[llvm] [SelectionDAG] Make `(a & x) | (~a & y) -> (a & (x ^ y)) ^ y` available for all targets (PR #137641)

[Iris Shi via llvm-commits]

================
@@ -0,0 +1,277 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc < %s -mtriple=s390x-linux-gnu -mcpu=z13 | FileCheck %s --check-prefix=NO-MISC3
+; RUN: llc < %s -mtriple=s390x-linux-gnu -mcpu=z15 | FileCheck %s --check-prefix=MISC3
+
+; test that masked-merge code is generated as "xor;and;xor" sequence or
+; "andn ; and; or" if and-not is available.
+
+define i32 @masked_merge0(i32 %a0, i32 %a1, i32 %a2) {
+; NO-MISC3-LABEL: masked_merge0:
+; NO-MISC3:       # %bb.0:
+; NO-MISC3-NEXT:    xr %r3, %r4
+; NO-MISC3-NEXT:    nr %r2, %r3
+; NO-MISC3-NEXT:    xr %r2, %r4
+; NO-MISC3-NEXT:    br %r14
+;
+; MISC3-LABEL: masked_merge0:
+; MISC3:       # %bb.0:
+; MISC3-NEXT:    nr %r3, %r2
+; MISC3-NEXT:    ncrk %r2, %r4, %r2
+; MISC3-NEXT:    or %r2, %r3
+; MISC3-NEXT:    br %r14
+  %and0 = and i32 %a0, %a1
+  %not = xor i32 %a0, -1
+  %and1 = and i32 %not, %a2
+  %or = or i32 %and0, %and1
+  ret i32 %or
+}
+
+define i16 @masked_merge1(i16 %a0, i16 %a1, i16 %a2) {
+; NO-MISC3-LABEL: masked_merge1:
+; NO-MISC3:       # %bb.0:
+; NO-MISC3-NEXT:    xr %r3, %r4
+; NO-MISC3-NEXT:    nr %r2, %r3
+; NO-MISC3-NEXT:    xr %r2, %r4
+; NO-MISC3-NEXT:    br %r14
+;
+; MISC3-LABEL: masked_merge1:
+; MISC3:       # %bb.0:
+; MISC3-NEXT:    ncrk %r0, %r4, %r2
+; MISC3-NEXT:    nr %r2, %r3
+; MISC3-NEXT:    or %r2, %r0
+; MISC3-NEXT:    br %r14
+  %and0 = and i16 %a0, %a1
+  %not = xor i16 %a0, -1
+  %and1 = and i16 %a2, %not
+  %or = or i16 %and0, %and1
+  ret i16 %or
+}
+
+define i8 @masked_merge2(i8 %a0, i8 %a1, i8 %a2) {
+; NO-MISC3-LABEL: masked_merge2:
+; NO-MISC3:       # %bb.0:
+; NO-MISC3-NEXT:    lr %r2, %r3
+; NO-MISC3-NEXT:    br %r14
+;
+; MISC3-LABEL: masked_merge2:
+; MISC3:       # %bb.0:
+; MISC3-NEXT:    lr %r2, %r3
+; MISC3-NEXT:    br %r14
+  %not = xor i8 %a0, -1
+  %and0 = and i8 %not, %a1
----------------
el-ev wrote:

The test was copied from X86 directory. Perhaps they thought the pattern `(a & x) | (~a & x)` was worth testing?
Also, the result isn't that trivial, as the optimization wasn't performed before.

There is a CI failure for lldb, I don't it's related to the changes here.

https://github.com/llvm/llvm-project/pull/137641