[uart] Share the fake UART, move the reader benchmarks to core and drive Modbus through loop()

script/cpp_benchmark.py

@@ -25,6 +25,9 @@ CORE_BENCHMARKS_DIR: Path = Path(root_path) / "tests" / "benchmarks" / "core"
 # allow benchmarks to compile on the host platform.
 STUBS_DIR: Path = Path(root_path) / "tests" / "benchmarks" / "stubs"
 
+# Shared test doubles, reached by -I so only the benchmarks that include them pull in their headers
+HELPERS_DIR: Path = Path(root_path) / "tests" / "benchmarks" / "helpers"
+
 BUILD_FLAGS = [
     "-Os",  # match firmware optimization level (detects inlining regressions)
     "-g",  # debug symbols for profiling
@@ -32,6 +35,7 @@ BUILD_FLAGS = [
     "-fdata-sections",  # required for dead-code stripping with -Os
     "-DUSE_BENCHMARK",  # disable WarnIfComponentBlockingGuard in finish()
     f"-I{STUBS_DIR}",  # stub headers for ESP32-only components
+    f"-I{HELPERS_DIR}",  # shared test doubles such as fake_uart.h
 ]
 
 

tests/benchmarks/components/ld2450/bench_ld2450_frame.cpp

@@ -0,0 +1,76 @@
+#include <benchmark/benchmark.h>
+
+#include "esphome/components/binary_sensor/binary_sensor.h"
+#include "esphome/components/ld2450/ld2450.h"
+#include "esphome/components/sensor/sensor.h"
+#include "fake_uart.h"
+
+namespace esphome::ld2450::benchmarks {
+
+using esphome::benchmarks::FakeUART;
+
+// Inner iteration count to amortize CodSpeed instrumentation overhead.
+static constexpr int kInnerIterations = 2000;
+static constexpr size_t FRAME_SIZE = 30;
+
+// Periodic frames with one moving target at two positions
+static constexpr uint8_t FRAME_A[FRAME_SIZE] = {0xAA, 0xFF, 0x03, 0x00, 0x0E, 0x03, 0xB1, 0x86, 0x10, 0x00,
+                                                0x40, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+                                                0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x55, 0xCC};
+static constexpr uint8_t FRAME_B[FRAME_SIZE] = {0xAA, 0xFF, 0x03, 0x00, 0x2A, 0x03, 0xC8, 0x86, 0x10, 0x00,
+                                                0x40, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+                                                0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x55, 0xCC};
+
+// --- loop() parsing one periodic frame, no entities attached ---
+
+static void LD2450_PeriodicFrame(benchmark::State &state) {
+  FakeUART uart;
+  LD2450Component radar;
+  radar.set_uart_parent(&uart);
+  radar.set_link_timeout(60000);
+  // Bring the link up outside the timed loop
+  uart.load(FRAME_A, FRAME_SIZE);
+  radar.loop();
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      uart.load(FRAME_A, FRAME_SIZE);
+      radar.loop();
+    }
+    benchmark::DoNotOptimize(uart.available());
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(LD2450_PeriodicFrame);
+
+// --- loop() parsing a moving target and publishing its sensors ---
+
+static void LD2450_PeriodicFrameSensors(benchmark::State &state) {
+  FakeUART uart;
+  LD2450Component radar;
+  radar.set_uart_parent(&uart);
+  radar.set_link_timeout(60000);
+  sensor::Sensor x, y, speed, distance, angle, count;
+  binary_sensor::BinarySensor presence;
+  radar.set_move_x_sensor(0, &x);
+  radar.set_move_y_sensor(0, &y);
+  radar.set_move_speed_sensor(0, &speed);
+  radar.set_move_distance_sensor(0, &distance);
+  radar.set_move_angle_sensor(0, &angle);
+  radar.set_target_count_sensor(&count);
+  radar.set_target_binary_sensor(&presence);
+  uart.load(FRAME_A, FRAME_SIZE);
+  radar.loop();
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      uart.load((i & 1) ? FRAME_B : FRAME_A, FRAME_SIZE);
+      radar.loop();
+    }
+    benchmark::DoNotOptimize(x.state);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(LD2450_PeriodicFrameSensors);
+
+}  // namespace esphome::ld2450::benchmarks

tests/benchmarks/components/ld2450/benchmark.yaml

@@ -0,0 +1,2 @@
+sensor:
+binary_sensor:

tests/benchmarks/components/modbus/bench_modbus_response.cpp

@@ -0,0 +1,67 @@
+#include <benchmark/benchmark.h>
+
+#include "esphome/components/modbus/modbus.h"
+#include "esphome/core/helpers.h"
+#include "fake_uart.h"
+
+namespace esphome::modbus::benchmarks {
+
+using esphome::benchmarks::FakeUART;
+
+// Inner iteration count to amortize CodSpeed instrumentation overhead.
+static constexpr int kInnerIterations = 2000;
+
+static constexpr uint8_t ADDRESS = 1;
+static constexpr uint8_t REGISTER_COUNT = 10;
+// Read 10 holding registers from address 0, as an energy meter poll does
+static constexpr uint8_t REQUEST_PDU[] = {0x03, 0x00, 0x00, 0x00, REGISTER_COUNT};
+// Address + function + byte count + 20 data bytes + CRC
+static constexpr size_t RESPONSE_SIZE = 3 + REGISTER_COUNT * 2 + 2;
+
+class SumDevice : public ModbusClientDevice {
+ public:
+  using ModbusClientDevice::ModbusClientDevice;
+  void on_read_holding_registers(uint16_t start_address, std::span<const uint16_t> registers,
+                                 ResponseStatus status) override {
+    for (uint16_t value : registers)
+      this->sum += value;
+  }
+  uint32_t sum{0};
+};
+
+class NoDelayHub : public ModbusClientHub {
+ protected:
+  // No wait before a send, so every loop() can send the next poll
+  int32_t tx_delay_remaining() override { return 0; }
+};
+
+// --- One continuous poll cycle per loop(): receive and CRC check the response, decode it, requeue and resend ---
+
+static void Modbus_ReadHoldingPollCycle(benchmark::State &state) {
+  // Address 1, function 3, byte count 20, 10 registers; the CRC goes in the last two bytes
+  uint8_t response[RESPONSE_SIZE] = {0x01, 0x03, 0x14, 0x09, 0x1F, 0x00, 0x00, 0x01, 0xF4, 0x00, 0x00, 0x13,
+                                     0x88, 0x00, 0x32, 0x03, 0xE8, 0x00, 0x00, 0x00, 0x64, 0x12, 0x34};
+  put_le16(&response[RESPONSE_SIZE - 2], crc16(response, RESPONSE_SIZE - 2));
+
+  FakeUART uart;
+  uart.set_baud_rate(9600);
+  NoDelayHub hub;
+  hub.set_uart_parent(&uart);
+  hub.setup();
+  SumDevice device(&hub, ADDRESS);
+  hub.queue_pdu(ADDRESS, REQUEST_PDU, &device, {.continuous = true});
+  // The first loop() sends the poll; every send rewinds the response, so the device answers each one
+  hub.loop();
+  uart.load(response, RESPONSE_SIZE);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      hub.loop();
+    }
+    benchmark::DoNotOptimize(device.sum);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(Modbus_ReadHoldingPollCycle);
+
+}  // namespace esphome::modbus::benchmarks

tests/benchmarks/components/uart/bench_uart_readers.cpp

@@ -0,0 +1,57 @@
+#include <benchmark/benchmark.h>
+
+#include "esphome/components/uart/uart.h"
+#include "fake_uart.h"
+
+namespace esphome::uart::benchmarks {
+
+using esphome::benchmarks::FakeUART;
+
+// Inner iteration count to amortize CodSpeed instrumentation overhead.
+static constexpr int kInnerIterations = 2000;
+
+// One second of a GPS module's NMEA output
+static constexpr char NMEA_BURST[] = "$GPGGA,123519,4807.038,N,01131.000,E,1,08,0.9,545.4,M,46.9,M,,*47\r\n"
+                                     "$GPRMC,123519,A,4807.038,N,01131.000,E,022.4,084.4,230394,003.1,W*6A\r\n"
+                                     "$GPGSA,A,3,04,05,,09,12,,,24,,,,,2.5,1.3,2.1*39\r\n"
+                                     "$GPGSV,2,1,08,01,40,083,46,02,17,308,41,12,07,344,39,14,22,228,45*75\r\n"
+                                     "$GPVTG,054.7,T,034.4,M,005.5,N,010.2,K*48\r\n";
+
+// --- read_available(): drain a whole NMEA burst through a byte-sum callback ---
+
+static void UART_ReadAvailable_NmeaBurst(benchmark::State &state) {
+  FakeUART uart;
+  UARTDevice dev(&uart);
+  uint32_t sum = 0;
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      uart.load(reinterpret_cast<const uint8_t *>(NMEA_BURST), sizeof(NMEA_BURST) - 1);
+      dev.read_available([&](uint8_t b) { sum += b; });
+    }
+    benchmark::DoNotOptimize(sum);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(UART_ReadAvailable_NmeaBurst);
+
+// --- discard_available(): drop a 256 byte backlog ---
+
+// Two virtual calls per byte, available() and read_byte(): this measures that per-byte call overhead
+static void UART_DiscardAvailable(benchmark::State &state) {
+  uint8_t backlog[256]{};
+  FakeUART uart;
+  UARTDevice dev(&uart);
+
+  for (auto _ : state) {
+    for (int i = 0; i < kInnerIterations; i++) {
+      uart.load(backlog, sizeof(backlog));
+      dev.discard_available();
+    }
+    benchmark::DoNotOptimize(uart.available());
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(UART_DiscardAvailable);
+
+}  // namespace esphome::uart::benchmarks

tests/benchmarks/core/bench_frame_readers.cpp

@@ -0,0 +1,93 @@
+#include <benchmark/benchmark.h>
+
+#include "esphome/core/frame_reader.h"
+#include "esphome/core/line_reader.h"
+
+namespace esphome::benchmarks {
+
+// Inner iteration count to amortize CodSpeed instrumentation overhead.
+static constexpr int kInnerIterations = 2000;
+
+static constexpr char NMEA_GGA[] = "$GPGGA,123519,4807.038,N,01131.000,E,1,08,0.9,545.4,M,46.9,M,,*47\r\n";
+
+// A 30 byte frame with a 4 byte header and a 2 byte tail
+static constexpr uint8_t FRAME_30[30] = {0xAA, 0xFF, 0x03, 0x00, 0x0E, 0x03, 0xB1, 0x86, 0x10, 0x00,
+                                         0x40, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+                                         0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x55, 0xCC};
+
+// Four A02YYUW distance frames among noise bytes; the doubled 0xFF fails the checksum and resyncs onto the
+// frame behind it. Ends on a whole frame, so every pass starts clean
+static constexpr uint8_t ULTRASONIC_NOISY[] = {0x12, 0xFF, 0x07, 0xA1, 0xA7, 0x34, 0xFF, 0xFF, 0x07, 0xA1, 0xA7,
+                                               0x00, 0xFF, 0x07, 0xA1, 0xA7, 0x56, 0xFF, 0x07, 0xA1, 0xA7};
+
+// --- LineReader: one GGA sentence fed byte by byte ---
+
+static void LineReader_NmeaLine(benchmark::State &state) {
+  LineReader<96> reader;
+  const auto *bytes = reinterpret_cast<const uint8_t *>(NMEA_GGA);
+  size_t total = 0;
+
+  for (auto _ : state) {
+    benchmark::DoNotOptimize(bytes);
+    for (int i = 0; i < kInnerIterations; i++) {
+      for (size_t j = 0; j < sizeof(NMEA_GGA) - 1; j++) {
+        size_t len;
+        if (reader.feed(bytes[j], len) != nullptr)
+          total += len;
+      }
+    }
+    benchmark::DoNotOptimize(total);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(LineReader_NmeaLine);
+
+// --- FrameReader: a 30 byte frame behind a 4 byte header ---
+
+static void FrameReader_30ByteFrame(benchmark::State &state) {
+  FrameReader<sizeof(FRAME_30), 0xAA, 0xFF, 0x03, 0x00> reader;
+  const uint8_t *bytes = FRAME_30;
+  uint32_t frames = 0;
+
+  for (auto _ : state) {
+    benchmark::DoNotOptimize(bytes);
+    for (int i = 0; i < kInnerIterations; i++) {
+      for (size_t j = 0; j < sizeof(FRAME_30); j++) {
+        if (reader.feed(bytes[j]) != nullptr)
+          frames++;
+      }
+    }
+    benchmark::DoNotOptimize(frames);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(FrameReader_30ByteFrame);
+
+// --- FrameReader: 4 byte checksummed frames, skipping noise and resyncing on a bad checksum ---
+
+static void FrameReader_UltrasonicNoisy(benchmark::State &state) {
+  FrameReader<4, 0xFF> reader;
+  const uint8_t *bytes = ULTRASONIC_NOISY;
+  uint32_t frames = 0;
+
+  for (auto _ : state) {
+    benchmark::DoNotOptimize(bytes);
+    for (int i = 0; i < kInnerIterations; i++) {
+      for (size_t j = 0; j < sizeof(ULTRASONIC_NOISY); j++) {
+        const uint8_t *frame = reader.feed(bytes[j]);
+        if (frame == nullptr)
+          continue;
+        if (frame[3] == static_cast<uint8_t>(frame[0] + frame[1] + frame[2])) {
+          frames++;
+        } else {
+          reader.resync();
+        }
+      }
+    }
+    benchmark::DoNotOptimize(frames);
+  }
+  state.SetItemsProcessed(state.iterations() * kInnerIterations);
+}
+BENCHMARK(FrameReader_UltrasonicNoisy);
+
+}  // namespace esphome::benchmarks

tests/benchmarks/helpers/fake_uart.h

@@ -0,0 +1,44 @@
+#pragma once
+
+#include <cstdint>
+#include <cstring>
+
+#include "esphome/components/uart/uart_component.h"
+
+namespace esphome::benchmarks {
+
+// Serves a fixed byte stream as received data. Every write rewinds it, so the stream answers each request.
+class FakeUART : public uart::UARTComponent {
+ public:
+  void load(const uint8_t *data, size_t len) {
+    this->data_ = data;
+    this->len_ = len;
+    this->pos_ = 0;
+  }
+  void rewind() { this->pos_ = 0; }
+
+  void write_array(const uint8_t *data, size_t len) override { this->rewind(); }
+  bool peek_byte(uint8_t *data) override {
+    if (this->pos_ >= this->len_)
+      return false;
+    *data = this->data_[this->pos_];
+    return true;
+  }
+  bool read_array(uint8_t *data, size_t len) override {
+    if (len > this->len_ - this->pos_)
+      return false;
+    std::memcpy(data, this->data_ + this->pos_, len);
+    this->pos_ += len;
+    return true;
+  }
+  size_t available() override { return this->len_ - this->pos_; }
+  uart::UARTFlushResult flush() override { return uart::UARTFlushResult::UART_FLUSH_RESULT_ASSUMED_SUCCESS; }
+  void check_logger_conflict() override {}
+
+ protected:
+  const uint8_t *data_{nullptr};
+  size_t len_{0};
+  size_t pos_{0};
+};
+
+}  // namespace esphome::benchmarks