SprivShader: Replace hand-rolled bitreverse with LLVM intrinsic

src/Pipeline/SpirvShader.cpp

@@ -2772,17 +2772,7 @@ namespace sw
 			}
 			case spv::OpBitReverse:
 			{
-				// TODO: Add an intrinsic to reactor. Even if there isn't a
-				// single vector instruction, there may be target-dependent
-				// ways to make this faster.
-				// https://graphics.stanford.edu/~seander/bithacks.html#ReverseParallel
-				SIMD::UInt v = src.UInt(i);
-				v = ((v >> 1) & SIMD::UInt(0x55555555)) | ((v & SIMD::UInt(0x55555555)) << 1);
-				v = ((v >> 2) & SIMD::UInt(0x33333333)) | ((v & SIMD::UInt(0x33333333)) << 2);
-				v = ((v >> 4) & SIMD::UInt(0x0F0F0F0F)) | ((v & SIMD::UInt(0x0F0F0F0F)) << 4);
-				v = ((v >> 8) & SIMD::UInt(0x00FF00FF)) | ((v & SIMD::UInt(0x00FF00FF)) << 8);
-				v = (v >> 16) | (v << 16);
-				dst.move(i, v);
+				dst.move(i, BitReverse(src.UInt(i)));
 				break;
 			}
 			case spv::OpBitCount:

src/Reactor/LLVMReactor.cpp

@@ -3214,6 +3214,12 @@ namespace rr
 		return RValue<Float4>(V(::builder->CreateCall(func, V(v.value))));
 	}
 
+	RValue<UInt4> BitReverse(RValue<UInt4> v)
+	{
+		auto func = llvm::Intrinsic::getDeclaration(::module, llvm::Intrinsic::bitreverse, { T(UInt4::getType()) } );
+		return RValue<UInt4>(V(::builder->CreateCall(func, { V(v.value) })));
+	}
+
 	RValue<UInt4> Ctlz(RValue<UInt4> v, bool isZeroUndef)
 	{
 #if REACTOR_LLVM_VERSION < 7

src/Reactor/Reactor.hpp

@@ -2232,6 +2232,7 @@ namespace rr
 
 	// Bit Manipulation functions.
 	// TODO: Currentlhy unimplemented for Subzero.
+	RValue<UInt4> BitReverse(RValue<UInt4> x);
 
 	// Count leading zeros.
 	// Returns 32 when: isZeroUndef && x == 0.

src/Reactor/SubzeroReactor.cpp

@@ -3359,6 +3359,18 @@ namespace rr
 		}
 	}
 
+	RValue<UInt4> BitReverse(RValue<UInt4> x)
+	{
+		// https://graphics.stanford.edu/~seander/bithacks.html#ReverseParallel
+		UInt4 v = x;
+		v = ((v >> 1) & UInt4(0x55555555)) | ((v & UInt4(0x55555555)) << 1);
+		v = ((v >> 2) & UInt4(0x33333333)) | ((v & UInt4(0x33333333)) << 2);
+		v = ((v >> 4) & UInt4(0x0F0F0F0F)) | ((v & UInt4(0x0F0F0F0F)) << 4);
+		v = ((v >> 8) & UInt4(0x00FF00FF)) | ((v & UInt4(0x00FF00FF)) << 8);
+		v = (v >> 16) | (v << 16);
+		return v;
+	}
+
 	Type *Float4::getType()
 	{
 		return T(Ice::IceType_v4f32);