Merge changes I1f5a3752,I95123051,I3bb724e0,Ie81077fa,Ic80f3c05, ...

* changes:
  ppc: tm predictor 16x16
  ppc: tm predictor 8x8
  ppc: horizontal predictor 32x32
  ppc: horizontal predictor 16x16
  ppc: vertical intrapred 16x16 and 32x32
  configure: Workaround clang not enabling altivec on -mvsx
  configure: Match power*64* as ppc64

build/make/Makefile

@@ -141,8 +141,8 @@ $(BUILD_PFX)%_avx2.c.d: CFLAGS += -mavx2
 $(BUILD_PFX)%_avx2.c.o: CFLAGS += -mavx2
 
 # POWER
-$(BUILD_PFX)%_vsx.c.d: CFLAGS += -mvsx
+$(BUILD_PFX)%_vsx.c.d: CFLAGS += -maltivec -mvsx
-$(BUILD_PFX)%_vsx.c.o: CFLAGS += -mvsx
+$(BUILD_PFX)%_vsx.c.o: CFLAGS += -maltivec -mvsx
 
 $(BUILD_PFX)%.c.d: %.c
 	$(if $(quiet),@echo "    [DEP] $@")

build/make/configure.sh

@@ -674,7 +674,6 @@ check_xcode_minimum_version() {
 process_common_toolchain() {
   if [ -z "$toolchain" ]; then
     gcctarget="${CHOST:-$(gcc -dumpmachine 2> /dev/null)}"
-
     # detect tgt_isa
     case "$gcctarget" in
       aarch64*)
@@ -697,6 +696,9 @@ process_common_toolchain() {
       *sparc*)
         tgt_isa=sparc
         ;;
+      power*64*-*)
+        tgt_isa=ppc64
+        ;;
       power*)
         tgt_isa=ppc
         ;;

test/test_intra_pred_speed.cc

@@ -312,6 +312,19 @@ INTRA_PRED_TEST(MSA, TestIntraPred32, vpx_dc_predictor_32x32_msa,
                 vpx_tm_predictor_32x32_msa)
 #endif  // HAVE_MSA
 
+#if HAVE_VSX
+INTRA_PRED_TEST(VSX, TestIntraPred8, NULL, NULL, NULL, NULL, NULL, NULL, NULL,
+                NULL, NULL, NULL, NULL, NULL, vpx_tm_predictor_8x8_vsx)
+
+INTRA_PRED_TEST(VSX, TestIntraPred16, NULL, NULL, NULL, NULL,
+                vpx_v_predictor_16x16_vsx, vpx_h_predictor_16x16_vsx, NULL,
+                NULL, NULL, NULL, NULL, NULL, vpx_tm_predictor_16x16_vsx)
+
+INTRA_PRED_TEST(VSX, TestIntraPred32, NULL, NULL, NULL, NULL,
+                vpx_v_predictor_32x32_vsx, vpx_h_predictor_32x32_vsx, NULL,
+                NULL, NULL, NULL, NULL, NULL, NULL)
+#endif  // HAVE_VSX
+
 // -----------------------------------------------------------------------------
 
 #if CONFIG_VP9_HIGHBITDEPTH

vpx_dsp/ppc/intrapred_vsx.c

@@ -0,0 +1,297 @@
+/*
+ *  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 "./vpx_dsp_rtcd.h"
+#include "vpx_dsp/ppc/types_vsx.h"
+
+void vpx_v_predictor_16x16_vsx(uint8_t *dst, ptrdiff_t stride,
+                               const uint8_t *above, const uint8_t *left) {
+  const uint8x16_t d = vec_vsx_ld(0, above);
+  int i;
+  (void)left;
+
+  for (i = 0; i < 16; i++, dst += stride) {
+    vec_vsx_st(d, 0, dst);
+  }
+}
+
+void vpx_v_predictor_32x32_vsx(uint8_t *dst, ptrdiff_t stride,
+                               const uint8_t *above, const uint8_t *left) {
+  const uint8x16_t d0 = vec_vsx_ld(0, above);
+  const uint8x16_t d1 = vec_vsx_ld(16, above);
+  int i;
+  (void)left;
+
+  for (i = 0; i < 32; i++, dst += stride) {
+    vec_vsx_st(d0, 0, dst);
+    vec_vsx_st(d1, 16, dst);
+  }
+}
+
+void vpx_h_predictor_16x16_vsx(uint8_t *dst, ptrdiff_t stride,
+                               const uint8_t *above, const uint8_t *left) {
+  const uint8x16_t d = vec_vsx_ld(0, left);
+  const uint8x16_t v0 = vec_splat(d, 0);
+  const uint8x16_t v1 = vec_splat(d, 1);
+  const uint8x16_t v2 = vec_splat(d, 2);
+  const uint8x16_t v3 = vec_splat(d, 3);
+
+  const uint8x16_t v4 = vec_splat(d, 4);
+  const uint8x16_t v5 = vec_splat(d, 5);
+  const uint8x16_t v6 = vec_splat(d, 6);
+  const uint8x16_t v7 = vec_splat(d, 7);
+
+  const uint8x16_t v8 = vec_splat(d, 8);
+  const uint8x16_t v9 = vec_splat(d, 9);
+  const uint8x16_t v10 = vec_splat(d, 10);
+  const uint8x16_t v11 = vec_splat(d, 11);
+
+  const uint8x16_t v12 = vec_splat(d, 12);
+  const uint8x16_t v13 = vec_splat(d, 13);
+  const uint8x16_t v14 = vec_splat(d, 14);
+  const uint8x16_t v15 = vec_splat(d, 15);
+
+  (void)above;
+
+  vec_vsx_st(v0, 0, dst);
+  dst += stride;
+  vec_vsx_st(v1, 0, dst);
+  dst += stride;
+  vec_vsx_st(v2, 0, dst);
+  dst += stride;
+  vec_vsx_st(v3, 0, dst);
+  dst += stride;
+  vec_vsx_st(v4, 0, dst);
+  dst += stride;
+  vec_vsx_st(v5, 0, dst);
+  dst += stride;
+  vec_vsx_st(v6, 0, dst);
+  dst += stride;
+  vec_vsx_st(v7, 0, dst);
+  dst += stride;
+  vec_vsx_st(v8, 0, dst);
+  dst += stride;
+  vec_vsx_st(v9, 0, dst);
+  dst += stride;
+  vec_vsx_st(v10, 0, dst);
+  dst += stride;
+  vec_vsx_st(v11, 0, dst);
+  dst += stride;
+  vec_vsx_st(v12, 0, dst);
+  dst += stride;
+  vec_vsx_st(v13, 0, dst);
+  dst += stride;
+  vec_vsx_st(v14, 0, dst);
+  dst += stride;
+  vec_vsx_st(v15, 0, dst);
+}
+
+#define H_PREDICTOR_32(v) \
+  vec_vsx_st(v, 0, dst);  \
+  vec_vsx_st(v, 16, dst); \
+  dst += stride
+
+void vpx_h_predictor_32x32_vsx(uint8_t *dst, ptrdiff_t stride,
+                               const uint8_t *above, const uint8_t *left) {
+  const uint8x16_t d0 = vec_vsx_ld(0, left);
+  const uint8x16_t d1 = vec_vsx_ld(16, left);
+
+  const uint8x16_t v0_0 = vec_splat(d0, 0);
+  const uint8x16_t v1_0 = vec_splat(d0, 1);
+  const uint8x16_t v2_0 = vec_splat(d0, 2);
+  const uint8x16_t v3_0 = vec_splat(d0, 3);
+  const uint8x16_t v4_0 = vec_splat(d0, 4);
+  const uint8x16_t v5_0 = vec_splat(d0, 5);
+  const uint8x16_t v6_0 = vec_splat(d0, 6);
+  const uint8x16_t v7_0 = vec_splat(d0, 7);
+  const uint8x16_t v8_0 = vec_splat(d0, 8);
+  const uint8x16_t v9_0 = vec_splat(d0, 9);
+  const uint8x16_t v10_0 = vec_splat(d0, 10);
+  const uint8x16_t v11_0 = vec_splat(d0, 11);
+  const uint8x16_t v12_0 = vec_splat(d0, 12);
+  const uint8x16_t v13_0 = vec_splat(d0, 13);
+  const uint8x16_t v14_0 = vec_splat(d0, 14);
+  const uint8x16_t v15_0 = vec_splat(d0, 15);
+
+  const uint8x16_t v0_1 = vec_splat(d1, 0);
+  const uint8x16_t v1_1 = vec_splat(d1, 1);
+  const uint8x16_t v2_1 = vec_splat(d1, 2);
+  const uint8x16_t v3_1 = vec_splat(d1, 3);
+  const uint8x16_t v4_1 = vec_splat(d1, 4);
+  const uint8x16_t v5_1 = vec_splat(d1, 5);
+  const uint8x16_t v6_1 = vec_splat(d1, 6);
+  const uint8x16_t v7_1 = vec_splat(d1, 7);
+  const uint8x16_t v8_1 = vec_splat(d1, 8);
+  const uint8x16_t v9_1 = vec_splat(d1, 9);
+  const uint8x16_t v10_1 = vec_splat(d1, 10);
+  const uint8x16_t v11_1 = vec_splat(d1, 11);
+  const uint8x16_t v12_1 = vec_splat(d1, 12);
+  const uint8x16_t v13_1 = vec_splat(d1, 13);
+  const uint8x16_t v14_1 = vec_splat(d1, 14);
+  const uint8x16_t v15_1 = vec_splat(d1, 15);
+
+  (void)above;
+
+  H_PREDICTOR_32(v0_0);
+  H_PREDICTOR_32(v1_0);
+  H_PREDICTOR_32(v2_0);
+  H_PREDICTOR_32(v3_0);
+
+  H_PREDICTOR_32(v4_0);
+  H_PREDICTOR_32(v5_0);
+  H_PREDICTOR_32(v6_0);
+  H_PREDICTOR_32(v7_0);
+
+  H_PREDICTOR_32(v8_0);
+  H_PREDICTOR_32(v9_0);
+  H_PREDICTOR_32(v10_0);
+  H_PREDICTOR_32(v11_0);
+
+  H_PREDICTOR_32(v12_0);
+  H_PREDICTOR_32(v13_0);
+  H_PREDICTOR_32(v14_0);
+  H_PREDICTOR_32(v15_0);
+
+  H_PREDICTOR_32(v0_1);
+  H_PREDICTOR_32(v1_1);
+  H_PREDICTOR_32(v2_1);
+  H_PREDICTOR_32(v3_1);
+
+  H_PREDICTOR_32(v4_1);
+  H_PREDICTOR_32(v5_1);
+  H_PREDICTOR_32(v6_1);
+  H_PREDICTOR_32(v7_1);
+
+  H_PREDICTOR_32(v8_1);
+  H_PREDICTOR_32(v9_1);
+  H_PREDICTOR_32(v10_1);
+  H_PREDICTOR_32(v11_1);
+
+  H_PREDICTOR_32(v12_1);
+  H_PREDICTOR_32(v13_1);
+  H_PREDICTOR_32(v14_1);
+  H_PREDICTOR_32(v15_1);
+}
+
+void vpx_tm_predictor_8x8_vsx(uint8_t *dst, ptrdiff_t stride,
+                              const uint8_t *above, const uint8_t *left) {
+  const int16x8_t tl = unpack_to_s16_h(vec_splat(vec_vsx_ld(-1, above), 0));
+  const int16x8_t l = unpack_to_s16_h(vec_vsx_ld(0, left));
+  const int16x8_t a = unpack_to_s16_h(vec_vsx_ld(0, above));
+  int16x8_t tmp, val;
+
+  tmp = unpack_to_s16_l(vec_vsx_ld(0, dst));
+  val = vec_sub(vec_add(vec_splat(l, 0), a), tl);
+  vec_vsx_st(vec_packsu(val, tmp), 0, dst);
+  dst += stride;
+
+  tmp = unpack_to_s16_l(vec_vsx_ld(0, dst));
+  val = vec_sub(vec_add(vec_splat(l, 1), a), tl);
+  vec_vsx_st(vec_packsu(val, tmp), 0, dst);
+  dst += stride;
+
+  tmp = unpack_to_s16_l(vec_vsx_ld(0, dst));
+  val = vec_sub(vec_add(vec_splat(l, 2), a), tl);
+  vec_vsx_st(vec_packsu(val, tmp), 0, dst);
+  dst += stride;
+
+  tmp = unpack_to_s16_l(vec_vsx_ld(0, dst));
+  val = vec_sub(vec_add(vec_splat(l, 3), a), tl);
+  vec_vsx_st(vec_packsu(val, tmp), 0, dst);
+  dst += stride;
+
+  tmp = unpack_to_s16_l(vec_vsx_ld(0, dst));
+  val = vec_sub(vec_add(vec_splat(l, 4), a), tl);
+  vec_vsx_st(vec_packsu(val, tmp), 0, dst);
+  dst += stride;
+
+  tmp = unpack_to_s16_l(vec_vsx_ld(0, dst));
+  val = vec_sub(vec_add(vec_splat(l, 5), a), tl);
+  vec_vsx_st(vec_packsu(val, tmp), 0, dst);
+  dst += stride;
+
+  tmp = unpack_to_s16_l(vec_vsx_ld(0, dst));
+  val = vec_sub(vec_add(vec_splat(l, 6), a), tl);
+  vec_vsx_st(vec_packsu(val, tmp), 0, dst);
+  dst += stride;
+
+  tmp = unpack_to_s16_l(vec_vsx_ld(0, dst));
+  val = vec_sub(vec_add(vec_splat(l, 7), a), tl);
+  vec_vsx_st(vec_packsu(val, tmp), 0, dst);
+}
+
+static void tm_predictor_16x8(uint8_t *dst, const ptrdiff_t stride, int16x8_t l,
+                              int16x8_t ah, int16x8_t al, int16x8_t tl) {
+  int16x8_t vh, vl, ls;
+
+  ls = vec_splat(l, 0);
+  vh = vec_sub(vec_add(ls, ah), tl);
+  vl = vec_sub(vec_add(ls, al), tl);
+  vec_vsx_st(vec_packsu(vh, vl), 0, dst);
+  dst += stride;
+
+  ls = vec_splat(l, 1);
+  vh = vec_sub(vec_add(ls, ah), tl);
+  vl = vec_sub(vec_add(ls, al), tl);
+  vec_vsx_st(vec_packsu(vh, vl), 0, dst);
+  dst += stride;
+
+  ls = vec_splat(l, 2);
+  vh = vec_sub(vec_add(ls, ah), tl);
+  vl = vec_sub(vec_add(ls, al), tl);
+  vec_vsx_st(vec_packsu(vh, vl), 0, dst);
+  dst += stride;
+
+  ls = vec_splat(l, 3);
+  vh = vec_sub(vec_add(ls, ah), tl);
+  vl = vec_sub(vec_add(ls, al), tl);
+  vec_vsx_st(vec_packsu(vh, vl), 0, dst);
+  dst += stride;
+
+  ls = vec_splat(l, 4);
+  vh = vec_sub(vec_add(ls, ah), tl);
+  vl = vec_sub(vec_add(ls, al), tl);
+  vec_vsx_st(vec_packsu(vh, vl), 0, dst);
+  dst += stride;
+
+  ls = vec_splat(l, 5);
+  vh = vec_sub(vec_add(ls, ah), tl);
+  vl = vec_sub(vec_add(ls, al), tl);
+  vec_vsx_st(vec_packsu(vh, vl), 0, dst);
+  dst += stride;
+
+  ls = vec_splat(l, 6);
+  vh = vec_sub(vec_add(ls, ah), tl);
+  vl = vec_sub(vec_add(ls, al), tl);
+  vec_vsx_st(vec_packsu(vh, vl), 0, dst);
+  dst += stride;
+
+  ls = vec_splat(l, 7);
+  vh = vec_sub(vec_add(ls, ah), tl);
+  vl = vec_sub(vec_add(ls, al), tl);
+  vec_vsx_st(vec_packsu(vh, vl), 0, dst);
+}
+
+void vpx_tm_predictor_16x16_vsx(uint8_t *dst, ptrdiff_t stride,
+                                const uint8_t *above, const uint8_t *left) {
+  const int16x8_t tl = unpack_to_s16_h(vec_splat(vec_vsx_ld(-1, above), 0));
+  const uint8x16_t l = vec_vsx_ld(0, left);
+  const int16x8_t lh = unpack_to_s16_h(l);
+  const int16x8_t ll = unpack_to_s16_l(l);
+  const uint8x16_t a = vec_vsx_ld(0, above);
+  const int16x8_t ah = unpack_to_s16_h(a);
+  const int16x8_t al = unpack_to_s16_l(a);
+
+  tm_predictor_16x8(dst, stride, lh, ah, al, tl);
+
+  dst += stride * 8;
+
+  tm_predictor_16x8(dst, stride, ll, ah, al, tl);
+}

vpx_dsp/ppc/types_vsx.h

@@ -13,8 +13,31 @@
 
 #include <altivec.h>
 
+typedef vector signed char int8x16_t;
+typedef vector unsigned char uint8x16_t;
 typedef vector signed short int16x8_t;
 typedef vector unsigned short uint16x8_t;
 typedef vector signed int int32x4_t;
+typedef vector unsigned int uint32x4_t;
+
+#ifdef WORDS_BIGENDIAN
+#define unpack_to_u16_h(v) \
+  (uint16x8_t) vec_mergeh(vec_splat_u8(0), (uint8x16_t)v)
+#define unpack_to_u16_l(v) \
+  (uint16x8_t) vec_mergel(vec_splat_u8(0), (uint8x16_t)v)
+#define unpack_to_s16_h(v) \
+  (int16x8_t) vec_mergeh(vec_splat_u8(0), (uint8x16_t)v)
+#define unpack_to_s16_l(v) \
+  (int16x8_t) vec_mergel(vec_splat_u8(0), (uint8x16_t)v)
+#else
+#define unpack_to_u16_h(v) \
+  (uint16x8_t) vec_mergeh((uint8x16_t)v, vec_splat_u8(0))
+#define unpack_to_u16_l(v) \
+  (uint16x8_t) vec_mergel((uint8x16_t)v, vec_splat_u8(0))
+#define unpack_to_s16_h(v) \
+  (int16x8_t) vec_mergeh((uint8x16_t)v, vec_splat_u8(0))
+#define unpack_to_s16_l(v) \
+  (int16x8_t) vec_mergel((uint8x16_t)v, vec_splat_u8(0))
+#endif
 
 #endif  // VPX_DSP_PPC_TYPES_VSX_H_

vpx_dsp/vpx_dsp.mk

@@ -51,6 +51,7 @@ DSP_SRCS-$(HAVE_SSE) += x86/intrapred_sse2.asm
 DSP_SRCS-$(HAVE_SSE2) += x86/intrapred_sse2.asm
 DSP_SRCS-$(HAVE_SSSE3) += x86/intrapred_ssse3.asm
 DSP_SRCS-$(HAVE_SSSE3) += x86/vpx_subpixel_8t_ssse3.asm
+DSP_SRCS-$(HAVE_VSX) += ppc/intrapred_vsx.c
 
 ifeq ($(CONFIG_VP9_HIGHBITDEPTH),yes)
 DSP_SRCS-$(HAVE_SSE)  += x86/highbd_intrapred_sse2.asm

vpx_dsp/vpx_dsp_rtcd_defs.pl

@@ -95,7 +95,7 @@ add_proto qw/void vpx_v_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const
 specialize qw/vpx_v_predictor_8x8 neon msa sse2/;
 
 add_proto qw/void vpx_tm_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vpx_tm_predictor_8x8 neon dspr2 msa sse2/;
+specialize qw/vpx_tm_predictor_8x8 neon dspr2 msa sse2 vsx/;
 
 add_proto qw/void vpx_dc_predictor_8x8/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
 specialize qw/vpx_dc_predictor_8x8 dspr2 neon msa sse2/;
@@ -119,7 +119,7 @@ add_proto qw/void vpx_d63_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride, c
 specialize qw/vpx_d63_predictor_16x16 ssse3/;
 
 add_proto qw/void vpx_h_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vpx_h_predictor_16x16 neon dspr2 msa sse2/;
+specialize qw/vpx_h_predictor_16x16 neon dspr2 msa sse2 vsx/;
 
 add_proto qw/void vpx_d117_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
 
@@ -130,10 +130,10 @@ add_proto qw/void vpx_d153_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride,
 specialize qw/vpx_d153_predictor_16x16 ssse3/;
 
 add_proto qw/void vpx_v_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vpx_v_predictor_16x16 neon msa sse2/;
+specialize qw/vpx_v_predictor_16x16 neon msa sse2 vsx/;
 
 add_proto qw/void vpx_tm_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vpx_tm_predictor_16x16 neon msa sse2/;
+specialize qw/vpx_tm_predictor_16x16 neon msa sse2 vsx/;
 
 add_proto qw/void vpx_dc_predictor_16x16/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
 specialize qw/vpx_dc_predictor_16x16 dspr2 neon msa sse2/;
@@ -157,7 +157,7 @@ add_proto qw/void vpx_d63_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, c
 specialize qw/vpx_d63_predictor_32x32 ssse3/;
 
 add_proto qw/void vpx_h_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vpx_h_predictor_32x32 neon msa sse2/;
+specialize qw/vpx_h_predictor_32x32 neon msa sse2 vsx/;
 
 add_proto qw/void vpx_d117_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
 
@@ -168,7 +168,7 @@ add_proto qw/void vpx_d153_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride,
 specialize qw/vpx_d153_predictor_32x32 ssse3/;
 
 add_proto qw/void vpx_v_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
-specialize qw/vpx_v_predictor_32x32 neon msa sse2/;
+specialize qw/vpx_v_predictor_32x32 neon msa sse2 vsx/;
 
 add_proto qw/void vpx_tm_predictor_32x32/, "uint8_t *dst, ptrdiff_t y_stride, const uint8_t *above, const uint8_t *left";
 specialize qw/vpx_tm_predictor_32x32 neon msa sse2/;