sl@0: /*
sl@0:  * LIBOIL - Library of Optimized Inner Loops
sl@0:  * Copyright (c) 2003,2004 David A. Schleef <ds@schleef.org>
sl@0:  * All rights reserved.
sl@0:  *
sl@0:  * Redistribution and use in source and binary forms, with or without
sl@0:  * modification, are permitted provided that the following conditions
sl@0:  * are met:
sl@0:  * 1. Redistributions of source code must retain the above copyright
sl@0:  *    notice, this list of conditions and the following disclaimer.
sl@0:  * 2. Redistributions in binary form must reproduce the above copyright
sl@0:  *    notice, this list of conditions and the following disclaimer in the
sl@0:  *    documentation and/or other materials provided with the distribution.
sl@0:  * 
sl@0:  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
sl@0:  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
sl@0:  * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
sl@0:  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT,
sl@0:  * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
sl@0:  * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
sl@0:  * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
sl@0:  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
sl@0:  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
sl@0:  * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
sl@0:  * POSSIBILITY OF SUCH DAMAGE.
sl@0:  */
sl@0: //Portions Copyright (c)  2008-2009 Nokia Corporation and/or its subsidiary(-ies). All rights reserved. 
sl@0: 
sl@0: #ifdef HAVE_CONFIG_H
sl@0: #include "config.h"
sl@0: #endif
sl@0: 
sl@0: #include <liboil/liboilfunction.h>
sl@0: #include "liboil/simdpack/simdpack.h"
sl@0: 
sl@0: #define ABS(x) ((x)>0 ? (x) : -(x))
sl@0: 
sl@0: #if 0
sl@0: static void
sl@0: abs_u16_s16_i386asm (uint16_t * dest, int dstr, int16_t * src, int sstr, int n)
sl@0: {
sl@0:   __asm__ __volatile__ ("\n"
sl@0:       "	.p2align 4,,15			\n"
sl@0:       "1:  movswl	(%0), %%eax		\n"
sl@0:       "    addl	$2, %0			\n"
sl@0:       "    movl	%%eax, %%edx		\n"
sl@0:       "    negl	%%edx			\n"
sl@0:       "    cmpl	$-1, %%eax		\n"
sl@0:       "    cmovle	%%edx, %%eax		\n"
sl@0:       "    movw	%%ax, (%1)		\n"
sl@0:       "    addl	$2, %1			\n"
sl@0:       "    decl	%2			\n"
sl@0:       "    testl	%2, %2			\n"
sl@0:       "    jg	1b			\n":"+r" (src), "+r" (dest), "+r" (n)
sl@0:       ::"eax", "edx");
sl@0: }
sl@0: 
sl@0: OIL_DEFINE_IMPL_FULL (abs_u16_s16_i386asm, abs_u16_s16, OIL_IMPL_FLAG_CMOV);
sl@0: #endif
sl@0: 
sl@0: #if 0
sl@0: /* The previous function after running through uberopt */
sl@0: static void
sl@0: abs_u16_s16_i386asm_uber4 (uint16_t * dest, int dstr, int16_t * src,
sl@0:     int sstr, int n)
sl@0: {
sl@0:   __asm__ __volatile__ ("\n"
sl@0:       "	.p2align 4,,15			\n"
sl@0:       "1:                               \n"
sl@0:       "    movswl	(%0), %%eax	\n" /* UBER 0:     */
sl@0:       "    addl	$2, %0			\n" /* UBER 1: 0   */
sl@0:       "    movl	%%eax, %%edx		\n" /* UBER 2: 0   */
sl@0:       "    decl	%2			\n" /* UBER 7:     */
sl@0:       "    negl	%%edx			\n" /* UBER 3: 2   */
sl@0:       "    cmpl	$-1, %%eax ; cmovle %%edx, %%eax \n" /* UBER 4: 3 */
sl@0:       "    movw	%%ax, (%1)		\n" /* UBER 5: 4   */
sl@0:       "    addl	$2, %1			\n" /* UBER 6: 5   */
sl@0:       "    testl	%2, %2		\n"
sl@0:       "    jg	1b			\n"
sl@0:       :"+r" (src), "+r" (dest), "+r" (n)
sl@0:       ::"eax", "edx");
sl@0: }
sl@0: OIL_DEFINE_IMPL_FULL (abs_u16_s16_i386asm_uber4, abs_u16_s16, OIL_IMPL_FLAG_CMOV);
sl@0: #endif
sl@0: 
sl@0: #if 0
sl@0: static void
sl@0: abs_u16_s16_i386asm2 (uint16_t * dest, int dstr, int16_t * src, int sstr, int n)
sl@0: {
sl@0:   __asm__ __volatile__ ("\n"
sl@0:       "	pushl	%%ebp			\n"
sl@0:       "	movl	%%eax, %%ebp		\n"
sl@0:       "	.p2align 4,,15			\n"
sl@0:       "1:	movswl	(%%edi), %%eax		\n"
sl@0:       "	addl	$2, %%edi		\n"
sl@0:       "	movl	%%eax, %%edx		\n"
sl@0:       "	negl	%%edx			\n"
sl@0:       "	cmpl	$-1, %%eax		\n"
sl@0:       "	cmovle	%%edx, %%eax		\n"
sl@0:       "	movw	%%ax, (%%ebp)		\n"
sl@0:       "	addl	$2, %%ebp		\n"
sl@0:       "	decl	%2			\n"
sl@0:       "	testl	%2, %2			\n"
sl@0:       "	jg	1b			\n"
sl@0:       "	popl	%%ebp			\n":"+D" (src), "+a" (dest), "+S" (n)
sl@0:       ::"ecx", "edx");
sl@0: }
sl@0: OIL_DEFINE_IMPL_FULL (abs_u16_s16_i386asm2, abs_u16_s16, OIL_IMPL_FLAG_CMOV);
sl@0: #endif
sl@0: 
sl@0: static void
sl@0: abs_u16_s16_i386asm3 (uint16_t * dest, int dstr, int16_t * src, int sstr, int n)
sl@0: {
sl@0: #if !defined(__WINSCW__) && !defined(__WINS__)      
sl@0:   __asm__ __volatile__ ("\n"
sl@0:       "	.p2align 4,,15			\n"
sl@0:       "1:  movswl (%1), %%eax           \n"
sl@0:       "    add %3, %1                   \n"
sl@0:       "    mov %%eax, %%edx             \n"
sl@0:       "    sar $0xf, %%ax               \n"
sl@0:       "    and %%edx, %%eax             \n"
sl@0:       "    add %%eax, %%eax             \n"
sl@0:       "    sub %%eax, %%edx             \n"
sl@0:       "    mov %%dx, (%0)               \n"
sl@0:       "    add %4, %0                   \n"
sl@0:       "    decl %2                      \n"
sl@0:       "    jne 1b                       \n"
sl@0:       : "+r" (dest), "+r" (src), "+m" (n)
sl@0:       : "m" (dstr), "m" (sstr)
sl@0:       : "eax", "edx");
sl@0: #endif
sl@0: }
sl@0: OIL_DEFINE_IMPL_ASM (abs_u16_s16_i386asm3, abs_u16_s16);
sl@0: 
sl@0: 
sl@0: 
sl@0: static void
sl@0: abs_u16_s16_mmx (uint16_t * dest, int dstr, int16_t * src, int sstr, int n)
sl@0: {
sl@0: #if !defined(__WINSCW__) && !defined(__WINS__)      
sl@0:   static const int16_t p[][4] = {
sl@0:     { -32768, -32768, -32768, -32768 },
sl@0:     { 32767, 32767, 32767, 32767 }
sl@0:   };
sl@0:   int16_t tmp[4];
sl@0: 
sl@0:   while (n & 3) {
sl@0:     *dest = ABS (*src);
sl@0:     OIL_INCREMENT (dest, dstr);
sl@0:     OIL_INCREMENT (src, sstr);
sl@0:     n--;
sl@0:   }
sl@0:   n /= 4;
sl@0:  
sl@0:   __asm__ __volatile__ ("\n"
sl@0:       "	movq	(%0), %%mm2		\n"
sl@0:       "	movq	8(%0), %%mm3		\n"
sl@0:       :: "r" (p));
sl@0: 
sl@0:   while (n--) {
sl@0:     tmp[0] = *src;
sl@0:     OIL_INCREMENT (src, sstr);
sl@0:     tmp[1] = *src;
sl@0:     OIL_INCREMENT (src, sstr);
sl@0:     tmp[2] = *src;
sl@0:     OIL_INCREMENT (src, sstr);
sl@0:     tmp[3] = *src;
sl@0:     OIL_INCREMENT (src, sstr);
sl@0:     __asm__ __volatile__ ("\n"
sl@0:         "	movq	(%0), %%mm1		\n"
sl@0:         "	movq	%%mm1, %%mm0		\n"
sl@0:         "	paddsw	%%mm2, %%mm0		\n"
sl@0:         "	paddsw	%%mm3, %%mm1		\n"
sl@0:         "	psubsw	%%mm2, %%mm0		\n"
sl@0:         "	psubsw	%%mm3, %%mm1		\n"
sl@0:         "	psubw	%%mm1, %%mm0		\n"
sl@0:         "	movq	%%mm0, (%0)		\n"
sl@0:         : : "r" (tmp)
sl@0:         : "memory" );
sl@0:     *dest = tmp[0];
sl@0:     OIL_INCREMENT (dest, dstr);
sl@0:     *dest = tmp[1];
sl@0:     OIL_INCREMENT (dest, dstr);
sl@0:     *dest = tmp[2];
sl@0:     OIL_INCREMENT (dest, dstr);
sl@0:     *dest = tmp[3];
sl@0:     OIL_INCREMENT (dest, dstr);
sl@0:   }
sl@0:   asm volatile ("emms");
sl@0: #endif  
sl@0: }
sl@0: 
sl@0: OIL_DEFINE_IMPL_FULL (abs_u16_s16_mmx, abs_u16_s16, OIL_IMPL_FLAG_MMX);
sl@0: 
sl@0: #if 0
sl@0: static void
sl@0: abs_u16_s16_mmxx (uint16_t * dest, int dstr, int16_t * src, int sstr, int n)
sl@0: {
sl@0:   short p[] = { -32768, -32768, -32768, -32768,
sl@0:     32767, 32767, 32767, 32767
sl@0:   };
sl@0: 
sl@0:   while (n & 7) {
sl@0:     *dest = ABS (*src);
sl@0:     OIL_INCREMENT (dest, dstr);
sl@0:     OIL_INCREMENT (src, sstr);
sl@0:     n--;
sl@0:   }
sl@0:   n /= 8;
sl@0:   __asm__ __volatile__ ("\n"
sl@0:       "	movq	(%3), %%mm2		\n"
sl@0:       "	movq	8(%3), %%mm3		\n"
sl@0:       "	.p2align 4,,15			\n"
sl@0:       "1:	movq	(%%edi), %%mm0		\n"
sl@0:       "	movq	(%%edi), %%mm1		\n"
sl@0:       "	paddsw	%%mm2, %%mm0		\n"
sl@0:       "	paddsw	%%mm3, %%mm1		\n"
sl@0:       "	psubsw	%%mm2, %%mm0		\n"
sl@0:       "	psubsw	%%mm3, %%mm1		\n"
sl@0:       "	psubw	%%mm1, %%mm0		\n"
sl@0:       "	movq	%%mm0, (%%eax)		\n"
sl@0:       "	 movq	8(%%edi), %%mm4		\n"
sl@0:       "	 movq	8(%%edi), %%mm5		\n"
sl@0:       "	 addl	$16, %%edi		\n"
sl@0:       "	 paddsw	%%mm2, %%mm4		\n"
sl@0:       "	 paddsw	%%mm3, %%mm5		\n"
sl@0:       "	 psubsw	%%mm2, %%mm4		\n"
sl@0:       "	 psubsw	%%mm3, %%mm5		\n"
sl@0:       "	 psubw	%%mm5, %%mm4		\n"
sl@0:       "	 movq	%%mm4, 8(%%eax)		\n"
sl@0:       "	 addl	$16, %%eax		\n"
sl@0:       "	decl	%2			\n"
sl@0:       "	testl	%2, %2			\n"
sl@0:       "	jg	1b			\n":"+D" (src), "+a" (dest), "+S" (n)
sl@0:       :"c" (p));
sl@0:   asm volatile ("emms");
sl@0: }
sl@0: OIL_DEFINE_IMPL_FULL (abs_u16_s16_mmxx, abs_u16_s16, OIL_IMPL_FLAG_MMX);
sl@0: #endif
sl@0: 
sl@0: #ifdef ENABLE_BROKEN_IMPLS
sl@0: static void
sl@0: abs_u16_s16_mmx2 (uint16_t * dest, int dstr, int16_t * src, int sstr, int n)
sl@0: {
sl@0:   while (n & 7) {
sl@0:     *dest = ABS (*src);
sl@0:     OIL_INCREMENT (dest, dstr);
sl@0:     OIL_INCREMENT (src, sstr);
sl@0:     n--;
sl@0:   }
sl@0:   n /= 8;
sl@0:   __asm__ __volatile__ ("\n"
sl@0:       "	pushl	%%ebp			\n"
sl@0:       "	movl	%%eax, %%ebp		\n"
sl@0:       "	.p2align 4,,15			\n"
sl@0:       "1:	movq	(%%edi), %%mm0		\n"
sl@0:       "	pxor	%%mm1, %%mm1		\n"
sl@0:       "	 movq	8(%%edi), %%mm2		\n"
sl@0:       "	 addl	$16, %%edi		\n"
sl@0:       "	psubw	%%mm0, %%mm1		\n"
sl@0:       "	 pxor	%%mm3, %%mm3		\n"
sl@0:       "	pmaxsw	%%mm0, %%mm1		\n"
sl@0:       "	 psubw	%%mm2, %%mm3		\n"
sl@0:       "	movq	%%mm1, (%%ebp)		\n"
sl@0:       "	 pmaxsw	%%mm2, %%mm3		\n"
sl@0:       "	 movq	%%mm3, 8(%%ebp)		\n"
sl@0:       "	 addl	$16, %%ebp		\n"
sl@0:       "	decl	%2			\n"
sl@0:       "	testl	%2, %2			\n"
sl@0:       "	jg	1b			\n"
sl@0:       "	popl	%%ebp			\n":"+D" (src), "+a" (dest), "+S" (n)
sl@0:       ::"ecx", "edx");
sl@0:   asm volatile ("emms");
sl@0: }
sl@0: OIL_DEFINE_IMPL_FULL (abs_u16_s16_mmx2, abs_u16_s16, OIL_IMPL_FLAG_MMXEXT);
sl@0: #endif
sl@0: 
sl@0: #ifdef ENABLE_BROKEN_IMPLS
sl@0: static void
sl@0: abs_u16_s16_sse2 (uint16_t * dest, int dstr, int16_t * src, int sstr, int n)
sl@0: {
sl@0:   while (n & 7) {
sl@0:     *dest = ABS (*src);
sl@0:     OIL_INCREMENT (dest, dstr);
sl@0:     OIL_INCREMENT (src, sstr);
sl@0:     n--;
sl@0:   }
sl@0:   n /= 8;
sl@0:   __asm__ __volatile__ ("\n"
sl@0:       "	pushl	%%ebp			\n"
sl@0:       "	movl	%%eax, %%ebp		\n"
sl@0:       "	.p2align 4,,15			\n"
sl@0:       "1:	movq	(%%edi), %%xmm0		\n"
sl@0:       "	addl	$16, %%edi		\n"
sl@0:       "	pxor	%%xmm1, %%xmm1		\n"
sl@0:       "	psubw	%%xmm0, %%xmm1		\n"
sl@0:       "	pmaxsw	%%xmm0, %%xmm1		\n"
sl@0:       "	movq	%%xmm1, (%%ebp)		\n"
sl@0:       "	addl	$16, %%ebp		\n"
sl@0:       "	decl	%2			\n"
sl@0:       "	testl	%2, %2			\n"
sl@0:       "	jg	1b			\n"
sl@0:       "	popl	%%ebp			\n":"+D" (src), "+a" (dest), "+S" (n)
sl@0:       ::"ecx", "edx");
sl@0: }
sl@0: OIL_DEFINE_IMPL_FULL (abs_u16_s16_sse2, abs_u16_s16, OIL_IMPL_FLAG_SSE2);
sl@0: #endif
sl@0: 
sl@0: 
sl@0: 
sl@0: #ifdef	__SYMBIAN32__
sl@0:  
sl@0: OilFunctionImpl* __oil_function_impl_abs_u16_s16_i386asm, abs_u16_s16() {
sl@0: 		return &_oil_function_impl_abs_u16_s16_i386asm, abs_u16_s16;
sl@0: }
sl@0: #endif
sl@0: 
sl@0: #ifdef	__SYMBIAN32__
sl@0:  
sl@0: OilFunctionImpl* __oil_function_impl_abs_u16_s16_i386asm_uber4, abs_u16_s16() {
sl@0: 		return &_oil_function_impl_abs_u16_s16_i386asm_uber4, abs_u16_s16;
sl@0: }
sl@0: #endif
sl@0: 
sl@0: #ifdef	__SYMBIAN32__
sl@0:  
sl@0: OilFunctionImpl* __oil_function_impl_abs_u16_s16_i386asm2, abs_u16_s16() {
sl@0: 		return &_oil_function_impl_abs_u16_s16_i386asm2, abs_u16_s16;
sl@0: }
sl@0: #endif
sl@0: 
sl@0: #ifdef	__SYMBIAN32__
sl@0:  
sl@0: OilFunctionImpl* __oil_function_impl_abs_u16_s16_mmx, abs_u16_s16() {
sl@0: 		return &_oil_function_impl_abs_u16_s16_mmx, abs_u16_s16;
sl@0: }
sl@0: #endif
sl@0: 
sl@0: #ifdef	__SYMBIAN32__
sl@0:  
sl@0: OilFunctionImpl* __oil_function_impl_abs_u16_s16_mmxx, abs_u16_s16() {
sl@0: 		return &_oil_function_impl_abs_u16_s16_mmxx, abs_u16_s16;
sl@0: }
sl@0: #endif
sl@0: 
sl@0: #ifdef	__SYMBIAN32__
sl@0:  
sl@0: OilFunctionImpl* __oil_function_impl_abs_u16_s16_mmx2, abs_u16_s16() {
sl@0: 		return &_oil_function_impl_abs_u16_s16_mmx2, abs_u16_s16;
sl@0: }
sl@0: #endif
sl@0: 
sl@0: #ifdef	__SYMBIAN32__
sl@0:  
sl@0: OilFunctionImpl* __oil_function_impl_abs_u16_s16_sse2, abs_u16_s16() {
sl@0: 		return &_oil_function_impl_abs_u16_s16_sse2, abs_u16_s16;
sl@0: }
sl@0: #endif
sl@0: 
sl@0: 
sl@0: 
sl@0: #ifdef	__SYMBIAN32__
sl@0:  
sl@0: OilFunctionImpl* __oil_function_impl_abs_u16_s16_i386asm3() {
sl@0: 		return &_oil_function_impl_abs_u16_s16_i386asm3;
sl@0: }
sl@0: #endif
sl@0: