[Mlir-commits] [mlir] [mlir][tosa] Add MXFP support for CONV3D (PR #230148)

[Luke Hutton]

================
@@ -2240,3 +2240,23 @@ func.func @test_cast_fp6e3m2_block_scaled(%arg0: tensor<4x32xf6E3M2FN>) -> tenso
   %0 = tosa.cast %arg0 : (tensor<4x32xf6E3M2FN>) -> tensor<4x32x!tosa.block_scaled<BLOCK_SHAPE_32:f8E8M0FNU:!tosa.mxint8>>
   return %0 : tensor<4x32x!tosa.block_scaled<BLOCK_SHAPE_32:f8E8M0FNU:!tosa.mxint8>>
 }
+
+// -----
+
+func.func @test_conv3d_mxfp_invalid_accumulator(%input: tensor<1x4x4x4x32x!tosa.block_scaled<BLOCK_SHAPE_32:f8E8M0FNU:f4E2M1FN>>, %weight: tensor<8x1x1x1x32x!tosa.block_scaled<BLOCK_SHAPE_32:f8E8M0FNU:f4E2M1FN>>, %bias: tensor<8xf32>) -> tensor<1x4x4x4x8xf32> {
+  %input_zp = "tosa.const"() <{values = dense<0.0> : tensor<1xf32>}> : () -> tensor<1xf32>
----------------
lhutton1 wrote:

nit: it's possible to use the non-generic version of const now 

https://github.com/llvm/llvm-project/pull/230148