Bug 1945602 - Update libyuv to 96f98f6915885f84b895dddcf8a553dc50a35a…

…ad r=ng

Differential Revision: https://phabricator.services.mozilla.com/D236578

media/libyuv/libyuv/BUILD.gn

@@ -395,6 +395,7 @@ if (libyuv_include_tests) {
 
     if (is_linux || is_chromeos) {
       cflags = [ "-fexceptions" ]
+
       # For enabling ASLR.
       ldflags = [ "-pie" ]
     }
@@ -406,9 +407,6 @@ if (libyuv_include_tests) {
     if (!is_ios && !libyuv_disable_jpeg) {
       defines += [ "HAVE_JPEG" ]
     }
-    if (is_android) {
-      deps += [ "//testing/android/native_test:native_test_native_code" ]
-    }
 
     # TODO(YangZhang): These lines can be removed when high accuracy
     # YUV to RGB to Neon is ported.

media/libyuv/libyuv/CMakeLists.txt

@@ -1,6 +1,6 @@
 # CMakeLists for libyuv
 # Originally created for "roxlu build system" to compile libyuv on windows
-# Run with -DTEST=ON to build unit tests
+# Run with -DUNIT_TEST=ON to build unit tests
 
 include(CheckCSourceCompiles)
 

media/libyuv/libyuv/README.chromium

@@ -1,9 +1,10 @@
 Name: libyuv
 URL: https://chromium.googlesource.com/libyuv/libyuv/
-Version: 1899
+Version: 1902
 License: BSD
 License File: LICENSE
 Shipped: yes
+Security Critical: yes
 
 Description:
 libyuv is an open source project that includes YUV conversion and scaling functionality.

media/libyuv/libyuv/build_overrides/build.gni

@@ -35,12 +35,13 @@ ignore_elf32_limitations = true
 
 # Use bundled hermetic Xcode installation maintained by Chromium,
 # except for local iOS builds where it is unsupported.
-if (host_os == "mac") {
+# Allow for mac cross compile on linux machines.
+if (host_os == "mac" || host_os == "linux") {
   _result = exec_script("//build/mac/should_use_hermetic_xcode.py",
                         [ target_os ],
                         "value")
   assert(_result != 2,
-         "Do not allow building targets with the default" +
+         "Do not allow building targets with the default " +
              "hermetic toolchain if the minimum OS version is not met.")
   use_system_xcode = _result == 0
 }

media/libyuv/libyuv/docs/environment_variables.md

@@ -29,6 +29,7 @@ By default the cpu is detected and the most advanced form of SIMD is used.  But
     LIBYUV_DISABLE_AVX512VBMI2
     LIBYUV_DISABLE_AVX512VBITALG
     LIBYUV_DISABLE_AVX10
+    LIBYUV_DISABLE_AVX10_2
     LIBYUV_DISABLE_AVXVNNI
     LIBYUV_DISABLE_AVXVNNIINT8
     LIBYUV_DISABLE_AMXINT8

media/libyuv/libyuv/include/libyuv/convert.h

@@ -598,6 +598,23 @@ int I400ToI420(const uint8_t* src_y,
                int width,
                int height);
 
+// Convert J420 to I420.
+LIBYUV_API
+int J420ToI420(const uint8_t* src_y,
+               int src_stride_y,
+               const uint8_t* src_u,
+               int src_stride_u,
+               const uint8_t* src_v,
+               int src_stride_v,
+               uint8_t* dst_y,
+               int dst_stride_y,
+               uint8_t* dst_u,
+               int dst_stride_u,
+               uint8_t* dst_v,
+               int dst_stride_v,
+               int width,
+               int height);
+
 // Convert I400 (grey) to NV21.
 LIBYUV_API
 int I400ToNV21(const uint8_t* src_y,
@@ -920,6 +937,19 @@ int RAWToJ420(const uint8_t* src_raw,
               int width,
               int height);
 
+// RGB big endian (rgb in memory) to J444.
+LIBYUV_API
+int RAWToJ444(const uint8_t* src_raw,
+              int src_stride_raw,
+              uint8_t* dst_y,
+              int dst_stride_y,
+              uint8_t* dst_u,
+              int dst_stride_u,
+              uint8_t* dst_v,
+              int dst_stride_v,
+              int width,
+              int height);
+
 // RGB16 (RGBP fourcc) little endian to I420.
 LIBYUV_API
 int RGB565ToI420(const uint8_t* src_rgb565,

media/libyuv/libyuv/include/libyuv/convert_from_argb.h

@@ -229,6 +229,19 @@ int ARGBToJ422(const uint8_t* src_argb,
                int width,
                int height);
 
+// Convert ARGB to J444.
+LIBYUV_API
+int ARGBToJ444(const uint8_t* src_argb,
+               int src_stride_argb,
+               uint8_t* dst_yj,
+               int dst_stride_yj,
+               uint8_t* dst_uj,
+               int dst_stride_uj,
+               uint8_t* dst_vj,
+               int dst_stride_vj,
+               int width,
+               int height);
+
 // Convert ARGB to J400. (JPeg full range).
 LIBYUV_API
 int ARGBToJ400(const uint8_t* src_argb,

media/libyuv/libyuv/include/libyuv/cpu_id.h

@@ -54,9 +54,10 @@ static const int kCpuHasAVX512VBMI = 0x200000;
 static const int kCpuHasAVX512VBMI2 = 0x400000;
 static const int kCpuHasAVX512VBITALG = 0x800000;
 static const int kCpuHasAVX10 = 0x1000000;
-static const int kCpuHasAVXVNNI = 0x2000000;
-static const int kCpuHasAVXVNNIINT8 = 0x4000000;
-static const int kCpuHasAMXINT8 = 0x8000000;
+static const int kCpuHasAVX10_2 = 0x2000000;
+static const int kCpuHasAVXVNNI = 0x4000000;
+static const int kCpuHasAVXVNNIINT8 = 0x8000000;
+static const int kCpuHasAMXINT8 = 0x10000000;
 
 // These flags are only valid on MIPS processors.
 static const int kCpuHasMIPS = 0x10;

media/libyuv/libyuv/include/libyuv/planar_functions.h

@@ -78,6 +78,16 @@ void Convert8To16Plane(const uint8_t* src_y,
                        int width,
                        int height);
 
+LIBYUV_API
+void Convert8To8Plane(const uint8_t* src_y,
+                      int src_stride_y,
+                      uint8_t* dst_y,
+                      int dst_stride_y,
+                      int scale,  // 220 for Y, 225 for U,V
+                      int bias,   // 16
+                      int width,
+                      int height);
+
 // Set a plane of data to a 32 bit value.
 LIBYUV_API
 void SetPlane(uint8_t* dst_y,

media/libyuv/libyuv/include/libyuv/row.h

@@ -125,9 +125,8 @@ extern "C" {
 
 // The following functions fail on gcc/clang 32 bit with fpic and framepointer.
 // caveat: clangcl uses row_win.cc which works.
-#if !defined(MOZ_PROFILING) && \
-     (defined(__x86_64__) || !defined(__pic__) || defined(__clang__) || \
-      defined(_MSC_VER))
+#if defined(__x86_64__) || !defined(__pic__) || defined(__clang__) || \
+    defined(_MSC_VER)
 // TODO(fbarchard): fix build error on android_full_debug=1
 // https://code.google.com/p/libyuv/issues/detail?id=517
 #define HAS_I422ALPHATOARGBROW_SSSE3
@@ -193,9 +192,8 @@ extern "C" {
 #define HAS_ARGBSUBTRACTROW_AVX2
 #define HAS_BLENDPLANEROW_AVX2
 
-#if !defined(MOZ_PROFILING) && \
-  (defined(__x86_64__) || !defined(__pic__) || defined(__clang__) || \
-   defined(_MSC_VER))
+#if defined(__x86_64__) || !defined(__pic__) || defined(__clang__) || \
+    defined(_MSC_VER)
 // TODO(fbarchard): fix build error on android_full_debug=1
 // https://code.google.com/p/libyuv/issues/detail?id=517
 #define HAS_I422ALPHATOARGBROW_AVX2
@@ -303,6 +301,7 @@ extern "C" {
 #define HAS_ARGBUNATTENUATEROW_AVX2
 #define HAS_CONVERT16TO8ROW_AVX2
 #define HAS_CONVERT8TO16ROW_AVX2
+#define HAS_CONVERT8TO8ROW_AVX2
 #define HAS_DETILEROW_16_AVX
 #define HAS_DIVIDEROW_16_AVX2
 #define HAS_HALFMERGEUVROW_AVX2
@@ -396,6 +395,7 @@ extern "C" {
 #define HAS_ARGBTORGB565DITHERROW_NEON
 #define HAS_ARGBTORGB565ROW_NEON
 #define HAS_ARGBTOUV444ROW_NEON
+#define HAS_ARGBTOUVJ444ROW_NEON
 #define HAS_ARGBTOUVJROW_NEON
 #define HAS_ARGBTOUVROW_NEON
 #define HAS_ARGBTOYJROW_NEON
@@ -407,6 +407,7 @@ extern "C" {
 #define HAS_BGRATOYROW_NEON
 #define HAS_BYTETOFLOATROW_NEON
 #define HAS_CONVERT16TO8ROW_NEON
+#define HAS_CONVERT8TO8ROW_NEON
 #define HAS_COPYROW_NEON
 #define HAS_DETILEROW_16_NEON
 #define HAS_DETILEROW_NEON
@@ -537,6 +538,7 @@ extern "C" {
 
 #define HAS_ARGBCOLORMATRIXROW_NEON_I8MM
 #define HAS_ARGBTOUV444ROW_NEON_I8MM
+#define HAS_ARGBTOUVJ444ROW_NEON_I8MM
 #endif
 
 // The following are available on AArch64 SVE platforms:
@@ -553,6 +555,7 @@ extern "C" {
 #define HAS_AYUVTOUVROW_SVE2
 #define HAS_AYUVTOVUROW_SVE2
 #define HAS_BGRATOUVROW_SVE2
+#define HAS_CONVERT8TO8ROW_SVE2
 #define HAS_DIVIDEROW_16_SVE2
 #define HAS_HALFFLOATROW_SVE2
 #define HAS_I210ALPHATOARGBROW_SVE2
@@ -595,6 +598,7 @@ extern "C" {
     defined(__aarch64__)
 #define HAS_ARGBMULTIPLYROW_SME
 #define HAS_CONVERT16TO8ROW_SME
+#define HAS_CONVERT8TO8ROW_SME
 #define HAS_COPYROW_SME
 #define HAS_I210ALPHATOARGBROW_SME
 #define HAS_I210TOAR30ROW_SME
@@ -1856,6 +1860,14 @@ void ARGBToUV444Row_NEON_I8MM(const uint8_t* src_argb,
                               uint8_t* dst_u,
                               uint8_t* dst_v,
                               int width);
+void ARGBToUVJ444Row_NEON(const uint8_t* src_argb,
+                          uint8_t* dst_u,
+                          uint8_t* dst_v,
+                          int width);
+void ARGBToUVJ444Row_NEON_I8MM(const uint8_t* src_argb,
+                               uint8_t* dst_u,
+                               uint8_t* dst_v,
+                               int width);
 void ARGBToUVRow_NEON(const uint8_t* src_argb,
                       int src_stride_argb,
                       uint8_t* dst_u,
@@ -2367,6 +2379,14 @@ void ARGBToUV444Row_Any_NEON_I8MM(const uint8_t* src_ptr,
                                   uint8_t* dst_u,
                                   uint8_t* dst_v,
                                   int width);
+void ARGBToUVJ444Row_Any_NEON(const uint8_t* src_ptr,
+                              uint8_t* dst_u,
+                              uint8_t* dst_v,
+                              int width);
+void ARGBToUVJ444Row_Any_NEON_I8MM(const uint8_t* src_ptr,
+                                   uint8_t* dst_u,
+                                   uint8_t* dst_v,
+                                   int width);
 void ARGBToUVRow_Any_NEON(const uint8_t* src_ptr,
                           int src_stride,
                           uint8_t* dst_u,
@@ -2684,6 +2704,11 @@ void ARGBToUV444Row_C(const uint8_t* src_argb,
                       uint8_t* dst_v,
                       int width);
 
+void ARGBToUVJ444Row_C(const uint8_t* src_argb,
+                       uint8_t* dst_u,
+                       uint8_t* dst_v,
+                       int width);
+
 void MirrorRow_AVX2(const uint8_t* src, uint8_t* dst, int width);
 void MirrorRow_SSSE3(const uint8_t* src, uint8_t* dst, int width);
 void MirrorRow_NEON(const uint8_t* src, uint8_t* dst, int width);
@@ -3643,6 +3668,42 @@ void Convert16To8Row_SME(const uint16_t* src_y,
                          int scale,
                          int width);
 
+void Convert8To8Row_C(const uint8_t* src_y,
+                      uint8_t* dst_y,
+                      int scale,
+                      int bias,
+                      int width);
+void Convert8To8Row_NEON(const uint8_t* src_y,
+                         uint8_t* dst_y,
+                         int scale,
+                         int bias,
+                         int width);
+void Convert8To8Row_Any_NEON(const uint8_t* src_ptr,
+                             uint8_t* dst_ptr,
+                             int scale,
+                             int bias,
+                             int width);
+void Convert8To8Row_SVE2(const uint8_t* src_y,
+                         uint8_t* dst_y,
+                         int scale,
+                         int bias,
+                         int width);
+void Convert8To8Row_SME(const uint8_t* src_y,
+                        uint8_t* dst_y,
+                        int scale,
+                        int bias,
+                        int width);
+void Convert8To8Row_AVX2(const uint8_t* src_y,
+                         uint8_t* dst_y,
+                         int scale,
+                         int bias,
+                         int width);
+void Convert8To8Row_Any_AVX2(const uint8_t* src_ptr,
+                             uint8_t* dst_ptr,
+                             int scale,
+                             int bias,
+                             int width);
+
 void CopyRow_SSE2(const uint8_t* src, uint8_t* dst, int width);
 void CopyRow_AVX(const uint8_t* src, uint8_t* dst, int width);
 void CopyRow_AVX512BW(const uint8_t* src, uint8_t* dst, int width);

media/libyuv/libyuv/include/libyuv/row_sve.h

@@ -1725,6 +1725,60 @@ static inline void I212ToARGBRow_SVE_SC(const uint16_t* src_y,
       : "cc", "memory", YUVTORGB_SVE_REGS);
 }
 
+#define CONVERT8TO8_SVE                                  \
+  "ld1b        {z0.b}, p0/z, [%[src]]                \n" \
+  "ld1b        {z1.b}, p1/z, [%[src], #1, mul vl]    \n" \
+  "incb        %[src], all, mul #2                   \n" \
+  "subs        %w[width], %w[width], %w[vl], lsl #1  \n" \
+  "umulh       z0.b, z0.b, z2.b                      \n" \
+  "umulh       z1.b, z1.b, z2.b                      \n" \
+  "prfm        pldl1keep, [%[src], 448]              \n" \
+  "add         z0.b, z0.b, z3.b                      \n" \
+  "add         z1.b, z1.b, z3.b                      \n" \
+  "st1b        {z0.b}, p0, [%[dst]]                  \n" \
+  "st1b        {z1.b}, p1, [%[dst], #1, mul vl]      \n" \
+  "incb        %[dst], all, mul #2                   \n"
+
+static inline void Convert8To8Row_SVE_SC(const uint8_t* src_y,
+                                         uint8_t* dst_y,
+                                         int scale,
+                                         int bias,
+                                         int width) STREAMING_COMPATIBLE {
+  uint64_t vl;
+  asm volatile(
+      "dup      z2.b, %w[scale]                         \n"
+      "dup      z3.b, %w[bias]                          \n"
+      "cntb     %[vl]                                   \n"
+      "subs     %w[width], %w[width], %w[vl], lsl #1    \n"
+      "b.lt     2f                                      \n"
+
+      // Run bulk of computation with all-true predicates to avoid predicate
+      // generation overhead.
+      "ptrue    p0.b                                    \n"
+      "ptrue    p1.b                                    \n"
+      "1:                                               \n"  //
+      CONVERT8TO8_SVE
+      "b.ge     1b                                      \n"
+
+      "2:                                               \n"
+      "adds     %w[width], %w[width], %w[vl], lsl #1    \n"
+      "b.eq     99f                                     \n"
+
+      // Calculate predicates for the final iteration to deal with the tail.
+      "whilelt     p0.b, wzr, %w2                       \n"
+      "whilelt     p1.b, %w[vl], %w2                    \n"  //
+      CONVERT8TO8_SVE
+
+      "99:                                              \n"
+      : [src] "+r"(src_y),    // %[src]
+        [dst] "+r"(dst_y),    // %[dst]
+        [width] "+r"(width),  // %[width]
+        [vl] "=&r"(vl)        // %[vl]
+      : [scale] "r"(scale),   // %[scale]
+        [bias] "r"(bias)      // %[bias]
+      : "cc", "memory", "z0", "z1", "z2", "z3", "p0", "p1");
+}
+
 #endif  // !defined(LIBYUV_DISABLE_SVE) && defined(__aarch64__)
 
 #ifdef __cplusplus

media/libyuv/libyuv/include/libyuv/version.h

@@ -11,6 +11,6 @@
 #ifndef INCLUDE_LIBYUV_VERSION_H_
 #define INCLUDE_LIBYUV_VERSION_H_
 
-#define LIBYUV_VERSION 1899
+#define LIBYUV_VERSION 1902
 
 #endif  // INCLUDE_LIBYUV_VERSION_H_

media/libyuv/libyuv/infra/config/commit-queue.cfg

@@ -91,13 +91,13 @@ config_groups {
       }
       builders {
         name: "libyuv/try/linux_msan"
+        experiment_percentage: 100
       }
       builders {
         name: "libyuv/try/linux_rel"
       }
       builders {
         name: "libyuv/try/linux_tsan2"
-        experiment_percentage: 100
       }
       builders {
         name: "libyuv/try/linux_ubsan"