Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

SPU LLVM: Use VDBPSADBW in SUMB #10937

Merged
merged 3 commits into from Sep 30, 2021
Merged

SPU LLVM: Use VDBPSADBW in SUMB #10937

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
b3e7e5d
SPU LLVM: Use VDBPSADBW in SUMB
Whatcookie Sep 26, 2021
06585d9
SPU LLVM: Optimize branches following byteswaps
Whatcookie Sep 29, 2021
e4624d8
LLVM DSL: reimplement vdbpsadbw
Whatcookie Sep 30, 2021
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Jump to
Jump to file
Failed to load files.
Diff view
Diff view
6 changes: 6 additions & 0 deletions rpcs3/Emu/CPU/CPUTranslator.h
View file
Open in desktop
Expand Up @@ -3564,6 +3564,12 @@ class cpu_translator
{
return llvm_calli<f32[4], T, U>{"llvm.x86.sse.min.ps", {std::forward<T>(a), std::forward<U>(b)}};
}

template <typename T, typename U, typename = std::enable_if_t<std::is_same_v<llvm_common_t<T, U>, u8[16]>>>
static auto vdbpsadbw(T&& a, U&& b, u8 c)
{
return llvm_calli<u8[16], T, U, llvm_const_int<u32>>{"llvm.x86.avx512.dbpsadbw.128", {std::forward<T>(a), std::forward<U>(b), llvm_const_int<u32>{c}}};
}
};

// Format llvm::SizeType
Expand Down
77 changes: 76 additions & 1 deletion rpcs3/Emu/Cell/SPURecompiler.cpp
View file
Open in desktop
Expand Up @@ -7022,6 +7022,23 @@ class spu_llvm_recompiler : public spu_recompiler_base, public cpu_translator

void SUMB(spu_opcode_t op)
{
if (m_use_avx512)
{
const auto [a, b] = get_vrs<u8[16]>(op.ra, op.rb);
const auto zeroes = splat<u8[16]>(0);

if (op.ra == op.rb && !m_interp_magn)
{
set_vr(op.rt, vdbpsadbw(a, zeroes, 0));
return;
}

const auto ax = vdbpsadbw(a, zeroes, 0);
const auto bx = vdbpsadbw(b, zeroes, 0);
set_vr(op.rt, shuffle2(ax, bx, 0, 8, 2, 10, 4, 12, 6, 14));
return;
}

if (m_use_vnni)
{
const auto [a, b] = get_vrs<u32[4]>(op.ra, op.rb);
Expand Down Expand Up @@ -9215,6 +9232,20 @@ class spu_llvm_recompiler : public spu_recompiler_base, public cpu_translator
{
if (m_block) m_block->block_end = m_ir->GetInsertBlock();

const auto rt = get_vr<u8[16]>(op.rt);

// Checking for zero doeesn't care about the order of the bytes,
// so load the data before it's byteswapped
if (auto [ok, as] = match_expr(rt, byteswap(match<u8[16]>())); ok)
{
m_block->block_end = m_ir->GetInsertBlock();
const auto cond = eval(extract(bitcast<u32[4]>(as), 0) == 0);
const auto addr = eval(extract(get_vr(op.ra), 3) & 0x3fffc);
const auto target = add_block_indirect(op, addr);
m_ir->CreateCondBr(cond.value, target, add_block_next());
return;
}

// Check sign bit instead (optimization)
if (match_vr<s32[4], s64[2]>(op.rt, [&](auto c, auto MP)
{
Expand Down Expand Up @@ -9246,6 +9277,21 @@ class spu_llvm_recompiler : public spu_recompiler_base, public cpu_translator
{
if (m_block) m_block->block_end = m_ir->GetInsertBlock();

const auto rt = get_vr<u8[16]>(op.rt);

// Checking for zero doeesn't care about the order of the bytes,
// so load the data before it's byteswapped
if (auto [ok, as] = match_expr(rt, byteswap(match<u8[16]>())); ok)
{
m_block->block_end = m_ir->GetInsertBlock();
const auto cond = eval(extract(bitcast<u32[4]>(as), 0) != 0);
const auto addr = eval(extract(get_vr(op.ra), 3) & 0x3fffc);
const auto target = add_block_indirect(op, addr);
m_ir->CreateCondBr(cond.value, target, add_block_next());
return;
}


// Check sign bit instead (optimization)
if (match_vr<s32[4], s64[2]>(op.rt, [&](auto c, auto MP)
{
Expand Down Expand Up @@ -9466,6 +9512,21 @@ class spu_llvm_recompiler : public spu_recompiler_base, public cpu_translator

const u32 target = spu_branch_target(m_pos, op.i16);

const auto rt = get_vr<u8[16]>(op.rt);

// Checking for zero doeesn't care about the order of the bytes,
// so load the data before it's byteswapped
if (auto [ok, as] = match_expr(rt, byteswap(match<u8[16]>())); ok)
{
if (target != m_pos + 4)
{
m_block->block_end = m_ir->GetInsertBlock();
const auto cond = eval(extract(bitcast<u32[4]>(as), 0) == 0);
m_ir->CreateCondBr(cond.value, add_block(target), add_block(m_pos + 4));
return;
}
}

// Check sign bit instead (optimization)
if (match_vr<s32[4], s64[2]>(op.rt, [&](auto c, auto MP)
{
Expand Down Expand Up @@ -9510,6 +9571,21 @@ class spu_llvm_recompiler : public spu_recompiler_base, public cpu_translator

const u32 target = spu_branch_target(m_pos, op.i16);

const auto rt = get_vr<u8[16]>(op.rt);

// Checking for zero doeesn't care about the order of the bytes,
// so load the data before it's byteswapped
if (auto [ok, as] = match_expr(rt, byteswap(match<u8[16]>())); ok)
{
if (target != m_pos + 4)
{
m_block->block_end = m_ir->GetInsertBlock();
const auto cond = eval(extract(bitcast<u32[4]>(as), 0) != 0);
m_ir->CreateCondBr(cond.value, add_block(target), add_block(m_pos + 4));
return;
}
}

// Check sign bit instead (optimization)
if (match_vr<s32[4], s64[2]>(op.rt, [&](auto c, auto MP)
{
Expand Down Expand Up @@ -9566,7 +9642,6 @@ class spu_llvm_recompiler : public spu_recompiler_base, public cpu_translator
m_block->block_end = m_ir->GetInsertBlock();
const auto a = get_vr<s8[16]>(op.rt);
const auto cond = eval((bitcast<s16>(trunc<bool[16]>(a)) & 0x3000) == 0);
//const auto cond = eval((m & 0x3000) == 0);
m_ir->CreateCondBr(cond.value, add_block(target), add_block(m_pos + 4));
return true;
}
Expand Down