From 7ecb1f36acab7b48d77991d58d456a34074a2d0e Mon Sep 17 00:00:00 2001 From: David Chase Date: Fri, 19 Dec 2025 15:50:57 -0500 Subject: simd/archsimd: add HasAVX2() guards to tests that need them This may not be complete, because some 256-bit float operations may be implemented with integer simd instead (in some cases there is no float op). (And some tests may have just been overlooked.) Fixes #76866. Change-Id: I90b1dfe039c322c484af916436fc6f6c50a31030 Reviewed-on: https://go-review.googlesource.com/c/go/+/731502 LUCI-TryBot-Result: Go LUCI Auto-Submit: David Chase Reviewed-by: Cherry Mui --- .../archsimd/internal/simd_test/binary_test.go | 195 ++++++++++++++------- 1 file changed, 129 insertions(+), 66 deletions(-) (limited to 'src/simd/archsimd/internal/simd_test/binary_test.go') diff --git a/src/simd/archsimd/internal/simd_test/binary_test.go b/src/simd/archsimd/internal/simd_test/binary_test.go index fa2b9511ca..28efdcb52f 100644 --- a/src/simd/archsimd/internal/simd_test/binary_test.go +++ b/src/simd/archsimd/internal/simd_test/binary_test.go @@ -17,23 +17,29 @@ func TestAdd(t *testing.T) { testFloat64x2Binary(t, archsimd.Float64x2.Add, addSlice[float64]) testFloat64x4Binary(t, archsimd.Float64x4.Add, addSlice[float64]) - testInt16x16Binary(t, archsimd.Int16x16.Add, addSlice[int16]) testInt16x8Binary(t, archsimd.Int16x8.Add, addSlice[int16]) testInt32x4Binary(t, archsimd.Int32x4.Add, addSlice[int32]) - testInt32x8Binary(t, archsimd.Int32x8.Add, addSlice[int32]) testInt64x2Binary(t, archsimd.Int64x2.Add, addSlice[int64]) - testInt64x4Binary(t, archsimd.Int64x4.Add, addSlice[int64]) testInt8x16Binary(t, archsimd.Int8x16.Add, addSlice[int8]) - testInt8x32Binary(t, archsimd.Int8x32.Add, addSlice[int8]) - testUint16x16Binary(t, archsimd.Uint16x16.Add, addSlice[uint16]) - testUint16x8Binary(t, archsimd.Uint16x8.Add, addSlice[uint16]) testUint32x4Binary(t, archsimd.Uint32x4.Add, addSlice[uint32]) - testUint32x8Binary(t, archsimd.Uint32x8.Add, addSlice[uint32]) testUint64x2Binary(t, archsimd.Uint64x2.Add, addSlice[uint64]) - testUint64x4Binary(t, archsimd.Uint64x4.Add, addSlice[uint64]) + testUint16x8Binary(t, archsimd.Uint16x8.Add, addSlice[uint16]) testUint8x16Binary(t, archsimd.Uint8x16.Add, addSlice[uint8]) - testUint8x32Binary(t, archsimd.Uint8x32.Add, addSlice[uint8]) + + if archsimd.X86.AVX2() { + testUint16x16Binary(t, archsimd.Uint16x16.Add, addSlice[uint16]) + testUint32x8Binary(t, archsimd.Uint32x8.Add, addSlice[uint32]) + testUint64x4Binary(t, archsimd.Uint64x4.Add, addSlice[uint64]) + testUint8x32Binary(t, archsimd.Uint8x32.Add, addSlice[uint8]) + } + + if archsimd.X86.AVX2() { + testInt16x16Binary(t, archsimd.Int16x16.Add, addSlice[int16]) + testInt32x8Binary(t, archsimd.Int32x8.Add, addSlice[int32]) + testInt64x4Binary(t, archsimd.Int64x4.Add, addSlice[int64]) + testInt8x32Binary(t, archsimd.Int8x32.Add, addSlice[int8]) + } if archsimd.X86.AVX512() { testFloat32x16Binary(t, archsimd.Float32x16.Add, addSlice[float32]) @@ -55,23 +61,29 @@ func TestSub(t *testing.T) { testFloat64x2Binary(t, archsimd.Float64x2.Sub, subSlice[float64]) testFloat64x4Binary(t, archsimd.Float64x4.Sub, subSlice[float64]) - testInt16x16Binary(t, archsimd.Int16x16.Sub, subSlice[int16]) - testInt16x8Binary(t, archsimd.Int16x8.Sub, subSlice[int16]) testInt32x4Binary(t, archsimd.Int32x4.Sub, subSlice[int32]) - testInt32x8Binary(t, archsimd.Int32x8.Sub, subSlice[int32]) + testInt16x8Binary(t, archsimd.Int16x8.Sub, subSlice[int16]) testInt64x2Binary(t, archsimd.Int64x2.Sub, subSlice[int64]) - testInt64x4Binary(t, archsimd.Int64x4.Sub, subSlice[int64]) testInt8x16Binary(t, archsimd.Int8x16.Sub, subSlice[int8]) - testInt8x32Binary(t, archsimd.Int8x32.Sub, subSlice[int8]) - testUint16x16Binary(t, archsimd.Uint16x16.Sub, subSlice[uint16]) - testUint16x8Binary(t, archsimd.Uint16x8.Sub, subSlice[uint16]) + if archsimd.X86.AVX2() { + testInt16x16Binary(t, archsimd.Int16x16.Sub, subSlice[int16]) + testInt32x8Binary(t, archsimd.Int32x8.Sub, subSlice[int32]) + testInt64x4Binary(t, archsimd.Int64x4.Sub, subSlice[int64]) + testInt8x32Binary(t, archsimd.Int8x32.Sub, subSlice[int8]) + } + testUint32x4Binary(t, archsimd.Uint32x4.Sub, subSlice[uint32]) - testUint32x8Binary(t, archsimd.Uint32x8.Sub, subSlice[uint32]) + testUint16x8Binary(t, archsimd.Uint16x8.Sub, subSlice[uint16]) testUint64x2Binary(t, archsimd.Uint64x2.Sub, subSlice[uint64]) - testUint64x4Binary(t, archsimd.Uint64x4.Sub, subSlice[uint64]) testUint8x16Binary(t, archsimd.Uint8x16.Sub, subSlice[uint8]) - testUint8x32Binary(t, archsimd.Uint8x32.Sub, subSlice[uint8]) + + if archsimd.X86.AVX2() { + testUint16x16Binary(t, archsimd.Uint16x16.Sub, subSlice[uint16]) + testUint32x8Binary(t, archsimd.Uint32x8.Sub, subSlice[uint32]) + testUint64x4Binary(t, archsimd.Uint64x4.Sub, subSlice[uint64]) + testUint8x32Binary(t, archsimd.Uint8x32.Sub, subSlice[uint8]) + } if archsimd.X86.AVX512() { testFloat32x16Binary(t, archsimd.Float32x16.Sub, subSlice[float32]) @@ -93,10 +105,13 @@ func TestMax(t *testing.T) { // testFloat64x2Binary(t, archsimd.Float64x2.Max, maxSlice[float64]) // nan is wrong // testFloat64x4Binary(t, archsimd.Float64x4.Max, maxSlice[float64]) // nan is wrong - testInt16x16Binary(t, archsimd.Int16x16.Max, maxSlice[int16]) testInt16x8Binary(t, archsimd.Int16x8.Max, maxSlice[int16]) testInt32x4Binary(t, archsimd.Int32x4.Max, maxSlice[int32]) - testInt32x8Binary(t, archsimd.Int32x8.Max, maxSlice[int32]) + + if archsimd.X86.AVX2() { + testInt16x16Binary(t, archsimd.Int16x16.Max, maxSlice[int16]) + testInt32x8Binary(t, archsimd.Int32x8.Max, maxSlice[int32]) + } if archsimd.X86.AVX512() { testInt64x2Binary(t, archsimd.Int64x2.Max, maxSlice[int64]) @@ -104,12 +119,18 @@ func TestMax(t *testing.T) { } testInt8x16Binary(t, archsimd.Int8x16.Max, maxSlice[int8]) - testInt8x32Binary(t, archsimd.Int8x32.Max, maxSlice[int8]) - testUint16x16Binary(t, archsimd.Uint16x16.Max, maxSlice[uint16]) + if archsimd.X86.AVX2() { + testInt8x32Binary(t, archsimd.Int8x32.Max, maxSlice[int8]) + } + testUint16x8Binary(t, archsimd.Uint16x8.Max, maxSlice[uint16]) testUint32x4Binary(t, archsimd.Uint32x4.Max, maxSlice[uint32]) - testUint32x8Binary(t, archsimd.Uint32x8.Max, maxSlice[uint32]) + + if archsimd.X86.AVX2() { + testUint16x16Binary(t, archsimd.Uint16x16.Max, maxSlice[uint16]) + testUint32x8Binary(t, archsimd.Uint32x8.Max, maxSlice[uint32]) + } if archsimd.X86.AVX512() { testUint64x2Binary(t, archsimd.Uint64x2.Max, maxSlice[uint64]) @@ -117,7 +138,10 @@ func TestMax(t *testing.T) { } testUint8x16Binary(t, archsimd.Uint8x16.Max, maxSlice[uint8]) - testUint8x32Binary(t, archsimd.Uint8x32.Max, maxSlice[uint8]) + + if archsimd.X86.AVX2() { + testUint8x32Binary(t, archsimd.Uint8x32.Max, maxSlice[uint8]) + } if archsimd.X86.AVX512() { // testFloat32x16Binary(t, archsimd.Float32x16.Max, maxSlice[float32]) // nan is wrong @@ -139,10 +163,13 @@ func TestMin(t *testing.T) { // testFloat64x2Binary(t, archsimd.Float64x2.Min, minSlice[float64]) // nan is wrong // testFloat64x4Binary(t, archsimd.Float64x4.Min, minSlice[float64]) // nan is wrong - testInt16x16Binary(t, archsimd.Int16x16.Min, minSlice[int16]) testInt16x8Binary(t, archsimd.Int16x8.Min, minSlice[int16]) testInt32x4Binary(t, archsimd.Int32x4.Min, minSlice[int32]) - testInt32x8Binary(t, archsimd.Int32x8.Min, minSlice[int32]) + + if archsimd.X86.AVX2() { + testInt16x16Binary(t, archsimd.Int16x16.Min, minSlice[int16]) + testInt32x8Binary(t, archsimd.Int32x8.Min, minSlice[int32]) + } if archsimd.X86.AVX512() { testInt64x2Binary(t, archsimd.Int64x2.Min, minSlice[int64]) @@ -150,12 +177,18 @@ func TestMin(t *testing.T) { } testInt8x16Binary(t, archsimd.Int8x16.Min, minSlice[int8]) - testInt8x32Binary(t, archsimd.Int8x32.Min, minSlice[int8]) - testUint16x16Binary(t, archsimd.Uint16x16.Min, minSlice[uint16]) + if archsimd.X86.AVX2() { + testInt8x32Binary(t, archsimd.Int8x32.Min, minSlice[int8]) + } + testUint16x8Binary(t, archsimd.Uint16x8.Min, minSlice[uint16]) testUint32x4Binary(t, archsimd.Uint32x4.Min, minSlice[uint32]) - testUint32x8Binary(t, archsimd.Uint32x8.Min, minSlice[uint32]) + + if archsimd.X86.AVX2() { + testUint16x16Binary(t, archsimd.Uint16x16.Min, minSlice[uint16]) + testUint32x8Binary(t, archsimd.Uint32x8.Min, minSlice[uint32]) + } if archsimd.X86.AVX512() { testUint64x2Binary(t, archsimd.Uint64x2.Min, minSlice[uint64]) @@ -163,7 +196,10 @@ func TestMin(t *testing.T) { } testUint8x16Binary(t, archsimd.Uint8x16.Min, minSlice[uint8]) - testUint8x32Binary(t, archsimd.Uint8x32.Min, minSlice[uint8]) + + if archsimd.X86.AVX2() { + testUint8x32Binary(t, archsimd.Uint8x32.Min, minSlice[uint8]) + } if archsimd.X86.AVX512() { // testFloat32x16Binary(t, archsimd.Float32x16.Min, minSlice[float32]) // nan is wrong @@ -180,23 +216,29 @@ func TestMin(t *testing.T) { } func TestAnd(t *testing.T) { - testInt16x16Binary(t, archsimd.Int16x16.And, andSlice[int16]) testInt16x8Binary(t, archsimd.Int16x8.And, andSlice[int16]) testInt32x4Binary(t, archsimd.Int32x4.And, andSlice[int32]) - testInt32x8Binary(t, archsimd.Int32x8.And, andSlice[int32]) testInt64x2Binary(t, archsimd.Int64x2.And, andSlice[int64]) - testInt64x4Binary(t, archsimd.Int64x4.And, andSlice[int64]) testInt8x16Binary(t, archsimd.Int8x16.And, andSlice[int8]) - testInt8x32Binary(t, archsimd.Int8x32.And, andSlice[int8]) - testUint16x16Binary(t, archsimd.Uint16x16.And, andSlice[uint16]) + if archsimd.X86.AVX2() { + testInt16x16Binary(t, archsimd.Int16x16.And, andSlice[int16]) + testInt32x8Binary(t, archsimd.Int32x8.And, andSlice[int32]) + testInt64x4Binary(t, archsimd.Int64x4.And, andSlice[int64]) + testInt8x32Binary(t, archsimd.Int8x32.And, andSlice[int8]) + } + testUint16x8Binary(t, archsimd.Uint16x8.And, andSlice[uint16]) testUint32x4Binary(t, archsimd.Uint32x4.And, andSlice[uint32]) - testUint32x8Binary(t, archsimd.Uint32x8.And, andSlice[uint32]) testUint64x2Binary(t, archsimd.Uint64x2.And, andSlice[uint64]) - testUint64x4Binary(t, archsimd.Uint64x4.And, andSlice[uint64]) testUint8x16Binary(t, archsimd.Uint8x16.And, andSlice[uint8]) - testUint8x32Binary(t, archsimd.Uint8x32.And, andSlice[uint8]) + + if archsimd.X86.AVX2() { + testUint16x16Binary(t, archsimd.Uint16x16.And, andSlice[uint16]) + testUint32x8Binary(t, archsimd.Uint32x8.And, andSlice[uint32]) + testUint64x4Binary(t, archsimd.Uint64x4.And, andSlice[uint64]) + testUint8x32Binary(t, archsimd.Uint8x32.And, andSlice[uint8]) + } if archsimd.X86.AVX512() { // testInt8x64Binary(t, archsimd.Int8x64.And, andISlice[int8]) // missing @@ -211,23 +253,29 @@ func TestAnd(t *testing.T) { } func TestAndNot(t *testing.T) { - testInt16x16Binary(t, archsimd.Int16x16.AndNot, andNotSlice[int16]) testInt16x8Binary(t, archsimd.Int16x8.AndNot, andNotSlice[int16]) testInt32x4Binary(t, archsimd.Int32x4.AndNot, andNotSlice[int32]) - testInt32x8Binary(t, archsimd.Int32x8.AndNot, andNotSlice[int32]) testInt64x2Binary(t, archsimd.Int64x2.AndNot, andNotSlice[int64]) - testInt64x4Binary(t, archsimd.Int64x4.AndNot, andNotSlice[int64]) testInt8x16Binary(t, archsimd.Int8x16.AndNot, andNotSlice[int8]) - testInt8x32Binary(t, archsimd.Int8x32.AndNot, andNotSlice[int8]) - testUint16x16Binary(t, archsimd.Uint16x16.AndNot, andNotSlice[uint16]) + if archsimd.X86.AVX2() { + testInt16x16Binary(t, archsimd.Int16x16.AndNot, andNotSlice[int16]) + testInt32x8Binary(t, archsimd.Int32x8.AndNot, andNotSlice[int32]) + testInt64x4Binary(t, archsimd.Int64x4.AndNot, andNotSlice[int64]) + testInt8x32Binary(t, archsimd.Int8x32.AndNot, andNotSlice[int8]) + } + + testUint8x16Binary(t, archsimd.Uint8x16.AndNot, andNotSlice[uint8]) testUint16x8Binary(t, archsimd.Uint16x8.AndNot, andNotSlice[uint16]) testUint32x4Binary(t, archsimd.Uint32x4.AndNot, andNotSlice[uint32]) - testUint32x8Binary(t, archsimd.Uint32x8.AndNot, andNotSlice[uint32]) testUint64x2Binary(t, archsimd.Uint64x2.AndNot, andNotSlice[uint64]) - testUint64x4Binary(t, archsimd.Uint64x4.AndNot, andNotSlice[uint64]) - testUint8x16Binary(t, archsimd.Uint8x16.AndNot, andNotSlice[uint8]) - testUint8x32Binary(t, archsimd.Uint8x32.AndNot, andNotSlice[uint8]) + + if archsimd.X86.AVX2() { + testUint16x16Binary(t, archsimd.Uint16x16.AndNot, andNotSlice[uint16]) + testUint32x8Binary(t, archsimd.Uint32x8.AndNot, andNotSlice[uint32]) + testUint64x4Binary(t, archsimd.Uint64x4.AndNot, andNotSlice[uint64]) + testUint8x32Binary(t, archsimd.Uint8x32.AndNot, andNotSlice[uint8]) + } if archsimd.X86.AVX512() { testInt8x64Binary(t, archsimd.Int8x64.AndNot, andNotSlice[int8]) @@ -242,23 +290,29 @@ func TestAndNot(t *testing.T) { } func TestXor(t *testing.T) { - testInt16x16Binary(t, archsimd.Int16x16.Xor, xorSlice[int16]) testInt16x8Binary(t, archsimd.Int16x8.Xor, xorSlice[int16]) testInt32x4Binary(t, archsimd.Int32x4.Xor, xorSlice[int32]) - testInt32x8Binary(t, archsimd.Int32x8.Xor, xorSlice[int32]) testInt64x2Binary(t, archsimd.Int64x2.Xor, xorSlice[int64]) - testInt64x4Binary(t, archsimd.Int64x4.Xor, xorSlice[int64]) testInt8x16Binary(t, archsimd.Int8x16.Xor, xorSlice[int8]) - testInt8x32Binary(t, archsimd.Int8x32.Xor, xorSlice[int8]) - testUint16x16Binary(t, archsimd.Uint16x16.Xor, xorSlice[uint16]) + if archsimd.X86.AVX2() { + testInt16x16Binary(t, archsimd.Int16x16.Xor, xorSlice[int16]) + testInt32x8Binary(t, archsimd.Int32x8.Xor, xorSlice[int32]) + testInt64x4Binary(t, archsimd.Int64x4.Xor, xorSlice[int64]) + testInt8x32Binary(t, archsimd.Int8x32.Xor, xorSlice[int8]) + } + testUint16x8Binary(t, archsimd.Uint16x8.Xor, xorSlice[uint16]) testUint32x4Binary(t, archsimd.Uint32x4.Xor, xorSlice[uint32]) - testUint32x8Binary(t, archsimd.Uint32x8.Xor, xorSlice[uint32]) testUint64x2Binary(t, archsimd.Uint64x2.Xor, xorSlice[uint64]) - testUint64x4Binary(t, archsimd.Uint64x4.Xor, xorSlice[uint64]) testUint8x16Binary(t, archsimd.Uint8x16.Xor, xorSlice[uint8]) - testUint8x32Binary(t, archsimd.Uint8x32.Xor, xorSlice[uint8]) + + if archsimd.X86.AVX2() { + testUint16x16Binary(t, archsimd.Uint16x16.Xor, xorSlice[uint16]) + testUint32x8Binary(t, archsimd.Uint32x8.Xor, xorSlice[uint32]) + testUint64x4Binary(t, archsimd.Uint64x4.Xor, xorSlice[uint64]) + testUint8x32Binary(t, archsimd.Uint8x32.Xor, xorSlice[uint8]) + } if archsimd.X86.AVX512() { // testInt8x64Binary(t, archsimd.Int8x64.Xor, andISlice[int8]) // missing @@ -273,23 +327,29 @@ func TestXor(t *testing.T) { } func TestOr(t *testing.T) { - testInt16x16Binary(t, archsimd.Int16x16.Or, orSlice[int16]) testInt16x8Binary(t, archsimd.Int16x8.Or, orSlice[int16]) testInt32x4Binary(t, archsimd.Int32x4.Or, orSlice[int32]) - testInt32x8Binary(t, archsimd.Int32x8.Or, orSlice[int32]) testInt64x2Binary(t, archsimd.Int64x2.Or, orSlice[int64]) - testInt64x4Binary(t, archsimd.Int64x4.Or, orSlice[int64]) testInt8x16Binary(t, archsimd.Int8x16.Or, orSlice[int8]) - testInt8x32Binary(t, archsimd.Int8x32.Or, orSlice[int8]) - testUint16x16Binary(t, archsimd.Uint16x16.Or, orSlice[uint16]) + if archsimd.X86.AVX2() { + testInt16x16Binary(t, archsimd.Int16x16.Or, orSlice[int16]) + testInt32x8Binary(t, archsimd.Int32x8.Or, orSlice[int32]) + testInt64x4Binary(t, archsimd.Int64x4.Or, orSlice[int64]) + testInt8x32Binary(t, archsimd.Int8x32.Or, orSlice[int8]) + } + testUint16x8Binary(t, archsimd.Uint16x8.Or, orSlice[uint16]) testUint32x4Binary(t, archsimd.Uint32x4.Or, orSlice[uint32]) - testUint32x8Binary(t, archsimd.Uint32x8.Or, orSlice[uint32]) testUint64x2Binary(t, archsimd.Uint64x2.Or, orSlice[uint64]) - testUint64x4Binary(t, archsimd.Uint64x4.Or, orSlice[uint64]) testUint8x16Binary(t, archsimd.Uint8x16.Or, orSlice[uint8]) - testUint8x32Binary(t, archsimd.Uint8x32.Or, orSlice[uint8]) + + if archsimd.X86.AVX2() { + testUint16x16Binary(t, archsimd.Uint16x16.Or, orSlice[uint16]) + testUint32x8Binary(t, archsimd.Uint32x8.Or, orSlice[uint32]) + testUint64x4Binary(t, archsimd.Uint64x4.Or, orSlice[uint64]) + testUint8x32Binary(t, archsimd.Uint8x32.Or, orSlice[uint8]) + } if archsimd.X86.AVX512() { // testInt8x64Binary(t, archsimd.Int8x64.Or, andISlice[int8]) // missing @@ -309,10 +369,13 @@ func TestMul(t *testing.T) { testFloat64x2Binary(t, archsimd.Float64x2.Mul, mulSlice[float64]) testFloat64x4Binary(t, archsimd.Float64x4.Mul, mulSlice[float64]) - testInt16x16Binary(t, archsimd.Int16x16.Mul, mulSlice[int16]) testInt16x8Binary(t, archsimd.Int16x8.Mul, mulSlice[int16]) testInt32x4Binary(t, archsimd.Int32x4.Mul, mulSlice[int32]) - testInt32x8Binary(t, archsimd.Int32x8.Mul, mulSlice[int32]) + + if archsimd.X86.AVX2() { + testInt16x16Binary(t, archsimd.Int16x16.Mul, mulSlice[int16]) + testInt32x8Binary(t, archsimd.Int32x8.Mul, mulSlice[int32]) + } // testInt8x16Binary(t, archsimd.Int8x16.Mul, mulSlice[int8]) // nope // testInt8x32Binary(t, archsimd.Int8x32.Mul, mulSlice[int8]) -- cgit v1.3