nlpodyssey/spago

View on GitHub
mat/internal/f64/asm64/gemvN_amd64.s

Summary

Maintainability
Test Coverage
// Copyright ©2017 The Gonum Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.

// +build !noasm,!gccgo,!safe

#include "textflag.h"

#define SIZE 8

#define M_DIM m+0(FP)
#define M CX
#define N_DIM n+8(FP)
#define N BX

#define TMP1 R14
#define TMP2 R15

#define X_PTR SI
#define X x_base+56(FP)
#define INC_X R8
#define INC3_X R9

#define Y_PTR DX
#define Y y_base+96(FP)
#define INC_Y R10
#define INC3_Y R11

#define A_ROW AX
#define A_PTR DI
#define LDA R12
#define LDA3 R13

#define ALPHA X15
#define BETA X14

#define INIT4 \
    XORPS X0, X0 \
    XORPS X1, X1 \
    XORPS X2, X2 \
    XORPS X3, X3

#define INIT2 \
    XORPS X0, X0 \
    XORPS X1, X1

#define INIT1 \
    XORPS X0, X0

#define KERNEL_LOAD4 \
    MOVUPS (X_PTR), X12       \
    MOVUPS 2*SIZE(X_PTR), X13

#define KERNEL_LOAD2 \
    MOVUPS (X_PTR), X12

#define KERNEL_LOAD4_INC \
    MOVSD  (X_PTR), X12           \
    MOVHPD (X_PTR)(INC_X*1), X12  \
    MOVSD  (X_PTR)(INC_X*2), X13  \
    MOVHPD (X_PTR)(INC3_X*1), X13

#define KERNEL_LOAD2_INC \
    MOVSD  (X_PTR), X12          \
    MOVHPD (X_PTR)(INC_X*1), X12

#define KERNEL_4x4 \
    MOVUPS (A_PTR), X4                \
    MOVUPS 2*SIZE(A_PTR), X5          \
    MOVUPS (A_PTR)(LDA*1), X6         \
    MOVUPS 2*SIZE(A_PTR)(LDA*1), X7   \
    MOVUPS (A_PTR)(LDA*2), X8         \
    MOVUPS 2*SIZE(A_PTR)(LDA*2), X9   \
    MOVUPS (A_PTR)(LDA3*1), X10       \
    MOVUPS 2*SIZE(A_PTR)(LDA3*1), X11 \
    MULPD  X12, X4                    \
    MULPD  X13, X5                    \
    MULPD  X12, X6                    \
    MULPD  X13, X7                    \
    MULPD  X12, X8                    \
    MULPD  X13, X9                    \
    MULPD  X12, X10                   \
    MULPD  X13, X11                   \
    ADDPD  X4, X0                     \
    ADDPD  X5, X0                     \
    ADDPD  X6, X1                     \
    ADDPD  X7, X1                     \
    ADDPD  X8, X2                     \
    ADDPD  X9, X2                     \
    ADDPD  X10, X3                    \
    ADDPD  X11, X3                    \
    ADDQ   $4*SIZE, A_PTR

#define KERNEL_4x2 \
    MOVUPS (A_PTR), X4         \
    MOVUPS (A_PTR)(LDA*1), X5  \
    MOVUPS (A_PTR)(LDA*2), X6  \
    MOVUPS (A_PTR)(LDA3*1), X7 \
    MULPD  X12, X4             \
    MULPD  X12, X5             \
    MULPD  X12, X6             \
    MULPD  X12, X7             \
    ADDPD  X4, X0              \
    ADDPD  X5, X1              \
    ADDPD  X6, X2              \
    ADDPD  X7, X3              \
    ADDQ   $2*SIZE, A_PTR

#define KERNEL_4x1 \
    MOVDDUP (X_PTR), X12        \
    MOVSD   (A_PTR), X4         \
    MOVHPD  (A_PTR)(LDA*1), X4  \
    MOVSD   (A_PTR)(LDA*2), X5  \
    MOVHPD  (A_PTR)(LDA3*1), X5 \
    MULPD   X12, X4             \
    MULPD   X12, X5             \
    ADDPD   X4, X0              \
    ADDPD   X5, X2              \
    ADDQ    $SIZE, A_PTR

#define STORE4 \
    MOVUPS (Y_PTR), X4       \
    MOVUPS 2*SIZE(Y_PTR), X5 \
    MULPD  ALPHA, X0         \
    MULPD  ALPHA, X2         \
    MULPD  BETA, X4          \
    MULPD  BETA, X5          \
    ADDPD  X0, X4            \
    ADDPD  X2, X5            \
    MOVUPS X4, (Y_PTR)       \
    MOVUPS X5, 2*SIZE(Y_PTR)

#define STORE4_INC \
    MOVSD  (Y_PTR), X4           \
    MOVHPD (Y_PTR)(INC_Y*1), X4  \
    MOVSD  (Y_PTR)(INC_Y*2), X5  \
    MOVHPD (Y_PTR)(INC3_Y*1), X5 \
    MULPD  ALPHA, X0             \
    MULPD  ALPHA, X2             \
    MULPD  BETA, X4              \
    MULPD  BETA, X5              \
    ADDPD  X0, X4                \
    ADDPD  X2, X5                \
    MOVLPD X4, (Y_PTR)           \
    MOVHPD X4, (Y_PTR)(INC_Y*1)  \
    MOVLPD X5, (Y_PTR)(INC_Y*2)  \
    MOVHPD X5, (Y_PTR)(INC3_Y*1)

#define KERNEL_2x4 \
    MOVUPS (A_PTR), X8               \
    MOVUPS 2*SIZE(A_PTR), X9         \
    MOVUPS (A_PTR)(LDA*1), X10       \
    MOVUPS 2*SIZE(A_PTR)(LDA*1), X11 \
    MULPD  X12, X8                   \
    MULPD  X13, X9                   \
    MULPD  X12, X10                  \
    MULPD  X13, X11                  \
    ADDPD  X8, X0                    \
    ADDPD  X10, X1                   \
    ADDPD  X9, X0                    \
    ADDPD  X11, X1                   \
    ADDQ   $4*SIZE, A_PTR

#define KERNEL_2x2 \
    MOVUPS (A_PTR), X8        \
    MOVUPS (A_PTR)(LDA*1), X9 \
    MULPD  X12, X8            \
    MULPD  X12, X9            \
    ADDPD  X8, X0             \
    ADDPD  X9, X1             \
    ADDQ   $2*SIZE, A_PTR

#define KERNEL_2x1 \
    MOVDDUP (X_PTR), X12       \
    MOVSD   (A_PTR), X8        \
    MOVHPD  (A_PTR)(LDA*1), X8 \
    MULPD   X12, X8            \
    ADDPD   X8, X0             \
    ADDQ    $SIZE, A_PTR

#define STORE2 \
    MOVUPS (Y_PTR), X4 \
    MULPD  ALPHA, X0   \
    MULPD  BETA, X4    \
    ADDPD  X0, X4      \
    MOVUPS X4, (Y_PTR)

#define STORE2_INC \
    MOVSD  (Y_PTR), X4          \
    MOVHPD (Y_PTR)(INC_Y*1), X4 \
    MULPD  ALPHA, X0            \
    MULPD  BETA, X4             \
    ADDPD  X0, X4               \
    MOVSD  X4, (Y_PTR)          \
    MOVHPD X4, (Y_PTR)(INC_Y*1)

#define KERNEL_1x4 \
    MOVUPS (A_PTR), X8       \
    MOVUPS 2*SIZE(A_PTR), X9 \
    MULPD  X12, X8           \
    MULPD  X13, X9           \
    ADDPD  X8, X0            \
    ADDPD  X9, X0            \
    ADDQ   $4*SIZE, A_PTR

#define KERNEL_1x2 \
    MOVUPS (A_PTR), X8    \
    MULPD  X12, X8        \
    ADDPD  X8, X0         \
    ADDQ   $2*SIZE, A_PTR

#define KERNEL_1x1 \
    MOVSD (X_PTR), X12 \
    MOVSD (A_PTR), X8  \
    MULSD X12, X8      \
    ADDSD X8, X0       \
    ADDQ  $SIZE, A_PTR

#define STORE1 \
    HADDPD X0, X0      \
    MOVSD  (Y_PTR), X4 \
    MULSD  ALPHA, X0   \
    MULSD  BETA, X4    \
    ADDSD  X0, X4      \
    MOVSD  X4, (Y_PTR)

// func GemvN(m, n int,
//    alpha float64,
//    a []float64, lda int,
//    x []float64, incX int,
//    beta float64,
//    y []float64, incY int)
TEXT ·GemvN(SB), NOSPLIT, $32-128
    MOVQ M_DIM, M
    MOVQ N_DIM, N
    CMPQ M, $0
    JE   end
    CMPQ N, $0
    JE   end

    MOVDDUP alpha+16(FP), ALPHA
    MOVDDUP beta+88(FP), BETA

    MOVQ x_base+56(FP), X_PTR
    MOVQ y_base+96(FP), Y_PTR
    MOVQ a_base+24(FP), A_ROW
    MOVQ incY+120(FP), INC_Y
    MOVQ lda+48(FP), LDA      // LDA = LDA * sizeof(float64)
    SHLQ $3, LDA
    LEAQ (LDA)(LDA*2), LDA3   // LDA3 = LDA * 3
    MOVQ A_ROW, A_PTR

    XORQ    TMP2, TMP2
    MOVQ    M, TMP1
    SUBQ    $1, TMP1
    IMULQ   INC_Y, TMP1
    NEGQ    TMP1
    CMPQ    INC_Y, $0
    CMOVQLT TMP1, TMP2
    LEAQ    (Y_PTR)(TMP2*SIZE), Y_PTR
    MOVQ    Y_PTR, Y

    SHLQ $3, INC_Y                // INC_Y = incY * sizeof(float64)
    LEAQ (INC_Y)(INC_Y*2), INC3_Y // INC3_Y = INC_Y * 3

    MOVSD  $0.0, X0
    COMISD BETA, X0
    JNE    gemv_start // if beta != 0 { goto gemv_start }

gemv_clear: // beta == 0 is special cased to clear memory (no nan handling)
    XORPS X0, X0
    XORPS X1, X1
    XORPS X2, X2
    XORPS X3, X3

    CMPQ incY+120(FP), $1 // Check for dense vector X (fast-path)
    JNE  inc_clear

    SHRQ $3, M
    JZ   clear4

clear8:
    MOVUPS X0, (Y_PTR)
    MOVUPS X1, 16(Y_PTR)
    MOVUPS X2, 32(Y_PTR)
    MOVUPS X3, 48(Y_PTR)
    ADDQ   $8*SIZE, Y_PTR
    DECQ   M
    JNZ    clear8

clear4:
    TESTQ  $4, M_DIM
    JZ     clear2
    MOVUPS X0, (Y_PTR)
    MOVUPS X1, 16(Y_PTR)
    ADDQ   $4*SIZE, Y_PTR

clear2:
    TESTQ  $2, M_DIM
    JZ     clear1
    MOVUPS X0, (Y_PTR)
    ADDQ   $2*SIZE, Y_PTR

clear1:
    TESTQ $1, M_DIM
    JZ    prep_end
    MOVSD X0, (Y_PTR)

    JMP prep_end

inc_clear:
    SHRQ $2, M
    JZ   inc_clear2

inc_clear4:
    MOVSD X0, (Y_PTR)
    MOVSD X1, (Y_PTR)(INC_Y*1)
    MOVSD X2, (Y_PTR)(INC_Y*2)
    MOVSD X3, (Y_PTR)(INC3_Y*1)
    LEAQ  (Y_PTR)(INC_Y*4), Y_PTR
    DECQ  M
    JNZ   inc_clear4

inc_clear2:
    TESTQ $2, M_DIM
    JZ    inc_clear1
    MOVSD X0, (Y_PTR)
    MOVSD X1, (Y_PTR)(INC_Y*1)
    LEAQ  (Y_PTR)(INC_Y*2), Y_PTR

inc_clear1:
    TESTQ $1, M_DIM
    JZ    prep_end
    MOVSD X0, (Y_PTR)

prep_end:
    MOVQ Y, Y_PTR
    MOVQ M_DIM, M

gemv_start:
    CMPQ incX+80(FP), $1 // Check for dense vector X (fast-path)
    JNE  inc

    SHRQ $2, M
    JZ   r2

r4:
    // LOAD 4
    INIT4

    MOVQ N_DIM, N
    SHRQ $2, N
    JZ   r4c2

r4c4:
    // 4x4 KERNEL
    KERNEL_LOAD4
    KERNEL_4x4

    ADDQ $4*SIZE, X_PTR

    DECQ N
    JNZ  r4c4

r4c2:
    TESTQ $2, N_DIM
    JZ    r4c1

    // 4x2 KERNEL
    KERNEL_LOAD2
    KERNEL_4x2

    ADDQ $2*SIZE, X_PTR

r4c1:
    HADDPD X1, X0
    HADDPD X3, X2
    TESTQ  $1, N_DIM
    JZ     r4end

    // 4x1 KERNEL
    KERNEL_4x1

    ADDQ $SIZE, X_PTR

r4end:
    CMPQ INC_Y, $SIZE
    JNZ  r4st_inc

    STORE4
    ADDQ $4*SIZE, Y_PTR
    JMP  r4inc

r4st_inc:
    STORE4_INC
    LEAQ (Y_PTR)(INC_Y*4), Y_PTR

r4inc:
    MOVQ X, X_PTR
    LEAQ (A_ROW)(LDA*4), A_ROW
    MOVQ A_ROW, A_PTR

    DECQ M
    JNZ  r4

r2:
    TESTQ $2, M_DIM
    JZ    r1

    // LOAD 2
    INIT2

    MOVQ N_DIM, N
    SHRQ $2, N
    JZ   r2c2

r2c4:
    // 2x4 KERNEL
    KERNEL_LOAD4
    KERNEL_2x4

    ADDQ $4*SIZE, X_PTR

    DECQ N
    JNZ  r2c4

r2c2:
    TESTQ $2, N_DIM
    JZ    r2c1

    // 2x2 KERNEL
    KERNEL_LOAD2
    KERNEL_2x2

    ADDQ $2*SIZE, X_PTR

r2c1:
    HADDPD X1, X0
    TESTQ  $1, N_DIM
    JZ     r2end

    // 2x1 KERNEL
    KERNEL_2x1

    ADDQ $SIZE, X_PTR

r2end:
    CMPQ INC_Y, $SIZE
    JNE  r2st_inc

    STORE2
    ADDQ $2*SIZE, Y_PTR
    JMP  r2inc

r2st_inc:
    STORE2_INC
    LEAQ (Y_PTR)(INC_Y*2), Y_PTR

r2inc:
    MOVQ X, X_PTR
    LEAQ (A_ROW)(LDA*2), A_ROW
    MOVQ A_ROW, A_PTR

r1:
    TESTQ $1, M_DIM
    JZ    end

    // LOAD 1
    INIT1

    MOVQ N_DIM, N
    SHRQ $2, N
    JZ   r1c2

r1c4:
    // 1x4 KERNEL
    KERNEL_LOAD4
    KERNEL_1x4

    ADDQ $4*SIZE, X_PTR

    DECQ N
    JNZ  r1c4

r1c2:
    TESTQ $2, N_DIM
    JZ    r1c1

    // 1x2 KERNEL
    KERNEL_LOAD2
    KERNEL_1x2

    ADDQ $2*SIZE, X_PTR

r1c1:

    TESTQ $1, N_DIM
    JZ    r1end

    // 1x1 KERNEL
    KERNEL_1x1

r1end:
    STORE1

end:
    RET

inc:  // Algorithm for incX != 1 ( split loads in kernel )
    MOVQ incX+80(FP), INC_X // INC_X = incX

    XORQ    TMP2, TMP2                // TMP2  = 0
    MOVQ    N, TMP1                   // TMP1 = N
    SUBQ    $1, TMP1                  // TMP1 -= 1
    NEGQ    TMP1                      // TMP1 = -TMP1
    IMULQ   INC_X, TMP1               // TMP1 *= INC_X
    CMPQ    INC_X, $0                 // if INC_X < 0 { TMP2 = TMP1 }
    CMOVQLT TMP1, TMP2
    LEAQ    (X_PTR)(TMP2*SIZE), X_PTR // X_PTR = X_PTR[TMP2]
    MOVQ    X_PTR, X                  // X = X_PTR

    SHLQ $3, INC_X
    LEAQ (INC_X)(INC_X*2), INC3_X // INC3_X = INC_X * 3

    SHRQ $2, M
    JZ   inc_r2

inc_r4:
    // LOAD 4
    INIT4

    MOVQ N_DIM, N
    SHRQ $2, N
    JZ   inc_r4c2

inc_r4c4:
    // 4x4 KERNEL
    KERNEL_LOAD4_INC
    KERNEL_4x4

    LEAQ (X_PTR)(INC_X*4), X_PTR

    DECQ N
    JNZ  inc_r4c4

inc_r4c2:
    TESTQ $2, N_DIM
    JZ    inc_r4c1

    // 4x2 KERNEL
    KERNEL_LOAD2_INC
    KERNEL_4x2

    LEAQ (X_PTR)(INC_X*2), X_PTR

inc_r4c1:
    HADDPD X1, X0
    HADDPD X3, X2
    TESTQ  $1, N_DIM
    JZ     inc_r4end

    // 4x1 KERNEL
    KERNEL_4x1

    ADDQ INC_X, X_PTR

inc_r4end:
    CMPQ INC_Y, $SIZE
    JNE  inc_r4st_inc

    STORE4
    ADDQ $4*SIZE, Y_PTR
    JMP  inc_r4inc

inc_r4st_inc:
    STORE4_INC
    LEAQ (Y_PTR)(INC_Y*4), Y_PTR

inc_r4inc:
    MOVQ X, X_PTR
    LEAQ (A_ROW)(LDA*4), A_ROW
    MOVQ A_ROW, A_PTR

    DECQ M
    JNZ  inc_r4

inc_r2:
    TESTQ $2, M_DIM
    JZ    inc_r1

    // LOAD 2
    INIT2

    MOVQ N_DIM, N
    SHRQ $2, N
    JZ   inc_r2c2

inc_r2c4:
    // 2x4 KERNEL
    KERNEL_LOAD4_INC
    KERNEL_2x4

    LEAQ (X_PTR)(INC_X*4), X_PTR
    DECQ N
    JNZ  inc_r2c4

inc_r2c2:
    TESTQ $2, N_DIM
    JZ    inc_r2c1

    // 2x2 KERNEL
    KERNEL_LOAD2_INC
    KERNEL_2x2

    LEAQ (X_PTR)(INC_X*2), X_PTR

inc_r2c1:
    HADDPD X1, X0
    TESTQ  $1, N_DIM
    JZ     inc_r2end

    // 2x1 KERNEL
    KERNEL_2x1

    ADDQ INC_X, X_PTR

inc_r2end:
    CMPQ INC_Y, $SIZE
    JNE  inc_r2st_inc

    STORE2
    ADDQ $2*SIZE, Y_PTR
    JMP  inc_r2inc

inc_r2st_inc:
    STORE2_INC
    LEAQ (Y_PTR)(INC_Y*2), Y_PTR

inc_r2inc:
    MOVQ X, X_PTR
    LEAQ (A_ROW)(LDA*2), A_ROW
    MOVQ A_ROW, A_PTR

inc_r1:
    TESTQ $1, M_DIM
    JZ    inc_end

    // LOAD 1
    INIT1

    MOVQ N_DIM, N
    SHRQ $2, N
    JZ   inc_r1c2

inc_r1c4:
    // 1x4 KERNEL
    KERNEL_LOAD4_INC
    KERNEL_1x4

    LEAQ (X_PTR)(INC_X*4), X_PTR
    DECQ N
    JNZ  inc_r1c4

inc_r1c2:
    TESTQ $2, N_DIM
    JZ    inc_r1c1

    // 1x2 KERNEL
    KERNEL_LOAD2_INC
    KERNEL_1x2

    LEAQ (X_PTR)(INC_X*2), X_PTR

inc_r1c1:
    TESTQ $1, N_DIM
    JZ    inc_r1end

    // 1x1 KERNEL
    KERNEL_1x1

inc_r1end:
    STORE1

inc_end:
    RET