test(freertos): add a test case for the DSP coprocessor

components/freertos/test_apps/freertos/port/test_dsp_in_task.c

@@ -0,0 +1,101 @@
+/*
+ * SPDX-FileCopyrightText: 2025 Espressif Systems (Shanghai) CO LTD
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+#include "sdkconfig.h"
+#include <math.h>
+#include "soc/soc_caps.h"
+#include "freertos/FreeRTOS.h"
+#include "freertos/task.h"
+#include "freertos/semphr.h"
+#include "unity.h"
+#include "test_utils.h"
+
+#if SOC_CPU_HAS_DSP
+
+#define TEST_NUM_TASKS       4
+
+typedef struct {
+    int32_t  id;
+    uint32_t result;
+    TaskHandle_t main;
+    SemaphoreHandle_t sem;
+} dsp_params_t;
+
+/**
+ * @brief Multiplies the given ID by a constant.
+ *
+ * @param id Value to multiply
+ */
+uint32_t dsp_id_mul(uint32_t id);
+
+/**
+ * @brief DSP Assembly routine need to access this constant, make it public.
+ * It will be used as a multiplier.
+ */
+const uint32_t g_dsp_constant = 100000;
+
+static void pinned_task(void *arg)
+{
+    dsp_params_t *param = (dsp_params_t*) arg;
+    ulTaskNotifyTake(pdTRUE, portMAX_DELAY);
+
+    param->result = dsp_id_mul(param->id);
+
+    /* Indicate done and wait to be deleted */
+    xSemaphoreGive((SemaphoreHandle_t)param->sem);
+    vTaskSuspend(NULL);
+}
+
+TEST_CASE("DSP: Usage in task", "[freertos]")
+{
+    TaskHandle_t task_handles[CONFIG_FREERTOS_NUMBER_OF_CORES][TEST_NUM_TASKS];
+    dsp_params_t params[CONFIG_FREERTOS_NUMBER_OF_CORES][TEST_NUM_TASKS];
+
+    SemaphoreHandle_t done_sem = xSemaphoreCreateCounting(TEST_NUM_TASKS, 0);
+    TEST_ASSERT_NOT_EQUAL(NULL, done_sem);
+
+    // Create test tasks for each core
+    for (int i = 0; i < CONFIG_FREERTOS_NUMBER_OF_CORES; i++) {
+        for (int j = 0; j < TEST_NUM_TASKS; j++) {
+            params[i][j] = (dsp_params_t) {
+                .id = i * TEST_NUM_TASKS + j + 1,
+                .sem = done_sem,
+            };
+            TEST_ASSERT_EQUAL(pdTRUE, xTaskCreatePinnedToCore(pinned_task, "task", 4096, (void *) &params[i][j], UNITY_FREERTOS_PRIORITY + 1, &task_handles[i][j], i));
+        }
+    }
+
+    // Start the created tasks
+    for (int i = 0; i < CONFIG_FREERTOS_NUMBER_OF_CORES; i++) {
+        for (int j = 0; j < TEST_NUM_TASKS; j++) {
+            xTaskNotifyGive(task_handles[i][j]);
+        }
+    }
+
+    // Wait for the tasks to complete
+    for (int i = 0; i < CONFIG_FREERTOS_NUMBER_OF_CORES * TEST_NUM_TASKS; i++) {
+        xSemaphoreTake(done_sem, portMAX_DELAY);
+    }
+
+    // Delete the tasks
+    for (int i = 0; i < CONFIG_FREERTOS_NUMBER_OF_CORES; i++) {
+        for (int j = 0; j < TEST_NUM_TASKS; j++) {
+            vTaskDelete(task_handles[i][j]);
+        }
+    }
+
+    // Check the values
+    for (int i = 0; i < CONFIG_FREERTOS_NUMBER_OF_CORES; i++) {
+        for (int j = 0; j < TEST_NUM_TASKS; j++) {
+            dsp_params_t* param = &params[i][j];
+            TEST_ASSERT_EQUAL(param->id * g_dsp_constant, param->result);
+        }
+    }
+
+    vSemaphoreDelete(done_sem);
+}
+
+#endif /* SOC_CPU_HAS_DSP */

components/freertos/test_apps/freertos/port/test_dsp_routines.S

@@ -0,0 +1,62 @@
+/*
+ * SPDX-FileCopyrightText: 2025 Espressif Systems (Shanghai) CO LTD
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+    #include "soc/soc_caps.h"
+
+#if SOC_CPU_HAS_DSP
+
+    #include "riscv/csr_dsp.h"
+
+    /**
+     * @brief Let's allow a small subset of registers to be used in the macro below
+     */
+    .set regnum_a0, 10
+    .set regnum_a1, 11
+    .set regnum_a2, 12
+    .set regnum_a3, 13
+    .set regnum_a4, 14
+    .set regnum_a5, 15
+    .set regnum_a6, 16
+    .set regnum_a7, 17
+
+    /**
+     * @brief The toolchain doesn't support DSP instructions yet, define it as a constant.
+     */
+    .macro macs32  _rs1, _rs2
+        .word ( ( regnum_\_rs1 << 15 ) | ( regnum_\_rs2 << 20) | 0b100<<12 | 0b1011011 )
+    .endm
+
+    .global g_dsp_constant
+
+    /**
+     * @brief Multiply the given ID by the global constant defined as g_dsp_constant.
+     * NOTE: The goal of the function is not to be fast and efficient, on the contrary, it needs to be
+     * slow and long so that it will be preempted.
+     *
+     * Parameters:
+     *  a0: 32-bit id
+     *
+     * Returns:
+     *  a0: multiplied value
+     */
+    .globl dsp_id_mul
+dsp_id_mul:
+    csrw 0x809, zero    // SAR = 0
+    csrw 0x806, zero    // XACC_L
+    csrw 0x807, zero    // XACC_H
+    li a1, 1
+    /* Load the constant in a2 */
+    lw a2, (g_dsp_constant)
+1:
+    // Perform ACC += a0 (=id) * a1 (=1)
+    macs32 a1, a0
+    addi a2, a2, -1
+    bnez a2, 1b
+    // Get the lowest bits
+    csrr a0, 0x806    // XACC_L
+    ret
+
+
+#endif // SOC_CPU_HAS_DSP