GitLab

Motivation

One would except these two definitions to have similar performance.

twiddleA, twiddleB :: Word8-> Bool
twiddleA w = w < 0x80
twiddleB w = not (testBit w 7)

They don't look that different at the STG level:

M.twiddleA :: GHC.Word.Word8 -> GHC.Types.Bool
[GblId, Arity=1, Caf=NoCafRefs, Str=<S,1*U(U)>, Unf=OtherCon []] =
    \r [w_s1Dd]
        case w_s1Dd of {
        GHC.Word.W8# x_s1Df [Occ=Once] ->
        case ltWord# [x_s1Df 128##] of sat_s1Dg [Occ=Once] {
        __DEFAULT -> tagToEnum# [sat_s1Dg];
        };
        };

M.twiddleB :: GHC.Word.Word8 -> GHC.Types.Bool
[GblId, Arity=1, Caf=NoCafRefs, Str=<S,1*U(U)>, Unf=OtherCon []] =
    \r [w_s1Dh]
        case w_s1Dh of {
        GHC.Word.W8# x#_s1Dj [Occ=Once] ->
        case and# [x#_s1Dj 128##] of {
          __DEFAULT -> GHC.Types.False [];
          0## -> GHC.Types.True [];
        };
        };

However in Cmm their core (ignoring unboxing overhead and some details) is:

//Variant A:
           R1 = I64[((I64[R1 + 7] < 128) << 3) + GHC.Types.Bool_closure_tbl];
           call (P64[Sp])(R1) args: 8, res: 0, upd: 8;
//Variant B:
       c1DJ: // global
           if (I64[R1 + 7] & 128 != 0) goto c1DX; else goto c1E3;
       c1DX: // global
           R1 = GHC.Types.False_closure+1;
           call (P64[Sp])(R1) args: 8, res: 0, upd: 8;
       c1E3: // global
           R1 = GHC.Types.True_closure+2;
           call (P64[Sp])(R1) args: 8, res: 0, upd: 8;

But we would really like to always get the first variant. Which I will confidently say is better even without benchmarking.

Proposal

The issue is that we can use tagToEnem# to go from Int# to Bool if the range is [0..1]. But the current testBit implementation uses .&. so there is no guarantee that the value will be in this range, so we end up with a regular comparison.

If we rewrite testBit to produce the Bool via tagToEnum# in haskell then we can fix this without touching codeGen at all.

Describe your proposed feature here.