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 AxpyInc(alpha complex64, x, y []complex64, n, incX, incY, ix, iy uintptr)
38 TEXT ·AxpyInc(SB), NOSPLIT, $0
39 MOVQ x_base+8(FP), SI // SI = &x
40 MOVQ y_base+32(FP), DI // DI = &y
41 MOVQ n+56(FP), CX // CX = n
42 CMPQ CX, $0 // if n==0 { return }
44 MOVQ ix+80(FP), R8 // R8 = ix
45 MOVQ iy+88(FP), R9 // R9 = iy
46 LEAQ (SI)(R8*8), SI // SI = &(x[ix])
47 LEAQ (DI)(R9*8), DI // DI = &(y[iy])
48 MOVQ DI, DX // DX = DI // Read/Write pointers
49 MOVQ incX+64(FP), R8 // R8 = incX
50 SHLQ $3, R8 // R8 *= sizeof(complex64)
51 MOVQ incY+72(FP), R9 // R9 = incY
52 SHLQ $3, R9 // R9 *= sizeof(complex64)
53 MOVSD alpha+0(FP), X0 // X0 = { 0, 0, imag(a), real(a) }
55 SHUFPS $0x11, X1, X1 // X1 = { 0, 0, real(a), imag(a) }
56 MOVAPS X0, X10 // Copy X0 and X1 for pipelining
59 ANDQ $3, CX // CX = n % 4
60 SHRQ $2, BX // BX = floor( n / 4 )
61 JZ axpyi_tail // if BX == 0 { goto axpyi_tail }
64 MOVSD (SI), X3 // X_i = { imag(x[i+1]), real(x[i+1]) }
66 LEAQ (SI)(R8*2), SI // SI = &(SI[incX*2])
70 // X_(i-1) = { imag(x[i]), imag(x[i]) }
76 // X_i = { real(x[i]), real(x[i]) }
82 // X_(i-1) = { real(a) * imag(x[i]), imag(a) * imag(x[i]) }
83 // X_i = { imag(a) * real(x[i]), real(a) * real(x[i]) }
94 // imag(result[i]): imag(a)*real(x[i]) + real(a)*imag(x[i]),
95 // real(result[i]): real(a)*real(x[i]) - imag(a)*imag(x[i]),
102 // X_i = { imag(result[i]) + imag(y[i]), real(result[i]) + real(y[i]) }
105 LEAQ (DX)(R9*2), DX // DX = &(DX[incY*2])
113 MOVSD X3, (DI) // y[i] = X_i
115 LEAQ (DI)(R9*2), DI // DI = &(DI[incDst])
118 LEAQ (SI)(R8*2), SI // SI = &(SI[incX*2])
119 LEAQ (DX)(R9*2), DX // DX = &(DX[incY*2])
120 LEAQ (DI)(R9*2), DI // DI = &(DI[incDst])
122 JNZ axpyi_loop // } while --BX > 0
123 CMPQ CX, $0 // if CX == 0 { return }
127 MOVSD (SI), X3 // X_i = { imag(x[i+1]), real(x[i+1]) }
128 MOVSHDUP_X3_X2 // X_(i-1) = { real(x[i]), real(x[i]) }
129 MOVSLDUP_X3_X3 // X_i = { imag(x[i]), imag(x[i]) }
131 // X_i = { imag(a) * real(x[i]), real(a) * real(x[i]) }
132 // X_(i-1) = { real(a) * imag(x[i]), imag(a) * imag(x[i]) }
137 // imag(result[i]): imag(a)*real(x[i]) + real(a)*imag(x[i]),
138 // real(result[i]): real(a)*real(x[i]) - imag(a)*imag(x[i]),
140 ADDSUBPS_X2_X3 // (ai*x1r+ar*x1i, ar*x1r-ai*x1i)
142 // X_i = { imag(result[i]) + imag(y[i]), real(result[i]) + real(y[i]) }
145 MOVSD X3, (DI) // y[i] = X_i
146 ADDQ R8, SI // SI += incX
147 ADDQ R9, DI // DI += incY
148 LOOP axpyi_tail // } while --CX > 0