fallensigh/ciciec2026_loongson
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

test(dma,fft): rewrite test programs for dma, fft and fft_dma

Browse Source
This commit is contained in:
FallenSigh
2026-04-13 17:26:12 +08:00
parent 63fb8f6cc1
commit 84155db987
3 changed files with 28 additions and 109 deletions
Download Patch File Download Diff File Expand all files Collapse all files

43
sdk/software/examples/dma/main.c
View File

@@ -1,9 +1,12 @@
#include <stdio.h> #include <stdio.h>
#include <string.h> #include <string.h>
#include <math.h> #include <math.h>
#include <stdbool.h>
#include <stdint.h> #include <stdint.h>
#include <common_func.h> #include <common_func.h>
#include <confreg_time.h> #include <confreg_time.h>
#include <dma.h>
// BSP板级支持包所需全局变量 // BSP板级支持包所需全局变量
unsigned long UART_BASE = 0xbf000000; unsigned long UART_BASE = 0xbf000000;
@@ -11,35 +14,12 @@ unsigned long CONFREG_TIMER_BASE = 0xbf20f100;
unsigned long CONFREG_CLOCKS_PER_SEC = 50000000L; unsigned long CONFREG_CLOCKS_PER_SEC = 50000000L;
unsigned long CORE_CLOCKS_PER_SEC = 33000000L; unsigned long CORE_CLOCKS_PER_SEC = 33000000L;
#define FFT_BASE 0xbf400000
#define FFT_IN_RE_BASE (FFT_BASE + 0x1000)
#define FFT_IN_IM_BASE (FFT_BASE + 0x2000)
#define FFT_OUT_RE_BASE (FFT_BASE + 0x3000)
#define FFT_OUT_IM_BASE (FFT_BASE + 0x4000)
#define FFT_CSR_REG (FFT_BASE + 0xF000)
#define FFT_CTRL_START (1 << 4)
#define FFT_STAT_DONE (1 << 1)
#define FFT_STAT_BUSY (1 << 0)
#define FFT_POINT_NUM 1024
#define DMA_BASE 0xbf300000
#define DMA_CTRL (DMA_BASE + 0x0000)
#define DMA_LEN (DMA_BASE + 0x0004)
#define DMA_SRC_ADDR (DMA_BASE + 0x0008)
#define DMA_DST_ADDR (DMA_BASE + 0x000c)
#define DMA_STATUS (DMA_BASE + 0x0010)
const float PI = 3.14159265358979323846;
// 加上 aligned(64) 是为了防止 DMA 突发传输时跨越缓存行或 AXI 非对齐边界 // 加上 aligned(64) 是为了防止 DMA 突发传输时跨越缓存行或 AXI 非对齐边界
uint32_t src_array[64] __attribute__((aligned(64))); uint32_t src_array[64] __attribute__((aligned(64)));
uint32_t dst_array[64] __attribute__((aligned(64))); uint32_t dst_array[64] __attribute__((aligned(64)));
int main(int argc, char** argv) int main(int argc, char** argv)
{ {
unsigned int dma_status = RegRead(DMA_STATUS);
printf("dma_status init = %x\n", dma_status);
uint32_t data_len = 64; uint32_t data_len = 64;
uint32_t byte_len = data_len * 4; uint32_t byte_len = data_len * 4;
@@ -65,22 +45,9 @@ int main(int argc, char** argv)
printf("uncached_src: %x\n", uncached_src); printf("uncached_src: %x\n", uncached_src);
printf("uncached_dst: %x\n", uncached_dst); printf("uncached_dst: %x\n", uncached_dst);
// 配置 DMA：DMA 只需要最纯粹的物理地址 // 配置 DMA：DMA 只需要最纯粹的物理地址
RegWrite(DMA_SRC_ADDR, phys_src_addr); dma_start_transfer(0, phys_src_addr, phys_dst_addr, byte_len);
RegWrite(DMA_DST_ADDR, phys_dst_addr); dma_wait_polling(0);
RegWrite(DMA_LEN, byte_len);
// burst_len = 15(16拍), burst_size = 2(4字节)
uint32_t burst_len = 15;
uint32_t burst_size = 2;
uint32_t ctrl_val = (burst_len << 6) | (burst_size << 3) | 0x01;
RegWrite(DMA_CTRL, ctrl_val);
// 等待 DMA 完成
while ((RegRead(DMA_STATUS) & 0x01) == 0) {
printf("polling...\n");
}
printf("dma passed!\n"); printf("dma passed!\n");

36
sdk/software/examples/fft/main.c
View File

@@ -1,8 +1,10 @@
#include <stdio.h> #include <stdio.h>
#include <string.h> #include <string.h>
#include <math.h> #include <math.h>
#include <common_func.h> #include <common_func.h>
#include <confreg_time.h> #include <confreg_time.h>
#include <fft.h>
// BSP板级支持包所需全局变量 // BSP板级支持包所需全局变量
unsigned long UART_BASE = 0xbf000000; unsigned long UART_BASE = 0xbf000000;
@@ -10,24 +12,6 @@ unsigned long CONFREG_TIMER_BASE = 0xbf20f100;
unsigned long CONFREG_CLOCKS_PER_SEC = 50000000L; unsigned long CONFREG_CLOCKS_PER_SEC = 50000000L;
unsigned long CORE_CLOCKS_PER_SEC = 33000000L; unsigned long CORE_CLOCKS_PER_SEC = 33000000L;
#define FFT_BASE 0xbf400000
#define FFT_IN_RE_BASE (FFT_BASE + 0x1000)
#define FFT_IN_IM_BASE (FFT_BASE + 0x2000)
#define FFT_OUT_RE_BASE (FFT_BASE + 0x3000)
#define FFT_OUT_IM_BASE (FFT_BASE + 0x4000)
#define FFT_CSR_REG (FFT_BASE + 0xF000)
#define FFT_CTRL_START (1 << 4)
#define FFT_STAT_DONE (1 << 1)
#define FFT_STAT_BUSY (1 << 0)
#define FFT_POINT_NUM 1024
#define DMA_BASE 0xbf300000
#define DMA_CTRL (DMA_BASE + 0x0000)
#define DMA_LEN (DMA_BASE + 0x0004)
#define DMA_SRC_ADDR (DMA_BASE + 0x0008)
#define DMA_DST_ADDR (DMA_BASE + 0x000c)
#define DMA_STATUS (DMA_BASE + 0x0010)
const float PI = 3.14159265358979323846; const float PI = 3.14159265358979323846;
// 读取定时器的当前Tick // 读取定时器的当前Tick
@@ -35,14 +19,12 @@ unsigned int get_timer_ticks() {
return RegRead(CONFREG_TIMER_BASE); return RegRead(CONFREG_TIMER_BASE);
} }
// ---------------------------------------------------------
// 软件FFT实现 (基2 DIT-FFT 算法) // 软件FFT实现 (基2 DIT-FFT 算法)
// ---------------------------------------------------------
void sw_fft(float re[], float im[], int n) { void sw_fft(float re[], float im[], int n) {
int i, j, k, l; int i, j, k, l;
float tr, ti, ur, ui, wr, wi; float tr, ti, ur, ui, wr, wi;
// 1. 比特翻转 (Bit Reversal) // 比特翻转 (Bit Reversal)
j = 0; j = 0;
for (i = 0; i < n - 1; i++) { for (i = 0; i < n - 1; i++) {
if (i < j) { if (i < j) {
@@ -58,7 +40,7 @@ void sw_fft(float re[], float im[], int n) {
j += k; j += k;
} }
// 2. 蝶形运算 (Butterfly Computation) // 蝶形运算 (Butterfly Computation)
for (l = 1; l < n; l *= 2) { for (l = 1; l < n; l *= 2) {
ur = 1.0; ur = 1.0;
ui = 0.0; ui = 0.0;
@@ -110,9 +92,7 @@ int main(int argc, char** argv)
unsigned int tick_start, tick_end; unsigned int tick_start, tick_end;
unsigned int hw_time, sw_time; unsigned int hw_time, sw_time;
// ========================================== // 硬件加速 FFT 测试
// 1. 硬件加速 FFT 测试
// ==========================================
printf("\n--- Starting Hardware FFT ---\n"); printf("\n--- Starting Hardware FFT ---\n");
tick_start = get_ns(); tick_start = get_ns();
@@ -122,10 +102,8 @@ int main(int argc, char** argv)
} }
RegWrite(FFT_CSR_REG, FFT_CTRL_START); fft_start();
while ((RegRead(FFT_CSR_REG) & FFT_STAT_DONE) == 0) { fft_wait();
// 等待硬件计算完成
}
tick_end = get_ns(); tick_end = get_ns();

58
sdk/software/examples/fft_dma/main.c
View File

@@ -1,9 +1,13 @@
#include <stdio.h> #include <stdio.h>
#include <string.h> #include <string.h>
#include <stdbool.h>
#include <math.h> #include <math.h>
#include <stdint.h> #include <stdint.h>
#include <common_func.h> #include <common_func.h>
#include <confreg_time.h> #include <confreg_time.h>
#include <dma.h>
#include <fft.h>
// BSP板级支持包所需全局变量 // BSP板级支持包所需全局变量
unsigned long UART_BASE = 0xbf000000; unsigned long UART_BASE = 0xbf000000;
@@ -11,42 +15,8 @@ unsigned long CONFREG_TIMER_BASE = 0xbf20f100;
unsigned long CONFREG_CLOCKS_PER_SEC = 50000000L; unsigned long CONFREG_CLOCKS_PER_SEC = 50000000L;
unsigned long CORE_CLOCKS_PER_SEC = 33000000L; unsigned long CORE_CLOCKS_PER_SEC = 33000000L;
#define FFT_BASE 0xbf400000
#define FFT_IN_RE_BASE (FFT_BASE + 0x1000)
#define FFT_IN_IM_BASE (FFT_BASE + 0x2000)
#define FFT_OUT_RE_BASE (FFT_BASE + 0x3000)
#define FFT_OUT_IM_BASE (FFT_BASE + 0x4000)
#define FFT_CSR_REG (FFT_BASE + 0xF000)
#define FFT_CTRL_START (1 << 4)
#define FFT_STAT_DONE (1 << 1)
#define FFT_STAT_BUSY (1 << 0)
#define FFT_POINT_NUM 1024
#define DMA_BASE 0xbf300000
#define DMA_CTRL (DMA_BASE + 0x0000)
#define DMA_LEN (DMA_BASE + 0x0004)
#define DMA_SRC_ADDR (DMA_BASE + 0x0008)
#define DMA_DST_ADDR (DMA_BASE + 0x000c)
#define DMA_STATUS (DMA_BASE + 0x0010)
const float PI = 3.14159265358979323846; const float PI = 3.14159265358979323846;
// DMA 传输通用封装函数 (阻塞等待模式)
void dma_transfer(uint32_t phys_src, uint32_t phys_dst, uint32_t byte_len) {
RegWrite(DMA_SRC_ADDR, phys_src);
RegWrite(DMA_DST_ADDR, phys_dst);
RegWrite(DMA_LEN, byte_len);
// burst_len = 15(16拍), burst_size = 2(4字节), start = 1
uint32_t ctrl_val = (15 << 6) | (2 << 3) | 0x01;
RegWrite(DMA_CTRL, ctrl_val);
// 轮询等待 DMA 搬运完成
while ((RegRead(DMA_STATUS) & 0x01) == 0) {
// CPU 空转等待
}
}
// 软件FFT实现 (基2 DIT-FFT 算法) // 软件FFT实现 (基2 DIT-FFT 算法)
void sw_fft(float re[], float im[], int n) { void sw_fft(float re[], float im[], int n) {
int i, j, k, l; int i, j, k, l;
@@ -152,18 +122,22 @@ int main(int argc, char** argv)
uint32_t transfer_bytes = FFT_POINT_NUM * 4; // 1024个点 * 4字节 uint32_t transfer_bytes = FFT_POINT_NUM * 4; // 1024个点 * 4字节
// MA 将数据从内存搬运到 FFT 输入外设 // MA 将数据从内存搬运到 FFT 输入外设
dma_transfer(((uint32_t)hw_in_re_arr & 0x1FFFFFFF), phys_fft_in_re, transfer_bytes); dma_start_transfer(0, ((uint32_t)hw_in_re_arr & 0x1FFFFFFF), phys_fft_in_re, transfer_bytes, 100);
dma_transfer(((uint32_t)hw_in_im_arr & 0x1FFFFFFF), phys_fft_in_im, transfer_bytes); dma_start_transfer(1, ((uint32_t)hw_in_im_arr & 0x1FFFFFFF), phys_fft_in_im, transfer_bytes, 200);
dma_wait_polling(0);
dma_wait_polling(1);
// 启动 FFT 并等待计算完成 // 启动 FFT 并等待计算完成
RegWrite(FFT_CSR_REG, FFT_CTRL_START); fft_start();
while ((RegRead(FFT_CSR_REG) & FFT_STAT_DONE) == 0) { fft_wait();
// poll
}
// DMA 将结果从 FFT 输出外设搬回内存 // DMA 将结果从 FFT 输出外设搬回内存
dma_transfer(phys_fft_out_re, ((uint32_t)hw_out_re_arr & 0x1FFFFFFF), transfer_bytes); dma_start_transfer(0, phys_fft_out_re, ((uint32_t)hw_out_re_arr & 0x1FFFFFFF), transfer_bytes, 10);
dma_transfer(phys_fft_out_im, ((uint32_t)hw_out_im_arr & 0x1FFFFFFF), transfer_bytes); dma_start_transfer(1, phys_fft_out_im, ((uint32_t)hw_out_im_arr & 0x1FFFFFFF), transfer_bytes, 20);
dma_wait_polling(0);
dma_wait_polling(1);
tick_end = get_ns(); // 结束计时 tick_end = get_ns(); // 结束计时
hw_time = tick_end - tick_start; hw_time = tick_end - tick_start;