AVX512 feature detects for cnl and icl

Key instruction sets added for each microarchitecture:

AVX512BW, AVX512VL, AVX512DQ - skylake server or later
AVX512_VBMI, AVX512_IFMA - cannon lake or later
AVX512_BITALG, AVX512_VBMI2, AVX512_VPOPCNTDQ, AVX512_VNNI, GFNI, VAES, VPCLMULQDQ - ice lake or later

Bug: libyuv:752
Test: ~/intelsde/sde -icl -- out/Release/libyuv_unittest --gtest_filter=*Cpu*
Change-Id: I9ee28904c90009d66721b9f805a440c5fc2da122
Reviewed-on: https://chromium-review.googlesource.com/755617Reviewed-by: Frank Barchard <fbarchard@google.com>
Reviewed-by: richard winterton <rrwinterton@gmail.com>

README.chromium

 Name: libyuv
 URL: http://code.google.com/p/libyuv/
-Version: 1675
+Version: 1676
 License: BSD
 License File: LICENSE
 

docs/environment_variables.md

@@ -14,7 +14,7 @@ By default the cpu is detected and the most advanced form of SIMD is used.  But
     LIBYUV_DISABLE_SSE42
     LIBYUV_DISABLE_AVX
     LIBYUV_DISABLE_AVX2
-    LIBYUV_DISABLE_AVX3
+    LIBYUV_DISABLE_AVX512BW
     LIBYUV_DISABLE_ERMS
     LIBYUV_DISABLE_FMA3
     LIBYUV_DISABLE_DSPR2

include/libyuv/cpu_id.h

@@ -36,15 +36,19 @@ static const int kCpuHasAVX = 0x200;
 static const int kCpuHasAVX2 = 0x400;
 static const int kCpuHasERMS = 0x800;
 static const int kCpuHasFMA3 = 0x1000;
-static const int kCpuHasAVX3 = 0x2000;
-static const int kCpuHasF16C = 0x4000;
-
-// 0x8000 reserved for future X86 flags.
+static const int kCpuHasF16C = 0x2000;
+static const int kCpuHasGFNI = 0x4000;
+static const int kCpuHasAVX512BW = 0x8000;
+static const int kCpuHasAVX512VL = 0x10000;
+static const int kCpuHasAVX512VBMI = 0x20000;
+static const int kCpuHasAVX512VBMI2 = 0x40000;
+static const int kCpuHasAVX512VBITALG = 0x80000;
+static const int kCpuHasAVX512VPOPCNTDQ = 0x100000;
 
 // These flags are only valid on MIPS processors.
-static const int kCpuHasMIPS = 0x10000;
-static const int kCpuHasDSPR2 = 0x20000;
-static const int kCpuHasMSA = 0x40000;
+static const int kCpuHasMIPS = 0x200000;
+static const int kCpuHasDSPR2 = 0x400000;
+static const int kCpuHasMSA = 0x800000;
 
 // Optional init function. TestCpuFlag does an auto-init.
 // Returns cpu_info flags.

include/libyuv/version.h

@@ -11,6 +11,6 @@
 #ifndef INCLUDE_LIBYUV_VERSION_H_
 #define INCLUDE_LIBYUV_VERSION_H_
 
-#define LIBYUV_VERSION 1675
+#define LIBYUV_VERSION 1676
 
 #endif  // INCLUDE_LIBYUV_VERSION_H_

source/cpu_id.cc

@@ -242,10 +242,17 @@ static SAFEBUFFERS int GetCpuFlags(void) {
 
     // Detect AVX512bw
     if ((GetXCR0() & 0xe0) == 0xe0) {
-      cpu_info |= (cpu_info7[1] & 0x40000000) ? kCpuHasAVX3 : 0;
+      cpu_info |= (cpu_info7[1] & 0x40000000) ? kCpuHasAVX512BW : 0;
+      cpu_info |= (cpu_info7[1] & 0x80000000) ? kCpuHasAVX512VL : 0;
+      cpu_info |= (cpu_info7[2] & 0x00000002) ? kCpuHasAVX512VBMI: 0;
+      cpu_info |= (cpu_info7[2] & 0x00000040) ? kCpuHasAVX512VBMI2: 0;
+      cpu_info |= (cpu_info7[2] & 0x00001000) ? kCpuHasAVX512VBITALG: 0;
+      cpu_info |= (cpu_info7[2] & 0x00004000) ? kCpuHasAVX512VPOPCNTDQ: 0;
+      cpu_info |= (cpu_info7[2] & 0x00000100) ? kCpuHasGFNI: 0;
     }
   }
 
+  // TODO(fbarchard): Consider moving these to gtest
   // Environment variable overrides for testing.
   if (TestEnv("LIBYUV_DISABLE_X86")) {
     cpu_info &= ~kCpuHasX86;
@@ -274,12 +281,12 @@ static SAFEBUFFERS int GetCpuFlags(void) {
   if (TestEnv("LIBYUV_DISABLE_FMA3")) {
     cpu_info &= ~kCpuHasFMA3;
   }
-  if (TestEnv("LIBYUV_DISABLE_AVX3")) {
-    cpu_info &= ~kCpuHasAVX3;
-  }
   if (TestEnv("LIBYUV_DISABLE_F16C")) {
     cpu_info &= ~kCpuHasF16C;
   }
+  if (TestEnv("LIBYUV_DISABLE_AVX512BW")) {
+    cpu_info &= ~kCpuHasAVX512BW;
+  }
 
 #endif
 #if defined(__mips__) && defined(__linux__)

unit_test/cpu_test.cc

@@ -21,38 +21,55 @@ namespace libyuv {
 TEST_F(LibYUVBaseTest, TestCpuHas) {
   int cpu_flags = TestCpuFlag(-1);
   printf("Cpu Flags %x\n", cpu_flags);
+#if defined(__arm__) || defined(__aarch64__)
   int has_arm = TestCpuFlag(kCpuHasARM);
   printf("Has ARM %x\n", has_arm);
   int has_neon = TestCpuFlag(kCpuHasNEON);
   printf("Has NEON %x\n", has_neon);
+#endif
   int has_x86 = TestCpuFlag(kCpuHasX86);
-  printf("Has X86 %x\n", has_x86);
   int has_sse2 = TestCpuFlag(kCpuHasSSE2);
-  printf("Has SSE2 %x\n", has_sse2);
   int has_ssse3 = TestCpuFlag(kCpuHasSSSE3);
-  printf("Has SSSE3 %x\n", has_ssse3);
   int has_sse41 = TestCpuFlag(kCpuHasSSE41);
-  printf("Has SSE4.1 %x\n", has_sse41);
   int has_sse42 = TestCpuFlag(kCpuHasSSE42);
-  printf("Has SSE4.2 %x\n", has_sse42);
   int has_avx = TestCpuFlag(kCpuHasAVX);
-  printf("Has AVX %x\n", has_avx);
   int has_avx2 = TestCpuFlag(kCpuHasAVX2);
-  printf("Has AVX2 %x\n", has_avx2);
   int has_erms = TestCpuFlag(kCpuHasERMS);
-  printf("Has ERMS %x\n", has_erms);
   int has_fma3 = TestCpuFlag(kCpuHasFMA3);
+  int has_f16c = TestCpuFlag(kCpuHasF16C); 
+  int has_gfni = TestCpuFlag(kCpuHasGFNI);
+  int has_avx512bw = TestCpuFlag(kCpuHasAVX512BW);
+  int has_avx512vl = TestCpuFlag(kCpuHasAVX512VL);
+  int has_avx512vbmi = TestCpuFlag(kCpuHasAVX512VBMI);
+  int has_avx512vbmi2 = TestCpuFlag(kCpuHasAVX512VBMI2);
+  int has_avx512vbitalg = TestCpuFlag(kCpuHasAVX512VBITALG);
+  int has_avx512vpopcntdq = TestCpuFlag(kCpuHasAVX512VPOPCNTDQ);
+  printf("Has X86 %x\n", has_x86);
+  printf("Has SSE2 %x\n", has_sse2);
+  printf("Has SSSE3 %x\n", has_ssse3);
+  printf("Has SSE4.1 %x\n", has_sse41);
+  printf("Has SSE4.2 %x\n", has_sse42);
+  printf("Has AVX %x\n", has_avx);
+  printf("Has AVX2 %x\n", has_avx2);
+  printf("Has ERMS %x\n", has_erms);
   printf("Has FMA3 %x\n", has_fma3);
-  int has_avx3 = TestCpuFlag(kCpuHasAVX3);
-  printf("Has AVX3 %x\n", has_avx3);
-  int has_f16c = TestCpuFlag(kCpuHasF16C);
   printf("Has F16C %x\n", has_f16c);
+  printf("Has GFNI %x\n", has_gfni);
+  printf("Has AVX512BW %x\n", has_avx512bw);
+  printf("Has AVX512VL %x\n", has_avx512vl);
+  printf("Has AVX512VBMI %x\n", has_avx512vbmi);
+  printf("Has AVX512VBMI2 %x\n", has_avx512vbmi2);
+  printf("Has AVX512VBITALG %x\n", has_avx512vbitalg);
+  printf("Has AVX512VPOPCNTDQ %x\n", has_avx512vpopcntdq);
+
+#if defined(__mips__)
   int has_mips = TestCpuFlag(kCpuHasMIPS);
   printf("Has MIPS %x\n", has_mips);
   int has_dspr2 = TestCpuFlag(kCpuHasDSPR2);
   printf("Has DSPR2 %x\n", has_dspr2);
   int has_msa = TestCpuFlag(kCpuHasMSA);
   printf("Has MSA %x\n", has_msa);
+#endif
 }
 
 TEST_F(LibYUVBaseTest, TestCpuCompilerEnabled) {

util/cpuid.c

@@ -79,20 +79,32 @@ int main(int argc, const char* argv[]) {
     int has_sse42 = TestCpuFlag(kCpuHasSSE42);
     int has_avx = TestCpuFlag(kCpuHasAVX);
     int has_avx2 = TestCpuFlag(kCpuHasAVX2);
-    int has_avx3 = TestCpuFlag(kCpuHasAVX3);
     int has_erms = TestCpuFlag(kCpuHasERMS);
     int has_fma3 = TestCpuFlag(kCpuHasFMA3);
-    int has_f16c = TestCpuFlag(kCpuHasF16C);
+    int has_f16c = TestCpuFlag(kCpuHasF16C); 
+    int has_gfni = TestCpuFlag(kCpuHasGFNI);
+    int has_avx512bw = TestCpuFlag(kCpuHasAVX512BW);
+    int has_avx512vl = TestCpuFlag(kCpuHasAVX512VL);
+    int has_avx512vbmi = TestCpuFlag(kCpuHasAVX512VBMI);
+    int has_avx512vbmi2 = TestCpuFlag(kCpuHasAVX512VBMI2);
+    int has_avx512vbitalg = TestCpuFlag(kCpuHasAVX512VBITALG);
+    int has_avx512vpopcntdq = TestCpuFlag(kCpuHasAVX512VPOPCNTDQ);
     printf("Has SSE2 %x\n", has_sse2);
     printf("Has SSSE3 %x\n", has_ssse3);
     printf("Has SSE4.1 %x\n", has_sse41);
     printf("Has SSE4.2 %x\n", has_sse42);
     printf("Has AVX %x\n", has_avx);
     printf("Has AVX2 %x\n", has_avx2);
-    printf("Has AVX3 %x\n", has_avx3);
     printf("Has ERMS %x\n", has_erms);
     printf("Has FMA3 %x\n", has_fma3);
     printf("Has F16C %x\n", has_f16c);
+    printf("Has GFNI %x\n", has_gfni);
+    printf("Has AVX512BW %x\n", has_avx512bw);
+    printf("Has AVX512VL %x\n", has_avx512vl);
+    printf("Has AVX512VBMI %x\n", has_avx512vbmi);
+    printf("Has AVX512VBMI2 %x\n", has_avx512vbmi2);
+    printf("Has AVX512VBITALG %x\n", has_avx512vbitalg);
+    printf("Has AVX512VPOPCNTDQ %x\n", has_avx512vpopcntdq);
   }
   return 0;
 }