add some code

managed_components/espressif__esp-dsp/modules/matrix/sub/float/dspm_sub_f32_ae32.S

@@ -0,0 +1,63 @@
+/*
+ * SPDX-FileCopyrightText: 2023 Espressif Systems (Shanghai) CO LTD
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+#include "dspm_sub_platform.h"
+#if (dspm_sub_f32_ae32_enabled == 1)
+
+// This is an sub function for sub-matrices for ESP32 processor
+    .text
+    .align  4
+    .global dspm_sub_f32_ae32
+    .type   dspm_sub_f32_ae32,@function
+// The function implements the following C code:
+// esp_err_t dspm_sub_f32_ansi(const float *input1, const float *input2, float *output, int rows, int cols, int padd1, int padd2, int padd_out, int step1, int step2, int step_out);
+
+dspm_sub_f32_ae32: 
+// input1           - a2
+// input2           - a3
+// output           - a4
+// rows             - a5
+// cols             - a6
+// padd1            - a7
+// padd2            - a8
+// padd_out         - a9
+// step1            - a10
+// step2            - a11
+// step_out         - a12
+
+    entry   a1, 16
+
+    l32i.n  a8,  a1, 16         // padd2
+    l32i.n  a9,  a1, 20         // padd_out
+    l32i.n  a10, a1, 24         // step1
+    l32i.n  a11, a1, 24         // step2
+    l32i.n  a12, a1, 24         // step_out
+
+    slli    a10, a10, 2         // a10  - step1 << 2
+    slli    a11, a11, 2         // a11  - step2 << 2
+    slli    a12, a12, 2         // a12  - step_out << 2
+
+    .outer_loop_sub_f32_ae32:
+
+        loopnez a6, .loop_sub_f32_ae32
+            lsxp     f0,  a2,  a10      // load input1 to f0, increment input1 (input1_ptr+=step1)
+            lsxp     f1,  a3,  a11      // load input2 to f1, increment input2 (input2_ptr+=step2)
+
+            sub.s    f2,  f0,  f1       // f2 = f0 - f1
+            ssxp     f2,  a4,  a12      // save result f2 to output a4, increment output (output_ptr+=step_out)
+        .loop_sub_f32_ae32:
+
+        addx4    a3,  a8,  a3           // input2_ptr += (padd2 << 2);
+        addx4    a2,  a7,  a2           // input1_ptr += (padd1 << 2);
+        addx4    a4,  a9,  a4           // output_ptr += (padd_out << 2);
+        addi.n   a5,  a5,  -1           // rows - 1
+
+    bnez a5, .outer_loop_sub_f32_ae32
+
+    movi.n  a2, 0 // return status ESP_OK
+    retw.n
+
+#endif // dspm_sub_f32_ae32_enabled

managed_components/espressif__esp-dsp/modules/matrix/sub/float/dspm_sub_f32_ansi.c

@@ -0,0 +1,63 @@
+/*
+ * SPDX-FileCopyrightText: 2023 Espressif Systems (Shanghai) CO LTD
+ *
+ * SPDX-License-Identifier: Apache-2.0
+ */
+
+#include "dspm_sub.h"
+
+esp_err_t dspm_sub_f32_ansi(const float *input1, const float *input2, float *output, int rows, int cols, int padd1, int padd2, int padd_out, int step1, int step2, int step_out)
+{
+    if (NULL == input1) {
+        return ESP_ERR_DSP_PARAM_OUTOFRANGE;
+    }
+    if (NULL == input2) {
+        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 (padd1 < 0) {
+        return ESP_ERR_DSP_PARAM_OUTOFRANGE;
+    }
+    if (padd2 < 0) {
+        return ESP_ERR_DSP_PARAM_OUTOFRANGE;
+    }
+    if (padd_out < 0) {
+        return ESP_ERR_DSP_PARAM_OUTOFRANGE;
+    }
+
+    if (step1 <= 0) {
+        return ESP_ERR_DSP_PARAM_OUTOFRANGE;
+    }
+    if (step2 <= 0) {
+        return ESP_ERR_DSP_PARAM_OUTOFRANGE;
+    }
+    if (step_out <= 0) {
+        return ESP_ERR_DSP_PARAM_OUTOFRANGE;
+    }
+
+    const int ptr_input1_step = cols + padd1;
+    const int ptr_input2_step = cols + padd2;
+    const int ptr_out_step = cols + padd_out;
+    float *ptr_input1 = (float *)input1;
+    float *ptr_input2 = (float *)input2;
+
+    for (int row = 0; row < rows; row++) {
+        for (int col = 0; col < cols; col++) {
+            output[col * step_out] = ptr_input1[col * step1] - ptr_input2[col * step2];
+        }
+        ptr_input1 += ptr_input1_step;
+        ptr_input2 += ptr_input2_step;
+        output += ptr_out_step;
+    }
+    return ESP_OK;
+}