fix max BF16 flop rate on CDNA2

Signed-off-by: Nicholas Curtis <nicurtis@amd.com>
Šī revīzija ir iekļauta:
Nicholas Curtis committed 2023-07-28 12:38:11 -04:00
1 vecāks 6487ba3853
revīzija 42248d5391
2 files changed
+3 -3

No files matched your search

+1 -1
Parādīt failu
@@ -339,7 +339,7 @@
"hide": false,
"rawQuery": true,
"refId": "A",
"target": "${Workload1}.pmc_perf.aggregate([\n {\"$match\": {\n \"Index\": { \"$in\": [${DispatchIDFilter:raw}] },\n \"gpu-id\": { \"$in\": [${gpuFilter:raw}] },\n \"KernelName\": { \"$in\": ${KernelNameFilter:json}}\n }},\n {\"$group\": {\n \"_id\": null,\n \"valu_flops_val\": {\n \"$avg\": { \"$divide\": [ { \"$add\": [\n { \"$multiply\": [64, { \"$add\": [\"&SQ_INSTS_VALU_ADD_F16\", \"&SQ_INSTS_VALU_MUL_F16\", \"&SQ_INSTS_VALU_TRANS_F16\", { \"$multiply\": [2, \"&SQ_INSTS_VALU_FMA_F16\"] }] }] },\n { \"$multiply\": [64, { \"$add\": [\"&SQ_INSTS_VALU_ADD_F32\", \"&SQ_INSTS_VALU_MUL_F32\", \"&SQ_INSTS_VALU_TRANS_F32\", { \"$multiply\": [2, \"&SQ_INSTS_VALU_FMA_F32\"] }] }] },\n { \"$multiply\": [64, { \"$add\": [\"&SQ_INSTS_VALU_ADD_F64\", \"&SQ_INSTS_VALU_MUL_F64\", \"&SQ_INSTS_VALU_TRANS_F64\", { \"$multiply\": [2, \"&SQ_INSTS_VALU_FMA_F64\"] }] }] }\n ]}\n ,{ \"$subtract\": [\"&EndNs\", \"&BeginNs\"] }]\n }},\n \n \"valu_intOps_val\": {\n \"$avg\": { \"$divide\": [{ \"$multiply\": [64, { \"$add\": [\"&SQ_INSTS_VALU_INT32\", \"&SQ_INSTS_VALU_INT64\"] }] }, { \"$subtract\": [\"&EndNs\", \"&BeginNs\"] }] }\n },\n \n \"mfma_flops_f16_val\": { \"$avg\": { \"$divide\": [{ \"$multiply\": [\"&SQ_INSTS_VALU_MFMA_MOPS_F16\", 512] }, { \"$subtract\": [\"&EndNs\", \"&BeginNs\"] }] } },\n \"mfma_flops_bf16_val\": { \"$avg\": { \"$divide\": [{ \"$multiply\": [\"&SQ_INSTS_VALU_MFMA_MOPS_BF16\", 512] }, { \"$subtract\": [\"&EndNs\", \"&BeginNs\"] }] } },\n \"mfma_flops_f32_val\": { \"$avg\": { \"$divide\": [{ \"$multiply\": [\"&SQ_INSTS_VALU_MFMA_MOPS_F32\", 512] }, { \"$subtract\": [\"&EndNs\", \"&BeginNs\"] }] } },\n \"mfma_flops_f64_val\": { \"$avg\": { \"$divide\": [{ \"$multiply\": [\"&SQ_INSTS_VALU_MFMA_MOPS_F64\", 512] }, { \"$subtract\": [\"&EndNs\", \"&BeginNs\"] }] } },\n \"mfma_flops_i8_val\": { \"$avg\": { \"$divide\": [{ \"$multiply\": [\"&SQ_INSTS_VALU_MFMA_MOPS_I8\", 512] }, { \"$subtract\": [\"&EndNs\", \"&BeginNs\"] }] } },\n \n \n \"salu_val\": {\n \"$avg\": { \"$divide\": [{ \"$multiply\": [100, \"&SQ_ACTIVE_INST_SCA\"] }, { \"$multiply\": [\"&GRBM_GUI_ACTIVE\", $numCU] }] }\n },\n \n \"valu_val\": {\n \"$avg\": { \"$divide\": [{ \"$multiply\": [100, \"&SQ_ACTIVE_INST_VALU\"] }, { \"$multiply\": [\"&GRBM_GUI_ACTIVE\", $numCU] }] }\n },\n \n \"mfma_val\": {\n \"$avg\": { \"$divide\": [{ \"$multiply\": [100, \"&SQ_VALU_MFMA_BUSY_CYCLES\"] }, { \"$multiply\": [\"&GRBM_GUI_ACTIVE\", $numCU, 4] }] }\n },\n\n \n \"lds_bconf\": {\n \"$avg\": {\"$cond\": [ {\"$ne\": [{\"$subtract\": [\"&SQ_LDS_IDX_ACTIVE\", \"&SQ_LDS_BANK_CONFLICT\"]}, 0]},\n {\"$divide\": [\n \"&SQ_LDS_BANK_CONFLICT\",\n {\"$subtract\": [\"&SQ_LDS_IDX_ACTIVE\", \"&SQ_LDS_BANK_CONFLICT\"]}\n ]},\n null\n ]\n } \n },\n \n \n \"lds_bw\": {\n \"$avg\": {\"$divide\":[ \n { \"$multiply\": [ { \"$subtract\": [\"&SQ_LDS_IDX_ACTIVE\", \"&SQ_LDS_BANK_CONFLICT\"]}, \n 4, \n {\"$toInt\": \"$L2Banks\"}\n ]}, \n {\"$subtract\": [\"&EndNs\", \"&BeginNs\"]}\n ]}\n },\n \n \"lds_bw_pop\": {\n \"$avg\": {\"$divide\": [\n {\"$divide\":[ \n { \"$multiply\": [ { \"$subtract\": [\"&SQ_LDS_IDX_ACTIVE\", \"&SQ_LDS_BANK_CONFLICT\"]}, \n 4, \n {\"$toInt\": \"$L2Banks\"}\n ]}, \n {\"$subtract\": [\"&EndNs\", \"&BeginNs\"]}\n ]}, \n {\"$multiply\": [$sclk, $numCU, 0.00128]}\n ]}\n },\n \n \"unpredthreads_val\": {\n \"$avg\": {\n \"$cond\": [\n {\"$ne\":[\"&SQ_ACTIVE_INST_VALU\" , 0]},\n { \"$divide\": [\"&SQ_THREAD_CYCLES_VALU\", \"&SQ_ACTIVE_INST_VALU\"] },\n null\n ]\n }\n },\n \n \"ipcIssue_val\": {\n \"$avg\": { \"$divide\": [{ \"$add\": [\"&SQ_INSTS_VALU\", \"&SQ_INSTS_VMEM\", \"&SQ_INSTS_SALU\", \"&SQ_INSTS_SMEM\", \"&SQ_INSTS_GDS\", \"&SQ_INLine truncated
"target": "${Workload1}.pmc_perf.aggregate([\n {\"$match\": {\n \"Index\": { \"$in\": [${DispatchIDFilter:raw}] },\n \"gpu-id\": { \"$in\": [${gpuFilter:raw}] },\n \"KernelName\": { \"$in\": ${KernelNameFilter:json}}\n }},\n {\"$group\": {\n \"_id\": null,\n \"valu_flops_val\": {\n \"$avg\": { \"$divide\": [ { \"$add\": [\n { \"$multiply\": [64, { \"$add\": [\"&SQ_INSTS_VALU_ADD_F16\", \"&SQ_INSTS_VALU_MUL_F16\", \"&SQ_INSTS_VALU_TRANS_F16\", { \"$multiply\": [2, \"&SQ_INSTS_VALU_FMA_F16\"] }] }] },\n { \"$multiply\": [64, { \"$add\": [\"&SQ_INSTS_VALU_ADD_F32\", \"&SQ_INSTS_VALU_MUL_F32\", \"&SQ_INSTS_VALU_TRANS_F32\", { \"$multiply\": [2, \"&SQ_INSTS_VALU_FMA_F32\"] }] }] },\n { \"$multiply\": [64, { \"$add\": [\"&SQ_INSTS_VALU_ADD_F64\", \"&SQ_INSTS_VALU_MUL_F64\", \"&SQ_INSTS_VALU_TRANS_F64\", { \"$multiply\": [2, \"&SQ_INSTS_VALU_FMA_F64\"] }] }] }\n ]}\n ,{ \"$subtract\": [\"&EndNs\", \"&BeginNs\"] }]\n }},\n \n \"valu_intOps_val\": {\n \"$avg\": { \"$divide\": [{ \"$multiply\": [64, { \"$add\": [\"&SQ_INSTS_VALU_INT32\", \"&SQ_INSTS_VALU_INT64\"] }] }, { \"$subtract\": [\"&EndNs\", \"&BeginNs\"] }] }\n },\n \n \"mfma_flops_f16_val\": { \"$avg\": { \"$divide\": [{ \"$multiply\": [\"&SQ_INSTS_VALU_MFMA_MOPS_F16\", 512] }, { \"$subtract\": [\"&EndNs\", \"&BeginNs\"] }] } },\n \"mfma_flops_bf16_val\": { \"$avg\": { \"$divide\": [{ \"$multiply\": [\"&SQ_INSTS_VALU_MFMA_MOPS_BF16\", 512] }, { \"$subtract\": [\"&EndNs\", \"&BeginNs\"] }] } },\n \"mfma_flops_f32_val\": { \"$avg\": { \"$divide\": [{ \"$multiply\": [\"&SQ_INSTS_VALU_MFMA_MOPS_F32\", 512] }, { \"$subtract\": [\"&EndNs\", \"&BeginNs\"] }] } },\n \"mfma_flops_f64_val\": { \"$avg\": { \"$divide\": [{ \"$multiply\": [\"&SQ_INSTS_VALU_MFMA_MOPS_F64\", 512] }, { \"$subtract\": [\"&EndNs\", \"&BeginNs\"] }] } },\n \"mfma_flops_i8_val\": { \"$avg\": { \"$divide\": [{ \"$multiply\": [\"&SQ_INSTS_VALU_MFMA_MOPS_I8\", 512] }, { \"$subtract\": [\"&EndNs\", \"&BeginNs\"] }] } },\n \n \n \"salu_val\": {\n \"$avg\": { \"$divide\": [{ \"$multiply\": [100, \"&SQ_ACTIVE_INST_SCA\"] }, { \"$multiply\": [\"&GRBM_GUI_ACTIVE\", $numCU] }] }\n },\n \n \"valu_val\": {\n \"$avg\": { \"$divide\": [{ \"$multiply\": [100, \"&SQ_ACTIVE_INST_VALU\"] }, { \"$multiply\": [\"&GRBM_GUI_ACTIVE\", $numCU] }] }\n },\n \n \"mfma_val\": {\n \"$avg\": { \"$divide\": [{ \"$multiply\": [100, \"&SQ_VALU_MFMA_BUSY_CYCLES\"] }, { \"$multiply\": [\"&GRBM_GUI_ACTIVE\", $numCU, 4] }] }\n },\n\n \n \"lds_bconf\": {\n \"$avg\": {\"$cond\": [ {\"$ne\": [{\"$subtract\": [\"&SQ_LDS_IDX_ACTIVE\", \"&SQ_LDS_BANK_CONFLICT\"]}, 0]},\n {\"$divide\": [\n \"&SQ_LDS_BANK_CONFLICT\",\n {\"$subtract\": [\"&SQ_LDS_IDX_ACTIVE\", \"&SQ_LDS_BANK_CONFLICT\"]}\n ]},\n null\n ]\n } \n },\n \n \n \"lds_bw\": {\n \"$avg\": {\"$divide\":[ \n { \"$multiply\": [ { \"$subtract\": [\"&SQ_LDS_IDX_ACTIVE\", \"&SQ_LDS_BANK_CONFLICT\"]}, \n 4, \n {\"$toInt\": \"$L2Banks\"}\n ]}, \n {\"$subtract\": [\"&EndNs\", \"&BeginNs\"]}\n ]}\n },\n \n \"lds_bw_pop\": {\n \"$avg\": {\"$divide\": [\n {\"$divide\":[ \n { \"$multiply\": [ { \"$subtract\": [\"&SQ_LDS_IDX_ACTIVE\", \"&SQ_LDS_BANK_CONFLICT\"]}, \n 4, \n {\"$toInt\": \"$L2Banks\"}\n ]}, \n {\"$subtract\": [\"&EndNs\", \"&BeginNs\"]}\n ]}, \n {\"$multiply\": [$sclk, $numCU, 0.00128]}\n ]}\n },\n \n \"unpredthreads_val\": {\n \"$avg\": {\n \"$cond\": [\n {\"$ne\":[\"&SQ_ACTIVE_INST_VALU\" , 0]},\n { \"$divide\": [\"&SQ_THREAD_CYCLES_VALU\", \"&SQ_ACTIVE_INST_VALU\"] },\n null\n ]\n }\n },\n \n \"ipcIssue_val\": {\n \"$avg\": { \"$divide\": [{ \"$add\": [\"&SQ_INSTS_VALU\", \"&SQ_INSTS_VMEM\", \"&SQ_INSTS_SALU\", \"&SQ_INSTS_SMEM\", \"&SQ_INSTS_GDS\", \"&SQ_INLine truncated
"type": "table"
},
{
@@ -45,9 +45,9 @@ Panel Config:
MFMA FLOPs (BF16):
value: AVG(((SQ_INSTS_VALU_MFMA_MOPS_BF16 * 512) / (EndNs - BeginNs)))
unit: GFLOP
peak: ((($sclk * $numCU) * 512) / 1000)
peak: ((($sclk * $numCU) * 1024) / 1000)
pop: ((100 * AVG(((SQ_INSTS_VALU_MFMA_MOPS_BF16 * 512) / (EndNs - BeginNs))))
/ ((($sclk * $numCU) * 512) / 1000))
/ ((($sclk * $numCU) * 1024) / 1000))
tips:
MFMA FLOPs (F16):
value: AVG(((SQ_INSTS_VALU_MFMA_MOPS_F16 * 512) / (EndNs - BeginNs)))