54 lines
1.8 KiB
C++
54 lines
1.8 KiB
C++
#include <stdint.h>
|
|
#include <vx_intrinsics.h>
|
|
#include <vx_spawn.h>
|
|
#include "common.h"
|
|
|
|
void kernel_body(int task_id, kernel_arg_t* __UNIFORM__ arg) {
|
|
const float *global_a = (const float *)arg->addr_a;
|
|
const float *global_b = (const float *)arg->addr_b;
|
|
float *global_c = (float *)arg->addr_c;
|
|
|
|
// assumes NT == NW == matrix_dim
|
|
const uint32_t dim_m = arg->dim_m;
|
|
const uint32_t dim_n = arg->dim_n;
|
|
const uint32_t dim_k = arg->dim_k;
|
|
const uint32_t block_dim = vx_num_warps();
|
|
const uint32_t local_row = vx_warp_id();
|
|
const uint32_t local_col = vx_thread_id();
|
|
|
|
// each thread generates one output element
|
|
float reg_c = 0.0f;
|
|
|
|
for (uint32_t k = 0; k < dim_k; k += block_dim) {
|
|
float *local_a = (float *)DEV_SMEM_START_ADDR;
|
|
float *local_b = (float *)DEV_SMEM_START_ADDR + (block_dim * block_dim);
|
|
|
|
// FIXME: assumes local block size is square shape
|
|
// TODO: "local_row" should be global_row
|
|
uint32_t offset_global_a = dim_k * local_row + (k + local_col);
|
|
uint32_t offset_global_b = dim_n * (local_row + k) + local_col;
|
|
local_a[block_dim * local_row + local_col] = global_a[offset_global_a];
|
|
local_b[block_dim * local_row + local_col] = global_b[offset_global_b];
|
|
|
|
vx_barrier(0, vx_num_warps());
|
|
vx_fence();
|
|
|
|
for (uint32_t local_k = 0; local_k < block_dim; local_k++) {
|
|
reg_c += local_a[block_dim * local_row + local_k] *
|
|
local_b[block_dim * local_k + local_col];
|
|
}
|
|
|
|
vx_barrier(0, vx_num_warps());
|
|
vx_fence();
|
|
}
|
|
|
|
global_c[dim_n * local_row + local_col] = reg_c;
|
|
}
|
|
|
|
int main() {
|
|
kernel_arg_t *arg = (kernel_arg_t *)KERNEL_ARG_DEV_MEM_ADDR;
|
|
int threads_per_core = vx_num_warps() * vx_num_threads();
|
|
vx_spawn_tasks(threads_per_core, (vx_spawn_tasks_cb)kernel_body, arg);
|
|
return 0;
|
|
}
|