Linux Nasm - Listing 16-04 : AvxPermute

main.cpp

#include <stdio.h>
#include "../../commonfiles/ymmval.h"
#include <math.h>

extern "C" void AvxPermuteInt32(YmmVal* des, YmmVal* src, YmmVal* ind);
extern "C" void AvxPermuteFloat(YmmVal* des, YmmVal* src, YmmVal* ind);
extern "C" void AvxPermuteFloatIl(YmmVal* des, YmmVal* src, YmmVal* ind);

void AvxPermuteInt32Cpp(void)
{
	__attribute__((aligned(32))) YmmVal des, src, ind;

    src.i32[0] = 10;        ind.i32[0] = 3;
    src.i32[1] = 20;        ind.i32[1] = 7;
    src.i32[2] = 30;        ind.i32[2] = 0;
    src.i32[3] = 40;        ind.i32[3] = 4;
    src.i32[4] = 50;        ind.i32[4] = 6;
    src.i32[5] = 60;        ind.i32[5] = 6;
    src.i32[6] = 70;        ind.i32[6] = 1;
    src.i32[7] = 80;        ind.i32[7] = 2;

    AvxPermuteInt32(&des, &src, &ind);

    printf("\nResults for AvxPermuteInt32()\n");
    for (int i = 0; i < 8; i++)
    {
        printf("des[%d]: %5d  ", i, des.i32[i]);
        printf("ind[%d]: %5d  ", i, ind.i32[i]);
        printf("src[%d]: %5d  ", i, src.i32[i]);
        printf("\n");
    }
}

void AvxPermuteFloatCpp(void)
{
	__attribute__((aligned(32))) YmmVal des, src, ind;

    // src1 indices must be between 0 and 7.
    src.r32[0] = 800.0f;       ind.i32[0] = 3;
    src.r32[1] = 700.0f;       ind.i32[1] = 7;
    src.r32[2] = 600.0f;       ind.i32[2] = 0;
    src.r32[3] = 500.0f;       ind.i32[3] = 4;
    src.r32[4] = 400.0f;       ind.i32[4] = 6;
    src.r32[5] = 300.0f;       ind.i32[5] = 6;
    src.r32[6] = 200.0f;       ind.i32[6] = 1;
    src.r32[7] = 100.0f;       ind.i32[7] = 2;

    AvxPermuteFloat(&des, &src, &ind);

    printf("\nResults for AvxPermuteFloat()\n");
    for (int i = 0; i < 8; i++)
    {
        printf("des[%d]: %8.1f  ", i, des.r32[i]);
        printf("ind[%d]: %5d  ", i, ind.i32[i]);
        printf("src[%d]: %8.1f  ", i, src.r32[i]);
        printf("\n");
    }
}

void AvxPermuteFloatIlCpp(void)
{
    __attribute__((aligned(32))) YmmVal des, src, ind;

    // Lower lane
    src.r32[0] = sqrt(10.0f);       ind.i32[0] = 3;
    src.r32[1] = sqrt(20.0f);       ind.i32[1] = 2;
    src.r32[2] = sqrt(30.0f);       ind.i32[2] = 2;
    src.r32[3] = sqrt(40.0f);       ind.i32[3] = 0;

    // Upper lane
    src.r32[4] = sqrt(50.0f);       ind.i32[4] = 1;
    src.r32[5] = sqrt(60.0f);       ind.i32[5] = 3;
    src.r32[6] = sqrt(70.0f);       ind.i32[6] = 3;
    src.r32[7] = sqrt(80.0f);       ind.i32[7] = 2;

    AvxPermuteFloatIl(&des, &src, &ind);

    printf("\nResults for AvxPermuteFloatIl()\n");
    for (int i = 0; i < 8; i++)
    {
        if (i == 0)
            printf("Lower lane\n");
        else if (i == 4)
            printf("Upper lane\n");

        printf("des[%d]: %8.4f  ", i, des.r32[i]);
        printf("ind[%d]: %5d  ", i, ind.i32[i]);
        printf("src[%d]: %8.4f  ", i, src.r32[i]);
        printf("\n");
    }
}

int main(int argc, char* argv[])
{
    AvxPermuteInt32Cpp();
    AvxPermuteFloatCpp();
    AvxPermuteFloatIlCpp();
    return 0;
}

avxpermute.asm

; Name:     avxpermute.asm
;
; Build:    g++ -c -m32 main.cpp -o main.o
;           nasm -f elf32 -o avxpermute.o avxpermute.asm
;           g++ -m32 -o avxpermute avxpermute.o main.o
;
; Source:   Modern x86 Assembly Language Programming p. 458

global AvxPermuteInt32
global AvxPermuteFloat
global AvxPermuteFloatIl

section .text

; extern "C" void AvxPermuteInt32(YmmVal* des, YmmVal* src, YmmVal* ind);
;
; Description:  The following function demonstrates use of the
;               vpermd instruction.
;
; Requires:     AVX2

%define des [ebp+8]
%define src [ebp+12]
%define ind [ebp+16]

AvxPermuteInt32:
    push    ebp
    mov     ebp,esp

; Load argument values
    mov     eax,des                    ;eax = ptr to des
    mov     ecx,src                    ;ecx = ptr to src
    mov     edx,ind                    ;edx = ptr to ind

; Perform dword permutation
    vmovdqa ymm1,[edx]                 ;ymm1 = ind
    vpermd  ymm0,ymm1,[ecx]
    vmovdqa [eax],ymm0                 ;save result
    vzeroupper
    pop     ebp
    ret

; extern "C" void AvxPermuteFloat(YmmVal* des, YmmVal* src, YmmVal* ind);
;
; Description:  The following function demonstrates use of the 
;               vpermps instruction.
;
; Requires:     AVX2

%define des [ebp+8]
%define src [ebp+12]
%define ind [ebp+16]

AvxPermuteFloat:
    push    ebp
    mov     ebp,esp

; Load argument values
    mov     eax,des                    ;eax = ptr to des
    mov     ecx,src                    ;ecx = ptr to src
    mov     edx,ind                    ;edx = ptr to ind

; Perform SPFP permutation
    vmovdqa ymm1,[edx]                 ;ymm1 = ind
    vpermps ymm0,ymm1,[ecx]
    vmovaps [eax],ymm0                 ;save result

    vzeroupper
    pop     ebp
    ret

; extern "C" void AvxPermuteFloatIl(YmmVal* des, YmmVal* src, YmmVal* ind);
;
; Description:  The following function demonstrates use of the 
;               vpermilps instruction.
;
; Requires:     AVX2

%define des [ebp+8]
%define src [ebp+12]
%define ind [ebp+16]

AvxPermuteFloatIl:
    push    ebp
    mov     ebp,esp

; Load argument values
    mov     eax,des                    ;eax = ptr to des
    mov     ecx,src                    ;ecx = ptr to src
    mov     edx,ind                    ;edx = ptr to ind

; Perform in-lane SPFP permutation.  Note that the second source
; operand of vpermilps specifies the indices.
    vmovdqa   ymm1,[ecx]               ;ymm1 = src
    vpermilps ymm0,ymm1,[edx]
    vmovaps   [eax],ymm0               ;save result

    vzeroupper
    pop     ebp
    ret

build

g++ -c -m32 main.cpp -o main.o -std=c++11
nasm -f elf32 -o avxpermute.o avxpermute.asm
g++ -m32 -o avxpermute avxpermute.o main.o ../../commonfiles/ymmval.o

Books

Listing 16-04 : AvxPermute

main.cpp

avxpermute.asm

build