Fix LLVM IR generation when loading Bool types in VCalls

Credits to @mcrescas for finding this

Author: njroussel

src/llvm_eval.cpp

@@ -344,6 +344,8 @@ void jitc_llvm_assemble_func(const char *name, uint32_t inst_id,
                           v, v, v, v);
 
             uint32_t offset = it->second - data_offset;
+            bool is_pointer_or_bool =
+                (vt == VarType::Pointer) || (vt == VarType::Bool);
             // Expand $<..$> only when we are compiling a recursive function call
             callable_depth--;
             fmt( "    $v_p1 = getelementptr inbounds i8, $<{i8*}$> %data, i32 $u\n"
@@ -353,12 +355,15 @@ void jitc_llvm_assemble_func(const char *name, uint32_t inst_id,
                 v, offset,
                 v, v,
                 v, v, v,
-                v, vt == VarType::Pointer ? "_p4" : "", v, v, v, v, v, v);
+                v, is_pointer_or_bool ? "_p4" : "", v, v, v, v, v, v);
             callable_depth++;
 
             if (vt == VarType::Pointer)
                 fmt("    $v = inttoptr <$w x i64> $v_p4 to <$w x {i8*}>\n",
                     v, v);
+            else if (vt == VarType::Bool)
+                fmt("    $v = trunc <$w x i8> $v_p4 to <$w x i1>\n",
+                    v, v);
         } else if (!v->is_stmt()) {
             jitc_llvm_render_var(sv.index, v);
         } else {

tests/vcall.cpp

@@ -916,3 +916,48 @@ TEST_BOTH(12_nested_with_side_effects) {
         jit_registry_trim();
     }
 }
+
+TEST_BOTH(13_load_bool_data) {
+    struct Base {
+        virtual Float f() = 0;
+    };
+    using BasePtr = Array<Base *>;
+
+    struct F1 : Base {
+        Mask cond = dr::opaque<Mask>(true);
+        Float val1 = dr::opaque<Float>(1);
+        Float val2 = dr::opaque<Float>(2);
+        Float f() override {
+            return select(cond, val1, val2);
+        }
+    };
+
+    struct F2 : Base {
+        Mask cond = dr::opaque<Mask>(false);
+        Float val1 = dr::opaque<Float>(3);
+        Float val2 = dr::opaque<Float>(4);
+        Float f() override {
+            return select(cond, val1, val2);
+        }
+    };
+
+    BasePtr self = arange<UInt32>(5) % 3;
+    for (uint32_t i = 0; i < 2; ++i) {
+        jit_set_flag(JitFlag::VCallOptimize, i);
+
+        F1 f1; F2 f2;
+        uint32_t i1 = jit_registry_put(Backend, "Base", &f1);
+        uint32_t i2 = jit_registry_put(Backend, "Base", &f2);
+        jit_assert(i1 == 1 && i2 == 2);
+
+        Float result = vcall(
+            "Base", [](Base *self2) { return self2->f(); }, self);
+
+        jit_var_schedule(result.index());
+        jit_assert(strcmp(result.str(), "[0, 1, 4, 0, 1]") == 0);
+
+        jit_registry_remove(Backend, &f1);
+        jit_registry_remove(Backend, &f2);
+        jit_registry_trim();
+    }
+}