use BNB_WARP_SIZE instead of warpSize in ops.hip

Author: electron271

csrc/ops.hip

@@ -5,6 +5,7 @@
 // This source code is licensed under the MIT license found in the
 // LICENSE file in the root directory of this source tree.
 
+#include <common_hip.cuh>
 #include <ops_hip.cuh>
 #include <kernels_hip.cuh>
 #include <hipcub/hipcub.hpp>
@@ -57,7 +58,7 @@ template <typename T, int STOCHASTIC, int DATA_TYPE> void quantizeBlockwise(floa
    hipLaunchKernelGGL(( kQuantizeBlockwise<T, 256, 2, 0, DATA_TYPE>), dim3(num_blocks), dim3(128), 0, 0, code, A, absmax, out, rand, rand_offset, n);
   else if(blocksize == 128)
    hipLaunchKernelGGL(( kQuantizeBlockwise<T, 128, 2, 0, DATA_TYPE>), dim3(num_blocks), dim3(64), 0, 0, code, A, absmax, out, rand, rand_offset, n);
-  else if(blocksize == 64 && warpSize == 32)
+  else if(blocksize == 64 && BNB_WARP_SIZE == 32)
    hipLaunchKernelGGL(( kQuantizeBlockwise<T, 64, 2, 0, DATA_TYPE>), dim3(num_blocks), dim3(32), 0, 0, code, A, absmax, out, rand, rand_offset, n);