summaryrefslogtreecommitdiffstats
path: root/src/shader_recompiler/ir_opt
diff options
context:
space:
mode:
authorReinUsesLisp <reinuseslisp@airmail.cc>2021-04-19 21:33:23 +0200
committerameerj <52414509+ameerj@users.noreply.github.com>2021-07-23 03:51:28 +0200
commit7018e524f5e6217b3259333acc4ea09ad036d331 (patch)
tree58e750b08d48e018accc4de9a05cb483d825904c /src/shader_recompiler/ir_opt
parentspirv: Fix ViewportMask (diff)
downloadyuzu-7018e524f5e6217b3259333acc4ea09ad036d331.tar
yuzu-7018e524f5e6217b3259333acc4ea09ad036d331.tar.gz
yuzu-7018e524f5e6217b3259333acc4ea09ad036d331.tar.bz2
yuzu-7018e524f5e6217b3259333acc4ea09ad036d331.tar.lz
yuzu-7018e524f5e6217b3259333acc4ea09ad036d331.tar.xz
yuzu-7018e524f5e6217b3259333acc4ea09ad036d331.tar.zst
yuzu-7018e524f5e6217b3259333acc4ea09ad036d331.zip
Diffstat (limited to 'src/shader_recompiler/ir_opt')
-rw-r--r--src/shader_recompiler/ir_opt/collect_shader_info_pass.cpp53
-rw-r--r--src/shader_recompiler/ir_opt/global_memory_to_storage_buffer_pass.cpp46
2 files changed, 61 insertions, 38 deletions
diff --git a/src/shader_recompiler/ir_opt/collect_shader_info_pass.cpp b/src/shader_recompiler/ir_opt/collect_shader_info_pass.cpp
index 0500a5141..cccf0909d 100644
--- a/src/shader_recompiler/ir_opt/collect_shader_info_pass.cpp
+++ b/src/shader_recompiler/ir_opt/collect_shader_info_pass.cpp
@@ -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:
@@ -373,7 +375,58 @@ 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;
diff --git a/src/shader_recompiler/ir_opt/global_memory_to_storage_buffer_pass.cpp b/src/shader_recompiler/ir_opt/global_memory_to_storage_buffer_pass.cpp
index 378a3a915..f294d297f 100644
--- a/src/shader_recompiler/ir_opt/global_memory_to_storage_buffer_pass.cpp
+++ b/src/shader_recompiler/ir_opt/global_memory_to_storage_buffer_pass.cpp
@@ -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;
}
}