From a88b321433647958d78764ff61ba307856d84a99 Mon Sep 17 00:00:00 2001
From: biondizzle <biondizzle@gmail.com>
Date: Sat, 30 May 2026 17:00:51 +0000
Subject: [PATCH] P6: Fix host-side BF16 conversion in test

---
 tests/unit/test_p6_tma_store.cu | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)

diff --git a/tests/unit/test_p6_tma_store.cu b/tests/unit/test_p6_tma_store.cu
index bce8a4cf..20768ea3 100644
--- a/tests/unit/test_p6_tma_store.cu
+++ b/tests/unit/test_p6_tma_store.cu
@@ -53,10 +53,12 @@ int main() {
     auto init_bf16 = [](bf16_t* d, int n) {
         float* h = new float[n];
         for (int i = 0; i < n; i++) h[i] = (float)rand() / RAND_MAX - 0.5f;
+        // Use host-side BF16 conversion
         for (int i = 0; i < n; i++) {
-            unsigned short us;
-            asm("cvt.rn.bf16.f32 %0, %1;" : "=h"(us) : "f"(h[i]));
-            d[i] = us;
+            uint32_t u;
+            memcpy(&u, &h[i], 4);
+            u = u >> 16;  // truncate FP32 to BF16 (rough but sufficient for test)
+            d[i] = (bf16_t)(u & 0xFFFF);
         }
         delete[] h;
     };