Symaptic: os/ossrv/genericopenlibs/liboil/src/math

     1 /*

     2  * Copyright (c) 2005

     3  *	Eric Anholt.  All rights reserved.

     4  *

     5  * Redistribution and use in source and binary forms, with or without

     6  * modification, are permitted provided that the following conditions

     7  * are met:

     8  * 1. Redistributions of source code must retain the above copyright

     9  *    notice, this list of conditions and the following disclaimer.

    10  * 2. Redistributions in binary form must reproduce the above copyright

    11  *    notice, this list of conditions and the following disclaimer in the

    12  *    documentation and/or other materials provided with the distribution.

    13  *

    14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND

    15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE

    16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE

    17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR BE LIABLE

    18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL

    19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS

    20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)

    21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT

    22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY

    23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF

    24  * SUCH DAMAGE.

    25  */

    26 //Portions Copyright (c)  2008-2009 Nokia Corporation and/or its subsidiary(-ies). All rights reserved.

    28 #ifdef HAVE_CONFIG_H

    29 #include "config.h"

    30 #endif

    31 #include <liboilclasses.h>

    32 #include <liboilfunction.h>

    33 #include <emmintrin.h>

    34 #include <xmmintrin.h>

    36 #define SSE_FUNCTION __attribute__((force_align_arg_pointer))

    38 SSE_FUNCTION static void

    39 add_f32_sse (float *dest, float *src1, float *src2, int n)

    40 {

    41   /* Initial operations to align the destination pointer */

    42   for (; ((long)dest & 15) && (n > 0); n--) {

    43     *dest++ = *src1++ + *src2++;

    44   }

    45   for (; n >= 4; n -= 4) {

    46     __m128 xmm0, xmm1;

    47     xmm0 = _mm_loadu_ps(src1);

    48     xmm1 = _mm_loadu_ps(src2);

    49     xmm0 = _mm_add_ps(xmm0, xmm1);

    50     _mm_store_ps(dest, xmm0);

    51     dest += 4;

    52     src1 += 4;

    53     src2 += 4;

    54   }

    55   for (; n > 0; n--) {

    56     *dest++ = *src1++ + *src2++;

    57   }

    58 }

    59 OIL_DEFINE_IMPL_FULL (add_f32_sse, add_f32, OIL_IMPL_FLAG_SSE);

    61 SSE_FUNCTION static void

    62 add_f64_sse2 (double *dest, double *src1, double *src2, int n)

    63 {

    64   __m128d xmm0, xmm1;

    65   while (((long)dest & 15) && (0 < n)) {

    66     *dest++ = *src1++ + *src2++;

    67     n--;

    68   }

    69   while (1 < n) {

    70     xmm0 = _mm_loadu_pd(src1);

    71     xmm1 = _mm_loadu_pd(src2);

    72     xmm0 = _mm_add_pd(xmm0, xmm1);

    73     _mm_store_pd(dest, xmm0);

    74     dest += 2;

    75     src1 += 2;

    76     src2 += 2;

    77     n -= 2;

    78   }

    79   while (0 < n) {

    80     *dest++ = *src1++ + *src2++;

    81     n--;

    82   }

    83 }

    84 OIL_DEFINE_IMPL_FULL (add_f64_sse2, add_f64, OIL_IMPL_FLAG_SSE2);

    86 SSE_FUNCTION static void

    87 add_f64_sse2_unroll (double *dest, double *src1, double *src2, int n)

    88 {

    89   __m128d xmm0, xmm1;

    90   while (((long)dest & 15) && (0 < n)) {

    91     *dest++ = *src1++ + *src2++;

    92     n--;

    93   }

    94   while (3 < n) {

    95     xmm0 = _mm_loadu_pd(src1);

    96     xmm1 = _mm_loadu_pd(src2);

    97     xmm0 = _mm_add_pd(xmm0, xmm1);

    98     _mm_store_pd(dest, xmm0);

   100     xmm0 = _mm_loadu_pd(src1+2);

   101     xmm1 = _mm_loadu_pd(src2+2);

   102     xmm0 = _mm_add_pd(xmm0, xmm1);

   103     _mm_store_pd(dest+2, xmm0);

   104     dest += 4;

   105     src1 += 4;

   106     src2 += 4;

   107     n -= 4;

   108   }

   109   while (1 < n) {

   110     xmm0 = _mm_loadu_pd(src1);

   111     xmm1 = _mm_loadu_pd(src2);

   112     xmm0 = _mm_add_pd(xmm0, xmm1);

   113     _mm_store_pd(dest, xmm0);

   114     dest += 2;

   115     src1 += 2;

   116     src2 += 2;

   117     n -= 2;

   118   }

   119   while (0 < n) {

   120     *dest++ = *src1++ + *src2++;

   121     n--;

   122   }

   123 }

   124 OIL_DEFINE_IMPL_FULL (add_f64_sse2_unroll, add_f64, OIL_IMPL_FLAG_SSE2);

   126 SSE_FUNCTION static void

   127 subtract_f32_sse (float *dest, float *src1, float *src2, int n)

   128 {

   129   /* Initial operations to align the destination pointer */

   130   for (; ((long)dest & 15) && (n > 0); n--) {

   131     *dest++ = *src1++ - *src2++;

   132   }

   133   for (; n >= 4; n -= 4) {

   134     __m128 xmm0, xmm1;

   135     xmm0 = _mm_loadu_ps(src1);

   136     xmm1 = _mm_loadu_ps(src2);

   137     xmm0 = _mm_sub_ps(xmm0, xmm1);

   138     _mm_store_ps(dest, xmm0);

   139     dest += 4;

   140     src1 += 4;

   141     src2 += 4;

   142   }

   143   for (; n > 0; n--) {

   144     *dest++ = *src1++ - *src2++;

   145   }

   146 }

   147 OIL_DEFINE_IMPL_FULL (subtract_f32_sse, subtract_f32, OIL_IMPL_FLAG_SSE);

   149 SSE_FUNCTION static void

   150 multiply_f32_sse (float *dest, float *src1, float *src2, int n)

   151 {

   152   /* Initial operations to align the destination pointer */

   153   for (; ((long)dest & 15) && (n > 0); n--) {

   154     *dest++ = *src1++ * *src2++;

   155   }

   156   for (; n >= 4; n -= 4) {

   157     __m128 xmm0, xmm1;

   158     xmm0 = _mm_loadu_ps(src1);

   159     xmm1 = _mm_loadu_ps(src2);

   160     xmm0 = _mm_mul_ps(xmm0, xmm1);

   161     _mm_store_ps(dest, xmm0);

   162     dest += 4;

   163     src1 += 4;

   164     src2 += 4;

   165   }

   166   for (; n > 0; n--) {

   167     *dest++ = *src1++ * *src2++;

   168   }

   169 }

   170 OIL_DEFINE_IMPL_FULL (multiply_f32_sse, multiply_f32, OIL_IMPL_FLAG_SSE);

   172 SSE_FUNCTION static void

   173 divide_f32_sse (float *dest, float *src1, float *src2, int n)

   174 {

   175   /* Initial operations to align the destination pointer */

   176   for (; ((long)dest & 15) && (n > 0); n--) {

   177     *dest++ = *src1++ / *src2++;

   178   }

   179   for (; n >= 4; n -= 4) {

   180     __m128 xmm0, xmm1;

   181     xmm0 = _mm_loadu_ps(src1);

   182     xmm1 = _mm_loadu_ps(src2);

   183     xmm0 = _mm_div_ps(xmm0, xmm1);

   184     _mm_store_ps(dest, xmm0);

   185     dest += 4;

   186     src1 += 4;

   187     src2 += 4;

   188   }

   189   for (; n > 0; n--) {

   190     *dest++ = *src1++ / *src2++;

   191   }

   192 }

   193 OIL_DEFINE_IMPL_FULL (divide_f32_sse, divide_f32, OIL_IMPL_FLAG_SSE);

   195 SSE_FUNCTION static void

   196 minimum_f32_sse (float *dest, float *src1, float *src2, int n)

   197 {

   198   /* Initial operations to align the destination pointer */

   199   for (; ((long)dest & 15) && (n > 0); n--) {

   200     *dest++ = *src1 < *src2 ? *src1 : *src2;

   201     src1++;

   202     src2++;

   203   }

   204   for (; n >= 4; n -= 4) {

   205     __m128 xmm0, xmm1;

   206     xmm0 = _mm_loadu_ps(src1);

   207     xmm1 = _mm_loadu_ps(src2);

   208     xmm0 = _mm_min_ps(xmm0, xmm1);

   209     _mm_store_ps(dest, xmm0);

   210     dest += 4;

   211     src1 += 4;

   212     src2 += 4;

   213   }

   214   for (; n > 0; n--) {

   215     *dest++ = *src1 < *src2 ? *src1 : *src2;

   216     src1++;

   217     src2++;

   218   }

   219 }

   220 OIL_DEFINE_IMPL_FULL (minimum_f32_sse, minimum_f32, OIL_IMPL_FLAG_SSE);

   222 SSE_FUNCTION static void

   223 maximum_f32_sse (float *dest, float *src1, float *src2, int n)

   224 {

   225   /* Initial operations to align the destination pointer */

   226   for (; ((long)dest & 15) && (n > 0); n--) {

   227     *dest++ = *src1 > *src2 ? *src1 : *src2;

   228     src1++;

   229     src2++;

   230   }

   231   for (; n >= 4; n -= 4) {

   232     __m128 xmm0, xmm1;

   233     xmm0 = _mm_loadu_ps(src1);

   234     xmm1 = _mm_loadu_ps(src2);

   235     xmm0 = _mm_max_ps(xmm0, xmm1);

   236     _mm_store_ps(dest, xmm0);

   237     dest += 4;

   238     src1 += 4;

   239     src2 += 4;

   240   }

   241   for (; n > 0; n--) {

   242     *dest++ = *src1 > *src2 ? *src1 : *src2;

   243     src1++;

   244     src2++;

   245   }

   246 }

   247 OIL_DEFINE_IMPL_FULL (maximum_f32_sse, maximum_f32, OIL_IMPL_FLAG_SSE);

   249 SSE_FUNCTION static void

   250 inverse_f32_sse (float *dest, float *src1, int n)

   251 {

   252   /* Initial operations to align the destination pointer */

   253   for (; ((long)dest & 15) && (n > 0); n--) {

   254     *dest++ = 1.0 / *src1++;

   255   }

   256   for (; n >= 4; n -= 4) {

   257     __m128 xmm0, xmm1;

   258     /* While _mm_rcp_ps sounds promising, the results it gives are rather

   259      * different from the 1.0 / src1 reference implementation, so do that.

   260      */

   261     xmm0 = _mm_set_ps1(1.0);

   262     xmm1 = _mm_loadu_ps(src1);

   263     xmm0 = _mm_div_ps(xmm0, xmm1);

   264     _mm_store_ps(dest, xmm0);

   265     dest += 4;

   266     src1 += 4;

   267   }

   268   for (; n > 0; n--) {

   269     *dest++ = 1.0 / *src1++;

   270   }

   271 }

   272 OIL_DEFINE_IMPL_FULL (inverse_f32_sse, inverse_f32, OIL_IMPL_FLAG_SSE);

   274 SSE_FUNCTION static void

   275 negative_f32_sse (float *dest, float *src1, int n)

   276 {

   277   /* Initial operations to align the destination pointer */

   278   for (; ((long)dest & 15) && (n > 0); n--) {

   279     *dest++ = -(*src1++);

   280   }

   281   for (; n >= 4; n -= 4) {

   282     __m128 xmm0, xmm1;

   283     xmm0 = _mm_setzero_ps();

   284     xmm1 = _mm_loadu_ps(src1);

   285     xmm0 = _mm_sub_ps(xmm0, xmm1);

   286     _mm_store_ps(dest, xmm0);

   287     dest += 4;

   288     src1 += 4;

   289   }

   290   for (; n > 0; n--) {

   291     *dest++ = -(*src1++);

   292   }

   293 }

   294 OIL_DEFINE_IMPL_FULL (negative_f32_sse, negative_f32, OIL_IMPL_FLAG_SSE);

   296 SSE_FUNCTION static void

   297 scalaradd_f32_ns_sse (float *dest, float *src1, float *val, int n)

   298 {

   299   __m128 xmm1;

   301   /* Initial operations to align the destination pointer */

   302   for (; ((long)dest & 15) && (n > 0); n--) {

   303     *dest++ = *src1++ + *val;

   304   }

   305   xmm1 = _mm_load_ps1(val);

   306   for (; n >= 4; n -= 4) {

   307     __m128 xmm0;

   308     xmm0 = _mm_loadu_ps(src1);

   309     xmm0 = _mm_add_ps(xmm0, xmm1);

   310     _mm_store_ps(dest, xmm0);

   311     dest += 4;

   312     src1 += 4;

   313   }

   314   for (; n > 0; n--) {

   315     *dest++ = *src1++ + *val;

   316   }

   317 }

   318 OIL_DEFINE_IMPL_FULL (scalaradd_f32_ns_sse, scalaradd_f32_ns, OIL_IMPL_FLAG_SSE);

   320 SSE_FUNCTION static void

   321 scalarmultiply_f32_ns_sse (float *dest, float *src1, float *val, int n)

   322 {

   323   __m128 xmm1;

   325   /* Initial operations to align the destination pointer */

   326   for (; ((long)dest & 15) && (n > 0); n--) {

   327     *dest++ = *src1++ * *val;

   328   }

   329   xmm1 = _mm_load_ps1(val);

   330   for (; n >= 4; n -= 4) {

   331     __m128 xmm0;

   332     xmm0 = _mm_loadu_ps(src1);

   333     xmm0 = _mm_mul_ps(xmm0, xmm1);

   334     _mm_store_ps(dest, xmm0);

   335     dest += 4;

   336     src1 += 4;

   337   }

   338   for (; n > 0; n--) {

   339     *dest++ = *src1++ * *val;

   340   }

   341 }

   342 OIL_DEFINE_IMPL_FULL (scalarmultiply_f32_ns_sse, scalarmultiply_f32_ns, OIL_IMPL_FLAG_SSE);

   344 SSE_FUNCTION static void

   345 scalarmultiply_f64_ns_sse2 (double *dest, double *src1, double *val, int n)

   346 {

   347   __m128d xmm1;

   349   /* Initial operations to align the destination pointer */

   350   for (; ((long)dest & 15) && (n > 0); n--) {

   351     *dest++ = *src1++ * *val;

   352   }

   353   xmm1 = _mm_load_pd1(val);

   354   for (; n >= 2; n -= 2) {

   355     __m128d xmm0;

   356     xmm0 = _mm_loadu_pd(src1);

   357     xmm0 = _mm_mul_pd(xmm0, xmm1);

   358     _mm_store_pd(dest, xmm0);

   359     dest += 2;

   360     src1 += 2;

   361   }

   362   for (; n > 0; n--) {

   363     *dest++ = *src1++ * *val;

   364   }

   365 }

   366 OIL_DEFINE_IMPL_FULL (scalarmultiply_f64_ns_sse2, scalarmultiply_f64_ns, OIL_IMPL_FLAG_SSE2);

   370 #ifdef	__SYMBIAN32__

   372 OilFunctionImpl* __oil_function_impl_add_f32_sse, add_f32() {

   373 		return &_oil_function_impl_add_f32_sse, add_f32;

   374 }

   375 #endif

   377 #ifdef	__SYMBIAN32__

   379 OilFunctionImpl* __oil_function_impl_add_f64_sse2, add_f64() {

   380 		return &_oil_function_impl_add_f64_sse2, add_f64;

   381 }

   382 #endif

   384 #ifdef	__SYMBIAN32__

   386 OilFunctionImpl* __oil_function_impl_add_f64_sse2_unroll, add_f64() {

   387 		return &_oil_function_impl_add_f64_sse2_unroll, add_f64;

   388 }

   389 #endif

   391 #ifdef	__SYMBIAN32__

   393 OilFunctionImpl* __oil_function_impl_subtract_f32_sse, subtract_f32() {

   394 		return &_oil_function_impl_subtract_f32_sse, subtract_f32;

   395 }

   396 #endif

   398 #ifdef	__SYMBIAN32__

   400 OilFunctionImpl* __oil_function_impl_multiply_f32_sse, multiply_f32() {

   401 		return &_oil_function_impl_multiply_f32_sse, multiply_f32;

   402 }

   403 #endif

   405 #ifdef	__SYMBIAN32__

   407 OilFunctionImpl* __oil_function_impl_divide_f32_sse, divide_f32() {

   408 		return &_oil_function_impl_divide_f32_sse, divide_f32;

   409 }

   410 #endif

   412 #ifdef	__SYMBIAN32__

   414 OilFunctionImpl* __oil_function_impl_minimum_f32_sse, minimum_f32() {

   415 		return &_oil_function_impl_minimum_f32_sse, minimum_f32;

   416 }

   417 #endif

   419 #ifdef	__SYMBIAN32__

   421 OilFunctionImpl* __oil_function_impl_maximum_f32_sse, maximum_f32() {

   422 		return &_oil_function_impl_maximum_f32_sse, maximum_f32;

   423 }

   424 #endif

   426 #ifdef	__SYMBIAN32__

   428 OilFunctionImpl* __oil_function_impl_inverse_f32_sse, inverse_f32() {

   429 		return &_oil_function_impl_inverse_f32_sse, inverse_f32;

   430 }

   431 #endif

   433 #ifdef	__SYMBIAN32__

   435 OilFunctionImpl* __oil_function_impl_negative_f32_sse, negative_f32() {

   436 		return &_oil_function_impl_negative_f32_sse, negative_f32;

   437 }

   438 #endif

   440 #ifdef	__SYMBIAN32__

   442 OilFunctionImpl* __oil_function_impl_scalaradd_f32_ns_sse, scalaradd_f32_ns() {

   443 		return &_oil_function_impl_scalaradd_f32_ns_sse, scalaradd_f32_ns;

   444 }

   445 #endif

   447 #ifdef	__SYMBIAN32__

   449 OilFunctionImpl* __oil_function_impl_scalarmultiply_f32_ns_sse, scalarmultiply_f32_ns() {

   450 		return &_oil_function_impl_scalarmultiply_f32_ns_sse, scalarmultiply_f32_ns;

   451 }

   452 #endif

   454 #ifdef	__SYMBIAN32__

   456 OilFunctionImpl* __oil_function_impl_scalarmultiply_f64_ns_sse2, scalarmultiply_f64_ns() {

   457 		return &_oil_function_impl_scalarmultiply_f64_ns_sse2, scalarmultiply_f64_ns;

   458 }

   459 #endif

author	sl@SLION-WIN7.fritz.box
	Fri, 15 Jun 2012 03:10:57 +0200
changeset 0	bde4ae8d615e
permissions	-rw-r--r--