ref: be60b142e734c0f24be50049e616c0cdf149a1b6
parent: 3e0f1508ad17b181b9252423655a15602497bd49
author: Michail Alvanos <malvanos@gmail.com>
date: Wed Oct 9 04:42:37 EDT 2019
Add VSX wiener filter implementation
--- a/src/looprestoration.h
+++ b/src/looprestoration.h
@@ -75,5 +75,6 @@
bitfn_decls(void dav1d_loop_restoration_dsp_init, Dav1dLoopRestorationDSPContext *c);
bitfn_decls(void dav1d_loop_restoration_dsp_init_arm, Dav1dLoopRestorationDSPContext *c);
bitfn_decls(void dav1d_loop_restoration_dsp_init_x86, Dav1dLoopRestorationDSPContext *c);
+bitfn_decls(void dav1d_loop_restoration_dsp_init_ppc, Dav1dLoopRestorationDSPContext *c);
#endif /* DAV1D_SRC_LOOPRESTORATION_H */
--- a/src/looprestoration_tmpl.c
+++ b/src/looprestoration_tmpl.c
@@ -580,6 +580,8 @@
#if HAVE_ASM
#if ARCH_AARCH64 || ARCH_ARM
bitfn(dav1d_loop_restoration_dsp_init_arm)(c);
+#elif ARCH_PPC64LE
+ bitfn(dav1d_loop_restoration_dsp_init_ppc)(c);
#elif ARCH_X86
bitfn(dav1d_loop_restoration_dsp_init_x86)(c);
#endif
--- a/src/meson.build
+++ b/src/meson.build
@@ -170,6 +170,7 @@
)
libdav1d_arch_tmpl_sources += files(
'ppc/cdef_init_tmpl.c',
+ 'ppc/looprestoration_init_tmpl.c',
)
endif
endif
--- /dev/null
+++ b/src/ppc/looprestoration_init_tmpl.c
@@ -1,0 +1,350 @@
+/*
+ * Copyright © 2019, VideoLAN and dav1d authors
+ * Copyright © 2019, Michail Alvanos
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright notice, this
+ * list of conditions and the following disclaimer.
+ *
+ * 2. Redistributions in binary form must reproduce the above copyright notice,
+ * this list of conditions and the following disclaimer in the documentation
+ * and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+ * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+ * DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+ * ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+ * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+ * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include "common/intops.h"
+#include "src/ppc/types.h"
+#include "src/cpu.h"
+#include "src/looprestoration.h"
+
+#if BITDEPTH == 8
+
+#define REST_UNIT_STRIDE (400)
+
+static inline i32x4 iclip_vec(i32x4 v, const i32x4 minv, const i32x4 maxv) {+ v = vec_max(minv, v);
+ v = vec_min(maxv, v);
+ return v;
+}
+
+#define APPLY_FILTER_H(v, f, ssum1, ssum2) do { \+ i16x8 ktmp_u16_high = (i16x8) u8h_to_u16(v); \
+ i16x8 ktmp_u16_low = (i16x8) u8l_to_u16(v); \
+ ssum1 = vec_madd(ktmp_u16_high, f, ssum1); \
+ ssum2 = vec_madd(ktmp_u16_low, f, ssum2); \
+} while (0)
+
+static void wiener_filter_h_vsx(int32_t *hor_ptr,
+ uint8_t *tmp_ptr,
+ const int16_t filterh[7],
+ const int w, const int h)
+{+ static const i32x4 zerov = vec_splats(0);
+ static const i32x4 seven_vec = vec_splats(7);
+ static const i32x4 bitdepth_added_vec = vec_splats(1 << 14);
+ static const i32x4 round_bits_vec = vec_splats(3);
+ static const i32x4 rounding_off_vec = vec_splats(1<<2);
+ static const i32x4 clip_limit_v = vec_splats((1 << 13) - 1);
+
+ i16x8 filterhvall = vec_vsx_ld(0, filterh);
+ i16x8 filterhv0 = vec_splat( filterhvall, 0);
+ i16x8 filterhv1 = vec_splat( filterhvall, 1);
+ i16x8 filterhv2 = vec_splat( filterhvall, 2);
+ i16x8 filterhv3 = vec_splat( filterhvall, 3);
+ i16x8 filterhv4 = vec_splat( filterhvall, 4);
+ i16x8 filterhv5 = vec_splat( filterhvall, 5);
+ i16x8 filterhv6 = vec_splat( filterhvall, 6);
+
+ for (int j = 0; j < h + 6; j++) {+ for (int i = 0; i < w; i+=16) {+ i32x4 sum1 = bitdepth_added_vec;
+ i32x4 sum2 = bitdepth_added_vec;
+ i32x4 sum3 = bitdepth_added_vec;
+ i32x4 sum4 = bitdepth_added_vec;
+
+ u8x16 tmp_v0 = vec_ld(0, &tmp_ptr[i]);
+ u8x16 tmp_v7 = vec_ld(0, &tmp_ptr[i+16]);
+
+ u8x16 tmp_v1 = vec_sld( tmp_v7, tmp_v0, 15);
+ u8x16 tmp_v2 = vec_sld( tmp_v7, tmp_v0, 14);
+ u8x16 tmp_v3 = vec_sld( tmp_v7, tmp_v0, 13);
+ u8x16 tmp_v4 = vec_sld( tmp_v7, tmp_v0, 12);
+ u8x16 tmp_v5 = vec_sld( tmp_v7, tmp_v0, 11);
+ u8x16 tmp_v6 = vec_sld( tmp_v7, tmp_v0, 10);
+
+ u16x8 tmp_u16_high = u8h_to_u16(tmp_v3);
+ u16x8 tmp_u16_low = u8l_to_u16(tmp_v3);
+
+ i32x4 tmp_expanded1 = i16h_to_i32(tmp_u16_high);
+ i32x4 tmp_expanded2 = i16l_to_i32(tmp_u16_high);
+ i32x4 tmp_expanded3 = i16h_to_i32(tmp_u16_low);
+ i32x4 tmp_expanded4 = i16l_to_i32(tmp_u16_low);
+
+ i16x8 ssum1 = (i16x8) zerov;
+ i16x8 ssum2 = (i16x8) zerov;
+
+ APPLY_FILTER_H(tmp_v0, filterhv0, ssum1, ssum2);
+ APPLY_FILTER_H(tmp_v1, filterhv1, ssum1, ssum2);
+ APPLY_FILTER_H(tmp_v2, filterhv2, ssum1, ssum2);
+ APPLY_FILTER_H(tmp_v3, filterhv3, ssum1, ssum2);
+ APPLY_FILTER_H(tmp_v4, filterhv4, ssum1, ssum2);
+ APPLY_FILTER_H(tmp_v5, filterhv5, ssum1, ssum2);
+ APPLY_FILTER_H(tmp_v6, filterhv6, ssum1, ssum2);
+
+ sum1 += i16h_to_i32(ssum1) + (tmp_expanded1 << seven_vec);
+ sum2 += i16l_to_i32(ssum1) + (tmp_expanded2 << seven_vec);
+ sum3 += i16h_to_i32(ssum2) + (tmp_expanded3 << seven_vec);
+ sum4 += i16l_to_i32(ssum2) + (tmp_expanded4 << seven_vec);
+
+ sum1 = (sum1 + rounding_off_vec) >> round_bits_vec;
+ sum2 = (sum2 + rounding_off_vec) >> round_bits_vec;
+ sum3 = (sum3 + rounding_off_vec) >> round_bits_vec;
+ sum4 = (sum4 + rounding_off_vec) >> round_bits_vec;
+
+ sum1 = iclip_vec(sum1, zerov, clip_limit_v);
+ sum2 = iclip_vec(sum2, zerov, clip_limit_v);
+ sum3 = iclip_vec(sum3, zerov, clip_limit_v);
+ sum4 = iclip_vec(sum4, zerov, clip_limit_v);
+
+ vec_st(sum1, 0, &hor_ptr[i]);
+ vec_st(sum2, 16, &hor_ptr[i]);
+ vec_st(sum3, 32, &hor_ptr[i]);
+ vec_st(sum4, 48, &hor_ptr[i]);
+ }
+ tmp_ptr += REST_UNIT_STRIDE;
+ hor_ptr += REST_UNIT_STRIDE;
+ }
+}
+
+static inline i16x8 iclip_u8_vec(i16x8 v) {+ static const i16x8 zerov = vec_splats((int16_t)0);
+ static const i16x8 maxv = vec_splats((int16_t)255);
+ v = vec_max(zerov, v);
+ v = vec_min(maxv, v);
+ return v;
+}
+
+#define APPLY_FILTER_V(index, f) do { \+ i32x4 v1 = vec_ld( 0, &hor[(j + index) * REST_UNIT_STRIDE + i]); \
+ i32x4 v2 = vec_ld(16, &hor[(j + index) * REST_UNIT_STRIDE + i]); \
+ i32x4 v3 = vec_ld(32, &hor[(j + index) * REST_UNIT_STRIDE + i]); \
+ i32x4 v4 = vec_ld(48, &hor[(j + index) * REST_UNIT_STRIDE + i]); \
+ sum1 = sum1 + v1 * f; \
+ sum2 = sum2 + v2 * f; \
+ sum3 = sum3 + v3 * f; \
+ sum4 = sum4 + v4 * f; \
+} while (0)
+
+#define LOAD_AND_APPLY_FILTER_V(sumpixelv, hor) do { \+ i32x4 v_1 = (i32x4) vec_ld( 0, &hor[(j + 3) * REST_UNIT_STRIDE + i]); \
+ i32x4 v_2 = (i32x4) vec_ld(16, &hor[(j + 3) * REST_UNIT_STRIDE + i]); \
+ i32x4 v_3 = (i32x4) vec_ld(32, &hor[(j + 3) * REST_UNIT_STRIDE + i]); \
+ i32x4 v_4 = (i32x4) vec_ld(48, &hor[(j + 3) * REST_UNIT_STRIDE + i]); \
+ i32x4 sum1 = -round_offset_vec; \
+ i32x4 sum2 = -round_offset_vec; \
+ i32x4 sum3 = -round_offset_vec; \
+ i32x4 sum4 = -round_offset_vec; \
+ APPLY_FILTER_V(0, filterv0); \
+ APPLY_FILTER_V(1, filterv1); \
+ APPLY_FILTER_V(2, filterv2); \
+ APPLY_FILTER_V(3, filterv3); \
+ APPLY_FILTER_V(4, filterv4); \
+ APPLY_FILTER_V(5, filterv5); \
+ APPLY_FILTER_V(6, filterv6); \
+ sum1 = (v_1 << seven_vec) + sum1 + rounding_off_vec; \
+ sum2 = (v_2 << seven_vec) + sum2 + rounding_off_vec; \
+ sum3 = (v_3 << seven_vec) + sum3 + rounding_off_vec; \
+ sum4 = (v_4 << seven_vec) + sum4 + rounding_off_vec; \
+ sum1 = sum1 >> round_bits_vec; \
+ sum2 = sum2 >> round_bits_vec; \
+ sum3 = sum3 >> round_bits_vec; \
+ sum4 = sum4 >> round_bits_vec; \
+ i16x8 sum_short_packed_1 = (i16x8) vec_pack( sum1, sum2 ); \
+ i16x8 sum_short_packed_2 = (i16x8) vec_pack( sum3, sum4 ); \
+ sum_short_packed_1 = iclip_u8_vec(sum_short_packed_1); \
+ sum_short_packed_2 = iclip_u8_vec(sum_short_packed_2); \
+ sum_pixel = (u8x16) vec_pack(sum_short_packed_1, sum_short_packed_2 ); \
+} while (0)
+
+static inline void wiener_filter_v_vsx(uint8_t *p,
+ const ptrdiff_t p_stride,
+ const int32_t *hor,
+ const int16_t filterv[7],
+ const int w, const int h)
+{+ static const i32x4 round_bits_vec = vec_splats(11);
+ static const i32x4 rounding_off_vec = vec_splats(1 << 10);
+ static const i32x4 round_offset_vec = vec_splats(1 << 18);
+ static const i32x4 seven_vec = vec_splats(7);
+
+ i32x4 filterv0 = vec_splats((int32_t) filterv[0]);
+ i32x4 filterv1 = vec_splats((int32_t) filterv[1]);
+ i32x4 filterv2 = vec_splats((int32_t) filterv[2]);
+ i32x4 filterv3 = vec_splats((int32_t) filterv[3]);
+ i32x4 filterv4 = vec_splats((int32_t) filterv[4]);
+ i32x4 filterv5 = vec_splats((int32_t) filterv[5]);
+ i32x4 filterv6 = vec_splats((int32_t) filterv[6]);
+
+ for (int j = 0; j < h; j++) {+ for (int i = 0; i <(w-w%16); i += 16) {+ u8x16 sum_pixel;
+ LOAD_AND_APPLY_FILTER_V(sum_pixel, hor);
+ vec_vsx_st(sum_pixel, 0, &p[j * PXSTRIDE(p_stride) + i]);
+ }
+ // remaining loop
+ if (w & 0xf){+ int i=w-w%16;
+ ALIGN_STK_16(uint8_t, tmp_out, 16,);
+ u8x16 sum_pixel;
+
+ LOAD_AND_APPLY_FILTER_V(sum_pixel, hor);
+ vec_vsx_st(sum_pixel, 0, tmp_out);
+
+ for (int k=0; i<w; i++, k++) {+ p[j * PXSTRIDE(p_stride) + i] = tmp_out[k];
+ }
+ }
+ }
+}
+
+static inline void padding(uint8_t *dst, const uint8_t *p,
+ const ptrdiff_t p_stride, const uint8_t (*left)[4],
+ const uint8_t *lpf, const ptrdiff_t lpf_stride,
+ int unit_w, const int stripe_h,
+ const enum LrEdgeFlags edges)
+{+ const int have_left = !!(edges & LR_HAVE_LEFT);
+ const int have_right = !!(edges & LR_HAVE_RIGHT);
+
+ // Copy more pixels if we don't have to pad them
+ unit_w += 3 * have_left + 3 * have_right;
+ uint8_t *dst_l = dst + 3 * !have_left;
+ p -= 3 * have_left;
+ lpf -= 3 * have_left;
+
+ if (edges & LR_HAVE_TOP) {+ // Copy previous loop filtered rows
+ const uint8_t *const above_1 = lpf;
+ const uint8_t *const above_2 = above_1 + PXSTRIDE(lpf_stride);
+ pixel_copy(dst_l, above_1, unit_w);
+ pixel_copy(dst_l + REST_UNIT_STRIDE, above_1, unit_w);
+ pixel_copy(dst_l + 2 * REST_UNIT_STRIDE, above_2, unit_w);
+ } else {+ // Pad with first row
+ pixel_copy(dst_l, p, unit_w);
+ pixel_copy(dst_l + REST_UNIT_STRIDE, p, unit_w);
+ pixel_copy(dst_l + 2 * REST_UNIT_STRIDE, p, unit_w);
+ if (have_left) {+ pixel_copy(dst_l, &left[0][1], 3);
+ pixel_copy(dst_l + REST_UNIT_STRIDE, &left[0][1], 3);
+ pixel_copy(dst_l + 2 * REST_UNIT_STRIDE, &left[0][1], 3);
+ }
+ }
+
+ uint8_t *dst_tl = dst_l + 3 * REST_UNIT_STRIDE;
+ if (edges & LR_HAVE_BOTTOM) {+ // Copy next loop filtered rows
+ const uint8_t *const below_1 = lpf + 6 * PXSTRIDE(lpf_stride);
+ const uint8_t *const below_2 = below_1 + PXSTRIDE(lpf_stride);
+ pixel_copy(dst_tl + stripe_h * REST_UNIT_STRIDE, below_1, unit_w);
+ pixel_copy(dst_tl + (stripe_h + 1) * REST_UNIT_STRIDE, below_2, unit_w);
+ pixel_copy(dst_tl + (stripe_h + 2) * REST_UNIT_STRIDE, below_2, unit_w);
+ } else {+ // Pad with last row
+ const uint8_t *const src = p + (stripe_h - 1) * PXSTRIDE(p_stride);
+ pixel_copy(dst_tl + stripe_h * REST_UNIT_STRIDE, src, unit_w);
+ pixel_copy(dst_tl + (stripe_h + 1) * REST_UNIT_STRIDE, src, unit_w);
+ pixel_copy(dst_tl + (stripe_h + 2) * REST_UNIT_STRIDE, src, unit_w);
+ if (have_left) {+ pixel_copy(dst_tl + stripe_h * REST_UNIT_STRIDE, &left[stripe_h - 1][1], 3);
+ pixel_copy(dst_tl + (stripe_h + 1) * REST_UNIT_STRIDE, &left[stripe_h - 1][1], 3);
+ pixel_copy(dst_tl + (stripe_h + 2) * REST_UNIT_STRIDE, &left[stripe_h - 1][1], 3);
+ }
+ }
+
+ // Inner UNIT_WxSTRIPE_H
+ for (int j = 0; j < stripe_h; j++) {+ pixel_copy(dst_tl + 3 * have_left, p + 3 * have_left, unit_w - 3 * have_left);
+ dst_tl += REST_UNIT_STRIDE;
+ p += PXSTRIDE(p_stride);
+ }
+
+ if (!have_right) {+ uint8_t *pad = dst_l + unit_w;
+ uint8_t *row_last = &dst_l[unit_w - 1];
+ // Pad 3x(STRIPE_H+6) with last column
+ for (int j = 0; j < stripe_h + 6; j++) {+ pixel_set(pad, *row_last, 3);
+ pad += REST_UNIT_STRIDE;
+ row_last += REST_UNIT_STRIDE;
+ }
+ }
+
+ if (!have_left) {+ // Pad 3x(STRIPE_H+6) with first column
+ for (int j = 0; j < stripe_h + 6; j++) {+ pixel_set(dst, *dst_l, 3);
+ dst += REST_UNIT_STRIDE;
+ dst_l += REST_UNIT_STRIDE;
+ }
+ } else {+ dst += 3 * REST_UNIT_STRIDE;
+ for (int j = 0; j < stripe_h; j++) {+ pixel_copy(dst, &left[j][1], 3);
+ dst += REST_UNIT_STRIDE;
+ }
+ }
+}
+
+
+// FIXME Could split into luma and chroma specific functions,
+// (since first and last tops are always 0 for chroma)
+// FIXME Could implement a version that requires less temporary memory
+// (should be possible to implement with only 6 rows of temp storage)
+static void wiener_filter_vsx(uint8_t *p, const ptrdiff_t p_stride,
+ const uint8_t (*const left)[4],
+ const uint8_t *lpf,
+ const ptrdiff_t lpf_stride,
+ const int w, const int h,
+ const int16_t filterh[7],
+ const int16_t filterv[7],
+ const enum LrEdgeFlags edges HIGHBD_DECL_SUFFIX)
+{+ // Wiener filtering is applied to a maximum stripe height of 64 + 3 pixels
+ // of padding above and below
+ ALIGN_STK_16(uint8_t, tmp, 70 /*(64 + 3 + 3)*/ * REST_UNIT_STRIDE,);
+ padding(tmp, p, p_stride, left, lpf, lpf_stride, w, h, edges);
+ ALIGN_STK_16(int32_t, hor, 70 /*(64 + 3 + 3)*/ * REST_UNIT_STRIDE + 64,);
+
+ wiener_filter_h_vsx(hor, tmp, filterh, w, h);
+ wiener_filter_v_vsx(p, p_stride, hor, filterv, w, h);
+
+}
+#endif
+
+COLD void bitfn(dav1d_loop_restoration_dsp_init_ppc)
+ (Dav1dLoopRestorationDSPContext *const c)
+{+ const unsigned flags = dav1d_get_cpu_flags();
+
+ if (!(flags & DAV1D_PPC_CPU_FLAG_VSX)) return;
+
+#if BITDEPTH == 8
+ c->wiener = wiener_filter_vsx;
+#endif
+}
+
+
--- a/src/ppc/types.h
+++ b/src/ppc/types.h
@@ -47,6 +47,8 @@
#define u8h_to_u16(v) ((u16x8) vec_mergeh((u8x16) v, vec_splat_u8(0)))
#define u8l_to_u16(v) ((u16x8) vec_mergel((u8x16) v, vec_splat_u8(0)))
#define u16h_to_i32(v) ((i32x4) vec_mergeh((u16x8) v, vec_splat_u16(0)))
+#define i16h_to_i32(v) ((i32x4) vec_unpackh((i16x8)v))
#define u16l_to_i32(v) ((i32x4) vec_mergel((u16x8) v, vec_splat_u16(0)))
+#define i16l_to_i32(v) ((i32x4) vec_unpackl((i16x8)v))
#endif /* DAV1D_SRC_PPC_TYPES_H */