[Mlir-commits] [mlir] [mlir][math] Propagate scalability in polynomial approximation (PR #84949)

[Cullen Rhodes]

================
@@ -151,6 +165,17 @@ func.func @exp_vector(%arg0: vector<8xf32>) -> vector<8xf32> {
   return %0 : vector<8xf32>
 }
 
+// CHECK-LABEL:   func @exp_scalable_vector
+// CHECK-NOT:  math.exp
+// CHECK-NOT:  vector<8xf32>
+// CHECK:      vector<[8]xf32>
+// CHECK-NOT:  vector<8xf32>
+// CHECK-NOT:  math.exp
----------------
c-rhodes wrote:

How about checking there are as many scalable vectors as scalars from the `@exp_scalar` test?
```suggestion
// CHECK-NOT: math.exp
// CHECK-COUNT-42: vector<[8]xf32>
// CHECK-NOT: math.exp
```

Seems a bit more useful than what it's currently doing (we know there will be a scalable vector because of the return). I understand this is based on the above test where the CHECK lines are close to useless.

https://github.com/llvm/llvm-project/pull/84949