SSE2 & SSE4.1 versions of GimpOperationNormalMode

* Add configure checks for SSE intrinsic code.
* Use SSE helper libraries in app/operations to compile
  code with different CFLAGS.
This commit is contained in:
Daniel Sabo 2013-06-25 03:25:19 -07:00
parent ff9c91ce6f
commit cccc3550af
7 changed files with 352 additions and 19 deletions

View File

@ -3,3 +3,6 @@
/.deps
/.libs
/libappoperations.a
/libappoperations-generic.a
/libappoperations-sse2.a
/libappoperations-sse4.a

View File

@ -12,9 +12,12 @@ AM_CPPFLAGS = \
$(GDK_PIXBUF_CFLAGS) \
-I$(includedir)
noinst_LIBRARIES = libappoperations.a
noinst_LIBRARIES = libappoperations-generic.a \
libappoperations-sse2.a \
libappoperations-sse4.a \
libappoperations.a
libappoperations_a_sources = \
libappoperations_generic_a_sources = \
operations-types.h \
gimp-operations.c \
gimp-operations.h \
@ -144,4 +147,28 @@ libappoperations_a_sources = \
gimplayermodefunctions.c \
gimplayermodefunctions.h
libappoperations_a_SOURCES = $(libappoperations_a_sources)
libappoperations_sse2_a_sources = \
gimpoperationnormalmode-sse2.c
libappoperations_sse4_a_sources = \
gimpoperationnormalmode-sse4.c
libappoperations_sse2_a_SOURCES = $(libappoperations_sse2_a_sources)
libappoperations_sse2_a_CFLAGS = $(SSE2_EXTRA_CFLAGS)
libappoperations_sse4_a_SOURCES = $(libappoperations_sse4_a_sources)
libappoperations_sse4_a_CFLAGS = $(SSE4_1_EXTRA_CFLAGS)
libappoperations_generic_a_SOURCES = $(libappoperations_generic_a_sources)
libappoperations.a: libappoperations-generic.a \
libappoperations-sse2.a \
libappoperations-sse4.a
$(AR) $(ARFLAGS) libappoperations.a \
$(libappoperations_generic_a_OBJECTS) \
$(libappoperations_sse2_a_OBJECTS) \
$(libappoperations_sse4_a_OBJECTS)
$(RANLIB) libappoperations.a

View File

@ -0,0 +1,117 @@
/* GIMP - The GNU Image Manipulation Program
* Copyright (C) 1995 Spencer Kimball and Peter Mattis
*
* gimpoperationnormalmode-sse2.c
* Copyright (C) 2013 Daniel Sabo
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; either version 2 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software
* Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
*/
#include "config.h"
#include <gegl-plugin.h>
#include "operations-types.h"
#include "gimpoperationnormalmode.h"
#if COMPILE_SSE2_INTRINISICS
/* SSE2 */
#include <emmintrin.h>
gboolean
gimp_operation_normal_mode_process_pixels_sse2 (gfloat *in,
gfloat *aux,
gfloat *mask,
gfloat *out,
gfloat opacity,
glong samples,
const GeglRectangle *roi,
gint level)
{
/* check alignment */
if ((((uintptr_t)in) | ((uintptr_t)aux) | ((uintptr_t)out)) & 0x0F)
{
return gimp_operation_normal_mode_process_pixels_core (in, aux, mask, out,
opacity, samples,
roi, level);
}
else
{
const __v4sf *v_in = (const __v4sf*) in;
const __v4sf *v_aux = (const __v4sf*) aux;
__v4sf *v_out = ( __v4sf*) out;
const __v4sf one = _mm_set1_ps (1.0f);
const __v4sf v_opacity = _mm_set1_ps (opacity);
while (samples--)
{
__v4sf rgba_in, rgba_aux, alpha;
rgba_in = *v_in++;
rgba_aux = *v_aux++;
/* expand alpha */
alpha = (__v4sf)_mm_shuffle_epi32 ((__m128i)rgba_aux,
_MM_SHUFFLE (3, 3, 3, 3));
if (mask)
{
__v4sf mask_alpha;
/* multiply aux's alpha by the mask */
mask_alpha = _mm_set1_ps (*mask++);
alpha = alpha * mask_alpha;
}
alpha = alpha * v_opacity;
if (_mm_ucomigt_ss (alpha, _mm_setzero_ps ()))
{
__v4sf dst_alpha, a_term, out_pixel, out_alpha, out_pixel_rbaa;
/* expand alpha */
dst_alpha = (__v4sf)_mm_shuffle_epi32 ((__m128i)rgba_in,
_MM_SHUFFLE (3, 3, 3, 3));
/* a_term = dst_a * (1.0 - src_a) */
a_term = dst_alpha * (one - alpha);
/* out(color) = src * src_a + dst * a_term */
out_pixel = rgba_aux * alpha + rgba_in * a_term;
/* out(alpha) = 1.0 * src_a + 1.0 * a_term */
out_alpha = alpha + a_term;
/* un-premultiply */
out_pixel = out_pixel / out_alpha;
/* swap in the real alpha */
out_pixel_rbaa = _mm_shuffle_ps (out_pixel, out_alpha, _MM_SHUFFLE (3, 3, 2, 0));
out_pixel = _mm_shuffle_ps (out_pixel, out_pixel_rbaa, _MM_SHUFFLE (2, 1, 1, 0));
*v_out++ = out_pixel;
}
else
{
*v_out++ = rgba_in;
}
}
}
return TRUE;
}
#endif /* COMPILE_SSE2_INTRINISICS */

View File

@ -0,0 +1,116 @@
/* GIMP - The GNU Image Manipulation Program
* Copyright (C) 1995 Spencer Kimball and Peter Mattis
*
* gimpoperationnormalmode-sse2.c
* Copyright (C) 2013 Daniel Sabo
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; either version 2 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software
* Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
*/
#include "config.h"
#include <gegl-plugin.h>
#include "operations-types.h"
#include "gimpoperationnormalmode.h"
#if COMPILE_SSE4_1_INTRINISICS
/* SSE4 */
#include <smmintrin.h>
gboolean
gimp_operation_normal_mode_process_pixels_sse4 (gfloat *in,
gfloat *aux,
gfloat *mask,
gfloat *out,
gfloat opacity,
glong samples,
const GeglRectangle *roi,
gint level)
{
/* check alignment */
if ((((uintptr_t)in) | ((uintptr_t)aux) | ((uintptr_t)out)) & 0x0F)
{
return gimp_operation_normal_mode_process_pixels_core (in, aux, mask, out,
opacity, samples,
roi, level);
}
else
{
const __v4sf *v_in = (const __v4sf*) in;
const __v4sf *v_aux = (const __v4sf*) aux;
__v4sf *v_out = ( __v4sf*) out;
const __v4sf one = _mm_set1_ps (1.0f);
const __v4sf v_opacity = _mm_set1_ps (opacity);
while (samples--)
{
__v4sf rgba_in, rgba_aux, alpha;
rgba_in = *v_in++;
rgba_aux = *v_aux++;
/* expand alpha */
alpha = (__v4sf)_mm_shuffle_epi32 ((__m128i)rgba_aux,
_MM_SHUFFLE (3, 3, 3, 3));
if (mask)
{
__v4sf mask_alpha;
/* multiply aux's alpha by the mask */
mask_alpha = _mm_set1_ps (*mask++);
alpha = alpha * mask_alpha;
}
alpha = alpha * v_opacity;
if (_mm_ucomigt_ss (alpha, _mm_setzero_ps ()))
{
__v4sf dst_alpha, a_term, out_pixel, out_alpha;
/* expand alpha */
dst_alpha = (__v4sf)_mm_shuffle_epi32 ((__m128i)rgba_in,
_MM_SHUFFLE (3, 3, 3, 3));
/* a_term = dst_a * (1.0 - src_a) */
a_term = dst_alpha * (one - alpha);
/* out(color) = src * src_a + dst * a_term */
out_pixel = rgba_aux * alpha + rgba_in * a_term;
/* out(alpha) = 1.0 * src_a + 1.0 * a_term */
out_alpha = alpha + a_term;
/* un-premultiply */
out_pixel = out_pixel / out_alpha;
/* swap in the real alpha */
out_pixel = _mm_blend_ps (out_pixel, out_alpha, 0x08);
*v_out++ = out_pixel;
}
else
{
*v_out++ = rgba_in;
}
}
}
return TRUE;
}
#endif /* COMPILE_SSE4_1_INTRINISICS */

View File

@ -23,10 +23,14 @@
#include <gegl-plugin.h>
#include <libgimpbase/gimpbase.h>
#include "operations-types.h"
#include "gimpoperationnormalmode.h"
GimpLayerModeFunction gimp_operation_normal_mode_process_pixels = NULL;
static gboolean gimp_operation_normal_parent_process (GeglOperation *operation,
GeglOperationContext *context,
@ -84,6 +88,18 @@ gimp_operation_normal_mode_class_init (GimpOperationNormalModeClass *klass)
operation_class->process = gimp_operation_normal_parent_process;
point_class->process = gimp_operation_normal_mode_process;
gimp_operation_normal_mode_process_pixels = gimp_operation_normal_mode_process_pixels_core;
#if COMPILE_SSE2_INTRINISICS
if (gimp_cpu_accel_get_support() & GIMP_CPU_ACCEL_X86_SSE2)
gimp_operation_normal_mode_process_pixels = gimp_operation_normal_mode_process_pixels_sse2;
#endif /* COMPILE_SSE2_INTRINISICS */
#if COMPILE_SSE4_1_INTRINISICS
if (gimp_cpu_accel_get_support() & GIMP_CPU_ACCEL_X86_SSE4_1)
gimp_operation_normal_mode_process_pixels = gimp_operation_normal_mode_process_pixels_sse4;
#endif /* COMPILE_SSE4_1_INTRINISICS */
}
static void
@ -162,14 +178,14 @@ gimp_operation_normal_mode_process (GeglOperation *operation,
}
gboolean
gimp_operation_normal_mode_process_pixels (gfloat *in,
gfloat *aux,
gfloat *mask,
gfloat *out,
gfloat opacity,
glong samples,
const GeglRectangle *roi,
gint level)
gimp_operation_normal_mode_process_pixels_core (gfloat *in,
gfloat *aux,
gfloat *mask,
gfloat *out,
gfloat opacity,
glong samples,
const GeglRectangle *roi,
gint level)
{
const gboolean has_mask = mask != NULL;

View File

@ -50,15 +50,33 @@ struct _GimpOperationNormalModeClass
GType gimp_operation_normal_mode_get_type (void) G_GNUC_CONST;
extern GimpLayerModeFunction gimp_operation_normal_mode_process_pixels;
gboolean gimp_operation_normal_mode_process_pixels (gfloat *in,
gfloat *aux,
gfloat *mask,
gfloat *out,
gfloat opacity,
glong samples,
const GeglRectangle *roi,
gint level);
gboolean gimp_operation_normal_mode_process_pixels_core (gfloat *in,
gfloat *aux,
gfloat *mask,
gfloat *out,
gfloat opacity,
glong samples,
const GeglRectangle *roi,
gint level);
gboolean gimp_operation_normal_mode_process_pixels_sse2 (gfloat *in,
gfloat *aux,
gfloat *mask,
gfloat *out,
gfloat opacity,
glong samples,
const GeglRectangle *roi,
gint level);
gboolean gimp_operation_normal_mode_process_pixels_sse4 (gfloat *in,
gfloat *aux,
gfloat *mask,
gfloat *out,
gfloat opacity,
glong samples,
const GeglRectangle *roi,
gint level);
#endif /* __GIMP_OPERATION_NORMAL_MODE_H__ */

View File

@ -686,6 +686,42 @@ LIBS=$gimp_save_LIBS
AC_SUBST(SOCKET_LIBS)
###################################
# Check for Intel vector intrinsics
###################################
intrinsics_save_CFLAGS="$CFLAGS"
#FIXME: Check the CFLAGS separatly
GIMP_DETECT_CFLAGS(SSE_MATH_CFLAG, '-mfpmath=sse')
GIMP_DETECT_CFLAGS(SSE2_CFLAG, '-msse2')
SSE2_EXTRA_CFLAGS="$SSE_MATH_CFLAG $SSE2_CFLAG"
CFLAGS="$intrinsics_save_CFLAGS $SSE2_EXTRA_CFLAGS"
AC_MSG_CHECKING(whether we can compile SSE2 intrinsics)
AC_LINK_IFELSE([AC_LANG_PROGRAM([#include <emmintrin.h>],[__m128i one = _mm_set1_epi32 (1);])],
AC_DEFINE(COMPILE_SSE2_INTRINISICS, 1, [Define to 1 if SSE2 intrinsics are available.])
AC_SUBST(SSE2_EXTRA_CFLAGS)
AC_MSG_RESULT(yes)
,
AC_MSG_RESULT(no)
AC_MSG_WARN([SSE2 intrinsics not available.])
)
GIMP_DETECT_CFLAGS(SSE4_1_CFLAG, '-msse4.1')
SSE4_1_EXTRA_CFLAGS="$SSE_MATH_CFLAG $SSE4_1_CFLAG"
CFLAGS="$intrinsics_save_CFLAGS $SSE4_1_EXTRA_CFLAGS"
AC_MSG_CHECKING(whether we can compile SSE4.1 intrinsics)
AC_LINK_IFELSE([AC_LANG_PROGRAM([#include <smmintrin.h>],[__v4sf a, b, c; c = _mm_blend_ps(a, b, 0x08);])],
AC_DEFINE(COMPILE_SSE4_1_INTRINISICS, 1, [Define to 1 if SSE4.1 intrinsics are available.])
AC_SUBST(SSE4_1_EXTRA_CFLAGS)
AC_MSG_RESULT(yes)
,
AC_MSG_RESULT(no)
AC_MSG_WARN([SSE4.1 intrinsics not available.])
)
CFLAGS="$intrinsics_save_CFLAGS"
########################
# Check for MMX assembly