T check_projection_accurate(const arma::Mat<T>& data,
const armapca::pca_result<T>& result) {
EXPECTS(data.n_rows >= 2);
EXPECTS(data.n_cols >= 1);
EXPECTS(result.eigenvectors.n_cols >= 1);
const arma::Mat<T> diff = ((data * result.eigenvectors) *
arma::trans(result.eigenvectors)) - data;
return T {1} - arma::sum(arma::sum(arma::abs(diff) )) / diff.n_elem;
}
std::vector<armapca::pca_result<T>> pca_bootstrap(const arma::Mat<T>& data,
bool compute_eigenvectors,
std::size_t n_bootstraps,
std::size_t random_seed) {
EXPECTS(data.n_rows >= 2);
EXPECTS(data.n_cols >= 1);
std::vector<armapca::pca_result<T>> result(n_bootstraps);
std::mt19937 gen{random_seed};
for (std::size_t i=0; i < n_bootstraps; ++i) {
const auto shuffled = armapca::shuffled_matrix(data, &gen);
result[i] = armapca::pca(shuffled, compute_eigenvectors);
}
return result;
}
armapca::pca_result<T> pca(const arma::Mat<T>& data,
bool compute_eigenvectors) {
EXPECTS(data.n_rows >= 2);
EXPECTS(data.n_cols >= 1);
const auto n_vars = data.n_cols;
armapca::pca_result<T> result;
arma::Mat<T> eigvec;
if (compute_eigenvectors) {
result.eigenvectors.set_size(n_vars, n_vars);
eigvec.set_size(n_vars, n_vars);
}
result.eigenvalues.set_size(n_vars);
arma::Col<T> eigval(n_vars);
const auto cov_mat = armapca::covariance_matrix(data);
if (compute_eigenvectors) {
arma::eig_sym(eigval, eigvec, cov_mat);
} else {
arma::eig_sym(eigval, cov_mat);
}
const arma::uvec indices = arma::sort_index(eigval, 1);
for (std::size_t i=0; i < n_vars; ++i) {
result.eigenvalues(i) = eigval(indices(i));
}
if (compute_eigenvectors) {
for (std::size_t i=0; i < n_vars; ++i) {
result.eigenvectors.col(i) = eigvec.col(indices(i));
}
armapca::enforce_positive_sign_by_column(&result.eigenvectors);
}
result.energy = arma::sum(result.eigenvalues);
result.eigenvalues *= T {1} / result.energy;
return result;
}
u64 write(const void* buffer, u64 count) override
{
// TODO (call WriteFile multiple times if count is too big)
const int size = ::narrow<int>(count, "Too big count" HERE);
EXPECTS(size >= 0);
DWORD nwritten;
if (!WriteFile(m_handle, buffer, size, &nwritten, NULL))
{
throw fmt::exception("Win32 error: %u." HERE, GetLastError());
}
return nwritten;
}
u32 get_vertex_type_size_on_host(vertex_base_type type, u32 size)
{
switch (type)
{
case vertex_base_type::s1:
case vertex_base_type::s32k:
switch (size)
{
case 1:
case 2:
case 4:
return sizeof(u16) * size;
case 3:
return sizeof(u16) * 4;
}
throw EXCEPTION("Wrong vector size");
case vertex_base_type::f: return sizeof(f32) * size;
case vertex_base_type::sf:
switch (size)
{
case 1:
case 2:
case 4:
return sizeof(f16) * size;
case 3:
return sizeof(f16) * 4;
}
throw EXCEPTION("Wrong vector size");
case vertex_base_type::ub:
switch (size)
{
case 1:
case 2:
case 4:
return sizeof(u8) * size;
case 3:
return sizeof(u8) * 4;
}
throw EXCEPTION("Wrong vector size");
case vertex_base_type::cmp: return sizeof(u16) * 4;
case vertex_base_type::ub256: EXPECTS(size == 4); return sizeof(u8) * 4;
}
throw EXCEPTION("RSXVertexData::GetTypeSize: Bad vertex data type (%d)!", type);
}
void thread::do_internal_task()
{
if (m_internal_tasks.empty())
{
std::this_thread::sleep_for(1ms);
}
else
{
EXPECTS(0);
//std::lock_guard<std::mutex> lock{ m_mtx_task };
//internal_task_entry &front = m_internal_tasks.front();
//if (front.callback())
//{
// front.promise.set_value();
// m_internal_tasks.pop_front();
//}
}
}
void CgBinaryDisasm::AddCodeAsm(const std::string& code)
{
EXPECTS(m_opcode < 70);
std::string op_name = "";
if (dst.dest_reg == 63)
{
m_dst_reg_name = fmt::format("RC%s, ", GetMask().c_str());
op_name = rsx_fp_op_names[m_opcode] + "XC";
}
else
{
m_dst_reg_name = fmt::format("%s%d%s, ", dst.fp16 ? "H" : "R", dst.dest_reg, GetMask().c_str());
op_name = rsx_fp_op_names[m_opcode] + std::string(dst.fp16 ? "H" : "R");
}
switch (m_opcode)
{
case RSX_FP_OPCODE_BRK:
case RSX_FP_OPCODE_CAL:
case RSX_FP_OPCODE_FENCT:
case RSX_FP_OPCODE_FENCB:
case RSX_FP_OPCODE_IFE:
case RSX_FP_OPCODE_KIL:
case RSX_FP_OPCODE_LOOP:
case RSX_FP_OPCODE_NOP:
case RSX_FP_OPCODE_REP:
case RSX_FP_OPCODE_RET:
m_dst_reg_name = "";
op_name = rsx_fp_op_names[m_opcode] + std::string(dst.fp16 ? "H" : "R");
break;
default: break;
}
m_arb_shader += (op_name + " " + m_dst_reg_name + FormatDisAsm(code) + ";" + "\n");
}
std::shared_ptr<fs::device_base> fs::set_virtual_device(const std::string& name, const std::shared_ptr<device_base>& device)
{
EXPECTS(name.size() > 2 && name[0] == '/' && name[1] == '/' && name.find('/', 2) == -1);
return get_device_manager().set_device(name, device);
}
arma::Mat<T> covariance_matrix(const arma::Mat<T>& matrix) {
EXPECTS(matrix.n_rows >= 2);
EXPECTS(matrix.n_cols >= 1);
return (arma::trans(matrix) * matrix) * (T {1} / (matrix.n_rows - 1));
}
void CgBinaryDisasm::TaskFP()
{
m_size = 0;
u32* data = (u32*)&m_buffer[m_offset];
EXPECTS((m_buffer_size - m_offset) % sizeof(u32) == 0);
for (u32 i = 0; i < (m_buffer_size - m_offset) / sizeof(u32); i++)
{
data[i] = se_storage<u32>::swap(data[i]); // WTF, cannot use be_t<> there?
}
enum
{
FORCE_NONE,
FORCE_SCT,
FORCE_SCB
};
int forced_unit = FORCE_NONE;
while (true)
{
for (auto found = std::find(m_end_offsets.begin(), m_end_offsets.end(), m_size);
found != m_end_offsets.end();
found = std::find(m_end_offsets.begin(), m_end_offsets.end(), m_size))
{
m_end_offsets.erase(found);
m_arb_shader += "ENDIF;\n";
}
for (auto found = std::find(m_loop_end_offsets.begin(), m_loop_end_offsets.end(), m_size);
found != m_loop_end_offsets.end();
found = std::find(m_loop_end_offsets.begin(), m_loop_end_offsets.end(), m_size))
{
m_loop_end_offsets.erase(found);
m_arb_shader += "ENDLOOP;\n";
}
for (auto found = std::find(m_else_offsets.begin(), m_else_offsets.end(), m_size);
found != m_else_offsets.end();
found = std::find(m_else_offsets.begin(), m_else_offsets.end(), m_size))
{
m_else_offsets.erase(found);
m_arb_shader += "ELSE;\n";
}
dst.HEX = GetData(data[0]);
src0.HEX = GetData(data[1]);
src1.HEX = GetData(data[2]);
src2.HEX = GetData(data[3]);
m_step = 4 * sizeof(u32);
m_opcode = dst.opcode | (src1.opcode_is_branch << 6);
auto SCT = [&]()
{
switch (m_opcode)
{
case RSX_FP_OPCODE_ADD: AddCodeAsm("$0, $1"); break;
case RSX_FP_OPCODE_DIV: AddCodeAsm("$0, $1"); break;
case RSX_FP_OPCODE_DIVSQ: AddCodeAsm("$0, $1"); break;
case RSX_FP_OPCODE_DP2: AddCodeAsm("$0, $1"); break;
case RSX_FP_OPCODE_DP3: AddCodeAsm("$0, $1"); break;
case RSX_FP_OPCODE_DP4: AddCodeAsm("$0, $1"); break;
case RSX_FP_OPCODE_DP2A: AddCodeAsm("$0, $1, $2"); break;
case RSX_FP_OPCODE_MAD: AddCodeAsm("$0, $1, $2"); break;
case RSX_FP_OPCODE_MAX: AddCodeAsm("$0, $1"); break;
case RSX_FP_OPCODE_MIN: AddCodeAsm("$0, $1"); break;
case RSX_FP_OPCODE_MOV: AddCodeAsm("$0"); break;
case RSX_FP_OPCODE_MUL: AddCodeAsm("$0, $1"); break;
case RSX_FP_OPCODE_RCP: AddCodeAsm("$0"); break;
case RSX_FP_OPCODE_RSQ: AddCodeAsm("$0"); break;
case RSX_FP_OPCODE_SEQ: AddCodeAsm("$0, $1"); break;
case RSX_FP_OPCODE_SFL: AddCodeAsm("$0, $1"); break;
case RSX_FP_OPCODE_SGE: AddCodeAsm("$0, $1"); break;
case RSX_FP_OPCODE_SGT: AddCodeAsm("$0, $1"); break;
case RSX_FP_OPCODE_SLE: AddCodeAsm("$0, $1"); break;
case RSX_FP_OPCODE_SLT: AddCodeAsm("$0, $1"); break;
case RSX_FP_OPCODE_SNE: AddCodeAsm("$0, $1"); break;
case RSX_FP_OPCODE_STR: AddCodeAsm("$0, $1"); break;
default:
return false;
}
return true;
};
auto SCB = [&]()
{
switch (m_opcode)
{
case RSX_FP_OPCODE_ADD: AddCodeAsm("$0, $1"); break;
case RSX_FP_OPCODE_COS: AddCodeAsm("$0"); break;
case RSX_FP_OPCODE_DP2: AddCodeAsm("$0, $1"); break;
case RSX_FP_OPCODE_DP3: AddCodeAsm("$0, $1"); break;
case RSX_FP_OPCODE_DP4: AddCodeAsm("$0, $1"); break;
case RSX_FP_OPCODE_DP2A: AddCodeAsm("$0, $1, $2"); break;
case RSX_FP_OPCODE_DST: AddCodeAsm("$0, $1"); break;
case RSX_FP_OPCODE_REFL: AddCodeAsm("$0, $1"); break;
case RSX_FP_OPCODE_EX2: AddCodeAsm("$0"); break;
case RSX_FP_OPCODE_FLR: AddCodeAsm("$0"); break;
case RSX_FP_OPCODE_FRC: AddCodeAsm("$0"); break;
case RSX_FP_OPCODE_LIT: AddCodeAsm("$0"); break;
case RSX_FP_OPCODE_LIF: AddCodeAsm("$0"); break;
case RSX_FP_OPCODE_LRP: AddCodeAsm("# WARNING"); break;
case RSX_FP_OPCODE_LG2: AddCodeAsm("$0"); break;
case RSX_FP_OPCODE_MAD: AddCodeAsm("$0, $1, $2"); break;
case RSX_FP_OPCODE_MAX: AddCodeAsm("$0, $1"); break;
case RSX_FP_OPCODE_MIN: AddCodeAsm("$0, $1"); break;
case RSX_FP_OPCODE_MOV: AddCodeAsm("$0"); break;
case RSX_FP_OPCODE_MUL: AddCodeAsm("$0, $1"); break;
case RSX_FP_OPCODE_PK2: AddCodeAsm("$0"); break;
case RSX_FP_OPCODE_PK4: AddCodeAsm("$0"); break;
case RSX_FP_OPCODE_PK16: AddCodeAsm("$0"); break;
case RSX_FP_OPCODE_PKB: AddCodeAsm("$0"); break;
case RSX_FP_OPCODE_PKG: AddCodeAsm("$0"); break;
case RSX_FP_OPCODE_SEQ: AddCodeAsm("$0, $1"); break;
case RSX_FP_OPCODE_SFL: AddCodeAsm("$0, $1"); break;
case RSX_FP_OPCODE_SGE: AddCodeAsm("$0, $1"); break;
case RSX_FP_OPCODE_SGT: AddCodeAsm("$0, $1"); break;
case RSX_FP_OPCODE_SIN: AddCodeAsm("$0"); break;
case RSX_FP_OPCODE_SLE: AddCodeAsm("$0, $1"); break;
case RSX_FP_OPCODE_SLT: AddCodeAsm("$0, $1"); break;
case RSX_FP_OPCODE_SNE: AddCodeAsm("$0, $1"); break;
case RSX_FP_OPCODE_STR: AddCodeAsm("$0, $1"); break;
default:
return false;
}
return true;
};
auto TEX_SRB = [&]()
{
switch (m_opcode)
{
case RSX_FP_OPCODE_DDX: AddCodeAsm("$0"); break;
case RSX_FP_OPCODE_DDY: AddCodeAsm("$0"); break;
case RSX_FP_OPCODE_NRM: AddCodeAsm("$0"); break;
case RSX_FP_OPCODE_BEM: AddCodeAsm("# WARNING"); break;
case RSX_FP_OPCODE_TEX: AddCodeAsm("$0, $t"); break;
case RSX_FP_OPCODE_TEXBEM: AddCodeAsm("# WARNING"); break;
case RSX_FP_OPCODE_TXP: AddCodeAsm("$0"); break;
case RSX_FP_OPCODE_TXPBEM: AddCodeAsm("# WARNING"); break;
case RSX_FP_OPCODE_TXD: AddCodeAsm("$0, $1, $t"); break;
case RSX_FP_OPCODE_TXB: AddCodeAsm("$0, $t"); break;
case RSX_FP_OPCODE_TXL: AddCodeAsm("$0, $t"); break;
case RSX_FP_OPCODE_UP2: AddCodeAsm("$0"); break;
case RSX_FP_OPCODE_UP4: AddCodeAsm("$0"); break;
case RSX_FP_OPCODE_UP16: AddCodeAsm("$0"); break;
case RSX_FP_OPCODE_UPB: AddCodeAsm("$0"); break;
case RSX_FP_OPCODE_UPG: AddCodeAsm("$0"); break;
default:
return false;
}
return true;
};
auto SIP = [&]()
{
switch (m_opcode)
{
case RSX_FP_OPCODE_BRK: AddCodeAsm("$cond"); break;
case RSX_FP_OPCODE_CAL: AddCodeAsm("$cond"); break;
case RSX_FP_OPCODE_FENCT: AddCodeAsm(""); break;
case RSX_FP_OPCODE_FENCB: AddCodeAsm(""); break;
case RSX_FP_OPCODE_IFE:
{
m_else_offsets.push_back(src1.else_offset << 2);
m_end_offsets.push_back(src2.end_offset << 2);
AddCodeAsm("($cond)");
}
break;
case RSX_FP_OPCODE_LOOP:
{
if (!src0.exec_if_eq && !src0.exec_if_gr && !src0.exec_if_lt)
{
AddCodeAsm(fmt::format("{ %u, %u, %u }", src1.end_counter, src1.init_counter, src1.increment));
}
else
{
m_loop_end_offsets.push_back(src2.end_offset << 2);
AddCodeAsm(fmt::format("{ %u, %u, %u }", src1.end_counter, src1.init_counter, src1.increment));
}
}
break;
case RSX_FP_OPCODE_REP:
{
if (!src0.exec_if_eq && !src0.exec_if_gr && !src0.exec_if_lt)
{
m_arb_shader += "# RSX_FP_OPCODE_REP_1\n";
}
else
{
m_end_offsets.push_back(src2.end_offset << 2);
m_arb_shader += "# RSX_FP_OPCODE_REP_2\n";
}
}
break;
case RSX_FP_OPCODE_RET: AddCodeAsm("$cond"); break;
default:
return false;
}
return true;
};
switch (m_opcode)
{
case RSX_FP_OPCODE_NOP: AddCodeAsm(""); break;
case RSX_FP_OPCODE_KIL: AddCodeAsm("$cond"); break;
default:
if (forced_unit == FORCE_NONE)
{
if (SIP()) break;
if (SCT()) break;
if (TEX_SRB()) break;
if (SCB()) break;
}
else if (forced_unit == FORCE_SCT)
{
forced_unit = FORCE_NONE;
if (SCT()) break;
}
else if (forced_unit == FORCE_SCB)
{
forced_unit = FORCE_NONE;
if (SCB()) break;
}
LOG_ERROR(RSX, "Unknown/illegal instruction: 0x%x (forced unit %d)", m_opcode, forced_unit);
break;
}
m_size += m_step;
if (dst.end)
{
m_arb_shader.pop_back();
m_arb_shader += " # last inctruction\nEND\n";
break;
}
ENSURES(m_step % sizeof(u32) == 0);
data += m_step / sizeof(u32);
}
}
