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
10 #define MOVSHDUP_X3_X2 BYTE $0xF3; BYTE $0x0F; BYTE $0x16; BYTE $0xD3
12 #define MOVSLDUP_X3_X3 BYTE $0xF3; BYTE $0x0F; BYTE $0x12; BYTE $0xDB
14 #define ADDSUBPS_X2_X3 BYTE $0xF2; BYTE $0x0F; BYTE $0xD0; BYTE $0xDA
17 #define MOVSHDUP_X5_X4 BYTE $0xF3; BYTE $0x0F; BYTE $0x16; BYTE $0xE5
19 #define MOVSLDUP_X5_X5 BYTE $0xF3; BYTE $0x0F; BYTE $0x12; BYTE $0xED
21 #define ADDSUBPS_X4_X5 BYTE $0xF2; BYTE $0x0F; BYTE $0xD0; BYTE $0xEC
24 #define MOVSHDUP_X7_X6 BYTE $0xF3; BYTE $0x0F; BYTE $0x16; BYTE $0xF7
26 #define MOVSLDUP_X7_X7 BYTE $0xF3; BYTE $0x0F; BYTE $0x12; BYTE $0xFF
28 #define ADDSUBPS_X6_X7 BYTE $0xF2; BYTE $0x0F; BYTE $0xD0; BYTE $0xFE
31 #define MOVSHDUP_X9_X8 BYTE $0xF3; BYTE $0x45; BYTE $0x0F; BYTE $0x16; BYTE $0xC1
33 #define MOVSLDUP_X9_X9 BYTE $0xF3; BYTE $0x45; BYTE $0x0F; BYTE $0x12; BYTE $0xC9
35 #define ADDSUBPS_X8_X9 BYTE $0xF2; BYTE $0x45; BYTE $0x0F; BYTE $0xD0; BYTE $0xC8
37 // func AxpyUnitaryTo(dst []complex64, alpha complex64, x, y []complex64)
38 TEXT ·AxpyUnitaryTo(SB), NOSPLIT, $0
39 MOVQ dst_base+0(FP), DI // DI = &dst
40 MOVQ x_base+32(FP), SI // SI = &x
41 MOVQ y_base+56(FP), DX // DX = &y
43 CMPQ y_len+64(FP), CX // CX = min( len(x), len(y), len(dst) )
44 CMOVQLE y_len+64(FP), CX
45 CMPQ dst_len+8(FP), CX
46 CMOVQLE dst_len+8(FP), CX
47 CMPQ CX, $0 // if CX == 0 { return }
49 MOVSD alpha+24(FP), X0 // X0 = { 0, 0, imag(a), real(a) }
50 SHUFPD $0, X0, X0 // X0 = { imag(a), real(a), imag(a), real(a) }
52 SHUFPS $0x11, X1, X1 // X1 = { real(a), imag(a), real(a), imag(a) }
54 MOVQ DX, BX // Align on 16-byte boundary for ADDPS
55 ANDQ $15, BX // BX = &y & 15
56 JZ caxy_no_trim // if BX == 0 { goto caxy_no_trim }
58 MOVSD (SI)(AX*8), X3 // X3 = { imag(x[i]), real(x[i]) }
59 MOVSHDUP_X3_X2 // X2 = { imag(x[i]), imag(x[i]) }
60 MOVSLDUP_X3_X3 // X3 = { real(x[i]), real(x[i]) }
61 MULPS X1, X2 // X2 = { real(a) * imag(x[i]), imag(a) * imag(x[i]) }
62 MULPS X0, X3 // X3 = { imag(a) * real(x[i]), real(a) * real(x[i]) }
64 // X3 = { imag(a)*real(x[i]) + real(a)*imag(x[i]), real(a)*real(x[i]) - imag(a)*imag(x[i]) }
66 MOVSD (DX)(AX*8), X4 // X3 += y[i]
68 MOVSD X3, (DI)(AX*8) // dst[i] = X3
71 JZ caxy_tail // if BX == 0 { goto caxy_tail }
74 MOVAPS X0, X10 // Copy X0 and X1 for pipelineing
77 ANDQ $7, CX // CX = n % 8
78 SHRQ $3, BX // BX = floor( n / 8 )
79 JZ caxy_tail // if BX == 0 { goto caxy_tail }
82 // X_i = { imag(x[i]), real(x[i]), imag(x[i+1]), real(x[i+1]) }
84 MOVUPS 16(SI)(AX*8), X5
85 MOVUPS 32(SI)(AX*8), X7
86 MOVUPS 48(SI)(AX*8), X9
88 // X_(i-1) = { imag(x[i]), imag(x[i]), imag(x[i]+1), imag(x[i]+1) }
94 // X_i = { real(x[i]), real(x[i]), real(x[i+1]), real(x[i+1]) }
100 // X_i = { imag(a) * real(x[i]), real(a) * real(x[i]),
101 // imag(a) * real(x[i+1]), real(a) * real(x[i+1]) }
102 // X_(i-1) = { real(a) * imag(x[i]), imag(a) * imag(x[i]),
103 // real(a) * imag(x[i+1]), imag(a) * imag(x[i+1]) }
114 // imag(result[i]): imag(a)*real(x[i]) + real(a)*imag(x[i]),
115 // real(result[i]): real(a)*real(x[i]) - imag(a)*imag(x[i]),
116 // imag(result[i+1]): imag(a)*real(x[i+1]) + real(a)*imag(x[i+1]),
117 // real(result[i+1]): real(a)*real(x[i+1]) - imag(a)*imag(x[i+1]),
124 // X_i = { imag(result[i]) + imag(y[i]), real(result[i]) + real(y[i]),
125 // imag(result[i+1]) + imag(y[i+1]), real(result[i+1]) + real(y[i+1]) }
127 ADDPS 16(DX)(AX*8), X5
128 ADDPS 32(DX)(AX*8), X7
129 ADDPS 48(DX)(AX*8), X9
130 MOVUPS X3, (DI)(AX*8) // y[i:i+1] = X_i
131 MOVUPS X5, 16(DI)(AX*8)
132 MOVUPS X7, 32(DI)(AX*8)
133 MOVUPS X9, 48(DI)(AX*8)
134 ADDQ $8, AX // i += 8
136 JNZ caxy_loop // } while BX > 0
137 CMPQ CX, $0 // if CX == 0 { return }
141 MOVSD (SI)(AX*8), X3 // X3 = { imag(x[i]), real(x[i]) }
142 MOVSHDUP_X3_X2 // X2 = { imag(x[i]), imag(x[i]) }
143 MOVSLDUP_X3_X3 // X3 = { real(x[i]), real(x[i]) }
144 MULPS X1, X2 // X2 = { real(a) * imag(x[i]), imag(a) * imag(x[i]) }
145 MULPS X0, X3 // X3 = { imag(a) * real(x[i]), real(a) * real(x[i]) }
147 // X3 = { imag(a)*real(x[i]) + real(a)*imag(x[i]),
148 // real(a)*real(x[i]) - imag(a)*imag(x[i]) }
150 MOVSD (DX)(AX*8), X4 // X3 += y[i]
152 MOVSD X3, (DI)(AX*8) // y[i] = X3
154 LOOP caxy_tail // } while --CX > 0