Sync-patch with libwebp ver 0.4.1-rc1.

Sync-patch with libwebp ver 0.4.1-rc1 (change#I5346984d2).
  - NEON assembly optimizations:
    - ~25% faster lossy decode / encode (-m 4)
    - ~10% faster lossless decode
    - ~5-10% faster lossless encode (-m 3/4)
  - Arch64 (arm64) & MIPS support/optimizations.

Change-Id: I855b65cec8fad5ec567c276b698e7714dc4bffd2

1 files changed, 53 insertions, 3 deletions

diff --git a/src/dsp/cpu.c b/src/dsp/cpu.c
index 0988df27..a2954981 100644
--- a/src/dsp/cpu.c
+++ b/src/dsp/cpu.c
@@ -14,7 +14,7 @@
 #include "./dsp.h"
 
 #if defined(__ANDROID__)
-#include "./cpu-features.h"
+#include "cpu-features.h"
 #endif
 
 //------------------------------------------------------------------------------
@@ -38,10 +38,41 @@ static WEBP_INLINE void GetCPUInfo(int cpu_info[4], int info_type) {
     : "=a"(cpu_info[0]), "=b"(cpu_info[1]), "=c"(cpu_info[2]), "=d"(cpu_info[3])
     : "a"(info_type));
 }
+#elif defined(_MSC_FULL_VER) && _MSC_FULL_VER >= 150030729  // >= VS2008 SP1
+#define GetCPUInfo(info, type) __cpuidex(info, type, 0)  // set ecx=0
 #elif defined(WEBP_MSC_SSE2)
 #define GetCPUInfo __cpuid
 #endif
 
+// NaCl has no support for xgetbv or the raw opcode.
+#if !defined(__native_client__) && (defined(__i386__) || defined(__x86_64__))
+static WEBP_INLINE uint64_t xgetbv(void) {
+  const uint32_t ecx = 0;
+  uint32_t eax, edx;
+  // Use the raw opcode for xgetbv for compatibility with older toolchains.
+  __asm__ volatile (
+    ".byte 0x0f, 0x01, 0xd0\n"
+    : "=a"(eax), "=d"(edx) : "c" (ecx));
+  return ((uint64_t)edx << 32) | eax;
+}
+#elif defined(_MSC_FULL_VER) && _MSC_FULL_VER >= 160040219  // >= VS2010 SP1
+#define xgetbv() _xgetbv(0)
+#elif defined(_M_IX86)
+static WEBP_INLINE uint64_t xgetbv(void) {
+  uint32_t eax_, edx_;
+  __asm {
+    xor ecx, ecx  // ecx = 0
+    // Use the raw opcode for xgetbv for compatibility with older toolchains.
+    __asm _emit 0x0f __asm _emit 0x01 __asm _emit 0xd0
+    mov eax_, eax
+    mov edx_, edx
+  }
+  return ((uint64_t)edx_ << 32) | eax_;
+}
+#else
+#define xgetbv() 0U  // no AVX for older x64 or unrecognized toolchains.
+#endif
+
 #if defined(__i386__) || defined(__x86_64__) || defined(WEBP_MSC_SSE2)
 static int x86CPUInfo(CPUFeature feature) {
   int cpu_info[4];
@@ -52,10 +83,23 @@ static int x86CPUInfo(CPUFeature feature) {
   if (feature == kSSE3) {
     return 0 != (cpu_info[2] & 0x00000001);
   }
+  if (feature == kAVX) {
+    // bits 27 (OSXSAVE) & 28 (256-bit AVX)
+    if ((cpu_info[2] & 0x18000000) == 0x18000000) {
+      // XMM state and YMM state enabled by the OS.
+      return (xgetbv() & 0x6) == 0x6;
+    }
+  }
+  if (feature == kAVX2) {
+    if (x86CPUInfo(kAVX)) {
+      GetCPUInfo(cpu_info, 7);
+      return ((cpu_info[1] & 0x00000020) == 0x00000020);
+    }
+  }
   return 0;
 }
 VP8CPUInfo VP8GetCPUInfo = x86CPUInfo;
-#elif defined(WEBP_ANDROID_NEON)
+#elif defined(WEBP_ANDROID_NEON)  // NB: needs to be before generic NEON test.
 static int AndroidCPUInfo(CPUFeature feature) {
   const AndroidCpuFamily cpu_family = android_getCpuFamily();
   const uint64_t cpu_features = android_getCpuFeatures();
@@ -66,7 +110,7 @@ static int AndroidCPUInfo(CPUFeature feature) {
   return 0;
 }
 VP8CPUInfo VP8GetCPUInfo = AndroidCPUInfo;
-#elif defined(__ARM_NEON__)
+#elif defined(WEBP_USE_NEON)
 // define a dummy function to enable turning off NEON at runtime by setting
 // VP8DecGetCPUInfo = NULL
 static int armCPUInfo(CPUFeature feature) {
@@ -74,6 +118,12 @@ static int armCPUInfo(CPUFeature feature) {
   return 1;
 }
 VP8CPUInfo VP8GetCPUInfo = armCPUInfo;
+#elif defined(__mips__)
+static int mipsCPUInfo(CPUFeature feature) {
+  (void)feature;
+  return 1;
+}
+VP8CPUInfo VP8GetCPUInfo = mipsCPUInfo;
 #else
 VP8CPUInfo VP8GetCPUInfo = NULL;
 #endif