Bug 1148639 - Update libvpx import script and patches. r=kinetik

From 3f94fd4fc68971f0dc2148aa19b4e146262d15af Mon Sep 17 00:00:00 2001 Matched against upstream commit afad1a84c1.
2015-03-27 14:38:35 -07:00 · 2015-03-27 14:38:35 -07:00 · c536fc79d4
--- a/media/libvpx/apple-clang.patch
+++ b/media/libvpx/apple-clang.patch
@ -1,29 +1,71 @@
 diff --git a/media/libvpx/vp9/common/x86/vp9_subpixel_8t_intrin_avx2.c b/media/libvpx/vp9/common/x86/vp9_subpixel_8t_intrin_avx2.c
+index 0ffb1bc..733693d 100644
 --- a/media/libvpx/vp9/common/x86/vp9_subpixel_8t_intrin_avx2.c
 +++ b/media/libvpx/vp9/common/x86/vp9_subpixel_8t_intrin_avx2.c
-@@ -27,21 +27,24 @@ DECLARE_ALIGNED(32, static const uint8_t
-   4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12
- };
- 
- DECLARE_ALIGNED(32, static const uint8_t, filt4_global_avx2[32]) = {
+@@ -28,6 +28,26 @@ DECLARE_ALIGNED(32, const unsigned char, filt4_global_avx2[32])= {
   6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12, 12, 13, 13, 14,
-   6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12, 12, 13, 13, 14
- };
+   6, 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12, 12, 13, 13, 14};
 
- #if defined(__clang__)
- # if __clang_major__ < 3 || (__clang_major__ == 3 && __clang_minor__ <= 3) || \
-      (defined(__APPLE__) && __clang_major__ == 5 && __clang_minor__ == 0)
-+      (defined(__APPLE__) && \
-+          (__clang_major__ == 4 && __clang_minor__ >= 0 && \
-+              __clang_minor__ <= 2) || \
-+          (__clang_major__ == 5 && __clang_minor__ == 0))
- #  define MM256_BROADCASTSI128_SI256(x) \
-        _mm_broadcastsi128_si256((__m128i const *)&(x))
- # else  // clang > 3.3, and not 5.0 on macosx.
- #  define MM256_BROADCASTSI128_SI256(x) _mm256_broadcastsi128_si256(x)
- # endif  // clang <= 3.3
- #elif defined(__GNUC__)
- # if __GNUC__ < 4 || (__GNUC__ == 4 && __GNUC_MINOR__ <= 6)
- #  define MM256_BROADCASTSI128_SI256(x) \
-        _mm_broadcastsi128_si256((__m128i const *)&(x))
- # elif __GNUC__ == 4 && __GNUC_MINOR__ == 7
+#if defined(__clang__)
+# if __clang_major__ < 3 || (__clang_major__ == 3 && __clang_minor__ <= 3) || \
+     (defined(__APPLE__) && __clang_major__ == 5 && __clang_minor__ == 0)
+#  define MM256_BROADCASTSI128_SI256(x) \
+         _mm_broadcastsi128_si256((__m128i const *)&(x))
+# else  // clang > 3.3, and not 5.0 on macosx.
+#  define MM256_BROADCASTSI128_SI256(x) _mm256_broadcastsi128_si256(x)
+# endif  // clang <= 3.3
+#elif defined(__GNUC__)
+# if __GNUC__ < 4 || (__GNUC__ == 4 && __GNUC_MINOR__ <= 6)
+#  define MM256_BROADCASTSI128_SI256(x) \
+         _mm_broadcastsi128_si256((__m128i const *)&(x))
+# elif __GNUC__ == 4 && __GNUC_MINOR__ == 7
+#  define MM256_BROADCASTSI128_SI256(x) _mm_broadcastsi128_si256(x)
+# else  // gcc > 4.7
+#  define MM256_BROADCASTSI128_SI256(x) _mm256_broadcastsi128_si256(x)
+# endif  // gcc <= 4.6
+#else  // !(gcc || clang)
+# define MM256_BROADCASTSI128_SI256(x) _mm256_broadcastsi128_si256(x)
+#endif  // __clang__
+ 
+ void vp9_filter_block1d16_h8_avx2(unsigned char *src_ptr,
+                                   unsigned int src_pixels_per_line,
+@@ -50,18 +70,7 @@ void vp9_filter_block1d16_h8_avx2(unsigned char *src_ptr,
+   // in both lanes of 128 bit register.
+   filtersReg =_mm_packs_epi16(filtersReg, filtersReg);
+   // have the same data in both lanes of a 256 bit register
+-#if defined (__GNUC__)
+-#if ( __GNUC__ < 4 || (__GNUC__ == 4 && \
+-(__GNUC_MINOR__ < 6 || (__GNUC_MINOR__ == 6 && __GNUC_PATCHLEVEL__ > 0))))
+-  filtersReg32 = _mm_broadcastsi128_si256((__m128i const *)&filtersReg);
+-#elif(__GNUC__ == 4 && (__GNUC_MINOR__ == 7 && __GNUC_PATCHLEVEL__ > 0))
+-  filtersReg32 = _mm_broadcastsi128_si256(filtersReg);
+-#else
+-  filtersReg32 = _mm256_broadcastsi128_si256(filtersReg);
+-#endif
+-#else
+-  filtersReg32 = _mm256_broadcastsi128_si256(filtersReg);
+-#endif
+  filtersReg32 = MM256_BROADCASTSI128_SI256(filtersReg);
+ 
+   // duplicate only the first 16 bits (first and second byte)
+   // across 256 bit register
+@@ -306,18 +315,7 @@ void vp9_filter_block1d16_v8_avx2(unsigned char *src_ptr,
+   // same data in both lanes of 128 bit register.
+   filtersReg =_mm_packs_epi16(filtersReg, filtersReg);
+   // have the same data in both lanes of a 256 bit register
+-#if defined (__GNUC__)
+-#if ( __GNUC__ < 4 || (__GNUC__ == 4 && \
+-(__GNUC_MINOR__ < 6 || (__GNUC_MINOR__ == 6 && __GNUC_PATCHLEVEL__ > 0))))
+-  filtersReg32 = _mm_broadcastsi128_si256((__m128i const *)&filtersReg);
+-#elif(__GNUC__ == 4 && (__GNUC_MINOR__ == 7 && __GNUC_PATCHLEVEL__ > 0))
+-  filtersReg32 = _mm_broadcastsi128_si256(filtersReg);
+-#else
+-  filtersReg32 = _mm256_broadcastsi128_si256(filtersReg);
+-#endif
+-#else
+-  filtersReg32 = _mm256_broadcastsi128_si256(filtersReg);
+-#endif
+  filtersReg32 = MM256_BROADCASTSI128_SI256(filtersReg);
+ 
+   // duplicate only the first 16 bits (first and second byte)
+   // across 256 bit register
--- a/media/libvpx/msvc2015.patch
+++ b/media/libvpx/msvc2015.patch
@ -5,11 +5,11 @@ diff --git a/media/libvpx/vp9/common/vp9_systemdependent.h b/media/libvpx/vp9/co
 #define VP9_COMMON_VP9_SYSTEMDEPENDENT_H_
 
 #ifdef _MSC_VER
- # include <math.h>  // the ceil() definition must precede intrin.h
 # if _MSC_VER > 1310 && (defined(_M_X64) || defined(_M_IX86))
 #  include <intrin.h>
 #  define USE_MSC_INTRIN
 # endif
+ # include <math.h>
 -# define snprintf _snprintf
 +# if _MSC_VER < 1900
 +#  define snprintf _snprintf
--- a/media/libvpx/stdint.patch
+++ b/media/libvpx/stdint.patch
@ -6,9 +6,9 @@
 
 +#if !defined(VPX_DONT_DEFINE_STDINT_TYPES)
 +
- #if defined(_MSC_VER)
- #define VPX_FORCE_INLINE __forceinline
- #define VPX_INLINE __inline
+ #if (defined(_MSC_VER) && (_MSC_VER < 1600)) || defined(VPX_EMULATE_INTTYPES)
+ typedef signed char  int8_t;
+ typedef signed short int16_t;
@@ -56,6 +58,8 @@
 
 #endif
--- a/media/libvpx/update.py
+++ b/media/libvpx/update.py
@ -217,8 +217,8 @@ files = {
    ],
    'X86-64_ASM': [
        'third_party/x86inc/x86inc.asm',
-        'vp8/common/x86/loopfilter_block_sse2_x86_64.asm',
-        'vp9/encoder/x86/vp9_quantize_ssse3_x86_64.asm',
+        'vp8/common/x86/loopfilter_block_sse2.asm',
+        'vp9/encoder/x86/vp9_quantize_ssse3.asm',
    ],
    'SOURCES': [
        'vp8/common/rtcd.c',
@ -226,6 +226,7 @@ files = {
        'vp8/encoder/bitstream.c',
        'vp8/encoder/onyx_if.c',
        'vp8/vp8_dx_iface.c',
+        'vp9/common/generic/vp9_systemdependent.c',
        'vp9/common/vp9_alloccommon.c',
        'vp9/common/vp9_blockd.c',
        'vp9/common/vp9_common_data.c',
@ -249,48 +250,39 @@ files = {
        'vp9/common/vp9_scale.c',
        'vp9/common/vp9_scan.c',
        'vp9/common/vp9_seg_common.c',
-        'vp9/common/vp9_thread.c',
        'vp9/common/vp9_tile_common.c',
        'vp9/decoder/vp9_decodeframe.c',
        'vp9/decoder/vp9_decodemv.c',
-        'vp9/decoder/vp9_decoder.c',
        'vp9/decoder/vp9_detokenize.c',
        'vp9/decoder/vp9_dsubexp.c',
        'vp9/decoder/vp9_dthread.c',
+        'vp9/decoder/vp9_onyxd_if.c',
        'vp9/decoder/vp9_reader.c',
        'vp9/encoder/vp9_bitstream.c',
-        'vp9/encoder/vp9_aq_complexity.c',
-        'vp9/encoder/vp9_aq_cyclicrefresh.c',
-        'vp9/encoder/vp9_aq_variance.c',
-        'vp9/encoder/vp9_context_tree.c',
-        'vp9/encoder/vp9_cost.c',
        'vp9/encoder/vp9_dct.c',
        'vp9/encoder/vp9_encodeframe.c',
        'vp9/encoder/vp9_encodemb.c',
        'vp9/encoder/vp9_encodemv.c',
-        'vp9/encoder/vp9_encoder.c',
        'vp9/encoder/vp9_extend.c',
        'vp9/encoder/vp9_firstpass.c',
        'vp9/encoder/vp9_lookahead.c',
        'vp9/encoder/vp9_mbgraph.c',
        'vp9/encoder/vp9_mcomp.c',
+        'vp9/encoder/vp9_onyx_if.c',
        'vp9/encoder/vp9_picklpf.c',
        'vp9/encoder/vp9_pickmode.c',
        'vp9/encoder/vp9_quantize.c',
        'vp9/encoder/vp9_ratectrl.c',
-        'vp9/encoder/vp9_rd.c',
        'vp9/encoder/vp9_rdopt.c',
        'vp9/encoder/vp9_resize.c',
        'vp9/encoder/vp9_sad.c',
        'vp9/encoder/vp9_segmentation.c',
-        'vp9/encoder/vp9_speed_features.c',
        'vp9/encoder/vp9_subexp.c',
-        'vp9/encoder/vp9_svc_layercontext.c',
        'vp9/encoder/vp9_temporal_filter.c',
        'vp9/encoder/vp9_tokenize.c',
        'vp9/encoder/vp9_treewriter.c',
+        'vp9/encoder/vp9_vaq.c',
        'vp9/encoder/vp9_variance.c',
-        'vp9/encoder/vp9_write_bit_buffer.c',
        'vp9/encoder/vp9_writer.c',
        'vp9/vp9_cx_iface.c',
        'vp9/vp9_dx_iface.c',
@ -307,9 +299,9 @@ manual = [
    # special case in moz.build
    'vp8/encoder/boolhuff.c',

-    # 64bit only
-    'vp8/common/x86/loopfilter_block_sse2_x86_64.asm',
-    'vp9/encoder/x86/vp9_quantize_ssse3_x86_64.asm',
+    # These 64-bit only files end up in X86_ASM. Filter them out.
+    'vp8/common/x86/loopfilter_block_sse2.asm',
+    'vp9/encoder/x86/vp9_quantize_ssse3.asm',

    # offsets are special cased in Makefile.in
    'vp8/encoder/vp8_asm_enc_offsets.c',
@ -530,7 +522,7 @@ def update_and_remove_files(prefix, libvpx_files, files):
 def apply_patches():
    # Patch to permit vpx users to specify their own <stdint.h> types.
    os.system("patch -p0 < stdint.patch")
-    # Patch to allow older versions of Apple's clang to build libvpx.
+    # Patch for AVX intrinsic support with Apple's clang.
    os.system("patch -p3 < apple-clang.patch")
    # Patch to allow MSVC 2015 to compile libvpx
    os.system("patch -p3 < msvc2015.patch")