From 8804a4eb23e0c4f3e4bab03dee7c204bd38bf21e Mon Sep 17 00:00:00 2001
From: Billy Laws <blaws05@gmail.com>
Date: Wed, 23 Nov 2022 21:30:23 +0000
Subject: shader_recompiler: Align SSBO offsets to meet host requirements

We can take advantage of SSBO addresses being passed in a constant bufer to account for the extra alignment requirements in the shader itself.
---
 src/shader_recompiler/frontend/maxwell/translate_program.cpp |  2 +-
 src/shader_recompiler/host_translate_info.h                  |  1 +
 .../ir_opt/global_memory_to_storage_buffer_pass.cpp          | 12 ++++++++----
 src/shader_recompiler/ir_opt/passes.h                        |  2 +-
 4 files changed, 11 insertions(+), 6 deletions(-)

diff --git a/src/shader_recompiler/frontend/maxwell/translate_program.cpp b/src/shader_recompiler/frontend/maxwell/translate_program.cpp
index ac159d24b..89fced84c 100644
--- a/src/shader_recompiler/frontend/maxwell/translate_program.cpp
+++ b/src/shader_recompiler/frontend/maxwell/translate_program.cpp
@@ -223,7 +223,7 @@ IR::Program TranslateProgram(ObjectPool<IR::Inst>& inst_pool, ObjectPool<IR::Blo
 
     Optimization::PositionPass(env, program);
 
-    Optimization::GlobalMemoryToStorageBufferPass(program);
+    Optimization::GlobalMemoryToStorageBufferPass(program, host_info);
     Optimization::TexturePass(env, program, host_info);
 
     if (Settings::values.resolution_info.active) {
diff --git a/src/shader_recompiler/host_translate_info.h b/src/shader_recompiler/host_translate_info.h
index d5d279554..7df090124 100644
--- a/src/shader_recompiler/host_translate_info.h
+++ b/src/shader_recompiler/host_translate_info.h
@@ -15,6 +15,7 @@ struct HostTranslateInfo {
     bool needs_demote_reorder{}; ///< True when the device needs DemoteToHelperInvocation reordered
     bool support_snorm_render_buffer{};  ///< True when the device supports SNORM render buffers
     bool support_viewport_index_layer{}; ///< True when the device supports gl_Layer in VS
+    u32 min_ssbo_alignment{};  ///< Minimum alignment supported by the device for SSBOs
 };
 
 } // namespace Shader
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 336338e62..f8d20af3c 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 "shader_recompiler/frontend/ir/breadth_first_search.h"
 #include "shader_recompiler/frontend/ir/ir_emitter.h"
 #include "shader_recompiler/frontend/ir/value.h"
+#include "shader_recompiler/host_translate_info.h"
 #include "shader_recompiler/ir_opt/passes.h"
 
 namespace Shader::Optimization {
@@ -402,7 +403,7 @@ void CollectStorageBuffers(IR::Block& block, IR::Inst& inst, StorageInfo& info)
 }
 
 /// Returns the offset in indices (not bytes) for an equivalent storage instruction
-IR::U32 StorageOffset(IR::Block& block, IR::Inst& inst, StorageBufferAddr buffer) {
+IR::U32 StorageOffset(IR::Block& block, IR::Inst& inst, StorageBufferAddr buffer, u32 alignment) {
     IR::IREmitter ir{block, IR::Block::InstructionList::s_iterator_to(inst)};
     IR::U32 offset;
     if (const std::optional<LowAddrInfo> low_addr{TrackLowAddress(&inst)}) {
@@ -415,7 +416,10 @@ IR::U32 StorageOffset(IR::Block& block, IR::Inst& inst, StorageBufferAddr buffer
     }
     // Subtract the least significant 32 bits from the guest offset. The result is the storage
     // buffer offset in bytes.
-    const IR::U32 low_cbuf{ir.GetCbuf(ir.Imm32(buffer.index), ir.Imm32(buffer.offset))};
+    IR::U32 low_cbuf{ir.GetCbuf(ir.Imm32(buffer.index), ir.Imm32(buffer.offset))};
+
+    // Align the offset base to match the host alignment requirements
+    low_cbuf = ir.BitwiseAnd(low_cbuf, ir.Imm32(~(alignment - 1U)));
     return ir.ISub(offset, low_cbuf);
 }
 
@@ -510,7 +514,7 @@ void Replace(IR::Block& block, IR::Inst& inst, const IR::U32& storage_index,
 }
 } // Anonymous namespace
 
-void GlobalMemoryToStorageBufferPass(IR::Program& program) {
+void GlobalMemoryToStorageBufferPass(IR::Program& program, const HostTranslateInfo& host_info) {
     StorageInfo info;
     for (IR::Block* const block : program.post_order_blocks) {
         for (IR::Inst& inst : block->Instructions()) {
@@ -534,7 +538,7 @@ void GlobalMemoryToStorageBufferPass(IR::Program& program) {
         const IR::U32 index{IR::Value{static_cast<u32>(info.set.index_of(it))}};
         IR::Block* const block{storage_inst.block};
         IR::Inst* const inst{storage_inst.inst};
-        const IR::U32 offset{StorageOffset(*block, *inst, storage_buffer)};
+        const IR::U32 offset{StorageOffset(*block, *inst, storage_buffer, host_info.min_ssbo_alignment)};
         Replace(*block, *inst, index, offset);
     }
 }
diff --git a/src/shader_recompiler/ir_opt/passes.h b/src/shader_recompiler/ir_opt/passes.h
index 1f8f2ba95..4ffad1172 100644
--- a/src/shader_recompiler/ir_opt/passes.h
+++ b/src/shader_recompiler/ir_opt/passes.h
@@ -15,7 +15,7 @@ namespace Shader::Optimization {
 void CollectShaderInfoPass(Environment& env, IR::Program& program);
 void ConstantPropagationPass(Environment& env, IR::Program& program);
 void DeadCodeEliminationPass(IR::Program& program);
-void GlobalMemoryToStorageBufferPass(IR::Program& program);
+void GlobalMemoryToStorageBufferPass(IR::Program& program, const HostTranslateInfo& host_info);
 void IdentityRemovalPass(IR::Program& program);
 void LowerFp16ToFp32(IR::Program& program);
 void LowerInt64ToInt32(IR::Program& program);
-- 
cgit v1.2.3