shader_recompiler: add byteswap pattern workaround for Nvidia

2 files changed, 80 insertions, 0 deletions

diff --git a/src/shader_recompiler/ir_opt/passes.h b/src/shader_recompiler/ir_opt/passes.h
index 629d18fa1..d4d5285e5 100644
--- a/src/shader_recompiler/ir_opt/passes.h
+++ b/src/shader_recompiler/ir_opt/passes.h
@@ -26,6 +26,7 @@ void SsaRewritePass(IR::Program& program);
 void PositionPass(Environment& env, IR::Program& program);
 void TexturePass(Environment& env, IR::Program& program, const HostTranslateInfo& host_info);
 void LayerPass(IR::Program& program, const HostTranslateInfo& host_info);
+void VendorWorkaroundPass(IR::Program& program);
 void VerificationPass(const IR::Program& program);
 
 // Dual Vertex
diff --git a/src/shader_recompiler/ir_opt/vendor_workaround_pass.cpp b/src/shader_recompiler/ir_opt/vendor_workaround_pass.cpp
new file mode 100644
index 000000000..08c658cb8
--- /dev/null
+++ b/src/shader_recompiler/ir_opt/vendor_workaround_pass.cpp
@@ -0,0 +1,79 @@
+// SPDX-FileCopyrightText: Copyright 2023 yuzu Emulator Project
+// SPDX-License-Identifier: GPL-2.0-or-later
+
+#include "shader_recompiler/frontend/ir/basic_block.h"
+#include "shader_recompiler/frontend/ir/ir_emitter.h"
+#include "shader_recompiler/frontend/ir/value.h"
+#include "shader_recompiler/ir_opt/passes.h"
+
+namespace Shader::Optimization {
+
+namespace {
+void AddingByteSwapsWorkaround(IR::Block& block, IR::Inst& inst) {
+    /*
+     * Workaround for an NVIDIA bug seen in Super Mario RPG
+     *
+     * We are looking for this pattern:
+     *   %lhs_bfe = BitFieldUExtract %factor_a, #0, #16
+     *   %lhs_mul = IMul32 %lhs_bfe, %factor_b           // potentially optional?
+     *   %lhs_shl = ShiftLeftLogical32 %lhs_mul, #16
+     *   %rhs_bfe = BitFieldUExtract %factor_a, #16, #16
+     *   %result  = IAdd32 %lhs_shl, %rhs_bfe
+     *
+     * And replacing the IAdd32 with a BitwiseOr32
+     *   %result  = BitwiseOr32 %lhs_shl, %rhs_bfe
+     *
+     */
+    IR::Inst* const lhs_shl{inst.Arg(0).TryInstRecursive()};
+    IR::Inst* const rhs_bfe{inst.Arg(1).TryInstRecursive()};
+    if (!lhs_shl || !rhs_bfe) {
+        return;
+    }
+    if (lhs_shl->GetOpcode() != IR::Opcode::ShiftLeftLogical32 ||
+        lhs_shl->Arg(1) != IR::Value{16U}) {
+        return;
+    }
+    if (rhs_bfe->GetOpcode() != IR::Opcode::BitFieldUExtract || rhs_bfe->Arg(1) != IR::Value{16U} ||
+        rhs_bfe->Arg(2) != IR::Value{16U}) {
+        return;
+    }
+    IR::Inst* const lhs_mul{lhs_shl->Arg(0).TryInstRecursive()};
+    if (!lhs_mul) {
+        return;
+    }
+    const bool lhs_mul_optional{lhs_mul->GetOpcode() == IR::Opcode::BitFieldUExtract};
+    if (lhs_mul->GetOpcode() != IR::Opcode::IMul32 &&
+        lhs_mul->GetOpcode() != IR::Opcode::BitFieldUExtract) {
+        return;
+    }
+    IR::Inst* const lhs_bfe{lhs_mul_optional ? lhs_mul : lhs_mul->Arg(0).TryInstRecursive()};
+    if (!lhs_bfe) {
+        return;
+    }
+    if (lhs_bfe->GetOpcode() != IR::Opcode::BitFieldUExtract) {
+        return;
+    }
+    if (lhs_bfe->Arg(1) != IR::Value{0U} || lhs_bfe->Arg(2) != IR::Value{16U}) {
+        return;
+    }
+    IR::IREmitter ir{block, IR::Block::InstructionList::s_iterator_to(inst)};
+    inst.ReplaceUsesWith(ir.BitwiseOr(IR::U32{inst.Arg(0)}, IR::U32{inst.Arg(1)}));
+}
+
+} // Anonymous namespace
+
+void VendorWorkaroundPass(IR::Program& program) {
+    for (IR::Block* const block : program.post_order_blocks) {
+        for (IR::Inst& inst : block->Instructions()) {
+            switch (inst.GetOpcode()) {
+            case IR::Opcode::IAdd32:
+                AddingByteSwapsWorkaround(*block, inst);
+                break;
+            default:
+                break;
+            }
+        }
+    }
+}
+
+} // namespace Shader::Optimization