esp-idf/components/esp_driver_ppa/test_apps/main/test_ppa.c

801 lines
29 KiB
C

/*
* SPDX-FileCopyrightText: 2024 Espressif Systems (Shanghai) CO LTD
*
* SPDX-License-Identifier: Apache-2.0
*/
#include <stdio.h>
#include <string.h>
#include <stdlib.h>
#include "unity.h"
#include "freertos/FreeRTOS.h"
#include "freertos/task.h"
#include "freertos/semphr.h"
#include "driver/ppa.h"
#include "esp_heap_caps.h"
#include "esp_err.h"
#include "ccomp_timer.h"
#include "hal/color_hal.h"
#include "esp_cache.h"
#define ALIGN_UP(num, align) (((num) + ((align) - 1)) & ~((align) - 1))
TEST_CASE("ppa_client_do_ppa_operation", "[PPA]")
{
const uint32_t w = 480;
const uint32_t h = 480;
const uint32_t buf_1_color_type_id = COLOR_TYPE_ID(COLOR_SPACE_ARGB, COLOR_PIXEL_ARGB8888);
const uint32_t buf_2_color_type_id = COLOR_TYPE_ID(COLOR_SPACE_ARGB, COLOR_PIXEL_ARGB8888);
color_space_pixel_format_t buf_1_cm = {
.color_type_id = buf_1_color_type_id,
};
color_space_pixel_format_t buf_2_cm = {
.color_type_id = buf_2_color_type_id,
};
uint32_t buf_1_size = ALIGN_UP(w * h * color_hal_pixel_format_get_bit_depth(buf_1_cm) / 8, 64);
uint32_t buf_2_size = ALIGN_UP(w * h * color_hal_pixel_format_get_bit_depth(buf_2_cm) / 8, 64);
uint8_t *buf_1 = heap_caps_aligned_calloc(4, buf_1_size, sizeof(uint8_t), MALLOC_CAP_SPIRAM | MALLOC_CAP_DMA); // cache alignment is implicited by MALLOC_CAP_DMA
TEST_ASSERT_NOT_NULL(buf_1);
uint8_t *buf_2 = heap_caps_aligned_calloc(4, buf_2_size, sizeof(uint8_t), MALLOC_CAP_SPIRAM | MALLOC_CAP_DMA);
TEST_ASSERT_NOT_NULL(buf_2);
// Register different types of PPA clients
ppa_client_handle_t ppa_client_srm_handle;
ppa_client_handle_t ppa_client_blend_handle;
ppa_client_handle_t ppa_client_fill_handle_a;
ppa_client_handle_t ppa_client_fill_handle_b;
ppa_client_config_t ppa_client_config = {
.oper_type = PPA_OPERATION_SRM,
};
TEST_ESP_OK(ppa_register_client(&ppa_client_config, &ppa_client_srm_handle));
ppa_client_config.oper_type = PPA_OPERATION_BLEND;
TEST_ESP_OK(ppa_register_client(&ppa_client_config, &ppa_client_blend_handle));
ppa_client_config.oper_type = PPA_OPERATION_FILL;
TEST_ESP_OK(ppa_register_client(&ppa_client_config, &ppa_client_fill_handle_a));
TEST_ESP_OK(ppa_register_client(&ppa_client_config, &ppa_client_fill_handle_b));
ppa_srm_oper_config_t srm_oper_config = {
.in.buffer = buf_1,
.in.pic_w = w,
.in.pic_h = h,
.in.block_w = w,
.in.block_h = h,
.in.block_offset_x = 0,
.in.block_offset_y = 0,
.in.srm_cm = buf_1_color_type_id,
.out.buffer = buf_2,
.out.buffer_size = buf_2_size,
.out.pic_w = w,
.out.pic_h = h,
.out.block_offset_x = 0,
.out.block_offset_y = 0,
.out.srm_cm = buf_2_color_type_id,
.rotation_angle = PPA_SRM_ROTATION_ANGLE_0,
.scale_x = 1.0,
.scale_y = 1.0,
.mode = PPA_TRANS_MODE_BLOCKING,
};
// A SRM client can request to do a SRM operation
TEST_ESP_OK(ppa_do_scale_rotate_mirror(ppa_client_srm_handle, &srm_oper_config));
// A non-SRM client can not request to do a SRM operation
TEST_ESP_ERR(ESP_ERR_INVALID_ARG, ppa_do_scale_rotate_mirror(ppa_client_blend_handle, &srm_oper_config));
ppa_blend_oper_config_t blend_oper_config = {
.in_bg.buffer = buf_1,
.in_bg.pic_w = w,
.in_bg.pic_h = h,
.in_bg.block_w = w,
.in_bg.block_h = h,
.in_bg.block_offset_x = 0,
.in_bg.block_offset_y = 0,
.in_bg.blend_cm = buf_1_color_type_id,
.in_fg.buffer = buf_2,
.in_fg.pic_w = w,
.in_fg.pic_h = h,
.in_fg.block_w = w,
.in_fg.block_h = h,
.in_fg.block_offset_x = 0,
.in_fg.block_offset_y = 0,
.in_fg.blend_cm = buf_2_color_type_id,
.out.buffer = buf_1,
.out.buffer_size = buf_1_size,
.out.pic_w = w,
.out.pic_h = h,
.out.block_offset_x = 0,
.out.block_offset_y = 0,
.out.blend_cm = buf_1_color_type_id,
.mode = PPA_TRANS_MODE_BLOCKING,
};
// A blend client can request to do a blend operation
TEST_ESP_OK(ppa_do_blend(ppa_client_blend_handle, &blend_oper_config));
// A non-blend client can not request to do a blend operation
TEST_ESP_ERR(ESP_ERR_INVALID_ARG, ppa_do_blend(ppa_client_fill_handle_b, &blend_oper_config));
ppa_fill_oper_config_t fill_oper_config = {
.out.buffer = buf_1,
.out.buffer_size = buf_1_size,
.out.pic_w = w,
.out.pic_h = h,
.out.block_offset_x = 0,
.out.block_offset_y = 0,
.out.fill_cm = buf_1_color_type_id,
.fill_block_w = w,
.fill_block_h = h,
.fill_argb_color = {
.val = 0xFF00FF00,
},
.mode = PPA_TRANS_MODE_NON_BLOCKING,
};
// A fill client can request to do a fill operation
TEST_ESP_OK(ppa_do_fill(ppa_client_fill_handle_a, &fill_oper_config));
// Another fill client can also request another fill operation at the same time
TEST_ESP_OK(ppa_do_fill(ppa_client_fill_handle_b, &fill_oper_config));
vTaskDelay(pdMS_TO_TICKS(500));
// Unregister all PPA clients
TEST_ESP_OK(ppa_unregister_client(ppa_client_srm_handle));
TEST_ESP_OK(ppa_unregister_client(ppa_client_blend_handle));
TEST_ESP_OK(ppa_unregister_client(ppa_client_fill_handle_a));
TEST_ESP_OK(ppa_unregister_client(ppa_client_fill_handle_b));
free(buf_1);
free(buf_2);
}
static bool ppa_trans_done_cb(ppa_client_handle_t ppa_client, ppa_event_data_t *event_data, void *user_data)
{
BaseType_t xHigherPriorityTaskWoken = pdFALSE;
SemaphoreHandle_t sem = (SemaphoreHandle_t)user_data;
xSemaphoreGiveFromISR(sem, &xHigherPriorityTaskWoken);
return (xHigherPriorityTaskWoken == pdTRUE);
}
TEST_CASE("ppa_pending_transactions_in_queue", "[PPA]")
{
// A big picture block takes longer time to process, desired for this test case
const uint32_t w = 1920;
const uint32_t h = 1080;
const uint32_t buf_1_color_type_id = COLOR_TYPE_ID(COLOR_SPACE_ARGB, COLOR_PIXEL_ARGB8888);
const uint32_t buf_2_color_type_id = COLOR_TYPE_ID(COLOR_SPACE_YUV, COLOR_PIXEL_YUV420);
color_space_pixel_format_t buf_1_cm = {
.color_type_id = buf_1_color_type_id,
};
color_space_pixel_format_t buf_2_cm = {
.color_type_id = buf_2_color_type_id,
};
uint32_t buf_1_size = w * h * color_hal_pixel_format_get_bit_depth(buf_1_cm) / 8;
uint32_t buf_2_size = ALIGN_UP(w * h * color_hal_pixel_format_get_bit_depth(buf_2_cm) / 8, 64);
uint8_t *buf_1 = heap_caps_aligned_calloc(4, buf_1_size, sizeof(uint8_t), MALLOC_CAP_SPIRAM | MALLOC_CAP_DMA);
TEST_ASSERT_NOT_NULL(buf_1);
uint8_t *buf_2 = heap_caps_aligned_calloc(4, buf_2_size, sizeof(uint8_t), MALLOC_CAP_SPIRAM | MALLOC_CAP_DMA);
TEST_ASSERT_NOT_NULL(buf_2);
// Register two PPA SRM clients with different max_pending_trans_num
ppa_client_handle_t ppa_client_a_handle;
ppa_client_handle_t ppa_client_b_handle;
ppa_client_config_t ppa_client_config = {
.oper_type = PPA_OPERATION_SRM,
};
TEST_ESP_OK(ppa_register_client(&ppa_client_config, &ppa_client_a_handle));
ppa_client_config.max_pending_trans_num = 3;
TEST_ESP_OK(ppa_register_client(&ppa_client_config, &ppa_client_b_handle));
ppa_event_callbacks_t cbs = {
.on_trans_done = ppa_trans_done_cb,
};
ppa_client_register_event_callbacks(ppa_client_a_handle, &cbs);
SemaphoreHandle_t sem = xSemaphoreCreateBinary();
ppa_srm_oper_config_t oper_config = {
.in.buffer = buf_1,
.in.pic_w = w,
.in.pic_h = h,
.in.block_w = w,
.in.block_h = h,
.in.block_offset_x = 0,
.in.block_offset_y = 0,
.in.srm_cm = buf_1_color_type_id,
.out.buffer = buf_2,
.out.buffer_size = buf_2_size,
.out.pic_w = w,
.out.pic_h = h,
.out.block_offset_x = 0,
.out.block_offset_y = 0,
.out.srm_cm = buf_2_color_type_id,
.rotation_angle = PPA_SRM_ROTATION_ANGLE_0,
.scale_x = 1.0,
.scale_y = 1.0,
.user_data = (void *)sem,
.mode = PPA_TRANS_MODE_NON_BLOCKING,
};
TEST_ESP_OK(ppa_do_scale_rotate_mirror(ppa_client_a_handle, &oper_config));
// Another transaction cannot be accept since client_a can only hold one transaction
TEST_ESP_ERR(ESP_FAIL, ppa_do_scale_rotate_mirror(ppa_client_a_handle, &oper_config));
// Wait for the last transaction finishes
xSemaphoreTake(sem, portMAX_DELAY);
// Then a new transaction can be accepted again
TEST_ESP_OK(ppa_do_scale_rotate_mirror(ppa_client_a_handle, &oper_config));
// Client can not be unregistered when there are unfinished transactions
TEST_ESP_ERR(ESP_ERR_INVALID_STATE, ppa_unregister_client(ppa_client_a_handle));
oper_config.mode = PPA_TRANS_MODE_BLOCKING;
TEST_ESP_OK(ppa_do_scale_rotate_mirror(ppa_client_b_handle, &oper_config));
// Every PPA engine can only process one operation at a time
// Transactions are being processed with First-In-First-Out
// So, at the moment, the new transaction requested by client_b has finished, the last transaction requested by client_a for sure has finished
TEST_ASSERT(xSemaphoreTake(sem, 0) == pdTRUE);
// client_b can accept more than one transactions
oper_config.mode = PPA_TRANS_MODE_NON_BLOCKING;
TEST_ESP_OK(ppa_do_scale_rotate_mirror(ppa_client_b_handle, &oper_config));
TEST_ESP_OK(ppa_do_scale_rotate_mirror(ppa_client_b_handle, &oper_config));
oper_config.mode = PPA_TRANS_MODE_BLOCKING;
TEST_ESP_OK(ppa_do_scale_rotate_mirror(ppa_client_b_handle, &oper_config));
// The last transaction requested is with BLOCKING mode, so the last call to ppa_do_scale_rotate_mirror returned means all transactions finished
// Unregister all PPA clients
TEST_ESP_OK(ppa_unregister_client(ppa_client_a_handle));
TEST_ESP_OK(ppa_unregister_client(ppa_client_b_handle));
vSemaphoreDelete(sem);
free(buf_1);
free(buf_2);
}
TEST_CASE("ppa_srm_basic_data_correctness_check", "[PPA]")
{
const uint32_t w = 4;
const uint32_t h = 4;
const uint32_t block_w = 3;
const uint32_t block_h = 3;
const uint32_t in_block_offset_x = 1;
const uint32_t in_block_offset_y = 1;
const uint32_t out_block_offset_x = 1;
const uint32_t out_block_offset_y = 0;
const ppa_srm_color_mode_t cm = PPA_SRM_COLOR_MODE_RGB565;
const ppa_srm_rotation_angle_t rotation = PPA_SRM_ROTATION_ANGLE_90; // CCW
const float scale_x = 1.0;
const float scale_y = 1.0;
color_space_pixel_format_t buf_cm = {
.color_type_id = cm,
};
const uint32_t buf_len = w * h * color_hal_pixel_format_get_bit_depth(buf_cm) / 8; // 32
uint32_t out_buf_size = ALIGN_UP(buf_len, 64);
uint8_t *out_buf = heap_caps_aligned_calloc(4, out_buf_size, sizeof(uint8_t), MALLOC_CAP_INTERNAL | MALLOC_CAP_8BIT | MALLOC_CAP_DMA);
TEST_ASSERT_NOT_NULL(out_buf);
esp_cache_msync((void *)out_buf, out_buf_size, ESP_CACHE_MSYNC_FLAG_DIR_C2M);
const uint16_t in_buf[16] = {
0xFFFF, 0xFFFF, 0xFFFF, 0xFFFF,
// /*******************************/
0xFFFF, /**/ 0x8080, 0x8080, 0x8080, /**/
0xFFFF, /**/ 0x8F80, 0x8F80, 0x8F80, /**/
0xFFFF, /**/ 0xFF80, 0xFF80, 0xFF80, /**/
// /*******************************/
};
// Expected SRM output
const uint16_t out_buf_expected[16] = {
// /*******************************/
0x0000, /**/ 0x8080, 0x8F80, 0xFF80, /**/
0x0000, /**/ 0x8080, 0x8F80, 0xFF80, /**/
0x0000, /**/ 0x8080, 0x8F80, 0xFF80, /**/
// /*******************************/
0x0000, 0x0000, 0x0000, 0x0000
};
ppa_client_handle_t ppa_client_handle;
ppa_client_config_t ppa_client_config = {
.oper_type = PPA_OPERATION_SRM,
.max_pending_trans_num = 1,
};
TEST_ESP_OK(ppa_register_client(&ppa_client_config, &ppa_client_handle));
ppa_srm_oper_config_t oper_config = {
.in.buffer = in_buf,
.in.pic_w = w,
.in.pic_h = h,
.in.block_w = block_w,
.in.block_h = block_h,
.in.block_offset_x = in_block_offset_x,
.in.block_offset_y = in_block_offset_y,
.in.srm_cm = cm,
.out.buffer = out_buf,
.out.buffer_size = out_buf_size,
.out.pic_w = w,
.out.pic_h = h,
.out.block_offset_x = out_block_offset_x,
.out.block_offset_y = out_block_offset_y,
.out.srm_cm = cm,
.rotation_angle = rotation,
.scale_x = scale_x,
.scale_y = scale_y,
.rgb_swap = 0,
.byte_swap = 0,
.mode = PPA_TRANS_MODE_BLOCKING,
};
TEST_ESP_OK(ppa_do_scale_rotate_mirror(ppa_client_handle, &oper_config));
// Check result
for (int i = 0; i < buf_len; i++) {
if (i % 8 == 0) {
printf("\n");
}
printf("0x%02X ", out_buf[i]);
}
printf("\n");
TEST_ASSERT_EQUAL_UINT8_ARRAY((void *)out_buf_expected, (void *)out_buf, buf_len);
TEST_ESP_OK(ppa_unregister_client(ppa_client_handle));
free(out_buf);
}
TEST_CASE("ppa_blend_basic_data_correctness_check", "[PPA]")
{
const uint32_t w = 2;
const uint32_t h = 2;
const uint32_t block_w = 1;
const uint32_t block_h = 1;
const uint32_t block_offset_x = 1;
const uint32_t block_offset_y = 1;
const ppa_blend_color_mode_t in_bg_cm = PPA_BLEND_COLOR_MODE_RGB888;
const ppa_blend_color_mode_t in_fg_cm = PPA_BLEND_COLOR_MODE_ARGB8888;
const ppa_blend_color_mode_t out_cm = PPA_BLEND_COLOR_MODE_ARGB8888;
color_space_pixel_format_t bg_buf_cm = {
.color_type_id = in_bg_cm,
};
const uint32_t bg_buf_len __attribute__((unused)) = w * h * color_hal_pixel_format_get_bit_depth(bg_buf_cm) / 8; // 12
color_space_pixel_format_t fg_buf_cm = {
.color_type_id = in_fg_cm,
};
const uint32_t fg_buf_len __attribute__((unused)) = w * h * color_hal_pixel_format_get_bit_depth(fg_buf_cm) / 8; // 16
const uint32_t out_buf_len = fg_buf_len; // 16
// We will make the output write to the in_fg_buffer, therefore, it has cache line alignment requirement
const uint32_t out_buf_size = 64;
// Alpha of BG will be scaled by 0.8
// Alpha of FG will be inverted
const float bg_alpha_scale_ratio = 0.8;
const uint8_t in_bg_buf[12] = {
0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
// /*************************/
0xFF, 0xFF, 0xFF, /**/ 0x80, 0x40, 0xA0, /**/
// /*************************/
};
uint8_t in_fg_buf[64] __attribute__((aligned(64))) = {
0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
// /*******************************/
0xFF, 0xFF, 0xFF, 0xFF, /**/ 0x00, 0x80, 0x80, 0xC0, /**/
// /* (B) (G) (R) (A) */
// /*******************************/
[16 ... 63] = 0,
};
uint8_t *out_buf = in_fg_buf;
// Expected blend output
// Alpha Blending calculation:
// A_bg' = (255 * 0.8) / 255 = 0.8, A_fg' = (255 - 0xC0) / 255 = 0.247
// A_out = 0.8 + 0.247 - 0.8 * 0.247 = 0.849 (216 -> 0xD8)
// C_out_b = (0x80 * 0.8 * (1 - 0.247) + 0x00 * 0.247) / 0.849 = 91 -> 0x5B
// C_out_g = (0x40 * 0.8 * (1 - 0.247) + 0x80 * 0.247) / 0.849 = 83 -> 0x53
// C_out_g = (0xA0 * 0.8 * (1 - 0.247) + 0x80 * 0.247) / 0.849 = 150 -> 0x96
const uint8_t out_buf_expected[16] = {
0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF,
// /*******************************/
0xFF, 0xFF, 0xFF, 0xFF, /**/ 0x5B, 0x53, 0x96, 0xD8, /**/
// /*******************************/
};
ppa_client_handle_t ppa_client_handle;
ppa_client_config_t ppa_client_config = {
.oper_type = PPA_OPERATION_BLEND,
.max_pending_trans_num = 1,
};
TEST_ESP_OK(ppa_register_client(&ppa_client_config, &ppa_client_handle));
ppa_blend_oper_config_t oper_config = {
.in_bg.buffer = in_bg_buf,
.in_bg.pic_w = w,
.in_bg.pic_h = h,
.in_bg.block_w = block_w,
.in_bg.block_h = block_h,
.in_bg.block_offset_x = block_offset_x,
.in_bg.block_offset_y = block_offset_y,
.in_bg.blend_cm = in_bg_cm,
.in_fg.buffer = in_fg_buf,
.in_fg.pic_w = w,
.in_fg.pic_h = h,
.in_fg.block_w = block_w,
.in_fg.block_h = block_h,
.in_fg.block_offset_x = block_offset_x,
.in_fg.block_offset_y = block_offset_y,
.in_fg.blend_cm = in_fg_cm,
.out.buffer = out_buf,
.out.buffer_size = out_buf_size,
.out.pic_w = w,
.out.pic_h = h,
.out.block_offset_x = block_offset_x,
.out.block_offset_y = block_offset_y,
.out.blend_cm = out_cm,
.bg_alpha_update_mode = PPA_ALPHA_SCALE,
.bg_alpha_scale_ratio = bg_alpha_scale_ratio,
.fg_alpha_update_mode = PPA_ALPHA_INVERT,
.bg_ck_en = false,
.fg_ck_en = false,
.mode = PPA_TRANS_MODE_BLOCKING,
};
TEST_ESP_OK(ppa_do_blend(ppa_client_handle, &oper_config));
// Check result
for (int i = 0; i < out_buf_len; i++) {
if (i % 8 == 0) {
printf("\n");
}
printf("0x%02X ", out_buf[i]);
}
printf("\n");
TEST_ASSERT_EQUAL_UINT8_ARRAY((void *)out_buf_expected, (void *)out_buf, out_buf_len);
TEST_ESP_OK(ppa_unregister_client(ppa_client_handle));
}
TEST_CASE("ppa_fill_basic_data_correctness_check", "[PPA]")
{
const uint32_t w = 80;
const uint32_t h = 120;
const uint32_t block_w = 80;
const uint32_t block_h = 80;
const uint32_t block_offset_x = 0;
const uint32_t block_offset_y = 40;
const ppa_fill_color_mode_t out_cm = PPA_FILL_COLOR_MODE_RGB565;
const color_pixel_argb8888_data_t fill_color = {.a = 0x80, .r = 0xFF, .g = 0x55, .b = 0xAA};
color_space_pixel_format_t out_pixel_format = {
.color_type_id = out_cm,
};
uint32_t out_pixel_depth = color_hal_pixel_format_get_bit_depth(out_pixel_format); // bits
uint32_t out_buf_len = w * h * out_pixel_depth / 8;
uint32_t out_buf_size = ALIGN_UP(out_buf_len, 64);
uint8_t *out_buf = heap_caps_aligned_calloc(4, out_buf_size, sizeof(uint8_t), MALLOC_CAP_INTERNAL | MALLOC_CAP_8BIT | MALLOC_CAP_DMA);
TEST_ASSERT_NOT_NULL(out_buf);
memset(out_buf, 0xFF, out_buf_len);
ppa_client_handle_t ppa_client_handle;
ppa_client_config_t ppa_client_config = {
.oper_type = PPA_OPERATION_FILL,
.max_pending_trans_num = 1,
};
TEST_ESP_OK(ppa_register_client(&ppa_client_config, &ppa_client_handle));
ppa_fill_oper_config_t oper_config = {
.out.buffer = out_buf,
.out.buffer_size = out_buf_size,
.out.pic_w = w,
.out.pic_h = h,
.out.block_offset_x = block_offset_x,
.out.block_offset_y = block_offset_y,
.out.fill_cm = out_cm,
.fill_block_w = block_w,
.fill_block_h = block_h,
.fill_argb_color = fill_color,
.mode = PPA_TRANS_MODE_BLOCKING,
};
TEST_ESP_OK(ppa_do_fill(ppa_client_handle, &oper_config));
// Check result
const color_pixel_rgb565_data_t fill_pixel_expected = {.r = fill_color.r >> 3,
.g = fill_color.g >> 2,
.b = fill_color.b >> 3,
};
TEST_ASSERT_EACH_EQUAL_UINT16(fill_pixel_expected.val, (void *)((uint32_t)out_buf + w * block_offset_y * out_pixel_depth / 8), block_w * block_h);
TEST_ESP_OK(ppa_unregister_client(ppa_client_handle));
free(out_buf);
}
/* All performance tests are tested under the following situations:
* - Testing PPA speed where in_buffer(s) and out_buffer all located in PSRAM
* - Only 2D-DMA is using PSRAM
* - 2D-DMA burst length is at maximum 128B
* - Input and output color mode is ARGB8888 (maximizing 2D-DMA data transafer amount)
*
* The time spend (T) to complete a PPA transaction is proportional to the amount of pixels (x) need to be processed.
* T = k * x + b
* k = (T - b) / x
*/
#define PPA_SRM_MIN_PERFORMANCE_PX_PER_SEC (21000 * 1000) // k_min
#define PPA_SRM_TIME_OFFSET (-26000) // b_approx
#define PPA_BLEND_MIN_PERFORMANCE_PX_PER_SEC (31500 * 1000) // k_min
#define PPA_BLEND_TIME_OFFSET (-37150) // b_approx
#define PPA_FILL_MIN_PERFORMANCE_PX_PER_SEC (150000 * 1000) // k_min
#define PPA_FILL_TIME_OFFSET (-106000) // b_approx
TEST_CASE("ppa_srm_performance", "[PPA]")
{
// Configurable parameters
const uint32_t w = 1920; // 1920 / 1280 / 800 / 640
const uint32_t h = 1080; // 1080 / 720 / 480
const uint32_t block_w = w;
const uint32_t block_h = h;
const ppa_srm_color_mode_t in_cm = PPA_SRM_COLOR_MODE_ARGB8888;
const ppa_srm_color_mode_t out_cm = PPA_SRM_COLOR_MODE_ARGB8888;
const ppa_srm_rotation_angle_t rotation = PPA_SRM_ROTATION_ANGLE_0;
const float scale_x = 1.0;
const float scale_y = 1.0;
color_space_pixel_format_t in_pixel_format = {
.color_type_id = in_cm,
};
color_space_pixel_format_t out_pixel_format = {
.color_type_id = out_cm,
};
uint32_t in_buf_size = w * h * color_hal_pixel_format_get_bit_depth(in_pixel_format) / 8;
uint32_t out_buf_size = ALIGN_UP(w * h * color_hal_pixel_format_get_bit_depth(out_pixel_format) / 8, 64);
uint8_t *out_buf = heap_caps_aligned_calloc(4, out_buf_size, sizeof(uint8_t), MALLOC_CAP_SPIRAM | MALLOC_CAP_DMA);
TEST_ASSERT_NOT_NULL(out_buf);
uint8_t *in_buf = heap_caps_aligned_calloc(4, in_buf_size, sizeof(uint8_t), MALLOC_CAP_SPIRAM | MALLOC_CAP_DMA);
TEST_ASSERT_NOT_NULL(in_buf);
uint8_t *ptr = in_buf;
for (int x = 0; x < in_buf_size; x++) {
ptr[x] = x;
}
ppa_client_handle_t ppa_client_handle;
ppa_client_config_t ppa_client_config = {
.oper_type = PPA_OPERATION_SRM,
.max_pending_trans_num = 1,
};
TEST_ESP_OK(ppa_register_client(&ppa_client_config, &ppa_client_handle));
uint32_t out_pic_w = (rotation == PPA_SRM_ROTATION_ANGLE_0 || rotation == PPA_SRM_ROTATION_ANGLE_180) ? w : h;
uint32_t out_pic_h = (rotation == PPA_SRM_ROTATION_ANGLE_0 || rotation == PPA_SRM_ROTATION_ANGLE_180) ? h : w;
ppa_srm_oper_config_t oper_config = {
.in.buffer = in_buf,
.in.pic_w = w,
.in.pic_h = h,
.in.block_w = block_w,
.in.block_h = block_h,
.in.block_offset_x = 0,
.in.block_offset_y = 0,
.in.srm_cm = in_cm,
.out.buffer = out_buf,
.out.buffer_size = out_buf_size,
.out.pic_w = out_pic_w,
.out.pic_h = out_pic_h,
.out.block_offset_x = 0,
.out.block_offset_y = 0,
.out.srm_cm = out_cm,
.rotation_angle = rotation,
.scale_x = scale_x,
.scale_y = scale_y,
.rgb_swap = 0,
.byte_swap = 0,
.mode = PPA_TRANS_MODE_BLOCKING,
};
ccomp_timer_start();
TEST_ESP_OK(ppa_do_scale_rotate_mirror(ppa_client_handle, &oper_config));
int64_t oper_time = ccomp_timer_stop(); // us
printf("PPA SRM - Process Time: %lld us\n", oper_time);
// Check performance
uint64_t num_pixels_processed = block_w * block_h;
uint64_t px_per_second = (num_pixels_processed - PPA_SRM_TIME_OFFSET) * 1000 * 1000 / oper_time;
printf("PPA SRM performance = %lld pixels/sec\n", px_per_second);
TEST_ASSERT_GREATER_THAN(PPA_SRM_MIN_PERFORMANCE_PX_PER_SEC, px_per_second);
TEST_ESP_OK(ppa_unregister_client(ppa_client_handle));
free(in_buf);
free(out_buf);
}
TEST_CASE("ppa_blend_performance", "[PPA]")
{
// Configurable parameters
const uint32_t w = 1280;
const uint32_t h = 720;
const uint32_t block_w = w;
const uint32_t block_h = h;
const ppa_blend_color_mode_t in_bg_cm = PPA_BLEND_COLOR_MODE_ARGB8888;
const ppa_blend_color_mode_t in_fg_cm = PPA_BLEND_COLOR_MODE_ARGB8888;
const ppa_blend_color_mode_t out_cm = PPA_BLEND_COLOR_MODE_ARGB8888;
color_space_pixel_format_t in_bg_pixel_format = {
.color_type_id = in_bg_cm,
};
color_space_pixel_format_t in_fg_pixel_format = {
.color_type_id = in_fg_cm,
};
color_space_pixel_format_t out_pixel_format = {
.color_type_id = out_cm,
};
uint32_t in_bg_buf_size = w * h * color_hal_pixel_format_get_bit_depth(in_bg_pixel_format) / 8;
uint32_t in_fg_buf_size = w * h * color_hal_pixel_format_get_bit_depth(in_fg_pixel_format) / 8;
uint32_t out_buf_size = ALIGN_UP(w * h * color_hal_pixel_format_get_bit_depth(out_pixel_format) / 8, 64);
uint8_t *out_buf = heap_caps_aligned_calloc(4, out_buf_size, sizeof(uint8_t), MALLOC_CAP_SPIRAM | MALLOC_CAP_DMA);
TEST_ASSERT_NOT_NULL(out_buf);
uint8_t *in_bg_buf = heap_caps_aligned_calloc(4, in_bg_buf_size, sizeof(uint8_t), MALLOC_CAP_SPIRAM | MALLOC_CAP_DMA);
TEST_ASSERT_NOT_NULL(in_bg_buf);
uint8_t *in_fg_buf = heap_caps_aligned_calloc(4, in_fg_buf_size, sizeof(uint8_t), MALLOC_CAP_SPIRAM | MALLOC_CAP_DMA);
TEST_ASSERT_NOT_NULL(in_fg_buf);
uint8_t *ptr = in_bg_buf;
for (int x = 0; x < in_bg_buf_size; x++) {
ptr[x] = x & 0x55;
}
ptr = in_fg_buf;
for (int x = 0; x < in_fg_buf_size; x++) {
ptr[x] = x & 0xAA;
}
ppa_client_handle_t ppa_client_handle;
ppa_client_config_t ppa_client_config = {
.oper_type = PPA_OPERATION_BLEND,
.max_pending_trans_num = 1,
};
TEST_ESP_OK(ppa_register_client(&ppa_client_config, &ppa_client_handle));
ppa_blend_oper_config_t oper_config = {
.in_bg.buffer = in_bg_buf,
.in_bg.pic_w = w,
.in_bg.pic_h = h,
.in_bg.block_w = block_w,
.in_bg.block_h = block_h,
.in_bg.block_offset_x = 0,
.in_bg.block_offset_y = 0,
.in_bg.blend_cm = in_bg_cm,
.in_fg.buffer = in_fg_buf,
.in_fg.pic_w = w,
.in_fg.pic_h = h,
.in_fg.block_w = block_w,
.in_fg.block_h = block_h,
.in_fg.block_offset_x = 0,
.in_fg.block_offset_y = 0,
.in_fg.blend_cm = in_fg_cm,
.out.buffer = out_buf,
.out.buffer_size = out_buf_size,
.out.pic_w = w,
.out.pic_h = h,
.out.block_offset_x = 0,
.out.block_offset_y = 0,
.out.blend_cm = out_cm,
.bg_ck_en = false,
.fg_ck_en = false,
.mode = PPA_TRANS_MODE_BLOCKING,
};
ccomp_timer_start();
TEST_ESP_OK(ppa_do_blend(ppa_client_handle, &oper_config));
int64_t oper_time = ccomp_timer_stop();
printf("PPA Blend - Process Time: %lld us\n", oper_time);
// Check performance
uint64_t num_pixels_processed = block_w * block_h;
uint64_t px_per_second = (num_pixels_processed - PPA_BLEND_TIME_OFFSET) * 1000 * 1000 / oper_time;
printf("PPA Blend performance = %lld pixels/sec\n", px_per_second);
TEST_ASSERT_GREATER_THAN(PPA_BLEND_MIN_PERFORMANCE_PX_PER_SEC, px_per_second);
TEST_ESP_OK(ppa_unregister_client(ppa_client_handle));
free(in_bg_buf);
free(in_fg_buf);
free(out_buf);
}
TEST_CASE("ppa_fill_performance", "[PPA]")
{
// Configurable parameters
const uint32_t w = 1280;
const uint32_t h = 720;
const uint32_t block_w = 800;
const uint32_t block_h = 480;
const ppa_fill_color_mode_t out_cm = PPA_FILL_COLOR_MODE_RGB565;
color_space_pixel_format_t out_pixel_format = {
.color_type_id = out_cm,
};
uint32_t out_buf_size = ALIGN_UP(w * h * color_hal_pixel_format_get_bit_depth(out_pixel_format) / 8, 64);
uint8_t *out_buf = heap_caps_aligned_calloc(4, out_buf_size, sizeof(uint8_t), MALLOC_CAP_SPIRAM | MALLOC_CAP_DMA);
TEST_ASSERT_NOT_NULL(out_buf);
ppa_client_handle_t ppa_client_handle;
ppa_client_config_t ppa_client_config = {
.oper_type = PPA_OPERATION_FILL,
.max_pending_trans_num = 1,
};
TEST_ESP_OK(ppa_register_client(&ppa_client_config, &ppa_client_handle));
ppa_fill_oper_config_t oper_config = {
.out.buffer = out_buf,
.out.buffer_size = out_buf_size,
.out.pic_w = w,
.out.pic_h = h,
.out.block_offset_x = 0,
.out.block_offset_y = 0,
.out.fill_cm = out_cm,
.fill_block_w = block_w,
.fill_block_h = block_h,
.fill_argb_color = {
.val = 0xFF00FFFF,
},
.mode = PPA_TRANS_MODE_BLOCKING,
};
ccomp_timer_start();
TEST_ESP_OK(ppa_do_fill(ppa_client_handle, &oper_config));
int64_t oper_time = ccomp_timer_stop();
printf("PPA Fill - Process Time: %lld us\n", oper_time);
// Check performance
uint64_t num_pixels_processed = block_w * block_h;
uint64_t px_per_second = (num_pixels_processed - PPA_FILL_TIME_OFFSET) * 1000 * 1000 / oper_time;
printf("PPA Blend performance = %lld pixels/sec\n", px_per_second);
TEST_ASSERT_GREATER_THAN(PPA_FILL_MIN_PERFORMANCE_PX_PER_SEC, px_per_second);
TEST_ESP_OK(ppa_unregister_client(ppa_client_handle));
free(out_buf);
}