// Define a list of AMD64 microarchitecture level features
// https://en.wikipedia.org/wiki/X86-64#Microarchitecture_levels

                     // SSE3     SSSE3    CMPXCHNG16 SSE4.1    SSE4.2    POPCNT
#define V2_FEATURES_CX (1 << 0 | 1 << 9 | 1 << 13  | 1 << 19 | 1 << 20 | 1 << 23)
                         // LAHF/SAHF
#define V2_EXT_FEATURES_CX (1 << 0)

	{APMULLW, ymm, Py1, opBytes{0xd5, Pe, 0xd5}},
	{APMULULQ, ymm, Py1, opBytes{0xf4, Pe, 0xf4}},
	{APOPAL, ynone, P32, opBytes{0x61}},
	{APOPAW, ynone, Pe, opBytes{0x61}},
	{APOPCNTW, yml_rl, Pef3, opBytes{0xb8}},
	{APOPCNTL, yml_rl, Pf3, opBytes{0xb8}},
	{APOPCNTQ, yml_rl, Pfw, opBytes{0xb8}},
	{APOPFL, ynone, P32, opBytes{0x9d}},
	{APOPFQ, ynone, Py, opBytes{0x9d}},
	{APOPFW, ynone, Pe, opBytes{0x9d}},

	}
	return false
}
func rewriteValueS390X_OpPopCount16(v *Value) bool {
	v_0 := v.Args[0]
	b := v.Block
	typ := &b.Func.Config.Types
	// match: (PopCount16 x)
	// result: (MOVBZreg (SumBytes2 (POPCNT <typ.UInt16> x)))
	for {
		x := v_0
		v.reset(OpS390XMOVBZreg)
		v0 := b.NewValue0(v.Pos, OpS390XSumBytes2, typ.UInt8)
		v1 := b.NewValue0(v.Pos, OpS390XPOPCNT, typ.UInt16)
		v1.AddArg(x)
		v0.AddArg(v1)

// CHECK:       %[[POP_CNT:.*]] = "tf.PopulationCount"(%[[ARG_0]]) : (tensor<8xi32>) -> tensor<8xui8>
// CHECK:       %[[RES:.*]] = "tf.Cast"(%[[POP_CNT]]) <{Truncate = false}> : (tensor<8xui8>) -> tensor<8xi32>
// CHECK:       return %[[RES]]
// CHECK:         }
func.func @convert_population_count_i32(%arg0: tensor<8xi32>) -> tensor<8xi32> {
  %0 = "mhlo.popcnt"(%arg0) : (tensor<8xi32>) -> tensor<8xi32>

	// load on condition
	{i: 25, as: ALOCGR, a1: C_SCON, a2: C_REG, a6: C_REG},

	// find leftmost one
	{i: 8, as: AFLOGR, a1: C_REG, a6: C_REG},

	// population count
	{i: 9, as: APOPCNT, a1: C_REG, a6: C_REG},

	// compare
	{i: 70, as: ACMP, a1: C_REG, a6: C_REG},
	{i: 71, as: ACMP, a1: C_REG, a6: C_LCON},
	{i: 70, as: ACMPU, a1: C_REG, a6: C_REG},
	{i: 71, as: ACMPU, a1: C_REG, a6: C_LCON},

			bFalse := s.f.NewBlock(ssa.BlockPlain)
			bEnd := s.f.NewBlock(ssa.BlockPlain)
			b.AddEdgeTo(bTrue)
			b.AddEdgeTo(bFalse)
			b.Likely = ssa.BranchLikely // most machines have popcnt nowadays

			// We have the intrinsic - use it directly.
			s.startBlock(bTrue)
			s.vars[n] = s.newValue1(op, types.Types[types.TINT], args[0])
			s.endBlock().AddEdgeTo(bEnd)

	}
	return false
}
func rewriteValueAMD64_OpPopCount16(v *Value) bool {
	v_0 := v.Args[0]
	b := v.Block
	typ := &b.Func.Config.Types
	// match: (PopCount16 x)
	// result: (POPCNTL (MOVWQZX <typ.UInt32> x))
	for {
		x := v_0
		v.reset(OpAMD64POPCNTL)
		v0 := b.NewValue0(v.Pos, OpAMD64MOVWQZX, typ.UInt32)
		v0.AddArg(x)
		v.AddArg(v0)
		return true
	}
}