preparing the field for tensor cores transposes

2019-07-17 13:20:33 -07:00
parent d2e116d057
commit bfa39b8992
9 changed files with 67 additions and 29 deletions
--- a/examples/python/tensorflow/dot.cpp
+++ b/examples/python/tensorflow/dot.cpp
@@ -49,7 +49,7 @@ class DotOp : public OpKernel {
    triton::driver::cu_buffer db(ctx, (CUdeviceptr)b.flat<Eigen::half>().data(), false);
    triton::driver::cu_buffer dc(ctx, (CUdeviceptr)c->flat<float>().data(), false);
    // template
-    triton::dnn::gemm dot(M, N, K, false, true, "fp16", "fp16", 4, 4);
+    triton::dnn::gemm dot(M, N, K, false, false, "fp16", "fp16", 4, 4);
    dot.enqueue(stream, {&da, &db, &dc});
  }