[Mlir-commits] [mlir] [mlir][Vector] Generate poison vectors in vector.shape_cast lowering (PR #125613)

[Jakub Kuderski]

================
@@ -83,17 +83,20 @@ func.func @vaddi_reduction(%arg0 : vector<8xi32>, %arg1 : vector<8xi32>) -> (i32
 // CHECK-LABEL: @transpose
 // CHECK-SAME: (%[[ARG0:.+]]: vector<3xi32>, %[[ARG1:.+]]: vector<3xi32>)
 func.func @transpose(%arg0 : vector<2x3xi32>) -> (vector<3x2xi32>) {
-  // CHECK: %[[CST:.*]] = arith.constant dense<0> : vector<2xi32>
+  // CHECK: %[[CST:.*]] = ub.poison : vector<1x2xi32>
----------------
kuhar wrote:

Ah, I must have forgotten about it 🙂 

>From the commit description:
> SPIR-V doesn't have poison, but poison can be converted to undef.

This is not correct in all contests and generally not safe. IMO we should convert to something like zero constants. cc: @Hardcode84

We can enable this conversion and revisit the exact conversion strategy separately.

https://github.com/llvm/llvm-project/pull/125613