[CIR] Add support for complex cast operations #758
Conversation
| return builder | ||
| .create<mlir::cir::TernaryOp>( | ||
| op.getLoc(), srcRealToBool, | ||
| [&](mlir::OpBuilder &, mlir::Location) { | ||
| builder.createYield(op.getLoc(), | ||
| builder.getTrue(op.getLoc()).getResult()); | ||
| }, | ||
| [&](mlir::OpBuilder &, mlir::Location) { | ||
| auto inner = | ||
| builder | ||
| .create<mlir::cir::TernaryOp>( | ||
| op.getLoc(), srcImagToBool, | ||
| [&](mlir::OpBuilder &, mlir::Location) { | ||
| builder.createYield( | ||
| op.getLoc(), | ||
| builder.getTrue(op.getLoc()).getResult()); | ||
| }, | ||
| [&](mlir::OpBuilder &, mlir::Location) { | ||
| builder.createYield( | ||
| op.getLoc(), | ||
| builder.getFalse(op.getLoc()).getResult()); | ||
| }) | ||
| .getResult(); | ||
| builder.createYield(op.getLoc(), inner); | ||
| }) | ||
| .getResult(); |
There was a problem hiding this comment.
This is not the first time I write code like these to implement a simple boolean binary op like a && b or a || b. I think a dedicated operation for binary logical operations are really necessary and they allow generation of much better LLVM IR, although these operations won't be generated from CIRGen directly.
There was a problem hiding this comment.
I have gone through the same feeling, we need a ternary that's basically a select operation. If you wanna add that it would be a great addition. cir.ternary_select or something, then a fold operation can even transform a cir.ternary into cir.ternary_select to handle all really simple cases like this one (some CIRGen will use generic code and might emit this pattern regardless of cases we actually know we want a select-like op).
There was a problem hiding this comment.
I have gone through the same feeling, we need a ternary that's basically a select operation.
Do we additionally need a boolean binary op, something like cir.binop logical_and ? This is preferable to me since it lowers to different code with select operation.
There was a problem hiding this comment.
I'm not sure how usable this would be, given that usually those have short-circuiting semantics. We have folders and other transformations passes that can help out with this, have you explored any of those solutions already?
There was a problem hiding this comment.
Let's track this in #785 .
|
Rebased onto the latest |
This patch adds support for complex cast operations. It adds the following new cast kind variants to the `cir.cast` operation: - `float_to_complex`, - `int_to_complex`, - `float_complex_to_real`, - `int_complex_to_real`, - `float_complex_to_bool`, - `int_complex_to_bool`, - `float_complex`, - `float_complex_to_int_complex`, - `int_complex`, and - `int_complex_to_float_complex`. CIRGen and LLVM IR support for these new cast variants are also included.
|
Rebased onto the latest |
This PR adds support for complex cast operations. It adds the following new cast kind variants to the `cir.cast` operation: - `float_to_complex`, - `int_to_complex`, - `float_complex_to_real`, - `int_complex_to_real`, - `float_complex_to_bool`, - `int_complex_to_bool`, - `float_complex`, - `float_complex_to_int_complex`, - `int_complex`, and - `int_complex_to_float_complex`. CIRGen and LLVM IR support for these new cast variants are also included.
This PR adds support for complex cast operations. It adds the following new cast kind variants to the `cir.cast` operation: - `float_to_complex`, - `int_to_complex`, - `float_complex_to_real`, - `int_complex_to_real`, - `float_complex_to_bool`, - `int_complex_to_bool`, - `float_complex`, - `float_complex_to_int_complex`, - `int_complex`, and - `int_complex_to_float_complex`. CIRGen and LLVM IR support for these new cast variants are also included.
This PR adds support for complex cast operations. It adds the following new cast kind variants to the `cir.cast` operation: - `float_to_complex`, - `int_to_complex`, - `float_complex_to_real`, - `int_complex_to_real`, - `float_complex_to_bool`, - `int_complex_to_bool`, - `float_complex`, - `float_complex_to_int_complex`, - `int_complex`, and - `int_complex_to_float_complex`. CIRGen and LLVM IR support for these new cast variants are also included.
This PR adds support for complex cast operations. It adds the following new cast kind variants to the `cir.cast` operation: - `float_to_complex`, - `int_to_complex`, - `float_complex_to_real`, - `int_complex_to_real`, - `float_complex_to_bool`, - `int_complex_to_bool`, - `float_complex`, - `float_complex_to_int_complex`, - `int_complex`, and - `int_complex_to_float_complex`. CIRGen and LLVM IR support for these new cast variants are also included.
This PR adds support for complex cast operations. It adds the following new cast kind variants to the `cir.cast` operation: - `float_to_complex`, - `int_to_complex`, - `float_complex_to_real`, - `int_complex_to_real`, - `float_complex_to_bool`, - `int_complex_to_bool`, - `float_complex`, - `float_complex_to_int_complex`, - `int_complex`, and - `int_complex_to_float_complex`. CIRGen and LLVM IR support for these new cast variants are also included.
This PR adds support for complex cast operations. It adds the following new cast kind variants to the
cir.castoperation:float_to_complex,int_to_complex,float_complex_to_real,int_complex_to_real,float_complex_to_bool,int_complex_to_bool,float_complex,float_complex_to_int_complex,int_complex, andint_complex_to_float_complex.CIRGen and LLVM IR support for these new cast variants are also included.