#0: Add comments about perf tests on run_paged_tests.sh

tests/tt_metal/tt_metal/perf_microbenchmark/dispatch/run_paged_tests.sh

@@ -128,17 +128,39 @@ done
 # PERF                                        #
 ###############################################
 
+
 DIR="TT_METAL_SLOW_DISPATCH_MODE=1 ${TT_METAL_HOME}/paged_write_tests_sanity_perf_dram_${SUFFIX}"
 mkdir -p $DIR
 
 # 3.3 GB/s whb0
 TT_METAL_SLOW_DISPATCH_MODE=1 ${TT_METAL_HOME}/build/test/tt_metal/perf_microbenchmark/dispatch/test_dispatcher -w 1000 -t 2 -wx 0 -wy 1 -min 1024 -max 1024 -lps 10 -pbs 2 -np 128 -c -i 1000 |& tee ${DIR}/perf_write_128_page_1024b_size_dispatch_buffer_1024b_pages_1000_iter_dram_pbs2.log
+# BW: 3.845 GB/s
+TT_METAL_SLOW_DISPATCH_MODE=1 build/test/tt_metal/perf_microbenchmark/dispatch/test_dispatcher -w 1 -t 2 -wx 0 -wy 1 -min 1024 -max 1024 -lps 10 -pbs 2 -np 128 -c -i 1 -pi 10000 |& tee perf_1kb_pi10000_warm_1.log
 
 # 5.7 GB/s whb0
 TT_METAL_SLOW_DISPATCH_MODE=1 ${TT_METAL_HOME}/build/test/tt_metal/perf_microbenchmark/dispatch/test_dispatcher -w 1000 -t 2 -wx 0 -wy 1 -min 2048 -max 2048 -lps 11 -pbs 2 -np 128 -c -i 1000 |& tee ${DIR}/perf_write_128_page_2048b_size_dispatch_buffer_2048b_pages_1000_iter_dram_pbs2.log
+# BW: 6.374 GB/s with device iterations.
+TT_METAL_SLOW_DISPATCH_MODE=1 ./build/test/tt_metal/perf_microbenchmark/dispatch/test_dispatcher -w 1 -t 2 -wx 0 -wy 1 -min 2048 -max 2048 -lps 11 -pbs 2 -np 128 -c -i 1 -pi 10000 |& tee perf_2kb_pi10000_warm_1.log
 
 # 8.6 GB/s whb0
 TT_METAL_SLOW_DISPATCH_MODE=1 ${TT_METAL_HOME}/build/test/tt_metal/perf_microbenchmark/dispatch/test_dispatcher -w 1000 -t 2 -wx 0 -wy 1 -min 4096 -max 4096 -lps 12 -pbs 2 -np 128 -c -i 1000 |& tee ${DIR}/perf_write_128_page_4096b_size_dispatch_buffer_4096b_pages_1000_iter_dram_pbs2.log
+# BW: 9.600 GB/s
+TT_METAL_SLOW_DISPATCH_MODE=1 ./build/test/tt_metal/perf_microbenchmark/dispatch/test_dispatcher -w 1 -t 2 -wx 0 -wy 1 -min 4096 -max 4096 -lps 12 -pbs 2 -np 128 -c -i 1 -pi 10000 |& tee perf_4kb_pi10000_warm_1.log
+
+# 11.7 GB/s whb0
+TT_METAL_SLOW_DISPATCH_MODE=1 ${TT_METAL_HOME}/build/test/tt_metal/perf_microbenchmark/dispatch/test_dispatcher -w 1000 -t 2 -wx 0 -wy 1 -min 8192 -max 8192 -lps 13 -pbs 2 -np 128 -c -i 1000 |& tee ${DIR}/perf_write_128_page_8192b_size_dispatch_buffer_8192b_pages_1000_iter_dram_pbs2.log
+
+
+# Hangs.... why? I was seeing mismatches at one point.
+TT_METAL_SLOW_DISPATCH_MODE=1 ./build/test/tt_metal/perf_microbenchmark/dispatch/test_dispatcher -w 1 -t 2 -wx 0 -wy 1 -min 8192 -max 8192 -lps 13 -pbs 2 -np 128 -c -i 1 -pi 10000 |& tee perf_8kb_pi10000_warm_1.log
+TT_METAL_SLOW_DISPATCH_MODE=1 ./build/test/tt_metal/perf_microbenchmark/dispatch/test_dispatcher -w 1 -t 2 -wx 0 -wy 1 -min 8192 -max 8192 -lps 13 -pbs 2 -np 128 -c -i 1 -pi 100 |& tee perf_8kb_pi100_warm_1.log
+# pi 10 passes
+# pi 100 hangs.
+
+# Mismatches
+TT_METAL_SLOW_DISPATCH_MODE=1 ./build/test/tt_metal/perf_microbenchmark/dispatch/test_dispatcher -w 0 -t 2 -wx 0 -wy 1 -min 8192 -max 8192 -lps 13 -pbs 2 -np 128 -c -i 1 -pi 20 |& tee perf_8kb_pi20_warm_0.log
+TT_METAL_SLOW_DISPATCH_MODE=1 ./build/test/tt_metal/perf_microbenchmark/dispatch/test_dispatcher -w 0 -t 2 -wx 0 -wy 1 -min 8192 -max 8192 -lps 13 -pbs 2 -np 128 -c -i 1 -pi 11 |& tee perf_8kb_pi11_warm_0.log
+# Passes:
+TT_METAL_LOGGER_LEVEL=Debug TT_METAL_SLOW_DISPATCH_MODE=1 ./build/test/tt_metal/perf_microbenchmark/dispatch/test_dispatcher -w 0 -t 2 -wx 0 -wy 1 -min 8192 -max 8192 -lps 13 -pbs 2 -np 128 -c -i 1 -pi 10 |& tee perf_8kb_pi10_warm_0.log
 
-# 11.7 GB/s whb0 - DRAM.   Have to reduce number of pages to not exceed 1MB L1 for GS.
-TT_METAL_SLOW_DISPATCH_MODE=1 ${TT_METAL_HOME}/build/test/tt_metal/perf_microbenchmark/dispatch/test_dispatcher -w 1000 -t 2 -wx 0 -wy 1 -min 8192 -max 8192 -lps 13 -pbs 2 -np 100 -c -i 1000 |& tee ${DIR}/perf_write_128_page_8192b_size_dispatch_buffer_8192b_pages_1000_iter_dram_pbs2.log
+# Theory - kernel bug somewhere, stuff getting corrupted. See if same results on latest.