fabi/Xenia-Canary
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

use much faster exp2/cos approximations in ffmpeg, large decrease in cpu usage on my machine on decoder thread

Browse Source 
properly byteswap r13 for spinlock
Add PPCOpcodeBits
stub out broken fpscr updating in ppc_hir_builder. it's just code that repeatedly does nothing right now.
add note about 0 opcode bytes being executed to ppc_frontend
Add assert to check that function end is greater than function start, can happen with malformed functions
Disable prefetch and cachecontrol by default, automatic hardware prefetchers already do the job for the most part
minor cleanup in simplification_pass, dont loop optimizations, let the pass manager do it for us
Add experimental "delay_via_maybeyield" cvar, which uses MaybeYield to "emulate" the db16cyc instruction
Add much faster/simpler way of directly calling guest functions, no longer have to do a byte by byte search through the generated code
Generate label string ids on the fly
Fix unused function warnings for prefetch on clang, fix many other clang warnings
Eliminated majority of CallNativeSafes by replacing them with naive generic code paths.
^ Vector rotate left, vector shift left, vector shift right, vector shift arithmetic right, and vector average are included
These naive paths are implemented small loops that stash the two inputs to the stack and load them in gprs from there, they are not particularly fast but should be an order of magnitude faster than callnativesafe
to a host function, which would involve a call, stashing all volatile registers, an indirect call, potentially setting up a stack frame for the arrays that the inputs get stashed to, the actual operations, a return, loading all volatile registers, a return, etc
Added the fast SHR_V128 path back in
Implement signed vector average byte, signed vector average word. previously we were emitting no code for them. signed vector average byte appears in many games
Fix bug with signed vector average 32, we were doing unsigned shift, turning negative values into big positive ones potentially
This commit is contained in:
chss95cs@gmail.com
2022-10-30 08:48:58 -07:00
parent d8b7b3ecec
commit 550d1d0a7c
26 changed files with 881 additions and 547 deletions
Download Patch File Download Diff File Expand all files Collapse all files

96
src/xenia/cpu/backend/x64/x64_emitter.cc
View File

@@ -46,10 +46,6 @@ DEFINE_bool(ignore_undefined_externs, true,
DEFINE_bool(emit_source_annotations, false,
"Add extra movs and nops to make disassembly easier to read.",
"CPU");
DEFINE_bool(resolve_rel32_guest_calls, true,
"Experimental optimization, directly call already resolved "
"functions via x86 rel32 call/jmp",
"CPU");
DEFINE_bool(enable_incorrect_roundingmode_behavior, false,
"Disables the FPU/VMX MXCSR sharing workaround, potentially "
@@ -78,7 +74,6 @@ using namespace xe::literals;
static const size_t kMaxCodeSize = 1_MiB;
static const size_t kStashOffset = 32;
// static const size_t kStashOffsetHigh = 32 + 32;
const uint32_t X64Emitter::gpr_reg_map_[X64Emitter::GPR_COUNT] = {
@@ -141,55 +136,6 @@ bool X64Emitter::Emit(GuestFunction* function, HIRBuilder* builder,
return true;
}
#pragma pack(push, 1)
struct RGCEmitted {
uint8_t ff_;
uint32_t rgcid_;
};
#pragma pack(pop)
#if 0
void X64Emitter::InjectCallAddresses(void* new_execute_address) {
for (auto&& callsite : call_sites_) {
RGCEmitted* hunter = (RGCEmitted*)new_execute_address;
while (hunter->ff_ != 0xFF || hunter->rgcid_ != callsite.offset_) {
hunter =
reinterpret_cast<RGCEmitted*>(reinterpret_cast<char*>(hunter) + 1);
}
hunter->ff_ = callsite.is_jump_ ? 0xE9 : 0xE8;
hunter->rgcid_ =
static_cast<uint32_t>(static_cast<intptr_t>(callsite.destination_) -
reinterpret_cast<intptr_t>(hunter + 1));
}
}
#else
void X64Emitter::InjectCallAddresses(void* new_execute_address) {
#if 0
RGCEmitted* hunter = (RGCEmitted*)new_execute_address;
std::map<uint32_t, ResolvableGuestCall*> id_to_rgc{};
for (auto&& callsite : call_sites_) {
id_to_rgc[callsite.offset_] = &callsite;
}
#else
RGCEmitted* hunter = (RGCEmitted*)new_execute_address;
for (auto&& callsite : call_sites_) {
while (hunter->ff_ != 0xFF || hunter->rgcid_ != callsite.offset_) {
hunter =
reinterpret_cast<RGCEmitted*>(reinterpret_cast<char*>(hunter) + 1);
}
hunter->ff_ = callsite.is_jump_ ? 0xE9 : 0xE8;
hunter->rgcid_ =
static_cast<uint32_t>(static_cast<intptr_t>(callsite.destination_) -
reinterpret_cast<intptr_t>(hunter + 1));
}
#endif
}
#endif
void* X64Emitter::Emplace(const EmitFunctionInfo& func_info,
GuestFunction* function) {
// To avoid changing xbyak, we do a switcharoo here.
@@ -207,10 +153,6 @@ void* X64Emitter::Emplace(const EmitFunctionInfo& func_info,
if (function) {
code_cache_->PlaceGuestCode(function->address(), top_, func_info, function,
new_execute_address, new_write_address);
if (cvars::resolve_rel32_guest_calls) {
InjectCallAddresses(new_execute_address);
}
} else {
code_cache_->PlaceHostCode(0, top_, func_info, new_execute_address,
new_write_address);
@@ -219,7 +161,6 @@ void* X64Emitter::Emplace(const EmitFunctionInfo& func_info,
ready();
top_ = old_address;
reset();
call_sites_.clear();
tail_code_.clear();
for (auto&& cached_label : label_cache_) {
delete cached_label;
@@ -336,7 +277,7 @@ bool X64Emitter::Emit(HIRBuilder* builder, EmitFunctionInfo& func_info) {
// Mark block labels.
auto label = block->label_head;
while (label) {
L(label->name);
L(std::to_string(label->id));
label = label->next;
}
@@ -418,7 +359,6 @@ void X64Emitter::EmitProfilerEpilogue() {
// actually... lets just try without atomics lol
// lock();
add(qword[r10], rdx);
}
#endif
}
@@ -534,44 +474,23 @@ void X64Emitter::Call(const hir::Instr* instr, GuestFunction* function) {
auto fn = static_cast<X64Function*>(function);
// Resolve address to the function to call and store in rax.
if (cvars::resolve_rel32_guest_calls && fn->machine_code()) {
ResolvableGuestCall rgc;
rgc.destination_ = uint32_t(uint64_t(fn->machine_code()));
rgc.offset_ = current_rgc_id_;
current_rgc_id_++;
if (fn->machine_code()) {
if (!(instr->flags & hir::CALL_TAIL)) {
mov(rcx, qword[rsp + StackLayout::GUEST_CALL_RET_ADDR]);
db(0xFF);
rgc.is_jump_ = false;
dd(rgc.offset_);
call((void*)fn->machine_code());
} else {
// tail call
EmitTraceUserCallReturn();
rgc.is_jump_ = true;
EmitProfilerEpilogue();
// Pass the callers return address over.
mov(rcx, qword[rsp + StackLayout::GUEST_RET_ADDR]);
add(rsp, static_cast<uint32_t>(stack_size()));
db(0xFF);
dd(rgc.offset_);
jmp((void*)fn->machine_code(), T_NEAR);
}
call_sites_.push_back(rgc);
return;
}
if (fn->machine_code()) {
// TODO(benvanik): is it worth it to do this? It removes the need for
// a ResolveFunction call, but makes the table less useful.
assert_zero(uint64_t(fn->machine_code()) & 0xFFFFFFFF00000000);
// todo: this should be changed so that we can actually do a call to
// fn->machine_code. the code will be emitted near us, so 32 bit rel jmp
// should be possible
mov(eax, uint32_t(uint64_t(fn->machine_code())));
} else if (code_cache_->has_indirection_table()) {
// Load the pointer to the indirection table maintained in X64CodeCache.
// The target dword will either contain the address of the generated code
@@ -1017,7 +936,10 @@ static const vec128_t xmm_consts[] = {
/*XMMSTVLShuffle*/
v128_setr_bytes(0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15),
/* XMMSTVRSwapMask*/
vec128b((uint8_t)0x83)};
vec128b((uint8_t)0x83), /*XMMVSRShlByteshuf*/
v128_setr_bytes(13, 14, 15, 8, 9, 10, 11, 4, 5, 6, 7, 0, 1, 2, 3, 0x80),
// XMMVSRMask
vec128b(1)};
void* X64Emitter::FindByteConstantOffset(unsigned bytevalue) {
for (auto& vec : xmm_consts) {