[Mlir-commits] [mlir] 8c7cfa3 - [mlir] fix LLVM IR translation of vector<... x index>

[Alex Zinenko]

Author: Alex Zinenko
Date: 2023-02-14T10:42:56Z
New Revision: 8c7cfa357280dd93d33b10bbba0fe33797e27d63

URL: https://github.com/llvm/llvm-project/commit/8c7cfa357280dd93d33b10bbba0fe33797e27d63
DIFF: https://github.com/llvm/llvm-project/commit/8c7cfa357280dd93d33b10bbba0fe33797e27d63.diff

LOG: [mlir] fix LLVM IR translation of vector<... x index>

When the translation was written, `vector<... x index>` was not allowed
at all. After it was added later, the translation was never adapted. It
kept working in the most common case of index-typed attributes using
64-bit storage and being converted to 64-bit integers in LLVM IR, but
not in the other cases that require truncation or extension, producing
wrong results when using the raw data storage of the dense attrbute to
construct the LLVM IR constant. When the storage size doesn't match,
fall back to the per-element constant construction, which is slower but
handles bitwidth differences correctly.

Fixes #60614.

Reviewed By: gysit

Differential Revision: https://reviews.llvm.org/D143993

Added: 
    

Modified: 
    mlir/lib/Target/LLVMIR/ModuleTranslation.cpp
    mlir/test/Target/LLVMIR/llvmir.mlir

Removed: 
    


################################################################################
diff  --git a/mlir/lib/Target/LLVMIR/ModuleTranslation.cpp b/mlir/lib/Target/LLVMIR/ModuleTranslation.cpp
index c928d78fcc558..a4906e4a75248 100644
--- a/mlir/lib/Target/LLVMIR/ModuleTranslation.cpp
+++ b/mlir/lib/Target/LLVMIR/ModuleTranslation.cpp
@@ -216,6 +216,16 @@ convertDenseElementsAttr(Location loc, DenseElementsAttr denseElementsAttr,
   if (type.getNumElements() == 0)
     return nullptr;
 
+  // Check that the raw data size matches what is expected for the scalar size.
+  // TODO: in theory, we could repack the data here to keep constructing from
+  // raw data.
+  // TODO: we may also need to consider endianness when cross-compiling to an
+  // architecture where it is 
diff erent.
+  unsigned elementByteSize = denseElementsAttr.getRawData().size() /
+                             denseElementsAttr.getNumElements();
+  if (8 * elementByteSize != innermostLLVMType->getScalarSizeInBits())
+    return nullptr;
+
   // Compute the shape of all dimensions but the innermost. Note that the
   // innermost dimension may be that of the vector element type.
   bool hasVectorElementType = type.getElementType().isa<VectorType>();

diff  --git a/mlir/test/Target/LLVMIR/llvmir.mlir b/mlir/test/Target/LLVMIR/llvmir.mlir
index 840b1f2caf04f..9e208570776f4 100644
--- a/mlir/test/Target/LLVMIR/llvmir.mlir
+++ b/mlir/test/Target/LLVMIR/llvmir.mlir
@@ -1251,6 +1251,20 @@ llvm.func @complexintconstantarray() -> !llvm.array<2 x !llvm.array<2 x !llvm.st
   llvm.return %1 : !llvm.array<2 x !llvm.array<2 x !llvm.struct<(i32, i32)>>>
 }
 
+// CHECK-LABEL: @indexconstantsplat
+llvm.func @indexconstantsplat() -> vector<3xi32> {
+  %1 = llvm.mlir.constant(dense<42> : vector<3xindex>) : vector<3xi32>
+  // CHECK: ret <3 x i32> <i32 42, i32 42, i32 42>
+  llvm.return %1 : vector<3xi32>
+}
+
+// CHECK-LABEL: @indexconstantarray
+llvm.func @indexconstantarray() -> vector<3xi32> {
+  %1 = llvm.mlir.constant(dense<[0, 1, 2]> : vector<3xindex>) : vector<3xi32>
+  // CHECK: ret <3 x i32> <i32 0, i32 1, i32 2> 
+  llvm.return %1 : vector<3xi32>
+}
+
 llvm.func @noreach() {
 // CHECK:    unreachable
   llvm.unreachable