add a simple abstraction for SIMD operations

src/.gitignore

@@ -58,3 +58,4 @@ thread-test
 utf8-test
 voltest
 start-pulseaudio-x11
+vector-test

src/Makefile.am

@@ -233,6 +233,7 @@ TESTS = \
 		strlist-test \
 		close-test \
 		voltest \
+		vector-test \
 		memblockq-test \
 		channelmap-test \
 		thread-mainloop-test \
@@ -263,6 +264,7 @@ TESTS_BINARIES = \
 		strlist-test \
 		close-test \
 		voltest \
+		vector-test \
 		memblockq-test \
 		sync-playback \
 		interpol-test \
@@ -408,6 +410,11 @@ voltest_CFLAGS = $(AM_CFLAGS)
 voltest_LDADD = $(AM_LDADD) libpulse.la
 voltest_LDFLAGS = $(AM_LDFLAGS) $(BINLDFLAGS)
 
+vector_test_SOURCES = tests/vector-test.c
+vector_test_CFLAGS = $(AM_CFLAGS)
+vector_test_LDADD = $(AM_LDADD) libpulsecore-@PA_MAJORMINORMICRO@.la libpulsecommon-@PA_MAJORMINORMICRO@.la
+vector_test_LDFLAGS = $(AM_LDFLAGS) $(BINLDFLAGS)
+
 channelmap_test_SOURCES = tests/channelmap-test.c
 channelmap_test_CFLAGS = $(AM_CFLAGS)
 channelmap_test_LDADD = $(AM_LDADD) libpulse.la
@@ -527,7 +534,7 @@ libpulsecommon_@PA_MAJORMINORMICRO@_la_SOURCES = \
 		pulsecore/llist.h \
 		pulsecore/lock-autospawn.c pulsecore/lock-autospawn.h \
 		pulsecore/log.c pulsecore/log.h \
-		pulsecore/macro.h \
+		pulsecore/macro.h pulsecore/vector.h \
 		pulsecore/mcalign.c pulsecore/mcalign.h \
 		pulsecore/memblock.c pulsecore/memblock.h \
 		pulsecore/memblockq.c pulsecore/memblockq.h \

src/pulsecore/vector.h

@@ -0,0 +1,97 @@
+/***
+  This file is part of PulseAudio.
+
+  Copyright 2004-2006 Lennart Poettering
+  Copyright 2006 Pierre Ossman <ossman@cendio.se> for Cendio AB
+
+  PulseAudio is free software; you can redistribute it and/or modify
+  it under the terms of the GNU Lesser General Public License as published
+  by the Free Software Foundation; either version 2 of the License,
+  or (at your option) any later version.
+
+  PulseAudio 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 Lesser General Public License
+  along with PulseAudio; if not, write to the Free Software
+  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307
+  USA.
+***/
+
+#include <inttypes.h>
+
+/* First, define HAVE_VECTOR if we have the gcc vector extensions at all */
+#if defined(__SSE2__) || defined(__ALTIVEC__)
+#define HAVE_VECTOR
+
+
+/* This is supposed to be portable to different SIMD instruction
+ * sets. We define vector types for different base types: uint8_t,
+ * int16_t, int32_t, float. The vector type is a union. The fields .i,
+ * .u, .f are arrays for accessing the separate elements of a
+ * vector. .v is a gcc vector type of the right format. .m is the
+ * vector in the type the SIMD extenstion specific intrinsics API
+ * expects. PA_xxx_VECTOR_SIZE is the size of the
+ * entries. PA_xxxx_VECTOR_MAKE constructs a gcc vector variable with
+ * the same value in all elements. */
+
+#ifdef __SSE2__
+
+#include <xmmintrin.h>
+#include <emmintrin.h>
+
+#define PA_UINT8_VECTOR_SIZE 16
+#define PA_INT16_VECTOR_SIZE 8
+#define PA_INT32_VECTOR_SIZE 4
+#define PA_FLOAT_VECTOR_SIZE 4
+
+#define PA_UINT8_VECTOR_MAKE(x) (pa_v16qi) { x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x }
+#define PA_INT16_VECTOR_MAKE(x) (pa_v8hi) { x, x, x, x, x, x, x, x }
+#define PA_INT32_VECTOR_MAKE(x) (pa_v4si) { x, x, x, x }
+#define PA_FLOAT_VECTOR_MAKE(x) (pa_v4fi) { x, x, x, x }
+
+#endif
+
+/* uint8_t vector */
+typedef uint8_t pa_v16qi __attribute__ ((vector_size (PA_UINT8_VECTOR_SIZE * sizeof(uint8_t))));
+typedef union pa_uint8_vector {
+    uint8_t u[PA_UINT8_VECTOR_SIZE];
+    pa_v16qi v;
+#ifdef __SSE2__
+    __m128i m;
+#endif
+} pa_uint8_vector_t;
+
+/* int16_t vector*/
+typedef int16_t pa_v8hi __attribute__ ((vector_size (PA_INT16_VECTOR_SIZE * sizeof(int16_t))));
+typedef union pa_int16_vector {
+    int16_t i[PA_INT16_VECTOR_SIZE];
+    pa_v8hi v;
+#ifdef __SSE2__
+    __m128i m;
+#endif
+} pa_int16_vector_t;
+
+/* int32_t vector */
+typedef int32_t pa_v4si __attribute__ ((vector_size (PA_INT32_VECTOR_SIZE * sizeof(int32_t))));
+typedef union pa_int32_vector {
+    int32_t i[PA_INT32_VECTOR_SIZE];
+    pa_v4si v;
+#ifdef __SSE2__
+    __m128i m;
+#endif
+} pa_int32_vector_t;
+
+/* float vector */
+typedef float pa_v4sf __attribute__ ((vector_size (PA_FLOAT_VECTOR_SIZE * sizeof(float))));
+typedef union pa_float_vector {
+    float f[PA_FLOAT_VECTOR_SIZE];
+    pa_v4sf v;
+#ifdef __SSE2__
+    __m128 m;
+#endif
+} pa_float_vector_t;
+
+#endif

src/tests/vector-test.c

@@ -0,0 +1,83 @@
+/***
+  This file is part of PulseAudio.
+
+  Copyright 2009 Lennart Poettering
+
+  PulseAudio is free software; you can redistribute it and/or modify
+  it under the terms of the GNU Lesser General Public License as
+  published by the Free Software Foundation; either version 2 of the
+  License, or (at your option) any later version.
+
+  PulseAudio 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 Lesser General Public
+  License along with PulseAudio; if not, write to the Free Software
+  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307
+  USA.
+***/
+
+#ifdef HAVE_CONFIG_H
+#include <config.h>
+#endif
+
+#include <pulsecore/vector.h>
+#include <pulsecore/log.h>
+
+int main(int argc, char *argv[]) {
+
+#ifdef __SSE2__
+    pa_int16_vector_t input, zero;
+    pa_int32_vector_t unpacked1, unpacked2;
+    pa_int32_vector_t volume1, volume2, volume1_hi, volume1_lo, volume2_hi, volume2_lo, reduce, mask;
+    pa_int16_vector_t output;
+
+    unsigned u;
+
+    zero.v = PA_INT16_VECTOR_MAKE(0);
+    reduce.v = PA_INT32_VECTOR_MAKE(0x10000);
+    volume1.v = volume2.v = PA_INT32_VECTOR_MAKE(0x10000*2+7);
+    mask.v = PA_INT32_VECTOR_MAKE(0xFFFF);
+
+    volume1_lo.m = _mm_and_si128(volume1.m, mask.m);
+    volume2_lo.m = _mm_and_si128(volume2.m, mask.m);
+    volume1_hi.m = _mm_srli_epi32(volume1.m, 16);
+    volume2_hi.m = _mm_srli_epi32(volume2.m, 16);
+
+    input.v = PA_INT16_VECTOR_MAKE(32000);
+
+    for (u = 0; u < PA_INT16_VECTOR_SIZE; u++)
+        pa_log("input=%i\n", input.i[u]);
+
+    unpacked1.m = _mm_unpackhi_epi16(zero.m, input.m);
+    unpacked2.m = _mm_unpacklo_epi16(zero.m, input.m);
+
+    for (u = 0; u < PA_INT32_VECTOR_SIZE; u++)
+        pa_log("unpacked1=%i\n", unpacked1.i[u]);
+
+    unpacked1.v /= reduce.v;
+    unpacked2.v /= reduce.v;
+
+    for (u = 0; u < PA_INT32_VECTOR_SIZE; u++)
+        pa_log("unpacked1=%i\n", unpacked1.i[u]);
+
+    for (u = 0; u < PA_INT32_VECTOR_SIZE; u++)
+        pa_log("volume1=%i\n", volume1.i[u]);
+
+    unpacked1.v = (unpacked1.v * volume1_lo.v) / reduce.v + unpacked1.v * volume1_hi.v;
+    unpacked2.v = (unpacked2.v * volume2_lo.v) / reduce.v + unpacked2.v * volume2_hi.v;
+
+    for (u = 0; u < PA_INT32_VECTOR_SIZE; u++)
+        pa_log("unpacked1=%i\n", unpacked1.i[u]);
+
+    output.m = _mm_packs_epi32(unpacked1.m, unpacked2.m);
+
+    for (u = 0; u < PA_INT16_VECTOR_SIZE; u++)
+        pa_log("output=%i\n", output.i[u]);
+
+#endif
+
+    return 0;
+}