ppc: vsx: Implement float_dsp

5 files changed, 175 insertions, 9 deletions

diff --git a/arch.mak b/arch.mak
index 4bfc883ffa..720062af72 100644
--- a/arch.mak
+++ b/arch.mak
@@ -5,6 +5,7 @@ OBJS-$(HAVE_VFP)     += $(VFP-OBJS)     $(VFP-OBJS-yes)
 OBJS-$(HAVE_NEON)    += $(NEON-OBJS)    $(NEON-OBJS-yes)
 
 OBJS-$(HAVE_ALTIVEC) += $(ALTIVEC-OBJS) $(ALTIVEC-OBJS-yes)
+OBJS-$(HAVE_VSX)     += $(VSX-OBJS) $(VSX-OBJS-yes)
 
 OBJS-$(HAVE_MMX)     += $(MMX-OBJS)     $(MMX-OBJS-yes)
 OBJS-$(HAVE_YASM)    += $(YASM-OBJS)    $(YASM-OBJS-yes)
diff --git a/libavutil/ppc/Makefile b/libavutil/ppc/Makefile
index 4fd8d6d57e..a0febf8d52 100644
--- a/libavutil/ppc/Makefile
+++ b/libavutil/ppc/Makefile
@@ -2,3 +2,5 @@ OBJS += ppc/cpu.o                                                       \
         ppc/float_dsp_init.o                                            \
 
 ALTIVEC-OBJS += ppc/float_dsp_altivec.o                                 \
+
+VSX-OBJS     += ppc/float_dsp_vsx.o                                     \
diff --git a/libavutil/ppc/float_dsp_init.c b/libavutil/ppc/float_dsp_init.c
index e81dc7a441..4407962d4d 100644
--- a/libavutil/ppc/float_dsp_init.c
+++ b/libavutil/ppc/float_dsp_init.c
@@ -24,18 +24,26 @@
 #include "libavutil/float_dsp.h"
 #include "libavutil/ppc/cpu.h"
 #include "float_dsp_altivec.h"
+#include "float_dsp_vsx.h"
 
 av_cold void ff_float_dsp_init_ppc(AVFloatDSPContext *fdsp, int bit_exact)
 {
-    if (!PPC_ALTIVEC(av_get_cpu_flags()))
-        return;
-#if HAVE_BIGENDIAN
-    fdsp->vector_fmul = ff_vector_fmul_altivec;
-    fdsp->vector_fmul_add = ff_vector_fmul_add_altivec;
-    fdsp->vector_fmul_reverse = ff_vector_fmul_reverse_altivec;
+    if (HAVE_BIGENDIAN && PPC_ALTIVEC(av_get_cpu_flags())) {
+        fdsp->vector_fmul = ff_vector_fmul_altivec;
+        fdsp->vector_fmul_add = ff_vector_fmul_add_altivec;
+        fdsp->vector_fmul_reverse = ff_vector_fmul_reverse_altivec;
 
-    if (!bit_exact) {
-        fdsp->vector_fmul_window = ff_vector_fmul_window_altivec;
+        if (!bit_exact) {
+            fdsp->vector_fmul_window = ff_vector_fmul_window_altivec;
+        }
+    }
+    if (PPC_VSX(av_get_cpu_flags())) {
+        fdsp->vector_fmul = ff_vector_fmul_vsx;
+        fdsp->vector_fmul_add = ff_vector_fmul_add_vsx;
+        fdsp->vector_fmul_reverse = ff_vector_fmul_reverse_vsx;
+
+        if (!bit_exact) {
+            fdsp->vector_fmul_window = ff_vector_fmul_window_vsx;
+        }
     }
-#endif
 }
diff --git a/libavutil/ppc/float_dsp_vsx.c b/libavutil/ppc/float_dsp_vsx.c
new file mode 100644
index 0000000000..4135db1391
--- /dev/null
+++ b/libavutil/ppc/float_dsp_vsx.c
@@ -0,0 +1,117 @@
+/*
+ * Copyright (c) 2015 Luca Barbato <lu_zero@gentoo.org>
+ *
+ * This file is part of Libav.
+ *
+ * Libav 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 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav 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
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "util_altivec.h"
+#include "float_dsp_vsx.h"
+
+void ff_vector_fmul_vsx(float *dst,
+                        const float *src0, const float *src1,
+                        int len)
+{
+    int i;
+    vec_f d0, d1, zero = (vec_f)vec_splat_u32(0);
+    for (i = 0; i < len - 7; i += 8) {
+        d0 = vec_vsx_ld( 0, src0 + i);
+        d1 = vec_vsx_ld(16, src0 + i);
+        d0 = vec_madd(d0, vec_vsx_ld( 0, src1 + i), zero);
+        d1 = vec_madd(d1, vec_vsx_ld(16, src1 + i), zero);
+        vec_vsx_st(d0,  0, dst + i);
+        vec_vsx_st(d1, 16, dst + i);
+    }
+}
+
+void ff_vector_fmul_window_vsx(float *dst, const float *src0,
+                               const float *src1, const float *win,
+                               int len)
+{
+    vec_f zero, t0, t1, s0, s1, wi, wj;
+    const vec_u8 reverse = vcprm(3, 2, 1, 0);
+    int i, j;
+
+    dst  += len;
+    win  += len;
+    src0 += len;
+
+    zero = (vec_f)vec_splat_u32(0);
+
+    for (i = -len * 4, j = len * 4 - 16; i < 0; i += 16, j -= 16) {
+        s0 = vec_vsx_ld(i, src0);
+        s1 = vec_vsx_ld(j, src1);
+        wi = vec_vsx_ld(i, win);
+        wj = vec_vsx_ld(j, win);
+
+        s1 = vec_perm(s1, s1, reverse);
+        wj = vec_perm(wj, wj, reverse);
+
+        t0 = vec_madd(s0, wj, zero);
+        t0 = vec_nmsub(s1, wi, t0);
+        t1 = vec_madd(s0, wi, zero);
+        t1 = vec_madd(s1, wj, t1);
+        t1 = vec_perm(t1, t1, reverse);
+
+        vec_vsx_st(t0, i, dst);
+        vec_vsx_st(t1, j, dst);
+    }
+}
+
+void ff_vector_fmul_add_vsx(float *dst, const float *src0,
+                            const float *src1, const float *src2,
+                            int len)
+{
+    int i;
+    vec_f d, s0, s1, s2;
+
+    for (i = 0; i < len - 3; i += 4) {
+        s0 = vec_vsx_ld(0, src0 + i);
+        s1 = vec_vsx_ld(0, src1 + i);
+        s2 = vec_vsx_ld(0, src2 + i);
+        d = vec_madd(s0, s1, s2);
+        vec_vsx_st(d, 0, dst + i);
+    }
+}
+
+void ff_vector_fmul_reverse_vsx(float *dst, const float *src0,
+                                const float *src1, int len)
+{
+    int i;
+    vec_f d, s0, s1, h0, l0, s2, s3;
+    vec_f zero = (vec_f)vec_splat_u32(0);
+
+    src1 += len - 4;
+    for (i = 0; i < len - 7; i += 8) {
+        s1 = vec_vsx_ld(0, src1 - i);              // [a,b,c,d]
+        s0 = vec_vsx_ld(0, src0 + i);
+        l0 = vec_mergel(s1, s1);               // [c,c,d,d]
+        s3 = vec_vsx_ld(-16, src1 - i);
+        h0 = vec_mergeh(s1, s1);               // [a,a,b,b]
+        s2 = vec_vsx_ld(16, src0 + i);
+        s1 = vec_mergeh(vec_mergel(l0, h0),    // [d,b,d,b]
+                        vec_mergeh(l0, h0));   // [c,a,c,a]
+        // [d,c,b,a]
+        l0 = vec_mergel(s3, s3);
+        d = vec_madd(s0, s1, zero);
+        h0 = vec_mergeh(s3, s3);
+        vec_vsx_st(d, 0, dst + i);
+        s3 = vec_mergeh(vec_mergel(l0, h0),
+                        vec_mergeh(l0, h0));
+        d = vec_madd(s2, s3, zero);
+        vec_vsx_st(d, 16, dst + i);
+    }
+}
diff --git a/libavutil/ppc/float_dsp_vsx.h b/libavutil/ppc/float_dsp_vsx.h
new file mode 100644
index 0000000000..32860dc2c9
--- /dev/null
+++ b/libavutil/ppc/float_dsp_vsx.h
@@ -0,0 +1,38 @@
+/*
+ * Copyright (c) 2015 Luca Barbato <lu_zero@gentoo.org>
+ *
+ * This file is part of Libav.
+ *
+ * Libav 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 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav 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
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#ifndef AVUTIL_PPC_FLOAT_DSP_VSX_H
+#define AVUTIL_PPC_FLOAT_DSP_VSX_H
+
+void ff_vector_fmul_vsx(float *dst, const float *src0,
+                        const float *src1, int len);
+
+void ff_vector_fmul_window_vsx(float *dst, const float *src0,
+                               const float *src1, const float *win,
+                               int len);
+
+void ff_vector_fmul_add_vsx(float *dst, const float *src0,
+                            const float *src1, const float *src2,
+                            int len);
+
+void ff_vector_fmul_reverse_vsx(float *dst, const float *src0,
+                                const float *src1, int len);
+
+#endif /* AVUTIL_PPC_FLOAT_DSP_VSX_H */