shader: Add support for "negative" and unaligned offsets

"Negative" offsets don't exist. They are shown as such due to a bug in
nvdisasm.

Unaligned offsets have been proved to read the aligned offset. For
example, when reading an U32, if the offset is 6, the offset read will
be 4.

src/shader_recompiler/backend/glasm/emit_glasm_context_get_set.cpp

@@ -18,6 +18,14 @@ void GetCbuf(EmitContext& ctx, IR::Inst& inst, const IR::Value& binding, ScalarU
         throw NotImplementedException("Indirect constant buffer loading");
     }
     const Register ret{ctx.reg_alloc.Define(inst)};
+    if (offset.type == Type::U32) {
+        // Avoid reading arrays out of bounds, matching hardware's behavior
+        const u32 imm_offset{offset.imm_u32};
+        if (offset.imm_u32 >= 0x10'000) {
+            ctx.Add("MOV.S {},0;", ret);
+            return;
+        }
+    }
     ctx.Add("LDC.{} {},c{}[{}];", size, ret, binding.U32(), offset);
 }
 

src/shader_recompiler/backend/spirv/emit_spirv_context_get_set.cpp

@@ -122,7 +122,7 @@ std::optional<OutAttr> OutputAttrPointer(EmitContext& ctx, IR::Attribute attr) {
 }
 
 Id GetCbuf(EmitContext& ctx, Id result_type, Id UniformDefinitions::*member_ptr, u32 element_size,
-           const IR::Value& binding, const IR::Value& offset, bool check_alignment = true) {
+           const IR::Value& binding, const IR::Value& offset) {
     if (!binding.IsImmediate()) {
         throw NotImplementedException("Constant buffer indexing");
     }
@@ -138,17 +138,14 @@ Id GetCbuf(EmitContext& ctx, Id result_type, Id UniformDefinitions::*member_ptr,
         const Id access_chain{ctx.OpAccessChain(uniform_type, cbuf, ctx.u32_zero_value, index)};
         return ctx.OpLoad(result_type, access_chain);
     }
-    if (check_alignment && offset.U32() % element_size != 0) {
-        throw NotImplementedException("Unaligned immediate constant buffer load");
-    }
+    // Hardware been proved to read the aligned offset (e.g. LDC.U32 at 6 will read offset 4)
     const Id imm_offset{ctx.Const(offset.U32() / element_size)};
     const Id access_chain{ctx.OpAccessChain(uniform_type, cbuf, ctx.u32_zero_value, imm_offset)};
     return ctx.OpLoad(result_type, access_chain);
 }
 
 Id GetCbufU32x4(EmitContext& ctx, const IR::Value& binding, const IR::Value& offset) {
-    return GetCbuf(ctx, ctx.U32[4], &UniformDefinitions::U32x4, sizeof(u32[4]), binding, offset,
-                   false);
+    return GetCbuf(ctx, ctx.U32[4], &UniformDefinitions::U32x4, sizeof(u32[4]), binding, offset);
 }
 
 Id GetCbufElement(EmitContext& ctx, Id vector, const IR::Value& offset, u32 index_offset) {

src/shader_recompiler/frontend/maxwell/translate/impl/impl.cpp

@@ -122,14 +122,14 @@ IR::F64 TranslatorVisitor::GetDoubleReg39(u64 insn) {
 static std::pair<IR::U32, IR::U32> CbufAddr(u64 insn) {
     union {
         u64 raw;
-        BitField<20, 14, s64> offset;
+        BitField<20, 14, u64> offset;
         BitField<34, 5, u64> binding;
     } const cbuf{insn};
 
     if (cbuf.binding >= 18) {
         throw NotImplementedException("Out of bounds constant buffer binding {}", cbuf.binding);
     }
-    if (cbuf.offset >= 0x10'000 || cbuf.offset < 0) {
+    if (cbuf.offset >= 0x10'000) {
         throw NotImplementedException("Out of bounds constant buffer offset {}", cbuf.offset);
     }
     const IR::Value binding{static_cast<u32>(cbuf.binding)};