[Mlir-commits] [mlir] [mlir][amdgpu] Adds make_dma_gather_base (PR #171857)

[Krzysztof Drewniak]

================
@@ -2276,72 +2276,105 @@ struct AMDGPUPermlaneLowering : public ConvertOpToLLVMPattern<PermlaneSwapOp> {
   }
 };
 
-struct AMDGPUMakeDmaBaseLowering
-    : public ConvertOpToLLVMPattern<MakeDmaBaseOp> {
-  using ConvertOpToLLVMPattern::ConvertOpToLLVMPattern;
+static Value setValueAtOffset(ConversionPatternRewriter &rewriter, Location loc,
+                              Value accumulator, Value value, int64_t shift) {
+  shift = shift % 32;
+  Value shiftAmount;
+  if (shift != 0) {
+    shiftAmount = createI32Constant(rewriter, loc, shift % 32);
+    value = LLVM::ShlOp::create(rewriter, loc, value, shiftAmount);
+  }
+
+  if (matchPattern(accumulator, mlir::m_Zero()))
+    return value;
+
+  return LLVM::OrOp::create(rewriter, loc, accumulator, value);
----------------
krzysz00 wrote:

In unrelated trivialities I just realized - all these `or`s are `disjoint` and could probably be marked as such.

https://github.com/llvm/llvm-project/pull/171857