sl@0: /* sl@0: * LIBOIL - Library of Optimized Inner Loops sl@0: * Copyright (c) 2003,2004 David A. Schleef 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: #include sl@0: #include sl@0: sl@0: sl@0: /* suboptimal */ sl@0: static void sl@0: conv_f32_s16_3dnow (float *dst, int dst_stride, int16_t * src, int src_stride, sl@0: int n) sl@0: { sl@0: #if !defined(__WINSCW__) && !defined(__WINS__) sl@0: int i; sl@0: sl@0: for (i = 0; i < n; i++) { sl@0: asm volatile (" movswl 0(%0), %%eax \n" sl@0: " movd %%eax, %%mm0 \n" sl@0: " pi2fd %%mm0, %%mm0 \n" " movd %%mm0, 0(%1) \n" sl@0: ::"r" (src), "r" (dst) sl@0: :"eax"); sl@0: sl@0: dst = OIL_OFFSET (dst, dst_stride); sl@0: src = OIL_OFFSET (src, src_stride); sl@0: } sl@0: asm volatile ("emms"); sl@0: #endif sl@0: } sl@0: sl@0: OIL_DEFINE_IMPL_FULL (conv_f32_s16_3dnow, conv_f32_s16, OIL_IMPL_FLAG_MMX|OIL_IMPL_FLAG_3DNOW); sl@0: sl@0: #ifdef ENABLE_BROKEN_IMPLS sl@0: /* suboptimal */ sl@0: /* This appears to fail because of differences in rounding of half sl@0: * integers. */ sl@0: static void sl@0: conv_s32_f32_3dnow (int32_t * dst, int dst_stride, float *src, int src_stride, sl@0: int n) sl@0: { sl@0: int i; sl@0: const float constants[][2] = { sl@0: { 0.5, 0.5}, sl@0: {-1.0, -1.0} sl@0: }; sl@0: sl@0: for (i = 0; i < n; i++) { sl@0: asm volatile ( sl@0: " movq 0(%0), %%mm0 \n" sl@0: " pxor %%mm1, %%mm1 \n" sl@0: " pfcmpgt %%mm0, %%mm1 \n" sl@0: " movq %%mm0, %%mm2 \n" sl@0: " pfadd 0(%2), %%mm0 \n" sl@0: " pfcmpeq %%mm0, %%mm2 \n" sl@0: " pandn %%mm1, %%mm2 \n" sl@0: " pf2id %%mm0, %%mm0 \n" sl@0: " paddd %%mm2, %%mm0 \n" sl@0: " movd %%mm0, 0(%1) \n" sl@0: : sl@0: :"r" (src), "r" (dst), "r" (constants) sl@0: ); sl@0: sl@0: dst = OIL_OFFSET (dst, dst_stride); sl@0: src = OIL_OFFSET (src, src_stride); sl@0: } sl@0: asm volatile ("emms"); sl@0: } sl@0: sl@0: OIL_DEFINE_IMPL_FULL (conv_s32_f32_3dnow, conv_s32_f32, OIL_IMPL_FLAG_MMX|OIL_IMPL_FLAG_3DNOW); sl@0: #endif sl@0: sl@0: sl@0: sl@0: #ifdef __SYMBIAN32__ sl@0: sl@0: OilFunctionImpl* __oil_function_impl_conv_f32_s16_3dnow, conv_f32_s16() { sl@0: return &_oil_function_impl_conv_f32_s16_3dnow, conv_f32_s16; sl@0: } sl@0: #endif sl@0: sl@0: #ifdef __SYMBIAN32__ sl@0: sl@0: OilFunctionImpl* __oil_function_impl_conv_s32_f32_3dnow, conv_s32_f32() { sl@0: return &_oil_function_impl_conv_s32_f32_3dnow, conv_s32_f32; sl@0: } sl@0: #endif sl@0: