Skip to content

Add approved SQ/GL1C/GRBM/TA/TCP counters for gfx1150-1153 #10041

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
vipangul merged 5 commits into from
Aug 24, 2026
Merged

Add approved SQ/GL1C/GRBM/TA/TCP counters for gfx1150-1153 #10041

Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
d9b8d84
Expose approved gfx11.5 (RDNA 3.5) PMCs in rocprofiler-sdk config
vipangul Aug 12, 2026
705f107
Merge remote-tracking branch 'origin' into users/vipangul/AIESW-38547…
vipangul Aug 12, 2026
69a5d64
style: format config.yaml counter descriptions per format-deps.py
vipangul Aug 12, 2026
0226d38
Extended GRBM_TC_BUSY for gfx11.5
vipangul Aug 18, 2026
a9ac524
Removed not user actionable counter
vipangul Aug 21, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
303 changes: 303 additions & 0 deletions projects/rocprofiler-sdk/source/share/rocprofiler-sdk/config.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -567,6 +567,51 @@ rocprofiler-sdk:
- gfx942
- gfx950
expression: simd_count/simd_per_cu
- name: GL1C_ARB_RET_LEVEL
description: "The sum of the number of requests sent to the GL1A for output arbitration in flight.
Average GL1A return arbitration latency = GL1C_PERF_SEL_GL1A_RET_LEVEL/GL1C_PERF_SEL_REQ"
properties: []
definitions:
- architectures:
- gfx1150
- gfx1151
- gfx1152
- gfx1153
block: GL1C
event: 3
- name: GL1C_REQ_READ_128B
description: "Count of read requests for 128B of data"
properties: []
definitions:
- architectures:
- gfx1150
- gfx1151
- gfx1152
- gfx1153
block: GL1C
event: 22
- name: GL1C_REQ_READ_32B
description: "Count of read requests for 32B of data"
properties: []
definitions:
- architectures:
- gfx1150
- gfx1151
- gfx1152
- gfx1153
block: GL1C
event: 23
- name: GL1C_REQ_READ_64B
description: "Count of read requests for 64B of data"
properties: []
definitions:
- architectures:
- gfx1150
- gfx1151
- gfx1152
- gfx1153
block: GL1C
event: 24
- name: SIMD_NUM
description: SIMD Number
properties: []
Expand Down Expand Up @@ -5350,6 +5395,19 @@ rocprofiler-sdk:
- gfx950
block: SQ
event: 55
- name: SQ_INST_LEVEL_TEX_LOAD
description: "Number of in-flight buffer load, image load, sample, or atomic (with return) instructions.
Set next counter to ACCUM_PREV and divide by INST_TEX_LOAD for average latency. Includes FLAT
instructions. {level, nondeterministic, C2}"
properties: []
definitions:
- architectures:
- gfx1150
- gfx1151
- gfx1152
- gfx1153
block: SQ
event: 90
- name: SQ_VALU_MFMA_COEXEC_CYCLES
description: Number of cycles in which MFMA VALU was busy and a normal VALU instruction was
issued (co-execution) (per-simd, nondeterministic)
Expand Down Expand Up @@ -6224,6 +6282,40 @@ rocprofiler-sdk:
- gfx950
block: SQ
event: 93
- name: SQ_VALU_STARVE
description: "Number of cycles VALU is starved while waves are present (cycles in which no sq_sp_cmd
credits are in use). {nondeterministic, C1}"
properties: []
definitions:
- architectures:
- gfx1150
- gfx1151
- gfx1152
- gfx1153
block: SQ
event: 115
- name: SQ_VMEM_BUS_ACTIVE
description: "Cycles sending address/data to LDS/TA, not windowed by perf_en"
properties: []
definitions:
- architectures:
- gfx1150
- gfx1151
- gfx1152
- gfx1153
block: SQ
event: 119
- name: SQ_VMEM_BUS_STALL
description: "Cycles stalled from sending address/data to LDS/TA"
properties: []
definitions:
- architectures:
- gfx1150
- gfx1151
- gfx1152
- gfx1153
block: SQ
event: 120
- name: SQ_WAIT_ANY
description: Number of wave-cycles spent waiting for anything (per-simd, nondeterministic).
Units in quad-cycles(4 cycles)
Expand Down Expand Up @@ -6272,6 +6364,42 @@ rocprofiler-sdk:
- gfx950
block: SQ
event: 106
- name: SQ_WAIT_CNT_LGKM
description: "Number of clock cycles spent waiting for LGKM counter. In units of cycles. {nondeterministic,
C1}"
properties: []
definitions:
- architectures:
- gfx1150
- gfx1151
- gfx1152
- gfx1153
block: SQ
event: 38
- name: SQ_WAIT_CNT_VMVS
description: "Number of clock cycles spent waiting for VM or VS counter. In units of cycles. {nondeterministic,
C1}"
properties: []
definitions:
- architectures:
- gfx1150
- gfx1151
- gfx1152
- gfx1153
block: SQ
event: 37
- name: SQ_WAIT_DELAY_ALU
description: "Number of clock cycles spent waiting for s_delay_alu to finish. {nondeterministic,
C1}"
properties: []
definitions:
- architectures:
- gfx1150
- gfx1151
- gfx1152
- gfx1153
block: SQ
event: 45
- name: SQ_WAIT_INST_ANY
description: Number of wave-cycles spent waiting for any instruction issue. Units in
quad-cycles(4 cycles).
Expand Down Expand Up @@ -6367,6 +6495,41 @@ rocprofiler-sdk:
- gfx950
block: SQ
event: 112
- name: SQ_WAIT_INST_SCA
description: "Number of clock cycles spent waiting for SALU or SMEM instruction issue. In units
of cycles. {nondeterministic, C1}"
properties: []
definitions:
- architectures:
- gfx1150
- gfx1151
- gfx1152
- gfx1153
block: SQ
event: 28
- name: SQ_WAIT_INST_VALU
description: "Number of clock cycles spent waiting for VALU instruction issue. In units of cycles.
{nondeterministic, C1}"
properties: []
definitions:
- architectures:
- gfx1150
- gfx1151
- gfx1152
- gfx1153
block: SQ
event: 27
- name: SQ_WAIT_INST_VMEM
description: "Sum of wait_inst_tex + wait_inst_flat + wait_inst_lds {nondeterministic}"
properties: []
definitions:
- architectures:
- gfx1150
- gfx1151
- gfx1152
- gfx1153
block: SQ
event: 32
- name: SQ_WAVE32_INSTS
description: Number of instructions issued by wave32 waves. Skipped instructions are not
counted. {emulated}
Expand Down Expand Up @@ -7135,6 +7298,40 @@ rocprofiler-sdk:
- architectures:
- gfx90a
expression: 100*reduce(GRBM_SPI_BUSY,max)/reduce(GRBM_GUI_ACTIVE,max)
- name: TA_ADDRESSER_BUSY
description: "Cycles addresser subblock is busy. Perf_Windowing not supported for this counter."
properties: []
definitions:
- architectures:
- gfx1150
- gfx1151
- gfx1152
- gfx1153
block: TA
event: 11
- name: TA_ADDRESSER_FIFO_BUSY
description: "Cycles sampler addresser ping-ping buffer, non-sampler balancing RAM, or aligner state
machine are busy. Perf_Windowing not supported for this counter."
properties: []
definitions:
- architectures:
- gfx1150
- gfx1151
- gfx1152
- gfx1153
block: TA
event: 12
- name: TA_ADDRESSER_STALLED_CYCLES
description: "Cycles Addresser stalled. Perf_Windowing not supported for this counter."
properties: []
definitions:
- architectures:
- gfx1150
- gfx1151
- gfx1152
- gfx1153
block: TA
event: 58
- name: TA_ADDR_STALLED_BY_TC_CYCLES
description: Number of cycles addr path stalled by TC. Perf_Windowing not supported for this
counter.
Expand Down Expand Up @@ -7203,6 +7400,28 @@ rocprofiler-sdk:
- gfx942
- gfx950
expression: reduce(TA_ADDR_STALLED_BY_TD_CYCLES,sum)
- name: TA_ALIGNER_BUSY
description: "Cycles aligner subblock is busy. Perf_Windowing not supported for this counter."
properties: []
definitions:
- architectures:
- gfx1150
- gfx1151
- gfx1152
- gfx1153
block: TA
event: 13
- name: TA_ALIGNER_CYCLES
description: "Cycles issued by the aligner state machine."
properties: []
definitions:
- architectures:
- gfx1150
- gfx1151
- gfx1152
- gfx1153
block: TA
event: 35
- name: TA_BUFFER_ATOMIC_WAVEFRONTS
description: Number of buffer atomic wavefronts processed by TA.
properties: []
Expand Down Expand Up @@ -7690,6 +7909,17 @@ rocprofiler-sdk:
- gfx1031
- gfx1032
expression: reduce(TA_FLAT_STORE_WAVEFRONTS,sum)
- name: TA_FLAT_TOTAL_CYCLES
description: "Flat cycles issued to TC."
properties: []
definitions:
- architectures:
- gfx1150
- gfx1151
- gfx1152
- gfx1153
block: TA
event: 42
- name: TA_FLAT_WAVEFRONTS
description: Number of flat opcode wavfronts processed by the TA.
properties: []
Expand Down Expand Up @@ -7752,6 +7982,28 @@ rocprofiler-sdk:
- gfx942
- gfx950
expression: reduce(TA_FLAT_WRITE_WAVEFRONTS,sum)
- name: TA_STORE_WRITE_DATA_INPUT_CYCLES
description: "Cycles received by write datapath from ta_input for store ops."
properties: []
definitions:
- architectures:
- gfx1150
- gfx1151
- gfx1152
- gfx1153
block: TA
event: 98
- name: TA_STORE_WRITE_DATA_OUTPUT_CYCLES
description: "Cycles sent from write datapath to TC for store ops."
properties: []
definitions:
- architectures:
- gfx1150
- gfx1151
- gfx1152
- gfx1153
block: TA
event: 99
- name: TA_TA_BUSY
description: TA block is busy. Perf_Windowing not supported for this counter.
properties: []
Expand Down Expand Up @@ -7905,6 +8157,39 @@ rocprofiler-sdk:
- architectures:
- gfx950
expression: reduce(TA_FLAT_COALESCEABLE_WAVEFRONTS, sum)
- name: TA_VMEMCMD_CYCLES
description: "Vmemcmd cycles destined for TA. Perf_Windowing not supported for this counter."
properties: []
definitions:
- architectures:
- gfx1150
- gfx1151
- gfx1152
- gfx1153
block: TA
event: 144
- name: TA_VMEMREQ_CYCLES
description: "Vmemreq cycles destined for TA. Perf_Windowing not supported for this counter."
properties: []
definitions:
- architectures:
- gfx1150
- gfx1151
- gfx1152
- gfx1153
block: TA
event: 145
- name: TA_WRITE_PATH_BUSY
description: "Cycles write datapath subblock is busy. Perf_Windowing not supported for this counter."
properties: []
definitions:
- architectures:
- gfx1150
- gfx1151
- gfx1152
- gfx1153
block: TA
event: 14
- name: TX_VCA_ADDR_STALLED_BY_VMW_CYCLES
description: Cycles any row addr path stalled by TC. Perf_windowing not supported for this
counter.
Expand Down Expand Up @@ -10651,6 +10936,13 @@ rocprofiler-sdk:
- gfx950
block: TCP
event: 21
- architectures:
- gfx1150
- gfx1151
- gfx1152
- gfx1153
block: TCP
event: 54
- name: TCP_PENDING_STALL_CYCLES_sum
description: Stall due to data pending from L2. Sum over TCP instances.
properties: []
Expand Down Expand Up @@ -10691,6 +10983,17 @@ rocprofiler-sdk:
- gfx942
- gfx950
expression: reduce(TCP_READ_TAGCONFLICT_STALL_CYCLES,sum)
- name: TCP_TA_REQ
description: Total requests from TA into TCP (L0). Tracks TA aligner throughput.
properties: []
definitions:
- architectures:
- gfx1150
- gfx1151
- gfx1152
- gfx1153
block: TCP
event: 2
- name: TCP_TA_TCP_STATE_READ
description: Number of state reads
properties: []
Expand Down
Loading