feat(amd64): encode the AVX-512 and BMI corpus families
Assisted-by: GLM 5.3 Flash
This commit is contained in:
@@ -721,3 +721,92 @@ func hexCompact(b []byte) string {
|
||||
}
|
||||
return string(out)
|
||||
}
|
||||
|
||||
// TestAvx512CorpusFamilies pins representative encodings of the AVX-512
|
||||
// families the toolchain's avx512enc corpus exercises: the bytes are the
|
||||
// go tool asm output for exactly these operands, and the same families are
|
||||
// covered end to end by the avx512_amd64.s differential kernel.
|
||||
func TestAvx512CorpusFamilies(t *testing.T) {
|
||||
vsib := func(base, idx string, scale int) Operand {
|
||||
return Idx(vreg(t, base), vreg(t, idx), scale, 0, 0)
|
||||
}
|
||||
cases := []struct {
|
||||
name string
|
||||
mnem string
|
||||
ops []Operand
|
||||
want string
|
||||
}{
|
||||
// AES rounds (EVEX NDS, VEX twin routed by operand width).
|
||||
{"VAESDEC Z", "VAESDEC", []Operand{vreg(t, "Z1"), vreg(t, "Z2"), vreg(t, "Z3")}, "62f26d48ded9"},
|
||||
// Integer VNNI and the bit algorithm group.
|
||||
{"VPDPBUSD", "VPDPBUSD", []Operand{vreg(t, "Z1"), vreg(t, "Z2"), vreg(t, "K2"), vreg(t, "Z3")}, "62f26d4a50d9"},
|
||||
{"VPOPCNTW", "VPOPCNTW", []Operand{vreg(t, "Z1"), vreg(t, "K3"), vreg(t, "Z2")}, "62f2fd4b54d1"},
|
||||
{"VPCONFLICTD", "VPCONFLICTD", []Operand{vreg(t, "Z1"), vreg(t, "K1"), vreg(t, "Z2")}, "62f27d49c4d1"},
|
||||
{"VPLZCNTQ masked", "VPLZCNTQ", []Operand{vreg(t, "Z7"), vreg(t, "K1"), vreg(t, "Z8")}, "6272fd4944c7"},
|
||||
{"VPERMT2B", "VPERMT2B", []Operand{vreg(t, "Z1"), vreg(t, "Z2"), vreg(t, "K1"), vreg(t, "Z3")}, "62f26d497dd9"},
|
||||
{"VPMULTISHIFTQB", "VPMULTISHIFTQB", []Operand{vreg(t, "Z1"), vreg(t, "Z2"), vreg(t, "K3"), vreg(t, "Z4")}, "62f2ed4b83e1"},
|
||||
{"VDBPSADBW", "VDBPSADBW", []Operand{Imm(3), vreg(t, "Z1"), vreg(t, "Z2"), vreg(t, "K3"), vreg(t, "Z3")}, "62f36d4b42d903"},
|
||||
{"VPSHUFBITQMB", "VPSHUFBITQMB", []Operand{vreg(t, "Z9"), vreg(t, "Z10"), vreg(t, "K3")}, "62d22d488fd9"},
|
||||
{"VPTESTNMQ", "VPTESTNMQ", []Operand{vreg(t, "Z13"), vreg(t, "Z14"), vreg(t, "K5")}, "62d28e4827ed"},
|
||||
// Permutations: immediate and register counts.
|
||||
{"VALIGNQ", "VALIGNQ", []Operand{Imm(3), vreg(t, "Z1"), vreg(t, "Z2"), vreg(t, "K1"), vreg(t, "Z3")}, "62f3ed4903d903"},
|
||||
{"VPERMQ imm", "VPERMQ", []Operand{Imm(1), vreg(t, "Z1"), vreg(t, "K2"), vreg(t, "Z2")}, "62f3fd4a00d101"},
|
||||
{"VPERMQ reg", "VPERMQ", []Operand{vreg(t, "Z3"), vreg(t, "Z4"), vreg(t, "K2"), vreg(t, "Z5")}, "62f2dd4a36eb"},
|
||||
{"VPERMPD reg", "VPERMPD", []Operand{vreg(t, "Z1"), vreg(t, "Z2"), vreg(t, "Z3")}, "62f2ed4816d9"},
|
||||
{"VPERMILPS imm", "VPERMILPS", []Operand{Imm(5), vreg(t, "Z9"), vreg(t, "K2"), vreg(t, "Z10")}, "62537d4a04d105"},
|
||||
{"VPERMILPS reg", "VPERMILPS", []Operand{vreg(t, "Z11"), vreg(t, "Z12"), vreg(t, "K2"), vreg(t, "Z13")}, "62521d4a0ceb"},
|
||||
// Shifts: immediate, register-count and memory-count forms; the
|
||||
// count source carries its own XMM tuple width.
|
||||
{"VPSLLW imm mask", "VPSLLW", []Operand{Imm(3), vreg(t, "Z1"), vreg(t, "K2"), vreg(t, "Z2")}, "62f16d4a71f103"},
|
||||
{"VPSLLD reg count", "VPSLLD", []Operand{vreg(t, "X1"), vreg(t, "Z2"), vreg(t, "K1"), vreg(t, "Z3")}, "62f16d49f2d9"},
|
||||
{"VPSLLDQ", "VPSLLDQ", []Operand{Imm(9), vreg(t, "Z7"), vreg(t, "Z8")}, "62f13d4873ff09"},
|
||||
{"VPSRLDQ mem", "VPSRLDQ", []Operand{Imm(11), Ptr(SI, 16, 16), vreg(t, "Z4")}, "62f15d48739e100000000b"},
|
||||
{"VPSRLVW", "VPSRLVW", []Operand{vreg(t, "Z3"), vreg(t, "Z4"), vreg(t, "K1"), vreg(t, "Z5")}, "62f2dd4910eb"},
|
||||
// Conversions and shuffles with the F2 prefix and no prefix.
|
||||
{"VCVTUDQ2PS", "VCVTUDQ2PS", []Operand{vreg(t, "Z1"), vreg(t, "K1"), vreg(t, "Z2")}, "62f17f497ad1"},
|
||||
{"VSHUFPS", "VSHUFPS", []Operand{Imm(2), vreg(t, "Z4"), vreg(t, "Z5"), vreg(t, "K1"), vreg(t, "Z6")}, "62f15449c6f402"},
|
||||
// Gather and scatter prefetch hints (memory-only, /digit in reg).
|
||||
{"VGATHERPF0DPD", "VGATHERPF0DPD", []Operand{vreg(t, "K5"), vsib("R10", "Y29", 8)}, "6292fd45c60cea"},
|
||||
{"VSCATTERPF1DPS", "VSCATTERPF1DPS", []Operand{vreg(t, "K2"), vsib("R10", "Z28", 4)}, "62927d42c634a2"},
|
||||
// Opmask broadcasts and the K logic.
|
||||
{"VPBROADCASTMB2Q", "VPBROADCASTMB2Q", []Operand{vreg(t, "K1"), vreg(t, "Z2")}, "62f2fe482ad1"},
|
||||
{"VPBROADCASTMW2D", "VPBROADCASTMW2D", []Operand{vreg(t, "K3"), vreg(t, "Z4")}, "62f27e483ae3"},
|
||||
{"KUNPCKWD", "KUNPCKWD", []Operand{vreg(t, "K6"), vreg(t, "K4"), vreg(t, "K1")}, "c5dc4bce"},
|
||||
{"KADDB", "KADDB", []Operand{vreg(t, "K2"), vreg(t, "K3"), vreg(t, "K5")}, "c5e54aea"},
|
||||
// Lane extracts to general registers (EVEX and VEX routes).
|
||||
{"VPEXTRB", "VPEXTRB", []Operand{Imm(3), vreg(t, "X26"), AX}, "62637d0814d003"},
|
||||
{"VPEXTRD", "VPEXTRD", []Operand{Imm(1), vreg(t, "X26"), vreg(t, "R9")}, "62437d0816d101"},
|
||||
{"VPEXTRD vex", "VPEXTRD", []Operand{Imm(1), vreg(t, "X2"), DI}, "c4e37916d701"},
|
||||
{"VPINSRQ", "VPINSRQ", []Operand{Imm(1), DI, vreg(t, "X3"), vreg(t, "X4")}, "c4e3e122e701"},
|
||||
// Moves: masked unaligned, masked scalar register form, half moves
|
||||
// and non-temporal stores.
|
||||
{"VMOVUPS mask", "VMOVUPS", []Operand{vreg(t, "Z1"), vreg(t, "K2"), vreg(t, "Z3")}, "62f17c4a11cb"},
|
||||
{"VMOVSD 3op", "VMOVSD", []Operand{vreg(t, "X14"), vreg(t, "X5"), vreg(t, "K3"), vreg(t, "X22")}, "6231d70b11f6"},
|
||||
{"VMOVSS 3op", "VMOVSS", []Operand{vreg(t, "X18"), vreg(t, "X3"), vreg(t, "K2"), vreg(t, "X25")}, "6281660a11d1"},
|
||||
{"VMOVHPS insert", "VMOVHPS", []Operand{Ptr(SI, 0, 8), vreg(t, "X18"), vreg(t, "X19")}, "62e16c00161e"},
|
||||
{"VMOVHPS store", "VMOVHPS", []Operand{vreg(t, "X20"), Ptr(SI, 8, 8)}, "62e17c08176601"},
|
||||
{"VMOVLHPS", "VMOVLHPS", []Operand{vreg(t, "X16"), vreg(t, "X5"), vreg(t, "X17")}, "62a1540816c8"},
|
||||
{"VMOVNTDQ", "VMOVNTDQ", []Operand{vreg(t, "Z7"), Ptr(SI, 0, 64)}, "62f17d48e73e"},
|
||||
{"VMOVNTDQA", "VMOVNTDQA", []Operand{Ptr(SI, 64, 64), vreg(t, "Z8")}, "62727d482a4601"},
|
||||
{"VMOVNTPS", "VMOVNTPS", []Operand{vreg(t, "Z9"), Ptr(SI, 0, 64)}, "62717c482b0e"},
|
||||
// Scalar compares with and without the 66 prefix.
|
||||
{"VCOMISD", "VCOMISD", []Operand{vreg(t, "X5"), vreg(t, "X6")}, "c5f92ff5"},
|
||||
{"VUCOMISS", "VUCOMISS", []Operand{vreg(t, "X7"), vreg(t, "X8")}, "c5782ec7"},
|
||||
// Floating point helpers.
|
||||
{"VSQRTSD", "VSQRTSD", []Operand{vreg(t, "X1"), vreg(t, "X2"), vreg(t, "K1"), vreg(t, "X3")}, "62f1ef0951d9"},
|
||||
{"VEXP2PD", "VEXP2PD", []Operand{vreg(t, "Z5"), vreg(t, "K1"), vreg(t, "Z6")}, "62f2fd49c8f5"},
|
||||
{"VRCP28SD", "VRCP28SD", []Operand{vreg(t, "X9"), vreg(t, "X8"), vreg(t, "K1"), vreg(t, "X10")}, "6252bd09cbd1"},
|
||||
{"VBROADCASTF32X2", "VBROADCASTF32X2", []Operand{vreg(t, "X1"), vreg(t, "K1"), vreg(t, "Z2")}, "62f27d4919d1"},
|
||||
{"VPCOMPRESSB", "VPCOMPRESSB", []Operand{vreg(t, "Z1"), vreg(t, "K1"), Ptr(SI, 0, 64)}, "62f27d49630e"},
|
||||
}
|
||||
for _, c := range cases {
|
||||
code, err := Encode(c.mnem, c.ops...)
|
||||
if err != nil {
|
||||
t.Errorf("%s: Encode: %v", c.name, err)
|
||||
continue
|
||||
}
|
||||
if got := hexCompact(code); got != c.want {
|
||||
t.Errorf("%s: got %s, want %s", c.name, got, c.want)
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user