EA_WRREQ_LEVEL_sum TCC_EA_ATOMIC_LEVEL_sum - -gpu: -kernel: - -range: diff --git a/src/rocprof_compute_soc/profile_configs/gfx906/pmc_tcp_perf.txt b/src/rocprof_compute_soc/profile_configs/gfx906/pmc_tcp_perf.txt deleted file mode 100644 index 83ee6402fd..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx906/pmc_tcp_perf.txt +++ /dev/null @@ -1,18 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_CYCLES SQ_BUSY_CYCLES SQ_BUSY_CU_CYCLES SQ_WAVES SQ_WAVE_CYCLES - -pmc: TCP_GATE_EN1_sum TCP_GATE_EN2_sum TCP_TD_TCP_STALL_CYCLES_sum TCP_TCR_TCP_STALL_CYCLES_sum -pmc: TCP_READ_TAGCONFLICT_STALL_CYCLES_sum TCP_WRITE_TAGCONFLICT_STALL_CYCLES_sum TCP_ATOMIC_TAGCONFLICT_STALL_CYCLES_sum TCP_TA_TCP_STATE_READ_sum -pmc: TCP_VOLATILE_sum TCP_TOTAL_ACCESSES_sum TCP_TOTAL_READ_sum TCP_TOTAL_WRITE_sum -pmc: TCP_TOTAL_ATOMIC_WITH_RET_sum TCP_TOTAL_ATOMIC_WITHOUT_RET_sum TCP_TOTAL_WRITEBACK_INVALIDATES_sum TCP_TOTAL_CACHE_ACCESSES_sum -pmc: TCP_UTCL1_TRANSLATION_MISS_sum TCP_UTCL1_TRANSLATION_HIT_sum TCP_UTCL1_PERMISSION_MISS_sum TCP_UTCL1_REQUEST_sum -pmc: TCP_TCP_LATENCY_sum TCP_TCC_READ_REQ_LATENCY_sum TCP_TCC_WRITE_REQ_LATENCY_sum TCP_TCC_READ_REQ_sum -pmc: TCP_TCC_WRITE_REQ_sum TCP_TCC_ATOMIC_WITH_RET_REQ_sum TCP_TCC_ATOMIC_WITHOUT_RET_REQ_sum TCP_TCC_NC_READ_REQ_sum -pmc: TCP_TCC_NC_WRITE_REQ_sum TCP_TCC_NC_ATOMIC_REQ_sum TCP_TCC_UC_READ_REQ_sum TCP_TCC_UC_WRITE_REQ_sum -pmc: TCP_TCC_UC_ATOMIC_REQ_sum TCP_TCC_CC_READ_REQ_sum TCP_TCC_CC_WRITE_REQ_sum TCP_TCC_CC_ATOMIC_REQ_sum -pmc: TCP_PENDING_STALL_CYCLES_sum - -#pmc: TCA_CYCLE_sum TCA_BUSY_sum - -gpu: -kernel: -range: diff --git a/src/rocprof_compute_soc/profile_configs/gfx906/pmc_td_perf.txt b/src/rocprof_compute_soc/profile_configs/gfx906/pmc_td_perf.txt deleted file mode 100644 index 91d20dfce0..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx906/pmc_td_perf.txt +++ /dev/null @@ -1,11 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_CYCLES SQ_BUSY_CYCLES SQ_BUSY_CU_CYCLES SQ_WAVES SQ_WAVE_CYCLES - -pmc: TD_TD_BUSY_sum TD_TC_STALL_sum -pmc: TD_COALESCABLE_WAVEFRONT_sum TD_LOAD_WAVEFRONT_sum -pmc: TD_ATOMIC_WAVEFRONT_sum TD_STORE_WAVEFRONT_sum - - - -gpu: -range: -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx908/pmc_cpc_perf.txt b/src/rocprof_compute_soc/profile_configs/gfx908/pmc_cpc_perf.txt deleted file mode 100644 index bd80137304..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx908/pmc_cpc_perf.txt +++ /dev/null @@ -1,13 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_CYCLES SQ_BUSY_CYCLES SQ_WAVES - -pmc: CPC_CPC_STAT_BUSY CPC_CPC_STAT_IDLE -pmc: CPC_CPC_TCIU_BUSY CPC_CPC_TCIU_IDLE -pmc: CPC_CPC_STAT_STALL CPC_UTCL1_STALL_ON_TRANSLATION -pmc: CPC_CPC_UTCL2IU_BUSY CPC_CPC_UTCL2IU_IDLE -pmc: CPC_CPC_UTCL2IU_STALL CPC_ME1_BUSY_FOR_PACKET_DECODE -pmc: CPC_ME1_DC0_SPI_BUSY - -range: -gpu: - -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx908/pmc_cpf_perf.txt b/src/rocprof_compute_soc/profile_configs/gfx908/pmc_cpf_perf.txt deleted file mode 100644 index e8119fff56..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx908/pmc_cpf_perf.txt +++ /dev/null @@ -1,11 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_CYCLES SQ_BUSY_CYCLES SQ_WAVES - -pmc: CPF_CPF_STAT_BUSY CPF_CPF_STAT_STALL -pmc: CPF_CPF_TCIU_BUSY CPF_CPF_TCIU_STALL -pmc: CPF_CPF_STAT_IDLE CPF_CPF_TCIU_IDLE -pmc: CPF_CMP_UTCL1_STALL_ON_TRANSLATION - -range: -gpu: - -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx908/pmc_spi_perf.txt b/src/rocprof_compute_soc/profile_configs/gfx908/pmc_spi_perf.txt deleted file mode 100644 index 4eb0d22656..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx908/pmc_spi_perf.txt +++ /dev/null @@ -1,10 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_CYCLES SQ_BUSY_CYCLES SQ_WAVES GRBM_SPI_BUSY - -pmc: SPI_CSN_WINDOW_VALID SPI_CSN_BUSY SPI_CSN_NUM_THREADGROUPS SPI_CSN_WAVE SPI_RA_REQ_NO_ALLOC SPI_RA_REQ_NO_ALLOC_CSN -pmc: SPI_RA_RES_STALL_CSN SPI_RA_TMP_STALL_CSN SPI_RA_WAVE_SIMD_FULL_CSN SPI_RA_VGPR_SIMD_FULL_CSN SPI_RA_SGPR_SIMD_FULL_CSN SPI_RA_LDS_CU_FULL_CSN -pmc: SPI_RA_BAR_CU_FULL_CSN SPI_RA_TGLIM_CU_FULL_CSN SPI_RA_WVLIM_STALL_CSN SPI_SWC_CSC_WR SPI_VWC_CSC_WR SPI_RA_BULKY_CU_FULL_CSN - -range: -gpu: - -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx908/pmc_sq_perf1.txt b/src/rocprof_compute_soc/profile_configs/gfx908/pmc_sq_perf1.txt deleted file mode 100644 index 72afa13fb4..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx908/pmc_sq_perf1.txt +++ /dev/null @@ -1,27 +0,0 @@ -#SQ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_INSTS_VMEM_WR SQ_INSTS_VMEM_RD SQ_INSTS_VMEM SQ_INSTS_SALU SQ_INSTS_VSKIPPED -pmc: SQ_INSTS_SMEM SQ_INSTS_FLAT SQ_INSTS_LDS SQ_INSTS_GDS SQ_INSTS_EXP_GDS SQ_INSTS_BRANCH SQ_INSTS_SENDMSG SQ_INSTS -pmc: SQ_WAVE_CYCLES SQ_WAIT_ANY SQ_WAIT_INST_ANY SQ_ACTIVE_INST_ANY SQ_CYCLES SQ_BUSY_CYCLES SQ_BUSY_CU_CYCLES SQ_INSTS_VALU -pmc: SQ_ACTIVE_INST_VMEM SQ_ACTIVE_INST_LDS SQ_ACTIVE_INST_VALU SQ_ACTIVE_INST_SCA SQ_ACTIVE_INST_EXP_GDS SQ_ACTIVE_INST_MISC SQ_ACTIVE_INST_FLAT SQ_INST_CYCLES_VMEM_WR -pmc: SQ_INST_CYCLES_VMEM_RD SQ_INST_CYCLES_SMEM SQ_INST_CYCLES_SALU SQ_THREAD_CYCLES_VALU SQ_IFETCH SQ_LDS_BANK_CONFLICT SQ_LDS_ADDR_CONFLICT SQ_LDS_UNALIGNED_STALL -pmc: SQ_WAVES SQ_WAVES_EQ_64 SQ_WAVES_LT_64 SQ_WAVES_LT_48 SQ_WAVES_LT_32 SQ_WAVES_LT_16 SQ_ITEMS SQ_INSTS_VSKIPPED -pmc: SQ_LDS_MEM_VIOLATIONS SQ_LDS_ATOMIC_RETURN SQ_LDS_IDX_ACTIVE SQ_WAVES_RESTORED SQ_WAVES_SAVED SQ_INSTS_SMEM_NORM - - -#SQ:MI200 -#pmc: SQ_INSTS_MFMA SQ_INSTS_VALU_MFMA_I8 SQ_INSTS_VALU_MFMA_F16 SQ_INSTS_VALU_MFMA_BF16 SQ_INSTS_VALU_MFMA_F32 SQ_INSTS_VALU_MFMA_F64 SQ_VALU_MFMA_BUSY_CYCLES -#pmc: SQ_INSTS_FLAT_LDS_ONLY SQ_INSTS_VALU_MFMA_MOPS_I8 SQ_INSTS_VALU_MFMA_MOPS_F16 SQ_INSTS_VALU_MFMA_MOPS_BF16 SQ_INSTS_VALU_MFMA_MOPS_F32 SQ_INSTS_VALU_MFMA_MOPS_F64 - -#SQC -pmc: SQC_TC_INST_REQ SQC_TC_DATA_READ_REQ SQC_TC_DATA_WRITE_REQ SQC_TC_DATA_ATOMIC_REQ SQC_TC_STALL SQC_TC_REQ SQC_DCACHE_REQ_READ_16 -pmc: SQC_ICACHE_REQ SQC_ICACHE_HITS SQC_ICACHE_MISSES SQC_ICACHE_MISSES_DUPLICATE SQC_DCACHE_INPUT_VALID_READYB SQC_DCACHE_ATOMIC SQC_DCACHE_REQ_READ_8 -pmc: SQC_DCACHE_REQ SQC_DCACHE_HITS SQC_DCACHE_MISSES SQC_DCACHE_MISSES_DUPLICATE SQC_DCACHE_REQ_READ_1 SQC_DCACHE_REQ_READ_2 SQC_DCACHE_REQ_READ_4 - - - -######################################## -# Filtering -######################################## -range: -gpu: -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx908/pmc_sq_perf2.txt b/src/rocprof_compute_soc/profile_configs/gfx908/pmc_sq_perf2.txt deleted file mode 100644 index 7175c0326c..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx908/pmc_sq_perf2.txt +++ /dev/null @@ -1,10 +0,0 @@ -################################################# -# VMEM latency -################################################# -pmc: SQ_INSTS_VMEM SQ_INST_LEVEL_VMEM SQ_ACCUM_PREV_HIRES - - -range: -gpu: - -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx908/pmc_sq_perf3.txt b/src/rocprof_compute_soc/profile_configs/gfx908/pmc_sq_perf3.txt deleted file mode 100644 index 8d36311af0..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx908/pmc_sq_perf3.txt +++ /dev/null @@ -1,11 +0,0 @@ -################################################# -# SMEM latency -################################################# -pmc: SQ_INSTS_SMEM SQ_INST_LEVEL_SMEM SQ_ACCUM_PREV_HIRES - - -range: -gpu: - - -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx908/pmc_sq_perf4.txt b/src/rocprof_compute_soc/profile_configs/gfx908/pmc_sq_perf4.txt deleted file mode 100644 index f8f29b4f54..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx908/pmc_sq_perf4.txt +++ /dev/null @@ -1,8 +0,0 @@ -################################################# -# ifetch latency -################################################# -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_WAVES SQ_IFETCH SQ_IFETCH_LEVEL SQ_ACCUM_PREV_HIRES - -range: -gpu: -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx908/pmc_sq_perf6.txt b/src/rocprof_compute_soc/profile_configs/gfx908/pmc_sq_perf6.txt deleted file mode 100644 index f9144576eb..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx908/pmc_sq_perf6.txt +++ /dev/null @@ -1,9 +0,0 @@ -################################################# -# LDS latency -################################################# -pmc: SQ_INSTS_LDS SQ_INST_LEVEL_LDS SQ_ACCUM_PREV_HIRES - -range: -gpu: - -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx908/pmc_sq_perf8.txt b/src/rocprof_compute_soc/profile_configs/gfx908/pmc_sq_perf8.txt deleted file mode 100644 index cb29ad2ce1..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx908/pmc_sq_perf8.txt +++ /dev/null @@ -1,6 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE CPC_ME1_BUSY_FOR_PACKET_DECODE SQ_CYCLES SQ_WAVES SQ_WAVE_CYCLES SQ_BUSY_CYCLES SQ_LEVEL_WAVES SQ_ACCUM_PREV_HIRES - -gpu: -range: - -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx908/pmc_sqc_perf1.txt b/src/rocprof_compute_soc/profile_configs/gfx908/pmc_sqc_perf1.txt deleted file mode 100644 index e602d35c1f..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx908/pmc_sqc_perf1.txt +++ /dev/null @@ -1,11 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_CYCLES SQ_BUSY_CYCLES SQ_BUSY_CU_CYCLES SQ_WAVES SQ_WAVE_CYCLES - - -pmc: SQC_TC_INST_REQ SQC_TC_DATA_READ_REQ SQC_TC_DATA_WRITE_REQ SQC_TC_DATA_ATOMIC_REQ SQC_TC_STALL SQC_TC_REQ SQC_DCACHE_REQ_READ_16 -pmc: SQC_ICACHE_REQ SQC_ICACHE_HITS SQC_ICACHE_MISSES SQC_ICACHE_MISSES_DUPLICATE SQC_DCACHE_INPUT_VALID_READYB SQC_DCACHE_ATOMIC SQC_DCACHE_REQ_READ_8 -pmc: SQC_DCACHE_REQ SQC_DCACHE_HITS SQC_DCACHE_MISSES SQC_DCACHE_MISSES_DUPLICATE SQC_DCACHE_REQ_READ_1 SQC_DCACHE_REQ_READ_2 SQC_DCACHE_REQ_READ_4 - -range: -gpu: - -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx908/pmc_ta_perf.txt b/src/rocprof_compute_soc/profile_configs/gfx908/pmc_ta_perf.txt deleted file mode 100644 index be544c627b..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx908/pmc_ta_perf.txt +++ /dev/null @@ -1,25 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_CYCLES SQ_BUSY_CYCLES SQ_BUSY_CU_CYCLES SQ_WAVES SQ_WAVE_CYCLES - -pmc: TA_TA_BUSY_sum - -# buffer access -pmc: TA_BUFFER_WAVEFRONTS_sum TA_BUFFER_READ_WAVEFRONTS_sum -pmc: TA_BUFFER_WRITE_WAVEFRONTS_sum TA_BUFFER_ATOMIC_WAVEFRONTS_sum -pmc: TA_BUFFER_TOTAL_CYCLES_sum -pmc: TA_BUFFER_COALESCED_READ_CYCLES_sum TA_BUFFER_COALESCED_WRITE_CYCLES_sum - - -# stalls -pmc: TA_ADDR_STALLED_BY_TC_CYCLES_sum TA_TOTAL_WAVEFRONTS_sum -pmc: TA_ADDR_STALLED_BY_TD_CYCLES_sum TA_DATA_STALLED_BY_TC_CYCLES_sum - -# flat accesses -pmc: TA_FLAT_WAVEFRONTS_sum TA_FLAT_READ_WAVEFRONTS_sum -pmc: TA_FLAT_WRITE_WAVEFRONTS_sum TA_FLAT_ATOMIC_WAVEFRONTS_sum - - -range: - -gpu: - -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx908/pmc_tcc2_perf.txt b/src/rocprof_compute_soc/profile_configs/gfx908/pmc_tcc2_perf.txt deleted file mode 100644 index c070fda797..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx908/pmc_tcc2_perf.txt +++ /dev/null @@ -1,26 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_CYCLES SQ_BUSY_CYCLES SQ_WAVES - -pmc: TCC_CYCLE[0] TCC_RW_REQ[0] TCC_HIT[0] TCC_MISS[0] TCC_CYCLE[1] TCC_RW_REQ[1] TCC_HIT[1] TCC_MISS[1] TCC_CYCLE[2] TCC_RW_REQ[2] TCC_HIT[2] TCC_MISS[2] TCC_CYCLE[3] TCC_RW_REQ[3] TCC_HIT[3] TCC_MISS[3] TCC_CYCLE[4] TCC_RW_REQ[4] TCC_HIT[4] TCC_MISS[4] TCC_CYCLE[5] TCC_RW_REQ[5] TCC_HIT[5] TCC_MISS[5] TCC_CYCLE[6] TCC_RW_REQ[6] TCC_HIT[6] TCC_MISS[6] TCC_CYCLE[7] TCC_RW_REQ[7] TCC_HIT[7] TCC_MISS[7] TCC_CYCLE[8] TCC_RW_REQ[8] TCC_HIT[8] TCC_MISS[8] TCC_CYCLE[9] TCC_RW_REQ[9] TCC_HIT[9] TCC_MISS[9] TCC_CYCLE[10] TCC_RW_REQ[10] TCC_HIT[10] TCC_MISS[10] TCC_CYCLE[11] TCC_RW_REQ[11] TCC_HIT[11] TCC_MISS[11] TCC_CYCLE[12] TCC_RW_REQ[12] TCC_HIT[12] TCC_MISS[12] TCC_CYCLE[13] TCC_RW_REQ[13] TCC_HIT[13] TCC_MISS[13] TCC_CYCLE[14] TCC_RW_REQ[14] TCC_HIT[14] TCC_MISS[14] TCC_CYCLE[15] TCC_RW_REQ[15] TCC_HIT[15] TCC_MISS[15] TCC_CYCLE[16] TCC_RW_REQ[16] TCC_HIT[16] TCC_MISS[16] TCC_CYCLE[17] TCC_RW_REQ[17] TCC_HIT[17] TCC_MISS[17] TCC_CYCLE[18] TCC_RW_REQ[18] TCC_HIT[18] TCC_MISS[18] TCC_CYCLE[19] TCC_RW_REQ[19] TCC_HIT[19] TCC_MISS[19] TCC_CYCLE[20] TCC_RW_REQ[20] TCC_HIT[20] TCC_MISS[20] TCC_CYCLE[21] TCC_RW_REQ[21] TCC_HIT[21] TCC_MISS[21] TCC_CYCLE[22] TCC_RW_REQ[22] TCC_HIT[22] TCC_MISS[22] TCC_CYCLE[23] TCC_RW_REQ[23] TCC_HIT[23] TCC_MISS[23] TCC_CYCLE[24] TCC_RW_REQ[24] TCC_HIT[24] TCC_MISS[24] TCC_CYCLE[25] TCC_RW_REQ[25] TCC_HIT[25] TCC_MISS[25] TCC_CYCLE[26] TCC_RW_REQ[26] TCC_HIT[26] TCC_MISS[26] TCC_CYCLE[27] TCC_RW_REQ[27] TCC_HIT[27] TCC_MISS[27] TCC_CYCLE[28] TCC_RW_REQ[28] TCC_HIT[28] TCC_MISS[28] TCC_CYCLE[29] TCC_RW_REQ[29] TCC_HIT[29] TCC_MISS[29] TCC_CYCLE[30] TCC_RW_REQ[30] TCC_HIT[30] TCC_MISS[30] TCC_CYCLE[31] TCC_RW_REQ[31] TCC_HIT[31] TCC_MISS[31] - - - -pmc: TCC_REQ[0] TCC_READ[0] TCC_WRITE[0] TCC_ATOMIC[0] TCC_REQ[1] TCC_READ[1] TCC_WRITE[1] TCC_ATOMIC[1] TCC_REQ[2] TCC_READ[2] TCC_WRITE[2] TCC_ATOMIC[2] TCC_REQ[3] TCC_READ[3] TCC_WRITE[3] TCC_ATOMIC[3] TCC_REQ[4] TCC_READ[4] TCC_WRITE[4] TCC_ATOMIC[4] TCC_REQ[5] TCC_READ[5] TCC_WRITE[5] TCC_ATOMIC[5] TCC_REQ[6] TCC_READ[6] TCC_WRITE[6] TCC_ATOMIC[6] TCC_REQ[7] TCC_READ[7] TCC_WRITE[7] TCC_ATOMIC[7] TCC_REQ[8] TCC_READ[8] TCC_WRITE[8] TCC_ATOMIC[8] TCC_REQ[9] TCC_READ[9] TCC_WRITE[9] TCC_ATOMIC[9] TCC_REQ[10] TCC_READ[10] TCC_WRITE[10] TCC_ATOMIC[10] TCC_REQ[11] TCC_READ[11] TCC_WRITE[11] TCC_ATOMIC[11] TCC_REQ[12] TCC_READ[12] TCC_WRITE[12] TCC_ATOMIC[12] TCC_REQ[13] TCC_READ[13] TCC_WRITE[13] TCC_ATOMIC[13] TCC_REQ[14] TCC_READ[14] TCC_WRITE[14] TCC_ATOMIC[14] TCC_REQ[15] TCC_READ[15] TCC_WRITE[15] TCC_ATOMIC[15] TCC_REQ[16] TCC_READ[16] TCC_WRITE[16] TCC_ATOMIC[16] TCC_REQ[17] TCC_READ[17] TCC_WRITE[17] TCC_ATOMIC[17] TCC_REQ[18] TCC_READ[18] TCC_WRITE[18] TCC_ATOMIC[18] TCC_REQ[19] TCC_READ[19] TCC_WRITE[19] TCC_ATOMIC[19] TCC_REQ[20] TCC_READ[20] TCC_WRITE[20] TCC_ATOMIC[20] TCC_REQ[21] TCC_READ[21] TCC_WRITE[21] TCC_ATOMIC[21] TCC_REQ[22] TCC_READ[22] TCC_WRITE[22] TCC_ATOMIC[22] TCC_REQ[23] TCC_READ[23] TCC_WRITE[23] TCC_ATOMIC[23] TCC_REQ[24] TCC_READ[24] TCC_WRITE[24] TCC_ATOMIC[24] TCC_REQ[25] TCC_READ[25] TCC_WRITE[25] TCC_ATOMIC[25] TCC_REQ[26] TCC_READ[26] TCC_WRITE[26] TCC_ATOMIC[26] TCC_REQ[27] TCC_READ[27] TCC_WRITE[27] TCC_ATOMIC[27] TCC_REQ[28] TCC_READ[28] TCC_WRITE[28] TCC_ATOMIC[28] TCC_REQ[29] TCC_READ[29] TCC_WRITE[29] TCC_ATOMIC[29] TCC_REQ[30] TCC_READ[30] TCC_WRITE[30] TCC_ATOMIC[30] TCC_REQ[31] TCC_READ[31] TCC_WRITE[31] TCC_ATOMIC[31] - - - -pmc: TCC_EA_RDREQ[0] TCC_EA_RDREQ_32B[0] TCC_EA_WRREQ[0] TCC_EA_WRREQ_64B[0] TCC_EA_RDREQ[1] TCC_EA_RDREQ_32B[1] TCC_EA_WRREQ[1] TCC_EA_WRREQ_64B[1] TCC_EA_RDREQ[2] TCC_EA_RDREQ_32B[2] TCC_EA_WRREQ[2] TCC_EA_WRREQ_64B[2] TCC_EA_RDREQ[3] TCC_EA_RDREQ_32B[3] TCC_EA_WRREQ[3] TCC_EA_WRREQ_64B[3] TCC_EA_RDREQ[4] TCC_EA_RDREQ_32B[4] TCC_EA_WRREQ[4] TCC_EA_WRREQ_64B[4] TCC_EA_RDREQ[5] TCC_EA_RDREQ_32B[5] TCC_EA_WRREQ[5] TCC_EA_WRREQ_64B[5] TCC_EA_RDREQ[6] TCC_EA_RDREQ_32B[6] TCC_EA_WRREQ[6] TCC_EA_WRREQ_64B[6] TCC_EA_RDREQ[7] TCC_EA_RDREQ_32B[7] TCC_EA_WRREQ[7] TCC_EA_WRREQ_64B[7] TCC_EA_RDREQ[8] TCC_EA_RDREQ_32B[8] TCC_EA_WRREQ[8] TCC_EA_WRREQ_64B[8] TCC_EA_RDREQ[9] TCC_EA_RDREQ_32B[9] TCC_EA_WRREQ[9] TCC_EA_WRREQ_64B[9] TCC_EA_RDREQ[10] TCC_EA_RDREQ_32B[10] TCC_EA_WRREQ[10] TCC_EA_WRREQ_64B[10] TCC_EA_RDREQ[11] TCC_EA_RDREQ_32B[11] TCC_EA_WRREQ[11] TCC_EA_WRREQ_64B[11] TCC_EA_RDREQ[12] TCC_EA_RDREQ_32B[12] TCC_EA_WRREQ[12] TCC_EA_WRREQ_64B[12] TCC_EA_RDREQ[13] TCC_EA_RDREQ_32B[13] TCC_EA_WRREQ[13] TCC_EA_WRREQ_64B[13] TCC_EA_RDREQ[14] TCC_EA_RDREQ_32B[14] TCC_EA_WRREQ[14] TCC_EA_WRREQ_64B[14] TCC_EA_RDREQ[15] TCC_EA_RDREQ_32B[15] TCC_EA_WRREQ[15] TCC_EA_WRREQ_64B[15] TCC_EA_RDREQ[16] TCC_EA_RDREQ_32B[16] TCC_EA_WRREQ[16] TCC_EA_WRREQ_64B[16] TCC_EA_RDREQ[17] TCC_EA_RDREQ_32B[17] TCC_EA_WRREQ[17] TCC_EA_WRREQ_64B[17] TCC_EA_RDREQ[18] TCC_EA_RDREQ_32B[18] TCC_EA_WRREQ[18] TCC_EA_WRREQ_64B[18] TCC_EA_RDREQ[19] TCC_EA_RDREQ_32B[19] TCC_EA_WRREQ[19] TCC_EA_WRREQ_64B[19] TCC_EA_RDREQ[20] TCC_EA_RDREQ_32B[20] TCC_EA_WRREQ[20] TCC_EA_WRREQ_64B[20] TCC_EA_RDREQ[21] TCC_EA_RDREQ_32B[21] TCC_EA_WRREQ[21] TCC_EA_WRREQ_64B[21] TCC_EA_RDREQ[22] TCC_EA_RDREQ_32B[22] TCC_EA_WRREQ[22] TCC_EA_WRREQ_64B[22] TCC_EA_RDREQ[23] TCC_EA_RDREQ_32B[23] TCC_EA_WRREQ[23] TCC_EA_WRREQ_64B[23] TCC_EA_RDREQ[24] TCC_EA_RDREQ_32B[24] TCC_EA_WRREQ[24] TCC_EA_WRREQ_64B[24] TCC_EA_RDREQ[25] TCC_EA_RDREQ_32B[25] TCC_EA_WRREQ[25] TCC_EA_WRREQ_64B[25] TCC_EA_RDREQ[26] TCC_EA_RDREQ_32B[26] TCC_EA_WRREQ[26] TCC_EA_WRREQ_64B[26] TCC_EA_RDREQ[27] TCC_EA_RDREQ_32B[27] TCC_EA_WRREQ[27] TCC_EA_WRREQ_64B[27] TCC_EA_RDREQ[28] TCC_EA_RDREQ_32B[28] TCC_EA_WRREQ[28] TCC_EA_WRREQ_64B[28] TCC_EA_RDREQ[29] TCC_EA_RDREQ_32B[29] TCC_EA_WRREQ[29] TCC_EA_WRREQ_64B[29] TCC_EA_RDREQ[30] TCC_EA_RDREQ_32B[30] TCC_EA_WRREQ[30] TCC_EA_WRREQ_64B[30] TCC_EA_RDREQ[31] TCC_EA_RDREQ_32B[31] TCC_EA_WRREQ[31] TCC_EA_WRREQ_64B[31] - - -pmc: TCC_EA_ATOMIC[0] TCC_EA_RDREQ_LEVEL[0] TCC_EA_WRREQ_LEVEL[0] TCC_EA_ATOMIC_LEVEL[0] TCC_EA_ATOMIC[1] TCC_EA_RDREQ_LEVEL[1] TCC_EA_WRREQ_LEVEL[1] TCC_EA_ATOMIC_LEVEL[1] TCC_EA_ATOMIC[2] TCC_EA_RDREQ_LEVEL[2] TCC_EA_WRREQ_LEVEL[2] TCC_EA_ATOMIC_LEVEL[2] TCC_EA_ATOMIC[3] TCC_EA_RDREQ_LEVEL[3] TCC_EA_WRREQ_LEVEL[3] TCC_EA_ATOMIC_LEVEL[3] TCC_EA_ATOMIC[4] TCC_EA_RDREQ_LEVEL[4] TCC_EA_WRREQ_LEVEL[4] TCC_EA_ATOMIC_LEVEL[4] TCC_EA_ATOMIC[5] TCC_EA_RDREQ_LEVEL[5] TCC_EA_WRREQ_LEVEL[5] TCC_EA_ATOMIC_LEVEL[5] TCC_EA_ATOMIC[6] TCC_EA_RDREQ_LEVEL[6] TCC_EA_WRREQ_LEVEL[6] TCC_EA_ATOMIC_LEVEL[6] TCC_EA_ATOMIC[7] TCC_EA_RDREQ_LEVEL[7] TCC_EA_WRREQ_LEVEL[7] TCC_EA_ATOMIC_LEVEL[7] TCC_EA_ATOMIC[8] TCC_EA_RDREQ_LEVEL[8] TCC_EA_WRREQ_LEVEL[8] TCC_EA_ATOMIC_LEVEL[8] TCC_EA_ATOMIC[9] TCC_EA_RDREQ_LEVEL[9] TCC_EA_WRREQ_LEVEL[9] TCC_EA_ATOMIC_LEVEL[9] TCC_EA_ATOMIC[10] TCC_EA_RDREQ_LEVEL[10] TCC_EA_WRREQ_LEVEL[10] TCC_EA_ATOMIC_LEVEL[10] TCC_EA_ATOMIC[11] TCC_EA_RDREQ_LEVEL[11] TCC_EA_WRREQ_LEVEL[11] TCC_EA_ATOMIC_LEVEL[11] TCC_EA_ATOMIC[12] TCC_EA_RDREQ_LEVEL[12] TCC_EA_WRREQ_LEVEL[12] TCC_EA_ATOMIC_LEVEL[12] TCC_EA_ATOMIC[13] TCC_EA_RDREQ_LEVEL[13] TCC_EA_WRREQ_LEVEL[13] TCC_EA_ATOMIC_LEVEL[13] TCC_EA_ATOMIC[14] TCC_EA_RDREQ_LEVEL[14] TCC_EA_WRREQ_LEVEL[14] TCC_EA_ATOMIC_LEVEL[14] TCC_EA_ATOMIC[15] TCC_EA_RDREQ_LEVEL[15] TCC_EA_WRREQ_LEVEL[15] TCC_EA_ATOMIC_LEVEL[15] TCC_EA_ATOMIC[16] TCC_EA_RDREQ_LEVEL[16] TCC_EA_WRREQ_LEVEL[16] TCC_EA_ATOMIC_LEVEL[16] TCC_EA_ATOMIC[17] TCC_EA_RDREQ_LEVEL[17] TCC_EA_WRREQ_LEVEL[17] TCC_EA_ATOMIC_LEVEL[17] TCC_EA_ATOMIC[18] TCC_EA_RDREQ_LEVEL[18] TCC_EA_WRREQ_LEVEL[18] TCC_EA_ATOMIC_LEVEL[18] TCC_EA_ATOMIC[19] TCC_EA_RDREQ_LEVEL[19] TCC_EA_WRREQ_LEVEL[19] TCC_EA_ATOMIC_LEVEL[19] TCC_EA_ATOMIC[20] TCC_EA_RDREQ_LEVEL[20] TCC_EA_WRREQ_LEVEL[20] TCC_EA_ATOMIC_LEVEL[20] TCC_EA_ATOMIC[21] TCC_EA_RDREQ_LEVEL[21] TCC_EA_WRREQ_LEVEL[21] TCC_EA_ATOMIC_LEVEL[21] TCC_EA_ATOMIC[22] TCC_EA_RDREQ_LEVEL[22] TCC_EA_WRREQ_LEVEL[22] TCC_EA_ATOMIC_LEVEL[22] TCC_EA_ATOMIC[23] TCC_EA_RDREQ_LEVEL[23] TCC_EA_WRREQ_LEVEL[23] TCC_EA_ATOMIC_LEVEL[23] TCC_EA_ATOMIC[24] TCC_EA_RDREQ_LEVEL[24] TCC_EA_WRREQ_LEVEL[24] TCC_EA_ATOMIC_LEVEL[24] TCC_EA_ATOMIC[25] TCC_EA_RDREQ_LEVEL[25] TCC_EA_WRREQ_LEVEL[25] TCC_EA_ATOMIC_LEVEL[25] TCC_EA_ATOMIC[26] TCC_EA_RDREQ_LEVEL[26] TCC_EA_WRREQ_LEVEL[26] TCC_EA_ATOMIC_LEVEL[26] TCC_EA_ATOMIC[27] TCC_EA_RDREQ_LEVEL[27] TCC_EA_WRREQ_LEVEL[27] TCC_EA_ATOMIC_LEVEL[27] TCC_EA_ATOMIC[28] TCC_EA_RDREQ_LEVEL[28] TCC_EA_WRREQ_LEVEL[28] TCC_EA_ATOMIC_LEVEL[28] TCC_EA_ATOMIC[29] TCC_EA_RDREQ_LEVEL[29] TCC_EA_WRREQ_LEVEL[29] TCC_EA_ATOMIC_LEVEL[29] TCC_EA_ATOMIC[30] TCC_EA_RDREQ_LEVEL[30] TCC_EA_WRREQ_LEVEL[30] TCC_EA_ATOMIC_LEVEL[30] TCC_EA_ATOMIC[31] TCC_EA_RDREQ_LEVEL[31] TCC_EA_WRREQ_LEVEL[31] TCC_EA_ATOMIC_LEVEL[31] - - - - -pmc: TCC_EA_RDREQ_IO_CREDIT_STALL[0] TCC_EA_RDREQ_GMI_CREDIT_STALL[0] TCC_EA_RDREQ_DRAM_CREDIT_STALL[0] TCC_EA_RDREQ_IO_CREDIT_STALL[1] TCC_EA_RDREQ_GMI_CREDIT_STALL[1] TCC_EA_RDREQ_DRAM_CREDIT_STALL[1] TCC_EA_RDREQ_IO_CREDIT_STALL[2] TCC_EA_RDREQ_GMI_CREDIT_STALL[2] TCC_EA_RDREQ_DRAM_CREDIT_STALL[2] TCC_EA_RDREQ_IO_CREDIT_STALL[3] TCC_EA_RDREQ_GMI_CREDIT_STALL[3] TCC_EA_RDREQ_DRAM_CREDIT_STALL[3] TCC_EA_RDREQ_IO_CREDIT_STALL[4] TCC_EA_RDREQ_GMI_CREDIT_STALL[4] TCC_EA_RDREQ_DRAM_CREDIT_STALL[4] TCC_EA_RDREQ_IO_CREDIT_STALL[5] TCC_EA_RDREQ_GMI_CREDIT_STALL[5] TCC_EA_RDREQ_DRAM_CREDIT_STALL[5] TCC_EA_RDREQ_IO_CREDIT_STALL[6] TCC_EA_RDREQ_GMI_CREDIT_STALL[6] TCC_EA_RDREQ_DRAM_CREDIT_STALL[6] TCC_EA_RDREQ_IO_CREDIT_STALL[7] TCC_EA_RDREQ_GMI_CREDIT_STALL[7] TCC_EA_RDREQ_DRAM_CREDIT_STALL[7] TCC_EA_RDREQ_IO_CREDIT_STALL[8] TCC_EA_RDREQ_GMI_CREDIT_STALL[8] TCC_EA_RDREQ_DRAM_CREDIT_STALL[8] TCC_EA_RDREQ_IO_CREDIT_STALL[9] TCC_EA_RDREQ_GMI_CREDIT_STALL[9] TCC_EA_RDREQ_DRAM_CREDIT_STALL[9] TCC_EA_RDREQ_IO_CREDIT_STALL[10] TCC_EA_RDREQ_GMI_CREDIT_STALL[10] TCC_EA_RDREQ_DRAM_CREDIT_STALL[10] TCC_EA_RDREQ_IO_CREDIT_STALL[11] TCC_EA_RDREQ_GMI_CREDIT_STALL[11] TCC_EA_RDREQ_DRAM_CREDIT_STALL[11] TCC_EA_RDREQ_IO_CREDIT_STALL[12] TCC_EA_RDREQ_GMI_CREDIT_STALL[12] TCC_EA_RDREQ_DRAM_CREDIT_STALL[12] TCC_EA_RDREQ_IO_CREDIT_STALL[13] TCC_EA_RDREQ_GMI_CREDIT_STALL[13] TCC_EA_RDREQ_DRAM_CREDIT_STALL[13] TCC_EA_RDREQ_IO_CREDIT_STALL[14] TCC_EA_RDREQ_GMI_CREDIT_STALL[14] TCC_EA_RDREQ_DRAM_CREDIT_STALL[14] TCC_EA_RDREQ_IO_CREDIT_STALL[15] TCC_EA_RDREQ_GMI_CREDIT_STALL[15] TCC_EA_RDREQ_DRAM_CREDIT_STALL[15] TCC_EA_RDREQ_IO_CREDIT_STALL[16] TCC_EA_RDREQ_GMI_CREDIT_STALL[16] TCC_EA_RDREQ_DRAM_CREDIT_STALL[16] TCC_EA_RDREQ_IO_CREDIT_STALL[17] TCC_EA_RDREQ_GMI_CREDIT_STALL[17] TCC_EA_RDREQ_DRAM_CREDIT_STALL[17] TCC_EA_RDREQ_IO_CREDIT_STALL[18] TCC_EA_RDREQ_GMI_CREDIT_STALL[18] TCC_EA_RDREQ_DRAM_CREDIT_STALL[18] TCC_EA_RDREQ_IO_CREDIT_STALL[19] TCC_EA_RDREQ_GMI_CREDIT_STALL[19] TCC_EA_RDREQ_DRAM_CREDIT_STALL[19] TCC_EA_RDREQ_IO_CREDIT_STALL[20] TCC_EA_RDREQ_GMI_CREDIT_STALL[20] TCC_EA_RDREQ_DRAM_CREDIT_STALL[20] TCC_EA_RDREQ_IO_CREDIT_STALL[21] TCC_EA_RDREQ_GMI_CREDIT_STALL[21] TCC_EA_RDREQ_DRAM_CREDIT_STALL[21] TCC_EA_RDREQ_IO_CREDIT_STALL[22] TCC_EA_RDREQ_GMI_CREDIT_STALL[22] TCC_EA_RDREQ_DRAM_CREDIT_STALL[22] TCC_EA_RDREQ_IO_CREDIT_STALL[23] TCC_EA_RDREQ_GMI_CREDIT_STALL[23] TCC_EA_RDREQ_DRAM_CREDIT_STALL[23] TCC_EA_RDREQ_IO_CREDIT_STALL[24] TCC_EA_RDREQ_GMI_CREDIT_STALL[24] TCC_EA_RDREQ_DRAM_CREDIT_STALL[24] TCC_EA_RDREQ_IO_CREDIT_STALL[25] TCC_EA_RDREQ_GMI_CREDIT_STALL[25] TCC_EA_RDREQ_DRAM_CREDIT_STALL[25] TCC_EA_RDREQ_IO_CREDIT_STALL[26] TCC_EA_RDREQ_GMI_CREDIT_STALL[26] TCC_EA_RDREQ_DRAM_CREDIT_STALL[26] TCC_EA_RDREQ_IO_CREDIT_STALL[27] TCC_EA_RDREQ_GMI_CREDIT_STALL[27] TCC_EA_RDREQ_DRAM_CREDIT_STALL[27] TCC_EA_RDREQ_IO_CREDIT_STALL[28] TCC_EA_RDREQ_GMI_CREDIT_STALL[28] TCC_EA_RDREQ_DRAM_CREDIT_STALL[28] TCC_EA_RDREQ_IO_CREDIT_STALL[29] TCC_EA_RDREQ_GMI_CREDIT_STALL[29] TCC_EA_RDREQ_DRAM_CREDIT_STALL[29] TCC_EA_RDREQ_IO_CREDIT_STALL[30] TCC_EA_RDREQ_GMI_CREDIT_STALL[30] TCC_EA_RDREQ_DRAM_CREDIT_STALL[30] TCC_EA_RDREQ_IO_CREDIT_STALL[31] TCC_EA_RDREQ_GMI_CREDIT_STALL[31] TCC_EA_RDREQ_DRAM_CREDIT_STALL[31] - - -pmc: TCC_EA_WRREQ_IO_CREDIT_STALL[0] TCC_EA_WRREQ_GMI_CREDIT_STALL[0] TCC_EA_WRREQ_DRAM_CREDIT_STALL[0] TCC_TOO_MANY_EA_WRREQS_STALL[0] TCC_EA_WRREQ_IO_CREDIT_STALL[1] TCC_EA_WRREQ_GMI_CREDIT_STALL[1] TCC_EA_WRREQ_DRAM_CREDIT_STALL[1] TCC_TOO_MANY_EA_WRREQS_STALL[1] TCC_EA_WRREQ_IO_CREDIT_STALL[2] TCC_EA_WRREQ_GMI_CREDIT_STALL[2] TCC_EA_WRREQ_DRAM_CREDIT_STALL[2] TCC_TOO_MANY_EA_WRREQS_STALL[2] TCC_EA_WRREQ_IO_CREDIT_STALL[3] TCC_EA_WRREQ_GMI_CREDIT_STALL[3] TCC_EA_WRREQ_DRAM_CREDIT_STALL[3] TCC_TOO_MANY_EA_WRREQS_STALL[3] TCC_EA_WRREQ_IO_CREDIT_STALL[4] TCC_EA_WRREQ_GMI_CREDIT_STALL[4] TCC_EA_WRREQ_DRAM_CREDIT_STALL[4] TCC_TOO_MANY_EA_WRREQS_STALL[4] TCC_EA_WRREQ_IO_CREDIT_STALL[5] TCC_EA_WRREQ_GMI_CREDIT_STALL[5] TCC_EA_WRREQ_DRAM_CREDIT_STALL[5] TCC_TOO_MANY_EA_WRREQS_STALL[5] TCC_EA_WRREQ_IO_CREDIT_STALL[6] TCC_EA_WRREQ_GMI_CREDIT_STALL[6] TCC_EA_WRREQ_DRAM_CREDIT_STALL[6] TCC_TOO_MANY_EA_WRREQS_STALL[6] TCC_EA_WRREQ_IO_CREDIT_STALL[7] TCC_EA_WRREQ_GMI_CREDIT_STALL[7] TCC_EA_WRREQ_DRAM_CREDIT_STALL[7] TCC_TOO_MANY_EA_WRREQS_STALL[7] TCC_EA_WRREQ_IO_CREDIT_STALL[8] TCC_EA_WRREQ_GMI_CREDIT_STALL[8] TCC_EA_WRREQ_DRAM_CREDIT_STALL[8] TCC_TOO_MANY_EA_WRREQS_STALL[8] TCC_EA_WRREQ_IO_CREDIT_STALL[9] TCC_EA_WRREQ_GMI_CREDIT_STALL[9] TCC_EA_WRREQ_DRAM_CREDIT_STALL[9] TCC_TOO_MANY_EA_WRREQS_STALL[9] TCC_EA_WRREQ_IO_CREDIT_STALL[10] TCC_EA_WRREQ_GMI_CREDIT_STALL[10] TCC_EA_WRREQ_DRAM_CREDIT_STALL[10] TCC_TOO_MANY_EA_WRREQS_STALL[10] TCC_EA_WRREQ_IO_CREDIT_STALL[11] TCC_EA_WRREQ_GMI_CREDIT_STALL[11] TCC_EA_WRREQ_DRAM_CREDIT_STALL[11] TCC_TOO_MANY_EA_WRREQS_STALL[11] TCC_EA_WRREQ_IO_CREDIT_STALL[12] TCC_EA_WRREQ_GMI_CREDIT_STALL[12] TCC_EA_WRREQ_DRAM_CREDIT_STALL[12] TCC_TOO_MANY_EA_WRREQS_STALL[12] TCC_EA_WRREQ_IO_CREDIT_STALL[13] TCC_EA_WRREQ_GMI_CREDIT_STALL[13] TCC_EA_WRREQ_DRAM_CREDIT_STALL[13] TCC_TOO_MANY_EA_WRREQS_STALL[13] TCC_EA_WRREQ_IO_CREDIT_STALL[14] TCC_EA_WRREQ_GMI_CREDIT_STALL[14] TCC_EA_WRREQ_DRAM_CREDIT_STALL[14] TCC_TOO_MANY_EA_WRREQS_STALL[14] TCC_EA_WRREQ_IO_CREDIT_STALL[15] TCC_EA_WRREQ_GMI_CREDIT_STALL[15] TCC_EA_WRREQ_DRAM_CREDIT_STALL[15] TCC_TOO_MANY_EA_WRREQS_STALL[15] TCC_EA_WRREQ_IO_CREDIT_STALL[16] TCC_EA_WRREQ_GMI_CREDIT_STALL[16] TCC_EA_WRREQ_DRAM_CREDIT_STALL[16] TCC_TOO_MANY_EA_WRREQS_STALL[16] TCC_EA_WRREQ_IO_CREDIT_STALL[17] TCC_EA_WRREQ_GMI_CREDIT_STALL[17] TCC_EA_WRREQ_DRAM_CREDIT_STALL[17] TCC_TOO_MANY_EA_WRREQS_STALL[17] TCC_EA_WRREQ_IO_CREDIT_STALL[18] TCC_EA_WRREQ_GMI_CREDIT_STALL[18] TCC_EA_WRREQ_DRAM_CREDIT_STALL[18] TCC_TOO_MANY_EA_WRREQS_STALL[18] TCC_EA_WRREQ_IO_CREDIT_STALL[19] TCC_EA_WRREQ_GMI_CREDIT_STALL[19] TCC_EA_WRREQ_DRAM_CREDIT_STALL[19] TCC_TOO_MANY_EA_WRREQS_STALL[19] TCC_EA_WRREQ_IO_CREDIT_STALL[20] TCC_EA_WRREQ_GMI_CREDIT_STALL[20] TCC_EA_WRREQ_DRAM_CREDIT_STALL[20] TCC_TOO_MANY_EA_WRREQS_STALL[20] TCC_EA_WRREQ_IO_CREDIT_STALL[21] TCC_EA_WRREQ_GMI_CREDIT_STALL[21] TCC_EA_WRREQ_DRAM_CREDIT_STALL[21] TCC_TOO_MANY_EA_WRREQS_STALL[21] TCC_EA_WRREQ_IO_CREDIT_STALL[22] TCC_EA_WRREQ_GMI_CREDIT_STALL[22] TCC_EA_WRREQ_DRAM_CREDIT_STALL[22] TCC_TOO_MANY_EA_WRREQS_STALL[22] TCC_EA_WRREQ_IO_CREDIT_STALL[23] TCC_EA_WRREQ_GMI_CREDIT_STALL[23] TCC_EA_WRREQ_DRAM_CREDIT_STALL[23] TCC_TOO_MANY_EA_WRREQS_STALL[23] TCC_EA_WRREQ_IO_CREDIT_STALL[24] TCC_EA_WRREQ_GMI_CREDIT_STALL[24] TCC_EA_WRREQ_DRAM_CREDIT_STALL[24] TCC_TOO_MANY_EA_WRREQS_STALL[24] TCC_EA_WRREQ_IO_CREDIT_STALL[25] TCC_EA_WRREQ_GMI_CREDIT_STALL[25] TCC_EA_WRREQ_DRAM_CREDIT_STALL[25] TCC_TOO_MANY_EA_WRREQS_STALL[25] TCC_EA_WRREQ_IO_CREDIT_STALL[26] TCC_EA_WRREQ_GMI_CREDIT_STALL[26] TCC_EA_WRREQ_DRAM_CREDIT_STALL[26] TCC_TOO_MANY_EA_WRREQS_STALL[26] TCC_EA_WRREQ_IO_CREDIT_STALL[27] TCC_EA_WRREQ_GMI_CREDIT_STALL[27] TCC_EA_WRREQ_DRAM_CREDIT_STALL[27] TCC_TOO_MANY_EA_WRREQS_STALL[27] TCC_EA_WRREQ_IO_CREDIT_STALL[28] TCC_EA_WRREQ_GMI_CREDIT_STALL[28] TCC_EA_WRREQ_DRAM_CREDIT_STALL[28] TCC_TOO_MANY_EA_WRREQS_STALL[28] TCC_EA_WRREQ_IO_CREDIT_STALL[29] TCC_EA_WRREQ_GMI_CREDIT_STALL[29] TCC_EA_WRREQ_DRAM_CREDIT_STALL[29] TCC_TOO_MANY_EA_WRREQS_STALL[29] TCC_EA_WRREQ_IO_CREDIT_STALL[30] TCC_EA_WRREQ_GMI_CREDIT_STALL[30] TCC_EA_WRREQ_DRAM_CREDIT_STALL[30] TCC_TOO_MANY_EA_WRREQS_STALL[30] TCC_EA_WRREQ_IO_CREDIT_STALL[31] TCC_EA_WRREQ_GMI_CREDIT_STALL[31] TCC_EA_WRREQ_DRAM_CREDIT_STALL[31] TCC_TOO_MANY_EA_WRREQS_STALL[31] - -gpu: -kernel: -range: diff --git a/src/rocprof_compute_soc/profile_configs/gfx908/pmc_tcc_perf.txt b/src/rocprof_compute_soc/profile_configs/gfx908/pmc_tcc_perf.txt deleted file mode 100644 index 89f81f65de..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx908/pmc_tcc_perf.txt +++ /dev/null @@ -1,18 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_CYCLES SQ_BUSY_CYCLES SQ_WAVES - -pmc: TCC_CYCLE_sum TCC_BUSY_sum TCC_PROBE_sum TCC_PROBE_ALL_sum -pmc: TCC_NC_REQ_sum TCC_UC_REQ_sum TCC_CC_REQ_sum TCC_RW_REQ_sum -pmc: TCC_REQ_sum TCC_STREAMING_REQ_sum TCC_HIT_sum TCC_MISS_sum -pmc: TCC_READ_sum TCC_WRITE_sum TCC_ATOMIC_sum TCC_WRITEBACK_sum -pmc: TCC_EA_WRREQ_sum TCC_EA_WRREQ_64B_sum TCC_EA_WR_UNCACHED_32B_sum TCC_EA_WRREQ_DRAM_sum -pmc: TCC_EA_WRREQ_STALL_sum TCC_EA_WRREQ_IO_CREDIT_STALL_sum TCC_EA_WRREQ_GMI_CREDIT_STALL_sum TCC_EA_WRREQ_DRAM_CREDIT_STALL_sum -pmc: TCC_EA_RDREQ_sum TCC_EA_RDREQ_32B_sum TCC_EA_RD_UNCACHED_32B_sum TCC_EA_RDREQ_DRAM_sum -pmc: TCC_EA_RDREQ_IO_CREDIT_STALL_sum TCC_EA_RDREQ_GMI_CREDIT_STALL_sum TCC_EA_RDREQ_DRAM_CREDIT_STALL_sum -pmc: TCC_TAG_STALL_sum TCC_NORMAL_WRITEBACK_sum TCC_ALL_TC_OP_WB_WRITEBACK_sum TCC_NORMAL_EVICT_sum -pmc: TCC_ALL_TC_OP_INV_EVICT_sum TCC_TOO_MANY_EA_WRREQS_STALL_sum TCC_EA_ATOMIC_sum -pmc: TCC_EA_RDREQ_LEVEL_sum TCC_EA_WRREQ_LEVEL_sum TCC_EA_ATOMIC_LEVEL_sum - -gpu: -kernel: - -range: diff --git a/src/rocprof_compute_soc/profile_configs/gfx908/pmc_tcp_perf.txt b/src/rocprof_compute_soc/profile_configs/gfx908/pmc_tcp_perf.txt deleted file mode 100644 index 9f614e4460..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx908/pmc_tcp_perf.txt +++ /dev/null @@ -1,18 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_CYCLES SQ_BUSY_CYCLES SQ_BUSY_CU_CYCLES SQ_WAVES SQ_WAVE_CYCLES - -pmc: TCP_GATE_EN1_sum TCP_GATE_EN2_sum TCP_TD_TCP_STALL_CYCLES_sum TCP_TCR_TCP_STALL_CYCLES_sum -pmc: TCP_READ_TAGCONFLICT_STALL_CYCLES_sum TCP_WRITE_TAGCONFLICT_STALL_CYCLES_sum TCP_ATOMIC_TAGCONFLICT_STALL_CYCLES_sum TCP_TA_TCP_STATE_READ_sum -pmc: TCP_VOLATILE_sum TCP_TOTAL_ACCESSES_sum TCP_TOTAL_READ_sum TCP_TOTAL_WRITE_sum -pmc: TCP_TOTAL_ATOMIC_WITH_RET_sum TCP_TOTAL_ATOMIC_WITHOUT_RET_sum TCP_TOTAL_WRITEBACK_INVALIDATES_sum TCP_TOTAL_CACHE_ACCESSES_sum -pmc: TCP_UTCL1_TRANSLATION_MISS_sum TCP_UTCL1_TRANSLATION_HIT_sum TCP_UTCL1_PERMISSION_MISS_sum TCP_UTCL1_REQUEST_sum -pmc: TCP_TCP_LATENCY_sum TCP_TCC_READ_REQ_LATENCY_sum TCP_TCC_WRITE_REQ_LATENCY_sum TCP_TCC_READ_REQ_sum -pmc: TCP_TCC_WRITE_REQ_sum TCP_TCC_ATOMIC_WITH_RET_REQ_sum TCP_TCC_ATOMIC_WITHOUT_RET_REQ_sum TCP_TCC_NC_READ_REQ_sum -pmc: TCP_TCC_NC_WRITE_REQ_sum TCP_TCC_NC_ATOMIC_REQ_sum TCP_TCC_UC_READ_REQ_sum TCP_TCC_UC_WRITE_REQ_sum -pmc: TCP_TCC_UC_ATOMIC_REQ_sum TCP_TCC_CC_READ_REQ_sum TCP_TCC_CC_WRITE_REQ_sum TCP_TCC_CC_ATOMIC_REQ_sum -pmc: TCP_TCC_RW_READ_REQ_sum TCP_TCC_RW_WRITE_REQ_sum TCP_TCC_RW_ATOMIC_REQ_sum TCP_PENDING_STALL_CYCLES_sum - -#pmc: TCA_CYCLE_sum TCA_BUSY_sum - -gpu: -kernel: -range: diff --git a/src/rocprof_compute_soc/profile_configs/gfx908/pmc_td_perf.txt b/src/rocprof_compute_soc/profile_configs/gfx908/pmc_td_perf.txt deleted file mode 100644 index 91d20dfce0..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx908/pmc_td_perf.txt +++ /dev/null @@ -1,11 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_CYCLES SQ_BUSY_CYCLES SQ_BUSY_CU_CYCLES SQ_WAVES SQ_WAVE_CYCLES - -pmc: TD_TD_BUSY_sum TD_TC_STALL_sum -pmc: TD_COALESCABLE_WAVEFRONT_sum TD_LOAD_WAVEFRONT_sum -pmc: TD_ATOMIC_WAVEFRONT_sum TD_STORE_WAVEFRONT_sum - - - -gpu: -range: -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx90a/pmc_cpc_perf.txt b/src/rocprof_compute_soc/profile_configs/gfx90a/pmc_cpc_perf.txt deleted file mode 100644 index bd80137304..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx90a/pmc_cpc_perf.txt +++ /dev/null @@ -1,13 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_CYCLES SQ_BUSY_CYCLES SQ_WAVES - -pmc: CPC_CPC_STAT_BUSY CPC_CPC_STAT_IDLE -pmc: CPC_CPC_TCIU_BUSY CPC_CPC_TCIU_IDLE -pmc: CPC_CPC_STAT_STALL CPC_UTCL1_STALL_ON_TRANSLATION -pmc: CPC_CPC_UTCL2IU_BUSY CPC_CPC_UTCL2IU_IDLE -pmc: CPC_CPC_UTCL2IU_STALL CPC_ME1_BUSY_FOR_PACKET_DECODE -pmc: CPC_ME1_DC0_SPI_BUSY - -range: -gpu: - -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx90a/pmc_cpf_perf.txt b/src/rocprof_compute_soc/profile_configs/gfx90a/pmc_cpf_perf.txt deleted file mode 100644 index e8119fff56..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx90a/pmc_cpf_perf.txt +++ /dev/null @@ -1,11 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_CYCLES SQ_BUSY_CYCLES SQ_WAVES - -pmc: CPF_CPF_STAT_BUSY CPF_CPF_STAT_STALL -pmc: CPF_CPF_TCIU_BUSY CPF_CPF_TCIU_STALL -pmc: CPF_CPF_STAT_IDLE CPF_CPF_TCIU_IDLE -pmc: CPF_CMP_UTCL1_STALL_ON_TRANSLATION - -range: -gpu: - -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx90a/pmc_spi_perf.txt b/src/rocprof_compute_soc/profile_configs/gfx90a/pmc_spi_perf.txt deleted file mode 100644 index 4eb0d22656..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx90a/pmc_spi_perf.txt +++ /dev/null @@ -1,10 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_CYCLES SQ_BUSY_CYCLES SQ_WAVES GRBM_SPI_BUSY - -pmc: SPI_CSN_WINDOW_VALID SPI_CSN_BUSY SPI_CSN_NUM_THREADGROUPS SPI_CSN_WAVE SPI_RA_REQ_NO_ALLOC SPI_RA_REQ_NO_ALLOC_CSN -pmc: SPI_RA_RES_STALL_CSN SPI_RA_TMP_STALL_CSN SPI_RA_WAVE_SIMD_FULL_CSN SPI_RA_VGPR_SIMD_FULL_CSN SPI_RA_SGPR_SIMD_FULL_CSN SPI_RA_LDS_CU_FULL_CSN -pmc: SPI_RA_BAR_CU_FULL_CSN SPI_RA_TGLIM_CU_FULL_CSN SPI_RA_WVLIM_STALL_CSN SPI_SWC_CSC_WR SPI_VWC_CSC_WR SPI_RA_BULKY_CU_FULL_CSN - -range: -gpu: - -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx90a/pmc_sq_perf1.txt b/src/rocprof_compute_soc/profile_configs/gfx90a/pmc_sq_perf1.txt deleted file mode 100644 index 5b59d03454..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx90a/pmc_sq_perf1.txt +++ /dev/null @@ -1,29 +0,0 @@ -#SQ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_INSTS_VALU_CVT SQ_INSTS_VMEM_WR SQ_INSTS_VMEM_RD SQ_INSTS_VMEM SQ_INSTS_SALU SQ_INSTS_VSKIPPED -pmc: SQ_INSTS SQ_INSTS_VALU SQ_INSTS_VALU_ADD_F16 SQ_INSTS_VALU_MUL_F16 SQ_INSTS_VALU_FMA_F16 SQ_INSTS_VALU_TRANS_F16 SQ_INSTS_VALU_ADD_F32 SQ_INSTS_VALU_MUL_F32 -pmc: SQ_INSTS_VALU_FMA_F32 SQ_INSTS_VALU_TRANS_F32 SQ_INSTS_VALU_ADD_F64 SQ_INSTS_VALU_MUL_F64 SQ_INSTS_VALU_FMA_F64 SQ_INSTS_VALU_TRANS_F64 SQ_INSTS_VALU_INT32 SQ_INSTS_VALU_INT64 -pmc: SQ_INSTS_SMEM SQ_INSTS_FLAT SQ_INSTS_LDS SQ_INSTS_GDS SQ_INSTS_EXP_GDS SQ_INSTS_BRANCH SQ_INSTS_SENDMSG -pmc: SQ_WAVE_CYCLES SQ_WAIT_ANY SQ_WAIT_INST_ANY SQ_ACTIVE_INST_ANY SQ_CYCLES SQ_BUSY_CYCLES SQ_BUSY_CU_CYCLES -pmc: SQ_ACTIVE_INST_VMEM SQ_ACTIVE_INST_LDS SQ_ACTIVE_INST_VALU SQ_ACTIVE_INST_SCA SQ_ACTIVE_INST_EXP_GDS SQ_ACTIVE_INST_MISC SQ_ACTIVE_INST_FLAT SQ_INST_CYCLES_VMEM_WR -pmc: SQ_INST_CYCLES_VMEM_RD SQ_INST_CYCLES_SMEM SQ_INST_CYCLES_SALU SQ_THREAD_CYCLES_VALU SQ_IFETCH SQ_LDS_BANK_CONFLICT SQ_LDS_ADDR_CONFLICT SQ_LDS_UNALIGNED_STALL -pmc: SQ_WAVES SQ_WAVES_EQ_64 SQ_WAVES_LT_64 SQ_WAVES_LT_48 SQ_WAVES_LT_32 SQ_WAVES_LT_16 SQ_ITEMS SQ_INSTS_VSKIPPED -pmc: SQ_LDS_MEM_VIOLATIONS SQ_LDS_ATOMIC_RETURN SQ_LDS_IDX_ACTIVE SQ_WAVES_RESTORED SQ_WAVES_SAVED SQ_INSTS_SMEM_NORM - - -#SQ:MI200 -pmc: SQ_INSTS_MFMA SQ_INSTS_VALU_MFMA_I8 SQ_INSTS_VALU_MFMA_F16 SQ_INSTS_VALU_MFMA_BF16 SQ_INSTS_VALU_MFMA_F32 SQ_INSTS_VALU_MFMA_F64 SQ_VALU_MFMA_BUSY_CYCLES -pmc: SQ_INSTS_FLAT_LDS_ONLY SQ_INSTS_VALU_MFMA_MOPS_I8 SQ_INSTS_VALU_MFMA_MOPS_F16 SQ_INSTS_VALU_MFMA_MOPS_BF16 SQ_INSTS_VALU_MFMA_MOPS_F32 SQ_INSTS_VALU_MFMA_MOPS_F64 - -#SQC -pmc: SQC_TC_INST_REQ SQC_TC_DATA_READ_REQ SQC_TC_DATA_WRITE_REQ SQC_TC_DATA_ATOMIC_REQ SQC_TC_STALL SQC_TC_REQ SQC_DCACHE_REQ_READ_16 -pmc: SQC_ICACHE_REQ SQC_ICACHE_HITS SQC_ICACHE_MISSES SQC_ICACHE_MISSES_DUPLICATE SQC_DCACHE_INPUT_VALID_READYB SQC_DCACHE_ATOMIC SQC_DCACHE_REQ_READ_8 -pmc: SQC_DCACHE_REQ SQC_DCACHE_HITS SQC_DCACHE_MISSES SQC_DCACHE_MISSES_DUPLICATE SQC_DCACHE_REQ_READ_1 SQC_DCACHE_REQ_READ_2 SQC_DCACHE_REQ_READ_4 - - - -######################################## -# Filtering -######################################## -range: -gpu: -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx90a/pmc_sq_perf2.txt b/src/rocprof_compute_soc/profile_configs/gfx90a/pmc_sq_perf2.txt deleted file mode 100644 index 7175c0326c..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx90a/pmc_sq_perf2.txt +++ /dev/null @@ -1,10 +0,0 @@ -################################################# -# VMEM latency -################################################# -pmc: SQ_INSTS_VMEM SQ_INST_LEVEL_VMEM SQ_ACCUM_PREV_HIRES - - -range: -gpu: - -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx90a/pmc_sq_perf3.txt b/src/rocprof_compute_soc/profile_configs/gfx90a/pmc_sq_perf3.txt deleted file mode 100644 index 8d36311af0..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx90a/pmc_sq_perf3.txt +++ /dev/null @@ -1,11 +0,0 @@ -################################################# -# SMEM latency -################################################# -pmc: SQ_INSTS_SMEM SQ_INST_LEVEL_SMEM SQ_ACCUM_PREV_HIRES - - -range: -gpu: - - -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx90a/pmc_sq_perf4.txt b/src/rocprof_compute_soc/profile_configs/gfx90a/pmc_sq_perf4.txt deleted file mode 100644 index f8f29b4f54..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx90a/pmc_sq_perf4.txt +++ /dev/null @@ -1,8 +0,0 @@ -################################################# -# ifetch latency -################################################# -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_WAVES SQ_IFETCH SQ_IFETCH_LEVEL SQ_ACCUM_PREV_HIRES - -range: -gpu: -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx90a/pmc_sq_perf6.txt b/src/rocprof_compute_soc/profile_configs/gfx90a/pmc_sq_perf6.txt deleted file mode 100644 index f9144576eb..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx90a/pmc_sq_perf6.txt +++ /dev/null @@ -1,9 +0,0 @@ -################################################# -# LDS latency -################################################# -pmc: SQ_INSTS_LDS SQ_INST_LEVEL_LDS SQ_ACCUM_PREV_HIRES - -range: -gpu: - -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx90a/pmc_sq_perf8.txt b/src/rocprof_compute_soc/profile_configs/gfx90a/pmc_sq_perf8.txt deleted file mode 100644 index cb29ad2ce1..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx90a/pmc_sq_perf8.txt +++ /dev/null @@ -1,6 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE CPC_ME1_BUSY_FOR_PACKET_DECODE SQ_CYCLES SQ_WAVES SQ_WAVE_CYCLES SQ_BUSY_CYCLES SQ_LEVEL_WAVES SQ_ACCUM_PREV_HIRES - -gpu: -range: - -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx90a/pmc_sqc_perf1.txt b/src/rocprof_compute_soc/profile_configs/gfx90a/pmc_sqc_perf1.txt deleted file mode 100644 index e602d35c1f..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx90a/pmc_sqc_perf1.txt +++ /dev/null @@ -1,11 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_CYCLES SQ_BUSY_CYCLES SQ_BUSY_CU_CYCLES SQ_WAVES SQ_WAVE_CYCLES - - -pmc: SQC_TC_INST_REQ SQC_TC_DATA_READ_REQ SQC_TC_DATA_WRITE_REQ SQC_TC_DATA_ATOMIC_REQ SQC_TC_STALL SQC_TC_REQ SQC_DCACHE_REQ_READ_16 -pmc: SQC_ICACHE_REQ SQC_ICACHE_HITS SQC_ICACHE_MISSES SQC_ICACHE_MISSES_DUPLICATE SQC_DCACHE_INPUT_VALID_READYB SQC_DCACHE_ATOMIC SQC_DCACHE_REQ_READ_8 -pmc: SQC_DCACHE_REQ SQC_DCACHE_HITS SQC_DCACHE_MISSES SQC_DCACHE_MISSES_DUPLICATE SQC_DCACHE_REQ_READ_1 SQC_DCACHE_REQ_READ_2 SQC_DCACHE_REQ_READ_4 - -range: -gpu: - -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx90a/pmc_ta_perf.txt b/src/rocprof_compute_soc/profile_configs/gfx90a/pmc_ta_perf.txt deleted file mode 100644 index be544c627b..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx90a/pmc_ta_perf.txt +++ /dev/null @@ -1,25 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_CYCLES SQ_BUSY_CYCLES SQ_BUSY_CU_CYCLES SQ_WAVES SQ_WAVE_CYCLES - -pmc: TA_TA_BUSY_sum - -# buffer access -pmc: TA_BUFFER_WAVEFRONTS_sum TA_BUFFER_READ_WAVEFRONTS_sum -pmc: TA_BUFFER_WRITE_WAVEFRONTS_sum TA_BUFFER_ATOMIC_WAVEFRONTS_sum -pmc: TA_BUFFER_TOTAL_CYCLES_sum -pmc: TA_BUFFER_COALESCED_READ_CYCLES_sum TA_BUFFER_COALESCED_WRITE_CYCLES_sum - - -# stalls -pmc: TA_ADDR_STALLED_BY_TC_CYCLES_sum TA_TOTAL_WAVEFRONTS_sum -pmc: TA_ADDR_STALLED_BY_TD_CYCLES_sum TA_DATA_STALLED_BY_TC_CYCLES_sum - -# flat accesses -pmc: TA_FLAT_WAVEFRONTS_sum TA_FLAT_READ_WAVEFRONTS_sum -pmc: TA_FLAT_WRITE_WAVEFRONTS_sum TA_FLAT_ATOMIC_WAVEFRONTS_sum - - -range: - -gpu: - -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx90a/pmc_tcc2_perf.txt b/src/rocprof_compute_soc/profile_configs/gfx90a/pmc_tcc2_perf.txt deleted file mode 100644 index c070fda797..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx90a/pmc_tcc2_perf.txt +++ /dev/null @@ -1,26 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_CYCLES SQ_BUSY_CYCLES SQ_WAVES - -pmc: TCC_CYCLE[0] TCC_RW_REQ[0] TCC_HIT[0] TCC_MISS[0] TCC_CYCLE[1] TCC_RW_REQ[1] TCC_HIT[1] TCC_MISS[1] TCC_CYCLE[2] TCC_RW_REQ[2] TCC_HIT[2] TCC_MISS[2] TCC_CYCLE[3] TCC_RW_REQ[3] TCC_HIT[3] TCC_MISS[3] TCC_CYCLE[4] TCC_RW_REQ[4] TCC_HIT[4] TCC_MISS[4] TCC_CYCLE[5] TCC_RW_REQ[5] TCC_HIT[5] TCC_MISS[5] TCC_CYCLE[6] TCC_RW_REQ[6] TCC_HIT[6] TCC_MISS[6] TCC_CYCLE[7] TCC_RW_REQ[7] TCC_HIT[7] TCC_MISS[7] TCC_CYCLE[8] TCC_RW_REQ[8] TCC_HIT[8] TCC_MISS[8] TCC_CYCLE[9] TCC_RW_REQ[9] TCC_HIT[9] TCC_MISS[9] TCC_CYCLE[10] TCC_RW_REQ[10] TCC_HIT[10] TCC_MISS[10] TCC_CYCLE[11] TCC_RW_REQ[11] TCC_HIT[11] TCC_MISS[11] TCC_CYCLE[12] TCC_RW_REQ[12] TCC_HIT[12] TCC_MISS[12] TCC_CYCLE[13] TCC_RW_REQ[13] TCC_HIT[13] TCC_MISS[13] TCC_CYCLE[14] TCC_RW_REQ[14] TCC_HIT[14] TCC_MISS[14] TCC_CYCLE[15] TCC_RW_REQ[15] TCC_HIT[15] TCC_MISS[15] TCC_CYCLE[16] TCC_RW_REQ[16] TCC_HIT[16] TCC_MISS[16] TCC_CYCLE[17] TCC_RW_REQ[17] TCC_HIT[17] TCC_MISS[17] TCC_CYCLE[18] TCC_RW_REQ[18] TCC_HIT[18] TCC_MISS[18] TCC_CYCLE[19] TCC_RW_REQ[19] TCC_HIT[19] TCC_MISS[19] TCC_CYCLE[20] TCC_RW_REQ[20] TCC_HIT[20] TCC_MISS[20] TCC_CYCLE[21] TCC_RW_REQ[21] TCC_HIT[21] TCC_MISS[21] TCC_CYCLE[22] TCC_RW_REQ[22] TCC_HIT[22] TCC_MISS[22] TCC_CYCLE[23] TCC_RW_REQ[23] TCC_HIT[23] TCC_MISS[23] TCC_CYCLE[24] TCC_RW_REQ[24] TCC_HIT[24] TCC_MISS[24] TCC_CYCLE[25] TCC_RW_REQ[25] TCC_HIT[25] TCC_MISS[25] TCC_CYCLE[26] TCC_RW_REQ[26] TCC_HIT[26] TCC_MISS[26] TCC_CYCLE[27] TCC_RW_REQ[27] TCC_HIT[27] TCC_MISS[27] TCC_CYCLE[28] TCC_RW_REQ[28] TCC_HIT[28] TCC_MISS[28] TCC_CYCLE[29] TCC_RW_REQ[29] TCC_HIT[29] TCC_MISS[29] TCC_CYCLE[30] TCC_RW_REQ[30] TCC_HIT[30] TCC_MISS[30] TCC_CYCLE[31] TCC_RW_REQ[31] TCC_HIT[31] TCC_MISS[31] - - - -pmc: TCC_REQ[0] TCC_READ[0] TCC_WRITE[0] TCC_ATOMIC[0] TCC_REQ[1] TCC_READ[1] TCC_WRITE[1] TCC_ATOMIC[1] TCC_REQ[2] TCC_READ[2] TCC_WRITE[2] TCC_ATOMIC[2] TCC_REQ[3] TCC_READ[3] TCC_WRITE[3] TCC_ATOMIC[3] TCC_REQ[4] TCC_READ[4] TCC_WRITE[4] TCC_ATOMIC[4] TCC_REQ[5] TCC_READ[5] TCC_WRITE[5] TCC_ATOMIC[5] TCC_REQ[6] TCC_READ[6] TCC_WRITE[6] TCC_ATOMIC[6] TCC_REQ[7] TCC_READ[7] TCC_WRITE[7] TCC_ATOMIC[7] TCC_REQ[8] TCC_READ[8] TCC_WRITE[8] TCC_ATOMIC[8] TCC_REQ[9] TCC_READ[9] TCC_WRITE[9] TCC_ATOMIC[9] TCC_REQ[10] TCC_READ[10] TCC_WRITE[10] TCC_ATOMIC[10] TCC_REQ[11] TCC_READ[11] TCC_WRITE[11] TCC_ATOMIC[11] TCC_REQ[12] TCC_READ[12] TCC_WRITE[12] TCC_ATOMIC[12] TCC_REQ[13] TCC_READ[13] TCC_WRITE[13] TCC_ATOMIC[13] TCC_REQ[14] TCC_READ[14] TCC_WRITE[14] TCC_ATOMIC[14] TCC_REQ[15] TCC_READ[15] TCC_WRITE[15] TCC_ATOMIC[15] TCC_REQ[16] TCC_READ[16] TCC_WRITE[16] TCC_ATOMIC[16] TCC_REQ[17] TCC_READ[17] TCC_WRITE[17] TCC_ATOMIC[17] TCC_REQ[18] TCC_READ[18] TCC_WRITE[18] TCC_ATOMIC[18] TCC_REQ[19] TCC_READ[19] TCC_WRITE[19] TCC_ATOMIC[19] TCC_REQ[20] TCC_READ[20] TCC_WRITE[20] TCC_ATOMIC[20] TCC_REQ[21] TCC_READ[21] TCC_WRITE[21] TCC_ATOMIC[21] TCC_REQ[22] TCC_READ[22] TCC_WRITE[22] TCC_ATOMIC[22] TCC_REQ[23] TCC_READ[23] TCC_WRITE[23] TCC_ATOMIC[23] TCC_REQ[24] TCC_READ[24] TCC_WRITE[24] TCC_ATOMIC[24] TCC_REQ[25] TCC_READ[25] TCC_WRITE[25] TCC_ATOMIC[25] TCC_REQ[26] TCC_READ[26] TCC_WRITE[26] TCC_ATOMIC[26] TCC_REQ[27] TCC_READ[27] TCC_WRITE[27] TCC_ATOMIC[27] TCC_REQ[28] TCC_READ[28] TCC_WRITE[28] TCC_ATOMIC[28] TCC_REQ[29] TCC_READ[29] TCC_WRITE[29] TCC_ATOMIC[29] TCC_REQ[30] TCC_READ[30] TCC_WRITE[30] TCC_ATOMIC[30] TCC_REQ[31] TCC_READ[31] TCC_WRITE[31] TCC_ATOMIC[31] - - - -pmc: TCC_EA_RDREQ[0] TCC_EA_RDREQ_32B[0] TCC_EA_WRREQ[0] TCC_EA_WRREQ_64B[0] TCC_EA_RDREQ[1] TCC_EA_RDREQ_32B[1] TCC_EA_WRREQ[1] TCC_EA_WRREQ_64B[1] TCC_EA_RDREQ[2] TCC_EA_RDREQ_32B[2] TCC_EA_WRREQ[2] TCC_EA_WRREQ_64B[2] TCC_EA_RDREQ[3] TCC_EA_RDREQ_32B[3] TCC_EA_WRREQ[3] TCC_EA_WRREQ_64B[3] TCC_EA_RDREQ[4] TCC_EA_RDREQ_32B[4] TCC_EA_WRREQ[4] TCC_EA_WRREQ_64B[4] TCC_EA_RDREQ[5] TCC_EA_RDREQ_32B[5] TCC_EA_WRREQ[5] TCC_EA_WRREQ_64B[5] TCC_EA_RDREQ[6] TCC_EA_RDREQ_32B[6] TCC_EA_WRREQ[6] TCC_EA_WRREQ_64B[6] TCC_EA_RDREQ[7] TCC_EA_RDREQ_32B[7] TCC_EA_WRREQ[7] TCC_EA_WRREQ_64B[7] TCC_EA_RDREQ[8] TCC_EA_RDREQ_32B[8] TCC_EA_WRREQ[8] TCC_EA_WRREQ_64B[8] TCC_EA_RDREQ[9] TCC_EA_RDREQ_32B[9] TCC_EA_WRREQ[9] TCC_EA_WRREQ_64B[9] TCC_EA_RDREQ[10] TCC_EA_RDREQ_32B[10] TCC_EA_WRREQ[10] TCC_EA_WRREQ_64B[10] TCC_EA_RDREQ[11] TCC_EA_RDREQ_32B[11] TCC_EA_WRREQ[11] TCC_EA_WRREQ_64B[11] TCC_EA_RDREQ[12] TCC_EA_RDREQ_32B[12] TCC_EA_WRREQ[12] TCC_EA_WRREQ_64B[12] TCC_EA_RDREQ[13] TCC_EA_RDREQ_32B[13] TCC_EA_WRREQ[13] TCC_EA_WRREQ_64B[13] TCC_EA_RDREQ[14] TCC_EA_RDREQ_32B[14] TCC_EA_WRREQ[14] TCC_EA_WRREQ_64B[14] TCC_EA_RDREQ[15] TCC_EA_RDREQ_32B[15] TCC_EA_WRREQ[15] TCC_EA_WRREQ_64B[15] TCC_EA_RDREQ[16] TCC_EA_RDREQ_32B[16] TCC_EA_WRREQ[16] TCC_EA_WRREQ_64B[16] TCC_EA_RDREQ[17] TCC_EA_RDREQ_32B[17] TCC_EA_WRREQ[17] TCC_EA_WRREQ_64B[17] TCC_EA_RDREQ[18] TCC_EA_RDREQ_32B[18] TCC_EA_WRREQ[18] TCC_EA_WRREQ_64B[18] TCC_EA_RDREQ[19] TCC_EA_RDREQ_32B[19] TCC_EA_WRREQ[19] TCC_EA_WRREQ_64B[19] TCC_EA_RDREQ[20] TCC_EA_RDREQ_32B[20] TCC_EA_WRREQ[20] TCC_EA_WRREQ_64B[20] TCC_EA_RDREQ[21] TCC_EA_RDREQ_32B[21] TCC_EA_WRREQ[21] TCC_EA_WRREQ_64B[21] TCC_EA_RDREQ[22] TCC_EA_RDREQ_32B[22] TCC_EA_WRREQ[22] TCC_EA_WRREQ_64B[22] TCC_EA_RDREQ[23] TCC_EA_RDREQ_32B[23] TCC_EA_WRREQ[23] TCC_EA_WRREQ_64B[23] TCC_EA_RDREQ[24] TCC_EA_RDREQ_32B[24] TCC_EA_WRREQ[24] TCC_EA_WRREQ_64B[24] TCC_EA_RDREQ[25] TCC_EA_RDREQ_32B[25] TCC_EA_WRREQ[25] TCC_EA_WRREQ_64B[25] TCC_EA_RDREQ[26] TCC_EA_RDREQ_32B[26] TCC_EA_WRREQ[26] TCC_EA_WRREQ_64B[26] TCC_EA_RDREQ[27] TCC_EA_RDREQ_32B[27] TCC_EA_WRREQ[27] TCC_EA_WRREQ_64B[27] TCC_EA_RDREQ[28] TCC_EA_RDREQ_32B[28] TCC_EA_WRREQ[28] TCC_EA_WRREQ_64B[28] TCC_EA_RDREQ[29] TCC_EA_RDREQ_32B[29] TCC_EA_WRREQ[29] TCC_EA_WRREQ_64B[29] TCC_EA_RDREQ[30] TCC_EA_RDREQ_32B[30] TCC_EA_WRREQ[30] TCC_EA_WRREQ_64B[30] TCC_EA_RDREQ[31] TCC_EA_RDREQ_32B[31] TCC_EA_WRREQ[31] TCC_EA_WRREQ_64B[31] - - -pmc: TCC_EA_ATOMIC[0] TCC_EA_RDREQ_LEVEL[0] TCC_EA_WRREQ_LEVEL[0] TCC_EA_ATOMIC_LEVEL[0] TCC_EA_ATOMIC[1] TCC_EA_RDREQ_LEVEL[1] TCC_EA_WRREQ_LEVEL[1] TCC_EA_ATOMIC_LEVEL[1] TCC_EA_ATOMIC[2] TCC_EA_RDREQ_LEVEL[2] TCC_EA_WRREQ_LEVEL[2] TCC_EA_ATOMIC_LEVEL[2] TCC_EA_ATOMIC[3] TCC_EA_RDREQ_LEVEL[3] TCC_EA_WRREQ_LEVEL[3] TCC_EA_ATOMIC_LEVEL[3] TCC_EA_ATOMIC[4] TCC_EA_RDREQ_LEVEL[4] TCC_EA_WRREQ_LEVEL[4] TCC_EA_ATOMIC_LEVEL[4] TCC_EA_ATOMIC[5] TCC_EA_RDREQ_LEVEL[5] TCC_EA_WRREQ_LEVEL[5] TCC_EA_ATOMIC_LEVEL[5] TCC_EA_ATOMIC[6] TCC_EA_RDREQ_LEVEL[6] TCC_EA_WRREQ_LEVEL[6] TCC_EA_ATOMIC_LEVEL[6] TCC_EA_ATOMIC[7] TCC_EA_RDREQ_LEVEL[7] TCC_EA_WRREQ_LEVEL[7] TCC_EA_ATOMIC_LEVEL[7] TCC_EA_ATOMIC[8] TCC_EA_RDREQ_LEVEL[8] TCC_EA_WRREQ_LEVEL[8] TCC_EA_ATOMIC_LEVEL[8] TCC_EA_ATOMIC[9] TCC_EA_RDREQ_LEVEL[9] TCC_EA_WRREQ_LEVEL[9] TCC_EA_ATOMIC_LEVEL[9] TCC_EA_ATOMIC[10] TCC_EA_RDREQ_LEVEL[10] TCC_EA_WRREQ_LEVEL[10] TCC_EA_ATOMIC_LEVEL[10] TCC_EA_ATOMIC[11] TCC_EA_RDREQ_LEVEL[11] TCC_EA_WRREQ_LEVEL[11] TCC_EA_ATOMIC_LEVEL[11] TCC_EA_ATOMIC[12] TCC_EA_RDREQ_LEVEL[12] TCC_EA_WRREQ_LEVEL[12] TCC_EA_ATOMIC_LEVEL[12] TCC_EA_ATOMIC[13] TCC_EA_RDREQ_LEVEL[13] TCC_EA_WRREQ_LEVEL[13] TCC_EA_ATOMIC_LEVEL[13] TCC_EA_ATOMIC[14] TCC_EA_RDREQ_LEVEL[14] TCC_EA_WRREQ_LEVEL[14] TCC_EA_ATOMIC_LEVEL[14] TCC_EA_ATOMIC[15] TCC_EA_RDREQ_LEVEL[15] TCC_EA_WRREQ_LEVEL[15] TCC_EA_ATOMIC_LEVEL[15] TCC_EA_ATOMIC[16] TCC_EA_RDREQ_LEVEL[16] TCC_EA_WRREQ_LEVEL[16] TCC_EA_ATOMIC_LEVEL[16] TCC_EA_ATOMIC[17] TCC_EA_RDREQ_LEVEL[17] TCC_EA_WRREQ_LEVEL[17] TCC_EA_ATOMIC_LEVEL[17] TCC_EA_ATOMIC[18] TCC_EA_RDREQ_LEVEL[18] TCC_EA_WRREQ_LEVEL[18] TCC_EA_ATOMIC_LEVEL[18] TCC_EA_ATOMIC[19] TCC_EA_RDREQ_LEVEL[19] TCC_EA_WRREQ_LEVEL[19] TCC_EA_ATOMIC_LEVEL[19] TCC_EA_ATOMIC[20] TCC_EA_RDREQ_LEVEL[20] TCC_EA_WRREQ_LEVEL[20] TCC_EA_ATOMIC_LEVEL[20] TCC_EA_ATOMIC[21] TCC_EA_RDREQ_LEVEL[21] TCC_EA_WRREQ_LEVEL[21] TCC_EA_ATOMIC_LEVEL[21] TCC_EA_ATOMIC[22] TCC_EA_RDREQ_LEVEL[22] TCC_EA_WRREQ_LEVEL[22] TCC_EA_ATOMIC_LEVEL[22] TCC_EA_ATOMIC[23] TCC_EA_RDREQ_LEVEL[23] TCC_EA_WRREQ_LEVEL[23] TCC_EA_ATOMIC_LEVEL[23] TCC_EA_ATOMIC[24] TCC_EA_RDREQ_LEVEL[24] TCC_EA_WRREQ_LEVEL[24] TCC_EA_ATOMIC_LEVEL[24] TCC_EA_ATOMIC[25] TCC_EA_RDREQ_LEVEL[25] TCC_EA_WRREQ_LEVEL[25] TCC_EA_ATOMIC_LEVEL[25] TCC_EA_ATOMIC[26] TCC_EA_RDREQ_LEVEL[26] TCC_EA_WRREQ_LEVEL[26] TCC_EA_ATOMIC_LEVEL[26] TCC_EA_ATOMIC[27] TCC_EA_RDREQ_LEVEL[27] TCC_EA_WRREQ_LEVEL[27] TCC_EA_ATOMIC_LEVEL[27] TCC_EA_ATOMIC[28] TCC_EA_RDREQ_LEVEL[28] TCC_EA_WRREQ_LEVEL[28] TCC_EA_ATOMIC_LEVEL[28] TCC_EA_ATOMIC[29] TCC_EA_RDREQ_LEVEL[29] TCC_EA_WRREQ_LEVEL[29] TCC_EA_ATOMIC_LEVEL[29] TCC_EA_ATOMIC[30] TCC_EA_RDREQ_LEVEL[30] TCC_EA_WRREQ_LEVEL[30] TCC_EA_ATOMIC_LEVEL[30] TCC_EA_ATOMIC[31] TCC_EA_RDREQ_LEVEL[31] TCC_EA_WRREQ_LEVEL[31] TCC_EA_ATOMIC_LEVEL[31] - - - - -pmc: TCC_EA_RDREQ_IO_CREDIT_STALL[0] TCC_EA_RDREQ_GMI_CREDIT_STALL[0] TCC_EA_RDREQ_DRAM_CREDIT_STALL[0] TCC_EA_RDREQ_IO_CREDIT_STALL[1] TCC_EA_RDREQ_GMI_CREDIT_STALL[1] TCC_EA_RDREQ_DRAM_CREDIT_STALL[1] TCC_EA_RDREQ_IO_CREDIT_STALL[2] TCC_EA_RDREQ_GMI_CREDIT_STALL[2] TCC_EA_RDREQ_DRAM_CREDIT_STALL[2] TCC_EA_RDREQ_IO_CREDIT_STALL[3] TCC_EA_RDREQ_GMI_CREDIT_STALL[3] TCC_EA_RDREQ_DRAM_CREDIT_STALL[3] TCC_EA_RDREQ_IO_CREDIT_STALL[4] TCC_EA_RDREQ_GMI_CREDIT_STALL[4] TCC_EA_RDREQ_DRAM_CREDIT_STALL[4] TCC_EA_RDREQ_IO_CREDIT_STALL[5] TCC_EA_RDREQ_GMI_CREDIT_STALL[5] TCC_EA_RDREQ_DRAM_CREDIT_STALL[5] TCC_EA_RDREQ_IO_CREDIT_STALL[6] TCC_EA_RDREQ_GMI_CREDIT_STALL[6] TCC_EA_RDREQ_DRAM_CREDIT_STALL[6] TCC_EA_RDREQ_IO_CREDIT_STALL[7] TCC_EA_RDREQ_GMI_CREDIT_STALL[7] TCC_EA_RDREQ_DRAM_CREDIT_STALL[7] TCC_EA_RDREQ_IO_CREDIT_STALL[8] TCC_EA_RDREQ_GMI_CREDIT_STALL[8] TCC_EA_RDREQ_DRAM_CREDIT_STALL[8] TCC_EA_RDREQ_IO_CREDIT_STALL[9] TCC_EA_RDREQ_GMI_CREDIT_STALL[9] TCC_EA_RDREQ_DRAM_CREDIT_STALL[9] TCC_EA_RDREQ_IO_CREDIT_STALL[10] TCC_EA_RDREQ_GMI_CREDIT_STALL[10] TCC_EA_RDREQ_DRAM_CREDIT_STALL[10] TCC_EA_RDREQ_IO_CREDIT_STALL[11] TCC_EA_RDREQ_GMI_CREDIT_STALL[11] TCC_EA_RDREQ_DRAM_CREDIT_STALL[11] TCC_EA_RDREQ_IO_CREDIT_STALL[12] TCC_EA_RDREQ_GMI_CREDIT_STALL[12] TCC_EA_RDREQ_DRAM_CREDIT_STALL[12] TCC_EA_RDREQ_IO_CREDIT_STALL[13] TCC_EA_RDREQ_GMI_CREDIT_STALL[13] TCC_EA_RDREQ_DRAM_CREDIT_STALL[13] TCC_EA_RDREQ_IO_CREDIT_STALL[14] TCC_EA_RDREQ_GMI_CREDIT_STALL[14] TCC_EA_RDREQ_DRAM_CREDIT_STALL[14] TCC_EA_RDREQ_IO_CREDIT_STALL[15] TCC_EA_RDREQ_GMI_CREDIT_STALL[15] TCC_EA_RDREQ_DRAM_CREDIT_STALL[15] TCC_EA_RDREQ_IO_CREDIT_STALL[16] TCC_EA_RDREQ_GMI_CREDIT_STALL[16] TCC_EA_RDREQ_DRAM_CREDIT_STALL[16] TCC_EA_RDREQ_IO_CREDIT_STALL[17] TCC_EA_RDREQ_GMI_CREDIT_STALL[17] TCC_EA_RDREQ_DRAM_CREDIT_STALL[17] TCC_EA_RDREQ_IO_CREDIT_STALL[18] TCC_EA_RDREQ_GMI_CREDIT_STALL[18] TCC_EA_RDREQ_DRAM_CREDIT_STALL[18] TCC_EA_RDREQ_IO_CREDIT_STALL[19] TCC_EA_RDREQ_GMI_CREDIT_STALL[19] TCC_EA_RDREQ_DRAM_CREDIT_STALL[19] TCC_EA_RDREQ_IO_CREDIT_STALL[20] TCC_EA_RDREQ_GMI_CREDIT_STALL[20] TCC_EA_RDREQ_DRAM_CREDIT_STALL[20] TCC_EA_RDREQ_IO_CREDIT_STALL[21] TCC_EA_RDREQ_GMI_CREDIT_STALL[21] TCC_EA_RDREQ_DRAM_CREDIT_STALL[21] TCC_EA_RDREQ_IO_CREDIT_STALL[22] TCC_EA_RDREQ_GMI_CREDIT_STALL[22] TCC_EA_RDREQ_DRAM_CREDIT_STALL[22] TCC_EA_RDREQ_IO_CREDIT_STALL[23] TCC_EA_RDREQ_GMI_CREDIT_STALL[23] TCC_EA_RDREQ_DRAM_CREDIT_STALL[23] TCC_EA_RDREQ_IO_CREDIT_STALL[24] TCC_EA_RDREQ_GMI_CREDIT_STALL[24] TCC_EA_RDREQ_DRAM_CREDIT_STALL[24] TCC_EA_RDREQ_IO_CREDIT_STALL[25] TCC_EA_RDREQ_GMI_CREDIT_STALL[25] TCC_EA_RDREQ_DRAM_CREDIT_STALL[25] TCC_EA_RDREQ_IO_CREDIT_STALL[26] TCC_EA_RDREQ_GMI_CREDIT_STALL[26] TCC_EA_RDREQ_DRAM_CREDIT_STALL[26] TCC_EA_RDREQ_IO_CREDIT_STALL[27] TCC_EA_RDREQ_GMI_CREDIT_STALL[27] TCC_EA_RDREQ_DRAM_CREDIT_STALL[27] TCC_EA_RDREQ_IO_CREDIT_STALL[28] TCC_EA_RDREQ_GMI_CREDIT_STALL[28] TCC_EA_RDREQ_DRAM_CREDIT_STALL[28] TCC_EA_RDREQ_IO_CREDIT_STALL[29] TCC_EA_RDREQ_GMI_CREDIT_STALL[29] TCC_EA_RDREQ_DRAM_CREDIT_STALL[29] TCC_EA_RDREQ_IO_CREDIT_STALL[30] TCC_EA_RDREQ_GMI_CREDIT_STALL[30] TCC_EA_RDREQ_DRAM_CREDIT_STALL[30] TCC_EA_RDREQ_IO_CREDIT_STALL[31] TCC_EA_RDREQ_GMI_CREDIT_STALL[31] TCC_EA_RDREQ_DRAM_CREDIT_STALL[31] - - -pmc: TCC_EA_WRREQ_IO_CREDIT_STALL[0] TCC_EA_WRREQ_GMI_CREDIT_STALL[0] TCC_EA_WRREQ_DRAM_CREDIT_STALL[0] TCC_TOO_MANY_EA_WRREQS_STALL[0] TCC_EA_WRREQ_IO_CREDIT_STALL[1] TCC_EA_WRREQ_GMI_CREDIT_STALL[1] TCC_EA_WRREQ_DRAM_CREDIT_STALL[1] TCC_TOO_MANY_EA_WRREQS_STALL[1] TCC_EA_WRREQ_IO_CREDIT_STALL[2] TCC_EA_WRREQ_GMI_CREDIT_STALL[2] TCC_EA_WRREQ_DRAM_CREDIT_STALL[2] TCC_TOO_MANY_EA_WRREQS_STALL[2] TCC_EA_WRREQ_IO_CREDIT_STALL[3] TCC_EA_WRREQ_GMI_CREDIT_STALL[3] TCC_EA_WRREQ_DRAM_CREDIT_STALL[3] TCC_TOO_MANY_EA_WRREQS_STALL[3] TCC_EA_WRREQ_IO_CREDIT_STALL[4] TCC_EA_WRREQ_GMI_CREDIT_STALL[4] TCC_EA_WRREQ_DRAM_CREDIT_STALL[4] TCC_TOO_MANY_EA_WRREQS_STALL[4] TCC_EA_WRREQ_IO_CREDIT_STALL[5] TCC_EA_WRREQ_GMI_CREDIT_STALL[5] TCC_EA_WRREQ_DRAM_CREDIT_STALL[5] TCC_TOO_MANY_EA_WRREQS_STALL[5] TCC_EA_WRREQ_IO_CREDIT_STALL[6] TCC_EA_WRREQ_GMI_CREDIT_STALL[6] TCC_EA_WRREQ_DRAM_CREDIT_STALL[6] TCC_TOO_MANY_EA_WRREQS_STALL[6] TCC_EA_WRREQ_IO_CREDIT_STALL[7] TCC_EA_WRREQ_GMI_CREDIT_STALL[7] TCC_EA_WRREQ_DRAM_CREDIT_STALL[7] TCC_TOO_MANY_EA_WRREQS_STALL[7] TCC_EA_WRREQ_IO_CREDIT_STALL[8] TCC_EA_WRREQ_GMI_CREDIT_STALL[8] TCC_EA_WRREQ_DRAM_CREDIT_STALL[8] TCC_TOO_MANY_EA_WRREQS_STALL[8] TCC_EA_WRREQ_IO_CREDIT_STALL[9] TCC_EA_WRREQ_GMI_CREDIT_STALL[9] TCC_EA_WRREQ_DRAM_CREDIT_STALL[9] TCC_TOO_MANY_EA_WRREQS_STALL[9] TCC_EA_WRREQ_IO_CREDIT_STALL[10] TCC_EA_WRREQ_GMI_CREDIT_STALL[10] TCC_EA_WRREQ_DRAM_CREDIT_STALL[10] TCC_TOO_MANY_EA_WRREQS_STALL[10] TCC_EA_WRREQ_IO_CREDIT_STALL[11] TCC_EA_WRREQ_GMI_CREDIT_STALL[11] TCC_EA_WRREQ_DRAM_CREDIT_STALL[11] TCC_TOO_MANY_EA_WRREQS_STALL[11] TCC_EA_WRREQ_IO_CREDIT_STALL[12] TCC_EA_WRREQ_GMI_CREDIT_STALL[12] TCC_EA_WRREQ_DRAM_CREDIT_STALL[12] TCC_TOO_MANY_EA_WRREQS_STALL[12] TCC_EA_WRREQ_IO_CREDIT_STALL[13] TCC_EA_WRREQ_GMI_CREDIT_STALL[13] TCC_EA_WRREQ_DRAM_CREDIT_STALL[13] TCC_TOO_MANY_EA_WRREQS_STALL[13] TCC_EA_WRREQ_IO_CREDIT_STALL[14] TCC_EA_WRREQ_GMI_CREDIT_STALL[14] TCC_EA_WRREQ_DRAM_CREDIT_STALL[14] TCC_TOO_MANY_EA_WRREQS_STALL[14] TCC_EA_WRREQ_IO_CREDIT_STALL[15] TCC_EA_WRREQ_GMI_CREDIT_STALL[15] TCC_EA_WRREQ_DRAM_CREDIT_STALL[15] TCC_TOO_MANY_EA_WRREQS_STALL[15] TCC_EA_WRREQ_IO_CREDIT_STALL[16] TCC_EA_WRREQ_GMI_CREDIT_STALL[16] TCC_EA_WRREQ_DRAM_CREDIT_STALL[16] TCC_TOO_MANY_EA_WRREQS_STALL[16] TCC_EA_WRREQ_IO_CREDIT_STALL[17] TCC_EA_WRREQ_GMI_CREDIT_STALL[17] TCC_EA_WRREQ_DRAM_CREDIT_STALL[17] TCC_TOO_MANY_EA_WRREQS_STALL[17] TCC_EA_WRREQ_IO_CREDIT_STALL[18] TCC_EA_WRREQ_GMI_CREDIT_STALL[18] TCC_EA_WRREQ_DRAM_CREDIT_STALL[18] TCC_TOO_MANY_EA_WRREQS_STALL[18] TCC_EA_WRREQ_IO_CREDIT_STALL[19] TCC_EA_WRREQ_GMI_CREDIT_STALL[19] TCC_EA_WRREQ_DRAM_CREDIT_STALL[19] TCC_TOO_MANY_EA_WRREQS_STALL[19] TCC_EA_WRREQ_IO_CREDIT_STALL[20] TCC_EA_WRREQ_GMI_CREDIT_STALL[20] TCC_EA_WRREQ_DRAM_CREDIT_STALL[20] TCC_TOO_MANY_EA_WRREQS_STALL[20] TCC_EA_WRREQ_IO_CREDIT_STALL[21] TCC_EA_WRREQ_GMI_CREDIT_STALL[21] TCC_EA_WRREQ_DRAM_CREDIT_STALL[21] TCC_TOO_MANY_EA_WRREQS_STALL[21] TCC_EA_WRREQ_IO_CREDIT_STALL[22] TCC_EA_WRREQ_GMI_CREDIT_STALL[22] TCC_EA_WRREQ_DRAM_CREDIT_STALL[22] TCC_TOO_MANY_EA_WRREQS_STALL[22] TCC_EA_WRREQ_IO_CREDIT_STALL[23] TCC_EA_WRREQ_GMI_CREDIT_STALL[23] TCC_EA_WRREQ_DRAM_CREDIT_STALL[23] TCC_TOO_MANY_EA_WRREQS_STALL[23] TCC_EA_WRREQ_IO_CREDIT_STALL[24] TCC_EA_WRREQ_GMI_CREDIT_STALL[24] TCC_EA_WRREQ_DRAM_CREDIT_STALL[24] TCC_TOO_MANY_EA_WRREQS_STALL[24] TCC_EA_WRREQ_IO_CREDIT_STALL[25] TCC_EA_WRREQ_GMI_CREDIT_STALL[25] TCC_EA_WRREQ_DRAM_CREDIT_STALL[25] TCC_TOO_MANY_EA_WRREQS_STALL[25] TCC_EA_WRREQ_IO_CREDIT_STALL[26] TCC_EA_WRREQ_GMI_CREDIT_STALL[26] TCC_EA_WRREQ_DRAM_CREDIT_STALL[26] TCC_TOO_MANY_EA_WRREQS_STALL[26] TCC_EA_WRREQ_IO_CREDIT_STALL[27] TCC_EA_WRREQ_GMI_CREDIT_STALL[27] TCC_EA_WRREQ_DRAM_CREDIT_STALL[27] TCC_TOO_MANY_EA_WRREQS_STALL[27] TCC_EA_WRREQ_IO_CREDIT_STALL[28] TCC_EA_WRREQ_GMI_CREDIT_STALL[28] TCC_EA_WRREQ_DRAM_CREDIT_STALL[28] TCC_TOO_MANY_EA_WRREQS_STALL[28] TCC_EA_WRREQ_IO_CREDIT_STALL[29] TCC_EA_WRREQ_GMI_CREDIT_STALL[29] TCC_EA_WRREQ_DRAM_CREDIT_STALL[29] TCC_TOO_MANY_EA_WRREQS_STALL[29] TCC_EA_WRREQ_IO_CREDIT_STALL[30] TCC_EA_WRREQ_GMI_CREDIT_STALL[30] TCC_EA_WRREQ_DRAM_CREDIT_STALL[30] TCC_TOO_MANY_EA_WRREQS_STALL[30] TCC_EA_WRREQ_IO_CREDIT_STALL[31] TCC_EA_WRREQ_GMI_CREDIT_STALL[31] TCC_EA_WRREQ_DRAM_CREDIT_STALL[31] TCC_TOO_MANY_EA_WRREQS_STALL[31] - -gpu: -kernel: -range: diff --git a/src/rocprof_compute_soc/profile_configs/gfx90a/pmc_tcc_perf.txt b/src/rocprof_compute_soc/profile_configs/gfx90a/pmc_tcc_perf.txt deleted file mode 100644 index 3c7c32c7e2..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx90a/pmc_tcc_perf.txt +++ /dev/null @@ -1,17 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_CYCLES SQ_BUSY_CYCLES SQ_WAVES - -pmc: TCC_CYCLE_sum TCC_BUSY_sum TCC_PROBE_sum TCC_PROBE_ALL_sum -pmc: TCC_NC_REQ_sum TCC_UC_REQ_sum TCC_CC_REQ_sum TCC_RW_REQ_sum -pmc: TCC_REQ_sum TCC_STREAMING_REQ_sum TCC_HIT_sum TCC_MISS_sum -pmc: TCC_READ_sum TCC_WRITE_sum TCC_ATOMIC_sum TCC_WRITEBACK_sum -pmc: TCC_EA_WRREQ_sum TCC_EA_WRREQ_64B_sum TCC_EA_WR_UNCACHED_32B_sum TCC_EA_WRREQ_DRAM_sum -pmc: TCC_EA_WRREQ_STALL_sum -pmc: TCC_EA_RDREQ_sum TCC_EA_RDREQ_32B_sum TCC_EA_RD_UNCACHED_32B_sum TCC_EA_RDREQ_DRAM_sum -pmc: TCC_TAG_STALL_sum TCC_NORMAL_WRITEBACK_sum TCC_ALL_TC_OP_WB_WRITEBACK_sum TCC_NORMAL_EVICT_sum -pmc: TCC_ALL_TC_OP_INV_EVICT_sum TCC_TOO_MANY_EA_WRREQS_STALL_sum TCC_EA_ATOMIC_sum -pmc: TCC_EA_RDREQ_LEVEL_sum TCC_EA_WRREQ_LEVEL_sum TCC_EA_ATOMIC_LEVEL_sum - -gpu: -kernel: - -range: diff --git a/src/rocprof_compute_soc/profile_configs/gfx90a/pmc_tcp_perf.txt b/src/rocprof_compute_soc/profile_configs/gfx90a/pmc_tcp_perf.txt deleted file mode 100644 index 8efbce13ec..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx90a/pmc_tcp_perf.txt +++ /dev/null @@ -1,18 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_CYCLES SQ_BUSY_CYCLES SQ_BUSY_CU_CYCLES SQ_WAVES SQ_WAVE_CYCLES - -pmc: TCP_GATE_EN1_sum TCP_GATE_EN2_sum TCP_TD_TCP_STALL_CYCLES_sum TCP_TCR_TCP_STALL_CYCLES_sum -pmc: TCP_READ_TAGCONFLICT_STALL_CYCLES_sum TCP_WRITE_TAGCONFLICT_STALL_CYCLES_sum TCP_ATOMIC_TAGCONFLICT_STALL_CYCLES_sum TCP_TA_TCP_STATE_READ_sum -pmc: TCP_VOLATILE_sum TCP_TOTAL_ACCESSES_sum TCP_TOTAL_READ_sum TCP_TOTAL_WRITE_sum -pmc: TCP_TOTAL_ATOMIC_WITH_RET_sum TCP_TOTAL_ATOMIC_WITHOUT_RET_sum TCP_TOTAL_WRITEBACK_INVALIDATES_sum TCP_TOTAL_CACHE_ACCESSES_sum -pmc: TCP_UTCL1_TRANSLATION_MISS_sum TCP_UTCL1_TRANSLATION_HIT_sum TCP_UTCL1_PERMISSION_MISS_sum TCP_UTCL1_REQUEST_sum -pmc: TCP_TCP_LATENCY_sum TCP_TCC_READ_REQ_LATENCY_sum TCP_TCC_WRITE_REQ_LATENCY_sum TCP_TCC_READ_REQ_sum -pmc: TCP_TCC_WRITE_REQ_sum TCP_TCC_ATOMIC_WITH_RET_REQ_sum TCP_TCC_ATOMIC_WITHOUT_RET_REQ_sum TCP_TCC_NC_READ_REQ_sum -pmc: TCP_TCC_NC_WRITE_REQ_sum TCP_TCC_NC_ATOMIC_REQ_sum TCP_TCC_UC_READ_REQ_sum TCP_TCC_UC_WRITE_REQ_sum -pmc: TCP_TCC_UC_ATOMIC_REQ_sum TCP_TCC_CC_READ_REQ_sum TCP_TCC_CC_WRITE_REQ_sum TCP_TCC_CC_ATOMIC_REQ_sum -pmc: TCP_TCC_RW_READ_REQ_sum TCP_TCC_RW_WRITE_REQ_sum TCP_TCC_RW_ATOMIC_REQ_sum TCP_PENDING_STALL_CYCLES_sum - -#pmc: TCA_CYCLE_sum TCA_BUSY_sum - -gpu: -kernel: -range: diff --git a/src/rocprof_compute_soc/profile_configs/gfx90a/pmc_td_perf.txt b/src/rocprof_compute_soc/profile_configs/gfx90a/pmc_td_perf.txt deleted file mode 100644 index 56601912be..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx90a/pmc_td_perf.txt +++ /dev/null @@ -1,12 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_CYCLES SQ_BUSY_CYCLES SQ_BUSY_CU_CYCLES SQ_WAVES SQ_WAVE_CYCLES - -pmc: TD_TD_BUSY_sum TD_TC_STALL_sum -pmc: TD_SPI_STALL_sum TD_LOAD_WAVEFRONT_sum -pmc: TD_ATOMIC_WAVEFRONT_sum TD_STORE_WAVEFRONT_sum -pmc: TD_COALESCABLE_WAVEFRONT_sum - - - -gpu: -range: -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx940/pmc_cpc_perf.txt b/src/rocprof_compute_soc/profile_configs/gfx940/pmc_cpc_perf.txt deleted file mode 100644 index bd80137304..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx940/pmc_cpc_perf.txt +++ /dev/null @@ -1,13 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_CYCLES SQ_BUSY_CYCLES SQ_WAVES - -pmc: CPC_CPC_STAT_BUSY CPC_CPC_STAT_IDLE -pmc: CPC_CPC_TCIU_BUSY CPC_CPC_TCIU_IDLE -pmc: CPC_CPC_STAT_STALL CPC_UTCL1_STALL_ON_TRANSLATION -pmc: CPC_CPC_UTCL2IU_BUSY CPC_CPC_UTCL2IU_IDLE -pmc: CPC_CPC_UTCL2IU_STALL CPC_ME1_BUSY_FOR_PACKET_DECODE -pmc: CPC_ME1_DC0_SPI_BUSY - -range: -gpu: - -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx940/pmc_cpf_perf.txt b/src/rocprof_compute_soc/profile_configs/gfx940/pmc_cpf_perf.txt deleted file mode 100644 index e8119fff56..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx940/pmc_cpf_perf.txt +++ /dev/null @@ -1,11 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_CYCLES SQ_BUSY_CYCLES SQ_WAVES - -pmc: CPF_CPF_STAT_BUSY CPF_CPF_STAT_STALL -pmc: CPF_CPF_TCIU_BUSY CPF_CPF_TCIU_STALL -pmc: CPF_CPF_STAT_IDLE CPF_CPF_TCIU_IDLE -pmc: CPF_CMP_UTCL1_STALL_ON_TRANSLATION - -range: -gpu: - -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx940/pmc_spi_perf.txt b/src/rocprof_compute_soc/profile_configs/gfx940/pmc_spi_perf.txt deleted file mode 100644 index 4eb0d22656..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx940/pmc_spi_perf.txt +++ /dev/null @@ -1,10 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_CYCLES SQ_BUSY_CYCLES SQ_WAVES GRBM_SPI_BUSY - -pmc: SPI_CSN_WINDOW_VALID SPI_CSN_BUSY SPI_CSN_NUM_THREADGROUPS SPI_CSN_WAVE SPI_RA_REQ_NO_ALLOC SPI_RA_REQ_NO_ALLOC_CSN -pmc: SPI_RA_RES_STALL_CSN SPI_RA_TMP_STALL_CSN SPI_RA_WAVE_SIMD_FULL_CSN SPI_RA_VGPR_SIMD_FULL_CSN SPI_RA_SGPR_SIMD_FULL_CSN SPI_RA_LDS_CU_FULL_CSN -pmc: SPI_RA_BAR_CU_FULL_CSN SPI_RA_TGLIM_CU_FULL_CSN SPI_RA_WVLIM_STALL_CSN SPI_SWC_CSC_WR SPI_VWC_CSC_WR SPI_RA_BULKY_CU_FULL_CSN - -range: -gpu: - -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx940/pmc_sq_perf1.txt b/src/rocprof_compute_soc/profile_configs/gfx940/pmc_sq_perf1.txt deleted file mode 100644 index b007f2876e..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx940/pmc_sq_perf1.txt +++ /dev/null @@ -1,33 +0,0 @@ -#SQ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_INSTS_VALU_CVT SQ_INSTS_VMEM_WR SQ_INSTS_VMEM_RD SQ_INSTS_VMEM SQ_INSTS_SALU SQ_INSTS_VSKIPPED -pmc: SQ_INSTS SQ_INSTS_VALU SQ_INSTS_VALU_ADD_F16 SQ_INSTS_VALU_MUL_F16 SQ_INSTS_VALU_FMA_F16 SQ_INSTS_VALU_TRANS_F16 SQ_INSTS_VALU_ADD_F32 SQ_INSTS_VALU_MUL_F32 -pmc: SQ_INSTS_VALU_FMA_F32 SQ_INSTS_VALU_TRANS_F32 SQ_INSTS_VALU_ADD_F64 SQ_INSTS_VALU_MUL_F64 SQ_INSTS_VALU_FMA_F64 SQ_INSTS_VALU_TRANS_F64 SQ_INSTS_VALU_INT32 SQ_INSTS_VALU_INT64 -pmc: SQ_INSTS_SMEM SQ_INSTS_FLAT SQ_INSTS_LDS SQ_INSTS_GDS SQ_INSTS_EXP_GDS SQ_INSTS_BRANCH SQ_INSTS_SENDMSG -pmc: SQ_WAVE_CYCLES SQ_WAIT_ANY SQ_WAIT_INST_ANY SQ_ACTIVE_INST_ANY SQ_CYCLES SQ_BUSY_CYCLES SQ_BUSY_CU_CYCLES -pmc: SQ_ACTIVE_INST_VMEM SQ_ACTIVE_INST_LDS SQ_ACTIVE_INST_VALU SQ_ACTIVE_INST_SCA SQ_ACTIVE_INST_EXP_GDS SQ_ACTIVE_INST_MISC SQ_ACTIVE_INST_FLAT SQ_INST_CYCLES_VMEM_WR -pmc: SQ_INST_CYCLES_VMEM_RD SQ_INST_CYCLES_SMEM SQ_INST_CYCLES_SALU SQ_THREAD_CYCLES_VALU SQ_IFETCH SQ_LDS_BANK_CONFLICT SQ_LDS_ADDR_CONFLICT SQ_LDS_UNALIGNED_STALL -pmc: SQ_WAVES SQ_WAVES_EQ_64 SQ_WAVES_LT_64 SQ_WAVES_LT_48 SQ_WAVES_LT_32 SQ_WAVES_LT_16 SQ_ITEMS SQ_INSTS_VSKIPPED -pmc: SQ_LDS_MEM_VIOLATIONS SQ_LDS_ATOMIC_RETURN SQ_LDS_IDX_ACTIVE SQ_WAVES_RESTORED SQ_WAVES_SAVED SQ_INSTS_SMEM_NORM - - -#SQ:MI200 -pmc: SQ_INSTS_MFMA SQ_INSTS_VALU_MFMA_I8 SQ_INSTS_VALU_MFMA_F16 SQ_INSTS_VALU_MFMA_BF16 SQ_INSTS_VALU_MFMA_F32 SQ_INSTS_VALU_MFMA_F64 SQ_VALU_MFMA_BUSY_CYCLES -pmc: SQ_INSTS_VALU_MFMA_MOPS_I8 SQ_INSTS_VALU_MFMA_MOPS_F16 SQ_INSTS_VALU_MFMA_MOPS_BF16 SQ_INSTS_VALU_MFMA_MOPS_F32 SQ_INSTS_VALU_MFMA_MOPS_F64 - -#SQ:MI300 -pmc: SQ_INSTS_VALU_MFMA_F8 -pmc: SQ_INSTS_VALU_MFMA_MOPS_F8 - -#SQC -pmc: SQC_TC_INST_REQ SQC_TC_DATA_READ_REQ SQC_TC_DATA_WRITE_REQ SQC_TC_DATA_ATOMIC_REQ SQC_TC_STALL SQC_TC_REQ SQC_DCACHE_REQ_READ_16 -pmc: SQC_ICACHE_REQ SQC_ICACHE_HITS SQC_ICACHE_MISSES SQC_ICACHE_MISSES_DUPLICATE SQC_DCACHE_INPUT_VALID_READYB SQC_DCACHE_ATOMIC SQC_DCACHE_REQ_READ_8 -pmc: SQC_DCACHE_REQ SQC_DCACHE_HITS SQC_DCACHE_MISSES SQC_DCACHE_MISSES_DUPLICATE SQC_DCACHE_REQ_READ_1 SQC_DCACHE_REQ_READ_2 SQC_DCACHE_REQ_READ_4 - - - -######################################## -# Filtering -######################################## -range: -gpu: -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx940/pmc_sq_perf2.txt b/src/rocprof_compute_soc/profile_configs/gfx940/pmc_sq_perf2.txt deleted file mode 100644 index 7175c0326c..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx940/pmc_sq_perf2.txt +++ /dev/null @@ -1,10 +0,0 @@ -################################################# -# VMEM latency -################################################# -pmc: SQ_INSTS_VMEM SQ_INST_LEVEL_VMEM SQ_ACCUM_PREV_HIRES - - -range: -gpu: - -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx940/pmc_sq_perf3.txt b/src/rocprof_compute_soc/profile_configs/gfx940/pmc_sq_perf3.txt deleted file mode 100644 index 8d36311af0..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx940/pmc_sq_perf3.txt +++ /dev/null @@ -1,11 +0,0 @@ -################################################# -# SMEM latency -################################################# -pmc: SQ_INSTS_SMEM SQ_INST_LEVEL_SMEM SQ_ACCUM_PREV_HIRES - - -range: -gpu: - - -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx940/pmc_sq_perf4.txt b/src/rocprof_compute_soc/profile_configs/gfx940/pmc_sq_perf4.txt deleted file mode 100644 index f8f29b4f54..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx940/pmc_sq_perf4.txt +++ /dev/null @@ -1,8 +0,0 @@ -################################################# -# ifetch latency -################################################# -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_WAVES SQ_IFETCH SQ_IFETCH_LEVEL SQ_ACCUM_PREV_HIRES - -range: -gpu: -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx940/pmc_sq_perf6.txt b/src/rocprof_compute_soc/profile_configs/gfx940/pmc_sq_perf6.txt deleted file mode 100644 index f9144576eb..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx940/pmc_sq_perf6.txt +++ /dev/null @@ -1,9 +0,0 @@ -################################################# -# LDS latency -################################################# -pmc: SQ_INSTS_LDS SQ_INST_LEVEL_LDS SQ_ACCUM_PREV_HIRES - -range: -gpu: - -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx940/pmc_sq_perf8.txt b/src/rocprof_compute_soc/profile_configs/gfx940/pmc_sq_perf8.txt deleted file mode 100644 index cb29ad2ce1..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx940/pmc_sq_perf8.txt +++ /dev/null @@ -1,6 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE CPC_ME1_BUSY_FOR_PACKET_DECODE SQ_CYCLES SQ_WAVES SQ_WAVE_CYCLES SQ_BUSY_CYCLES SQ_LEVEL_WAVES SQ_ACCUM_PREV_HIRES - -gpu: -range: - -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx940/pmc_sqc_perf1.txt b/src/rocprof_compute_soc/profile_configs/gfx940/pmc_sqc_perf1.txt deleted file mode 100644 index e602d35c1f..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx940/pmc_sqc_perf1.txt +++ /dev/null @@ -1,11 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_CYCLES SQ_BUSY_CYCLES SQ_BUSY_CU_CYCLES SQ_WAVES SQ_WAVE_CYCLES - - -pmc: SQC_TC_INST_REQ SQC_TC_DATA_READ_REQ SQC_TC_DATA_WRITE_REQ SQC_TC_DATA_ATOMIC_REQ SQC_TC_STALL SQC_TC_REQ SQC_DCACHE_REQ_READ_16 -pmc: SQC_ICACHE_REQ SQC_ICACHE_HITS SQC_ICACHE_MISSES SQC_ICACHE_MISSES_DUPLICATE SQC_DCACHE_INPUT_VALID_READYB SQC_DCACHE_ATOMIC SQC_DCACHE_REQ_READ_8 -pmc: SQC_DCACHE_REQ SQC_DCACHE_HITS SQC_DCACHE_MISSES SQC_DCACHE_MISSES_DUPLICATE SQC_DCACHE_REQ_READ_1 SQC_DCACHE_REQ_READ_2 SQC_DCACHE_REQ_READ_4 - -range: -gpu: - -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx940/pmc_ta_perf.txt b/src/rocprof_compute_soc/profile_configs/gfx940/pmc_ta_perf.txt deleted file mode 100644 index be544c627b..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx940/pmc_ta_perf.txt +++ /dev/null @@ -1,25 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_CYCLES SQ_BUSY_CYCLES SQ_BUSY_CU_CYCLES SQ_WAVES SQ_WAVE_CYCLES - -pmc: TA_TA_BUSY_sum - -# buffer access -pmc: TA_BUFFER_WAVEFRONTS_sum TA_BUFFER_READ_WAVEFRONTS_sum -pmc: TA_BUFFER_WRITE_WAVEFRONTS_sum TA_BUFFER_ATOMIC_WAVEFRONTS_sum -pmc: TA_BUFFER_TOTAL_CYCLES_sum -pmc: TA_BUFFER_COALESCED_READ_CYCLES_sum TA_BUFFER_COALESCED_WRITE_CYCLES_sum - - -# stalls -pmc: TA_ADDR_STALLED_BY_TC_CYCLES_sum TA_TOTAL_WAVEFRONTS_sum -pmc: TA_ADDR_STALLED_BY_TD_CYCLES_sum TA_DATA_STALLED_BY_TC_CYCLES_sum - -# flat accesses -pmc: TA_FLAT_WAVEFRONTS_sum TA_FLAT_READ_WAVEFRONTS_sum -pmc: TA_FLAT_WRITE_WAVEFRONTS_sum TA_FLAT_ATOMIC_WAVEFRONTS_sum - - -range: - -gpu: - -kernel: diff --git a/src/rocprof_compute_soc/profile_configs/gfx940/pmc_tcc2_perf.txt b/src/rocprof_compute_soc/profile_configs/gfx940/pmc_tcc2_perf.txt deleted file mode 100644 index 67fe878348..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx940/pmc_tcc2_perf.txt +++ /dev/null @@ -1,22 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_CYCLES SQ_BUSY_CYCLES SQ_WAVES - -pmc: TCC_CYCLE[0] TCC_RW_REQ[0] TCC_HIT[0] TCC_MISS[0] TCC_CYCLE[1] TCC_RW_REQ[1] TCC_HIT[1] TCC_MISS[1] TCC_CYCLE[2] TCC_RW_REQ[2] TCC_HIT[2] TCC_MISS[2] TCC_CYCLE[3] TCC_RW_REQ[3] TCC_HIT[3] TCC_MISS[3] TCC_CYCLE[4] TCC_RW_REQ[4] TCC_HIT[4] TCC_MISS[4] TCC_CYCLE[5] TCC_RW_REQ[5] TCC_HIT[5] TCC_MISS[5] TCC_CYCLE[6] TCC_RW_REQ[6] TCC_HIT[6] TCC_MISS[6] TCC_CYCLE[7] TCC_RW_REQ[7] TCC_HIT[7] TCC_MISS[7] TCC_CYCLE[8] TCC_RW_REQ[8] TCC_HIT[8] TCC_MISS[8] TCC_CYCLE[9] TCC_RW_REQ[9] TCC_HIT[9] TCC_MISS[9] TCC_CYCLE[10] TCC_RW_REQ[10] TCC_HIT[10] TCC_MISS[10] TCC_CYCLE[11] TCC_RW_REQ[11] TCC_HIT[11] TCC_MISS[11] TCC_CYCLE[12] TCC_RW_REQ[12] TCC_HIT[12] TCC_MISS[12] TCC_CYCLE[13] TCC_RW_REQ[13] TCC_HIT[13] TCC_MISS[13] TCC_CYCLE[14] TCC_RW_REQ[14] TCC_HIT[14] TCC_MISS[14] TCC_CYCLE[15] TCC_RW_REQ[15] TCC_HIT[15] TCC_MISS[15] - - - -pmc: TCC_REQ[0] TCC_READ[0] TCC_WRITE[0] TCC_ATOMIC[0] TCC_REQ[1] TCC_READ[1] TCC_WRITE[1] TCC_ATOMIC[1] TCC_REQ[2] TCC_READ[2] TCC_WRITE[2] TCC_ATOMIC[2] TCC_REQ[3] TCC_READ[3] TCC_WRITE[3] TCC_ATOMIC[3] TCC_REQ[4] TCC_READ[4] TCC_WRITE[4] TCC_ATOMIC[4] TCC_REQ[5] TCC_READ[5] TCC_WRITE[5] TCC_ATOMIC[5] TCC_REQ[6] TCC_READ[6] TCC_WRITE[6] TCC_ATOMIC[6] TCC_REQ[7] TCC_READ[7] TCC_WRITE[7] TCC_ATOMIC[7] TCC_REQ[8] TCC_READ[8] TCC_WRITE[8] TCC_ATOMIC[8] TCC_REQ[9] TCC_READ[9] TCC_WRITE[9] TCC_ATOMIC[9] TCC_REQ[10] TCC_READ[10] TCC_WRITE[10] TCC_ATOMIC[10] TCC_REQ[11] TCC_READ[11] TCC_WRITE[11] TCC_ATOMIC[11] TCC_REQ[12] TCC_READ[12] TCC_WRITE[12] TCC_ATOMIC[12] TCC_REQ[13] TCC_READ[13] TCC_WRITE[13] TCC_ATOMIC[13] TCC_REQ[14] TCC_READ[14] TCC_WRITE[14] TCC_ATOMIC[14] TCC_REQ[15] TCC_READ[15] TCC_WRITE[15] TCC_ATOMIC[15] - - - -pmc: TCC_EA0_RDREQ[0] TCC_EA0_RDREQ_32B[0] TCC_EA0_WRREQ[0] TCC_EA0_WRREQ_64B[0] TCC_EA0_RDREQ[1] TCC_EA0_RDREQ_32B[1] TCC_EA0_WRREQ[1] TCC_EA0_WRREQ_64B[1] TCC_EA0_RDREQ[2] TCC_EA0_RDREQ_32B[2] TCC_EA0_WRREQ[2] TCC_EA0_WRREQ_64B[2] TCC_EA0_RDREQ[3] TCC_EA0_RDREQ_32B[3] TCC_EA0_WRREQ[3] TCC_EA0_WRREQ_64B[3] TCC_EA0_RDREQ[4] TCC_EA0_RDREQ_32B[4] TCC_EA0_WRREQ[4] TCC_EA0_WRREQ_64B[4] TCC_EA0_RDREQ[5] TCC_EA0_RDREQ_32B[5] TCC_EA0_WRREQ[5] TCC_EA0_WRREQ_64B[5] TCC_EA0_RDREQ[6] TCC_EA0_RDREQ_32B[6] TCC_EA0_WRREQ[6] TCC_EA0_WRREQ_64B[6] TCC_EA0_RDREQ[7] TCC_EA0_RDREQ_32B[7] TCC_EA0_WRREQ[7] TCC_EA0_WRREQ_64B[7] TCC_EA0_RDREQ[8] TCC_EA0_RDREQ_32B[8] TCC_EA0_WRREQ[8] TCC_EA0_WRREQ_64B[8] TCC_EA0_RDREQ[9] TCC_EA0_RDREQ_32B[9] TCC_EA0_WRREQ[9] TCC_EA0_WRREQ_64B[9] TCC_EA0_RDREQ[10] TCC_EA0_RDREQ_32B[10] TCC_EA0_WRREQ[10] TCC_EA0_WRREQ_64B[10] TCC_EA0_RDREQ[11] TCC_EA0_RDREQ_32B[11] TCC_EA0_WRREQ[11] TCC_EA0_WRREQ_64B[11] TCC_EA0_RDREQ[12] TCC_EA0_RDREQ_32B[12] TCC_EA0_WRREQ[12] TCC_EA0_WRREQ_64B[12] TCC_EA0_RDREQ[13] TCC_EA0_RDREQ_32B[13] TCC_EA0_WRREQ[13] TCC_EA0_WRREQ_64B[13] TCC_EA0_RDREQ[14] TCC_EA0_RDREQ_32B[14] TCC_EA0_WRREQ[14] TCC_EA0_WRREQ_64B[14] TCC_EA0_RDREQ[15] TCC_EA0_RDREQ_32B[15] TCC_EA0_WRREQ[15] TCC_EA0_WRREQ_64B[15] - - -pmc: TCC_EA0_ATOMIC[0] TCC_EA0_RDREQ_LEVEL[0] TCC_EA0_WRREQ_LEVEL[0] TCC_EA0_ATOMIC_LEVEL[0] TCC_EA0_ATOMIC[1] TCC_EA0_RDREQ_LEVEL[1] TCC_EA0_WRREQ_LEVEL[1] TCC_EA0_ATOMIC_LEVEL[1] TCC_EA0_ATOMIC[2] TCC_EA0_RDREQ_LEVEL[2] TCC_EA0_WRREQ_LEVEL[2] TCC_EA0_ATOMIC_LEVEL[2] TCC_EA0_ATOMIC[3] TCC_EA0_RDREQ_LEVEL[3] TCC_EA0_WRREQ_LEVEL[3] TCC_EA0_ATOMIC_LEVEL[3] TCC_EA0_ATOMIC[4] TCC_EA0_RDREQ_LEVEL[4] TCC_EA0_WRREQ_LEVEL[4] TCC_EA0_ATOMIC_LEVEL[4] TCC_EA0_ATOMIC[5] TCC_EA0_RDREQ_LEVEL[5] TCC_EA0_WRREQ_LEVEL[5] TCC_EA0_ATOMIC_LEVEL[5] TCC_EA0_ATOMIC[6] TCC_EA0_RDREQ_LEVEL[6] TCC_EA0_WRREQ_LEVEL[6] TCC_EA0_ATOMIC_LEVEL[6] TCC_EA0_ATOMIC[7] TCC_EA0_RDREQ_LEVEL[7] TCC_EA0_WRREQ_LEVEL[7] TCC_EA0_ATOMIC_LEVEL[7] TCC_EA0_ATOMIC[8] TCC_EA0_RDREQ_LEVEL[8] TCC_EA0_WRREQ_LEVEL[8] TCC_EA0_ATOMIC_LEVEL[8] TCC_EA0_ATOMIC[9] TCC_EA0_RDREQ_LEVEL[9] TCC_EA0_WRREQ_LEVEL[9] TCC_EA0_ATOMIC_LEVEL[9] TCC_EA0_ATOMIC[10] TCC_EA0_RDREQ_LEVEL[10] TCC_EA0_WRREQ_LEVEL[10] TCC_EA0_ATOMIC_LEVEL[10] TCC_EA0_ATOMIC[11] TCC_EA0_RDREQ_LEVEL[11] TCC_EA0_WRREQ_LEVEL[11] TCC_EA0_ATOMIC_LEVEL[11] TCC_EA0_ATOMIC[12] TCC_EA0_RDREQ_LEVEL[12] TCC_EA0_WRREQ_LEVEL[12] TCC_EA0_ATOMIC_LEVEL[12] TCC_EA0_ATOMIC[13] TCC_EA0_RDREQ_LEVEL[13] TCC_EA0_WRREQ_LEVEL[13] TCC_EA0_ATOMIC_LEVEL[13] TCC_EA0_ATOMIC[14] TCC_EA0_RDREQ_LEVEL[14] TCC_EA0_WRREQ_LEVEL[14] TCC_EA0_ATOMIC_LEVEL[14] TCC_EA0_ATOMIC[15] TCC_EA0_RDREQ_LEVEL[15] TCC_EA0_WRREQ_LEVEL[15] TCC_EA0_ATOMIC_LEVEL[15] - -pmc: TCC_TAG_STALL[0] TCC_BUBBLE[0] TCC_TOO_MANY_EA_WRREQS_STALL[0] TCC_TAG_STALL[1] TCC_BUBBLE[1] TCC_TOO_MANY_EA_WRREQS_STALL[1] TCC_TAG_STALL[2] TCC_BUBBLE[2] TCC_TOO_MANY_EA_WRREQS_STALL[2] TCC_TAG_STALL[3] TCC_BUBBLE[3] TCC_TOO_MANY_EA_WRREQS_STALL[3] TCC_TAG_STALL[4] TCC_BUBBLE[4] TCC_TOO_MANY_EA_WRREQS_STALL[4] TCC_TAG_STALL[5] TCC_BUBBLE[5] TCC_TOO_MANY_EA_WRREQS_STALL[5] TCC_TAG_STALL[6] TCC_BUBBLE[6] TCC_TOO_MANY_EA_WRREQS_STALL[6] TCC_TAG_STALL[7] TCC_BUBBLE[7] TCC_TOO_MANY_EA_WRREQS_STALL[7] TCC_TAG_STALL[8] TCC_BUBBLE[8] TCC_TOO_MANY_EA_WRREQS_STALL[8] TCC_TAG_STALL[9] TCC_BUBBLE[9] TCC_TOO_MANY_EA_WRREQS_STALL[9] TCC_TAG_STALL[10] TCC_BUBBLE[10] TCC_TOO_MANY_EA_WRREQS_STALL[10] TCC_TAG_STALL[11] TCC_BUBBLE[11] TCC_TOO_MANY_EA_WRREQS_STALL[11] TCC_TAG_STALL[12] TCC_BUBBLE[12] TCC_TOO_MANY_EA_WRREQS_STALL[12] TCC_TAG_STALL[13] TCC_BUBBLE[13] TCC_TOO_MANY_EA_WRREQS_STALL[13] TCC_TAG_STALL[14] TCC_BUBBLE[14] TCC_TOO_MANY_EA_WRREQS_STALL[14] TCC_TAG_STALL[15] TCC_BUBBLE[15] TCC_TOO_MANY_EA_WRREQS_STALL[15] - - - -gpu: -kernel: -range: diff --git a/src/rocprof_compute_soc/profile_configs/gfx940/pmc_tcc_perf.txt b/src/rocprof_compute_soc/profile_configs/gfx940/pmc_tcc_perf.txt deleted file mode 100644 index 390f04cd1c..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx940/pmc_tcc_perf.txt +++ /dev/null @@ -1,17 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_CYCLES SQ_BUSY_CYCLES SQ_WAVES - -pmc: TCC_CYCLE_sum TCC_BUSY_sum TCC_PROBE_sum TCC_PROBE_ALL_sum -pmc: TCC_NC_REQ_sum TCC_UC_REQ_sum TCC_CC_REQ_sum TCC_RW_REQ_sum -pmc: TCC_REQ_sum TCC_STREAMING_REQ_sum TCC_HIT_sum TCC_MISS_sum -pmc: TCC_READ_sum TCC_WRITE_sum TCC_ATOMIC_sum TCC_WRITEBACK_sum -pmc: TCC_EA0_WRREQ_sum TCC_EA0_WRREQ_64B_sum TCC_EA0_WR_UNCACHED_32B_sum TCC_EA0_WRREQ_DRAM_sum -pmc: TCC_EA0_RDREQ_sum TCC_EA0_RDREQ_32B_sum TCC_BUBBLE_sum -pmc: TCC_EA0_RD_UNCACHED_32B_sum TCC_EA0_RDREQ_DRAM_sum -pmc: TCC_TAG_STALL_sum TCC_NORMAL_WRITEBACK_sum TCC_ALL_TC_OP_WB_WRITEBACK_sum TCC_NORMAL_EVICT_sum -pmc: TCC_ALL_TC_OP_INV_EVICT_sum TCC_TOO_MANY_EA_WRREQS_STALL_sum TCC_EA0_ATOMIC_sum -pmc: TCC_EA0_RDREQ_LEVEL_sum TCC_EA0_WRREQ_LEVEL_sum TCC_EA0_ATOMIC_LEVEL_sum TCC_EA0_WRREQ_STALL_sum - -gpu: -kernel: - -range: diff --git a/src/rocprof_compute_soc/profile_configs/gfx940/pmc_tcp_perf.txt b/src/rocprof_compute_soc/profile_configs/gfx940/pmc_tcp_perf.txt deleted file mode 100644 index 432c69f771..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx940/pmc_tcp_perf.txt +++ /dev/null @@ -1,18 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_CYCLES SQ_BUSY_CYCLES SQ_BUSY_CU_CYCLES SQ_WAVES SQ_WAVE_CYCLES - -pmc: TCP_GATE_EN1_sum TCP_GATE_EN2_sum TCP_TD_TCP_STALL_CYCLES_sum TCP_TCR_TCP_STALL_CYCLES_sum -pmc: TCP_READ_TAGCONFLICT_STALL_CYCLES_sum TCP_WRITE_TAGCONFLICT_STALL_CYCLES_sum TCP_ATOMIC_TAGCONFLICT_STALL_CYCLES_sum TCP_TA_TCP_STATE_READ_sum -pmc: TCP_VOLATILE_sum TCP_TOTAL_ACCESSES_sum TCP_TOTAL_READ_sum TCP_TOTAL_WRITE_sum -pmc: TCP_TOTAL_ATOMIC_WITH_RET_sum TCP_TOTAL_ATOMIC_WITHOUT_RET_sum TCP_TOTAL_WRITEBACK_INVALIDATES_sum TCP_TOTAL_CACHE_ACCESSES_sum -pmc: TCP_UTCL1_TRANSLATION_MISS_sum TCP_UTCL1_TRANSLATION_HIT_sum TCP_UTCL1_PERMISSION_MISS_sum TCP_UTCL1_REQUEST_sum -pmc: TCP_TCC_READ_REQ_sum -pmc: TCP_TCC_WRITE_REQ_sum TCP_TCC_ATOMIC_WITH_RET_REQ_sum TCP_TCC_ATOMIC_WITHOUT_RET_REQ_sum TCP_TCC_NC_READ_REQ_sum -pmc: TCP_TCC_NC_WRITE_REQ_sum TCP_TCC_NC_ATOMIC_REQ_sum TCP_TCC_UC_READ_REQ_sum TCP_TCC_UC_WRITE_REQ_sum -pmc: TCP_TCC_UC_ATOMIC_REQ_sum TCP_TCC_CC_READ_REQ_sum TCP_TCC_CC_WRITE_REQ_sum TCP_TCC_CC_ATOMIC_REQ_sum -pmc: TCP_TCC_RW_READ_REQ_sum TCP_TCC_RW_WRITE_REQ_sum TCP_TCC_RW_ATOMIC_REQ_sum TCP_PENDING_STALL_CYCLES_sum - -#pmc: TCA_CYCLE_sum TCA_BUSY_sum - -gpu: -kernel: -range: diff --git a/src/rocprof_compute_soc/profile_configs/gfx940/pmc_td_perf.txt b/src/rocprof_compute_soc/profile_configs/gfx940/pmc_td_perf.txt deleted file mode 100644 index 56601912be..0000000000 --- a/src/rocprof_compute_soc/profile_configs/gfx940/pmc_td_perf.txt +++ /dev/null @@ -1,12 +0,0 @@ -pmc: GRBM_COUNT GRBM_GUI_ACTIVE SQ_CYCLES SQ_BUSY_CYCLES SQ_BUSY_CU_CYCLES SQ_WAVES SQ_WAVE_CYCLES - -pmc: TD_TD_BUSY_sum TD_TC_STALL_sum -pmc: TD_SPI_STALL_sum TD_LOAD_WAVEFRONT_sum -pmc: TD_ATOMIC_WAVEFRONT_sum TD_STORE_WAVEFRONT_sum -pmc: TD_COALESCABLE_WAVEFRONT_sum - - - -gpu: -range: -kernel: diff --git a/src/rocprof_compute_soc/soc_base.py b/src/rocprof_compute_soc/soc_base.py index 277157286f..2e4a34e705 100644 --- a/src/rocprof_compute_soc/soc_base.py +++ b/src/rocprof_compute_soc/soc_base.py @@ -27,25 +27,27 @@ import math import os import re import shutil -import sys import threading -from abc import ABC, abstractmethod -from collections import OrderedDict +from abc import abstractmethod from pathlib import Path -import numpy as np import pandas as pd import yaml +import config from utils.mi_gpu_spec import get_gpu_model, get_gpu_series from utils.parser import build_in_vars, supported_denom from utils.utils import ( + capture_subprocess_output, console_debug, console_error, console_log, + console_warning, convert_metric_id_to_panel_idx, demarcate, - get_default_accumulate_counter_file_ymal, + detect_rocprof, + get_submodules, + is_tcc_channel_counter, total_xcds, using_v3, ) @@ -63,7 +65,6 @@ class OmniSoC_Base: self.__perfmon_config = ( {} ) # Per IP block max number of simulutaneous counters. GFX IP Blocks - self.__section_counters = set() # hw counters corresponding to filtered sections self.__soc_params = {} # SoC specifications self.__compatible_profilers = [] # Store profilers compatible with SoC self.populate_mspec() @@ -82,9 +83,6 @@ class OmniSoC_Base: def set_perfmon_config(self, config: dict): self.__perfmon_config = config - def get_workload_perfmon_dir(self): - return str(Path(self.__perfmon_dir).parent.absolute()) - def get_soc_param(self): return self.__soc_params @@ -191,45 +189,85 @@ class OmniSoC_Base: ) @demarcate - def section_filter(self): + def detect_counters(self): """ Create a set of counters required for the selected report sections. Parse analysis report configuration files based on the selected report sections to be filtered. """ - args = self.__args - for section in self.__filter_metric_ids: + counters = set() + config_filenames = { + filename: [] + for filename in os.listdir( + Path(self.get_args().config_dir).joinpath(self.__arch) + ) + if filename.endswith(".yaml") + } + metric_ids = [ + name + for name, type in self.get_args().filter_blocks.items() + if type == "metric_id" + ] + file_ids = [] + for section in metric_ids: section_num = convert_metric_id_to_panel_idx(section) file_id = str(section_num // 100) # Convert "4" to "04" if len(file_id) == 1: file_id = f"0{file_id}" - # Identify yaml file corresponding to file_id - config_filename = [ - filename - for filename in os.listdir(Path(args.config_dir).joinpath(self.__arch)) - if filename.endswith(".yaml") and filename.startswith(file_id) - ][0] + file_ids.append(file_id) + # Apply sub section filtering + for config_filename in config_filenames: + if config_filename.startswith(file_id) and section_num % 100: + config_filenames[config_filename].append(section_num) + + # Apply section filters only if metric ids have been provided for filtering + if metric_ids: + # Identify yaml files corresponding to file_ids + config_filenames = { + filename: subsections + for filename, subsections in config_filenames.items() + if filename.startswith(tuple(file_ids)) + } + + for config_filename, subsections in config_filenames.items(): # Read the yaml file with open( - Path(args.config_dir).joinpath(self.__arch, config_filename), "r" + Path(self.get_args().config_dir).joinpath(self.__arch, config_filename), + "r", ) as stream: section_config = yaml.safe_load(stream) # Extract subsection if section is of the form 4.52 - if section_num % 100: + if subsections: section_config_text = "\n".join( [ # Convert yaml to string yaml.dump(subsection) for subsection in section_config["Panel Config"]["data source"] - if subsection["metric_table"]["id"] == section_num + if subsection["metric_table"]["id"] in subsections ] ) else: # Convert yaml to string section_config_text = yaml.dump(section_config) - self.__section_counters = self.__section_counters.union( - parse_counters(section_config_text) - ) + counters = counters.union(self.parse_counters(section_config_text)) + + # Handle TCC channel counters: if hw_counter_matches has elements ending with '[' + # Expand and interleve the TCC channel counters + # e.g. TCC_HIT[0] TCC_ATOMIC[0] ... TCC_HIT[1] TCC_ATOMIC[1] ... + for counter_name in counters.copy(): + if counter_name.startswith("TCC") and counter_name.endswith("["): + counters.remove(counter_name) + counter_name = counter_name.split("[")[0] + counters = counters.union( + { + f"{counter_name}[{i}]" + for i in range( + int(self._mspec.num_xcd) * int(self._mspec._l2_banks) + ) + } + ) + + return counters @demarcate def perfmon_filter(self, roofline_perfmon_only: bool): @@ -240,31 +278,177 @@ class OmniSoC_Base: ): return + if roofline_perfmon_only: + counters = set() + for fname in glob.glob(self.__perfmon_dir + "/" + "pmc_roof_perf.txt"): + lines = open(fname, "r").read().splitlines() + for line in lines: + # Strip all comments, skip empty lines + stext = line.split("#")[0].strip() + if not stext: + continue + # all pmc counters start with "pmc:" + m = re.match(r"^pmc:(.*)", stext) + if m is None: + continue + # de-duplicate counters + counters = counters.union(set(m.group(1).split())) + else: + counters = self.detect_counters() + # Perfmon hardware block filtering + filter_hardware_blocks = [ + name + for name, type in self.get_args().filter_blocks.items() + if type == "hardware_block" + ] + if filter_hardware_blocks: + counters = { + counter_name + for counter_name in counters + if counter_name.startswith(tuple(filter_hardware_blocks)) + } + + if using_v3(): + # Counters not supported in rocprof v3 + counters = counters - {"TCC_BUBBLE"} + else: + # Counters not supported in rocprof v1 / v2 + counters = counters - {"SQ_INSTS_VALU_MFMA_F8", "SQ_INSTS_VALU_MFMA_MOPS_F8"} + + # Following counters are not supported + # TCP_TCP_LATENCY_sum (except for gfx908 and gfx90a) + # SQC_DCACHE_INFLIGHT_LEVEL + counters = counters - {"SQC_DCACHE_INFLIGHT_LEVEL"} + if self.__arch not in ("gfx908", "gfx90a"): + counters = counters - {"TCP_TCP_LATENCY_sum"} + + # Coalesce and writeback workload specific perfmon + self.perfmon_coalesce(counters) + + @demarcate + def parse_counters(self, config_text): + """ + Create a set of all hardware counters mentioned in the given config file content string + """ + hw_counter_matches, variable_matches = self.parse_counters_text(config_text) + + # get hw counters and variables for all supported denominators + for formula in supported_denom.values(): + hw_counter_matches_denom, variable_matches_denom = self.parse_counters_text( + formula + ) + hw_counter_matches.update(hw_counter_matches_denom) + variable_matches.update(variable_matches_denom) + + # get hw counters corresponding to variables recursively + while variable_matches: + subvariable_matches = set() + for var in variable_matches: + if var in build_in_vars: + hw_counter_matches_vars, variable_matches_vars = ( + self.parse_counters_text(build_in_vars[var]) + ) + hw_counter_matches.update(hw_counter_matches_vars) + subvariable_matches.update(variable_matches_vars) + # process new found variables + variable_matches = subvariable_matches - variable_matches + + return hw_counter_matches + + def parse_counters_text(self, text): + """Parse out hardware counters and variables from given text""" + # hw counter name should start with ip block name + # hw counter name should have all capital letters or digits and should not end with underscore + # he counter name can either optionally end with '[' or '_sum' + hw_counter_regex = ( + r"(?:SQ|SQC|TA|TD|TCP|TCC|CPC|CPF|SPI|GRBM)_[0-9A-Z_]*[0-9A-Z](?:\[|_sum)*" + ) + # only capture the variable name after $ using capturing group + variable_regex = r"\$([0-9A-Za-z_]*[0-9A-Za-z])" + hw_counter_matches = set(re.findall(hw_counter_regex, text)) + variable_matches = set(re.findall(variable_regex, text)) + # variable matches cannot be counters + hw_counter_matches = hw_counter_matches - variable_matches + return hw_counter_matches, variable_matches + + def get_rocprof_supported_counters(self): + rocprof_cmd = detect_rocprof() + rorcprof_counters = set() + + if str(rocprof_cmd).endswith("rocprof"): + command = [rocprof_cmd, "--list-basic"] + success, output = capture_subprocess_output(command, enable_logging=False) + # return code should be 1 so success should be False + if success: + console_error( + f"Failed to list rocprof supported counters using command: {command}" + ) + for line in output.splitlines(): + if "gpu-agent" in line: + counters, _ = self.parse_counters_text(line.split(":")[1].strip()) + rorcprof_counters.update(counters) + + command = [rocprof_cmd, "--list-derived"] + success, output = capture_subprocess_output(command, enable_logging=False) + # return code should be 1 so success should be False + if success: + console_error( + f"Failed to list rocprof supported counters using command: {command}" + ) + for line in output.splitlines(): + if "gpu-agent" in line: + counters, _ = self.parse_counters_text(line.split(":")[1].strip()) + rorcprof_counters.update(counters) + + elif str(rocprof_cmd).endswith("rocprofv2"): + command = [rocprof_cmd, "--list-counters"] + success, output = capture_subprocess_output(command, enable_logging=False) + # return code should be 1 so success should be False + if success: + console_error( + f"Failed to list rocprof supported counters using command: {command}" + ) + for line in output.splitlines(): + if "gfx" in line: + counters, _ = self.parse_counters_text(line.split(":")[2].strip()) + rorcprof_counters.update(counters) + + elif str(rocprof_cmd).endswith("rocprofv3"): + command = [rocprof_cmd, "--list-avail"] + success, output = capture_subprocess_output(command, enable_logging=False) + # return code should be 0 so success should be True + if not success: + console_error( + f"Failed to list rocprof supported counters using command: {command}" + ) + for line in output.splitlines(): + if "Name:" in line: + counters, _ = self.parse_counters_text(line.split(":")[1].strip()) + rorcprof_counters.update(counters) + + else: + console_error( + "Incompatible profiler: %s. Supported profilers include: %s" + % (rocprof_cmd, get_submodules("rocprof_compute_profile")) + ) + + return rorcprof_counters + + @demarcate + def perfmon_coalesce(self, counters): + """Sort and bucket all related performance counters to minimize required application passes""" + + # Create workload directory # In some cases (i.e. --specs) path will not be given - if hasattr(self.__args, "path"): - if self.__args.path == str(Path(os.getcwd()).joinpath("workloads")): + if hasattr(self.get_args(), "path"): + if self.get_args().path == str(Path(os.getcwd()).joinpath("workloads")): workload_dir = str( - Path(self.__args.path).joinpath( - self.__args.name, self._mspec.gpu_model + Path(self.get_args().path).joinpath( + self.get_args().name, self._mspec.gpu_model ) ) else: - workload_dir = self.__args.path - - workload_perfmon_dir = workload_dir + "/perfmon" - - self.__filter_hardware_blocks = [ - name - for name, type in self.get_args().filter_blocks.items() - if type == "hardware_block" - ] - self.__filter_metric_ids = [ - name - for name, type in self.get_args().filter_blocks.items() - if type == "metric_id" - ] - - self.section_filter() + workload_dir = self.get_args().path # Initialize directories if not Path(workload_dir).is_dir(): @@ -274,48 +458,223 @@ class OmniSoC_Base: else: os.unlink(workload_dir) + workload_perfmon_dir = workload_dir + "/perfmon" os.makedirs(workload_perfmon_dir) - if not roofline_perfmon_only: - ref_pmc_files_list = glob.glob(self.__perfmon_dir + "/" + "pmc_*perf*.txt") - ref_pmc_files_list += glob.glob( - self.__perfmon_dir + "/" + self.__arch + "/pmc_*_perf*.txt" + # Sanity check whether counters are supported by underlying rocprof tool + rocprof_counters = self.get_rocprof_supported_counters() + # rocprof does not support TCC channel counters, so remove channel suffix for comparison + not_supported_counters = { + counter.split("[")[0] if is_tcc_channel_counter(counter) else counter + for counter in counters + } - rocprof_counters + if not_supported_counters: + console_warning( + f"Following counters might not be supported by rocprof: {', '.join(not_supported_counters)} " + ) + # We might be providing definitions of unsupported counters, so still try to collect them + if not counters: + console_error( + "profiling", + "No performance counters to collect, please check the provided profiling filters", + ) + else: + console_log(f"Collecting following counters: {', '.join(counters)} ") + + output_files = [] + + accu_file_count = 0 + # Create separate perfmon file for LEVEL counters without _sum suffix + # TCC LEVEL counters are handled channel wise, so ignore them + for counter in counters.copy(): + if ( + "LEVEL" in counter + and not counter.endswith("_sum") + and not is_tcc_channel_counter(counter) + ): + counters.remove(counter) + output_files.append(CounterFile(counter + ".txt", self.__perfmon_config)) + output_files[-1].add(counter) + if using_v3(): + # v3 does not support SQ_ACCUM_PREV_HIRES. Instead we defined our own + # counters in counter_defs.yaml that use the accumulate() function. These + # use the name of the accumulate counter with _ACCUM appended to them. + output_files[-1].add(counter + "_ACCUM") + else: + output_files[-1].add("SQ_ACCUM_PREV_HIRES") + accu_file_count += 1 + + file_count = 0 + # Store all channels for a TCC channel counter in the same file + tcc_channel_counter_file_map = dict() + for ctr in counters: + # Store all channels for a TCC channel counter in the same file + if is_tcc_channel_counter(ctr): + output_file = tcc_channel_counter_file_map.get(ctr.split("[")[0]) + if output_file: + output_file.add(ctr) + continue + # Add counter to first file that has room + added = False + for i in range(len(output_files)): + if output_files[i].add(ctr): + added = True + if is_tcc_channel_counter(ctr): + tcc_channel_counter_file_map[ctr.split("[")[0]] = output_files[i] + break + + # All files are full, create a new file + if not added: + output_files.append( + CounterFile( + "pmc_perf_{}.txt".format(file_count), self.__perfmon_config + ) + ) + file_count += 1 + output_files[-1].add(ctr) + + console_debug("profiling", "perfmon_coalesce file_count %s" % file_count) + + # TODO: rewrite the above logic for spatial_multiplexing later + if self.get_args().spatial_multiplexing: + + # TODO: more error checking + if len(self.get_args().spatial_multiplexing) != 3: + console_error( + "profiling", + "multiplexing need provide node_idx node_count and gpu_count", + ) + + node_idx = int(self.get_args().spatial_multiplexing[0]) + node_count = int(self.get_args().spatial_multiplexing[1]) + gpu_count = int(self.get_args().spatial_multiplexing[2]) + + old_group_num = file_count + accu_file_count + new_bucket_count = node_count * gpu_count + groups_per_bucket = math.ceil( + old_group_num / new_bucket_count + ) # It equals to file num per node + max_groups_per_node = groups_per_bucket * gpu_count + + group_start = node_idx * max_groups_per_node + group_end = min((node_idx + 1) * max_groups_per_node, old_group_num) + + console_debug( + "profiling", + "spatial_multiplexing node_idx %s, node_count %s, gpu_count: %s, old_group_num %s, " + "new_bucket_count %s, groups_per_bucket %s, max_groups_per_node %s, " + "group_start %s, group_end %s" + % ( + node_idx, + node_count, + gpu_count, + old_group_num, + new_bucket_count, + groups_per_bucket, + max_groups_per_node, + group_start, + group_end, + ), ) - # Perfmon list filtering - if self.__filter_hardware_blocks: - hardware_blocks = [ - block.lower() for block in self.__filter_hardware_blocks - ] - mpattern = "pmc_([a-zA-Z0-9_]+)_perf*" + for f_idx in range(groups_per_bucket): + file_name = str( + Path(workload_perfmon_dir).joinpath( + "pmc_perf_" + "node_" + str(node_idx) + "_" + str(f_idx) + ".txt" + ) + ) - pmc_files_list = [] - for fname in ref_pmc_files_list: - fbase = Path(fname).stem - ip = re.match(mpattern, fbase).group(1) - if ip in hardware_blocks: - pmc_files_list.append(fname) - console_log("fname: " + fbase + ": Added") - else: - console_log("fname: " + fbase + ": Skipped") + pmc = [] + for g_idx in range( + group_start + f_idx * gpu_count, + min(group_end, group_start + (f_idx + 1) * gpu_count), + ): + gpu_idx = g_idx % gpu_count + for block_name in output_files[g_idx].blocks.keys(): + for ctr in output_files[g_idx].blocks[block_name].elements: + pmc.append(ctr + ":device=" + str(gpu_idx)) + + stext = "pmc: " + " ".join(pmc) + + # Write counters to file + fd = open(file_name, "w") + fd.write(stext + "\n\n") + fd.close() - else: - # default: take all perfmons - pmc_files_list = ref_pmc_files_list else: - ref_pmc_files_list = glob.glob(self.__perfmon_dir + "/" + "pmc_roof_perf.txt") - pmc_files_list = ref_pmc_files_list + # Output to files + with open( + str( + Path(config.rocprof_compute_home).joinpath( + "rocprof_compute_soc", + "profile_configs", + "accum_counters.yaml", + ) + ), + "r", + ) as fp: + accum_counters_def = yaml.safe_load(fp) - # Coalesce and writeback workload specific perfmon - perfmon_coalesce( - pmc_files_list, - self.__perfmon_config, - workload_dir, - self.get_args().spatial_multiplexing, - self.__section_counters, - self._mspec, - self.__arch, - ) + for f in output_files: + file_name_txt = str(Path(workload_perfmon_dir).joinpath(f.file_name_txt)) + file_name_yaml = str( + Path(workload_perfmon_dir).joinpath(f.file_name_yaml) + ) + + pmc = [] + counter_def = dict() + for ctr in [ + ctr + for block_name in f.blocks + for ctr in f.blocks[block_name].elements + ]: + pmc.append(ctr) + if using_v3(): + if ctr in accum_counters_def: + counter_def[ctr] = accum_counters_def[ctr] + # Add TCC channel counters definitions + if is_tcc_channel_counter(ctr): + counter_name = ctr.split("[")[0] + idx = int(ctr.split("[")[1].split("]")[0]) + xcd_idx = idx // int(self._mspec._l2_banks) + channel_idx = idx % int(self._mspec._l2_banks) + counter_def.update( + { + ctr: { + "architectures": { + self.__arch: { + "expression": f"select({counter_name},[DIMENSION_XCC=[{xcd_idx}], DIMENSION_INSTANCE=[{channel_idx}]])", + } + }, + "description": f"{counter_name} on {xcd_idx}th XCC and {channel_idx}th channel", + } + } + ) + + stext = "pmc: " + " ".join(pmc) + # Write counters to file + fd = open(file_name_txt, "w") + fd.write(stext + "\n\n") + fd.write("gpu:\n") + fd.write("range:\n") + fd.write("kernel:\n") + fd.close() + + # Write counter definitions to file + if using_v3(): + with open(file_name_yaml, "w") as fp: + if counter_def: + fp.write(yaml.dump(counter_def)) + + # Add a timestamp file + # TODO: Does v3 need this? + if not using_v3(): + fd = open(str(Path(workload_perfmon_dir).joinpath("timestamps.txt")), "w") + fd.write("pmc:\n\n") + fd.write("gpu:\n") + fd.write("range:\n") + fd.write("kernel:\n") + fd.close() # ---------------------------------------------------- # Required methods to be implemented by child classes @@ -336,10 +695,6 @@ class OmniSoC_Base: console_debug("analysis", "perform SoC analysis setup for %s" % self.__arch) -def getblock(counter): - return counter.split("_")[0] - - # Set with limited size class LimitedSet: def __init__(self, maxsize) -> None: @@ -349,13 +704,15 @@ class LimitedSet: def add(self, e) -> None: if e in self.elements: return True - elif self.avail <= 0: - return False - - self.avail -= 1 - self.elements.append(e) - - return True + # Store all channels for a TCC channel counter in the same file + if e.split("[")[0] in {element.split("[")[0] for element in self.elements}: + self.elements.append(e) + return True + if self.avail > 0: + self.avail -= 1 + self.elements.append(e) + return True + return False # Represents a file that lists PMC counters. Number of counters for each @@ -368,435 +725,10 @@ class CounterFile: self.blocks = {b: LimitedSet(v) for b, v in perfmon_config.items()} def add(self, counter) -> bool: - block = getblock(counter) + block = counter.split("_")[0] # SQ and SQC belong to the same IP block if block == "SQC": block = "SQ" return self.blocks[block].add(counter) - - -# FIXME: This is a HACK -def using_v3(): - return "ROCPROF" in os.environ.keys() and os.environ["ROCPROF"].endswith("rocprofv3") - - -@demarcate -def parse_counters(config_text): - """ - Create a set of all hardware counters mentioned in the given config file content string - """ - # hw counter name should start with ip block name - hw_counter_regex = r"(?:SQ|SQC|TA|TD|TCP|TCC|CPC|CPF|SPI|GRBM)_[0-9A-Za-z_]+" - # only capture the variable name after $ using capturing group - variable_regex = r"\$([0-9A-Za-z_]+)" - hw_counter_matches = set(re.findall(hw_counter_regex, config_text)) - variable_matches = set(re.findall(variable_regex, config_text)) - # get hw counters and variables for all supported denominators - for formula in supported_denom.values(): - hw_counter_matches.update(re.findall(hw_counter_regex, formula)) - variable_matches.update(re.findall(variable_regex, formula)) - # get hw counters corresponding to variables recursively - while variable_matches: - subvariable_matches = set() - for var in variable_matches: - if var in build_in_vars: - hw_counter_matches.update( - re.findall(hw_counter_regex, build_in_vars[var]) - ) - subvariable_matches.update(re.findall(variable_regex, build_in_vars[var])) - # process new found variables - variable_matches = subvariable_matches - variable_matches - return list(hw_counter_matches) - - -@demarcate -def perfmon_coalesce( - pmc_files_list, - perfmon_config, - workload_dir, - spatial_multiplexing, - section_counters, - mspec, - arch, -): - """Sort and bucket all related performance counters to minimize required application passes""" - workload_perfmon_dir = workload_dir + "/perfmon" - - # Will be 2D array - accumulate_counters = [] - - normal_counters = OrderedDict() - - for fname in pmc_files_list: - - lines = open(fname, "r").read().splitlines() - - for line in lines: - - # Strip all comments, skip empty lines - stext = line.split("#")[0].strip() - if not stext: - continue - - # all pmc counters start with "pmc:" - m = re.match(r"^pmc:(.*)", stext) - if m is None: - continue - - counters = m.group(1).split() - - if "SQ_ACCUM_PREV_HIRES" in counters: - # Accumulate counters - accumulate_counters.append(counters.copy()) - else: - # Normal counters - for ctr in counters: - # Remove me later: - # v1 and v2 don't support these counters - if not using_v3(): - if ctr.startswith("SQ_INSTS_VALU_MFMA_F8") or ctr.startswith( - "SQ_INSTS_VALU_MFMA_MOPS_F8" - ): - continue - - # Channel counter e.g. TCC_ATOMIC[0] - if "[" in ctr: - - # FIXME: - # Remove channel number, append "_sum" so rocprof will - # sum the counters for us instead of specifying every - # channel. - channel = int(ctr.split("[")[1].split("]")[0]) - if channel == 0: - counter_name = ( - ctr.split("[")[0] + "_sum" - if using_v3() - else ctr.split("[")[0] + "_expand" - ) - - try: - normal_counters[counter_name] += 1 - except: - normal_counters[counter_name] = 1 - else: - try: - normal_counters[ctr] += 1 - except: - normal_counters[ctr] = 1 - - # De-duplicate. Remove accumulate counters from normal counters - for accus in accumulate_counters: - for accu in accus: - if accu in normal_counters: - del normal_counters[accu] - - # If section report filters have been provided, only collect counters necessary for those section reports - # Remove _sum and _expand suffixes while matching - def remove_suffixes(string): - for suffix in ["_sum", "_expand"]: - if string.endswith(suffix): - string = string[: -len(suffix)] - break - return string - - section_counters = {remove_suffixes(counter) for counter in section_counters} - ignored_counters = list() - - if section_counters: - # Remove unnecessary normal counters - for counter_name in list(normal_counters.keys()): - if remove_suffixes(counter_name) not in section_counters: - del normal_counters[counter_name] - ignored_counters.append(counter_name) - - # Remove unnecessary accumulate counters - filtered_accumlate_counters = list() - for counters in accumulate_counters: - if any( - remove_suffixes(counter_name) in section_counters - for counter_name in counters - ): - filtered_accumlate_counters.append(counters) - else: - ignored_counters.extend(counter_name) - accumulate_counters = filtered_accumlate_counters - - if ignored_counters: - console_log( - f"Not collecting following counters per provided filter: {', '.join(ignored_counters)} " - ) - - # Throw error if no counters to be collected - if len(normal_counters) == 0 and len(accumulate_counters) == 0: - console_error( - "profiling", - "No performance counters to collect, please check the provided profiling filters", - ) - - output_files = [] - - accu_file_count = 0 - # Each accumulate counter is in a different file - for ctrs in accumulate_counters: - - ctr_name = ctrs[ctrs.index("SQ_ACCUM_PREV_HIRES") - 1] - - if using_v3(): - # v3 does not support SQ_ACCUM_PREV_HIRES. Instead we defined our own - # counters in counter_defs.yaml that use the accumulate() function. These - # use the name of the accumulate counter with _ACCUM appended to them. - ctrs.remove("SQ_ACCUM_PREV_HIRES") - - accum_name = ctr_name + "_ACCUM" - - ctrs.append(accum_name) - - # Use the name of the accumulate counter as the file name - output_files.append(CounterFile(ctr_name, perfmon_config)) - for ctr in ctrs: - output_files[-1].add(ctr) - accu_file_count += 1 - - file_count = 0 - for ctr in normal_counters.keys(): - - # Add counter to first file that has room - added = False - for f in output_files: - if f.add(ctr): - added = True - break - - # All files are full, create a new file - if not added: - output_files.append( - CounterFile("pmc_perf_{}.txt".format(file_count), perfmon_config) - ) - file_count += 1 - output_files[-1].add(ctr) - - console_debug("profiling", "perfmon_coalesce file_count %s" % file_count) - - # TODO: rewrite the above logic for spatial_multiplexing later - if spatial_multiplexing: - - # TODO: more error checking - if len(spatial_multiplexing) != 3: - console_error( - "profiling", - "multiplexing need provide node_idx node_count and gpu_count", - ) - - node_idx = int(spatial_multiplexing[0]) - node_count = int(spatial_multiplexing[1]) - gpu_count = int(spatial_multiplexing[2]) - - old_group_num = file_count + accu_file_count - new_bucket_count = node_count * gpu_count - groups_per_bucket = math.ceil( - old_group_num / new_bucket_count - ) # It equals to file num per node - max_groups_per_node = groups_per_bucket * gpu_count - - group_start = node_idx * max_groups_per_node - group_end = min((node_idx + 1) * max_groups_per_node, old_group_num) - - console_debug( - "profiling", - "spatial_multiplexing node_idx %s, node_count %s, gpu_count: %s, old_group_num %s, " - "new_bucket_count %s, groups_per_bucket %s, max_groups_per_node %s, " - "group_start %s, group_end %s" - % ( - node_idx, - node_count, - gpu_count, - old_group_num, - new_bucket_count, - groups_per_bucket, - max_groups_per_node, - group_start, - group_end, - ), - ) - - for f_idx in range(groups_per_bucket): - file_name = str( - Path(workload_perfmon_dir).joinpath( - "pmc_perf_" + "node_" + str(node_idx) + "_" + str(f_idx) + ".txt" - ) - ) - - pmc = [] - for g_idx in range( - group_start + f_idx * gpu_count, - min(group_end, group_start + (f_idx + 1) * gpu_count), - ): - gpu_idx = g_idx % gpu_count - for block_name in output_files[g_idx].blocks.keys(): - for ctr in output_files[g_idx].blocks[block_name].elements: - pmc.append(ctr + ":device=" + str(gpu_idx)) - - stext = "pmc: " + " ".join(pmc) - - # Write counters to file - fd = open(file_name, "w") - fd.write(stext + "\n\n") - fd.close() - - else: - # Output to files - for f in output_files: - file_name_txt = str(Path(workload_perfmon_dir).joinpath(f.file_name_txt)) - file_name_yaml = str(Path(workload_perfmon_dir).joinpath(f.file_name_yaml)) - - pmc = [] - for block_name in f.blocks.keys(): - if block_name == "TCC": - if using_v3(): - # TODO: implement mehcanisim for muti channel TCC counter for v3 - # Expand and interleve the TCC channel counters - # e.g. TCC_HIT[0] TCC_ATOMIC[0] ... TCC_HIT[1] TCC_ATOMIC[1] ... - from utils.specs import total_xcds - - channel_counters = [] - xcds = total_xcds(mspec.gpu_model, mspec.compute_partition) - tcc_channel_per_xcd = int(mspec._l2_banks) - - for ctr in f.blocks[block_name].elements: - if "_sum" in ctr: - channel_counters.append(ctr.split("_sum")[0]) - pmc.append(ctr) - - channel_counters = ( - pd.Series(channel_counters).drop_duplicates().to_list() - ) - - lock = threading.Lock() - - def generate_yaml_config_per_pmc( - raw_counter_name, - tcc_counter_1d_index, - xcd_index, - channel_index, - ): - # Ensure that the keys exist before trying to assign values - yaml_data = {} - if tcc_counter_1d_index not in yaml_data: - yaml_data[tcc_counter_1d_index] = { - "architectures": {}, - "description": "", - } - - if ( - arch - not in yaml_data[tcc_counter_1d_index]["architectures"] - ): - yaml_data[tcc_counter_1d_index]["architectures"][ - arch - ] = {} - - yaml_data[tcc_counter_1d_index]["architectures"][arch][ - "expression" - ] = "select({},[DIMENSION_XCC=[{}], DIMENSION_INSTANCE=[{}]])".format( - raw_counter_name, xcd_index, channel_index - ) - yaml_data[tcc_counter_1d_index]["description"] = ( - "{} on {}th XCC and {}th channel".format( - raw_counter_name, xcd_index, channel_index - ) - ) - - lock.acquire() - pmc.append(tcc_counter_1d_index) - with open(file_name_yaml, "a") as file_yaml: - yaml.dump( - yaml_data, - file_yaml, - default_flow_style=False, - allow_unicode=True, - ) - lock.release() - - threads_edit_yaml = [] - - for i in range(0, xcds): - for j in range(0, tcc_channel_per_xcd): - for c in channel_counters: - tcc_counter_1d_index = "{}[{}]".format( - c, (i * tcc_channel_per_xcd) + j - ) - - thread = threading.Thread( - target=generate_yaml_config_per_pmc, - args=[c, tcc_counter_1d_index, i, j], - ) - threads_edit_yaml.append(thread) - thread.start() - - for thread in threads_edit_yaml: - thread.join() - - # Handle the rest of the TCC counters - for ctr in f.blocks[block_name].elements: - if "_expand" not in ctr and "_sum" not in ctr: - pmc.append(ctr) - - else: - # Expand and interleve the TCC channel counters - # e.g. TCC_HIT[0] TCC_ATOMIC[0] ... TCC_HIT[1] TCC_ATOMIC[1] ... - channel_counters = [] - for ctr in f.blocks[block_name].elements: - if "_expand" in ctr: - channel_counters.append(ctr.split("_expand")[0]) - for i in range(0, perfmon_config["TCC_channels"]): - for c in channel_counters: - pmc.append("{}[{}]".format(c, i)) - # Handle the rest of the TCC counters - for ctr in f.blocks[block_name].elements: - if "_expand" not in ctr: - pmc.append(ctr) - - else: - if using_v3(): - yaml_global_config_dir = ( - get_default_accumulate_counter_file_ymal() - ) - with open(yaml_global_config_dir, "r") as file_read: - with open(file_name_yaml, "a") as file_out: - dic_read = yaml.safe_load(file_read) - for ctr in f.blocks[block_name].elements: - if ctr in dic_read: - section_to_dump = {ctr: dic_read[ctr]} - yaml.dump( - section_to_dump, - file_out, - default_flow_style=False, - allow_unicode=True, - ) - pmc.append(ctr) - - else: - for ctr in f.blocks[block_name].elements: - pmc.append(ctr) - - stext = "pmc: " + " ".join(pmc) - - # Write counters to file - fd = open(file_name_txt, "w") - fd.write(stext + "\n\n") - fd.write("gpu:\n") - fd.write("range:\n") - fd.write("kernel:\n") - fd.close() - - # Add a timestamp file - # TODO: Does v3 need this? - if not using_v3(): - fd = open(str(Path(workload_perfmon_dir).joinpath("timestamps.txt")), "w") - fd.write("pmc:\n\n") - fd.write("gpu:\n") - fd.write("range:\n") - fd.write("kernel:\n") - fd.close() diff --git a/src/rocprof_compute_soc/soc_gfx906.py b/src/rocprof_compute_soc/soc_gfx906.py index 5bdbe13e49..31d8653fcc 100644 --- a/src/rocprof_compute_soc/soc_gfx906.py +++ b/src/rocprof_compute_soc/soc_gfx906.py @@ -33,15 +33,6 @@ class gfx906_soc(OmniSoC_Base): def __init__(self, args, mspec): super().__init__(args, mspec) self.set_arch("gfx906") - self.set_perfmon_dir( - str( - Path(str(config.rocprof_compute_home)).joinpath( - "rocprof_compute_soc", - "profile_configs", - self.get_arch(), - ) - ) - ) self.set_compatible_profilers(["rocprofv1", "rocscope"]) # Per IP block max number of simultaneous counters. GFX IP Blocks self.set_perfmon_config( @@ -56,7 +47,6 @@ class gfx906_soc(OmniSoC_Base): "SPI": 2, "GRBM": 2, "GDS": 4, - "TCC_channels": 16, } ) diff --git a/src/rocprof_compute_soc/soc_gfx908.py b/src/rocprof_compute_soc/soc_gfx908.py index 9cf8a46d8a..9c668a424e 100644 --- a/src/rocprof_compute_soc/soc_gfx908.py +++ b/src/rocprof_compute_soc/soc_gfx908.py @@ -33,15 +33,6 @@ class gfx908_soc(OmniSoC_Base): def __init__(self, args, mspec): super().__init__(args, mspec) self.set_arch("gfx908") - self.set_perfmon_dir( - str( - Path(str(config.rocprof_compute_home)).joinpath( - "rocprof_compute_soc", - "profile_configs", - self.get_arch(), - ) - ) - ) self.set_compatible_profilers(["rocprofv1", "rocscope", "rocprofv3"]) # Per IP block max number of simultaneous counters. GFX IP Blocks self.set_perfmon_config( @@ -56,7 +47,6 @@ class gfx908_soc(OmniSoC_Base): "SPI": 2, "GRBM": 2, "GDS": 4, - "TCC_channels": 32, } ) diff --git a/src/rocprof_compute_soc/soc_gfx90a.py b/src/rocprof_compute_soc/soc_gfx90a.py index 734ea0505b..58a3ffa6ce 100644 --- a/src/rocprof_compute_soc/soc_gfx90a.py +++ b/src/rocprof_compute_soc/soc_gfx90a.py @@ -45,16 +45,6 @@ class gfx90a_soc(OmniSoC_Base): ) ) ) - else: - self.set_perfmon_dir( - str( - Path(str(config.rocprof_compute_home)).joinpath( - "rocprof_compute_soc", - "profile_configs", - self.get_arch(), - ) - ) - ) self.set_compatible_profilers(["rocprofv1", "rocscope", "rocprofv2", "rocprofv3"]) # Per IP block max number of simultaneous counters. GFX IP Blocks self.set_perfmon_config( @@ -69,7 +59,6 @@ class gfx90a_soc(OmniSoC_Base): "SPI": 2, "GRBM": 2, "GDS": 4, - "TCC_channels": 32, } ) self.roofline_obj = Roofline(args, self._mspec) diff --git a/src/rocprof_compute_soc/soc_gfx940.py b/src/rocprof_compute_soc/soc_gfx940.py index 626dee1753..055473ed87 100644 --- a/src/rocprof_compute_soc/soc_gfx940.py +++ b/src/rocprof_compute_soc/soc_gfx940.py @@ -45,17 +45,6 @@ class gfx940_soc(OmniSoC_Base): ) ) ) - else: - # NB: We're using generalized Mi300 perfmon configs - self.set_perfmon_dir( - set( - Path(str(config.rocprof_compute_home)).joinpath( - "rocprof_compute_soc", - "profile_configs", - "gfx940", - ) - ) - ) self.set_compatible_profilers(["rocprofv1", "rocprofv2", "rocprofv3"]) # Per IP block max number of simultaneous counters. GFX IP Blocks self.set_perfmon_config( @@ -70,7 +59,6 @@ class gfx940_soc(OmniSoC_Base): "SPI": 2, "GRBM": 2, "GDS": 4, - "TCC_channels": 32, } ) self.roofline_obj = Roofline(args, self._mspec) diff --git a/src/rocprof_compute_soc/soc_gfx941.py b/src/rocprof_compute_soc/soc_gfx941.py index 31e9efd4ad..2ff0325ab7 100644 --- a/src/rocprof_compute_soc/soc_gfx941.py +++ b/src/rocprof_compute_soc/soc_gfx941.py @@ -45,17 +45,6 @@ class gfx941_soc(OmniSoC_Base): ) ) ) - else: - # NB: We're using generalized Mi300 perfmon configs - self.set_perfmon_dir( - str( - Path(str(config.rocprof_compute_home)).joinpath( - "rocprof_compute_soc", - "profile_configs", - "gfx940", - ) - ) - ) self.set_compatible_profilers(["rocprofv1", "rocprofv2", "rocprofv3"]) # Per IP block max number of simultaneous counters. GFX IP Blocks self.set_perfmon_config( @@ -70,7 +59,6 @@ class gfx941_soc(OmniSoC_Base): "SPI": 2, "GRBM": 2, "GDS": 4, - "TCC_channels": 32, } ) self.roofline_obj = Roofline(args, self._mspec) diff --git a/src/rocprof_compute_soc/soc_gfx942.py b/src/rocprof_compute_soc/soc_gfx942.py index 425fedffd8..c8c1e2d2ab 100644 --- a/src/rocprof_compute_soc/soc_gfx942.py +++ b/src/rocprof_compute_soc/soc_gfx942.py @@ -45,17 +45,6 @@ class gfx942_soc(OmniSoC_Base): ) ) ) - else: - # NB: We're using generalized Mi300 perfmon configs - self.set_perfmon_dir( - str( - Path(str(config.rocprof_compute_home)).joinpath( - "rocprof_compute_soc", - "profile_configs", - "gfx940", - ) - ) - ) self.set_compatible_profilers(["rocprofv1", "rocprofv2", "rocprofv3"]) # Per IP block max number of simultaneous counters. GFX IP Blocks self.set_perfmon_config( @@ -70,7 +59,6 @@ class gfx942_soc(OmniSoC_Base): "SPI": 2, "GRBM": 2, "GDS": 4, - "TCC_channels": 16, } ) self.roofline_obj = Roofline(args, self._mspec) diff --git a/src/utils/utils.py b/src/utils/utils.py index 22daf56f7e..61d2861d0f 100644 --- a/src/utils/utils.py +++ b/src/utils/utils.py @@ -48,23 +48,14 @@ rocprof_cmd = "" rocprof_args = "" -# TODO: This is a HACK +def is_tcc_channel_counter(counter): + return counter.startswith("TCC") and counter.endswith("]") + + def using_v3(): return "ROCPROF" in os.environ.keys() and "rocprofv3" in os.environ["ROCPROF"] -# TODO: This is a HACK -def get_default_accumulate_counter_file_ymal(): - """Return the path of the default derivative counters' definatin's yaml file that we current use to store accumulated counters' defination. It will possibly be removed later on""" - return str( - config.rocprof_compute_home.joinpath( - "rocprof_compute_soc", - "profile_configs", - "accum_counters.yaml", - ) - ) - - def demarcate(function): def wrap_function(*args, **kwargs): logging.trace("----- [entering function] -> %s()" % (function.__qualname__)) @@ -205,11 +196,10 @@ def store_app_cmd(args): rocprof_args = args -def capture_subprocess_output(subprocess_args, new_env=None, profileMode=False): - global rocprof_args - # Format command for debug messages, formatting for rocprofv1 and rocprofv2 - command = " ".join(rocprof_args) - console_debug("subprocess", "Running: " + command + " " + " ".join(subprocess_args)) +def capture_subprocess_output( + subprocess_args, new_env=None, profileMode=False, enable_logging=True +): + console_debug("subprocess", "Running: " + " ".join(subprocess_args)) # Start subprocess # bufsize = 1 means output is line buffered # universal_newlines = True is required for line buffering @@ -241,10 +231,11 @@ def capture_subprocess_output(subprocess_args, new_env=None, profileMode=False): # line to read when this function is called line = stream.readline() buf.write(line) - if profileMode: - console_log(rocprof_cmd, line.strip(), indent_level=1) - else: - console_log(line.strip()) + if enable_logging: + if profileMode: + console_log(rocprof_cmd, line.strip(), indent_level=1) + else: + console_log(line.strip()) except UnicodeDecodeError: # Skip this line pass diff --git a/tests/test_profile_general.py b/tests/test_profile_general.py index eed22c4df9..221ad193b8 100644 --- a/tests/test_profile_general.py +++ b/tests/test_profile_general.py @@ -62,12 +62,6 @@ ALL_CSVS = sorted( "pmc_perf_2.csv", "pmc_perf_3.csv", "pmc_perf_4.csv", - "pmc_perf_5.csv", - "pmc_perf_6.csv", - "pmc_perf_7.csv", - "pmc_perf_8.csv", - "pmc_perf_9.csv", - "pmc_perf_10.csv", "sysinfo.csv", "timestamps.csv", ] @@ -87,9 +81,6 @@ ALL_CSVS_MI200 = sorted( "pmc_perf_4.csv", "pmc_perf_5.csv", "pmc_perf_6.csv", - "pmc_perf_7.csv", - "pmc_perf_8.csv", - "pmc_perf_9.csv", "sysinfo.csv", "timestamps.csv", ] @@ -109,8 +100,6 @@ ALL_CSVS_MI300 = sorted( "pmc_perf_4.csv", "pmc_perf_5.csv", "pmc_perf_6.csv", - "pmc_perf_7.csv", - "pmc_perf_8.csv", "sysinfo.csv", "timestamps.csv", ] @@ -615,9 +604,6 @@ def test_block_SQ(binary_handler_profile_rocprof_compute): "pmc_perf_0.csv", "pmc_perf_1.csv", "pmc_perf_2.csv", - "pmc_perf_3.csv", - "pmc_perf_4.csv", - "pmc_perf_5.csv", "sysinfo.csv", "timestamps.csv", ] @@ -636,8 +622,6 @@ def test_block_SQ(binary_handler_profile_rocprof_compute): "pmc_perf_4.csv", "pmc_perf_5.csv", "pmc_perf_6.csv", - "pmc_perf_7.csv", - "pmc_perf_8.csv", "sysinfo.csv", "timestamps.csv", ] @@ -665,7 +649,6 @@ def test_block_SQC(binary_handler_profile_rocprof_compute): "pmc_perf_0.csv", "pmc_perf_1.csv", "pmc_perf_2.csv", - "pmc_perf_3.csv", "sysinfo.csv", "timestamps.csv", ] @@ -801,27 +784,9 @@ def test_block_TCC(binary_handler_profile_rocprof_compute): "pmc_perf_5.csv", "pmc_perf_6.csv", "pmc_perf_7.csv", - "pmc_perf_8.csv", - "pmc_perf_9.csv", - "pmc_perf_10.csv", "sysinfo.csv", "timestamps.csv", ] - if soc == "MI200" or "MI300" in soc: - expected_csvs = [ - "pmc_perf.csv", - "pmc_perf_0.csv", - "pmc_perf_1.csv", - "pmc_perf_2.csv", - "pmc_perf_3.csv", - "pmc_perf_4.csv", - "pmc_perf_5.csv", - "pmc_perf_6.csv", - "pmc_perf_7.csv", - "pmc_perf_8.csv", - "sysinfo.csv", - "timestamps.csv", - ] assert sorted(list(file_dict.keys())) == sorted(expected_csvs) @@ -851,7 +816,6 @@ def test_block_SPI(binary_handler_profile_rocprof_compute): "pmc_perf_5.csv", "pmc_perf_6.csv", "pmc_perf_7.csv", - "pmc_perf_8.csv", "sysinfo.csv", "timestamps.csv", ] @@ -881,7 +845,6 @@ def test_block_CPC(binary_handler_profile_rocprof_compute): "pmc_perf_2.csv", "pmc_perf_3.csv", "pmc_perf_4.csv", - "pmc_perf_5.csv", "sysinfo.csv", "timestamps.csv", ] @@ -937,9 +900,6 @@ def test_block_SQ_CPC(binary_handler_profile_rocprof_compute): "pmc_perf_0.csv", "pmc_perf_1.csv", "pmc_perf_2.csv", - "pmc_perf_3.csv", - "pmc_perf_4.csv", - "pmc_perf_5.csv", "sysinfo.csv", "timestamps.csv", ] @@ -958,8 +918,6 @@ def test_block_SQ_CPC(binary_handler_profile_rocprof_compute): "pmc_perf_4.csv", "pmc_perf_5.csv", "pmc_perf_6.csv", - "pmc_perf_7.csv", - "pmc_perf_8.csv", "sysinfo.csv", "timestamps.csv", ] @@ -992,9 +950,6 @@ def test_block_SQ_TA(binary_handler_profile_rocprof_compute): "pmc_perf_0.csv", "pmc_perf_1.csv", "pmc_perf_2.csv", - "pmc_perf_3.csv", - "pmc_perf_4.csv", - "pmc_perf_5.csv", "sysinfo.csv", "timestamps.csv", ] @@ -1013,8 +968,6 @@ def test_block_SQ_TA(binary_handler_profile_rocprof_compute): "pmc_perf_4.csv", "pmc_perf_5.csv", "pmc_perf_6.csv", - "pmc_perf_7.csv", - "pmc_perf_8.csv", "sysinfo.csv", "timestamps.csv", ] @@ -1043,9 +996,6 @@ def test_block_SQ_SPI(binary_handler_profile_rocprof_compute): "pmc_perf_0.csv", "pmc_perf_1.csv", "pmc_perf_2.csv", - "pmc_perf_3.csv", - "pmc_perf_4.csv", - "pmc_perf_5.csv", "sysinfo.csv", "timestamps.csv", ] @@ -1064,8 +1014,6 @@ def test_block_SQ_SPI(binary_handler_profile_rocprof_compute): "pmc_perf_4.csv", "pmc_perf_5.csv", "pmc_perf_6.csv", - "pmc_perf_7.csv", - "pmc_perf_8.csv", "sysinfo.csv", "timestamps.csv", ] @@ -1099,7 +1047,6 @@ def test_block_SQ_SQC_TCP_CPC(binary_handler_profile_rocprof_compute): "pmc_perf_2.csv", "pmc_perf_3.csv", "pmc_perf_4.csv", - "pmc_perf_5.csv", "sysinfo.csv", "timestamps.csv", ] @@ -1118,8 +1065,6 @@ def test_block_SQ_SQC_TCP_CPC(binary_handler_profile_rocprof_compute): "pmc_perf_4.csv", "pmc_perf_5.csv", "pmc_perf_6.csv", - "pmc_perf_7.csv", - "pmc_perf_8.csv", "sysinfo.csv", "timestamps.csv", ] @@ -1148,9 +1093,6 @@ def test_block_SQ_SPI_TA_TCC_CPF(binary_handler_profile_rocprof_compute): "pmc_perf_0.csv", "pmc_perf_1.csv", "pmc_perf_2.csv", - "pmc_perf_3.csv", - "pmc_perf_4.csv", - "pmc_perf_5.csv", "sysinfo.csv", "timestamps.csv", ] @@ -1169,8 +1111,6 @@ def test_block_SQ_SPI_TA_TCC_CPF(binary_handler_profile_rocprof_compute): "pmc_perf_4.csv", "pmc_perf_5.csv", "pmc_perf_6.csv", - "pmc_perf_7.csv", - "pmc_perf_8.csv", "sysinfo.csv", "timestamps.csv", ]