Change Details

Although using `__builtin_shufflevector` and the `shufflevector` instruction works fine, they are not opaque to the optimizer. As a result, DAGCombine can potentially reduce the number of shuffles and change the shuffle masks. This is unexpected behavior for users of the WebAssembly SIMD intrinsics, since it is impossible to tell if combining shuffles will be a performance win without breaking the WebAssembly abstraction and reasoning about the underlying platforms. This CL solves the problem by adding a new shuffle intrinsic that is opaque to the optimizers.

Although using `__builtin_shufflevector` and the `shufflevector` instruction works fine, they are not opaque to the optimizer. As a result, DAGCombine can potentially reduce the number of shuffles and change the shuffle masks. This is unexpected behavior for users of the WebAssembly SIMD intrinsics who have crafted their shuffles to optimize the code generated by engines. This patch solves the problem by adding a new shuffle intrinsic that is opaque to the optimizers in line with the decision of the WebAssembly SIMD contributors at https://github.com/WebAssembly/simd/issues/196#issuecomment-622494748. In the future we may implement custom DAG combines to properly optimize shuffles and replace this solution.

Although using `__builtin_shufflevector` and the `shufflevector` instruction works fine, they are not opaque to the optimizer. As a result, DAGCombine can potentially reduce the number of shuffles and change the shuffle masks. This is unexpected behavior for users of the WebAssembly SIMD intrinsics, since it is impossible to tell if combining who have crafted their shuffles to shuffles will be a performance win without breakingoptimize the code generated by engines. This patch solves the WebAssemblyproblem abstraction and reasoning about the underlying platforms. This CL solvesby adding a new shuffle intrinsic that is opaque to the optimizers in the problem by adding a new shuffle intrinsic that is opaque to theline with the decision of the WebAssembly SIMD contributors at https://github.com/WebAssembly/simd/issues/196#issuecomment-622494748. In the future we may implement custom DAG combines to properly optimize shuffles and replace this solution. optimizers.