guard bundled tbb against gcc cpuid.h / mingw intrin.h conflict (#248)

GCC's <cpuid.h> defines a function-like '__cpuid' macro. If a downstream
package includes <cpuid.h> before any TBB header on Windows (mingw), the
macro mangles the __cpuid() declarations in mingw's <intrin.h>, which
oneTBB's detail/_machine.h includes, and compilation fails, e.g.:

  intrin-impl.h:2013:42: error: macro "__cpuid" requires 5 arguments, but only 2 given

Suppress the macro via push_macro/pop_macro while including <intrin.h>,
so the two headers can coexist in any include order. Record the change
in patches/mingw_cpuid.diff so it can be reapplied on oneTBB updates.

This was observed in the wild with the 'robscale' package on r-universe:
https://github.com/r-universe/cran/actions/runs/30028399599/job/89335144753

Author: kevinushey

NEWS.md

@@ -1,5 +1,11 @@
 # RcppParallel (development version)
 
+* The bundled oneTBB headers now guard against GCC's `<cpuid.h>` being
+  included before `<intrin.h>` on Windows (mingw). Previously, translation
+  units including `<cpuid.h>` before any TBB header would fail to compile,
+  as the `__cpuid` macro from `<cpuid.h>` conflicts with the `__cpuid()`
+  function declared by mingw's `<intrin.h>`.
+
 * When building the bundled copy of oneTBB, RcppParallel no longer searches
   for hwloc, and so no longer tries to build the optional 'tbbbind' library.
   This fixes build failures on machines where a static hwloc library is

patches/mingw_cpuid.diff

@@ -0,0 +1,24 @@
+diff --git a/src/tbb/include/oneapi/tbb/detail/_machine.h b/src/tbb/include/oneapi/tbb/detail/_machine.h
+index ca481380..8174f3d3 100644
+--- a/src/tbb/include/oneapi/tbb/detail/_machine.h
++++ b/src/tbb/include/oneapi/tbb/detail/_machine.h
+@@ -26,7 +26,19 @@
+ #include <cstddef>
+ 
+ #ifdef _WIN32
++
++// GCC's <cpuid.h> defines a function-like '__cpuid' macro that mangles the
++// __cpuid() declarations in mingw's <intrin.h>. Hide the macro while
++// including <intrin.h> so the two headers can coexist in any order.
++#if defined(__MINGW32__) && defined(__cpuid)
++#pragma push_macro("__cpuid")
++#undef __cpuid
++#include <intrin.h>
++#pragma pop_macro("__cpuid")
++#else
+ #include <intrin.h>
++#endif
++
+ #ifdef __TBBMALLOC_BUILD
+ #define WIN32_LEAN_AND_MEAN
+ #ifndef NOMINMAX

src/tbb/include/oneapi/tbb/detail/_machine.h

@@ -26,7 +26,19 @@
 #include <cstddef>
 
 #ifdef _WIN32
+
+// GCC's <cpuid.h> defines a function-like '__cpuid' macro that mangles the
+// __cpuid() declarations in mingw's <intrin.h>. Hide the macro while
+// including <intrin.h> so the two headers can coexist in any order.
+#if defined(__MINGW32__) && defined(__cpuid)
+#pragma push_macro("__cpuid")
+#undef __cpuid
+#include <intrin.h>
+#pragma pop_macro("__cpuid")
+#else
 #include <intrin.h>
+#endif
+
 #ifdef __TBBMALLOC_BUILD
 #define WIN32_LEAN_AND_MEAN
 #ifndef NOMINMAX