================
@@ -55,6 +60,156 @@ namespace mlir {
namespace {
+//===----------------------------------------------------------------------===//
+// x86_64 System V classifier bridge (scalar types)
+//
+// Maps scalar CIR types to llvm::abi::Type, runs the LLVM ABI Lowering
+// Library's SysV x86_64 classifier, and converts the result back into the
+// dialect-agnostic mlir::abi::FunctionClassification that CIRABIRewriteContext
+// consumes. Only integer / pointer / bool / f32 / f64 signatures are handled;
+// aggregates and other leaf types are reported NYI by classifyX86_64 so an
+// unsupported signature fails the pass instead of being misclassified.
+//===----------------------------------------------------------------------===//
+
+/// llvm::Align requires a power of two; DataLayout can report non-power-of-two
+/// alignments for unusual types.
----------------
adams381 wrote:
E.g. `i24`. `IntType::getABIAlignment` returns bare `width / 8` for
non-`_BitInt` widths, so alignment is 3, not a power of two.
Probably, yes. But that changes alignment for every caller, not just this
classifier, so I'd want to look at that separately rather than fold it
into this PR.
https://github.com/llvm/llvm-project/pull/209636
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
