Merge branch 'fix/spi_polling_speed_test' into 'master'

spi: speed up polling performance and fix several tests

Closes #33

See merge request idf/esp-idf!3530
This commit is contained in:
Angus Gratton 2018-10-26 12:41:00 +08:00
commit 5bcd864430
5 changed files with 71 additions and 45 deletions

View file

@ -135,7 +135,6 @@ We have two bits to control the interrupt:
#include "freertos/semphr.h"
#include "freertos/xtensa_api.h"
#include "freertos/task.h"
#include "freertos/ringbuf.h"
#include "soc/soc.h"
#include "soc/soc_memory_layout.h"
#include "soc/dport_reg.h"
@ -709,7 +708,7 @@ static SPI_MASTER_ISR_ATTR esp_err_t device_acquire_bus_internal(spi_device_t *h
/* This function check for whether the ISR is done, if not, block until semaphore given.
*/
static inline esp_err_t device_wait_for_isr_idle(spi_device_t *handle, TickType_t wait)
static inline SPI_MASTER_ISR_ATTR esp_err_t device_wait_for_isr_idle(spi_device_t *handle, TickType_t wait)
{
//quickly skip if the isr is already free
if (!handle->host->isr_free) {
@ -767,7 +766,7 @@ static SPI_MASTER_ISR_ATTR void device_release_bus_internal(spi_host_t *host)
}
}
static inline bool device_is_polling(spi_device_t *handle)
static inline SPI_MASTER_ISR_ATTR bool device_is_polling(spi_device_t *handle)
{
return atomic_load(&handle->host->acquire_cs) == handle->id && handle->host->polling;
}
@ -1088,7 +1087,7 @@ static void SPI_MASTER_ISR_ATTR spi_intr(void *arg)
if (do_yield) portYIELD_FROM_ISR();
}
static esp_err_t check_trans_valid(spi_device_handle_t handle, spi_transaction_t *trans_desc)
static SPI_MASTER_ISR_ATTR esp_err_t check_trans_valid(spi_device_handle_t handle, spi_transaction_t *trans_desc)
{
SPI_CHECK(handle!=NULL, "invalid dev handle", ESP_ERR_INVALID_ARG);
spi_host_t *host = handle->host;
@ -1112,7 +1111,7 @@ static esp_err_t check_trans_valid(spi_device_handle_t handle, spi_transaction_t
return ESP_OK;
}
static SPI_MASTER_ATTR void uninstall_priv_desc(spi_trans_priv_t* trans_buf)
static SPI_MASTER_ISR_ATTR void uninstall_priv_desc(spi_trans_priv_t* trans_buf)
{
spi_transaction_t *trans_desc = trans_buf->trans;
if ((void *)trans_buf->buffer_to_send != &trans_desc->tx_data[0] &&
@ -1131,7 +1130,7 @@ static SPI_MASTER_ATTR void uninstall_priv_desc(spi_trans_priv_t* trans_buf)
}
}
static SPI_MASTER_ATTR esp_err_t setup_priv_desc(spi_transaction_t *trans_desc, spi_trans_priv_t* new_desc, bool isdma)
static SPI_MASTER_ISR_ATTR esp_err_t setup_priv_desc(spi_transaction_t *trans_desc, spi_trans_priv_t* new_desc, bool isdma)
{
*new_desc = (spi_trans_priv_t) { .trans = trans_desc, };

View file

@ -29,7 +29,6 @@
#include "freertos/semphr.h"
#include "freertos/xtensa_api.h"
#include "freertos/task.h"
#include "freertos/ringbuf.h"
#include "soc/soc.h"
#include "soc/soc_memory_layout.h"
#include "soc/dport_reg.h"

View file

@ -507,7 +507,6 @@ TEST_CASE("SPI Master no response when switch from host1 (HSPI) to host2 (VSPI)"
TEST_ASSERT(spi_bus_free(host) == ESP_OK);
}
IRAM_ATTR static uint32_t data_iram[80];
DRAM_ATTR static uint32_t data_dram[80]={0};
//force to place in code area.
static const uint8_t data_drom[320+3] = {
@ -544,13 +543,24 @@ TEST_CASE("SPI Master DMA test, TX and RX in different regions", "[spi]")
#ifdef CONFIG_SPIRAM_SUPPORT
//test psram if enabled
ESP_LOGI(TAG, "testing PSRAM...");
uint32_t* data_malloc = (uint32_t*)heap_caps_calloc(1, 324, MALLOC_CAP_SPIRAM);
uint32_t* data_malloc = (uint32_t*)heap_caps_malloc(324, MALLOC_CAP_SPIRAM);
TEST_ASSERT(esp_ptr_external_ram(data_malloc));
#else
uint32_t* data_malloc = (uint32_t*)heap_caps_calloc(1, 324, MALLOC_CAP_DMA);
uint32_t* data_malloc = (uint32_t*)heap_caps_malloc(324, MALLOC_CAP_DMA);
TEST_ASSERT(esp_ptr_in_dram(data_malloc));
#endif
TEST_ASSERT(data_malloc != NULL);
//refer to soc_memory_layout.c
uint32_t* data_iram = (uint32_t*)heap_caps_malloc(324, MALLOC_CAP_EXEC);
TEST_ASSERT(data_iram != NULL);
ESP_LOGI(TAG, "iram: %p, dram: %p", data_iram, data_dram);
ESP_LOGI(TAG, "drom: %p, malloc: %p", data_drom, data_malloc);
TEST_ASSERT(esp_ptr_in_dram(data_dram));
TEST_ASSERT(esp_ptr_in_iram(data_iram));
TEST_ASSERT(esp_ptr_in_drom(data_drom));
srand(52);
for (int i = 0; i < 320/4; i++) {
data_iram[i] = rand();
@ -577,8 +587,6 @@ TEST_CASE("SPI Master DMA test, TX and RX in different regions", "[spi]")
static spi_transaction_t trans[TEST_REGION_SIZE];
int x;
ESP_LOGI(TAG, "iram: %p, dram: %p", data_iram, data_dram);
ESP_LOGI(TAG, "drom: %p, malloc: %p", data_drom, data_malloc);
memset(trans, 0, sizeof(trans));
@ -619,6 +627,7 @@ TEST_CASE("SPI Master DMA test, TX and RX in different regions", "[spi]")
TEST_ASSERT(spi_bus_remove_device(spi) == ESP_OK);
TEST_ASSERT(spi_bus_free(HSPI_HOST) == ESP_OK);
free(data_malloc);
free(data_iram);
}
//this part tests 3 DMA issues in master mode, full-duplex in IDF2.1
@ -1350,9 +1359,26 @@ static void sorted_array_insert(uint32_t* array, int* size, uint32_t item)
#define TEST_TIMES 11
TEST_CASE("spi_speed","[spi]")
static IRAM_ATTR void spi_transmit_measure(spi_device_handle_t spi, spi_transaction_t* trans, uint32_t* t_flight)
{
RECORD_TIME_PREPARE();
spi_device_transmit(spi, trans); // prime the flash cache
RECORD_TIME_START();
spi_device_transmit(spi, trans);
RECORD_TIME_END(t_flight);
}
static IRAM_ATTR void spi_transmit_polling_measure(spi_device_handle_t spi, spi_transaction_t* trans, uint32_t* t_flight)
{
RECORD_TIME_PREPARE();
spi_device_polling_transmit(spi, trans); // prime the flash cache
RECORD_TIME_START();
spi_device_polling_transmit(spi, trans);
RECORD_TIME_END(t_flight);
}
TEST_CASE("spi_speed","[spi]")
{
uint32_t t_flight;
//to get rid of the influence of randomly interrupts, we measured the performance by median value
uint32_t t_flight_sorted[TEST_TIMES];
@ -1372,16 +1398,13 @@ TEST_CASE("spi_speed","[spi]")
//record flight time by isr, with DMA
t_flight_num = 0;
for (int i = 0; i < TEST_TIMES; i++) {
spi_device_transmit(spi, &trans); // prime the flash cache
RECORD_TIME_START();
spi_device_transmit(spi, &trans);
RECORD_TIME_END(&t_flight);
spi_transmit_measure(spi, &trans, &t_flight);
sorted_array_insert(t_flight_sorted, &t_flight_num, t_flight);
}
TEST_PERFORMANCE_LESS_THAN(SPI_PER_TRANS_NO_POLLING, "%d us", t_flight_sorted[(TEST_TIMES+1)/2]);
for (int i = 0; i < TEST_TIMES; i++) {
ESP_LOGI(TAG, "%d", t_flight_sorted[i]);
}
TEST_PERFORMANCE_LESS_THAN(SPI_PER_TRANS_NO_POLLING, "%d us", t_flight_sorted[(TEST_TIMES+1)/2]);
//acquire the bus to send polling transactions faster
ret = spi_device_acquire_bus(spi, portMAX_DELAY);
@ -1390,16 +1413,13 @@ TEST_CASE("spi_speed","[spi]")
//record flight time by polling and with DMA
t_flight_num = 0;
for (int i = 0; i < TEST_TIMES; i++) {
spi_device_polling_transmit(spi, &trans); // prime the flash cache
RECORD_TIME_START();
spi_device_polling_transmit(spi, &trans);
RECORD_TIME_END(&t_flight);
spi_transmit_polling_measure(spi, &trans, &t_flight);
sorted_array_insert(t_flight_sorted, &t_flight_num, t_flight);
}
TEST_PERFORMANCE_LESS_THAN(SPI_PER_TRANS_POLLING, "%d us", t_flight_sorted[(TEST_TIMES+1)/2]);
for (int i = 0; i < TEST_TIMES; i++) {
ESP_LOGI(TAG, "%d", t_flight_sorted[i]);
}
TEST_PERFORMANCE_LESS_THAN(SPI_PER_TRANS_POLLING, "%d us", t_flight_sorted[(TEST_TIMES+1)/2]);
//release the bus
spi_device_release_bus(spi);
@ -1410,16 +1430,13 @@ TEST_CASE("spi_speed","[spi]")
//record flight time by isr, without DMA
t_flight_num = 0;
for (int i = 0; i < TEST_TIMES; i++) {
spi_device_transmit(spi, &trans); // prime the flash cache
RECORD_TIME_START();
spi_device_transmit(spi, &trans);
RECORD_TIME_END(&t_flight);
spi_transmit_measure(spi, &trans, &t_flight);
sorted_array_insert(t_flight_sorted, &t_flight_num, t_flight);
}
TEST_PERFORMANCE_LESS_THAN( SPI_PER_TRANS_NO_POLLING_NO_DMA, "%d us", t_flight_sorted[(TEST_TIMES+1)/2]);
for (int i = 0; i < TEST_TIMES; i++) {
ESP_LOGI(TAG, "%d", t_flight_sorted[i]);
}
TEST_PERFORMANCE_LESS_THAN( SPI_PER_TRANS_NO_POLLING_NO_DMA, "%d us", t_flight_sorted[(TEST_TIMES+1)/2]);
//acquire the bus to send polling transactions faster
ret = spi_device_acquire_bus(spi, portMAX_DELAY);
@ -1427,16 +1444,13 @@ TEST_CASE("spi_speed","[spi]")
//record flight time by polling, without DMA
t_flight_num = 0;
for (int i = 0; i < TEST_TIMES; i++) {
spi_device_polling_transmit(spi, &trans); // prime the flash cache
RECORD_TIME_START();
spi_device_polling_transmit(spi, &trans);
RECORD_TIME_END(&t_flight);
spi_transmit_polling_measure(spi, &trans, &t_flight);
sorted_array_insert(t_flight_sorted, &t_flight_num, t_flight);
}
TEST_PERFORMANCE_LESS_THAN(SPI_PER_TRANS_POLLING_NO_DMA, "%d us", t_flight_sorted[(TEST_TIMES+1)/2]);
for (int i = 0; i < TEST_TIMES; i++) {
ESP_LOGI(TAG, "%d", t_flight_sorted[i]);
}
TEST_PERFORMANCE_LESS_THAN(SPI_PER_TRANS_POLLING_NO_DMA, "%d us", t_flight_sorted[(TEST_TIMES+1)/2]);
//release the bus
spi_device_release_bus(spi);

View file

@ -59,16 +59,18 @@
#define APP_CPU_NUM (1)
/* Overall memory map */
#define SOC_IROM_LOW 0x400D0000
#define SOC_IROM_HIGH 0x40400000
#define SOC_DROM_LOW 0x3F400000
#define SOC_DROM_HIGH 0x3F800000
#define SOC_RTC_IRAM_LOW 0x400C0000
#define SOC_RTC_IRAM_HIGH 0x400C2000
#define SOC_RTC_DATA_LOW 0x50000000
#define SOC_RTC_DATA_HIGH 0x50002000
#define SOC_EXTRAM_DATA_LOW 0x3F800000
#define SOC_EXTRAM_DATA_HIGH 0x3FC00000
#define SOC_IROM_LOW 0x400D0000
#define SOC_IROM_HIGH 0x40400000
#define SOC_DROM_LOW 0x3F400000
#define SOC_DROM_HIGH 0x3F800000
#define SOC_DRAM_LOW 0x3FAE0000
#define SOC_DRAM_HIGH 0x40000000
#define SOC_RTC_IRAM_LOW 0x400C0000
#define SOC_RTC_IRAM_HIGH 0x400C2000
#define SOC_RTC_DATA_LOW 0x50000000
#define SOC_RTC_DATA_HIGH 0x50002000
#define SOC_EXTRAM_DATA_LOW 0x3F800000
#define SOC_EXTRAM_DATA_HIGH 0x3FC00000
#define DR_REG_DPORT_BASE 0x3ff00000
@ -130,7 +132,7 @@
//Registers Operation {{
#define ETS_UNCACHED_ADDR(addr) (addr)
#define ETS_CACHED_ADDR(addr) (addr)
#define ETS_CACHED_ADDR(addr) (addr)
#ifndef __ASSEMBLER__
#define BIT(nr) (1UL << (nr))

View file

@ -173,3 +173,15 @@ inline static bool IRAM_ATTR esp_ptr_internal(const void *p) {
inline static bool IRAM_ATTR esp_ptr_external_ram(const void *p) {
return ((intptr_t)p >= SOC_EXTRAM_DATA_LOW && (intptr_t)p < SOC_EXTRAM_DATA_HIGH);
}
inline static bool IRAM_ATTR esp_ptr_in_iram(const void *p) {
return ((intptr_t)p >= SOC_IRAM_LOW && (intptr_t)p < SOC_IRAM_HIGH);
}
inline static bool IRAM_ATTR esp_ptr_in_drom(const void *p) {
return ((intptr_t)p >= SOC_DROM_LOW && (intptr_t)p < SOC_DROM_HIGH);
}
inline static bool IRAM_ATTR esp_ptr_in_dram(const void *p) {
return ((intptr_t)p >= SOC_DRAM_LOW && (intptr_t)p < SOC_DRAM_HIGH);
}