[Mlir-commits] [mlir] [mlir][sparse] update block24 example (PR #70145)
llvmlistbot at llvm.org
llvmlistbot at llvm.org
Tue Oct 24 16:36:08 PDT 2023
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-mlir
Author: Aart Bik (aartbik)
<details>
<summary>Changes</summary>
Removes TODO, shows how to define 8-bit crd (lacking 2-bit for now)
---
Full diff: https://github.com/llvm/llvm-project/pull/70145.diff
1 Files Affected:
- (modified) mlir/test/Dialect/SparseTensor/roundtrip_encoding.mlir (+4-16)
``````````diff
diff --git a/mlir/test/Dialect/SparseTensor/roundtrip_encoding.mlir b/mlir/test/Dialect/SparseTensor/roundtrip_encoding.mlir
index ea8217ab6e3f233..db844312d02b999 100644
--- a/mlir/test/Dialect/SparseTensor/roundtrip_encoding.mlir
+++ b/mlir/test/Dialect/SparseTensor/roundtrip_encoding.mlir
@@ -130,19 +130,6 @@ func.func private @sparse_slice(tensor<?x?xf64, #CSR_SLICE>)
// -----
-// TODO: It is probably better to use [dense, dense, 2:4] (see NV_24 defined using new syntax
-// below) to encode a 2D matrix, but it would require dim2lvl mapping which is not ready yet.
-// So we take the simple path for now.
-#NV_24= #sparse_tensor.encoding<{
- map = (d0, d1) -> (d0 : dense, d1 : block2_4)
-}>
-
-// CHECK-LABEL: func private @sparse_2_out_of_4(
-// CHECK-SAME: tensor<?x?xf64, #sparse_tensor.encoding<{ map = (d0, d1) -> (d0 : dense, d1 : block2_4) }>>
-func.func private @sparse_2_out_of_4(tensor<?x?xf64, #NV_24>)
-
-// -----
-
#BSR = #sparse_tensor.encoding<{
map = ( i, j ) ->
( i floordiv 2 : dense,
@@ -204,11 +191,12 @@ func.func private @BSR_explicit(%arg0: tensor<?x?xf64, #BSR_explicit>) {
( i : dense,
j floordiv 4 : dense,
j mod 4 : block2_4
- )
+ ),
+ crdWidth = 8 // we would even like just 2-bits
}>
// CHECK-LABEL: func private @NV_24(
-// CHECK-SAME: tensor<?x?xf64, #sparse_tensor.encoding<{ map = (d0, d1) -> (d0 : dense, d1 floordiv 4 : dense, d1 mod 4 : block2_4) }>>
+// CHECK-SAME: tensor<?x?xf64, #sparse_tensor.encoding<{ map = (d0, d1) -> (d0 : dense, d1 floordiv 4 : dense, d1 mod 4 : block2_4), crdWidth = 8 }>>
func.func private @NV_24(%arg0: tensor<?x?xf64, #NV_24>) {
return
}
@@ -245,4 +233,4 @@ func.func private @NV_24(%arg0: tensor<?x?x?xf64, #NV_24>) {
// CHECK-SAME: tensor<?x?x?xf64, #sparse_tensor.encoding<{ map = (d0, d1, d2) -> (d0 : dense, d2 floordiv 4 : dense, d1 : dense, d2 mod 4 : block2_4) }>>
func.func private @NV_24(%arg0: tensor<?x?x?xf64, #NV_24>) {
return
-}
\ No newline at end of file
+}
``````````
</details>
https://github.com/llvm/llvm-project/pull/70145
More information about the Mlir-commits
mailing list