Merge branch 'kernels' of https://github.com/hansungk/vortex-private into kernels

tests/regression/sgemm_gemmini_dma/.gitignore

@@ -1,5 +1,5 @@
 *.bin
 *.dump
 *.elf
-sgemm_wg
+sgemm_gemmini_dma
 .depend

tests/regression/sgemm_gemmini_dma/common.h

@@ -3,7 +3,7 @@
 
 #include <cstdint>
 
-#define KERNEL_ARG_DEV_MEM_ADDR 0x7fff0000
+#define KERNEL_ARG_DEV_MEM_ADDR 0x9fff0000
 #define DEV_SMEM_START_ADDR 0xff000000
 
 typedef struct {

tests/regression/sgemm_gemmini_dma/kernel.cpp

@@ -51,7 +51,7 @@ inline void threadblock_barrier(unsigned int barrier_id, unsigned int count) {
 void thread_block_matmul_gemmini(kernel_arg_t *__UNIFORM__ arg,
                                  const uint32_t threadblock_id,
                                  const uint32_t tid_in_threadblock) {
-  __asm__("matmul_start:");
+  asm volatile ("matmul_start_%=:" :: );
   const float * const A = (const float * const) arg->addr_a;
   const float * const B = (const float * const) arg->addr_b;
   float * const C = (float * const) arg->addr_c;
@@ -178,4 +178,4 @@ int main() {
   vx_spawn_tasks_contiguous(grid_size, (vx_spawn_tasks_cb)kernel_body, arg);
 #endif
   return 0;
-}
+}

tests/regression/sgemm_gemmini_dma/main.cpp

@@ -193,8 +193,8 @@ int main(int argc, char *argv[]) {
   {
     std::cout << "upload kernel argument" << std::endl;
     auto buf_ptr = staging_buf.data();
-    kernel_arg.addr_a = (uint64_t) 0x20000;
-    kernel_arg.addr_b = (uint64_t) 0x28000;
+    kernel_arg.addr_a = (uint64_t) 0xa0000000ULL;
+    kernel_arg.addr_b = (uint64_t) 0xa1000000ULL;
     kernel_arg.addr_c = (uint64_t) 0xc0000000ULL;
     memcpy(buf_ptr, &kernel_arg, sizeof(kernel_arg_t));