4 files changed, 10 insertions, 1 deletions

diff --git a/src/lib_jit.c b/src/lib_jit.c
index 402ca35c..c6330c49 100644
--- a/src/lib_jit.c
+++ b/src/lib_jit.c
@@ -668,6 +668,11 @@ static uint32_t jit_cpudetect(lua_State *L)
       if (fam >= 0x00000f00)  /* K8, K10. */
         flags |= JIT_F_PREFER_IMUL;
     }
+    if (vendor[0] >= 7) {
+      uint32_t xfeatures[4];
+      lj_vm_cpuid(7, xfeatures);
+      flags |= ((xfeatures[1] >> 8)&1) * JIT_F_BMI2;
+    }
 #endif
   }
   /* Check for required instruction set support on x86 (unnecessary on x64). */
diff --git a/src/lj_jit.h b/src/lj_jit.h
index 5a51357e..2d2e833a 100644
--- a/src/lj_jit.h
+++ b/src/lj_jit.h
@@ -19,10 +19,11 @@
 #define JIT_F_SSE4_1            0x00000040
 #define JIT_F_PREFER_IMUL       0x00000080
 #define JIT_F_LEA_AGU           0x00000100
+#define JIT_F_BMI2              0x00000200
 
 /* Names for the CPU-specific flags. Must match the order above. */
 #define JIT_F_CPU_FIRST         JIT_F_SSE2
-#define JIT_F_CPUSTRING         "\4SSE2\4SSE3\6SSE4.1\3AMD\4ATOM"
+#define JIT_F_CPUSTRING         "\4SSE2\4SSE3\6SSE4.1\3AMD\4ATOM\4BMI2"
 #elif LJ_TARGET_ARM
 #define JIT_F_ARMV6_            0x00000010
 #define JIT_F_ARMV6T2_          0x00000020
diff --git a/src/vm_x64.dasc b/src/vm_x64.dasc
index bc2419c1..99ebb9b9 100644
--- a/src/vm_x64.dasc
+++ b/src/vm_x64.dasc
@@ -2639,6 +2639,7 @@ static void build_subroutines(BuildCtx *ctx)
   |  mov eax, CARG1d
   |  .if X64WIN; push rsi; mov rsi, CARG2; .endif
   |  push rbx
+  |  xor ecx, ecx
   |  cpuid
   |  mov [rsi], eax
   |  mov [rsi+4], ebx
diff --git a/src/vm_x86.dasc b/src/vm_x86.dasc
index fcef660c..f108c0b5 100644
--- a/src/vm_x86.dasc
+++ b/src/vm_x86.dasc
@@ -3026,6 +3026,7 @@ static void build_subroutines(BuildCtx *ctx)
   |  mov eax, CARG1d
   |  .if X64WIN; push rsi; mov rsi, CARG2; .endif
   |  push rbx
+  |  xor ecx, ecx
   |  cpuid
   |  mov [rsi], eax
   |  mov [rsi+4], ebx
@@ -3049,6 +3050,7 @@ static void build_subroutines(BuildCtx *ctx)
   |  mov eax, [esp+4]                   // Argument 1 is function number.
   |  push edi
   |  push ebx
+  |  xor ecx, ecx
   |  cpuid
   |  mov edi, [esp+16]                  // Argument 2 is result area.
   |  mov [edi], eax