gonum.org/v1/gonum@v0.14.0/internal/asm/f32/axpyincto_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  // func AxpyIncTo(dst []float32, incDst, idst uintptr, alpha float32, x, y []float32, n, incX, incY, ix, iy uintptr)
    10  TEXT ·AxpyIncTo(SB), NOSPLIT, $0
    11  	MOVQ  n+96(FP), CX       // CX = n
    12  	CMPQ  CX, $0             // if n==0 { return }
    13  	JLE   axpyi_end
    14  	MOVQ  dst_base+0(FP), DI // DI = &dst
    15  	MOVQ  x_base+48(FP), SI  // SI = &x
    16  	MOVQ  y_base+72(FP), DX  // DX = &y
    17  	MOVQ  ix+120(FP), R8     // R8 = ix  // Load the first index
    18  	MOVQ  iy+128(FP), R9     // R9 = iy
    19  	MOVQ  idst+32(FP), R10   // R10 = idst
    20  	LEAQ  (SI)(R8*4), SI     // SI = &(x[ix])
    21  	LEAQ  (DX)(R9*4), DX     // DX = &(y[iy])
    22  	LEAQ  (DI)(R10*4), DI    // DI = &(dst[idst])
    23  	MOVQ  incX+104(FP), R8   // R8 = incX
    24  	SHLQ  $2, R8             // R8 *= sizeof(float32)
    25  	MOVQ  incY+112(FP), R9   // R9 = incY
    26  	SHLQ  $2, R9             // R9 *= sizeof(float32)
    27  	MOVQ  incDst+24(FP), R10 // R10 = incDst
    28  	SHLQ  $2, R10            // R10 *= sizeof(float32)
    29  	MOVSS alpha+40(FP), X0   // X0 = alpha
    30  	MOVSS X0, X1             // X1 = X0  // for pipelining
    31  	MOVQ  CX, BX
    32  	ANDQ  $3, BX             // BX = n % 4
    33  	SHRQ  $2, CX             // CX = floor( n / 4 )
    34  	JZ    axpyi_tail_start   // if CX == 0 { goto axpyi_tail_start }
    35  
    36  axpyi_loop: // Loop unrolled 4x   do {
    37  	MOVSS (SI), X2        // X_i = x[i]
    38  	MOVSS (SI)(R8*1), X3
    39  	LEAQ  (SI)(R8*2), SI  // SI = &(SI[incX*2])
    40  	MOVSS (SI), X4
    41  	MOVSS (SI)(R8*1), X5
    42  	MULSS X1, X2          // X_i *= a
    43  	MULSS X0, X3
    44  	MULSS X1, X4
    45  	MULSS X0, X5
    46  	ADDSS (DX), X2        // X_i += y[i]
    47  	ADDSS (DX)(R9*1), X3
    48  	LEAQ  (DX)(R9*2), DX  // DX = &(DX[incY*2])
    49  	ADDSS (DX), X4
    50  	ADDSS (DX)(R9*1), X5
    51  	MOVSS X2, (DI)        // dst[i] = X_i
    52  	MOVSS X3, (DI)(R10*1)
    53  	LEAQ  (DI)(R10*2), DI // DI = &(DI[incDst*2])
    54  	MOVSS X4, (DI)
    55  	MOVSS X5, (DI)(R10*1)
    56  	LEAQ  (SI)(R8*2), SI  // SI = &(SI[incX*2])  // Increment addresses
    57  	LEAQ  (DX)(R9*2), DX  // DX = &(DX[incY*2])
    58  	LEAQ  (DI)(R10*2), DI // DI = &(DI[incDst*2])
    59  	LOOP  axpyi_loop      // } while --CX > 0
    60  	CMPQ  BX, $0          // if BX == 0 { return }
    61  	JE    axpyi_end
    62  
    63  axpyi_tail_start: // Reset loop registers
    64  	MOVQ BX, CX // Loop counter: CX = BX
    65  
    66  axpyi_tail: // do {
    67  	MOVSS (SI), X2   // X2 = x[i]
    68  	MULSS X1, X2     // X2 *= a
    69  	ADDSS (DX), X2   // X2 += y[i]
    70  	MOVSS X2, (DI)   // dst[i] = X2
    71  	ADDQ  R8, SI     // SI = &(SI[incX])
    72  	ADDQ  R9, DX     // DX = &(DX[incY])
    73  	ADDQ  R10, DI    // DI = &(DI[incY])
    74  	LOOP  axpyi_tail // } while --CX > 0
    75  
    76  axpyi_end:
    77  	RET
    78