diff --git a/tests/unit/test_pv_ss_128.cu b/tests/unit/test_pv_ss_128.cu index 8689b7da..3d0c9d16 100644 --- a/tests/unit/test_pv_ss_128.cu +++ b/tests/unit/test_pv_ss_128.cu @@ -72,7 +72,7 @@ test_pv_ss_128() { uint64_t dv = make_umma_desc_kmajor_none(__cvta_generic_to_shared(sV), 16); uint32_t idesc = make_idesc(BLOCK_MN, HD); - for (int kt = 1; kt < 2; kt++) { + for (int kt = 0; kt < 2; kt++) { bf16_t* sp = sP + kt * 2048; uint64_t dp = make_umma_desc_kmajor_none(__cvta_generic_to_shared(sp), BLOCK_MN); if (tid == 0) umma_ss_f16(tb, dp, dv, idesc, true); // ALL accumulate