1 // Copyright ©2017 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 MOVSLDUP_XPTR_IDX_8__X3 LONG $0x1C120FF3; BYTE $0xC6 // MOVSLDUP (SI)(AX*8), X3
10 #define MOVSLDUP_16_XPTR_IDX_8__X5 LONG $0x6C120FF3; WORD $0x10C6 // MOVSLDUP 16(SI)(AX*8), X5
11 #define MOVSLDUP_32_XPTR_IDX_8__X7 LONG $0x7C120FF3; WORD $0x20C6 // MOVSLDUP 32(SI)(AX*8), X7
12 #define MOVSLDUP_48_XPTR_IDX_8__X9 LONG $0x120F44F3; WORD $0xC64C; BYTE $0x30 // MOVSLDUP 48(SI)(AX*8), X9
14 #define MOVSHDUP_XPTR_IDX_8__X2 LONG $0x14160FF3; BYTE $0xC6 // MOVSHDUP (SI)(AX*8), X2
15 #define MOVSHDUP_16_XPTR_IDX_8__X4 LONG $0x64160FF3; WORD $0x10C6 // MOVSHDUP 16(SI)(AX*8), X4
16 #define MOVSHDUP_32_XPTR_IDX_8__X6 LONG $0x74160FF3; WORD $0x20C6 // MOVSHDUP 32(SI)(AX*8), X6
17 #define MOVSHDUP_48_XPTR_IDX_8__X8 LONG $0x160F44F3; WORD $0xC644; BYTE $0x30 // MOVSHDUP 48(SI)(AX*8), X8
19 #define MOVSHDUP_X3_X2 LONG $0xD3160FF3 // MOVSHDUP X3, X2
20 #define MOVSLDUP_X3_X3 LONG $0xDB120FF3 // MOVSLDUP X3, X3
22 #define ADDSUBPS_X2_X3 LONG $0xDAD00FF2 // ADDSUBPS X2, X3
23 #define ADDSUBPS_X4_X5 LONG $0xECD00FF2 // ADDSUBPS X4, X5
24 #define ADDSUBPS_X6_X7 LONG $0xFED00FF2 // ADDSUBPS X6, X7
25 #define ADDSUBPS_X8_X9 LONG $0xD00F45F2; BYTE $0xC8 // ADDSUBPS X8, X9
38 // func DotcUnitary(x, y []complex64) (sum complex64)
39 TEXT ·DotcUnitary(SB), NOSPLIT, $0
40 MOVQ x_base+0(FP), X_PTR // X_PTR = &x
41 MOVQ y_base+24(FP), Y_PTR // Y_PTR = &y
42 PXOR SUM, SUM // SUM = 0
43 PXOR P_SUM, P_SUM // P_SUM = 0
44 MOVQ x_len+8(FP), LEN // LEN = min( len(x), len(y) )
45 CMPQ y_len+32(FP), LEN
46 CMOVQLE y_len+32(FP), LEN
47 CMPQ LEN, $0 // if LEN == 0 { return }
49 XORQ IDX, IDX // i = 0
51 SHUFPS $0, NEG1, NEG1 // { -1, -1, -1, -1 }
54 ANDQ $15, DX // DX = &x & 15
55 JZ dotc_aligned // if DX == 0 { goto dotc_aligned }
57 MOVSD (X_PTR)(IDX*8), X3 // X_i = { imag(x[i]), real(x[i]) }
58 MOVSHDUP_X3_X2 // X_(i-1) = { imag(x[i]), imag(x[i]) }
59 MOVSLDUP_X3_X3 // X_i = { real(x[i]), real(x[i]) }
60 MOVSD (Y_PTR)(IDX*8), X10 // X_j = { imag(y[i]), real(y[i]) }
61 MULPS NEG1, X2 // X_(i-1) = { -imag(x[i]), imag(x[i]) }
62 MULPS X10, X3 // X_i = { imag(y[i]) * real(x[i]), real(y[i]) * real(x[i]) }
63 SHUFPS $0x1, X10, X10 // X_j = { real(y[i]), imag(y[i]) }
64 MULPS X10, X2 // X_(i-1) = { real(y[i]) * imag(x[i]), imag(y[i]) * imag(x[i]) }
67 // imag(result[i]): imag(y[i])*real(x[i]) + real(y[i])*imag(x[i]),
68 // real(result[i]): real(y[i])*real(x[i]) - imag(y[i])*imag(x[i]) }
71 MOVAPS X3, SUM // SUM = X_i
74 JZ dotc_ret // if LEN == 0 { goto dotc_ret }
78 ANDQ $7, TAIL // TAIL = LEN % 8
79 SHRQ $3, LEN // LEN = floor( LEN / 8 )
80 JZ dotc_tail // if LEN == 0 { return }
81 MOVUPS NEG1, P_NEG1 // Copy NEG1 for pipelining
84 MOVSLDUP_XPTR_IDX_8__X3 // X_i = { real(x[i]), real(x[i]), real(x[i+1]), real(x[i+1]) }
85 MOVSLDUP_16_XPTR_IDX_8__X5
86 MOVSLDUP_32_XPTR_IDX_8__X7
87 MOVSLDUP_48_XPTR_IDX_8__X9
89 MOVSHDUP_XPTR_IDX_8__X2 // X_(i-1) = { imag(x[i]), imag(x[i]), imag(x[i+1]), imag(x[i+1]) }
90 MOVSHDUP_16_XPTR_IDX_8__X4
91 MOVSHDUP_32_XPTR_IDX_8__X6
92 MOVSHDUP_48_XPTR_IDX_8__X8
94 // X_j = { imag(y[i]), real(y[i]), imag(y[i+1]), real(y[i+1]) }
95 MOVUPS (Y_PTR)(IDX*8), X10
96 MOVUPS 16(Y_PTR)(IDX*8), X11
97 MOVUPS 32(Y_PTR)(IDX*8), X12
98 MOVUPS 48(Y_PTR)(IDX*8), X13
100 // X_(i-1) = { -imag(x[i]), -imag(x[i]), -imag(x[i]+1), -imag(x[i]+1) }
106 // X_i = { imag(y[i]) * real(x[i]), real(y[i]) * real(x[i]),
107 // imag(y[i+1]) * real(x[i+1]), real(y[i+1]) * real(x[i+1]) }
113 // X_j = { real(y[i]), imag(y[i]), real(y[i+1]), imag(y[i+1]) }
114 SHUFPS $0xB1, X10, X10
115 SHUFPS $0xB1, X11, X11
116 SHUFPS $0xB1, X12, X12
117 SHUFPS $0xB1, X13, X13
119 // X_(i-1) = { real(y[i]) * imag(x[i]), imag(y[i]) * imag(x[i]),
120 // real(y[i+1]) * imag(x[i+1]), imag(y[i+1]) * imag(x[i+1]) }
127 // imag(result[i]): imag(y[i]) * real(x[i]) + real(y[i]) * imag(x[i]),
128 // real(result[i]): real(y[i]) * real(x[i]) - imag(y[i]) * imag(x[i]),
129 // imag(result[i+1]): imag(y[i+1]) * real(x[i+1]) + real(y[i+1]) * imag(x[i+1]),
130 // real(result[i+1]): real(y[i+1]) * real(x[i+1]) - imag(y[i+1]) * imag(x[i+1]),
143 ADDQ $8, IDX // IDX += 8
145 JNZ dotc_loop // } while --LEN > 0
147 ADDPS SUM, P_SUM // P_SUM = { P_SUM[1] + SUM[1], P_SUM[0] + SUM[0] }
148 XORPS SUM, SUM // SUM = 0
150 CMPQ TAIL, $0 // if TAIL == 0 { return }
155 SHRQ $1, LEN // LEN = floor( LEN / 2 )
156 JZ dotc_tail_one // if LEN == 0 { goto dotc_tail_one }
158 dotc_tail_two: // do {
159 MOVSLDUP_XPTR_IDX_8__X3 // X_i = { real(x[i]), real(x[i]), real(x[i+1]), real(x[i+1]) }
160 MOVSHDUP_XPTR_IDX_8__X2 // X_(i-1) = { imag(x[i]), imag(x[i]), imag(x[i]+1), imag(x[i]+1) }
161 MOVUPS (Y_PTR)(IDX*8), X10 // X_j = { imag(y[i]), real(y[i]) }
162 MULPS NEG1, X2 // X_(i-1) = { -imag(x[i]), imag(x[i]) }
163 MULPS X10, X3 // X_i = { imag(y[i]) * real(x[i]), real(y[i]) * real(x[i]) }
164 SHUFPS $0xB1, X10, X10 // X_j = { real(y[i]), imag(y[i]) }
165 MULPS X10, X2 // X_(i-1) = { real(y[i]) * imag(x[i]), imag(y[i]) * imag(x[i]) }
168 // imag(result[i]): imag(y[i])*real(x[i]) + real(y[i])*imag(x[i]),
169 // real(result[i]): real(y[i])*real(x[i]) - imag(y[i])*imag(x[i]) }
172 ADDPS X3, SUM // SUM += X_i
174 ADDQ $2, IDX // IDX += 2
176 JNZ dotc_tail_two // } while --LEN > 0
178 ADDPS SUM, P_SUM // P_SUM = { P_SUM[1] + SUM[1], P_SUM[0] + SUM[0] }
179 XORPS SUM, SUM // SUM = 0
185 MOVSD (X_PTR)(IDX*8), X3 // X_i = { imag(x[i]), real(x[i]) }
186 MOVSHDUP_X3_X2 // X_(i-1) = { imag(x[i]), imag(x[i]) }
187 MOVSLDUP_X3_X3 // X_i = { real(x[i]), real(x[i]) }
188 MOVSD (Y_PTR)(IDX*8), X10 // X_j = { imag(y[i]), real(y[i]) }
189 MULPS NEG1, X2 // X_(i-1) = { -imag(x[i]), imag(x[i]) }
190 MULPS X10, X3 // X_i = { imag(y[i]) * real(x[i]), real(y[i]) * real(x[i]) }
191 SHUFPS $0x1, X10, X10 // X_j = { real(y[i]), imag(y[i]) }
192 MULPS X10, X2 // X_(i-1) = { real(y[i]) * imag(x[i]), imag(y[i]) * imag(x[i]) }
195 // imag(result[i]): imag(y[i])*real(x[i]) + real(y[i])*imag(x[i]),
196 // real(result[i]): real(y[i])*real(x[i]) - imag(y[i])*imag(x[i]) }
199 ADDPS X3, SUM // SUM += X_i
202 ADDPS P_SUM, SUM // SUM = { P_SUM[0] + SUM[0] }
203 MOVHLPS P_SUM, P_SUM // P_SUM = { P_SUM[1], P_SUM[1] }
204 ADDPS P_SUM, SUM // SUM = { P_SUM[1] + SUM[0] }
207 MOVSD SUM, sum+48(FP) // return SUM