Drop using intrinsics on ARM until after they actually got gdb to run on it.

It is all a waste of time anyhow, because when my computer is literally a
factor 100 times faster at building, what frame rate can one really expect?
What should it run at? 320x200?

tools/drawergen/drawergen.cpp

@@ -89,7 +89,7 @@ int main(int argc, char **argv)
 		std::cout << "Target triple is " << triple << std::endl;
 
 #ifdef __arm__
-		std::string cpuName = "armv8";
+		std::string cpuName = llvm::sys::getHostCPUName(); // "armv8";
 #else
 		std::string cpuName = "pentium4";
 #endif

tools/drawergen/ssa/ssa_float.cpp

@@ -57,7 +57,8 @@ llvm::Type *SSAFloat::llvm_type()
 SSAFloat SSAFloat::rsqrt(SSAFloat f)
 {
 #ifdef ARM_TARGET
-	return SSAFloat::from_llvm(SSAScope::builder().CreateCall(SSAScope::intrinsic(llvm::Intrinsic::aarch64_neon_frsqrts), f.v, SSAScope::hint()));
+	//return SSAFloat::from_llvm(SSAScope::builder().CreateCall(SSAScope::intrinsic(llvm::Intrinsic::aarch64_neon_frsqrts), f.v, SSAScope::hint()));
+	return SSAFloat(1.0f) / (f * SSAFloat(0.01f));
 #else
 	llvm::Value *f_ss = SSAScope::builder().CreateInsertElement(llvm::UndefValue::get(SSAVec4f::llvm_type()), f.v, llvm::ConstantInt::get(SSAScope::context(), llvm::APInt(32, (uint64_t)0)));
 	f_ss = SSAScope::builder().CreateCall(SSAScope::intrinsic(llvm::Intrinsic::x86_sse_rsqrt_ss), f_ss, SSAScope::hint());

tools/drawergen/ssa/ssa_vec16ub.cpp

@@ -73,9 +73,15 @@ SSAVec16ub::SSAVec16ub(SSAVec8s s0, SSAVec8s s1)
 : v(0)
 {
 #ifdef ARM_TARGET
+	/*
 	llvm::Value *int8x8_i0 = SSAScope::builder().CreateCall(SSAScope::intrinsic(llvm::Intrinsic::arm_neon_vqmovnsu), s0.v, SSAScope::hint());
 	llvm::Value *int8x8_i1 = SSAScope::builder().CreateCall(SSAScope::intrinsic(llvm::Intrinsic::arm_neon_vqmovnsu), s1.v, SSAScope::hint());
 	v = shuffle(from_llvm(int8x8_i0), from_llvm(int8x8_i1), 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15).v;
+	*/
+	// To do: add some clamping here
+	llvm::Value *int8x8_i0 = SSAScope::builder().CreateTrunc(s0.v, llvm::VectorType::get(llvm::Type::getInt8Ty(SSAScope::context()), 8));
+	llvm::Value *int8x8_i1 = SSAScope::builder().CreateTrunc(s1.v, llvm::VectorType::get(llvm::Type::getInt8Ty(SSAScope::context()), 8));
+	v = shuffle(from_llvm(int8x8_i0), from_llvm(int8x8_i1), 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15).v;
 #else
 	llvm::Value *values[2] = { s0.v, s1.v };
 	v = SSAScope::builder().CreateCall(SSAScope::intrinsic(llvm::Intrinsic::x86_sse2_packuswb_128), values, SSAScope::hint());

tools/drawergen/ssa/ssa_vec8s.cpp

@@ -63,9 +63,15 @@ SSAVec8s::SSAVec8s(SSAVec4i i0, SSAVec4i i1)
 : v(0)
 {
 #ifdef ARM_TARGET
+	/*
 	llvm::Value *int16x4_i0 = SSAScope::builder().CreateCall(SSAScope::intrinsic(llvm::Intrinsic::arm_neon_vqmovns), i0.v, SSAScope::hint());
 	llvm::Value *int16x4_i1 = SSAScope::builder().CreateCall(SSAScope::intrinsic(llvm::Intrinsic::arm_neon_vqmovns), i1.v, SSAScope::hint());
 	v = shuffle(from_llvm(int16x4_i0), from_llvm(int16x4_i1), 0, 1, 2, 3, 4, 5, 6, 7).v;
+	*/
+	// To do: add some clamping here
+	llvm::Value *int16x4_i0 = SSAScope::builder().CreateTrunc(i0.v, llvm::VectorType::get(llvm::Type::getInt16Ty(SSAScope::context()), 4));
+	llvm::Value *int16x4_i1 = SSAScope::builder().CreateTrunc(i1.v, llvm::VectorType::get(llvm::Type::getInt16Ty(SSAScope::context()), 4));
+	v = shuffle(from_llvm(int16x4_i0), from_llvm(int16x4_i1), 0, 1, 2, 3, 4, 5, 6, 7).v;
 #else
 	llvm::Value *values[2] = { i0.v, i1.v };
 	v = SSAScope::builder().CreateCall(SSAScope::intrinsic(llvm::Intrinsic::x86_sse2_packssdw_128), values, SSAScope::hint());