my %config = ();
while (<CONFIG_FILE>) {
- next if !/^CONFIG_/;
+ next if !/^(?:CONFIG_|HAVE_)/;
chomp;
my @pair = split /=/;
$config{$pair[0]} = $pair[1];
make_tuple(64, 64, &convolve8_ssse3)));
#endif
-#if HAVE_AVX2
+#if HAVE_AVX2 && HAVE_SSSE3
const ConvolveFunctions convolve8_avx2(
vp9_convolve8_horiz_avx2, vp9_convolve8_avg_horiz_ssse3,
vp9_convolve8_vert_avx2, vp9_convolve8_avg_vert_ssse3,
make_tuple(64, 32, &convolve8_avx2),
make_tuple(32, 64, &convolve8_avx2),
make_tuple(64, 64, &convolve8_avx2)));
-#endif
+#endif // HAVE_AVX2 && HAVE_SSSE3
#if HAVE_NEON_ASM
const ConvolveFunctions convolve8_neon(
$avx_x86_64 = $avx2_x86_64 = '';
}
+# optimizations which depend on multiple features
+if ((vpx_config("HAVE_AVX2") eq "yes") && (vpx_config("HAVE_SSSE3") eq "yes")) {
+ $avx2_ssse3 = 'avx2';
+} else {
+ $avx2_ssse3 = '';
+}
+
#
# RECON
#
$vp9_convolve_avg_neon_asm=vp9_convolve_avg_neon;
add_proto qw/void vp9_convolve8/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
-specialize qw/vp9_convolve8 sse2 ssse3 neon_asm dspr2 avx2/;
+specialize qw/vp9_convolve8 sse2 ssse3 neon_asm dspr2/, "$avx2_ssse3";
$vp9_convolve8_neon_asm=vp9_convolve8_neon;
add_proto qw/void vp9_convolve8_horiz/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
-specialize qw/vp9_convolve8_horiz sse2 ssse3 neon_asm dspr2 avx2/;
+specialize qw/vp9_convolve8_horiz sse2 ssse3 neon_asm dspr2/, "$avx2_ssse3";
$vp9_convolve8_horiz_neon_asm=vp9_convolve8_horiz_neon;
add_proto qw/void vp9_convolve8_vert/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h";
-specialize qw/vp9_convolve8_vert sse2 ssse3 neon_asm dspr2 avx2/;
+specialize qw/vp9_convolve8_vert sse2 ssse3 neon_asm dspr2/, "$avx2_ssse3";
$vp9_convolve8_vert_neon_asm=vp9_convolve8_vert_neon;
add_proto qw/void vp9_convolve8_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";
filter_x, x_step_q4, filter_y, y_step_q4, w, h); \
} \
}
-#if HAVE_AVX2
+#if HAVE_AVX2 && HAVE_SSSE3
filter8_1dfunction vp9_filter_block1d16_v8_avx2;
filter8_1dfunction vp9_filter_block1d16_h8_avx2;
filter8_1dfunction vp9_filter_block1d4_v8_ssse3;
-#if (ARCH_X86_64)
+#if ARCH_X86_64
filter8_1dfunction vp9_filter_block1d8_v8_intrin_ssse3;
filter8_1dfunction vp9_filter_block1d8_h8_intrin_ssse3;
filter8_1dfunction vp9_filter_block1d4_h8_intrin_ssse3;
#define vp9_filter_block1d8_v8_avx2 vp9_filter_block1d8_v8_intrin_ssse3
#define vp9_filter_block1d8_h8_avx2 vp9_filter_block1d8_h8_intrin_ssse3
#define vp9_filter_block1d4_h8_avx2 vp9_filter_block1d4_h8_intrin_ssse3
-#else
+#else // ARCH_X86
filter8_1dfunction vp9_filter_block1d8_v8_ssse3;
filter8_1dfunction vp9_filter_block1d8_h8_ssse3;
filter8_1dfunction vp9_filter_block1d4_h8_ssse3;
#define vp9_filter_block1d8_v8_avx2 vp9_filter_block1d8_v8_ssse3
#define vp9_filter_block1d8_h8_avx2 vp9_filter_block1d8_h8_ssse3
#define vp9_filter_block1d4_h8_avx2 vp9_filter_block1d4_h8_ssse3
-#endif
+#endif // ARCH_X86_64 / ARCH_X86
filter8_1dfunction vp9_filter_block1d16_v2_ssse3;
filter8_1dfunction vp9_filter_block1d16_h2_ssse3;
filter8_1dfunction vp9_filter_block1d8_v2_ssse3;
// const int16_t *filter_y, int y_step_q4,
// int w, int h);
FUN_CONV_2D(, avx2);
-#endif
+#endif // HAVE_AX2 && HAVE_SSSE3
#if HAVE_SSSE3
-#if (ARCH_X86_64)
+#if ARCH_X86_64
filter8_1dfunction vp9_filter_block1d16_v8_intrin_ssse3;
filter8_1dfunction vp9_filter_block1d16_h8_intrin_ssse3;
filter8_1dfunction vp9_filter_block1d8_v8_intrin_ssse3;
#define vp9_filter_block1d8_v8_ssse3 vp9_filter_block1d8_v8_intrin_ssse3
#define vp9_filter_block1d8_h8_ssse3 vp9_filter_block1d8_h8_intrin_ssse3
#define vp9_filter_block1d4_h8_ssse3 vp9_filter_block1d4_h8_intrin_ssse3
-#else
+#else // ARCH_X86
filter8_1dfunction vp9_filter_block1d16_v8_ssse3;
filter8_1dfunction vp9_filter_block1d16_h8_ssse3;
filter8_1dfunction vp9_filter_block1d8_v8_ssse3;
filter8_1dfunction vp9_filter_block1d8_h8_ssse3;
filter8_1dfunction vp9_filter_block1d4_v8_ssse3;
filter8_1dfunction vp9_filter_block1d4_h8_ssse3;
-#endif
+#endif // ARCH_X86_64 / ARCH_X86
filter8_1dfunction vp9_filter_block1d16_v8_avg_ssse3;
filter8_1dfunction vp9_filter_block1d16_h8_avg_ssse3;
filter8_1dfunction vp9_filter_block1d8_v8_avg_ssse3;
// int w, int h);
FUN_CONV_2D(, ssse3);
FUN_CONV_2D(avg_ , ssse3);
-#endif
+#endif // HAVE_SSSE3
#if HAVE_SSE2
filter8_1dfunction vp9_filter_block1d16_v8_sse2;
// int w, int h);
FUN_CONV_2D(, sse2);
FUN_CONV_2D(avg_ , sse2);
-#endif
+#endif // HAVE_SSE2