10 files changed, 37 insertions, 13 deletions

diff --git a/src/shader_recompiler/backend/glasm/emit_glasm_memory.cpp b/src/shader_recompiler/backend/glasm/emit_glasm_memory.cpp
index 2705ab140..9319ea007 100644
--- a/src/shader_recompiler/backend/glasm/emit_glasm_memory.cpp
+++ b/src/shader_recompiler/backend/glasm/emit_glasm_memory.cpp
@@ -5,6 +5,7 @@
 #include "shader_recompiler/backend/glasm/glasm_emit_context.h"
 #include "shader_recompiler/frontend/ir/program.h"
 #include "shader_recompiler/frontend/ir/value.h"
+#include "shader_recompiler/profile.h"
 #include "shader_recompiler/runtime_info.h"
 
 namespace Shader::Backend::GLASM {
@@ -35,7 +36,9 @@ void GlobalStorageOp(EmitContext& ctx, Register address, bool pointer_based, std
             continue;
         }
         const auto& ssbo{ctx.info.storage_buffers_descriptors[index]};
-        ctx.Add("LDC.U64 DC.x,c{}[{}];"    // ssbo_addr
+        const u64 ssbo_align_mask{~(ctx.profile.min_ssbo_alignment - 1U)};
+        ctx.Add("LDC.U64 DC.x,c{}[{}];"    // unaligned_ssbo_addr
+                "AND.U64 DC.x,DC.x,{};"    // ssbo_addr = unaligned_ssbo_addr & ssbo_align_mask
                 "LDC.U32 RC.x,c{}[{}];"    // ssbo_size_u32
                 "CVT.U64.U32 DC.y,RC.x;"   // ssbo_size = ssbo_size_u32
                 "ADD.U64 DC.y,DC.y,DC.x;"  // ssbo_end = ssbo_addr + ssbo_size
@@ -44,8 +47,8 @@ void GlobalStorageOp(EmitContext& ctx, Register address, bool pointer_based, std
                 "AND.U.CC RC.x,RC.x,RC.y;" // cond = a && b
                 "IF NE.x;"                 // if cond
                 "SUB.U64 DC.x,{}.x,DC.x;", // offset = input_addr - ssbo_addr
-                ssbo.cbuf_index, ssbo.cbuf_offset, ssbo.cbuf_index, ssbo.cbuf_offset + 8, address,
-                address, address);
+                ssbo.cbuf_index, ssbo.cbuf_offset, ssbo_align_mask, ssbo.cbuf_index,
+                ssbo.cbuf_offset + 8, address, address, address);
         if (pointer_based) {
             ctx.Add("PK64.U DC.y,c[{}];"      // host_ssbo = cbuf
                     "ADD.U64 DC.x,DC.x,DC.y;" // host_addr = host_ssbo + offset
diff --git a/src/shader_recompiler/backend/glsl/glsl_emit_context.cpp b/src/shader_recompiler/backend/glsl/glsl_emit_context.cpp
index 9ff4028c2..b2ceeefc4 100644
--- a/src/shader_recompiler/backend/glsl/glsl_emit_context.cpp
+++ b/src/shader_recompiler/backend/glsl/glsl_emit_context.cpp
@@ -1,6 +1,7 @@
 // SPDX-FileCopyrightText: Copyright 2021 yuzu Emulator Project
 // SPDX-License-Identifier: GPL-2.0-or-later
 
+#include "common/div_ceil.h"
 #include "shader_recompiler/backend/bindings.h"
 #include "shader_recompiler/backend/glsl/glsl_emit_context.h"
 #include "shader_recompiler/frontend/ir/program.h"
@@ -431,9 +432,11 @@ void EmitContext::DefineConstantBuffers(Bindings& bindings) {
     }
     for (const auto& desc : info.constant_buffer_descriptors) {
         const auto cbuf_type{profile.has_gl_cbuf_ftou_bug ? "uvec4" : "vec4"};
+        const u32 cbuf_used_size{Common::DivCeil(info.constant_buffer_used_sizes[desc.index], 16U)};
+        const u32 cbuf_binding_size{info.uses_global_memory ? 0x1000U : cbuf_used_size};
         header += fmt::format("layout(std140,binding={}) uniform {}_cbuf_{}{{{} {}_cbuf{}[{}];}};",
                               bindings.uniform_buffer, stage_name, desc.index, cbuf_type,
-                              stage_name, desc.index, 4 * 1024);
+                              stage_name, desc.index, cbuf_binding_size);
         bindings.uniform_buffer += desc.count;
     }
 }
@@ -601,7 +604,10 @@ std::string EmitContext::DefineGlobalMemoryFunctions() {
             addr_xy[i] = fmt::format("ftou({}[{}].{})", cbuf, addr_loc / 16, Swizzle(addr_loc));
             size_xy[i] = fmt::format("ftou({}[{}].{})", cbuf, size_loc / 16, Swizzle(size_loc));
         }
-        const auto addr_pack{fmt::format("packUint2x32(uvec2({},{}))", addr_xy[0], addr_xy[1])};
+        const u32 ssbo_align_mask{~(static_cast<u32>(profile.min_ssbo_alignment) - 1U)};
+        const auto aligned_low_addr{fmt::format("{}&{}", addr_xy[0], ssbo_align_mask)};
+        const auto aligned_addr{fmt::format("uvec2({},{})", aligned_low_addr, addr_xy[1])};
+        const auto addr_pack{fmt::format("packUint2x32({})", aligned_addr)};
         const auto addr_statment{fmt::format("uint64_t {}={};", ssbo_addr, addr_pack)};
         func += addr_statment;
 
diff --git a/src/shader_recompiler/backend/spirv/emit_spirv.cpp b/src/shader_recompiler/backend/spirv/emit_spirv.cpp
index 34592a01f..0031fa5fb 100644
--- a/src/shader_recompiler/backend/spirv/emit_spirv.cpp
+++ b/src/shader_recompiler/backend/spirv/emit_spirv.cpp
@@ -407,7 +407,7 @@ void SetupCapabilities(const Profile& profile, const Info& info, EmitContext& ct
         }
         ctx.AddCapability(spv::Capability::DemoteToHelperInvocation);
     }
-    if (info.stores[IR::Attribute::ViewportIndex]) {
+    if (info.stores[IR::Attribute::ViewportIndex] && profile.support_multi_viewport) {
         ctx.AddCapability(spv::Capability::MultiViewport);
     }
     if (info.stores[IR::Attribute::ViewportMask] && profile.support_viewport_mask) {
diff --git a/src/shader_recompiler/backend/spirv/emit_spirv_context_get_set.cpp b/src/shader_recompiler/backend/spirv/emit_spirv_context_get_set.cpp
index 1d77426e0..e5a78a914 100644
--- a/src/shader_recompiler/backend/spirv/emit_spirv_context_get_set.cpp
+++ b/src/shader_recompiler/backend/spirv/emit_spirv_context_get_set.cpp
@@ -84,6 +84,10 @@ std::optional<OutAttr> OutputAttrPointer(EmitContext& ctx, IR::Attribute attr) {
         }
         return std::nullopt;
     case IR::Attribute::ViewportIndex:
+        if (!ctx.profile.support_multi_viewport) {
+            LOG_WARNING(Shader, "Ignoring viewport index store on non-supporting driver");
+            return std::nullopt;
+        }
         if (ctx.profile.support_viewport_index_layer_non_geometry ||
             ctx.stage == Shader::Stage::Geometry) {
             return OutAttr{ctx.viewport_index, ctx.U32[1]};
diff --git a/src/shader_recompiler/backend/spirv/spirv_emit_context.cpp b/src/shader_recompiler/backend/spirv/spirv_emit_context.cpp
index 57df6fc34..3350f1f85 100644
--- a/src/shader_recompiler/backend/spirv/spirv_emit_context.cpp
+++ b/src/shader_recompiler/backend/spirv/spirv_emit_context.cpp
@@ -891,7 +891,9 @@ void EmitContext::DefineGlobalMemoryFunctions(const Info& info) {
             const Id ssbo_size_pointer{OpAccessChain(uniform_types.U32, cbufs[ssbo.cbuf_index].U32,
                                                      zero, ssbo_size_cbuf_offset)};
 
-            const Id ssbo_addr{OpBitcast(U64, OpLoad(U32[2], ssbo_addr_pointer))};
+            const u64 ssbo_align_mask{~(profile.min_ssbo_alignment - 1U)};
+            const Id unaligned_addr{OpBitcast(U64, OpLoad(U32[2], ssbo_addr_pointer))};
+            const Id ssbo_addr{OpBitwiseAnd(U64, unaligned_addr, Constant(U64, ssbo_align_mask))};
             const Id ssbo_size{OpUConvert(U64, OpLoad(U32[1], ssbo_size_pointer))};
             const Id ssbo_end{OpIAdd(U64, ssbo_addr, ssbo_size)};
             const Id cond{OpLogicalAnd(U1, OpUGreaterThanEqual(U1, addr, ssbo_addr),
diff --git a/src/shader_recompiler/frontend/maxwell/translate_program.cpp b/src/shader_recompiler/frontend/maxwell/translate_program.cpp
index 8fac6bad3..321ea625b 100644
--- a/src/shader_recompiler/frontend/maxwell/translate_program.cpp
+++ b/src/shader_recompiler/frontend/maxwell/translate_program.cpp
@@ -298,7 +298,7 @@ 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 7d2ded907..1b53404fc 100644
--- a/src/shader_recompiler/host_translate_info.h
+++ b/src/shader_recompiler/host_translate_info.h
@@ -16,6 +16,7 @@ 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
     bool support_geometry_shader_passthrough{}; ///< True when the device supports geometry
                                                 ///< passthrough shaders
     bool support_conditional_barrier{}; ///< True when the device supports barriers in conditional
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 d1e59f22e..0cea79945 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 {
@@ -408,7 +409,7 @@ 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)}) {
@@ -421,7 +422,10 @@ 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);
 }
 
@@ -516,7 +520,7 @@ 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()) {
@@ -540,7 +544,8 @@ 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 d4d5285e5..1e637cb23 100644
--- a/src/shader_recompiler/ir_opt/passes.h
+++ b/src/shader_recompiler/ir_opt/passes.h
@@ -16,7 +16,7 @@ void CollectShaderInfoPass(Environment& env, IR::Program& program);
 void ConditionalBarrierPass(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 LowerFp64ToFp32(IR::Program& program);
 void LowerFp16ToFp32(IR::Program& program);
diff --git a/src/shader_recompiler/profile.h b/src/shader_recompiler/profile.h
index 38d820db2..66901a965 100644
--- a/src/shader_recompiler/profile.h
+++ b/src/shader_recompiler/profile.h
@@ -43,6 +43,7 @@ struct Profile {
     bool support_gl_sparse_textures{};
     bool support_gl_derivative_control{};
     bool support_scaled_attributes{};
+    bool support_multi_viewport{};
 
     bool warp_size_potentially_larger_than_guest{};
 
@@ -84,6 +85,8 @@ struct Profile {
 
     /// Maxwell and earlier nVidia architectures have broken robust support
     bool has_broken_robust{};
+
+    u64 min_ssbo_alignment{};
 };
 
 } // namespace Shader