volk: avx support for MSVC

3 files changed, 20 insertions, 4 deletions

diff --git a/volk/gen/compilers.xml b/volk/gen/compilers.xml
index 70c82e555..005eda2aa 100644
--- a/volk/gen/compilers.xml
+++ b/volk/gen/compilers.xml
@@ -2,9 +2,21 @@
 <grammar>
 
 <compiler name="MSVC">
+  <!-- remap the following flags to SSE -->
   <remap name="mmmx">arch:SSE</remap>
   <remap name="msse">arch:SSE</remap>
+
+  <!-- remap the following flags to SSE2 -->
   <remap name="msse2">arch:SSE2</remap>
+
+  <!-- remap the following flags to AVX -->
+  <remap name="msse3">arch:AVX</remap>
+  <remap name="mssse3">arch:AVX</remap>
+  <remap name="msse4.1">arch:AVX</remap>
+  <remap name="msse4.2">arch:AVX</remap>
+  <remap name="mpopcnt">arch:AVX</remap>
+  <remap name="mavx">arch:AVX</remap>
+
   <prefix>/</prefix>
 </compiler>
 
@@ -15,4 +27,4 @@
 
 
 
-</grammar>
\ No newline at end of file
+</grammar>
diff --git a/volk/gen/machines.xml b/volk/gen/machines.xml
index b872b9fb1..9c19c91c6 100644
--- a/volk/gen/machines.xml
+++ b/volk/gen/machines.xml
@@ -15,9 +15,9 @@
 -->
 
 <!--
-Create an SSE2 only machine (without 64/32 inline assembly support).
+Create an SSE2 and AVX only machine (without 64/32 inline assembly support).
 This machine is intended to support the MSVC compiler on x86/amd64.
-The MSVC compiler has intrinsic support for SSE and SSE2,
+The MSVC compiler has intrinsic support for SSE, SSE2, AVX
 however it does not support the gcc style inline assembly.
 -->
 
@@ -57,6 +57,10 @@ however it does not support the gcc style inline assembly.
 <archs>generic 32|64 mmx sse sse2 sse3 ssse3 sse4_1 sse4_2 popcount avx</archs>
 </machine>
 
+<machine name="avx_only">
+<archs>generic mmx sse sse2 sse3 ssse3 sse4_1 sse4_2 popcount avx</archs>
+</machine>
+
 <machine name="altivec">
 <archs>generic altivec</archs>
 </machine>
diff --git a/volk/gen/make_cpuid_c.py b/volk/gen/make_cpuid_c.py
index 7281f45a3..2be1123a8 100644
--- a/volk/gen/make_cpuid_c.py
+++ b/volk/gen/make_cpuid_c.py
@@ -39,7 +39,7 @@ struct VOLK_CPU volk_cpu;
 //implement get cpuid for gcc compilers using a copy of cpuid.h
 #if defined(__GNUC__)
 #include <gcc_x86_cpuid.h>
-#define cpuid_x86(op, r) __get_cpuid(op, r+0, r+1, r+2, r+3)
+#define cpuid_x86(op, r) __get_cpuid(op, (unsigned int *)r+0, (unsigned int *)r+1, (unsigned int *)r+2, (unsigned int *)r+3)
 
 //implement get cpuid for MSVC compilers using __cpuid intrinsic
 #elif defined(_MSC_VER)