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 AxpyIncTo(dst []complex64, incDst, idst uintptr, alpha complex64, x, y []complex64, n, incX, incY, ix, iy uintptr)
38 TEXT ·AxpyIncTo(SB), NOSPLIT, $0
39 MOVQ dst_base+0(FP), DI // DI = &dst
40 MOVQ x_base+48(FP), SI // SI = &x
41 MOVQ y_base+72(FP), DX // DX = &y
42 MOVQ n+96(FP), CX // CX = n
43 CMPQ CX, $0 // if n==0 { return }
45 MOVQ ix+120(FP), R8 // Load the first index
48 LEAQ (SI)(R8*8), SI // SI = &(x[ix])
49 LEAQ (DX)(R9*8), DX // DX = &(y[iy])
50 LEAQ (DI)(R10*8), DI // DI = &(dst[idst])
51 MOVQ incX+104(FP), R8 // Incrementors*8 for easy iteration (ADDQ)
55 MOVQ incDst+24(FP), R10
57 MOVSD alpha+40(FP), X0 // X0 = { 0, 0, imag(a), real(a) }
59 SHUFPS $0x11, X1, X1 // X1 = { 0, 0, real(a), imag(a) }
60 MOVAPS X0, X10 // Copy X0 and X1 for pipelining
63 ANDQ $3, CX // CX = n % 4
64 SHRQ $2, BX // BX = floor( n / 4 )
65 JZ axpyi_tail // if BX == 0 { goto axpyi_tail }
68 MOVSD (SI), X3 // X_i = { imag(x[i]), real(x[i]) }
70 LEAQ (SI)(R8*2), SI // SI = &(SI[incX*2])
74 // X_(i-1) = { imag(x[i]), imag(x[i]) }
80 // X_i = { real(x[i]), real(x[i]) }
86 // X_(i-1) = { real(a) * imag(x[i]), imag(a) * imag(x[i]) }
87 // X_i = { imag(a) * real(x[i]), real(a) * real(x[i]) }
98 // imag(result[i]): imag(a)*real(x[i]) + real(a)*imag(x[i]),
99 // real(result[i]): real(a)*real(x[i]) - imag(a)*imag(x[i]),
106 // X_i = { imag(result[i]) + imag(y[i]), real(result[i]) + real(y[i]) }
109 LEAQ (DX)(R9*2), DX // DX = &(DX[incY*2])
117 MOVSD X3, (DI) // y[i] = X_i
118 MOVSD X5, (DI)(R10*1)
119 LEAQ (DI)(R10*2), DI // DI = &(DI[incDst])
121 MOVSD X9, (DI)(R10*1)
122 LEAQ (SI)(R8*2), SI // SI = &(SI[incX*2])
123 LEAQ (DX)(R9*2), DX // DX = &(DX[incY*2])
124 LEAQ (DI)(R10*2), DI // DI = &(DI[incDst])
126 JNZ axpyi_loop // } while --BX > 0
127 CMPQ CX, $0 // if CX == 0 { return }
131 MOVSD (SI), X3 // X_i = { imag(x[i]), real(x[i]) }
132 MOVSHDUP_X3_X2 // X_(i-1) = { imag(x[i]), imag(x[i]) }
133 MOVSLDUP_X3_X3 // X_i = { real(x[i]), real(x[i]) }
135 // X_i = { imag(a) * real(x[i]), real(a) * real(x[i]) }
136 // X_(i-1) = { real(a) * imag(x[i]), imag(a) * imag(x[i]) }
141 // imag(result[i]): imag(a)*real(x[i]) + real(a)*imag(x[i]),
142 // real(result[i]): real(a)*real(x[i]) - imag(a)*imag(x[i]),
146 // X_i = { imag(result[i]) + imag(y[i]), real(result[i]) + real(y[i]) }
149 MOVSD X3, (DI) // y[i] = X_i
150 ADDQ R8, SI // SI += incX
151 ADDQ R9, DX // DX += incY
152 ADDQ R10, DI // DI += incDst
153 LOOP axpyi_tail // } while --CX > 0