Author: post
Date: 2012-01-31 18:16:46 +0100 (Tue, 31 Jan 2012)
New Revision: 4117
Added:
trunk/plugins/lensfun/lensfun-sse4.c
Modified:
trunk/plugins/lensfun/Makefile.am
trunk/plugins/lensfun/lensfun.c
Log:
Add SSE4 version of Lensfun distortion correction.
Modified: trunk/plugins/lensfun/Makefile.am
===================================================================
--- trunk/plugins/lensfun/Makefile.am 2012-01-31 16:56:02 UTC (rev 4116)
+++ trunk/plugins/lensfun/Makefile.am 2012-01-31 17:16:46 UTC (rev 4117)
@@ -18,10 +18,10 @@
libdir = $(datadir)/rawstudio/plugins/
-lensfun_la_LIBADD = @PACKAGE_LIBS@ @LENSFUN_LIBS@ lensfun-avx.lo
lensfun-sse2.lo lensfun-c.lo
+lensfun_la_LIBADD = @PACKAGE_LIBS@ @LENSFUN_LIBS@ lensfun-avx.lo
lensfun-sse2.lo lensfun-sse4.lo lensfun-c.lo
lensfun_la_LDFLAGS = -module -avoid-version
lensfun_la_SOURCES = lensfun-version.c lensfun-version.h
-EXTRA_DIST = lensfun-avx.c lensfun-sse2.c lensfun.c
+EXTRA_DIST = lensfun-avx.c lensfun-sse2.c lensfun-sse4.c lensfun.c
lensfun-c.lo: lensfun.c
$(LTCOMPILE) -o lensfun-c.lo -c $(top_srcdir)/plugins/lensfun/lensfun.c
@@ -41,3 +41,11 @@
AVX_FLAG=
endif
$(LTCOMPILE) $(AVX_FLAG) -c $(top_srcdir)/plugins/lensfun/lensfun-avx.c
+
+lensfun-sse4.lo: lensfun-sse4.c
+if CAN_COMPILE_SSE4_1
+SSE4_FLAG=-msse4.1
+else
+SSE4_FLAG=
+endif
+ $(LTCOMPILE) $(SSE4_FLAG) -c
$(top_srcdir)/plugins/lensfun/lensfun-sse4.c
Added: trunk/plugins/lensfun/lensfun-sse4.c
===================================================================
--- trunk/plugins/lensfun/lensfun-sse4.c (rev 0)
+++ trunk/plugins/lensfun/lensfun-sse4.c 2012-01-31 17:16:46 UTC (rev
4117)
@@ -0,0 +1,177 @@
+/*
+ * * Copyright (C) 2006-2011 Anders Brander <[email protected]>,
+ * * Anders Kvist <[email protected]> and Klaus Post <[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(__SSE4_1__)
+
+#include <smmintrin.h>
+
+static gfloat twofiftytwo_ps[4] __attribute__ ((aligned (16))) = {256.0f,
256.0f, 256.0f, 0.0f};
+static gint _zero12[4] __attribute__ ((aligned (16))) = {0,1,2,0};
+
+gboolean is_sse4_compiled(void)
+{
+ return TRUE;
+}
+
+void
+rs_image16_bilinear_nomeasure_sse4(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
+ }
+
+ // to x2x2 x1x0
+ __m128 xf = _mm_shuffle_ps(p0, p1, _MM_SHUFFLE(0,0,2,0));
+ // to y2y2 y1y0
+ __m128 yf = _mm_shuffle_ps(p0, p1, _MM_SHUFFLE(1,1,3,1));
+
+ __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);
+
+ /* Clamping */
+ x = _mm_min_epi32(x, _m_w);
+ y = _mm_min_epi32(y, _m_h);
+
+ __m128i zero = _mm_setzero_si128();
+ x = _mm_max_epi32(x, zero);
+ y = _mm_max_epi32(y, zero);
+
+ __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));
+
+ /* Check that 'next' pixels are in bounds */
+ _m_w = _mm_srai_epi32(_m_w, 8);
+ _m_h = _mm_srai_epi32(_m_h, 8);
+
+ nx = _mm_min_epi32(nx, _m_w);
+ ny = _mm_min_epi32(ny, _m_h);
+
+ int xfer[16] __attribute__ ((aligned (16)));
+
+ /* Pitch as pixels */
+ __m128i pitch = _mm_set1_epi32(in->rowstride >> 2);
+
+ /* Remove remainder */
+ __m128i tx = _mm_srai_epi32(x, 8);
+ __m128i ty = _mm_srai_epi32(y, 8);
+
+ /* Multiply y by pitch */
+ ty = _mm_mullo_epi32(ty, pitch);
+ ny = _mm_mullo_epi32(ny, pitch);
+
+ /* Add pitch and x offset */
+ __m128i a_offset = _mm_add_epi32(tx, ty);
+ __m128i b_offset = _mm_add_epi32(nx, ty);
+ __m128i c_offset = _mm_add_epi32(tx, ny);
+ __m128i d_offset = _mm_add_epi32(nx, ny);
+
+ /* Multiply by pixelsize and add RGB offsets */
+ __m128i zero12 = _mm_load_si128((__m128i*)_zero12);
+ a_offset = _mm_add_epi32(zero12, _mm_slli_epi32(a_offset, 2));
+ b_offset = _mm_add_epi32(zero12, _mm_slli_epi32(b_offset, 2));
+ c_offset = _mm_add_epi32(zero12, _mm_slli_epi32(c_offset, 2));
+ d_offset = _mm_add_epi32(zero12, _mm_slli_epi32(d_offset, 2));
+
+ _mm_store_si128((__m128i*)xfer, a_offset);
+ _mm_store_si128((__m128i*)&xfer[4], b_offset);
+ _mm_store_si128((__m128i*)&xfer[8], c_offset);
+ _mm_store_si128((__m128i*)&xfer[12], d_offset);
+
+ gushort* pixels[12];
+
+ /* Loop unrolled, allows agressive instruction reordering */
+ /* Red, then G & B */
+ pixels[0] = in->pixels + xfer[0]; // a
+ pixels[1] = in->pixels + xfer[4]; // b
+ pixels[2] = in->pixels + xfer[8]; // c
+ pixels[3] = in->pixels + xfer[12]; // d
+
+ pixels[4] = in->pixels + xfer[1+0]; // a
+ pixels[5] = in->pixels + xfer[1+4]; // b
+ pixels[6] = in->pixels + xfer[1+8]; // c
+ pixels[7] = in->pixels + xfer[1+12]; // d
+
+ pixels[8] = in->pixels + xfer[2+0]; // a
+ pixels[9] = in->pixels + xfer[2+4]; // b
+ pixels[10] = in->pixels + xfer[2+8]; // c
+ pixels[11] = in->pixels + xfer[2+12]; // 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 // NO SSE4
+
+gboolean is_sse4_compiled(void)
+{
+ return FALSE;
+}
+
+
+void
+rs_image16_bilinear_nomeasure_sse4(RS_IMAGE16 *in, gushort *out, gfloat *pos)
+{
+}
+
+#endif // defined (__SSE4_1__)
Modified: trunk/plugins/lensfun/lensfun.c
===================================================================
--- trunk/plugins/lensfun/lensfun.c 2012-01-31 16:56:02 UTC (rev 4116)
+++ trunk/plugins/lensfun/lensfun.c 2012-01-31 17:16:46 UTC (rev 4117)
@@ -91,6 +91,8 @@
extern gboolean is_sse2_compiled(void);
extern void rs_image16_bilinear_full_sse2(RS_IMAGE16 *in, gushort *out, gfloat
*pos, const gint *current_xy, const gint* min_max_xy);
extern void rs_image16_bilinear_nomeasure_sse2(RS_IMAGE16 *in, gushort *out,
gfloat *pos);
+extern gboolean is_sse4_compiled(void);
+extern void rs_image16_bilinear_nomeasure_sse4(RS_IMAGE16 *in, gushort *out,
gfloat *pos);
extern gboolean is_avx_compiled(void);
extern void rs_image16_bilinear_full_avx(RS_IMAGE16 *in, gushort *out, gfloat
*pos, const gint *current_xy, const gint* min_max_xy);
extern void rs_image16_bilinear_nomeasure_avx(RS_IMAGE16 *in, gushort *out,
gfloat *pos);
@@ -337,13 +339,12 @@
}
gboolean sse2_available = !!(rs_detect_cpu_features() &
RS_CPU_FLAG_SSE2) && is_sse2_compiled();
+ gboolean sse4_available = !!(rs_detect_cpu_features() &
RS_CPU_FLAG_SSE4_1) && is_sse4_compiled();
gboolean avx_available = !!(rs_detect_cpu_features() & RS_CPU_FLAG_AVX)
&& is_avx_compiled();
if (t->input->pixelsize != 4)
- sse2_available = avx_available = FALSE;
+ sse2_available = sse4_available = avx_available = FALSE;
-
-
if (t->stage == 3)
{
/* Do TCA and distortion */
@@ -366,6 +367,15 @@
l_pos += 6;
}
}
+ else if (sse4_available)
+ {
+ for(x = 0; x < t->roi->width ; x++)
+ {
+
rs_image16_bilinear_nomeasure_sse4(t->input, target, l_pos);
+ target += 4;
+ l_pos += 6;
+ }
+ }
else if (sse2_available)
{
for(x = 0; x < t->roi->width ; x++)
_______________________________________________
Rawstudio-commit mailing list
[email protected]
http://rawstudio.org/cgi-bin/mailman/listinfo/rawstudio-commit
