Quarter Wide Int Arith Pass

lib/Dialect/Arith/Conversions/ArithToModArith/ArithToModArith.cpp

@@ -98,14 +98,14 @@ struct ConvertConstant : public OpConversionPattern<mlir::arith::ConstantOp> {
   }
 };
 
-struct ConvertExt : public OpConversionPattern<mlir::arith::ExtSIOp> {
+struct ConvertExt : public OpConversionPattern<mlir::arith::ExtUIOp> {
   ConvertExt(mlir::MLIRContext *context)
-      : OpConversionPattern<mlir::arith::ExtSIOp>(context) {}
+      : OpConversionPattern<mlir::arith::ExtUIOp>(context) {}
 
   using OpConversionPattern::OpConversionPattern;
 
   LogicalResult matchAndRewrite(
-      ::mlir::arith::ExtSIOp op, OpAdaptor adaptor,
+      ::mlir::arith::ExtUIOp op, OpAdaptor adaptor,
       ConversionPatternRewriter &rewriter) const override {
     ImplicitLocOpBuilder b(op.getLoc(), rewriter);
 

lib/Dialect/Arith/Transforms/BUILD

@@ -0,0 +1,40 @@
+load("@heir//lib/Transforms:transforms.bzl", "add_heir_transforms")
+
+package(
+    default_applicable_licenses = ["@heir//:license"],
+    default_visibility = ["//visibility:public"],
+)
+
+cc_library(
+    name = "Transforms",
+    hdrs = ["Passes.h"],
+    deps = [
+        ":QuarterWideInt",
+        ":pass_inc_gen",
+        "@llvm-project//mlir:ArithDialect",
+        "@llvm-project//mlir:IR",
+    ],
+)
+
+cc_library(
+    name = "QuarterWideInt",
+    srcs = ["QuarterWideInt.cpp"],
+    hdrs = ["QuarterWideInt.h"],
+    deps = [
+        ":pass_inc_gen",
+        "@heir//lib/Utils:ConversionUtils",
+        "@llvm-project//mlir:ArithDialect",
+        "@llvm-project//mlir:FuncDialect",
+        "@llvm-project//mlir:FuncTransforms",
+        "@llvm-project//mlir:IR",
+        "@llvm-project//mlir:Pass",
+        "@llvm-project//mlir:TensorDialect",
+        "@llvm-project//mlir:Transforms",
+    ],
+)
+
+add_heir_transforms(
+    header_filename = "Passes.h.inc",
+    pass_name = "Arith",
+    td_file = "Passes.td",
+)

lib/Dialect/Arith/Transforms/Passes.h

@@ -0,0 +1,17 @@
+#ifndef LIB_DIALECT_ARITH_TRANSFORMS_PASSES_H_
+#define LIB_DIALECT_ARITH_TRANSFORMS_PASSES_H_
+
+#include "lib/Dialect/Arith/Transforms/QuarterWideInt.h"
+
+namespace mlir {
+namespace heir {
+namespace arith {
+
+#define GEN_PASS_REGISTRATION
+#include "lib/Dialect/Arith/Transforms/Passes.h.inc"
+
+}  // namespace arith
+}  // namespace heir
+}  // namespace mlir
+
+#endif  // LIB_DIALECT_ARITH_TRANSFORMS_PASSES_H_

lib/Dialect/Arith/Transforms/Passes.td

@@ -0,0 +1,26 @@
+#ifndef LIB_DIALECT_ARITH_TRANSFORMS_PASSES_TD_
+#define LIB_DIALECT_ARITH_TRANSFORMS_PASSES_TD_
+
+include "mlir/Pass/PassBase.td"
+
+def QuarterWideInt : Pass<"arith-quarter-wide-int"> {
+
+  let summary = "Convert high precision arithmetic operations to a sequence of lower precision operations";
+  let description = [{
+    This pass converts high precision arithmetic operations, i.e. operations on 32 bit integer,
+    into a sequence of lower precision operations, i.e 8b operations.
+    Currently, the pass splits the 32b integer into four 8b integers, using the tensor dialect.
+    These smaller integers are stored in an 16b integer, so that we don't lose the carry information.
+
+    Based on the `arith-emulate-wide-int` pass from the MLIR arith dialect.
+
+    General assumption: the first element in the tensor is also the LSB element.
+  }];
+  let dependentDialects = [
+    "mlir::arith::ArithDialect",
+    "mlir::vector::VectorDialect",
+    "mlir::tensor::TensorDialect",
+  ];
+}
+
+#endif  // LIB_DIALECT_ARITH_TRANSFORMS_PASSES_TD_