[Mlir-commits] [mlir] [mlir][arith] mul operation regressions (PR #96975)

[Andrzej Warzyński]

================
@@ -0,0 +1,44 @@
+// Tests mul operations and their variants (e.g. extended).
+// These tests are intended to be target agnostic: they should yield the same results 
+// regardless of the target platform.
+
+// RUN: mlir-opt %s --convert-scf-to-cf --convert-cf-to-llvm --convert-vector-to-llvm \
+// RUN:             --convert-func-to-llvm --convert-arith-to-llvm | \
+// RUN:   mlir-cpu-runner -e entry -entry-point-result=void \
+// RUN:                   --shared-libs=%mlir_c_runner_utils | \
+// RUN:   FileCheck %s --match-full-lines
+
+func.func @mulsi_extended_on_i1() {
+  // mulsi_extended on i1, tests for overflow bit
+  // mulsi_extended 1, 1 : i1 = (1, 0)
+  // CHECK:      1
+  // CHECK-NEXT: 0
+  %true = arith.constant true
+  %low, %high = arith.mulsi_extended %true, %true : i1
+  vector.print %low : i1
+  vector.print %high : i1
+  return
+}
+
+func.func @mulsi_mului_extended_overflows() {
+  // mulsi and mului extended versions, with overflow
+  // mulsi_extended -100, -100 : i8 = (16, 39); mului_extended -100, -100 : i8 = (16, 95)
+  // CHECK-NEXT:  16
+  // CHECK-NEXT:  39
+  // CHECK-NEXT:  16
+  // CHECK-NEXT:  95
+  %c-100_i8 = arith.constant -100 : i8
+  %low, %high = arith.mulsi_extended %c-100_i8, %c-100_i8 : i8
+  vector.print %low : i8
+  vector.print %high : i8
+  %low_0, %high_1 = arith.mului_extended %c-100_i8, %c-100_i8 : i8
----------------
banach-space wrote:

Why `low_0` but `high_1` (i.e. 0 vs 1)?

Also, why complicate the unsigned version and pass negative numbers? IIUC, these will be interpreted as 156, so the actual multiplication will be `156 * 156`. What's the benefit of obfuscating that as `-100 * -100`? I do see the value of verifying that `arith.mului` gives the same result for:
*  `-100 * -100`, and
* `156 * 156`, and
* `-100 * 156`

but that's not what's being tested here. Also, I hope that I got that ^^^ correctly? 

https://github.com/llvm/llvm-project/pull/96975