[PyTorch] `te.Linear` FP8 DGRAD+RS output bugfix (#1412)

* corrected RS overlap BF16 output clashing with Float8Tensor constructor
Signed-off-by: Alp Dener <adener@nvidia.com>

* fixed empty dgrad buffer dtype at initialization
Signed-off-by: Alp Dener <adener@nvidia.com>

* [pre-commit.ci] auto fixes from pre-commit.com hooks

for more information, see https://pre-commit.ci



---------
Signed-off-by: Alp Dener <adener@nvidia.com>
Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>

transformer_engine/pytorch/module/linear.py

@@ -506,13 +506,6 @@ class _Linear(torch.autograd.Function):
                     ub_obj_wgrad = get_ub(ctx.ub_name + "_wgrad")
                     dgrad = ub_obj_wgrad.get_ubuf_output(1)
 
-            if dgrad is None:
-                if ctx.parallel_mode == "column" and ctx.sequence_parallel:
-                    dgrad_shape[0] = dgrad_shape[0] * tp_world_size
-                dgrad = torch.empty(
-                    dgrad_shape, dtype=ctx.activation_dtype, device=grad_output.device
-                )
-
             (
                 grad_output,
                 grad_output_c,
@@ -550,6 +543,7 @@ class _Linear(torch.autograd.Function):
                 fp8_dtype_forward = get_fp8_te_dtype(ctx.fp8_meta["recipe"], fprop_tensor=True)
                 fp8_dtype_backward = get_fp8_te_dtype(ctx.fp8_meta["recipe"], fprop_tensor=False)
 
+            output_dtype = ctx.activation_dtype
             if ctx.requires_dgrad:
                 if ctx.fp8:
                     if ctx.is_input_fp8 or (
@@ -570,6 +564,14 @@ class _Linear(torch.autograd.Function):
                             None,
                             ctx.activation_dtype,
                         )
+
+                if dgrad is None:
+                    if ctx.parallel_mode == "column" and ctx.sequence_parallel:
+                        dgrad_shape[0] = dgrad_shape[0] * tp_world_size
+                    dgrad = torch.empty(dgrad_shape, dtype=output_dtype, device=grad_output.device)
+
+            if ctx.requires_dgrad:
+                if ctx.fp8:
                     _ = fp8_gemm(
                         weight_fp8.transpose_2d(),
                         weight_fp8._scale_inv,
@@ -593,8 +595,7 @@ class _Linear(torch.autograd.Function):
 
                     if ctx.ub_overlap_rs_dgrad:
                         dgrad = rs_out
-
-                    if output_dtype == torch.uint8:
+                    elif output_dtype == torch.uint8:
                         dgrad = Float8Tensor(
                             data=dgrad,
                             fp8_meta=ctx.fp8_meta,