From df08e67e063a505c8fe5e88d804769d05577e0db Mon Sep 17 00:00:00 2001
From: James Zern <jzern@google.com>
Date: Fri, 16 May 2014 18:43:11 -0700
Subject: [PATCH] dsp/cpu: add AVX detection

currently unused.
https://software.intel.com/en-us/articles/introduction-to-intel-advanced-vector-extensions

similar checks exist in ffmpeg, libyuv. the visual studio inline asm is
based off of libyuv.

Change-Id: I3e233de3492172434e482607a94b99c617f11aad
---
 src/dsp/cpu.c | 35 +++++++++++++++++++++++++++++++++++
 src/dsp/dsp.h |  1 +
 2 files changed, 36 insertions(+)

diff --git a/src/dsp/cpu.c b/src/dsp/cpu.c
index 66008c7c..0dccf5b3 100644
--- a/src/dsp/cpu.c
+++ b/src/dsp/cpu.c
@@ -42,6 +42,34 @@ static WEBP_INLINE void GetCPUInfo(int cpu_info[4], int info_type) {
 #define GetCPUInfo __cpuid
 #endif
 
+#if 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];
@@ -52,6 +80,13 @@ 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;
+    }
+  }
   return 0;
 }
 VP8CPUInfo VP8GetCPUInfo = x86CPUInfo;
diff --git a/src/dsp/dsp.h b/src/dsp/dsp.h
index 072a6a0a..7181eb4f 100644
--- a/src/dsp/dsp.h
+++ b/src/dsp/dsp.h
@@ -55,6 +55,7 @@ extern "C" {
 typedef enum {
   kSSE2,
   kSSE3,
+  kAVX,
   kNEON,
   kMIPS32
 } CPUFeature;