zdoom/qzdoom
0
0
Fork 0
mirror of https://github.com/ZDoom/qzdoom.git synced 2024-11-11 15:22:16 +00:00
Code Issues Projects Releases Packages Wiki Activity

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

Browse source 
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?
This commit is contained in:
Magnus Norddahl 2017-01-10 13:11:12 +01:00
parent 3487be2c40
commit 2f64bfa5af
4 changed files with 15 additions and 2 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
tools/drawergen/drawergen.cpp
View file

@ -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

3
tools/drawergen/ssa/ssa_float.cpp
View file

@ -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());

6
tools/drawergen/ssa/ssa_vec16ub.cpp
View file

@ -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());

6
tools/drawergen/ssa/ssa_vec8s.cpp
View file

@ -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());