MSPI throughput performance test

Overview

This sample demonstrates MSPI throughput performance testing by measuring transfer rates across different buffer sizes and packet configurations. The sample performs a series of tests to determine optimal MSPI transfer parameters for maximum throughput and efficiency.

The sample tests: * Multiple buffer sizes: 4, 16, 64, 256, 1024, 1500, and 2304 bytes * Multiple packet counts: 1, 4, and 16 packets per transfer * Total transfer size: 32 KB per test case * Performance metrics: throughput (MB/s), efficiency (%), CPU load (%), and latency.

NOTE: The latency is calculated based on the efficiency loss and is an estimated value and is not printed in the summary table, it is printed in the detailed results for each test case.

The sample supports both DMA and CPU-driven polling modes. When CONFIG_CPU_LOAD is enabled, it reports CPU load (percentage) using the CPU load subsystem (supported on Cortex-M, RISC-V, and Cortex-A; single-core only).

Goals and future work

The current sample measures controller-side throughput and efficiency so that you can see how the driver performs with different packet sizes and packet counts, and compare DMA vs FIFO and CPU utilization. This is useful for characterizing any MSPI controller regardless of what is on the other end (flash, slave, or host).

In the future, the sample may be extended so that two SoCs run it—one in controller mode and one in peripheral mode—with a simple protocol to measure full link throughput. That work depends on a defined protocol and MSPI peripheral support.

Configuration

The sample is configured for:

MSPI device parameters (clock frequency, I/O mode, data rate, endianness, command/address length, etc.) come from devicetree on the target device node via MSPI_DEVICE_CONFIG_DT(). See the board overlay for the specific values.

Total transfer size per test case is set by Kconfig CONFIG_SAMPLE_MSPI_THROUGHPUT_TOTAL_TRANSFER_SIZE (default: 32 KB).

Hardware Setup

This sample requires a board overlay that defines alias dev0 to the MSPI target device node (e.g. a dummy or peripheral device with reg = <0> under the controller). The sample obtains the controller as the bus of dev0 and uses MSPI_DEVICE_ID_DT(DT_ALIAS(dev0)) for the device index and chip-select.

Building and Running

west build -b nrf7120dk_nrf7120_cpuapp samples/drivers/mspi/mspi_throughput
west build -t Test
west build -t MSPI
west build -t throughput

Sample Output

The sample outputs detailed performance metrics for each test configuration.

Example output of running nRF7120 at 32MHz clock for 32KB total transfer size:

=== TEST RESULTS === BufSize PktCnt Time Throughput Efficiency CPU Load (bytes) (us) (MB/s) (%) (%) ——————————————————————- 4 1 87432 0.536 3.51 100.00 4 4 79175 0.444 2.91 100.00 4 16 76135 0.423 2.77 100.00 16 1 31622 1.112 7.29 100.00 16 4 28646 1.125 7.37 100.00 16 16 28037 1.123 7.36 100.00 64 1 7910 4.074 26.70 99.99 64 4 7158 4.400 28.83 99.99 64 16 6956 4.501 29.50 99.99 256 1 3610 8.724 57.17 71.36 256 4 3460 9.049 59.31 69.28 256 16 3424 9.131 59.84 68.87 1024 1 2441 12.827 84.06 26.50 1024 4 2403 13.011 85.27 25.00 1024 16 2393 13.061 85.59 24.63 1500 1 2319 13.494 88.43 19.20 1500 4 2295 13.622 89.27 18.15 1500 16 2289 13.654 89.48 17.89 2304 1 2234 14.001 91.76 13.64 2304 4 2218 14.093 92.36 12.83 2304 16 2212 14.128 92.59 12.55

=== SUMMARY REPORT === Best TX Performance: 14.13 MB/s (buffer size=2304, packet count=16) === MSPI Throughput Test Complete ===

The output includes: * BufSize: Buffer size in bytes used for each transfer * PktCnt: Number of packets per transfer * Time: Total transfer time in microseconds (CPU cycles) * Throughput: Measured throughput in MB/s * Efficiency: Percentage of ideal throughput achieved * CPU Load: CPU utilization percentage (when CONFIG_CPU_LOAD is enabled)

The summary report identifies the configuration that achieved the highest throughput.

Performance Analysis

The sample calculates several performance metrics:

Throughput: Actual data transfer rate in MB/s, calculated from total bytes transferred (including command and address overhead) divided by transfer time.
Efficiency: Percentage of ideal throughput achieved. Ideal throughput is calculated as (clock_frequency * 4) / 8 for quad mode, accounting for 4 data lines and 8 bits per byte.
CPU Load: When CONFIG_CPU_LOAD is enabled, the sample uses the CPU load subsystem to measure utilization during transfers (architecture-agnostic where supported).
Estimated Latency: Calculated per-buffer overhead based on efficiency loss, helping identify optimal buffer sizes.

The sample helps identify: * Optimal buffer sizes for maximum throughput * Impact of packet batching on performance * CPU overhead of different transfer configurations