[Mlir-commits] [mlir] [mlir][ABI] Fix scalable vector mapping in ABITypeMapper (PR #206617)

[Andrzej Warzyński]

================
@@ -170,4 +170,34 @@ TEST_F(ABITypeMapperTest, MapUnsignedI32) {
   EXPECT_FALSE(intTy->isSigned());
 }
 
+TEST_F(ABITypeMapperTest, MapScalableVectorOf4xF32) {
+  DataLayout dl(module);
+  ABITypeMapper mapper(dl);
+
+  auto f32 = Float32Type::get(&ctx);
+  auto vec = VectorType::get({4}, f32, /* scalableDims=*/true);
+  const llvm::abi::Type *result = mapper.map(vec);
+
+  ASSERT_NE(result, nullptr);
+  EXPECT_TRUE(result->isVector());
+
+  auto *vecTy = llvm::cast<llvm::abi::VectorType>(result);
+  EXPECT_TRUE(vecTy->getNumElements().isScalable());
+  EXPECT_EQ(vecTy->getNumElements().getKnownMinValue(), 4u);
+  EXPECT_TRUE(vecTy->getElementType()->isFloat());
+}
+
+TEST_F(ABITypeMapperTest, MapMultiScalableVectorReturnsNull) {
+  DataLayout dl(module);
+  ABITypeMapper mapper(dl);
+
+  auto f32 = Float32Type::get(&ctx);
+  auto vec = VectorType::get({2, 4}, f32, /*scalableDims=*/{true, true});
+  const llvm::abi::Type *result = mapper.map(vec);
+
+  // (vscale*2)*(vscale*4) = vscale^2*8,
----------------
banach-space wrote:

In general, `vscale^2` doesn't really make sense and I would refrain from using it at all. Instead, I would write something like this:
> LLVM supports at most 1 scalable dimension, hence this case should return nullptr.

Or similar. Feel free to use.

https://github.com/llvm/llvm-project/pull/206617