diff --git a/source/devices/AM62AX/linux/RT_Linux_Performance_Guide.rst b/source/devices/AM62AX/linux/RT_Linux_Performance_Guide.rst index 7c65ee6bf..020eaccfa 100644 --- a/source/devices/AM62AX/linux/RT_Linux_Performance_Guide.rst +++ b/source/devices/AM62AX/linux/RT_Linux_Performance_Guide.rst @@ -57,14 +57,22 @@ Test commands used for running stress-ng and cyclictest together stress-ng --cpu-method=all -c 4 & cyclictest -m -Sp80 -D6h -h400 -i200 -M -q -The latencies where observed with this SDK are summarized below: +The latencies observed with the default yocto SDK are summarized below: + +.. note:: + + A known issue in this SDK release is affecting this benchmark. + Applying `this patch`_ on Yocto's meta-ti layer will restore the + context switch latencies to the values you see below: .. csv-table:: :header: "Latencies","CPU 0","CPU 1","CPU 2","CPU 3" "Minimum (us)","5","5","5","5" - "Average (us)","5","6","6","5" - "Maximum (us)","68","31","34","28" + "Average (us)","6","6","6","6" + "Maximum (us)","31","29","30","30" .. image:: img/rt-cpu-method-all-latency-histogram.png + +.. _this patch: https://lists.yoctoproject.org/g/meta-ti/message/19089 diff --git a/source/devices/AM62AX/linux/img/rt-cpu-method-all-latency-histogram.png b/source/devices/AM62AX/linux/img/rt-cpu-method-all-latency-histogram.png index e2f0c31c0..57d67a854 100644 Binary files a/source/devices/AM62AX/linux/img/rt-cpu-method-all-latency-histogram.png and b/source/devices/AM62AX/linux/img/rt-cpu-method-all-latency-histogram.png differ diff --git a/source/devices/AM62LX/linux/RT_Linux_Performance_Guide.rst b/source/devices/AM62LX/linux/RT_Linux_Performance_Guide.rst index 9f47d9972..1dbe0921e 100644 --- a/source/devices/AM62LX/linux/RT_Linux_Performance_Guide.rst +++ b/source/devices/AM62LX/linux/RT_Linux_Performance_Guide.rst @@ -56,13 +56,21 @@ Test commands used for running stress-ng and cyclictest together stress-ng --cpu-method=all -c 4 & cyclictest -m -Sp80 -D6h -h400 -i200 -M -q -The latencies observed with this SDK are summarized below: +The latencies observed with the default yocto SDK are summarized below: + +.. note:: + + A known issue in this SDK release is affecting this benchmark. + Applying `this patch`_ on Yocto's meta-ti layer will restore the + context switch latencies to the values you see below: .. csv-table:: :header: "Latencies","CPU 0","CPU 1" "Minimum (us)","5","5" - "Average (us)","6","8" - "Maximum (us)","47","53" + "Average (us)","8","7" + "Maximum (us)","57","48" .. image:: img/rt-cpu-method-all-latency-histogram.png + +.. _this patch: https://lists.yoctoproject.org/g/meta-ti/message/19089 diff --git a/source/devices/AM62LX/linux/img/rt-cpu-method-all-latency-histogram.png b/source/devices/AM62LX/linux/img/rt-cpu-method-all-latency-histogram.png index f1e689e0a..942afe78d 100644 Binary files a/source/devices/AM62LX/linux/img/rt-cpu-method-all-latency-histogram.png and b/source/devices/AM62LX/linux/img/rt-cpu-method-all-latency-histogram.png differ diff --git a/source/devices/AM62PX/linux/RT_Linux_Performance_Guide.rst b/source/devices/AM62PX/linux/RT_Linux_Performance_Guide.rst index 780c34920..ce7e5b6ff 100644 --- a/source/devices/AM62PX/linux/RT_Linux_Performance_Guide.rst +++ b/source/devices/AM62PX/linux/RT_Linux_Performance_Guide.rst @@ -57,13 +57,21 @@ Test commands used for running stress-ng and cyclictest together stress-ng --cpu-method=all -c 4 & cyclictest -m -Sp80 -D6h -h400 -i200 -M -q -The latencies observed with this SDK are summarized below: +The latencies observed with the default yocto SDK are summarized below: + +.. note:: + + A known issue in this SDK rlease is affecting this benchmark. + Applying `this patch`_ on Yocto's meta-ti layer will restore the + context switch latencies to the values you see below. .. csv-table:: :header: "Latencies","CPU 0","CPU 1","CPU 2","CPU 3" "Minimum (usec)","5","5","5","5" - "Average (usec)","6","6","5","6" - "Maximum (usec)","30","33","31","35" + "Average (usec)","6","6","7","6" + "Maximum (usec)","36","34","37","34" .. image:: img/rt-cpu-method-all-latency-histogram.png + +.. _this patch: https://lists.yoctoproject.org/g/meta-ti/message/19089 diff --git a/source/devices/AM62PX/linux/img/rt-cpu-method-all-latency-histogram.png b/source/devices/AM62PX/linux/img/rt-cpu-method-all-latency-histogram.png index 6433c5b37..35abdd7db 100644 Binary files a/source/devices/AM62PX/linux/img/rt-cpu-method-all-latency-histogram.png and b/source/devices/AM62PX/linux/img/rt-cpu-method-all-latency-histogram.png differ diff --git a/source/devices/AM62X/linux/RT_Linux_Performance_Guide.rst b/source/devices/AM62X/linux/RT_Linux_Performance_Guide.rst index 35de6053f..ed8fafc48 100644 --- a/source/devices/AM62X/linux/RT_Linux_Performance_Guide.rst +++ b/source/devices/AM62X/linux/RT_Linux_Performance_Guide.rst @@ -61,13 +61,22 @@ Test commands used for running stress-ng and cyclictest together stress-ng --cpu-method=all -c 4 & cyclictest -m -Sp80 -D6h -h400 -i200 -M -q -The latencies observed with this SDK are summarized below: +The latencies observed with the default yocto SDK image on +https://www.ti.com/tool/SK-AM62B-P1 are summarized below: + +.. note:: + + A known issue in this SDK release is affecting this benchmark. + Applying `this patch`_ on Yocto's meta-ti layer will restore the + context switch latencies to the values you see below. .. csv-table:: :header: "Latencies","CPU 0","CPU 1","CPU 2","CPU 3" - "Minimum (usec)","5","5","4","4" - "Average (usec)","7","5","6","6" - "Maximum (usec)","66","41","73","70" + "Minimum (usec)","5","5","5","5" + "Average (usec)","6","6","6","6" + "Maximum (usec)","62","46","47","52" .. image:: img/rt-cpu-method-all-latency-histogram.png + +.. _this patch: https://lists.yoctoproject.org/g/meta-ti/message/19089 diff --git a/source/devices/AM62X/linux/img/rt-cpu-method-all-latency-histogram.png b/source/devices/AM62X/linux/img/rt-cpu-method-all-latency-histogram.png index a36483b58..fb0c4ae0b 100644 Binary files a/source/devices/AM62X/linux/img/rt-cpu-method-all-latency-histogram.png and b/source/devices/AM62X/linux/img/rt-cpu-method-all-latency-histogram.png differ diff --git a/source/devices/AM64X/linux/RT_Linux_Performance_Guide.rst b/source/devices/AM64X/linux/RT_Linux_Performance_Guide.rst index 1c0429250..6f5319f14 100644 --- a/source/devices/AM64X/linux/RT_Linux_Performance_Guide.rst +++ b/source/devices/AM64X/linux/RT_Linux_Performance_Guide.rst @@ -60,15 +60,23 @@ Test commands used for running stress-ng and cyclictest together The latencies observed with this DK are summarized below: +.. note:: + + A known issue in this SDK release is affecting this benchmark. + Applying `this patch`_ on Yocto's meta-ti layer will restore the + context switch latencies to the values you see below. + .. csv-table:: :header: "Latencies","CPU 0","CPU 1" "Minimum (usec)","6","6" - "Average (usec)","8","8" - "Maximum (usec)","77","105" + "Average (usec)","10","9" + "Maximum (usec)","77","52" .. image:: img/rt-cpu-method-all-latency-histogram.png +.. _this patch: https://lists.yoctoproject.org/g/meta-ti/message/19089 + | LMBench @@ -791,8 +799,8 @@ RP Message Inter-Processor Communication (IPC) Latency RP Message latency Performance ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ -RP Message latency is the delay measured from sending a round trip echo message from -a Linux application to a remote processor and back. The following measurements use +RP Message latency is the delay measured from sending a round trip echo message from +a Linux application to a remote processor and back. The following measurements use a RP message length of 1 byte and 490 bytes for comparison. The Linux user space application `rpmsg_char_benchmark `__ captures these latency values. diff --git a/source/devices/AM64X/linux/img/rt-cpu-method-all-latency-histogram.png b/source/devices/AM64X/linux/img/rt-cpu-method-all-latency-histogram.png index 54a41d8d5..eff07aff2 100644 Binary files a/source/devices/AM64X/linux/img/rt-cpu-method-all-latency-histogram.png and b/source/devices/AM64X/linux/img/rt-cpu-method-all-latency-histogram.png differ