platform.h: fully define bitcast template (#1635)

Ported from OIIO. This lets us not have to depend on the specific
definition in OIIO's fmath.h, and in particular not worry about
whether it's a new enough version of OIIO that the old bit_cast is now
marked as deprecated.

Signed-off-by: Larry Gritz <lg@larrygritz.com>

src/include/OSL/oslnoise.h

@@ -144,12 +144,7 @@ inline OSL_HOSTDEVICE float bits_to_01 (unsigned int bits) {
 OSL_FORCEINLINE OSL_HOSTDEVICE int
 bitcast_to_uint (float x)
 {
-#if OPENIMAGEIO_VERSION >= 20500
-    return OIIO::bitcast<unsigned int, float>(x);
-#else
-    // obsolete call
-    return OIIO::bit_cast<float, unsigned int>(x);
-#endif
+    return OSL::bitcast<unsigned int, float>(x);
 }
 
 

src/include/OSL/platform.h

@@ -23,6 +23,9 @@
 
 #include <OSL/oslversion.h>
 
+#if defined(__x86_64__) && !defined(__CUDA_ARCH__)
+#   include <x86intrin.h>
+#endif
 
 /////////////////////////////////////////////////////////////////////////
 // Detect which compiler and version we're using
@@ -516,6 +519,42 @@ OSL_FORCEINLINE OSL_HOSTDEVICE To bitcast(const From& src) noexcept {
     return dst;
 }
 
+#if defined(__x86_64__) && !defined(__CUDA_ARCH__) && \
+    (defined(__INTEL_COMPILER) || defined(__INTEL_LLVM_COMPILER) \
+     || OSL_CLANG_VERSION >= 100000 || OSL_APPLE_CLANG_VERSION >= 130000)
+// On x86/x86_64 for certain compilers we can use Intel CPU intrinsics for
+// some common bitcast cases that might be even more understandable to the
+// compiler and generate better code without its getting confused about the
+// memcpy in the general case. We're a bit conservative with the compiler
+// version checks here, it may be that some earlier versions support these
+// intrinsics.
+
+template<> OSL_FORCEINLINE uint32_t bitcast<uint32_t, float>(const float& val) noexcept {
+    return static_cast<uint32_t>(_castf32_u32(val));
+}
+template<> OSL_FORCEINLINE int32_t bitcast<int32_t, float>(const float& val) noexcept {
+    return static_cast<int32_t>(_castf32_u32(val));
+}
+template<> OSL_FORCEINLINE float bitcast<float, uint32_t>(const uint32_t& val) noexcept {
+    return _castu32_f32(val);
+}
+template<> OSL_FORCEINLINE float bitcast<float, int32_t>(const int32_t& val) noexcept {
+    return _castu32_f32(val);
+}
+template<> OSL_FORCEINLINE uint64_t bitcast<uint64_t, double>(const double& val) noexcept {
+    return static_cast<uint64_t>(_castf64_u64(val));
+}
+template<> OSL_FORCEINLINE int64_t bitcast<int64_t, double>(const double& val) noexcept {
+    return static_cast<int64_t>(_castf64_u64(val));
+}
+template<> OSL_FORCEINLINE double bitcast<double, uint64_t>(const uint64_t& val) noexcept {
+    return _castu64_f64(val);
+}
+template<> OSL_FORCEINLINE double bitcast<double, int64_t>(const int64_t& val) noexcept {
+    return _castu64_f64(val);
+}
+#endif
+
 
 
 #if OSL_CPLUSPLUS_VERSION >= 20

src/include/OSL/wide.h

@@ -980,12 +980,7 @@ template<int WidthT> struct alignas(VecReg<WidthT>) Block<ustringhash, WidthT> {
 #ifdef OIIO_USTRING_HAS_CTR_FROM_USTRINGHASH
         return ustringhash::from_hash(str[lane]);
 #else
-        // Dumb workaround if we are on old OIIO
-#    if OPENIMAGEIO_VERSION >= 20500
-        return OIIO::bitcast<ustringhash>(str[lane]);
-#    else
-        return OIIO::bit_cast<size_t, ustringhash>(str[lane]);
-#    endif
+        return OSL::bitcast<ustringhash>(str[lane]);
 #endif
     }
 

src/liboslexec/wide/wide_ophash.cpp

@@ -28,12 +28,7 @@ namespace {
 OSL_FORCEINLINE OSL_HOSTDEVICE int
 bitcast_to_uint(float x)
 {
-#if OPENIMAGEIO_VERSION >= 20500
-    return OIIO::bitcast<unsigned int, float>(x);
-#else
-    // obsolete call
-    return OIIO::bit_cast<float, unsigned int>(x);
-#endif
+    return OSL::bitcast<unsigned int>(x);
 }