AVX-512 swizzle in native assembly for k1om (Intel Xeon Phi Knights Corner)

test_swizzle.c

// AVX-512 swizzle in native assembly for k1om (Intel Xeon Phi Knights Corner)
//
// (c) 2016 Dmitry Mikushin dmitry@parallel-compute.org
//
// $ icc -no-gcc -mmic -O3 -std=c99 test_swizzle.c -o test_swizzle
// $ micnativeloadex ./test_swizzle
// 1.000000 2.000000 3.000000 4.000000 5.000000 6.000000 7.000000 8.000000 
// 8.000000 8.000000 8.000000 8.000000 8.000000 8.000000 8.000000 8.000000

#include <immintrin.h>
#include <stdio.h>

#ifdef __GNUC__
/* Constants for register swizzle primitives. */
typedef enum {
    _MM_SWIZ_REG_NONE,      /* hgfe dcba - Nop */
#define _MM_SWIZ_REG_DCBA _MM_SWIZ_REG_NONE
    _MM_SWIZ_REG_CDAB,      /* ghef cdab - Swap pairs */
    _MM_SWIZ_REG_BADC,      /* fehg badc - Swap with two-away */
    _MM_SWIZ_REG_AAAA,      /* eeee aaaa - broadcast a element */
    _MM_SWIZ_REG_BBBB,      /* ffff bbbb - broadcast b element */
    _MM_SWIZ_REG_CCCC,      /* gggg cccc - broadcast c element */
    _MM_SWIZ_REG_DDDD,      /* hhhh dddd - broadcast d element */
    _MM_SWIZ_REG_DACB       /* hegf dacb - cross-product */
} _MM_SWIZZLE_ENUM;
#endif

inline __m512d mm512_swizzle_pd(__m512d val, _MM_SWIZZLE_ENUM imm)
{
#if 1
	switch (imm)
	{
	case _MM_SWIZ_REG_DCBA : break;
	case _MM_SWIZ_REG_CDAB : __asm__ ("vmovapd %0{{cdab}}, %0" : "+x"(val)); break;
	case _MM_SWIZ_REG_BADC : __asm__ ("vmovapd %0{{badc}}, %0" : "+x"(val)); break;
	case _MM_SWIZ_REG_AAAA : __asm__ ("vmovapd %0{{aaaa}}, %0" : "+x"(val)); break;
	case _MM_SWIZ_REG_BBBB : __asm__ ("vmovapd %0{{bbbb}}, %0" : "+x"(val)); break;
	case _MM_SWIZ_REG_CCCC : __asm__ ("vmovapd %0{{cccc}}, %0" : "+x"(val)); break;
	case _MM_SWIZ_REG_DDDD : __asm__ ("vmovapd %0{{dddd}}, %0" : "+x"(val)); break;
	case _MM_SWIZ_REG_DACB : __asm__ ("vmovapd %0{{dacb}}, %0" : "+x"(val)); break;
	}

	return val;
#else
	return _mm512_swizzle_pd(val, _MM_SWIZ_REG_CDAB);
#endif
}

static void _mm512_print_pd(__m512d val)
{
	double values[8] __attribute__((aligned(64)));
	_mm512_store_pd(values, val);
	for (int i = 0; i < 8; i++)
		printf("%f ", values[i]);
	printf("\n");
}

int main()
{
	double values[8] __attribute__((aligned(64))) = { 1, 2, 3, 4, 5, 6, 7, 8 };
	__m512d val = _mm512_load_pd(values);
	_mm512_print_pd(val);
	val = mm512_swizzle_pd(val, _MM_SWIZ_REG_CDAB);
	_mm512_print_pd(val); 

	return 0;
}