Author: post
Date: 2009-12-31 00:14:23 +0100 (Thu, 31 Dec 2009)
New Revision: 2930
Added:
branches/rawstudio-ng-color/plugins/lensfun/lensfun-sse2.c
Modified:
branches/rawstudio-ng-color/plugins/lensfun/Makefile.am
branches/rawstudio-ng-color/plugins/lensfun/lensfun.c
Log:
Lensfun: Put SSE2 into separate file, so we don't risk SSE2 in lensfun.c
Modified: branches/rawstudio-ng-color/plugins/lensfun/Makefile.am
===================================================================
--- branches/rawstudio-ng-color/plugins/lensfun/Makefile.am 2009-12-30
22:39:52 UTC (rev 2929)
+++ branches/rawstudio-ng-color/plugins/lensfun/Makefile.am 2009-12-30
23:14:23 UTC (rev 2930)
@@ -1,15 +1,9 @@
plugindir = $(libdir)/rawstudio/plugins
-if CAN_COMPILE_SSE2
-SSE_FLAG=-msse2
-else
-SSE_FLAG=
-endif
-
AM_CFLAGS =\
-Wall\
-O4\
- $(SSE_FLAG)
+ -funroll-loops
AM_CXXFLAGS = $(AM_CFLAGS)
@@ -23,6 +17,16 @@
libdir = $(datadir)/rawstudio/plugins/
-lensfun_la_LIBADD = @PACKAGE_LIBS@
+lensfun_la_LIBADD = @PACKAGE_LIBS@ lensfun-sse2.lo
lensfun_la_LDFLAGS = -module -avoid-version
-lensfun_la_SOURCES = lensfun.c
+lensfun.lo: lensfun.c
+ $(LTCOMPILE) -DEXIT_CODE=0 -c lensfun.c
+
+lensfun-sse2.lo: lensfun-sse2.c
+if CAN_COMPILE_SSE2
+SSE_FLAG=-msse2
+else
+SSE_FLAG=
+endif
+ $(LTCOMPILE) $(SSE_FLAG) -DEXIT_CODE=1 -c lensfun-sse2.c
+
Added: branches/rawstudio-ng-color/plugins/lensfun/lensfun-sse2.c
===================================================================
--- branches/rawstudio-ng-color/plugins/lensfun/lensfun-sse2.c
(rev 0)
+++ branches/rawstudio-ng-color/plugins/lensfun/lensfun-sse2.c 2009-12-30
23:14:23 UTC (rev 2930)
@@ -0,0 +1,157 @@
+/*
+ * Copyright (C) 2006-2009 Anders Brander <[email protected]> and
+ * Anders Kvist <[email protected]>
+ *
+ * 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., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301,
USA.
+ */
+
+/* Plugin tmpl version 4 */
+
+#include <rawstudio.h>
+#include <lensfun.h>
+
+#if defined (__SSE2__)
+
+#include <emmintrin.h>
+
+static gfloat twofiftytwo_ps[4] __attribute__ ((aligned (16))) = {256.0f,
256.0f, 256.0f, 0.0f};
+
+gboolean is_sse2_compiled()
+{
+ return TRUE;
+}
+
+void
+rs_image16_bilinear_full_sse2(RS_IMAGE16 *in, gushort *out, gfloat *pos)
+{
+ const gint m_w = (in->w-1);
+ const gint m_h = (in->h-1);
+
+ __m128 p0, p1;
+ if ((uintptr_t)pos & 15)
+ {
+ p0 = _mm_loadu_ps(pos); // y1x1 y0x0
+ p1 = _mm_loadu_ps(pos+4); // ---- y2x2
+ } else
+ {
+ p0 = _mm_load_ps(pos); // y1x1 y0x0
+ p1 = _mm_load_ps(pos+4); // ---- y2x2
+ }
+
+ __m128 xf = _mm_shuffle_ps(p1, p0, _MM_SHUFFLE(0,2,2,0));
+ __m128 yf = _mm_shuffle_ps(p1, p0, _MM_SHUFFLE(1,3,1,1));
+
+ __m128 fl256 = _mm_load_ps(twofiftytwo_ps);
+ xf = _mm_mul_ps(xf, fl256);
+ yf = _mm_mul_ps(yf, fl256);
+ __m128i x = _mm_cvttps_epi32(xf);
+ __m128i y = _mm_cvttps_epi32(yf);
+
+ __m128i _m_w = _mm_slli_epi32(_mm_set1_epi32(m_w), 8);
+ __m128i _m_h = _mm_slli_epi32(_mm_set1_epi32(m_h), 8);
+
+ __m128i x_gt, y_gt;
+
+ /* If positions from lensfun is properly clamped this should not be
needed */
+ /* Enable, if crashes begin occuring here here */
+#if 0
+ x_gt = _mm_cmpgt_epi32(x, _m_w);
+ y_gt = _mm_cmpgt_epi32(y, _m_h);
+
+ x = _mm_or_si128(_mm_andnot_si128(x_gt, x), _mm_and_si128(_m_w, x_gt));
+ y = _mm_or_si128(_mm_andnot_si128(y_gt, y), _mm_and_si128(_m_h, y_gt));
+
+ __m128i zero = _mm_setzero_si128();
+ __m128i x_lt = _mm_cmplt_epi32(x, zero);
+ __m128i y_lt = _mm_cmplt_epi32(y, zero);
+ x = _mm_andnot_si128(x_lt, x);
+ y = _mm_andnot_si128(y_lt, y);
+#endif
+ __m128i one = _mm_set1_epi32(1);
+ __m128i nx = _mm_add_epi32(one, _mm_srai_epi32(x, 8));
+ __m128i ny = _mm_add_epi32(one, _mm_srai_epi32(y, 8));
+
+ _m_w = _mm_srai_epi32(_m_w, 8);
+ _m_h = _mm_srai_epi32(_m_h, 8);
+
+ x_gt = _mm_cmpgt_epi32(nx, _m_w);
+ y_gt = _mm_cmpgt_epi32(ny, _m_h);
+
+ nx = _mm_or_si128(_mm_andnot_si128(x_gt, nx), _mm_and_si128(_m_w,
x_gt));
+ ny = _mm_or_si128(_mm_andnot_si128(y_gt, ny), _mm_and_si128(_m_h,
y_gt));
+
+ int xfer[16] __attribute__ ((aligned (16)));
+
+ _mm_store_si128((__m128i*)xfer, _mm_srai_epi32(x, 8));
+ _mm_store_si128((__m128i*)&xfer[4], _mm_srai_epi32(y, 8));
+ _mm_store_si128((__m128i*)&xfer[8], nx);
+ _mm_store_si128((__m128i*)&xfer[12], ny);
+
+ gushort* pixels[12];
+
+ /* Loop unrolled, allows agressive instruction reordering */
+ /* Red, then G & B */
+ pixels[0] = GET_PIXEL(in, xfer[0], xfer[4]); // a
+ pixels[1] = GET_PIXEL(in, xfer[8], xfer[4]); // b
+ pixels[2] = GET_PIXEL(in, xfer[0], xfer[12]); // c
+ pixels[3] = GET_PIXEL(in, xfer[8], xfer[12]); // d
+
+ pixels[4] = GET_PIXEL(in, xfer[1], xfer[1+4]) + 1; // a
+ pixels[4+1] = GET_PIXEL(in, xfer[1+8], xfer[1+4]) + 1; // b
+ pixels[4+2] = GET_PIXEL(in, xfer[1], xfer[1+12]) + 1; // c
+ pixels[4+3] = GET_PIXEL(in, xfer[1+8], xfer[1+12]) + 1; // d
+
+ pixels[2*4] = GET_PIXEL(in, xfer[2], xfer[2+4]) + 2; // a
+ pixels[2*4+1] = GET_PIXEL(in, xfer[2+8], xfer[2+4]) + 2; // b
+ pixels[2*4+2] = GET_PIXEL(in, xfer[2], xfer[2+12]) + 2; // c
+ pixels[2*4+3] = GET_PIXEL(in, xfer[2+8], xfer[2+12]) + 2; // d
+
+ /* Calculate distances */
+ __m128i twofiftyfive = _mm_set1_epi32(255);
+ __m128i diffx = _mm_and_si128(x, twofiftyfive);
+ __m128i diffy = _mm_and_si128(y, twofiftyfive);
+ __m128i inv_diffx = _mm_andnot_si128(diffx, twofiftyfive);
+ __m128i inv_diffy = _mm_andnot_si128(diffy, twofiftyfive);
+
+ /* Calculate weights */
+ __m128i aw = _mm_srai_epi32(_mm_mullo_epi16(inv_diffx, inv_diffy),1);
+ __m128i bw = _mm_srai_epi32(_mm_mullo_epi16(diffx, inv_diffy),1);
+ __m128i cw = _mm_srai_epi32(_mm_mullo_epi16(inv_diffx, diffy),1);
+ __m128i dw = _mm_srai_epi32(_mm_mullo_epi16(diffx, diffy),1);
+
+ _mm_store_si128((__m128i*)xfer, aw);
+ _mm_store_si128((__m128i*)&xfer[4], bw);
+ _mm_store_si128((__m128i*)&xfer[8], cw);
+ _mm_store_si128((__m128i*)&xfer[12], dw);
+
+ gushort** p = pixels;
+ /* Loop unrolled */
+ out[0] = (gushort) ((xfer[0] * *p[0] + xfer[4] * *p[1] + xfer[8] *
*p[2] + xfer[12] * *p[3] + 16384) >> 15 );
+ p+=4;
+ out[1] = (gushort) ((xfer[1] * *p[0] + xfer[1+4] * *p[1] + xfer[1+8] *
*p[2] + xfer[1+12] * *p[3] + 16384) >> 15 );
+ p+=4;
+ out[2] = (gushort) ((xfer[2] * *p[0] + xfer[2+4] * *p[1] + xfer[2+8] *
*p[2] + xfer[2+12] * *p[3] + 16384) >> 15 );
+}
+#else
+gboolean is_sse2_compiled()
+{
+ return FALSE;
+}
+
+void
+rs_image16_bilinear_full_sse2(RS_IMAGE16 *in, gushort *out, gfloat *pos)
+{
+}
+#endif // defined (__SSE2__)
Modified: branches/rawstudio-ng-color/plugins/lensfun/lensfun.c
===================================================================
--- branches/rawstudio-ng-color/plugins/lensfun/lensfun.c 2009-12-30
22:39:52 UTC (rev 2929)
+++ branches/rawstudio-ng-color/plugins/lensfun/lensfun.c 2009-12-30
23:14:23 UTC (rev 2930)
@@ -87,9 +87,8 @@
static RSFilterResponse *get_image(RSFilter *filter, const RSFilterRequest
*request);
static void inline rs_image16_nearest_full(RS_IMAGE16 *in, gushort *out,
gfloat *pos);
static void inline rs_image16_bilinear_full(RS_IMAGE16 *in, gushort *out,
gfloat *pos);
-#if defined (__SSE2__)
-static void inline rs_image16_bilinear_full_sse2(RS_IMAGE16 *in, gushort *out,
gfloat *pos);
-#endif
+extern gboolean is_sse2_compiled();
+extern void rs_image16_bilinear_full_sse2(RS_IMAGE16 *in, gushort *out, gfloat
*pos);
static RSFilterClass *rs_lensfun_parent_class = NULL;
G_MODULE_EXPORT void
@@ -343,7 +342,8 @@
}
}
- gboolean sse2_available = !!(rs_detect_cpu_features() &
RS_CPU_FLAG_SSE2);
+ gboolean sse2_available = !!(rs_detect_cpu_features() &
RS_CPU_FLAG_SSE2) && is_sse2_compiled();
+
if (t->stage == 3)
{
/* Do TCA and distortion */
@@ -356,7 +356,7 @@
lf_modifier_apply_subpixel_geometry_distortion(t->mod,
t->roi->x, (gfloat) y, t->roi->width, 1, pos);
target = GET_PIXEL(t->output, t->roi->x, y);
gfloat* l_pos = pos;
-#if defined (__SSE2__)
+
if (sse2_available)
{
for(x = 0; x < t->roi->width ; x++)
@@ -366,13 +366,14 @@
l_pos += 6;
}
} else
-#endif
- for(x = 0; x < t->roi->width ; x++)
{
- rs_image16_bilinear_full_sse2(t->input, target,
l_pos);
- rs_image16_bilinear_full(t->input, target,
l_pos);
- target += pixelsize;
- l_pos += 6;
+ for(x = 0; x < t->roi->width ; x++)
+ {
+ rs_image16_bilinear_full_sse2(t->input,
target, l_pos);
+ rs_image16_bilinear_full(t->input,
target, l_pos);
+ target += pixelsize;
+ l_pos += 6;
+ }
}
}
g_free(pos);
@@ -711,121 +712,3 @@
out[i] = (gushort) ((a[i]*aw + b[i]*bw + c[i]*cw + d[i]*dw
+ 16384) >> 15 );
}
}
-
-
-#if defined (__SSE2__)
-static gfloat twofiftytwo_ps[4] __attribute__ ((aligned (16))) = {256.0f,
256.0f, 256.0f, 0.0f};
-
-static void inline
-rs_image16_bilinear_full_sse2(RS_IMAGE16 *in, gushort *out, gfloat *pos)
-{
- const gint m_w = (in->w-1);
- const gint m_h = (in->h-1);
-
- __m128 p0, p1;
- if ((uintptr_t)pos & 15)
- {
- p0 = _mm_loadu_ps(pos); // y1x1 y0x0
- p1 = _mm_loadu_ps(pos+4); // ---- y2x2
- } else
- {
- p0 = _mm_load_ps(pos); // y1x1 y0x0
- p1 = _mm_load_ps(pos+4); // ---- y2x2
- }
-
- __m128 xf = _mm_shuffle_ps(p1, p0, _MM_SHUFFLE(0,2,2,0));
- __m128 yf = _mm_shuffle_ps(p1, p0, _MM_SHUFFLE(1,3,1,1));
-
- __m128 fl256 = _mm_load_ps(twofiftytwo_ps);
- xf = _mm_mul_ps(xf, fl256);
- yf = _mm_mul_ps(yf, fl256);
- __m128i x = _mm_cvttps_epi32(xf);
- __m128i y = _mm_cvttps_epi32(yf);
-
- __m128i _m_w = _mm_slli_epi32(_mm_set1_epi32(m_w), 8);
- __m128i _m_h = _mm_slli_epi32(_mm_set1_epi32(m_h), 8);
-
- __m128i x_gt, y_gt;
-
- /* If positions from lensfun is properly clamped this should not be
needed */
- /* Enable, if crashes begin occuring here here */
-#if 0
- x_gt = _mm_cmpgt_epi32(x, _m_w);
- y_gt = _mm_cmpgt_epi32(y, _m_h);
-
- x = _mm_or_si128(_mm_andnot_si128(x_gt, x), _mm_and_si128(_m_w, x_gt));
- y = _mm_or_si128(_mm_andnot_si128(y_gt, y), _mm_and_si128(_m_h, y_gt));
-
- __m128i zero = _mm_setzero_si128();
- __m128i x_lt = _mm_cmplt_epi32(x, zero);
- __m128i y_lt = _mm_cmplt_epi32(y, zero);
- x = _mm_andnot_si128(x_lt, x);
- y = _mm_andnot_si128(y_lt, y);
-#endif
- __m128i one = _mm_set1_epi32(1);
- __m128i nx = _mm_add_epi32(one, _mm_srai_epi32(x, 8));
- __m128i ny = _mm_add_epi32(one, _mm_srai_epi32(y, 8));
-
- _m_w = _mm_srai_epi32(_m_w, 8);
- _m_h = _mm_srai_epi32(_m_h, 8);
-
- x_gt = _mm_cmpgt_epi32(nx, _m_w);
- y_gt = _mm_cmpgt_epi32(ny, _m_h);
-
- nx = _mm_or_si128(_mm_andnot_si128(x_gt, nx), _mm_and_si128(_m_w,
x_gt));
- ny = _mm_or_si128(_mm_andnot_si128(y_gt, ny), _mm_and_si128(_m_h,
y_gt));
-
- int xfer[16] __attribute__ ((aligned (16)));
-
- _mm_store_si128((__m128i*)xfer, _mm_srai_epi32(x, 8));
- _mm_store_si128((__m128i*)&xfer[4], _mm_srai_epi32(y, 8));
- _mm_store_si128((__m128i*)&xfer[8], nx);
- _mm_store_si128((__m128i*)&xfer[12], ny);
-
- gushort* pixels[12];
-
- /* Loop unrolled, allows agressive instruction reordering */
- /* Red, then G & B */
- pixels[0] = GET_PIXEL(in, xfer[0], xfer[4]); // a
- pixels[1] = GET_PIXEL(in, xfer[8], xfer[4]); // b
- pixels[2] = GET_PIXEL(in, xfer[0], xfer[12]); // c
- pixels[3] = GET_PIXEL(in, xfer[8], xfer[12]); // d
-
- pixels[4] = GET_PIXEL(in, xfer[1], xfer[1+4]) + 1; // a
- pixels[4+1] = GET_PIXEL(in, xfer[1+8], xfer[1+4]) + 1; // b
- pixels[4+2] = GET_PIXEL(in, xfer[1], xfer[1+12]) + 1; // c
- pixels[4+3] = GET_PIXEL(in, xfer[1+8], xfer[1+12]) + 1; // d
-
- pixels[2*4] = GET_PIXEL(in, xfer[2], xfer[2+4]) + 2; // a
- pixels[2*4+1] = GET_PIXEL(in, xfer[2+8], xfer[2+4]) + 2; // b
- pixels[2*4+2] = GET_PIXEL(in, xfer[2], xfer[2+12]) + 2; // c
- pixels[2*4+3] = GET_PIXEL(in, xfer[2+8], xfer[2+12]) + 2; // d
-
- /* Calculate distances */
- __m128i twofiftyfive = _mm_set1_epi32(255);
- __m128i diffx = _mm_and_si128(x, twofiftyfive);
- __m128i diffy = _mm_and_si128(y, twofiftyfive);
- __m128i inv_diffx = _mm_andnot_si128(diffx, twofiftyfive);
- __m128i inv_diffy = _mm_andnot_si128(diffy, twofiftyfive);
-
- /* Calculate weights */
- __m128i aw = _mm_srai_epi32(_mm_mullo_epi16(inv_diffx, inv_diffy),1);
- __m128i bw = _mm_srai_epi32(_mm_mullo_epi16(diffx, inv_diffy),1);
- __m128i cw = _mm_srai_epi32(_mm_mullo_epi16(inv_diffx, diffy),1);
- __m128i dw = _mm_srai_epi32(_mm_mullo_epi16(diffx, diffy),1);
-
- _mm_store_si128((__m128i*)xfer, aw);
- _mm_store_si128((__m128i*)&xfer[4], bw);
- _mm_store_si128((__m128i*)&xfer[8], cw);
- _mm_store_si128((__m128i*)&xfer[12], dw);
-
- gushort** p = pixels;
- /* Loop unrolled */
- out[0] = (gushort) ((xfer[0] * *p[0] + xfer[4] * *p[1] + xfer[8] *
*p[2] + xfer[12] * *p[3] + 16384) >> 15 );
- p+=4;
- out[1] = (gushort) ((xfer[1] * *p[0] + xfer[1+4] * *p[1] + xfer[1+8] *
*p[2] + xfer[1+12] * *p[3] + 16384) >> 15 );
- p+=4;
- out[2] = (gushort) ((xfer[2] * *p[0] + xfer[2+4] * *p[1] + xfer[2+8] *
*p[2] + xfer[2+12] * *p[3] + 16384) >> 15 );
-}
-
-#endif // defined (__SSE2__)
_______________________________________________
Rawstudio-commit mailing list
[email protected]
http://rawstudio.org/cgi-bin/mailman/listinfo/rawstudio-commit
