Fix all warnings when compiling with HLSL202x due to microsoft/Direct…

…XShaderCompiler#6867 (Conforming Literals proposal 0017 change in HLSL)

include/nbl/builtin/hlsl/sort/counting.hlsl

@@ -22,7 +22,7 @@ template<
     typename ValueAccessor,
     typename HistogramAccessor,
     typename SharedAccessor,
-    typename key_t = decltype(impl::declval < KeyAccessor > ().get(0)),
+    typename key_t = decltype(impl::declval<KeyAccessor>().get(0)),
     bool robust=false
 >
 struct counting
@@ -88,8 +88,8 @@ struct counting
         uint32_t sum = inclusive_scan(histogram_value, sdata);
         histogram.atomicAdd(tid, sum);
 
-        const bool is_last_wg_invocation = tid == (GroupSize - 1);
-        const static uint16_t RoundedKeyBucketCount = (KeyBucketCount - 1) / GroupSize + 1;
+        const bool is_last_wg_invocation = tid == (GroupSize-_static_cast<uint16_t>(1));
+        const static uint16_t RoundedKeyBucketCount = (KeyBucketCount-_static_cast<uint16_t>(1))/GroupSize+_static_cast<uint16_t>(1);
 
         for (int i = 1; i < RoundedKeyBucketCount; i++)
         {

include/nbl/builtin/hlsl/workgroup/arithmetic.hlsl

@@ -87,9 +87,9 @@ uint16_t ballotCountedBitDWORD(NBL_REF_ARG(BallotAccessor) ballotAccessor)
         uint32_t bitfield;
         ballotAccessor.get(index, bitfield);
         // strip unwanted bits from bitfield of the last item
-        const uint16_t Remainder = ItemCount&31;
+        const uint16_t Remainder = ItemCount&_static_cast<uint16_t>(31);
         if (Remainder!=0 && index==DWORDCount-1)
-            bitfield &= (0x1u<<Remainder)-1;
+            bitfield &= (0x1<<Remainder)-1;
         return uint16_t(countbits(bitfield));
     }
     return 0;

include/nbl/builtin/hlsl/workgroup/ballot.hlsl

@@ -26,7 +26,7 @@ uint16_t getDWORD(uint16_t invocation)
 // essentially means 'how many DWORDs are needed to store ballots in bitfields, for each invocation of `itemCount`
 uint16_t BallotDWORDCount(const uint16_t itemCount)
 {
-    return getDWORD(itemCount+31); // round up, in case all items don't fit in even number of DWORDs
+    return getDWORD(itemCount+_static_cast<uint16_t>(31)); // round up, in case all items don't fit in even number of DWORDs
 }
 
 // this silly thing exists only because we can't make the above `constexpr`

include/nbl/builtin/hlsl/workgroup/scratch_size.hlsl

@@ -34,7 +34,7 @@ struct trunc_geom_series<N,K,W,true> : integral_constant<uint32_t,0> {};
 template<uint16_t ContiguousItemCount>
 struct scratch_size_ballot
 {
-	NBL_CONSTEXPR_STATIC_INLINE uint16_t value = (ContiguousItemCount+31)>>5;
+	NBL_CONSTEXPR_STATIC_INLINE uint16_t value = (ContiguousItemCount+uint16_t(31))>>5;
 };
 
 // you're only writing one element

include/nbl/builtin/hlsl/workgroup/shared_scan.hlsl

@@ -26,8 +26,8 @@ struct reduce
     template<class Accessor>
     void __call(NBL_CONST_REF_ARG(type_t) value, NBL_REF_ARG(Accessor) scratchAccessor)
     {
-        const uint16_t lastInvocation = ItemCount-1;
-        const uint16_t subgroupMask = uint16_t(glsl::gl_SubgroupSize()-1u);
+        const uint16_t lastInvocation = ItemCount-_static_cast<uint16_t>(1);
+        const uint16_t subgroupMask = _static_cast<uint16_t>(glsl::gl_SubgroupSize()-1u);
 
         subgroup::inclusive_scan<BinOp,device_capabilities> subgroupOp;
 
@@ -47,16 +47,16 @@ struct reduce
         // Consequently, those first gl_SubgroupSz^2 invocations will store their results on gl_SubgroupSz scratch slots 
         // and the next level will follow the same + the previous as an `offset`.
 
-        const uint16_t loadStoreIndexDiff = scanLoadIndex-uint16_t(glsl::gl_SubgroupID());
+        const uint16_t loadStoreIndexDiff = scanLoadIndex-_static_cast<uint16_t>(glsl::gl_SubgroupID());
 
         // to cancel out the index shift on the first iteration
         if (lastInvocationInLevel>subgroupMask)
-             scanLoadIndex -= lastInvocationInLevel-1;
+             scanLoadIndex -= lastInvocationInLevel-_static_cast<uint16_t>(1);
         // TODO: later [unroll(scan_levels<ItemCount,subgroup::MinSubgroupSize>::value-1)]
         [unroll(1)]
         while (lastInvocationInLevel>subgroupMask)
         {
-            scanLoadIndex += lastInvocationInLevel+1;
+            scanLoadIndex += lastInvocationInLevel+_static_cast<uint16_t>(1);
             // only invocations that have the final value of the subgroupOp (inclusive scan) store their results
             if (participate && (SubgroupContiguousIndex()==lastInvocationInLevel || isLastSubgroupInvocation))
                 scratchAccessor.set(scanLoadIndex-loadStoreIndexDiff,scan); // For subgroupSz = 32, first 512 invocations store index is [0,15], 512-1023 [16,31] etc.
@@ -93,10 +93,10 @@ struct scan// : reduce<BinOp,ItemCount> https://github.com/microsoft/DirectXShad
 
         const uint16_t subgroupID = uint16_t(glsl::gl_SubgroupID());
         // abuse integer wraparound to map 0 to 0xffffu
-        const uint16_t prevSubgroupID = subgroupID-1;
+        const uint16_t prevSubgroupID = subgroupID-_static_cast<uint16_t>(1);
 
         // important check to prevent weird `firstbithigh` overlflows
-        const uint16_t lastInvocation = ItemCount-1;
+        const uint16_t lastInvocation = ItemCount-_static_cast<uint16_t>(1);
         if(lastInvocation>=uint16_t(glsl::gl_SubgroupSize()))
         {
             const uint16_t subgroupSizeLog2 = uint16_t(glsl::gl_SubgroupSizeLog2());
@@ -121,7 +121,7 @@ struct scan// : reduce<BinOp,ItemCount> https://github.com/microsoft/DirectXShad
                     if (logShift!=initialLogShift) // but the top level doesn't have any level above itself
                     {
                         // this is fine if on the way up you also += under `if (participate)`
-                        scanStoreIndex -= __base.lastInvocationInLevel+1;
+                        scanStoreIndex -= __base.lastInvocationInLevel+_static_cast<uint16_t>(1);
                         type_t higherLevelEPS;
                         scratchAccessor.get(scanStoreIndex, higherLevelEPS);
                         __base.lastLevelScan = binop(__base.lastLevelScan,higherLevelEPS);