add some code

managed_components/espressif__esp-dsp/modules/matrix/addc/float/dspm_addc_f32_ae32.S

@@ -0,0 +1,58 @@
+/*
+ * SPDX-FileCopyrightText: 2023 Espressif Systems (Shanghai) CO LTD
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+#include "dspm_addc_platform.h"
+#if (dspm_addc_f32_ae32_enabled == 1)
+
+// This is an add function for sub-matrices for ESP32 processor
+    .text
+    .align  4
+    .global dspm_addc_f32_ae32
+    .type   dspm_addc_f32_ae32,@function
+// The function implements the following C code:
+// esp_err_t dspm_addc_f32_ansi(const float *input, float *output, float C, int rows, int cols, int padd_in, int padd_out, int step_in, int step_out);
+
+dspm_addc_f32_ae32: 
+// input            - a2
+// output           - a3
+// C                - a4
+// rows             - a5
+// cols             - a6
+// padd_in          - a7
+// padd_out         - a8
+// step_in          - a9
+// step_out         - a10
+
+    entry   a1, 16
+
+    l32i.n  a8,  a1, 16         // padd_out
+    l32i.n  a9,  a1, 20         // step_in
+    l32i.n  a10, a1, 24         // step_out
+
+    slli    a9,  a9,  2         // a9   - step_in << 2
+    slli    a10, a10, 2         // a10  - step_out << 2
+
+    wfr     f0,  a4             // a4   - load to the f0
+
+    .outer_loop_addc_f32_ae32:
+
+        loopnez a6, .loop_addc_f32_ae32
+            lsxp     f1,  a2,  a9       // load input to f1, increment input (input_ptr+=step_in)
+
+            add.s    f2,  f0,  f1       // f2 = f0 + f1
+            ssxp     f2,  a3,  a10      // save result f2 to output a3, increment output (output_ptr+=step_out)
+        .loop_addc_f32_ae32:
+
+        addx4    a2,  a7,  a2           // input1_ptr += (padd_in << 2);
+        addx4    a3,  a8,  a3           // output_ptr += (padd_out << 2);
+        addi.n   a5,  a5,  -1           // rows - 1
+
+    bnez a5, .outer_loop_addc_f32_ae32
+
+    movi.n  a2, 0 // return status ESP_OK
+    retw.n
+
+#endif // dspm_add_f32_ae32_enabled

managed_components/espressif__esp-dsp/modules/matrix/addc/float/dspm_addc_f32_ansi.c

@@ -0,0 +1,51 @@
+/*
+ * SPDX-FileCopyrightText: 2023 Espressif Systems (Shanghai) CO LTD
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+#include "dspm_addc.h"
+
+esp_err_t dspm_addc_f32_ansi(const float *input, float *output, float C, int rows, int cols, int padd_in, int padd_out, int step_in, int step_out)
+{
+    if (NULL == input) {
+        return ESP_ERR_DSP_PARAM_OUTOFRANGE;
+    }
+    if (NULL == output) {
+        return ESP_ERR_DSP_PARAM_OUTOFRANGE;
+    }
+
+    if (rows <= 0) {
+        return ESP_ERR_DSP_PARAM_OUTOFRANGE;
+    }
+    if (cols <= 0) {
+        return ESP_ERR_DSP_PARAM_OUTOFRANGE;
+    }
+
+    if (padd_in < 0) {
+        return ESP_ERR_DSP_PARAM_OUTOFRANGE;
+    }
+    if (padd_out < 0) {
+        return ESP_ERR_DSP_PARAM_OUTOFRANGE;
+    }
+
+    if (step_in <= 0) {
+        return ESP_ERR_DSP_PARAM_OUTOFRANGE;
+    }
+    if (step_out <= 0) {
+        return ESP_ERR_DSP_PARAM_OUTOFRANGE;
+    }
+
+    const int ptr_step_in = cols + padd_in;
+    const int ptr_step_out = cols + padd_out;
+    float *ptr_input = (float *)input;
+
+    for (int row = 0; row < rows; row++) {
+        for (int col = 0; col < cols; col++) {
+            output[col * step_out] = ptr_input[col * step_in] + C;
+        }
+        output += ptr_step_out;
+        ptr_input += ptr_step_in;
+    }
+    return ESP_OK;
+}