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 MOVDDUP_X2_X3 BYTE $0xF2; BYTE $0x0F; BYTE $0x12; BYTE $0xDA
12 #define MOVDDUP_X4_X5 BYTE $0xF2; BYTE $0x0F; BYTE $0x12; BYTE $0xEC
14 #define MOVDDUP_X6_X7 BYTE $0xF2; BYTE $0x0F; BYTE $0x12; BYTE $0xFE
16 #define MOVDDUP_X8_X9 BYTE $0xF2; BYTE $0x45; BYTE $0x0F; BYTE $0x12; BYTE $0xC8
19 #define ADDSUBPD_X2_X3 BYTE $0x66; BYTE $0x0F; BYTE $0xD0; BYTE $0xDA
21 #define ADDSUBPD_X4_X5 BYTE $0x66; BYTE $0x0F; BYTE $0xD0; BYTE $0xEC
23 #define ADDSUBPD_X6_X7 BYTE $0x66; BYTE $0x0F; BYTE $0xD0; BYTE $0xFE
25 #define ADDSUBPD_X8_X9 BYTE $0x66; BYTE $0x45; BYTE $0x0F; BYTE $0xD0; BYTE $0xC8
27 // func AxpyIncTo(dst []complex128, incDst, idst uintptr, alpha complex128, x, y []complex128, n, incX, incY, ix, iy uintptr)
28 TEXT ·AxpyIncTo(SB), NOSPLIT, $0
29 MOVQ dst_base+0(FP), DI // DI = &dst
30 MOVQ x_base+56(FP), SI // SI = &x
31 MOVQ y_base+80(FP), DX // DX = &y
32 MOVQ n+104(FP), CX // CX = n
33 CMPQ CX, $0 // if n==0 { return }
35 MOVQ ix+128(FP), R8 // R8 = ix // Load the first index
36 SHLQ $4, R8 // R8 *= sizeof(complex128)
37 MOVQ iy+136(FP), R9 // R9 = iy
38 SHLQ $4, R9 // R9 *= sizeof(complex128)
39 MOVQ idst+32(FP), R10 // R10 = idst
40 SHLQ $4, R10 // R10 *= sizeof(complex128)
41 LEAQ (SI)(R8*1), SI // SI = &(x[ix])
42 LEAQ (DX)(R9*1), DX // DX = &(y[iy])
43 LEAQ (DI)(R10*1), DI // DI = &(dst[idst])
44 MOVQ incX+112(FP), R8 // R8 = incX
45 SHLQ $4, R8 // R8 *= sizeof(complex128)
46 MOVQ incY+120(FP), R9 // R9 = incY
47 SHLQ $4, R9 // R9 *= sizeof(complex128)
48 MOVQ incDst+24(FP), R10 // R10 = incDst
49 SHLQ $4, R10 // R10 *= sizeof(complex128)
50 MOVUPS alpha+40(FP), X0 // X0 = { imag(a), real(a) }
52 SHUFPD $0x1, X1, X1 // X1 = { real(a), imag(a) }
53 MOVAPS X0, X10 // Copy X0 and X1 for pipelining
56 ANDQ $3, CX // CX = n % 4
57 SHRQ $2, BX // BX = floor( n / 4 )
58 JZ axpyi_tail // if BX == 0 { goto axpyi_tail }
61 MOVUPS (SI), X2 // X_i = { imag(x[i]), real(x[i]) }
63 LEAQ (SI)(R8*2), SI // SI = &(SI[incX*2])
68 // X_(i+1) = { real(x[i], real(x[i]) }
74 // X_i = { imag(x[i]), imag(x[i]) }
80 // X_i = { real(a) * imag(x[i]), imag(a) * imag(x[i]) }
81 // X_(i+1) = { imag(a) * real(x[i]), real(a) * real(x[i]) }
92 // imag(result[i]): imag(a)*real(x[i]) + real(a)*imag(x[i]),
93 // real(result[i]): real(a)*real(x[i]) - imag(a)*imag(x[i])
100 // X_(i+1) = { imag(result[i]) + imag(y[i]), real(result[i]) + real(y[i]) }
103 LEAQ (DX)(R9*2), DX // DX = &(DX[incY*2])
106 MOVUPS X3, (DI) // dst[i] = X_(i+1)
107 MOVUPS X5, (DI)(R10*1)
110 MOVUPS X9, (DI)(R10*1)
111 LEAQ (SI)(R8*2), SI // SI = &(SI[incX*2])
112 LEAQ (DX)(R9*2), DX // DX = &(DX[incY*2])
113 LEAQ (DI)(R10*2), DI // DI = &(DI[incDst*2])
115 JNZ axpyi_loop // } while --BX > 0
116 CMPQ CX, $0 // if CX == 0 { return }
120 MOVUPS (SI), X2 // X_i = { imag(x[i]), real(x[i]) }
121 MOVDDUP_X2_X3 // X_(i+1) = { real(x[i], real(x[i]) }
122 SHUFPD $0x3, X2, X2 // X_i = { imag(x[i]), imag(x[i]) }
123 MULPD X1, X2 // X_i = { real(a) * imag(x[i]), imag(a) * imag(x[i]) }
124 MULPD X0, X3 // X_(i+1) = { imag(a) * real(x[i]), real(a) * real(x[i]) }
127 // imag(result[i]): imag(a)*real(x[i]) + real(a)*imag(x[i]),
128 // real(result[i]): real(a)*real(x[i]) - imag(a)*imag(x[i])
132 // X_(i+1) = { imag(result[i]) + imag(y[i]), real(result[i]) + real(y[i]) }
134 MOVUPS X3, (DI) // y[i] X_(i+1)
135 ADDQ R8, SI // SI += incX
136 ADDQ R9, DX // DX += incY
137 ADDQ R10, DI // DI += incDst
138 LOOP axpyi_tail // } while --CX > 0