ref: 63860ba7b8aa0133049dddd9b9dd39b033636b79
parent: ef5918098d5c7f8ffda960274e3f8e38f02cb487
author: Luca Barbato <lu_zero@gentoo.org>
date: Wed Apr 26 11:43:19 EDT 2017
ppc: Add convolve_copy Change-Id: Ie26d6dbe090e711d84bac01ba7da270db983f405
--- a/test/convolve_test.cc
+++ b/test/convolve_test.cc
@@ -1221,4 +1221,16 @@
INSTANTIATE_TEST_CASE_P(MSA, ConvolveTest,
::testing::ValuesIn(kArrayConvolve8_msa));
#endif // HAVE_MSA
+
+#if HAVE_VSX
+const ConvolveFunctions convolve8_vsx(
+ vpx_convolve_copy_vsx, vpx_convolve_avg_c, vpx_convolve8_horiz_c,
+ vpx_convolve8_avg_horiz_c, vpx_convolve8_vert_c, vpx_convolve8_avg_vert_c,
+ vpx_convolve8_c, vpx_convolve8_avg_c, vpx_scaled_horiz_c,
+ vpx_scaled_avg_horiz_c, vpx_scaled_vert_c, vpx_scaled_avg_vert_c,
+ vpx_scaled_2d_c, vpx_scaled_avg_2d_c, 0);
+const ConvolveParam kArrayConvolve_vsx[] = { ALL_SIZES(convolve8_vsx) };+INSTANTIATE_TEST_CASE_P(VSX, ConvolveTest,
+ ::testing::ValuesIn(kArrayConvolve_vsx));
+#endif // HAVE_VSX
} // namespace
--- /dev/null
+++ b/vpx_dsp/ppc/vpx_convolve_vsx.c
@@ -1,0 +1,85 @@
+/*
+ * Copyright (c) 2017 The WebM project authors. All Rights Reserved.
+ *
+ * Use of this source code is governed by a BSD-style license
+ * that can be found in the LICENSE file in the root of the source
+ * tree. An additional intellectual property rights grant can be found
+ * in the file PATENTS. All contributing project authors may
+ * be found in the AUTHORS file in the root of the source tree.
+ */
+#include <string.h>
+#include "./vpx_dsp_rtcd.h"
+#include "vpx_dsp/ppc/types_vsx.h"
+
+// TODO(lu_zero): unroll
+static inline void copy_w16(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride, int32_t h) {+ int i;
+
+ for (i = h; i--;) {+ vec_vsx_st(vec_vsx_ld(0, src), 0, dst);
+ src += src_stride;
+ dst += dst_stride;
+ }
+}
+
+static inline void copy_w32(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride, int32_t h) {+ int i;
+
+ for (i = h; i--;) {+ vec_vsx_st(vec_vsx_ld(0, src), 0, dst);
+ vec_vsx_st(vec_vsx_ld(16, src), 16, dst);
+ src += src_stride;
+ dst += dst_stride;
+ }
+}
+
+static inline void copy_w64(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride, int32_t h) {+ int i;
+
+ for (i = h; i--;) {+ vec_vsx_st(vec_vsx_ld(0, src), 0, dst);
+ vec_vsx_st(vec_vsx_ld(16, src), 16, dst);
+ vec_vsx_st(vec_vsx_ld(32, src), 32, dst);
+ vec_vsx_st(vec_vsx_ld(48, src), 48, dst);
+ src += src_stride;
+ dst += dst_stride;
+ }
+}
+
+void vpx_convolve_copy_vsx(const uint8_t *src, ptrdiff_t src_stride,
+ uint8_t *dst, ptrdiff_t dst_stride,
+ const int16_t *filter_x, int32_t filter_x_stride,
+ const int16_t *filter_y, int32_t filter_y_stride,
+ int32_t w, int32_t h) {+ (void)filter_x;
+ (void)filter_y;
+ (void)filter_x_stride;
+ (void)filter_y_stride;
+
+ switch (w) {+ case 16: {+ copy_w16(src, src_stride, dst, dst_stride, h);
+ break;
+ }
+ case 32: {+ copy_w32(src, src_stride, dst, dst_stride, h);
+ break;
+ }
+ case 64: {+ copy_w64(src, src_stride, dst, dst_stride, h);
+ break;
+ }
+ default: {+ int i;
+ for (i = h; i--;) {+ memcpy(dst, src, w);
+ src += src_stride;
+ dst += dst_stride;
+ }
+ break;
+ }
+ }
+}
--- a/vpx_dsp/vpx_dsp.mk
+++ b/vpx_dsp/vpx_dsp.mk
@@ -144,6 +144,8 @@
DSP_SRCS-$(HAVE_DSPR2) += mips/convolve8_horiz_dspr2.c
DSP_SRCS-$(HAVE_DSPR2) += mips/convolve8_vert_dspr2.c
+DSP_SRCS-$(HAVE_VSX) += ppc/vpx_convolve_vsx.c
+
# loop filters
DSP_SRCS-yes += loopfilter.c
--- a/vpx_dsp/vpx_dsp_rtcd_defs.pl
+++ b/vpx_dsp/vpx_dsp_rtcd_defs.pl
@@ -332,7 +332,7 @@
# Sub Pixel Filters
#
add_proto qw/void vpx_convolve_copy/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
-specialize qw/vpx_convolve_copy neon dspr2 msa sse2/;
+specialize qw/vpx_convolve_copy neon dspr2 msa sse2 vsx/;
add_proto qw/void vpx_convolve_avg/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
specialize qw/vpx_convolve_avg neon dspr2 msa sse2/;