Skip to content

[wasm] Add PackedSimd saturating integer arith. #85043

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
radekdoulik merged 2 commits into from
Apr 24, 2023
Merged

[wasm] Add PackedSimd saturating integer arith. #85043

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
94193a7
[wasm] Add PackedSimd saturating integer arith.
radekdoulik Apr 14, 2023
30ae02a
Add doc comments
radekdoulik Apr 24, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
37 changes: 37 additions & 0 deletions ...tem.Private.CoreLib/src/System/Runtime/Intrinsics/Wasm/PackedSimd.PlatformNotSupported.cs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -123,6 +123,43 @@ public abstract class PackedSimd
public static Vector128<int> AddPairwiseWidening(Vector128<short> value) { throw new PlatformNotSupportedException(); }
public static Vector128<uint> AddPairwiseWidening(Vector128<ushort> value) { throw new PlatformNotSupportedException(); }

// Saturating integer arithmetic

public static Vector128<sbyte> AddSaturate(Vector128<sbyte> left, Vector128<sbyte> right) { throw new PlatformNotSupportedException(); }
public static Vector128<byte> AddSaturate(Vector128<byte> left, Vector128<byte> right) { throw new PlatformNotSupportedException(); }
public static Vector128<short> AddSaturate(Vector128<short> left, Vector128<short> right) { throw new PlatformNotSupportedException(); }
public static Vector128<ushort> AddSaturate(Vector128<ushort> left, Vector128<ushort> right) { throw new PlatformNotSupportedException(); }

public static Vector128<sbyte> SubtractSaturate(Vector128<sbyte> left, Vector128<sbyte> right) { throw new PlatformNotSupportedException(); }
public static Vector128<byte> SubtractSaturate(Vector128<byte> left, Vector128<byte> right) { throw new PlatformNotSupportedException(); }
public static Vector128<short> SubtractSaturate(Vector128<short> left, Vector128<short> right) { throw new PlatformNotSupportedException(); }
public static Vector128<ushort> SubtractSaturate(Vector128<ushort> left, Vector128<ushort> right) { throw new PlatformNotSupportedException(); }

public static Vector128<short> MultiplyRoundedSaturateQ15(Vector128<short> left, Vector128<short> right) { throw new PlatformNotSupportedException(); }

public static Vector128<sbyte> Min(Vector128<sbyte> left, Vector128<sbyte> right) { throw new PlatformNotSupportedException(); }
public static Vector128<byte> Min(Vector128<byte> left, Vector128<byte> right) { throw new PlatformNotSupportedException(); }
public static Vector128<short> Min(Vector128<short> left, Vector128<short> right) { throw new PlatformNotSupportedException(); }
public static Vector128<ushort> Min(Vector128<ushort> left, Vector128<ushort> right) { throw new PlatformNotSupportedException(); }
public static Vector128<int> Min(Vector128<int> left, Vector128<int> right) { throw new PlatformNotSupportedException(); }
public static Vector128<uint> Min(Vector128<uint> left, Vector128<uint> right) { throw new PlatformNotSupportedException(); }

public static Vector128<sbyte> Max(Vector128<sbyte> left, Vector128<sbyte> right) { throw new PlatformNotSupportedException(); }
public static Vector128<byte> Max(Vector128<byte> left, Vector128<byte> right) { throw new PlatformNotSupportedException(); }
public static Vector128<short> Max(Vector128<short> left, Vector128<short> right) { throw new PlatformNotSupportedException(); }
public static Vector128<ushort> Max(Vector128<ushort> left, Vector128<ushort> right) { throw new PlatformNotSupportedException(); }
public static Vector128<int> Max(Vector128<int> left, Vector128<int> right) { throw new PlatformNotSupportedException(); }
public static Vector128<uint> Max(Vector128<uint> left, Vector128<uint> right) { throw new PlatformNotSupportedException(); }

public static Vector128<byte> AverageRounded(Vector128<byte> left, Vector128<byte> right) { throw new PlatformNotSupportedException(); }
public static Vector128<ushort> AverageRounded(Vector128<ushort> left, Vector128<ushort> right) { throw new PlatformNotSupportedException(); }

public static Vector128<sbyte> Abs(Vector128<sbyte> value) { throw new PlatformNotSupportedException(); }
public static Vector128<short> Abs(Vector128<short> value) { throw new PlatformNotSupportedException(); }
public static Vector128<int> Abs(Vector128<int> value) { throw new PlatformNotSupportedException(); }
public static Vector128<long> Abs(Vector128<long> value) { throw new PlatformNotSupportedException(); }
public static Vector128<nint> Abs(Vector128<nint> value) { throw new PlatformNotSupportedException(); }

// Bit shifts

public static Vector128<sbyte> ShiftLeft(Vector128<sbyte> value, int count) { throw new PlatformNotSupportedException(); }
Expand Down
121 changes: 121 additions & 0 deletions src/libraries/System.Private.CoreLib/src/System/Runtime/Intrinsics/Wasm/PackedSimd.cs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -515,6 +515,127 @@ public abstract class PackedSimd
[Intrinsic]
public static Vector128<uint> AddPairwiseWidening(Vector128<ushort> value) => AddPairwiseWidening(value);

// Saturating integer arithmetic

/// <summary>
/// i8x16.add.sat.s
/// </summary>
public static Vector128<sbyte> AddSaturate(Vector128<sbyte> left, Vector128<sbyte> right) => AddSaturate(left, right);
/// <summary>
/// i8x16.add.sat.u
/// </summary>
public static Vector128<byte> AddSaturate(Vector128<byte> left, Vector128<byte> right) => AddSaturate(left, right);
/// <summary>
/// i16x8.add.sat.s
/// </summary>
public static Vector128<short> AddSaturate(Vector128<short> left, Vector128<short> right) => AddSaturate(left, right);
/// <summary>
/// i16x8.add.sat.u
/// </summary>
public static Vector128<ushort> AddSaturate(Vector128<ushort> left, Vector128<ushort> right) => AddSaturate(left, right);

/// <summary>
/// i8x16.sub.sat.s
/// </summary>
public static Vector128<sbyte> SubtractSaturate(Vector128<sbyte> left, Vector128<sbyte> right) => SubtractSaturate(left, right);
/// <summary>
/// i8x16.sub.sat.u
/// </summary>
public static Vector128<byte> SubtractSaturate(Vector128<byte> left, Vector128<byte> right) => SubtractSaturate(left, right);
/// <summary>
/// i16x8.sub.sat.s
/// </summary>
public static Vector128<short> SubtractSaturate(Vector128<short> left, Vector128<short> right) => SubtractSaturate(left, right);
/// <summary>
/// i16x8.sub.sat.u
/// </summary>
public static Vector128<ushort> SubtractSaturate(Vector128<ushort> left, Vector128<ushort> right) => SubtractSaturate(left, right);

/// <summary>
/// i16x8.q15mulr.sat.s
/// </summary>
public static Vector128<short> MultiplyRoundedSaturateQ15(Vector128<short> left, Vector128<short> right) => MultiplyRoundedSaturateQ15(left, right);

/// <summary>
/// i8x16.min.s
/// </summary>
public static Vector128<sbyte> Min(Vector128<sbyte> left, Vector128<sbyte> right) => Min(left, right);
/// <summary>
/// i8x16.min.u
/// </summary>
public static Vector128<byte> Min(Vector128<byte> left, Vector128<byte> right) => Min(left, right);
/// <summary>
/// i16x8.min.s
/// </summary>
public static Vector128<short> Min(Vector128<short> left, Vector128<short> right) => Min(left, right);
/// <summary>
/// i16x8.min.u
/// </summary>
public static Vector128<ushort> Min(Vector128<ushort> left, Vector128<ushort> right) => Min(left, right);
/// <summary>
/// i32x4.min.s
/// </summary>
public static Vector128<int> Min(Vector128<int> left, Vector128<int> right) => Min(left, right);
/// <summary>
/// i32x4.min.u
/// </summary>
public static Vector128<uint> Min(Vector128<uint> left, Vector128<uint> right) => Min(left, right);

/// <summary>
/// i8x16.max.s
/// </summary>
public static Vector128<sbyte> Max(Vector128<sbyte> left, Vector128<sbyte> right) => Max(left, right);
/// <summary>
/// i8x16.max.u
/// </summary>
public static Vector128<byte> Max(Vector128<byte> left, Vector128<byte> right) => Max(left, right);
/// <summary>
/// i16x8.max.s
/// </summary>
public static Vector128<short> Max(Vector128<short> left, Vector128<short> right) => Max(left, right);
/// <summary>
/// i16x8.max.u
/// </summary>
public static Vector128<ushort> Max(Vector128<ushort> left, Vector128<ushort> right) => Max(left, right);
/// <summary>
/// i32x4.max.s
/// </summary>
public static Vector128<int> Max(Vector128<int> left, Vector128<int> right) => Max(left, right);
/// <summary>
/// i32x4.max.u
/// </summary>
public static Vector128<uint> Max(Vector128<uint> left, Vector128<uint> right) => Max(left, right);

/// <summary>
/// i8x16.avgr.u
/// </summary>
public static Vector128<byte> AverageRounded(Vector128<byte> left, Vector128<byte> right) => AverageRounded(left, right);
/// <summary>
/// i16x8.avgr.u
/// </summary>
public static Vector128<ushort> AverageRounded(Vector128<ushort> left, Vector128<ushort> right) => AverageRounded(left, right);

/// <summary>
/// i8x16.abs
/// </summary>
public static Vector128<sbyte> Abs(Vector128<sbyte> value) => Abs(value);
/// <summary>
/// i16x8.abs
/// </summary>
public static Vector128<short> Abs(Vector128<short> value) => Abs(value);
/// <summary>
/// i32x4.abs
/// </summary>
public static Vector128<int> Abs(Vector128<int> value) => Abs(value);
/// <summary>
/// i64x2.abs
/// </summary>
public static Vector128<long> Abs(Vector128<long> value) => Abs(value);
/// <summary>
/// i32x4.abs
/// </summary>
public static Vector128<nint> Abs(Vector128<nint> value) => Abs(value);

// Bit shifts

/// <summary>
Expand Down
28 changes: 28 additions & 0 deletions src/libraries/System.Runtime.Intrinsics/ref/System.Runtime.Intrinsics.cs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5745,6 +5745,34 @@ public abstract partial class PackedSimd
public static Vector128<ushort> AddPairwiseWidening(Vector128<byte> value) { throw null; }
public static Vector128<int> AddPairwiseWidening(Vector128<short> value) { throw null; }
public static Vector128<uint> AddPairwiseWidening(Vector128<ushort> value) { throw null; }
public static Vector128<sbyte> AddSaturate(Vector128<sbyte> left, Vector128<sbyte> right) { throw null; }
public static Vector128<byte> AddSaturate(Vector128<byte> left, Vector128<byte> right) { throw null; }
public static Vector128<short> AddSaturate(Vector128<short> left, Vector128<short> right) { throw null; }
public static Vector128<ushort> AddSaturate(Vector128<ushort> left, Vector128<ushort> right) { throw null; }
public static Vector128<sbyte> SubtractSaturate(Vector128<sbyte> left, Vector128<sbyte> right) { throw null; }
public static Vector128<byte> SubtractSaturate(Vector128<byte> left, Vector128<byte> right) { throw null; }
public static Vector128<short> SubtractSaturate(Vector128<short> left, Vector128<short> right) { throw null; }
public static Vector128<ushort> SubtractSaturate(Vector128<ushort> left, Vector128<ushort> right) { throw null; }
public static Vector128<short> MultiplyRoundedSaturateQ15(Vector128<short> left, Vector128<short> right) { throw null; }
public static Vector128<sbyte> Min(Vector128<sbyte> left, Vector128<sbyte> right) { throw null; }
public static Vector128<byte> Min(Vector128<byte> left, Vector128<byte> right) { throw null; }
public static Vector128<short> Min(Vector128<short> left, Vector128<short> right) { throw null; }
public static Vector128<ushort> Min(Vector128<ushort> left, Vector128<ushort> right) { throw null; }
public static Vector128<int> Min(Vector128<int> left, Vector128<int> right) { throw null; }
public static Vector128<uint> Min(Vector128<uint> left, Vector128<uint> right) { throw null; }
public static Vector128<sbyte> Max(Vector128<sbyte> left, Vector128<sbyte> right) { throw null; }
public static Vector128<byte> Max(Vector128<byte> left, Vector128<byte> right) { throw null; }
public static Vector128<short> Max(Vector128<short> left, Vector128<short> right) { throw null; }
public static Vector128<ushort> Max(Vector128<ushort> left, Vector128<ushort> right) { throw null; }
public static Vector128<int> Max(Vector128<int> left, Vector128<int> right) { throw null; }
public static Vector128<uint> Max(Vector128<uint> left, Vector128<uint> right) { throw null; }
public static Vector128<byte> AverageRounded(Vector128<byte> left, Vector128<byte> right) { throw null; }
public static Vector128<ushort> AverageRounded(Vector128<ushort> left, Vector128<ushort> right) { throw null; }
public static Vector128<sbyte> Abs(Vector128<sbyte> value) { throw null; }
public static Vector128<short> Abs(Vector128<short> value) { throw null; }
public static Vector128<int> Abs(Vector128<int> value) { throw null; }
public static Vector128<long> Abs(Vector128<long> value) { throw null; }
public static Vector128<nint> Abs(Vector128<nint> value) { throw null; }
public static Vector128<sbyte> ShiftLeft(Vector128<sbyte> value, int count) { throw null; }
public static Vector128<byte> ShiftLeft(Vector128<byte> value, int count) { throw null; }
public static Vector128<short> ShiftLeft(Vector128<short> value, int count) { throw null; }
Expand Down
17 changes: 13 additions & 4 deletions src/mono/mono/mini/llvm-intrinsics.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -242,16 +242,18 @@ INTRINS(AESNI_AESDECLAST, x86_aesni_aesdeclast, X86)
INTRINS(AESNI_AESENC, x86_aesni_aesenc, X86)
INTRINS(AESNI_AESENCLAST, x86_aesni_aesenclast, X86)
INTRINS(AESNI_AESIMC, x86_aesni_aesimc, X86)
INTRINS_OVR(SSE_SADD_SATI8, sadd_sat, Generic, v128_i1_t)
INTRINS_OVR(SSE_UADD_SATI8, uadd_sat, Generic, v128_i1_t)
INTRINS_OVR(SSE_SADD_SATI16, sadd_sat, Generic, v128_i2_t)
INTRINS_OVR(SSE_UADD_SATI16, uadd_sat, Generic, v128_i2_t)

INTRINS_OVR(SSE_SSUB_SATI8, ssub_sat, Generic, v128_i1_t)
INTRINS_OVR(SSE_USUB_SATI8, usub_sat, Generic, v128_i1_t)
INTRINS_OVR(SSE_SSUB_SATI16, ssub_sat, Generic, v128_i2_t)
INTRINS_OVR(SSE_USUB_SATI16, usub_sat, Generic, v128_i2_t)
#endif
#if defined(TARGET_AMD64) || defined(TARGET_X86) || defined(TARGET_WASM)
INTRINS_OVR(SSE_SADD_SATI8, sadd_sat, Generic, v128_i1_t)
INTRINS_OVR(SSE_UADD_SATI8, uadd_sat, Generic, v128_i1_t)
INTRINS_OVR(SSE_SADD_SATI16, sadd_sat, Generic, v128_i2_t)
INTRINS_OVR(SSE_UADD_SATI16, uadd_sat, Generic, v128_i2_t)
#endif
#if defined(TARGET_WASM)
INTRINS_OVR(WASM_EXTADD_PAIRWISE_SIGNED_V16, wasm_extadd_pairwise_signed, Wasm, sse_i2_t)
INTRINS_OVR(WASM_EXTADD_PAIRWISE_SIGNED_V8, wasm_extadd_pairwise_signed, Wasm, sse_i4_t)
Expand All @@ -265,6 +267,8 @@ INTRINS_OVR(WASM_ANYTRUE_V16, wasm_anytrue, Wasm, sse_i1_t)
INTRINS_OVR(WASM_ANYTRUE_V8, wasm_anytrue, Wasm, sse_i2_t)
INTRINS_OVR(WASM_ANYTRUE_V4, wasm_anytrue, Wasm, sse_i4_t)
INTRINS_OVR(WASM_ANYTRUE_V2, wasm_anytrue, Wasm, sse_i8_t)
INTRINS_OVR(WASM_AVERAGE_ROUNDED_V16, wasm_avgr_unsigned, Wasm, sse_i1_t)
INTRINS_OVR(WASM_AVERAGE_ROUNDED_V8, wasm_avgr_unsigned, Wasm, sse_i2_t)
INTRINS_OVR(WASM_BITMASK_V16, wasm_bitmask, Wasm, sse_i1_t)
INTRINS_OVR(WASM_BITMASK_V8, wasm_bitmask, Wasm, sse_i2_t)
INTRINS_OVR(WASM_BITMASK_V4, wasm_bitmask, Wasm, sse_i4_t)
Expand All @@ -276,7 +280,12 @@ INTRINS_OVR_2_ARG(WASM_NARROW_SIGNED_V16, wasm_narrow_signed, Wasm, sse_i1_t, ss
INTRINS_OVR_2_ARG(WASM_NARROW_SIGNED_V8, wasm_narrow_signed, Wasm, sse_i2_t, sse_i4_t)
INTRINS_OVR_2_ARG(WASM_NARROW_UNSIGNED_V16, wasm_narrow_unsigned, Wasm, sse_i1_t, sse_i2_t)
INTRINS_OVR_2_ARG(WASM_NARROW_UNSIGNED_V8, wasm_narrow_unsigned, Wasm, sse_i2_t, sse_i4_t)
INTRINS(WASM_Q15MULR_SAT_SIGNED, wasm_q15mulr_sat_signed, Wasm)
INTRINS(WASM_SHUFFLE, wasm_shuffle, Wasm)
INTRINS_OVR(WASM_SUB_SAT_SIGNED_V16, wasm_sub_sat_signed, Wasm, sse_i1_t)
INTRINS_OVR(WASM_SUB_SAT_SIGNED_V8, wasm_sub_sat_signed, Wasm, sse_i2_t)
INTRINS_OVR(WASM_SUB_SAT_UNSIGNED_V16, wasm_sub_sat_unsigned, Wasm, sse_i1_t)
INTRINS_OVR(WASM_SUB_SAT_UNSIGNED_V8, wasm_sub_sat_unsigned, Wasm, sse_i2_t)
INTRINS(WASM_SWIZZLE, wasm_swizzle, Wasm)
#endif
#if defined(TARGET_ARM64)
Expand Down
73 changes: 73 additions & 0 deletions src/mono/mono/mini/simd-intrinsics.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4891,17 +4891,23 @@ static SimdIntrinsic wasmbase_methods [] = {
};

static SimdIntrinsic packedsimd_methods [] = {
{SN_Abs, OP_VECTOR_IABS},
{SN_Add},
{SN_AddPairwiseWidening},
{SN_AddSaturate},
{SN_And, OP_XBINOP_FORCEINT, XBINOP_FORCEINT_AND},
{SN_AverageRounded},
{SN_Bitmask, OP_WASM_SIMD_BITMASK},
{SN_CompareEqual},
{SN_CompareNotEqual},
{SN_ConvertNarrowingSignedSaturate},
{SN_ConvertNarrowingUnsignedSaturate},
{SN_Dot, OP_XOP_X_X_X, INTRINS_WASM_DOT},
{SN_ExtractLane},
{SN_Max, OP_XBINOP, OP_IMIN, OP_XBINOP, OP_IMIN_UN},
{SN_Min, OP_XBINOP, OP_IMAX, OP_XBINOP, OP_IMAX_UN},
{SN_Multiply},
{SN_MultiplyRoundedSaturateQ15, OP_XOP_X_X_X, INTRINS_WASM_Q15MULR_SAT_SIGNED},
{SN_MultiplyWideningLower, OP_WASM_EXTMUL_LOWER, 0, OP_WASM_EXTMUL_LOWER_U},
{SN_MultiplyWideningUpper, OP_WASM_EXTMUL_UPPER, 0, OP_WASM_EXTMUL_UPPER_U},
{SN_Negate},
Expand All @@ -4912,6 +4918,7 @@ static SimdIntrinsic packedsimd_methods [] = {
{SN_Shuffle, OP_WASM_SIMD_SHUFFLE},
{SN_Splat},
{SN_Subtract},
{SN_SubtractSaturate},
{SN_Swizzle, OP_WASM_SIMD_SWIZZLE},
{SN_get_IsSupported},
};
Expand Down Expand Up @@ -5018,6 +5025,48 @@ emit_wasm_supported_intrinsics (

return NULL;
}
case SN_AddSaturate: {
op = OP_XOP_X_X_X;

switch (arg0_type) {
case MONO_TYPE_I1:
c0 = INTRINS_SSE_SADD_SATI8;
break;
case MONO_TYPE_I2:
c0 = INTRINS_SSE_SADD_SATI16;
break;
case MONO_TYPE_U1:
c0 = INTRINS_SSE_UADD_SATI8;
break;
case MONO_TYPE_U2:
c0 = INTRINS_SSE_UADD_SATI16;
break;
}

// continue with default emit
if (c0 != 0)
break;

return NULL;
}
case SN_AverageRounded: {
op = OP_XOP_X_X_X;

switch (arg0_type) {
case MONO_TYPE_U1:
c0 = INTRINS_WASM_AVERAGE_ROUNDED_V16;
break;
case MONO_TYPE_U2:
c0 = INTRINS_WASM_AVERAGE_ROUNDED_V8;
break;
}

// continue with default emit
if (c0 != 0)
break;

return NULL;
}
case SN_CompareEqual:
return emit_simd_ins_for_sig (cfg, klass, type_enum_is_float (arg0_type) ? OP_XCOMPARE_FP : OP_XCOMPARE, CMP_EQ, arg0_type, fsig, args);
case SN_CompareNotEqual:
Expand Down Expand Up @@ -5074,6 +5123,30 @@ emit_wasm_supported_intrinsics (
g_assert (fsig->param_count == 1 && mono_metadata_type_equal (fsig->params [0], etype));
return emit_simd_ins (cfg, klass, type_to_expand_op (etype->type), args [0]->dreg, -1);
}
case SN_SubtractSaturate: {
op = OP_XOP_X_X_X;

switch (arg0_type) {
case MONO_TYPE_I1:
c0 = INTRINS_WASM_SUB_SAT_SIGNED_V16;
break;
case MONO_TYPE_I2:
c0 = INTRINS_WASM_SUB_SAT_SIGNED_V8;
break;
case MONO_TYPE_U1:
c0 = INTRINS_WASM_SUB_SAT_UNSIGNED_V16;
break;
case MONO_TYPE_U2:
c0 = INTRINS_WASM_SUB_SAT_UNSIGNED_V8;
break;
}

// continue with default emit
if (c0 != 0)
break;

return NULL;
}
}

// default emit path for cases with op set
Expand Down
2 changes: 2 additions & 0 deletions src/mono/mono/mini/simd-methods.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -634,10 +634,12 @@ METHOD(MultiplyRoundedDoublingScalarBySelectedScalarAndSubtractSaturateHigh)
// Arm.Dp
METHOD(DotProductBySelectedQuadruplet)
// Wasm
METHOD(AverageRounded)
METHOD(Bitmask)
METHOD(ConvertNarrowingSignedSaturate)
METHOD(ConvertNarrowingUnsignedSaturate)
METHOD(ExtractLane)
METHOD(MultiplyRoundedSaturateQ15)
METHOD(ReplaceLane)
METHOD(ShiftLeft)
METHOD(Splat)
Expand Down