[PATCH] D20591: [InstCombine] Catch more bswap cases missed due to zext and truncs.

[Chad Rosier via llvm-commits]

mcrosier added inline comments.

================
Comment at: test/Transforms/InstCombine/bswap.ll:106
@@ +105,3 @@
+  %conv = zext i16 %a to i32
+  %shr = lshr i16 %a, 8
+  %shl = shl i32 %conv, 8
----------------
gberry wrote:
> Shouldn't this be an i32 lshr of %conv and the second zext below deleted if you want the IR to look like it would for a target without i16 as a legal type?
With the exception of relabeling the variables and dropping irrelevant nsw/nuws, this is the code that is generated from this C test case when targeting AArch64:

  unsigned short test16(unsigned short a) {
    unsigned short b = (a & 0xff00) >> 8;
    unsigned short c = (a & 0x00ff) << 8;
    return b | c;
  }

Produces:

  define i16 @test16(i16 %a) #0 {
  entry:
    %conv = zext i16 %a to i32
    %shr11 = lshr i16 %a, 8
    %and3 = shl nuw nsw i32 %conv, 8
    %conv5 = zext i16 %shr11 to i32
    %or = or i32 %conv5, %and3
    %conv7 = trunc i32 %or to i16
    ret i16 %conv7
  }

Regardless, I'm happy to add a test case similar to what you're suggesting.


http://reviews.llvm.org/D20591