Performance data on Ryzen 7 5700G iGPU

4 files changed, 184 insertions, 0 deletions

diff --git a/SampleClips/columbia-large-vega8.txt b/SampleClips/columbia-large-vega8.txt
new file mode 100644
index 0000000..1f78e04
--- /dev/null
+++ b/SampleClips/columbia-large-vega8.txt
@@ -0,0 +1,46 @@
+    CPU Tasks
+LoadModel       7.28719 seconds
+RunComplete     144.925 seconds
+Run     144.859 seconds
+Callbacks       13.7721 milliseconds, 44 calls, 313.002 microseconds average
+Spectrogram     784.186 milliseconds, 41 calls, 19.1265 milliseconds average
+Sample  84.0784 milliseconds, 527 calls, 159.542 microseconds average
+Encode  82.3195 seconds, 9 calls, 9.14661 seconds average
+Decode  62.5221 seconds, 9 calls, 6.9469 seconds average
+DecodeStep      62.4374 seconds, 527 calls, 118.477 milliseconds average
+    GPU Tasks
+LoadModel       5.03066 seconds
+Run     144.417 seconds
+Encode  83.3841 seconds, 9 calls, 9.2649 seconds average
+EncodeLayer     70.9038 seconds, 288 calls, 246.194 milliseconds average
+Decode  61.0329 seconds, 9 calls, 6.78143 seconds average
+DecodeStep      61.0324 seconds, 527 calls, 115.811 milliseconds average
+DecodeLayer     57.3642 seconds, 16864 calls, 3.40158 milliseconds average
+    Compute Shaders
+mulMatTiledEx   64.7225 seconds, 2880 calls, 22.4731 milliseconds average
+mulMatByRowTiled        24.9768 seconds, 166278 calls, 150.211 microseconds average
+mulMatTiled     18.9281 seconds, 3465 calls, 5.46265 milliseconds average
+mulMatByRowTiledEx      15.2747 seconds, 33152 calls, 460.748 microseconds average
+softMaxFixed    4.00524 seconds, 17152 calls, 233.515 microseconds average
+convolutionMain2Fixed   2.65685 seconds, 9 calls, 295.205 milliseconds average
+norm    1.98729 seconds, 51704 calls, 38.4359 microseconds average
+matReshapePanels        1.95102 seconds, 1737 calls, 1.12321 milliseconds average
+addRepeat       1.62049 seconds, 68896 calls, 23.5207 microseconds average
+addRepeatGelu   1.52277 seconds, 17170 calls, 88.6878 microseconds average
+copyConvert     1.0661 seconds, 34880 calls, 30.5649 microseconds average
+fmaRepeat1      1.04679 seconds, 51704 calls, 20.2457 microseconds average
+softMax 812.506 milliseconds, 17391 calls, 46.7199 microseconds average
+scaleInPlace    805.156 milliseconds, 17152 calls, 46.9424 microseconds average
+addInPlace      672.184 milliseconds, 34304 calls, 19.5949 microseconds average
+copyTranspose   634.985 milliseconds, 34304 calls, 18.5105 microseconds average
+addRepeatScale  500.301 milliseconds, 33728 calls, 14.8334 microseconds average
+add     384.464 milliseconds, 16873 calls, 22.7858 microseconds average
+convolutionMain 369.129 milliseconds, 9 calls, 41.0143 milliseconds average
+diagMaskInf     220.886 milliseconds, 16864 calls, 13.0981 microseconds average
+convolutionPrep1        44.2511 milliseconds, 18 calls, 2.45839 milliseconds average
+convolutionPrep2        38.6289 milliseconds, 18 calls, 2.14605 milliseconds average
+addRows 1.8634 milliseconds, 527 calls, 3.53586 microseconds average
+    Memory Usage
+Model   892.591 KB RAM, 2.8815 GB VRAM
+Context 92.2617 MB RAM, 1.27432 GB VRAM
+Total   93.1334 MB RAM, 4.15582 GB VRAM
diff --git a/SampleClips/columbia-medium-vega8.txt b/SampleClips/columbia-medium-vega8.txt
new file mode 100644
index 0000000..7652bd3
--- /dev/null
+++ b/SampleClips/columbia-medium-vega8.txt
@@ -0,0 +1,46 @@
+    CPU Tasks
+LoadModel       1.02959 seconds
+RunComplete     82.2062 seconds
+Run     82.1216 seconds
+Callbacks       9.3817 milliseconds, 37 calls, 253.559 microseconds average
+Spectrogram     727.782 milliseconds, 42 calls, 17.3281 milliseconds average
+Sample  86.0688 milliseconds, 511 calls, 168.432 microseconds average
+Encode  47.9066 seconds, 10 calls, 4.79066 seconds average
+Decode  34.2022 seconds, 10 calls, 3.42022 seconds average
+DecodeStep      34.1156 seconds, 511 calls, 66.7624 milliseconds average
+    GPU Tasks
+LoadModel       521.215 milliseconds
+Run     81.7841 seconds
+Encode  48.6521 seconds, 10 calls, 4.86521 seconds average
+EncodeLayer     41.1815 seconds, 240 calls, 171.589 milliseconds average
+Decode  33.132 seconds, 10 calls, 3.3132 seconds average
+DecodeStep      33.1315 seconds, 511 calls, 64.8366 milliseconds average
+DecodeLayer     29.8613 seconds, 12264 calls, 2.43488 milliseconds average
+    Compute Shaders
+mulMatTiledEx   34.7452 seconds, 2400 calls, 14.4772 milliseconds average
+mulMatByRowTiled        14.1115 seconds, 120741 calls, 116.874 microseconds average
+mulMatTiled     12.7533 seconds, 2890 calls, 4.4129 milliseconds average
+mulMatByRowTiledEx      6.27617 seconds, 24048 calls, 260.985 microseconds average
+softMaxFixed    2.7865 seconds, 12504 calls, 222.849 microseconds average
+convolutionMain2Fixed   1.89585 seconds, 10 calls, 189.585 milliseconds average
+matReshapePanels        1.31263 seconds, 1450 calls, 905.261 microseconds average
+addRepeat       1.11903 seconds, 50256 calls, 22.2666 microseconds average
+addRepeatGelu   949.823 milliseconds, 12524 calls, 75.8403 microseconds average
+softMax 766.059 milliseconds, 12775 calls, 59.9654 microseconds average
+copyConvert     762.917 milliseconds, 25488 calls, 29.9324 microseconds average
+fmaRepeat1      701.015 milliseconds, 37793 calls, 18.5488 microseconds average
+normFixed       672.62 milliseconds, 37793 calls, 17.7975 microseconds average
+addInPlace      543.904 milliseconds, 25008 calls, 21.7492 microseconds average
+scaleInPlace    517.351 milliseconds, 12504 calls, 41.3749 microseconds average
+addRepeatScale  496.454 milliseconds, 24528 calls, 20.2403 microseconds average
+copyTranspose   411.612 milliseconds, 25008 calls, 16.4592 microseconds average
+convolutionMain 338.765 milliseconds, 10 calls, 33.8765 milliseconds average
+add     272.422 milliseconds, 12274 calls, 22.195 microseconds average
+diagMaskInf     144.462 milliseconds, 12264 calls, 11.7793 microseconds average
+convolutionPrep2        60.9698 milliseconds, 20 calls, 3.04849 milliseconds average
+convolutionPrep1        33.0132 milliseconds, 20 calls, 1.65066 milliseconds average
+addRows 1.5961 milliseconds, 511 calls, 3.12348 microseconds average
+    Memory Usage
+Model   877.966 KB RAM, 1.42785 GB VRAM
+Context 91.0721 MB RAM, 893.634 MB VRAM
+Total   91.9295 MB RAM, 2.30054 GB VRAM
diff --git a/SampleClips/jfk-large-vega8.txt b/SampleClips/jfk-large-vega8.txt
new file mode 100644
index 0000000..22c55bb
--- /dev/null
+++ b/SampleClips/jfk-large-vega8.txt
@@ -0,0 +1,46 @@
+    CPU Tasks
+LoadModel       1.7639 seconds
+RunComplete     12.2094 seconds
+Run     12.1434 seconds
+Callbacks       781.9 microseconds, 4 calls, 195.475 microseconds average
+Spectrogram     48.9234 milliseconds, 3 calls, 16.3078 milliseconds average
+Sample  4.7225 milliseconds, 27 calls, 174.907 microseconds average
+Encode  9.21833 seconds
+Decode  2.92391 seconds
+DecodeStep      2.91915 seconds, 27 calls, 108.117 milliseconds average
+    GPU Tasks
+LoadModel       1.15875 seconds
+Run     12.0632 seconds
+Encode  9.30063 seconds
+EncodeLayer     7.88068 seconds, 32 calls, 246.271 milliseconds average
+Decode  2.76256 seconds
+DecodeStep      2.76254 seconds, 27 calls, 102.316 milliseconds average
+DecodeLayer     2.59817 seconds, 864 calls, 3.00715 milliseconds average
+    Compute Shaders
+mulMatTiledEx   6.73307 seconds, 320 calls, 21.0409 milliseconds average
+mulMatTiled     1.4207 seconds, 385 calls, 3.69012 milliseconds average
+mulMatByRowTiled        1.22465 seconds, 8346 calls, 146.735 microseconds average
+mulMatByRowTiledEx      746.587 milliseconds, 1664 calls, 448.67 microseconds average
+softMaxFixed    386.055 milliseconds, 896 calls, 430.865 microseconds average
+convolutionMain2Fixed   278.426 milliseconds
+addRepeat       191.992 milliseconds, 3616 calls, 53.0952 microseconds average
+matReshapePanels        159.243 milliseconds, 193 calls, 825.094 microseconds average
+norm    156.746 milliseconds, 2684 calls, 58.4003 microseconds average
+addRepeatGelu   109.01 milliseconds, 898 calls, 121.392 microseconds average
+copyConvert     102.255 milliseconds, 1856 calls, 55.0943 microseconds average
+fmaRepeat1      92.7262 milliseconds, 2684 calls, 34.5478 microseconds average
+scaleInPlace    90.3307 milliseconds, 896 calls, 100.816 microseconds average
+addInPlace      72.1176 milliseconds, 1792 calls, 40.2442 microseconds average
+copyTranspose   67.2686 milliseconds, 1792 calls, 37.5383 microseconds average
+softMax 55.0367 milliseconds, 891 calls, 61.7696 microseconds average
+addRepeatScale  45.4057 milliseconds, 1728 calls, 26.2764 microseconds average
+convolutionMain 43.7905 milliseconds
+add     33.058 milliseconds, 865 calls, 38.2173 microseconds average
+diagMaskInf     16.9767 milliseconds, 864 calls, 19.649 microseconds average
+convolutionPrep1        5.2747 milliseconds, 2 calls, 2.63735 milliseconds average
+convolutionPrep2        4.6813 milliseconds, 2 calls, 2.34065 milliseconds average
+addRows 80.8 microseconds, 27 calls, 2.99259 microseconds average
+    Memory Usage
+Model   892.591 KB RAM, 2.8815 GB VRAM
+Context 1.98427 MB RAM, 1.13175 GB VRAM
+Total   2.85594 MB RAM, 4.01325 GB VRAM
diff --git a/SampleClips/jfk-medium-vega8.txt b/SampleClips/jfk-medium-vega8.txt
new file mode 100644
index 0000000..070522f
--- /dev/null
+++ b/SampleClips/jfk-medium-vega8.txt
@@ -0,0 +1,46 @@
+    CPU Tasks
+LoadModel       3.693 seconds
+RunComplete     6.3734 seconds
+Run     6.2892 seconds
+Callbacks       436.4 microseconds, 4 calls, 109.1 microseconds average
+Spectrogram     40.3373 milliseconds, 3 calls, 13.4458 milliseconds average
+Sample  4.5591 milliseconds, 28 calls, 162.825 microseconds average
+Encode  4.80886 seconds
+Decode  1.47958 seconds
+DecodeStep      1.47499 seconds, 28 calls, 52.6782 milliseconds average
+    GPU Tasks
+LoadModel       2.13832 seconds
+Run     6.20864 seconds
+Encode  4.82799 seconds
+EncodeLayer     4.05025 seconds, 24 calls, 168.761 milliseconds average
+Decode  1.38065 seconds
+DecodeStep      1.38062 seconds, 28 calls, 49.3078 milliseconds average
+DecodeLayer     1.23477 seconds, 672 calls, 1.83746 milliseconds average
+    Compute Shaders
+mulMatTiledEx   3.25092 seconds, 240 calls, 13.5455 milliseconds average
+mulMatTiled     852.787 milliseconds, 289 calls, 2.95082 milliseconds average
+mulMatByRowTiled        721.566 milliseconds, 6507 calls, 110.891 microseconds average
+mulMatByRowTiledEx      324.607 milliseconds, 1296 calls, 250.469 microseconds average
+softMaxFixed    224.749 milliseconds, 696 calls, 322.915 microseconds average
+convolutionMain2Fixed   179.754 milliseconds
+matReshapePanels        96.6348 milliseconds, 145 calls, 666.447 microseconds average
+addRepeat       89.3072 milliseconds, 2808 calls, 31.8046 microseconds average
+addRepeatGelu   66.8906 milliseconds, 698 calls, 95.8318 microseconds average
+scaleInPlace    60.7223 milliseconds, 696 calls, 87.2447 microseconds average
+copyConvert     53.7138 milliseconds, 1440 calls, 37.3012 microseconds average
+normFixed       48.1435 milliseconds, 2093 calls, 23.0022 microseconds average
+softMax 46.5807 milliseconds, 700 calls, 66.5439 microseconds average
+fmaRepeat1      45.484 milliseconds, 2093 calls, 21.7315 microseconds average
+addInPlace      36.4279 milliseconds, 1392 calls, 26.1695 microseconds average
+convolutionMain 32.6121 milliseconds
+copyTranspose   30.096 milliseconds, 1392 calls, 21.6207 microseconds average
+addRepeatScale  18.6507 milliseconds, 1344 calls, 13.877 microseconds average
+add     11.0739 milliseconds, 673 calls, 16.4545 microseconds average
+convolutionPrep2        5.4643 milliseconds, 2 calls, 2.73215 milliseconds average
+diagMaskInf     4.0519 milliseconds, 672 calls, 6.02961 microseconds average
+convolutionPrep1        3.5447 milliseconds, 2 calls, 1.77235 milliseconds average
+addRows 76.9 microseconds, 28 calls, 2.74643 microseconds average
+    Memory Usage
+Model   877.966 KB RAM, 1.42785 GB VRAM
+Context 1.9836 MB RAM, 771.354 MB VRAM
+Total   2.84099 MB RAM, 2.18113 GB VRAM