1 // Copyright ©2016 The Gonum Authors. All rights reserved.
2 // Use of this source code is governed by a BSD-style
3 // license that can be found in the LICENSE file.
5 //+build !noasm,!appengine
9 #define MOVDDUP_XPTR_IDX_8__X3 LONG $0x1C120FF2; BYTE $0xC6 // MOVDDUP (SI)(AX*8), X3
10 #define MOVDDUP_16_XPTR_IDX_8__X5 LONG $0x6C120FF2; WORD $0x10C6 // MOVDDUP 16(SI)(AX*8), X5
11 #define MOVDDUP_32_XPTR_IDX_8__X7 LONG $0x7C120FF2; WORD $0x20C6 // MOVDDUP 32(SI)(AX*8), X7
12 #define MOVDDUP_48_XPTR_IDX_8__X9 LONG $0x120F44F2; WORD $0xC64C; BYTE $0x30 // MOVDDUP 48(SI)(AX*8), X9
14 #define MOVDDUP_XPTR_IIDX_8__X2 LONG $0x14120FF2; BYTE $0xD6 // MOVDDUP (SI)(DX*8), X2
15 #define MOVDDUP_16_XPTR_IIDX_8__X4 LONG $0x64120FF2; WORD $0x10D6 // MOVDDUP 16(SI)(DX*8), X4
16 #define MOVDDUP_32_XPTR_IIDX_8__X6 LONG $0x74120FF2; WORD $0x20D6 // MOVDDUP 32(SI)(DX*8), X6
17 #define MOVDDUP_48_XPTR_IIDX_8__X8 LONG $0x120F44F2; WORD $0xD644; BYTE $0x30 // MOVDDUP 48(SI)(DX*8), X8
19 #define ADDSUBPD_X2_X3 LONG $0xDAD00F66 // ADDSUBPD X2, X3
20 #define ADDSUBPD_X4_X5 LONG $0xECD00F66 // ADDSUBPD X4, X5
21 #define ADDSUBPD_X6_X7 LONG $0xFED00F66 // ADDSUBPD X6, X7
22 #define ADDSUBPD_X8_X9 LONG $0xD00F4566; BYTE $0xC8 // ADDSUBPD X8, X9
35 // func DotcUnitary(x, y []complex128) (sum complex128)
36 TEXT ·DotcUnitary(SB), NOSPLIT, $0
37 MOVQ x_base+0(FP), X_PTR // X_PTR = &x
38 MOVQ y_base+24(FP), Y_PTR // Y_PTR = &y
39 MOVQ x_len+8(FP), LEN // LEN = min( len(x), len(y) )
40 CMPQ y_len+32(FP), LEN
41 CMOVQLE y_len+32(FP), LEN
42 PXOR SUM, SUM // sum = 0
43 CMPQ LEN, $0 // if LEN == 0 { return }
45 XORPS P_SUM, P_SUM // psum = 0
47 SHUFPD $0, NEG1, NEG1 // { -1, -1 }
48 XORQ IDX, IDX // i := 0
49 MOVQ $1, I_IDX // j := 1
51 ANDQ $3, TAIL // TAIL = floor( TAIL / 4 )
52 SHRQ $2, LEN // LEN = TAIL % 4
53 JZ dot_tail // if LEN == 0 { goto dot_tail }
55 MOVAPS NEG1, P_NEG1 // Copy NEG1 to P_NEG1 for pipelining
58 MOVDDUP_XPTR_IDX_8__X3 // X_(i+1) = { real(x[i], real(x[i]) }
59 MOVDDUP_16_XPTR_IDX_8__X5
60 MOVDDUP_32_XPTR_IDX_8__X7
61 MOVDDUP_48_XPTR_IDX_8__X9
63 MOVDDUP_XPTR_IIDX_8__X2 // X_i = { imag(x[i]), imag(x[i]) }
64 MOVDDUP_16_XPTR_IIDX_8__X4
65 MOVDDUP_32_XPTR_IIDX_8__X6
66 MOVDDUP_48_XPTR_IIDX_8__X8
68 // X_i = { -imag(x[i]), -imag(x[i]) }
74 // X_j = { imag(y[i]), real(y[i]) }
75 MOVUPS (Y_PTR)(IDX*8), X10
76 MOVUPS 16(Y_PTR)(IDX*8), X11
77 MOVUPS 32(Y_PTR)(IDX*8), X12
78 MOVUPS 48(Y_PTR)(IDX*8), X13
80 // X_(i+1) = { imag(a) * real(x[i]), real(a) * real(x[i]) }
86 // X_j = { real(y[i]), imag(y[i]) }
92 // X_i = { real(a) * imag(x[i]), imag(a) * imag(x[i]) }
99 // imag(result[i]): imag(a)*real(x[i]) + real(a)*imag(x[i]),
100 // real(result[i]): real(a)*real(x[i]) - imag(a)*imag(x[i])
113 ADDQ $8, IDX // IDX += 8
114 ADDQ $8, I_IDX // I_IDX += 8
116 JNZ dot_loop // } while --LEN > 0
117 ADDPD P_SUM, SUM // sum += psum
118 CMPQ TAIL, $0 // if TAIL == 0 { return }
122 MOVDDUP_XPTR_IDX_8__X3 // X_(i+1) = { real(x[i]) , real(x[i]) }
123 MOVDDUP_XPTR_IIDX_8__X2 // X_i = { imag(x[i]) , imag(x[i]) }
124 MULPD NEG1, X2 // X_i = { -imag(x[i]) , -imag(x[i]) }
125 MOVUPS (Y_PTR)(IDX*8), X10 // X_j = { imag(y[i]) , real(y[i]) }
126 MULPD X10, X3 // X_(i+1) = { imag(a) * real(x[i]), real(a) * real(x[i]) }
127 SHUFPD $0x1, X10, X10 // X_j = { real(y[i]) , imag(y[i]) }
128 MULPD X10, X2 // X_i = { real(a) * imag(x[i]), imag(a) * imag(x[i]) }
131 // imag(result[i]): imag(a)*real(x[i]) + real(a)*imag(x[i]),
132 // real(result[i]): real(a)*real(x[i]) - imag(a)*imag(x[i])
135 ADDPD X3, SUM // SUM += result[i]
136 ADDQ $2, IDX // IDX += 2
137 ADDQ $2, I_IDX // I_IDX += 2
139 JNZ dot_tail // } while --TAIL > 0
142 MOVUPS SUM, sum+48(FP)