[GitHub] [tvm] ANSHUMAN87 commented on a change in pull request #7148: [Frontend][Tensorflow] Sparse_Dense Op CSR scheduling issue resolved for Cuda & X86

[GitBox <gi...@apache.org>]

ANSHUMAN87 commented on a change in pull request #7148:
URL: https://github.com/apache/tvm/pull/7148#discussion_r547985023



##########
File path: python/tvm/topi/x86/sparse.py
##########
@@ -28,15 +28,17 @@ def schedule_sparse_dense(outs):
 
     def _callback(op):
         simd_width = get_fp32_len()
-        if op.tag == "sparse_dense_csrmm" and op != outs[0].op:
-            (_, v_i) = s[op].op.axis
-            s[op].vectorize(v_i)
-            (y_o, y_i) = s[outs[0].op].split(s[outs[0].op].op.axis[1], 2 * simd_width)
-            s[op].compute_at(s[outs[0]], y_o)
-            s[outs[0].op].vectorize(y_i)
-        if op.tag == "sparse_dense_bsrmm":
+        if op.tag == "sparse_dense_csrmm_v2" or op.tag == "sparse_dense_csrmm_v1":
+            (y_o, y_i) = s[op].split(s[op].op.axis[1], 2)
+            fused = s[op].fuse(s[op].op.axis[0], y_o)
+            s[op].parallel(fused)
+            s[op].vectorize(y_i)
+        elif op.tag == "sparse_dense_bsrmm_v2" or op.tag == "sparse_dense_bsrmm_v1":

Review comment:
       Basically the difference is whether lhs is sparse or rhs is sparse.
   May be we change name like "sparse_dense_bsrmm_v2" --> "sparse_dense_bsrmm_sparse_rhs"  ?




----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

For queries about this service, please contact Infrastructure at:
users@infra.apache.org