1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/os/ossrv/genericopenlibs/liboil/src/splat_sse.c Fri Jun 15 03:10:57 2012 +0200
1.3 @@ -0,0 +1,155 @@
1.4 +/*
1.5 + * Copyright (c) 2005
1.6 + * Eric Anholt. All rights reserved.
1.7 + *
1.8 + * Redistribution and use in source and binary forms, with or without
1.9 + * modification, are permitted provided that the following conditions
1.10 + * are met:
1.11 + * 1. Redistributions of source code must retain the above copyright
1.12 + * notice, this list of conditions and the following disclaimer.
1.13 + * 2. Redistributions in binary form must reproduce the above copyright
1.14 + * notice, this list of conditions and the following disclaimer in the
1.15 + * documentation and/or other materials provided with the distribution.
1.16 + *
1.17 + * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND
1.18 + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
1.19 + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
1.20 + * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR BE LIABLE
1.21 + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
1.22 + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
1.23 + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
1.24 + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
1.25 + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
1.26 + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
1.27 + * SUCH DAMAGE.
1.28 + */
1.29 +//Portions Copyright (c) 2008-2009 Nokia Corporation and/or its subsidiary(-ies). All rights reserved.
1.30 +
1.31 +#ifdef HAVE_CONFIG_H
1.32 +#include "config.h"
1.33 +#endif
1.34 +#include <liboil/liboilclasses.h>
1.35 +#include <liboil/liboilfunction.h>
1.36 +#include <emmintrin.h>
1.37 +
1.38 +#define SSE_FUNCTION __attribute__((force_align_arg_pointer))
1.39 +
1.40 +SSE_FUNCTION static void
1.41 +splat_u32_ns_sse (uint32_t *dest, const uint32_t *param, int n)
1.42 +{
1.43 + __m128i v;
1.44 +
1.45 + v = _mm_set1_epi32(*param);
1.46 +
1.47 + for (; ((long)dest & 15) && (n > 0); n--) {
1.48 + *dest++ = *param;
1.49 + }
1.50 + for (; n >= 4; n -= 4) {
1.51 + _mm_store_si128((__m128i *)dest, v);
1.52 + dest += 4;
1.53 + }
1.54 + for (; n > 0; n--) {
1.55 + *dest++ = *param;
1.56 + }
1.57 +}
1.58 +OIL_DEFINE_IMPL_FULL (splat_u32_ns_sse, splat_u32_ns, OIL_IMPL_FLAG_SSE2);
1.59 +
1.60 +SSE_FUNCTION static void
1.61 +splat_u32_ns_sse_unroll2 (uint32_t *dest, const uint32_t *param, int n)
1.62 +{
1.63 + __m128i v;
1.64 +
1.65 + v = _mm_set1_epi32(*param);
1.66 +
1.67 + for (; ((long)dest & 15) && (n > 0); n--) {
1.68 + *dest++ = *param;
1.69 + }
1.70 + for (; n >= 8; n -= 8) {
1.71 + _mm_store_si128((__m128i *)dest, v);
1.72 + _mm_store_si128((__m128i *)(dest + 4), v);
1.73 + dest += 8;
1.74 + }
1.75 + for (; n >= 4; n -= 4) {
1.76 + _mm_store_si128((__m128i *)dest, v);
1.77 + dest += 4;
1.78 + }
1.79 + for (; n > 0; n--) {
1.80 + *dest++ = *param;
1.81 + }
1.82 +}
1.83 +OIL_DEFINE_IMPL_FULL (splat_u32_ns_sse_unroll2, splat_u32_ns, OIL_IMPL_FLAG_SSE2);
1.84 +
1.85 +SSE_FUNCTION static void
1.86 +splat_u8_ns_sse (uint8_t *dest, const uint8_t *param, int n)
1.87 +{
1.88 + __m128i v;
1.89 +
1.90 + v = _mm_set1_epi8(*param);
1.91 +
1.92 + for (; ((long)dest & 15) && (n > 0); n--) {
1.93 + *dest++ = *param;
1.94 + }
1.95 + for (; n >= 16; n -= 16) {
1.96 + _mm_store_si128((__m128i *)dest, v);
1.97 + dest += 16;
1.98 + }
1.99 + for (; n > 0; n--) {
1.100 + *dest++ = *param;
1.101 + }
1.102 +}
1.103 +OIL_DEFINE_IMPL_FULL (splat_u8_ns_sse, splat_u8_ns, OIL_IMPL_FLAG_SSE2);
1.104 +
1.105 +SSE_FUNCTION static void
1.106 +splat_u8_ns_sse_unroll2 (uint8_t *dest, const uint8_t *param, int n)
1.107 +{
1.108 + __m128i v;
1.109 +
1.110 + v = _mm_set1_epi8(*param);
1.111 +
1.112 + for (; ((long)dest & 15) && (n > 0); n--) {
1.113 + *dest++ = *param;
1.114 + }
1.115 + for (; n >= 32; n -= 32) {
1.116 + _mm_store_si128((__m128i *)dest, v);
1.117 + _mm_store_si128((__m128i *)(dest + 16), v);
1.118 + dest += 32;
1.119 + }
1.120 + for (; n >= 16; n -= 16) {
1.121 + _mm_store_si128((__m128i *)dest, v);
1.122 + dest += 16;
1.123 + }
1.124 + for (; n > 0; n--) {
1.125 + *dest++ = *param;
1.126 + }
1.127 +}
1.128 +OIL_DEFINE_IMPL_FULL (splat_u8_ns_sse_unroll2, splat_u8_ns, OIL_IMPL_FLAG_SSE2);
1.129 +
1.130 +
1.131 +#ifdef __SYMBIAN32__
1.132 +
1.133 +OilFunctionImpl* __oil_function_impl_splat_u32_ns_sse, splat_u32_ns() {
1.134 + return &_oil_function_impl_splat_u32_ns_sse, splat_u32_ns;
1.135 +}
1.136 +#endif
1.137 +
1.138 +#ifdef __SYMBIAN32__
1.139 +
1.140 +OilFunctionImpl* __oil_function_impl_splat_u32_ns_sse_unroll2, splat_u32_ns() {
1.141 + return &_oil_function_impl_splat_u32_ns_sse_unroll2, splat_u32_ns;
1.142 +}
1.143 +#endif
1.144 +
1.145 +#ifdef __SYMBIAN32__
1.146 +
1.147 +OilFunctionImpl* __oil_function_impl_splat_u8_ns_sse, splat_u8_ns() {
1.148 + return &_oil_function_impl_splat_u8_ns_sse, splat_u8_ns;
1.149 +}
1.150 +#endif
1.151 +
1.152 +#ifdef __SYMBIAN32__
1.153 +
1.154 +OilFunctionImpl* __oil_function_impl_splat_u8_ns_sse_unroll2, splat_u8_ns() {
1.155 + return &_oil_function_impl_splat_u8_ns_sse_unroll2, splat_u8_ns;
1.156 +}
1.157 +#endif
1.158 +