Merge pull request Qiskit#4 from atilag/simd-fix-win

Fix compilation in other compilers / platforms

CMakeLists.txt

@@ -233,9 +233,11 @@ else() # Standalone build
 
 	# We build SIMD filed separately, because they will be reached only if the
 	# machine running the code has SIMD support
+	set(SIMD_SOURCE_FILE "${PROJECT_SOURCE_DIR}/src/simulators/statevector/qubitvector_avx2.cpp")
 	if(APPLE OR UNIX)
-		set(SIMD_SOURCE_FILE "${PROJECT_SOURCE_DIR}/src/simulators/statevector/qubitvector_avx2.cpp")
 		set_source_files_properties(${SIMD_SOURCE_FILE} PROPERTIES COMPILE_FLAGS "-mfma -mavx2")
+	elseif(MSVC)
+		set_source_files_properties(${SIMD_SOURCE_FILE} PROPERTIES COMPILE_FLAGS "/arch:AVX2")
 	endif()
 
 	set(CMAKE_INSTALL_RPATH_USE_LINK_PATH TRUE)

qiskit/providers/aer/backends/wrappers/CMakeLists.txt

@@ -7,20 +7,24 @@ find_package(Pybind11 REQUIRED)
 # shared library.
 string(REPLACE " -static " "" CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS}")
 
-# Controllers
+
 if(CUDA_FOUND)
     set_source_files_properties(bindings.cc PROPERTIES
         CUDA_SOURCE_PROPERTY_FORMAT OBJ)
 endif()
 
 # We build SIMD filed separately, because they will be reached only if the
 # machine running the code has SIMD support
+set(SIMD_SOURCE_FILE "../../../../../src/simulators/statevector/qubitvector_avx2.cpp")
+
 if(APPLE OR UNIX)
-    set(SIMD_SOURCE_FILE "../../../../../src/simulators/statevector/qubitvector_avx2.cpp")
     set_source_files_properties(${SIMD_SOURCE_FILE} PROPERTIES COMPILE_FLAGS "-mfma -mavx2")
+elseif(MSVC)
+    set_source_files_properties(${SIMD_SOURCE_FILE} PROPERTIES COMPILE_FLAGS "/arch:AVX2")
 endif()
 
-basic_pybind11_add_module(controller_wrappers bindings.cc ${SIMD_SOURCE_FILE})
+
+basic_pybind11_add_module(controller_wrappers ${SIMD_SOURCE_FILE} bindings.cc)
 target_include_directories(controller_wrappers PRIVATE ${AER_SIMULATOR_CPP_SRC_DIR}
                                                PRIVATE ${AER_SIMULATOR_CPP_EXTERNAL_LIBS})
 target_link_libraries(controller_wrappers ${AER_LIBRARIES})

src/controllers/qasm_controller.hpp

@@ -21,11 +21,14 @@
 #include "simulators/matrix_product_state/matrix_product_state.hpp"
 #include "simulators/stabilizer/stabilizer_state.hpp"
 #include "simulators/statevector/statevector_state.hpp"
+#include "simulators/statevector/qubitvector.hpp"
+#include "simulators/statevector/qubitvector_avx2.hpp"
 #include "simulators/superoperator/superoperator_state.hpp"
 #include "transpile/delay_measure.hpp"
 #include "transpile/fusion.hpp"
 #include "framework/avx2_detect.hpp"
 
+
 namespace AER {
 namespace Simulator {
 

src/controllers/statevector_controller.hpp

@@ -17,6 +17,7 @@
 
 #include "controller.hpp"
 #include "simulators/statevector/statevector_state.hpp"
+#include "simulators/statevector/qubitvector_avx2.hpp"
 
 namespace AER {
 namespace Simulator {

src/framework/avx2_detect.hpp

@@ -15,51 +15,74 @@
 #ifndef _aer_controller_avx2_detect_hpp_
 #define _aer_controller_avx2_detect_hpp_
 
-#ifdef _WIN64
+#ifdef _MSC_VER
 #include <intrin.h>
 #elif defined(__GNUC__)
 #include <cpuid.h>
 #endif
+#include <array>
+#include <vector>
+#include <bitset>
 
-namespace AER {
+namespace {
+inline void ccpuid(int cpu_info[4], int function_id){
+#ifdef _MSC_VER
+  __cpuid(cpu_info, function_id);
+#elif defined(__GNUC__)
+  __cpuid(function_id,
+    cpu_info[0],
+    cpu_info[1],
+    cpu_info[2],
+    cpu_info[3]);
+#else // We don't support this platform intrinsics
+  cpu_info[0] = cpu_info[1] = cpu_info[2] = cpu_info[3] = 0;
+#endif
+}
 
-#ifdef __GNUC__
-static void get_cpuid(void* p) {
-  int* a = (int*) p;
-  __cpuid(1, a[0], a[1], a[2], a[3]);
+inline void cpuidex(int cpu_info[4], int function_id, int subfunction_id){
+#ifdef _MSC_VER
+  __cpuidex(cpu_info, function_id, subfunction_id);
+#elif defined(__GNUC__)
+  __cpuid_count(function_id, subfunction_id, cpu_info[0], cpu_info[1], cpu_info[2], cpu_info[3]);
+#else // We don't support this platform intrinsics
+   cpu_info[0] = cpu_info[1] = cpu_info[2] = cpu_info[3] = 0;
+#endif
 }
-static void get_cpuid_count(void* p) {
-  int* a = (int*) p;
-  __cpuid_count(0x00000007, 0, a[0], a[1], a[2], a[3]);
 }
-#endif
 
-inline bool is_avx2_supported() {
-  static bool cached = false;
-  static bool is_supported = false;
-  if(cached)
-    return is_supported;
+namespace AER {
 
-#if defined(__GNUC__)
-  int info[4] = {0};
-  get_cpuid(info);
-  bool fma = (info[2] >> 12 & 1);
-  bool avx = (info[2] >> 28 & 1);
-  if (!fma || !avx){
-    cached = true;
+inline bool is_avx2_supported(){
+	static bool cached = false;
+	static bool is_supported = false;
+	if(cached)
+		return is_supported;
+
+	std::array<int, 4> cpui;
+ ccpuid(cpui.data(), 0);
+	auto num_ids = cpui[0];
+	if(num_ids < 7){
+		cached = true;
     is_supported = false;
     return false;
+	}
+
+  std::vector<std::array<int, 4>> data;
+  for (int i = 0; i <= num_ids; ++i){
+      cpuidex(cpui.data(), i, 0);
+      data.push_back(cpui);
   }
-  get_cpuid_count(info);
-  bool avx2 = (info[1] >> 5 & 1);
-  cached = true;
-  is_supported = avx2;
-  return avx2;
-#else
-  return false;
-#endif
-}
 
+  std::bitset<32> f_1_ECX = data[1][2];
+  std::bitset<32> f_7_EBX = data[7][1];
+
+  bool is_fma_supported = (f_1_ECX[12] & 1);
+  bool is_avx2_supported = (f_7_EBX[5] & 1);
+
+	cached = true;
+  is_supported = is_fma_supported && is_avx2_supported;
+	return is_supported;
+}
 // end namespace AER
 }
 #endif

src/simulators/statevector/qubitvector_avx2.cpp

@@ -19,6 +19,7 @@
 
 #include "qubitvector_avx2.hpp"
 #include "qubitvector.hpp"
+#include "qvintrin_avx.hpp"
 
 using namespace QV;
 

src/simulators/statevector/qubitvector_avx2.hpp

@@ -27,8 +27,7 @@
 #include <iostream>
 #include <sstream>
 #include <stdexcept>
-
-#include "qvintrin_avx.hpp"
+#include <memory>
 
 namespace QV {
 

src/simulators/statevector/qvintrin_avx.hpp

@@ -217,8 +217,6 @@ inline void reorder(QV::areg_t<N>& qregs, QV::cvector_t<FloatType>& mat) {
   for(size_t i = 0; i < dim; ++i) {
     size_t index = 0U;
     for(size_t j = 0; j < N; ++j) {
-
-
       if(i & (1U << j))
         index |= masks[j];
     }
@@ -719,7 +717,7 @@ void _apply_lambda(uint64_t data_size, const uint64_t skip, Lambda&& func, const
   std::sort(qubits_sorted.begin(), qubits_sorted.end());
 
 #pragma omp parallel for if (omp_threads > 1) num_threads(omp_threads)
-  for (uint64_t k = 0; k < END; k += skip) {
+  for (int64_t k = 0; k < END; k += skip) {
     const auto inds = _indexes(qubits, qubits_sorted, k);
     std::forward < Lambda > (func)(inds, params);
   }

src/simulators/statevector/statevector_state.hpp

@@ -23,7 +23,6 @@
 #include "framework/json.hpp"
 #include "simulators/state.hpp"
 #include "qubitvector.hpp"
-#include "qubitvector_avx2.hpp"
 #ifdef AER_THRUST_SUPPORTED
 #include "qubitvector_thrust.hpp"
 #endif