mirror of
https://github.com/jessicanataliagta/PSPRecomp
synced 2026-09-26 16:49:34 -04:00
Jessica_Natalia
184 lines
12 KiB
C++
184 lines
12 KiB
C++
#include "psprecomp/allegrex_context.hpp"
|
|
#include <bit>
|
|
#include <cstdint>
|
|
#include <cstdlib>
|
|
#include <iostream>
|
|
#include <random>
|
|
using psprecomp::AllegrexContext;
|
|
static std::mt19937 rng(0x56435332u);
|
|
static AllegrexContext make_ctx(){
|
|
AllegrexContext c{};
|
|
std::uniform_real_distribution<float> f(-100.0f,100.0f);
|
|
for(auto &v:c.vfpu) v=f(rng);
|
|
c.vfpu_ctrl[0]=rng() & 0x000FFFFFu;
|
|
c.vfpu_ctrl[1]=rng() & 0x000FFFFFu;
|
|
c.vfpu_ctrl[2]=rng() & 0x00000FFFu;
|
|
c.vfpu_ctrl[3]=rng();
|
|
for(unsigned i=4;i<c.vfpu_ctrl.size();++i)c.vfpu_ctrl[i]=rng();
|
|
return c;
|
|
}
|
|
static bool same(const AllegrexContext&a,const AllegrexContext&b){
|
|
for(size_t i=0;i<a.vfpu.size();++i)
|
|
if(std::bit_cast<std::uint32_t>(a.vfpu[i])!=std::bit_cast<std::uint32_t>(b.vfpu[i])) return false;
|
|
return a.vfpu_ctrl==b.vfpu_ctrl;
|
|
}
|
|
[[noreturn]] static void fail(const char*n){std::cerr<<"DIFF "<<n<<"\n";std::exit(2);}
|
|
template<unsigned M,unsigned S> void test_matrix_read(){for(int n=0;n<8;++n){auto a=make_ctx();float x[16]{},y[16]{};a.read_vfpu_matrix(x,M,S);a.read_vfpu_matrix_ct<M,S>(y);for(int i=0;i<16;++i)if(std::bit_cast<std::uint32_t>(x[i])!=std::bit_cast<std::uint32_t>(y[i]))fail("matrix_read");}}
|
|
template<unsigned M,unsigned S> void test_matrix_write(){for(int n=0;n<8;++n){auto a=make_ctx(),b=a;float x[16]{};std::uniform_real_distribution<float>f(-50,50);for(float&v:x)v=f(rng);a.write_vfpu_matrix(x,M,S);b.write_vfpu_matrix_ct<M,S>(x);if(!same(a,b))fail("matrix_write");}}
|
|
template<unsigned D,unsigned S,unsigned T,unsigned L> void test_cross(){for(int n=0;n<32;++n){auto a=make_ctx(),b=a;if((n&1)==0){a.vfpu_ctrl[0]=b.vfpu_ctrl[0]=0xE4u;a.vfpu_ctrl[1]=b.vfpu_ctrl[1]=0xE4u;a.vfpu_ctrl[2]=b.vfpu_ctrl[2]=0u;}a.execute_vfpu_cross_quat(D,S,T,L);b.execute_vfpu_cross_quat_ct<D,S,T,L>();if(!same(a,b))fail("cross");}}
|
|
template<unsigned S,unsigned T,unsigned L,unsigned C> void test_vcmp(){for(int n=0;n<64;++n){auto a=make_ctx(),b=a;if((n&1)==0){a.vfpu_ctrl[0]=b.vfpu_ctrl[0]=0xE4u;a.vfpu_ctrl[1]=b.vfpu_ctrl[1]=0xE4u;a.vfpu_ctrl[2]=b.vfpu_ctrl[2]=0u;}a.execute_vfpu_vcmp(S,T,L,C);b.execute_vfpu_vcmp_ct<S,T,L,C>();if(!same(a,b))fail("vcmp");}}
|
|
template<unsigned D,unsigned S,unsigned L,unsigned C,bool F> void test_vcmov(){for(int n=0;n<64;++n){auto a=make_ctx(),b=a;if((n&1)==0){a.vfpu_ctrl[0]=b.vfpu_ctrl[0]=0xE4u;a.vfpu_ctrl[1]=b.vfpu_ctrl[1]=0xE4u;a.vfpu_ctrl[2]=b.vfpu_ctrl[2]=0u;}a.execute_vfpu_vcmov(D,S,L,C,F);b.execute_vfpu_vcmov_ct<D,S,L,C,F>();if(!same(a,b))fail("vcmov");}}
|
|
template<unsigned D,unsigned S,unsigned L,unsigned I> void test_vi2f(){for(int n=0;n<64;++n){auto a=make_ctx(),b=a;if((n&1)==0){a.vfpu_ctrl[0]=b.vfpu_ctrl[0]=0xE4u;a.vfpu_ctrl[1]=b.vfpu_ctrl[1]=0xE4u;a.vfpu_ctrl[2]=b.vfpu_ctrl[2]=0u;}float src[4]{},dst[4]{};a.read_vfpu_vector(src,S,L);a.apply_vfpu_source_prefix(src,L,0u);const float scale=std::ldexp(1.0f,-static_cast<int>(I));for(unsigned lane=0;lane<L;++lane){const auto integer=static_cast<std::int32_t>(std::bit_cast<std::uint32_t>(src[lane]));dst[lane]=static_cast<float>(integer)*scale;}a.write_vfpu_vector_with_destination_prefix(dst,D,L);b.execute_vfpu_vi2f_ct<D,S,L,I>();if(!same(a,b))fail("vi2f");}}
|
|
template<unsigned D,unsigned M,unsigned T,unsigned Side,unsigned In> void test_vtfm(){for(int n=0;n<64;++n){auto a=make_ctx(),b=a;if((n&1)==0){a.vfpu_ctrl[0]=b.vfpu_ctrl[0]=0xE4u;a.vfpu_ctrl[1]=b.vfpu_ctrl[1]=0xE4u;a.vfpu_ctrl[2]=b.vfpu_ctrl[2]=0u;}float matrix[16]{},raw[4]{},target[4]{},result[4]{};a.read_vfpu_matrix(matrix,M,Side);a.read_vfpu_vector(raw,T,Side);for(unsigned i=0;i<4;++i)target[i]=i<In?raw[i]:0.0f;if(Side-1u>=In)target[Side-1u]=1.0f;for(unsigned row=0;row+1u<Side;++row){float sum=0.0f;for(unsigned col=0;col<Side;++col)sum+=matrix[row*4u+col]*target[col];result[row]=sum;}float final_row[4]{matrix[(Side-1u)*4u+0u],matrix[(Side-1u)*4u+1u],matrix[(Side-1u)*4u+2u],matrix[(Side-1u)*4u+3u]};a.apply_vfpu_source_prefix(final_row,4u,0u);a.apply_vfpu_source_prefix(target,4u,1u);for(unsigned col=0;col<4u;++col)result[Side-1u]+=final_row[col]*target[col];const auto dp=a.vfpu_ctrl[2];const unsigned last=Side-1u;a.vfpu_ctrl[2]=((dp&(1u<<8u))<<last)|((dp&3u)<<(last*2u));a.write_vfpu_vector_with_destination_prefix(result,D,Side);b.execute_vfpu_vtfm_ct<D,M,T,Side,In>();if(!same(a,b))fail("vtfm");}}
|
|
template<unsigned D,unsigned S,unsigned L> void test_vh2f(){for(int n=0;n<8;++n){auto a=make_ctx(),b=a;a.execute_vfpu_vh2f(D,S,L);b.execute_vfpu_vh2f_ct<D,S,L>();if(!same(a,b))fail("vh2f");}}
|
|
template<unsigned D,unsigned S,unsigned T,unsigned L> void test_vhdp(){for(int n=0;n<8;++n){auto a=make_ctx(),b=a;a.execute_vfpu_vhdp(D,S,T,L);b.execute_vfpu_vhdp_ct<D,S,T,L>();if(!same(a,b))fail("vhdp");}}
|
|
template<unsigned D,unsigned S,unsigned L,unsigned O> void test_vx2i(){for(int n=0;n<8;++n){auto a=make_ctx(),b=a;a.execute_vfpu_vx2i(D,S,L,O);b.execute_vfpu_vx2i_ct<D,S,L,O>();if(!same(a,b))fail("vx2i");}}
|
|
template<unsigned D,unsigned S,unsigned T,unsigned L,bool M> void test_minmax(){for(int n=0;n<8;++n){auto a=make_ctx(),b=a;a.execute_vfpu_vminmax(D,S,T,L,M);b.execute_vfpu_vminmax_ct<D,S,T,L,M>();if(!same(a,b))fail("minmax");}}
|
|
template<unsigned D,unsigned S,unsigned T,unsigned L,unsigned O> void test_cmp3(){for(int n=0;n<8;++n){auto a=make_ctx(),b=a;a.execute_vfpu_compare3(D,S,T,L,O);b.execute_vfpu_compare3_ct<D,S,T,L,O>();if(!same(a,b))fail("cmp3");}}
|
|
template<unsigned D,unsigned S,unsigned L,unsigned I> void test_vrot(){for(int n=0;n<8;++n){auto a=make_ctx(),b=a;a.execute_vfpu_vrot(D,S,L,I);b.execute_vfpu_vrot_ct<D,S,L,I>();if(!same(a,b))fail("vrot");}}
|
|
|
|
|
|
template<unsigned D,unsigned S,unsigned T,unsigned L,unsigned O> void test_vec3_fastlane(){
|
|
for(int n=0;n<64;++n){
|
|
auto a=make_ctx(),b=a;
|
|
if((n&1)==0){a.vfpu_ctrl[0]=b.vfpu_ctrl[0]=0xE4u;a.vfpu_ctrl[1]=b.vfpu_ctrl[1]=0xE4u;a.vfpu_ctrl[2]=b.vfpu_ctrl[2]=0u;}
|
|
float sv[4]{},tv[4]{},dv[4]{};
|
|
a.read_vfpu_vector_with_source_prefix(sv,S,L,0u);
|
|
a.read_vfpu_vector_with_source_prefix(tv,T,L,1u);
|
|
for(unsigned i=0;i<L;++i){if constexpr(O==0)dv[i]=sv[i]+tv[i];else if constexpr(O==1)dv[i]=sv[i]-tv[i];else if constexpr(O==2)dv[i]=sv[i]*tv[i];else dv[i]=sv[i]/tv[i];}
|
|
a.write_vfpu_vector_with_destination_prefix(dv,D,L);
|
|
b.execute_vfpu_vec3_ct<D,S,T,L,O>();
|
|
if(!same(a,b))fail("vec3_fastlane");
|
|
}
|
|
}
|
|
template<unsigned D,unsigned S,unsigned L,unsigned O> void test_unary_fastlane(){
|
|
for(int n=0;n<32;++n){
|
|
auto a=make_ctx(),b=a;
|
|
if((n&1)==0){a.vfpu_ctrl[0]=b.vfpu_ctrl[0]=0xE4u;a.vfpu_ctrl[2]=b.vfpu_ctrl[2]=0u;}
|
|
float sv[4]{},dv[4]{};a.read_vfpu_vector_with_source_prefix(sv,S,L,0u);
|
|
for(unsigned i=0;i<L;++i)dv[i]=AllegrexContext::vfpu_unary_lane<O>(sv[i]);
|
|
a.write_vfpu_vector_with_destination_prefix(dv,D,L);
|
|
b.execute_vfpu_unary_ct<D,S,L,O>();
|
|
if(!same(a,b))fail("unary_fastlane");
|
|
}
|
|
}
|
|
template<unsigned D,unsigned S,unsigned T,unsigned L> void test_vdot_fastlane(){for(int n=0;n<64;++n){auto a=make_ctx(),b=a;if((n&1)==0){a.vfpu_ctrl[0]=b.vfpu_ctrl[0]=0xE4u;a.vfpu_ctrl[1]=b.vfpu_ctrl[1]=0xE4u;a.vfpu_ctrl[2]=b.vfpu_ctrl[2]=0u;}a.execute_vfpu_vdot(D,S,T,L);b.execute_vfpu_vdot_ct<D,S,T,L>();if(!same(a,b))fail("vdot_fastlane");}}
|
|
template<unsigned D,unsigned S,unsigned T,unsigned L> void test_vscl_fastlane(){for(int n=0;n<64;++n){auto a=make_ctx(),b=a;if((n&1)==0){a.vfpu_ctrl[0]=b.vfpu_ctrl[0]=0xE4u;a.vfpu_ctrl[1]=b.vfpu_ctrl[1]=0xE4u;a.vfpu_ctrl[2]=b.vfpu_ctrl[2]=0u;}a.execute_vfpu_vscl(D,S,T,L);b.execute_vfpu_vscl_ct<D,S,T,L>();if(!same(a,b))fail("vscl_fastlane");}}
|
|
template<unsigned D,unsigned L> void test_dest_prefix(){for(int n=0;n<64;++n){auto a=make_ctx(),b=a;float x[4]{};std::uniform_real_distribution<float>f(-16,16);for(float&v:x)v=f(rng);a.write_vfpu_vector_with_destination_prefix(x,D,L);b.write_vfpu_vector_with_destination_prefix_ct<D,L>(x);if(!same(a,b))fail("dest_prefix");}}
|
|
template<unsigned D,unsigned S,unsigned L> void test_vf2h(){for(int n=0;n<32;++n){auto a=make_ctx(),b=a;a.execute_vfpu_vf2h(D,S,L);b.execute_vfpu_vf2h_ct<D,S,L>();if(!same(a,b))fail("vf2h");}}
|
|
template<unsigned D,unsigned S,unsigned L,bool A> void test_horizontal(){for(int n=0;n<32;++n){auto a=make_ctx(),b=a;a.execute_vfpu_horizontal(D,S,L,A);b.execute_vfpu_horizontal_ct<D,S,L,A>();if(!same(a,b))fail("horizontal");}}
|
|
template<unsigned D,unsigned S,unsigned L> void test_vocp(){for(int n=0;n<32;++n){auto a=make_ctx(),b=a;a.execute_vfpu_vocp(D,S,L);b.execute_vfpu_vocp_ct<D,S,L>();if(!same(a,b))fail("vocp");}}
|
|
template<unsigned D,unsigned S,unsigned L> void test_vmmov(){for(int n=0;n<32;++n){auto a=make_ctx(),b=a;a.execute_vfpu_vmmov(D,S,L);b.execute_vfpu_vmmov_ct<D,S,L>();if(!same(a,b))fail("vmmov");}}
|
|
template<unsigned D,unsigned S,unsigned T,unsigned L> void test_vmscl(){for(int n=0;n<32;++n){auto a=make_ctx(),b=a;a.execute_vfpu_vmscl(D,S,T,L);b.execute_vfpu_vmscl_ct<D,S,T,L>();if(!same(a,b))fail("vmscl");}}
|
|
template<unsigned D,unsigned L,unsigned O> void test_matrix_init(){for(int n=0;n<32;++n){auto a=make_ctx(),b=a;a.execute_vfpu_matrix_init(D,L,O);b.execute_vfpu_matrix_init_ct<D,L,O>();if(!same(a,b))fail("matrix_init");}}
|
|
int main(){
|
|
test_matrix_read<0,4>();
|
|
test_matrix_read<16,3>();
|
|
test_matrix_read<24,4>();
|
|
test_matrix_read<32,3>();
|
|
test_matrix_read<32,4>();
|
|
test_matrix_read<36,3>();
|
|
test_matrix_read<36,4>();
|
|
test_matrix_read<40,4>();
|
|
test_matrix_read<44,3>();
|
|
test_matrix_read<48,3>();
|
|
test_matrix_read<4,3>();
|
|
test_matrix_read<4,4>();
|
|
test_matrix_read<56,3>();
|
|
test_matrix_read<56,4>();
|
|
test_matrix_read<8,4>();
|
|
test_matrix_write<12,4>();
|
|
test_matrix_write<16,4>();
|
|
test_matrix_write<20,4>();
|
|
test_matrix_write<32,4>();
|
|
test_matrix_write<40,4>();
|
|
test_cross<0,1,2,3>();
|
|
test_cross<14,11,13,3>();
|
|
test_cross<15,11,8,3>();
|
|
test_cross<16,10,12,3>();
|
|
test_cross<17,10,9,3>();
|
|
test_cross<18,9,12,3>();
|
|
test_cross<19,9,10,3>();
|
|
test_cross<2,4,1,3>();
|
|
test_vh2f<0,0,1>();
|
|
test_vh2f<13,12,2>();
|
|
test_vh2f<1,0,2>();
|
|
test_vh2f<1,16,2>();
|
|
test_vh2f<1,32,1>();
|
|
test_vh2f<20,32,1>();
|
|
test_vh2f<21,12,2>();
|
|
test_vh2f<2,64,2>();
|
|
test_vh2f<2,80,2>();
|
|
test_vhdp<2,0,48,4>();
|
|
test_vhdp<2,0,49,4>();
|
|
test_vhdp<2,0,50,4>();
|
|
test_vhdp<2,0,51,4>();
|
|
test_vhdp<34,1,52,4>();
|
|
test_vhdp<34,1,53,4>();
|
|
test_vhdp<34,1,54,4>();
|
|
test_vhdp<34,1,55,4>();
|
|
test_vx2i<0,2,2,3>();
|
|
test_vx2i<12,13,2,3>();
|
|
test_vx2i<13,14,1,3>();
|
|
test_vx2i<1,66,2,3>();
|
|
test_vx2i<2,8,2,3>();
|
|
test_vx2i<3,9,2,3>();
|
|
test_vx2i<4,10,2,3>();
|
|
test_vx2i<4,8,2,3>();
|
|
test_vx2i<5,9,2,3>();
|
|
test_vx2i<6,10,2,3>();
|
|
test_minmax<111,118,114,1,false>();
|
|
test_minmax<15,15,47,1,true>();
|
|
test_minmax<15,15,79,1,true>();
|
|
test_minmax<2,0,1,3,false>();
|
|
test_minmax<3,0,1,3,true>();
|
|
test_minmax<98,8,40,1,true>();
|
|
test_cmp3<12,5,2,3,7>();
|
|
test_cmp3<13,3,4,3,7>();
|
|
test_vrot<1,64,2,4>();
|
|
test_vrot<4,0,4,11>();
|
|
test_dest_prefix<0,3>();
|
|
test_dest_prefix<1,3>();
|
|
test_dest_prefix<3,3>();
|
|
test_dest_prefix<5,3>();
|
|
test_dest_prefix<5,4>();
|
|
test_dest_prefix<13,4>();
|
|
test_dest_prefix<14,4>();
|
|
test_vocp<16,16,1>();
|
|
test_vocp<52,20,1>();
|
|
test_vocp<32,0,1>();
|
|
test_matrix_init<0,4,3>();
|
|
test_matrix_init<32,4,3>();
|
|
test_matrix_init<32,3,3>();
|
|
test_horizontal<8,36,3,true>();
|
|
test_horizontal<40,37,3,true>();
|
|
test_horizontal<72,38,3,true>();
|
|
test_vf2h<64,0,2>();
|
|
test_vmmov<32,40,4>();
|
|
test_vmmov<24,0,4>();
|
|
test_vmscl<32,36,8,3>();
|
|
test_vcmp<14,0,4,3>();
|
|
test_vcmp<1,2,3,9>();
|
|
test_vcmov<66,98,1,0,false>();
|
|
test_vcmov<14,11,3,6,true>();
|
|
test_vi2f<2,1,2,3>();
|
|
test_vi2f<66,32,1,0>();
|
|
test_vi2f<12,36,4,16>();
|
|
test_vtfm<14,36,12,4,3>();
|
|
test_vtfm<0,36,1,3,3>();
|
|
test_vtfm<32,40,8,4,4>();
|
|
test_vec3_fastlane<0,1,2,3,0>();
|
|
test_vec3_fastlane<32,36,40,4,1>();
|
|
test_vec3_fastlane<14,11,13,3,2>();
|
|
test_vec3_fastlane<3,4,5,2,3>();
|
|
test_unary_fastlane<0,1,3,0>();
|
|
test_unary_fastlane<32,36,4,2>();
|
|
test_unary_fastlane<14,11,3,17>();
|
|
test_unary_fastlane<5,9,2,22>();
|
|
test_vdot_fastlane<2,0,48,4>();
|
|
test_vdot_fastlane<34,1,52,3>();
|
|
test_vscl_fastlane<32,36,8,3>();
|
|
test_vscl_fastlane<0,1,2,4>();
|
|
std::cout<<"vfpu_tier2_tests: PASS\n";
|
|
}
|