Merge pull request #99 from Devsh-Graphics-Programming/exclusive_min_fix

workgroupExclusiveMin fix and a fix of the floating point And workgroup op identity

Author: devshgraphicsprogramming

examples_tests/48.ArithmeticUnitTest/main.cpp

@@ -9,13 +9,28 @@ using namespace core;
 using namespace video;
 using namespace asset;
 
+template <class To, class From>
+typename std::enable_if_t<
+	std::is_trivially_copyable_v<From> &&
+	std::is_trivially_copyable_v<To>,
+	To>
+	// constexpr support needs compiler magic
+	bit_cast(const From& src) noexcept
+{
+	static_assert(std::is_trivially_constructible_v<To>,
+		"This implementation additionally requires destination type to be trivially constructible");
+
+	To dst;
+	std::memcpy(&dst, &src, sizeof(To));
+	return dst;
+}
 
 
 template<typename T>
 struct and
 {
 	using type_t = T;
-	_NBL_STATIC_INLINE_CONSTEXPR T IdentityElement = ~0ull; // this should be a reinterpret cast
+	static inline const T IdentityElement = bit_cast<T>(~0ull); // until C++20 std::bit_cast this cannot be a constexpr
 
 	inline T operator()(T left, T right) { return left & right; }
 	_NBL_STATIC_INLINE_CONSTEXPR bool runOPonFirst = false;
@@ -25,7 +40,7 @@ template<typename T>
 struct xor
 {
 	using type_t = T;
-	_NBL_STATIC_INLINE_CONSTEXPR T IdentityElement = 0ull; // this should be a reinterpret cast
+	static inline const T IdentityElement = bit_cast<T>(0ull); // until C++20 std::bit_cast this cannot be a constexpr
 
 	inline T operator()(T left, T right) { return left ^ right; }
 	_NBL_STATIC_INLINE_CONSTEXPR bool runOPonFirst = false;
@@ -35,7 +50,7 @@ template<typename T>
 struct or
 {
 	using type_t = T;
-	_NBL_STATIC_INLINE_CONSTEXPR T IdentityElement = 0ull; // this should be a reinterpret cast
+	static inline const T IdentityElement = bit_cast<T>(0ull); // until C++20 std::bit_cast this cannot be a constexpr
 
 	inline T operator()(T left, T right) { return left | right; }
 	_NBL_STATIC_INLINE_CONSTEXPR bool runOPonFirst = false;

include/nbl/builtin/glsl/subgroup/basic_portability.glsl

@@ -155,7 +155,7 @@ bool nbl_glsl_subgroupElect()
 	#ifdef NBL_GL_KHR_shader_subgroup_basic_subgroup_elect
 	return subgroupElect();
 	#else
-	// TODO: do a bunch of `atomicXor` on a shared memory address
+	// TODO: do a bunch of `atomicXor` on a shared memory address OR do a ballotARB?
 	atomicXor(ADDRESS,nbl_glsl_SubgroupEqMask);
 	memoryBarrierShared();
 	return (ADDRESS&nbl_glsl_SubgroupLeMask)==nbl_glsl_SubgroupEqMask;

include/nbl/builtin/glsl/subgroup/shuffle_portability.glsl

@@ -5,6 +5,7 @@
 #include <nbl/builtin/glsl/subgroup/shared_shuffle_portability.glsl>
 
 
+// TODO: A SPIRV-Cross contribution so we can set NBL_GL_KHR_shader_subgroup_shuffle when AMD_gcn_shader or NVidia shuffle extensions are available 
 #ifdef NBL_GL_KHR_shader_subgroup_shuffle
 
 

include/nbl/builtin/glsl/workgroup/arithmetic.glsl

@@ -52,15 +52,15 @@
 
 uint nbl_glsl_workgroupAnd_noBarriers(in uint val)
 {
-	NBL_GLSL_WORKGROUP_REDUCE(nbl_glsl_identityFunction,nbl_glsl_subgroupInclusiveAnd_impl,val,~0u,nbl_glsl_identityFunction);
+	NBL_GLSL_WORKGROUP_REDUCE(nbl_glsl_identityFunction,nbl_glsl_subgroupInclusiveAnd_impl,val,0xffFFffFFu,nbl_glsl_identityFunction);
 }
 int nbl_glsl_workgroupAnd_noBarriers(in int val)
 {
 	return int(nbl_glsl_workgroupAnd_noBarriers(uint(val)));
 }
 float nbl_glsl_workgroupAnd_noBarriers(in float val)
 {
-	NBL_GLSL_WORKGROUP_REDUCE(uintBitsToFloat,nbl_glsl_subgroupInclusiveAnd_impl,val,0.0,floatBitsToUint);
+	return uintBitsToFloat(nbl_glsl_workgroupAnd_noBarriers(floatBitsToUint(val)));
 }
 
 DECLARE_OVERLOAD_WITH_BARRIERS(uint,workgroupAnd)
@@ -185,7 +185,7 @@ DECLARE_OVERLOAD_WITH_BARRIERS(float,workgroupMax)
 
 uint nbl_glsl_workgroupInclusiveAnd_noBarriers(in uint val)
 {
-	NBL_GLSL_WORKGROUP_SCAN(false,nbl_glsl_identityFunction,nbl_glsl_and,nbl_glsl_subgroupInclusiveAnd_impl,val,~0u,nbl_glsl_identityFunction);
+	NBL_GLSL_WORKGROUP_SCAN(false,nbl_glsl_identityFunction,nbl_glsl_and,nbl_glsl_subgroupInclusiveAnd_impl,val,0xffFFffFFu,nbl_glsl_identityFunction);
 }
 int nbl_glsl_workgroupInclusiveAnd_noBarriers(in int val)
 {
@@ -203,7 +203,7 @@ DECLARE_OVERLOAD_WITH_BARRIERS(float,workgroupInclusiveAnd)
 
 uint nbl_glsl_workgroupExclusiveAnd_noBarriers(in uint val)
 {
-	NBL_GLSL_WORKGROUP_SCAN(true,nbl_glsl_identityFunction,nbl_glsl_and,nbl_glsl_subgroupInclusiveAnd_impl,val,~0u,nbl_glsl_identityFunction);
+	NBL_GLSL_WORKGROUP_SCAN(true,nbl_glsl_identityFunction,nbl_glsl_and,nbl_glsl_subgroupInclusiveAnd_impl,val,0xffFFffFFu,nbl_glsl_identityFunction);
 }
 int nbl_glsl_workgroupExclusiveAnd_noBarriers(in int val)
 {
@@ -392,7 +392,7 @@ uint nbl_glsl_workgroupExclusiveMin_noBarriers(in uint val)
 }
 int nbl_glsl_workgroupExclusiveMin_noBarriers(in int val)
 {
-	NBL_GLSL_WORKGROUP_SCAN(true,int,min,nbl_glsl_subgroupExclusiveMin_impl,val,INT_MAX,uint);
+	NBL_GLSL_WORKGROUP_SCAN(true,int,min,nbl_glsl_subgroupInclusiveMin_impl,val,INT_MAX,uint);
 }
 float nbl_glsl_workgroupExclusiveMin_noBarriers(in float val)
 {