add some code

managed_components/espressif__esp-dsp/modules/dotprod/float/dspi_dotprod_f32_ansi.c

@@ -0,0 +1,47 @@
+// Copyright 2018-2019 Espressif Systems (Shanghai) PTE LTD
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "dspi_dotprod.h"
+
+esp_err_t dspi_dotprod_f32_ansi(image2d_t *in_image, image2d_t *filter, float *out_value, int count_x, int count_y)
+{
+    if (in_image->step_x * count_x > in_image->stride_x) {
+        return ESP_ERR_DSP_PARAM_OUTOFRANGE;
+    }
+    if (in_image->step_y * count_y > in_image->stride_y) {
+        return ESP_ERR_DSP_PARAM_OUTOFRANGE;
+    }
+    if (filter->step_x * count_x > filter->stride_x) {
+        return ESP_ERR_DSP_PARAM_OUTOFRANGE;
+    }
+    if (filter->step_y * count_y > filter->stride_y) {
+        return ESP_ERR_DSP_PARAM_OUTOFRANGE;
+    }
+
+    float *i_data =  (float *)in_image->data;
+    float *f_data =  (float *)filter->data;
+    int i_step = in_image->stride_x * in_image->step_y;
+    int f_step = filter->stride_x * filter->step_y;
+
+    float acc = 0;
+    for (int y = 0; y < count_y; y++) {
+        for (int x = 0; x < count_x; x++) {
+            acc += i_data[in_image->step_x * x] * f_data[filter->step_x * x];
+        }
+        i_data += i_step;
+        f_data += f_step;
+    }
+    *out_value = acc;
+    return ESP_OK;
+}

managed_components/espressif__esp-dsp/modules/dotprod/float/dspi_dotprod_off_f32_ansi.c

@@ -0,0 +1,47 @@
+// Copyright 2018-2019 Espressif Systems (Shanghai) PTE LTD
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "dspi_dotprod.h"
+
+esp_err_t dspi_dotprod_off_f32_ansi(image2d_t *in_image, image2d_t *filter, float *out_value, int count_x, int count_y, float offset)
+{
+    if (in_image->step_x * count_x > in_image->stride_x) {
+        return ESP_ERR_DSP_PARAM_OUTOFRANGE;
+    }
+    if (in_image->step_y * count_y > in_image->stride_y) {
+        return ESP_ERR_DSP_PARAM_OUTOFRANGE;
+    }
+    if (filter->step_x * count_x > filter->stride_x) {
+        return ESP_ERR_DSP_PARAM_OUTOFRANGE;
+    }
+    if (filter->step_y * count_y > filter->stride_y) {
+        return ESP_ERR_DSP_PARAM_OUTOFRANGE;
+    }
+
+    float *i_data =  (float *)in_image->data;
+    float *f_data =  (float *)filter->data;
+    int i_step = in_image->stride_x * in_image->step_y;
+    int f_step = filter->stride_x * filter->step_y;
+
+    float acc = 0;
+    for (int y = 0; y < count_y; y++) {
+        for (int x = 0; x < count_x; x++) {
+            acc += i_data[in_image->step_x * x] * (f_data[filter->step_x * x] + offset);
+        }
+        i_data += i_step;
+        f_data += f_step;
+    }
+    *out_value = acc;
+    return ESP_OK;
+}

managed_components/espressif__esp-dsp/modules/dotprod/float/dsps_dotprod_f32_ae32.S

@@ -0,0 +1,62 @@
+// Copyright 2018-2019 Espressif Systems (Shanghai) PTE LTD
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License. 
+
+#include "dsps_dotprod_platform.h"
+#if (dotprod_f32_ae32_enabled == 1)
+
+#include "dsps_dotprod_f32_m_ae32.S"
+
+// This is dot product function for ESP32 processor.
+	.text
+	.align  4
+	.global dsps_dotprod_f32_ae32
+	.global .dsps_dotprod_f32_ae32_body
+	.type   dsps_dotprod_f32_ae32,@function
+// The function implements the following C code:
+//esp_err_t dsps_dotprod_f32_ae32(const float* src1, const float* src2, float* dest, int len)
+//{
+//    float acc = 0;
+//    for (int i=0 ; i< len ; i++)
+//    {
+//        acc += src1[i]*src2[i];
+//    }
+//    *dest = acc;
+//    return ESP_OK;
+//}
+
+dsps_dotprod_f32_ae32: 
+// src1 - a2
+// src2 - a3
+// dest - a4
+// len  - a5
+
+	entry	a1, 16
+.dsps_dotprod_f32_ae32_body:	
+	// Array increment for floating point data should be 4
+	movi.n	a8, 4
+	// Clear initial state of the result register
+	movi.n	a9, 0
+	wfr	    f1, a9
+	// a2 - input1
+	// a3 - input2
+	// a5 - length
+	// a8 - 4,  step in arrays
+	dotprod_f32_ae32 a2, a3, a5, a9, a8;
+	
+	ssi	f1, a4, 0 // Store result from f1 to memory at a4
+	
+	movi.n	a2, 0 // return status ESP_OK
+	retw.n
+
+#endif // dotprode_f32_ae32_enabled

managed_components/espressif__esp-dsp/modules/dotprod/float/dsps_dotprod_f32_aes3.S

@@ -0,0 +1,85 @@
+// Copyright 2018-2019 Espressif Systems (Shanghai) PTE LTD
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License. 
+
+#include "dsps_dotprod_platform.h"
+#if (dsps_dotprod_f32_aes3_enabled == 1)
+
+// This is dot product function for ESP32 processor.
+	.text
+	.align  4
+	.global dsps_dotprod_f32_aes3
+	.global .dsps_dotprod_f32_ae32_body
+	.type   dsps_dotprod_f32_aes3,@function
+// The function implements the following C code:
+//esp_err_t dsps_dotprod_f32_ae32(const float* src1, const float* src2, float* dest, int len)
+//{
+//    float acc = 0;
+//    for (int i=0 ; i< len ; i++)
+//    {
+//        acc += src1[i]*src2[i];
+//    }
+//    *dest = acc;
+//    return ESP_OK;
+//}
+
+dsps_dotprod_f32_aes3: 
+// src1 - a2
+// src2 - a3
+// dest - a4
+// len  - a5
+
+	entry	a1, 16
+	// Check length and align
+	movi.n	a10, 3
+	and a10, a10, a5
+	movi.n	a9, 15
+	or  a11, a3, a2
+	and a11, a9, a11
+	or  a10, a10, a11
+	beqz  a10, .dsps_dotprod_f32_aes3_body
+	// Call Esp32 function
+	J 	.dsps_dotprod_f32_ae32_body
+	
+.dsps_dotprod_f32_aes3_body:
+	// Clear initial state of the result register
+	movi.n	a9, 0
+	wfr	    f0, a9
+	wfr	    f1, a9
+	wfr	    f2, a9
+	wfr	    f3, a9
+	// a2 - input1
+	// a3 - input2
+	// a5 - length
+
+	srli	    a6, a5, 2		// N count
+//	lsx	    f0, a2,  a9
+	loopnez a6, .loop_mac_end_m_ae32
+		EE.LDF.128.IP  f11, f10, f9, f8, a2, 16
+		EE.LDF.128.IP  f7, f6, f5, f4, a3, 16
+		madd.s	f0, f4, f8		// f0 = X11*Y11
+		madd.s	f1, f5, f9		// f1 = X12*Y11
+		madd.s	f2, f6, f10		// f2 = X13*Y11
+		madd.s	f3, f7, f11		// f3 = X14*Y11
+	.loop_mac_end_m_ae32:
+
+	add.s   f0, f0, f1
+	add.s   f0, f0, f2
+	add.s   f0, f0, f3
+
+	ssi	f0, a4, 0 // Store result from f1 to memory at a4
+	
+	movi.n	a2, 0 // return status ESP_OK
+	retw.n
+
+#endif // dotprode_f32_ae32_enabled

managed_components/espressif__esp-dsp/modules/dotprod/float/dsps_dotprod_f32_ansi.c

@@ -0,0 +1,25 @@
+// Copyright 2018-2019 Espressif Systems (Shanghai) PTE LTD
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "dsps_dotprod.h"
+
+esp_err_t dsps_dotprod_f32_ansi(const float *src1, const float *src2, float *dest, int len)
+{
+    float acc = 0;
+    for (int i = 0 ; i < len ; i++) {
+        acc += src1[i] * src2[i];
+    }
+    *dest = acc;
+    return ESP_OK;
+}

managed_components/espressif__esp-dsp/modules/dotprod/float/dsps_dotprod_f32_arp4.S

@@ -0,0 +1,77 @@
+// Copyright 2024 Espressif Systems (Shanghai) PTE LTD
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License. 
+
+#include "dsps_dotprod_platform.h"
+#if (dsps_dotprod_f32_arp4_enabled == 1)
+
+    .text
+    .align  4
+    .global dsps_dotprod_f32_arp4
+    .type   dsps_dotprod_f32_arp4,@function
+// The function implements the following C code:
+//esp_err_t dsps_dotprod_f32(const float* src1, const float* src2, float* dest, int len)
+//{
+//    float acc = 0;
+//    for (int i=0 ; i< len ; i++)
+//    {
+//        acc += src1[i]*src2[i];
+//    }
+//    *dest = acc;
+//    return ESP_OK;
+//}
+
+dsps_dotprod_f32_arp4:
+// src1 - a0
+// src2 - a1
+// dest - a2
+// len  - a3
+    add sp,sp,-16
+
+    fmv.w.x fa2,zero
+
+
+    flw     fa0, 0(a0)
+    flw     fa1, 0(a1)
+    add     a0, a0, 4
+    add     a1, a1, 4
+    li      a4, 2
+    ble     a3, a4, .loop_less_2 
+
+// Loop when len > 2
+    esp.lp.setup    0, a3, .dotprod_loop
+        fmadd.s fa2, fa0, fa1, fa2
+        flw     fa0, 0(a0)
+        flw     fa1, 0(a1)
+        add     a0, a0, 4
+.dotprod_loop:  add     a1, a1, 4
+    fsw     fa2, 0(a2)
+
+    add sp,sp,16
+    li  a0,0
+    ret
+// Loop when len <=2
+.loop_less_2:
+    fmadd.s fa2, fa0, fa1, fa2
+    flw     fa0, 0(a0)
+    flw     fa1, 0(a1)
+    add     a0, a0, 4
+    add     a1, a1, 4
+    add     a3, a3, -1
+    bnez    a3, .loop_less_2
+    fsw     fa2, 0(a2)
+    add sp,sp,16
+    li  a0,0
+    ret
+
+#endif // dotprode_f32_arp4_enabled

managed_components/espressif__esp-dsp/modules/dotprod/float/dsps_dotprod_f32_m_ae32.S

@@ -0,0 +1,42 @@
+// Copyright 2018-2019 Espressif Systems (Shanghai) PTE LTD
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License. 
+
+
+.macro dotprod_f32_ae32 x1 x2 count step1  step2
+// This macro calculates floating point dot product for count float samples
+// x1, x2 - input arrays
+// count - amount of samples
+// step1 - start step 
+//,step2 - A register for array step increment. (should be divided by 4)
+// f1 - contains initial value 
+//
+// result in f1
+// 
+// Macros body:
+// f1 += x1[i*step1]*x2[i*step2]; i: 0..counter-1
+// affected: f0, f1, f2
+// Example: dotprod_f32_ae32 a2 a3 a5 a8 a9
+// a8 == 4, step is 4 bytes
+// a5 == 32, length of array is 32
+// 
+//    mov     \step1, \step2
+	lsx	    f0, \x2,  \step1
+//    sub	    \x1, \x1, \step1 // To compensate first increment
+	loopnez \count, .loop_mac_end_m_ae32
+		lsx    f2, \x1, \step1
+		madd.s  f1, f2, f0
+		add.n	\step1, \step1, \step2
+		lsx    f0, \x2, \step1
+	.loop_mac_end_m_ae32:
+.endm

managed_components/espressif__esp-dsp/modules/dotprod/float/dsps_dotprode_f32_ae32.S

@@ -0,0 +1,64 @@
+// Copyright 2018-2019 Espressif Systems (Shanghai) PTE LTD
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License. 
+
+#include "dsps_dotprod_platform.h"
+#if (dotprode_f32_ae32_enabled == 1)
+
+#include "dsps_dotprode_f32_m_ae32.S"
+
+// This is dot product function for ESP32 processor.
+	.text
+	.align  4
+	.global dsps_dotprode_f32_ae32
+	.type   dsps_dotprode_f32_ae32,@function
+// The function implements the following C code:
+//esp_err_t dsps_dotprod_f32_ae32(const float* src1, const float* src2, float* dest, int len)
+//{
+//    float acc = 0;
+//    for (int i=0 ; i< len ; i++)
+//    {
+//        acc += src1[i]*src2[i];
+//    }
+//    *dest = acc;
+//    return ESP_OK;
+//}
+
+dsps_dotprode_f32_ae32: 
+// src1 - a2
+// src2 - a3
+// dest - a4
+// len  - a5
+// step1- a6
+// step2- a7
+
+	entry	a1, 16
+	// Array increment for floating point data should be 4
+	
+	slli    a6,a6, 2
+	slli    a7,a7, 2
+	// Clear initial state of the result register
+	movi.n	a9, 0
+	wfr	    f1, a9
+	// a2 - input1
+	// a3 - input2
+	// a5 - length
+	// a6,a7,  step in arrays
+	dotprode_f32_ae32 a2, a3, a5, a6, a7;
+	
+	ssi	f1, a4, 0 // Store result from f1 to memory at a4
+	
+	movi.n	a2, 0 // return status ESP_OK
+	retw.n
+
+#endif //dotprode_f32_ae32_enabled

managed_components/espressif__esp-dsp/modules/dotprod/float/dsps_dotprode_f32_ansi.c

@@ -0,0 +1,25 @@
+// Copyright 2018-2019 Espressif Systems (Shanghai) PTE LTD
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "dsps_dotprod.h"
+
+esp_err_t dsps_dotprode_f32_ansi(const float *src1, const float *src2, float *dest, int len, int step1, int step2)
+{
+    float acc = 0;
+    for (int i = 0 ; i < len ; i++) {
+        acc += src1[i * step1] * src2[i * step2];
+    }
+    *dest = acc;
+    return ESP_OK;
+}

managed_components/espressif__esp-dsp/modules/dotprod/float/dsps_dotprode_f32_arp4.S

@@ -0,0 +1,78 @@
+// Copyright 2024 Espressif Systems (Shanghai) PTE LTD
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License. 
+
+#include "dsps_dotprod_platform.h"
+#if (dsps_dotprod_f32_arp4_enabled == 1)
+
+	.text
+	.align  4
+	.global dsps_dotprode_f32_arp4
+	.type   dsps_dotprode_f32_arp4,@function
+// The function implements the following C code:
+//esp_err_t dsps_dotprode_f32(const float *src1, const float *src2, float *dest, int len, int step1, int step2)
+//{
+//    float acc = 0;
+//    for (int i = 0 ; i < len ; i++) {
+//        acc += src1[i * step1] * src2[i * step2];
+//    }
+//    *dest = acc;
+//    return ESP_OK;
+//}
+
+dsps_dotprode_f32_arp4:
+// src1 - a0
+// src2 - a1
+// dest - a2
+// len  - a3
+    add	sp,sp,-16
+
+    fmv.w.x	fa2,zero
+    slli    a4, a4, 2  // step address increment by 4
+    slli    a5, a5, 2  // step address increment by 4
+
+	flw		fa0, 0(a0)
+	flw		fa1, 0(a1)
+	add		a0, a0, a4
+	add		a1, a1, a5
+	li		a6, 2
+	ble		a3, a6, .loop_less_2 
+
+// Loop when len > 2
+    esp.lp.setup    0, a3, .dotprod_loop
+		fmadd.s fa2, fa0, fa1, fa2
+		flw		fa0, 0(a0)
+		flw		fa1, 0(a1)
+		add		a0, a0, a4
+.dotprod_loop:	add		a1, a1, a5
+	fsw		fa2, 0(a2)
+
+    add	sp,sp,16
+    li	a0,0
+    ret
+// Loop when len <=2
+.loop_less_2:
+	fmadd.s fa2, fa0, fa1, fa2
+	flw		fa0, 0(a0)
+	flw		fa1, 0(a1)
+	add		a0, a0, a4
+	add		a1, a1, a5
+	add		a3, a3, -1
+	bnez	a3, .loop_less_2
+
+	fsw		fa2, 0(a2)
+    add	sp,sp,16
+    li	a0,0
+    ret
+
+#endif // dotprode_f32_arp4_enabled

managed_components/espressif__esp-dsp/modules/dotprod/float/dsps_dotprode_f32_m_ae32.S

@@ -0,0 +1,41 @@
+// Copyright 2018-2019 Espressif Systems (Shanghai) PTE LTD
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License. 
+
+
+.macro dotprode_f32_ae32 x1 x2 count step1  step2
+// This macro calculates floating point dot product for count float samples
+// x1, x2 - input arrays
+// count - amount of samples
+// step1,step2 - A register for array step. (should be divided by 4)
+// f1 - contains initial value 
+//
+// result in f1
+// 
+// Macros body:
+// f1 += x1[i*step1]*x2[i*step2]; i: 0..counter-1
+// affected: f0, f1, f2
+// Example: dotprod_f32_ae32 a2 a3 a5 a8 a9
+// a8 == 4, step is 4 bytes
+// a5 == 32, length of array is 32
+// 
+	lsi	    f0, \x2, 0
+	sub	    \x1, \x1, \step1 // To compensate first increment
+	loopnez \count, .loop_mace_end_m_ae32
+		add.n	\x1, \x1, \step1
+		lsi	    f2, \x1, 0
+		madd.s	f1, f2, f0
+		add.n	\x2, \x2, \step2
+		lsi	     f0, \x2, 0
+	.loop_mace_end_m_ae32:
+.endm