[Mlir-commits] [mlir] [MLIR] Implement emulation of static indexing subbyte type vector stores (PR #115922)

[Andrzej Warzyński]

================
@@ -0,0 +1,104 @@
+// RUN: mlir-opt --test-emulate-narrow-int="arith-compute-bitwidth=1 memref-load-bitwidth=8 atomic-store=false" --cse --split-input-file %s | FileCheck %s
+
+// TODO: remove memref.alloc() in the tests to eliminate noises.
+// memref.alloc exists here because sub-byte vector data types such as i2
+// are currently not supported as input arguments.
+
+func.func @vector_store_i2_const_rmw(%arg0: vector<3xi2>) {
+    %0 = memref.alloc() : memref<3x3xi2>
+    %c0 = arith.constant 0 : index
+    %c2 = arith.constant 2 : index
+    vector.store %arg0, %0[%c2, %c0] :memref<3x3xi2>, vector<3xi2>
+    return
+}
+// CHECK: func @vector_store_i2_const_rmw(
+// CHECK-SAME: %[[ARG0:.+]]: vector<3xi2>)
+// CHECK: %[[ALLOC:.+]] = memref.alloc() : memref<3xi8>
+// CHECK: %[[C1:.+]] = arith.constant 1 : index
+// CHECK: %[[CST:.+]] = arith.constant dense<[false, false, true, true]>
+// CHECK: %[[CST0:.+]] = arith.constant dense<0> : vector<4xi2>
+// CHECK: %[[EXTRACT:.+]] = vector.extract_strided_slice %[[ARG0]]
+// CHECK-SAME: {offsets = [0], sizes = [2], strides = [1]} : vector<3xi2> to vector<2xi2>
+// CHECK: %[[INSERT:.+]] = vector.insert_strided_slice %[[EXTRACT]], %[[CST0]]
+// CHECK-SAME: {offsets = [2], strides = [1]} : vector<2xi2> into vector<4xi2>
+// CHECK: %[[LOAD:.+]] = vector.load
+
+// Actual part to do RMW sequence
+// CHECK: %[[UPCAST:.+]] = vector.bitcast %[[LOAD]] : vector<1xi8> to vector<4xi2>
+// CHECK: %[[SELECT:.+]] = arith.select %[[CST]], %[[UPCAST]], %[[INSERT]]
+// CHECK: %[[DOWNCAST:.+]] = vector.bitcast %[[SELECT]]
+// CHECK: vector.store %[[DOWNCAST]], %[[ALLOC]][%[[C1]]]
----------------
banach-space wrote:

Shouldn't there be `memref.generic_atomic_rmw`?

https://github.com/llvm/llvm-project/pull/115922