INSANEMODE/yuzu-android
1
0
Fork 0
mirror of https://github.com/yuzu-emu/yuzu-android.git synced 2025-06-12 07:08:04 -05:00
Code Issues Packages Projects Releases Wiki Activity

shader: Add NVN storage buffer fallbacks

Browse Source 
When we can't track the SSBO origin of a global memory instruction,
leave it as a global memory operation and assume these pointers are in
the NVN storage buffer slots, then apply a linear search in the shader's
runtime.
This commit is contained in:
ReinUsesLisp
2021-04-19 16:33:23 -03:00
committed by ameerj
parent 6325601947
commit 7018e524f5
9 changed files with 214 additions and 62 deletions
Download Patch File Download Diff File Expand all files Collapse all files

53
src/shader_recompiler/ir_opt/collect_shader_info_pass.cpp
View File

@ -187,6 +187,8 @@ void VisitUsages(Info& info, IR::Inst& inst) {
case IR::Opcode::FPUnordGreaterThanEqual16:
case IR::Opcode::FPIsNan16:
case IR::Opcode::GlobalAtomicAddF16x2:
case IR::Opcode::GlobalAtomicMinF16x2:
case IR::Opcode::GlobalAtomicMaxF16x2:
case IR::Opcode::StorageAtomicAddF16x2:
case IR::Opcode::StorageAtomicMinF16x2:
case IR::Opcode::StorageAtomicMaxF16x2:
@ -373,12 +375,63 @@ void VisitUsages(Info& info, IR::Inst& inst) {
case IR::Opcode::StorageAtomicAnd64:
case IR::Opcode::StorageAtomicOr64:
case IR::Opcode::StorageAtomicXor64:
case IR::Opcode::StorageAtomicExchange64:
info.uses_int64 = true;
break;
default:
break;
}
switch (inst.GetOpcode()) {
case IR::Opcode::LoadGlobalU8:
case IR::Opcode::LoadGlobalS8:
case IR::Opcode::LoadGlobalU16:
case IR::Opcode::LoadGlobalS16:
case IR::Opcode::LoadGlobal32:
case IR::Opcode::LoadGlobal64:
case IR::Opcode::LoadGlobal128:
case IR::Opcode::WriteGlobalU8:
case IR::Opcode::WriteGlobalS8:
case IR::Opcode::WriteGlobalU16:
case IR::Opcode::WriteGlobalS16:
case IR::Opcode::WriteGlobal32:
case IR::Opcode::WriteGlobal64:
case IR::Opcode::WriteGlobal128:
case IR::Opcode::GlobalAtomicIAdd32:
case IR::Opcode::GlobalAtomicSMin32:
case IR::Opcode::GlobalAtomicUMin32:
case IR::Opcode::GlobalAtomicSMax32:
case IR::Opcode::GlobalAtomicUMax32:
case IR::Opcode::GlobalAtomicInc32:
case IR::Opcode::GlobalAtomicDec32:
case IR::Opcode::GlobalAtomicAnd32:
case IR::Opcode::GlobalAtomicOr32:
case IR::Opcode::GlobalAtomicXor32:
case IR::Opcode::GlobalAtomicExchange32:
case IR::Opcode::GlobalAtomicIAdd64:
case IR::Opcode::GlobalAtomicSMin64:
case IR::Opcode::GlobalAtomicUMin64:
case IR::Opcode::GlobalAtomicSMax64:
case IR::Opcode::GlobalAtomicUMax64:
case IR::Opcode::GlobalAtomicAnd64:
case IR::Opcode::GlobalAtomicOr64:
case IR::Opcode::GlobalAtomicXor64:
case IR::Opcode::GlobalAtomicExchange64:
case IR::Opcode::GlobalAtomicAddF32:
case IR::Opcode::GlobalAtomicAddF16x2:
case IR::Opcode::GlobalAtomicAddF32x2:
case IR::Opcode::GlobalAtomicMinF16x2:
case IR::Opcode::GlobalAtomicMinF32x2:
case IR::Opcode::GlobalAtomicMaxF16x2:
case IR::Opcode::GlobalAtomicMaxF32x2:
info.uses_int64 = true;
info.uses_global_memory = true;
info.used_constant_buffer_types |= IR::Type::U32 | IR::Type::U32x2;
info.used_storage_buffer_types |= IR::Type::U32 | IR::Type::U32x2 | IR::Type::U32x4;
break;
default:
break;
}
switch (inst.GetOpcode()) {
case IR::Opcode::DemoteToHelperInvocation:
info.uses_demote_to_helper_invocation = true;
break;

46
src/shader_recompiler/ir_opt/global_memory_to_storage_buffer_pass.cpp
View File

@ -11,6 +11,7 @@
#include <boost/container/flat_set.hpp>
#include <boost/container/small_vector.hpp>
#include "common/alignment.h"
#include "shader_recompiler/frontend/ir/basic_block.h"
#include "shader_recompiler/frontend/ir/breadth_first_search.h"
#include "shader_recompiler/frontend/ir/ir_emitter.h"
@ -244,39 +245,6 @@ bool MeetsBias(const StorageBufferAddr& storage_buffer, const Bias& bias) noexce
storage_buffer.offset < bias.offset_end;
}
/// Discards a global memory operation, reads return zero and writes are ignored
void DiscardGlobalMemory(IR::Block& block, IR::Inst& inst) {
IR::IREmitter ir{block, IR::Block::InstructionList::s_iterator_to(inst)};
const IR::Value zero{u32{0}};
switch (inst.GetOpcode()) {
case IR::Opcode::LoadGlobalS8:
case IR::Opcode::LoadGlobalU8:
case IR::Opcode::LoadGlobalS16:
case IR::Opcode::LoadGlobalU16:
case IR::Opcode::LoadGlobal32:
inst.ReplaceUsesWith(zero);
break;
case IR::Opcode::LoadGlobal64:
inst.ReplaceUsesWith(IR::Value{ir.CompositeConstruct(zero, zero)});
break;
case IR::Opcode::LoadGlobal128:
inst.ReplaceUsesWith(IR::Value{ir.CompositeConstruct(zero, zero, zero, zero)});
break;
case IR::Opcode::WriteGlobalS8:
case IR::Opcode::WriteGlobalU8:
case IR::Opcode::WriteGlobalS16:
case IR::Opcode::WriteGlobalU16:
case IR::Opcode::WriteGlobal32:
case IR::Opcode::WriteGlobal64:
case IR::Opcode::WriteGlobal128:
inst.Invalidate();
break;
default:
throw LogicError("Invalid opcode to discard its global memory operation {}",
inst.GetOpcode());
}
}
struct LowAddrInfo {
IR::U32 value;
s32 imm_offset;
@ -350,6 +318,10 @@ std::optional<StorageBufferAddr> Track(const IR::Value& value, const Bias* bias)
.index{index.U32()},
.offset{offset.U32()},
};
if (!Common::IsAligned(storage_buffer.offset, 16)) {
// The SSBO pointer has to be aligned
return std::nullopt;
}
if (bias && !MeetsBias(storage_buffer, *bias)) {
// We have to blacklist some addresses in case we wrongly
// point to them
@ -372,19 +344,17 @@ void CollectStorageBuffers(IR::Block& block, IR::Inst& inst, StorageInfo& info)
// Track the low address of the instruction
const std::optional<LowAddrInfo> low_addr_info{TrackLowAddress(&inst)};
if (!low_addr_info) {
DiscardGlobalMemory(block, inst);
// Failed to track the low address, use NVN fallbacks
return;
}
// First try to find storage buffers in the NVN address
const IR::U32 low_addr{low_addr_info->value};
std::optional storage_buffer{Track(low_addr, &nvn_bias)};
std::optional<StorageBufferAddr> storage_buffer{Track(low_addr, &nvn_bias)};
if (!storage_buffer) {
// If it fails, track without a bias
storage_buffer = Track(low_addr, nullptr);
if (!storage_buffer) {
// If that also failed, drop the global memory usage
// LOG_ERROR
DiscardGlobalMemory(block, inst);
// If that also fails, use NVN fallbacks
return;
}
}