fallensigh/ciciec2026
Archived
0
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Avoid DSP in matmul master path

Browse Source
This commit is contained in:
FallenSigh
2026-07-01 23:30:11 +08:00
parent fd03aaaaa1
commit baf3f6f696
3 changed files with 76 additions and 287 deletions
Download Patch File Download Diff File Expand all files Collapse all files

31
sdk/software/examples/asm/user-sample.c
View File

@@ -5,7 +5,6 @@
*/
#include <dma.h>
#include <confreg_time.h>
/* ─── UART 16550 direct access (bypasses slow printf/picolibc) ──────── */
#define UART_THR (*(volatile unsigned char *)0xbf000000)
@@ -29,24 +28,6 @@ static void uart_puthex(unsigned int val) {
}
}
static void uart_putdec(unsigned int val) {
char buf[10];
int i = 0;
if (val == 0) {
uart_tx('0');
return;
}
while (val != 0) {
buf[i++] = (char)('0' + (val % 10));
val /= 10;
}
while (i > 0) {
uart_tx((unsigned char)buf[--i]);
}
}
/* ─── Configurable group count (default 5000) ──────────────────────── */
#ifndef MATMUL_GROUP_NUM
#define MATMUL_GROUP_NUM 5000
@@ -54,9 +35,6 @@ static void uart_putdec(unsigned int val) {
/* ─── BSP globals ─────────────────────────────────────────────────── */
unsigned long UART_BASE = 0xbf000000;
unsigned long CONFREG_TIMER_BASE = 0xbf20f100;
unsigned long CONFREG_CLOCKS_PER_SEC = 50000000L;
unsigned long CORE_CLOCKS_PER_SEC = 33000000L;
/* ─── Memory layout ────────────────────────────────────────────────── */
#define EXTRAM_BASE_PHYS 0x1c400000
@@ -107,11 +85,9 @@ int main(void)
volatile unsigned int *mat_crc = (volatile unsigned int *)MATMUL_CRC;
unsigned int crc;
unsigned int cycle_start, cycle_end, cycles, time_us;
/* ── Start marker (must be first output) ──────────────────── */
uart_puts("MATMUL_START\r\n");
cycle_start = get_clock_count();
*(volatile unsigned int *)MATMUL_CRC = 0;
*mat_status = STATUS_DMA_DONE | STATUS_DONE | STATUS_ERROR;
@@ -123,19 +99,12 @@ int main(void)
*mat_ctrl = CTRL_DMA_START;
while (!(*mat_status & STATUS_DMA_DONE)) {}
*mat_status = STATUS_DMA_DONE;
cycle_end = get_clock_count();
cycles = cycle_end - cycle_start;
time_us = cycles / (CORE_CLOCKS_PER_SEC / 1000000UL);
/* ── Read hardware CRC32 ──────────────────────────────────── */
crc = *mat_crc;
uart_puts("MATMUL_CRC32=");
uart_puthex(crc);
uart_puts("\r\nMATMUL_CYCLES=");
uart_putdec(cycles);
uart_puts("\r\nMATMUL_TIME_US=");
uart_putdec(time_us);
uart_puts("\r\nMATMUL_DONE\r\n");
return 0;