matmul_nbits: Use GPU_WARP_SIZE_HOST for host side code

For ROCm device, the host side code needs to call GPU_WARP_SIZE_HOST to
query warpsize of the underlying GPU device.

Fixes MatMulNBits tests on Navi.

Signed-off-by: Jagadish Krishnamoorthy <[email protected]>

onnxruntime/contrib_ops/cuda/quantization/matmul_nbits.cu

@@ -288,6 +288,7 @@ bool TryMatMul4Bits(
   if (n % kColsPerThreadBlock != 0 || k % 8 != 0 || m > 1) {
     return false;
   }
+  const int kWarpSize = GPU_WARP_SIZE_HOST;
   dim3 blocks((n + kColsPerThreadBlock - 1) / kColsPerThreadBlock, m);
   dim3 threads(kWarpSize, kColsPerThreadBlock);
   int blocks_per_K = (k + block_size - 1) / block_size;