wu-arch/kernels
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Repeat vx_wmma issue & hardcode dst address

Browse Source
This commit is contained in:
Hansung Kim
2024-05-08 13:22:26 -07:00
parent 7775830814
commit 5821bfd10d
2 changed files with 8 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
tests/kernel/reductions/main.cpp
View File

@@ -138,6 +138,7 @@ void test_maxu_reduce() {
y = reduced; y = reduced;
} }
// assumes NUM_THREADS == 4
unsigned bit_vectors[4] = {0b11010110000111001100010100100110, 0b10010100011010001010000000001110, 0b10001001010111110001110000000010, 0b00010011010100101101110111001111}; unsigned bit_vectors[4] = {0b11010110000111001100010100100110, 0b10010100011010001010000000001110, 0b10001001010111110001110000000010, 0b00010011010100101101110111001111};
void test_and_reduce() { void test_and_reduce() {

6
tests/kernel/tensor/main.cpp
View File

@@ -65,6 +65,7 @@ float results[32*8];
void store_wmma_result() { void store_wmma_result() {
int tid = vx_thread_id(); int tid = vx_thread_id();
float *results = reinterpret_cast<float *>(0xc0000000UL);
asm volatile ("fsw f16, %0" :: "m"(results[tid*8+0])); asm volatile ("fsw f16, %0" :: "m"(results[tid*8+0]));
asm volatile ("fsw f17, %0" :: "m"(results[tid*8+1])); asm volatile ("fsw f17, %0" :: "m"(results[tid*8+1]));
asm volatile ("fsw f18, %0" :: "m"(results[tid*8+2])); asm volatile ("fsw f18, %0" :: "m"(results[tid*8+2]));
@@ -87,7 +88,10 @@ int main()
{ {
vx_tmc(-1); vx_tmc(-1);
vx_wmma_load(); vx_wmma_load();
vx_wmma(); #pragma GCC unroll 100
for (int i = 0; i < 100; i++) {
vx_wmma();
}
store_wmma_result(); store_wmma_result();
vx_tmc(1); vx_tmc(1);
// print_wmma_result(); // print_wmma_result();