fix: add cuInit(0) for CUDA driver API

This commit is contained in:
2026-05-29 04:43:24 +00:00
parent 6cc2f61431
commit 0c9245b4d2

View File

@@ -10,6 +10,9 @@
typedef unsigned short bf16_t;
int main() {
// Initialize CUDA driver API — REQUIRED before any cuTensorMap calls
cuInit(0);
printf("=== TMA Descriptor Creation Test ===\n");
bf16_t* d_data;