[wasm] Add extending conversions to PackedSimd

[radekdoulik]

No description provided.

[dotnet-issue-labeler]

Note regarding the new-api-needs-documentation label:

This serves as a reminder for when your PR is modifying a ref *.cs file and adding/modifying public APIs, please make sure the API implementation in the src *.cs file is documented with triple slash comments, so the PR reviewers can sign off that change.

[ghost]

Tagging subscribers to this area: @dotnet/area-system-runtime-intrinsics
See info in area-owners.md if you want to be subscribed.

Issue Details

---

null

Author:	radekdoulik
Assignees:	-
Labels:	area-System.Runtime.Intrinsics, new-api-needs-documentation
Milestone:	-

[radekdoulik]

Test method:

    static void TestExtConversions()
    {
        Vector128<int> v1int = Vector128.Create(RandomInt, RandomInt, RandomInt, RandomInt);
        Vector128<uint> v1uint = Vector128.Create((uint)Random.Shared.Next(), (uint)Random.Shared.Next(), (uint)Random.Shared.Next(), (uint)Random.Shared.Next());
        Vector128<byte> v1byte = Vector128.Create(0, (byte)Random.Shared.Next(), (byte)Random.Shared.Next(), (byte)Random.Shared.Next(), (byte)Random.Shared.Next(), (byte)Random.Shared.Next(), (byte)Random.Shared.Next(), (byte)Random.Shared.Next(), (byte)Random.Shared.Next(), (byte)Random.Shared.Next(), (byte)Random.Shared.Next(), (byte)Random.Shared.Next(), (byte)Random.Shared.Next(), (byte)Random.Shared.Next(), (byte)Random.Shared.Next(), (byte)Random.Shared.Next());
        Vector128<sbyte> v1sbyte = Vector128.Create(RandomSByte, RandomSByte, RandomSByte, RandomSByte, RandomSByte, RandomSByte, RandomSByte, RandomSByte, RandomSByte, RandomSByte, RandomSByte, RandomSByte, RandomSByte, RandomSByte, RandomSByte, RandomSByte);
        Vector128<short> v1short = Vector128.Create(RandomShort, RandomShort, RandomShort, RandomShort, RandomShort, RandomShort, RandomShort, RandomShort);
        Vector128<ushort> v1ushort = Vector128.Create((ushort)Random.Shared.Next(), (ushort)Random.Shared.Next(), (ushort)Random.Shared.Next(), (ushort)Random.Shared.Next(), (ushort)Random.Shared.Next(), (ushort)Random.Shared.Next(), (ushort)Random.Shared.Next(), (ushort)Random.Shared.Next());

        Console.WriteLine($"v1sbyte: {v1sbyte}");
        Console.WriteLine($"v1byte: {v1byte}");
        Console.WriteLine($"v1short: {v1short}");
        Console.WriteLine($"v1ushort: {v1ushort}");
        Console.WriteLine($"v1int: {v1int}");
        Console.WriteLine($"v1uint: {v1uint}");

        Console.WriteLine($"sbyte to short sign ext (lower): {PackedSimd.SignExtendWideningLower(v1sbyte)}");
        Console.WriteLine($"sbyte to short sign ext (upper): {PackedSimd.SignExtendWideningUpper(v1sbyte)}");
        Console.WriteLine($"sbyte to short zero ext (lower): {PackedSimd.ZeroExtendWideningLower(v1sbyte)}");
        Console.WriteLine($"sbyte to short zero ext (upper): {PackedSimd.ZeroExtendWideningUpper(v1sbyte)}");

        Console.WriteLine($"byte to ushort sign ext (lower): {PackedSimd.SignExtendWideningLower(v1byte)}");
        Console.WriteLine($"byte to ushort sign ext (upper): {PackedSimd.SignExtendWideningUpper(v1byte)}");
        Console.WriteLine($"byte to ushort zero ext (lower): {PackedSimd.ZeroExtendWideningLower(v1byte)}");
        Console.WriteLine($"byte to ushort zero ext (upper): {PackedSimd.ZeroExtendWideningUpper(v1byte)}");

        Console.WriteLine($"short to int sign ext (lower): {PackedSimd.SignExtendWideningLower(v1short)}");
        Console.WriteLine($"short to int sign ext (upper): {PackedSimd.SignExtendWideningUpper(v1short)}");
        Console.WriteLine($"short to int zero ext (lower): {PackedSimd.ZeroExtendWideningLower(v1short)}");
        Console.WriteLine($"short to int zero ext (upper): {PackedSimd.ZeroExtendWideningUpper(v1short)}");

        Console.WriteLine($"ushort to uint sign ext (lower): {PackedSimd.SignExtendWideningLower(v1ushort)}");
        Console.WriteLine($"ushort to uint sign ext (upper): {PackedSimd.SignExtendWideningUpper(v1ushort)}");
        Console.WriteLine($"ushort to uint zero ext (lower): {PackedSimd.ZeroExtendWideningLower(v1ushort)}");
        Console.WriteLine($"ushort to uint zero ext (upper): {PackedSimd.ZeroExtendWideningUpper(v1ushort)}");

        Console.WriteLine($"int to long sign ext (lower): {PackedSimd.SignExtendWideningLower(v1int)}");
        Console.WriteLine($"int to long sign ext (upper): {PackedSimd.SignExtendWideningUpper(v1int)}");
        Console.WriteLine($"int to long zero ext (lower): {PackedSimd.ZeroExtendWideningLower(v1int)}");
        Console.WriteLine($"int to long zero ext (upper): {PackedSimd.ZeroExtendWideningUpper(v1int)}");

        Console.WriteLine($"uint to ulong sign ext (lower): {PackedSimd.SignExtendWideningLower(v1uint)}");
        Console.WriteLine($"uint to ulong sign ext (upper): {PackedSimd.SignExtendWideningUpper(v1uint)}");
        Console.WriteLine($"uint to ulong zero ext (lower): {PackedSimd.ZeroExtendWideningLower(v1uint)}");
        Console.WriteLine($"uint to ulong zero ext (upper): {PackedSimd.ZeroExtendWideningUpper(v1uint)}");
    }

output:

  v1sbyte: <-2, -25, 81, 85, 94, -114, -46, 18, 1, -66, 46, -73, 76, -4, 124, 88>
  v1byte: <0, 229, 13, 234, 228, 107, 238, 119, 4, 190, 183, 84, 87, 127, 46, 135>
  v1short: <-21942, 12726, -31745, -3459, 11206, -25452, 15233, -19319>
  v1ushort: <58688, 38751, 4831, 63651, 29450, 49653, 62398, 19778>
  v1int: <-330723718, 654137123, 666254210, -764101589>
  v1uint: <1734002338, 80703557, 1326853359, 991892688>
  sbyte to short sign ext (lower): <-2, -25, 81, 85, 94, -114, -46, 18>
  sbyte to short sign ext (upper): <1, -66, 46, -73, 76, -4, 124, 88>
  sbyte to short zero ext (lower): <254, 231, 81, 85, 94, 142, 210, 18>
  sbyte to short zero ext (upper): <1, 190, 46, 183, 76, 252, 124, 88>
  byte to ushort sign ext (lower): <0, 65509, 13, 65514, 65508, 107, 65518, 119>
  byte to ushort sign ext (upper): <4, 65470, 65463, 84, 87, 127, 46, 65415>
  byte to ushort zero ext (lower): <0, 229, 13, 234, 228, 107, 238, 119>
  byte to ushort zero ext (upper): <4, 190, 183, 84, 87, 127, 46, 135>
  short to int sign ext (lower): <-21942, 12726, -31745, -3459>
  short to int sign ext (upper): <11206, -25452, 15233, -19319>
  short to int zero ext (lower): <43594, 12726, 33791, 62077>
  short to int zero ext (upper): <11206, 40084, 15233, 46217>
  ushort to uint sign ext (lower): <4294960448, 4294940511, 4831, 4294965411>
  ushort to uint sign ext (upper): <29450, 4294951413, 4294964158, 19778>
  ushort to uint zero ext (lower): <58688, 38751, 4831, 63651>
  ushort to uint zero ext (upper): <29450, 49653, 62398, 19778>
  int to long sign ext (lower): <-330723718, 654137123>
  int to long sign ext (upper): <666254210, -764101589>
  int to long zero ext (lower): <3964243578, 654137123>
  int to long zero ext (upper): <666254210, 3530865707>
  uint to ulong sign ext (lower): <1734002338, 80703557>
  uint to ulong sign ext (upper): <1326853359, 991892688>
  uint to ulong zero ext (lower): <1734002338, 80703557>
  uint to ulong zero ext (upper): <1326853359, 991892688>

emitted code check:

> wa-info -v -d -f TestExtConv src/mono/sample/wasm/console-node/bin/Release/AppBundle/dotnet.native.wasm|grep SIMD|egrep 'ext'
      i16x8.extend.low.i8x16.s    [SIMD]
      i16x8.extend.high.i8x16.s    [SIMD]
      i16x8.extend.low.i8x16.u    [SIMD]
      i16x8.extend.low.i8x16.u    [SIMD]
      i16x8.extend.low.i8x16.s    [SIMD]
      i16x8.extend.high.i8x16.s    [SIMD]
      i16x8.extend.low.i8x16.u    [SIMD]
      i16x8.extend.low.i8x16.u    [SIMD]
      i32x4.extend.low.i16x8.s    [SIMD]
      i32x4.extend.high.i16x8.s    [SIMD]
      i32x4.extend.low.i16x8.u    [SIMD]
      i32x4.extend.low.i16x8.u    [SIMD]
      i32x4.extend.low.i16x8.s    [SIMD]
      i32x4.extend.high.i16x8.s    [SIMD]
      i32x4.extend.low.i16x8.u    [SIMD]
      i32x4.extend.low.i16x8.u    [SIMD]
      i64x2.extend.low.i32x4.s    [SIMD]
      i64x2.extend.high.i32x4.s    [SIMD]
      i64x2.extend.low.i32x4.u    [SIMD]
      i64x2.extend.low.i32x4.u    [SIMD]
      i64x2.extend.low.i32x4.s    [SIMD]
      i64x2.extend.high.i32x4.s    [SIMD]
      i64x2.extend.low.i32x4.u    [SIMD]
 i64x2.extend.low.i32x4.u    [SIMD]

[radekdoulik]

Looks like the SN_ZeroExtendWideningUpper is either not emitted right or reported wrong by wa-info, I will look into that.

[radekdoulik]

Looks like the SN_ZeroExtendWideningUpper is either not emitted right or reported wrong by wa-info, I will look into that.

Looks like llvm is optimizing the code to use i16x8.extend.low.i8x16.u instead of i16x8.extend.high.i8x16.u. Not sure why. The actual output is right though.

The "optimized" code:

...
 i8x16.shuffle 0x00000000000000000f0e0d0c0b0a0908    [SIMD]
 i16x8.extend.low.i8x16.u    [SIMD]
...