[Mlir-commits] [mlir] [mlir][spirv] Add 8-bit float type emulation (PR #148811)

[Jakub Kuderski]

================
@@ -39,6 +39,10 @@ struct SPIRVConversionOptions {
   /// The number of bits to store a boolean value.
   unsigned boolNumBits{8};
 
+  /// Whether to emulate unsupported floats with integer types of same bit
+  /// width.
+  bool emulateUnsupportedFloatTypes{true};
----------------
kuhar wrote:

I'm concerned about cases when you ended up with unsupported types by accident -- this often comes up with all the variants of fp8 and smaller fp types. This is much easier to diagnose when you can see the original type. IMO, dialect conversion should error out by default, unless someone opts into these types being handled in some other way.

https://github.com/llvm/llvm-project/pull/148811