tensor: Change B in-memory layout to column-major

2024-08-12 15:20:55 -07:00
parent 07dd9e35a0
commit 95e3e96c6c
6 changed files with 42 additions and 24 deletions
--- a/tests/regression/sgemm_tcore/kernel.cpp
+++ b/tests/regression/sgemm_tcore/kernel.cpp
@@ -572,7 +572,7 @@ void kernel_body(int task_id, kernel_arg_t *__UNIFORM__ arg) {
  const uint32_t problem_size = (dim_m * dim_n) / (ELEM_PER_THREAD);
  const uint32_t num_threadblocks = problem_size / threads_per_threadblock;

-  using float_type = float;
+  using float_type = float16_t;

  // "static" shared memory allocation.  This would determine threadblock
  // occupancy of a single cluster