github.com/gopherd/gonum@v0.0.4/internal/asm/c128/dotcunitary_amd64.s (about) 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. 4 5 // +build !noasm,!gccgo,!safe 6 7 #include "textflag.h" 8 9 #define MOVDDUP_XPTR_IDX_8__X3 LONG $0x1C120FF2; BYTE $0xC6 // MOVDDUP (SI)(AX*8), X3 10 #define MOVDDUP_16_XPTR_IDX_8__X5 LONG $0x6C120FF2; WORD $0x10C6 // MOVDDUP 16(SI)(AX*8), X5 11 #define MOVDDUP_32_XPTR_IDX_8__X7 LONG $0x7C120FF2; WORD $0x20C6 // MOVDDUP 32(SI)(AX*8), X7 12 #define MOVDDUP_48_XPTR_IDX_8__X9 LONG $0x120F44F2; WORD $0xC64C; BYTE $0x30 // MOVDDUP 48(SI)(AX*8), X9 13 14 #define MOVDDUP_XPTR_IIDX_8__X2 LONG $0x14120FF2; BYTE $0xD6 // MOVDDUP (SI)(DX*8), X2 15 #define MOVDDUP_16_XPTR_IIDX_8__X4 LONG $0x64120FF2; WORD $0x10D6 // MOVDDUP 16(SI)(DX*8), X4 16 #define MOVDDUP_32_XPTR_IIDX_8__X6 LONG $0x74120FF2; WORD $0x20D6 // MOVDDUP 32(SI)(DX*8), X6 17 #define MOVDDUP_48_XPTR_IIDX_8__X8 LONG $0x120F44F2; WORD $0xD644; BYTE $0x30 // MOVDDUP 48(SI)(DX*8), X8 18 19 #define ADDSUBPD_X2_X3 LONG $0xDAD00F66 // ADDSUBPD X2, X3 20 #define ADDSUBPD_X4_X5 LONG $0xECD00F66 // ADDSUBPD X4, X5 21 #define ADDSUBPD_X6_X7 LONG $0xFED00F66 // ADDSUBPD X6, X7 22 #define ADDSUBPD_X8_X9 LONG $0xD00F4566; BYTE $0xC8 // ADDSUBPD X8, X9 23 24 #define X_PTR SI 25 #define Y_PTR DI 26 #define LEN CX 27 #define TAIL BX 28 #define SUM X0 29 #define P_SUM X1 30 #define IDX AX 31 #define I_IDX DX 32 #define NEG1 X15 33 #define P_NEG1 X14 34 35 // func DotcUnitary(x, y []complex128) (sum complex128) 36 TEXT ·DotcUnitary(SB), NOSPLIT, $0 37 MOVQ x_base+0(FP), X_PTR // X_PTR = &x 38 MOVQ y_base+24(FP), Y_PTR // Y_PTR = &y 39 MOVQ x_len+8(FP), LEN // LEN = min( len(x), len(y) ) 40 CMPQ y_len+32(FP), LEN 41 CMOVQLE y_len+32(FP), LEN 42 PXOR SUM, SUM // sum = 0 43 CMPQ LEN, $0 // if LEN == 0 { return } 44 JE dot_end 45 XORPS P_SUM, P_SUM // psum = 0 46 MOVSD $(-1.0), NEG1 47 SHUFPD $0, NEG1, NEG1 // { -1, -1 } 48 XORQ IDX, IDX // i := 0 49 MOVQ $1, I_IDX // j := 1 50 MOVQ LEN, TAIL 51 ANDQ $3, TAIL // TAIL = floor( TAIL / 4 ) 52 SHRQ $2, LEN // LEN = TAIL % 4 53 JZ dot_tail // if LEN == 0 { goto dot_tail } 54 55 MOVAPS NEG1, P_NEG1 // Copy NEG1 to P_NEG1 for pipelining 56 57 dot_loop: // do { 58 MOVDDUP_XPTR_IDX_8__X3 // X_(i+1) = { real(x[i], real(x[i]) } 59 MOVDDUP_16_XPTR_IDX_8__X5 60 MOVDDUP_32_XPTR_IDX_8__X7 61 MOVDDUP_48_XPTR_IDX_8__X9 62 63 MOVDDUP_XPTR_IIDX_8__X2 // X_i = { imag(x[i]), imag(x[i]) } 64 MOVDDUP_16_XPTR_IIDX_8__X4 65 MOVDDUP_32_XPTR_IIDX_8__X6 66 MOVDDUP_48_XPTR_IIDX_8__X8 67 68 // X_i = { -imag(x[i]), -imag(x[i]) } 69 MULPD NEG1, X2 70 MULPD P_NEG1, X4 71 MULPD NEG1, X6 72 MULPD P_NEG1, X8 73 74 // X_j = { imag(y[i]), real(y[i]) } 75 MOVUPS (Y_PTR)(IDX*8), X10 76 MOVUPS 16(Y_PTR)(IDX*8), X11 77 MOVUPS 32(Y_PTR)(IDX*8), X12 78 MOVUPS 48(Y_PTR)(IDX*8), X13 79 80 // X_(i+1) = { imag(a) * real(x[i]), real(a) * real(x[i]) } 81 MULPD X10, X3 82 MULPD X11, X5 83 MULPD X12, X7 84 MULPD X13, X9 85 86 // X_j = { real(y[i]), imag(y[i]) } 87 SHUFPD $0x1, X10, X10 88 SHUFPD $0x1, X11, X11 89 SHUFPD $0x1, X12, X12 90 SHUFPD $0x1, X13, X13 91 92 // X_i = { real(a) * imag(x[i]), imag(a) * imag(x[i]) } 93 MULPD X10, X2 94 MULPD X11, X4 95 MULPD X12, X6 96 MULPD X13, X8 97 98 // X_(i+1) = { 99 // imag(result[i]): imag(a)*real(x[i]) + real(a)*imag(x[i]), 100 // real(result[i]): real(a)*real(x[i]) - imag(a)*imag(x[i]) 101 // } 102 ADDSUBPD_X2_X3 103 ADDSUBPD_X4_X5 104 ADDSUBPD_X6_X7 105 ADDSUBPD_X8_X9 106 107 // psum += result[i] 108 ADDPD X3, SUM 109 ADDPD X5, P_SUM 110 ADDPD X7, SUM 111 ADDPD X9, P_SUM 112 113 ADDQ $8, IDX // IDX += 8 114 ADDQ $8, I_IDX // I_IDX += 8 115 DECQ LEN 116 JNZ dot_loop // } while --LEN > 0 117 ADDPD P_SUM, SUM // sum += psum 118 CMPQ TAIL, $0 // if TAIL == 0 { return } 119 JE dot_end 120 121 dot_tail: // do { 122 MOVDDUP_XPTR_IDX_8__X3 // X_(i+1) = { real(x[i]) , real(x[i]) } 123 MOVDDUP_XPTR_IIDX_8__X2 // X_i = { imag(x[i]) , imag(x[i]) } 124 MULPD NEG1, X2 // X_i = { -imag(x[i]) , -imag(x[i]) } 125 MOVUPS (Y_PTR)(IDX*8), X10 // X_j = { imag(y[i]) , real(y[i]) } 126 MULPD X10, X3 // X_(i+1) = { imag(a) * real(x[i]), real(a) * real(x[i]) } 127 SHUFPD $0x1, X10, X10 // X_j = { real(y[i]) , imag(y[i]) } 128 MULPD X10, X2 // X_i = { real(a) * imag(x[i]), imag(a) * imag(x[i]) } 129 130 // X_(i+1) = { 131 // imag(result[i]): imag(a)*real(x[i]) + real(a)*imag(x[i]), 132 // real(result[i]): real(a)*real(x[i]) - imag(a)*imag(x[i]) 133 // } 134 ADDSUBPD_X2_X3 135 ADDPD X3, SUM // SUM += result[i] 136 ADDQ $2, IDX // IDX += 2 137 ADDQ $2, I_IDX // I_IDX += 2 138 DECQ TAIL 139 JNZ dot_tail // } while --TAIL > 0 140 141 dot_end: 142 MOVUPS SUM, sum+48(FP) 143 RET